CLASSIFIERS

OXFORD STUDIES IN TYPOLOGY AND LINGUISTIC

THEORY

SERIES EDITORS: Ronnie Cann, University of Edinburgh, William Croft, University of

Manchester, Mark Durie University of Melbourne, Anna Siewierska, University of
Lancaster
This series offers a forum for orginal and accessible books on language typology
and linguistic universals. Works published will be theoretically innovative and
informed and will seek to link theory and empirical research in ways that are
mutually productive. Each volume will also provide the reader with a wide range
of cross-linguistic data. The series is open to typological work in semantics, syntax,
phonology, and phonetics or at the interfaces between these fields.

Published:
Indefinite Pronouns
Martin Haspelmath
Intransitive Predication
Leon Stassen
Classifiers: A Typology of Noun Categorization Devices
Alexandra A. Aikhenvald
Anaphora
Yan Huang
In preparation:
The Noun Phrase
Jan Rijkhoff
Double Object Constructions
Maria Polinsky
CLASSIFIERS

A Typology of Noun
Categorization Devices
ALEXANDRA Y. AIKHENVALD

OXFORD
UNIVERSITY PRESS
 OXFORD
UNIVERSITY PRESS
Great Clarendon Street, Oxford 0x2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University's objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Athens Auckland Bangkok Bogota Buenos Aires Calcutta
Cape Town Chennai Dar es Salaam Delhi Florence Hong Kong Istanbul
Karachi Kuala Lumpur Madrid Melbourne Mexico City Mumbai
Nairobi Paris Sao Paulo Singapore Taipei Tokyo Toronto Warsaw
and associated companies in Berlin Ibadan
Oxford is a registered trade mark of Oxford University Press
in the UK and certain other countries
Published in the United States
by Oxford University Press Inc., New York
© A. Y. Aikhenvald 2000
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2000
All rights reserved, No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organizations. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same conditions on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data applied for

ISBN 0-19-823886-X
1 3 5 7 9 1 08 6 4 2
Typeset in Times
by J&L Composition Ltd, Filey, North Yorkshire
Printed in Great Britain
on acid-free paper by
Biddies Ltd, Guildford & King's Lynn
For Bob, okojibotee
This page intentionally left blank
 Preface

This book aims at providing a cross-linguistic analysis of noun classification

systems across the languages of the world, also dealing with a variety of
other problems such as the morphological status of the markers of these
categories, agreement phenomena, and the syntactic and semantic
classification of adjectives and numbers. It is generally accepted that
linguistic categorization of nouns is a reflection of human mind and culture.
The present study thus has far-reaching implications for cross-cultural as
well as cross-linguistic studies of human cognition, and will provide new
insights concerning the mechanisms by which human language functions.
Languages with extensive systems of noun classification devices,
especially those which combine classifiers and genders, present a true
challenge for the typologist. My first encounter with these unusual systems
was through fieldwork on Tariana and Baniwa, two closely related North
Arawak languages spoken in Northwest Amazonia. The more I worked on
the topic, the more exotic and unusual systems I encountered, especially
among little-known South American languages, and languages of the
South Pacific. This book came into being as an attempt to integrate these
systems into a cross-linguistically based typological framework.
This study is an up-to-date introduction to the field, and will be of value
not only to a wide variety of linguists and linguistic students but also to
anthropologists, cognitive psychologists, and philosophers who are interested
in language and the mind. It can be used both as a sourcebook for further
typological studies, and as a textbook. The discussion in the book is in
terms of basic linguistic theory, the framework of linguistic analysis in
terms of which most grammars are cast, and in terms of which significant
typological generalizations are postulated. (I have avoided using any of the
more specific formalisms, which come and go with such frequency.)
Some terminological clarifications are in order. First, my conception of a
lexical entry for 'noun' roughly corresponds to the notion of 'lexeme' as
outlined by Lyons (1977 vol. 1: 19). Second, throughout the book 'linguistic
categorization of a noun' is used to mean 'linguistic categorization of the
referent of a noun', just as in many linguistic usages 'human noun' is a short
way of saying 'noun with a human referent'. Third, the term 'noun categor-
ization' is used here in a sense close to the 'noun classification' (cf. Craig
1986a; Derbyshire and Payne 1990) or 'nominal classification' (cf. Harvey and
Reid 1997) employed by other authors. The term 'classifier system' refers to a
grammatical system of noun categorization device(s) in a particular language.
viii Preface
In order to limit the book to a reasonable size, I have only been able to
refer to a portion of the available literature. There are many other sources
that I have consulted, which only provide additional exemplification for
points that are already well covered. When a language is introduced for the
first time, its genetic affiliation and the source of information on it are
given in parentheses; further on, this information is only repeated where
relevant. Examples, tables and diagrams are numbered separately within
each chapter.
The orthography used in the examples and language names follows that
of the sources (unless indicated otherwise).
A study like this could only be definitive when good and thorough
descriptions have been provided for most of the world's languages; we are
at present a long way from this situation. Nevertheless, I hope that this
study will provide a framework within which fieldworkers and typologists
will be able to work, and which can be amended and adjusted as new data
and new insights emerge.
It is my hope that this book will encourage people to study noun
classification devices, especially in little-known or undescribed languages,
going out into the field and documenting languages threatened by extinc-
tion (before it is too late to do so).
 Acknowledgements

My gratitude goes to all those native speakers who taught me their

languages and their unusual classifier and gender systems: Candido, Jose,
Jovino, Graciliano, and Olivia Brito (Tariana); Humberto Baltazar and
Pedro Angelo Tomas (Warekena); the late Candelario da Silva (Bare);
Afonso, Albino and Joao Fontes, Celestino da Silva, Cecilia and Laureano
da Silva, and the late Marcilia Rodrigues (Baniwa); the late Tiago Cardoso
(Desano, Piratapuya); Alfredo Fontes (Tucano); Marilda and Carlito
Paumari (Paumari); Raimunda Palikur (Palikur); Simone Nientao
(Tamachek) and—last but not least—Pauline and James Laki (Manambu).
I am also indebted to students in the Federal University of Santa Catarina,
Brazil, and in the Australian National University. I learned a lot from
working with Rute Amorim, Lilias Chun, Christiane Cunha de Oliveira,
Tim Curnow, Michael Dunn, Catriona Hyslop, Dorothy Jauncey, Yunseok
Lee, Eva Lindstrom, Peita Littleton, Rina Marnita, Silvana Martins,
Kazuko Obata, Kristina Sands, Eva Tatrai, Angela Terrill, Simoni
Valadares, and Jacki Wicks. My warmest thanks go to Silvana and Valteir
Martins and Lenita and Elias Coelho de Assis, without whose friendship
and assistance a great deal of my fieldwork would have been impossible.
Special gratitude goes to Diana Green, who revealed to me the beauty of
genders and classifiers in Palikur.
I am most grateful to those people who helped me by sending copies of
their papers, answering my questions and commenting on various parts of
this manuscript: Jose Alvarez, Mengistu Amberber, Felix Ameka, Peter
van Baarle, Janet Barnes, Candida Barros, Edith Bavin, Walter Bisang,
Kim Blewett, Paula Boley, John Boyle, Friederike Braun, Lea Brown, Bill
Callister, Lyle Campbell, Eugene Casad, Meiyun Chang-Smith, Adam
Chapman, Shirley Chapman, Hilary Chappell, Helen Charters, Bernard
Comrie, Bob Conrad, Grev Corbett, Tim Curnow, Des Derbyshire, Connie
Dickinson, Tony Diller, Gerritt Dimmendaal, Mark Donohue, Nancy
Dorian, Mark Durie, Tom Dutton, Nora England, Nick Evans, Cindi
and Jim Farr, Bill Foley, Lys Ford, David Foris, Paul Frank, David Gil,
Cliff Goddard, Elsa Gomez-Imbert, Ian Green, Rebecca Green, Colette
Grinevald (Craig), Geoff Haig, Mark Harvey, Rie Hasada, Bernd Heine,
Debbie Hill, Chu-Ren Huang, Rodney Huddleston, Suanu Ikoro, Liisa
Jarvinen, Jae Jung Song, Aleksandr J. Kibrik, Harold Koch, Antonina I.
Koval', Randy LaPolla, Jason Lee, Jennie Lee, Jeff Leer, W. P. Lehmann,
Adrienne Lehrer, Frank Lichtenberk, Eva Lindstrom, Elizabeth Lobel,
x Acknowledgements
Ivan Lowe, Harriet Manelis-Klein, Jack Martin, Marianne Mithun,
Catherine McGuckin, Ulrike Mosel, Otto Nekitel, Bee Chin Ng, Johanna
Nichols, Masayuki Onishi, Patricia Pacioni, Helma Pasch, Peter Paul,
Andrew Pawley, David Payne, Vladimir Plungian, Bill Poser, Kostantin
Pozdnjakov, Bruce Rigsby, Phil Quick, Susan Quigley, Katya Rakhilina,
Henri Ramirez, Nick Reid, Keren Rice, Aryon Rodrigues, Malcolm Ross,
Carl Rubino, Alan Rumsey, Filomena Sandalo, Risto Sarsa, Hansjakob
Seiler, Lucy Seki, Gunter Senft, Beatriz and Rodolfo Senn, Gi-Hyun Shin,
Tim Shopen, Edgar Suter, Chad Thompson, Irina Toporova, Joe Tsonope,
Ione Vasconcelos, Alejandra Vidal, Tiit-Rein Viitso, Viktor A. Vinogradov,
Julie Waddy, Bruce Waters, Laurel Watkins, Anna Wierzbicka, Mary Ruth
Wise, Stephen Wurm, and Roberto Zavala.
I am most grateful to those who read through the whole draft of this book,
or parts of it, and provided comments, corrections and ideas—Peter Denny,
Nancy Dorian, David Foris, Cliff Goddard, Nikolaus Himmelmann, Frank
Lichtenberk, Edith Moravcsik, Patricia Pacioni, Helma Pasch, Doris Payne,
Nick Reid, Malcolm Ross, Fritz Serzisko, and Roberto Zavala. Invaluable
comments on almost every page came from R. M. W. Dixon, Keith Allan,
Walter Bisang, Lyle Campbell, Gerritt Dimmendaal, Mark Durie, Ulrike
Mosel, and Gunter Senft.
My deepest gratitude goes to the SIL Library in Ukarumpa, and to Paul
Frank, the director of the Summer Institute of Linguistics (Colombia),
who provided me with invaluable materials on languages of Papua New
Guinea and Colombia respectively.
I am also grateful to the members of Eesti Noorte Grupp of Canberra—
Reet Bergman, Krista Gardiner, and Reet Vallak—who helped me realize
that communicating in a language without genders or classifiers can be
great fun.
Suzanne Kite carefully read through several drafts of this book and
corrected it with her usual skill, dedication, and good humour. Thanks
are equally due to her.
Jennifer Elliott provided a wonderful working atmosphere at the
Research Centre for Linguistic Typology. This book would have been
scarcely possible without her.
 Plea

This book is far from being the last word on noun categorization devices. I
welcome reactions, counterexamples, new ideas and data, to further
develop, refine, and improve the generalizations put forward here. Please
send them to me at Research Centre for Linguistic Typology, La Trobe
University, Bundoora Vic., 3083 Australia.
This page intentionally left blank
 Contents

List of Maps xx
List of Tables xxi
List of Diagrams xxiv
List of Abbreviations xxv
1. Preliminaries 1
1.1 General remarks 1
1.2 Classifiers: an illustration 1
1.3 Theoretical framework, data, and sources 4
1.4 Approaches to the typology of classifiers 5
1.5 Parameters for the typology of classifiers 13
1.6 The structure of this book 16
2. Noun Class and Gender Systems 19
2.1 General remarks 19
2.2 Properties of noun class systems 20
2.3 Principles of noun class assignment 22
2.3.1 Semantic assignment 22
2.3.2 Morphological assignment 25
2.3.3 Phonological assignment 25
2.3.4 Mixed principles of assignment 25
2.4 Noun classes and agreement 28
2.4.1 A working definition of agreement and
agreement properties 28
2.4.2 Principles of noun class agreement 31
2.4.3 Variability in noun class assignment and
variable agreement 41
2.4.4 Determining the number of noun classes in
a language 45
2.5 Markedness and resolution in noun classes 50
2.5.1 Markedness 50
2.5.2 Noun class resolution 52
2.5.3 Markedness relationships in noun classes 54
2.6 Realization of noun classes 56
2.6.1 Overt and covert noun class marking 57
2.6.2 Morphological realization of noun classes 58
2.6.3 Double marking of noun classes 63
xiv Contents
2.7 Languages with more than one kind of noun class 67
2.7.1 Nominal and pronominal noun class 68
2.7.2 Different kinds of noun class in the same
environment 70
2.7.3 Languages with more than one kind of noun class:
a summary 76
2.8 Distribution of noun classes in the languages of the
world 77
3. Noun Classifiers 81
3.1 Properties of noun classifiers 81
3.2 Noun classifiers: discussion and exemplification 82
3.2.1 The choice of noun classifiers and the
cooccurrence of several classifiers within one
noun phrase 82
3.2.2 Semantic functions of noun classifiers 84
3.2.3 Size of inventory and degree of
grammaticalization of noun classifiers 84
3.2.4 Syntactic functions of noun classifiers 87
3.3 Noun classifiers and numeral classifiers 90
3.4 Realization and grammaticalization of noun classifiers 91
3.5 Overt noun class marking and noun classifiers 92
3.6 Distribution of noun classifiers in the languages of
the world 97
4. Numeral Classifiers 98
4.1 Properties of numeral classifiers 98
4.2 Numeral classifier constructions and morphological
realization of numeral classifiers 101
4.2.1 Numeral classifiers as independent lexemes 101
4.2.2 Numeral classifiers attached to numerals 105
4.2.3 Numeral classifiers attached to the head noun 110
4.3 Languages with more than one morphological type of
numeral classifier 112
4.3.1 Different types of numeral classifier in
complementary distribution 112
4.3.2 Different types of numeral classifier which
occur together 113
4.4 Problems with numeral classifiers 114
4.4.1 Mensural and sortal classifiers: distinguishing
classifiers from quantifying expressions 114
4.4.2 Incipient numeral classifiers 120
4.5 Distribution of numeral classifiers in the languages of
the world 121
 Contents xv
5. Classifiers in Possessive Constructions 125
5.1 Categorization in possessive constructions 125
5.2 Possessed classifiers 126
5.3 Relational classifiers 133
5.3.1 Relational classifiers and their properties 133
5.3.2 Types of possession and relational classifiers 137
5.4 Possessor classifiers 139
5.5 Interaction of possessed and relational classifiers 140
5.5.1 Integrating relational and possessive classifiers 140
5.5.2 Languages with two types of classifier in
possessive constructions 142
5.6 Contrasting classifiers in possessive constructions 144
5.7 Distribution of classifiers in possessive constructions
in the languages of the world 147

6. Verbal Classifiers 149

6.1 Properties of verbal classifiers 149
6.2 Realization of verbal classifiers 149
6.2.1 Classificatory noun incorporation 150
6.2.2 Verbal classifiers as affixes 152
6.2.3 Suppletive 'classificatory verbs' 153
6.2.4 The interaction of the three types of verbal
classifier 160
6.3 Verbal classifiers and syntactic function of the argument 162
6.4 Combinations of different types of verbal classifier 163
6.4.1 Different types of verbal classifier in complementary
distribution 163
6.4.2 Distinct systems of verbal classifiers 167
6.5 Distribution of verbal classifiers 169
7. Locative and Deictic Classifiers 172
7.1 The structure of this chapter 172
7.2 Properties of locative classifiers 172
7.3 Properties of deictic classifiers 176
7.3.1 Examples of deictic classifiers 177
7.3.2 Conclusions and discussion 181
8. Different Classifier Types in One Language 184
8.1 General observations 184
8.2 Coexisting classifier sets in different environments 185
8.3 Different classifier sets in the same environment 198
8.4 Conclusions 201
xvi Contents
9. Multiple Classifier Languages 204
9.1 Noun categorization in multiple classifier languages 204
9.2 Multiple classifier languages and noun class agreement
on multiple targets 228
9.3 Fuzzy types: overlapping classifiers in multiple
environments 230
9.3.1 Multiple classifiers in Baniwa 230
9.3.2 Multiple classifiers in Tariana 235
9.3.3 Fuzzy types and borderline cases 240
10. Classifiers and Other Grammatical Categories 242
10.1 Classifiers and number 243
10.1.1 Noun classes and number 243
10.1.2 Number and other classifier types 249
10.2 Classifiers and person 252
10.3 Classifiers and grammatical function 255
10.3.1 Noun classes and grammatical function 255
10.3.2 Verbal classifiers and grammatical function 257
10.4 Classifiers and types of possession 257
10.4.1 Noun classes and types of possession 258
10.4.2 Classifiers in possessive constructions and types
of possession 259
10.5 Classifiers and politeness 260
10.6 Classifiers and declensional classes 262
10.7 Classifiers and verbal categories 263
10.8 Classifiers and deictic categories 266
10.9 Classifiers, derivation, and lexicon 266
10.10 Conclusions 268
11. Semantics of Noun Categorization Devices 271
11.1 Semantic parameters in noun categorization 271
11.1.1 Basic parameters of categorization 271
11.1.2 Additional semantic characteristics 274
11.1.3 Semantic relationship between a classifier and
the referent 275
11.2 Semantics of classifier types 275
11.2.1 Semantics of noun classes 275
11.2.2 Semantics of noun classifiers 283
11.2.3 Semantics of numeral classifiers 286
11.2.4 Semantics of classifiers in possessive
constructions 293
11.2.5 Semantics of verbal classifiers 295
 Contents xvii
11.2.6 Semantics of locative and of deictic classifiers 300
11.2.7 Semantic parameters in languages with several
different types of classifier 301
11.2.8 Semantic parameters in multiple classifier
systems 303
11.2.9 Conclusions 305
12. Semantic Organization and Functions of Noun
Categorization 307
12.1 Semantic organization and functions of classifier
systems 307
12.1.1 Semantic complexity in classifier systems 308
12.1.2 Semantic roles of classifiers 317
12.1.3 Discourse-pragmatic functions of classifiers 320
12.1.4 Applicability of classifiers and default classes 334
12.2 Human cognition and classifiers 337
12.2.1 Perceptual correlates of noun categorization 337
12.2.2 Cognitive mechanisms and noun categorization 339
12.3 Social and cultural issues in noun categorization 340
12.3.1 Social structure in noun categorization 342
12.3.2 Environment and culture in noun categorization 343
12.3.3 Culture-specific metaphorical extensions 346
12.3.4 Socio-cultural motivations for change in noun
categorization 347
12.4 Conclusions 350

13. Origin and Development of Noun Categorization Devices 352

13.1 Lexical sources for classifiers 353
13.1.1 From a noun to a classifier 353
13.1.2 Repeater phenomena and the origin of classifier
constructions 361
13.1.3 From a verb to a classifier 362
13.1.4 Classifiers from deverbal nominalizations 365
13.1.5 Classifiers of mixed origin 366
13.2 From a closed class to a noun categorization system 367
13.3 Languages with several classifier types, and the relative
age of noun categorization devices 370
13.4 Internal evolution of noun categorization 372
13.5 Grammaticalization and reanalysis in noun
categorization systems 374
13.5.1 Grammaticalization in the development of noun
categorization 374
13.5.2 Reanalysis in noun categorization 377
xviii Contents
13.6 Reduction and loss of noun categorization devices 379
13.7 Language-external motivations for the development
and decay of noun categorization 382
13.7.1 Language contact and noun categorization 383
13.7.2 Creolization and noun categorization 388
13.7.3 Language obsolescence and noun categorization 389
13.7.4 Language-external motivations and their impact
on noun categorization 391
13.8 Development and loss of agreement 391
13.8.1 The genesis and development of agreement 391
13.8.2 Decline and loss of agreement 398
13.9 Semantic changes in noun categorization devices 400
13.9.1 From lexical item to classifier: principles of
semantic change 401
13.9.2 Further changes in noun categorization devices 407
13.10 Sources of noun categorization devices: a summary 411
14. Noun Categorization Devices in Language Acquisition and
Dissolution 413
14.1 Acquisition and development of noun classes 413
14.2 Acquisition of numeral classifiers 417
14.3 Dissolution of noun classes and of numeral classifiers 422
14.4 Conclusions 423

15. Conclusions 425

15.1 Properties of classifier types 425
15.2 Cooccurrence of classifier types and multiple classifier
languages; prototypes and continua 432
15.3 Prospects for future studies 434
Appendix 1. Noun Categorization by Means Other than
Classifiers 436
Appendix 2. From Nouns to Classifiers: Further Examples of
Semantic Change 442
(A) Body parts as sources for classifiers: semantic
extensions 442
(B) Sources for shape-based numeral classifiers 446
Appendix 3. Fieldworker's Guide to Classifier Languages 447
References 452
List of Languages 489
List of Language Families, Linguistic Areas, and Proto-languages 504
 Contents xix
Index of Languages, Linguistic Areas, and Language Families 509
Index of Authors 519
Subject Index 525
 List of Maps

1. Distribution of noun classes and genders in the languages

of the world 78
2. Distribution of noun classifiers in the languages of the world 96
3. Distribution of numeral classifiers in the languages of the
world 122
4. Distribution of classifiers in possessive constructions in the
languages of the world 148
5. Distribution of verbal classifiers in the languages of the world 170
 List of Tables

1.1 Shopping list in Japanese 2

1.2 Differences between noun classes and classifiers 6
1.3 Classifiers, their morphological realization, and semantics 7
1.4 Classifiers and their functions 8
2.1 Semantics of noun classes in Proto-Bantu 24
2.2 Semantic features for the gender assignment of inanimate nouns
in Cantabrian Spanish 27
2.3 Inflection and derivation 30
2.4 Gender marking in Rumanian 46
2.5 Gender marking in Telugu 46
2.6 Gender agreement in Khinalug 47
2.7 A fragment of the Russian nominal paradigm 48
2.8 Noun classes in Ingush 49
2.9 Noun classes in Ndali 64
2.10 'Pronominal' and 'nominal' noun class systems 68
2.11 Ka-class assignment in Paumari 73
2.12 Two types of noun class (genders) in Mba 75
4.1 Numeral classifiers used with humans in Assamese 102
4.2 Numeral classifiers in Telugu 108
4.3 Classifiers fused with numerals in Kusaiean 108
4.4 Numeral classifiers in Nivkh 109
4.5 Numeral classifiers in Warekena 109
4.6 Numeral classifiers in Squamish 110
5.1 Possessed classifiers in Panare 128
5.2 Systems of two relational classifiers 134
5.3 A system of relational classifiers in Boumaa Fijian 134
5.4 Sample of classifiers in Puluwat 141
5.5 Differences between possessed and relational classifiers 145
5.6 Relational, possessed, and possessor classifier; a comparison 146
6.1 Mescalero Apache classificatory verb categories 155
6.2 Classificatory verbs in Ojibway 155
6.3 Classificatory verbs in Ika 156
6.4 Examples of the use of 'give' in Mescalero Apache 157
6.5 Classificatory verbs in Koyukon: an example 157
6.6 Classificatory verbs in Nevome 158
6.7 Classificatory verbs in Enga 159
6.8 Classificatory verbs in Ku Waru 159
6.9 Verbal classifiers in Waris 166
xxii List of Tables
6.10 Classificatory verbs in Waris 167
6.11 Affixed verbal classifiers in Koyukon 167
7.1 Locative Classificatory suffixes in three Carib languages 175
7.2 Article classifiers in Ponca 178
7.3 A sample of demonstratives in Proto-Guaicuruan 181
7.4 Deictic classifiers in Eskimo 182
8.1 Noun classes and noun classifiers in Ngan'gityemerri 186
8.2 Classifiers in Mokilese 187
8.3 Numeral classifiers in Akatek 188
8.4 Noun classifiers in Akatek 188
8.5 Numeral, verbal, and locative classifiers in Palikur 193
8.6 Demonstratives in Palikur (singular) 194
8.7 Gender marking on verbs in Palikur 194
8.8 Properties of classifiers and genders in Palikur 195
8.9 Semantic and functional properties and origin of classifiers
in Palikur 197
8.10 Different classifier sets in different environments in one language 202
8.11 Different classifier sets in the same environment in one language 202
9.1 Same set of classifiers in several environments (A–D) 207
9.2 Locative and verbal classifiers in Eyak 209
9.3 Same morphemes in several classifier environments 225
9.4 Environments in which genders and classifiers are used in Baniwa
and Tariana 230
9.5 Classifiers in Baniwa 232-4
9.6 Classifiers in Tariana 236-8
9.7 Agreement forms of kwa- in Tariana 240
10.1 Personal pronouns in Tamachek 245
10.2 Personal pronouns in Lithuanian 245
10.3 Personal pronouns in Slovene 246
10.4 Personal pronouns in Resigaro and in Bora 247
10.5 Gender in Malto 248
10.6 Classificatory verbs in Tewa and their semantics 251
10.7 Animacy marking on verbs in Jarawara (Arawa) 251
10.8 Personal pronouns in Minangkabau (singular) 253
10.9 Personal pronouns in Spanish 254
10.10 Personal pronouns in Tariana 254
10.11 Paradigm of Latin is 'this' 256
10.12 Possessed classifiers and speech styles in Ponapean 261
10.13 Examples of possessed classifiers in common and humiliative
speech in Ponapean 261
10.14 Paradigm of dela-t' 'do' in past tense in Russian 264
10.15 Paradigm of katav 'write' in Modern Hebrew 264
10.16 Variable noun class assignment in Maung 267
10.17 Classifiers and their interaction with other grammatical categories 269
 List of Tables xxiii
11.1 Examples of physical properties in noun class assignment 277
11.2 Semantic basis of gender choice in German: an illustration 280
11.3 Noun classes in Bantu 282
11.4 Shape-based classes in ChiBemba 282
11.5 Size-based classes in ChiBemba 282
11.6 Noun classifiers for humans and deities in Jacaltec 284
11.7 Noun classifiers for non-humans in Jacaltec 285
11.8 Hole classifiers in Tzeltal 289
11.9 Function-based classifiers in Burmese 291
11.10 Verbal classifiers in Ojibway and Cree 297
11.11 Chipewyan (Athabaskan) classificatory verbs 298
11.12 Classificatory existential verbs in Kamoro (Asmat) 299
11.13 Preferred semantic parameters in classifiers 306
12.1 Burmese numeral classifiers for inanimate objects 312
12.2 Animate classifiers in Burmese 315
12.3 Verified superordinate-subordinate pairs in Japanese numeral
classifiers 317
12.4 Reclassification of an inanimate noun in Burmese 319
13.1 Groups of nouns which tend to develop to classifiers 354
13.2 Numeral classifers from body parts in Totonac 356
13.3 Noun classifiers derived from common nouns in Mam 357
13.4 Generic classifiers in Minangkabau 359
13.5 Semantic groups of verbs which develop into classifiers 362
13.6 A sample of verbal classifiers in Imonda 363
13.7 Historical changes in Mandarin Chinese classifiers 410
13.8 Typical sources for noun categorization devices 412
15.1 Scope of classifier types 427
15.2 Assignment of classifiers 429
15.3 Morphological realization of classifiers 430
 List of Diagrams

2.1 Tendencies for animacy-based and shape-based noun classes 76

6.1 Verbal classifiers in Palikur 164
7.1 Locative classifiers in Palikur 173
7.2 Classifiers with spatial semantics in Toba 179
8.1 Semantics of numeral classifiers in Minangkabau 189
9.1 Semantics and form of demonstratives with classifiers
in Tariana 239
9.2 Semantics and form of articles with classifiers in Tariana 240
10.1 Animacy hierarchy and expression of number 247
11.1 Gender assignment in Manambu 278
11.2 Extendedness in Proto-Bantu noun classes 283
11.3 Numeral classifiers in Totonac 289
11.4 Numeral classifiers for inanimates in Palikur 290
11.5 Interaction of semantic domains in numeral classifiers in
Minangkabau 292
11.6 Classificatory verbs in Western Apache 298
11.7 Semantics of fourteen numeral classifiers in Akatek 302
11.8 Semantics of fourteen noun classifiers in Akatek 302
11.9 Polygrammaticalization of batang 'tree' in Minangkabau 302
11.10 Classifiers in Nambiquara 304
11.11 Semantics of classifiers in Kilivila 304
12.1 Structure of the tua category in Thai 314
12.2 Factors regulating the use of classifiers in Malay 324
12.3 Semantic network of the nge-class in Maasina Fulfulde 346
12.4 Gender pronouns in former and contemporary prescribed
English usage 350
13.1 Evolution of gender markers (1) 367
13.2 Evolution of gender markers (2) 368
13.3 Phonological reduction in the development of noun
classifiers in Mixtec 376
13.4 Singular and plural noun classes in Grebo 377
13.5 Gender in Proto-Dravidian (singular) 378
13.6 Gender in Proto-South-Dravidian (singular) 378
 List of Abbreviations

A subject of a DEF definite

transitive verb DEIC deictic
ABL ablative DEM demonstrative
ABS absolutive DER derivational
ACC accusative DET determiner
ADJ adjectivizer DIM diminutive
ADV adverb DIR directional
AFF affix DIST distal
AGR agreement DS different
ALIM alimentary subject
possession DU dual
AN, ANIM animate DUR durative
ANA anaphoric EMPH emphatic
ARG.MAN argument manip- ERG ergative
ulating derivation EXCL exclusive
ART article EXT extended
ASP aspect EYEW.PRES eyewitness
ATT attributive present
AUG augmented
F, FEM, f fern feminine
AUX auxiliary
FRUST frustrative
FUT future
BENEFACT benefactive GEN generic
CAUS causative GN genitive
CL classifier HAB habitual
CL:HAB classifier: habitat HON honorific
CMPL completed HORIZ horizontal
COLL collective HUM human
COMIT comitative HUMIL humiliative
COMPL completive IMAG imaginary
CONC concordial IMP impersonal
CONJ conjunctive IMPF imperfective
CONT continuous INAN, INANIM inanimate
CONTR contrast INCL, incl inclusive
CURV curved INDEF indefinite
CV connective INS instrumental
CYLIDR cylindrical INT intensifier
DAT dative INTER interrogative
DEC declarative IRREG irregular form
or shape
xxvi List of Abbreviations
LIG ligature PRED predicative
vowel PREF prefix
LINK linker PRES present
LIV.BEING living prim primary
being PRO 1/2/3 person
LOC locative proform
M, MASC, m, masc masculine PROB probability
MENS mensural PROGR progressive
classifier pron pronoun
MIN minimal PURP purposive
MOD modal PX proximity
N, NEUT neuter QUAL qualifier
NCL noun class QUANT quantifier
NEG negative RE referential
NF, nf non- REC reciprocal
feminine REFL reflexive
NOM nominative REL relativizer
NONPOSS, NPOSS non- REL.CL relational
classifier
possessed
REM.P.INFR remote past
NP nonpast
inferred
NUM.CL numeral
RES resultative
classifier
S subject of an
0 object of a
intransitive verb
transitive
sec secondary
verb
sg, SG singular
OBJ object
sp species
PART participle
SUBJ subject
PASS passive SUBORD subordinating
pcl paucal or SUFF suffix
plural TA tense-aspect
number marker
PERF perfective TAM tense-aspect-
Pf perfect mood marker
PI past TH thematic
imperfective THEM.CONTR thematic contrast
PL plural TNS tense
POSS possessive TOP topic
POSS.CL possessive TOP.ADV topic advancing
classifier voice
PP past TOP.O topical O
perfective VB verbalizer
PRECONT precontem- VCL verbal classifier
porary tense VERT vertical
 1 Preliminaries

1.1. General remarks

Almost all languages have some grammatical means for the linguistic
categorization of nouns and nominals. The term 'classifiers' will be used
here as an umbrella label for a wide range of noun categorization devices.
Different types of classifier can be distinguished by their grammatical
status, degree of grammaticalization, conditions for use, meaning, kinds
of origin, mode of acquisition, and tendencies towards loss.
Classifiers and noun categorization devices have long been a particular
focus of interest in functional typology. The urgent need to establish a
comprehensive typology of classifiers is motivated by a number of factors.
First, a large amount of new data on classifier systems has been produced
during the past decades; on the one hand, this data needs to be system-
atized, and on the other hand, its existence creates the opportunity of
providing a typology with reasonable scope and validity. Second, due to
the lack of an overarching unified analysis of classifier systems in the
languages of the world, there exists a pervasive terminological confusion
in the literature which makes difficult the cross-linguistic comparison of
noun categorization devices as well as the analysis of new data. This book
is an attempt to provide such a comprehensive approach insofar as this is
possible at our present stage of knowledge about the structure and
mechanisms of human languages and human cognition. The book is also
intended to serve as a guide for analytic work on previously undescribed
languages and their mechanisms for noun categorization.
Examples of different kinds of classifier are provided in §1.2. In §1.3 I
briefly describe the theoretical framework used in this study, together with
the database and sources. The next section provides a short overview of
previous approaches to noun categorization which are precursors to the
approach adopted here. The methodological basis for this approach is
outlined in §1.5. The structure of this book is outlined in §1.6.

1.2. Classifiers: an illustration

Classifiers come in different guises.
Some languages have grammatical agreement classes, based on such core
semantic characteristics as animacy, sex, or humanness. These are called
2 Classifiers
NOUN CLASSES, or GENDERS. The number of noun classes varies—from two,
as in Portuguese (examples below), to ten, as in Bantu, or even to several
dozen, as in some South American languages. Examples 1.1 and 1.2, from
Portuguese, illustrate masculine and feminine genders which are marked on
the noun itself and on the accompanying article and adjective.
1.1. o menin-o bonit-o
ART:MASC.SG child-MASC.SG beautiful-MASC.SG
'the beautiful boy'
1.2. a menin-a bonit-a
ART:FEM.SG child-FEM.SG beautiful-FEM.SG
'the beautiful girl'
A classifier can just categorize the noun by itself, as in the following
example from Yidiny, an Australian language (Dixon 1982: 192 ff.). This is
a NOUN CLASSIFIER.

1.3. bama waguja

CL:PERSON man
'a man'
Other languages have special morphemes which only appear next to a
numeral, or a quantifier. They may categorize the referent of a noun in
terms of its animacy, shape, and other inherent properties. These are
NUMERAL CLASSIFIERS. The way they are used is exemplified with a shopping
list in Japanese (Rie Hasada 1995) given in Table 1.1.

TABLE 1.1. Shopping list in Japanese

Shopping list Numeral Classifier Meaning of classifier
nasu (eggplant) nana (7) -ko CL:SMALL.EQUIDIMENSIONAL
kyuuri (cucumber) hachi (8) -hon CL:ELONGATED
hamu (ham) juu (10) -mai CL:SHEETLIKE

A special morpheme may characterize a possessed noun in a possessive

construction, as in 1.4, from Tariana, a South American language from the
Arawak family. This is a POSSESSED classifier.
1.4. tsinu nu-ite
dog lSG-CL: ANIMATE
'my dog'
A special morpheme in a possessive construction may characterize the
way in which the referent of a possessed noun relates to that of the
possessor. This is illustrated in 1.5 and 1.6, from Fijian, an Austronesian
 Preliminaries 3
language (Lichtenberk 1983a: 157-8). Such morphemes, underlined in 1.5
and 1.6, are called RELATIONAL CLASSIFIERS.
1.5. na me-qu yaqona
ART CL:DRlNKABLE-my kava
'my kava' (which I intend to drink)
1.6. na no-qu yaqona
ART CL:GENERAL-my kava
'my kava' (that I grew, or that I will sell)
VERBAL CLASSIFIERS appear on the verb, but they categorize a noun, which
is typically in S (intransitive subject) or O (direct object) function, in terms
of its shape, consistency, and animacy. Example 1.7, from Waris, a Papuan
language (Brown 1981: 96), shows how the classifier put- 'round objects' is
used with the verb 'get' to characterize its O argument, 'coconut'.
1.7. sa ka-m put-ra-ho-o
coconut 1SG-to VCL:ROUND-GET-BENEFACT-IMPERATIVE
'Give me a coconut' (lit. 'coconut to-me round.one-give')
There are two more, much rarer, kinds of classifiers. Those which occur
on locative adpositions, are called LOCATIVE CLASSIFIERS. This is illustrated
with 1.8 and 1.9, from Palikur, an Arawak language from Brazil.
1.8. pi-wan min
2SG-arm on+VERT
'on your (vertical) arm'
1.9. ah peu
tree on+BRANCH.LiKE
'on (branch-like) tree'
Classifiers which are associated with deictics and articles are called
DEICTIC CLASSIFIERS. Examples of deictic classifiers, from Mandan, a Siouan
language (Barron and Serzisko 1982: 99), are given in 1.10 and 1.11.
1.10. de-mak
'this one (lying)'
1.11. de-nak
'this one (sitting)'
The term 'classifier systems' is used to denote a continuum of methods of
noun categorization. Well-known systems, such as the lexical numeral
classifiers of Southeast Asia, on the one hand, and the highly grammati-
calized gender agreement classes of Indo-European languages, on the
other, are the extremes of this continuum. They can have a similar semantic
4 Classifiers
basis; and one type can develop out of the other. Parameters used for the
proposed typology of classifiers are discussed in §1.5.

1.3. Theoretical framework, data, and sources

The aim of this book is to present a functional-typological, empirically

based account of noun categorization devices across the languages of the
world. The analysis is cast in terms of basic linguistic theory, 'the funda-
mental theoretical apparatus that underlies all work in describing lan-
guages and formulating universals about the nature of human language',
where 'justification must be given for every piece of analysis, with a full
train of argumentation' (Dixon 1997: 132; see also Dixon 1994: p. xvi). The
categories, and their properties, considered here are developed inductively.1
This study is based on examination of the grammars of about 500
languages representing each major language family and each linguistic
area across the globe. A large database has been used, since the presence
or absence of a particular kind of classifier system is often an inherited
property of a language family or a diffusional property of a linguistic
area. Special attention has been paid to data that has recently become
available on the languages of South America (which by and large have not
been included in previous typological studies of classifier systems). Data on
the following languages come from my own fieldwork: Tariana, Baniwa,
Warekena, Bare (Arawak family), Tucano, Piratapuya (East Tucano
family), Paumari (Arawa family), from Brazil; and Manambu (Ndu family,
East Sepik) from Papua New Guinea.
I have not restricted myself to considering just some samples of the
available set of languages. Rather, I have looked at every language on which
I could find data and which has noun categorization devices. This approach
(sometimes called 'sample of convenience') allowed me to make the typol-
ogy proposed here as comprehensive as it could be at our present level of
knowledge about the languages of the world, without imposing artificial
limitations dictated by this or that 'sampling strategy'. Owing to limitations
of space, I could not cite all the examples of occurrence of every particular
phenomenon. I usually provide a particularly illustrative example, and
mention others. If a certain phenomenon is found in more than half of
the languages under consideration I call it 'relatively frequent'; if it is found
1
Cf. Bloomfleld (1933: 20) 'The only useful generalizations about language are inductive
generalizations. Features which we think ought to be universal may be absent from the very
next language that becomes accessible. . . . The fact that some features are, at any rate,
widespread, is worthy of notice and calls for an explanation; when we have adequate data
about many languages, we shall have to return to the problem of general grammar and to
explain these similarities and divergences, but this study, when it comes, will not be speculative
but inductive.'
 Preliminaries 5
in a restricted number of languages (one to ten), I cite all of them and
indicate its rarity. Note, however, that what appears rare to us at the
present stage of knowledge may turn out to be frequent when we start
learning more about hitherto little-known languages and areas. This is
the reason why I choose not to give any statistical counts at this stage.
Five hundred is no more than about one-tenth of all human languages,
and it seems most judicious to follow a qualitative approach at the
present time, postponing quantitative analysis until more data is available
and can be assessed.
Lists of languages, of language families, and of linguistic areas consid-
ered, are given in the index. I chose not to enumerate classifier types found
in each particular language referred to in the index in order not to impose
my analytic solution onto a language which is not my area of expertise
(readers can do this for themselves). Examples which come from my own
work are not followed by the indication of a source. I preserve the ortho-
graphy of the source (or use an accepted practical orthography, transcrip-
tion, or transliteration) unless otherwise indicated.

1.4. Approaches to the typology of classifiers

Classifiers and noun categorization systems have long been a particular

focus of interest in functional typology. They provide a unique insight
into how people categorize the world through their language. The study
of classifiers and noun categorization systems is intrinsically connected
with many issues which are crucial in modern linguistics, such as
agreement; processes in language development and obsolescence; the
distinction between inflection and derivation; and types of possessive
construction.
Noun classes and genders, on the one hand, and numeral classifiers, on
the other, have been the object of linguistic investigation for as long as
languages with these categories have been studied. The first overview full of
fascinating insights—albeit preliminary—was provided by Royen (1929). A
number of linguists have had ideas about similarities between different
systems of noun categorization devices; for instance, Worsley (1954)
pointed out functional similarities between Bantu-type noun class systems
and noun classes and numeral and verbal classifiers in Anindilyakwa, an
Australian language.
The systematic typological study of classifiers started only about two
decades ago. Studies of classifiers divide into two kinds: attempts to create
a general typological picture, and studies of individual types. The two
cannot be easily separated, since each discovery of a new type provides
feedback into the general typological picture.
6 Classifiers
During the last two decades, there have been a number of proposals for a
semantic and grammatical typology of noun categorization systems (often
also called 'noun classification'; e.g. Dixon 1968; 1982; Denny 1976; Allan
1977; Craig 1986a). Recently the typological parameters of classifiers and
other agreement categories have had to be revised in the light of new data,
especially those from previously undescribed South American Indian
languages (e.g. Derbyshire and Payne 1990; Craig 1992, forthcoming;
Corbett 1991).
Greenberg undertook a pioneering study of classifiers, in his paper on
numeral classifiers and substantival number (1972). Though this paper
does not overtly suggest any typology of noun categorization devices,
various classificatory phenomena are mentioned alongside numeral classi-
fiers (e.g. relational classifiers in Oceanic languages and verbal classifiers);
he also suggested a correlation between the existence of numeral classifiers
in a language and other grammatical categories, such as obligatory expres-
sion of number.
Further attempts at global typologies of classifiers include Adams and
Conklin (1973), Denny (1976), Allan (1977), and Serzisko (1982). Dixon
(1982) put forward an important suggestion for distinguishing between the
two extremes of noun categorization devices: obligatory grammatical noun
class systems, and semi-open lexical-like systems of classifiers (e.g. noun
classifiers and numeral classifiers). Dixon (1982; 1986) was also the first to
have explicitly stated a correlation between language type and noun cat-
egorization devices (that classifiers tend to be a property of isolating lan-
guages, while noun classes tend to be present in fusional and agglutinating
languages); he showed how one type (noun classifiers) can develop into
another (noun classes). The distinctions he drew between noun classes and
classifiers are shown in Table 1.2.

TABLE 1.2. Differences between noun classes and classifiers

Noun classes Classifiers
Size Small finite set Large number
Realization Closed grammatical system Free forms
Scope Marking is never entirely within Never any reference outside the
the noun word noun phrase
Source: Dixon (1982; 1986).

Allan (1977) provided a useful overview of noun categorization, for the

first time explicitly stating that the following types of noun categoriza-
tion device belong to the same domain: noun classes (or concordial
classifiers), numeral classifiers, verbal classifiers (including separate mor-
 Preliminaries 7
phemes and suppletive classificatory verbs), possessive and intralocative
classifiers.
Serzisko (1982) considered gender, noun class, and numeral categoriza-
tion as a part of a continuum of 'classificatory techniques' (under the
typological dimension of 'apprehension'), working out correlations
between these and other categories (such as number), and comparing
them as to their grammaticality, semantic complexity, and variability.
A very important though frequently underestimated contribution to the
typology of noun categorization is found in Seiler and Stachowiak (1982) and
Seiler and Lehmann (1982), followed by a summary in Seiler (1986), and also
in Seller's (1983) book on possession. These volumes are full of insightful case
studies; also, Baron and Serzisko (1982), followed by a summary in Seiler
(1986), provided the first consistent evidence in favour of the existence of
deictic (or article) classifiers in Siouan languages. Seiler (1986) was the first
to put forward the view of various kinds of classificatory techniques—
including numeral classifiers, verbal classifiers, noun classes and 'article'
classifiers—as continua within the broad dimension of apprehension.
Craig (1986a) was a major contribution to typological studies on noun
categorization, their role in cognition and culture. In particular, noun
classifiers as a special type have been established on the basis of her work.
A new view on the typology of noun categorization devices was provided by
Derbyshire and Payne (1990) in their survey of typologically unusual
systems of noun categorization devices in Lowland Amazonian languages.
Amazonian languages were shown to systematically allow more than one—
and often more than two—types of noun categorization simultaneously.
Further typological studies on classifiers include Nichols (1989b),
Kiyomi (1992), and Croft (1994). These focused on different parameters.
Nichols concentrated on the morphosyntactic realization of classifiers,
pointing out the differences between agreeing and non-agreeing noun
categorization devices. Kiyomi (1992) attempted to establish morpho-
syntactic correlates of classifier realization (with free or with bound
morphemes) for the main classifier types, and argued that neither animacy
nor shape can be established as defining semantic parameters for a typol-
ogy of noun categorization devices. See Table 1.3.

TABLE 1.3. Classifiers, their morphological realization, and semantics

Free morpheme classifiers Bound morpheme classifiers
Numeral classifiers (Animate, Shape) Concordial classifiers (Animate, Shape)
Non-numeral classifiers (Animate only) Predicate classifiers (Animate, Shape)
Intralocative classifiers (Shape only)
Source: Kiyomi (1992: 33).
8 Classifiers
Croft (1994) reanalysed classifier types, associating each of them with
semantic and pragmatic functions (he disregarded a few problematic clas-
sifier types, such as locative and deictic classifiers, and systems of verbal
classifiers with no animacy distinctions; see Chapters 6, 7, and 11 below).
See Table 1.4.

TABLE 1.4. Classifiers and their functions

Classifier type Semantic/pragmatic function
Noun class Determination (reference)
Numeral classifiers Enumeration
Possessive classifiers Possession
Predicate classifiers Spatial predication
Source: Croft (1994: 147).

Recent overviews of the typology of classifier systems can be found in an

in-depth study of Japanese numeral classifiers by Downing (1996), and in a
detailed analysis of classifiers in Kilivila (Austronesian) by Senft (1996).
Craig (1992; forthcoming) argued for the existence of the following types
of classifiers based primarily on the morphosyntactic loci in which they
occur: numeral classifiers; noun classifiers; noun class and gender; verbal
classifiers; genitive classifiers. Craig (forthcoming: 42) also mentioned
the existence of a 'marginal' classifier type—classifiers which occur with
articles or deictics. Further arguments in favour of this morphosyntactic
typology include cooccurrence of types within one language, different
semantics for distinct classifier types, and different degrees of grammati-
calization of classifiers. Importantly, classifiers are not presented as discrete
types, but rather as focal points on various continua. This prototype-
continuum approach, which implies a gradient rather than categorical
treatment of properties of classifier systems, is taken up in the present
study (see §1.5).
Craig's approach was elaborated upon in a case study of classifiers in
Tariana by Aikhenvald (1994a), and in Palikur by Aikhenvald and Green
(1998). The typology proposed in this book is largely based on the schema
established by Craig.
However, the current literature is somewhat confusing as far as generally
adopted definitions and concepts are concerned. The way linguists of
different traditions and theoretical trends use different terms, such as
GENDER, NOUN CLASS, CLASSIFIER, can be misleading.
The terms GENDER and NOUN CLASS are sometimes used interchangeably
(see §2.1). Corbett (1991) uses 'gender' as a cover term for agreement
classes, while Evans (1997: 109) opts for 'noun class' to cover the same
 Preliminaries 9
phenomenon. GENDER has also been used in a quite different way. In the
Athabaskan linguistic tradition the term 'gender' is used to refer to verbal
classifiers which mark agreement with intransitive subject or transitive
object, and characterize the referent noun in terms of shape and form
(Thompson 1993). In the Bantuist tradition, the term 'noun class' is
used to refer to a set of singular and of corresponding plural forms of a
noun and the agreement markers they trigger on modifiers and on the
predicate, while the pairs of singular and plural markers are considered
'genders'. For instance, Singular Noun Class 1 forms one 'gender' with its
plural counterpart, Noun Class 2 (see e.g. Table 2.1).
The term 'verbal classifier' is sometimes used by Australianists (Silverstein
1986; Green 1989; Reid 1990 and p.c.; Rumsey 1982; Donaldson 1980 and
others) to refer to a closed class of inflected verbs which typically carry
grammatical marking, and 'classify' the lexical verb by delimiting its aspect
or scope (e.g. 'do something on the surface', 'do something with hands',
'do moving up'). There is typically a small class of inflected verbs with
fairly generic meanings (often called 'simple' verbs, e.g. Rumsey 1982, for
Ungarinjin; Silverstein 1986) which together with a 'main verb' (or 'co-
verb') form a complex verb. In Ngiyambaa (Donaldson 1980: 201-24;
Dixon forthcoming: §6.2) a main verb 'dig', 'sew', or 'spear' takes the
classifier 'pierce'; and a verb such as 'take', or 'pick up' requires a classifier
'do with hands'. This usage of 'classifier' has some similarity with noun
categorization via generic noun classifiers: a simple, or 'classifier' verb
defines the generic scope of action, and the main verb specifies it; similarly,
a noun classifier indicates general reference (e.g. 'person' for people or
'animal' for animates), and the specific noun following it further specifies
this reference. This usage is completely different from the one adopted
here; however, as pointed out by Ian Green and Nicholas Reid (p.c.),
simple verbs may develop further semantic specifications whereby they
start being used to characterize the particular kind of instrument or loca-
tion. Further study is needed to delineate 'noun classifying' functions of
simple verbs in Australian languages.
The term 'verbal classifiers' is used in another, completely different way
by some specialists in the languages of South and Southeast Asia. Haas
(1942: 205) calls 'words indicating how many times an event takes place'
verbal classifiers; i.e. in a sentence like 'he ran twice' 'twice' is considered a
verbal classifier. This term is employed in a similar way for Newari (Tibeto-
Burman) by Bhaskararao and Joshi (1985: 17), and for Mulao, a Tai
language, by Jun and Guoqiao (1993: 48).
The term 'classifier' is used in yet another way in the Athabaskan
linguistic tradition, where it refer to markers of voice and change of
transitivity which have nothing to do with categorization of nouns.
Some authors simply avoid the term 'classifier'. Moussay (1981) uses the
10 Classifiers
term 'specificatif' for numeral classifiers and 'categoriel' for noun classifiers
in Minangkabau.
A number of the general statements about different types of classifier
have recently been shown to be erroneous. Some of the previously accepted
universals and general tendencies do not, in fact, hold. For example, Dixon
(1982: 220) suggested that languages could not have classifiers and gender
as separate categories, and stated: 'no example is known of a language with
two distinct systems of noun classes' (see also Craig 1986a; 1986b; 1986c).
Recent work on South American and Papuan languages has shown that
classifiers and genders do cooccur, and that languages can have two distinct
systems of noun classes. For instance, Baniwa, a North Arawak language
from Brazil, has a system of two genders, and also a system of over 40 noun
classes (see Chapters 8 and 9 below).
The dichotomy between a concordial noun class as an 'obligatory gram-
matical system where each noun chooses one from a small number of
possibilities' and noun categorization as a system where 'noun classifiers
are always separate lexemes which may be included with a noun in certain
syntactic environment' (Dixon 1986: 105) appears to be rather simplistic,
especially in the light of the data from Amazonian languages. The presence
of noun classes had often been associated with a fusional or agglutinating
morphological type, and classifiers (especially numeral classifiers) were
viewed as a typical property of isolating languages—a premise that also
appears to be a little simplistic when viewed cross-linguistically.
Finally, particular terms, such as 'classifier', 'noun classifier', or 'noun
categorization system', are frequently used by different authors either in a
different way for different types of system or as a cover term for any kind
of system. Thus, it is not always clear what is a classifier and what is a
concordial noun class in each particular case.
During the last two decades, a number of studies of specific classifier
types and individual languages have made an important contribution to an
overall typological picture. Corbett's (1991) book on GENDER (which is used
as a cover term for NOUN CLASS SYSTEMS) is an important, almost encyclo-
pedic, overview of this type of noun categorization. It is almost impossible
to enumerate all the studies of noun class systems in African languages;
however, the collection La Classification nominale dans les langues negro-
africaines (1967) and Hyman's (1980) book on noun classes in Grasslands
Bantu languages remain the main reference on the subject. Heine's (1982a)
article 'African Noun Class Systems' remains the main reference for the
typology of noun classes in African languages. Noun class systems in
Papuan languages are described by Foley (1986); some of these are ex-
tremely unusual—their assignment may be largely based on phonological
form (see Foley 1986; 1991; Conrad 1996; Nekitel forthcoming). Work on
noun classes in Australian languages includes ground-breaking studies by
 Preliminaries 11
Dixon (1972; 1982), Sands (1995), and the papers in Harvey and Reid
(1997).
NOUN CLASSIFIERS have been introduced into the typological picture
comparatively recently; their properties have been discussed at length by
Dixon (1982), Craig (1986a; 1992; forthcoming), and in more specific case
studies, e.g. Zavala (1992; forthcoming), Reid (1997), and Sands (1995). In
addition, Payne (1990) and Derbyshire and Payne (1990) considered the
problem of noun classifiers in Amazonian languages.
There is an immense corpus of literature on NUMERAL CLASSIFIERS,
especially in Southeast and East Asian languages. Adams (1989) and
Downing (1996) provide in-depth discussions of problems relevant for the
cross-linguistic definition of numeral classifiers (further, more language-
specific, or area-specific case studies include Barz and Diller 1985; Goral
1978; Bisang 1993; 1996; Pe 1965; T'sou 1976; Conklin 1981).
The existence of CLASSIFIERS IN POSSESSIVE CONSTRUCTIONS in Oceanic
languages was first recognized by Codrington (1885). The credit for the
first systematic study of relational classifiers and how they differ from
numeral classifiers in Oceanic languages goes to Lichtenberk (1983a);
among further studies one must mention Harrison (1976) for Mokilese;
Dixon (1988) for Fijian; Pawley and Sayaba (1990) for Wayan, a Western
Fijian dialect; and Rehg (1981) and Keating (1997) for Ponapean. Seiler
(1983) provides an insightful analysis of noun categorization in possessive
constructions and of the differences which can be noted between classifica-
tion devices which characterize the ways in which nouns can be possessed,
or handled (relational classifiers) and devices which describe properties of
possessed nouns (possessed classifiers, in our terminology). Carlson and
Payne (1989) attempted a broader survey of relational classifiers in some
North American Indian languages (Yuman, Uto-Aztecan) and some South
American Indian languages (some Carib, Tupi-Guarani, and Je languages);
further data on relational and possessive classifiers in South American
languages can be found in Rodrigues (1997, on Kipea, an extinct language
of the Kariri family, Macro-Je, South America), Rodrigues (1999), Barnes
(1990), Martins (1994), and Aikhenvald (1994a).
VERBAL CLASSIFIERS and SUPPLETIVE CLASSIFICATORY VERBS have been the
subject of extensive study based on the facts of specific language families.
Seminal studies of classificatory verbs in Athabaskan languages include
Hoijer (1945), Davidson et al. (1963), Krauss (1968), Basso (1968), Carter
(1976), and Thompson (1993); also see Mithun (1986) and Seiler (1986).
There is extensive literature on classificatory verbs in other North American
Indian languages, e.g. Kiowa-Tanoan (Speirs 1974) and Cherokee
(Blankenship 1996; 1997). Verbal classifiers in South American languages
are discussed in Derbyshire and Payne (1990); verbal classifiers in Papuan
languages are considered by Lang (1975), Brown (1981), and Merlan et al.
12 Classifiers
(1997). For verbal classifiers in Mesoamerican and South American Indian
languages see also Suarez (1983), Goncalves (1987), and Mithun (1984;
1986).
A few studies have been undertaken on rare and problematic classifier
types. The existence of DEICTIC classifiers as a special type has been shown
by Klein (1979), Vidal (1995; 1997), and Ceria and Sandalo (1995), for the
languages of the Guiacuruan family of Argentina and Brazil. Barron and
Serzisko (1982) describe article classifiers for Siouan languages. Further
data from South American languages in support of the existence of deictic
classifiers are given by Aikhenvald (1994a; forthcoming b). The existence
of a special type of LOCATIVE classifier was first suggested by Allan (1977)
(the term he used was 'intralocative'); his results were criticized by Croft
(1994). In fact, locative classifiers have only been found in a limited number
of South American languages (to which Croft did not have access), e.g.
Palikur (Arawak), Daw (Maku), and Carib languages (see Aikhenvald
1994a; 1996b).
Up until now no systematic attempt has been made to consider multiple
classifier systems in a cross-linguistic perspective (see Chapter 8 below).
Previous studies have not taken account of the unusual types of multiple
classifier system found in South American Indian languages. (Systems of
this kind were only briefly mentioned by Dixon 1982; Craig 1992; forth-
coming; and Lichtenberk 1983a.) Among descriptions of multiple classi-
fier systems from other parts of the world one should mention Hurd
(1977) on the Nasioi language from Bougainville, and Worsley (1954) on
Anindilyakwa, an Australian language from Groote Eylandt. Recently, the
number of studies of multiple classifier languages has increased, e.g.
Goncalves (1987), on Munduruku, a Tupi language from Amazonia;
Bisang (1993) on Hmong, a Miao-Yao language from China; Onishi
(1994) on Motuna, a Papuan language; Foris (forthcoming) on Sochiapan
Chinantec from Mexico; Vidal (1997) on Pilaga, a Guaicuruan language
from Argentina; Shepard (1997) on Machiguenga, an Arawak language
from Peru; and the survey of multiple classifier systems in Arawak
languages in Aikhenvald (1996b). Another problem for multiple classifier
systems is 'fuzzy' boundaries between types which makes it difficult to
attribute a given language to a particular type (see e.g. the discussion in
Vidal 1997).
To summarize—in spite of the considerable work already accomplished,
a new, integrated typological framework is needed to account for all the
types of noun categorization device and the new language data which have
appeared on the linguistic scene during the last decades. This is attempted
in the present volume.
 Preliminaries 13
1.5. Parameters for the typology of classifiers

All human languages have some ways of categorizing nouns and their
referents in terms of their semantic and syntactic properties. The purpose
of this book is to investigate how languages employ classifiers to provide a
semantically based categorization, which may have far-reaching implica-
tions concerning human cognitive mechanisms.
Classifiers are defined as morphemes which occur 'in surface structures
under specifiable conditions', denote 'some salient perceived or imputed
characteristics of the entity to which an associated noun refers' (Allan
1977: 285), and are restricted to particular construction types known as
'classifier constructions'. Classifier constructions are understood as
morphosyntactic units (which may be noun phrases of different kinds,
verb phrases, or clauses) which require the presence of a particular kind
of a morpheme, the choice of which is dictated by the semantic character-
istics of the referent of the head of a noun phrase.
Nouns and their referents can also be categorized in various other ways,
e.g. by choosing different number forms for nouns with different semantics;
by assigning the nouns to different declension classes; or by using different
pronominalization strategies. These strategies of noun categorization
(sometimes also called 'noun classification') are not considered classifiers.
However, they may be used in a way functionally similar to classifiers, and
they often reflect comparable semantic parameters. Historically, they may
go back to classifier systems. Examples are given in Appendix 1.
The main purpose of this book is to present a typology of classifiers
primarily based on the morphosyntactic loci (or environments) of classifier
morphemes (following the approach in Craig 1992; forthcoming). This
implies establishing types of noun categorization system which acquire sur-
face realization in natural languages. As a result, the typology is inclusive in
that it covers types of classifier morpheme and construction types in which
they are required, and categorization types. We start with a typology of
classifier morphemes and the constructions in which they are employed,
and then proceed to uncover a link between these and universal and language
specific parameters of categorization types. This is the basis for distinguish-
ing definitional properties and contingent characteristics of classifier types.
The terminology chosen for each classifier type relies as much as possible
on currently accepted terminology. If there are several terms in use, I
employ the one which is most current and most transparently describes
the morphosyntactic locus of a classifier type (e.g. I use 'verbal classifier'
rather than 'verb-incorporated classifier').
Following Craig (forthcoming: 43), classifier types are not viewed as
discrete entities, but rather as focal points on continua of various properties
used for the present typology (see below). As the result, definitional as well
14 Classifiers
as secondary, or contingent, properties of different classifier types will be
shown to be gradient rather than categorical; this accounts for the exis-
tence of instances of classifier systems which 'do not fit squarely into any of
the types' (Craig forthcoming: 43). As Frawley (1992: 30) puts it,
if we look at ordinary language, we find that it is full of gradient phenomena, more
technically known as fuzziness. . . . The insight behind fuzziness indicates that
categories have vague boundaries and are internally organised from central focal
values, the prototype (Rosch 1973, 1975a, b), to less focal instances and fringe
values. As the centrality of the category fades, . . . criteria for membership in the
category are less decisively applied, and categories merge into each other.

Consequently, classifier types outlined and argued for in this study cor-
respond to prototypes, or focal instances, which display all the definitional
and most of the contingent properties of a type. Less focal instances
represent various points on continua for different parameters of a typology
of noun categorization; these display varying degrees of the prototypical
properties of each type. In describing and analysing the data on noun
categorization devices in a given language, it is important to situate them
within the continua of various gradient properties rather than to try and fit
them into the mould of cross-linguistically established 'types'.
This prototype-continuum approach is also justified by historical facts
about classifier systems—it is well known that distinct classifier types
'blend into one another through time' (Craig forthcoming: 43). These
points will be amply illustrated within the present study; they are summar-
ized in Chapter 15.
The following dimensions will be employed to establish focal points on
the typological continuum of noun categorization devices.
(A) Morphosyntactic locus of coding
A noun categorization device can be realized in different morphosyntactic
loci, that is, on the head, or on all—or just some—of the dependents. We
will pay particular attention to languages which use different sets of clas-
sifier morphemes (often with different semantic and other properties) in
several morphosyntactic environments. The coexistence of these sets in one
language constitutes a strong argument in favour of the proposed typology,
since this indicates the independent existence and independent develop-
ment of different noun categorization devices in several morphosyntactic
environments in one language.
Some kinds of noun categorization device have several distinct subtypes
coexisting within one language: one set of noun classes may be used in one
environment, and a somewhat different set in another. For instance, many
Arawak languages of South America have a small system based on the
masculine/feminine distinction realized on verbal cross-referencing markers
 Preliminaries 15
and on demonstratives, while adjectival modifiers show a large system of
agreement noun classes. Systems of this kind are called 'split' systems; they
may represent potential new 'focal points' for developing further classifier
types.
(B) Scope, or domain of categorization
Noun categorization devices can refer to nouns within noun phrases of
different structures (modifier-head, possessive noun phrases, or adposi-
tional noun phrases), or within a verb phrase. They can also refer to
different constituents (e.g. possessed noun or possessor; A, S, O, or an
oblique argument). Thus, one can say that in 1.7 it is the O constituent,
'coconut', that is being categorized by the morpheme put 'classifier: round',
and thus it constitutes the scope, or domain, of this classifier morpheme.
(C) Principles of choice, or 'assignment' of noun categorization devices
The choice of a classifier may depend on some semantic properties of the
referent of the noun they categorize. However, it can also depend on other
properties of a noun (e.g. morphological or phonological).
(D) Kind of surface realization
Some noun categorization devices are realized with an affix or a clitic,
while others often appear as separate words.
(E) Agreement
Some noun categorization devices involve agreement, and some do not.
Agreement is understood as a requirement in covariance between gram-
matical meanings of grammatical morphemes (cf. Steele 1978: 610;
Lehmann 1982: 203; see §2.4 below). Categories which involve agreement
are 'syntactic' (or 'inflectional') in nature.
(F) Markedness Relations
Some noun categorization devices have a functionally and/or a formally
unmarked term; while others tend not to.
(G) Degree of Grammaticalization and Lexicalization
Some noun categorization devices are highly grammaticalized closed sets
while others tend to involve a lexical choice. A more lexical kind of noun
categorization can become grammaticalized.
(H) Interaction with other grammatical categories
Different types of noun categorization device tend to show different depen-
dencies with other grammatical categories (such as number, or case, or
verbal categories).
16 Classifiers
(I) Semantic organization of the system
Noun categorization in the languages of the world is based on a number of
universal parameters (e.g. 'human' versus 'non-human'). However, noun
categorization devices differ in terms of a number of other parameters,
termed their 'preferred semantics'. They also differ as to the organization
of their systems: in some, but not in others, every noun has to be assigned
a classifier. They also differ in the degree of their semantic transparency
and in the syntactic and discourse-pragmatic functions they perform.
Classifiers of different types differ in how they respond to socio-cultural
influence.
(J) Evolution and decay
Distinct types of noun categorization devices differ in their etymological
sources, and in the ways they develop and how they fall out of use.
Classifiers of one type can develop into another.
(K) Language Acquisition and Dissolution
Distinct noun categorization devices show fundamental differences in how
they are acquired by children, and what processes they undergo under
language dissolution in aphasia.
Properties (A-G) are definitional properties of classifiers, in agreement
with the morphosyntax-prior approach to classifiers adopted here. Proper-
ties (H-K) are contingent properties. Once the types of classifiers are
established with respect to characteristics (A-G), they will be shown to
display correlations with properties (H-K).
Quite a few languages use different sets of morphemes in different
classifier environments. Many languages employ the same (or almost the
same) set of classifier morphemes in different morphosyntactic loci. In this
case, the question to ask is whether we should consider them as instances of
distinct, albeit homophonous, classifier types, or as basically one type
extended to other environments. These and other related issues will be
discussed together with the problems of multiple classifier systems.
The structure of this book, as outlined in the next section, follows the
above order: we discuss the definitional properties of classifiers first, and
then proceed to consider the contingent ones.

1.6. The structure of this book

We will first discuss the proposed types, or 'focal points' on the continuum
of noun categorization devices with respect to their definitional properties
(A-G above) in the following order.
 Preliminaries 17
NOUN CLASSES and GENDERS are noun categorization devices realized
outside the noun itself within a head-modifier noun phrase. They are
realized, as agreement markers, on modifiers such as adjectives, but may
also appear on modifiers from closed classes such as demonstratives and
interrogatives. They can also be realized outside the noun phrase, e.g. be
marked on the predicate, or even on adverbs. They are most often affixes.
They usually contain reference to inherent properties of nouns, such as
animacy and sex, and sometimes also shape, structure etc. Some languages
have a special smallish set of noun classes/genders restricted to closed
classes of modifiers (demonstratives, and others) along with a different
set which appears on modifiers from other classes. These are discussed in
Chapter 2.
NOUN CLASSIFIERS are associated with the noun itself, and are indepen-
dent of any other element in an NP, or in a clause. They may be independent
words, or, more rarely, affixes attached to nouns. They refer to inherent
properties of nouns. Noun classifiers are free forms. Noun classes and noun
classifiers differ in their synchronic properties; however, noun classes often
develop from noun classifiers. These are discussed in Chapter 3.
NUMERAL CLASSIFIERS are another kind of noun categorization device
which operate within an attributive NP. These are realized outside the
noun in a numeral NP, and/or in expressions of quantity. Numeral classi-
fiers can be free forms, or affixes, typically to the numeral or quantifier.
They refer to the noun in terms of its inherent properties. These are
discussed in Chapter 4.
Noun categorization devices which operate within a possessive NP are
considered in Chapter 5. They can be of three kinds:
(i) The scope of categorization is the possessive relation itself, i.e. the way a
noun can be possessed, or treated. These markers are called RELATIONAL
CLASSIFIERS; they refer to the function of a noun, and not to its inherent
properties.
(ii) The scope of categorization is the possessed noun itself. Classifiers
which categorize the possessed noun are called POSSESSED CLASSIFIERS. The
noun is categorized in terms of its inherent properties.
(iii) The scope of categorization is the possessor, and its inherent proper-
ties. These are POSSESSOR CLASSIFIERS.
Another type of classifiers which have a clause as their scope are VERBAL
(or VERB-INCORPORATED) CLASSIFIERS discussed in Chapter 6. Their scope is
an argument of the predicate, usually in S/O function, more rarely in an
oblique function, and they are realized on the verb. They refer to inherent
properties of the noun; and may also convey information on its position in
space.
18 Classifiers
There are a few further, rare and rather problematic kinds of noun
categorization devices with an NP as their scope. LOCATIVE CLASSIFIERS
appear in adpositional NPs attached to an adposition, and characterize
the head noun in terms of its inherent properties. Some languages have
DEICTIC classifiers—morphemes which appear on deictics within an NP and
qualify the noun in terms of its inherent properties and its orientation, such
as horizontal or vertical. These are considered in Chapter 7.
Some languages have more than one kind of noun categorization—these
are discussed in Chapter 8. The same set of morphemes can be used in
several classifier environments—see discussion in Chapter 9.
We then consider contingent properties of classifiers. The ways in which
different classifier types interact with other grammatical categories are
discussed in Chapter 10. Parameters for the semantic categorization of
referents of nouns and the preferred semantics of different classifiers are
considered in Chapter 11.
The semantic organization of classifier systems and their functions are
dealt with in Chapter 12, together with a discussion of socio-cultural
parameters and mechanisms of human cognition reflected in noun categor-
ization. This chapter demonstrates the unitary basis for noun categoriza-
tion devices, providing support for considering them as variant realizations
of one phenomenon.
The origins, evolution, and decay of different noun categorization de-
vices are discussed in Chapter 13. The processes noun categorization
devices undergo in language acquisition and dissolution are considered in
Chapter 14. The results of the proposed typology and perspectives for
further studies are given in the concluding Chapter 15.
Appendix 1 describes noun categorization by means other than classi-
fiers, i.e. through marking number, grammatical relations, and other cat-
egories. Appendix 2 contains additional examples of semantic changes in
the process of development from nouns to classifiers.
Suggestions for linguists undertaking fieldwork on classifier languages
are provided in Appendix 3.
 2 Noun Class and Gender Systems

2.1. General remarks

NOUN CLASSES and GENDERS are grammaticalized agreement systems which
correlate—at least in part—with certain semantic characteristics (particu-
larly in the domain of human and animate referents). They are sometimes
called concordial classes; they include grammaticalized 'gender' systems of
the Indo-European type. They are realized outside the noun itself, usually
on modifiers which most often include adjectives, but may also include
modifiers from closed classes (demonstratives, interrogatives, possessives,
etc). They can also be realized outside the noun phrase, i.e. be marked on
the predicate, or even on adverbs. Some languages have a special smallish
set of noun classes/genders restricted to closed classes of modifiers (demon-
stratives, and others) along with a different set which appears on modifiers
from other classes.
A terminological clarification is in order. The term GENDER was first
used in the 5th century BC by the Greek philosopher Protagoras, when he
divided Greek nouns into three classes: 'feminine', 'masculine', and 'in-
animate' (nowadays called 'neuter'). This is a typical gender system,
which is found in many Indo-European languages. Latin had a similar
system; later, neuter nouns were redistributed between the other two
genders, giving the modern system of masculine and feminine in French
and Italian.
When Europeans came to study African languages, they discovered
larger gender-like systems with eight or more distinctions in languages
like Swahili; these often did not include a masculine/feminine distinction.
The term NOUN CLASS came to be used for systems of this type.
NOUN CLASS, GENDER, and sometimes GENDER CLASS are often used inter-
changeably, depending on the linguistic tradition (some examples are given
in §1.4). Here I shall use 'noun class' as a cover term for noun class and
gender. In agreement with the linguistic tradition, I shall reserve the term
gender for small systems of two to three distinctions (always including
masculine and feminine), like the ones typically found in Indo-European,
Afroasiatic, and Dravidian languages.
Since gender systems show some correlation with sex, many non-
linguists (and a few linguists) erroneously confuse 'linguistic' gender and
sex. However, sex represents biological categorization, and gender repre-
sents grammatical categorization. Feminine and masculine genders often
20 Classifiers
include inanimate nouns with no connection to female or male sex, e.g.
French maison 'house' (feminine), chateau 'castle' (masculine).
The languages of the world differ in the number of noun classes they
have, how much semantic transparency there is to noun class assignment,
where and how noun class gets expressed, and whether it is possible to
change the noun class of a given noun.
Noun class systems are typically found in languages with a fusional or
agglutinating (not an isolating) profile. Noun class agreement is often a
major criterion for distinguishing nouns from other word classes. In a
language where noun and adjective have similar morphology, an adjective
can usually take any noun class marking whereas a noun is normally
restricted to one class.
Because of the limitations of space, it is impossible to cover all the
literature on noun class systems. To avoid an overlap with Corbett's
(1991) study of noun class systems (for which he uses a cover term
'gender'), I will concentrate on the issues and examples which have not
been considered there, and briefly mention the ones for which Corbett
(1991) provides detailed coverage.
The properties of noun class systems are considered in §2.2. Noun class
assignment is discussed in §2.3. Noun class agreement is dealt with in §2.4.
The next section discusses markedness relations and resolution in noun
class systems. Realization of noun classes is analysed in §2.6. Some
languages have two noun class systems. These may be in a complementary
distribution with respect to the modifiers with which they are used; they
may display peculiar agreement properties—see §2.7. Finally, §2.8 surveys
the distribution of noun classes in the languages of the world.

2.2. Properties of noun class systems

A noun class system is the most grammatical means a language can use for
the semantic categorization of nouns. As we shall see later, other noun
categorization mechanisms are more lexical and often more semantically
based.
Noun class systems have the following definitional properties.
1. Some constituent outside the noun itself must agree in noun class with a
noun. Agreement can be with other words in the noun phrase (adjectives,
numbers, demonstratives, articles etc.) and/or with the predicate of the
clause or with an adverb. That is, noun class can be realized in a number
of morphosyntactic loci (depending on the agreement rules in the
language) and its scope can be a noun phrase and/or a clause. Noun classes
are defined syntactically. They constitute a closed obligatory grammatical
 Noun Class and Gender Systems 21
system (which often arises as the result of grammaticalization of some
other noun categorization device: see Chapter 13).
Noun classes are realized with affixes or with clitics, and in most cases
there is a limited, countable number of noun classes (see §2.6).
2. Noun class membership is assigned on semantic—and sometimes also
morphological and phonological—principles. Each noun in the language
belongs to one (or occasionally more than one) class(es).
There is always some semantic basis to the grouping of nouns into
classes, but languages vary in how much semantic transparency there is.
This semantic basis usually includes animacy, humanness and sex, and
sometimes also shape and size. We will return to this in Chapter 11.
In some languages, in addition to the realization of noun classes through
agreement, there is a marker of noun class on the noun itself, or on some
nouns; in other languages nouns bear no overt marker.
Languages often have portmanteau morphemes combining information
about noun class with number, person, case, etc. This is considered in
Chapter 10.
Some systems based on animacy and sex (and traditionally called 'gen-
der systems') do not, in fact, satisfy the criteria set out here. English
distinguishes three genders just in 3rd person pronouns, helshelit. They
involve the opposition: male, female, inanimate. There are a few conven-
tionalized metaphorical extensions, e.g. ships are commonly referred to
with the feminine pronoun she (see further examples in §12.3.3). There is
no gender agreement within a noun phrase or with a verb in a clause.1
Gender markers in English simply have an anaphoric function, as they also
do in Japanese where masculine and feminine forms are distinguished only
in 3rd person pronouns with a human referent kare 'he', kanojo 'she'
(recently introduced, possibly, under the influence of European languages:
Walter Bisang, p.c.).2 Many languages of the world also have animacy-
based distinctions in interrogative and indefinite pronouns (see Haspelmath
1997), e.g. English anybody or anything which can be used anaphorically.
Strictly speaking, these are not noun classes.
The presence of agreement is the main definitional property of a noun
class. Some languages have singular/plural alternations which can be
shown as at least partially conditioned by the semantics of nouns. However,
these pairings do not correspond to different agreement classes. Such
appears to be the case with singular/plural alternations in Eastern Sudanic

1
However, if we follow Lehmann's (1982: 219) view of agreement, the use of these pronouns
to 'agree' in animacy/sex with their antecedent can be considered 'anaphoric agreement'; see
§2.4.1.
2
There can also be complicated relations between sex of speaker and form of other
pronouns, e.g. first person; these relate to the category of politeness: cf. Chapter 10.
22 Classifiers
languages (Dimmendaal forthcoming). This is a system of classifying
nouns and their referents; however, it cannot be considered a system of
noun classes. In contrast, Bantu languages have large systems of noun class
affixes which are portmanteau morphemes of noun class with number;
since they appear both on the noun itself and on the agreeing constituents
they 'qualify' as noun classes. Modern Hebrew distinguishes two genders,
masculine and feminine, both in the singular and in the plural (see §2.3.4)
which are realized in agreement within a noun phrase and on the verb.
Nouns also fall into several classes depending on their number and case
forms (Aikhenvald 1990: 48); this second kind of classification lies outside
the scope of the present study.

2.3. Principles of noun class assignment

The principles by which nouns are 'assigned' to different classes can be
governed by semantics (§2.3.1), or formal morphological (§2.3.2) or phono-
logical (§2.3.3) properties of a noun, or a combination of these (§2.3.4)
(also see Corbett 1991: 7-69). In a sense all systems of noun class assign-
ment are mixed, since there is always a semantic core which involves the
universal semantic parameters (see §11.2.1) of sex, humanness, animacy
but this is never the entire story.
Noun class systems were defined above as obligatory grammatical sys-
tems, such that every noun has to belong to a noun class. However, noun
class assignment is sometimes impossible for a smallish group of nouns.
These are exceptions to the statement that every noun in a language with a
noun class system has to be assigned to a noun class. For example, Russian
does not distinguish genders in the plural. Then, the gender of pluralia
tantum, i.e. nouns which are used only in the plural and always have plural
agreement, cannot be determined, e.g. sani 'sledge', brjuki 'trousers', seni
'entrance into a hut'. 3
The psycholinguistic reality of gender assignment has been confirmed by
recent studies of child language acquisition (Connelly 1984; Mills 1986;
Tsonope 1988). We shall return to this in Chapter 14.

2.3.1. Semantic assignment

In languages with purely semantic assignment the class of a noun can be
inferred from its meaning. In Tamil (Dravidian) all nouns divide into what
are traditionally called 'rational' and 'non-rational' classes. 'Rational'
nouns comprise humans, gods, and demons (Asher 1985: 136). Other
3
These nouns do distinguish animacy; see (C) in §2.4.4 on subgenders in Russian.
 Noun Class and Gender Systems 23
Dravidian languages, Malto, Kolami, Ollari, and Parji distinguish male
humans as distinct from other nouns which refer to 'rational' beings.
From the Northeast Caucasian family, Godoberi, Akhvakh, and Bagval
distinguish male 'rational', female 'rational', and the rest (Corbett 1994b).
Diyari (Australian), Kaingang (Je), and the North Arawak subgroup
divide nouns into female humans and the rest.
Semantic assignment can be more complex. Dyirbal (Dixon 1972: 306-
12) has four classes. Three are associated with one or more basic concepts:
gender 1—male humans, non-human animates; gender 2—female humans,
water, fire, fighting; gender 3—non-flesh food. Gender 4 is a residue class,
covering everything else. There are also two rules for 'transferring' gender
membership. By the first, an object can be assigned to a gender by its
mythological association rather than by its actual semantics. Birds are
classed as feminine by mythological association since women's souls are
believed to enter birds after death. The second transfer rule is that if a
subset of a certain group of objects has a particular important property,
e.g., being dangerous, it can be assigned to a different class from the other
nouns in that group. Most trees without edible parts belong to gender 4,
but stinging trees are placed in gender 2.
Mythological association plays an important role in class assignment in
other languages, too. In the Western Torres Strait language all nouns
denoting males are masculine, with the remainder being feminine. How-
ever, the moon is masculine, due to its mythological association with
masculinity (Bani 1987). This is also characteristic of other Australian
languages. The assignment of masculine and feminine noun classes in
Abu' Arapesh (Papuan; Nekitel 1985, 1986, forthcoming) can often be
explained by mythological associations; for instance, cassowary is feminine
because it used to be a mythological woman, and the moon is masculine—
it is a mythological man who engages in a sexual intercourse with women
(making them menstruate).
In Ket (Krejnovic 1961, Dul'son 1968: 62ff.), all sex-differentiable nouns
are masculine or feminine. Among non-sex-differentiable nouns, those
which show a higher degree of activity or are particularly important for
Ket culture are masculine, e.g. wood, large wooden objects, growing trees.
Gender assignment of sun (feminine) and moon (masculine) is determined
by their role in myths.4 Other inanimate nouns are treated as neuter.
The degree of semantic motivation for noun classes varies from language
to language. Systems with a larger number of noun classes tend to have
4
Also see Harvey (1997) for an insightful overview of semantic parameters employed in
gender assignment in Australian languages. In this and other cases it can also be argued that
the mythological role was originally determined by gender assignment, and not the other way
round. Indeed, in many cases it is impossible to prove which line of argument is better
founded.
24 Classifiers
more semantic motivation; however, this is not necessarily so. Languages of
the Ndu family (East Sepik, Papua New Guinea) have two semantically
assigned genders (see §2.4.3). In Babungo (Grassfields Bantu, Benue-
Congo: Schaub 1985: 172; Croft 1995) there are significant correlations
between 14 noun classes and semantic categories, but none of them is
absolute. The meanings ascribed to some reconstructed Proto-Bantu
noun classes by Denny and Creider (1986: 232-9) are shown in Table 2.1
(a somewhat different version is given in Table 11.3). In the Bantuist
tradition, every countable noun is assigned to two classes: one singular
and one plural (see §1.5 on terminology and §10.1.1 on the correlations
between noun classes and number).
TABLE 2.1. Semantics of noun classes in Proto-Bantu
Noun class (SG/PL) Semantics
Class 1/2 Human, person
Class 3/4 Extended (long) (e.g. body, river)
Class 5/6 Fruits; non-extended (e.g. stone, spot, nose)
Class 7/8 Utilitarian artefacts, despised objects and beings
Class 9/10 Animal
Class 14/6 Differentiated internal structure (e.g. bridge, bow, canoe)

In modern Bantu languages, however, noun class assignment is often

much less semantically motivated, though the semantic 'nucleus' is still
discernible. Thus, in Babungo, class 1/2 is basically human; however, it is a
much bigger class than it was in Proto-Bantu, and also contains many
animals, some birds and insects, body parts, plants, household and other
objects, e.g. necklace, pot, book, rainbow (Schaub 1985: 175) (also see
Tables 11.4 and 11.5, and Diagram 11.2).
It has often been stated that there is no real semantic basis for gender
assignment of the better-known Indo-European languages. However, in a
seminal study, Zubin and Kopcke (1986) provided a semantic rationale for
gender assignment of nouns of different semantic groups in German (see
(D) in §11.2.1). Masculine and feminine genders mark the terms for male
and female adults of each species of domestic and game animals (following
the 'natural sex' principle), and neuter is assigned to non-sex-specific
generic and juvenile terms. Masculine gender is used for types of cloth,
of precipitation and wind, and of minerals. Types of knowledge and
discipline have feminine gender, and games and types of metal—with the
exception of alloys—have neuter gender.5
5
Paul (1972) also demonstrated a partial semantic motivation for the gender assignment of
English borrowings into German; for instance, drinks are mostly masculine, while fruits and
flowers are feminine.
 Noun Class and Gender Systems 25
2.3.2. Morphological assignment
A connection between derivational suffix and noun class can form a
morphological basis for noun class assignment. In German at least some
derivational affixes are each associated with one gender, e.g. -ung 'action
noun' is feminine and -chen 'diminutive' is neuter (Zubin and Kopcke 1986;
Plank 1986). In Portuguese, a number of derivational suffixes (e.g. -cao
'action noun', as in marca-cao 'marking') indicate feminine gender.
In a language with a number of nominal declensions, each may correlate
with a gender. In Russian the semantic assignment is restricted to human
and higher animate referents. Otherwise, gender assignment is linked to
declension: all nouns of declension 1 are masculine, nouns of declensions 2
and 3 are feminine, and all the rest are neuter (Corbett 1991: 40).

2.3.3. Phonological assignment

No noun class system in the world is assigned by phonological principles
only. The application of phonological principles of assignment is usually
restricted to nouns with inanimate referents. In some languages every noun
which ends or begins with a certain vowel or consonant must belong to a
particular gender. In Qafar (Saho-Afar, East Cushitic: Corbett 1991: 51-2)
nouns with inanimate reference whose citation form ends in a vowel are
feminine; all the rest are masculine. In Hausa all non-sex-differentiable
nouns which end in -aa are feminine. In Katcha (Kordofanian: Heine
1982a: 200), any noun—unless it has a male referent—belongs to the
feminine gender if it begins with m-. Phonological noun class assignment
is found in Limilngan (Australian: Harvey forthcoming: §3.2): nouns whose
initial segment is /1/ or /d/ tend to be assigned to Class 2 (which includes
animals), and nouns with the initial /m/ tend to be assigned to Class 3
(which covers plants), even if their reference lies outside the semantic
domain of these classes.

2.3.4. Mixed principles of assignment

No system of noun classes is completely devoid of semantic motivation. If
a language has non-semantic principles of noun class assignment the
assigment principles will be mixed, since there is always a 'core' where
semantics operates. This 'core' includes humans in some languages and
animates in others (see §11.2.1).
In the Harar dialect of Oromo (East Cushitic: Clamons 1993: 271) nouns
referring to females are feminine, and noun referring to males are mascu-
line. Nouns referring to inanimates, or animates for which sex is not
important, are feminine if they end in a non-low vowel; otherwise they
26 Classifiers
are masculine. Russian has predominantly morphological gender assign-
ment (§2.3.2). However, sex-differentiable nouns are assigned gender
according to their semantics, and not their form. Thus, nouns like muzcina
'man' or detina 'big man' belong to the second declension, and should be
feminine; but in fact they are masculine as far as agreement goes.6
An interesting interaction of semantic and phonological principles is
found in Yimas (Sepik, Papua New Guinea, Foley 1986: 86 ff; 1991),
where the first four classes are assigned by their semantics: I—human
males; II—human females; III—animals; class IV—culturally important
plants. Classes 5-11 are motivated phonologically: the agreeing constituent
repeats the last consonant of the nominal root.
Arapesh languages (Torricelli phylum, Papua New Guinea) appear to
have a similar system. The assignment of two human classes, which
comprise males and females respectively, is considered semantic by most
scholars. Nekitel (1985; 1986; forthcoming) has convincingly argued
against a 'purely phonological' assignment of noun classes in Abu'
Arapesh (cf. Fortune 1942 and the analysis of Muhiang Arapesh by
Conrad 1978: 92).7 A strong argument in favour of the semantic assign-
ment of the human classes comes from loans. Most nouns which belong to
Class 1 'masculine' contain either a final or an initial segment n (e.g.
aleman 'man', Nekitel 'male name'). However, words like Spiritu Santu
'Holy Spirit' and ankelo 'angel' are attributed to Class 1 'masculine',
although they do not contain the 'marker' n; similarly, siste 'nun' (from
English sister) is assigned to the Class 2 'feminine' (though it does not
contain ?- or kw-, initial sounds typical of this class). Morphological and
phonological assignment may be hard to distinguish. There is a strong
tendency to assign nouns which end in -n (e.g. aun 'moon') to the masculine
class (unless they have a female referent, e.g. nes 'nurse', a loan from
English); this is sometimes interpreted as a fossilized affix (cf. Conrad
1996).
Many languages display more complicated mixed principles of 'assign-
ment' mingling semantic, morphological, and phonological criteria. Iraqw
(South Cushitic; Heine 1982a: 200) has masculine and feminine genders.
All nouns denoting singular male and female animates and male and
female agentive nouns are masculine and feminine respectively. Singular
6
Loan words provide a few more exceptions to morphological assignment in Russian.
Kofe 'coffee' is indeclinable, ends in -e, and does not denote a sex-differentiable being; these
are usual properties of neuter nouns. The archaic form of this word (borrowed from Dutch
or English: Vasmer 1953) was kofej, and it was assigned masculine gender since nouns
ending in -ej are masculine; later it became kofe. In prescriptive Russian grammar, this
noun is still considered masculine, while in substandard colloquial language it triggers neuter
agreement.
7
For an attempt to account for Arapesh noun classes as an exclusively phonologically
based system, see Aronoff (1991; 1994).
 Noun Class and Gender Systems 27
nouns ending in -mo and -anw are masculine. Nouns derived from Class 1
verbs are masculine while those from Class II verbs are feminine.8
A more 'grammatical' system of gender assignment can move towards a
more 'semantically' oriented one. In Cantabrian Spanish a number of seman-
tic features have been introduced for the assignment of inanimates to mascu-
line or feminine gender (Holmquist 1991: 69), e.g. feature large/small: masc.
monton 'stack of hay' vs fern, montona 'very big stack of hay'; feature depre-
catory/approbatory, neutral: masc. carreteru 'a bad road' vs. fem. carretera 'a
road'; feature coarse/smooth: masc. espinu 'mountain thorn, dark coarse
bark' vs. fem. espina 'mountain thorn, light smooth bark'—see Table 2.2.
These oppositions are mostly absent from standard Spanish.

TABLE 2.2. Semantic features for the gender assignment

of inanimate nouns in Cantabrian Spanish
Masculine Feminine
Male Female
Small Large
Narrow Wide
Vertical Horizontal
Tall Squat
Phallic Supine
Coarse Smooth
Dark Light
Deprecatory Approbatory, neutral
Derived Primary
Occasional Familiar

For nouns with animate reference, semantic assignment often overrides

morphological or other principles. In Alamblak (Lower Sepik) (Bruce
1984: 97) all nouns which denote females and short, squat, or wide objects
are feminine and can have a form marked with a feminine suffix. There is
one exception, the word for canoe, doh-t, which has a feminine marker -t,
but is always treated as masculine in agreement, in accordance with its
typical 'masculine-like' slender dimensions.9
8
A similar system is found in Punjabi (Indo-Aryan: Bantia 1993: 216-18). In Chechen and
Ingush (Nakh-Daghestanian: Nichols 1989a) the distribution of nouns into five classes is
partially semantically motivated; there is also a dependency between class assignment and the
initial consonant of the noun. Semantic and phonological principles of noun class assignment
interact in Bowili (Togo Remnant, Eastern Ghana: Heine 1982a: 199-200). See also Aikhenvald
and Green (1998) for a detailed description of mixed semantic and phonological principles of
gender assignment in Palikur, a North Arawak language from Brazil.
9
The word Madchen 'girl' in German may be considered an exception to this. It is neuter
since it contains a diminutive suffix -chen which, like other diminutive suffixes, determines
the assignment of neuter. However, the situation in the modern spoken language is more
28 Classifiers
Semantic, morphological, and phonological principles account for the
great majority of the assignment of nouns to gender classes in any given
language, but there will often be a small residue of unexplained exceptions.
Modern Hebrew has two genders, masculine and feminine. The principles
of assignment are semantic and morphological. Sex-differentiable nouns
are assigned gender in agreement with their semantics; nouns which
contain suffixes -t and -a are feminine. Nouns which refer to cities and
land, and paired and some non-paired body parts, are feminine. There are
a few nouns which belong to feminine gender as unexplainable exceptions:
even 'stone', kos 'goblet', es 'fire' (Aikhenvald 1990: 44).I0
Nouns which display a conflict between different rules of gender/noun
class assignment are called HYBRIDS (Corbett 1991). In Russian muzcina
'man' is feminine by its morphology (it belongs to the 2nd declension: see
§2.3.2) but masculine according to semantics. Semantics often takes pre-
cedence in agreement, as it does here. In Russian, in the case of most nouns
denoting professions which are morphologically masculine (they belong to
the first declension), such as professor 'professor', the agreement is
feminine when focusing on the female sex of the person. In Portuguese,
some nouns denoting professional occupations which end in -a or -e can
also be assigned either gender depending on the sex of the referent, e.g.
dentista 'dentist', estudante 'student'. The issue of variable noun class
assignment and agreement is taken up in §2.4.3.

2.4. Noun classes and agreement

The presence of agreement is a definitional property of noun classes which
distinguishes them from a number of other noun categorization devices.
The presence of agreement is linked to the opposition between inflection
and derivation. A working definition of agreement is given in §2.4.1.
Principles of noun class agreement are discussed in §2.4.2. Variability in
agreement, and variable noun class assignment are considered in §2.4.3.
Then, in §2.4.4, I discuss the problem of how to determine the number of
noun classes within a language.

2.4.1. A working definition of agreement and agreement properties

Agreement is defined by Matthews (1997: 12) as a 'syntactic relation
between words and phrases which are compatible, in a given construction,
complicated: Madchen may trigger feminine agreement with a relative clause marker and with
a possessive pronoun; neuter is preferred when the antecedent is a child rather than a grown-
up girl (Fritz Serzisko, Geoff Haig, Helma Pasch p.c.).
10
See also Clamons (1995) on semantic residue in gender assignment in Oromo, a Cushitic
language.
 Noun Class and Gender Systems 29
11
by virtue of inflections carried by at least one of them'. Agreement
implies a systematic covariance between the grammatical meanings of
grammatical morphemes (cf. Durie 1986). In a study of agreement systems
the questions to be answered are:
(A) Domain of agreement
What elements agree with what elements in what grammatical configurations?
It is useful to distinguish two basic types of agreement domain: (a)
agreement within an NP between modifiers and heads (head-modifier
type), and (b) agreement within a clause between a predicate and its
arguments (predicate-argument type).12
Morphosyntactic loci on which agreement markers appear are called
AGREEMENT TARGETS (see Corbett 1991). Noun classes can have a noun
phrase, and/or a clause, as their domain of agreement.
(B) Features and principles of agreement
In what grammatical properties do grammatical elements agree and how is
it marked?
The principles of agreement are linked to the assignment of agreement
features. The assignment may be either purely semantic, or mixed semantic
and syntactic (cf. the distinction between 'grammatical' agreement and
'notional' agreement in Matthews 1997). These issues are discussed in
§2.4.2. Conditions which may allow neutralization or variation in the
agreement, and limitations on the agreement, are considered in §2.4.3.

11
A similar definition of agreement is provided by Steele (1978: 610): The term agreement
commonly refers to some systematic covariance between a semantic or formal property of one
element and a formal property of another. For example, adjectives may take some formal
indication of the number and gender of the noun they modify.' See further attempts at defining
agreement by Keenan (1978: 167); Lehmann (1982: 203; 1988); further analysis of the basic
parameters in terms of which agreement phenomena can or should be characterized is given by
Barlow and Ferguson (1988a: 3); also see Lapointe (1985: 84), and discussion in Anderson
(1992: 103-18). Agreement can be taken in a wider sense to include the so-called anaphoric
agreement, i.e. the 'determination of the form of personal and relative pronouns' by their
antecedents (Corbett 1991: 112; cf. Lehmann 1982). Barlow (1992) has shown that there are no
reasons to make a sharp distinction between agreement within a noun phrase, and antecedent-
anaphora relations. Historically, grammatical agreement often comes from grammaticalized
anaphoric markers (see §13.8; also see Given 1976; Bresnan and McChombo 1986). A number
of languages, including English, distinguish different forms of personal pronouns conditioned
by the gender and animacy of the antecedent. If agreement is understood in a wider sense,
English can be considered a language with genders (as it was done by Corbett 1991: 112, 169).
12
Following the distinction in Anderson (1992: 106 ff.). The important difference between
Anderson's approach, and the one suggested here lies in the treatment of adpositional and
possessive constructions. For the reasons which will become obvious in the course of this
chapter, I will consider the principles of agreement within a noun phrase under head-modifier
type; and the principles of agreement within a clause under predicate-argument type (unlike
Anderson, who groups together agreement of verb with its arguments, and agreement in
possessive and adpositional constructions).
30 Classifiers
Some languages have different noun class/gender agreement systems
depending on the domain of agreement (head-modifier vs. predicate-
argument) and on the morphological class of the agreeing element. These
systems, called 'split agreement', are discussed in §2.7.
Correlations between noun classes and other grammatical categories
which may also influence the ways agreement operates are considered in
Chapter 10; correlations with discourse-pragmatic functions are discussed
in Chapter 12.
An important distinction in the morphology of many languages is that
between inflectional and derivational processes. These are summarized in
Table 2.3 (cf. Payne 1990; Anderson 1992: 77 ff.; Aikhenvald forthcoming d).

TABLE 2.3. Inflection and derivation

Inflection Derivation
1. Usually obligatory Optional
2. Final process (if affix, on rim of word) Pre-final process (if affix, between root
and inflection)
3. Forms a complete word Derives a stem which takes inflections
4. Defining characteristic of a word class Usually specific to a word class
(e.g. nouns inflect for case)
5. Do not change word class May derive a stem of a different word
class, or may add some semantic
specification to a root without
changing class
May indicate grammatical Never indicate grammatical
relationship between words, and/or relationship between words or
participate in agreement participate in agreement
Tend to be smallish systems May be large systems
Tend to have high frequency in Likely to have lower frequency
language
Tend to be monosyllabic likely to May be longer and are less likely to
undergo phonological processes when undergo phonological processes
combined with stem (such as
assimilation, or fusion)

By virtue of being realized as agreement markers, noun classes have to

be treated as an inflectional category. Note that noun classes marked on
the head noun ('head classes': see §2.6.1, and Evans 1997) can have a
derivational function, since they may derive a stem of a different word
class, e.g. Kikongo (Bantu) 0lba -bakala (CLl/2-male) 'man', ki-bakala

13
Note that here and in other examples from Bantu languages the two numbers correspond
to singular and plural class markers.
 Noun Class and Gender Systems 31
(CL7/8-male) 'maleness'; Swahili kubwa 'big', u-kubwa 'size' (Mufwene
1980: 248-9). Similarly, in Portuguese—and in numerous other Indo-
European languages—gender is used to mark agreement, e.g. agua branc-
a (water: FEM white-FEM.so) 'white water'; it is also used as a derivational
device, e.g. professor (teacher: MASC.SG) 'he-teacher', professor-a (teacher-
FEM.SG) 'she-teacher'; ministro 'he-minister', ministr-a 'she-minister'.

2.4.2. Principles of noun class agreement

Noun class agreement can be a property of a noun phrase or of a clause. In
A-E below we consider possible morphosyntactic loci of agreement mark-
ing and interrelations between them. Semantic versus syntactic agreement
is discussed in F. Factors which may constrain agreement are summarized
under G.
(A) Noun class agreement in head-modifier (attributive) noun phrases
Noun class agreement can be marked on any type of modifier—adjectives,
including numeral and deverbal adjectives (participles), demonstratives, or
articles.
The head determines noun class agreement within an NP (there may not
necessarily be an overt noun class marker on the head noun itself—see
§2.6.2). This is a major criterion for recognizing which word is the head of
an NP (Nichols 1986). In 2.1, from Baniwa (North Arawak), ina3u
'woman' is the head since it determines the form of the proximate
demonstrative.
2.1. sua ina3u
DEM: FEM woman
'this woman'
In 2.2, from Mayali (Australian: Evans 1997: 129), an adjective 'good'
shows noun class agreement with the noun.
2.2. al-makkawarri al-mak
CLII(FEM)-lesser.salmon.catfish CLII(FEM)-good
'that good male lesser salmon catfish'
(B) Noun class agreement in possessive noun phrases
Noun class agreement is rarer in possessive constructions. In numerous
Bantu languages, a possessive, or 'associative' morpheme is marked for the
noun class of the possessed noun, e.g. 'knife', as in 2.3 from Swahili
(Welmers 1973: 175).
32 Classifiers
2.3. kisu ch-a Hamisi
NCL7-knife NCL7-POSS Hamisi
'Hamisi's knife'
Possessed nouns can show noun class agreement with the possessor. This
is illustrated with 2.4, from Paumari (Arawa: my own data). The possessed
noun, 'house', shows agreement with the feminine possessor, Bajara.14
2.4. Bajara gora-ni
female.name house-3SG.FEM
'Bajara's house'
The possessor noun can take a marker of agreement with the possessed
one. This is the case in Chamalal (Northeast Caucasian: Magomedbekova
1967: 388). Chamalal has five noun classes. In 2.5, the possessor, hek'wa
'man', belongs to Class 1 and the possessed noun, isa 'cheese', belongs to
Class 4; the Class 4 agreement marker appears on the possessor.15
2.5. hek'wa-ssu-1 Isa
man(cLl)-GN-CL4 cheese (CL4)
'man's cheese'
In Manambu, a Ndu language from the East Sepik province of New Guinea,
possessive constructions mark noun class agreement with the possessor, and
with the possessed noun within a a possessive morpheme (cf. Aikhenvald
1998a). This is illustrated with 2.6. Agreement is shown with arrows.

2.6. Pauline
Pauline(feminine) 3sGFEM-POSS-3 SGMASC story(masculine)
'Pauline's story'
Double agreement—with the possessor and with the possessed noun—is
also found in a few Bantu languages. In Shona, the possessive morpheme is
marked for the class of both possessor and possessed (Welmers 1973: 178).
Tuvana 'child' belongs to class 13 (diminutive plural), and imbwa 'dogs'
belongs to class 10 (animal plural).

2.7. tu-vana nembwa dz-a-tw-o

PL:CLl3-child and+dogs:CL10 CLlO-POSS-CLl3-their
'the little children and their dogs'
14
A similar phenomenon is found in Abkhaz (Northwest Caucasian: Hewitt 1979: 116;
Corbett 1991: 108), and in Wari' (Everett and Kern 1997).
15
A similar agreement pattern is reported for Kuot (non-Austronesian, New Ireland: Eva
Lindstrom p.c.), where there are two genders.
 Noun Class and Gender Systems 33
Noun class agreement in possessive constructions may depend on the
morphological subclass of a noun. In many Australian languages inalien-
ably possessed nouns display unusual agreement patterns and unusual
correlations between overtly marked noun classes and agreement types
(Evans 1994: 2; Kirton 1971: 13).16 The agreement of an inalienably
possessed noun 'name' with possessor in noun class is illustrated with
2.8a and 2.8b, from Yanyuwa:
2.8a. nya-ganymarda niya-wini
MALE-two CL:MASC-name
'his two names'
2.8b. rra-mangaji nanda-wini
FEMALE-that CL:FEM-name
'that name of hers'
In some languages noun class agreement is marked on adpositions, as in
Abkhaz (Northwest Caucasian: Hewitt 1979: 113-14), e.g. Axra ye-ze
(Axra 3SG.HUMAN.MALE-for) 'for Axra'.
Noun classes are very rarely marked on interrogative pronouns. This is
the case in Tariana (North Arawak), e.g. ku-ite tfari-tha (what-CL:AN man-
PRES.INTER.VISUAL) 'what (person) is this?', kwaka-whya namia-ni-whya
(what:INAN-CL:CANOE 3PL+sink-TOP.ADV-CL:CANOE) 'Which canoe was (the
canoe) they sank?' In Baining, a non-Austronesian language from East
New Britain, interrogatives 'who' and 'which' take noun class agreement
markers (Parker and Parker 1977: 21).17
(C) Noun class agreement outside a noun phrase
If there is noun class agreement outside a noun phrase, the verb agrees with
core constituents: subject (S and/or A) and/or direct object. There is often
agreement with just one core argument, more frequently with the subject
than with a direct object (see Corbett 1991: 110 ff. for more examples).
Bantu languages show consistent gender agreement with subject and direct
object; the subject agreement marker usually conies before the tense

16
There may also be occasional lexical 'exceptions'. In Tiwi, like in Yanyuwa, body parts
are assigned the gender/noun class of their possessor (Osborne 1974), but genitals are assigned
the gender of the opposite sex (Evans 1994: 2). Some Australian languages have several
patterns of agreement depending on body part lexemes. One of the most striking and com-
plicated examples is Gurr-goni (Evans 1994: 6; and R. Green 1995: 109 ff.).
17
Gender and number are marked on interrogatives in Bine, a Fly river language of New
Guinea with two genders (Fleischmann and Turpeinen 1975: 13) and in Kuot (Eva Lind-
strom p.c.); also see Pasch (1986: 154–5; 175) for the agreement on interrogatives in Mba
(Ubangi, Niger-Congo). Most Australian languages with agreement noun classes mark them
on at least some interrogatives; e.g. in Ngalakan noun classes are marked only on 'who',
while in Nunggubuyu they are marked on 'where'; see Dixon (forthcoming) for a detailed
description.
34 Classifiers
marker, while the object agreement marker comes between the tense prefix
and the root. In Swahili, the occurrence of the object prefix is optional
when the object is inanimate, but obligatory when it is animate. Example
2.9 shows verb agreement with animate subject (Maryamu) and object
('children'), from Swahili (Bresnan and McChombo 1986: 293).
2.9. Maryamu A a-li-wa-onyesha wa-totoo ki-su
Maryamu CLlA-PAST-CL2o-show CL20-children CL7-knife
'Maryamu showed the children a/the knife.'
Noun class agreement with both subject (A/S) and object is found in
some prefixing Australian languages (Dixon forthcoming). In the North
Kimberley languages (Worrorra, Ungarinjin, Wunambal) agreement is
found with S/O, and not with A, while Jingulu has noun class agreement
with A, and not with S/O (see §10.7, for interactions between clausal or
predicate categories and noun class).
In Paumari (Arawa), there is agreement either with A, or with S/O,
depending on the type of construction and constituent order (Aikhenvald
MS; Chapman and Derbyshire 1991; see §2.7.2 below).
Noun class agreement with a peripheral constituent is rare. In Lak (North-
east Caucasian), there is noun class agreement on adverbs. This is illustrated
with 2.10: ars 'son' belongs to Noun Class 1 which includes human males;
the numeral 'two' takes Noun Class 1 agreement marker -j-, while the
locative adverb 'at home' takes the agreement marker -w-; the marker on
the verb 'be' is 0 (markers are underlined) (Khaidakov 1980: 206).
2.10. k'i-j'a ars sa-wa 0-usar
two-NCLl:MALE son at.home-NCLl:MALE NCLl:MALE-be
'Two sons are at home'
Agreement is very rarely marked on complementizers. However, these
agree in gender with the subject of the complement clause in West Flemish
(Corbett 1991: 113).18
Noun class agreement may take place with a topical constituent,
independently of its syntactic function. Motuna (non-Austronesian,
Bougainville: Onishi 1994 and p.c.) has five noun classes. The verb in a
clause takes obligatory subject and object cross-referencing, and also
agrees in noun class with the topical constituent. In 2.11, the object nii 'I

18
Garifuna (North Arawak, Central America) is reported to have neutral, or default
agreement on complementizers; thus complementizers can be said to agree with the clause
they introduce (Munroe MS: 7):
buse-tina 1-un n-abinaha
want-TI.SERIES:lSG 3SG.NF-DATivE ISG-dance
'I want to dance'
 Noun Class and Gender Systems 35
(masculine)' is topical and its masculine noun class is cross-referenced on
the verb (Masa Onishi, p.c.).
2.11. nii Aanih-ki tangu-mu-u-ng
I(MASC):ABS Aanih(FEM)-ERG slap-lSGO + 3SGA-NEAR.PAST-MASC
'Aanih (a female name) slapped me (topic).'
In contrast, in 2.12 the subject, Aanih, is topical and its feminine gender
is cross-referenced on the verb (Masa Onishi, p.c.).
2.12. Aanih nii tangu-mu-i-na
Aanih(FEM) I(MASC:ABS slap-lSGO+3sGA-NEAR.PAST-FEM
'Aanih (a female name) (topic) slapped me.'
(D) Noun class agreement on several targets
The majority of languages mark noun class in more than one place in the
clause. Dyirbal marks gender on determiners and interrogatives. Burush-
aski marks gender on pronouns, adjectival modifiers, and verbs. In
Anindilyakwa (Australian: Leeding 1989) gender is marked on head nouns,
verbs, adjectives, and all types of pronoun. Bantu, some West Atlantic
languages, and the North Kimberley languages of Australia (Worrorra,
Ungarinjin, and Wunambal) mark gender on every type of noun modifier
and on the verb.
A remarkable property of Bantu languages is ALLITERATIVE CONCORD,
whereby the same noun class marker is repeated on modifiers and on the
predicate, as in 2.13, from Swahili (Corbett 1991: 117; Welmers 1973: 171).
2.13. ki-kapu ki-kubwa ki-moja ki-li-anguka
CL7-basket CL7-large CL7-one CL7-PAST-fall
'One large basket fell.'
!X66, a Southern Khoisan language, has five noun classes (Traill 1994:
20-2). They are marked by suffixes on nouns and also realized through
agreement on various targets (adjectives, relativizers, object markers).
Noun suffixes bear a strong phonological similarity to agreement markers;
an example of such an 'alliterative' concord is given in 2.14 (Traill 1994:
21). Noun class markers are underlined.
2.14. n a |na-i |a-i !xa-i t-i
I PAST see-NCLl:O lion-NCLl big-NCLl which-NCLl
| aa
I'aa |ii
lii k-i
k-i

dead is which-NCLl
'I saw a large dead lion'
Only some nouns from classes 1-4 can take a noun class prefix. Others
do not, e.g. dao 'road' in 2.15; concord is then not alliterative.
36 Classifiers
2.15. n a |na-i dao xa-i t-i
I past see-NCLl:O road big-NCLl which-NCLl
!nolisi |ii k-i
broken is which-NCL1
'I saw the broken up big road.'

In many Bantu languages the original alliterative character of the con-

cord has been obscured by phonological changes, and so prefixes retain
some phonological similarities without being exactly the same, as in Zulu
(Herbert 1991: 106):
2.16. uku-dla kw-ethu Koike ku-phelile
CLl5-eat CLl5-our CLl5-all CLl5-is.finished
'All our food is finished.'

Lowland Amazonian languages with large systems of noun classes which

can be overtly marked on the head noun as nominalizers (see (F) in §9.1) do
not allow multiple affixation on agreeing constituents. To mark agreement,
just the last classifier on the head noun is repeated on all the agreeing
constituents. Example 2.17 is from Tariana (North Arawak). The last
classifier, -puna 'stretch' (underlined), appears as an agreement marker
on the adjective and on the verb.
2.17. kara-ka-whya-puna hanu-puna
REL + fly-THEME-CL:CANOE-CL:STRETCH big-CL:STRETCH
na-ni-ni-puna
3PL-do-TOP.ADV-CL: STRETCH
'A big flying strip (i.e. a stretch (of land) for flying canoes = planes)
was made.'
The presence of agreement on multiple targets does not imply the exist-
ence of a multiple classifier system. We return to this in §9.2.
(E) Interrelations between different morphosyntactic loci of noun class
agreement
For the time being, it appears impossible to establish any hierarchy in
agreement preferences between members of closed classes, e.g. demonstra-
tives or personal pronouns, such that 'if agreement is marked in class X, it
must also be marked on class Y'. In many languages of the world noun
classes/genders are found only in personal pronouns (e.g. English;
Kaingang, a Je language from South Brazil: Wiesemann 1972; Rikbaktsa,
a Macro-Je language from Central Brazil: Wiesemann 1986; Guahibo
languages from Colombia). In other languages, such as Dyirbal, noun
classes are restricted to determiners, deictics and locative interrogatives.
 Noun Class and Gender Systems 37
It should be noted, however, that 3rd person pronouns and deictics are
often used interchangeably.
The existence of correlations between different morphosyntactic loci of
noun class agreement is another intriguing question. According to
Greenberg's Universal 31 (Greenberg 1963: 112; Corbett 1991: 111), 'a
language which has agreement of the verb in gender with subject or object
will also have agreement of the adjective with its head noun'. In other
words, the existence of predicate-argument noun class/gender agreement
presupposes the existence of head-modifier agreement. This appears to be
true in the vast majority of cases. In just a few cases noun class agreement
is found outside a noun phrase but not within it. Most such examples come
from prefixing Australian languages: in Marrithiyel (Green 1989) and
Murrinhpatha (Walsh 1976) noun classes (masculine and feminine) are
marked only on dative bound pronouns and on free pronouns, while in
Iwaidja (Pym and Larrimore 1979) noun classes are marked exclusively on
bound pronouns referring to the transitive subject (A).
This dependency may take a more complicated form. For instance, not
all modifiers may agree in gender with their head noun; there may also be
different systems of noun classes/genders in head-modifier and predicate-
argument agreement. We will return to this in §2.7.
If there is head-modifier agreement, then the following 'hierarchy'
operates: agreement on modifiers from open classes > agreement on
modifiers from closed classes > agreement in possessive constructions >
agreement on peripheral constituents.
These correlations do not necessarily mean that the agreement rules are
the same in terms of strictness or variability, or in terms of correlation with
discourse-pragmatic and semantic functions. Thus, Manambu distin-
guishes two genders both in head-modifier and in predicate-argument
construction; a correlation between topicality of a noun and its agreement
in gender is restricted to non-subject (A/S) participants in predicate-
argument constructions, and head-modifier constructions. In Baniwa,
agreement in pronominal noun class correlates with topicality of a noun
only in predicate-argument constructions, but not in head-modifier ones. In
this case different agreement types differ in their discourse-pragmatic
properties and variability—though they are marked in a similar way and
involve the same semantic oppositions (see Aikhenvald 1995b and (G)
below).
(F) Semantic and non-semantic agreement
Noun class can be assigned on the basis of semantic parameters, or a
combination of formal and semantic features (see §2.3). These principles
of assignment correspond to the notion of 'semantic' and 'syntactic', or
'mechanical' agreement, respectively (Corbett 1991: 225-6; Heine 1982a:
38 Classifiers
194). A typical example of semantic agreement would be feminine agree-
ment of a modifier with a noun like Russian vrac 'doctor'—a noun which
can equally refer to a man or to a woman. In this case, there is a choice
between feminine and masculine agreement; the feminine agreement will be
preferred if the sex of a particular doctor is known, and is important for the
speaker. Otherwise the functionally unmarked masculine agreement will be
used (see §2.5 on markedness). Semantic agreement is found in many Bantu
and non-Bantu languages of Africa, and in Papuan languages. We often get
semantic agreement with animate nouns and syntactic agreement with
other nouns, e.g. Swahili ki-faru m-kubwa (CL7-rhinoceros CLl-big) 'a big
rhinoceros' where the head noun, 'rhinoceros', is marked for Class 7 but it
governs agreement in Class 1 which covers animates. Syntactic or 'mechan-
ical' agreement is agreement which goes with non-semantic principles of
assignment (see Heine 1982a: 194).
The problem of a choice between semantic and syntactic agreement often
arises in the case of morphological 'mismatches' between the noun class
marked on the noun itself, and the agreement class it is assigned to.19 In
Mayali (Australian: Evans forthcoming: 107-16) the noun classes marked
on nouns and agreement class markers on modifiers are often isomorphic,
na-rangem na-kimuk (CL:MASC-boy CL:MASC-big) 'big boy', ngal-kohbanj
ngal-kimuk (CL:FEM-old woman CL:FEM-big) 'big old woman'. Nouns with-
out prefixes are assigned to the appropriate agreement classes according to
their semantics, e.g. bininj na-kimuk (man CL:MASC-big) 'big man' (or on
other principles: for instance, manner adverbials derived from nouns and
adjectives belong to Class 3 (which is basically the vegetable class) (Evans
forthcoming: 105-7)).
The regular 'mismatch' between noun class marked on the noun and the
agreement class it belongs to has been referred to as 'quirky' agreement
(Evans forthcoming; R. Green 1995). In Mayali quite a few nouns are
assigned agreement Class 1 (masculine) or 3 (non-flesh food) which is
different from their overt noun class marking. For instance, kun-(r)ak
'fire' has a Class 4 prefix but always requires Class 3 agreement (Evans
forthcoming: 109). (The quirky agreement is one of the indications in
favour of functional unmarkedness of Class 1, for animates, and Class 3,
for inanimates; see §2.5.3.) In Gurr-goni (Australian: R. Green 1995: 62)
'quirky' agreement shows a tendency towards a semantic 'regularization' of
an otherwise opaque noun class system. The majority of plants and plant
parts belong to Class 3, non-flesh food; but a very small number are in
Class 1 and 2, e.g. wartbirritji 'a white yam', gurlpuru 'a yellow yam'. All
these nouns require Class 3 agreement. Green reports an opposite tendency

19
To account for them, Evans (1997) suggested a useful distinction between head class
(overtly marked on the noun itself) and agreement class (realized in agreement).
 Noun Class and Gender Systems 39
as well. Yams which do not have any overt marking of class usually belong
to Class 3. Some speakers occasionally use Class 1 agreement markers for
these yams. According to R. Green (1995: 63), 'both patterns of reclassi-
fication act to unite yams within one noun class'. 'Double' agreement in
noun class is another instance of non-semantic agreement; it will be
discussed in §2.6.3, together with double noun class marking.
The Agreement Hierarchy suggested by Corbett (1979: 204; 1991: 226-
42) presents constraints for the choices between semantic and non-seman-
tic agreement. There are four types of agreement position (or agreement
target): attributive < predicate < relative pronoun < personal pronoun,
and 'as we move rightwards along the hierarchy, the likelihood of semantic
agreement will increase monotonically (that is, with no intervening
decrease)' (Corbett 1991: 226). The Agreement Hierarchy accounts for
the use of neuter on adjectives with Madchen 'girl' in German, and the
occasional anaphoric use of the personal pronoun sie 'feminine'.20
(G) Constraints on agreement
Agreement can be constrained by various language specific factors. In
some languages agreement only occurs in certain constructions. In North
Berber languages which have two genders (masculine and feminine) the
verb agrees in gender with its subject; however, there is no agreement in
cleft constructions (Laoust 1928: 201-3). In Palikur (North Arawak:
Aikhenvald and Green 1998) gender agreement shows up only in some
tense-aspect forms. Members of a given word class can show different
gender distinctions depending on their syntactic function; in Tamazight
and Kabyle (North Berber: Laoust 1928: 40; Vincennes and Dallet 1960)
demonstratives used as modifiers do not distinguish gender; they do
distinguish genders when used as noun phrase heads.
'Referential' constraints involve dependencies between agreement and
such discourse properties of nouns as definiteness and topicality. In some
Arawak languages of South America the form of noun class agreement on a
modifier is associated with the topicality of the head noun and with focusing
on a particular property. This is often the case in languages with a large
number of noun classes. In Baniwa (North Arawak: Aikhenvald 1996c),
-napi 'bone' can be considered part of the human body, in which case it
triggers agreement in the 'human' noun class, as in 2.18. If it is considered
as a vertical object, it triggers agreement in the 'vertical' class, as in 2.19;
and if it is looked at as a long object (e.g. a bone used to make a flute), it
triggers agreement in the noun class for 'long' objects, as in 2.20.

20
See Corbett (1991: 235-41) for further examples; reformulations of the Agreement
Hierarchy have been suggested by Barlow (1992).
40 Classifiers
2.18. ri-napi maka-dari
3sGNF-bone big-NCL:NON.FEM.ANIM
'a bone' (seen as an attribute of human body)
2.19. ri-napi maka-ne
3SGNF-bone big-NCL:VERT
'a big bone' (considered as a long vertical object, e.g. a leg bone)
2.20. ri-napi maka-pi
3sGNF-bone big-NCL:LONG.THIN
'a long bone'
In North Arawak languages, and in Abkhaz-Abaza, agreement in gender
is neutralized if the subject is preposed to the verb, as a means of focusing
the agent (Aikhenvald 1995b). In Garifuna, from the North Arawak sub-
group, predicate-argument agreement with the subject is obligatory;
agreement with an object depends on its definiteness: only definite objects
agree (Munroe MS: 2).
In Motuna (see 2.11 and 2.12) gender agreement on the verb depends on
the topicality of a noun. In classical Arabic agreement in gender in
predicate-argument constructions depends both on constituent order and
on definiteness. In verb-initial sentences, agreement in gender is optional
(Corbett 1991: 125; Russell 1984: 124-5); it is more likely to occur if the
noun phrase is definite. If the subject precedes the predicate, agreement in
gender is obligatory.21
There may be morphological constraints; for instance, agreeing modi-
fiers can divide into different morphological classes which determine their
noun class agreement possibilities. In Latin, adjectives fall into three
classes: those which distinguish three genders (masculine, feminine,
neuter), those which distinguish two (neuter vs. non-neuter), and those
which have just one form for all genders.
Phonological constraints on agreement imply that there is no agreement
because of some phonological property of a modifier. As pointed out by
Corbett (1991: 134), many adjectives in the spoken French of Paris which
end in a vowel do not distinguish genders. In Tsez (Northeast Caucasian)
noun classes are overtly marked only on verbs which begin with a vowel
(Bernard Comrie, p.c.). In Portuguese, adjectives which end in -e do not
distinguish gender agreement forms, e.g. urn homem gigante 'a giant man',
uma mulher gigante 'a giant woman'.
Finally, there may be lexical constraints; some modifiers simply fail to
show agreement, often due to some 'historical accident'. Many examples of
this sort are found among numerals. Cross-linguistically, smaller numbers
21
See also Corbett (1991: 124), for examples of referential constraints on agreement in
Swedish.
 Noun Class and Gender Systems 41
are more likely to display agreement in gender/noun class than larger ones.
In most Berber languages of North Africa, only the numerals 'one' and
'two' agree with the head noun in gender. In the Nakh languages of the
Caucasus (Chechen, Ingush, and Tsova-Tush) only the numeral 'four'
shows gender agreement. In a number of Dravidian languages (Kolami,
Parji, Naiki) which have two basic genders (male human and the rest) lower
numerals have special female human forms. In Russian odin 'one' distin-
guishes three genders, as do all adjectives, but dva 'two' and oba 'both'
distinguish only two genders (one form is used for masculine/neuter, the
other one for feminine). Historically they are residues of duals, and fewer
gender distinctions were present in the dual than in the singular (see
§10.1.1). Other numerals do not distinguish genders.22

2.4.3. Variability in noun class assignment and variable agreement

Variability in noun class agreement is often due to the semantic reassign-
ment of noun class, for instance, in the case of HYBRID nouns (see §2.3.4)
denoting professions.23
Variability in noun class assignment is found in systems with semanti-
cally transparent noun classes. In Australian languages, nouns with sex-
differentiable referents can trigger agreement in accordance with the
natural gender (note that the noun class marked on the noun itself remains
the same), e.g. Mayali na-garndegin na-rangem (CL:MASC-dingo CL:MASC-male)
'male dingo'; na-garndegin al-daluk (CL:MASC-dingo CL:FEM-female) 'female
dingo' (Evans 1997: 128).
(A) Semantic groups of nouns with variable noun classes
Many languages allow variable noun class marking on a number of nouns
like 'baby' and 'child', on some kinship terms and some inanimates. In
Dyirbal, bimu is both 'father's elder sister' (taking a feminine gender
marker) and 'father's elder brother' (masculine marker). Jaja 'baby' can
be specified as either masculine or feminine. (Alternately, these may be
viewed as different lexemes.)
22
In a way, one can say that numbers from 2 to 10 agree in animacy with the head noun
(Corbett 1991: 135). They have special collective forms optionally used with animate nouns,
e.g. dva muzciny (two:MASC/NEUT man-so.COUNTING.FORM) 'two men', dve zentschin-y (two:FEM
woman-SG.COUNTING FORM) 'two women', dvoe muzcinlzentschin (TWO:ANIM COLL man/woman-
PL.COUNTING.fORM), *dvoe okon 'tWO:ANIM.COLL windows+PL.COUNTING.FORM'. Another group
of constraints on agreement forms comes from interaction between classifiers and other
categories, e.g. noun class and number. See Chapter 10.
23
Some generic terms for animate beings take one agreement form though they denote
beings of either sex (termed 'epicenes'). It is up to the language as to what gender epicenes are
formally assigned to. In Russian, kit 'whale' takes masculine agreement, and akula 'shark'
always takes feminine agreement, though both can denote beings of either sex (Corbett 1991:
67).
42 Classifiers
Noun classes and genders can vary in different ways for different semantic
groups of nouns. In Brazilian Portuguese, this is widespread for nouns with
a human referent, especially the ones which refer to professions. Masculine
forms also have generic reference and are used when the sex of a referent is
not in focus: juiz 'judge (male, or generic)', juiz-a 'female judge'; ministro
'minister (male, or generic)', ministra 'female minister'. In the colloquial
language, variable gender is sometimes extended to sex-differentiable non-
humans, e.g. peixe 'fish (male or generic)', peix-a 'female fish (which has
caviar inside it)'.
In Ket gender can vary only for inanimates (Dul'son 1968: 62 ff.,
Krejnovic 1961). Then, for instance, a growing tree is masculine, a cut-
down tree is inanimate; an upright tree is masculine, and a tree with a
curved trunk is feminine. In languages with shape-based genders spoken in
the East Sepik region of New Guinea, change in (covert) gender of most
inanimate nouns signals change in shape (see Bruce 1984, on Alamblak). In
Manambu (Ndu family) nouns which denote male humans and higher
animates and long and thin inanimate objects are masculine, while those
which denote female humans and higher animates, and short and round
objects, are feminine. Consequently, nema-de wi (big-MASC house) is used to
refer to 'a big house' (which is extraordinarily long, or high, or both); and
nema wi (big+FEM house) describes 'a big house' (which is not necessarily
long, or high).24
Along similar lines, in Kxoe (Khoisan; Heine 1982a: 198) an inanimate
noun stem can be allocated to masculine or feminine gender depending on
its shape: masculine is associated with big, long, rectangular, and feminine
with small, round, broad, e.g. ngu 'hut', ngu-ma 'big rectangular hut
(masc.)'; ngu-he 'small, round hut (fem.)'.
Semantic choices may be more complex. Turkana (East Nilotic:
Dimmendaal 1983: 220) has three genders: masculine, feminine, and
neuter. By changing the gender of an animate noun, specific reference
can be made to a male, female or a young of the species:
fern, a-gete' 'female antelope'
masc. e-gete 'male antelope'
neut. i-gete 'small antelope of either sex'
A number of nouns referring to inanimates allow variable gender. Some
have two forms—masculine and feminine, e.g. 'grass' and 'tree'. Masculine

24
A smallish woman-like man can occasionally be treated as feminine, and a largeish
woman can be treated as masculine. The use of a different gender is impossible when the
shape cannot be changed (turtles are 'round' and always feminine), or when the 'masculinity'
is culturally important. Descent is strictly patrilineal, and so the word gwalugw 'patrilineal
clan' is masculine. Morphologically, gwal-ugw is the plural form of gwal which means 'father's
child (female or male)' and 'father's father'.
 Noun Class and Gender Systems 43
forms mean 'growing, green', and feminine forms mean 'dead, dry state',
e.g.
masc. ni-na 'green grass'
fem. na-na' 'dry grass'
Some inanimate nouns have three gender forms. The opposition of the
three is by size, e.g.
masc. e-mor-u 'rocky mountain, big stone'
fem. a-mor-u 'hill stone'
neut. i-mor-u 'pebble'
The degree of variability in agreement class assignment depends on the
language. In Dyirbal, variable class assignment is restricted to sex-
differentiable animals; the Class I/Class 2 correlation with male/female is
obligatory for humans. Each name of an animal has a fixed class member-
ship; however, exceptionally, noun class assignment can be changed to
stress the sex of a particular animal, e.g. 'to point out that a certain dog
is male bayi guda can be used' (Dixon 1982: 182). Usually, guda 'dog'
belongs to Class 2 (Dixon 1982: 180), and so the 'unmarked' usage would
be balan guda. Very occasionally, changing noun class can create a
pragmatic effect. In Dyirbal, yara 'man' belongs to Class 1, and so would
be referred to as bayi yara. However, Dixon (1982: 166) reports that a
hermaphrodite was once jokingly referred to as balan yara, with a feminine
Class 2 marker, pointing out his female characteristics. In this case, the
manipulation of noun class realized in agreement has pragmatic, as well as
semantic effect (see C below).
In these cases variability of noun class assignment is linked to a semantic
superclassing of nouns (see C in §2.4.4), most often into animate and
inanimate, or human and non-human.
Change in noun class agreement can be employed to distinguish dis-
tinct lexical entries. In Anindilyakwa (Australian) dirija 'dress' is treated
as feminine when understood as a piece of female clothing; it is treated as
a member of inanimate w-class when seen as a material (Julie Waddy,
p.c.).25 These nouns are said to have double or multiple gender, depend-
ing on their semantics. In Russian, Portuguese, or Anindilyakwa, only a
limited number of nouns can be assigned more than one gender. In other
languages, such 'reclassification' (which is reminiscent of the use of noun
classifiers—see Chapter 3—for disambiguating polysemous referents) is
much more widespread; this typically happens in languages with semantic
25
Just as in Turkana discussed above. Similarly, in Archi (Northeast Caucasian), lo can be
assigned to three classes, with a corresponding change in meaning. When assigned to Class 1,
it means 'boy'; when assigned to Class 2, it means 'girl', and when assigned to Class 4, it means
'young animal'.
44 Classifiers
gender assignment which involves parameters other than just sex or
animacy.
Variability in 'overt' noun class marking on the same root is the way of
creating new words, in languages with overt noun class marking. In Bantu
languages, e.g. Swahili, most stems usually occur with a prefix of one class.
Prefixes can be substituted to mark a characteristic of an object. M-zee
means 'old person' and has the human class prefix m-. It can be replaced by
ki- (inanimate class) to yield ki-zee 'scruffy old person' (Dixon 1982: 166;
also see Shepardson 1982 on the correlations between diminution and
choice of class prefix in Swahili).

(B) Limits of variability of noun classes

Variable gender assignment may be restricted to a few random exceptions,
and have no semantic content. In Hebrew, a few nouns can have either
feminine or masculine gender, without any change in semantics, e.g. dereh
'road, way', lason 'language', ruah 'spirit, wind'.26 Variable gender can be
found with recent loans. In Russian, kivi 'kiwi-fruit', a recent loanword, is
sometimes treated as neuter (since it is an undeclinable noun which ends in
a vowel), and sometimes as masculine, by association with the super-
ordinate noun frukt 'fruit' which is masculine.
(C) Functions of noun class variation
The choice of noun class agreement depends on what aspect of the noun is
highlighted. Some northern Australian languages allow different agree-
ment possibilities for certain nouns depending on the 'viewpoint' from
which the referent is seen (Evans forthcoming: 108). In Gunwinjgu kukku
'water, drink' triggers Class 4 ('neuter') agreement when it is seen as a part
of the landscape:
2.21. kun-ekke kukku kun-bo-gimuk
CLIV:NEUTER-that Water CLIV:NEUTER-CL:LIQUID-big
That water (i.e. river) is big.'
When considered a drink, it is assigned to Class 3 (which also covers
vegetable food).
2.22. yun yi-bongu-n man-ih kukku
don't 2sG-drink-NON.PAST CLIII:NON.FLESH.FOOD-this.here water
'Don't drink this water!'

26
Garifuna (North Arawak) is reported to have a different kind of variable gender assign-
ment. A subclass of nouns which covers plant names and body parts changes gender according
to the sex of the speaker (Taylor 1952). This requires further investigation.
 Noun Class and Gender Systems 45
Variable agreement is reported for other semantic groups. Kun-waral
'spirit' usually belongs to neuter Class 4 (which is also its 'head class',
kun- being the prefix of this class). When it refers to 'certain types of spirit',
especially malignant spirits, this noun takes agreement with Class 1, since
malignant spirits are a subclass of this class. Along similar lines, delek
'white ochre, white clay' takes Class 3 agreement when it is not associated
with art; but in association with painting, which is considered a typically
male activity, it takes masculine Class 1.
Gender variation is often used metaphorically to describe unusual situa-
tions (see A above, on the use of the word yara 'man' with the feminine
class marker, instead of the masculine one, to point out the female
characteristics of a hermaphrodite, in Dyirbal). In Manambu, ab 'head'
is usually feminine because of its round shape, but it is treated as masculine
when a person has a headache, since then the head feels heavy and un-
usually big (properties associated with the masculine gender). Spontaneous
manipulation of gender variation is widely used in jokes, and to describe
unusual situations, in Cantabrian Spanish. A speaker may use the mascu-
line hiju miu 'my son' in a deprecatory reference to a young girl, i.e. a
female not yet fully developed, or masculine oveju 'male sheep' with a
reference to a particularly meagre and unattractive animal independently
of its sex (Holmquist 1991: 59-60; see Table 2.2 above).
Different gender agreement systems for different word classes can coexist
in one language. Languages with distinct noun class systems used for
different types of modifier often allow variable assignment only for one
of the systems; usually, the largest one (see §2.7).

2.4.4. Determining the number of noun classes in a language

The number of genders in a language is determined by the types of agree-
ment (Corbett 1991; 1994a; 1994b; Chapter 10 below). For instance,
Spanish and Portuguese have two types of agreement, which correspond
to feminine and masculine genders. However, a number of problems may
create difficulties with establishing how many noun classes and noun class
agreement types a language has.
(A) Target genders and controller genders
The number of surface realizations of genders (TARGET GENDERS: Corbett
1991: §6.3) can be different from types of agreement (CONTROLLER GENDERS:
Corbett 1991: §6.3). In Rumanian nouns divide into three gender classes.
There are two surface markers of genders in singular and two in the plural,
but three combinations of these (Corbett 1991: 151). See Table 2.4.
46 Classifiers
TABLE 2.4. Gender marking in Rumanian
SG PL

Thus, nouns divide in three classes: (I) those which take -0 in the
singular and -i' in the plural, e.g. barbat 'man' as in barbatul e bun (man.DEF
is good) 'the man is good'; barbatii sint buni (men.DEF are good) 'the men
are good'; (II) those which take -0 in the singular and -e in the plural, e.g.
scaun 'chair' as in scaunul e bun (chair.DEF is good) 'the chair is good',
scaunele sint bune (chairs.DEF are good) 'the chairs are good'; and (III)
nouns which take -a in the singular and -e in the plural, e.g. fata 'girl' as in
fata e bun-a (girl.DEF is good) 'the girl is good',fetele sint bune (girls.DEF are
good) 'the girls are good' (Corbett 1991: 150-1). Thus, there are two target
genders and three controller genders (or agreement classes).
The gender system in the Dravidian language Telugu (Krishnamurti and
Gwynn 1985: 56-8) works on a similar principle (Table 2.5).

TABLE 2.5. Gender marking in Telugu

SG PL
-du males, moon, sun -rn
females
-ei yi

This is illustrated with the agreement on the verb 'to be' in 2.23-6.
2.23. waad(u) unnaadu
he be+MASC.SG
'He is'
2.24. ad(i) unnadi
She/it be + FEM/NEUTER.SG
'She/it is'
2.25. waaru unnaaru
they be+MASC/FEM.PL
'They (those persons: M,F) are'
2.26. aw(i) unnaayi
they be+NEUTER.PL
'They (those things) are'
 Noun Class and Gender Systems 47
Khinalug (Northeast Caucasian: Kibrik et al. 1972: 154-5; 118 ff.) has a
more complicated system. Nouns divide into four classes: 1: males; 2:
females; 3: most non-human animates and some inanimates; 4: all the
rest (including abstract nouns and other nouns, e.g. nimts 'louse'). Gender
agreement is found on demonstrative pronouns, headless adjectives, and
verbs. There are two sets of surface agreement markers. Set 1 is used with
past and future tenses of resultative and non-resultative aspects. Set 2 is
used to mark gender/number agreement on different verbs depending on
their morphological class (Table 2.6).

TABLE 2.6. Gender agreement in Khinalug

Gender Set 1 Set 2
Singular Plural Singular Plural
1 -du -dur 0/j/h- b/v/f-
2 -dae -dur b/v/f-
z/z/s-b/v/f-
3 -dae -3i(th) 0/j/h-
4 -3i -3i(th) 0/j/h- 0/j/h-

In Set 1 singular, genders 2 and 3 have identical marking. In Set 2

singular, genders 1 and 4 group together. And in both plurals, Genders 1
and 2, and 3 and 4, have the same marking. See further discussion in
Corbett (1991: 147-70).
(B) Restricted and residual genders
Some languages have genders with a very small number of members. In
Archi (Northeast Caucasian) there is a class which takes Gender 3 agree-
ment markers when singular and Gender 1/2 when plural; it consists of just
two nouns ('people' and 'population'). In Tsova-Tush (Nakh) there is a
gender which contains only four nouns (Corbett 1991: 171).
In Spanish, Romansch, and Portuguese there is a residual neuter gender
which is only used for anaphoric agreement (cf. Corbett 1991: 214-15).
Portuguese has two genders realized in agreement within a noun phrase
(see examples 1.1 and 1.2). However, demonstrative pronouns have three
forms: masculine, feminine, and the third one, which goes back to an
original neuter in Latin, e.g. masculine este, feminine esta, neuter isso
'this'. These neuter forms are used as resumptive anaphoric pronouns for
'default' agreement (see §2.5). They are also used in fixed expressions, e.g.
por isso (for this:NEUT) 'this is why', isto e 'that is, i.e.', isto! 'This is what it
is, you are right'. Obviously, neither Portuguese, Spanish, nor Romansch
can be said to have three genders synchronically; however, the residual
48 Classifiers
'anaphoric' gender is a useful indication of the previous stage of the
language.
In some languages gender distinctions are restricted to a subclass of
nouns and adjectives. In Ayacucho Quechua only a few nouns with human
referents require agreement with a closed class of adjectives borrowed from
Spanish, e.g. loko maqta 'crazy:MASC boy', loka sipas 'crazy:FEM girl'. A few
borrowed nouns with a human referent distinguish feminine and masculine
forms, e.g. biyudo 'widower', biyuda 'widow' (Parker 1969: 34-5). There is
no agreement elsewhere in the language. It is problematic whether such
'exceptions' should be considered separate noun classes at all.27

(C) Subgenders and superclassing

Further grouping of noun classes and genders involves SUBGENDERS and
SUPERCLASSING. Subgenders can be defined as 'minor agreement classes
which control minimally different sets of agreement' (Corbett 1991: 163).
Innovative subgenders based on the feature of animateness and personality
are characteristic of Slavic languages (Laskowski 1988). Russian has three
genders: masculine, feminine, and neuter. Masculine nouns fall into two
further subclasses, animate and inanimate. In the animate subclass, accu-
sative singular (on nouns and modifiers) coincides with genitive singular.
In the inanimate subclass, accusative singular (on nouns and modifiers)
coincides with nominative (as is also the case for neuter nouns) (see
Table 2.7).

TABLE 2 .7. A fragment of the Russian nominal paradigm

Masculine animate Masculine inanimate Neuter Feminine

NOM SG bol's-oj mal'cik bol's-oj dom bol's-oe selo bol's-aja kosk-a
'a big boy' 'a big house' 'a big village' 'a big cat'
ACC SG bol's-ogo mal'cik-a bol's-oj dom bol's-oe selo bol's-uju kosk-u
GN SG bol's-ogo mal'cik-a bol's-ogo dom-a bol's-ogo sel-a bol's-oj kosk-i

Northwest Caucasian languages (Abkhaz-Abaza) have a basic distinc-

tion between human and non-human noun classes. This distinction is
manifested in the choice of plural morpheme and agreement with numerals
and with adjectives in the plural. Male and female subclasses are distin-
guished within 3rd person possessive and cross-referencing morphemes
(Hewitt 1979: 44-8).

27
A similar phenomenon in Tagalog has been pointed out to me by Randy LaPolla. Certain
nouns referring to humans and adjectives used to modify them, most or possibly all of which
are loans from Spanish, distinguish two genders, e.g. loko-ng Pinoy (crazy:MASC-ATT Philippine)
'a crazy Philippine man', loka-ng Pinay (crazy:FEM-ATT Philippine) 'a crazy Philippine woman'.
 Noun Class and Gender Systems 49
Chechen and Ingush (Nichols 1989a) has five lexically determined noun
classes, formally marked by the appearance of an agreement prefix. There
are two human classes, which can be grouped into a human 'macroclass' (or
macrogender: Nichols 1989a: 162). For other classes, only a few semantic
generalizations can be made (e.g. fruiting trees and fruits and wild animals
are in the J class; non-fruiting trees are in the D class; names of manufac-
tured items are distributed among the D, J, and B classes). On the whole,
semantics is not a good predictor of gender. Surface markers used for these
classes partially overlap (see Table 2.8).

TABLE 2.8. Noun classes in Ingush

Class Prefix singular Prefix plural
Human feminine b(ld)-
Human masculine V- b(ld)-
J j- j-
D d-
Bd b- d-
Bb b- b-

An unmarked agreement class may replace the inherent class of the

noun. In Ngalakan (Merlan 1983) demonstratives can agree with the
head noun, or can simply show the unmarked prefix rnu- 'masculine
gender', unless the referent of the head noun is female.
This is described as CONCORDIAL SUPERCLASSING (Evans 1997; Harvey
1997; Sands 1995: 264-5). Concordial superclassing may result in the
creation of 'macroclasses', which are superimposed onto the system of
noun agreement classes. For example, Gaagudju (Australian: Harvey
1997: 153) has four noun agreement classes, with distinct class prefixes:
1: human males, most animates, European material objects, rain; 2: human
females, some animates; 3: plants and their parts, weapons; 4: abstract
entities, body parts, fire, geographical features, temporals. Noun class is
marked on adjectives, demonstratives, quantifiers, and in absolutive pro-
nominal prefixes on the verb. With demonstratives as modifiers, all nouns
with an animate referent tend to have Class 1 agreement, and all the
inanimates Class 3 agreement. Thus, Gaagudju can be said to have two
macroclasses: animate and inanimate.28

28
Alternatively, one may say that there are homophonous class markers used just with
demonstratives. This solution does not help us solve the problem concerning the 'number' of
noun classes in this language.
50 Classifiers
Superclassing does not necessarily reduce the number of classes. In some
languages 'superclassing' indicates the overlap of different classes.
Lokono (North Arawak: Pet 1987: 25-7) distinguishes two genders,
masculine and feminine, in the singular (see (C) in §11.2.1, on the semantics
of gender in Lokono). There are three distinctions in singular and in plural
which are based on an interaction of feminine and masculine gender, and
the feature human/non-human. All non-humans regardless of their sex or
number are referred to with an anaphoric pronoun tho and require femi-
nine agreement on modifiers and on the verb. Plural humans are referred to
with a plural anaphoric pronoun ne and require plural agreement. Singular
masculine nouns are referred to with li and require masculine agreement,
while singular feminine nouns require feminine agreement and are referred
to with tho.

2.5. Markedness and resolution in noun classes

Markedness relations in noun class systems are different from those in

other noun classification devices. I discuss markedness in §2.5.1. The
notion of noun class resolution is explained in §2.5.2. Then, in §2.5.3, I
consider different markedness relations in noun class systems.

2.5.1. Markedness
There are two main types of markedness—formal and functional. A term
in a system is formally unmarked if it has zero realization or a zero
allomorph. If the terms in a system, save one, are only used in specified
circumstances, and the remaining term is used in all other circumstances,
then it is said to be functionally unmarked (cf. Dixon 1994: 56-7;
Aikhenvald and Dixon 1998: 60).
A useful summary of criteria found relevant to markedness was provided
by Greenberg (1966); also see Croft (1990: 71).29 Only criteria relevant to
functional markedness are given here (Greenberg 1966: 25-30).

29
See Croft (1996) and Andrews (1990) on further issues concerning markedness relations.
When grammatical categories have more than two values, one value is marked relative to
another (cf. Croft 1990: 66; 1996). Markedness relations may be dependent on other categories
which correlate with them, and may be restricted to grammatical subclasses. For instance, in
the vast majority of the world's languages singular numbers are less marked than non-singular
ones (e.g. Tsonope 1988, on the relative markedness of singular and plural noun classes in
Setswana, a Bantu language). In some languages, the unmarked form is the collective one; and
the singular noun has a special singulative marker. This is an areal feature shared by a number
of families in the Northwest Amazon—Arawak (Resigaro, Tariana), East Tucano, and
Guahibo (e.g. Cuiba: Kerr 1995). There, noun classifiers are used as singulative markers;
and a noun can be pluralized only if it contains a noun classifier. In this case, formally
 Noun Class and Gender Systems 51
(i) The unmarked value of the form will refer to either value (marked or
unmarked one) in certain contexts—e.g. the unmarked term can be used for
a supercategory which covers all the terms.
(ii) In certain grammatical environments, only the unmarked value will
appear (see below on gender resolution and neutral, or default agreement).
(iii) The unmarked category is the one most frequently used (or the one
that is used at least as frequently as each marked one).
(iv) The marked category displays syncretization of its inflectional possi-
bilities with respect to the unmarked member; i.e. there are 'at least as
many distinct forms in the paradigm with the unmarked value as in the
paradigm with the marked value' (Greenberg 1966: 27).
(v) The unmarked category is realized in neutralized contexts.
Formal markedness and functional markedness may correlate, but they
do not necessarily always go together (see Corbett 1991: 291; Hayward and
Corbett 1988, for discussion of Qafar, an East Cushitic language; Bulygina
and Shmelev 1996, for Russian30). There are systems in which all noun
classes are equally formally marked and no relations of functional marked-
ness can be established.
A noun class can be considered functionally unmarked under the follow-
ing conditions:
(a) It is used as a generic term and for indefinite reference.
(b) It is used when the noun class distinction is neutralized or is of no
relevance.
(c) It may be used in default, or neutral agreement (see Corbett 1991:
206 ff., and below).
A functionally unmarked noun class is likely to be one of the largest
classes.
Noun class systems may have a special agreement form with heads, or
controllers which are not specified for any agreement class. This is called
neutral or default agreement (Corbett 1991: 203 ff.). These controllers are
called 'non-prototypical'. The range of non-prototypical controllers varies.
These may include infinitive phrases, nominalizations, and dummy
unmarked nouns are also functionally unmarked: they are used if number does not have to be
specified. This is called 'local markedness' or 'markedness reversal' by Croft (1990: 66, 144-5).
This may be limited to certain semantic subclasses. In some languages this 'reversal' is limited
to objects that 'naturally occur in groups and are difficult to individuate' (Croft 1990: 66, 144-
5), as in Semitic, Berber and Nilo-Saharan languages. In Amazonian languages, this is found
with inanimate nouns whose referents are easily individuated, and never with nouns which
have a collective referent. There may be different markedness relations between noun classes
and classifiers for animate and inanimate nouns.
30
For further discussion of relative markedness of Russian genders, see Jakobson (1984).
52 Classifiers
elements. The use of one of the noun classes for default agreement is an
indication of the functional unmarkedness of this class. In a two-gender
system, Hebrew uses masculine form for neutral agreement. Masculine
gender is unmarked—both formally, and functionally. In Zayse (Omotic:
Hayward 1989), feminine gender is unmarked and is used for default
agreement (see Fraser and Corbett 1997, for the notion of default gender
in Arapesh).
There may be unique forms used just for default agreement, e.g. neutral
resumptive pronouns in Spanish and Portuguese (discussed in §2.4.4; cf.
discussion of Spanish in Corbett 1991: 214-15).

2.5.2. Noun class resolution

Languages differ in the ways in which they deal with the coordination of
nouns belonging to different noun classes/genders, or associated with differ-
ent classifiers. This choice of a noun class for coordination can be referred to
as a 'resolution rule' (Givon 1970). The choice of the agreeing form has to do
with the functional markedness of one, or more than one, form.31 Resolution
rules in coordinating nouns which belong to different classes also provide
evidence in favour of the relative markedness of noun classes.
The problem of noun class/gender resolution is sometimes solved
semantically, and sometimes syntactically, or by a combination of these
methods. In some languages it cannot be solved at all; then markedness
relations cannot be established.
SEMANTIC RESOLUTION involves exclusive reference to the meaning of the
conjoined noun phrases (Corbett 1991: 269-78). It usually involves group-
ing nouns into larger classes, based on more general semantic features,
usually dividing nouns into animate, or human and the rest.
In Luganda and ChiBemba conjoined nouns with human referents are
treated as plural class 2 (human), even if none of the referents actually belongs
to this class (Givon 1970: 253–4; 1971: 38-9): see 2.27, from ChiBemba.
2.27. omu-kazi es-sajja ne olu-ana
CLl-woman CL5-fat.man and CLll-thin.child
ba-alabwa omusajja
CL2-were.seen (by).the.man
'The woman, the fat man and the thin child were seen (by) the man.'
If neither of the conjuncts denotes a human, class 8 is used on the
predicate:
31
Features which require resolution are person, number, and gender/noun classes. Person
resolution rules and person hierarchies have been discussed at length by Zwicky (1977: 718,
725), and summarized by Corbett (1991: 262-3). Number resolution rules are stated in
Corbett (1991: 263-4).
 Noun Class and Gender Systems 53
2.28. en-te omu-su eki-be ne ely-ato
CL9-COW CL3-wild.cat CL7-jackal and CL5-canoe
bi-alabwa omusajja
CL8-were.seen (by).the.man
'The cow, the wild cat, the jackal and the canoe were seen (by) the
man.'
Conjoining nouns denoting a human and a non-human is considered
'less natural', or 'less grammatical' and tends to be avoided (see Givon
1970: 253).
In some Northern Australian languages resolution principles are directly
associated with superclassing (also see C in §2.4.4). The functionally
unmarked masculine noun class 1 is used to mark agreement of conjoined
plurals of any class in Mayali (Nick Evans, p.c.).
In SYNTACTIC RESOLUTION, one of the existing forms will be used, as in
Portuguese, French, Modern Hebrew, or Hindi, and it can then be con-
sidered the unmarked one. In Portuguese, when the conjuncts are of
different genders, the masculine form of the adjective is used:32
2.29. um menino e uma menina bonit-os
INDEF:MASC boy and INDEF:FEM girl beautiful-PL:MASC
'beautiful boy and girl'
The use of a default, or neutral form is another means of resolution.
Lama (Gur) has a default class—the resumptive wa is used to refer to
conjoined elements of any two classes if they undergo left dislocation, as in
2.30 (Yu 1988: 332-3):
2.30. yo nka na nawu nku me hem wa
child DET.CL5 and bull DET.cL.3 I pulled PRONOUN
'The child and the bull, I pulled them.'
There may be MIXED SEMANTIC AND SYNTACTIC RESOLUTION. In Latin,
semantic resolution is found with nouns denoting persons; and syntactic
resolution is used with other nouns. Masculine agreement is used when
conjuncts of different genders denote persons; if not, neutral agreement is
used (Corbett 1991: 284 ff.).33

32
Another strategy which is alternative to syntactic resolution is agreement with just one
conjunct (see a discussion on Swahili in Corbett 1991: 265-6; see also Steinberg and Caskey
1988: 301; Bokamba 1985).
33
Languages can combine several strategies of syntactic resolution, as does Hausa
(Schwartz et al. 1988), or several semantic and syntactic principles, as does Spanish (Steinberg
and Caskey 1988). A very complicated example of mixed semantic and syntactic noun class
resolution is found in Gurr-goni (Australian: R. Green 1995). Gender resolution may even be
determined on phonological principles, as in Xhosa, a Bantu language; see Voeltz (1971) and
Pullum and Zwicky (1986).
54 Classifiers
Some languages have no resolution strategies: nouns which belong to
different classes cannot be conjoined. Then, markedness relations cannot
be established (see C in §2.5.3).

2.5.3. Markedness relationships in noun classes

The following situations can be distinguished with respect to the functional
markedness of noun classes.
(A) In a system with several classes, one is marked and the others are
unmarked
The unmarked gender is often masculine, as in Indo-European or
Afroasiatic languages; more rarely feminine, as in Guajiro, Lokono
(Arawak: Aikhenvald 1999a), Jarawara (Arawa: Dixon 1995); Wangkumara
and Wagaya (Australian: Alpher 1987: 174; Breen 1976a: 336; 1976b: 340,
590).
A term—or an agreement form used as generic, and for indefinite
reference—may be functionally unmarked. In traditional English, the
masculine pronoun he was used as an unmarked generic term for human
reference (see the discussion in Alpher 1987), and so was functionally
unmarked. In French, masculine gender can be considered functionally
unmarked since it is used in the case of generic reference. Masculine plural
like les americains can refer to a mixture of the sexes, while feminine plural
les americaines refers only to female Americans (Schane 1970; cf. also
Greenberg 1966: 30-1; Corbett 1991: 291). Indefinites and interrogatives
require masculine agreement in Manambu, a Ndu language from the East
Sepik region of New Guinea, with two genders (masculine and feminine).
This shows that masculine is functionally unmarked—see 2.31. The use of
the feminine agreement form (ha-l) would imply that the speaker already
has some idea about the shape, or the sex of the referent.
2.31. agua-jap ha-d
what-thing DEM-MASC
'What is this?'
In Russian, the choice of masculine agreement with interrogative and
indefinite pronouns indicates the functional unmarkedness of masculine
(Bulygina and Shmelev 1996: 103).
In Gurr-goni, a northern Australian language with a 'typical' Australian
system of four noun classes, class prefixes used on the indeterminates -nji
'what/which/any thing' and -njatbu 'whatsitsname' provide interesting
evidence in favour of the relative functional markedness of noun classes.
All the four noun classes are formally marked with prefixes (R. Green 1995:
64-7). If the semantic domain of a referent is known, the appropriate noun
 Noun Class and Gender Systems 55
class prefix is used. Class I 'masculine' is used only when the referent is not
known, as in 2.32.
2.32. a-nji nji-na-ni
3CLl-what 2MIN.A.3MIN.O-See-PRECONT
'What did you see?'
In contrast, in 2.33 the indefinite/interrogative takes the Class 3 prefix
mu- 'vegetable food', because the domain of the referent is known: the
speaker is trying to remember a plant name. The indefinite/interrogative
form is underlined.
2.33. mu-njatbu muwu-me-nji
3cL3-whatsit 3AUG.A.3CL3:O-get-PKECONT
awurr-ni-0
3AUGS-be-PRECONT
'What's that CL.3 thing (vegetable food) they were getting?'
Class 1 agreement is also used as 'generic' agreement with reference to a
group of referents of a different class. In 2.34 the speaker enumerates
members of Class 3, 'vegetable food', but refers to them with the Class 1
prefix on the indefinite pronoun (underlined).
2.34. njiwu-ba-rri mitja a-nji
lAUG.A.3MiNO-eat-pRECONT vegetable.food(cL3) 3cLl-what/which
djuka dilip . . .
sugar(CL3) tea-leaves(cL3)
'We ate vegetable food, which one (kind), sugar, tea-leaves . . .'
If a language has concordial superclassing (when one agreement class
can be used to replace some, or all other), the class which is used as the
superclass can be considered unmarked (Sands 1995: 264, and C under
§2.4.4). Australian languages with superclassing typically use masculine as
the superclass (cf. (C3) in §12.1.3).
(B) Noun classes can have different properties with respect to markedness
relations depending on semantics, and on syntactic contexts
Concordial superclassing in Australian languages often provides evidence
in favour of different functionally unmarked classes for animates and for
inanimates. Mayali (Evans 1997) has a 'typical' Australian system of four
noun classes. In the case of a mismatch between head class and agreement
class the class assigned to animates is always Class 1 'masculine', and the
class assigned to inanimates is always Class 3 'vegetable food'.
Different classes of modifiers can have different markedness relations. In
Gurr-goni Class I 'masculine' is used as the functionally unmarked choice
with indefinites and interrogatives. But there are also instances in which
56 Classifiers
Class IV is used as a functionally unmarked one (R. Green 1995: 66). There
are two adjectives whose 'scope of reference is universal: everything or
nothing': -marrman 'good, well' and -yalang 'not having anything'. Thus,
gun-yalang (3CLIV-not.having.anything) can be used to deny the possession
or existence of something from any of the four noun classes. In the same
way, gun-marrman (3CLIV-good, well) can be used to refer to the state of any
person, or anything independently of its class.
Some languages provide contradictory evidence for markedness. The
Western Torres Strait language has two semantically assigned 'genders',
feminine and masculine. The feminine gender is the obvious 'candidate' to
be considered an unmarked gender, since it is the one used for the majority
of instances of default agreement (Bani 1987). All plurals are treated as
'feminine'. However, agreement with an unidentified human being (at a
distance) will always be masculine (for comparable data from Slovene, a
Slavic language with a three-gender system, see Corbett 1991: 291).
Different word classes can display different relations of markedness. In
Ngalakan masculine gender is used as a superclass only for demonstrative
modifiers (Merlan 1983).
(C) None of the noun classes can be considered unmarked
This is the case for Ungarinjin (Rumsey 1982), Tiwi (Osborne 1974),
Alamblak (Bruce 1984), or Dyirbal (Dixon 1972). In Tamil, rational nouns
(which include humans, gods, and other mythical beings) cannot be con-
joined with irrational ones. The Australian language Ungarinjin does not
allow conjunction of nouns belonging to different classes. Then the same
verb must simply be repeated with each conjunct, as in 2.35 (Rumsey
1982: 137).
2.35. ungalu m-ininani
beet.like. tuber M.CLASS.OBJECT-she.put
banimbum w-ininani
carrot.like.tuber W.CLASS.OBJECT-she.put
'She put down a beet-like tuber and a carrot-like tuber (lit. she put
down a beet-like tuber, she put down a carrot-like tuber).'

2.6. Realization of noun classes

In a language with noun classes, noun class is realized through agreement

outside the noun itself. In some languages the noun class of some, or all
nouns can be inferred from its form. These languages are said to have an
'overt' noun class. In contrast, languages in which the form of a noun tells
us nothing about its class are said to have a 'covert' noun class. These, and
 Noun Class and Gender Systems 57
related problems, are discussed in §2.6.1. Morphological means used to
mark noun classes are discussed in §2.6.2. Double marking of noun class is
considered in §2.6.3.

2.6.1. Overt and covert noun class marking

Overt noun class marking is common in Bantu languages. They have large
systems of affixes which are typically portmanteau morphemes of noun
class and number (see D in §10.1.1), e.g. Sesotho mo-tho'person' (Class 1),
ba-tho 'people' (Class 2), se-liba 'spring, well' (Class 7), li-liba 'springs,
wells' (Class 8) (Demuth et al. 1986: 455); Kikuyu ha-ndu 'place' (Class 16),
u-ndu 'event' (Class 14) (Denny and Creider 1986: 217).
In Apurina (Arawak, Brazil) feminine nouns tend to end in -ro, and
masculine nouns tend to end in -ri (Facundes 1994: 38). In Portuguese,
similar to many other Romance languages, nouns which end in -a are
feminine, and those which end in -o are masculine (with a few exceptions,
e.g. dia 'day' which is masculine, and mao 'hand', which is feminine).
Chechen-Ingush languages have a strong correlation between the initial
consonant of a noun and its gender. For instance, nouns beginning with d
or t favour D-class, and nouns beginning with b or m favour B-class
(Nichols 1989a: 165; see §2.3.4).
Overt and covert marking can be viewed as two extremes of a continuum.
The degree of 'overtness' of a noun class may depend on case and number.
In Russian, gender is almost always shown in the nominative case. In
oblique cases there is a certain amount of syncretism between the inflec-
tions of nouns of different genders, and so it is more difficult to tell from
the form of the noun what gender it belongs to.
Overt marking of noun class may depend on other factors. In numerous
Australian languages the degree of overtness of noun class marking is
related to its morphological transparency. In Wardaman (Australian)
many nouns have a prefix of one of three classes: yi- 'animate and human
beings, meat, body parts'; ma- 'flora', wu- 'the rest'. However, prefixes are
often morphologically fused with the root, and are then synchronically
inseparable from it, e.g. wuja 'fire', magulu 'cheeky yam'; only a few nouns
can occur with either ma- or wu- prefix (Merlan et al. 1997). And numerous
nouns, e.g. names for animals, trees, and flora, do not occur with a class
prefix (Merlan 1994: 61).
Not all nouns in a language have to be overtly marked. Babungo
(Grassfields Bantu, Benue-Congo; Schaub 1985: 172) has fourteen noun
classes which differ as to the degree of their overt marking on the noun.
Only eight of them require a noun class prefix, one has a noun class suffix,
and five classes have no overt marker on the noun at all.
In Anindilyakwa (Australian) only inherited nouns (not loans) are
58 Classifiers
overtly marked for noun class; the borrowings receive no overt class mark-
ing. This may be an indication that overt noun class marking was produc-
tive at an earlier stage, but then ceased to be productive; any noun which
was borrowed into the language after the noun class marking stopped
being productive does not have a prefix (Leeding 1989; Sands 1995: 260).
Some languages seldom or never mark gender on the noun itself, e.g.
Arawa languages (South Amazon), Ndu languages (East Sepik region of
Papua New Guinea), Zande (Niger-Congo), or !Xu (North Khoisan)
(Heine 1982a: 193). This is 'covert' gender.
In some languages nouns can be either marked or unmarked for gender.
The marked form tends to be more specific than the unmarked form. The
presence of a gender marker may correlate with definiteness in discourse
(as in Gola: West Atlantic; see A in §12.1.3). In Turkana (Eastern Nilotic:
Dimmendaal 1983: 221) the gender prefix is frequently omitted from names
of animals in folk tales when the names are used generically. Overt gender
marking in Alamblak (Lower Sepik) is used to focus on the sex of an
animate referent, or indicate a change in size of an inanimate one.
The overt noun class marking on the noun can be omitted in Baniwa,
Tariana (North Arawak) and in Tucano languages. The overt marking
correlates with the individuation of the item and the focusing of a criterial
property, usually related to shape and size. In Tariana, sawari means
'thread (in general)'; and sawari-kha (thread-CL:CURVILINEAR) means 'long
and thin thread'. In Baniwa, tfinu means 'dog (of any size or sex, usually
non-feminine)', tfinu-da (dog-CL:ROUND) means 'smallish dog'.
Languages with semantically transparent noun class assignment tend not
to mark class on the noun. In languages of this type, only nouns with
human and, more rarely, with animate referents are likely to have their
gender overtly marked. For instance, in Manambu masculine and feminine
gender (assigned according to transparent semantic principles) are not
marked on the noun.34
'Mismatches' between the overt noun class marking and the agreement
class were discussed under F in §2.4.2.

2.6.2. Morphological realization of noun classes

Noun classses are never marked with free morphemes (unlike other noun
categorization devices). Various types of morphological processes are used
for noun class marking: (A) external affixation, (B) apophony, or vowel
34
The lexemes ta:kw 'woman' and du 'man' are used with a few kinship nouns and a few
other nouns with a human referent for specification of sex, e.g. yanan 'grandchild', yanan-takw
'granddaughter', yanan-du 'grandson'. In Tariana (North Arawak) a number of derivational
affixes are used to distinguish masculine and feminine nouns with human referents, e.g. ha-nifi
'father (parent-MASC)', ha-dua 'mother (parent-FEM)'.
 Noun Class and Gender Systems 59
changes. Suprasegmental processes (tone patterns, or change of stress) are
almost never used (see (C)). Noun classes are never realized suppletively, or
via reduplication. Another, very rare, method of morphological realization
of noun classes is via repeaters (see (D)).
(A) External affixation
Suffixes and prefixes are the most frequent realizations of noun classifica-
tion. In the majority of Indo-European and South American Indian
languages, noun class agreement in noun phrases is realized with suffixes.
Some languages use a prefix and a suffix together (sometimes called
circumfixes, confixes, or simulfixes), e.g. t- . . . -t 'feminine marker' in
Berber languages.
Suffixes are more common than prefixes across the world's languages,
and noun class suffixes are found more frequently than noun class prefixes.
Languages sometimes employ either suffixes or prefixes depending on the
type of agreement target and the type of modifier. In Papuan languages of
the Torricelli family, head-modifier agreement on adjectives is marked with
suffixes, and predicate-argument agreement is realized through prefixes
(e.g. Arapesh, Yimas, Monumbo, Olo: Foley 1986: 85). In Bainouk (West
Atlantic) some demonstratives take prefixed agreement, and some take
suffixed agreement in noun class; agreement markers with adjectives,
numerals, and interrogatives are prefixes (see (D) below). Limilngan
(Australian: Harvey forthcoming) has four noun classes: one for humans,
one for animals, one for plants, and one residual. Noun class agreement is
marked with prefixes on adjectives and pronominal possessives, and with
suffixes on demonstratives. Tiwi (Australian: Osborne 1974: 51) has two
noun classes, masculine and feminine; these are marked with suffixes on
nouns, adjectives and the interrogative 'who/what', and with prefixes to
verbs and demonstratives. In Nunggubuyu, an Australian language with
four noun classes (Heath 1984: 163, 274), noun class is marked with
prefixes to adjectives, nouns, and demonstratives, and with suffixes to
demonstratives used as predicates (Heath 1984: 272).
Prefixes mark noun classes in most Benue-Congo, Togo Remnant, West
Atlantic and Eastern Nilotic languages. Suffixes are used for marking noun
class in Gur (Voltaic), some West Atlantic (e.g. Ful), a few Niger-Congo,
Khoisan, and Afroasiatic languages (Heine 1982a: 194).
(B) Apophony, or vowel changes
Languages rarely use vowel change to mark noun class agreement. How-
ever, in Jarawara (Arawa) vowel changes are used to mark feminine and
masculine agreement (Dixon 1995; Dixon and Vogel forthcoming). Typic-
ally for an Amazonian language, gender is not overtly marked on nouns. It
is marked on some adjectival and deictic modifiers, and on some possessed
60 Classifiers
nouns within a possessive NP—these cross-reference the gender of the
possessor. There is also gender cross-referencing on most types of verbal
suffix (and on a verb root itself when no suffix follows). For possessed
nouns, there are six types of gender marking which involve vowel alter-
nation and external affixation, e.g. ilo alternation in the last syllable, e.g.
fem, noki, masc. noko 'eye, face', or e/a alternation in non-final syllable(s),
e.g. fem, tame, masc. teme 'foot'; fem, anate, masc. enete 'chin'.35
Another notable example of noun classes marked by vowel alternation is
Marind, spoken in southern Irian Jaya (Drabbe 1955: 22-3; Foley 1986: 82-
3). Marind has four noun classes which trigger head-modifier type agree-
ment with demonstratives and adjectives. Noun class is indicated by the
vowel of the stem-final syllable of some nouns, and of modifiers. The first
class, with a characteristic vowel e, contains male humans; the second class,
with vowel u, contains female humans and animals; the third class mainly
consists of plants and trees, and is characterized by e, a, or o. The fourth
class is a residual class which contains decorations, clothing, body parts,
some plants and trees, etc., and its characteristic vowel is i, e.g. hazez 'weak'
(1 class), hazuz 'weak' (2 class), hazaz 'weak' (3 class), haziz 'weak' (4
class).36
Elements of ablaut are used for gender marking in several Pamir
languages (Southeastern Iranian languages spoken in Tadjikistan and
Afganistan: Rastorgueva 1978), e.g. Roshani rast mawn 'red'.FEM apple',
rost kurta 'red:MASC shirt' (Sokolova 1966: 372).37
Warekena (North Arawak: Aikhenvald 1998b) has one of the most
morphologically varied systems of gender agreement marking. Distal
spatial demonstratives obligatorily mark gender agreement with the head
noun through an infix: masc. eta (from ayta), fem. ay-u-ta 'that'. The
proximate demonstrative is semi-suppletive: masc. eni, fem. ayupalu
'this'. Agreement with adjectives is marked with a suffix, masc. 0 fem.
-yawa; predicate-argument agreement is marked with prefixes for A and Sa
and with suffixes for O and So.
(C) Suprasegmental processes (tone patterns, or change of stress)
Suprasegmental processes are very rarely used in noun classification mark-
ing. An example of a Suprasegmental realization of noun classes marked on
the noun itself comes from Rendille (East Cushitic: Heine 1982a: 201). In
35
This unusual fusional pattern of gender marking in possessed nouns in Jarawara can be
explained in terms of a number of regular diachronic changes from an entirely agglutinative,
suffixing structure in Proto-Arawa (Dixon 1995: 281), with possessed nouns marked with the
suffixes -ni 'feminine' and -ne 'masculine'.
36
There are further examples for how this ablaut pattern works in complex adjectives in
Drabbe (1955: 23).
37
Vowel ablaut and infixation mark absolutive gender agreement in Budukh (Lezgian
branch of Daghestanian, Northeast Caucasian: Nichols 1989a: 159).
 Noun Class and Gender Systems 61
this language every noun is overtly marked for masculine or feminine
gender. Masculine nouns have the tone pattern high-low, and feminine
nouns have low-high pattern on the last two syllables; all preceding
syllables are low in both genders, e.g. maxabal 'man, husband', maxabal
'woman, wife' (where ' marks high tone and low tone is unmarked).
(D) 'Repeaters' as noun class agreement markers
In very few languages noun class agreement is marked via partial or total
'repetition' of a noun. ('Repeaters' of this sort are a rare and frequently
neglected means of marking agreement.) Similarly to free forms used as
repeaters for numeral classifiers (see Chapter 4), or in multiple classifier
systems (see Chapter 9), they are never the only agreement type, being
limited to a semantically or morphologically defined subset of nouns.
'Repeaters' are a kind of counterexception to the statement that there is
a limited countable number of noun classes (see §2.2).
This phenomenon has been described for Bainouk, a West Atlantic
language spoken in Senegal and in Guinea Bisau (Sauvageot 1967).38
Bainouk has two structural types of noun—those which contain overt
noun class prefixes39 and those which do not. Agreement in noun class is
obligatory in head-modifier constructions (a modifier can be a demonstra-
tive, an adjective, an interrogative, or a numeral up to nine). If a noun is
marked for class, 'head class' prefixes are also used as agreement markers.
Then they are suffixed or prefixed to demonstratives of different types, as in
2.36, and prefixed to adjectives, interrogatives and numerals, as in 2.37
(Sauvageot 1967: 230-1).
2.36. si-den-o in-si
NCL4-canoe-DEM this-NCL4
'this canoe'
2.37. gu-scl gu-fer
NCL7-boubou NCL7-white
'a white boubou (a kind of clothing)'
Some nouns with no overt noun class marking40 require prefix a- with
adjectives and numerals, and (-)no(-) with demonstratives. Examples given
38
A similar system has been mentioned for a few other West Atlantic languages, e.g. Badyar-
anke and Landuma (Volodja Plungian, Antonina I. Koval', and Viktor A. Vinogradov, p.c.).
39
As in many other West Atlantic languages, noun classes interact with number: there are
fewer classes in the plural than there are in the singular (Sauvageot 1967: 227). The assignment
of noun classes is opaque in most cases, e.g. si-den (NCL4-canoe) 'a canoe'; in some cases,
however, semantics is straightforward, e.g. da-den (NCLl3:AUGMENTATIVE-canoe) 'a big canoe';
ko-den (NCL11:DIM-canoe) 'a small canoe' (Sauvageot 1967: 229).
40
There are about 200 nouns which have no prefixes. They constitute a quarter of the
vocabulary on which Sauvageot's analysis is based (1967: 229). No information is given about
how many prefixless nouns require repeating agreement.
62 Classifiers
by Sauvageot (1967: 232) include nouns with animate referents, e.g. dibon
a-fer 'a white horse', dibon in-no 'this horse'.
Other prefixless nouns mark agreement on modifiers by REPEATING the
first CV sequence of their stem on the agreeing constituent, e.g. 2.38-41.
All the examples given by Sauvageot (1967: 232) are nouns with inanimate
referents.
2.38. katama-nc in-ka
river-DEM this-REPEATER
'this river'
2.39. katama ka-wayi
river REPEATER-big
'a wide river'
2.40. dapcn-c in-da
grass-DEM this-REPEATER
'this grass'
2.41. dapcn da-wuri
grass REPEATER-long
'long grass'
Thus, in Bai'nouk phonologically determined 'repeating' agreement is
restricted to a morphologically defined subclass of nouns.41 According to
Sauvageot (1967: 233), prefixless nouns appear to be loans, or at least
'foreign-sounding' words.
This example of repeaters used for noun class agreement shows that:
(a) repeaters are just one of the mechanisms of marking agreement; it will
be shown in Chapter 9 that no example has been found so far of a language
where this is the only mechanism;42
(b) repeaters have some semantic and/or morphological constraints on
their use (and sometimes discourse constraints as well: see Chapter 9);
they are often used just with inanimates.

41
But see objections to this raised by Doneux (1967: 235).
42
'Repeaters' as an agreement device have some similarities with phonologically condi-
tioned noun class agreement systems, e.g. those found in the languages of the Arapesh family
in the East Sepik province of New Guinea (Arapesh: Fortune 1942; Aronoff 1991; Conrad
1996; Bukiyip: Conrad and Wogiga 1991; Mufian: Conrad 1978; 1996), and in Yimas (Foley
1986). Elements of repetition are also found in the noun class assignment of loan words in
Wolof (described as 'the copy process' by McLaughlin 1997: 16-17), e.g. galaas gi 'the ice'
(from French glace), waliis wi 'the suitcase' (from French valise), soble si 'the onion' (from
Portuguese cebola).
 Noun Class and Gender Systems 63
2.6.3. Double marking of noun classes
Some languages allow double marking of noun class. Examples of these
unusual strategies are found (A) in a few Bantu languages and (B) in some
Australian languages. Their properties are summarized in (C).
(A) Double marking of noun classes in Bantu languages
A number of Bantu languages with multiple noun class systems allow two
noun classes to be marked on a morphological word. Certain nouns that
already contain a noun class prefix which determines their agreement
properties may take a further noun class prefix.
In Kikuyu, if a noun is diminutivized, it can receive two noun class
prefixes: if a noun of noun class 1/2 (singular/plural) is diminutivized, it
acquires the diminutive class 12/13 which is then prefixed to the original 1/2
class marker, e.g. mu-ndu (Prefix l:CL1U/2-person) 'a person'; ka-mu-ndu
(Prefix2:CLl2/13-Prefixl:cLl/2-person) 'a small person' (Stump 1993:
171). A similar strategy is found in Swahili (Shepardson 1982; Helma
Pasch, p.c).
Not all noun class prefixes can occupy the pre-prefixal position. The set
of prefixes which can occur in this position (termed 'secondary' prefixes by
Vail 1974: 24) are typically associated with degree, or value: they include
diminutives and augmentatives, and sometimes pejoratives and honorifics.
Ndali (a Bantu language spoken in Malawi and in Tanzania) has 21
noun class prefixes which correspond to 14 classes: of the 21 prefixes, 14 are
singular vs. plural pairs, and thus correspond to seven classes; the remain-
ing seven classes do not have plural counterparts (they contain abstract
nouns and locatives). The system of noun classes in Ndali is illustrated in
Table 2.9 (adapted from Vail 1974: 25-47). Noun classes are numbered
in accordance with Vail (1974: 25).
As in the majority of Bantu languages, noun class assignment is only
partly based on semantic principles, and is partly opaque; there are also
some morphological principles in operation, e.g. all verbal infinitives
belong to Class 15. It can be seen from this table that secondary noun
prefixes correspond to augmentatives, pejoratives, and diminutives, and
also locatives, e.g. u - k a b w a 'dog (cL1a)', icokabw a (from ic(i)-u-
kafiwva) 'very dirty dog' (sec. CL?) (Vail 1974: 33). If the same prefix is
used as a primary, and as a secondary marker, the semantics of the primary
marker may be opaque; but the semantics of the secondary one is always
associated with augmentatives/pejoratives, or diminutives, as is the case
with Classes 3/4, 7/8, 12/13 above.
Double prefix structures in Ndali, and a few other languages (e.g.
Kikuyu: Stump 1993) are restricted to marking the overt noun classes;
this is to say, they appear on head nouns only.
64 Classifiers
TABLE 2.9. Noun classes in Ndali
No. Prefix SG Prefix PL Status Semantics
1/2 umu aba prim Persons; kinship terms
la/2b u0 awo prim Mostly animates and persons
3/4 umu imi prim, sec Prim.: inanimates, natural phenomena
Sec: augmentative, pejorative
5/6 10 ama prim Natural phenomena, body parts, plant
names, etc.
7/8 id ifi prim, sec Prim: miscellaneous, generally
impersonal
Sec: pejorative
9/10 iN iN43 prim Impersonal; animals; tools,
implements, etc.
11 ulu prim Impersonal objects, body parts, plants,
insects, abstract concepts
12/13 aka utu prim, sec Prim: body parts, manners of action
Sec: diminutives
14 ubu prim Abstract nouns; names of
geographical areas, miscellaneous
15 uku - Verbal infinitives
16 pa sec Motion to/from, situation; proximity
to someone or something near the
speaker
17 ku prim Motion to/from, situation; proximity
to someone or something far from the
speaker
18 mu - prim Situation inside something
21 ili sec Augmentation, pejorative

The preprefixed structures in Kikuyu display the following possibilities

of agreement.
(i) Most frequently, the prefix 2, or the 'outmost' one, overrides the prefix
1 and determines agreement of any type, as in 2.42 from Kikuyu (Stump
1993: 174):
2.42. tu-mi-ruuthi tu-nini (*mi-nini)
CLl3:pL-ci.2:pL-lion CLl3-little (*CL4-little)
'little lions'
(ii) There are a few cases when the agreement properties of a noun with
double prefixing are determined either by prefix 1, or by prefix 2. In all
these cases prefix 2 marks a locative noun class. Variable agreement with a
43
N stands for a homorganic nasal.
 Noun Class and Gender Systems 65
demonstrative modifier 'that' in Ndali is shown in 2.43 and 2.44 (Vail 1974:
42). There is no information as to semantic consequences of the differences
in agreement.44
2.43. p. -i0-lirjga il-yo
LOC.CLl6-CL5-fortress CL5-that
'at that fortress'
2.44. p. -i0-linja apo
LOC.CL 16-CL5-fortress LOC.CL 16+that
'at that fortress'
(iii) A few Bantu languages have 'pre-prefixation' in marking agreement.
In Nyanja, a Bantu language of Malawi, some adjectives regularly take one
agreement prefix, as in 2.45 (Stump 1993: 175).
2.45. ci-manga ca-bwino
NCL7-maize NCL7-good
'good maize'
Other adjectives mark agreement with two sets of prefixes if the head
noun has two prefixes (Stump 1993: 176). One is labelled qualifying
(restricted to head-modifier agreement) and the other concordial (also
used in verb-argument agreement). This is illustrated with 2.46. The division
of adjectives into two agreement types appears to be lexically determined.
2.46. ka-n-khuku ka-ka-kulu
CL 12-CL9-chicken QUAL 12-coNC 12-large
'a large chicken'
All these cases can be considered instances of non-semantic agreement.
(B) Double marking of noun class in Australian languages
A similar but somewhat different example of double marking of noun class
on a head noun comes from Australian languages. Gender/noun class assign-
ment to body parts and other inalienably possessed items in Australian
languages is often problematic (Evans 1994). The noun class can be assigned
according to the 'intrinsic' properties of a body part. If a body part belongs to
neuter gender, it is assigned neuter gender and takes the appropriate gender
prefix, e.g. Maung kun-ngey (CLIV: NEUT RESIDUAL-name) 'her name', but not
*ngal-ngey (CLII:FEM-name) (Evans 1994: 1). In other languages inalienably
possessed items take the 'inherited' gender/noun class, i.e. the noun class of
the possessor (which is then to be considered the head of a possessive NP),
e.g. 2.8 from Yanyuwa. Yet another strategy is found in Nungali. In this

Similar examples from the Zezuru dialect of Shona are given by Stump (1993: 173).
66 Classifiers
language, possessed body parts take two prefixes—the 'inner' prefix (i.e. the
one which comes closer to the root) corresponds to the noun class/gender
of the possessor, while the 'outer' prefix (the one which precedes the inner
prefix) corresponds to the gender of the possessed noun itself. In 2.47
(Evans 1994: 3; Bolt et al. 1971: 70) the possessed noun 'ear' has two
noun class prefixes: the Class 4 prefix marks its inherent class and the
Class 1 prefix marks agreement with the possessor, 'man'.
2.47. ni-ya-manga d-ununin
CL4:NEUT-CLl :MASc-ear CL! :MASc:ABS-man
'the man's ear'
The examples of double marking of noun classes with body parts in
Nungali are limited to masculine vs. feminine 'inherited' gender, and veget-
able and neuter 'intrinsic' gender. This is perfectly understandable from the
semantics of morphosyntactic contexts in which possessed body part terms
occur. Typically, possessors are animate (and thus belong to the masculine
or feminine class), and body part terms belong to either the neutral or
vegetable class.
In Yanyuwa (Evans 1994: 2; Kirton 1971) and Anindilyakwa (Leeding
1996) the multiple-marking strategy is found with kin terms. This multiple
marking of noun class is restricted to overt noun class marking only.
Languages differ as to the strategies of agreement: some agree with the
'intrinsic' gender only (i.e. the gender of the possessed noun, not that of the
possessor), as in Nungali, and some show more complicated agreement
patterns, as do Nunggubuyu, and Gurr-goni. This double marking is
related to the head noun; it does not correspond to agreement.
(C) Double marking of noun classes: a summary
Overt noun class morphemes in Bantu languages which allow multiple
noun class markers fall into two groups which correspond to primary
and secondary prefixes. In a way, they can be considered as two noun class
systems which coexist in a language. They can be marked simultaneously
on a head noun. However, they differ from systems described in §2.7 in that
they are used in the same morphosyntactic environments.
In Nungali semantic restrictions on cooccurrence of noun class prefixes
with possessed body parts which take double noun class marking are linked
to the semantics of possessive constructions. Unlike Bantu languages, only
body parts (which are inalienably possessed) have this double marking;
double marking is associated with a specific type of possessive construction
and a specific type of noun semantics.
It will be shown in §2.7 that, unlike the cases described here, 'split' noun
class systems allow 'double' agreement which is not attested in Bantu, or in
Australian languages.
 Noun Class and Gender Systems 67
2.7. Languages with more than one kind of noun class

It was possible at one time to state that 'no example is known of a language
with two distinct systems of noun classes' (Dixon 1982: 220; cf. Craig
1986a; 1986b; 1986c). The discussion in this section is based mainly on
the data from languages which have been described only recently. This is an
example of a certain 'progress' in typology: the access to new, previously
unknown data broadens the scope of the typological generalizations we can
achieve.
Languages which have more than one type of noun class fall into two
groups. On the one hand, there can be more than one noun class type, with
different semantics, used with different modifiers, and/or for different
agreement types. This type is sometimes referred to as 'nominal' and
'pronominal' noun classes (or genders)—see §2.7.1. On the other hand,
there can be more than one system of noun classes, with different semantics,
which are used, at least partly, in the same environment, e.g. with the same
modifiers—see §2.7.2. Both kinds of systems involve 'split agreement'; i.e.
different agreement rules operate for different morphological classes (cf.
Aikhenvald 1994a). A summary is given in §2.7.3.
The possibility of coexistence of 'noun classes' and 'semantic' gender
was in fact discussed by Dixon (1982: 169), for Mba (Niger-Congo) and for
Wogamusin and Chenapian (Papuan, based on Laycock and Z'graggen
1975: 743–4), and by Corbett (1991: 168-75). Corbett considers cases
like the South Dravidian languages Kolami, Parji, Naiki, which have two
genders (male human and others). In these languages, some lower numerals
have additional forms for female humans. Similarly, Wogamusin and
Chenapian have at least five noun classes, shown by different forms of
numerals; in addition, the number 'one' has distinct masculine and
feminine forms (Laycock and Z'graggen 1975: 744). Given the scarcity of
information, these data can be reinterpreted in such a way as to avoid
admitting the existence of more than one noun class, as Dixon does for
Wogamusin and Chenapian (positing 'ten noun classes, arranged in
masculine/feminine pairs, with neutralisation between pairs beyond the
number "one"').45 Corbett considers similar cases as 'overdifferentiated
targets', with a somewhat 'exceptional' behaviour.
The data from South American, Papuan, African, and Australian
languages discussed below show quite a few regularities as to the ways in
which more than one noun class system can operate in a language.

45
Another possibility would be to say that these languages have a set of numeral classifiers
fused with numbers 1–4 and an independent set of noun classes, or genders. This analysis was
adopted by Lock (forthcoming) for Abau, an isolate spoken in the same area as Wogamusin
and Chenapian.
68 Classifiers
2.7.1. Nominal and pronominal noun class
Different noun class systems can coexist in one language for different types
of modifier; this distinction was first outlined by Heine (1982a) for African
languages. One system of noun classes is used with personal, demonstra-
tive, and other pronouns, and for verbal cross-referencing; this is called
'pronominal' gender/noun class (Heine 1982a: 195). The other is used with
adjectives (and sometimes other modifiers, such as numerals); this is called
'nominal' gender/noun class (Heine 1982a: 195).
(A) Properties of pronominal and nominal noun classes
In each case the two types have all the properties of noun class systems.
They differ, with respect to:
(i) morphosyntactic loci (i.e. grammatical context of occurrence);
(ii) size of system;
(iii) semantics;
(iv) transparency of semantic basis;
(v) variability in assignment;
(vi) overlap with other classifier types in multiple classifier systems (see
Chapter 9);
(vii) possible interrelations with other categories.
These differences are summarized in Table 2.10.

TABLE 2.10. 'Pronominal' and 'nominal' noun class systems

Properties Pronominal noun class Nominal noun class
(i) Morphosyntactic Personal pronouns and Adjectives, more rarely
loci demonstratives as modifiers; numerals, as modifiers
verbal cross-referencing
(ii) Size of system Smaller systems Bigger systems
(iii) Semantics Animacy/sex/humanness Animacy/sex/shape/size
(iv) Transparency of Always clear May be opaque
semantic basis
(v) Variability in None Possible
assignment
(vi) Use in multiple None Possible
classifier systems
(vii) Interrelations with Interrelate with other Interrelate only with
other categories nominal and verbal number (Baniwa,
categories, e.g. number, Tariana: see §10.1)
person, and tense (see
Chapter 10).
 Noun Class and Gender Systems 69
(B) Examples of coexisting noun class systems
In Motuna (Papuan language of Southern Bougainville, Papua New
Guinea: Onishi 1994) a five-term gender opposition (masculine, feminine,
diminutive, local and manner) governs agreement in head-modifier con-
structions with articles, demonstratives and some adjectives. Gender agree-
ment with a topical constituent in verb-argument agreement was illustrated
in 2.11 and 2.12. A system of 51 agreement morphemes, based on shape,
size, and animacy, is used with other modifiers (see (D) in §9.1).
Many languages of South America—including languages of the Arawak,
Tucano, and Harakmbet families, and Yagua—have a small gender system
for personal pronouns, articles, and verbal agreement, and a largeish
system of classes for adjectival and numeral modifiers. The smaller system
of classes involves sex and animacy distinctions. Among Arawak
languages, Tariana has a feminine/non-feminine distinction; Palikur and
Ignaciano have feminine, masculine, and inanimate (see Aikhenvald and
Green 1998 on gender marking and gender assignment principles in
Palikur). The larger system consists of several dozen classes based on shape
and sometimes also animacy and sex; the classifying morphemes can be
used in other classifier environments (see §8.3).
Demonstratives may behave differently in different languages that have
two types of noun class system. In the majority of Arawak languages,
demonstratives usually distinguish two (more rarely three) genders. In
Tariana and Resigaro (both Arawak), Tucano and Bora-Witoto languages,
and in Yagua (Payne 1990: 139) they take noun class agreement markers.
Baniwa (North Arawak: Aikhenvald 1995a; see Table 9.5 below) has two
genders, feminine/non-feminine in demonstratives, personal pronouns, and
verbal cross-referencing markers. A closed set of 44 noun classes is used with
adjectives; these are based on shape, form, structure, sex, and animacy; there
is a feminine class which is marked differently from the way feminine is
marked in demonstratives. Variable noun class assignment only occurs
within a larger system, and is only possible for inanimate nouns. It is used
to focus on a particular shape. 2.48 and 2.49 illustrate the possibility of
variable noun class assignment, depending on the shape of a river in Baniwa.
2.48. hliehe uni maka-peki
DEM:NF Water big-NCL:EXTENDED.LONG.STRETCH
'this long river'
2.49. hliehe uni maka-khay
DEM:NF Water big-NCL:CURVILINEAR
'this long (curved) river'
2.49a is ungrammatical: the system of 'pronominal' noun class does not
allow variability in class assignment.
70 Classifiers
2.49a. *Juaha uni
DEM.F water
Two systems of noun class—one restricted to pronominal modifiers and
verbal cross-referencing, and the other used for the remaining modifiers—
can be reconstructed for Proto-Arawak (Aikhenvald 1996b).
In the Australian language Gaagudju a two-term gender system is
restricted to verb-argument agreement and free pronouns, and a larger
system of four classes is used with adjectives. In Gurr-goni (R. Green
1995) and Warray (Harvey 1987) a system of four classes is found in all
head-modifier and verb-argument constructions, and a smaller system with
only two distinctions is found in independent personal pronouns. Malak-
Malak (Birk 1976: 97-101) has four noun classes, with obligatory agree-
ment in some head-modifier and all verb-argument constructions, and a
feminine/non-feminine distinction in 3rd person singular pronouns and
object bound pronouns.
In African languages agreement in pronominal gender is most likely to
be semantic, while nominal gender agreement is most likely to be syntactic,
as in Iraqw (South Cushitic: Heine 1982a: 195-6).
There are usually no more than two coexisting noun class systems. A
possible case of three systems is found in Barasano (East Tucano: Jones
and Jones 1991: 29, 74). This language distinguishes masculine, feminine,
and inanimate for personal pronouns (used anaphorically); masculine,
feminine, animate, and inanimate in verbal cross-referencing; a large
system of shape and animacy based classes is used with other modifiers,
and in other classifier environments (as in numerous other East Tucano
languages—cf. examples 9.54-7 below).
This situation is totally unlike concordial superclassing (see §2.4.4),
because different systems are used for different modifiers and agree-
ment types, and their semantics may partially overlap. In concordial
superclassing, different choices are made from essentially the same
system.

2.7.2. Different kinds of noun class in the same environment

There are fewer examples of languages with different noun classes in the
same environment than of those with two noun class systems in com-
plementary distribution. The only example appears to be Paumari
(Arawa) (possibly, a similar system is found in Deni, from the same
family: Paula Boley, p.c.). It will be discussed in some detail in (A). Two
noun class systems which partially overlap are found in Mba; this is
discussed in (B).
 Noun Class and Gender Systems 71
(A) Noun classes in Paumari
Paumari (Arawa: Chapman and Derbyshire 1991: 254 ff. and my field data)
has two types of agreement class. One is based on a feminine/masculine
opposition, feminine being the unmarked term. This is called 'gender' by
Chapman and Derbyshire (1991). The other, the so-called ka- vs. non-ka-
noun class, has its semantics partially based on shape and structure. Both
systems are only partly semantically motivated. Both have no overt marker
and are realized only through agreement.
'Gender' agreement in Paumari is marked on demonstratives (e.g. ada
'this:MASc'; ida 'this:FEM'), on suffixes of inalienably possessed nouns which
agree in gender with the possessor (the head of a construction), and on just
three verbal suffixes. Gender agreement in an NP denoting inalienable
possession is illustrated in 2.50. Gender agreement through verbal suffixes
is illustrated in 2.51. Note that the functionally unmarked agreement is
feminine.46
2.50. kahami 'dama
palm.tree(NON-KA,MASC) foot(FEM) + MASC
'foot of a palm tree'
2.51. voroni-'a-ha ada kahami
fall-ASP-TH:MASC this + MASC palm.tree(NON-KA.MASC)
The palm tree fell down.'
The other system of noun classes involves the inclusion or lack of
inclusion of ka- as a prefix to the verb, marking agreement with the S or
O constituent. The ka- class agreement with the S constituent is illustrated
in 2.52. Ka- is also used to mark agreement with the possessor in a
construction of inalienable possession, as illustrated in 2.53. The absence
of ka- agreement with a non-fca-class noun is illustrated in 2.54.
2.52. ka-voroni-'a-hi ida ojoro
NCL-fall-ASP-TH:FEM this + FEM turtle(KA-CLASS.FEM)
'The turtle fell down.'
2.53. kaira ka-bono-ni
guava(KA-CLASS.FEM) NCL-fruit(NON-KA-CLASS,FEM)-FEM
'guava fruit'
2.54. sinari bono-ni
buriti.palm(NON-KA-CLASS,FEM) fruit(NON-K.A-CLASS,FEM)-FEM
'fruit of a buriti palm'

46
First and second person (singular and plural) pronouns have inherent feminine gender:
that is, they trigger feminine agreement; and those of third person plural have inherent
masculine gender (trigger masculine agreement).
72 Classifiers
(A1) Gender assigment in Paumari All nouns with female referents
belong to the feminine gender. All body parts, parts of plants and the
vast majority of artefacts and their parts are feminine. Of 52 names for
cultivated plants (Chapman MS: A.g.l), 17 are feminine. Of 92 names for
wild plants and trees (A.g.4), 53 are feminine. Out of 45 mammals, 27 are
feminine; out of 80 bird names, 41 are feminine; out of 15 reptiles, one is
feminine; out of 34 amphibians (turtles, lizards, crocodiles), 25 are femi-
nine; all the 8 molluscs are masculine; out of 74 insects, 14 are feminine;
and out of 84 fish, 8 are feminine. Nouns denoting terrestrial natural
phenomena, such as 'earth', 'lake', 'beach', are feminine. Nouns denoting
celestial bodies are mostly masculine (e.g. 'star', 'moon', 'sun'); the word
for 'sky' is feminine. Some nouns denoting weather phenomena are mascu-
line (e.g. 'rain', 'fog'), and some are feminine ('rainbow', 'wind'); all names
of seasons are feminine.
Many nouns which denote higher animals can be of either gender
according to the sex of the referent, e.g. ojoro 'turtle', hotairi 'deer'.
Thus, gender assignment in Paumari is only partially semantically
motivated.
(A2) The ka-class assignment in Paumari Nouns with human referents,
abstract nouns (e.g. nominalizations) and nouns which refer to natural
objects and phenomena, never belong to the ka-c\ass. The assignment is
partially semantic, and the principles vary, depending on the semantic field
to which the referent belongs. These principles are given in Table 2.11.
Some body parts distinguish two forms—a ka- class and a non-to- class
one (see (1) in Table 2.11).
The masculine-feminine gender and to-class interact in certain ways.
The vast majority of nouns with non-human inanimate referents which
belong to the to-class are inherently feminine. Few to-class nouns are
inherently masculine; these include kasi'i 'crocodile', vahajari 'alligator',
maoba 'the ritual building'.
Gender and to-class cooccur with non-demonstrative modifiers and on
verbs. (Note that all numerals are stative verbs in Paumari.) Demonstra-
tives distinguish only feminine and masculine forms (they do not take the
to- prefix).
2.55. ada kawina hoara-na
DEM:MASC monkey(NON-KA-CLASS.MASC) One-MASC
'one monkey'
2.56. ida hotairi hoara-ni
DEM:FEM deer(NON-KA-CLASS.FEM) One-FEM
'this one deer'
 Noun Class and Gender Systems 73
TABLE 2.11. Ka-class assignment in Paumari
Semantic group Ka-class Non-ka-class
1. Body parts a. Whole, extended: sa'ay 'hand', Singular parts: sa'ay
'damay 'foot' 'finger', 'damay 'toe'
b. Inner (i.e. more vital) organ: Outer organ: moroboy
moroboy 'inner ear', viridi 'inner 'outer part of the ear',
part of the nose', kajoi viridi 'outer part of the
'intestines'. nose'
2. Plants, fruit, a. Larger size/extension and/or Small, thin, long objects,
artefacts flat in shape, e.g. mesa 'table', e.g. dono 'pestle', hado
sandalia 'sandals', vanami 'knife'
'paddle', and most containers,
e.g. kanawa 'canoe', kojira
'spoon', carro 'car'
b. Substances which consist of Other substances and fruit,
small particles, e.g. ka'ija e.g. paha 'water', simaka
'pepper', jokira 'salt'; or are thick 'manioc', sipatihi 'banana'
in texture, e.g. kojahari 'banana
mash'; fruit with many seeds, e.g.
barasia 'watermelon', jaro'oa
'corn'
3. Animals Big, flat, e.g. ojoro 'turtle', Others, e.g. hotairi 'deer',
ba'dana 'lizard' jomahi 'jaguar'

2.57. ada kasi'i ka-hoara-na

DEM:MASC crocodile(KA-CLASS,MASc) NCL-One-MASC
'one crocodile'
2.58. ida kanawa ka-hoara-ni
DEM:FEM Canoe(KA-CLASS,FEM) NCL-One-FEM
'this one canoe'
Some stative verbs used as modifiers have the ka-class agreement, but no
gender agreement.
2.59. vanami ka-pororo-ki
paddle(KA-CLASS,FEM) NCL-black-DESCRIPTIVE
'a black paddle'
Further differences concern the agreement in gender, and ka- noun class
predicate argument constructions. Agreement in genders and in the ka-
class goes along different lines. The predicate agrees in gender with O in
one type of transitive construction (of structure AVO) illustrated in 2.60,
where sinari bono-ni 'fruit of buriti palm' is feminine, and it triggers
feminine agreement on the aspect and thematic markers -'i-hi (underlined).
74 Classifiers
2.60. ada ojoro bi-kamitha-'i-hi
DEM:MASC turtle(KA-CLASS, MASC) 3SGA-hear-ASPF-THF
ida sinari bono-ni
DEM:FEM buriti.palm(NON-KA-CLASS, FEM) fruit(NON-KA-CLASS, FEM)
-POSSF
'The turtle heard the buriti-fruit fall.'
In a different type of transitive construction, of structure OVA, it is the
A that triggers gender agreement. In 2.61, makhira 'man' is masculine, and
it triggers masculine agreement on the aspect and thematic marker -a-ha
(underlined).
2.61. ida ojoro-ra ka-karaga-'a-ha ada
DEM:FEM turtle(KA-CLASS, FEM)-OBJ NCL-find-ASPM-THM DEM:MASC
makhira
man(NON-KA-CLASS, MASC)
A man found a turtle.'
In intransitive sentences, gender agreement on the verb is with S:
2.62. oniaroa ajihi-'a-ha ada jomahi
then go away-ASP-THM DEM:MASC jaguar(NON-KA-CLASS, MASC)
'Then the jaguar went away.'
Ka- noun class agreement is always obligatory for O and S only. There is
never ka- class agreement with A. This can be seen in 2.52 and in 2.61
(agreement with S and with O is signalled by ka- on the predicate). Note
that in 2.60 there is no agreement with the ka- class noun, ojoro 'turtle', in
A function.
Another basic difference between gender and ka-class agreement in
Paumari is that gender agreement applies only to singular nouns.
Masculine and feminine are neutralized in the plural. In contrast, ka- class
agreement does not depend on number, i.e. there is here no neutralization
of noun class agreement in the plural. Thus, the ka- class marking is
independent of plural, and cooccurs with it. This is shown in 2.63.
2.63. ada-ni kasi'i vi-ko-bami-ki
DEM:MASC-PL Crocodile PL-NCL-two-DESCRIPTIVE
'those two crocodiles (ka-class, masculine)' (-ko here is a phonologic-
ally conditioned allomorph of ka-.)
The two types of noun classification are only partially semantically
motivated and constitute distinct agreement systems. They correlate with
the grammatical categories of number in different ways (see §10.1).
The ka- agreement class is more likely to be lost first in a gradual
language death situation (Aikhenvald MS). Gender agreement is more
 Noun Class and Gender Systems 75
stable in the situation of language death. A tendency towards maintaining
gender agreement, but not noun class agreement may be due (wholly or
partly) to the influence of Portuguese in which all the young speakers of
Paumari are bilingual; Portuguese has masculine and feminine genders, but
nothing corresponding to the ka-class.
(B) Noun classes in Mba
Mba (Ubangi, Niger-Congo: Serzisko 1981: 114 ff.; Heine 1982a: 208-9;
Pasch 1986) has two noun class systems which partially overlap. It distin-
guishes seven 'gender' classes in the singular, and three pronominal
genders. Nominal genders have opaque semantics; nouns indicating human
beings occur in Classes 1, 2, 7; body parts occur in Classes 3 and 4.
Pronominal genders have a clear semantic basis: (i) masculine, (ii) feminine,
(iii) non-human animate and inanimate. They are used only to mark
agreement with animate nouns. Agreement in pronominal gender is semantic,
and agreement with nominal gender is syntactic. The distribution of
genders with different modifiers is shown in Table 2.12, where + indicates
obligatory occurrence of gender in the given environment and (+) indicates
optional occurrence.

TABLE 2.12. Two types of noun class (genders) in Mba

Nominal gender Pronominal gender (with
animate NP head)
Nouns +
Adjectives +
Demonstratives + (+)
Genitive (+) -
Interrogative 'which?'
Interrogative 'how much?'
Numerals (+) +
Personal pronouns - +

The two systems of gender marking can cooccur on the same modifier in
a noun phrase (Serzisko 1982: 115 ff.). In 2.64, the numeral 'one' has a
prefix bi- showing pronominal gender agreement followed by a prefix it-
showing nominal gender agreement.
2.64. ju bi-u-ma
Woman PRON: FEM-NOMINAL.GENDER-One
'one woman'
Pronominal gender agreement is said to 'intensify' a demonstrative.
76 Classifiers
2.7.3. Languages with more than one kind of noun class: a summary
We have considered two kinds of situation. Different systems of noun
classes involving agreement can be independent and employed with
different kinds of modifiers used in distinct morphosyntactic contexts;
two different noun classes cannot then be marked within one morpho-
logical word. This is the case in African, Australian, and a number of South
American languages (§2.7.1). Alternatively, different noun class systems
may partially overlap; then two noun classes can be marked within one
morphological word, as in Paumari (Arawa), or Mba (Ubangi, Niger-
Congo); see §2.7.2.
These two kinds of systems are similar in that, in each case, one type of
noun class is used for pronouns and a different type for other modifiers.
If a language has a closed set of animacy-based noun classes, these are
more likely to appear associated with third person pronouns and demon-
stratives (see §10.2 on the link between person and noun class). A larger set
of animacy and shape-based noun classes are more likely to be used for
common nouns. This tendency, indicated in Table 2.10, is shown in
Diagram 2.1 (Aikhenvald 1994a).

3rd person Demonstratives Common

pronouns nouns

Animacy-based noun classes

<
Shape-based noun classes
>-
DIAGRAM 2.1. Tendencies for animacy-based and shape-based noun classes

If a language distinguishes a system of nominal and of pronominal noun

classes, members of a closed class of modifiers (e.g. demonstratives,
articles, etc.) will agree in pronominal noun classes (with a smallish number
of members). For instance, Waura (Xinguan Arawak: Richards 1988,
Aikhenvald 1996a) distinguishes two genders—masculine and feminine—
only in deictics. Gender marking is fused with demonstratives, e.g. masc.
eze, fern, izi 'this'. There are no gender oppositions elsewhere in the
system, unlike for other Arawak languages. Waura also has a largeish set
of morphemes used as numeral, verbal, and noun classifiers, and possibly,
as noun class markers on modifiers from open classes.
In languages with two systems of noun classes there is also a correlation
between head-modifier and predicate-argument agreement. In a language
with split agreement the predicate will tend to agree with its arguments in
pronominal noun class, while nominal noun class is more likely to be found
 Noun Class and Gender Systems 77
in head-modifier constructions. This is typically the case in North
Amazonian languages with large multiple classifier systems and two
genders (e.g. Baniwa: Aikhenvald 1996c, or Tariana: Aikhenvald 1994a).

2.8. Distribution of noun classes in the languages of the world

NOUN CLASSES (which include GENDER) constitute one of the most frequent
types of classifier system—see Map 1.
A system of two or three genders is present in most Indo-European and
Northwest Caucasian languages. Some Indo-European languages
(Armenian, some Indie, e.g., Bengali, and some Iranian, e.g. Persian)
have lost gender agreement. The vast majority of Afroasiatic languages
have two genders: masculine and feminine. Munda languages, spoken in
Northeast India, have animate and inanimate genders.
More complicated systems of three to five genders are present in North-
east Caucasian languages (Comrie 1981: 208). Burushaski (an isolate
spoken in Northwestern Kashmir) has four genders. The Dravidian
languages of South India have from two to four genders (Krishnamurti
1975). Ket (a Paleosiberian isolate) has three genders.
Noun classes and genders are present in the majority of African
languages. Eastern Nilotic languages distinguish masculine and feminine
gender (Dimmendaal 1983). Khoisan languages distinguish three to four
genders (Welmers 1973: 248; Kohler 1962; 1971). The majority of Niger-
Congo languages have extensive noun class systems (up to 20 agreement
noun classes combined with number). Some Nilo-Saharan languages do not
have gender, but instead employ systems of noun classification through
nominal derivational suffixes and plural markers which do not trigger
agreement. Some African languages lack noun classes, but have categoriza-
tion of nouns which goes back to noun class systems in the proto-languages
(Pozdniakov 1995; Ducos 1979). For instance, Mande languages have an
opposition of alienable and inalienable nouns, Yoruba of human and non-
human, and Igbo of animate and inanimate (Welmers 1973: 211–21; Heine
1982a: 190), but there is no agreement, and thus these are not noun classes.
Noun classes are widespread in Papuan languages of the Sepik basin and
adjoining lowland areas of Irian Jaya. Languages spoken in the Sepik
area—e.g. all languages of the Ndu family and distantly related Kwoma/
Washkuk (Kooyers n.d.), Sepik Hill family, or Abau (Lock forthcoming)—
tend to have two genders, feminine and masculine, which correlate with
shape and size of the referent noun. Ok languages (see e.g. the description
of Yonggom by Christensen 1995: 9-10) spoken in the Western province of
Papua New Guinea and the adjacent areas of Irian Jaya have two genders,
as does Marind (Marind family, Southern Irian Jaya). The languages of the
 noun classes and genders found
in continuous areas
fossilized noun class and gender markers
A isolated instances of noun classes
and genders

MAP 1. Distribution of noun classes and genders in the languages of the world
 Noun Class and Gender Systems 79
Torricelli family and some of those of the Lower Sepik family have fas-
cinating systems of about a dozen classes (see Foley 1986: 85 ff.; 1991 for
information on Yimas; Fortune 1942; Nekitel 1985; 1986; Conrad 1978
on the Arapesh languages). Angan languages spoken at the junction of
the Eastern Highlands, Morobe, and Gulf Provinces, have up to ten noun
classes (see Whitney n.d., for Akoye; Speece n.d., for Angave; Carlson
1991 for Taenae). Extensive systems with several dozen agreement classes
are found in the Papuan languages of Southern Bougainville: Nasioi,
Motuna (Foley 1986: 83 ff.; Onishi 1994) and in those of Central
Bougainville: Napues (or Kunua), Eywo, Keriaka and Rotokas (Kim
Blewett, p.c.), and in Reef-Santa Cruzan languages (Wurm 1992a; 1992b).
Small gender systems (with two or three members) are found in the non-
Austronesian languages of the Solomon Islands (Bilua: Kazuko Obata,
p.c.; Lavukaleve: Angela Terrill, p.c.; Savosavo), in Kuot, the only Papuan
language of New Ireland (Eva Lindstrom, p.c.) and in most languages of
the Fly river (Bine: Fleischmann and Turpeinen 1975; Wipi: Dondorp and
Shim 1997; Wara: Risto Sarsa, p.c.). Some non-Austronesian languages of
East New Britain (e.g. Baining and possibly also Taulil) have a complicated
system of agreement noun classes (Parker and Parker 1977 for Baining;
Lindrud and Nicholson n.d., for Taulil).
The typical noun class system in Australian languages contains four
terms which can be broadly labelled as masculine, feminine, edible
vegetable, and residual. Individual languages range from two noun classes
to six. The majority of languages with noun classes are spoken in a
continuous region of the central north 'prefixing' area with two outliers,
Dyirbal and Banjalang, on the east coast (Sands 1995; Dixon forth-
coming). A few other languages have a gender distinction just in the third
person singular pronoun (Sands 1995: 257).
American Indian languages north of Mexico (Sherzer 1976; Campbell
1997) divide into a number of linguistic areas, some of which show genders/
noun classes and classifiers, and some of which do not. Gender systems are
typically small (with two to three members). The most frequent opposi-
tions are masculine/feminine, animate/inanimate, or human/non-human.
Algonquian languages—spoken in the Eastern Subarctic, the Northeast,
and in the Plains linguistic area—distinguish animate and inanimate
genders. Chemakuan and Salish languages of the Northwest Coast
linguistic area distinguish masculine and feminine genders, and Lower
Chinook distinguishes between human and non-human. The distinction
between masculine and feminine is an areal feature of the Northwest Coast
area (Campbell 1997: 332). In the Plateau area, Upper Chinook has an
opposition between masculine and feminine. Among Californian
languages, only Pomoan distinguishes feminine and masculine (this is
considered an innovation). Animate and inanimate nouns are distinguished
80 Classifiers
in the plural in Pacific Coast Athabaskan, in Yuki (Yukian), in Karok and
Yana, and in Miwok and Costanoan.
Of the Pueblo languages, Hopi (Uto-Aztecan) distinguishes three
genders (animate, inanimate, and vegetable) in the plural. Kiowa-Tanoan
languages (with Kiowa spoken in the Great Plains) have up to four classes
of nouns based on distinctions between number, animacy, and individua-
tion (Watkins 1984, 1995). In the Great Basin area, Southern Paiute
(Numic branch of Uto-Aztecan) has animate and inanimate 3rd person
pronouns. In the Plains, the opposition between animate and inanimate
genders is (as mentioned above) present in all Algonquian languages, in a
number of Siouan languages, and in Comanche (Numic branch of Uto-
Aztecan). Iroquoian languages fall into two subgroups: those of the North-
east have masculine, feminine, and neuter for the 3rd person pronoun,
while Cherokee, in the Southeast, distinguishes only masculine and
feminine. In the Southeast linguistic area, masculine and feminine are
distinguished in nouns and pronouns in Tunica (Gulf), and only in 3rd
person pronouns in Yuchian languages.
In Central America, two genders (animate vs. inanimate) are present in a
few Otomanguean languages and in Tequislatec (Costenla 1991: 117).
Many Central American languages have numeral classifiers and a few
have noun classifiers, rather than genders/noun classes (see Chapters 3
and 4).
More than half of the languages of South America show genders and/or
noun classes. A system of two genders, masculine and feminine, is
characterictic of languages of the Je, Guahibo, and Arawa families, some
Arawak languages, the languages of Gran Choco and related families, e.g.
Guaicuruan and Maka. Chapacuran languages distinguish three genders
(Everett and Kern 1997). The coexistence of two noun class systems for
different types of modifier is characteristic of Arawak and Tucano
languages, Yagua (Peba-Yagua), Zaparoan, Bora-Witoto and Harakmbet,
and a number of isolates, such as Waorani, Saliba, Itonama and Movima
(see Derbyshire and Payne 1990; Dixon and Aikhenvald 1999). One of
these noun class systems has two or three members, while the other is
larger. The Tupi, Pano, Carib, Yanomami, Maku,47 Tacana, Quechua,
Piraha, Aymara, Jivaro and Ticuna families, isolates Aikana, Koaia, and
four of the Arawak languages (Terena, Amuesha, Chamicuro, Bahwana) of
South America have no genders or noun classes; however, they often have
classifiers of other types. No noun classes or classifiers of any other type
have been reported for the isolates Jabuti, Trumai, Yaruro, and Warao (see
Aikhenvald and Dixon 1999).

47
Kakua is the only Maku language which is reported to have a gender distinction in
pronominal cross-referencing (Martins and Martins 1999).
 3 Noun Classifiers

3.1. Properties of noun classifiers

Noun classifiers characterize the noun and cooccur with it in a noun

phrase. They have been recognized in Australian (Dixon 1977; 1982;
Wilkins 1989; Sands 1995: 269-70) and in Mesoamerican languages (Craig
1986b; 1986c). Their properties have been discussed by Craig (1992; forth-
coming; §3.2.3 below). In the Australianist tradition, they are called
'generic classifiers', or 'generics' (Sands 1995: 269-70; Harvey and Reid
1997: 9-10).
The definitional property of noun classifiers is that their presence in a
noun phrase is independent of other constituents inside or outside it. Their
'scope' is a noun phrase. They are a type of non-agreeing noun categoriza-
tion device, their choice being determined by lexical selection, and not by
matching any inflectional properties of nouns with any other constituents
of a noun phrase.
Additional, contingent, properties of noun classifiers are:
(i) The choice of a noun classifier is based on semantics. Every noun in a
language does not necessarily take a noun classifier (§3.2.1).
(ii) Languages may allow the cooccurrence of several noun classifiers
within one noun phrase (§3.2.1).
(iii) One noun can be used with different classifiers, with a change in
meaning (§3.2.2).
(iv) The size of the inventory of noun classifiers can vary, from a fairly
small closed set to a fairly large open set. Consequently, noun classifiers
can be grammaticalized to varying extents (§3.2.3).
(v) Noun classifiers are often used anaphorically; they may grammaticalize
as markers of syntactic functions (§3.2.4).
§3.3 deals with the relationship between noun classifiers and numeral
classifiers. The realization of noun classifiers is considered in §3.4, and their
functional similarities and historical links with overt noun class marking in
§3.5. The distribution of noun classifiers across the languages of the world
is surveyed in §3.6.
82 Classifiers
3.2. Noun classifiers: discussion and exemplification

Noun classifiers are found in a fair number of Australian languages, as well

as in some Mesoamerican languages and in some Western Austronesian
and Amazonian languages. They are also found in a few isolating lan-
guages of East and Southeast Asia1 (see §3.6). The existence of noun
classifiers does not seem to correlate with the degree of synthesis or
head- or dependent-marking morphology in a language.

3.2.1. The choice of noun classifiers and the cooccurrence of several

classifiers within one noun phrase
The choice of a noun classifier is semantic (unlike noun classes described
in §2.3). Example 3.1 illustrates the use of a noun classifier in Jacaltec
(Kanjobalan Mayan: Craig 1992: 284). The noun classifiers, naj 'man' and
no7 'animal', categorize the nouns with which they form one NP as belong-
ing to a class of 'humans' and of 'animals', respectively.
3.1. xil [naj xuwan] [no7 lab'a]
saw CL:MAN John CL:ANIMAL snake
'(man) John saw the (animal) snake'
Noun classifiers correlate with inherent semantic characteristics of
nouns, such as 'animal', 'human', 'plant'; form, shape and structure
of the referent. There is often a generic-specific relationship between
the classifier and the noun (see §11.2.2). Classifiers can also refer to the
social status, or a kinship relation of people. Some Australian lan-
guages have 'social status' noun classifiers with the meanings such as
'initiated man' (see Goddard 1985: 94 for Yankuntjatjara, and a
slightly different system in Mparntwe Arrernte described by Wilkins
1989: 106). Social status classifiers are widespread in East and Southeast
Asian languages (e.g. Adams 1989: 47 ff. for human classification in
Palaungic). DeLancey (1998) discusses noun classifiers referring to social
status in Tibetan.
The choice of a classifier is usually semantically transparent; in some
cases, however, the semantic link between a noun classifier and a noun is
not obvious. In most languages of the Daly area in Australia 'honey' takes
the noun classifier for 'flesh food' (maybe due to the way in which 'honey'
' An interesting example of noun classifiers of a different sort are 'graphical' classifiers, also
known as 'semantic determiners', in the hieroglyphic writing of Ancient Egyptian and cunei-
form writing systems of ancient languages of the Middle East (Rude 1986). These languages
had no classifiers. Semantic determiners were used to disambiguate polysemous signs, or to
specify the meaning of a noun, e.g. Hittite KUR Hani '(land of) Hatti, URU Hatti city of
Haiti'. Apparently, they were not pronounced.
 Noun Classifiers 83
is conceived as a source of energy: Lys Ford, p.c.). The choice of noun
classifier in Jacaltec is often 'obscured' by extension through perceptual
analogy; for instance, 'ice' is assigned to 'rock' class (see Craig 1986c: 275-
6). Similarly to numeral classifiers (Chapter 4), proficiency in noun classi-
fiers can vary from one speaker to another.
The most thoroughly described system of classifiers in an Australian
language is that in Yidiny (Dixon 1977: 480 ff.; 1982: 192 ff.).
Yidiny has a closed set of about twenty classifiers. They fall into two
groups. INHERENT NATURE classifiers divide into humans (waguja 'man',
bunya 'woman', and a superordinate bama 'person'); fauna (jarruy 'bird',
mangum 'frog', munyimunyi 'ant'); flora (jugi 'tree', narra 'vine'); natural
objects (buri 'fire', walba 'stone', jabu 'earth'); and artefacts (gala 'spear',
bundu 'bag', baji 'canoe'). FUNCTION classifiers are minya 'edible flesh food',
mayi 'edible non-flesh food', bulmba 'habitable', bana 'drinkable', wirra
'movable', gugu 'purposeful noise'.
3.2. jarruy durrguu 'mopoke owl'
CL:BIRD Owl

3.3. buri birmar '(hot) charcoal'

CL:FIRE charcoal
Among classifiers referring to humans, bama 'person' can cooccur with
waguja or with bunya, as shown in bama waguuja wurgun 'a teenage boy',
lit. CL:PERSON CL:MAN boy, in 3.4. This is the only instance of a 'hierarchical'
relation among generic classifiers in Yidiny (Dixon 1977: 484).
3.4. nanyji bama waguuja
We+NOM CL:PERSON+ABS CL:MAN+ABS
wurgun muynga gunda:alna
pubescent.boy+ABS cicatrice+ABS cut+PURP
'We must cut tribal marks [on] the teenage boy.' (lit. person man
pubescent boy)
The classification is cross-cutting. This means that different classifiers
have to be used depending on whether the focus is an inherent property,
or on function. Thus, all ants are covered by the inherent nature
classifier munyimunyi; only some ant species are edible, and these are
also covered with function classifier minya 'edible flesh food' (Dixon 1982:
198).
Cooccurrence of classifiers is governed by the following principle. If two
classifiers cooccur, one of them must be an inherent nature classifier, and
the other has to refer to 'function/use' (Dixon 1982: 203), e.g. bulmba walba
malan (CL:HABITABLE CL:STONE flat rock) 'a flat rock for camping' (Dixon
84 Classifiers
1982: 200); jugi mayi badil (CL:TREE CL:EDIBLE.NON.FLESH.FOOD rickety nut)
'rickety nut tree and its fruit' (Dixon 1982: 199).2
Noun classifiers can fall into different subgroups, as in Yidiny (and see
Wilkins 1989: 106, on how social status noun classifiers differ from noun
classifiers of other groups). However, so far no language has been reported
to have several distinct sets of noun classifiers.
The cooccurrence of two noun classifiers is not permitted in all languages;
for instance, it does not happen in Mayan languages, or Minangkabau.

3.2.2. Semantic functions of noun classifiers

In languages with noun classifiers, distinct classifiers can be used with the
same noun to specify its meaning, e.g. Minangkabau batang limau (CL:TREE
lemon) 'lemon-tree', buah limau (CL:FRUIT lemon) 'lemon-fruit' (Marnita
1996). Similarly, in Yidiny, 'a piece of hot charcoal (nirgil), say, can be
described as buri (CL: FIRE) or as wirra (CL:MOVABLE.OBJECT), and a plot of
ground either as jabu (CL:GROUND) or as bulmba (CL:HABITABLE) (Dixon
1977: 203), focusing on different properties of the referent of the noun.
Cooccurrences of nouns with different classifiers may be less obvious; 'a
tree species, diwiy, holds water inside its bark (which can be tapped and
drunk) yielding bana (CL:DRINKABLE) diwiy in addition to jugi (CL:TREE)
diwiy' (Dixon 1982: 203). Walsh (1997: 275) provides further examples
of changing meanings of a polysemous noun with noun classifiers in
Murrinhpatha: nanthi (CL:GENERIC) kamarl (eye) 'eye/face', kura (CL:AQUATIC)
kamarl (eye) 'water-hole', kardu (CL:HUMAN) kamarl (eye) 'sweetheart', mi
(CL:VEGETABLE) kamarl (eye) 'seed'.
These noun classifiers are not semantically redundant (pace Craig 1992:
292). Their behaviour is similar to the derivational functions of noun class
markers (see §2.6; also see §3.5).

3.2.3. Size of inventory and degree of grammaticalization of noun classifiers

Systems of noun classifiers differ in the size of the inventory and their
degree of grammaticalization. We have seen that Yidiny has a closed class
of around twenty classifiers. Mparntwe Arrernte (Australian: Wilkins
1989) has 19; Ngan'gityemerri (Australian: Reid 1997) has about twelve
'generic' noun classifiers, and Murrinhpatha (Australian: Walsh 1997) has
ten. In contrast, Emmi (Australian: Ford 1998) has two, and Patjtjamalh
has three (Lys Ford, p.c.). Acehnese (Western Austronesian) has eleven
noun classifiers. Some languages of East and Southeast Asia have as many
as several hundred noun classifiers.
2
The order of two classifiers appears to be relatively free. There are no examples of
cooccurrence of more than two classifiers (Dixon 1982: 203).
 Noun Classifiers 85
In other languages, almost any generic noun can be used as a classifier.
This appears to be the case in Minangkabau (Austronesian: Marnita 1996)
and in Daw (Maku: Martins 1994), where any noun with generic reference
can be used as a noun classifier. A generic noun classifier daw 'human' in
Daw is illustrated in 3.5 (Martins 1994: 51).
3.5. daw tuum
NOUN.CL:HUMAN eye
'a human eye'
Language-internal criteria have to be established to distinguish a free
form classifier from a noun. In Yidiny, a test for what can be used as a
classifier is provided by the way interrogative-indefinite pronouns are
used. Wanyi 'what, something' is used 'to refer to an object concerning
which nothing is known—it enquires about the genus' (Dixon 1982: 190-
1). There is another interrogative pronoun, wanyirra, which is used 'when
it is known which generic classifier the object comes under, and the actual
species name is being sought'. That is, wanyirra occurs with a generic
classifier, and 'a criterion for whether a word belonged to the set of
generic classifiers was whether it could occur with wanyirra' (Dixon
1982: 191). This is illustrated with 3.6. The classifier minya 'animal' is
underlined.
3.6. person A: wanyi gali-n
what.genus-ABS go-PRES
'What is that going [along there]?'
person B: minya gali-n
CL:ANIMAL-ABS go-PRES
'It's an animal going [along].'
person A: [wanyirra minya ] junga-n
what.species-ABS CL:ANIMAL-ABS run-PRES
'What sort of animal is it running [along there]?'
person B: [minya-ABS ganguul] warri-n
CL:ANIMAL-ABS Wallaby-ABS jump-PRES
'It's a wallaby jumping [along].'
Further syntactic properties of noun classifiers in Yidiny are discussed in
§3.2.4.
In Minangkabau (Western Austronesian: Marnita 1996), generic noun
classifiers can be distinguished from parts of compounds. A generic noun
classifier can be omitted under certain discourse conditions (see Chapter
12), but a part of a compound can not be. Thus, buruang balam 'NCL:BIRD
turtledove' is a construction Classifier—Noun; buruang can be omitted if
the referent has been previously established. Buruang antu 'ghost bird'
86 Classifiers
(lit. bird ghost) is a compound noun since the omission of buruang is
impossible.3
It is often hard to decide whether a language has established noun
classifiers, or whether there is just a discourse device which consists in
occasional pairing of generic and specific nouns. Awa Pit (Barbacoan:
Curnow 1997: 121) has a generic-specific construction which can be used
just for plants and animates (it is never obligatory). For instance, wisha is a
generic term for 'person', and so a white person can be referred to as wisha
awa (white.person person). More study of their discourse and syntactic
properties is needed to decide whether they are noun classifiers, or just a
stylistic device (also see discussion in Sands 1995: 270). It is often difficult
to ascertain whether a language has a given type of classifier or not,
because of the limited productivity of a classifying device. For instance,
Pilaga (Vidal 1997: 61) has generic-specific noun compounding which
seems similar to noun classifiers, e.g. pagela-lapaxat (wasp-insect) 'a
wasp', piyoq-lapa at (flea-insect) 'flea'. These are comparable to deriva-
tional components in class nouns, such as berry in English strawberry,
blackberry, etc.; because of their limited productivity, high degree of
lexicalization, and the fact that they are restricted to a closed class of
noun roots they should not be considered part of a classifier system.4
Similar problems arise with respect to class nouns and noun classifiers in
languages of other types (cf. DeLancey 1986); in each case a decision
should be made based on language-internal criteria. Problems of the
same sort appear when one has to decide whether a language has classifi-
catory verbs or just a set of lexical verbs the choice of which is determined
by the properties of S or O (e.g. drink, or chew: see §6.2.3).
The decisive criterion is how obligatory the generics are, and whether it is
possible to formulate explicit rules of their omission. For instance, in the
Australian language Emmi (Ford 1998) generic classifiers have evolved into
a grammatical device, since there are identifiable discourse conditions
under which either generics, or specific nouns can be omitted. Incipient
structures of this sort can be found in Indo-European languages. For
instance, in English it is possible to use a proper name together with a
descriptive noun phrase, such as that evil man Adolf Hitler, but this type of
apposition is rather marked and used to achieve rhetorical effect.5 Lexico-

3
Walsh (1997) provides criteria distinguishing noun classifiers from compounds in
Murrinhpatha.
4
Denny Moore (p.c.) informs that noun classifiers in Gaviao (Tupi) are also better analysed
as markers of class nouns (see Moore 1984: 203–4), pace Carlson and Payne (1989).
5
In a famous poem 'Lullaby' ('Kolybel'naya'), Nikolai Zabolocki used generic-specific
combinations such as 'fish flounder' (ryba kambala), 'plant potato' (rastenie kartoska) and
'animal dog' (zhivotnoe sobaka) in describing a fantastic and absurd reality. This 'generic-
specific' construction is just a lexico-syntactic mechanism employed for stylistic effect.
 Noun Classifiers 87
syntactic mechanisms of this kind may well be a historical source of noun
classification devices.
Noun classifiers are also different from nouns such as 'berry' in the
English combinations straw-berry, blue-berry, black-berry. These are some-
times called class nouns, and are usually restricted to few lexical fields
(generally a few floral and/or faunal domains) while noun classifiers are
more extensive in that they cover most of the lexicon. Class nouns do not
constitute a syntactic construction as classifier-noun constructions do, and
they do not have the contingent properties (i-iv) outlined in §3.1.

3.2.4. Syntactic functions of noun classifiers

Noun classifiers are typically used with anaphoric function, as in 3.7 (Craig
1992: 284), where the corresponding nouns, 'John' and 'snake', are omitted.
3.7. xil naj no7
Saw CL:MAN CL:ANIMAL
'he (man) John saw it (animal)'6
Classifiers are employed anaphorically in Yidiny. They are often used in
answers to a question, in order to avoid the repetition of the head noun. In
3.8, the question contains a noun duguur 'house'. The answer contains the
corresponding function classifier bulmba 'habitable' (Dixon 1982: 187).
3.8. Question: nyundu duguur-mu gada-any
you+NOM house-ABL come-PAST
'Have you just come from the house?'
Answer. (yiyi) ngayu bulmba-m
(yes) I + NOM CL:HABITABLE-ABL
gada-any
come-PAST
'(Yes), I've just come from the camp'
In subordinate clauses, a specific noun is often stated in the main clause,
and its classifier in the subordinate clause, as in 3.9, or vice versa (Dixon
1982: 188-9).
3.9. ngayu ganguul bugaany nyundu
I+NOM wallaby+ABS eat+PAST you+NOM
minya baga-lnyunda
CL:EDIBLE.FLESH.FOOD+ABS Spear-SUBORD
'I ate the wallaby, which animal you speared.'
6
Mam (Mayan: England 1983: 158-9) has a dozen noun classifiers which are only used
anaphorically when the head noun is omitted. They involve such meanings as baby, non-
human, young man, and young woman.
88 Classifiers
Noun classifiers are used anaphorically in most other Australian lan-
guages (see Wilkins 1989 for Mparntwe Arrernte, and Goddard 1985 for
Yankuntjatjara).
Akatek (Kanjobal Mayan: Zavala 1993: 25-7, and p.c.) has 14 noun
classifiers; they are obligatory as nominal adjuncts, and can be used ana-
phorically. 3.10 illustrates the use of the noun classifier no? 'animal' in a
noun phrase.
3.10. no? tsitam tu?
CL:ANIMAL pig distal
'those pigs'
3.11 illustrates nax 'CL::MAN' used anaphorically. The head noun tsownom
'merchant' is introduced in the first sentence, and is later referred to just by
a noun classifier.
3.11. yeesin si nax tsonwom s?ey tsotan
all right said CLIMAN merchant sat down
nax smaxa syetsmane
CL:MAN waiting waited
nax satk'al k'am tsen tumin
CL:MAN long-time there.was.no CL:ROCK money
'All right, said the merchant, and sat down to wait, he waited, he
waited a long time, but there was no money.'
Noun classifiers can have a number of syntactic functions as the result of
their grammaticalization as anaphoric devices.
In a number of languages classifiers realized as independent words can
be used to mark relative causes, in a function similar to relative pronouns.
There is a link between an anaphoric function of classifiers and their use as
heads of relative clauses: the passage from anaphoric demonstrative pro-
noun to a relative pronoun is typologically well attested (see Zalizniak and
Paducheva 1976).
An extension of anaphoric and pronominal use of noun classifiers in
Jacaltec is their function as markers of coreferentiality (Craig 1986b: 276).
Pronominally used noun classifiers can be deleted under coreferentiality
within certain syntactic boundaries, e.g. 3.12a, b.
3.12a. xil [naj pel] [s-mam naj]
Saw CL:MALE.NON.KIN Peter POSS-father CL:MALE.NON.KIN
'Peter; saw hisj father' (cannot mean 'hisi father')
3.12b. xil [naj pel] [s-mam]
saw CL:MALE.NON.KiN Peter poss-father
'Peter; saw hisi father' (cannot mean 'his; father')
 Noun Classifiers 89
The anaphoric classifier cannot be coreferential with the NP to its left.
Classifier deletion is important in determining complement clause bound-
aries, and it is the major way of encoding a relative clause. 3.13a shows how
a classifier inside the relative clause is deleted under coreference with a
classifier in the main clause (Craig 1986b: 279; example from Craig 1977:
165). 3.13b shows an underlying full clause which corresponds to the
relative clause in 3.13a.
3.13a. mat yohtajoj ix
not knows CL:FEMALE.NON.KIN
[naj xmakni { }
CL:MALE.NON.KIN hit {Noun Classifier DELETION}
yul parce]
in park
'She; does not know the man who hit heri, in the park.'
3.13b. smak naj ix yul
hit CL:MALE.NON.KIN/he CL:FEMALE.NON.KIN/her in
parce
park
'He hit her in the park.'
Morphemes which are used in noun classifier constructions can be used
to mark relative clauses. For instance, in Lao (Chapman 1996) classifiers—
which function as numeral classifiers and as noun classifiers (see Chapter
9)—are used to mark non-restrictive relative clauses.
3.14. [sua too puan l w]Relative clause mn sua khj
shirt CL:BODY dirty finish be shirt ISG
'The shirt that's already dirty is mine.'
The obligatoriness of noun classifiers in a noun phrase differs from
language to language. We have seen above that in Yidiny (Australian),
noun classifiers and full nouns alternate in relative and other subordinate
clauses: if a specific noun is stated in the main clause, its classifier can
appear in the subordinate clause (see 3.9). Their use almost always depends
on whether the referent is contextually established. In Ngan'gityemerri,
nouns can 'stand alone without classifiers'; 'there is a preference, once a
specific noun is contextually established', to refer to the noun with a
classifier (Reid 1997: 167). In Murrinhpatha (Walsh 1997: 264), noun
classifiers are often omitted 'when the specific referent has a norm associa-
tion with that noun class', as in 3.15.
3.15. (ku) kulerrkurrk murntak ngala pangu
(CL:ANIM) brolga old big that
That big old brolga'
90 Classifiers
Noun classifiers in Murrinhpatha are always included with a noun when
it is being cited (Lys Ford, p.c.). Similarly, in Kugu-Ngancara (Australian:
Smith and Johnson 1999), the generic is 'normally given along with the
noun when vocabulary is being elicited or discussed', but 'may not be
present when the noun occurs in a sentence, especially when it has already
been mentioned in the discourse'.
The noun can be considered the semantic head of a classifier-noun
construction. The question of syntactic headship in noun phrases which
consist of a noun classifier and a noun has to be established on a language
specific basis. Since the noun can often be omitted (see examples above;
and also Walsh 1997: 262-3 for further evidence from Murrinhpatha), the
classifier can be used as head. The classifier may alternatively be omitted
(as in 3.15 above); then the noun is the head.

3.3. Noun classifiers and numeral classifiers

Noun classifiers are often considered as a subtype of numeral classifiers in

the languages of Southeast Asia (e.g. Craig forthcoming; Lichtenberk
1983a; see Chapter 4 for numeral classifiers). The basic difference between
the two types is that numeral classifiers occur in numerical—and often in
other quantifying—expressions. Noun classifiers occur independently of
the presence of other modifiers in a noun phrase. Quite a number of
languages have numeral classifiers, and noun classifiers as separate systems.
Noun classifiers will then differ from numeral classifiers in a number of
grammatical and semantic properties. Noun classifiers may not be obliga-
tory in a noun phrase, unlike numeral classifiers which are generally obli-
gatory in a numerical NP (see Chapter 8).
Minangkabau has numeral classifiers and noun classifiers; the two sets
differ in a number of properties such as anaphoric usage and obligatori-
ness. Noun classifiers can be more readily omitted than numeral classifiers
(see §8.2) and are more often used anaphorically. A numeral classifier and a
noun classifier may cooccur in one noun phrase, as in 3.16.
3.16. sa-batang batang pisang
one-NUM.CL:LONG.VERT NOUN.CL:TREE banana
'one banana tree'
In Akatek (Kanjobalan Mayan: Zavala 1992)—a Mayan language with
numeral classifiers and noun classifiers—noun classifiers are independent
lexemes, and numeral classifiers are suffixed to numerical and quantifying
expressions (see §8.2). In this language numeral classifiers can also cooccur
with noun classifiers in one NP (see example 4.31).
Grammaticalization of generic nouns in different classifier constructions
 Noun Classifiers 91
can follow different paths. In Acehnese (Western Austronesian: Durie
1985) eleven nouns may be used as generic classifiers (e.g. aneuk 'child',
boh 'fruit, egg', ie 'liquid', ureueng 'person': Durie 1985: 135). These nouns
can also be used by themselves. There are nine numeral classifiers, some of
which are related to generic nouns, e.g. yue 'NUM.CL:LEAF.OF.BANANA.OR.PALM',
cf. generic noun yue with the same meaning. A few generic nouns gram-
maticalize as numeral classifiers and as noun classifiers, with meaning
difference, e.g. generic noun bak 'stem trunk', 'NOUN.CL:STEM,TRUNK',
NUM.CL:LONG.CYLINDRICAL.THINGS'; generic noun on 'leafy plant', 'NOUN.CL:
LEAFY.PLANT', 'NUM.CL:FLAT.FLEXIBLE.THIN.THINGS' (Durie 1985: 135, 139).
(See §13.1 for discussion of the lexical sources for classifiers.)

3.4. Realization and grammaticalization of noun classifiers

The noun classifiers considered so far in this chapter are all free mor-
phemes. They can be cliticized to the noun, as in Acehnese (Durie 1985:
139), and Akatek (Zavala forthcoming). Noun classifiers can be a subclass
of nouns (as in Yidiny or a number of other Australian languages: Sands
1995), or constitute a class of morphemes on their own. For instance, free-
form classifiers in Mayan languages are often derived from independent
nouns. Of fifteen free-form noun classifiers in Kanjobal Mayan, twelve
come from independent nouns (Zavala forthcoming).7
No system has been found with 'repeaters' employed as noun classifiers.
Noun classifiers can undergo grammaticalization and phonological
reduction and become affixes to nouns. This has happened in some Aus-
tralian languages (Dixon 1982: 207; Sands 1995: 252). In Olgolo, some of
the optional prefixes to nouns are reduced forms of generic classifiers, e.g.
y- is based on uyu- 'fish' and nh- comes from inha- 'animal'. The evolution
of overt noun class prefixes from free classifier forms in Olgolo has a
phonological motivation: the language is eliminating vowel-initial words
(which appeared as the result of consonant-initial dropping) (Dixon 1982:
207-10).
Noun classifiers, once they become affixes to nouns, can further give
rise to noun class agreement (see Sands 1995: 253–4; see §13.4 below).
The grammaticalization chain: noun classifiers —> overt noun class
markers -> agreement markers, and the genesis of agreement are discussed
in §13.8.

7
The development of classifiers from independent nouns involves a few phonological
processes (Craig 1986b: 255 ff.; cf. §13.5.1) such as segmental reduction, e.g. loss of the first
syllable, as in winax 'man' > nax 'CL:MALE' and cliticization, i.e. loss of independent stress.
92 Classifiers
3.5. Overt noun class marking and noun classifiers

Overt noun class markers have a number of functional similarities with

noun classifiers (from which they may have derived in the first place: see
§3.4, and Chapter 13). Overt noun class markers occur on a noun inde-
pendently of other constituents in a noun phrase. They often have a similar
semantics.8 Lama Lama dialects (Laycock 1969; Sands 1995) have overt
noun class markers (probably derived from grammaticalized classifiers),
e.g. nja- 'meat food', mun- 'starch food', aR- 'body parts and nature', ku,
kuR- 'animals', kuR-, ku- 'trees'. Combinations of different noun classifiers
with the same root result in the creation of new lexical items, e.g. ku-won
'kangaroo (living animal)', nja-won 'kangaroo (as game)' (cf. the behaviour
of noun classifiers discussed in §3.2.2). Overt noun class prefixes are more
grammaticalized than noun classifiers, they usually constitute a closed set,
and are often less semantically transparent (see e.g. Evans 1997 on overt
noun class prefixes in Mayali).
Similarly to noun classifiers which are often used anaphorically and have
discourse functions (see §12.1.3), affixed noun class markers have a dis-
course role in some Australian languages, for example, Nunggubuyu
(Heath 1984: 169-70), Warray (Harvey 1987: 53), and Wardaman (Merlan
et al. 1997).
Ngan'gityemerri (Australian: Reid 1997) has agreement noun classes,
overt marking of a noun class on the noun itself and an independent system
of generic noun classifiers. Functionally, one of the most striking properties
of Ngan'gityemerri noun class markers is that, similarly to noun classifiers,
noun class proclitics can nominalize or relativize a clausal constituent or
even a whole clause. 3.17 is a 'simple' example of this (Reid 1997: 203):
3.17. wa=[de-pi kerre]
MALE=body-head big
'the boss' (big-headed man)
The nominal constituent ('big head') is in square brackets; the male noun
class proclitic, wa=, is used to nominalize this constituent. This device is

8
Numerous languages have several sets of nominal derivational affixes which are semanti-
cally similar to genders, or noun classes. In Afroasiatic languages gender markers have
derivational functions. Feminine gender is frequently used to form diminutives, as in the
majority of Berber langauges (Aikhenvald 1984), and Oromo (Cushitic: Clamons 1993). There
is another set of noun classifiers, suffixes to nouns used to mark semantic subgroups of non-
humans, such as *-(a)b 'non-domesticated or harmful animals', *-r 'domesticated animals'
(Diakonoff 1988: 57). Both can cooccur in one word. For instance, Hebrew ?arn(e)-b-(e)t
'hare' contains a Proto-Afroasiatic noun classifier *-(a)b 'non-domesticated or harmful
animals', and -t 'feminine marker'. Tariana (North Arawak) has several sets of derivational
gender-sensitive suffixes (see Aikhenvald in prep.). Their analysis lies outside the scope of this
book.
 Noun Classifiers 93
productively used to create new names out of nominalized clauses. A more
complicated example is in 3.18, with the Ngan'gityemerri name for 'metal
detector' (Reid 1997: 205).
3.18. yerr=[tyagani-merrendi gentyerrmi-gi-baty knife]
TREE/THING=[something-LEST 2pL.S:Aux-ou-hold knife]
'metal detector' (lit. a thing in case you might have something like a
knife)
This usage is very similar to the syntactic functions of noun classifiers as
relativizers and anaphoric pronouns described in §3.2.4, indicating an
'intermediary' status for Ngan'gityemerri noun classification devices, on
a grammaticalization continuum between noun classifiers and noun classes.
The Ngan'gityemerri system can be regarded as, in Reid's words, a 'system
in transition', from noun classifiers to noun classes.
In most languages with classifiers in multiple environments (Chapter 9)
classifier morphemes occur on the noun itself, adding new meaning to it. In
Arawak and Tucanoan languages they have individualizing functions, e.g.
Tariana (North Arawak): episi 'iron (as substance); iron in general'; episi-
da (iron-CL:ROUND) 'axe'; episi-aphi (iron-CL:HOLLOW) 'iron pan'; episi-kha
(iron-CL:CURVED) 'long thin piece of iron'; episi-pukwi (iron-CL:RING.LIKE)
'metal ring' (see Aikhenvald 1994a). Thus, these morphemes are used as
derivational affixes and as noun class agreement markers, and could be
analysed as derivational or as inflectional (see §9.1).
In a number of multiple classifier languages from Amazonia, classifiers
are also employed as relativizers. In Tuyuca (Central Tucano: Barnes 1990:
286; p.c.), a classifier can be suffixed to a nominalized verb, to yield the
predicate of a relative clause. In 3.19 the relative predicate, with a classifier,
is underlined.
3.19. ti-ba-re ado-pe kii
that-CL:PATH-RE here-THEM.CONTR 3msg
ati-a-ri-ba-pi
come-recently-SG.NOM-CL:PATH-LOC
hoa-wa-yigi
start.down.path-go-3msg.PAST.EVIDENTiAL
'He started down that path over here [that he had recently come on].'
These examples illustrate a functional similarity between noun classifiers
and overt class markers on nouns themselves (which can be considered
derivational). Similar examples from Nambiquara (Lowe 1999) are given in
§9.1.
We have seen in §3.2.1 that in some languages, e.g. Yidiny, several noun
classifiers can cooccur within one NP In Lowland Amazonian languages of
the Tucano, North Arawak, and possibly a few other families (such as
94 Classifiers
Guahibo and Witoto), head nouns can take more than one classifier suffix
simultaneously. Several (up to three) classifiers can cooccur on nouns with
inanimate referents. Examples below are from Tariana. (Note that 'aero-
plane' comes within the scope of the 'canoe' classifier -hwya.)
3.20. kara-ka-hwya-puna hanu-puna
REL + fly-TH-CL:CANOE-CL:STRETCH big-CL:STRETCH
'a big airstrip'
3.21. kar a-ka-hwya-puna-way
REL + fly-TH-CL:CANOE-CL:STRETCH-CL:CORNER
'corner of an airstrip'
There seem to be no semantic restrictions on which classifiers may and
which may not cooccur. The order of classifiers depends on the type of
morphological word. If a noun with classifiers is to be presented as a head-
modifier construction, the affix which is the head of the morphological
construction always occupies the last place in the string of derivational
affixes. It also triggers the agreement on an adjective (as in 3.20).
In 3.22, the last derivational affix -maka 'cloth-like' is the head of a
morphological construction, and so it occupies the last place in the string
of morphemes.
3.22. kuda-ma-maka
garment-CL:FEM-CL:CLOTH.LIKE
'woman's garment'
In 3.23 the derivational affix -da 'round object' is used twice. The first
occurrence of -da can be interpreted as a modifier to the root hipa 'ground,
earth', and its second occurrence as a modifier to the derived noun hipa-da
'stone'. An alternative ordering of morphemes is possible, but it changes
the meaning of the word (as illustrated in 3.24 and 3.25) since the semantics
of a derived word in Tariana is connected with bracketing (see Anderson
1992: 264 ff.; and Aikhenvald 1999b, for a fuller account of the possibilities
of variable morpheme ordering in Tariana).
3.23. hipa-da-da
ground-CL:ROUND-CL:ROUND
'gravel, i.e. a small round stone'
3.24. nu-kapima-da hanu-da
1SG-hand + CL:SIDE-CL: ROUND big-CL:ROUND
'the big palm of my hand'
or: nu-kapima hanu-da
1SG-hand + CL:SIDE big-CL:ROUND
'the big side of my hand'
 Noun Classifiers 95
3.25. nu-kapi-da-ma hanu-ma
1 SG-hand-CL:ROUND-CL:SIDE big-CL:SIDE
'the big side of my finger'
In all the Amazonian languages where this kind of 'stacking' of classi-
fiers occurs, if the same—or almost the same—set of classifiers is used as
noun class agreement markers, the noun is assigned to just one agreement
class and this is marked on the adjective, as shown in 3.23 to 3.25.
The multiple occurrences of classifiers are reminiscent of 'double mark-
ing' of head classes attested in Bantu and in some Northern Australian
languages discussed in §2.6.3.
In spite of the functional similarities and possible historical connection,
we prefer not to consider the derivational (overt) noun class markers, as in
Bantu, Australian, or Amazonian languages (see §2.6.1 and §2.6.3) as
instances of noun classifiers, since their scope is a noun, and not a noun
phrase. Synchronically, there are significant differences between noun clas-
sifiers and overt noun class markers on head nouns in the rare languages
which have both. In Ngan'gityemerri (Australian: Reid 1997) noun classi-
fiers are optional and their usage depends on the discourse; the overt noun
class markers are obligatory. This is shown in 3.26 (Reid 1997: 175).
3.26. (gagu) a-matyi bengin-da
CL: ANIMAL NCL:ANIM-kangaroo 3sG.S:AUX-hit
'He shot a kangaroo.'
Note that the overt noun class markers in Ngan'gityemerri are most
often the same as agreement noun class markers on adjectives, as shown in
3.27 (note that = marks a clitic boundary) (Reid 1997: 176).
3.27. a-matyi a=kerre
NCL:ANIM-kangaroo NCL:ANIM-big
'a big kangaroo'
There is no such correlation between noun classifiers and agreement, at
least synchronically. (Note that the absence of agreement is the definitional
property of noun classifiers.) However, if we look at the relationships
between noun classifiers and noun class markers from a historical perspec-
tive the two may be considered as extreme points of a grammaticalization
continuum, from the lexico-syntactic mechanism of noun classifiers to
closed sets of grammaticalized noun classes (cf. Sands 1995: 249). This
shows that the distinct types of noun categorization devices discussed in
this book are not independent of each other. We will return to this in
Chapter 13.
 noun classifiers found in continuous areas

isolated instances of noun classifiers

MAP 2. Distribution of noun classifiers in the languages of the world

 Noun Classifiers 97
3.6. Distribution of noun classifiers in the languages of the
world

Map 2 shows the distribution of noun classifiers in the languages of the

world. Numerous Australian languages have noun classifiers (Sands 1995:
269-70; Dixon forthcoming). In particular, noun classifier systems are
found in the Daly area languages, e.g. Murrinhpatha (Walsh 1997),
Emmi (Ford 1998), and Pattjamalh (Ford 1990). In Ngan'gityemerri
(Reid 1997) and Marrithiyel (Green 1997), noun classifier systems coexist
with noun classes. Noun classifiers are widespread in the languages spoken
along the east coast of the Cape York peninsula of Australia (Dixon 1977:
496). The best described system of noun classifiers in an Australian lan-
guage is that of Yidiny (Dixon 1977; 1982).
Noun classifiers are found in some Mesoamerican languages, e.g. Mayan
languages of the Kanjobalan branch and, possibly, in Chibchan languages
(Craig forthcoming).
Examples of noun classifiers are not known for North American Indian
or Papuan languages. In South America, noun classifiers are found in a few
Maku and possibly in some Je languages, and maybe also in Awa Pit
(Barbacoan: Curnow 1997: 121).
Noun classifiers are found in a number of Western Austronesian lan-
guages (such as Minangkabau: Marnita 1996, or Acehnese: Durie 1985;
also see Conklin 1981), and in some Oceanic languages (e.g. Yapese: Jensen
1977). They are also found in Tai languages (DeLancey 1986), in Tibetan
(DeLancey 1998), and in a number of Austroasiatic languages (Adams
1989).
 4 Numeral Classifiers

4.1. Properties of numeral classifiers

Numeral classifiers are perhaps the most commonly recognized type of

classifier system. They appear contiguous to numerals in numeral
noun phrases and expressions of quantity. Numeral classifiers do not
have to appear on any constituent outside the numeral NP; thus, there
is no agreement in numeral classifier between the noun and another
constituent.
Numeral classifiers have other, contingent properties.
(i) The choice of a numeral classifier is predominantly semantic.
(ii) Numeral classifier systems differ in the extent to which they are gram-
maticalized. Numeral classifiers can be an open lexical class.
In a language with a large set of numeral classifiers, the way they are
used often varies from speaker to speaker, depending on their social status
and competence (Adams 1989). It is much more similar to the use of lexical
items than to the use of a limited set of noun classes. Zavala (1992: 140)
points out the variation in the inventory of independent numeral classifiers
for speakers of Akatek (Kanjobal Mayan). In Minangkabau (Rina
Marnita, p.c.), several specific classifiers (for example, 'thread-like
ornaments', 'arms') are not known to younger people.
(iii) In some numeral classifier languages not every noun can be associated
with a numeral classifier. Some nouns take no classifier at all; other nouns
may have alternative choices of classifier, depending on which property of
the noun is in focus.
The range of semantic oppositions employed in numeral classifiers
varies; it most often involves animacy, shape, size, and structure. There
may be one 'generic' classifier which can be used with any—or almost
any—noun, replacing other more specific classifiers. This happens with
the classifier ge in Mandarin, which can replace specific classifiers for
quite a few speakers (Li and Thompson 1981: 112). Numeral classifiers
have varying functions, and may be used anaphorically. These issues will be
discussed in Chapter 12.
The presence of numeral classifiers in a language is traditionally
associated with a number of typological properties.
Numeral classifiers are frequently independent lexemes, but can be affixes
 Numeral Classifiers 99
to numerals. Classifiers of these two kinds share almost all properties,
except for the fact that in isolating languages numeral classifiers are usually
independent items.
A typical example is 4.1, from Mandarin Chinese (Li and Thompson
1981: 104). The classifier is here an independent form.
4.1. san ge ren
three CL:GENERIC person
'three people'
Numeral classifiers are often found in languages with an isolating typo-
logical profile (see Dixon 1982; Adams and Conklin 1973). They are also
encountered in non-isolating agglutinating languages, such as Japanese or
some Niger-Congo languages; in polysynthetic languages, such as Lowland
Amazonian languages (Derbyshire and Payne 1990); and in a few fusional
languages, e.g. Indie and South Dravidian.
Example 4.2 illustrates a numeral classifier construction in Ejagham
(Benue-Congo: Walters 1981: 310). Ejagham has about nineteen noun
classes and five numeral classifiers. Numeral classifiers are marked for
noun classes, and the numeral agrees in noun class with the classifier. There
is a genitive linker (manifested with a floating tone') between the classifier
and the noun.
4.2. a-m g ' i-ckud
NCLl/6-CL:SMALL.ROUND GN NCLl9/3-orange.seed
a-ba'e
NCLl/6-tWO
'two orange seeds'
In exceptional cases, numeral classifiers can form a constituent with the
noun rather than with the numeral (as in Kana), or appear fused with a
modifier within a numerical construction (as in Nauru). Numeral classifiers
can constitute an almost open set, due to the presence of 'repeaters'. The
structure of numeral classifier constructions and the morphological real-
ization of numeral classifiers are considered in §4.2.
The existence of numeral classifiers in a language presupposes that
numerals are a special word class. There are two kinds of numeral system
in the languages of the world. Some languages have a large class of
numerals, within which it is possible to indicate as high a number as
required. Languages of this kind often have numeral classifiers, e.g. isolat-
ing languages of East and Southeast Asia, Dravidian, and some agglut-
inating Turkic languages. Other languages have very few numerals,
restricted to 'one', 'two', perhaps 'three' and 'many'. This is the case in
numerous languages of New Guinea, in the majority of Australian lan-
guages, and in many South American Indian languages. These languages
100 Classifiers
rely on some form of deictic indication for higher numbers (frequently
employing body parts), or else borrow higher numbers from neighbouring
languages if required. Languages with a small number of numerals tend not
to have numeral classifiers. A few South American languages are an excep-
tion to this statement. All Arawak languages have two to three numerals,
and the vast majority of them also have numeral classifiers (see the discus-
sion of Palikur, Tariana, Baniwa, and Warekena in this and the following
chapters).
There are languages which do not have numerals in a strict sense: that is,
numerals are not used for counting. Australian languages are often said to
have 'one', 'two', perhaps 'three' (which is frequently a compound form)
(Dixon 1980: 108). Hale (1975) suggested, however, that in many languages
these forms are not really numerals; rather, they are indefinite determiners,
comparable to English a and some. In Jarawara and Paumari (Arawa) the
verb meaning 'be alone' came to be used as a numeral 'one' and the verb
meaning 'be a pair' came to be used as a numeral 'two' in the situation of
contact with Portuguese culture (R. M. W. Dixon, p.c.).
If numerals belong to a major word class, this may partly account for the
absence of numeral classifiers. In some Australian languages, numerals are
distributed between the adjective and noun classes; for instance, Gurr-goni
(Rebecca Green, p.c.) has only two numerals; 'one' is an adjective, and
'two' a noun. 'One' as an adjective takes noun class agreement. In Bantu
languages numerals are a subclass of nouns (Gerrit Dimmendaal, p.c.), and
they take noun class markers. These languages have no numeral classifiers.
In languages with elaborate systems of numerals, the use of numeral
classifiers can depend on the numeral. Classifiers are likely to be restricted
to use with smaller numbers. In many languages, they are obligatory with
small numbers, and optional with larger ones. In Minangkabau (Austro-
nesian), numeral classifiers are obligatory with numerals one to three, and
optional with other numerals (Marnita 1996). In Nung (Tai; Saul and
Wilson 1980: 27) classifiers are optional with multiples often. In Burmese
classifiers do not occur with multiples of ten, and in Thai classifiers are not
used with large numbers like 1000, unless individuation is implied. Classi-
fiers are not used with numerals bigger than ten in Telugu (Emeneau 1964:
649). The possibility of classifier omission is linked to the semantic
organization of the system (see §12.1.4).
The absence of obligatory plural marking on nouns and of plural agree-
ment (Greenberg 1972; Sanches and Slobin 1973) is a typological property
usually associated with the existence of numeral classifiers in a language.
However, South Dravidian languages, Nivkh (Paleosiberian isolate: Panfilov
1968), Algonquian languages (Peter Denny, p.c.) as well as some South
American languages (Tucano, North Arawak) and a number of languages
which combine numeral classifiers with noun class systems (e.g. Ejagham:
 Numeral Classifiers 101
see 4.2) are exceptions to this generalization, since number is obligatory in
those languages (see §10.1.2).
Numeral classifiers are often an areal feature. Such is the case for the
languages of East and Southeast Asia (Bisang forthcoming), and for
Mesoamerica. Numeral classifiers are extremely rare in African and in
Australian languages. Kana and other Kegboid languages (Cross River,
Benue-Congo, Ikoro 1994) and Anindilyakwa (Australian, Groote Eylandt)
are exceptions.
Languages can have more than one kind of numeral classifiers. See
§4.3.
A number of questions typically arise with respect to numeral classifiers.
In many languages, distinguishing between numeral classifiers and quanti-
fiying expressions is problematic. Some languages can be said to have
incipient systems of numeral classifiers. These issues will be treated in §4.4.
In the last section of this chapter, §4.5, we consider the distribution of
numeral classifiers in the languages of the world.

4.2. Numeral classifier constructions and morphological

realization of numeral classifiers
Morphologically, numeral classifiers come in one of three forms.
(i) They may be independent lexemes. This happens often, but not always,
in languages with an isolating structure. See §4.2.1.
(ii) They may be affixes, or clitics, attached to, or fused with, numerals. See
§4.2.2.
(iii) They may be attached to, or fused with, the head noun. This extremely
rare situation is discussed in §4.2.3.

4.2.1. Numeral classifiers as independent lexemes

Numeral classifiers as independent lexemes often comprise semi-open,
lexical classes of morphemes; classifiers of this kind are generally found
in isolating languages.
An example of numeral classifiers as independent lexemes from an
isolating language of Southeast Asia Mai (Mon-Khmer: Wajanarat
1979: 295-6) is given below. The classifier follows the numeral which is
postposed to the noun. Example 4.3 illustrates a classifier which refers to
shape, and 4.4 illustrates an animacy-based classifier. Mal has a few
dozen numeral classifiers, as do many other languages of Southeast
Asia (Adams 1989).
1 02 Classifiers
4.3. 4.3 ?n ?i phe? l?
I have pot three CL:ROUND.THINGS
'I have three pots.'

4.4. ?n ?ui khwan thiat phoon l

I have child four CL:PERSON
'I have four children.'
Other languages have fewer classifiers. Numeral classifiers are indepen-
dent lexemes in Uzbek (Turkic: Beckwith 1998; see also Vietze 1979;
Scherbak 1977), an agglutinating language. Examples 4.5 and 4.6 are
from Tashkent Uzbek, which has fourteen classifiers (Beckwith 1998:
131–2). Example 4.5 illustrates a special classifier for humans. Inanimate
objects are classified by their form, as shown in 4.6.
4.5. bir nafar adam
one CL: HUMAN person
'one person'
4.6. bir bas karam
one CL:HEAD.SHAPED cabbage
'one (head of) cabbage'
It is claimed that Hungarian (Finno-Ugric: Beckwith 1992: 201) also has
numeral classifiers (about six) used with inanimate objects. Classifiers
categorize the nouns with respect to their shape and form, e.g. egy szal
gyertya (one CL:LONG.CYLINDRICAL candle) 'one candle'. Not all nouns
require a classifier; for example, there are no classifiers for human
nouns — one just says egy ember 'one person'.
Numeral classifiers are separate lexemes in the fusional languages of the
Magadhan subgroup of Indo-Aryan (Bengali, Assamese, Oriya, Bihari,
Marathi, some Hindi dialects, Nepali: see Barz and Diller 1985; Emeneau
1964) and in South Dravidian languages, e.g. Malto.
Assamese (Barz and Diller 1985: 169) has about ten classifiers used as
independent lexemes. Classifiers z n, z ni, zsna, and g raki are used with
humans (see Table 4.1). They combine reference to sex, animacy, and
politeness.

TABLE 4.1. Numeral classifiers used with humans in Assamese

Human males of Female animals; High-status humans Humans of either
normal rank human females of any sex sex (respectful)
(respectful) (disrespectful)
zon zoni zna goraki
 Numeral Classifiers 103
These classifiers are illustrated in 4.7–4.9.
4.7. tini Zn xokhi
three CL:HUMAN.MALE friend
'three friends' (respectful)
4.8. tini zoni sowali
three CL:FEM.DISRESPECTFUL girl
'three girls' (disrespectful)
4.9. tini goraki mohila
three CL:HUMAN.RESPECTFUL woman
'three women' (respectful)
There is a tendency for isolating languages to have a largeish number of
numeral classifiers. According to some estimates (e.g. Hundius and Kolver
1983, for Thai), Thai and Burmese have around 200 classifiers; Vietnamese
has about 140 classifiers (Craig forthcoming, Adams 1989, Burling 1965).
In contrast, agglutinating languages of the Turkic group and Hungarian,
and fusional languages, such as Indie and Dravidian, have much smaller
sets of classifiers. However, this is only a tendency. Tzeltal (Mayan: Berlin
1968), an agglutinating language, has several hundred classifiers. Nung, an
isolating language from the Tai family, has only four numeral classifiers:
ahn 'inanimate', ohng 'human', tu 'animate', and cah 'general' (Saul and
Wilson 1980: 25 ff.).
Numeral classifiers can be considered an open class in languages which
use repeaters (sometimes called 'self-classifiers', or 'auto-classifiers'1). A
'repeater' appears when 'the specific object itself (or part of it) [is] used as a
numerative' (see Pe 1965: 166; Benton 1968: 115). In a number of East and
Southeast Asian languages, such as Lao and Thai, virtually any noun can
be used in the numeral or quantifier NP; so that 'classifiers straddle the
boundary between closed class and open class words' (Carpenter 1992:
138). Example 4.10 illustrates a repeater construction from Thai (Hundius
and Kolver 1983: 190).
4.10. pratheet saam pratheet
land three CL:LAND
'three countries'
If a classifier construction contains a compound, only the head is
'repeated'; this is known as 'semi-repeater' construction. Example 4.11 is
from Thai (Bisang forthcoming: 14).

Other terms are: 'echo classifiers' and 'identical classifiers' (Burling 1965: 249).
104 Classifiers
4.11. raan-?aahaan saam raan
shop-meal,food three CL:SHOP
'three restaurants'
The use of repeaters makes the system of classifiers almost open-ended,
'to the point of absurdity' (Kolver 1982a: 178). However, no language has
been found so far where repeaters are the only type of classifier (cf. Senft
1996: 7). In every language with repeaters these represent a subclass of a
classifier; they are often used for otherwise 'non-classifiable' items.2 For
instance, in Mal, a Mon-Khmer language (Wajanarat 1979), there are ten
regular unit numeral classifiers, fourteen group classifiers, and two action
classifiers (see 4.3 and 4.4). A subset of nouns can be used as repeaters. All
of these have an inanimate referent. They include otherwise unclassifiable
nouns, e.g. cia 'house', duup 'hut', boh 'mountain', e 'day'. Example 4.12
illustrates a repeater (Wajanarat 1979: 298)
4.12. ?n ?ui cian ba ciarn
I have house one CL:HOUSE
'I have one house.'
Grammaticalized repeaters often give rise to classifiers as a closed or
semi-closed class of affixes to a numeral (Senft 1996: 353). In Kilivila, iga
'name' is a phonologically 'depleted' (shortened) form of the repeater
yegila 'name' in a classifier function (Senft 1996: 171).3 See §13.1.2.
The status of numeral classifiers with respect to word class has to be
established on language internal grounds. For most languages they are
treated as a separate word class (e.g. Bisang 1993; forthcoming; cf. Cohen
1976, for Jeh (Mon-Khmer), and Saul and Wilson 1980, for Nung (Tai)).
In some languages numeral classifiers are treated as a subclass of adjec-
tives, e.g. Malto (South Dravidian: Mahapatra 1979: 121).
Parameters of variation in numeral classifier constructions include
constituent order and constituency relations in classifier constructions.
Constituent order in classifier constructions usually depends on the general
syntactic rules of the language (see Greenberg 1963; 1972; Dryer 1992).
Greenberg (1972) established four possible constituent orders in numeral
classifier constructions:
(i) [NUM-CL]-N: e.g. Chinese, Vietnamese, Hmong, Miao of Wei Ning,
Uzbek, Hungarian;
(ii) N-[NUM-CL]: e.g. Thai, Khmer, Mal (above);
(iii) [CL-NUM]-N: e.g. Ibibio (Niger-Congo: Greenberg 1972);
2
Cf. Jones (1970: 2): 'it is interesting to speculate on the possibility that such usage arises
from an inadequate supply of classifiers once their use becomes firmly established.'
3
In Kilivila, of 88 classifiers which are most frequently used 25 are repeaters.
 Numeral Classifiers 105
(iv) N-[CL-NUM]: possibly, Bodo (Sino-Tibetan: Greenberg 1972).
These orders exhaust all the possibilities—numerals and classifiers are
always adjacent. As Greenberg pointed out (1990: 228), orders (i) and (ii)
are much more frequent than orders (iii) and (iv). (Note that he did not
distinguish classifiers as independent lexemes from classifiers as affixes.
Suffixes are much more frequent than prefixes in the world's languages;
so the preference for suffixed rather than prefixed classifiers could be linked
to this.) Languages also permit variation in order between types (i) and (ii)
(see Greenberg 1990: 236; 1972). In most Thai languages classifiers usually
follow numerals except for the number 'one' which precedes (Bisang forth-
coming: 5; Greenberg 1972).4
The classifier usually forms a constituent with the numeral. The noun,
and not the classifier, is generally the head (see Greenberg 1972 for discus-
sion and examples). However, in isolating languages it is often difficult to
work out syntactic criteria for heads, especially since either a classifier or a
noun can be omitted under specifiable discourse conditions (see §12.1.3).
According to Greenberg, no language has the order Classifier-Noun-
Numeral or Numeral-Noun-Classifier (where the numeral and the classifier
are separated by the noun). The former order is in fact found in Ejagham
(Walters 1981: see example 4.2 above). In Ejagham the classifier can be
shown to form a constituent with the noun rather than with the numeral
(at least prosodically). The classifier, and not the noun, is the syntactic
head of the NP since it triggers noun class agreement on the number
word.

4.2.2. Numeral classifiers attached to numerals

Numeral classifiers can be attached to numerals as suffixes, or, more rarely,
as prefixes. Numeral classifier systems of this kind are found in South and
North American Indian languages and in inflecting Indie languages. There
can be from just two to several hundred classifiers.
Bengali (Masa Onishi, p.c.) has five numeral classifiers which are suffixed
to the numeral: -ta 'countable non-human'; -ti 'diminutive of -ta; -jan
'human'; -khana 'solid objects with rectangular or flat shape'; -khdni
'diminutive of -khana', for example,
4.13. ek-ta bai
one-CL:NON.HUMAN book
'one book'

4
In Bodo (Sino-Tibetan) there are two subsystems: the 'indigenous' one has the order CL-
NUM, while the one borrowed from Assamese has NUM-CL order (Greenberg 1972).
106 Classifiers
4.14. ek-ti bai
One-CL:NON.HUMAN.DIM book
'one beautiful small book'
Other nouns do not take any classifier.
Numeral classifiers in Marathi (Emeneau 1964: 648) distinguish mascu-
line and feminine forms: masc. jan, fem, jani 'human'.
Yucuna (North Arawak: Schauer and Schauer 1978) has eight numeral
classifiers referring to form (round, cylindrical, plain, side; symmetrical;
concave) and animacy (human and non-human animate). Yucuna, like all
Arawak languages, has just a small system of numerals (one, two, and
three); see 4.15.
4.15. pajluhua-na yahui
one-CL:ANIM dog
'one dog'
Japanese has several hundred classifiers attached to numerals, however,
speakers of the language typically use only about thirty-eight classifiers
(Downing 1986: 346; Masa Onishi, p.c.; also see Denny 1979b: 317). There
are two syntactic constructions in which numeral classifiers are used
(Denny 1979b: 318). One is a numeral classifier noun phrase; it involves
a linker, or 'genitive' particle -no (called 'basic' by Martin 1975: 777, and
'individualizing' by Downing 1984: 194), e.g. ni-dai-no kuruma (two-
NUM.CL:VEHICLE-LINKER car) 'two cars'. The other construction, called
'adverbial' by Denny (1979b), involves a copula clause, e.g. enpitsu ga
san-bon aru (pencil SUBJECT three-long there are) 'there are three pencils'.
Similarly, in Korean attributive or genitive -uy is typically used to accom-
pany a numeral+classifier construction, e.g. sey calwu-uy yenphil (three
NUM.CL:LONG.SLENDER-ATT pencil) 'three pencils'. The form -uy can be
omitted in a more colloquial speech register (Lee 1997; also see Downing
1984: 199, on genitive particle deletion in Japanese). There are several
possible orders in numeral classifier constructions; however, the classifier
must remain within the same constituent as the numeral. Japanese has two
possibilities of constituent order (Bisang forthcoming: 5): the classifier-
numeral constituent can occur in prenominal position (with the linker no)
as in 4.16, or following the noun (without -no) as in 4.17.
4.16. ni-dai-no kuruma o kai-mashi-ta
two-NUM.CL:VEHICLE-LINKER Car ACC buy-HON-PAST
'(S/he) bought two cars.'
4.17. kuruma o ni-dai kai-mashi-ta
Car ACC two-NUM.CL:VEHICLE buy-HON-PAST
'(S/he) bought two cars.'
 Numeral Classifiers 107
In Korean there are three possibilities. 4.18 shows the unmarked order in
a numeral classifier construction (Lee 1997):
4.18. sey calwu-uy yenphil-lul
three NUM.CL:LONG.SLENDER-ATT pencil-Ace
sa-ss-ta
buy-PAST-DEC
'(1) bought three pencils.'
The classifier can immediately follow the head noun and takes case
marking (Sohn 1994: 272), as in 4.19.
4.19. Minca-nun ecey chayk sey kwen-ul
Minca-TOP yesterday book three CL:BOOK-ACC
sa-ss-ta
buy-PAST-DEC
'Minca bought three books yesterday.'
4.20 shows a different constituent order, used to focus on the head noun
(note the absence of the attributive -uy) (Lee 1997). The classifier follows
the head noun and both are case-marked. This could be considered a
discontinuous noun phrase.
4.20. yenphil-lul ecey sey calwu-lul
pencil-Ace yesterday three NUM.CL:LONG.SLENDER-ACC
sa-ss-ta
buy-PAST-DEC
'I bought three pencils (not other items) yesterday.'
While variations in the ordering of the head noun and the numeral +
classifier are acceptable (especially in the colloquial speech register:
Yunseok Lee, p.c.), the ordering of the numeral and the classifier is rigid,
and no constituent can be inserted between them.
In Yagua (Derbyshire and Payne 1990: 253–4; Payne and Payne 1990:
445 ff.) classifiers are infixed to numerals (usually small numbers), as illu-
strated in 4.21. (The same morphemes are suffixed to demonstratives,
adjectives, and verbs; see §9.1.)
4.21. ta-dasiy-quii
One-CL:THIN.POLE-One
'one (shotgun, blowgun, palm trunk, etc.)'
Classifiers sometimes have a different morphological status depending
on whether they occur with a number, or with a quantifier. Bahwana, a
nearly extinct North Arawak language from Northern Brazil (Ramirez
1992: 55) has twenty-six classifiers obligatorily suffixed to quantifiers
(4.22), and infixed to numbers one and two (4.23).
108 Classifiers
4.22. yaoa-da karaka
how.much-CL:ROUND,OR.HOLLOW hen
'How many hens (are there)?'
4.23. a-da-rini karaka
one-CL:ROUND.OR.HOLLow-one hen
'one hen'
Numeral classifiers can be fused with a number. This often happens in
languages with fusional characteristics. Telugu, a fusional language, has two
classifiers, human and non-human, fused with the numeral (Krishnamurti
and Gwynn 1985: 106-7). Table 4.2 features a sample of numeral classifiers
in Telugu.

TABLE 4.2. Numeral classifiers in Telugu

Numbers Non-human Human
'two' rendu iddaru
'three' muudu mugguru
'four' naalugu naluguru
'five' aydu ayduguru

Languages with only little fusion can have numeral classifiers fused with
numerals. Kusaiean (Micronesian: Lee 1975) has two sets of cardinal
numerals. Set A is used in counting fishes, insects, four-legged animals,
plants, means of transportation, and long, pointed objects. Set B is used for
everything else; these numerals are morphologically unanalysable and can
be considered suppletive—see Table 4.3. Some nouns are used with numerals
of both sets with a difference in meaning, e.g. paip yoko 'four cigarettes', paip
ahkorr 'four packs of cigarettes'.

TABLE 4.3. Classifiers fused with numerals in Kusaiean

Set A Set B
'one' soko sie
'two' lukoac luo
'three' tolko(e) tolu
'four' yoko ahkorr

Nivkh (Paleosiberian isolate: Panfilov 1968: 414-15), a polysynthetic

language, has twenty-six numeral classifiers fused with a numeral; a sample
of these is given in Table 4.4. Most of them are used with numerals from
 Numeral Classifiers 109
5
one to five, and some with numerals up to ten. At least some numeral
classifiers in Nivkh originate from nouns, e.g. ar 'bunch of firewood', n'-ar
'one (bundle of firewood)'; cu 'family', riiz-cu 'one (family)' (cf. Chapter
13).

TABLE 4.4. Numeral classifiers in Nivkh

Thin flat Small People Animals Objects of Sledges Bundles Fishes Families Fishnets
objects round different of strung on
objects forms firewood twigs

'one' n'rah n'ik n'en n'tn n'aqp n'irs n'ar n'gak n'izcu nvor
'two* merah mik men mor meqp mirs mer mengaq mizcu mevor

Warekena (North Arawak, Brazil, and Venezuela: Aikhenvald 1998b)

has six numeral classifiers; some are realized with prefixes, and some are
portmanteau with a numeral (see Table 4.5). This morphological mechan-
ism is peculiar in a predominantly suffixing agglutinating language like
Warekena. Note that there are only two native numerals, 'one' and 'two';6
the others are loans from Portuguese.

TABLE 4.5. Numeral classifiers in Warekena

Semantics 'One' 'Two'
Human male peya e-naba
Human female peya tuwa-naba
Animals pamina pamina-naba
Fish pe-ceyaru ere-naba
Curvilinear objects pa-puriaruni e-naba
Periods of time ba-buya, pa-puya bu-naba

In Squamish (Salish: Kuipers 1967: 149-51) one of the two sets of

numeral classifiers (see §4.3.2) is realized by reduplication. There are three
numeral classifiers: for objects, animals, and humans. The form used for
classifying inanimate objects is formally unmarked. The numeral classifier
used with animals is marked with reduplication of the first consonant, and
the one used with humans is marked via reduplication of the first two
consonants (Table 4.6).

5
Young speakers tend not to use classifiers at all (Vladimir P. Nedjalkov, p.c.); it is
unknown which form of numerals is used.
6
Classifiers are used with numerals 'one' and 'two' in only one dialect of Warekena. Other
dialects use the human masculine form peya with all nouns.
110 Classifiers
TABLE 4.6. Numeral classifiers in Squamish

Objects Animals Humans

'one' nc'u? ni'-nc'u? nc'-nc'u'?

'two' ?a'n?us ?a'n.n.?us ?n-?a'n?us
'three' ca'nat ca'-cn?at cn-ca'nat

Numeral classifiers can attach to a modifier other than a numeral. In

Nauru (Micronesian) numeral classifiers are fused with demonstratives in a
numerical noun phrase (Kayser 1993: 41 ff.), e.g. n-aiquon-oe 'one this one
here' (inanimate), n-airan-e 'one this one here' (a flat object).
Repeater affixes to numerals can be used as numeral classifiers, as in
Truquese (Melanesian: Benton 1968: 115-17).
4.24. linu-rjaf naaf
three-CL:FATHOM fathom
'three fathoms'
An example of partial repeaters used as numeral classifiers comes from
Movima (an isolate from Bolivia). This language has three genders (mascu-
line, feminine, and neuter or inanimate) in personal pronouns with cross-
referencing, and numeral classifiers. Preliminary fieldwork by Craig (1996)
showed that there are a few classifiers the choice of which is based on the
semantics of a noun: -poy 'quadruped animals', e.g. fox, crocodile, tapir;
-mo 'biped animals', e.g. rooster, duck, owl; and -ba 'fruit', e.g. papaya,
guava, orange (cf. also Key 1979: 67-8). For other native Movima nouns,
the last syllable is repeated on the numeral as an agreement device, e.g. -d'o
for chad'o 'plate'; -mas for d'imas 'hay'; -pi, for sukapi 'belt'. For borrowed
nouns, the last two syllables are repeated if a noun consists of more than two
syllables, e.g. -misa for kamisa 'shirt', and -pato for zapato 'shoe' (both loans
are from Spanish). If a loan consists of just two syllables, the reduplicated
last syllable is used, e.g. -sasa for mesa 'table', -yaya for siya 'seat, chair'.
The use of repeaters as agreement devices in Movima depends on the
semantics of nouns and on their origin: agreement with loanwords follows
different principles from that with native words.
Affixed repeaters are also used with numerals and in other contexts in
multiple classifier languages such as Tucano (East Tucano), Tariana (North
Arawak), and Kilivila (Austronesian); see Chapter 9.

4.2.3. Numeral classifiers attached to the head noun

The only hitherto known instances of numeral classifiers which form a
constituent with the head noun rather than with a numeral are Kegboid
languages such as Kana (Cross River, Benue-Congo: Ikoro 1994).
 Numeral Classifiers 111
Classifiers in Kana form a single morphological and phonological word
with the head noun, and not with the numeral. The order in the NP is
Numeral-Classifier-Head. The following criteria confirm that the classifier
forms one phonological and morphological word with the noun, and not
with the numeral.

(A) Rule of tone sandhi in Kana (Ikoro 1994: 19-21)

The rule of tone sandhi in a numeral NP shows that the classifier forms a
single phonological word with the head noun. A tone-lowering sandhi rule
operates between components of one phonological word, so that the first
component of a phonological word never has raised tone. Classifiers never
have raised tone, while numerals do.

(B) Formation of diminutives and adjectives in Kana (Ikoro 1994: 21-3)

The diminutive proclitic i is attached to the classifier, and not to the noun,
in a numeral-classifier-noun construction, e.g.

4.25. i nuu
DIM rat
'small rat'
4.26. zii i ka nuu
one DIM CL:GENERIC rat
'one small rat'

A numeral classifier is placed before a noun-based adjective, e.g.

4.27. zii ka kpaa-bee nee

one CL:GENERIC bald-head person
'one bald person'

Ikoro (1994: 23-5) suggests an areal origin for numeral classifiers in

Kana. Proto-Benue-Congo had a noun class system, marked on nouns;
Kegboid languages lost this, and acquired numeral classifiers. Noun cat-
egorization is still associated with the head noun, as a vestige of the noun
class system of the proto-language. This may explain the unusual system of
numeral classifiers in this and other Kegboid languages.7

7
Alternatively, it is possible that the numeral-classifier construction in Kana historically
goes back to a genitive constituent, similar to the one in Ejagham (see example 4.2) (Gerrit
Dimmendaal, p.c.).
112 Classifiers
4.3. Languages with more than one morphological type of
numeral classifier

Languages with more than one morphological type of numeral classifier

are rare. Similarly to languages with more than one type of noun class
system (§2.7), there are two possibilities: different types of numeral classi-
fier may be in a complementary distribution (see §4.3.1), or different types
of numeral classifier may cooccur and display different properties (§4.3.2).

4.3.1. Different types of numeral classifier in complementary distribution

Different systems of numeral classifiers can be in complementary distribu-
tion depending on the numeral, or on the semantics of the classifier.
In Malto (South Dravidian: Mahapatra 1979: 120) numeral classifiers
are independent lexemes when used with the number 'three' or more. The
order in a numeral phrase is Numeral-Classifier-Noun:
4.28. tini maq o:ydu
three CL:INAN cow
'three cows'
With the numbers 'one' and 'two', non-human classifiers attach to
numerals as prefixes, and the structure of the numeral phrase is: Classi-
fier + Numeral-Noun (4.29). For human nouns, portmanteau forms are
used: ort 'one (human)', jo:rond or irw 'two (humans)'.
4.29. maq-ond o:ydu
CL:INAN-One COW
'one cow'
Some dialects of Telugu described by Emeneau (1964: 649-50) have a
numeral classifier for humans which appears in different forms with differ-
ent numbers. Suffix -aru appears with the number 'two', e.g. idd-aru man-
usulu 'two people'. With numbers from three to seven, -guru is used, and it
can be suffixed to the numeral, as in mug-guru manusulu 'three people', or
be an independent word, as in nalu guru manusulu' four people'. With
numerals from eight to ten, there is an independent lexeme mandi
'CL:PERSON', as in enimidi mandi manusulu 'eight men'. (Classifiers are not
used with numerals bigger than ten.) In Nias (Western Austronesian: Lea
Brown, p.c.), numeral classifiers for human nouns are suffixes to the
number 'one' but prefixes to other numbers, e.g. sa-mosa niha (OHC-NUM.CL:
HUMAN person) 'one person', da-rua niha (NUM.CL:HUMAN-tWO person) 'two
people'. In many languages classifiers for humans (and/or animates) have
different forms with different numbers; see also Aikhenvald and Green
(1998) for examples from Palikur.
 Numeral Classifiers 113
A few languages use quite different classifier sets for different numbers.
To'aba'ita (Austronesian) uses an optional classifier fa 'small objects' with
numbers below ten. There are different words for 'ten' whose selection is
determined by the semantics of the counted noun, taafulu, taafuli 'ten (gen-
eral)', qada 'ten (coconuts)', akwala 'ten (people, porpoise teeth)', finta 'ten
(tubers, fruit, etc.)', lama 'ten (birds)', kobi 'ten (strings of shell money)', ai 'ten
(bamboo containers full of canarium nuts)' (Lichtenberk 1995),8 e.g. 4.30.
4.30. te'e lama'e sobe ma lima fa sobe
one ten:BIRDS bird.species and five CL bird.species
'15 (one ten and five) sobe birds'
A few languages distinguish several sets of numeral classifiers according
to their origin. Korean has three sets of classifiers (native, Sino-Korean,
and loans); native classifiers are used to enumerate 'natural objects' or
items which reflect traditional culture, while Sino-Korean and loan classi-
fiers are used with nouns referring to 'products of modern civilization'
(Sohn 1994: 272); see Downing (1996: 46) for a discussion of classifiers
of Sino-Japanese origin in Japanese.

4.3.2. Different types of numeral classifier which occur together

In Akatek (Kanjobal of San Miguel Acatan; Mayan: Zavala 1992) there are
three classifiers which are affixed to the numeral; more than ten classifiers
are used as independent lexemes (see Table 8.3). Affixed classifiers are: -wan
'human'; -k'on 'animate non-human'; -eb' 'inanimate' (Zavala 1992: 130-
6). These classifiers are obligatory in a numeral, or quantifier phrase, unless
there is another modifier present. Affixed classifiers may cooccur with
noun classifiers in a noun phrase (noun classifiers in Akatek are discussed
in §3.2.4; see also §8.2), for example:
4.31. no? ?os-k'on ts'i
NOUN.CL:ANIMAL three-NUM.CL:ANIMATE.NON.HUMAN dog
'three dogs'
Independent classifiers differ from affixed classifiers in the following
ways (Zavala 1992: 139-50). First, the inventory of independent classifiers
differs from one speaker to another, while the inventory of affixed classi-
fiers does not. Independent classifiers refer to the shape and form of an
object; affixed classifiers characterize the referent of the noun in terms of
8
This is reminiscent of Fijian, which does not have numeral classifiers (Dixon 1988). Old
Fijian had special forms, for 'ten' and 'hundred', used for counting different objects; these can
be considered as fused numeral classifiers. Churchward (1941: 66-7) lists twenty-one forms for
'ten' and two for 'hundred', e.g. bi 'ten (turtles)', mata 'ten (fish)', sava 'ten (pots)', uduudu 'ten
(canoes)'; bola 'a hundred (canoes)', koro 'a hundred (coconuts)'.
114 Classifiers
its animacy. There is more freedom in the choice of an independent classi-
fier. One noun can combine with more than one independent classifier,
depending on a particular, shape-related, property of the referent which is
in focus. This is not so for affixed classifiers.
A number phrase may contain both an independent and an affixed
numeral classifier plus a noun classifier, as in 4.32 (Zavala 1992: 144).
4.32. ?os-eb' soyan ?isim
three-NUM.CL:INAN NUM.CL:ROUND NOUN.CL:MAIZE
paat
tortilla
'three tortillas'
Independent classifiers may be omitted (Zavala 1992: 134), as in 4.33.
4.33. te? ?os-eb' te?
NOUN.CL:WOOD three-NUM.CL:INAN WOOd
'three trees'
An affixed numeral classifier also can be omitted; then the independent
numeral classifier forms one phonological word with the numeral, as 4.34.
According to Zavala (1992: 144), 4.32 and 4.34 are synonymous.
4.34. ?os-soyan ?isim paat
three-NUM.CL:ROUND NOUN.CL:MAIZE tortilla
'three tortillas'
The two classifier sets in Akatek also differ in what numbers they are used
with. Suffixal numeral classifiers occur with all numbers except for 'one'
(Roberto Zavala, p.c.), while the other set is used with all the numbers.
Squamish (Salish: Kuipers 1967: 149-52) also has two sets of numeral
classifiers which may cooccur in the same environment. Numerals and
numerical interrogative 'how much' distinguish three forms: objects, animals,
and humans (see Table 4.6). In addition, numerals cooccur with one of the
seven so-called 'lexical suffixes' the choice of which depends on the seman-
tics of the noun, e.g. /-qs/ 'small oblong object' as in /xa?u'cn-qs ca'mx/
(four:OBJECT-CL:SMALL.OBLONG wood) 'four pieces of wood'. (No further
data is provided concerning the differences between the two sets.)

4.4. Problems with numeral classifiers

4.4.1. Mensural and sortal classifiers: distinguishing classifiers from

quantifying expressions
Two basic types of numeral classifier have been distinguished: sortal
classifiers and mensural classifiers. A sortal classifier is 'the one which
 Numeral Classifiers 115
individuates whatever it refers to in terms of the kind of entity that it is'
(Lyons 1977: 463). Examples of sortal numeral classifiers were given
above. A mensural classifier is 'the one which individuates in terms of
quantity' (Lyons 1977: 463). While sortal classifiers categorize nouns in
terms of their inherent properties such as animacy, shape, consistency (see
§11.2.3), mensural classifiers are used for measuring units of countable and
mass nouns. The choice of a mensural classifier is conditioned by two
factors: the quantity, or measure, of an entity, and its physical properties
(permanent or, more often, temporary ones). The mensural classifier han in
Korean is used exclusively for measuring rice wine in terms of an institu-
tionalized measuring cup (Lee 1997).
4.35. makkeli han mal
rice.wine one MENS.NUM.CL:rice.wine
'one measure of makkeli (rice wine)'
Mensural classifiers differ from sortal classifiers in their semantics (see
(F) in §11.2.3). Since the choice of a mensural classifier is often determined
by the temporary state of an object (its quantity, or the arrangement it
occurs in) there may be more freedom in choosing a mensural classifier
than in choosing a sortal one. For instance, in Tzeltal (Mayan: Berlin 1968:
175) the noun lagrio 'brick' is used with just one sortal classifier peck
'rectangular, non-flexible object'; when counted it can occur with several
different mensural classifiers depending on the arrangement: classifier latz
is used to refer to a stack of bricks, chol to aligned bricks, and bus is used
for a pile of bricks.
There are further points on which mensural and sortal classifiers differ in
their behaviour. Different preferences for the lexical sources of the two
kinds of numeral classifiers are discussed in §13.1, and the differences in the
modes of obsolescence of mensural and sortal classifiers are discussed in
§13.7.
Another problem is distinguishing between numeral classifiers, especially
those of the mensural type, and quantifiers (sometimes called 'measure
words'). As Ahrens (1994: 204) put it, 'classifiers can only classify over a
limited and specific group of nouns, while measure words can be used as a
measure for a wide variety of nouns'. Almost every language, whether it has
numeral classifiers or not, has quantifiers, the choice of which may depend
on the semantics of the noun (e.g. in English much is used with non-
countable nouns and many with countable nouns). Examples include
'head' in English five head of cattle and Russian golova 'head' in its
equivalent pjatj golov skota; stack in English three stacks of books; csepp
'drop' in Hungarian egy csepp mez 'one drop of honey'. These quantifying
expressions are not numeral classifiers for the following reasons.
116 Classifiers
(a) They do not fill an obligatory slot in the numeral-noun construction.
(b) They often have a lexical meaning of their own (unlike mensural
classifiers).
(c) Their usage is often related to the distinction between mass and count
nouns; for instance, 'honey', a noun with mass reference in Hungarian, has
to occur with a quantifier in order to be counted (*egy mez '?one honey'
instead of egy csepp mez 'one drop of honey' which is ungrammatical); see
Lyons (1977) and Craig (forthcoming) for the semantic distinctions
involved in dividing nouns into mass and count.
(d) They are used in a type of construction which is also employed for
other purposes. For instance, quantifier constructions in English three
heads of cattle are in fact a subtype of genitive constructions.9 This is the
main reason why English is not a numeral classifier language (see further
discussion in Lehrer 1986).
(e) There is a restricted number of such words in a non-classifying lan-
guage; they also have a restricted distribution (for more discussion, see
Dixon 1982: 211; Adams 1989: 5 ff.; Lehrer 1986; Beckwith 1992).
In some languages numeral classifiers may also be used with a quanti-
fying meaning, such as 'half, 'quarter', 'slice'. In Baniwa of Icana (North
Arawak) these 'quantifying' classifiers are used as agreement markers on
other types of modifiers (Aikhenvald 1996c). This language also has
quantifiers as a closed word class. These do not take classifiers, or
show any agreement with a noun they refer to, e.g. manupe inu 'many
dogs', manupe panti 'many houses' (see Table 9.5 for classifiers in
Baniwa).
Classifiers and quantifiers may be hard to distinguish if they occupy the
same slot in a noun phrase, as appears to be the case in a number of
Austroasiatic languages (cf. Adams 1989: 3).10 Numeral classifiers, then,
have to be distinguished from quantifiers on the basis of language specific
criteria.
Quantifying expressions and numeral classifiers can have semantic and
grammatical differences, which are considered in (A) and (B) below
(further criteria are discussed by T'sou 1976; Bisang 1993: 8–14; Pe
1965).

The same construction type is employed independently whether the measure words just
quantify the referent of a noun (as in half of) or contain some reference to arrangement (as in
row of corn).
10
There are also significant differences in the use of such terms as (numeral) classifier and
quantifier. Burling (1965) uses the term 'numeral classifier' to refer to all the items which occur
in the slot adjacent to a number, as a cover term for both classifiers and measure words.
Huffman (1970) refers to this group of items as 'specifiers'.
 Numeral Classifiers 117
(A) Semantic and pragmatic criteria for distinguishing classifiers from
quantifiying expressions
Classifiers use the unit provided by a count noun, while quantifiers
establish the unit to be counted. Classifiers categorize nouns in terms
of their size, shape and animacy; they provide no information as to
'quantity' (Allan 1977; Becker 1975; Adams 1989: 6). Quantifiying
expressions have fewer restrictions than classifiers on the type of noun
they can cooccur with. In Minangkabau (Rina Marnita, p.c.), 'half is a
quantifying expression, and not a numeral classifier, since it can be used
with any countable noun.
A distinction between classifiers and quantifying expressions is usually
linked to the division of nouns into countable and mass (or uncountable).
Mass nouns can only be combined with a numeral through the use of a
quantifier (cf. Adams 1989: 9). This is illustrated with Comaltepec Chinan-
tec (Otomanguean: Anderson 1989: 61). Example 4.36 shows a countable
noun 'orange' in a numeral phrase, and 4.37 shows a mass noun 'paper'
with a classifier (M and L indicate middle and low tones).
4.36. geM hj?L
seven orange
'seven oranges'
4.37. tu,M ma?L maL hiL
two CL:LEAF paper
'two sheets of paper'
Unlike quantifying expressions, classifiers can have pragmatic uses. For
example, in Assamese classifiers mark definiteness-indefiniteness (Barz and
Diller 1985), while classifiers in Vietnamese signal definiteness and refer-
entiality of the noun (Lobel forthcoming: 45; cf. §12.1.2). Quantifying
expressions usually do not have such functions.
The occurrence of classifiers in numeral phrases may not be obligatory.
In Minangkabau classifiers are often omitted in everyday language, and
this does not change the semantics of a numeral phrase. In contrast, the
omission of a quantifying expression does affect the meaning.
In Khmer, classifiers are obligatory in formal standard language, but not
in informal language (the only classifier which is used more or less con-
sistently is the human classifier: Walter Bisang, p.c.); quantifying expres-
sions are 'obligatory' in the sense that their omission alters the sense
(Adams 1989: 9).
Unlike quantifying expressions, classifiers may be optional for some
numbers, usually for big ones (see §4.1; and §4.3.2 on Akatek).
118 Classifiers
(B) Grammatical criteria for distinguishing classifiers from quantifying
expressions
Classifiers and quantifying expressions can differ in possibilities of use in
other classifier environments; in anaphoric use; and in agreement.
The different morphosyntactic behaviour of classifiers and 'quantifying
expressions' in Comaltepec Chinantec (Otomanguean: Anderson 1989: 58)
is shown by the fact that quantifying expressions agree with the head noun
in gender (animate vs. inanimate), while classifiers do not.
In Nung (Thai: Saul and Wilson 1980: 25-9) only classifiers, not quan-
tifiers, can be used anaphorically, i.e. as a 'substitute' for a head noun. A
classifier construction is illustrated in 4.38.
4.38. slam ahn boc
three CL: GENERIC flower
'three flowers'
Anaphoric use of a classifier is illustrated in 4.39. The classifier ohng
'HUMAN' is repeated, and it has distributive meaning 'everyone'.
4.39. ohng ohng to ma cheu
CL:HUMAN CL:HUMAN also Come look
'Everyone also came to look.'
In multiple classifier systems, the same morphemes may be used with
numerals and in other environments; quantifiers are not used this way. In
Nung classifiers are also used in possessive and demonstrative NPs (Saul
and Wilson 1980: 25). In Chinese and Vietnamese classifiers occur with
demonstratives (Adams 1989: 10; Goral 1978).
In Akatek (Zavala 1992: 145), some independent numeral classifiers can
be used as quantifying expressions (§4.3.2) while affixed numeral classifiers
cannot. Example 4.40 illustrates the use of kupan 'semi-circle' as a numeral
classifier. An affixed numeral classifier can be omitted.
4.40. ?os-eb' kupan ?isim
three-NUM.CL:INAN NUM.CL:SEMI.CIRCULAR NOUN.CL:MAIZE
paat
tortilla
'three quesadillas (taco of folded tortillas which form a semi-circle)'
Example 4.41 illustrates the use of the same item as a quantifying
expression, 'a heap of semi-circular objects' (Zavala 1992: 145-6). Here,
the affixed numeral classifier is required.
 Numeral Classifiers 119
4.41. ?os-eb' mimex kupan
three-NUM.CL:INAN big QUANT:SEMI.CIRCULAR.HEAP
?isim paat
NOUN.CL:MAIZE tortilla
'three big heaps of tortillas'
There are other morphosyntactic differences between quantifying
expressions and numeral classifiers in Akatek (Zavala 1992: 143). For
instance, a suffixed numeral classifier agrees with the head noun in a number
phrase which does not contain a quantifying expression, as illustrated in
4.42.
4.42. ?os-wan k'itan eb'
three-NUM.CL:HUMAN NUM.CL:SEPARATE PL:HUMAN
nax winax
NOUN.CL:MAN man
'three men' (viewed separately)
If a number phrase contains a quantifying expression, the affixed
numeral classifier agrees with it. This is illustrated in 4.43. Classifier -eb'
'inanimate' refers to tinan 'conglomerate'.
4.43. ?os-eb' tinan eb'
three-NUM.CL:INAN QUANT:CONGLOMERATE PL:HUMAN
nax winax
NOUN.CL:MAN man
'three groups of men' (group is 'inanimate')
These examples from Akatek clearly show that while a quantifier 'provides'
the unit that is counted, the classifier categorizes this unit.
Lobel (forthcoming: 30) shows that the connection between a quantifier
and a noun is not as strong as that between a classifier and a classified
noun, in spite of the surface similarity between classifier phrases and
measure phrases. A stative verb, e.g. day 'be full of or a noun ru'o'i 'half
of can come between a quantifier and a noun, as illustrated in 4.44.
4.44. mot can ru'o'i cho
on pound half.of dog
'one and a half pounds of dog (meat)' (at the butchers)
This is not possible in classifier phrases; then, these forms have to follow
the entire classifier phrase, as shown in 4.45.
4.45. mot con cho ru'o'i
one CL:ANIMAL dog half
'one and a half dogs' (at the butcher's)
120 Classifiers
Difficulties with discriminating between quantifiers and classifiers may
be due to the fact that these are better viewed as extremes of a continuum.
Note Becker's (1975: 114) suggestion that quantity and quality are possibly
not discrete semantic classes but rather 'polarities in a semantic contin-
uum'. Similarly to mensural classifiers, quantifiers cooccur with numerals
and their choice may also correlate with the properties of the units enum-
erated (Downing 1996: 13). Consequently, in some languages the two
categories simply cannot be clearly distinguished (this appears to be the
case in Korean). Some properties which proved to be useful for distinguish-
ing quantifiers and classifiers have been given above; internal linguistic
criteria should always be used to establish this differentiation.
There may be a historically attested diachronic process of change from a
numeral classifier to a quantifier with a corresponding meaning shift. In
Minangkabau miya was originally just used as a classifier for 'bread
crumbs'; in the modern language this item is used as a quantifier 'a little
bit' (Rina Marnita, p.c.; Conklin 1981).11

4.4.2. Incipient numeral classifiers

Non-classifier languages can have incipient systems of numeral classifiers.
Numeral classifiers are emerging in the Omani-Zanzibar and in the
Egyptian variety of Arabic (Greenberg 1990: 178 ff.; Classical Arabic has
no classifiers). Omani Arabic has four classifiers used in noun phrases with
numerals: ra:s 'head' is used for some animals usable as meat, root crops,
and 'slave'; qarn 'horn' is used for corn-shaped edibles; so:b 'fruit' is used
to classify fruits, and 'o:d 'branch' to classify flowers. Not all nouns in the
language take classifiers. The numeral precedes the classifier, agreeing with
it in gender. The construction numeral + classifier is similar to the con-
struction numeral + noun. The classified noun follows the classifier and
does not change its form, e.g. ra:s finda:1 '(one) potato' (lit. head potato);
thala:thit rwa:s finda:l 'three potatoes' (lit. three head:?L potato). These
classifiers are restricted to limited classes of nouns, and they are optional.
In Russian (Greenberg 1990: 181–3) celovek, the genitive plural form of
'man', can be used with numerals bigger than four for a restricted set of
'' Numeral classifiers may be difficult to distinguish from compounding. In the following
example from Vietnamese it seems to be virtually impossible to tell a classifier from a part of a
nominal compound in a phrase like (i) (Adams 1989: 11). See also Bisang (forthcoming).
(i) mot ngu o i la 'a stranger'
one person strange
In this example ngudi is a general classifier for 'person', and also a member of compound
meaning 'stranger'. According to Lobel (forthcoming: 48–51), stress is decisive here: in the
case of a compound it is on the first element, while in a classifier construction the last element
receives the stress.
 Numeral Classifiers 121
human referents. The classifying noun takes genitive case (as do nouns
'governed' by numerals bigger than four), e.g. pjatj celovek detej (five
peopleroN.PL children:GN.PL) 'five children' (lit. five people children). In
this construction, the 'classifier' is governed by the numeral exactly as a
head noun is governed by the numeral. This construction may go back to
the individualizing use of generics with collective nouns. Similarly to
Arabic, the classifier use of celovek 'man:GN.PL' is optional. Noun phrases
lacking it, e.g. pjatero detej 'five:COLL children:GN.PL', or pjatj detej 'five
children:GN.PL', are perfectly grammatical. Also, the classifier is used for a
very limited class of nouns.
These examples provide evidence in favour of the emergence of numeral
classifiers from independent lexical items in inflecting languages (see
Chapter 13).

4.5. Distribution of numeral classifiers in the languages of the

world
Numeral classifiers are the second most frequent type of classifiers, after
noun classes and genders—see Map 3.
Numeral classifiers are widespread across the languages of East and
Southeast Asia and Oceania. They are present in many Tibeto-Burman
languages, in Chinese languages, and in most Austroasiatic languages. In
the Far East, Japanese, Korean, and Ainu have numeral classifiers.
Numeral classifiers are not found in languages of the Afroasiatic family.
Among Indo-European languages, some of the Indie and Iranian lan-
guages (e.g, Marathi, Hindi, Persian) have numeral classifiers, as do
many Dravidian languages. (This is an areal phenomenon in South Asia:
see Emeneau 1964.)
Among Uralic languages, Hungarian is said to have numeral classifiers
(Beckwith 1992), similar to many Turkic languages (Vietze 1979; Beckwith
1998). Nivkh (or Gilyak; Paleosiberian isolate) has a large set of numeral
classifiers.
Numeral classifiers are found in scattered pockets across North America
(Sherzer 1976; Campbell 1997). The inventories of classifiers are typically
small, and the semantics of their classifiers are based on shape and
animacy. In the Eastern subarctic linguistic area, Ojibway (Algonquian)
has numeral classifiers (as well as verbal classifiers: Denny 1979a). Numeral
classifiers are a central areal trait of the Northwest Coast linguistic area,
present in those languages of the Eyak-Athabaskan, Haida, Tlingit,
Wakashan, Chemakuan, and Salish families which are spoken in the
area (Sherzer 1976: 74). Tsimshian apparently acquired them through
contact with its neighbours. The Salishan languages of the Plateau area
 numeral classifiers found in continuous areas

isolated instances of numeral classifiers

MAP 3. Distribution of numeral classifiers in the languages of the world

 Numeral Classifiers 123
have shape and form-based numeral classifiers. Human/non-human is dis-
tinguished for numerals in Nez Perce, Sahaptin, and Upper Chinook and in
one Salish language, Colville. In Sahaptin, number marking in adverbial
cases depends on whether or not the nominal refers to a human (Rigsby
and Rude 1996).
Numeral classifiers, based on the shape and form of the object, are an
areal feature of the extreme northwestern region of California; they are
present in Yurok and Wiyot (Algonquian-Ritwan), and in Karok; Hupa
(Athabaskan) distinguishes different numerals for humans and non-
humans. In the Great Basin linguistic area, numerals distinguish persons,
non-persons, and immutable objects only in Washo (isolate). The only
language in the Plains with numeral classifiers is Blackfoot (Algonquian),
which has two sets of numerals, one for animate and the other for inani-
mate nouns. In the Northeast linguistic area, Menomini and Potawatomi
(Algonquian) have form/shape-based numeral classifiers. Absence of
numeral classifiers is a feature of the Pueblo and Southeast linguistic areas.
Numerous Mexican and Central American languages have numeral
classifiers, e.g. Aztec (Uto-Aztecan), Huave (isolate), Totonac (isolate),
Sierra Popoluca (Zoquean), Zapotec (Otomanguean), Nahuatl (Campbell
et al. 1986: 550; Costenla Umana 1991: 116; Campbell 1997). Mayan
languages (especially these of the Kanjobalan branch) have extensive sys-
tems of numeral and noun classifiers. Some instances of numeral classifiers
in Mayan languages may be the result of diffusion (Hopkins 1970).
Large sets of numeral classifiers are present in numerous languages of
South America, especially those of Lowland Amazonia, including the
languages of such families as Arawak, Tucano, Guahibo, Peba-Yagua,
Chapahuan, Harakmbet, Bora-Witoto, Nambiquara, Tsafiki (Barbacoan:
Connie Dickinson, p.c.) and a few isolates, e.g. Waorani and Saliba (see
Derbyshire and Payne 1990; Dixon and Aikhenvald 1999). Some Tupi
languages, e.g. Munduruku, also have numeral classifiers. Chimila (Amaya
1997: 139) appears to be the only language with numeral classifiers among
the Chibchan languages of Colombia.
Numeral classifiers are found in a number of Papuan languages. In the
East Sepik province, Iwam (Conrad and Conrad n.d.; Laycock and Z'graggen
1975: 744) has five numeral classifiers, and Abau (Lock forthcoming;
Laycock and Z'graggen 1975: 744) has twelve. Chambri (Lower Sepik
area) has five numeral classifiers used with numbers from one to five, while
Wogamusin and Chenapian (Laycock and Z'graggen 1975: 743–4) have five
numeral classifiers used with numbers up to four. In the Gulf province,
some Angan languages have numeral classifiers, also used in other envir-
onments (i.e. with adjectives and demonstratives: Speece n.d, for Angave,
and Carlson 1991, for Taenae). In the Highlands, Folopa (Podopa), of the
Teberan family, has numeral classifiers (Anderson and Anderson 1976). In
124 Classifiers
the Morobe province, the closely related Wantoat (Davis n.d.) and Award
(Susan Quigley forthcoming and p.c.) have a largeish set of numeral clas-
sifiers used with the numbers 'one' and 'two' (also employed with demon-
stratives and as derivational affixes on nouns).
Most Western Austronesian and Oceanic languages12 have numeral
classifiers. However, numeral classifiers appear to be absent from the
Austronesian languages spoken in Taiwan. Some Oceanic languages
spoken in the Papua New Guinea region have extensive numeral classifier
systems (for example, Loniu, an Oceanic language spoken in the Admiral-
ties, has thirty numeral classifiers: Hamel 1994). In contrast, Oceanic
languages spoken in Bougainville (e.g. Teop, Halia, or Petals: Ulrike Mosel,
Ruth Spriggs, Evelyn Boxall, p.c.) lack numeral classifiers.
Numeral classifiers are very rare in Africa and absent from Australia. In
Africa, they are found in a few Kegboid languages (Cross River: Benue-
Congo: Ikoro 1994), Ejagham (Walters 1981), and in a few Grassfields
languages from Cameroon (e.g. Ngyemboon: Viktor Vinogradov, p.c.).
12
Pawley (1973) provides a reconstruction of the Proto-Oceanic numeral classifers.
 5 Classifiers in Possessive
Constructions

5.1 Categorization in possessive constructions

There are three ways of categorizing nouns in possessive constructions and
these correspond to three kinds of noun categorization devices.
(A) Categorizing the possessed noun
The choice of classifiers in possessive constructions can be determined by
the nature of the referent of the possessed noun in terms of its animacy,
shape, form, etc. (cf. Craig forthcoming: 32; 1992). I shall call these classi-
fiers 'possessed' classifiers (see §5.2).
(B) Categorizing the semantic nature of a relation between the possessee
and the possessor in a possessive construction
The choice of possessive marker can be determined by the way the pos-
sessor handles or owns or otherwise relates to the possessee. I shall call this
type of categorization 'relational' classifiers, following Lichtenberk (1983a:
148): 'The crucial property of relational classifiers is that their use is
determined ... by the semantic relation between two linguistic elements',
e.g., how the possessor might use the possessed.
Relational classifiers (found in Austronesian and in a few American
Indian languages; see Carlson and Payne 1989; Croft 1994) are considered
in §5.3.
(C) Categorizing the possessor
'Possessor classifiers'—morphemes the choice of which is conditioned by
the properties of possessor—are very rare; see §5.4.
The three types of categorization in possessive constructions1 are relatively
independent of each other. Relational classifiers may combine with pos-
sessed classifiers to create an integrated system which includes both kinds
of categorization (since the way a noun can be 'possessed' may correlate
with its inherent properties). This is the case in some Micronesian lan-
guages with large systems of classifiers in possessive constructions. There
1
See Heine (1997a), H. Seller (1986), and Chappell and McGregor (1996) for detailed
discussions of possession.
126 Classifiers
are other languages which have distinct sets of relational and possessed
classifiers (see §5.5).
The different types of classifiers in possessive constructions are con-
trasted in §5.6.2 Their distribution in the languages of the world is outlined
in §5.7.

5.2. Possessed classifiers

Possessed classifiers characterize the possessed term in a possessive con-

struction. They do not involve agreement, and their choice is strictly
semantic. They share the following properties with numeral classifiers
and noun classifiers.
(i) They characterize nouns in terms of their animacy, shape, size, and
structure.
(ii) They are not expressed outside the possessive NP.
(iii) Every noun in a language may not necessarily be able to take a
possessed classifier.
(iv) Some languages can have a 'generic' possessed classifier which replaces
other, more specific, classifiers.
The size of the inventory of possessed classifiers can vary. Morpho-
logically they can be independent words or affixes to the possessed noun
or the possession marker.
Since possessed classifiers do not characterize the type of relationship
between the possessor and the possessed, they do not necessarily interrelate
with, or depend upon, the distinction between alienable and inalienable
possession. In some languages possessed classifiers may be used only with
different subclasses of alienably possessed nouns (A); in other languages
possessed classifiers are used independently of whether a noun is alienably
or inalienably possessed (B). Possessed classifiers of group A are not used
in any other environments. Possessed classifiers of group B often are.
(A) Possessed classifiers used only with subclasses of alienably possessed
nouns
Possessed classifiers are used with alienably possessed nouns in Yuman and
Uto-Aztecan languages.
In Yuman languages inalienable possession is indicated by a prefix on the
possessed noun, or a prefix on the possessed noun plus a pronoun. Possession
2
The existing literature is somewhat confusing in distinguishing between different classifier
types in possessive constructions. Most typologies consider the three types distinguished here
as just one kind of noun categorization (e.g. Lichtenberk 1983a; Craig forthcoming).
 Classifiers in Possessive Constructions 127
of alienably possessed nouns is expressed either by a noun followed by a
classifier with affixes indicating possession, or by a noun with possessive
suffixes attached to it. There are two classifiers, one for 'pets and domestic
animals', and the other a general one (Hualapai, Yavapai, Maricopa -hat
'pet'; Yavapai wi, Hualapai -wi:nych, Maricopa nywish 'general posses-
sion'). A possessed classifier for 'pet' in Yavapai is exemplified with 5.1;
a general classifier in Hualapai is shown in 5.2 (Carlson and Payne 1989).
5.1. qoleyaw ?-n-hat
chicken ISG-GENITIVE-CL:PET
'my chicken' (chicken-my pet)
5.2. mad" ma m-wi:nych
land your 2SG-CL:GENERAL
'your land'
According to Langacker (1977), 'an important feature of Uto-Aztecan
languages is a "classifier" construction'. To indicate alienable possession,
possessive prefixes are attached to the classifier. Languages typically have
two or three classifiers, which distinguish animacy and humanness. Cora
(Casad 1982: 236) distinguishes human, animate, and inanimate; Papago
and Northern Tepehuan (Bascom 1982; Saxton 1982) have animate and
inanimate classifiers.3 Papago classifier soi- with an animate noun is illu-
strated in 5.3 (Saxton 1982: 186–7).
5.3. has-cu ?oi-g-j g huan
what-thing CL:ANIMATE-ALIENABLE-GENITIVE ART Juan
'What kind of animal does Juan have?' (lit. what kind of animal of
Juan?)
Possessed classifiers in Uto-Aztecan can be used headlessly, in a kind of
anaphoric function. This is illustrated in 5.4, from Papago (Langacker
1977: 92), where the noun miisa 'table' is included in copula subject func-
tion; it is omitted from the NP in the function of a copula complement.
5.4. iida miisa o=d t-'ini-ga
this table be our-CL:INANIMATE-possessed
This table is ours'
Panare (Carib: Mattei-Muller 1974; Carlson and Payne 1989) has
twenty-one classifiers which are used with alienably possessed nouns
(Carlson and Payne 1989: 19). Classifiers characterize the possessed
noun in terms of its shape, structure, and consistency. There are seven
3
Some languages have just one possessed classifier, e.g. Luiseno classifier 'aac for pets.
Chemehuevi (Southern Numic branch of Uto-Aztecan) has two possessed classifiers: -putjku
'pet' and -igapt 'domesticated plant" (Press 1979: 60–1).
128 Classifiers
classifiers which are used with a single item each. Four of these are repea-
ters, i.e. they coincide with the noun which they categorize in a possessive
construction: see Table 5.1.

TABLE 5.1. Possessed classifiers in Panare

Semantics Classifier Examples
General iyu Soap, gasoline, hammer, paper, table etc.
Edible 1 yung Food items in paste form: mango, corn
(mash), manioc
Edible 2 ta'ma Something softer than yung: egg, rice, soup
Edible 3 empa Fruit with its pulp, not squeezed out
Edible 4 yo'lare Meat (cooked or raw): fish, chicken, beef
Drinkable/liquid uku Liquids: coffee, milk, blood
Animal yiki Live animals, including domesticated ones
Vehicle kanowa Canoe, truck, helicopter, car
Hunting arms ko Spear, arrow, harpoon
Musical instrument ntyen Flute, guitar, violin
Body paint yanoe Body paints
Clothing po' Necklace, shirt
Container mara pi Gourd, etc.
Artificial light uyung Lantern, flashlight
Single item classes ewi' House
tipi' Swidden garden
ichi' Hammock
Repeaters pata Village, community
chiste Hatchet
wata Blowgun
wa'to Fire

5.5 illustrates the use of the possessed classifier uku 'liquid'.

5.5. y-uku-n wane
1 so-CL: LIQUID-GN honey
'my honey (mixed with water for drinking)'
In Macushi and Apalai (further Carib languages), the classifiers—which
are used with alienably possessed nouns—are themselves in fact a subclass
of inalienably possessed nouns. Inalienably possessed nouns can take pos-
sessive affixes directly, while the alienably possessed ones cannot. Classi-
fiers are generic terms, which specify nouns that are to be classified. 5.6
illustrates a generic term for food in Apalai (Koehn and Koehn 1986:
85 ff.).
 Classifiers in Possessive Constructions 129
5.6. a-napy-ry
2sG-fruit/vegetable-poss
'your fruit, vegetable'
Names of plants, animals, and natural phenomena cannot take posses-
sive affixes directly; these must be added to an accompanying possessed
classifier with a 'generic' semantics (cf. 'generics' used as noun classifiers:
§3), as in 5.7.
5.7. a-napy-ry paruru
2SG-fruit/vegetable-poss banana
'your banana' (lit. 'your fruit banana')
Similar constructions from Macushi (Abbott 1991: 85-6) are shown in
5.8 and 5.9.
5.8. u-yekin kaware
Iso-pet horse
'my horse'
5.9. u-yekkari ma'piya
1 SG-FRUIT:FOOD papaya
'my papaya'
Classifiers of this kind, in both Macushi and Apalai, are a semi-open
class, since any generic noun can be used as a 'classifier' in a possessive
construction. These languages do have an interrelation between possession
type and the use of possessed classifier. However, this operates on a
different basis from Yuman and Uto-Aztecan languages: both types of
possession are involved in a classifier construction in the Carib languages,
while possessed classifiers are restricted to constructions with alienable
possession in Yuman and Uto-Aztecan. Possessed classifiers which are a
subclass of nouns with generic semantics are used with alienably possessed
items in other South American languages. Northern Je languages, such as
Timbira (Canela, Kraho), Kayapo, and Panara, have one generic classifier
for all alienably possessed items; Bororo (Macro-Je) has one classifier for
pets and another for all other alienably possessed nouns; Tupi-Guarani
languages typically have one classifier for pets and another for game
(Rodrigues 1997: 72–3).4
(B) Possessed classifiers used independently of possession type
Possessed classifiers can be used independently of possession type both in
languages which have a grammatical distinction between alienable and
inalienable possession, and in languages which have no such distinction.
4
For possessed classifiers in Urubu-Kaapor (Tupi-Guarani), see Kakumasu (1986: 371).
130 Classifiers
'Dongo-ko (Mba group, Ubangi branch of Niger-Congo: Pasch 1985:
75 ff.; 1986: 240 ff.) has ten possessed classifiers. There is an elaborate clas-
sification of nouns with classifiers used just in possessive constructions.
Example 5.10 illustrates the use of the classifier da 'possession of animals
and some inanimate objects' (Pasch 1986: 248). (Note that 'Dongo-ko, simi-
larly to other Mba languages, also has noun class agreement; see §2.4.2.)
5.10. bi-go a da re
leopard-CL7 CL1-POSS CL:ANIMAL.POSS 1SG
'my leopard'
5.11 illustrates the use of the classifier k6 which refers to the possession
of 'inseparable' body parts (Pasch 1986: 249).
5.11. nzi m-a k5 i-go
blood CL11-poss CL:BODY.PART.Poss leopard-CL7
'leopard's blood'
Guaicuruan languages (Brazil and Argentina) have no grammatical dis-
tinction between alienable and inalienable possession. Some of them do
have possessed classifiers (Ceria and Sandalo 1995: 14). For instance, Toba
and Mocovi each have one possessed classifier (lo 'animal classifier'), and
Kadiweu has two classifiers; one is used with domestic animals (wigadi) and
the other is used with other nouns (nebi 'generic classifier'). The animal
possessive classifier agrees in gender (feminine/non-feminine) and number
(singular, plural) with the possessed noun, as shown in 5.12, from Kadiweu.
5.12. li-wigagi nigidagiwaga
3-CL:ANIMATE:NON.FEM.SG pig(NON.FEM)
'his (male) pig'
The generic classifier, illustrated in 5.13, shows no agreement (Griffiths
and Griffiths 1976: 101–3).5
5.13. go-nebi leyeema
1PL-CL:GENERIC wheat
'our wheat'
Possession in Kadiweu can also be expressed with a possessive prefix
attached to the noun; for emphasis, the possessive prefix may attach both
to the classifier and to the possessed noun. This is illustrated in 5.14.
5.14. jabeyagi ga-nibole ga-nebi-tiwaji
already.spoilt 2pL-meat 2pL-CL:GENERIC-PL
'Your meat is already spoilt.'
5
Griffiths and Griffiths (1976: 103) note a certain degree of interchangeability between
animate and inanimate classifiers for animate referents.
 Classifiers in Possessive Constructions 131
Possessive prefixes and possessed classifiers occur together in equational
copula clauses, illustrated in 5.15. (Examples like this can also be inter-
preted as instances of the headless use of possessive classifiers.)
5.15. go-dacilo go-nebi
1 PL-head 1PL-CL:GENERIC
The head is ours.'
Classifiers characterize the possessed noun in possessive constructions
independently of possession type in a number of multiple classifier lan-
guages, including Tariana and Baniwa (North Arawak), East Tucano,
Hmong (Miao-Yao: Bisang 1993), and Motuna and Nasioi (Papuan).
Baniwa and Tariana and the East Tucano languages distinguish alien-
ably and inalienably possessed nouns. Kinship terms, body parts, and a few
other items, e.g. home, are inalienably possessed; other nouns are alienably
possessed. These languages have large sets of classifiers which characterize
the noun in terms of animacy, shape, structure, size; classifier morphemes
can be used in several environments (see §9.3; Aikhenvald 1994a; 1996c).
Possessed classifiers are often employed anaphorically.
Inalienable possession in Tariana is marked with possessive prefixes, as
shown in 5.16. Alienable possession is marked by juxtaposition of posses-
sor and possessed, as in 5.17. Note that 5.17a, with a possessive prefix, is
ungrammatical.
5.16. nu-pana
IsG-home
'my home'
5.17. nuha tfinu
I dog
'my dog'
5.17a. * n u - n u
Possessed classifiers can be used with any noun if the fact of a noun
being possessed is to be focused on. Example 5.18 illustrates the use of a
possessed classifier with an inalienably possessed noun, panisi 'home', in
Tariana. Example 5.19 illustrates the use of a possessed classifier with an
alienably possessed noun, tfinu 'dog'. The head noun is in parentheses
(since it can be omitted).
5.18. nu-ya-dapana (panisi)
1SG-POSS-CL:HOUSE (home)
'my (home)'
132 Classifiers
5.19. nu-ite (tfinu)
lSG-POSS+CL:ANIM (dog)
'my (dog)'
Two Papuan languages of Southern Bougainville, Nasioi Hurd 1977)
and Motuna (Onishi 1994), make no grammatical distinction between
alienable and inalienable possession. Possessed classifiers are used in a
special 'classifier' possessive construction type, and also in other classifier
environments (see §9.1). (§5.4 discusses another possessive construction in
Nasioi which is similar to possessor classifiers.) A possessed classifier in
Motuna is illustrated in 5.20 (Onishi 1994: 131), and a classifier used in a
possessive construction in Nasioi is shown in 5.21.
5.20. ong moo ngo-no-mung
DEM.MASC coconut 1 SG.POSS-LINK-CL:PLANT/FRUIT
'this/that coconut, (which is) my plant/fruit'

5.21. n-ee-ka-na-va
US-DU-INTENSE-DER.SUFF-CL:HOUSE
'our (house)' (Hurd 1977: 155)
In Hmong, a Miao language of China—and in some other Miao lan-
guages of the region—possessed classifiers are used with both alienably and
inalienably possessed nouns. (The same set of classifiers is used with
demonstratives and with numerals: see §9.1.) The possessed classifier can
only be omitted from possessive constructions with inalienable possession
(Bisang 1993: 29-30). A possessed classifier is used with an inalienably
possessed noun in 5.22. It is omitted in 5.23, in a construction with
inalienable possession.
5.22. nws tus txiv ntxawm
he CL:Liv.BEING uncle
tus ntxhais
CL:LIV.BEING daughter
'the daughter of his uncle'

5.23. kuv txiv

I father
'my father'
Example 5.24 illustrates the use of a possessed classifier with an alienably
possessed noun 'sword'. Here, the possessed classifier cannot be omitted.
5.24. nws rab riam ntaj
he CL:ARTEFACT sword
'his sword'
 Classifiers in Possessive Constructions 133
5.3. Relational classifiers

Relational classifiers are unlike any other classifier type in that, instead of
just characterizing a noun, they characterize a possessive relation between
nouns. Like possessed classifiers, they do not involve agreement.
They are not marked outside the possessive NP. Every noun in a lan-
guage does not necessarily take a relational classifier. There may be a
generic classifier, and the size of the inventory can vary. Morphologically,
they can be realized as independent words or as affixes to the possessed
noun or the possession marker.
Relational classifiers are almost always restricted to constructions of
alienable possession. I consider examples of relational classifiers in
§5.3.1. The distinction between alienable and inalienable possession is
another way of categorizing a possessive relationship. In §5.3.2, relational
classifiers are compared with the alienable/inalienable distinction.
Unlike possessed classifiers, relational classifiers are not used in other
classifier constructions within multiple classifier systems. Their semantics is
often different from that of other classifier types: a parameter such as value
is more frequent in relational classifiers than in other classifier types (see
§11.2.4).

5.3.1. Relational classifiers and their properties

Relational classifiers are found in the Oceanic subgroup of Austronesian,6
and in a few South American Indian languages (Kipea, from the Kipea-
Kariri family, South America, extinct: Rodrigues 1997; Baniwa of Icana,
from the North Arawak subgroup). Large systems of relational classifiers
(integrated with possessed classifiers) will be considered in §5.5.1.
All Oceanic languages distinguish between alienable (indirect) and
inalienable (direct) possession. In most languages, inalienable possession
is restricted to kinship nouns and parts of a whole; it is indicated by suffixes
attached to possessed nouns, e.g. Standard Fijian (Lichtenberk 1983a: 153):
5.25. na tama-dratou
ART father-their
'their (paucal) father'
5.26. na ulu-qu
ART head-my
'my head'
6
The existence of relational classifiers in Oceanic languages was first recognized by
Codrington (1885); see Harrison (1976), for Mokilese; Dixon (1988), for Boumaa Fijian;
Pawley and Sayaba (1990), for Wayan, a Western Fijian language; for a more general discus-
sion, see Lichtenberk (1983a); Carlson and Payne (1989).
134 Classifiers
To indicate alienable possession, pronominal affixes are attached to
relational classifiers, the special morphemes which indicate the type of
relationship between the possessor and the possessed, e.g. Boumaa Fijian
(Dixon 1988: 137):
5.27. a o-mu da'ai
ART CL-2SG gun
'your gun' (which belongs to you)
5.28. a e-mu da'ai
ART CL-2so gun
'your gun' (which will be used to shoot you)
Oceanic languages have the following types of systems of relational
classifiers (Lichtenberk 1983a).7
(a) A system of two relational classifiers involving the opposition: general
versus alimentary/consumable (cf. Heine 1997a). This is illustrated for
Manam (Lichtenberk 1983a: 151) and Kaliai-Kove (Counts 1969: 100) in
Table 5.2. (Two relational classifiers in Baniwa are discussed in §5.5.2.)

TABLE 5.2. Systems of two relational classifiers

Language General Alimentary

Manam ne ?ana
Kaliai-Kove le a

(b) A system of four relational classifiers is illustrated in Table 5.3 for

Boumaa Fijian (Dixon 1988: 136):

Table 5.3. A system of relational classifiers in Boumaa Fijian

Meaning Form

(a) consumed (i) drunk/sucked/licked me-

(ii) eaten/chewed/smoked 'e-
(b) not consumed (ii) relating to the 'possessor' e-, as in 5.28
(iii) owned by the 'possessor' we-/o-, as in 5.27

7
According to Lynch (1993), the distinction between direct and indirect possession and
possessive markers, i.e. relational classifiers, can be reconstructed for Proto-Oceanic. Austro-
nesianists (see Lichtenberk 1985) reconstruct three 'indirect' possessive markers: (a) drink
possession, Proto-Oceanic *ma-; (b) food and passive possession, Proto-Oceanic *ka- (c)
general possession, marked by Proto-Oceanic *na-. According to Lynch (1993), they have a
verbal origin. This is compatible with the derivation of classifiers from independent lexical
items; see Chapter 13.
 Classifiers in Possessive Constructions 135
Classifier 'e- is used for anything which undergoes change of state as it is
being consumed; and me- is used for anything which does not undergo a
change of shape or state as it is being consumed (including a pill that is
swallowed).
In Raga, another Oceanic language spoken in Vanuatu, there is a
five-term system, including a special classifier for valued possession
(Lichtenberk 1983a: 154). This is illustrated in 5.29.
5.29. qoe pila-ma
pig CL:VALUABLE.POSSESSION-2SG
'your valued pig'
Kipea-Kariri (Rodrigues 1995; 1997) had a system of twelve classifiers
characterizing the relationship between the possessor and the possessed in
terms of how the possessed could be handled by the possessor. Classifiers
used with food items categorize them in terms of (a) acquisition: (i)
gathering of wild plants: classifier uapru; (ii) raising of animals: enki;
(iii) cultivation of manioc: uanhi; (iv) cultivation of other plants: udje; and
(b) preparation: (i) boiling: ude; (ii) roasting: upodo; (iii) maturation at
home: ubo. Other items are also categorized in terms of (a) acquisition: (i)
finding: uito; (ii) sharing: ukisi; (iii) gift from outsiders: uba; (iv) booty:
boronunu, or (b) transportation—carried goods: e.
Examples 5.30-3 (Rodrigues 1995) show the same item, sabuka 'fowl',
with different relational classifiers, to reflect the different ways in which it
can be handled by the possessor.
5.30. dz-upodo do sabuka
1 SG-CL:ROASTED POSS fowl
'my fowl (roasted)'
5.31. dz-ude do sabuka
1 SG-CL: BOILED POSS fowl
'my fowl (boiled)'
5.32. dz-ukisi do sabuka
1 SG-CL:SHARING POSS fowl
'my fowl (that was my share)'
5.33. dz-uba do sabuka
lSG-CL:GIFT. FROM.OUTSIDERS POSS fowl
'my fowl (that was given me)'
Polynesian languages have two relational classifiers (Lichtenberk 1983a:
162), commonly o and a as basic forms (these may occur with various
prefixes, usually an article). A-possession implies the control of the pos-
sessor over the relationship, or over the initiation of the relationship
136 Classifiers
(Wilson 1982), and o-possession implies the lack of control of the posses-
sor over the relationship.8 A and O relational classifiers can be used with
the same noun, e.g. examples 5.34 and 5.35 from Hawaiian (Lichtenberk
1983a: 163; examples (84) and (85)).
5.34. k-o-'u inoa
ART-CL-my name
'my name (that represents me)'
5.35. k-a-'u inoa
ART-CL-my name
'my name (that I bestow on someone)'
Relational classifiers can be realized as affixes on possession markers,
and as independent words or as affixes to nouns. They may form one
morphological word with the marker of possession, as in Tolai (Oceanic:
Ulrike Mosel, p.c.), some Micronesian languages (e.g. Truquese: §5.5.1)
and Kipea. They may form one morphological word with the possessed
noun, as in Baniwa (§5.5.2). Truquese has 'repeaters' (see §5.5.1).
Which relational classifier is to be used in a possessive construction
depends on the way the possessed item is to be treated. This is why different
relational classifiers can be employed with the same noun.
In Boumaa Fijian, different classifiers can be used with certain nouns,
especially with some newly introduced items, such as 'jelly (or jello)', and
also some traditional ones, such as tobacco. According to Dixon (1988:
136), 'informants vary as to whether jelly should be be- or me-, depending
on whether or not they chew it before swallowing'. As far as tobacco is
concerned, 'one should use the classifier 'e- to describe tobacco itself, since
it is likely to be rolled and smoked, but me- for a pipe which is sucked—
thus a 'e-na tapa'o 'his tobacco'; but a me-na paipo ni tapa'o 'his tobacco
pipe' (see Table 5.3). A tripartite system of relational classifiers in Kilivila
involves the division of nouns into edible (Series I: 5.36), consumable or
closely associated with the possessor (Series II: 5.37), or more distantly
associated with the possessor (Series III: 5.38) (Senft 1986: 49-54).
5.36. kagu tetu
my:REL.CL:SERIES.I yam
'my yams (to eat)'

8
These relationships are described as 'objective' or 'subjective', for Tongan by Churchward
(1953: 78). According to some interpretations (H. Seiler 1986; Lichtenberk 1983a), the two
relational classifiers in Polynesian can be interpreted as corresponding to alienable and
inalienable possession. In some languages, e.g. Tongan, the two different classifiers are used
for subject and object nominalizations, e.g. 'e-ne taki 'his leading', ho-no laki 'his being led'
(Churchward 1953: 78). This shows how distinction between relational classifiers can also be
employed elsewhere in the grammar.
 Classifiers in Possessive Constructions 137
5.37. agu tetu
my:REL.CL:SERIES.II yam
'my yams (planted in the garden)'
5.38. ula tobaki
my:REL.CL:SERIEs.III tobacco
'my tobacco (that I will trade or give away)'

5.3.2. Types of possession and relational classifiers

The way possessive relationships can be categorized with relational classi-
fiers shows certain similarities to the grammatical distinction between
alienable and inalienable possession. Nichols (1992: 134–5) considers sys-
tems with an opposition between alienable and inalienable possession as a
subtype of 'non-agreeing' classification.
The similarities between alienable and inalienable possession and rela-
tional classifiers can be summarized as follows.
(i) All languages with relational classifiers distinguish alienable and
inalienable possession. Relational classifiers are restricted to alienably pos-
sessed nouns.
(ii) In languages with a grammatical distinction between alienable and
inalienable possession, some nouns can be either alienably or inalienably
possessed; there may be a semantic difference. Relational classifiers in these
languages are used in a way similar to the alienable/inalienable distinction.
The distinction between alienable and inalienable possession can be used
in Kilivila (Oceanic: Senft 1986: 49–54) to indicate different ways in which
nouns are possessed (see §5.3.2). Here doba 'grass skirt' is treated as
inalienably possessed in 5.39, and then it is understood as a piece of
clothing.
5.39. doba-gu
grass, skirt-my
'my grass skirt' (when wearing it)
It is understood as a piece of cloth in 5.40, where it is alienably
possessed.
5.40. ula doba
my:REL.CL:SERIES.III grass.skirt
'my grass skirt' (my grass-skirt material)
Similarly, 5.41, from Manam (Oceanic: Lichtenberk 1983a: 158), shows
head as an inalienably possessed body part. Examples 5.42 and 5.43 show
the same noun as alienably possessed with different relational classifiers.
138 Classifiers
5.41. parjana-gu
head-my
'my head' (body part; inalienably possessed)
5.42. parjana ?ana-gu
head CLASS:ALIM-my
'my head' (e.g. fish head: alimentary possession)
5.43. parjana ne-gu
head CLASS:GENERAL-my
'my head' (the one I found: general possession)
(iii) In some languages with a grammatical distinction between alien-
able and inalienable possession, but without relational classifiers, some
body parts can be marked with either type of possession with a
difference in meaning. In Tariana, kare means 'breath, heart' when it
is inalienably possessed (nu-kare 'my breath, my heart'), and 'wind'
when alienably possessed. In Paumari (Arawa: Aikhenvald MS; Shirley
Chapman p.c.) bodi means 'mouthhole' if inalienably possessed, and
'hole, lair (of an animal)' if alienably possessed. In Boumaa Fijian yaca
means 'name' when inalienably possessed and 'namesake' when alienably
possessed.
However, the analogy between relational classifiers and alienable vs.
inalienable possession is not complete. The important difference
between relational classifiers and the distinction between alienable and
inalienable possession is that relational classifiers are usually semantically
transparent; the distinction between alienable and inalienable possession is
often much more grammaticalized. For instance, in Paumari (Arawa)
almost all the kinship nouns and body parts are inalienably possessed;
but there are three body parts and two kinship nouns which are alienably
possessed.
Another difference is that relational classifiers are capable of specifying
various distinct ways of handling an object (not just possessive relations).
This obviously correlates with properties of the possessed noun, and it
results in an interaction between possessed and relational classifiers (see
§5.5.1).
Languages which do not have any grammatical opposition between
alienable and inalienable possession may have other means of marking a
'closer' possession. 'Closer' possession is claimed to be marked by posses-
sive suffixes in Hebrew (according to Berman 1978): sifri 'my book; the one
I have written'; with a 'less close' possession being marked by possessive
pronouns: sefer feli 'my book; the one I own'. So-called possessive adjec-
tives in Russian are used to express a 'closer' possession than possessive
genitives (see the final part of §5.4).
 Classifiers in Possessive Constructions 139
5.4. Possessor classifiers

Possessor classifiers are used to categorize the possessor in a possessive

construction. They are extremely rare in the languages of the world.9
Possessor classifiers are found in Daw (Martins 1994) and, possibly, in
Hupda (Moore and Franklin 1979), languages of the small Maku family
spoken on the Brazil/Colombia border.
Daw distinguishes inalienably and alienably possessed nouns. The first
group includes body parts, parts of plants and some kinship terms (Mar-
tins 1994: 46). Inalienable possession is expressed by simple juxtaposition
of nouns, as shown in 5.44.
5.44. 'yam dum
tail dog
'dog's tail'
Constructions with alienably possessed nouns require classifiers. There
are two classifiers: -dee' is used for inanimate possessor, and -ej for animate
possessor. Both morphemes are clitics (Martins 1994: 138–41).
5.45. yud daw tog-ej
clothing human daughter-CL:ANIMATE,POSSESSOR
The clothing is girl's; the girl's clothing'
5.46. yak kaw-wa'-dee'
manioc garden-up-CL:INANIMATE.POSSESSOR
'manioc of a garden'
Daw is unusual in that an inanimate possessor can be used in an alien-
ably possessed construction, as in 5.46. In many languages with a gram-
matical distinction between alienable and inalienable possession, the
possessor in an alienable construction has to be animate.
In Nasioi (Papuan, Bougainville: Hurd 1977), a language with multiple
classifiers, classifiers can categorize the possessor in one subtype of pos-
sessive constructions: when the possessor is not a personal pronoun.
According to Hurd (1977: 138), 'these possessives are the inverse of the
possessive pronouns ... in that the [classifiers] of the latter stand for that
which is possessed whereas the [classifiers] of the inverse possessives stand
for the possessor'.
These constructions, in which classifiers refer to the possessor, and not to
the possessed noun have a special possession marker -po'-. Example 5.47
shows a possessive construction in which the classifier, nono 'village', refers
to the possessor, and so can be interpreted as a possessor classifier. The
9
It will be shown in Chapter 9 that the agreement of possessed nouns with the possessor is
not an instance of possessor classifiers.
140 Classifiers
same set of morphemes are used in the function of possessed classifiers, e.g.
-va 'house' as a possessed classifier in 5.21 above.
5.47. niikanamono toire' mmau'-po'-na-nono
our.village children many-poss-DS-CL:VILLAGE
'Our village has many children.'
Categorization of possessor10 in possessive constructions is found in
other, non-classifying languages. Possessive marker 's in English is pre-
dominantly used with a human or at least animate possessor (e.g. man's leg,
but 1 table's leg). In Russian, possessive adjectives in -ov (masculine pos-
sessor) and -in (feminine possessor) can be used only if the possessor is
animate. One can say otc-ov-o kreslo (father-POSS.MASC.ADJ-NEUT.SG.NOM
armchair+NEUT) 'father's armchair' (with a human possessor), or kosk-
in-dom(cat(FEM)-(POSS.FEM.ADJ-MASC.SG.NOMhouse+MASc) 'cat's house'
(animate possessor). In contrast, *dom-ov-a krysa (house(MASc)-poss.
MASC.ADJ-FEM.SG.NOM roof+FEM) '*house's roof'?, with an inanimate pos-
sessor, is unacceptable.11 Possessive adjectives show gender agreement with
the head noun (as all adjectives do in Russian). Boumaa Fijian (Dixon
1988: 120) distinguishes several possessive constructions depending on the
type of possessor. All these techniques have functional similarities with
possessor classifiers.

5.5. Interaction of possessed and relational classifiers

Possessed and relational classifiers can interact in two ways. Relational
classifiers may combine with possessed classifiers to create a single inte-
grated system which marks both kinds of categorization as shown in §5.5.1.
Languages which have relational and possessed classifiers as independent
systems are considered in §5.5.2.

5.5.1. Integrating relational and possessive classifiers

Largeish systems of classifiers in possessive constructions often character-
ize both the way an item can be possessed or handled and the item itself.
This is the case in a number of Micronesian languages. Puluwat (Elbert
10
Macushi (Carlson and Payne 1989) appears to have a possessor classifier: pi- prefix,
which is reported to indicate items which can be 'owned by animate beings'. Mam (Mayan:
England 1983: 68) has a possession marker which is restricted to nouns possessible by humans.
This can also be interpreted as a kind of possessor classification.
11
An analytic possessive construction would have to be used: krysa dom-a (roof(FEM)
house(MASc)-GN) 'a roof of the house'. Note that analytic possessive constructions are equally
acceptable as variants for 'father's armchair' and 'cat's house' respectively (kreslo otca, dom
koski), the differences being pragmatic.
 Classifiers in Possessive Constructions 141
1974: 59-60) has several dozen classifiers, mostly derived from verbal roots.
They are a closed class, with the following semantic parameters: general/
default classifier; age with respect to the possessor, kinship; edible/drink-
able, preparation of food; function: ornaments, loincloth, etc. A selection is
given in Table 5.4. Similarly to Kilivila (5.36-8), nouns in Puluwat can be
assigned different classifiers depending on the ways in which they can be
handled. Generic classifier yd- is used in ya-an (CL:GEN-ATT) Pen wuur
'bananas raised, owned, or given by Ben'. Classifier wor 'raw food, meat'
is used in wor-dy (CL:RAW.FOOD-ATT) Pen wuur 'raw bananas eaten by Ben',
and classifier yan 'cooked food' is shown in yan-an (CL:COOKED.FOOD-ATT)
Pen wuur 'cooked bananas eaten by Ben' (Elbert 1974: 60-1).

Table 5.4. Sample of classifiers in Puluwat

Classifier Semantics
na- General classifier for every noun excluding some artefacts
ya- General classifier for artefacts not taking na-
haam- Senior persons
pwi- Classificatory sibling of ego's sex
wor- Raw food, meat
wuniim- Beverages, tobacco (things that are not chewed?)
yan- Cooked food, coconuts
kiy- Pandanus mat
la- Bracelets
lim Sharp tool
mengaak Clothes
paar Hat, umbrella
wa- Canoe, vehicle
wook Cane, club, stick, spear
yanuk- Ropes
ydpel Loincloth

Another Micronesian language, Truquese (Benton 1968: 123 ff.), also has
a large system of classifiers which refer to the ways nouns can be possessed,
or handled, and also to the physical properties of referents. Since some
classifiers are of the repeater type, they form a quasi-open class. Example
5.48 illustrates the use of a classifier wunuma 'potentially drinkable' which
relates to the possessed noun, 'water' as something drinkable; it is also
similar to relational classifiers such as the ones illustrated for Fijian in
Table 5.3 in that it specifies the way in which a referent can be handled.
Example 5.49 illustrates the use of repeater kuusa 'blanket' in the same
slot in a possessive construction (Benton 1968: 124-5).
142 Classifiers
5.48. wumima-yi we kkoniki
CL:DRINKABLE-my DEM Water
'my water'
5.49. kuusa-yi we kuusa
CL:BLANKET-my DEM blanket
'my blanket'
Tinrin, an Austronesian language of New Caledonia (Osumi 1996: 438),
has seven classifiers used in possessive constructions which combine prop-
erties of relational and possessed classifiers. They are: e- 'starches, to be
eaten', ere- 'fruit, to be eaten', hwee- 'meat or eggs, to be eaten', odho-
'things to be drunk', hwiie- 'things to be chewed', ee- 'plants growing on
his/her land, to be planted', hee- 'belongings'.
Palikur (North Arawak; Aikhenvald and Green 1998) has a small set of
possessed classifiers. Unlike other classifiers and genders, not all nouns in
the language require a possessive classifier. Their use is restricted to alien-
ably possessed referents of nouns which cannot take possessive affixes.
Possessed classifiers are in a generic-specific relationship with the noun
they refer to (similarly to Apalai and Macushi: 5.7-9). Referents are
classified depending on their functions, or the ways in which they can be
handled: fruit can be eaten, or planted; animals can be domesticated, or
caught for food. For instance, -pig 'pet' is used with domesticated animals,
as in gi-pig pewru (3M-pet dog) 'his dog'; gi-pig mutom 'his sheep'; -win is
used with animals that are caught to eat, e.g. nu-win arudiki (1SG-catch
tapir) 'my catch-tapir' (the tapir I caught); and -kamkayh 'child' is used
with children, e.g. nu-kamkayh awayg (1SG-child man) 'my son'. The same
noun can be used with different classifiers depending on the way it is going
to be treated, e.g. pi-mana uwas (2SG-food orange) 'your orange' (the
orange you eat), n-amutra uwas (ISG-plant orange) 'my plant-orange'
(the orange I plant).

5.5.2. Languages with two types of classifier in possessive constructions

Languages which have two kinds of classifier in possessive constructions
are rare. An example of a language with separate sets of possessed and
relational classifiers is Baniwa of Icana (North Arawak: Aikhenvald
1994a).
In Baniwa, a multiple classifier language (§9.3), classifiers are used in
possessive constructions independently of possession type; they are very
similar to those in Tariana exemplified in 5.16-17. Possessed classifiers in
Baniwa are mostly used predicatively. The head is often omitted, as in
5.50.
 Classifiers in Possessive Constructions 143
5.50. (nu-hwida) nu-dza-da
(ISG-head) 1SG-POSS-CUROUND
'The head (lit. 'my head') is mine' or 'it is my (own) head'
Baniwa also has relational classifiers which are used only with alienably
possessed nouns. These classifiers have no other classifying functions. In
the Hohodene dialect of Baniwa, alienably possessed nouns with animate
reference can take either suffix -ni or suffix -te, depending on the type of
possessive relation. In 5.51 inu 'dog' has a fairly intimate relationship with
the possessor; it takes the -ni suffix.
5.51. n u - n u - n i
1 SG-dog-poss.1
'my dog (the one I brought up)'
In 5.52, nu 'dog' is less close to the possessor, and it takes the -te suffix.
5.52. n u - n u - t e
1SG-poss.2
'my dog (the one I found)'
The opposition between relational classifiers is lexicalized in other dia-
lects of Baniwa, e.g. Siuci and Kurripaco; every inalienably possessed noun
occurs with just one possessive suffix.12
Another example of a language which probably has two independent sets
of classifiers in possessive constructions is Cahuilla (Takic branch of Uto-
Aztecan: Seiler 1983: 61; Seiler 1977: 299-305). Classifiers in Cahuilla fall
into two types. There is a set called 'temporary' ones; these relate to the
way an item is 'handled', and are similar to relational classifiers (as
illustrated above for Oceanic languages). Certain items can occur with
different classifiers, depending on the way they are treated. See 5.53–4
(Seiler 1983: 37).
5.53. ne-?ay-?a menikis
1SG-pluck-ABSTRACT mesquite. beans
'my (fresh) mesquite beans (to be plucked from the tree)'

12
Possibly, an opposition of this kind goes back to Proto-Arawak. David Payne (1991a)
reconstructs the Proto-Arawak possessive markers -ni, -te, -te, labelling them as 'noun classes'.
This opposition is blurred in the majority of modern languages, but one cannot exclude the
possibility that it might go back to some sort of relational classifiers. A similar situation has
been discussed by Facundes (1994) for Apurina (Pre-Andine Arawak). In some North Arawak
languages, e.g., Warekena, the two markers of alienable possession are reinterpreted as
proximate vs. remote (Aikhenvald 1998b). Other Arawak languages of the Upper Rio Negro
do not use possessive suffixes as relational classifiers; the suffixes are used with different
groups of nouns.
144 Classifiers
5.54. ne-ci-?a menikis
1SG-pick.up-ABSTRACT mesquite.beans
'my mesquite beans (to be picked from the ground)'
Another set of classifiers refers to 'inherent' properties of a possessed
nouns. Such a classifier, 'pet' for dog, is illustrated in 5.55. Similar pos-
sessed classifiers in other Uto-Aztecan languages were discussed in §5.2
(examples 5.1–2).
5.55. ne-?as ?awal
1SG-CL:PET dog
'my dog'
The system of relational classifiers in Cahuilla is considered a recent
innovation by Langacker (1977: 91).

5.6. Contrasting classifiers in possessive constructions

The main difference between relational classifiers and possessed classifiers

is that possessed classifiers characterize the possessed item while relational
classifiers characterize the possessive relationship. Thus, relational classi-
fiers do not constitute a noun categorization device in the way possessed
classifiers do. Often, however, the way a noun can be possessed, or handled,
depends on its inherent properties (e.g. a liquid is drinkable, and not
chewable). This is why a classifier system may integrate the semantics of
relational and of possessed classifiers.
Further differences between relational and possessed classifiers are:
(i) Languages with relational classifiers always have a grammatical distinc-
tion between alienable and inalienable possession. This is not always so for
languages with possessed classifiers.
(ii) Relational classifiers are necessarily integrated into the system for the
marking of possession; they are restricted to alienably possessed nouns.
This is not necessarily so for possessed classifiers.
(iii) Possessed classifiers which can be used independently of possession
type and integrated systems of possessed and relational classifiers are
widespread in multiple classifier languages (e.g. Hmong, and other Miao
languages, according to Bisang 1993, as well as the Papuan languages of
Bougainville, and South American languages). This is not true of mor-
phemes which are used just as relational classifiers.
(iv) Relational classifiers never occur in other classifier environments in
multiple classifier languages (see Chapter 9).
 Classifiers in Possessive Constructions 145
We have also seen that relational and possessed classifiers can coexist as
distinct types in one language. Differences between relational and pos-
sessed classifiers are summarized in Table 5.5.

TABLE 5.5. Differences between possessed and relational classifiers

Properties Relational classifiers Possessed classifiers
Which component of a Type of possessive Possessed noun
possessive construction is relation
categorized?
Link with alienable/ Used only with alienable Used independently of
inalienable possession possession possession type
Use in multiple classifier No Yes
languages

Relational classifiers can be independent lexemes; if not, they may

attach to the possessive marker or the noun itself (Baniwa). Possessed
classifiers can attach to the possessive marker, or they may be independent
lexemes (Hmong). No example has been found so far of a language in
which possessed classifiers attach to the possessed noun.
Possessed classifiers are more frequent across the languages of the world
than relational classifiers. Indeed, relational classifiers are almost restricted
to one genetic group—Oceanic languages, with a very few examples in
South America. Possessor classifiers are even rarer.
Inventories of possessed classifiers vary—from smallish, with two to
three terms, to largeish, as in South American languages. Inventories of
relational classifiers are small; they categorize the way in which items are
typically handled or possessed. In languages where relational classifiers are
also used as possessed classifiers, such as in Puluwat or Truquese, the
inventory can be large. The number of possessor classifiers is always very
small; the distinction is between animates and inanimates, or humans and
non-humans.
These differences between the three types of classifier in a possessive
construction are summarized in Table 5.6.
The differences between the three types of classifier in possessive con-
structions, in frequency and in size of inventory, are not accidental. They
relate to the semantics of these classifiers, and to the type of constituent
they categorize.
Possessed classifiers categorize possessed nouns in terms of their shape,
size, animacy, and other inherent properties; objects do vary a lot accord-
ing to these properties, and so possessed classifiers would be expected to be
rather frequent, and to come in larger inventories.
146 Classifiers
TABLE 5.6. Relational, possessed, and possessor classifier: a comparison
Type of classifier Frequency in Size of Semantics
languages inventory
Possessor classifiers Rare Small Animacy/humanness
Relational classifiers Medium Small The way a noun can be
possessed
Relational/possessed Medium May be large The way a noun can be
classifiers possessed, animacy,
shape, size, etc.
Possessed classifiers Common May be large Animacy, shape, size, etc.

Relational classifiers categorize kinds of institutionalized, culturally

relevant relationships between the possessor and the possessed, and these
appear to allow fewer possibilities; this is why inventories of relational
classifiers are smaller than those of possessed classifiers. Languages outside
Oceania and a few languages in South America tend only to use distinc-
tions such as alienable vs. inalienable possession, and kinship possession vs.
possession of other kinds.
Inherent properties of possessed nouns do correlate with the ways in
which they can be handled, or possessed; this is why possessed and rela-
tional classifiers can be integrated into one system. The inventory of these
morphemes tends to be largeish, for the same reason as the inventory of
possessed classifiers: inherent properties of nouns allow much more varia-
bility in categorization.
Finally, the rarity of possessor classifiers in the languages of the world
may be due to the fact that the nature of the possessor seems to allow
little variation. A prototypical possessor is animate, or at least human
(cf. Seiler 1986; and especially Heine 1997a: 5, 39; Taylor 1989). In
many languages inanimate nouns cannot be used as possessors in con-
structions with alienable possession13 (a part-whole structure is used
instead). Thus, there is little need to categorize the possessor; and if
such categorization takes place, the only distinction that appears to be
likely to occur in a classifier construction is that of animate/inanimate and/
or human/non-human.

13
It is true that in many languages the possessed noun is the head in the possessive
construction. Classifiers would be expected to characterize the head of the construction, as
happens in modifier constructions. Possessor classifiers which characterize the possessor
would be expected in languages where the possessor has at least some properties of the
head in possessive constructions. However, this is not necessarily so. In Paumari possessor is
the head of constructions with inalienable possession; Paumari does not have any possessor
classifiers.
 Classifiers in Possessive Constructions 147
5.7. Distribution of classifiers in possessive constructions in
the languages of the world

Classifiers in possessive constructions are rarer across the world's lan-

guages than noun classes or numeral classifiers—see Map 4 below.
POSSESSED CLASSIFIERS are found in some North American Indian
languages (Yuman, Uto-Aztecan), a number of South American Indian
languages (Nadeb, from the Maku family; Carib, Tupi-Guarani, Je, some
North Arawak and some Guaicuruan languages), and in 'Dongo-ko (Mba,
Ubangi; Niger-Congo). Classifiers are used in possessive constructions in
some multiple classifier languages, e.g. Hmong and other Miao-Yao lan-
guages spoken in Northern China and Indochina, and Papuan languages
of Central and Southern Bougainville (Kim Blewett, p.c.).
RELATIONAL CLASSIFIERS are found in Oceanic languages and in Micro-
nesian languages (where they often coexist with numeral classifiers), and in
two South American languages—Kipea (Kipea-Kariri family; extinct) and
Baniwa of Icana (North Arawak). Reduced sets of relational classifiers are
found in Austronesian languages spoken in Papua New Guinea (e.g. Takia:
Bruce Waters, p.c., Missima: Bill Callister, p.c., and Gapapaiwa: Catherine
McGuckin, p.c.).
POSSESSOR CLASSIFIERS are found only in Daw and, possibly, Hupda
(Maku languages of Northwest Amazonia).
Complicated systems of classifiers in possessive constructions which
categorize possessive relationship, possessed noun, and possessor, are
reported for Reef-Santa Cruzan languages spoken in the Solomons (Ste-
phen Wurm, p.c.).
No special classifiers in possessive constructions are found in Australian
or Eurasian languages.
 A relational classifiers
O possessed classifiers
• possessor classifiers

MAP 4. Distribution of classifiers in possessive constructions in the languages of the world

 6 Verbal Classifiers

6.1. Properties of verbal classifiers

Verbal classifiers appear on the verb, categorizing the referent of its argu-
ment in terms of its shape, consistency, size, structure, position, and ani-
macy (cf. example 1.7, from Waris, a Papuan language). Verbal classifiers
always refer to a predicate argument (usually, S in an intransitive or O in a
transitive clause) and can cooccur with it. Their choice is predominantly
semantic. Every noun in a language does not necessarily require a verbal
classifier; some nouns may be associated with more than one classifier.
The inventory of verbal classifiers varies from two to several score. The
choice of verbal classifiers is based on lexical selection rather than gram-
matical agreement (see §2.4; cf. Rushforth 1991: 255). Verbal classifiers are
thus a subclass of non-agreeing noun categorization devices. The use of
verbal classifiers is often limited to certain semantic groups of verbs.
The ways verbal classifiers are realized are dealt with in §6.2. Syntactic
functions of arguments categorized by verbal classifiers are considered in
§6.3. Some languages have more than one kind of verbal classifier—see
§6.4. The distribution of verbal classifiers in the languages of the world is
presented in §6.5.

6.2. Realization of verbal classifiers

Verbal classifiers come in three forms. Classificatory noun incorporation,
whereby a noun is incorporated into a verb to categorize an extra-predicate
argument, is discussed in §6.2.1.Verbal classifiers affixed to the verb are
described in §6.2.2. Suppletive 'classificatory verbs' are considered in
§6.2.3.
Affixed verbal classifiers often develop from classificatory noun incor-
poration. Sometimes the two coexist, reflecting different stages of gram-
maticalization. Classificatory noun incorporation, affixed verbal classifiers,
and how they can interact, are discussed in §6.2.4.
In many languages, verbal classifiers are optional, and are determined by
the discourse function of the extra-predicate noun. Verbal classifiers can be
used to maintain reference to the noun within a narrative. Participants can
be reintroduced with the help of verbal classifiers in Papuan languages
(Merlan et al. 1997). In South American Indian languages, verbal classifiers
150 Classifiers
are often used anaphorically (see Aikhenvald 1996b, for examples from the
Arawak language family; also see Aikhenvald and Green 1998); see §12.1.3.

6.2.1. Classificatory noun incorporation

Classificatory noun incorporation is the incorporating of a noun into the
verb to characterize an external argument, usually, in S or O function. This
phenomenon was first recognized by Mithun (1984: her Type 4). There is
frequently a relationship of generic-specific between the incorporated NP
and the 'external' NP which accompanies it (Mithun 1984: 863; see Dixon
1980: 436-7; forthcoming). This relation is reminiscent of noun classifiers,
such as in Yidiny and other Australian languages discussed in Chapter 3.
Otherwise incorporated verbal classifiers characterize the referent of a
noun in terms of its animacy, shape, and consistency.
In some prefixing languages from Northern Australia, a generic noun—
which specifies the corresponding specific noun in S or O function—can be
incorporated into the verb. Mayali (Australian: Evans 1996; forthcoming)
has a closed class of about forty verbal classifiers. Example 6.1 shows a
verbal classifier referring to a noun in O function, and 6.2 shows a verbal
classifier referring to an S.1
6.1. ga-yaw-garrm-e al-daluk
3/3-GEN.CL-.BABY-have-NP cui-woman
'She has a baby girl.'
6.2. ga-rrulk-di an-dubang
3NP-GEN.CL:TREE-stand(Np) CLIII-ironwood.tree
'An ironwood tree is there' (lit. a tree-is there an ironwoodtree)
6.3, from Tiwi, illustrates an incorporated verbal classifier referring to S,
and 6.4 illustrates a classifier referring to O (Jennifer Lee, p.c.).
6.3. warta a-watu-wuji-ngi-mangi-rr-akupuraji yiripuwartas
bush 3sG.MASc-morning-coNT-cv-vcL:WATERs-cv-fall high, tide
'The high tide is falling (lit. water-falling) [exposing the] land (bush).'
6.4. ninkiyi arlitunga nga-ri-kunjingi-kirimi
then skewer 1 PL.AUG-CV-CL:COOKED.FOODo-make
[awarra yinkiti jirraka]
that.masc food wallaby
'Then we put the cooked food (wallaby) on skewers.'
1
Incorporated nouns may be used to categorize themselves, e.g. the classifier rruga 'stone'
in Nunggubuyu (Australian: Sands 1995: 273; Heath 1984: 463), used to refer to the S
argument: ma-rruga-Tju-burra? mana-rruga (3SG+NCLIII-stone-linker-sits NCLIII-stone) 'The
stone sits'.
 Verbal Classifiers 151
Noun incorporation and verbal classifiers can be shown to be distinct
categories. This is the case in Mayali (Evans 1996; forthcoming). This
language has non-productive lexical compounding (Mithun 1984: Type
1), body part incorporation, and also verbal classifiers (called 'generic'
classificatory incorporation by Evans 1996). Lexical compounding (Evans
1996: 72–4) is a non-productive process, and constitutes an obligatory
closed system, since 'stems with lexically incorporated nominals lack unin-
corporated paraphrases' (p. 73). In contrast, body part incorporation and
verbal classifiers ('generic incorporation') are optional: the choice between
incorporated and unincorporated constructions depends on the discourse
status of the constituent. The differences between body part incorporation
and verbal classifiers are as follows.
(i) Verbal classifiers are a closed class of about 40 members the majority of
which are inanimate (the exception being yaw 'child, baby', and, used more
rarely, daluk 'woman' and binij 'man'), while body parts are a semi-open
class.
(ii) Two body parts in a part-whole relationship may be incorporated
together into one verb, as in 6.5, while two classifiers cannot cooccur.
6.5. ngan-garre+mok-bukka-ng
3/1M-calf+sore-show-pp
'He showed me the sore on his calf
Lexical compounding and generic incorporation can cooccur in the same
word, as in 6.5 (Evans forthcoming: 266; example 7-537). Of the incorp-
orated nominals, one, -mim 'fruit, seed of, can be omitted but the other
one, bo- 'liquid', cannot be since it forms a part of a lexical compound.
6.6. an-barnadja ngarri-mim-bo+wo-ni
CL:in-Owenia.vernicosa lA-fruit-water+put(=put.in.water)-pi
'We used to put the fruit of Owenia vernicosa in the water (to poison
the fish)'
Besides the syntactic differences mentioned above, incorporated body
parts terms and incorporated verbal classifiers have a number of semantic
contrasts, e.g. gun-gaj as a body part noun means 'flesh, muscle', and as a
classifier it means 'meat' (Evans 1996: 78).2

2
In other cases, it may be more difficult to distinguish incorporated verbal classifiers from
other cases of incorporation and compounding. In Anindilyakwa, a prefixing Australian
language from Groote Eylandt, about 100 'bound roots' are used as verbal classifiers and
as numeral classifiers. The same morphemes are used to form compounds. There are semantic
and syntactic differences which allow one to distinguish the two processes—see Leeding (1996)
for details.
152 Classifiers
6.2.2. Verbal classifiers as affixes
Verbal classifiers can be realized with prefixes (in Waris: see 1.7, or in
Athabaskan languages discussed in §6.4), or with suffixes (in North
Arawak languages). Verbal classifiers are never expressed with repeaters.
Imonda (Papuan, Waris language family: W. Seiler 1985: 120 ff., 132 ff.;
1986: 197 ff.) has about 100 verbal classifiers; not all verbs take a classifier.
Classifiers characterize a noun, in O or in S function, in terms of its
inherent properties, such as shape. In 6.7 a classifier pot 'fruit which can
be picked from trees' refers to O. There is also a verb pot which means 'pick
fruit from trees'; the classifier pot is derived from this verb (see §13.1,3;
Table 13.6). It appears that verbal classifiers in Imonda developed from the
reanalysis of serial verb constructions.
6.7. sa ka-m pot-ai-h-u
coconut 1SG-GOAL CL:FRUIT-give-RECIPIENT-IMPERATIVE
'Give me the coconut.'
Systems of verbal classifiers as affixes vary in their size and semantics.
Terena (South Arawak; Ekdahl and Butler 1979) has several dozen verbal
classifiers which characterize the S/O argument in terms of its shape, size,
form, and animacy, e.g. -pu'i 'round' in 6.8.
6.8. oye-pu'i-co-ti
COOk-CL:ROUND-THEME-PROGR
'He is cooking (round things).'
In multiple classifier languages, classifiers are often used on verbs (see
§9.1). Munduruku (Tupi; Goncalves 1987: 42) has a multiple classifier
system of over 100 morphemes which refer to shape and form. In 6.9 the
classifier -bo4 'banana-like' refers to S, and in 6.9 it refers to O. The same
morphemes appear suffixed on the nouns (-ba4 'LONG RIGID OBJECT'3 is used
as a suffix with the root a2ko3- 'banana').
6.9. a2ko3-ba4 i3-ba2-dom3
banana-CL:LONG.RIGID 3SG.POSS-CL:LONG.RIGID-Stay + FUT
ko4be3 be3
canoe LOC
'A banana will remain in the canoe.'
6.10. be3kit2kit2 a2ko3-ba4 o'3-su2-ba2-do3bu2xik3
child banana-CL:LONG.RIGID 3sG-poss-CL:LONG.RIGID-find
'A child found a banana.'

Here and in other examples raised numbers indicate tones.

 Verbal Classifiers 153
6.2.3. Suppletive 'classificatory verbs'
Suppletive verbal stems, usually referred to as CLASSIFICATORY VERBS, are
comparatively rare in the languages of the world. The selection of a stem is
conditioned by properties of the referent of the S or O constituent.
Unlike the verbal classifiers discussed in §6.2.2, classificatory verbs can-
not be used in other classifier functions. However, classificatory verbs do
show certain correlations with number and with other verbal inflectional
categories (see §10.1.2).
Classificatory verbs fall into two categories.
(i) They can be used to categorize the S/O argument in terms of its
inherent properties (shape, animacy, etc.; see also §11.2.5).
(ii) They can be used to categorize the S/O argument in terms of its
orientation or stance in space, with its inherent properties.
That classificatory verbs should combine reference to inherent properties
of referents and to their orientation is not surprising. Shape, form, and
other inherent properties of objects correlate with their stance in space.
Certain positions and states are only applicable to objects of certain kinds;
for instance, a tree usually 'stands', and only liquids can 'flow'. This
interaction between the inherent properties of a referent and the spatial
position it typically occupies is reminiscent of the way some languages
combine typical properties of possessed and of relational classifiers in
one system (see §5.5.1).
Classificatory verbs differ from the lexical selection of a verb in terms of
physical properties or the position of an object. Most languages have lexical
items similar to English drink (which implies a liquid O), or chew (which
implies an O of chewable consistency). Unlike these verbs, classificatory verbs
(which typically include verbs of location and handling: see below and
§11.2.5) make consistent paradigmatic distinctions in the choice of semantic
features for their S/O argument throughout the verbal lexicon. In other
words, while English distinguishes liquid and non-liquid objects only for
verbs of drinking, classificatory verbs provide a set of paradigmatic opposi-
tions for the choice of verb sets depending on the physical properties of S/O.
Similarly, posture verbs in many languages tend to occur with objects of a
certain shape. For instance, in Russian, long vertical objects usually 'stand',
and long ones 'lie' (Rakhilina forthcoming); see Borneto (1996), on similar
phenomena in German. However, the correlations between the choice of the
verb and the physical properties of the object are not paradigmatic; these
verbs cannot be considered 'classificatory' (see also §13.1.3).4
4
For some languages more work is needed to determine whether they have classiflcatory
verbs, like Athabaskan, or just lexical pairs, like English (e.g. the Mesoamerican languages
cited by Suarez 1983: 90–1).
154 Classifiers
(A) Classificatory verbs categorizing the S/O argument according to its
inherent properties
Classificatory verbs of this sort are predominantly found in North
American Indian languages, in Tibeto-Burman languages, and in Ika, a
Chibchan language from Northern Colombia (Frank 1990).
According to Carter (1976: 24), suppletive Classificatory verbs in
Athabaskan languages refer to concrete objects, and they describe 'objects
at rest, in motion, being handled, being dropped, or falling'. Similarly, in
Ika Classificatory verbs have to do with the way of handling physical
objects ('put', 'carry', 'drop', etc.) or with their position and location, or
existence ('lie', 'hang', 'fall', etc.). In Cora (Uto-Aztecan: Casad 1996: 246)
Classificatory verbs 'indicate the shape of an object that is being trans-
ported or the manner in which the object is being moved'; transportable
entities are grouped into six classes: (i) long rigid things; (ii) round things;
(iii) flat, squareish rigid things; (iv) long, flexible things; (v) domesticated
animals; (vi) humans.
Suppletive Classificatory verbs in Athabaskan languages go back to
Proto-Eyak-Athabaskan (Krauss 1968: 200); they did not develop from
affixed classifiers. They categorize the S/O constituent in terms of its shape,
form, animacy, number, and consistency (see Chapter 11). Chipewyan
(Carter 1976) distinguishes the following semantic oppositions in classi-
ficatory verb stems: round objects, long or stick-like objects, living beings
(animate or human), containers with contents, fabric-like objects, a col-
lection of objects or rope-like objects, granular mass, dough-like or mud-
like object, piled-up fabric (blankets). Similar semantic oppositions are
also found in Slave (Rice 1989: 779 ff.). Examples 6.11 and 6.12, from
Chiricahua Apache (Hoijer 1945: 14), illustrate the use of Classificatory
verbs.
6.11. ha-n-?ah
out of-2SUB.JMPF-handle.a.round.object:IMPF.MOMENTANEOUS
'you take a round object (out of enclosed space)'
6.12. ha-n-t-cos
out.of-2SUBJ.IMPF-VOICE.MARKER-
handle a. fabric. like, object: IMPF.MOMENTANEOUS
'you take a fabric-like object (out of enclosed space)'
Table 6.1 illustrates Classificatory verb categories from Mescalero
Apache (Rushforth 1991: 253). (The stem given here means 'be located').
In Ojibway (Algonquian: Denny 1979a: 106–7), the semantic character-
ization in Classificatory verbs is similar to Athabaskan: the choice of the
verb is conditioned by the shape of a noun. Examples of Classificatory
verbs in Ojibway are in Table 6.2.
 Verbal Classifiers 155
TABLE 6.1. Mescalero Apache classificatory verb categories
1. -'a 'single, solid, round inanimate object'
2. -ti 'single animate object'
3. -la 'dual objects of any kind; a ropelike object'
4. -ta 'elongated, rigid object; a stick-like object'
5. -1-tsuus 'flexible object; a cloth-like object'
6. -ka 'contents of a shallow, open container; a cup- or dish-like object with its
contents; a rigid container with its contents'
7. -jaash 'plural objects of any kind; uncontained dry and loose or granular
substance, uncontained sand- or flour-like substance; a dry mass
8. -tie 'uncontained wet or damp mass; dough- or mud-like substance'
9. -l-ta 'flexible container with its contents'
10. -'a 'indefinitely shaped single solid object'

TABLE 6.2. Classificatory verbs in Ojibway

Verb Semantics
sak-Tk-inan (extended two-dimensional flexible) to hold on to something sheet-like
sakit-api-ssin (extended one-dimensional flexible be sticking out [string-like object]
kotako-minak-ipiton (non-extended) to roll over something round-like
kotako-minak-isse (non-extended) something round-like rolls over

In Tibeto-Burman languages classificatory existential or locative verbs

typically distinguish animate and inanimate forms if there are only two
of them (as in Idu, with i55 'animate existential verb' and kha55 'inan-
imate existential'—LaPolla 1994: 75). Other languages distinguish up to
seven existential verbs; for instance, Queyu has t iss for animals; t y13
for location in a vessel or enclosed area; Ro 31 , for non-movable objects
i13 for movable objects, lo13 for an object mixed with another object,
ru13 for abstract objects; and t e13, for possession by a person (LaPolla
1994: 75). (Animate/inanimate distinction in existential verbs is also
found in Japanese; this may be an areal typological feature shared with
Tibeto-Burman languages—LaPolla 1994: 76.)
Qiang (LaPolla forthcoming: 120–1) has four existential verbs: for
inanimate referents, zi for animate referents, le for referents located in a
container of some type, and we for immovable referents or referents
inalienably connected to a larger entity. Their use is illustrated in the
following examples. In 6.13, the 'inanimate' existential verb is used to refer
to a book.
6.13. tssudts-mq-ta l z-e-pen
table-top-Loc:ON book-one-CL exist:INANIM
'There is a book on the table.'
156 Classifiers
In 6.14, the animate verb is used to refer to fish.
6.14. ts-Ra RZ zi
water-LOC:IN fish exist:ANIM
'There are fish in the water.'
Different existential verbs can be used with the same referent, with a
difference in meaning. In 6.14 the fish is considered an animate being; while
in 6.15 the situation is presented 'from the point of view of the fish being in
the water' (LaPolla forthcoming: 121):
6.15. ts-Ra RZ le
Water-LOC:IN fish exist:REFERENT.LOCATED.IN.A.CONTAINER
'There are fish in the water.'
Hani (LaPolla 1994: 75) has six classificatory existential verbs: dza33 'gen-
eral existential', dzo55 'an existential for people and animals', bo33 'an exis-
tential for people and their body parts', d31 'an existential for liquids', de31
'an existential for animates', and k31 'existence within a group'.
This phenomenon is widespread in Tibeto-Burman languages; for an
account of this as a recent innovation, see LaPolla (1994: 75).5
In the cases illustrated above classificatory verbs are suppletive and not
analysable synchronically. In Ika (Chibchan, Colombia: Frank 1990: 55)
classificatory verbs (used for location and handling) are partly analysable
(one could argue that this is an intermediary case between classificatory
verbs and affixed verbal classifiers). In Ika, different semantic parameters
interact to yield semantically complicated systems where dimensionality
and shape correlate with consistency, form and directionality (see §11.2.5,
Table 6.3, and Examples 6.16 and 6.17.)

TABLE 6.3. Classificatory verbs in Ika

Long Flat Three- Liquid Holders Upright

dimenional

Exist/1oc gaka pa sa _ — tso

Be in a?-geikua a?-p nkua a?-nikua a?kua a?zu a?nuk
Be up on i-geikua i-pAnkua i-nikua - izu i-nuk
Be on geikua p nkua nikua - - nuk
Put up on igeika i-pan isa idos - itso
Put down gaka (gako) pan (pa) sa dos - tso?s
Put in k gaka k pas k ssa k dos kAZUS k tso?s

5
As pointed out by William Croft, these systems are different from those in Athabaskan
languages in that they classifiy the Ground as well as the Figure, or even the Path (following
the terminology of Talmy 1985).
 Verbal Classifiers 157
6.16. k n gako u
stick put.down:LONG.OBJECT AUX
'Put down the stick!'

6.17. ribru pa u
book put.down:FLAT.OBJECT AUX
'Put down the book!'
Classificatory verbs can be used anaphorically as referent tracking
devices (Rushforth 1991: 255-7). They are also employed to highlight
different meanings of a polysemous noun in a similar way to noun classi-
fiers discussed in §3.2.2 (see examples from Mescalero Apache in Table 6.4).
Different 'arrangements' of tobacco are reflected in the form of the classi-
ficatory stems of the verb with the basic meaning 'give'.

TABLE 6.4. Examples of the use of 'give' in Mescalero

Apache
1. Nat'uhi shdn'aa 'Give me (a plug of) tobacco'
2. Nat'uhi shdnkaa 'Give me (a can, box, pack) of tobacco'
3. Nat'uhi shdnitii 'Give me (a bag) of tobacco'
4. Nat'uhi shdntii 'Give me (a stick) of tobacco'
5. Nat'uhi shdnjaash 'Give me (loose, plural) tobacco'
Source: Rushforth (1991: 254).

Table 6.5 gives an idea of the range of meaning of classificatory verbs in

Koyukon (Northern Athabaskan: Axelrod forthcoming: 5), and how they
can be used to specify the referent of boogee 'flour, dough'.

TABLE 6.5. Classificatory verbs in Koyukon: an example

Example Meaning
boogee le'onh 'a lump of flour is there"
boogee daal'onh 'a canister of flour is there'
boogee daaltonh 'a box of flour is there'
boogee etltonh 'a bag of flour is there'
boogee lekkonh 'a bowl of flour is there'
boogee ledlo 'lumps of flour are there'
boogee etlkoot 'flour is there (stored as provision)'
boogee eietlaakk 'dough is there'
boogee daatenokk 'loose flour is there'

'Non-literal' uses of classificatory verbs can be of two kinds. The use of

classificatory verbs can be extended to apply to abstract nouns as a kind
158 Classifiers
of metaphorical extension; or an unexpected classificatory verb is used, for
instance, the stem for 'single round object' is employed in reference to a
person (Rushforth 1991: 262–4), as a kind of 'pun' (Sapir 1932). This is
reminiscent of the metaphorical use of gender variation (see (C) in §2.4.3);
cf. §12.3.3.
(B) Classificatory verbs characterizing the S/O argument in terms of its
orientation in space and its inherent properties
Suppletive classificatory verbs involve reference to the orientation and
stance of S/O in a number of North American Indian languages, e.g.
Dakota (Siouan: Boas and Deloria 1941: 126; Croft 1994: 157), and
Nevome (Uto-Aztecan: Shaul 1986: 12). The categorization involves refer-
ence to the orientation and stance, standing or lying, and to the animate-
ness of the S/O argument (Table 6.6).

TABLE 6.6. Classificatory verbs in Nevome

Inanimate referent Animate referent
be lying catulvutu voho/vopo
be standing cuhcaltutu cuhcalguguhuca

Complicated systems of suppletive existential verbs are found in

Muskogean languages (also see Haas 1978: 306). In Koasati (Kimball
1991: 452-9) five different positional verbs are used depending on the
shape of the referent. These verbs have suppletive plural forms. The verb
'stand' is used for tall, vertical things, such as posts, trees, and objects with
legs; 'sit' is used for globular objects, such as ball, hill, sun, and heaps of
objects; three different verbs, all glossed as 'lie' are used for large, long,
thin things; for small, long, thin things; and for objects covering a broad
area. In Creek the use of suppletive dual and plural forms of some verbs
correlates with shape and consistency of objects (Haas 1978; Jack Martin,
p.c.; see §10.1.2). According to Haas (1978: 306):
the only categories of classification that have developed are those which distinguish
cloth-like objects from liquids, and these in turn from all other types of objects.
Cloth-like objects are treated as duals when the verb has a distinctive dual form;
otherwise they are treated as plurals. Liquids are treated as plurals. All other types
of objects are treated as singulars, duals or plurals depending upon the actual
number of entities involved.
Classificatory verbs which combine reference to the orientation of the
noun and to its inherent properties are widespread in Papuan languages of
the Engan family (New Guinea Highlands: Foley 1986: 89 ff). Enga itself
(Lang 1975; Foley 1986: 89-91) has seven classificatory verbs, shown in
 Verbal Classifiers 159
Table 6.7. When used as existential predicates they classify a noun (S
argument) in terms of its orientation in space and its inherent properties.
(Their lexical meaning is given in brackets in the first column of the table.)

TABLE 6.7. Classificatory verbs in Enga

Verb Semantics of classified nouns Examples of nouns

katenge ('stand') Referents judged to be tall, men, house, tree
large, strong, powerful,
standing or supporting
pentenge ('sit') Referents judged to be small, woman, possum, pond
squat, horizontal and weak
lyingi ('hang') Referents hanging or wasp, fruit, seed
protruding out of another
object
palenge ('lie inside') Referents internal or worm, heart, sweet potato
subterranean
epenge ('come') Referents which are rain, fur, blood
intermittent, but capable of
growth; or liquid or gas
singe ('lie') Referents which are orifices, ground, eels, mouth
locations, or crawling or
aquatic
mandenge ('carry') Referents of sexual production penis, vagina, testicles

Ku Waru (Waris family, Papuan: Merlan et al. 1997: 75) has a similar,
somewhat smaller system of classificatory existential verbs (Table 6.8).
Unlike Engan, these are not employed as just posture verbs.

TABLE 6.8. Classificatory verbs in Ku Waru

Verb Semantics of nouns
mol Liquids, certain inanimate objects, e.g. shoes when being worn; almost all
living things (including plants)
pe Abstract nouns referring to sources of trouble (e.g. 'grievance', 'property
dispute', 'trouble')
le Inanimate nouns, prototypically nouns referring to wealth objects
angaly Body parts and certain artefacts (e.g. 'house')

When classificatory verbs in Ku Waru are used with human beings,

they indicate only the orientation of the referent, e.g. mol: 'staying', pe:
'lying/sleeping', le: 'lying prostrate/dead', angaly: 'standing'; mol: is also
used as a marker of habitual aspect. They are also developing into aspect
markers.
160 Classifiers
6.2.4. The interaction of the three types of verbal classifier
Classificatory noun incorporation is a frequent source of verbal classifiers
as affixes (also see §13.1.1). According to Mithun (1986: 388), 'all classi-
ficatory stems begin life as nouns', being used with a relatively narrow
scope, and then being used as generics. Grammaticalization of incorp-
orated nouns may further result in the creation of suppletive classificatory
stems.
Unlike verbal classifiers as affixes, classificatory noun incorporation is
not used in other classifier environments in multiple classifier languages; it
is generally restricted to just this grammatical context. Nouns are incorp-
orated into verbs, and the verbs retain the same argument structure. The
incorporated noun is used as a means of categorizing an overtly expressed
argument.
More than half of the 100 verbal classifiers in Munduruku (Tupi:
Goncalves 1987: see examples 6.9 and 6.10 above) coincide in form with
nouns. In Terena (South Arawak: Ekdahl and Butler 1979), some verbal
classifiers come from incorporated nouns which underwent phonological
shortening; these include the classifier -hi- 'leaf, grass' and the root uhi
'grass' in 6.18.
6.18. moyo-hi-ti-rau uhi-ti
dry-CL:LEAF-PROGR-DEM grass-NPOSS
'Grass is dry'
Other classifiers bear no resemblance to nouns—such as the classifier
-pu'i 'round' in 6.8.
The connection between classificatory noun incorporation, verbal clas-
sifiers and nouns can be more complex. One such case is Anindilyakwa
(Australian). In this language 'bound forms' of nouns belonging to
certain semantic groups (such as body parts) are incorporated into the
predicate to signal the S/O argument. 6.19 illustrates an incorporated
body part yakwi 'chest', which refers to the O argument, athalyima 'river'
(Leeding 1989: 347), to categorize the river with respect to its extended,
flat shape.
6.19. wirra+yakwi+yiwi+rni ana athalyima
2NON.SG + CL:CHEST + follow + TNS DEM river
'Follow the river!'
Classifiers in Anindilyakwa are also used as incorporated nouns.
According to Leeding (1996: 201), only a relatively small proportion of
the incorporated roots show any similarity with corresponding free form
nouns. Similarly, in Tiwi (Australian: Osborne 1974: 48-9) most incorp-
orated verbal classifiers are in a suppletive relationship with a noun they
 Verbal Classifiers 161
categorize. This is dramatically different from the case of Terena, where the
connection between some classifiers and free nouns is transparent. In these
cases it is impossible to decide whether classifiers 'begin their life' as
incorporated, or as free nouns.
Many prefixing languages from northern Australia show the incipient
reanalysis of incorporated nouns as classifiers. In Emmi (Australian: Ford
1998) incorporated body parts are often used 'almost' as classifiers, refer-
ring to a particular shape of the referent of an argument. For instance, mart
'belly' is used as a classifier for 'concave-shaped interiors', e.g. dishes and
buckets, while miri 'eye' is used for circular openings (such as doors or
windows).
A complicated system of verbal classifiers (called 'class morphemes') and
classificatory verbs in Cherokee (Iroquoian: Blankenship 1996; Mithun
1986) shows an intermediate stage in the process of evolution from incor-
porated classifiers to suppletive classificatory verbs. Cherokee has at least
forty verbs for which the choice of the form is determined by the nature of
its S/O argument (Blankenship 1996; 1997; Haas 1978). The classes of
referents distinguished are LIVING (or ANIMATE), LIQUID, FLEXIBLE, LONG,
and COMPACT. The following examples illustrate the use of classificatory
verbs (Blankenship 1997: 92).
6.20. Weesa ga-kaa-nee'a
Cat 3SG.A + 3sG.O-CL:LIVING-give:PRES
'She is giving him a cat.'
6.21. Ama ga-neeh-nee'a
water 3sG.A+3saO-CL:LIQUID-give:PRES
'She is giving him water.'
6.22. Ahnawo ga-nvv-nee'a
Shirt 3SG.A + 3sG.O-CL:FLEXIBLE-give:PRES
'She is giving him a shirt.'

6.23. Gansda aa-d-ee'a

Stick 3sG.A + 3SG.O-CL:LONG-give:PRES
'She is giving him a stick.'
6.24. Kwana aa-h-nee'a
peach 3sG.A+3so.O-CL:COMPACT-give:PRES
'She is giving him a peach.'
Examples 6.20–4 show the presence of classifier-like morphemes which
refer to an S/O constituent.6 Synchronically suppletive classificatory verbs
6
Unlike Terena, Mundurukii, or Anindilyakwa, different sets of class morphemes are used
with different groups of verbs in Cherokee (Blankenship 1997: 97).
162 Classifiers
in Cherokee have been shown to result from the depletion and fusion of
verb-incorporated classifiers (Mithun 1984: 884, 392; Blankenship 1996:
97). Cherokee also has non-productive noun incorporation used with verbs
which refer to putting on and changing clothes. Noun incorporation can
change the transitivity of a verb (Mithun's type I); or not; this involves the
advancement of an oblique argument into the direct object position
'vacated' by the incorporated constituent (Mithun's type II: Blankenship
1996: 68; 1997).
The three types of realization for verbal classifiers can be seen as distinct
points on a grammaticalization continuum (see Chapter 13).

6.3. Verbal classifiers and syntactic function of the argument

Verbal classifiers signal the presence of a surface NP. In every language this
NP may be in S and in O function, that is, they operate on an 'absolutive'
basis (Keenan 1984).7 In a few languages, verbal classifiers can also refer to
peripheral arguments. However, no examples have been found of a classi-
ficatory verb stem referring to a peripheral constituent.
In Motuna (Papuan: Southern Bougainville) verbal classifiers can signal
the presence of an S argument (as in 6.25), an O (as in 6.26), or a peripheral
argument (as in 6.27) (Onishi 1994: 175-6).
6.25. . . . hoo koto honna rii-kui-no-wori
ART:MASC Up big be. 3s-IMAG-LINK-CL:ANIMATE
'. . . the elder one (animate) who would be big'
6.26. ong topo inak-i-heeto-no-uru
DEM:MASC Well look.after-3o + 2A-FUT-LINK-CL:HUMAN
'This is one (lit. human male) you will look after well.'

7
No exceptions have been found to Keenan's (1984) generalization concerning the abso-
lutive basis of verbal classifiers. In Motuna, a verbal classifier can refer to the A argument of
the predicate of the participle clause. Onishi (1994: 176) gives one such example: hoo-no jii
eejee nil minno-wah-no-wori (ART:M-COMIT and my.opposite.sex.sibling me follow-PART-UNK-
CLIANIMATE) '. . . and with one (animate) of my brother who was following me (age-wise)'.
He mentions that such examples are extremely rare in Motuna texts. Here, the A argument of
the participle clause ('who was following me') is coreferential with an oblique constituent of
the main clause. Thus, it can be argued that examples of this sort in Motuna are not true
exceptions to Keenan's generalization.
The sentence Tent toire tareuri-ma-0-i bau'uri-ma-0-i (The.FEM children care.for-PRES HAB-
DER.suFF-cL:FEM feed-PRES.PROGR-DER.SUFF-CL:FEM) The lady who cares for the children (is) the
one who feeds them', from Nasioi (Papuan, Bougainville: Hurd 1977: 144) shows agreement of
the predicate in verbal classifier with A function in both main and subordinate clause. This
example could qualify as a counterexception to Keenan's generalization unless the predicate
'the one who feeds them' should be analysed as a deverbal nominalization (Masa Onishi, p.c.).
 Verbal Classifiers 163
6.27. u'kisa hoo kitori hoo kongsi'
long ago ART:MASC children ART:MASC mango
haaro'-ki-no-mori hoo kongsi'
fall + 3S-HAB.PAST-L1NK-CL:SEASON ART:MASC
u'w-a-hee uwi-ki-ng
pick.from.ground-3o+3pCL.A-DEF.FUT go+3pCL.s-HAB.PAsr-MAsc
'Long ago, in the season when mangoes fall, the children went to
pick mangoes from the ground.'
Verbal classifiers in Tarascan (Southwest Mexico isolate: Friedrich 1970:
390) frequently refer to a locative argument. Tarascan has thirty-two verbal
suffixes of 'locative' space, which 'signify the features of a location, often
including its dimension and shape'. See 6.28 (Friedrich 1970: 393).
6.28. inca-hpa-mu-ku-nta-ni
to.enter-Suddenly:ADV-SPATIAL.SUFF:ORIFICE.EDGE-ACTIVE-
RECURRENT-NON. FINITE
'(And then the cantor) suddenly re-entered the building by the door.'

6.4. Combinations of different types of verbal classifier

Languages which have several distinct systems of verbal classifier in com-

plementary distribution are considered in §6.4.1. Languages which have
affixed verbal classifiers and classificatory verbs as distinct systems are
discussed in §6.4.2.
No example has been found of a language which combines classificatory
noun incorporation and affixed verbal classifiers as different systems. An
intermediary stage which could be accounted for by grammaticalization of
incorporated nouns as verbal classifiers was described in §6.2.4 for Cherokee.
No language is known to combine classificatory noun incorporation and
classificatory verb stems.

6.4.1. Different types of verbal classifier in complementary distribution

The only known example of a language with two sets of verbal classifier
affixes in complementary distribution is Palikur (North Arawak: Aikhenvald
and Green 1998; also see §8.2 below). There are two largely overlapping
sets of verbal classifiers. One set is used on stative verbs to refer to the S, or
to the head noun if a stative verb is used as a modifier. The other is used on
transitive verbs, to refer to the O argument; the same set refers to the
derived S of detransitivized passive verbs.8
8
Some multiple classifier languages have two sets of classifier morphemes: a closed set of
affixes and an open class of repeaters. In Tariana both sets are employed similarly to verbal
classifiers. See §9.3.
164 Classifiers
The assignment of verbal classifiers is semantic, and shape-based; there
are no distinctions based on animacy, all animate nouns being treated as
'irregular-shaped'. Verbal classifiers can be used without the overt NP as
referent-tracking devices. Similarly to numeral classifiers, an inanimate
noun can be used with different classifiers depending on the aspect of
the S/O constituent that is involved in the action.
All verbal classifiers are optional. They are used (a) if the corresponding
constituent (S or O) is fully involved in the activity, or displays the full
degree of a 'property'; or (b) the action/state involves the whole surface of
the object.
The forms and semantics of verbal classifiers are given in Diagram 6.1.
If a classifier is used with both stative, and transitive verbs, the first form
given is the one used with a stative verb and the second is the one used
with a transitive verb (surface differences are due to morphophonological
processes). Classifiers used with stative verbs only are marked with an
asterisk.
Verbal classifiers are used on stative verbs of the following semantic
types (following Dixon 1982: 16):
(i) dimension, e.g. pugum 'thick', imu 'tall';
(ii) physical property, e.g. mtibdi 'soft', ivat 'hard', kiki 'smooth', kiyaw
'sharp, abrasive (of cloth)', miyaw 'blunt', barew 'clean; pretty', patauh
'dirty, not pretty', dax 'stained';
(iii) colour, e.g. puhi 'black', sey 'white', duruweh 'red', ayeweye 'blue,
green', kuwikwiye 'yellow'.

DIAGRAM 6.1. Verbal classifiers in Palikur

 Verbal Classifiers 165
Stative verbs of other semantic groups—such as speed, age, difficulty,
qualification, human propensity, and value—do not cooccur with classi-
fiers. Verbal classifiers are only used if the stative verb describes the com-
plete involvement of the S, or of the head of a head-modifier construction.
In 6.29, a classifier is used to indicate the complete blackness of the bird's
feathers. (In the examples below, classifiers are underlined.)
6.29. gu-sipri puhi-pti-ye
3sGF-feather black-V.CL:IRREG-DUR.NF
'Her (bird's) feathers are completely black.'
Verbal classifiers are used with transitive verbs which imply direct
physical contact with the object. These are:
(i) physical actions such as 'grab', 'wash', 'dry', 'hit', 'rub'; or
(ii) positional verbs such as 'hang', 'stand', or 'lie'.
Classifiers are employed with telic verbs such as 'look' (as opposed to
'see'). Accordingly, they are not used with verbs denoting mental processes,
such as 'think' or 'remember', or with verbs which do not involve direct
physical contact with the object, such as 'hear', or 'say'. The restriction
on classifier use depending on the semantics of verbs is reminiscent of
Athabaskan languages and Ika (see §6.2.3 and §11.2.5).
Verbal classifiers are only used if the object does not have to be, of
necessity, completely involved in the action. They are not used with the
verb 'kill', since it always involves the whole object—'non-complete' killing
is not killing at all. Verbal classifiers may be used to refer to an O, as in
6.30, or to the derived S of a passive, as in 6.31.
6.30. yak-pit-apa-e-gu-kis nikwe
sting-v.CL:IRREG-TOTAL-coMPL-3F-PL therefore
'So (the killer bees) stung them all over (their bodies).'
6.31. gu-apitiw wanak-pita-ka a-kak mawru
3F-head tie-V.CL:IRREG-PASS 3N-with cotton
'I see [that] the head [of the rattle] is tied with cotton.'
The origin of most verbal classifiers is unknown. However, at least
three come from parts of the body, or parts of a plant: -kig 'pointed' is
related to -kig 'nose'; -pewal-peru 'branch-like' is related to -peru 'branch'
(cf. a-peru 'on a branch'), and -kat 'trunk' is related to akat 'trunk (of a
tree)'. Palikur also has body part incorporation. Note that incorporated
body parts cannot cooccur with verbal classifiers since they go in the
same slot (see Green and Green 1972, on verb structure in Palikur). Incor-
porated body parts and verbal classifiers differ in their morphosyntactic
166 Classifiers
behaviour and in their semantics (see Aikhenvald and Green 1998: §5, for
details).9
Waris (Waris family, Papuan: Foley 1986: 90–1; Brown 1981) has affixed
verbal classifiers and also 'stance' classificatory verbs (see (B) in §6.2.3 on
Ku Waru) which are used to categorize the argument as to its orientation in
space, in combination with its inherent properties.
Prefixed verbal classifiers in Waris are similar to classifiers in Imonda
(see example 6.7) both in semantics and in origin: they derive historically
from compounded verbs. The choice of verbal classifiers is dictated by such
properties of the direct object as consistency, shape, function, arrangement,
and measure (Brown 1981: 101-3). There are two 'unique' classifiers: see
Table 6.9, and (B) in §11.2.5.

TABLE 6.9. Verbal classifiers in Waris

A. Consistency D. Arrangement
Soft and pliable: mwan- Objects inside a container: vela-
Leaf-like with soft stem: le- Container: ev-
Leaf-like with hard stem: pola- Bundle: selvo-
E. Form, nature and measures
B. Shape Pieces cut from longer lengths:
Spherical (balls, fruit): put- tuvv-
„ . , Cut lengths of vine: kov-
C. Function ana nature
Cooked food: ninge- F. Unique classifiers
Food removed from the fire ready to eat: vet- For bunch of betelnut: sengeit-
Dead game: vend- For pitpit fruit: si-

The classificatory verb system for existential verbs is similar to that of

Enga (cf. Table 6.7 above, and Kamoro: Table 11.12 below). Classificatory
verbs (see Table 6.10) refer to the position of the S argument, and to its
shape. The two systems of verbal classification in Waris have a different
origin; prefixed classifiers are a later development in the language than
classificatory verbs (Brown 1981; cf. §13.1.3).

9
While the use of a classifier is linked, basically, to the completeness of involvement of the
O/S in the action, the use of an incorporated body part implies the lack of individuation of the
noun which is in the O/S function, and its non-focused status.
Another important difference in behaviour between incorporated body parts and verbal
classifiers concerns possibilities of lexicalization of the former. Only incorporated body parts
can get lexicalized with certain verbs. That is, they may result in the creation of unique
idiomatic expressions in which the meaning of the whole cannot be determined from the
meanings of the parts. This happens both with transitive verbs, e.g. kamax-duka (grab-CHEST
+REFL) 'He had a quick snack' (lit. he grabbed his own chest), and with stative verbs, e.g. nah
barew-wok(1SGclean-HAND) 'I am poor, destitute' (lit. I am clean-handed)'. Nothing of this sort
ever happens with verbal classifiers.
 Verbal Classifiers 167
TABLE 6.10. Classificatory verbs in Waris
Verb Semantics Nouns it occurs with
lohv be standing Vertical or standing things: man, tree, garden,
dog, pig, fish in water, sugar cane, sun, sky
av be sitting Small roundish things: woman, small animal,
insect, taro in garden, bunch of betelnut on
tree
liv be lying prone Water, liquids, yam in garden, snake
dihilv lie or sit Axe, road, tractor
nalohv lie or sit (orderly) Firewood
diav lie or sit (disorderly) Fallen trees in newly-cut garden
endv be hanging Fruit, rattan, peanuts on stem underneath the
earth
vilv lie crumpled or folded Net bag, towel

Unlike Athabaskan languages (see §6.4.2), the two kinds of verbal classifier
are in a complementary distribution with respect to the argument they refer
to, one set referring to the O argument and the other to the S argument.

6.4.2. Distinct systems of verbal classifiers

A few North Athabaskan languages have two kinds of affixed verbal
classifier, and they also have classificatory verbs (§6.2.3). A typical system
is found in Koyukon10 (Axelrod forthcoming: 4; Thompson 1993: 315-16);
one set of verbal classifiers in Koyukon is shown in Table 6.11. (Note that
Athabaskan linguistic tradition calls verbal classifiers 'genders'; the term
'classifier' is, misleadingly, used for voice-like markers.)

TABLE 6.11. Affixed verbal classifiers in Koyukon

Classifier Meaning
ne- Round things: beads, berries, face, eyes, strings and ropes
de- Wood, plants, rigid containers, stiff clothing, language
dene- Round, heavy objects: animal heads, cabbages, apples, rocks; long
cylindrical objects: pipe, bridge, pencil
hude- Weather
0- People, animals

10
The system found in Koyukon is fairly typical of the majority of Athabaskan languages,
though the details vary. The two most widespread verbal classifiers are d 'long, slender objects'
and n 'round objects'. Verbal classifiers are highly productive in Koyukon, Dena'ina, Ahtna,
Tanana, and Carrier (Thompson 1993), but are also found in others including Navajo and
Slave (Keren Rice, p.c.). See Kari (1990: 34) on the semantics of verbal classifiers ('genders') in
Ahtna.
168 Classifiers
These verbal classifier prefixes are used to refer to the S or O argument of
the verb. Another kind of verbal classifier, termed the 'areal' prefix, hu-, is
used to refer to extended objects, places, events or abstractions. The verbal
classifier prefixes listed in Table 6.11, and the areal prefix, share a number
of morphosyntactic properties. They are used to categorize the S/O argu-
ment in terms of its shape.
A verbal classifier is shown to refer to the S argument in 6.32, and to the
O argument in 6.33. The areal prefix is used to refer to the S argument in
6.34. Examples are from Thompson (1993: 316-17).
6.32. tl'oel n-aal'onh
rope CL-be.there
'A rope (ne- classifier) is there.'

6.33. tlool n-aan-s-'onh

rope CL-PREF-1SG-arrive. carrying
'I arrived carrying a rope (ne- classifier).'

6.34. yeh h-ool'onh

house AREAL-be.there
'A house (hu- areal prefix) is there.'
The areal prefix differs from other verbal classifiers in the following ways
(Thompson 1993: 317-18).
(i) Unlike other classifiers, the areal prefix may be used in a further
morphosyntactic context: it can cross-reference the argument of a loca-
tional postposition, as shown in 6.35.
6.35. Fairbanks ba-ts'e taalyo
Fairbanks AREAL-to went
S/he went to Fairbanks.'
(ii) The areal prefix occupies a slot in the verb different from that of
classifiers (Thompson 1993: 319-21);11 it behaves like prefixes which belong
to the pronominal slot; the other classifiers occupy a quite different 'qua-
lifier' slot (see further discussion in Thompson 1993).
(iii) The areal prefix has other functions, e.g. that of impersonal and
proximate subject. Verbal classifiers do not have such functions.
(iv) In some Athabaskan languages with verbal classifiers these cannot
cooccur with each other, but they can cooccur with the areal prefix; this
happens in Carrier (Morice 1932: 141–4; Bill Poser, p.c.).

See Kari (1989) on prefix positions in Athabaskan languages.

 Verbal Classifiers 169
(v) Some Athabaskan languages do not have verbal classifiers other than
the areal prefix, e.g. Gwich'in, Holikachuk (Thompson 1993: 318).12
A diachronic study of Athabaskan suggests different origins for verbal
classifiers de- and ne-, and for the areal prefix. According to Thompson
(1993: 332), 'the history of the areal prefix suggests that it has always been
an inflectional and productive mopheme'. According to Jeff Leer (p.c.), the
verbal classifiers de- and ne- come from incorporated nouns (i.e. the type
discussed in §6.2.1) in Proto-Eyak-Athabaskan, and these later grammati-
calized as classifiers.
Whatever the whole story is, the two types of verbal classifier in
Athabaskan languages have different synchronic functions and different
origins.

6.5. Distribution of verbal classifiers

Map 5 shows the distribution of verbal classifiers in the languages of the

world.
In North America, verbal classifiers are found in all Eyak-Athabaskan
languages and in Haida. Northern Athabaskan languages have suppletive
classificatory verbs and verbal classifiers (termed 'genders' in the Athabaskan
linguistic tradition: see Thompson 1993). Classificatory verbs are an areal
feature of the Southeast (including all the Muskogean languages, cf.
Kimball 1991: 452-8 and Haas 1978, on Koasati; Campbell 1997: 342,
on Tunica, Natchez, Atakapa, Chitimacha, Yuchi, Biloxi, and Dhegiha). In
the Great Basin area Nevome (Piman: Uto-Aztecan) has suppletive verbal
classifiers. Among Californian languages, Pomoan languages have verbal
classifiers, as do Tarascan (isolate, Southwest Mexico: Costenla Umana
1991: 116; Friedrich 1971), Wakashan, Hokan, some Salish, Iroquoian,
Caddoan languages, and Dakota (Siouan) (Campbell 1997). The existence
of classificatory verbs in a few Mesoamerican languages has been suggested
by Suarez (1983: 90-1); however, the evidence is not clear and more work is
needed (see the criticism of Suarez by Quinn forthcoming and Campbell
1985).
In Australia, verbal classifiers are found in some northern Australian
prefixing languages, e.g. Mayali (Gunwinjgu), Nunggubuyu, Ngandi, Tiwi,
and Anindilyakwa (Sands 1995: 272 ff; Dixon forthcoming).
Verbal classifiers are found in Papuan languages of southern and central
12
Verbal classifiers of and n also occur in some nouns as a kind of derivational marker, e.g.
Slave dechi 'stick' (with d marker) (Keren Rice, p.c.); in Ahtna n is said to occur 'in compound
nouns that are roundish or rope-like', e.g. c'enluu 'unripe berries', -nts'ese' 'seed, pit' (Kari
1990: 285) (the classifier n is in bold type).
 verbal classifiers found in continuous areas

isolated instances of verbal classifiers

MAP 5. Distribution of verbal classifiers in the languages of the world

 Verbal Classifiers 171
Bougainville, and in the Waris language family (Brown 1981; Seiler 1983).
They may also be found in Reef-Santa Cruzan languages (see Wurm 1992a;
1992b). Classificatory existential verbs are extremely widespread in Papua
New Guinea—they are found in Engan, Waris, Asmat, Kiwaian families, in
Chimbu, Melpa, and in Huon languages (see Lang 1975: 115–20, for an
overview; W. Seiler 1986; 1989; Foley 1986; Lang 1975; Brown 1981;
Merlan et al, 1997); and also in some Ok languages (Christensen 1995).
Quite a few languages of Lowland Amazonia have verbal classifiers. These
include North Arawak, Yagua, Harakmbet, a few Tupi languages (e.g.
Munduruku and Karo), and some isolates, such as Waorani (Derbyshire
and Payne 1990). Partly analysable classificatory verbs are found in Ika
(Chibchan: Frank 1990).
There are no verbal classifiers in the languages of Africa or Eurasia or in
the Austronesian family.
 7 Locative and Deictic Classifiers

7.1. The structure of this chapter

This chapter describes two further types of classifier which have a noun
phrase as their scope: locative classifiers which occur in locative noun
phrases (§7.2) and deictic classifiers which occur on deictic modifiers
and/or articles in head-modifier noun phrases (§7.3). More examples of
these classifiers in the languages of the world would need to be discovered
before their typological profile could be fully established.

7.2. Properties of locative classifiers

Locative classifiers are morphemes which occur in locative noun

phrases. Their choice is determined by the semantic character of the
noun involved—in all the cases described here, it is the argument of a
locative adposition. Classifiers of this type are rare; all the examples
discussed here come from South American Indian languages on which
information has only recently become available: Palikur and Lokono
(North Arawak), in Carib languages and in Daw (Maku). This is the
reason why they have rarely been considered in previous typologies of
classifiers.1
In all the known cases, locative classifiers are 'fused' with an adposition
(preposition or postposition). The choice of adposition then depends on
physical properties of the head noun, e.g. shape, or consistency. (Following
the analogy of classificatory verbs (§6.2.3), they could be called classifica-
tory adpositions.) The choice of a locative classifier is semantic.
Palikur (North Arawak; Aikhenvald and Green 1998) offers the only
clear-cut example of locative classifiers. Palikur has three genders, and also
five distinct types of classifier: numeral classifiers, two subsets of verbal
classifiers (those occurring on stative verbs, which are frequently used as
modifiers in NPs, and those occurring on transitive verbs: see §6.4), as well
as possessive classifiers (generic nouns used in possessive constructions
with some alienably possessed nouns (§5.5.1) (and discussion in §8.2).
The choice of locative classifiers is based on the shape, dimensionality
and boundedness of the head noun (see Diagram 7.1); it does not involve
1
Locative classifiers were first introduced as a distinct subtype by Allan (1977: 286).
 Locative and Deictic Classifiers 173
animacy distinctions. These morphemes can be used as locative adpositions
by themselves meaning 'on' or 'in'. Whether they cross-reference person,
number, and gender of the head noun and whether they are used as pre-
positions or as adpositions depends on the pragmatic status of the head
noun (see details in Aikhenvald and Green 1998). Note that the shape and
dimension terms are the same as in Diagram 6.1, but one boundedness
related term and the specific classifiers are different. Those morphemes in
Diagram 7.1 which are the same as verbal classifiers (see Diagram 6.1) are
marked with an asterisk (*).

DIAGRAM 7.1. Locative classifiers in Palikur

Classificatory locative adpositions cooccur with the locative suffixes to

form directionals, elatives, and perlatives: -t 'directional: into, to', -tak
'elative: from', -iu 'perlative: along'. In 7.1 -min is used to refer to the
vertical location: the arm. In 7.2 -peru is used to refer to a branch-like
location, a tree.
7.1. pis keh paha-t arab p1-wan
2SG make one-NUM.CL:VERT shield 2so-arm
min
On.VERT
'You make a shield on your arm.'
174 Classifiers
7.2. ig-kis ute-e-gi ig motye
3M-PL find-COMPL-3M 3M wasp
ay-h-te a-peru ah
there-INT-DISTAL 3n-on.BRANCH LIKE tree
'They found the wasps on the tree.'
Example 7.3 shows a directional marker on locative classifier -hakwa
'in. WATER'.
7.3. wis-uh tarak-e-gu a-hakwa-t un
IPL-EXCL push-COMPL-3F 3n-in.wATER-DIR water
'We push it (the canoe) into the water.'
In an adpositional phrase, the head can be omitted. For example, a
locative classifier ('on.flat', i.e. on the flat surface of the rattle) is used
headlessly, as in 7.4.
7.4. ka-daha-ni warukma gu-madka
ATT-for-Poss big.star 3F-on.FLAT
'It (rattle: feminine, flat) had a big star on it.'
The adposition -bet used for substances (e.g. mud, clay, faeces) also plays
the role of a residual classifier: it is used for otherwise unclassifiable items.
These include abstract nouns, such as thoughts, darkness, coolness or
suffering.
A smallish system of five locative classifiers is found in Lokono
(Arawak) (Aikhenvald 1996b; Pet 1987: 37-8). One, koborokon 'inside
of an animate body, among living beings', correlates with the animacy
of the referent of the head noun; three correlate with consistency (loko
'inside a hollow or solid object'; rakon 'in a fluid'; kolokon 'in fire or light'),
and one with interioricity and dimensionality (roko 'on the inside surface
of).
Daw (Maku, Northwest Amazonia: Martins 1994: 53 ff.) has five loca-
tive classifiers. Their choice depends on the physical properties of the
referent of the head: kcd 'inside a bounded object', mi' 'inside liquid, or
fire', bit 'underneath an object with an upper boundary', wa? 'above an
unbounded object', a 'inside a mixture'. Locative classifiers are fused
with locational markers, just like in Palikur. In 7.5, -k d 'in:HOLLOW' is used
with the noun canoe. In 7.6 -mi' 'in:LIQUID' is used with the noun 'river'.
7.6a is ungrammatical.
7.5. xoo-ked
Canoe-IN:HOLLOW
'in a canoe'
 Locative and Deictic Classifiers 175
7.6. naax-pis-mi'
water-small-IN:LIQUID
'in a small river'
7.6a. *xoo-mi'?
Canoe-IN:LIQUID

Locative classifiers are also found in Carib languages (Derbyshire forth-

coming) where the choice of a locative classifier depends on the dimension-
ality and consistency of a referent. See Table 7.1 (Derbyshire 1999) and 7.7,
from Hixkaryana.

TABLE 7.1. Locative classificatory suffixes in three Carib languages

Apalai Hixkaryana Macushi
Liquid, in kua-o kwa-wo ka
into kua-ka kwa-ka ka-ta
Flat surface, on po ho po
to po-na ho-na po-na
Open area, on ta-o ta-wolya-wo ya
to ta-ka ta-kalya-ka ya-pth
Enclosed place, in a-o ya-wo ta
to a-ka ya-ka ta-pih

7.7. asama y-ahe-tawo

trail GENITIVE-edge-CL:ON.OPEN.AREA
'at the edge of the trail'
The locative classifiers in Lokono, Daw, and in Carib languages are not
used elsewhere in these languages. In contrast, three of the locative classi-
fiers in Palikur are also used as verbal classifiers (see §8.2). No language
with locative classifier affixes has been found so far.2
Directionality and location are semantic parameters widely used in
verbal classifiers (see §11.2.5). The choice of a classificatory verb may
correlate with the type of location referred to. For instance, in Cherokee
(Blankenship 1997: 95) 'for materials which can assume the shape of their
container, the quality of the container often determines the class. Sugar in
a bowl is COMPACT, but sugar in a bag is FLEXIBLE.' One of the four
classificatory existential verbs in Qiang (Tibeto-Burman: LaPolla forth-
coming: 120-1) is used with referents located within a container of some

2
A possible example of locative classifiers as affixes is Kadiweu (Guaicuruan family:
Griffiths and Griffiths 1976: 111), where some classifiers are prefixed to locational adverbs.
This issue deserves more in-depth study; for an alternative interpretation of this issue in
Kadiweu, see Sandalo (1996).
176 Classifiers
type (see §6.2.3). Verbal classifiers can be used in locative expressions.
The so-called 'areal' prefix in Athabaskan languages can be used on
locative prepositions in Koyukon (example 6.35: Thompson 1993). Verbal
classifiers appear on locatives in Eyak (Krauss 1968; §9.1, Table 9.2
below).
Distinctions comparable to locative classifiers may be found in other
languages. In Archi (Northeast Caucasian: Aleksandr J. Kibrik, p.c.) the
choice between the two locative cases (glossed as IN, e.g. inside a con-
tainer, and INTER, e.g. among) appears to depend on the semantics of the
noun. These cases are, by and large, used with different nouns; but there
are a few instances of a reclassification of the same item. So, for instance,
'village-in' means 'in a village as a place', 'village-inter' means 'among
villagers'. The two locative cases have some similarity with locative classi-
fiers. Similar distinctions are found in other Northeast Caucasian lan-
guages, e.g. Tsez (Bernard Comrie, p.c.). These issues deserve further
investigation. Locative meanings are often expressed with other noun
categorization devices; for example, agreement classes with locative mean-
ings are found in Bantu languages (cf. e.g. Givon 1969; Bresnan and
McChombo 1986), in Motuna (Onishi 1994: 76-7), and in Nasioi (Hurd
1977: 137).
In many languages, including English and Tongan (see Broschart
1997), the choice of a preposition or of a locational expression depends,
to a varying extent, on the properties of the head noun. For instance, the
referent of a noun has to have a surface for the preposition 'on' to be
used with it; and have an 'inside' for 'in' to be used. However, this lexical
choice is different from locative classifiers—in the same way that the
choice of verbs like 'drink' and 'chew' in English (see §6.2.3) is different
from the choice of classificatory verbs in Athabaskan languages. In
Palikur and other languages with locative classifiers described in this
section the obligatory choice of an adposition is made depending strictly
on the properties of the referent noun; there are paradigmatic relations
between the types of nouns and the choice of an adposition. This is not
so in English.

7.3. Properties of deictic classifiers

Deictic classifiers obligatorily occur with deictic elements such as articles
and demonstratives. Their choice is semantic, and they categorize the noun
in terms of its shape, animacy, and position in space; they do not always
appear on the noun itself. Deictic classifiers were first identified by Barren
and Serzisko (1982) and then by H. Seiler (1986). Craig (forthcoming: 42–3)
also mentions the existence of 'classifying systems' in articles in some
 Locative and Deictic Classifiers 177
North American Indian languages. In §7.3.1 I consider examples of deictic
classifiers, and provide justification for considering them a classifier type.
Discussion and conclusions are in §7.3.2.

7.3.7. Examples of deictic classifiers

Deictic classifiers have been described for (A) The North American lan-
guages Yuchi and the Siouan family, (B) Guaicuruan languages of South
America, and (C) Eskimo, where they are fused with demonstratives
(unlike in A and B).
(A) Deictic classifiers in North American languages
Classifiers are obligatorily used with deictics and articles in a number of
Siouan languages, especially Mandan and Ponca (Barren and Serzisko
1982, and a resume in H. Seiler 1986: 87 ff.). The classifier morphemes
come from grammaticalized stance verbs 'sit', 'stand', and 'lie' (cf. Rankin
1976). They indicate the stance of the antecedent, as well as the form of the
antecedent: one-dimensional (long, vertical, or 'standing'), two-dimen-
sional (horizontal, or 'lying'), or three-dimensional (round, or 'sitting'),
e.g. Mandan (Barren and Serzisko 1982: 99; Watkins 1976: 30):
7.8. d-mak
this-DEICTIC.CL:LYING
'this one (lying)'
7.9. de-nak
this-DEICTIC.CL:SITTING
'this one (sitting)'
7.10. de-hak
this-DEICTIC.CL:STANDING
'this one (standing)'
Stance verbs in North American languages with deictic classifiers are
also used as classificatory verbs the choice of which correlates with the
shape of the intransitive subject: long objects 'stand', round objects 'sit'
and flat objects 'lie' (Watkins 1976: 30, for Siouan; H. Seiler 1986: 89).
Ponca (Siouan: Watkins 1976: 33–4; Barron and Serzisko 1982: 86) has
two sets of definite articles the choice of which is determined by the
animacy of the referent, its position, and, for an animate referent, its
number and whether it is at rest or in motion. (If the noun is indefinite
it is used without an article or wi is used irrespectively of the properties of
the referent.) See Table 7.2.
178 Classifiers
TABLE 7.2. Article classifiers in Ponca

Inanimate articles Animate articles

k'e 'horizontal' ak'a 'sg.subj at rest'

t'e 'standing; collective' ama 'sg.subj in motion; pi.'
ca 'rounded' t'a 'sg.object standing'
ge 'scattered' ci 'sg.object moving'
ma 'pl.object' (see ama)
(cink'e 'sg.object sitting'
(cank'a 'pl.object sitting'

Examples 7.11a-c (Barron and Serzisko 1982: 93) illustrate how the use
of different article classifiers results in varying interpretations of a poly-
semous noun. This is similar to the semantic functions of other classifiers,
e.g. numeral classifiers and noun classifiers (cf. §3.2.2).
7.11 a. ni t'e
water ARTSTANDING, COLLECTIVE
'the water'
7.11b. ni ca
water ART:ROUNDED
'the handful of water (cupped)'
7.11c. ni k'e
water ART:HORIZONTAL
'the (line of) water, the stream'
The same set of morphemes are also used predicatively, as locative
expressions (Barron and Serzisko 1982: 94-5).
Yuchi (isolate: Watkins 1976: 35; Barron and Serzisko 1982: 96–7) has a
largeish set of morphemes the choice of which is determined by whether the
object is one-dimensional (long, vertical), two-dimensional (horizontal), or
three-dimensional. There is also a special plural class. These morphemes
appear to be obligatorily marked on the demonstrative and on the noun
itself and have an article-like meaning, e.g. ya-fa (wood-ART:VERTICAL) 'the
tree', ya-'e (wood-ART:HORIZONTAL) 'the log'; ne-fa ya-fa (this-ART:VERTICAL
wood-ART:VERTICAL) 'this tree'.
(B) Deictic classifiers in Guaicuruan languages
Classifiers which indicate spatial position/location of an object, its absence
vs. presence in the visual field and its form (extendedness vs. non-extend-
edness; horizontal vs. vertical extension) are used in Toba and Pilaga
(Guaicuruan family, Argentina).
 Locative and Deictic Classifiers 179
Toba distinguishes six classifiers (termed 'locative particles' by Klein
1979: 89-91; 1978: 151 ff.); they are obligatory with headless demonstrative
pronouns (which distinguish masculine and feminine genders) and option-
ally cooccur with nouns. The system of classifiers in Toba is shown in
Diagram 7.2 (Klein 1979: 91).
Presence in the visual field Absence from visual field

Anticipated presence Realized presence Anticipated absence Realized absence

in visual field in visual field from visual field from visual field

Non-extended ka

Vertical extension Horizontal extension

ra ji
DIAGRAM 7.2. Classifiers with spatial semantics in Toba

Examples of demonstrative and third person pronouns with classifiers

and gender markers (0 for masculine and ha- for feminine) in Toba are
given in 7.12-15 (Klein 1978: 155-7 and p.c.).
7.12. 0-ra-mari
MASC-VERTICAL.EXTENSION-3RD.PERSON.PRONOMINAL.BASE
'he (standing)'
7.13. ha-ka-mari
FEM-REALIZED.ABSENCE.FROM.VISUAL.FIELD-3RDPERSON.PRONOMINAL.BASE
'she (unseen or unknown to the speaker)'
7.14. 0-ra-ra
MASC-VERTICAL.EXTENSION-REDUPLICATION
'this (one masculine) standing'
A classifier with a noun is shown in 7.15 (Klein 1978: 151).
7.15. hinakta 0-ra-ngoto-lek
is.biting MASC-VERTICAL.EXTENSION-yOUth-MASC
0-ni-mpi'oq
MASC-NON.EXTENDED-dog
'A (sitting) dog is biting a (standing) boy.'
180 Classifiers
Classifiers in Toba do not qualify as locative classifiers, since they do not
cooccur with spatial adverbs or adpositions (see Croft 1994). They cat-
egorize nouns; and at the same time they obligatorily cooccur with deictics.
Classifiers in Pilaga, another language of the Guaicuruan family (Vidal
1994; 1997), are similar to those in Toba; they are preposed to nouns, and
refer both to the position of the noun in space and to its form. Pilaga has
six classifiers: three 'positional' and three 'deictic' (Vidal 1997: 75, 78). The
positional classifiers are: da7 'vertically extended, long' (e.g. humans, trees,
horses), nil 'sitting/non-extended, rounded' (mammals, snakes, insects,
buildings, fruits), di7 'lying/horizontally extended' (dead people, fishes,
towns). The deictic classifiers are na7 'coming/proximal', so7 'going
away/past', ga7 'absent/distal'. In 7.16 the classifier da7 'vertical' refers to
a person (vertical position is considered 'inherent for people and animals').
The second occurrence of the same classifier refers to the knife (in a vertical
position) (Vidal 1997: 76).
7.16. da7 siyawa di-kiyana-a
DEIC.CL:VERT person 3sG-eat-OBJ.AGR
da7 ganaat
DEIC.CL:VERT knife
'That person (standing) is eating something with a knife.' (i.e. He/she
shows the knife which is in vertical position.)
Classifier ni7 'sitting, non-extended' is used with buildings and with
mammals, birds and insects, as in 7.17. It can also be extended to humans
in a sitting position, as shown in 7.18.
7.17. ni7 siyaq netawe ni7 emek
CL:NON.ExT animal LOC CL:NON.EXT house
'The animal is inside the house.'
7.18. ni'7-ca7 weta di7 noik sekaet
CL:NON.EXT-PRO LOC CL:HORIZ town yesterday
'That one who is sitting (far from me . . .) was in the town
yesterday.'
Classifier di7 is used with objects which are perceived as 'horizontally
extended', such as 'town' in 7.18, or 'fire' in 7.19.
7.19. an-toni-igi di7-m7e dole
2sG-warm-MOD CL:HORIZ-DEM fire
'Warm yourself up by the fire.'
Like Toba, Pilaga also has derivational markers which refer to the sex of
human referents, and physical properties of non-human referents (Vidal
1997: 63-5). An optional marker of feminine gender, ha, appears attached
to demonstratives, pronouns, and deictic classifiers (cf. ha- as the feminine
 Locative and Deictic Classifiers 181
marker in Toba in 7.13), e.g. ha-ni7-m7e yaw7o (FEM-CL:NON.EXT-DEM
woman) 'that woman' (sitting or non-extended) (Vidal 1997: 68).3
Similarly to Siouan languages, the classifiers used with deictics in
Guaicuruan languages come from verbs. Some of the positional deictics
reconstructed by Ceria and Sandalo (1995: 181) for Proto-Guaicuruan, the
ancestor language of Toba, Pilaga (and Kadiweu), are given in Table 7.3.
According to Filomena Sandalo (p.c.), deictic classifiers in Toba and Pilaga
correspond to existential verbs in Kadiweu.

TABLE 7.3. A sample of demonstratives in Proto-Guaicuruan

Gender Presence/absence from the visual field Position Demonstrative
Masculine Absent General *k:ae
Standing *(e)-d:a
Sitting *(e)-n:i
Present Lying *(e)-d:i
Coming *(e)-n:a
Going *(e)-dyu
Feminine Absent General *a-kae
Standing *a-d:a
Sitting *a-n:i
Present Lying *a-d:i
Coming *a-n:a
Going *a-dyu

(C) Deictic classifiers in Eskimo

In Eskimo, classifiers which combine reference to boundedness and visi-
bility of the objects appear fused with demonstratives. Central Yup'ik
classifiers fused with deictics (data from the Chevak dialect: Woodbury
1981: 237-8) are shown in Table 7.4.
The use of a classifier is illustrated with a following example from Gagne
(1966), cited by Denny (1979a: 103):
7.20. unaaq parnna aiguk
harpoon EXTENDE:one.up.there get
'Get that [visible, extended] harpoon up there.'

7.5.2 Conclusions and discussion

We saw in §7.3.1 that in (A) Siouan and Yuchi, and in (B) Toba and Pilaga,
deictic classifiers combine reference to the position of the noun categorized
3
This feminine marker can also be used to refer to inanimate referents; the rules of
assignment are unclear (Vidal 1997: 68).
182 Classifiers
TABLE 7.4. Deictic classifiers in Eskimo

Restricted Extended Obscured

Refer to objects or Refer to objects or areas which are Refer to what is far
areas which are unbounded in one or more away or invisible
bounded in all directions: a line, long object, or
dimensions: point-like, something in linear motion
rounded, stationary, or through time or space (one
confined unbounded dimension); an
expanse (two unbounded
dimensions); or a space or vast
object (three unbounded
dimensions)

and to its shape. In Eskimo their choice is determined by extendedness and

visibility of the object. In (A) and (B) deictic classifiers have arisen as the
result of the grammaticalization of posture verbs. At the same time, dif-
ferent classifiers may be used with the same referent if its position is varied
(compare examples 7.16 and 7.18, from Pilaga); they can also be used to
distinguish different meanings of a polysemous noun (7.11 from Ponca).
The use of deictic classifiers in North American languages is linked to
definiteness. H. Seiler (1986: 94) pointed out the 'intermediary' status of
'article classifiers': they have a verbal origin but are 'confined' to a noun
phrase and have 'particularly close affinities' with definiteness. We con-
clude that there is probably enough ground to postulate deictic classifiers
as a separate type, whose defining properties are:
(i) the morphosyntactic environment in which they appear: on deictics and/
or articles with scope over a noun phrase;
(ii) categorization of the head noun in terms of shape, extendedness, posi-
tion, and also animacy.
Deictic classifiers in Eskimo are different. Their choice is conditioned by
the shape of the referent noun in terms of 'typical' deictic categories, such
as visibility. More studies are needed in order to provide a convincing
categorization of Eskimo deictics which are obviously different from the
situation described for Siouan and for Guaicuruan.
We have seen in §2.7 that in languages with distinct systems of nominal
and pronominal noun classes, demonstratives would pattern with the latter.
Demonstratives may be the only modifiers to distinguish genders or noun
classes. Waura (Xinguan Arawak: Richards 1988; Aikhenvald 1996b) pro-
vides a dramatic example of this: two genders—masculine and feminine—
are distinguished only in deictics and nowhere else. In fact, gender marking
 Locative and Deictic Classifiers 183
is synchronically fused with deictics, e.g. MASC. eze, FEM. izi 'this'. There are
no gender oppositions elsewhere in the system, unlike other Arawak lan-
guages. (Waura also has a largeish set of morphemes used as numeral and
as verbal classifiers.) The gender opposition in Waura which only exists in
the system of demonstratives can also be considered analogous to deictic
classifiers in other languages. However, if a language has animacy or
humanness distinctions marked only on articles or on demonstratives
this does not necessarily imply the existence of deictic classifiers, since it
may be just a case of noun class agreement on multiple targets (cf. §9.2).4
Numeral classifiers are often, but not always, also used with demonstra-
tives, as in Mandarin Chinese or in Hmong. Even if the same set of
classifiers is used with numerals and with deictics, they can behave differ-
ently (see Chapter 9). We will show in §9.3 that Tariana, a language with
classifiers used in multiple environments, employs somewhat distinct sets
with various types of deictic, with interrogatives, and with articles, and that
this does not mean that every morphosynctactic context involves a distinct
type of classifiers. However, the fact that demonstratives and articles can
have a distinct agreement system, or employ a distinct subset of classifier
forms, may be considered a prerequisite for the emergence of deictic
classifiers as a distinct type, as in Siouan, in Yuchi, and in Guaicuruan
languages.
4
Teop (Austronesian, Bougainville: Mosel and Spriggs forthcoming) has three genders
marked on articles. This system cannot be considered that of article classifiers because it
involves agreement, and thus is better considered a noun class system.
 8 Different Classifier Types in One
Language

8.1. General observations

Several distinct classifier types may coexist in one language. The cooccur-
rence of different classifier systems in different morphosyntactic environ-
ments constitutes a strong argument in favour of the proposed typology of
classifiers based on the morphosyntactic locus of coding of noun categor-
ization devices (together with their scope of categorization, principles of
assignment, and the kind of surface realization: see §1.5).
The different sets of morphemes used in distinct classifier environments
may partly overlap in their form and/or semantics. Sometimes their seman-
tics is the same, but the form is different; sometimes it is the other way
round. The most frequent combination of distinct classifiers within one
language is numeral classifiers and noun class systems. Numeral classifiers
may also coexist with noun classifiers or with relational classifiers. Verbal,
locative, and noun classifiers may coexist with noun classes. Different sets
of classifiers interact with other grammatical categories, such as number, in
different ways (see Chapter 10).
In Chapters 2–61 discussed languages with more than one (in most cases,
just two) subsets of noun categorization devices of the same type. The most
frequent situation is for a language to have two types of noun class system:
a 'nominal', and a 'pronominal' one (see §2.7). There is rarely more than
one variety of other kinds of classifier (see §4.3 on numeral classifiers, §5.5.2
on classifiers in possessive constructions, and §6.4 on verbal classifiers). No
language has been found with two kinds of noun classifiers, or of locative
or deictic classifiers.1
Two varieties of the same classifier type (e.g. noun classes, numeral or
verbal classifiers) can be in a complementary distribution. In §2.7.1 we
discussed distinct noun class agreement systems the choice of which
depends on the kinds of modifiers (adjectives may have one agreement
system, and deictics another). Different kinds of verbal classifier may
categorize different arguments: one is used for O and the other is used
for S (e.g. Waris, in §6.4.1). Distinct kinds of classifier may cooccur in the
same environment. The two noun class systems in Paumari can be
1
This is to be expected due to the rarity of these types.
 Different Classifiers in One Language 185
marked on the same modifier and on the predicate (§2.7.2); in Athabaskan
languages different verbal classifiers coexist in the same environments
(§6.4.2).
All these cases are very similar to the ones discussed here. In each
instance, every subtype of the 'same' kind of classifier is likely to have its
own semantic and functional properties, as well as its own history; or they
may correspond to different historical stages of the system. They constitute
'incipient' types of classifiers, grammaticalizing distinct 'focal' points on
the continuum of noun categorization.
In other languages the same set of morphemes can be used in more than
one classifier environment (see Chapter 9). These morphemes can then have
somewhat different grammatical realization, or be more or less obligatory
depending on what classifier construction they are in. Then there are more
problematic systems—such as Tariana and Baniwa (§9.3), where classifiers
in several different environments overlap just slightly. These are fuzzy
types, with tendencies towards the grammaticalization of various focal
points on the continuum of noun categorization devices. As we will show
in §9.3, in most instances—but not always—these focal points correspond
to already established classifier types.
We first consider coexisting classifier sets in different contexts (§8.2). In
§8.3 we deal with languages where distinct classifier sets cooccur in the
same environment. Some conclusions are given in §8.4.

8.2. Coexisting classifier sets in different environments

Two sets of classifiers in one language is by far the most frequent case.
Combinations of two classifier systems are discussed in (A). Three or four
sets of classifiers are rarer—see (B) and (C) respectively.
(A) Two sets of classifiers
NOUN CLASSES and NUMERAL CLASSIFIERS coexist in a few Indie (Rastorgueva
et al. 1978), Dravidian (Emeneau 1964), Iranian (Zograf 1976), and some
Arawak languages (e.g. Warekena, Yucuna, Achagua). Malto (South
Dravidian: Mahapatra 1979) has about thirty numeral classifiers, and
two noun classes (genders). Classifiers are used to categorize nouns in
terms of animacy, form, shape, and size; there are also a few unique
classifiers each used with a single noun. Nouns are divided into two
genders: male human vs. the rest in singular; human vs. non-human in
plural (see §10.1).
NOUN CLASSIFIERS and NOUN CLASS systems coexist in some prefixing
languages of Northern Australia, such as Ngan'gityemerri (Reid 1990;
1997; Sands 1995: 281-2). The systems of noun class markers and classifiers
186 Classifiers
in Ngan'gityemerri are shown in Table 8.1; they overlap semantically. Noun
classifiers are optional, while noun class markers are obligatory in noun
phrases. Some functions of the noun class markers in Ngan'gityemerri were
discussed in §3.5. Examples 3.26 and 3.27 illustrate how noun class markers
and noun classifiers are different in their form, and obligatoriness. As Reid
(1997) has convincingly shown, historically noun class prefixes developed
as the result of grammaticalization and reanalysis of generic noun classifier
lexemes (also see §13.8). Ngan'gityemerri also distinguishes masculine and
feminine genders in third person singular pronouns.

TABLE 8.1. Noun classes and noun classifiers in

Ngan'gityemerri
Gloss Noun class marker Classifier
Male wa-
Female wur-
Human group awa-
Body parts da-, 0-, a-
Canine wu-
Animal a- gagu
Plant food, vegetable mi- miyi
Trees/things yerr- yawurr
Bamboo spears yeli- yawul
Strikers syiri
Fire yenggi
Liquid kuru
Digging sticks kini
Large woomeras tyin
Canegrass spears kurum

NUMERAL classifiers and RELATIONAL classifiers coexist in numerous

Micronesian languages. Mokilese has four numeral classifier suffixes and
around fourteen relational classifiers which are independent nouns
(Harrison 1976: 95, 128-31)—see Table 8.2. Relational classifiers are
chosen according to the way the possessed noun is 'possessed', or handled
by the possessor, and also according to the properties of possessed nouns.
Thus, they share the properties of relational and possessed classifiers (see
§5.5.1).
Ponapean (Micronesian: Rehg 1981; Carlson and Payne 1989) has
around thirty numeral classifiers; there are also relational classifiers.
Numeral classifiers are used to categorize objects in term of their form
and shape; relational classifiers combine the reference to the way in which
items can be possessed, and to their functional properties. Twenty-two
 Different Classifiers in One Language 187
TABLE 8.2. Classifiers in Mokilese

Semantics Numeral classifier Relational classifier

Animate -men
Long objects -pas
Things which have pieces, parts -kij
General -w ah
valuable, child, pet nah
Food kanah
Drink nimah
Chaw ngidah
Vehicle warah
House imwah
Garland mwarah
Ear decoration dapah
Earring siah
Mat kiah
Land japwah
Sheet upah
Pillow wilingah

relational classifiers are used in the 'common' speech register, while the
'honorific' and the 'humiliative' registers employ fewer classifiers (Keating
1997; see §10.5). No such distinctions are found in numeral classifiers in
this language.
Truquese (Micronesian: Benton 1968) has a virtually open class of rela-
tional classifiers, which combine properties of possessed classifiers (see
examples 5.48, 5.49: §5.5.1). It also has a large inventory of numeral
classifiers, some of which are also repeaters (example 4.24).
NOUN CLASSIFIERS coexist with NUMERAL CLASSIFIERS in Akatek (Kanjobal
Mayan) and in Minangkabau (Western Austronesian). Akatek has two sets
of numeral classifiers: a small closed set of affixed classifiers and a larger set
of numeral classifiers as independent lexemes (see examples 4.31–4, and
§4.3.2 on the difference between the two types). Akatek also has fourteen
noun classifiers (see example 4.31, and also 3.10 and 3.11): see Tables 8.3 and
8.4 (Zavala 1992: 131, 140, 152). The sets of numeral and noun classifiers
show a partial semantic overlap. Affixed numeral classifiers characterize the
noun in terms of humanness and animacy; independent numeral classifiers
characterize the noun with respect to its shape, size, and form. Noun
classifiers categorize the nouns in terms of what generic class it belongs to.
Akatek also has a morphological opposition human vs. the rest realized
in the plural marker, eb\ used to mark human plurals, shown in 8.1 (Zavala
1992: 119). This marker cannot be used with a non-human referent, cf. 8.2.
188 Classifiers
TABLE 8.3. Numeral classifiers in Akatek
Semantics of classifiers Affixed numeral Independent
classifiers numeral classifiers
Human -wan
Non-human animate -k'on
Inanimate -eb'
Long vertical wa?an
Separate k'itan
Curved kupan
Round soyan
Big flat patsan
Extended xenan
Round three-dimensional k'olan
Small round three-dimensional b'ilan
Big round or oval three-dimensional pilan
Round two- or three-dimensional silan
Short three-dimensional xilan

TABLE 8.4. Noun classifiers in Akatek

Semantics Noun classifier Noun from which it is derived
Man nax winax
Woman ?is ?is
Human, known k'o 7
Human, appreciative yab' ?
Animal no? no?
Wood te? te?
Stone ts'en ts'een
Maize (?i)sim ?isim
Kind of rope ts'an ts'an
Earth ts'ots' tsoots'
Salt (?a)ts'am ?ats'am
Water (x)a? xa?
Plant ?an 7
Fire q'a?lk'a? q'a?

.1. xa?-0 eb' nax winax

ACTUALIZER-ABS PL:HUMAN NCL:MAN man
'There are the men.'
.2. xa?-0 no? tsee
ACTUALIZER-ABS NCL:ANIMAL horse
There are/is horse(s).'
 Different Classifiers in One Language 189
This human vs. non-human distinction in the plural marker in Akatek is
reminiscent of noun classes. (It will be shown in §10.1 that it is typologi-
cally unusual for a language to have more noun class-like distinctions in
non-singular numbers than in the singular; cf. Croft 1990: 214-16.) The
distinction between human and non-human nouns in Akatek is reminiscent
of languages like Chinantec (see B), which combine numeral classifiers,
noun classifiers, and noun classes. Unlike Akatek, Chinantec has a much
more regular animacy agreement in adjectives.
Minangkabau (Western Austronesian: Marnita 1996) has noun classi-
fiers and just one set of numeral classifiers. Any noun with generic seman-
tics can be used as a noun classifier to accompany a specific noun (see §3.3).
Numeral classifiers are a closed class. Both numeral classifiers and noun
classifiers come from independent nouns. The differences between the two
types of classifier involve the following:
(i) semantics, especially of the same morphemes used as noun classifiers
and as numeral classifiers;
(ii) the use of different noun classifiers and numeral classifiers for nouns of
the same semantic group;
(iii) morphosyntactic behaviour;
(iv) obligatoriness.
(Al) Semantics Numeral classifiers have the semantic oppositions shown
in Diagram 8.1 with reference to animacy, humanness, and form and shape
for inanimate nouns.

<
Human: urang

Non-human: ikua

Flat foldable objects, e.g. paper; hair: alai

Round, hollow objects, e.g rings: bantuak
Long vertical objects, e.g. logs, trees: batang
Flat long thin objects, e.g. timber: bilah
Round objects, e.g. fruit: incek
Solid objects, e.g. house, and abstract notions: buah
* Specific classifiers: thread-like ornaments, arms, words,
letter, rice field, flower
DIAGRAM 8.1. Semantics of numeral classifiers in Minangkabau
190 Classifiers
Noun classifiers are generic nouns; unlike numeral classifiers, they signal
that the referent of a noun belongs to a certain class of objects, e.g. buruang
'bird', ula 'snake', ikan 'fish', kayu 'wood', e.g. ikan hiu (NOUN.CL:FISH
shark) 'a shark', kayu candano (NOUN.CL:WOOD cendana) 'cendana tree'. A
few morphemes can be used both as numeral classifiers and as noun
classifiers, with a difference in meaning. Batang, when used as a noun
classifier, means 'trees as a class'. As a numeral classifier, it is used to refer
to long vertical objects, e.g. trees. As an independent noun, it means 'tree
trunk'. A numeral classifier and a noun classifier can cooccur in one noun
phrase, and will have different meanings, as in 8.3 (= 3.16).
8.3. sa-batang batang pisang
one-NUM.CL:LONG.VERT NOUN.CL:TREE banana
'one banana tree'2
(A2) The use of different noun classifiers and numeral classifiers for nouns of
the same semantic group Nouns of a given semantic group may take
different noun classifiers and numeral classifiers. Names of flowers take
bungo 'flower' as a generic noun classifier. The specific numeral classifier
for flowers is tangkai, e.g.
8.4. sa-tangkai bungo
one-NUM.CL:FLOWER flower
'one flower'
A specific name of a flower can be added, as in 8.5.
8.5. sa-tangkai bungo ros
one-NUM.CL:FLOWER flower rose
'one rose'
(A3) Morphosyntactic behaviour Numeral classifiers, which are attached
to numerals, are used with small numbers, 'one' and 'two', and form one
phonological and morphological word with the number (a special semi-
suppletive form sa.of numeral 'one', satu, is used with numeral classifiers).
Noun classifiers are independent phonological words.
(A4) Obligatoriness Numeral classifiers are always optional; in fact,
young people nowadays rarely use them (Rina Marnita, p.c.). Noun
classifiers are often obligatory. There are nouns (mostly plant names)
which cannot be used without a noun classifier, e.g. bungo ros (NOUN.CL:
FLOWER) 'a rose', *ros 'rose'.

2
Similar examples can be found in the closely related Acehnese (Durie 1985; p.c.). The
homophonous noun classifier boh 'fruit, egg' and numeral classifier boh 'generic' can cooccur
within one noun phrase, but their semantics is different, e.g. dua boh boh mamplam (two
NUM.CL:GENERIC NOUN.CL:FRUIT mango) 'two mangoes'.
 Different Classifiers in One Language 191
RELATIONAL CLASSIFIERS may be distinct from another set of classifiers used
in several different environments. This is the case in Kilivila (see §9.1).
Relational classifiers in this language are exemplified in 5.36-8.
NOUN CLASSES can be distinct from another set of classifiers which occurs
in several environments. Then they can usually cooccur—see the examples
from Anindilyakwa (Australian: Sands 1995: 277; Leeding 1989) and
Machiguenga (Campa, Peruvian Arawak: Shepard 1997) in §8.3 below.
DEICTIC CLASSIFIERS and POSSESSED CLASSIFIERS coexist in Toba (Guaicur-
uan). The deictic classifiers (see (B) in §7.3.1) combine information about
the shape and the position of a referent noun, and there is one possessed
classifier (lo 'animal possession') (see (B) in §5.2). Cora combines possessed
classifiers ((A) in §5.2) with classificatory verbs (see (A) in §6.2.3).
(B) Three sets of classifiers
Three distinct sets of classifiers are comparatively rare in the languages of
the world.
NUMERAL and NOUN CLASSIFIERS coexist in Chinantec languages (Mexico)
together with NOUN CLASSES. In Sochiapan Chinantec (Foris forthcoming:
256 ff.) numeral classifiers and noun classifiers are independent words.
There are ten noun classifiers which refer to sex and shape; there are
also two diminutive classifiers and one classifier with the meaning 'old,
disused'. Some of these derive from independent nouns. There are several
dozen numeral classifiers (Foris forthcoming: 255, 314-21). Numeral and
noun classifiers can cooccur in a noun phrase, as in 8.6.
8.6. hna3 mai3 (mi3) jlai2
five NUM.CL:SPHERE (NOUN.CL:SPHERICAL) egg
'five eggs'3
In examples like 8.6, the noun classifier can be omitted; the numeral
classifier is obligatory. Chinantec languages also distinguish animate and
inanimate noun classes; this distinction is realized in agreement with
modifiers and with predicates. It is illustrated in the following examples
(Foris forthcoming: 280-1). Agreeing adjectives are underlined.
8.7. tia2 ni1-ni1tsin21 jna 13 tsa3cua1 jlinh 1 ni2
not INT-mount+TA+lsg I horse wet+AN that
'I don't want to mount that wet horse.'
8.8. tia2 re2 cau32 cuo1 jlih21
not well burn+PRES firewood wet+INAN
'Wet firewood does not burn well.'

Here, and in other examples from Chinantec, numbers stand for tones.
192 Classifiers
NOUN CLASSES coexist with NOUN CLASSIFIERS and POSSESSED CLASSIFIERS in
'Dongo-ko (Mba, Ubangi branch of Niger-Congo: Pasch 1985). Possessed
classifiers in 'Dongo-ko were illustrated in 5.10 and 5.11. Alongside ten
possessed classifiers, 'Dongo-ko also has nine noun classes and nine noun
classifiers (Pasch 1985: 70 ff). These have the same semantics, but differ in
form and in syntactic conditions of use. A noun can belong to a noun class
which is distinct from the noun classifier it takes. In 5.10, b-go 'leopard'
has a class 7 noun class; it shows agreement with a possessive marker in
noun class la: 0-a 'CL1-poss'. Difference in form between a noun classifier
and a noun class marker is illustrated in 8.9.
8.9. pi-wo w-i r
arm-NCL1 CL1-POSS ISG
'my arm'
Daw (Maku: Martins 1994) has three distinct sets of classifiers: POSSESSOR
CLASSIFIERS, LOCATIVE CLASSIFIERS, and NOUN CLASSIFIERS. They differ in form
and semantics.
Daw has two possessor classifiers: -dee' is used for inanimate possessor,
and -ej for animate possessor (see examples 5.45 and 5.46 in §5.4; Martins
1994: 138–41). There are also locative classifiers (Martins 1994: 53ff.) which
categorize nouns with respect to their physical properties, and animacy (see
examples 7.5 and 7.6 in §7.2). Daw also has an open class of noun classi-
fiers. They are independent lexemes; apparently, any noun with a generic
reference can be used as a noun classifier (Martins 1994: 51; see example
3.5 in §3.2.3).
Possessor and locative classifiers are obligatory. Noun classifiers are
optional; they can be omitted if the referent has been established in the
previous discourse. Also, unlike possessor and locative classifiers, they are
often employed for referent tracking in discourse.
Baniwa (North Arawak; Aikhenvald 1996c) has at least three distinct
sets of classifiers. One of them is the 'pronominal' noun class, used in
verbal cross-referencing and with deictics (see 2.1; §2.7.1; and (E) in
§2.4.2). There are also relational classifiers (see §5.5.2). Yet another set of
classifiers is used with numerals, verbs and in possessive constructions;
there are also noun class markers some of which are different from all
these sets, and some which are not (see Table 9.5). Depending on the
interpretation of noun class markers in Baniwa, this language can be
said to have three or four sets of classifiers (see §9.3.1).
(C) Four or more sets of classifiers
Palikur (North Arawak: Aikhenvald and Green 1998) has the world's
richest system of noun classification devices. Palikur has genders (with
two or three agreement forms depending on the construction type), and
 Different Classifiers in One Language 193
the following classifier types: numeral classifiers; verbal classifiers with two
subtypes—the ones used with stative verbs and the ones used with transi-
tive verbs (discussed in §6.4.1—see Diagram 6.1); locative classifiers (§7.2,
Diagram 7.1); and possessive classifiers (which combine properties of
relational and of possessed classifiers: see §5.5.1). Only numeral classifiers,
the two subtypes of verbal classifier, and locative classifiers show some
similarities—i.e. the same forms are used in some instances. There is a
significant overlap between the inventories of verbal and locative classifiers,
and some overlap with numeral classifiers: see Table 8.5 (further details are
in Aikhenvald and Green 1998).
Genders and possessive classifiers are completely independent. The
possessive classifiers in Palikur are -pig 'pet' used with domesticated
animals, -win used with animals that are caught to eat, and -kamkayh
'child' used with children, e.g. nu-kamkayh awayg (1sg-child man) 'my
son' (see §5.5.1). Gender in Palikur is realized through agreement of the
head-modifier kind, and also of the predicate-argument kind. Typically
for an Amazonian language, gender is usually not marked on the head
noun itself. Gender agreement is obligatory and every noun has a fixed
gender.

TABLE 8.5. Numeral, verbal and locative classifiers in Palikur

Semantics Numeral Verbal classifiers Locative
classifiers classifiers
With stative With transitive
verbs verbs

Animate -P -pit -pit -pit

Round, square -ul-so -pit -pit -pit
Irregular shape -al-sa -pit -pit -pit
Side -al-sa -muh -muh -pit
Vertical objects -tl-ta- -min -min -min
Rigid, thin -tl-ta- -ah -min -min
Flat -kl-ka-l-bu -boha -bo -madka
Concave; numeral classifier: -mkul-muk -apa -ap -madka
metal
Edge -mkul-muk -kiya -kig -kigbi(-mna)
Pointed -mkul-muk -kisa -kig -kigsa
Linear; numeral classifier: -tral-tahr-l-bu -buka -buk -buhku(-mna)
long and extended
Road, river -tral-tahr-l-bu -buka -buk -vigku
The inside part of; NUM.CL: -ikul-rik -eku -ik -iku
extended with boundaries
Tree, plant, trunk -ktil-kat -kat -min 'vertical' -pew
Tree, branch-like -ktil-kat -pewa -peru -peru
Water — -pit -pit -hakwa
194 Classifiers
Three gender agreement forms (masculine, feminine, and neuter) occur
on demonstratives in head-modifier constructions (see Table 8.6); they are
also found in predicate-argument constructions on third person cross-
referencing affixes, and on independent pronouns.

TABLE 8.6. Demonstratives in Palikur (singular)

In speaker's Near to Far from Far from Very far from
hand speaker and speaker and both, visible both, not
to hearer near hearer visible
or vice versa
Masc. ner ner nop netra nere
Fern. no no nop notra nore
Neut. inin ini nop inetra inere

Two gender agreement forms (feminine, and masculine-neuter or non-

feminine) are found with a number of verbal suffixes in predicate-argument
agreement with the subject (A/S) and in head-modifier agreement if a
modifier is a stative verb. These 'gender-sensitive' suffixes are shown in
Table 8.7. For a few individual lexical items gender agreement is realized
via internal vowel change (o 'feminine'; e or a 'masculine/neuter').

TABLE 8.7. Gender marking on verbs in Palikur

Masculine/neuter (NF) Feminine (F)
Continuative -ne -no
Continuative prolonged -nene -nano
Non-completed frustrated action -pa-ri -pa-ru
Inchoative -pi-ye -pi-yo
Durative -ye -yo
Individual lexical items miyap 'he/it dies' miyop 'she dies'
nemnik 'approach' nomnik 'approach'
nawenewa 'different' nawenowa 'different'

The coexisting systems of noun categorization devices in Palikur differ

on a number of points, but they also have some properties in common.
These properties are summarized and contrasted in Table 8.8.
The differences between the classifiers in Palikur are discussed below.
(I) Morphological form
Noun categorization devices in Palikur differ in morphological complexity.
Genders have two or three agreement forms depending on the type of
construction they are used in. There are also a few morphologically
TABLE 8.8. Properties of classifiers and genders in Palikur
Genders Numeral Verbal Locative Possessive
classifiers classifiers classifiers classifiers
(2 sets)
A Different morphological forms depending on yes no no no no
construction type
Different morphological forms depending on no yes no no no
classifier
Formal overlap with other noun classification no partly yes yes yes no
devices
Irregular forms yes yes no no no
Bound morphemes: yes yes yes no no
Prefixes, suffixes, infixes prefixes, suffixes, suffixes, infixes suffixes - -
infixes
Derivational functions (limited) no (limited) yes no
B Every noun 'classified' yes yes yes yes no
Obligatory use yes yes no yes yes
C Cooccurrence with other noun classification yes no yes yes no
devices in one morphological word
196 Classifiers
irregular gender forms. Numeral classifiers fall into several subgroups
depending on what numbers they are used with. Some are used just with
number 'one', e.g. -imku 'classifier for wrapped objects'; some with num-
bers 'one' and 'two', e.g. classifier -t 'vertical objects'; and some with other
numbers as well, e.g. -mkul-muk 'classifier for concave objects'. Unlike
gender systems, restrictions on the number of forms a numeral classifier
has are idiosyncratic for each particular classifier. Several numeral classi-
fiers have suppletive forms, e.g. the classifier for irregular-shaped objects is
-a with number 'one', and -sa with 'two' (Aikhenvald and Green 1998).
Classifiers of other types do not have variant forms, or show any
morphological irregularities.
Classifiers differ in their morphological status. Verbal and numeral
classifiers are bound morphemes. Numeral classifiers may be suffixes, or
infixes (to number 'two'). Verbal classifiers are suffixes. Gender markers
can be suffixes, or prefixes, or infixes. Possessive and locative classifiers are
independent morphemes.
Locative classifiers are used as derivational suffixes, while genders and
verbal classifiers have limited use as derivational markers. Other classifiers
are not used this way.
(II) Function and usage
Not all noun classification devices in Palikur—including gender and pos-
sessive classifiers—have to cooccur with an overt NP, that is they can be
used anaphorically. All of them allow variable classification of nouns
depending on which shape characteristic is in focus. This shows that
classifiers are not semantically redundant; they add information to the
noun (cf. Denny 1986; Downing 1996: 93).
Every noun in Palikur is assigned a gender, or a numeral, verbal, or
locative classifier. Only some nouns are assigned a possessive classifier.
The use of gender, numeral classifiers, possessive classifiers, and locative
classifiers are obligatory, while whether verbal classifiers are employed or
not depends on whether the S, or the O constituent is completely involved
in the action (see §6.4.1).
(III) Cooccurrence with other noun categorization devices
Gender marking can cooccur with verbal classifiers and with locative
classifiers, in one morphological word. It does not cooccur with possessive
classifiers, or with numeral classifiers (since there is a gender distinction in
numeral classifiers—however, this applies only to animate nouns and is
more semantically transparent). This is discussed in §8.3.
Table 8.9 summarizes the semantic, pragmatic and functional properties
of noun classification devices in Palikur. It also shows their different
origins.
TABLE 8.9. Semantic and functional properties and origin of classifiers in Palikur
Classifier Function Scope Semantics Pragmatic effect Origin
Genders Head-modifier NP: three genders Animacy, physical - Deictic/3rd person
agreement in NP with pronouns and properties enclitic: Proto-
A/S agreement in demonstratives; Arawak
clause two genders with
stative verbs as
modifiers
Outside NP (two
genders)
Numeral Quantification, NP Animacy, physical -
classifiers enumeration properties, nature
Verbal classifiers O/S agreement Clause Physical properties, Complete involvement
with transitive parts of objects or topicality/contrast Some come from
verbs and verbal of O/S nouns, e.g. body
classifiers with parts
stative verbs
Locative classifiers Location NP Physical properties,
'typical' locations
Possessive Possession NP Function, way of - Generic nouns
classifiers handling
198 Classifiers
Thus, Palikur can be said to have five distinct sets of classifiers.4 We have
no explanation so far as to why this language has such a uniqely compli-
cated system.5

8.3. Different classifier sets in the same environment

If a language has several classifier sets which cooccur in the same environ-
ment one of these is always noun classes. This is due to the fact that only
noun classes can have scope over a noun phrase, and/or over a whole
clause.
In a few languages which have both NOUN CLASSES and NUMERAL
CLASSIFIERS both can be marked on numeral modifiers. This is the case in
Achagua, a North Arawak language spoken in Venezuela (Wilson 1992:
61-3). Achagua distinguishes two genders—feminine and non-feminine.
The gender agreement is obligatory in verb-argument constructions and
in head-modifier constructions, e.g. auli maanu-i (dog big-so.MASc) 'a big
dog'. There are also twelve numeral classifiers, used with numbers from one
to three (such as 'recipient', 'wide objects', 'long and thin objects', 'round
objects', 'humans' etc.), e.g. 8.10.
8.10. aaba-hiza kaasta
one-CL:LONG.THIN paper
'one sheet of paper'
If a number is used with an animate noun, it takes both a classifier and a
gender marker, as in 8.11. There is no special 'inanimate' classifier; classi-
fier for mammals, -na-, is the only one to cooccur with gender markers.
8.11. aaba-na-i auli
One-CL:MAMMAL-SG.MASC dog
'one (male) dog'
Palikur also has an obligatory 'double marking'—of animacy and of
gender—on the cardinal numbers 'one' and 'two' when they accompany an
4
Diana Green reports that the modern Palikur tell that they have descended from eight
groups which came to live together. We may hypothesize that this unusual system could have
emerged as the result of different areal influence, and probably language mixing. For instance,
locative classifiers and classifiers in possessive constructions are reminiscent of those found in
neighbouring Carib languages. Bearing in mind long-term (and none too peaceful) contacts of
Palikurs with Carib peoples, one may hypothesize that these classifiers could have been
acquired as the result of the Carib influence. Genders are of the Arawak origin (Aikhenvald
1999a).
5
Kadiweu (Guaicuruan: Griffiths and Griffiths 1976: 111-12) can be said to have distinct
sets of noun classes, numeral classifiers, relational classifiers, locative classifiers and deictic
classifiers; Sandalo (1996) provides a different interpretation.
 Different Classifiers in One Language 199
animate head noun. Palikur has around eighteen numeral classifiers and
three genders (masculine, feminine, and neuter) marked on adjectives and
demonstratives as modifiers, and on the verb. The assignment of gender
when governing agreement on cardinal numerals is much more semanti-
cally transparent than the assignment of gender in other environments: it is
predominantly sex-based: males (including 'mythological' males such as
heavenly bodies, e.g. sun, moon) belong to the masculine gender, and
females belong to the feminine gender. There is no marking for gender
on classifiers used with inanimate nouns (gender assignment to inanimates
in Palikur is semantically very complex: see Aikhenvald and Green 1998).
The way gender agreement operates on numeral classifiers used with
animate masculine and feminine nouns is illustrated in 8.12 and 8.13.
8.12. no paha-p-ru tino
this:F one-NUM.CL:ANIM-F woman
'one woman'
8.13. ner paha-p-ri awayg
this:M one-NUM.CL:ANIM-M man
'one man'
An inanimate head noun cannot trigger gender agreement on numeral
classifiers, even though it may agree in gender with a demonstrative or a
predicate. The noun 'path' has neuter gender (shown in the form of the
demonstrative 'this'). There is no gender agreement on the numeral. That
is, there is a generic 'animate' classifier, -p; but there is no generic inanimate
form.
8.14. ini paha-tra ahin
this:NEUT one-NUM.CL:EXTENDED path
'one path'
The noun warik 'river' is feminine. Example 8.15 shows how this noun
does not take a gender agreement marker with a numeral because it is
inanimate.
8.15. no paha-tra warik
this:? one-NUM.CL:EXTENDED river
'one river'
Machiguenga (Peruvian Arawak: Derbyshire and Payne 1990; Wise
1986; Shepard 1997) distinguishes 'masculine' and 'feminine' genders in
verbal cross-referencing. The feminine gender is assigned to females and
most inanimate objects, while animates are assigned the masculine gender.
In addition, the sun, the moon, lightning, stars, rainbows, and other salient
natural phenomena are also treated as 'masculine'. There is also a large
200 Classifiers
system of classifiers used with verbs and with numerals (see §9.1), the
choice of which is based on the shape and animacy of the referent noun.
Unlike Achagua and Palikur, gender specification on numerals obligatorily
cooccurs with a numeral classifier. This is illustrated with 8.16 and 8.17
(Shepard 1997: 35-6). (Note that numeral 'one' is a discontinuous mor-
pheme: the classifiers are infixed.)
8.16. pa-poa-t-iro sekatsi
one-CL:CYLINDRICAL.OBJECT-INAN-one manioc.plant
'one manioc plant'
8.17. pa-poa-n-iro parari
one-CL:CYLINDRICAL.OBJECT-AN-one river.otter
'one river otter'
NOUN CLASSES and VERBAL CLASSIFIERS coexist in a number of the Aus-
tralian languages of Arnhem Land. Nunggubuyu (Heath 1984) and Gun-
barlang (Coleman 1982) have five noun classes, and also verbal classifiers.
In Gunbarlang, the noun classes have the following semantics:
Class 1: males and items associated with males.
Class 1: females and items associated with females.
Class 3: plants and plant parts, fire, implements not associated with men,
prepared or cooked food, some body parts.
Class 4: non-living things associated with the landscape, bodily excretions,
places, and abstract nouns.
Class 5: some body parts.
Verbal classifiers categorize the S/O constituent in terms of its shape and
function. The following example from Gunbarlang shows a verbal classifier
cooccurring with a noun class marker on a verb:
8.18. balangit na-njambi-bulerj
blanket NCL1-CL:CLOTH-dry
'The blanket is dry.'
In Anindilyakwa (Australian) A, S, and O are obligatorily cross-referenced
for noun class on each verb. The use of verbal classifiers, to characterize
the S/O constituent, is optional (Leeding 1989; Sands 1995: 276 ff.).6 In
Tiwi (Australian: Osborne 1974: 38; 46-50) the two genders, masculine and
feminine, are cross-referenced on the verb (only subject forms of non-
imperative verbs distinguish two genders). There are also verbal classifiers

6
Some languages use noun class prefixes on modifiers, and verbal classifiers with verbs
(Evans forthcoming).
 Different Classifiers in One Language 201
which mark agreement with the S or O constituent (6.3 and 6.4). Verbal
classifiers (or classificatory noun-incorporation) also coexist with noun
class cross-referencing on verbs in Iroquoian and Caddoan languages
(Mithun 1984: 864-8). And in Palikur, gender marking on the verb can
cooccur with verbal classifiers (see 6.29).
Verbal classifiers and predicate-argument agreement in pronominal noun
classes also coexist in a few Lowland Amazonian languages. In Baniwa and
Tariana, predicate-argument agreement in pronominal noun classes is
found with the A/Sa constituent and—only in Baniwa—with O/S0; the
agreement in verbal classifier is restricted to very few grammatical con-
structions; it is always with the subject in Tariana, while in Baniwa it may
be with either subject or object (see Aikhenvald 1995b). In Baniwa, NOUN
CLASSES (obligatorily marked in possessive cross-referencing prefixes) co-
occur with RELATIONAL CLASSIFIERS (see ex. 5.51 and 5.52). NOUN CLASSES
(cross-referencing the argument of an adposition) can cooccur with LOCATIVE
classifiers in Lokono (Arawak: Pet 1987) and Palikur (7.3 and 7.4).
A number of languages employ the same, or almost the same, set of
morphemes in different classifier environments, and also possess a small
system of 'pronominal' genders marked in verbal cross-referencing and
often also on demonstratives (see Chapter 9); they often cooccur. The
following example, from Baniwa (Arawak), shows a possessive construc-
tion which contains a 3rd person feminine cross-referencing marker and
also a classifier suffix, -da 'round objects', used as a possessed classifier:
8.19. inu inasu-da 3u-dza-da
dog Woman-ROUND 3SGF-POSS-CL:ROUND
'A female dog is hers.' (or 'She has a female dog.')

8.4. Conclusions
Several types of classifier—which represent different focal points on the
continuum of noun categorization devices—can coexist in one language. In
languages with a number of different sets of classifiers in varying morpho-
syntactic environments, these can differ in their grammatical properties.
Thus, noun classifiers are optional in Wardaman and Ngan'gityemerri, but
noun class marking is obligatory (cf. Reid 1997). Noun classifiers are
optional and numeral classifiers are obligatory in Sochiapan Chinantec.
In Palikur, verbal classifiers are optional, but classifiers of other types and
noun classes are obligatory.
The attested combinations of different sets of classifiers in different
environments are shown in Table 8.10. Table 8.11 shows the attested
combinations of different classifiers in the same environment.
TABLE 8.10. Different classifier sets in different environments in one language
Noun classes Numeral Noun c1 Verbal Relational Possessor Possessed Locative Deictic Examples of languages

X y - - - - - Dravidian, Indie, Arawak

X y - - - - Wardaman, Ngan'gityemerri
X y - - Ponapean, Mokilese, Truquese
x y - - - - Kanjobal, Minangkabau
x - y - Cora
- y z Toba
X y z - - - - Chinantec
X y - - z - 'Dongo-ko
X — - y - z - Daw
x, y z z z u - z - y Baniwa
x, y z u v y Palikur
Different letters (x, y, z, u, v) refer to the different classifier sets.

TABLE 8.11. Different classifier sets in the same environment in one language
Noun classes Numeral Verbal Relational Locative Examples of languages
classifiers classifiers classifiers classifiers
X x Palikur, Achagua, Machiguenga
X x Gunbarlang, Nunggubuyu, Anindilyakwa,
Iroquoian, Caddoan
X x Baniwa
X x Palikur, Lokono
 Different Classifiers in One Language 203
No language has been found which has seven distinct sets of classifiers
corresponding to all the types of classifiers outlined in Chapters 2-7. The
largest number of formally different sets of classifiers is found in Palikur
(North Arawak); only some of the functionally different classifiers are
identical in their semantics and in their origin.
 9 Multiple Classifier Languages

The same, or almost the same, set of morphemes can be used in more than
one classifier environment. These morphemes may have different gramma-
tical properties, or be more or less obligatory depending on what classifier
environment they are in. We label them 'multiple classifier' systems. Noun
categorization in multiple classifier languages is discussed in §9.1. In §9.2
the multiple classifier systems are compared with noun class agreement on
different targets. In §9.3 we discuss some borderline cases.

9.1. Noun categorization in multiple classifier languages

The same set of classifier morphemes can be used in up to six different

morphosyntactic environments. An example of the same classifier
morphemes used with a numeral, with a deictic, and with an adjective is
9.1, from Kilivila (Austronesian, Trobriand Islands: Senft 1996: 18).
Classifiers are prefixed to the numeral and to the adjective, and infixed
to the demonstrative.
9.1. ke-yu waga ma-ke-si-na
CL:WOODEN-two Canoe this-CL:WOODEN-PL-this
ke-manabweta (le-kota-si)
CL:WOODEN-beautiful (3p.pAST-arrive-PL)
These two beautiful canoes (arrived).'
In 9.2, from Tariana (North Arawak), the classifier -dapana 'HOUSE', is
used with a demonstrative, a numeral, a possessive, an adjective, and the
predicate; it also appears as a part of the head noun 'hospital' itself as a
derivational suffix.
9.2. ha-dapana pa-dapana na-tape-dapana
DEM:INAN-CL:HOUSE One-CL:HOUSE 3PL-medicine-CL:HOUSE
na-ya-dapana hanu-dapana heku
3PL-POSS-CL:HOUSE big-CL:HOUSE WOOd
na-ni-ni-dapana-mahka
3pl-make-TOPIC.ADVANCING.VOICE-CL:HOUSE-RECENT.PAST.NON.VISUAL
This one big hospital of theirs has been made of wood.'
Systems of this kind are attested in numerous South American languages
(e.g. Arawak, Tucano, Guahibo, Tupi, Yagua, Nambiquara), in some
 Multiple Classifier Languages 205
Papuan languages, and in some Austronesian languages, as well as in a
number of the languages of East and Southeast Asia such as Chinese, Thai,
Vietnamese, or Hmong.
The same classifier morphemes in different environments can differ in
their realization and in at least some syntactic properties. In some cases it is
possible to decide that one environment is the 'primary' one. If there is no
reason to take one classifier environment as primary and the others as
secondary, we say that this is just a 'multiple classifier' language.
Languages of agglutinating and isolating profiles tend to use the same
morphemes in several classifier environments more often than do fusional
languages. There are several restrictions on what classifier environments
can be marked with the same morpheme.
(i) If a morpheme is used as a RELATIONAL classifier, it is never used in any
other classifier environment (except for the large systems of classifiers
which combine the properties of relational and possessed classifiers
discussed in §5.5.1).
(ii) If classifiers of any type are fused with the head noun, or with a
modifier, they cannot be used in multiple classifier environments. For
instance, suppletive classificatory verbs in Athabaskan languages are
never used as another type of classifier; in contrast, an affixed verbal
classifier (the so-called 'areal prefix') can be used as a locative classifier in
some North Athabaskan languages (§6.4.2). Similarly, numeral classifiers
which are fused with a numeral in Squamish are not used in other
classifier environments. In Nauru (Micronesian) numeral classifiers are
fused with demonstratives, and are not employed in any other classifier
environment.
(iii) If a language has more than one subtype of a certain type of classifier,
it is likely that only one of these may be used in (one or more) other
classifier environment(s). Thus, in languages of the Arawak family which
have two noun class systems: one for pronouns, and the other for modifiers
of other kinds, only the latter is also used in other classifier environments,
such as numeral and verbal classifiers. 'Pronominal' genders are never used
in other classifier environments (see Table 2.10, §2.7.1).
(iv) There is a correlation between the size of a classifier system, and the
likelihood that it will be used in more than one classifier environment.
Larger classifier systems tend to be used in more than one environment;
such is the case in a few languages of East Asia, such as Chinese, and the
Arawak and Tucano families, as well as in such languages as Munduruku
and Yagua, of South America. In languages with two noun class systems
(§2.7), the larger one tends to be used in other classifier environments; this
happens in numerous South American languages.
206 Classifiers
However, this is by no means true for all languages. Nivkh (Paleosiberian
isolate) and Malto (South Dravidian) have largeish systems of numeral
classifiers used just in this environment. Hmong has only seven classifiers
employed with numerals, and these are also used with demonstratives and
in possessive constructions; the same morphemes are also employed as
noun classifiers. In Eyak (Eyak-Athabaskan) a small set of classifiers is
used as locative and as verbal classifiers.
The maximum number of classifier environments for one set of mor-
phemes established so far is five. Combinations of different classifier envir-
onments for one set of morphemes are shown in Table 9.1.
In some of these languages the same classifiers can display differences in
their properties when used in different environments. These differences may
concern the type of morpheme (e.g. the same morphemes can be infixed in
some classifier environments, and suffixed in others), obligatoriness of use,
and other characteristics. Examples of different multiple classifier lan-
guages are discussed under (A-D). Further functions of classifiers in mul-
tiple classifier languages as derivational affixes and nominalizers are
discussed in (E). The use of repeaters in multiple classifier languages is
discussed in (F). Some conclusions and generalizations are given in (G).
The question of whether one classifier environment can be considered
'primary' is dealt with under (H).
(A) One set of morphemes in two environments
The same classifier morphemes are often used with numerals and with
demonstratives.1 This is found in many languages of East and Southeast
Asia including Mandarin Chinese (Li and Thompson 1981: 104, 112). 9.3
shows classifier ge in a numeral phrase; 9.4 shows the same classifier in a
demonstrative phrase.
9.3. san ge ren
three CL person
'three people'
9.4. nei-ge cai
that-CL course of food
'that course of food'
A demonstrative can cooccur with a numeral in a classifier construction,
and there is then also one occurrence of the classifier, e.g. 9.5.

1
Cf. Craig (forthcoming: 24): 'this type (numeral) of classifiers may also appear on other
elements than numerals, i.e. . . . demonstratives.1
TABLE 9.1. Same set of classifiers in several environments (A-D)
Adjectival Numeral Noun Verbal Possessed Locative Deictic Examples of languages
modifier classifier classifier classifier classifier classifier classifier
A _ X _ _ -_ X Mandarin Chinese
- X X - - Chayahuita, Anindilyakwa
- - X - X - Eyak, Koyukon
X - X - - Terena, Yawalapiti
B X X - - - X Kilivila
- X X - - X Vietnamese, Awara
- X - - X - X Nung
X X - - - - X Newari
X X X - - Waura, Ignaciano, Machiguenga
C X X X - - - X Thai
X X - - X X Cantonese, Dulong-Rawang
- X X - X X Hmong
X X X - X Yagua, Munduruku, Waorani
X X - - X X Tucano, Tuyuca
D X X X X X Kubeo, Nasioi, Motuna
208 Classifiers
9.5. nei san ge ren
that three CL person
'those three people' (Helen Charters, p.c.)
Classifiers show certain differences in behaviour, when used with numer-
als or other quantifying expressions, and when used with demonstratives.
The plural suffix -men, which occurs on some nouns with human reference,
can occur with the plural classifier xie following a demonstrative but not
following a numeral (other than 'one') (Charters 1995). Example 9.6 is
grammatical; 9.6a is not.
9.6. nei xie haizi-men
that CL:PLURAL child-PL
'those children'
9.6a. *san xie haizi-men
three CL: PLURAL child-PL
Classifiers are optional with demonstratives if the head noun has a
human referent while numeral classifiers are always obligatory (Helen
Charters, p.c.). It will have been noted that ge 'generic classifier' is written
as a separate word in 9.3 and in 9.5, where it is used as a numeral classifier;
but in 9.4 it is written as a suffix to the demonstrative. This reflects the
intuition of speakers; they have a 'feeling' that a classifier forms a closer unit
with a demonstrative than it does with a numeral (Helen Charters, p.c.).2
The same morphemes may be used as NUMERAL CLASSIFIERS and as
VERBAL CLASSIFIERS in a few languages. An example of the same set of
morphemes used with numerals and with verbs comes from Chayahuita
(Cahuapanan: Derbyshire and Payne 1990: 258-9). Example 9.7 illustrates
rin 'long, flexible' as a numeral classifier, and 9.8 shows the same
morpheme as a verbal classifier.
9.7. cara-rin
three-CL:LONG. FLEXIBLE
'three' (e.g. pieces of string, vine)
9.8. i'sho-rin-in
peel-CL:LONG.FLEXIBLE-3SG
'He peeled vines'
There is no information concerning any differences in the behaviour of
the two types.
The same morphemes are used in the functions of LOCATIVE CLASSIFIER
and VERBAL CLASSIFIER in very few languages. The areal prefix in Koyukon is

2
Similar usage was observed in Wei Ning (Wang 1972); see Bisang (forthcoming).
 Multiple Classifier Languages 209
used as a verbal classifier (see example 6.34), and as a locative classifier (see
example 6.35).
The same classifiers are used with locational expressions and with verbs
in Eyak (Eyak-Athabaskan: Krauss 1968: 195) exemplified in Table 9.2.
Classifiers in Eyak are not semantically transparent.

TABLE 9.2. Locative and verbal classifiers in Eyak

Classifier Locative classifier Verbal classifiers
Residual ?uq 'on it (stick)' (dkh) ?u'd s t a h l 'it (a stick)
lies there'
d-classifier ?uda*q 'on it' (e.g., a board) ( c i ) ?u*d desestahl 'it (a board)
lies there'
xd-classifier uxeda*q 'on it' (e.g. on a log) (Ite'sK) ?u'd xedesetahl 'it (a
log) lies there'
ti-classifier ?uti'na?q 'on it' (e.g. a leaf, ( K u f t a h l ?u*d ti'setahl 'it (a
feather) feather, leaf) lies there'

The same set of morphemes are used with verbs and with adjectival
modifiers to mark agreement in Terena (South Arawak) and Yawalapiti
(Xinguan Arawak).
9.9 shows how pu'i 'CL:ROUND' is used as a noun class marker on an
adjective in Terena (Ekdahl and Butler 1979; Aikhenvald 1996a). The same
classifier used with a verb to characterize the O is given in 9.10 (= 6.8).
9.9. tuti puru-pu'i
head big-CL:ROUND
'a round head'
9.10. oye-pu'i-co-ti
COOK-CL:ROUND-THEME-PROGR
'He is cooking (round things).'
9.11, from Yawalapiti, illustrates the use of classifier -pana 'CL:LEAF.LIKE'
as a noun class marker, with the adjective 'green'. Note that the demon-
strative ifutifa 'that+FEM' does not take a classifier.
9.11. i utia ata-pana irula-pana
that + FEM tree-CL:LEAF.LIKE green-CL:LEAF.LIKE
'that green leaf
In 9.12 -pana 'CL:LEAF.LIKE' is used on the verb, to refer to the S.
9.12. ata-pana kuka iu kama-pana
tree-CL: LEAF LIKE PAST DEM + FEM die-CL:LEAF.LIKE
'The leaf which died (i.e. a dry leaf) is there.'
210 Classifiers
Terena is one of the few Arawak languages which have lost the gender
opposition in the pronominal system. The difference between classifier
systems in Terena and Yawalapiti is that Yawalapiti distinguishes feminine
and non-feminine genders in demonstratives and third person pronouns
(Mujica 1992; Seki and Aikhenvald forthcoming).
Neither Terena nor Yawalapiti uses classifiers with numerals. This may
be an innovation, since the same morphemes are used as numeral classifiers
in other, closely related languages, e.g. Ignaciano (South Arawak) and
Waura (Xinguan Arawak); these are considered under (B) below.
No other combinations of classifiers in just two environments have so far
been found.
(B) One set of morphemes in three classifier environments
The same morphemes are used in the function of NOUN CLASS agreement
markers with adjectives and with demonstratives, and also with numerals in
Kilivila (Austronesian) (Lawton 1993; Senft 1986; 1996).
Kilivila has about 200 classifiers (called 'classificatory particles' by Senft
1986; 1997, following Malinowski 1920). It appears to be difficult to
establish the exact number of classifiers in this language, because nouns
with inanimate referents can be used as 'repeaters'; which makes classifiers
a virtually open class (Gunter Senft, p.c.). In 9.13 (Senft 1986: 77) a
classifier is used with a numeral, and in 9.14 (Senft 1986: 64) with a
demonstrative. Classifiers used with demonstratives are infixes; they are
prefixes in all the other environments.
9.13. te-tala tau
CL:MAN-one man
'one man'
9.14. mi-na-na vivila
this-CL:WOMAN-this woman
'this woman'
Adjectives fall into three classes, depending on whether they can be used
with classifiers, and whether this is obligatory. 9.15 illustrates an adjective
which requires a classifier (Senft 1986: 85).
9.15. valu kwe-manabweta
village CL:THING-beautiful
'beautiful village'
Classifiers are used with all the numerals; in contrast, not all adjectives
take classifiers.3
3
Lawton (1993) reports that classifiers are used with demonstratives more frequently than
in other environments; see criticism of this statement by Senft (1996: 179).
 Multiple Classifier Languages 211
The same set of morphemes is used with numerals, with demonstratives
and as noun classifiers in Vietnamese (Goral 1978: 11 ff). This situation is
fairly typical for languages of Southeast Asia (Goral 1978). Example 9.16
illustrates a numeral classifier in Vietnamese.
9.16. ba cuon sach
three CL:BOOK book
'three books'
The first occurrence of cuon 'CL:BOOK' in 9.17 is as a noun classifier, and
the second is as a deictic classifier.
9.17. anh muon cuon sach nao cuon kia
you want CL:BOOK book which CL:BOOK that
'Which book do you want? That one.'
A 'generic' noun classifier is exemplified in 9.18 (Lobel forthcoming: 13-
24).
9.18. cai cay
CL:NON.LIVING.THINGS tree/plant
'a tree/plant'
All the classifiers are used for referent tracking and marking definiteness
(see §12.1.3, and also Goral 1978: 14, and Lobel forthcoming). Unlike
classifiers with numerals, classifiers are not obligatory with demonstratives
(Goral 1978: 15). Generic classifiers are often omitted in general state-
ments, e.g. in proverbs (Goral 1978: 14):
9.19. meo so chuot
cat fear mouse
'Cats are afraid of mice.'
Awara, a Papuan language from the Morobe province in New Guinea,
appears to have a somewhat similar system (Susan Quigley, p.c.). Classi-
fiers are prefixed to the numbers 'one' and 'two' and suffixed to demon-
stratives; they appear to be obligatory with numbers, but not with
demonstratives. When used as noun classifiers, they are independent
words.
The same morphemes are used with numerals, demonstratives and in
possessive constructions in Nung (Tai: Saul and Wilson 1980: 25 ff.). There
are four classifiers. The generic classifier ahn as a numeral classifier was
illustrated in 4.38. The use of of ahn as a possessed classifier in a predicative
construction (the head noun is omitted) is shown in 9.20. Example 9.21
illustrates the 'animate' classifier tu in a non-predicative possessive con-
struction. The classifier NP is in square brackets.
212 Classifiers
9.20. mi su' [ahn hau]
not correct CL:GEN myself
'It's really not mine.'
9.21. ha [slong tu luhc bao] mu'hn va
tell two CL:ANIM child boy he say
'Tell his two sons that . . .'
9.22 illustrates tu 'animate classifier' with a demonstrative.
9.22. leo [tu te] chihng ma
then CL:ANIM that then come
'Then that one came.'
Similarly to Vietnamese, classifiers can occur with nouns, e.g. tu me
(CL:ANIMATE wife) 'the wife'.
Classifiers with numerals are used somewhat differently from classifiers in
other environments. They are optional with powers of ten; no such restric-
tions are found for other classifier environments (Saul and Wilson 1980: 27).4
Newari (Tibeto-Burman: Bhaskararao and Joshi 1985) has 53 classifiers
used with numerals and interrogative quantifiers ('how much', 'how many')
(9.23), with demonstratives (9.24) and with two adjectival modifiers which
refer to dimensions (i.e. 'big' and 'small', as in 9.25). Classifiers tend not to
be used with higher numerals (Bhaskakarao and Joshi 1985: 24); in other
environments they are obligatory.
9.23. ni-mhA khica
two-CL:ANIMATE dog
'two dogs'
9.24. thwA:-ma swa
this.many-CL:PLANT plant
'this many plants'
9.25. tA:-pa-gu mAri
big-CL:SWEET.FOOD.ITEMS-CONNECTOR bread
'big bread'
The same nouns can require different classifiers when used with a
numeral, or with an adjectival modifier. Nouns which normally require
a general classifier -gu take the classifier rhA when accompanied by an
adjectival modifier 'big' or 'small'. In 9.26, -gu is a classifier on a number
ni-gu (note that in tA:-rhA-gu gu is a homophonous morpheme, a con-
nector).
4
Bisang (forthcoming) discusses further morphosyntactic differences between classifiers
used with numbers and with demonstratives in Chinese, Vietnamese, Thai, and a number of
Miao-Yao languages.
 Multiple Classifier Languages 213
9.26. ni-gu tA:-rhA-gu sAphu:
two-CL:GEN big-CL:GEN-CONNECTOR book
'two big books'
One set of morphemes is used as agreement markers on adjectival
modifiers, with numbers, and with verbs (similarly to verbal classifiers) in
a few South American Indian languages, e.g. Waura (Xinguan Arawak).
9.27 illustrates a numeral classifier in Waura (Jackson 1966). Example 9.28
shows a classifier as an agreement marker on an adjective, and 9.29 features
a verbal classifier (Richards 1973).
9.27. mepiawa-pa ita
two-CL:POINTED horn
'two horns'
9.28. atakahi ityula-pana
grass green-CL:LEAF.LIKE
'green grass'
9.29. i-tsitya-pi-tsa
CAUS-entwine-CL:LINEAR-CAUS
'entwine linear objects'
A similar system is found in Ignaciano (South Arawak) and in Machi-
guenga (Peruvian Arawak: Shepard 1997). The two-way gender distinction
in Machiguenga was discussed in §8.3. The difference between Waura and
Ignaciano is that Waura employs the masculine and feminine distinction
only for demonstratives (see §7.3.2); Ignaciano has a three-way masculine,
feminine, and inanimate distinction in third person pronouns, cross-refer-
encing affixes and demonstratives (Ott and Ott 1983; Aikhenvald 1996b).
There is not enough information on classifiers in Waura to evaluate the
possible differences in behaviour of classifiers of different types. In Machi-
guenga, classifiers are infixes to numbers and suffixes in other environ-
ments.

(C) One set of morphemes in four environments

The same set of morphemes is used with numerals, demonstratives, a
limited class of adjectival modifiers and also as noun classifiers in Thai.
Example 9.30 illustrates the use of a classifier with a numeral, 9.31 with a
demonstrative, and 9.32 with an adjectival modifier (referring to dimension
and colour) (Hundius and Kolver 1983).
9.30. rom saam khan
umbrella three CL: LONG. HANDLED
'three umbrellas'
214 Classifiers
9.31. rom khan nii
umbrella CL:LONG.HANDLED this
'this umbrella'
9.32. nok tua jaj
bird CL:BODY big
'the big bird'
The order is numeral-classifier, classifier-demonstrative, and classifier-
adjective. Further restrictions of ordering in classifier constructions in Thai
are considered by Hundius and Kolver (1983: 177-81). When there are
several adjectives or demonstrative modifiers, the classifier is repeated with
each, as in 9.33 and 9.34.
9.33. nok tua sii-khiaw tua jaj
bird CL:BODY green CL:BODY big
'the big green bird'
9.34. rom khan sii-khiaw
umbrella CL:ILONG.HANDLED green
khan jaj khan nii
CL:LONG.HANDLED big CL:LONG.HANDLED this
'this big green umbrella'
In contrast, the classifier is used only once in a numeral noun phrase, as
shown in 9.35.
9.35. rom saam khan nii
umbrella three CL::LONG.HANDLED this
'these three umbrellas'
Noun classifiers are optional, and they are used to distinguish different
meanings of the same noun, e.g. bay-phluu (CL:LEAF-betel) 'betel leaf, ton-
phluu (CL:VINE-betel) 'betel vine' (DeLancey 1986: 440). They have been
frequently interpreted as 'class nouns', e.g. bay-maay (CL:LEAF-wood) 'leaf
(DeLancey 1986: 440-1).
Hundius and Kolver (1983: 169, 172) point out that classifiers are
optional with adjectives and with demonstratives; they are obligatory
with numerals. They attempt to provide a unified account for classifier
uses, arguing for the 'primary' character of classifiers used with numerals.
We return to this in (H) below.
In Cantonese classifiers are used with numerals and in quantifiying
expressions, with demonstratives, with a limited class of 'size' adjectives
and in possessive constructions (see Pacioni 1997; forthcoming). While
classifiers are obligatory with numerals and in quantifying expressions
and with demonstratives, they appear to be optional in other contexts
 Multiple Classifier Languages 215
where their presence or absence correlates with the specificity of the refer-
ent. Thus, these forms in Cantonese can be said to have two functions: that
of numeral classifiers, and that of 'specificity' markers—which span a
number of morphosyntactic environments—but are most regularly found
as deictic classifiers.5 Similarly, in Dulong-Rawang (Tibeto-Burman:
Randy LaPolla, p.c.) classifiers are obligatory with numerals. They are
used with nouns (as noun classifiers) to indicate specificity; they may
optionally appear with possessives and with most adjectives in the same
function—as specificity markers. Their primary function is that of numeral
classifiers, and as deictic classifiers (also see (H) below).
Hmong (Miao-Yao: Bisang 1993; Jaisser 1987) uses the same set of
morphemes with numerals (9.36), demonstratives (9.37), in possessive con-
structions and as noun classifiers.6 The classifier follows the number in
9.36, and precedes the noun—followed by a demonstrative.
9.36. Lawv muaj rau tus me nyuam
they have six CL:LIVING.BEING child
'They have six children' (Jaisser 1987: 172)
9.37. lub tsev no
CL:OBJECT house this
'this house' (Jaisser 1987: 171)
In 9.38 the classifier is used in a possessive construction to characterize
the possessed noun.
9.38. nws rab riam ntaj
he CL:MAN.MADE.OBJECT sword
'his sword' (Bisang 1993: 29)
Classifiers can be used with nouns by themselves as noun classifiers, as
shown in 9.39 (Bisang 1993: 20-2).
9.39. zaj teev ntuj
CL:STORY,POEM pray heaven
'prayer'
The use of noun classifiers correlates with the definiteness of the refer-
ent—cf. 9.40 (Jaisser 1987: 171). The omission of a classifier is ungramma-
tical if a definite referent is implied (9.40a). (In actual fact, the correlation
5
A similar situation has been reported for Yue dialects, Hakka, and Hokkien: see Pacioni
(1997). According to Bisang (1993: 30) a similar system is found in other Miao-Yao languages
of China.
6
Conklin (1981: 187) showed that the numeral classifier morphemes are used with demon-
stratives, possessives, adjectival modifiers, and as noun classifiers in White Tai and Dioi; their
frequency of use varies. While in Dioi classifiers are usually employed in all these environments,
in White Tai they appear 'at least sometimes'.
216 Classifiers
between degrees of definiteness and the use of classifiers in Hmong is not
that simple; see Bisang 1993: 26 ff., for details.)
9.40. tus tsov tshaib tshaib plab
CL:LIVING.BEING tiger be.hungry be.hungry stomach
'The tiger was very hungry.'
9.40a. *tsov tshaib tshaib plab
The same classifier can be used twice in the same noun phrase thus
forming several classifier constructions, as in 9.41. The classifier may be
used just once, as in 9.42 where the classifier zaj is employed with a numeral
and as a noun classifier (Bisang 1993: 21).
9.41. ob zaj zaj tshoob
TWO CL:STORY,POEM CL:STORY,POEM SONGS
'two wedding songs'
9.42. nws hais ib zaj teev ntuj
he say one CL:STORY,POEM pray heaven
'He is praying once (one prayer).'
Unlike classifiers in other environments, possessed classifiers in Hmong
are used differently depending on whether the head noun is alienably or
inalienably possessed (see 5.22 and 5.23).
The same morphemes are used in the function of agreement marker on
adjectives, numeral classifiers, verbal classifiers, and deictic classifiers in a
few languages of Lowland Amazonia. One such example is Mundurukii
(Tupi: Goncalves 1987; Crofts 1973; 1985). Mundurukii has over 100
classifiers, at least half of which coincide, or show resemblances with, a
corresponding nominal root (see §13.1, on how classifiers could originate
from full nouns). Examples 6.9 and 6.10 above illustrate the classifier ba4
long rigid object' as a verbal classifier, and as a derivational suffix on a
noun itself, in a2KO3-ba4 'banana'. 9.43 shows ba4 as a numeral classifier;
9.44 shows the same morpheme as a deictic classifier (Goncalves 1987: 35).
(In the examples below, pa4 and ba2 are related by regular phonological
processes described in Goncalves 1987: 64.)
9.43. xep3xep3-pa4 a2ko3-ba4
TWO-CL:LONG.RIGID banana-CL:LONG.RIGID
'two bananas'
9.44. J2ja3-ba2 a2ko3-ba4
this-CL: LONG.RIGID banana-CL:LONGRIGID
'this banana'
 Multiple Classifier Languages 217
4
9.45 illustrates ba as a noun class marker with a descriptive verb in
attributive function (example from Crofts 1973: 87).7
9.45. a2ko3-ba4 i3-ba4-dip3
banana-CL:LONG.RIGID 3pRON-CL:LONG.RIGID-beautiful
'a beautiful banana'
The same morphemes in different classifier environments show morpho-
logical differences. Classifiers in Mundurukii fall into two groups according
to their position in the morphological word. Those used with numerals and
demonstratives and as derivational markers on nouns are suffixes; verbal
classifiers and markers on adjectival modifiers are prefixes.
In Yagua a single set of classifier morphemes can be used in the same
environments as in Mundurukii. Yagua has around forty classifiers (Payne
and Payne 1990; Payne 1990). 9.46 illustrates a verbal classifier, 9.47 a
deictic classifier.
9.46. suduu-bii-numaa navaa-bii
be. ripe-CL:SPROUT-now banana-CL:SPROUT
'The stalk of bananas is now ripe.' (Derbyshire and Payne 1990: 254)
9.47. jiy-see muuy-see
DEM-CL:STICK Write-CL:STICK
'this pencil' (Payne 1990: 135)
Classifiers used as agreement markers on adjectives, as in 9.48, differ
from other classifiers. In this context, classifiers are said to be optional
(Payne 1990: 133). Both 9.48 and 9.49 are acceptable.
9.48. rab'ii runay-bii 'its red flower'
its.flower red-CL:FLOWER
9.49. rabi'i runay 'its red flower'
its.flower red
Classifiers are obligatorily infixed to the numerals 'one' and 'two' (see
example 4.21) but suffixed to larger numerals, e.g. vuyajuuy-nu niinu
(ten-CL:POLE tree) 'ten trees' (Payne and Payne 1990: 349).
Waorani, an isolate from Equador (Derbyshire and Payne 1990: 259;
Peeke 1968) has 35 classifier affixes used with numerals, adjectives, demon-
stratives and verbs. Classifiers are suffixes, when used with demonstratives
(9.50), with adjectives (9.51), and with verbs (9.52).

7
Mundurukii, apparently, does not have a morphological class of adjectives; descriptive
verbs are used as modifiers in noun phrases.
218 Classifiers
9.51. giyie-ka 'small stone'
Sniall-CL:STONE

9.52. ko-wa 'to pierce a foot'

pierce-CL:FOOT
When used with numerals, classifiers are infixes (9.53), similarly to Yagua.
9.53. ado-ba-ke
One-CL:PALM.LEAF-One
'one palm leaf
Thus, in Yagua, Waorani, and Mundurukii the same morphemes behave
differently in different classifier environments. In Mundurukii, noun class
markers and verbal classifiers occupy a position in the word different from
other classifiers. In Yagua, noun class markers are not obligatory, unlike
other classifier types; in Yagua and Waorani, classifiers are infixes with
(some) numerals, but suffixes otherwise.
The same set of morphemes is used as agreement markers on adjectives
and demonstratives, with numerals and in possessive constructions in
numerous East Tucano languages.8 These languages have a virtually open
set of classifiers used with inanimate nouns. A gender-like distinction
(masculine vs. feminine) applies to animate nouns. Examples 9.54-7 illus-
trate the use of classifiers in Tucano as adjectival agreement markers, as
numeral classifiers, as possessed classifiers, and as deictic classifiers respec-
tively, as well as use on the noun itself (West 1980; my data). A similar
system has been described for Tuyuca by Barnes (1990).
9.54. pino phai-gi
snake big-CL:ANIM.MASC
'a big snake'
9.55. ni'ca-wa watique'a
one-CL:EXTENDED.FORM manioc.squeczer
'one manioc squeezer'
9.56. ati-wi'i numio-ya-wi'i
this-CL: HOUSE Woman-POSS-CL:HOUSE
'this house is a woman's; woman's house'

8
Papuan languages of Central Bougainville appear to also use the same classifier mor-
phemes as noun class markers, numeral classifiers, possessed and deictic classifiers, and also as
noun classifiers in nominalizations (Kim Blewett, p.c.). At least some of these functions are
combined in multiple classifier system of Reef-Santa Cruzan languages spoken in the
Solomons (Wurm 1981; 1987; 1992a; 1992b; p.c.). More studies are needed on multiple
classifiers in languages of the South Pacific, such as Award and Wantoat (Morobe province),
Angan languages (Angave, Taenae), and the two non-Austronesian languages of East New
Britain, Baining and Taulil.
 Multiple Classifier Languages 219
9.57. ati-gl (pu-gi)
DEM:INAN-CL:LARGE (hammock-CL:LARGE)
'this hammock'
In East Tucano languages, classifiers are not used with finite verbs.
However, they can be suffixed to nominalized verbs, to mark the predicate
of a relative clause. This is illustrated with 3.19, from Tuyuca (Barnes*
1990).
(D) One morpheme in five environments
Only rather rarely is the same set of morphemes used as agreement markers
on adjectives and demonstratives, with numerals, in possessive construc-
tions and with verbs. One such language is Kubeo (Central Tucano:
Gomez-Imbert 1996).
Kubeo has a system of classifiers which is very similar to East Tucano
languages, exemplified in 9.54–7. Unlike them, Kubeo also uses classifiers
with main clause predicates. The classifier -ki is used in 9.58 with a demon-
strative and with a verb.9
9.58. i-ki hoe-ki bea-ki-bu
this-CL:CYLINDR axe-CL:CYLINDR good-CL:CYLINDR-INAN
'This axe is good.'
Nasioi (Papuan of Southern Bougainville: Hurd 1977) has over a hun-
dred classifiers used in the same environments as in Kubeo. A possessed
classifier in Nasioi is shown in 5.21, and a verbal classifier in note 7 to
Chapter 6. Example 9.59 illustrates classifiers with a numeral and with a
demonstrative.
9.59. nto-na-ru' bee-ru'-pi
Water-DER.SUFF-CL:UNIT.OF.LIQUID three-CL:UNIT.OF.LIQUID-PL
a-ru'-daang
this-CL:UNIT.OF.LIQUID-inland
'These three inland lakes.'
In 9.60, a classifier marks the agreement of an adjective with a head
noun.
9.60. tamp-a-u'
good-DER.SUFF-CL:TARO
'a good (taro)' (Hurd 1977: 132)
Motuna, from the same language family, has 51 classifiers used in the
same environments as Nasioi (Onishi 1994). A possessive classifier in
9
The verbal classifiers in Kubeo, unusual from a Tucano perspective, are due to the areal
influence of Baniwa (see Gomez-Imbert 1996).
220 Classifiers
Nasioi is illustrated in 5.21; verbal classifiers are discussed in note 7 to
Chapter 6. Example 9.61 (Onishi 1994: 163) illustrates a classifier used with
a numeral.
9.61. no-uru
One-CL:HUMAN
'one human'
Classifiers are also used with several determiners, such as the demon-
strative o-, the article ti-, muuko 'other', and jee 'what'. 9.62 (Onishi 1994:
164) illustrates a classifier with a demonstrative.
9.62. o-'ri
DEM-CL:ROUND
'this round object' (Malayan apple)
A few adjectives require a classifier in attributive constructions, as in
9.63 (Onishi 1994: 173) (in Motuna, classifiers are only obligatory with
some adjectives).
9.63. tii miru mohko-muru
ART:DIM String short-CL:PIECE.OF.LONG.OBJECT
'a/the short piece of string'
It is possible that classifier morphemes in Nasioi and Motuna may also be
used as possessor classifiers (see 5.47); this requires further investigation.
(E) Derivational functions of classifiers in multiple classifier systems
Classifiers in multiple classifier languages are often used as derivational
affixes and as nominalizers (see Payne 1990, for a discussion of the role of
classifiers in Yagua, a multiple classifier language, in inflection and deriva-
tion). Examples of classifiers as derivational devices in Tariana (North
Arawak) were given in §3.5. In this language several classifiers can cooccur
on a head noun; this is a productive derivational process, e.g. kara-ka-
whya (REL+fly-THEMATic-CL:CANOE) 'plane', kara-ka-whya-puna (REL+fly-
THEMATic-CL:CANOE-CL:STRETCH) 'airstrip'. A similar use of classifiers is
attested in Waura (Xinguan Arawak: (B) above), e.g. kunuma-tai (cotton-
CL:CURVED) 'thread' (Jackson and Richards 1966), cf. Yagua ruu-dasiy
(blow-CL:THIN,POLE) 'blow gun' (Payne and Payne 1990: 446); cf. pu-gt
(hammock-CL:LARGE) 'a hammock' in 9.57, from Tucano. In Terena, the
same set of morphemes is used on verbs and as agreement markers on
adjectives (see 9.9 and 9.10); they are also used as derivational suffixes, e.g.
ope-pu'i (bone-CL:ROUND) 'skull'.
Classifiers employed as derivational markers on the head noun can trans-
form mass or collective nouns into countable nouns. This happens in a
number of North Amazonian languages, including Guahibo, Bora-Witoto,
 Multiple Classifier Languages 221
Tucano, and North Arawak, e.g. Tariana heku 'wood', heku-na (wood-
CLrVERi) 'tree', heku-da (wood-CL:ROUND) 'fruit'. Only a noun which con-
tains a classifier can be pluralized with a suffix -pe, e.g. heku-na-pe 'trees',
heku-da-pe 'fruits', but not *heku-pe (see also West 1980, for Tucano; Kerr
1995, for Cuiba (Guahibo); Thiesen 1996, for Bora (Bora-Witoto family)).
Classifiers often mark nominalizations and predicates of relative clauses
(cf. 3.19, from Tuyuca, an East Tucano language). In 9.64, from Yagua
(Peba-Yagua: Payne 1990: 132), the classifier -ra, 'neuter', derives a noun
from an inherently verbal root (this example is given in underlying form).
9.64. mach99~-ra-numaa riy-rooriy
remain-CL:NEUT-now 3pL-house
'Their house was what remained' (lit. their house was now a remaining
thing).
In Waorani classifiers function as deverbal nominalizers, as in 9.65.
9.65. 6ki-be
3so.make-CL:VINE
'string which he will make'
Example 9.66 demonstrates nominalization with a classifier in Tucano.
The verb 'be big' is nominalized with -ri, and then the classifier, of a
repeater type, is attached (West 1980: 195-6).
9.66. pa-ri-sawero waa-yiro
be.big-SG.NOM-CL:EAR.SHAPED go-INAN.PAST. EVIDENTIAL
'An ear-shaped area that was big was left (in the manioc bread after
a piece had been torn out)'.
Classifiers can also be used as nominalizers in Motuna (Masa Onishi,
p.c.), as shown in 9.67.
9.67. ong poti kongsi' haaro'ko-no-mori
DEM:MASC time mangO fall + PRES-LINKER-CL:SEASON
roki manni tokotokohah
really hot
The season when mangoes fall is really hot.'
Nambiquara (Nambiquara family: Lowe 1999) has fifteen classifiers
used in several distinct environments (their form and semantics are given
in Diagram 11.10). They are prefixed to numbers, as in 9.68, but suffixed to
adjectives as agreement markers, as in 9.69.
9.68. a3la3a2 so11?i3 ki3-ha1li1 hut3tit3-ta'-he3-ra2
parrot only CL:ROUND-two shoot-1 SG-PAST-PERFECTIVE
'I shot only two parrots.'
222 Classifiers
9.69 wa2la2 wi3win3-ka3lo3-a2
cloth blue-CL:FLAT.SHEET-definite
'the blue cloth'
Classifiers are also used as derivational suffixes on nouns, e.g huk-en'
-su2 (shooter-CL:HOLE.LiKE-indefinite) 'shotgun' (the gun barrel being a
hole), huk3-ki3-su2 (shooter-CL:ROUND-indefinite) 'bow' (the arc of the
bow is round). They are also used as relativizers, e.g. s?i2ha2 ?yau3 -ain1-
thi3 -na2 (house live-they-CL:HOUSE-DEFINITE) 'the house they live in'; wan3ta2
e3e3-k?i2-sain'-jau?3-ai2na2 (word speak-to-they.to.me-CL:WORD-this.defi-
nite) 'this word that they spoke to me'.
We saw in §3.5 that classifiers realized as independent words can also be
used to mark relative clauses, in a function similar to relative pronouns, e.g.
3.14, from Lao.
(F) Repeaters in multiple classifier environments
Repeaters are used in multiple classifier environments in at least three
groups of languages of Lowland Amazonia: (1) the East Tucano sub-
group,10 (2) Tariana (North Arawak), and (3) the Guahibo family. Their
choice depends on the morphological structure of the noun; it has
pragmatic effect.
(F1) Repeaters in the East Tucano languages The same set of classifier
morphemes is used with numbers, with demonstratives, in possessive con-
structions, and as derivational markers and nominalizers (see (C) and (E)
above); there are also three genders in cross-referencing.
Nouns fall into three morpho-semantic classes according to what classi-
fier morphemes are used with each. Examples below are from the Tucano
language.
(i) Animate nouns divide into feminine and non-feminine; agreement suf-
fixes are -gi 'non-feminine' (including 'sun/moon'), -go 'feminine', e.g. pino
phai-gi (snake big-CL:AN.NON.FEM) 'a big snake'.
(ii) Inanimate nouns have one of the following affixes: -kal-ga 'spherical';
-ti, -ri 'containers, e.g. pots'; -kilgi 'long, large'; -pil-wi 'transport'; -pal-wa
'extended'; -poro 'banana-like'; -phi 'long, e.g. knife'; -ro 'general object';
-pe 'hole, seed'; -kwi 'plain'; -ra 'plain (water)'; -pa 'plain (land)'; -se
'uncountable' (West 1980: 119; my field data). They use the same affixes
as agreement markers, e.g. ati-gi pu-gi (DEM:INAN-CL:LARGE hammock-
CL:LARGE) 'this hammock'.
10
This may also apply to West Tucano, but data is lacking. Repeaters are also found in
Yagua (Payne and Payne 1990: 448); however, the data and descriptions are not sufficient to
analyse them; and possibly in Bora-Witoto languages of Northeastern Peru and adjacent areas
of Colombia (Thiesen 1996; Wise 1999).
 Multiple Classifier Languages 223
(iii) To mark agreement with all other nouns with inanimate referents, the
whole of the noun is repeated on the modifier, e.g. 9.70 and 9.71.
9.70. ati-wi'i numio-ya-wi'i
this-CL:HOUSE Woman-POSS-CL:HOUSE
'this house is a woman's; a woman's house'
9.71. (kahsero) phairi-kahsero
bark big-CL:BARK
'big (piece of) bark'
If a noun is a compound, or contains a derivational suffix, the last two
syllables or the suffix is used for agreement, e.g. 9.72.
9.72. (yuhki-gi-dihpj) phairi-dihpi
tree-CL:LARGE-branch big-CL:BRANCH
'a big branch'

(F2) Tariana Repeaters in Tariana (North Arawak) are used to mark

agreement on modifiers (noun classes), demonstratives (deictic classifiers),
and verbs (verbal classifiers) as an alternative to non-repeater classifiers
(some of which are listed in Table 9.6) under three conditions:
(i) the noun must have an inanimate referent;
(ii) the noun must not contain a derivational suffix;
(iii) the noun must be in contrastive focus in discourse (see Aikhenvald
1994a, for further discussion).11
In 9.73, a regular classifier -dapana 'CL:HABITAT' is used when the noun
'house' is not in focus. The head noun is omitted, as often happens in
Tariana discourse (see §12.1.2).
9.73. nuha hnuta nu-dia nu-a-de du-a-pidana
I 1SG+take I SG-return 1SG-go-FUT 3SGNF-say-REM.p.iNFR
nu-ya-dapana-se du-a-pidana namu
1 SG-POSS-CL:HABITAT-LOC 3SGF-Say-REM.P.INFR evil.Spirit
i-sa-du
INDEF-Spouse-FEM
'I shall take you back to my house, said the wife of the evil spirit.'
In 9.74 the same noun, 'house', is in contrastive focus, and so the
repeater (identical to the noun panisi) is used. The head noun is omitted.

1
' In Kilivila, repeaters are used as an alternative 'way' of classification, to indicate more
emphasis of the 'expressed nominal concept' (Gunter Senft, p.c.).
224 Classifiers
9.74. kayi du-ni dhuta du-dia du-a-pidana
so 3SGF-do 3sgF+take 3SGF-return 3SGF-go-REM.p.iNFR
du-ya-panisi-se
3SGF-POSS-CL:HOUSE-LOC
'After she did so, she took (them) back to her very house.'
If a noun is derived, its derivational suffix is used as an agreement
marker, e.g. hala-yawa hanu-yawa (open-DER.AFF:HOLE big-CL:HOLE) 'a big
hole'.
(F3) Guahibo languages These languages have large complex systems of
classifiers. In Cuiba (Kerr 1995) the set of classifiers is virtually open, due
to the existence of repeaters. The same morphemes are used as noun class
agreement markers on adjectives, e.g. peru-nae (old-CL:WOODEN.THINGs) 'old
(canoe)'; with numerals, e.g. cae-bo (one-CL:HOUSE) 'one (house)'; with
demonstratives, e.g. barapo-bo (this-CL:HOUSE) 'this (house)' (Kerr 1995:
143), and in possessive constructions, e.g. piya-nae jera (3SMASC.SG.POSS-
OUWOODEN.THINGS canoe) 'his canoe' (Kerr 1995: 134).
Unlike Tucano languages and Tariana, the use of repeaters is confined
to inalienably possessed nouns (Kerr 1995: 132, 336-7), e.g. cou 'track',
pe-cou (3sG.MASc.poss-CL:TRACK) 'his track'. More work is needed to
describe them fully.
The examples of repeaters discussed above show that:
(i) repeaters are usually only one of the mechanisms of marking agree-
ment; no example has been found so far of a language where this is the only
way of marking;
(ii) repeaters always have semantic, morphological, and sometimes
discourse constraints; they are often used with inanimates only.
(G) Conclusions and generalizations
We have seen that the same set of classifier morphemes can be used in up to
five environments. Classifier morphemes often display differences in their
morphological form and in how obligatory they are in different contexts.
Differences in the behaviour of the same12 set of morphemes in different
classifier constructions are summarized in Table 9.3.
Classifiers used with numerals are a different morpheme type from other
classifiers in Yagua, Waorani, and Nambiquara. In contrast, in Munduruku
classifier suffixes are used with numerals, with demonstratives and as
derivational markers, and prefixes are used with verbs and adjectives.
This may be due to the fact that adjectives are morphologically a subclass

12
The same morphemes in multiple classifier environments can be considered polysemous
morphemes.
 Multiple Classifier Languages 225
13
TABLE 9.3. Same morphemes in several classifier environments
Language Place of morpheme Obligatoriness
Kilivila (B) Deictic classifiers are infixes; Deictic classifiers more
numeral and adjectival obligatory than others;
classifiers are prefixes classifiers are not used with
some classes of adjectives
Yagua, Waorani (C) Numeral classifiers infixed,
others suffixed
Mundurukii (C) Verbal classifiers and adjectival -
markers are prefixes, numeral,
deictic classifiers and classifiers
in derivational function are
suffixes
Hmong (C) - Possessed classifiers not
always obligatory
Nambiquara (E) Numeral classifiers are prefixes, -
agreement markers and
relativizers are suffixes

of verbs in this language, and verbs have a highly prefixing profile. In

Kilivila, classifiers are infixes with demonstratives, and prefixes with
numerals and adjectives.
The fact that different kinds of behaviour correspond to different kinds
of morphosyntactic loci of classifiers may be considered an additional
argument in favour of the validity of a morphosyntactically based typology
of classifiers.
Some preliminary generalizations can be made as to the use of the same
set of classifiers in distinct environments.
(i) Classifiers in multiple environments have derivational, relativizing, and
nominalizing functions if also used as noun classifiers (see §3.2.4, for some
examples from Jacaltec and from Lao), or if also used as agreement
markers on adjectives (as in Tariana, Yagua, Tucano, Mundurukii, or
Nambiquara). Verbal, relational, possessed, or locative classifiers are not
used as derivational markers. For instance, classifiers used in locative expres-
sions and in verbal classifier constructions in Eyak play no role in derivation.
Only in Palikur (Aikhenvald and Green 1998) are the distinct sets of verbal
and of locative classifiers also used as derivational affixes. Verbal classifiers
added to some nouns transform them into colour terms, e.g. ahamna-bo-ye

13
Letters in brackets after language names indicate the subsection of §9.1 where classifiers
in these languages are considered.
226 Classifiers
(leaf-v.CL:FLAT-DURNF) 'leaf-coloured', or 'green'. Locative classifiers are
sometimes employed as derivational affixes, e.g. paraw-hakwa (waves-
in.wATER) 'ocean', pi-duk-madka-ya (2SG-chest-on.FLAT-PERTAINING) 'the
flat part of your chest, your breast plate', a-kigbimna-ya (3N-on.EDGE-
PERTAINING) 'its frame'. However, these patterns have limited productivity.14
(ii) In languages with classifiers as independent words, these are used with
a subset of adjectives only if they are also used with demonstratives (as is
the case in Cantonese and in Thai). If classifiers are bound morphemes,
they may be used with adjectives but not with demonstratives (e.g. the
Arawak languages discussed above, such as Waura, where a small class of
pronominal genders is marked on demonstratives).
(iii) In multiple classifier languages classifiers are not used as relational
classifiers; however, they can be used in possessive constructions in a
function similar to that of possessed classifiers (i.e. categorizing the
possessed).
(iv) In all the attested examples, classifiers are used in possessive construc-
tions only if they are also used in head-modifier noun phrases.

(H) Which classifier environment is primary?

The question of which (if any) classifier environment is a 'primary' one,
and which one is 'derived', can only be answered on the basis of language
internal criteria.
The first question is which morphosyntactic environment is historically
primary, and which one is a later development. In the history of Chinese
the usage of classifiers with numerals appears to predate usage with demon-
stratives; the use of classifiers in possessive constructions in some Chinese
languages such as Cantonese, Hakka, and Hokkien is historically an even
later development (cf. Hashimoto 1977; Pacioni 1997).
The analysis of the patterns of genetic inheritance and areal diffusion
in some Arawak languages from Northern Amazonia shows that some
morphosyntactic environments for classifier use emerged historically later
than others. Tariana and Resigaro 'extended' the use of classifiers to
demonstratives under the areal influence of neighbouring languages,
Tucano and Bora-Witoto respectively; genetically close Arawak languages
use classifiers with adjectival modifiers, numerals, and verbs, but not with
demonstratives (see Aikhenvald 1996a; forthcoming c). In Kubeo, a

14
Another possible exception to the above claim is Pareci (Xingu-Pareci subgroup of
Arawak: Rowan and Burgess 1979; Aikhenvald 1996b: 164). Pareci appears to have verbal
classifiers, e.g. -koa 'flat' in aolaka-koa-tya (burn-VCL:FLAT-VB) 'burn (e.g. a field, a garden)'.
Classifier morphemes are also employed as derivational suffixes, e.g. mare-koa (field-CL:FLAT)
'platform'. However, we have no information about the productivity of classifiers as deriva-
tional devices; the sketchy grammar of Pareci provides little information on the NP structure.
 Multiple Classifier Languages 227
Central Tucano language, classifiers began to be used with verbs under
the influence of Baniwa, a neighbouring Arawak language (Gomez-
Imbert 1996). This historical information is obtained as the result of
comparison within each language family and with the neighbouring
languages; in many cases we do not have access to any information of
this sort.15
The second question is whether any synchronic functional priority can
be given to one morphosyntactic environment. The use of classifiers with
demonstratives in Chinese and in Austroasiatic languages is often consid-
ered an 'extension' of numeral classifiers (see Hashimoto 1977; Conklin
1981: 186). In Chinese as well as in Austroasiatic languages classifiers are
indeed obligatory with numerals, but can be omitted in other noun phrases
(see (A) above, for Chinese). In some languages only a subset of classifiers
can be used with modifiers other than numerals. In Nung and Black Tai
only the general classifiers for animates and inanimates are used with
demonstratives, adjectives and possessives (Conklin 1981: 188). This may
also be an indication in favour of the primary use of classifiers with
numerals. However, in Northern Tai languages, according to Conklin
(1981: 190), the numeral phrase usage 'is not the primary usage' of classi-
fiers, since classifiers appear regularly with nouns in any context 'serving as
their substitutes'.16
The subset of adjectives which require classifiers always includes the
dimension type (as in Cantonese and other Chinese languages), and also
the colour type (as in Thai). This restriction on the semantics of adjectives
which require classifiers in isolating languages could provide an additional
argument in favour of quantification as a primary function of classifiers
(and thus, possibly, the usage of classifiers with numerals being the
'primary' one). The behaviour of classifiers used with adjectives in isolating
languages like Chinese or Thai deserves further in-depth study (which is
15
Derbyshire and Payne (1990: 266) hypothesize that classifiers in Mundurukii and other
multiple classifier languages such as Waorani and Chayahuita developed from classificatory
noun incorporation; i.e. that verbal classifiers developed first and numeral and adjectival
agreement markers developed later. However, there is no language-internal evidence in favour
of that. We have seen that languages which use classifiers with numerals and demonstratives
are likely also to use them with adjectives; verbal classifiers are a much rarer type, and their
existence does not seem to correlate with the use of classifier morphemes in other environ-
ments. Until more historical evidence becomes available there is no reason to believe that any
one of the classifier environments in Mundurukii, Waorani, Chayahuita, Waura, or Ignaciano
is historically 'primary'.
16
Hundius and Kolver (1983: 181) make an attempt at providing a functional explanation
for the use of classifiers in multiple environments in Thai. They define the function of
classifiers as indicating that an 'NP applies to specified or specifiable numbers of individual
objects'. Then, the classifier is obligatorily used in quantitative expressions, while it is optional
with determiners which do not necessarily relate to quantity (such as adjectives). This hypoth-
esis deserves further justification, especially since the question of priority of each environment
is not addressed.
228 Classifiers
lacking from the existing literature, mainly, because of the presupposition
that these are nothing but a 'funny' parasitic extension of numeral classifiers
to another environment).17

9.2. Multiple classifier languages and noun class agreement on

multiple targets
Noun class agreement on multiple targets is functionally similar to multiple
classifier systems in that it may include adjectives, numerals, and other
modifiers, possessives, the noun itself and the predicate (cf. (D) in §2.4.2,
and example 2.13 from Swahili).
Often, different subsets of noun classes are used with different modifiers
and with the predicate. We have seen that Palikur distinguishes three
gender forms in demonstratives as modifiers (see Table 8.8) and two gender
forms in active verbs and stative verbs (the latter can be used as predicates
and as modifiers) (see Table 8.9). Genders also differ in their realization
depending on the morphosyntactic environment they occur in. More agree-
ment forms are distinguished in demonstratives and personal pronouns
than in gender-sensitive verbal affixes, in conformity with an almost
universal tendency to distinguish more gender agreement forms with 3rd
person pronouns and/or with demonstratives than with modifiers from
open classes.
In many languages of the world a gender opposition is found only in
personal pronouns (e.g. English; Kaingang, a Je language from South Brazil;
Rikbaktsa, a Macro-Je language from Central Brazil; Kakua, a Maku
language from Colombia: see Aikhenvald 1999a; numerous languages
from Southeast and east-central Australia, e.g. Pittapitta, Yandruwanhtha,
Awabakal, and Gadjang: Dixon forthcoming). In other languages, such as
Dyirbal, an Australian language from North Queensland, noun classes are
restricted to determiners, the demonstrative, and the locational interroga-
tive. Waura and Yawalapiti, two Xinguan Arawak languages from Brazil
(from the same family as Palikur), distinguish masculine and feminine
genders only in demonstratives, which are also used in the function of
3rd person pronouns (Aikhenvald 1996b: 165-7).
The use of one form for both masculine and neuter gender agreement
with gender-sensitive verbal suffixes reminds us of a phenomenon known
as 'concordial superclassing' ((C) in §2.4.4) in Australian linguistics (Sands
1995: 264–5), where fewer agreement forms are used with some modifiers,
17
These restrictions are reminiscent of restrictions on the use of verb types in languages
with verbal classifiers, discussed in Chapter 6: there, verbs which require classifiers are mostly
posture and location verbs, and also verbs which imply handling objects.
 Multiple Classifier Languages 229
such as demonstratives.18 Further examples of different subsets of noun
classes/genders with different modifiers are given in (G) in §2.4.2.
Noun class agreement on varying targets sometimes involves different
morphological realizations (e.g. prefixes in one case and suffixes in another:
see (A) in §2.6.2), or the use of the same set of morphemes with different
properties. Numeral classifiers sometimes have different realization
depending on the choice of a numeral or of a quantifier. For instance, in
Bahwana (Ramirez 1992: 55) classifiers are obligatorily suffixed to quanti-
fiers (4.22), and infixed to numbers one and two (4.23). In Palikur, numeral
classifiers are suffixes to the numeral 'one' and infixes to the numeral 'two';
and often different forms of classifiers are used with 'one' and with 'two'
(see Aikhenvald and Green 1998). In multiple classifier systems it is fre-
quently the case that the same set of classifiers in distinct environments
display different properties (see Table 9.3).
The main differences between multiple classifier languages and noun
class agreement realized on different targets are the following.
(i) Noun class agreement realized on different targets is a grammaticalized
agreement system whose semantic motivation varies. This involves both
semantic and non-semantic agreement (see (F) in §2.4.2). In contrast, mul-
tiple classifier systems may involve lexical selection rather than agreement;
this is especially striking in the case when multiple classifiers are realized
through independent lexemes. The choice of a classifier in a multiple classi-
fier language is always semantically based.
(ii) Noun class agreement on different targets retains all the properties of
noun class systems (see §2.2); in particular, every noun must be assigned a
noun class, and the semantic motivation involves animacy and/or sex.
Multiple classifier systems are different in that not all the nouns require
a classifier; their semantic motivation is more comparable to that of
numeral or verbal classifiers.19
(iii) Noun class agreement on different targets obeys the Agreement Hier-
archy (see (F) in §2.4.2). The choice of classifier environments in multiple
classifier systems does not. It appears to be difficult to establish any hier-
archy in motivating the choices for classifier environment in multiple clas-
sifier languages; some preliminary generalizations have been suggested in
(G) above.

18
Unlike Palikur, however, superclassing in Australian languages often has a discourse
function: more general agreement forms are used when the head noun has a general reference,
or is backgrounded.
19
In Kariri (Kipea-Kariri family: Rodrigues 1997: 69-72) the same morphemes are used
with numerals and with adjectives referring to dimension, consistency, and colour. However,
our knowledge of this extinct language is not sufficient to decide whether this is noun class
agreement or a multiple classifier system.
230 Classifiers
(iv) Noun class agreement markers used on different targets have the
same origin as other noun class systems, and show the same principles for
their interaction with other categories as do noun classes. In contrast,
multiple classifier systems pattern in these respects more like other classifier
types (see Chapter 10).

9.3. Fuzzy types: overlapping classifiers in multiple

environments

Languages can use almost but not quite the same classifier sets in several
different environments. In this section we present two case studies, from
genetically closely related languages. In §9.3.1 we consider classifiers in
Baniwa (Arawak: Aikhenvald 1996c); and in §9.3.2 we deal with classifiers
in Tariana. Some conclusions as to the possible analysis of these transi-
tional systems are given in §9.3.3.
Both languages have large sets of classifiers, and also two pronominal
genders (feminine and non-feminine). The environments in which classi-
fiers and genders are employed in the two languages are given in Table 9.4.
Unlike Tariana, Baniwa has a closed set of 44 classifier morphemes; while
Tariana employs 'repeaters' (see (F) in §9.1) alongside a large set of over
sixty 'established' classifiers (see Aikhenvald 1994a). Classifiers are not
employed on locatives. Also unlike Tariana, Baniwa has relational classi-
fiers (see 5.51 and 5.52).

TABLE 9.4. Environments in which genders and classifiers are used in Baniwa and
Tariana

Language Genders Classifiers

Baniwa Demonstratives, cross-referencing Noun classes, numerals, verbs,

affixes, personal pronouns head nouns, possessives
Tariana Cross-referencing affixes, personal Noun classes, numerals, verbs,
head nouns, possessives,
demonstratives

9.3.1. Multiple classifiers in Baniwa

A complete list of classifiers in Baniwa is given in Table 9.5 (arranged
semantically, according to the parameters in Chapter 11). The inventories
of (a) noun class markers and (b) numeral, noun, verbal and possessed
classifiers only partly overlap.
Some adjectival agreement markers on adjectives are 'derived' from a
classifier with the help of an adjectivizing morpheme -y, or feminine -3u, or
 Multiple Classifier Languages 231
non-feminine -ri. These are marked with * in the table. Noun class markers
differ from classifiers in the way they interact with plural number (see §10.1).
Plural noun class markers which are not used as classifiers are given under (G)
in Table 9.5. The two classifiers in (H) are not used as noun class markers.
A possessed classifier in Baniwa was shown in 5.50. Example 9.75
illustrates a classifier with a numeral.
9.75. apa-kha a:pi
one-CL:CURVED snake
'one snake'
Classifiers are employed with the predicates of relative clauses,20 where
their use depends on whether the noun is in focus. In 9.76 the classifier -kha
'curved' is used to focus the snake. It is not used in 9.77 since the snake is
not in focus.
9.76. a:pi nu-inua-ri-kha awakada-riku
snake lSG-kill-REL-CL:CURVED bush-LOC
'The very snake I killed is in the bush.'
9.77. a:pi nu-inua-ri awakada-riku
snake 1SG-Kill-REL bush-LOC
'A snake I killed is in the bush.'
A noun class marker which consists of a classifier + adjectivizer -y is
shown in 9.78. A noun class marker which coincides with a classifier is
shown in 9.79.
9.78. a:pi maka-khay
snake big-NCL.CURVED
'a big snake'
9.79. pan-ti maka-dapana
house-NON.POSSESSED big-CL:HOUSE
'a big house'
Examples 9.80 and 9.81 show the derivational use of classifiers.
9.80. i-thi-da
INDEF-eye-CL:ROUND
'eye'
9.81. i-thi-maka
INDEF-eye-CL:PIECE.OF.CLOTH
'eye lid'
20
And also on verbs in the form of the topic-advancing voice, as in bula a na-iyi-nita-mi
(na-i3a-ni-ita-mi (underlying form)) (biscuit 3pL-consume-ARGUMENT.MANIPULATING-CL:ANIM-
PAST) 'biscuits have been eaten'.
TABLE 9.5. Classifiers in Baniwa
Classifier Adjectival Gloss Examples
agreement
marker
A. Animate, feminine
1.* -ita ite < ita-y, Non-feminine animate inu 'dog', nukapi
-da-ri Animate human 'hand'
attributes
2* -hipa same as cl.l Non-feminine human wa i-pari 'young man',
iafi 'man'
3.* -ma -da-3u Feminine human ina3u 'woman', haddua
'mother'
B. Form and shape
4.* -da -da-ri Round objects; generic idza 'rain', hnuwida 'my
head'
5. -api -api Hollow objects tfipara:pi 'pan',
ma:wipi 'blow gun'
6. -aphi -aphi Limited space kiniki 'garden', karita
'lake'
7. -apu -apu Long thin stick-like haikwapu 'stick',
object dzawithyapu 'bow'
8. -hiku -hiku Pointed, long objects ri-api 'his bone', garafa
'bottle'
9. -him -hiwi Thin sharp objects tfikure 'a hair',
nukapida 'my finger'
10. -i/i -ifi Small seed-like objects pipefi 'seed of a palm
tree fruit', a:[i 'pepper'
11. -ku -ku, -ki Pieces of cloth; piete 'hammock',
hammock enikarya 'cobweb'
12.* -kwa -kwe < -kway Flat, round, extended kanari 'looking glass',
kaida 'beach'
13.* -kha -khay < kha-y Curvilinear objects a:pi 'snake', hinipu
'road'
14.* -na -ne, -nay Vertical objects haiku 'tree', dzawi
<- na-y 'jaguar'
15. -pi -pi Long thin objects teruripi 'manioc
squeezer'
16.* -phe -phe, phay Thin flat foldable objects yamakati 'cloth',
pape3a 'paper'
17. -wa -wa Small holes nu-numa 'my mouth',
hipadaphi 'hole in a
stone'
18.* -0 -yafi Hollow smallish objects a:ta 'cup', kureya
'spoon'
 Classifier Adjectival Gloss Examples
agreement
marker
19. -yawa -yawa Holes, open spaces pantinuma 'door',
harayawa 'hole'
C. Functional p roperties
20.* -apa -apa-ri Non-feminine flying piti3i 'bat', kepi3eni
animate; semi-oval 'bird',
objects ainidzu 'mosquito',
parana 'banana'
21.* -apa -apa-3u Feminine flying animate karaka 'hen'
22* -maka -make < Piece of cloth tsaia 'skirt', dzawiya
maka-y 'jaguar's skin'
23. -peku, -peku, -peki Extended long stretches hinipu 'road', u:ni
-peki 'river'
D. Structure
24.* -ahna -anha-y Liquids u:ni 'water', kutfiaka
'drink of manioc flour'
E. Quantification
25. -hipada -hipada Piece, half parahnepada
makhepada 'piece of
banana'
26. -ida -ida Half apaida mawiju 'half a
pineapple'
21* -ima -ima-ri One side nukapi makemari 'one
side of my hand'
28. -naku, -naku, -naki Bundle of thin long lapi '(bundle of)
-naki objects pencils'
29.* -pa -pa-fi Bundle, box, parcel apa-pa itsa maka-pa-ri
'a box of
fishing hooks'
30. -puku, -puku, Bundle of long objects apa-puki parana 'a
-puki -puki bundle of bananas'
31. -tawahre -tawahre Joint apa-tawahre
nukaphiwida 'finger-
joint'
32. -wai -wari Time cycle hamufi 'year', hekwapi
'day'
33.* -wana -wane < Thin slice apawana kuphe
wana-y makawane 'a big thin
slice of fish'
34.* -wata -wate < Bundle ready for parana makawate 'a
wata-y carrying bundle of bananas'
234 Classifiers
TABLE 9.5. (contd.)
Classifier Adjectival Gloss Examples
agreement
marker
F. Specific classes (each applying to a single noun)
35.* 0 -a, -afi Canoe i:ta 'canoe'
36. -dapana -dapana House panti 'house'
37. -ihwe -ihwe Egg, stone (of a fruit) karakehwe 'hen's egg'
38. -hipani -hipani Waterfall hipa 'waterfall'
39.* -pawa -pawani Big river u:ni 'river'
40.* -fa -fa-fi Excrement i/a 'excrement'
41. -tuhwya -tuhwya Room tuhwya 'room'
42.* -ya -ya, -yari Skin dzawiya 'jaguar's skin'
G. Morphemes used only as noun class markers
1. -pe3i Inanimate; collectives tfikure yapi-pey 'long
hair'
2. -peni Animate nouns; their dzawithipa matfia-peni
attributes 'beautiful footsteps of a
jaguar' (an ornament)
H. Morphemes used only as numeral and verbal classifiers
1. -i Bundle ape pipi3i 'a bundle of
pupunha' (palm tree
fruit)
2. -itfia Bundle of small fruit ape tfia manakhe 'a
bundle of acai'

The choice of a classifier in a given environment depends on what aspect

of the referent is highlighted (cf. examples 2.18-20). Classifiers with verbs
in Baniwa are less obligatory than classifiers in other environments; their
use depends on pragmatic factors. Otherwise, classifiers behave in a similar
manner. The system of classifiers in Baniwa can be analysed in two distinct
ways. One can say that it is a multiple classifier system with the same set of
morphemes used with adjectives, numerals, verbs and as derivational suf-
fixes; the formal differences between the 'adjectival' classifiers and the rest
are due to the presence of additional attributive markers which appear on
adjectives with some classifiers (but not all of them).
Alternatively, Baniwa can be said to have noun classes marked on adjec-
tives as a type of noun categorization device which is distinct from classifiers
employed with numerals and with verbs, as well as in a derivational
function.
The argument against the second solution is that semantically, noun
classes and classifiers coincide; and they are also very similar formally.
 Multiple Classifier Languages 235
The argument in its favour lies in the existence of morphemes which are
used only as noun class agreement markers ((G) in Table 9.5), and of
morphemes used only as numeral and verbal classifiers ((H) in Table 9.5).
Both analyses are equally acceptable. Baniwa is the case of a language in
transition from a system with classifiers in multiple environments to a
system with noun classes distinct from other noun categorization devices.

9.3.2. Multiple classifiers in Tariana

Classifiers in Tariana (also from the North Arawak subgroup) are used as
noun class agreement markers on adjectives; with numbers (one to four) as
possessed classifiers in possessive NPs, where they attach to the possessive
marker and characterize the possessed noun (see examples 5.18 and 5.19);
with verbs; with demonstratives, articles, and other modifiers from closed
classes; and on head nouns, as derivational suffixes. Classifier morphemes fall
into three groups: (i) classifiers used in all these environments with a distinct
form in each; (ii) classifiers used—with the same form—only in two environ-
ments: as noun classes and as derivational markers on the head noun itself;
and (iii) classifiers used in all the environments with the same form in each.
In the last group, the classifier -ma 'feminine' is not used with the article.
The classifier -dapana 'HOUSE', from group (iii), was illustrated with a
demonstrative, a numeral, a possessive, an adjective and the predicate; and
on the head noun 'hospital' itself as a derivational suffix in 9.2 above.
We can now consider the classifiers of group (i). The 'animate' noun class
marker is -ite; the form -ita21 is used with numerals and verbs: see 9.82.
Noun class markers also differ from other classifier types in the ways they
correlate with the grammatical category of number (see §10.1), e.g. 9.82 and
its plural counterpart, 9.83.
9.82. pa-ita tfari hanu-ite
One-CL:ANIM man big-N.CL:SG.ANIM
'one big man'
9.83. atfa hanu-peni
man:pL big-N.CL:PL.ANIM
'big men'
For group (i), the same form is used with numerals, verbs, in possessive
constructions and on the noun itself (except for -ital-ite). There is also a
21
-ite in Tariana comes from -ita-y, -y being the adjectivizing morpheme used on some
noun class markers in Baniwa. The difference between this noun class marker and the corre-
sponding classifier in Tariana is thus reminiscent of the one in Baniwa. It can be considered an
archaism, due to the areal influence of Tucano languages on Tariana (Aikhenvald 1994a;
1996b).
TABLE 9.6. Classifiers in Tariana
(i) Classifiers with distinct forms in different environments

Semantics Examples Noun class Noun class Numeral Noun Possessed Verbal Classifier with
singular plural classifier classifier classifier classifier deictics
Generic tfari 'man', inaru 'woman', -ite -peni -ita -ite -ite -ita Animate form of
Animate tfinu 'dog', a:pi 'snake'; demonstrative
human attributes: siruli
'trousers'
Human tfari 'man', inaru 'woman' -ite -peni -hipa -ite ite -ita As above

(ii) Classifiers used in only two environments

Semantics Examples Noun classes and noun classifiers
Collective pumeni-peri 'sugar', u:ni 'water', sana-peri 'sweat' -peri
Abstract nouns, places ehkwapi 'day, weather, world', iya 'rain', panisi-wani 'place -wani
where home is'

(iii) Classifiers with the same forms in all classifier functions

A. Gender classifier (not used with specifier article)

Classifier Semantics Examples
Feminine inaru 'woman', kabueta-ma 'female teacher'
B. Shape and form classifiers (a sample out of about 80)
-da Round objects mawina 'pineapple', dithi 'eye', heku-da 'fruit'
-hiwi Thin long objects itfa 'hair', nuthiwi 'eye-lashes'
-ipu Long, hollow, bundle-like haiku-pu 'log', nawiki-pu 'grave'
-kha Curvilinear kule-kha 'fishing line', hewya-pi-kha 'rainbow'
-khi Thin curved maka-khi 'rope'
-maka Extended cloth, cloth-like yarumakasi 'cloth', hitisi-maka 'funeral mask' (lit. mask of tears)
-mapha Completely covered disinuma 'his beard', itfima 'mane of hair'
-na Long vertical heku-na 'tree'
-pa Largeish and long de i 'banana', pesanini-pa 'ladder'
-peku Thin stretch leka-peku 'a broken longish piece'
-pi Long, thin, vertical; cycle of time hirjna 'manioc squeezer (tipiti)', deji-pi 'banana tree'; ke:ri, ke:ri-pi 'month'
-pukwi Round and hollow episi-pukwi 'metal ring'
-puna A length, stretch hinipu 'road', karaka-hwya-puna 'air-strip'
-phe Leaf-like dina-phe 'feather', papera-phe 'leaf of paper'
-(a)phi Small, hollow surupe-phi 'clay pot', episi-aphi 'metal pot'
-yawa Holes hala-yawa 'hole', dithaku-yawa 'his nostril'
C. Quantifier-like classifiers
-apa Pair of diphema 'wings'
-itfi Bundle of manaketfi 'bundle of acai', dejitfi 'bundle of bananas'
-ima A paired object; one side of two iphema '(insect's)wing', diaranima '(bird's)wing'
-iphina A quarter pethe-iphina 'a quarter of a loaf of a manioc bread'
-pada Piece of maka-pada 'half
-piu Time pa-piu 'once'
-sawa Group mare-sawa 'a group of birds'
-yami Piece (of cloth) yarumakasi-yami 'piece of cloth'
TABLE 9.6. (contd.)
D. Function classifiers
Classifier Semantics Examples
-ita Instruments marie 'knife', kanari 'mirror', hekuta 'paddle'
-dapana Habitation panisi 'house', ditape-dapana 'hospital'
-whya Canoe, transport ita-whya 'canoe', ka-koloka-whya 'car'
E. Specific classifiers (each used a single noun)
-dawa Corner, limited space maka-dawa 'room; still part of a river', karisana-dawa 'part of a lake with
stagnant water'
-ithi Seed ithi 'seed', iwi 'grain of salt'
-iwai Trap; wall nehpaniwai 'a made up trap', panisiwai 'house-wall'
-kada A day pa-kada 'one day'
-kawa Leg, handle, anything leg-like sidu-kawa 'the long part of an arrow', huni-kawa 'manioc trunk'
-kena Branch heku-kena 'tree branch', dikawana-kena 'crab's leg'
-kiyere Island kewere 'island', maka-kyere 'a big island'
 Multiple Classifier Languages 239
numeral classifier, -hipa 'animate', used in this, and in no other environ-
ment. This classifier is interchangeable with -ita 'animate'; there is a mild
preference to use -hipa for human male referents (see further details in
Aikhenvald in prep.), cf. 9.84 and 9.85.
9.84 phepa (< pa- + -hipa) tfari
one+NUM.CL.HUMAN man
'one man'
or
9.85. pa-ita tfari
one-NUM.CL:ANIM man
'one man'
The animate classifier -ita is acceptable with a non-human referent, such
as a:pi 'snake' (9.86). Example 9.87, with the classifier -hipa, is acceptable
only if the snake is personified, in a traditional legend.
9.86. pa-ita a:pi
one-NUM.CL:ANIM snake
'one snake'
9.87. ?phepa a:pi
one+NUM.CL:HUMAN snake
'one snake'
Only noun class markers and the classifier morphemes in the deriva-
tional function allow 'stacking', that is, the occurrence of more than one
classifier morpheme in one word (see examples in §3.5; Aikhenvald 1999b).
Slightly different subsets of classifiers are used with modifiers from
closed classes (proximate and distal demonstratives, the demonstrative
'this big', articles, the interrogative-distributive 'which, every', and the
interrogative quantifier 'how many?'). The animate noun class marker
(see (i) in Table 9.6) -ite is used only with the interrogative-distributive
'which, every'. The articles have a feminine form and do not take the
feminine classifier -ma. Proximate and distal demonstratives differ in
how they mark animacy, and in the classifiers they take. The semantics
and form of demonstratives with classifiers are shown in Diagram 9.1.

Animate Inanimate

+fem Generic Generic + shape

Proximate ha-ma hi ha-CLASSIFIER

Distal hane-ma hane hane-CLASSlFlER
DIAGRAM 9.1. Semantics and form of demonstratives with classifiers in Tariana
240 Classifiers
Diagram 9.2 features the semantics and the form of classifiers with
articles (Aikhenvald forthcoming b).

Singular

> Animate Inanimates

s
Feminine' Non-feminine Generic +shape
duha - diha -

Plural

Animate - Inanimate

naha diha-CLASSIFIERS(-pe)
DIAGRAM 9.2. Semantics and form of articles with classifiers in Tariana

The agreement forms of the interrogative-distributive kwa 'which, every'

are given in Table 9.7 (see Aikhenvald in preparation for further details).

TABLE 9.7. Agreement forms of kwa- in Tariana

Animate Animate Animate Inanimate Inanimate Inanimate

singular singular plural unmarked plural form-specific
head modifier non-plural

kwana ku-ite ku-ita-peni kwaka kwe-peji kwaka-

INANIMATE CL

The question arises: how many classifier sets does Tariana have? If one
takes a 'maximalist' approach, every morphosyntactic locus (including
every type of modifier from a closed classs) may be considered as repre-
sentative of a separate type, since each of them (with the possible exception
of verbal classifiers) can be shown to employ a slightly different set of
forms. Another way of approaching this would be to say that, similarly to
Baniwa, Tariana is another, more complicated instance of a language in
transition to a system with noun classes and numeral classifiers being
distinct from other noun categorization devices, and also of several distinct
classifier types developing in several further morphosyntactic loci.

9.3.5. Fuzzy types and borderline cases

The classifier types proposed here are defined by the morphosyntactic
contexts in which they appear. Each has a set of properties, and there
 Multiple Classifier Languages 241
can be variations on them. It is not always an easy matter to assign a
particular system of classification unambiguously to just one of the types,
as we have seen for Baniwa and Tariana. These languages present examples
of fuzzy types, or borderline cases found in systems which can be called
transitional in the sense that the classifier types are not as clear-cut as they
are in other instances. Systems like Tariana, with distinct subtypes of
classifiers for different modifier classes, can develop into further new
classifier types (each being used with a variety of closed-class modifiers).
 10 Classifiers and Other
Grammatical Categories

Noun categorization devices interact with other grammatical and lexical

categories in the following ways.
(i) There may be dependencies between noun categorization devices and
other categories: the semantics or form of noun categorization devices may
correlate with the choice made in another grammatical category. For
instance, there may be fewer noun class distinctions in non-singular
numbers than in the singular.
(ii) Noun categorization devices may be realized as portmanteau mor-
phemes with other categories, e.g. case or number.
(iii) The meanings encoded through noun categorization devices in some
languages can be realized through different categories in others; for
instance, the choice of case marking may depend on the animacy or
humanness of the referent even if a language does not have an overt
classifier system (see Appendix 1).
Distinct classifier types interact in different ways with verbal and nom-
inal grammatical categories; they may also interact with derivational
devices. Their presence may influence the structure of the lexicon.
The degree of interaction between classifiers and other categories also
depends on the morphological type of the language, and the morphological
realization of classifiers. We expect more interactions between categories in
a fusional language, because of the higher probability of portmanteau
realization of categories (cf. Aikhenvald and Dixon 1998). Consequently,
noun classes which are always realized with bound morphemes (see
Chapter 2) are likely to display more interdependences with other
categories than numeral classifiers realized as independent words.
We also expect most interactions between noun categorization devices and
nominal categories; we have seen in previous chapters that most classifiers
have an NP as their scope. Since some do have a clause as their scope, we
also expect some interaction between classifiers and predicate categories.'

1
We understand predicate categories to be categories typically realized on the predicate, or
on the verb, e.g. tense/aspect, or voice. It is sometimes argued that categories such as tense/
aspect are clausal rather than just verbal. We will leave this question open here; for further
 Classifiers and Other Categories 243
Interactions between noun categorization devices and the grammatical
categories of number, person, marking of grammatical relations, posses-
sion, and politeness are considered in §10.1-5. Noun categorization devices
also interact with the organization of declensional paradigms—§10.6. In
§10.7 we discuss the ways in which classifiers interact with verbal
categories; §10.8 deals with the categories characteristic of deictics (e.g.
proximal vs. distal) and classifiers. §10.9 describes how classifiers interact
with derivational devices and how their existence may affect the lexicon.2
Conclusions are given in §10.10.

10.1. Classifiers and number

The semantics of number usually depends on the meaning of a noun; this is
the reason why classifiers tend to interact with number in many ways. Noun
classes display the most complicated interactions and interdependencies
with number: §10.1.1. Interdependencies of number with other types of
classifier are considered in §10.1.2.

10.1.1. Noun classes and number

Noun classes and number show mutual interdependencies: (a) noun class
choice depends on number; or (b) number choice depends on noun class; or
(c) there may be different noun class systems in different numbers, or, finally,
(d) number and noun class can be expressed with a portmanteau morpheme
(see Aikhenvald and Dixon 1998, for further discussion of (a) and (b)).
(A) Noun class depends on number
The choices available for noun classes may depend on the choices made in the
number system.3 In many languages, gender distinctions are made only in the
singular and not in the plural—for example, Indo-European languages such
as German, Russian, Danish; Manambu (Ndu: Papuan); North Berber
languages such as Zemmur and Izayan; and the East Berber language
Siwa. Many African languages distinguish fewer noun classes in the plural
than in the singular—for example, Ful (Heine 1982a: 207) has twenty-three
singular but only five plural classes, and Mba (Ubangi: Heine 1982a: 208) has
six singular and three plural classes. Seneca (Iroquoian: Chafe 1967: 13-14)
discussion of dependencies between gramamtical systems, see Aikhenvald and Dixon (1998),
and references therein.
2
The ways classifiers which interact with definiteness are linked to their discourse functions
are discussed in §12.1.3.
3
Note that quite a few languages have the same noun class/gender distinctions in all
numbers, e.g. Portuguese and Hebrew.
244 Classifiers
has three genders in the singular and two in the dual and the plural. Polish
(West Slavic) distinguishes three genders in singular and two in plural.
Other non-singular numbers often behave differently. There may be more
gender distinctions in dual than in plural. In Old Church Slavonic three
genders (masculine, feminine, and neuter) were distinguished in the singu-
lar, two in the dual (masculine and neuter having fallen together), while no
gender distinctions were made in plural. East Slavic languages have lost the
dual, and gender is shown just in the singular. In the two languages which
do have dual, Upper Serbian (West Slavic) distinguishes three genders in
singular and two in dual and plural (feminine and neuter having fallen
together); Slovene (South Slavic) distinguishes three genders in singular
and plural, and two genders in dual (masculine and feminine-neuter)
(DeBray 1951). In Manambu (Ndu, Papuan) there are two genders in
singular but none in dual and plural.
Different choices in different numbers are found in Baining (East New
Britain: Parker and Parker 1977: 18), where demonstratives distinguish
masculine and feminine genders in singular and dual; in plural, there are
distinct forms for human and non-human referents.
Some languages have the same gender distinctions in singular and in
dual, but none in plural, e.g. Lavukaleve (Papuan, Solomons: Todd 1975:
813; Angela Terrill, p.c.) and also Angave (Angan: Speece n.d.: 113).
This dependency relates to Greenberg's Universal 37 (1963: 95): 'A
language never has more gender categories in non-singular numbers than
in the singular', and Universal 45: 'If there are gender distinctions in the
plural of the pronoun, there are some gender distinctions in the singular
also.' There is, however, a set of counterexamples to these generalizations.
A number of Austronesian languages distinguish animate/non-animate
forms in the plural only. Biak (South Halmahera, West New Guinea sub-
group: Steinhauer 1986) has animate/inanimate forms for 3rd person plural
pronouns and demonstratives; there are no such distinctions in the singu-
lar, dual, or trial. In a seminal article, Pawley (1973: 100) reconstructs an
animate/inanimate opposition for 3rd person plural in Southeast Solomonic
languages.4 Further counterexamples concern the interaction between
noun class, number, and person (see (B) in §10.2).

4
A few further counterexamples are given in Plank and Schellinger (1997). Thus, Greenberg's
Universals 37 and 45 can be considered tendencies rather than universal statements. It appears
true that if there are gender distinctions marked on personal pronouns, there are never more
gender distinctions in 2nd person than there are in 3rd person; there is typically the same set of
distinctions in all persons in non-singular numbers. It is rare for a language to mark gender on 1 st
person singular pronouns; more work has to be done to provide an explanation for this.
Greenberg's (1963: 94) Universal 32 claims that 'whenever the verb agrees with a nominal
subject or nominal object in gender, it also agrees in number'. There are significant exceptions
to this claim, e.g. Caucasian languages (Corbett 1991: 198ff.) and numerous South American
languages (e.g. Arawak and Je) and Papuan languages (e.g. the Ndu family).
 Classifiers and Other Categories 245
Independent personal pronouns in Tamachek (Tuareg Berber: Prasse
and agg-albostan ag-Sidiyan 1985: 8) distinguish masculine and feminine
genders in all three persons in the plural, but only in the 2nd person for
singular number. Direct and indirect object pronouns are clitics; they
distinguish genders in the 2nd person singular and 2nd and 3rd person
plural. (See Table 10.1.)
TABLE 10.1. Personal pronouns in Tamachek
Independent Direct object Indirect object
pronouns pronouns pronouns
1SG M, F nak -(a) hi -ahi
2 SGM kay -kay -ak
F kam -kam -am
3 SG M, F anta -t -as
1 PL M nakkaned \ -anagh \ -anagh
F nakkanated
2 PL M kawaned -kawan -awan
F kamated -kamat -akmat
3 PLM sntaned -tan -asan
F enimated -tanat -asndt

Languages may acquire more gender distinctions in non-singular num-

bers than in the singular for historical reasons. In some Baltic and Slavic
languages the dual arose as the result of the grammaticalization of the
number 'two' which distinguished genders. Consequently, gender distinc-
tions are found in all persons in dual, and only in 3rd person in singular
and plural. This is illustrated from Lithuanian in Table 10.2 (Peterson
1955: 48–9).

TABLE 10.2. Personal pronouns in Lithuanian

so DU PL
1 M
\ as mudu (cf. masc,. du 'two') \ mes
F mudvi (cf. fem. dvi 'two')
2M
tu judu iu
F judvi
3M Jis juodu jie
F ji jiedvi Jos

In Slovene (Priestly 1993: 406-8) two genders are distinguished in 1st

and 2nd person dual and plural, and in 3rd person dual; there are three
gender forms in 3rd person singular and two in plural, and none in 1st and
246 Classifiers
2nd singular (Table 10.3). Thus, there are more gender distinctions in the
1st and 2nd person non-singular than in the singular forms for the same
persons.

TABLE 10.3. Personal pronouns in Slovene

SG DU PL

1M jaz midva mi
medve, midve me
FEM/NEUT
2M
ti vidva Vi
FEM/NEUT vedve, vidve ve
M on onadva one
3 F ona \ onidve, onedve ona
N ono

In Bora-Witoto languages (which have two gender distinctions in pro-

nouns and large multiple classifier systems), feminine and masculine are
distinguished for all persons only in the dual, but only for 3rd person in
singular and none in plural (e.g. Minor and Minor 1971: 10, on Murui
Witoto). According to Aschmann (1993: 133), such a system can be recon-
structed for Proto-Bora-Witoto. In Bora (Thiesen 1996: 33) masculine
and feminine genders are distinguished in all persons in dual (with the
exception of dual inclusive) and in 3rd person singular. This system has
been calqued into Resigaro (Allin 1975) an Arawak language which under-
went drastic restructuring under the Bora influence (see Aikhenvald forth-
coming b). Table 10.4 features personal pronouns in Resigaro and in Bora.
The morphemes which have been borrowed from Bora into Resigaro are in
bold.
In languages with multiple noun class systems, pronominal noun classes
tend to develop more interactions with number than nominal ones. In
Paumari (Arawa) pronominal gender depends on number (two genders in
the singular are neutralized in the plural); there is no such dependency for
the ka- noun class (see §2.7.2).
In Tariana and Baniwa, North Arawak languages with two pronominal
genders and a large set of noun classes, both depend on number: all
pronominal genders and some noun classes are neutralized in plural
(see §9.3).
(B) Number depends on noun class
In some languages number marking is used only on nouns with animate or
human referents. This is the case in a few Australian languages. Ngandi
 Classifiers and Other Categories 247
TABLE 10.4. Personal pronouns in Resigaro and in Bora

Resigaro Bora
lSG no 00
2SG phit uu
3SG MASCULINE tsu diibye
3SG FEMININE tso diille
1 INCL DUAL MASCULINE fa-musi
mee
1 INCL DUAL FEMININE fa-mupi
1 EXCL DUAL MASCULINE muu-musi muhtsi
1 EXCL DUAL FEMININE muu-mupi muhpi
2 DU DUAL MASCULINE ha-musi a-muhtsi
2 DUAL FEMININE ha-mupi a-muhpt
3 DUAL MASCULINE na-musi diitye-tsi
3 DUAL FEMININE na-mupi diitye-pi
1 INCLUSIVE fa?a mee
1 EXCLUSIVE muu-?a muuha
2PL ha-?a amuuha
3PL na-?a diitye, adtye

(Heath 1978: 35) has seven noun classes; those with non-human referents
have the same prefix for all numbers (ni-, na-, a-, gu-, and ma-). Masculine
and feminine classes have prefixes (ni-, na-) in the singular. Prefix ban-
covers masculine dual; prefix ba- is used for masculine and feminine plural,
feminine dual, and mixed masculine/feminine dual.
Number marking and number agreement follows the Nominal Hierarchy
(Dixon 1994: 85; Silverstein 1976): see Diagram 10.1. The arrow points at
the direction of likelihood of overt number expression, or number agree-
ment (also see Smith-Stark 1974; Stebbins 1997).

Pronouns Proper names/Kinship nouns < Common nouns

1<2<3 Humans < Other animates < Inanimates

DIAGRAM 10.1. Animacy hierarchy and expression of number

In Anindilyakwa (Australian: Leeding 1996), number is obligatorily

expressed only with personified nouns. In numerous South American
languages with non-obligatory expression of number, agreement is more
frequently encountered with humans and higher animates than with in-
animates (e.g. North Arawak languages Tariana, Warekena, Bare). In
Palikur (North Arawak: Aikhenvald and Green 1998), number marking
and agreement is restricted to human nouns. Bora (Bora-Witoto: Thiesen
1996: 33–4) uses different forms for animate and inanimate plurals. In
248 Classifiers
Tamil number marking is obligatory on nouns denoting 'rational' beings
(adult humans, gods and demons); irrational beings (which include infants
and animals) do not have to be marked for number (Asher 1985: 135).
(Further examples are given in Chapter 2; e.g. Table 2.6, for Khinalug.)
The way in which nouns from different noun classes have different
number distinctions may involve superclassing (see (C) in §2.4.4). For
instance, Motuna (Papuan, Bougainville) distinguishes four numbers for
nouns with animate and human referents (singular, dual, paucal, plural)
and two numbers (singular and non-singular) for nouns with inanimate
referents.
(C) There may be different noun class systems in different numbers
If noun classes have a different semantic basis in different numbers, there is
no hierarchical dependency of one upon the other. For instance, Malto
(South Dravidian) has male vs. the rest in the singular and human vs. non-
human in the plural (Mahapatra 1979: 60), as shown in Table 10.5.

Table 10.5. Gender in Malto

Singular Plural
Male referent Male referent
Female referent Female referent
Non-human referent Non-human referent

(D) Noun class is realized as a portmanteau with number marking5

This is the case in many fusional languages, e.g. Bantu, Indo-European,
and Afroasiatic, and in predominantly agglutinating or polysynthetic lan-
guages, e.g. Arawak or Tucano. For instance, portmanteau gender/number
endings in Modern Hebrew are:
MASC SG MASC PL FEM SG FEM PL
-0 -im -a, -Vt -ot
In Kiowa-Tanoan languages nouns are divided into a number of classes
on the basis of their number marking. In Kiowa, for one noun class the
singular/dual are unmarked and plural is shown by the suffix -go; all
animate nouns belong to this class; for a second one the singular and
plural are unmarked and the singular shown by -go; and for a third class
dual is unmarked with the singular and plural shown by -go.

5
Some typologists, e.g. Serzisko (1982), define the notion of a concordial class through its
cooccurrence with number, which is typically the case in Bantu languages (see Chapter 2).
 Classifiers and Other Categories 249
10.1.2. Number and other classifier types
NUMERAL CLASSIFIERS may interact with number in an indirect way. Number
is normally not expressed in numeral classifier constructions, being marked
elsewhere (e.g. on modifiers within an NP, or on the predicate).6 The use of
numeral classifiers may depend on the countability of a noun; classifiers
tend to be employed with countable rather than with mass nouns. In some
languages, e.g. Malay, mass nouns may be left unclassified (see Hopper
1986: 313, 316). In numeral classifier languages, the distinction between
countable and uncountable nouns is realized through classifiers and quan-
tifiers, instead of overt number marking on nouns (De Leon 1987: 27).
It has been claimed that numeral classifier languages always lack 'com-
pulsory number' (Greenberg 1990: 188; Sanches and Slobin 1973; cf. §4.1
above). This means that number marking is optional or it is restricted to a
set of nouns, most frequently humans or animates. There are a number of
exceptions to this claim; for instance, Yuki, Nootka, Tlingit, and Dravidian
languages have numeral classifiers and obligatory number marking, and so
does Ejagham, a Benue-Congo language (see 4.2). De Leon (1987) showed
that the number category became more obligatory in Tzotzil, a language
with a large system of numeral classifiers, under Spanish influence.
Numeral classifiers and the expression of number share a number of
semantic parameters. The distinction between mass and unit, and between
countable and uncountable nouns, is important for both. There are some-
times special classifiers for mass nouns; in addition, some classifiers are
restricted to countable nouns only. In harmony with this, number systems
tend to apply to countable nouns expressing units.
The choice of numeral classifier may occasionally depend on number.
Bengali has five numeral classifiers in the singular; in the definite plural all
classifiers are replaced by one marker -gulo (cf. Greenberg 1990: 188). This
can be explained in terms of the origin of classifiers in Bengali. We will see
in §13.5 that numeral classifiers in Bengali are thought to have developed as
the result of a reanalysis of a noun class system.
VERBAL CLASSIFIERS and CLASSIFICATORY VERBS may interact with number in
several distinct ways, reminiscent of how noun classes interact with number.
First, there may be a classificatory verb stem used just to refer to non-
singular objects. That is, fewer shape-based semantic distinctions are pres-
ent in classificatory verbs which refer to plural objects than in those which
refer to singular ones. In Hupa and Chipewyan (Haas 1967: 360; Carter
1976: 25, 27) one classificatory verb stem is used to refer to all plural
objects, or objects in sets (e.g. stem la in Chipewyan: Table 11.11).
6
There are a few cases in which number is marked in numeral phrases in classifier
languages. In Tariana and Tucano languages, nouns in phrases with numbers bigger than 3
obligatorily take a plural marker, e.g. Tariana nama-da heku-da (two-CL:ROUND tree-CL:ROUND)
'two fruit', kehpunipe-da-pe heku-da-pe (four-CL:ROUND-PL tree-CL:ROUND-PL) 'four fruit'.
250 Classifiers
Second, the choice of a non-singular classincatory stem may correlate
with the shape of objects. Navajo (Hoijer 1945) has a classincatory stem for
an aggregate of non-singular objects, and another one for a set of parallel,
long objects. In Slave the choice of classincatory verb stem for non-singular
referents correlates with number and with shape: there is a classincatory
verb stem for plural objects and aggregates7 (-tl'ih, also used for liquids),
and for dual objects (-keh, also used for rope-like objects) (Rice 1989: 788),
and a number of classincatory stems used to refer to masses (e.g. -tle 'to
handle an uncontained wet mass, e.g. mud, or dough': Rice 1989: 789; -deh
'fall (used of a two dimensional, changing—but maintaining unity—mass':
Rice 1989: 788).8
Other Athabaskan languages make different choices in their classifica-
tory stems for non-singular objects. In Western Apache there is a stem for
pairs of non-animal, non-contained objects. In Chiricahua Apache (Hoijer
1945: 19-21; Carter 1976: 28) there are three stems which refer to non-
singular objects: a set of objects, a set of parallel objects, and a set of two
objects (also used for rope-like objects). Dogrib distinguishes between
plurality of similar, and of disparate objects (Carter 1976: 28). This varia-
tion within the Athabaskan family led Carter (1976: 28) to a conclusion
that 'notions of non-singularity were not an integral part of the Proto-
Athabaskan classincatory verb system, but that non-singularity was
handled by semantic extension of the core system, typically by an extension
of the rope-like object category'.
Choices of classincatory verb stems may also depend on the animacy or
inherent nature of a non-singular referent. Koyukon has a classincatory
verb stem for general plural, one for plural objects scattered about and one
for plural animate (Chad Thompson, p.c.).
Tewa (Kiowa-Tanoan) has a limited set of classincatory verb stems with
the meaning 'to be in a place' (or 'to have') and 'to put or set down' (Speirs
1974). The S or O is characterized in terms of its position in space and shape
(like classincatory verbs in Enga, or Waris: Tables 6.7 and 6.10). Six classi-
ncatory verbs are used with singular and dual objects, and only five with
plural objects—the distinction between 'be sitting' and 'be lying' is
neutralized with one classificatory stem being used for both.9 See Table 10.6.

7
These stems are used for falling objects rather than for location of objects (Keren Rice,
p.c).
The different choices for classificatory verb stems referring to non-singular objects are
independent of so-called 'verb themes indicating number' (Rice 1989: 791-2), which are
mostly found for semantic groups of verbs which are different from classificatory verbs.
9
'Be lying' and 'be sitting' are also neutralized in the plural in Kiowa (Watkins 1984: 154);
however, it is not clear whether positional verbs have the same classificatory function in Kiowa
as they have in Tewa. Similarly to Kiowa (see §12.5.2), nouns in Tewa belong to six different
classes conditioned by their choice of number forms; the interrelation with classificatory verbs
which are also number-sensitive provides an unusual cross-classification.
 Classifiers and Other Categories 251
TABLE 10.6. Classificatory verbs in Tewa and their semantics
Singular/dual Plural Translation Objects classified
-?an } -k'wQ 'be sitting' Squat bulky objects, e.g. clock
-k'o: I 'be lying' Long objects, e.g. pencil
-ca, -sa: 'be attached, be a Attached or rooted objects,
container (in normal e.g. tree; or container, e.g.
position), be in a dish
container'
-win -win 'be standing' Upright objects, e.g. shovel
-yi? -yi? 'be walking about' Moving objects, e.g. dog
-na -na 'existence of time or Phenomena of nature, events,
event' institutions, time

In Creek (Muskogean), suppletive dual and plural verbs are developing

into classificatory verbs, and there is a correlation between consistency and
the physical nature of the object, and the number form of the verb (see (B)
in§6.2.3).10
An interaction between number expressed on verbs and animacy of the S
(or S/O) is frequently attested in languages with number-sensitive verbs.
Jarawara, an Arawa language from Brazil, has two genders (Dixon 1995).
The verbs 'sit' and 'stand' distinguish between animate and inanimate S in
plural, but not in singular or dual. Jarawara displays a dependency between
number and animacy which goes in the opposite direction to Tewa, where
fewer classificatory stem choices are available in plural than in singular or
dual. In contrast, the Jarawara verbs shown in Table 10.7 have more
animacy distinctions in plural than in singular or dual.

TABLE 10.7. Animacy marking on verbs in Jarawara (Arawa)

Verb Singular Dual Plural

Animate S Inanimate S
Sit -ita joro -na- -naho- sii na-
Stand -wa(a)-

Verbal classifiers often do not have to make any special reference to the
number of objects (cf. Table 6.7 and Table 6.11).
DEICTIC CLASSIFIERS can be fused with number if the number has to be
marked in an NP on a modifier, e.g. Pilaga (Guaicuruan: Vidal 1997: 85)

10
Jack Martin (p.c.) informs me that this is a typical phenomenon in many languages of the
Southwest of the USA.
252 Classifiers
da7m7e 'CL:VERTICAL:SINGULAR/COLLECTIVE', da:m7e 'CL:VERTICAL:PAUCAL'.
However, unlike noun classes, there is no dependency between the choice
of a deictic classifier and a choice made in a number system.
NOUN CLASSIFIERS or CLASSIFIERS IN POSSESSIVE CONSTRUCTIONS do not inter-
act with number since number tends to be marked elsewhere in NPs;
neither do LOCATIVE CLASSIFIERS.
Classifiers in multiple functions can be employed as derivational markers
on the head noun; they may also transform mass or collective nouns into
countable nouns—see (E) in §9.1.

10.2. Classifiers and person

The interaction between classifiers and person is limited to pronominal

elements—independent personal pronouns and cross-referencing bound
pronouns. That is, in many languages noun class and person are realized
as portmanteau morphemes.
Correlations between noun categorization and person have been
observed mostly for noun classes11 and for those noun classifiers which
tend to be used anaphorically for pronominalization.
(A) Noun class distinctions often depend on person
Most frequently, NOUN CLASS specification is made just in 3rd person.
Examples are numerous—Indo-European, Caucasian, and a number of
Australian languages, e.g. Yaigin (Australian: Crowley 1979); some lan-
guages do not have a 3rd person pronoun, and use demonstratives in this
function. Then, gender distinctions are found in demonstratives, e.g. in
Paumari (Arawa) (Chapman and Derbyshire 1991: 267). First and second
person are uniquely specified and their sex is presumably known, so gender
specification here is communicatively redundant.
Pronominal usage has been reported for NOUN CLASSIFIERS, for instance,
in Mayan languages (Mam, Jacaltec, or Akatek), and in Mixtec (De Leon
1987: 175). Since noun classifiers are used as 3rd person pronouns only, it
can be said that more classificatory distinctions are made for 3rd person
than for 1st and 2nd persons.
A few languages also show gender in 2nd (but not in 1st) person. If
gender oppositions are found in 2nd person, they will normally also be
there in 3rd, and if they are found in 1 st, which is rare, they will normally
also be there in 2nd and 3rd. The majority of Afroasiatic languages
(Semitic, Berber) distinguish masculine and feminine genders in 2nd and
'' See Chapter 2, on how there tend to be more noun class distinctions with personal and
other pronouns than elsewhere.
 Classifiers and Other Categories 253
3rd person singular and plural. latmul and Abelam, languages of the Ndu
family (East Sepik region of New Guinea), distinguish masculine and
feminine genders in 2nd and 3rd persons only, while Manambu and Ngala,
from the same family, distinguish feminine and masculine in all persons.12
There are a few exceptions. Two genders are distinguished in 1 st (but not
2nd or 3rd) person just in possessive pronouns in Kalaw Kawaw Ya, the
Western Torres Strait language (Ford and Ober 1991: 138), where a mascu-
line possessor will use ngaw and a feminine will use nguzu for 'my'. In
Maka (Mataguayo: Gerzenstein 1994: 175) feminine gender is marked on
the 1st person inclusive personal pronoun only; other personal pronouns
have no gender marking. Minangkabau (Western Austronesian) distin-
guishes two gender forms of 2nd person singular pronouns; there are no
gender distinctions for other pronouns—see Table 10.8 (adapted from
Marnita 1996: 64).

TABLE 10.8. Personal pronouns in Minangkabau (singular)

Person Masculine Feminine

1st (a)den
2nd waang/ang kau
3rd inyo

(B) Noun class distinctions often depend on a combination of categories

which include person
Noun class distinctions can correlate with person and with number. Berber
languages Siwa (Laoust 1931: 108) and Izayan (Laoust 1928: 187) distin-
guish masculine and feminine genders in 2nd and 3rd person singular but
not in the plural.
A few languages distinguish masculine and feminine genders in all
persons in plural pronouns, but only in 3rd person for singular number.
Examples are Spanish (Table 10.9) and Tariana (North Arawak; Table
10.10). In Spanish, 1st and 2nd person plural pronouns have been derived
from a combination of a personal pronoun + indefinite adjective otro
'other' (masculine), otra (feminine) (Corominas 1954: 523).13 In Tariana
feminine plural pronouns consist of a personal pronoun plus feminine
marker -ma plus plural -pe. In both languages 'masculine' plural pronouns
are used if feminine gender does not have to be specified.

12
In Manambu genders are distinguished in 1st person in the remote past, but not in other
tense forms.
13
Spanish also has an old neuter form, ello, used in fixed expressions, and as a neutral
agreement form (Corbett 1991: 214).
254 Classifiers
TABLE 10.9. Personal pronouns in Spanish
SG PL

MASC. FEM. MASC. FEM.

1 io nosotros nosotras
2 tulvos vosotros vosotras
3 el ella* ellos ellas

TABLE 10.10. Personal pronouns in Tariana

SG PL

MASC. FEM. MASC. FEM.

1 nuha waha waha-ma-pe

2 piha iha iha-ma-pe
3 diha duha naha naha-ma-pe

Further counterexamples to Greenberg's Universal 45 (see (A) in

§10.1.1) concern the mutual dependency between person, noun class
and number. In a few North Berber languages, independent pronouns
distinguish masculine and feminine genders in 2nd and 3rd person singular
and plural, and also in 1st person plural, but not in 1st singular (e.g.
Tazerwalt Shilh: Stumme 1899: 85, and closely related Ntifa: Laoust
1918: 211). Tuareg languages (South Berber, e.g. Tamachek: Prasse and
agg-albostan ag-Sidiyan 1985: 8; Simone Nientao, p.c.; Ahaggar: Prasse
1972) distinguish two gender forms for all persons in plural and in 2nd
person singular; there are no distinctions in 1st and 3rd person singular
(see Table 10.1). Nama (Khoisan: Hagman 1977) distinguishes three
genders (feminine, masculine, and common) in 2nd and 3rd person singu-
lar, and in all persons in plural; dual has two gender forms (masculine and
feminine-common) for all persons, while 1st person singular has no gender
marking.
The choice of noun class distinctions may correlate with tense and
person. Non-present tense forms of the verb in Hebrew have distinct
masculine and feminine for both 2nd and 3rd person in the singular; in
the plural we find: 2nd person masculine, 3rd person masculine, and a
single form covering both 2nd and 3rd person feminine. That is, the con-
trast between 2nd and 3rd persons is neutralized in feminine gender and
 Classifiers and Other Categories 255
plural number. See the future paradigm of katav 'write' in Table 10.15 (see
also example from Tucano languages in §10.7).

10.3. Classifiers and grammatical function

Of all the noun categorization devices, only noun classes and verbal
classifiers interact with the marking of the grammatical function of an
NP in a clause, since both classifier types have the clause as their scope.
Grammatical functions can be marked with (a) nominal case (in depen-
dent marking languages) or (b) verbal cross-referencing (in head marking
languages) (see Nichols 1986). The two types of marking interact with
noun classes.

10.3.1. Noun classes and grammatical function

There are mutual dependencies between choices available in noun class
systems and those in case systems.
(A) Dependencies between case and noun class
The choices available in case systems may depend on the choice made in noun
class system (cf. Aikhenvald and Dixon 1998). In Latin (and many other
Indo-European languages), case distinctions depend on gender. All nouns
belonging to masculine and feminine genders (whether their reference is
human, animate, or inanimate) distinguish nominative and accusative cases,
but this distinction is neutralized for all nouns belonging to neuter gender.
Some determiners, e.g. demonstratives, have the same ablative form for
masculine and neuter singular, and a different form for feminine singular
(cf. Lyons 1968: 293).
Further interactions involve noun classes, case, and number. In Latin,
demonstratives do not distinguish any gender forms in dative and ablative
plural; and only two forms (masculine-neuter and feminine) are distinguished
in genitive plural; all three genders are distinguished in nominative plural.
More complicated interactions can involve noun classes, case, person
and number. In Lithuanian, masculine and feminine genders are distin-
guished in nominative, accusative, and vocative cases in 1 st and 2nd person
dual; for 3rd person dual, they are distinguished in all cases except for
genitive (Peterson 1955: 48-9).
The realization of noun classes in different persons may correlate with
grammatical function. In Berber languages there are three series of pro-
nouns: independent subject, direct object, and indirect object (also used
with prepositions and as possessive pronouns). Genders are regularly dis-
tinguished in 2nd and 3rd persons in all the three series in the singular. The
256 Classifiers
only exception is the 3rd person indirect object pronoun, which employs
just one form for both genders.14 Berber languages show a correlation
between noun classes, person, case forms, and number. We have seen above
that some Berber languages (e.g. Siwa, Izayan, Zemmur) do not distinguish
genders in plural; Kabyle and Tamachek distinguish genders in 2nd and
3rd person in the plural in direct and indirect object pronouns, and in all
persons in independent pronouns. Shilh distinguishes two genders in 2nd
and 3rd person singular and all persons in plural only in independent
subject pronouns. Direct and indirect object pronouns have gender distinc-
tions for all persons and numbers excluding 1st person plural.
(B) Portmanteau realization of noun class and case
This is widespread in fusional languages of the Indo-European type, e.g.
Latin, Sanskrit, or Russian, where case forms of all nominals usually
convey information about noun class/gender, case, and number. The para-
digm of is 'this' in Latin is given in Table 10.11.

TABLE 10.11. Paradigm of Latin is 'this'

Case SG PL

MASC. FEM. MASC. NEUT. FEM.

Nominative is id ea ii ea eae
Accusative eum id earn eos ea eas
Dative ei Us
Ablative eo ea Us
Genitive eius eorum earum

(C) Dependencies between cross-referencing and noun class

In head-marking languages, noun classes can be marked on verbs to cross-
reference S, A, and O; then cross-referencing affixes combine the reference
to noun categorization and syntactic function. Often, the choice of noun
class depends on the choice made in the person system. For instance, in
Anindilyakwa (Australian) different series of verbal prefixes cross-refer-
ence A, S, and O, and four noun class forms are distinguished in bound
pronouns, while five noun class forms are distinguished in free forms
(Leeding 1989).
The choice of noun class in cross-referencing may depend on person and
number. In the majority of Arawak languages feminine/non-feminine
pronominal gender distinctions are only made in 3rd person singular
14
For an attempt at a reconstruction of Proto-Berber personal pronouns, see Aikhenvald
(1986).
 Classifiers and Other Categories 257
cross-referencing prefixes (typically used for A/Sa) and suffixes (used for
0/S0).
The choice of noun class can depend on the grammatical function
realized in cross-referencing. Chinantec languages distinguish animate
and inanimate gender realized in agreement with modifiers in NPs; verbs
are inflected to agree with S/O depending on whether it is animate or not.
There are no animacy distinctions for A (see Westley 1991: 11, 23, for
Tepetotutla Chinantec, and Anderson 1989: 6, 18, for Comaltepec
Chinantec). In Ungarinjin (Australian) four noun classes are distinguished
in S/O bound pronouns, and none is distinguished for A pronouns, while in
Alawa and Tiwi two noun classes are distinguished for A/S and none for O
(Dixon forthcoming).
'Oro Nao (Chapacuran: Everett and Kern 1997: 329) always distin-
guishes masculine, feminine and neuter genders of third person O. At least
in some of the verb paradigms gender is not distinguished for A/S. In
Lavukaleve (Papuan, Solomon Islands: Angela Terrill p.c.) three genders
are distinguished in cross-referencing prefixes which mark O, but not for
those which mark S/A. In Emmi (Australian: Ford 1998) gender distinc-
tions are neutralized in 'minimal' transitive subject (A) cross-referencing
markers, but not in other numbers or functions.
In a language with two noun class systems these may relate to different
case functions. In Paumari (Arawa: §2.7.2) there are two types of tran-
sitive construction. In one of these masculine/feminine gender of the
subject (A) is cross-referenced on the verb, and in the other that of
the object (O) gets cross-referenced. In both constructions, however,
the verb cross-references the shape-based ka-noun class of the object
(O). In intransitive clauses, both gender and ka-noun class cross-reference
the subject (S).

10.3.2. Verbal classifiers and grammatical function

Verbal classifiers and classificatory verbs always contain reference to the
grammatical function of an argument. This function is usually S/O (see
Keenan 1984 and §6.3). In Palikur (Aikhenvald and Green 1998) the
classifiers refer to O and derived S of a passive, or to the S of a stative
intransitive verb (see §6.4.1).

10.4. Classifiers and types of possession

The way a noun can be possessed may correlate with the intrinsic proper-
ties of its referent. The division of nouns into alienably and inalienably
possessed may be considered as a kind of noun categorization device (see
258 Classifiers
Nichols 1992: 134-5, and the discussion in §5.3.2).15 Of all the noun
categorization devices only noun classes and the three types of possessive
classifier show an interaction with types of possession (since only these
classifier types can have possessive NPs as their scope).

10.4.1. Noun classes and types of possession

Noun class realization and agreement in a possessive NP can depend on the
type of possessive construction. Some languages make more noun class
distinctions in constructions with INALIENABLE possession than in those
with ALIENABLE possession. Jarawara (Arawa) distinguishes alienably and
inalienably possessed nouns (see Dixon 1995; Dixon and Vogel forthcom-
ing). The gender of the possessor is cross-referenced on the inalienably
possessed noun, and not on an alienably possessed one. Examples 10.1 and
10.2 show that the gender of the possessor is marked on the inalienably
possessed noun 'home'.
10.1. Okomobi taboro
Okomobi(masculine) home+MASC
'Okomobi's home'
10.2. Sasa tabori
Sasa(feminine) home + FEM
'Sasa's home'
The noun 'house' in 10.3 and 10.4 is alienably possessed, and there is no
gender agreement with a possessor.
10.3. Okomobi kaa jobe
Okomobi(masculine) POSS house
'Okomobi's house'
10.4. Sasa kaa jobe
Sasa(feminine) POSS house
'Sasa's house'
Interrelations between possession type and gender and noun class agree-
ment may be more complicated. In 'Oro Nao (Chapacuran: Everett and
Kern 1997: 299) gender is marked on possessed forms of inalienably pos-
sessed nouns, as in Arawa languages, e.g. capija-in wao' (mouth-3neut
small.basket:neuter) 'rim of the basket'; capija-con worn (mouth-3masc
cotton:masculine) 'hem of the skirt'; capija-cam narima" (mouth-3fem
15
Possession types may be more complicated than just alienable and inalienable. There may
be special type of 'unpossessable' nouns, or 'body part' or 'kinship possession'. For a useful
overview, see Chappell and McGregor (1996).
 Classifiers and Other Categories 259
woman: feminine) 'woman's mouth'. Unlike Arawa, Chapacuran languages
have three genders: feminine (used for human females, collective nouns,
mixed gender groups), masculine (for human males, high animates, and
culturally significant objects, and also for nature phenomena, insects, fish,
etc.), and neuter (used for most inanimate objects, newly introduced
objects/animals/plants, loans, nominalizations). Gender marking is realized
on modifiers, possessed forms and in verbal cross-referencing.
In East Tucano languages and in Tariana (North Arawak) more noun
class distinctions are found in constructions with ALIENABLE possession
than in those with INALIENABLE possession. These languages have a closed
set of pronominal noun classes, and a large set of morphemes used as
noun class markers and as numeral and noun classifiers (see Chapter 9;
Aikhenvald 1994a: 427; 1996a). Possessed classifiers in alienable possessive
constructions appear on the possessive morpheme ya. Thus:
10.5. Tucano: mi't ya-gi pu-gi
2SG POSS-CL:LARGE hammock-CL:LARGE
'your hammock'
10.6. Tariana: pi-ya-ku ama-ku
2SG-POSS-CL:EXTENDED hammock-CL:EXTENDED
'your hammock'

10.4.2. Classifiers in possessive constructions and types of possession

The choice of a RELATIONAL CLASSIFIER usually depends on possession type
(see §5.3.2). Relational classifiers which contain information on how an
object can be handled, or whether it is consumable or not, are restricted to
alienably possessed nouns.
In numerous cases, POSSESSED CLASSIFIERS are also used exclusively with
alienably possessed nouns. This is the case in Yuman and Uto-Aztecan
languages in North America, and in Carib, Tupi-Guarani, Macro-Je,
Nade'b (Maku) and a few Arawak languages (Island Carib, Bahwana,
Achagua and Palikur) in South America (see §5.2).
Possessed classifiers are used independently of possession type only in a
few North Arawak and Tucano languages and in the Mba group (Ubangi,
Niger-Congo) (see §5.2).
It is hard to make any generalizations about possessor classifiers. In Daw
(Maku: Martins 1994) discussed in §5.4, possessor classifiers can only be
used with alienably possessed nouns. In Nasioi (Papuan: Bougainville),
which may feature another example of possessor classifiers, they are also
restricted to just one type of possessive constructions (see 5.47; Hurd 1977:
138).
260 Classifiers
10.5. Classifiers and politeness

There may be a grammatical category indicating politeness or social status

marked on nouns, or on verbs. Politeness can be expressed in the same
morphosyntactic contexts as classifiers. Consequently, one expects inter-
relations between classifiers and politeness. The choice of a politeness form
often correlates with the animacy or humanness of the referent: polite
forms, or forms sensitive to social status, are applicable only to human
(or, more rarely, to animate) referents. Social status is frequently expressed
in systems of NUMERAL CLASSIFIERS for humans (see Chapter 11).16 For
instance some Austroasiatic languages have elaborate systems of 'honorific'
numeral classifiers based on social status. So do some Indo-Aryan lan-
guages (e.g. Assamese: see Table 4.1), Korean (Lee 1997), Burmese and
Thai (Bisang forthcoming).
Social status is often expressed in a system of NOUN CLASSIFIERS (see
§11.2.2 and Chapter 3; also see the discussion on noun classifiers and social
deixis in Mixtec in De Leon 1987: 171-2).
Occasionally, the choice of a possessed classifier can depend on the social
status of speakers, or the speech style. Ponapean (Micronesian: Keating
1997) has twenty-two possessed classifiers used in the common speech
register. Only objects related to rank in the society are reclassified in the
honorific register. This is shown in Table 10.12 (based on Keating 1997:
252-3, 255, 258).
In contrast, in the humiliative register (used to address socially inferior
people) there is just one kind of possessive classifier, a combination of
general classifier ah and a specific humiliative classifier tungoal (lit. 'food');
all the semantic distinctions present in honorific and common style classi-
fiers are neutralized (see §12.3, for a socio-cultural explanation). Table
10.13 provides examples of noun categorization in humiliative speech
and in the common speech register (Keating 1997: 255).
As in other Micronesian languages, possessive classifiers in Ponapean
appear only in constructions with alienably possessed items. This is, how-
ever, not true for the possessive classifier used in humiliative speech which
is used with any item in humiliative speech, independently of whether it is
treated in other speech styles as alienably or inalienably possessed; thus,
body parts which are inalienably possessed in common speech, e.g.
moahng-ei (head-lsg.poss) 'my head', take the only possessive classifier,
ah tungoal, in humiliative speech, e.g. ei tungoal moahng (POSS.CL:GEN+
1sg.poss POSS.CL.HUMIL:FOOD head) 'my (lower status) head' (Keating 1997:
256).

16
See also Senft (1996: 202, 203 and passim) on the interaction between classifiers and
sociolinguistic variables in Kilivila.
 Classifiers and Other Categories 261
TABLE 10.12. Possessed classifiers and speech styles in Ponapean

Common Semantics Honorific Semantics Humiliative

ah General classifier
nah General, dominance of sapwellime General classifier
possessor
kene Edible koanoat Possession of food/drink by
paramount chief
pwenieu Id. by paramount
chieftainess
nime Drinkable sahk Id. by secondary chief
sapwe Land nillime Land
imwe Buildings tehnpese Dwellings
were Vehicles tehnwere Vehicles
kie Sleeping pads moatoare Things to sleep on ah tungoal
ipe Sleeping covers
ulunge Pillows
rie Siblings
kiseh Relatives
ullepe Maternal uncles
wahwah Nieces, nephews
sawi Clan members Same as in common speech
pelie Peers
seike Catch
pwekidah Share of feast food
mware Name, title, garland
ede Names
tie Earrings
dewe Location

TABLE 10.13. Examples of possessed classifiers in common and humiliative speech in

Ponapean

Common Humiliative Translation

kene mwahng (POSS.CL:EDIBLE taro) ah tungoal mwahng (POSS.CLIGEN.POSS 'her taro'

CL HUMIL:FOOD taro)
nahpwihk (POSS.CL:GEN pig) ah tungoal pwihk 'her pig'
were sidohsa (POSS.CL: VEHICLE car) ah tungoal sidohsa 'her car'

The existence of an elaborate system of politeness in a language does not

presuppose classifiers which refer to social status; politeness is important as
a grammatical category in Japanese, however, there is just one honorific,
unlike Korean—see §12.4.
In some languages the choice between various first and second person
pronouns tends to be made according to the social status of the speaker,
or of the addressee. Then politeness and gender can be considered
262 Classifiers
portmanteau. In Minangkabau 2nd person pronouns combine information
on the sex and social status and age of the referent: waang, or ang, '2nd
person singular masculine' and kau '2nd person feminine' are used to refer
to people younger than the speaker, while uda '2nd person singular mascu-
line' and uni '2nd person singular feminine' are used to refer to people older
than the speaker and higher on the social hierarchy (Marnita 1996: 64).17
Meanings related to politeness, such as social status and respect, can be
expressed through NOUN CLASS. In Jarawara, a woman can be referred to
with the marked masculine gender instead of the unmarked feminine as a
sign of respect (R. M. W. Dixon, p.c.). In Amharic, the marked feminine
gender is used for respectful reference to men (Mengistu Amberber, p.c.).
'Respect' vs. 'disdain' can be an extension of positive vs. negative values
assigned to genders. In Lokono (Arawak: Pet 1987: 26–7; (C) in §2.4.4;
§11.2.1), masculine gender covers (a) all males of the speaker's tribe, except
if they are despised; (b) males who are not of the speaker's tribe, if they are
friends of the speaker or if a relationship of mutual respect exists with the
speaker, and also (c) animals, objects and spirits 'considered to be good and
desirable or when they are protagonists in stories'. Feminine gender includes
(a) all females and males who are despised; (b) males who are not of the
speaker's tribe; and (c) animals, objects, and spirits not included in masculine
gender. The 3rd person human plural is used 'for male and female humans
who are Arawaks or with whom the speaker is in sympathy' (Pet 1987: 27).

10.6. Classifiers and declensional classes

Declensional classes and the organization of paradigms is usually indepen-
dent of classifier systems. However, NOUN CLASS and declension correlate in
languages with morphological gender assignment, e.g. Indo-European lan-
guages such as Latin, Lithuanian, or Russian. For instance, in Latin 2nd
declension nouns are usually masculine, and 1st declension nouns are
usually feminine. In Modern Irish, some nouns choose 'feminine' or 'mas-
culine' declension forms due to the historical merging of paradigms (O
Siadhail 1989: 145-6).18
17
Some Asian languages, e.g. Thai, Burmese, Vietnamese, and Japanese, have different sets
of personal pronouns (for all persons) the choice of which is determined by the sex of the
speaker and their status (cf. Cooke 1986). These systems cannot be considered on a par with
noun classification devices, but are similar to other elements of men's and women's speech,
such as different lexical choices or pitch. Using different kinship terms of address for female
and male relatives is not treated here as a gender distinction.
18
Noun class assignment in Arapesh was analysed as a function of a morphological
declensional class by Aronoff (1991). However, Nekitel (1985; 1986; forthcoming) presents
convincing arguments in favour of semantic assignment of at least some classes in Abu'
Arapesh.
 Classifiers and Other Categories 263
There are no correlations between other classifier types and declensional
classes. This may be explained by the fact that only noun classes can be
assigned on morphological principles (see §2.3).
A few languages which have lost genders have retained traces of gender-
like distinctions in declensions. Lezgian, a Northeast Caucasian language
which lost its gender system, has oblique stem suffixes which correlate
with the semantics of nouns: -Adi is used with nouns which denote a non-
discrete mass, -rA is used with most nouns which denote animals, or
people (Haspelmath 1993: 76-7; see Alexeyev 1985: 27 ff, for other Lezgic
languages).
In Armenian, which lost Indo-European gender distinctions, the opposi-
tion 'person'/'non-person' is present in inflectional paradigms (Garibian
1976). Further examples of a correlation between the organization of
paradigms and categorization are given by Nichols (1992: 134–6), under
the heading of'non-agreeing classification'. This appears to be the property
of the organization of paradigms in Indo-European languages, where con-
cordial gender displays a correlation with the organization of declensions.
Gender and declensional classes seem to be independent and orthogonal
systems in some Daghestanian languages. Chechen and Ingush have
genders, declension classes marked by thematic suffixes and declension
subclasses marked by different ergative case and/or plural endings. Human
and/or animate can be used in the description of the semantic basis of all
three, but all these systems are independent (Nichols 1992: 134).

10.7. Classifiers and verbal categories

Of all the classifier types, only NOUN CLASSES and VERBAL CLASSIFIERS interact
with verbal categories.
NOUN CLASSES can interact with tense. The choices available in the noun
class system may depend on choices made in the tense system. Then noun
classes are distinguished in some tenses, but not in others.19 For example, in
Manambu gender is distinguished in 1 st person singular in the immediate
past tense, but not in the remote past.20 In Russian the verb in non-past
tenses inflects for person and number of subject but not gender; in the past
tense the verb marks singular and plural and, within singular, masculine,
feminine, and neuter gender, but not person. The person system thus
applies only in non-past; three genders are distinguished only in the past
singular. The paradigm of dela-t' 'do' is shown in Table 10.14.
19
In all these cases, noun class distinctions also depend on person and number, alongside
tense. See Aikhenvald and Dixon (1998).
20
Manambu has complicated correlations between gender marking on verbs, mood, and
tense. See Aikhenvald (1998a).
264 Classifiers
TABLE 10.14. Paradigm of dela-t' 'do' in past tense in Russian
Present Past
1SG delaju IPL delaem MASC. SG delal PL. delali
2SG delaesh 2PL delaete FEM. SG delala
3SG delaet 3PL delajut NEUT. SG delalo

In Hebrew, with two genders, the verb in present tense is inflected

for gender and number (the gender distinction is made in both singular
and plural, but not person). In other tenses the verb inflects for
number and person, with gender being specified just for 2nd and 3rd
person. That is, person is only distinguished in non-present, and gender
is restricted to second and third person in non-present. This is illustrated
with the paradigm of the verb katav 'write' in past and present in Table
10.15.

TABLE 10.15. Paradigm of katav 'write' in Modern Hebrew

Past Future Present
1 SG katavti IPL katavnu extov nixtov FEM. SG kotevet
2SG F katavt 2PL F ktavten tixtavu tixtovna MASC. SG kotev
2SG M katavta 2PL M ktavtem tixtov tixtavu FEM. PL kotvot
3SG F katva tixtavi tixtovna MASC. PL kotvim
3SG M katav 3PL katvu yixtov yixtevu

There are diachronic reasons for this split in grammatical marking. The
past in Russian and the present in Hebrew have both developed out of
deverbal adjectives, and inflect in a manner typical of nominals.
An interaction between noun class, person, and tense is observed in
Tucano (West 1980), which has two systems of noun categorization (simi-
larly to other languages mentioned in §2.7): the 'pronominal' one (a mascu-
line/feminine/inanimate opposition) is used with verbal cross-referencing
and personal pronouns; the 'nominal' one, which is much larger, is used
with the remaining modifiers—numerals, deictics, adjectives. Pronominal
classes are distinguished for all three persons in future tenses; in past and
present tenses they are distinguished only for 3rd person.
Noun class distinctions can depend on clausal polarity (see the discussion
in Aikhenvald and Dixon 1998). In some languages all specifications of
person, number, and gender are neutralized in negative forms, e.g. Tariana,
or Manambu. Similarly, in Palikur, two genders are distinguished in certain
aspects; but these aspects are not used in negative clauses: there are fewer
 Classifiers and Other Categories 265
aspect choices in negative clauses than in positive ones. Consequently,
gender is not contrastive under negation. Gender agreement is marked
on the durative suffix which is typically used with stative verbs (10.7). In
10.8, the negative counterpart of 10.7, it is suppressed.21
10.7. tino barew-yo
woman be.pretty/clean-DUR.F
'The woman is pretty.'
10.8. tino ka-barew
woman NEG-be.pretty/clean
'The woman is not pretty.'
The choice of VERBAL CLASSIFIERS may depend on (a) verb class, or (b)
other verbal categories, e.g. the choice made in the tense-aspect system, or
in the voice system.22
(A) The choice of verbal classifiers depends on verb class
For example, in Koyukon (Athabaskan: Axelrod forthcoming) verbal clas-
sifiers (traditionally named 'genders' in Athabaskan linguistic tradition)
never occur with attributive verbs (e.g. 'be happy'). Suppletive classifica-
tory verbs are never used attributively.
Palikur has two sets of verbal classifiers—one used with transitive verbs,
and the other with intransitive stative verbs. Intransitive active verbs do
not take classifiers.23
(B) Verbal classifiers may be restricted to certain verbal categories
In Tariana and Baniwa (North Arawak) classifiers are used only with
relativized verbs, verbs in purposive mood, and in the form of argument
manipulating derivation. In Palikur, verbal classifiers can only be used if
the O of a transitive verb, or the S of a stative verb are completely
involved in action or state; consequently, classifiers are almost always
used with verbs marked for completive aspect (see further examples in
§6.4.1).

21
However, the gender agreement is obligatory in negative clauses with emphatic contras-
tive negation (marked by both negative prefix ka- and a negative suffix -ma), e.g. tino ka-
barew-yo-ma (woman NEG-be.pretty/clean-DUR.F-NEG) 'The woman is not pretty at all'.
22
An analogy between numeral classifiers (also used with demonstratives and interroga-
tives, and with some adjectives) in Newari and morphemes which marks repetitive action with
verbs depending on the type of action performed and thus, in a way, 'classifiying' verbal
actions, is drawn by Bhaskararao and Joshi (1985) for Newari (Tibeto-Burman). This analogy
is radically different from the dependencies considered in this section.
23
This dependency has to do with semantic restrictions on classes of verbs used with
classifiers; see Chapter 11.
266 Classifiers
10.8. Classifiers and deictic categories
Categories typically found in deictics (e.g. distance from speaker or hearer,
or visibility) interact with noun categorization devices expressed in deictics—
NOUN CLASSES and DEICTIC CLASSIFIERS.
Noun classes can be portmanteau with deictic categories, e.g. Warekena
(North Arawak) eni 'this: masculine', ayupalu 'this: feminine'. The choice
of noun class may depend on degree of distance realized in deictics. Palikur
(Table 8.6) has five sets of deictics; three genders are distinguished in all but
the series which refers to objects located far from speaker and near hearer.
In Bare (North Arawak: Aikhenvald 1995a: 22) masculine and feminine
genders are regularly distinguished for the distal deictic only; no distinction
has been observed for proximal ('near to speaker') or medium distance
('nearer to hearer').
Deictic classifiers combine reference to visibility, distance, and proximity
of the referent, and its shape, directionality, or other nature-based proper-
ties (see §11.2.6). In Eskimo deictic classifiers interact with visibility; in
Toba and Pilaga they contain additional reference to the presence or
absence of an object in a visual field.

10.9. Classifiers, derivation, and lexicon

Classifiers interact with derivational devices; in this way they can enrich the
lexicon. Noun classes and genders often combine inflectional (i.e. concor-
dial) and derivational properties (see Payne 1990; §2.4.1).24 Gender mar-
kers are productively used in derivation in Brazilian Portuguese, e.g.
ministr-o 'minister', ministr-a 'she-minister'; juiz 'judge', juiz-a 'she-judge'.
In Bantu languages, e.g. Swahili, most stems usually occur with a prefix of
one class. Prefixes can be substituted to mark a characteristic of an object.
M-zee means 'old person' and normally takes the human class prefix m-. It
can be replaced by ki- (inanimate class) to yield ki-zee 'scruffy old person'
(Dixon 1982: 166). Classifiers in multiple environments in Amazonian
languages are often used as derivational devices (see (E) in §9.1).
In other languages there may be a number of genders expressed by
derivational affixes, but only a subset of these operates in an agreement
system. Gondi (South Dravidian) has feminine, masculine, and neuter in
derivation, but only two genders (masculine and non-masculine) in agree-
ment (Subrahmanyam 1968) (see also Evans forthcoming; 1997, for further
differences between head and agreement classes in Mayali).
24
The distinction between head classes and agreement classes (see §2.6.1; Evans 1997)
illustrates the difference between the two functions.
 Classifiers and Other Categories 267
Classifiers of all types provide cross-categorization of nouns, and offer
additional information about a noun's referent (see §12.1.2). Noun classi-
fiers are often used to highlight different aspects of the meaning of a
polysemous noun (see §3.2.2). This variable classification is a characteristic
property of noun classifiers, numeral classifiers, and verbal classifiers (see
Table 12.4, for an example of the 'reclassification' of an inanimate noun in
Burmese, and discussion in Chapter 12).
Variable noun class marking in Maung (Capell and Hinch 1970: 47-52)
has a similar semantic effect. Maung has five noun classes the assignment
of which is partially based on semantics. Class 1 consists of names of male
beings; Class II of names of female beings; Class III includes objects
associated with the ground (except plants); Class IV consists of trees and
their parts; and Class V of vegetable foods and plants. Noun classes are
overtly marked on nouns, and also realized through subject and object
agreement on verbs and/or agreement on adjectives and possessives.
Changing the noun class of some nouns may result in a change of meaning.
The noun -nimi expresses the idea of a 'long bone'. Used with a 'human
male' Class 1 prefix, (j) i-nimi means 'his backbone'; with a Class II, female
human class, ninj-imi means 'her backbone'. With a Class IV prefix, ma-
nimi means 'trunk of a tree', and with a class V prefix ad-imi means 'main
radicle' of a potato plant or yam vine (Capell and Hinch 1970: 47). Table
10.16 illustrates how noun classes are used with another noun root, mawur
'arm, long part of.

TABLE 10.16. Variable noun class assignment in Maung

Class I i-mawur 'man's arm'
Class II ninj-mawur 'woman's arm'
Class III u-mawur 'tributary of a river'
Class IV ma-mawur 'branch of a tree'
Class V a-bawur 'tendril of a vine'

In systems with the same or overlapping sets of classifiers in different

functions, classifier morphemes are typically used in the same way. In
Baniwa (North Arawak), a noun can be associated with more than one
classifier in accordance with which the semantic aspect of the referent the
speaker wishes to stress—see 2.18-20.
The taxonomies expressed by classifier systems differ not only from
those encoded by nouns; they also differ from one another (see §11.1).
As Benton (1968: 142-3) put it, in his analysis of the system of two
classifier types in Truquese, an Austronesian language, 'the classifiers . . .
thus at the same time provide a means for ordering the universe, and a
method for structuring concepts without multiplying vocabulary'.
268 Classifiers
This is yet another way in which the presence of classifiers in a language
may affect its lexicon. Languages with classifiers, the semantics of which
includes reference to shape, dimensionality, and other nature-based proper-
ties of entities, tend to have fewer lexical items—adjectives, or functionally
related categories—which describe shapes or dimensions. Along with a rich
system of shape-based classifiers, Palikur has very few stative verbs cor-
responding to adjectives in other languages for such semantic groups as
dimension. The same lexical item, pugum-lpugub-lpugu- can be translated
as 'thick', 'broad', 'large', and 'big' with further semantic distinctions
being made by classifiers. Thai and Lao have one stative verb meaning
'big' or 'large'; further subtleties referring to shape are expressed with
classifiers (Adam Chapman, p.c.). Tariana has just one word for 'big'
(covering the same semantic range as in Palikur); unlike Thai and Lao,
there is no adjective meaning 'round' since the concept is expressed through
a classifier for round objects. (Note that Bare and Warekena—languages
from the same North Arawak family without extensive classifier systems—
have lexical items for these concepts.)
The ways the presence of classifiers in a language can reduce the actual
lexicon by expressing the same concepts through grammar relates to a well-
known facet of the interface between grammar and lexicon. Another
problem is which concepts tend always to be realized in grammar, and
which can 'shift' into the lexicon, or be realized in both. The human and
non-human distinction is widespread in classifier systems; however, all
languages, including the ones with classifiers, also have other—more
lexical—ways of distinguishing humans from non-humans, or persons
from non-persons. Thus, grammar and lexicon feed each other in the
semantic domains covered by classifiers.

10.10. Conclusions

How classifiers of different types interact with nominal and verbal gram-
matical categories is summarized in Table 10.17.
NOUN CLASSES interact with all the categories described here, with the
exception of politeness (though they also interact with social status in a
limited way). They show mutual interdependencies with number, person,
marking of grammatical relations, possession, verbal, and deictic categories;
they can also be realized as portmanteau morphemes with these categories,
and with declension markers and derivational morphemes.
NUMERAL CLASSIFIERS show no correlations with categories other than
politeness and, in a limited way, number.
NOUN CLASSIFIERS interact with person and politeness; they can also be
used for derivational purposes. POSSESSIVE CLASSIFIERS interact just with
TABLE 10.17. Classifiers and their interaction with other grammatical categories
Noun Numeral Noun Possessive Verbal Locative Deictic
classes classifiers classifiers classifiers classifiers classifiers classifiers
Number yes limited no no yes no (yes)
Person yes no no no no no no
Grammatical function yes no no no yes no no
Types of possession yes no no yes no no no
Politeness limited yes yes no no no no
Declensions yes no no no no no no
Verbal/clausal categories yes no no no yes no no
Deictic categories yes no no no no no yes
Derivational categories yes no yes no no no no
Symbols used: 'yes' indicates the existence of a dependency, or of a correlation, 'no' indicates its absence, and (yes) means that further investigation is
required.
270 Classifiers
types of possession. VERBAL CLASSIFIERS show interdependencies with num-
ber, with the ways grammatical relations are expressed, and with various
verbal categories. DEICTIC CLASSIFIERS interact with deictic categories (such
as distance and visibility), and sometimes with number. LOCATIVE CLASSIFIERS
show no interdependencies at all.
The ways different classifier types interact with other categories depend
on (a) their scope and (b) possibilities of their realization as bound mor-
phemes or as free morphemes. NOUN CLASSES can have an argument NP and
a clause as their scope; this accounts for the maximum amount of inter-
actions for this classifier type. They are often realized as portmanteau
morphemes with other categories—note that noun class markers are never
free morphemes, so they are likely to fuse with other markers.
The scope of VERBAL CLASSIFIERS is the clause; they also categorize the
argument, hence interactions with number, with grammatical relations, and
with verbal categories (some of which could alternately be considered
clausal; see Aikhenvald and Dixon 1998). Again, they are never free mor-
phemes, so one would expect a considerable amount of fusion with other
categories. CLASSIFIERS IN POSSESSIVE CONSTRUCTIONS have a possessive NP as
their scope; their interaction with possessive categories is expected.
NUMERAL and NOUN CLASSIFIERS show fewer correlations with other cat-
egories, and so do DEICTIC and LOCATIVE CLASSIFIERS. This can be accounted
for by limitations of their scope; they can also be realized as independent
morphemes.
In languages with more than one noun class system, different noun
classes may interact with different categories. All types of classifier show
some interaction with the lexicon—this topic requires further study.
The ways in which classifiers used in different morphosyntactic environ-
ments interact with different categories is a direct consequence of their
morphosyntactic loci of marking, and their scope. We return to this topic
in Chapter 15.
 11 Semantics of Noun Categorization
Devices

All noun categorization devices have some semantic basis. Different

classifier types tend to correlate with different semantic parameters.
Many systems allow variable choice of classifiers; then classifiers may
specify the meaning of a polysemous noun. The semantic parameters
employed in noun categorization are discussed in §11.1. The semantics of
each type of classifier is considered in §11.2. The semantic organization of
classifier systems and their functions are considered in the next chapter.

11.1. Semantic parameters in noun categorization

All classifier systems employ a number of basic semantic parameters
(§11.1.1). There are additional semantic properties of objects which may
be relevant for noun categorization (§11.1.2).1 The semantic relationships
between a classifier and a noun are dealt with in §11.1.3.

11.1.1. Basic parameters of categorization

A number of basic semantic parameters tend to be encoded in different
types of classifiers.2 These parameters fall into three large classes: animacy,
physical properties, and function.3
1
The semantic features discussed here are probably universal, or close to universal (this is a
topic for a future study). The semantic parameters used in noun categorization are framed
within a typological approach followed by Frawley (1992) and Dixon (1982).
2
There have been various attempts at identifying cross-linguistic universal semantic proper-
ties of noun classifications devices (see Adams and Conklin 1973; Conklin 1981; Denny 1976;
Allan 1977; Dixon 1982; Craig 1991; forthcoming; Frawley 1992; Kiyomi 1992; Croft 1994).
Croft (1994) made an attempt to establish a number of implicational hierarchies of semantic
distinctions associated with different types of classifier. The main problem with this kind of
approach appears to be the possibility of using the same morpheme in multiple classifier
environments. See §11.2.8.
3
Further semantic characteristics may be considered concomitant with each of these.
Frawley (1992: 68 ff.) considers eight classes of entities which cover different semantic divi-
sions of nouns only some of which are reflected in classifier semantics. In particular, SPECIFICITY
(Frawley 1992: 69) is only indirectly connected with classifier systems. BOUNDEDNESS (Frawley
1992: 81) is associated with countability, and is particularly relevant for numeral classifiers
(§11.2.3). Other classes discussed by Frawley are ANIMACY, GENDER, KINSHIP, SOCIAL STATUS,
PHYSICAL PROPERTIES, and FUNCTION. See discussion in this chapter.
272 Classifiers
Animate nouns4 can be further divided into human/non-human, person/
non-person, or relating to sex. Humans can be classified according to their
social status and function, age, and kinship relationship. The notion of
animacy plays a significant role in the morphology and syntax of many
languages (see Comrie 1981; Frawley 1992: 89). Though an animate/
inanimate distinction is found in the lexicon and grammar of most lan-
guages, language-internal criteria have to be applied to decide how entities
are divided into animates and inanimates. This binary division can be found
in Russian (where all living beings, within the masculine gender, belong to
the animate declension: see (C) in §2.4.4, and Table 2.7), or in Jarawara
(Arawa) (where all animate beings must be referred to with the third person
plural pronoun mee: R. M. W. Dixon, p.c.). On the other hand, Baniwa
(Arawak) has a classifier for humans, while animals are classified by shape,
similarly to inanimate nouns; personal pronouns can be used with humans
and with higher animals (e.g. mammals) and birds, but not with insects.
Languages also differ in how they treat supernatural beings: benevolent
gods and angels are frequently personified, i.e., treated on a par with
humans, or at least with animates, while malevolent spirits, e.g. ghosts,
may be treated like inanimates, or even be completely outside the system.
Thus, some Austroasiatic languages simply have separate classifiers for
supernatural beings (see examples of Mon in Adams 1989: 58, Khmuic
in Adams 1989: 60). In Minangkabau the human numeral classifier is used
with angels; but bad spirits such as ghosts are simply not classified.
Inanimate nouns—and, more rarely, animates—can be classified accord-
ing to their physical properties and function. The most common properties
reflected in classifier systems are listed below. Properties (a-h) relate to the
time-stable or inherent properties of an entity. Properties (i) and (j) relate
to its temporary state, e.g. configuration or arrangement;5 they are often
encountered in mensural numeral classifiers. The way objects can be
handled or used (property (h)) may correlate with their inherent as well
as temporary properties. Each property is exemplified in §11.2.
(a) EXTENDEDNESS has two sub-parameters: SHAPE/DIMENSIONALITY and
DIRECTION.
DIMENSIONALITY has three values: one-dimensional (or long), two-

4
Animate' is defined by Trask (1993: 16) as 'denoting a noun or noun phrase which is
perceived as referring to a conscious, volitional entity, a human or higher animal' (cf. also
Matthews 1997: 19). In some languages insects and fish are also considered animate. Semantic
parameters of animacy and humanness are universal; they are found in most systems of noun
categorization, i.e. classifiers of distinct types (Kiyomi 1992), or elsewhere in the grammar (see
Frawley 1992: 89 ff. and App. 1). Further discussion of the universality of animacy is given by
Frawley (1992: 89-91).
5
This distinction is similar to the opposition of 'permanent' and 'temporary' character-
istics in the classifier system of Tzeltal introduced by Berlin (1968).
 Semantics of Noun Categorization 273
dimensional (or flat), and three-dimensional (spherical). Dimensionality
and shape often occur together, but they can also be separate. Many
languages differentiate more shapes for one- and two-dimensional objects
(Hundius and Kolver 1983: 206; Denny 1979a). Three-dimensional objects
can also be classified in terms of their different shapes (e.g., round and
irregular shape). SHAPE is a cover term for other, form-related properties
(e.g. curved, linear, pointed, or blunt).
DIRECTION, or ORIENTATION, refers to a distinction between objects which
are vertically extended and those which are horizontally extended.
(b) INTERIORICITY refers to the way an entity 'differentiates its inside from
its outside', e.g. the distinction between rings and holes. BOUNDEDNESS is a
related parameter, and it indicates whether or not outlined entities have a
delimitation (for instance, flat objects can be bounded, e.g. tortillas, or
unbounded, e.g. plains).
(c) SIZE has two values, large and small.
(d) CONSISTENCY refers to the plasticity of the object under manipulation.
The two most frequent values are flexible and rigid. Other possible values
include viscosity of a liquid, or its surface tackiness (Frawley 1992: 128).
(e) CONSTITUTION, or STATE, refers to the physical state of an entity, such as
liquid or solid. It is often fused with CONSISTENCY, DIMENSIONALITY, and/or
SHAPE. For instance, many classifier systems have a single term for liquids
while classifiers for solid objects always include reference to whether the
object is long or round.
(f) MATERIAL out of which an object is made may be reflected by classifiers;
there may be classifiers for wooden and for metal objects.
(g) Other INHERENT NATURE or TIME-STABLE properties used in noun cat-
egorization have to do with the material and function of the items. Often,
there are special classifiers for plants, houses, canoes, verses of poetry,
books, and so on. Inherent nature properties are often realized through
SPECIFIC, or UNIQUE classifiers which combine with just one noun, e.g. the
Korean classifier hwan 1 for 'ball-shaped Chinese traditional medicinal pill'
(Lee 1997: 55). These classifiers are generally culture-specific (see §12.3).
(h) FUNCTION, or functional interaction6 classifiers refer to specific uses of
objects, or kinds of action which are typically performed on them.
Languages tend to encode culture-specific functional properties in classi-
fiers. Objects can be classified depending on whether they can be consumed
(eaten, drunk), or whether they can be planted, or domesticated; there are
often classifiers for means of transport, clothing, and housing. Actions

6
Functional interaction is, in a sense, parallel to social interaction in classifiers which refer
to social status (cf. Denny 1976).
274 Classifiers
performed on objects and encoded in classifiers may involve cutting, pier-
cing, harvesting, peeling, and so on (see Berlin 1968, for examples of highly
idiosyncratic function-based numeral classifiers in Tzeltal).
Objects can be classified by their VALUE, which is ultimately determined
by their functional properties. Thus, among relational classifiers there is
often a term for 'valuable' possessions.
(i) ARRANGEMENT refers to the configuration of objects, e.g. coil of rope, or
objects strung together. Arrangement very often correlates with consis-
tency, and material (e.g., Nivkh has a special classifier for fishes strung
on twigs: see Table 4.4).
(j) QUANTA is similar to ARRANGEMENT and refers to number, or quantity of
objects, e.g. cluster, set, flock, bunch. Quanta correlate with consistency
and material. For instance, a 'cluster' refers to a quantum which is also
irregularly shaped and dense (Frawley 1992: 128).
Some nouns in a language may lie outside the scope of existing classi-
fiers. A RESIDUE, or DEFAULT classifier can then be used with otherwise
'unclassifiable' nouns; the functions of a 'default' member of a classifier
system are discussed in §12.1.4.

11.1.2. Additional semantic characteristics

A number of other properties of objects are relevant for the semantic
organization of different kinds of classifiers. COUNTABILITY involves the
distinction between count and mass nouns (this property may also cor-
relate with the distinction of boundedness/unboundedness: Frawley 1992:
84-8). Different classifiers which refer to configurations and quanta of
objects can be used with a noun depending on whether it is countable or
not.7 The distinction between count and mass nouns also affects correla-
tions between classifiers and other categories, e.g. number.8 The mass/unit
distinction is important for numeral classifiers. In addition, one of the
functions of affixed noun classifiers in some South American languages
is to make mass, uncountable, or collective nouns into countable ones.
7
Countability, and how it is important for distinguishing between classifiers and quantifiers
in noun phrases with numerals, was discussed in Chapter 4. Classifiers used with countable
nouns which are 'natural units' have an individualizing function, while classifiers with uncoun-
table nouns are mainly used to 'create' units by measuring them (Croft 1994). Classifiers which
refer to quanta and configuration can also be used to create 'new' units out of count nouns,
e.g. one bunch of carrots. Depending on language-internal criteria, measure terms may or may
not be considered a subtype of classifiers. In many cases, there can be a continuum between
'true' classifiers and quantifiers.
8
The count/mass distinction is closely tied to grammatical number; however, the correlation
between mass and singularity, and count and plurality, is not straightforward (see Allan 1980;
Wierzbicka 1985, for an analysis of an iconic relationship between grammar of plural vs.
singular marking in mass nouns and the internal structure of referents; see also Stebbins 1997).
 Semantics of Noun Categorization 275
The division of nouns into ABSTRACT and CONCRETE is another property
relevant for the structure of classifier systems. Often, abstract nouns are
not classified; some may take specific classifiers.

11.1.3. Semantic relationship between a classifier and the referent

The semantic relationship between the referent of a noun and a classifier
can be of two kinds.
(i) A classifier is chosen according to the time-stable, or temporary proper-
ties of a referent noun.
(ii) A classifier is a superordinate term which indicates a larger class of
'prototypical' referents to which the noun belongs as a subordinate
member.
Then, classifier and noun are in a generic-specific relation. The differentia-
tion between the superordinate level of noun categorization (animal, plant,
vehicle, etc.) and the subordinate level is basic to this relationship. The
generic-specific relationship is basic to the organization of the lexicon.
Generic terms usually cover inherent nature properties ((a-g) above) and
functional properties, (h).

11.2. Semantics of classifier types

Our next step is to investigate the semantics of different noun categoriza-
tion devices, to see whether there are any significant correlations or
preferences between classifier types and the semantic parameters used.
The semantics of different noun categorization devices is considered in
§11.2.1-6. The semantics of different classifiers coexisting in one language
is dealt with in §11.2.7, and the semantics of multiple classifier systems is
discussed in §11.2.8.

11.2.1. Semantics of noun classes

Nouns are assigned to noun classes on a semantic basis, or on a combina-
tion of semantic and other—morphological or phonological—properties of
nouns. Because of the highly grammatical status of noun classes, their
assignment is often at least to some degree semantically opaque (see
§2.3.1), and every noun tends to belong to just one class (see §2.4.3 on
variability in noun class assignment).9
9
However, as we have shown in §2.3.4, even if noun class assignment is based on a
combination of different properties, there is always some ultimate semantic basis to it. The
276 Classifiers
The semantic basis of noun classes involves animate/inanimate, human/
non-human, person/non-person, and sex: see (A) below. Additional para-
meters for noun class assignment of inanimates may include (B) physical
and (C) other properties. Semantic complexity and opacity in noun class
assignment are discussed under (D), and semantic parameters in split noun
class systems are considered in (E).
(A) Animate/inanimate, human/non-human, person/non-person, and sex
in noun class systems
These parameters interrelate in different ways depending on the system.
Some Dravidian languages divide nouns into human and non-human (or
person/non-person). Siouan and Algonquian languages divide nouns into
animate and inanimate classes. In many Afroasiatic, Eastern Nilotic, and
Central Khoisan languages, nouns are divided into feminine and masculine
depending on their sex. 'Rationality' can be an extension of 'humanness';
nouns divide into rational (humans, gods, demons) and non-rational in
Tamil and other Dravidian languages. (The term 'neuter' is often used to
refer to 'irrational', inanimate gender, or as a residue gender with no clear
semantic basis.)
Languages can combine these parameters. Zande and Ma (Ubangi,
Niger-Congo) distinguish masculine, feminine, non-human animate, and
inanimate. Godoberi (Northeast Caucasian) has feminine, masculine, and
non-rational genders (for further examples, see §2.3.1; Corbett 1991;
Aikhenvald 1999a).
(B) Physical properties in noun class systems
Inanimate nouns can be assigned genders based on their shape, size, and
also position and consistency. Physical properties employed in noun class
assignment, summarized in Table 11.1,10 are iconic and nature-based:
women are usually smaller and rounder than men and are perceived to
sit more than men, so round shape and squat position tend to be associated
with feminine gender.

pervasive character of animacy-based distinctions is best illustrated with languages which have
phonological noun class assignment. Noun class agreement in Arapesh and Bukiyip, Papuan
languages from the East Sepik, is partly phonologically based. A more detailed analysis of the
way noun classes are assigned shows its semantic motivation. In Bukiyip, nouns divide into
eighteen classes. Four of these classes include animate nouns: males, females, non-human
animates and personal names; and all the rest include inanimates classified according to their
shape and other physical properties (see also §2.3, for an example of semantic and phonolo-
gical principles of noun class assignment in Yimas).
10
The semantics of gender in the languages of the Sepik region is discussed by Bruce
(1984), for Alamblak and by Aikhenvald (1998a) for Manambu; for Oromo, see Clamons
(1993); for Cushitic, East Nilotic, and Khoisan, see Heine (1982a); for Tiwi, see Osborne
(1974) and Lee (1987); see data on Cantabrian Spanish in Table 2.2.
 Semantics of Noun Categorization 211
TABLE 11.1. Examples of physical properties in noun class assignment
Parameter Masculine Feminine Example
Size Large (and wide) Small (and narrow) Some Afroasiatic
languages, e.g. Dasenech,
Oromo, Amharic,
Turkana, Camus
(Eastern Nilotic)
Narrow (and small) Wide (and large) Cantabrian Spanish,
Central Khoisan, small
in Tiwi (Australian)
Shape Straight Round Alamblak, Manambu,
Central Khoisan, Tiwi
Position Vertical Horizontal, squat Manambu, Cantabrian
Spanish
Solidity Solid Hollow, deep, concave Katcha (Kadugli-Kongo)

Further examples of shape-based gender assignment to inanimates are

found in Papuan languages of New Guinea. In Yonggom, an Ok language
spoken in the Western province (Christensen 1995: 9-10), feminine gender
assignment to inanimates is associated with large size, with masculine
gender assignment being linked with elongated shape; a similar system is
found in Olo (Torricelli phylum: Laycock 1975: 770; McGregor and
McGregor 1982: 55). In Wara (Fly river; Risto Sarsa, p.c.) long inanimate
objects are assigned masculine gender, while round objects and objects
consisting of multiple parts are feminine. In Abau (isolate; East Sepik
province) feminine gender assignment is associated with flat shape (Arjen
Lock, p.c.).
A cluster of physical properties is typically used for gender assignment.
Dimensionality is never the only physical property on which gender
assignment for inanimates is based; rather, it always correlates with other
properties, such as size or position.
If gender assignment to inanimates is SHAPE-based and SIZE is secondary,
then masculine gender is associated with narrow, vertical, longish items. In
contrast, feminine items get associated with the opposite: wide, horizontal,
round items (e.g., Tiwi, Cantabrian Spanish, Central Khoisan).
If gender assignment is SIZE-based, and SHAPE is secondary, masculine
items are typically associated with large size, and feminine items are con-
sidered small (e.g. Oromo and a few East Nilotic languages).
In some shape-based gender assignments to inanimates it is impossible
to decide whether size or shape is primary, since the choice of a semantic
parameter depends on the semantic group of the noun. The principles of
masculine and feminine gender assignment to animates and to inanimates
278 Classifiers
in Manambu (Aikhenvald 1998a) are shown in Diagram 11.1. The gender
assignment for humans depends on their sex. Gender assignment to ani-
mals is determined by their size, while gender choice for inanimates and
lower animates (e.g. mosquitoes or flies) is determined by their size and
shape. Gender choice for mass nouns (blood, dirt) is determined by their
quantity; for natural phenomena it is determined by their 'extent': for
instance, a very dark night or a very strong wind are masculine, while a
moonlit night or a light breeze are feminine.

Males and males by association: masculine

Human-
Females: feminine
Large: masculine (e.g. large mammals)
- Animal - - Size
Non-humant Small: feminine (e.g. small mammals)
- Inanimates - Size and Long and large: masculine (e.g. a stick)
Lower animates shape Round and small: feminine (e.g. a turtle)
Large: masculine (e.g. a big pool of blood)
Massnouns -Quantity
Small: feminine (e.g. a small pool of blood)
Complete: masculine (e.g. a very dark night)
'Natural phenomena Extent
Non-complete: feminine (e.g. a moonlit night)
DIAGRAM 11.1. Gender assignment in Manambu

This complicated interaction between the semantic group of a noun and

the way it is assigned a gender shows that the shape and size in gender
assignment to inanimates in this language requires a more complex seman-
tic explanation than just 'stereotypical semantic associations with each sex'
(cf. Croft 1994, in his debate with Kiyomi 1992).
Physical properties are only rarely employed to assign genders to ani-
mates. In Manambu (Ndu), a man can exceptionally be referred to with
feminine gender if used pejoratively to refer to a fat, round, or squat
person. In Palikur, if the sex of a person is unknown, gender is assigned
by size: an adult is masculine, and a child is feminine. Some non-human
animates (fish and birds) are assigned gender depending on their size.
(C) Additional oppositions
Noun class systems can use additional semantic features. Northeast
Caucasian languages with four classes often have a special class to include
animals (e.g. Lak or Andi: Corbett 1994b).
Function is rarely used in noun classes. A notable exception is Australian
languages where one class, that of non-flesh (or vegetable) foods, is based
on the function of the referent. A typical Australian noun class system has
 Semantics of Noun Categorization 279
four classes: masculine (I), feminine (II), vegetable (III), and neuter, or
'residue' (IV) (cf. Evans 1997; forthcoming).
There may be special genders for locative nouns, as in Bantu languages
(see Chapter 13), e.g. ChiBemba Class 17 uku- 'target of motion or adher-
ence', Class 18 umu- 'in ground, in water'. Motuna (Papuan, Bougainville)
has distinct noun classes for locational and for manner nouns (Onishi
1994).
Genders often have affective values. In Oromo and Berber languages
feminine gender is also used for diminutives, and often implies endearment.
This is especially striking in systems which allow variability in gender
assignment. In Amharic, close male friends and relatives may address
each other in feminine gender; feminine can also be used with reference
to males to express admiration, e.g. tesfaye gobez nec (Tesfaye smart/
clever/brave be.PF.3FEM.SG) Tesfaye (a man) is smart/clever/brave' (Men-
gistu Amberber, p.c.). In contrast, in Polish expressive masculine forms of
feminine names have endearing value. The masculine gender 'signals an
attitude of affectionate jocularity' (Wierzbicka 1996: 398). Secondary value
distinctions are often found of Bantu noun classes—see Table 11.3.11 In
Palikur, gender assignment to non-human animates depends on their value
and on speakers' attitude. Feminine gender is associated with positive
value, while masculine goes together with negative feelings. The rat is a
small animal; however, it is assigned masculine gender because it is looked
upon as dirty and bad. But a cute little baby rat would be referred to as
feminine. Along similar lines, turtles are usually feminine; but a turtle
which is a nuisance and has to be got rid of would be referred to as
masculine; all insects are masculine in spite of their small size, according
to an explanation by a native consultant, 'because none of them are any
good for food and all they do is bother people, eat crops and cause
sickness' (Aikhenvald and Green 1998).
In Lokono (Arawak: Baarle and Sabajo 1997; cf. §10.5) just male
humans belong to the masculine gender, and all the rest are feminine.
But there is much more to it: to be assigned masculine gender, a male
person must also belong to the Arawak community (all foreign men are
treated as feminine). Masculine gender assignment is associated with posi-
tive value, while feminine gender goes together with negative feelings.
Animals and birds which are thought of as having a 'positive personality'
are masculine—they include turtles and hummingbirds. Domestic animals
to which speakers have a special attachment, for instance, a dog, are
masculine; however, one's neighbour's dog is more likely to be feminine.
Attractive animals are masculine, while bigger animals are feminine: the
tapir is feminine just because of its big size.
11
And see Aksenova and Toporova (1994: 81) for data on Kuria.
280 Classifiers
Some semantic parameters appear never to be used in noun class sys-
tems, e.g. kin and social status for humans, or material and value for
inanimates.12 Colour is never used as a basis for noun categorization.13
(D) Semantic complexity and opacity in noun class assignment
It has been often stated that there is no real semantic basis for gender
assignment in some well-known Indo-European languages. In a seminal
study, Zubin and Kopcke (1986) provided a semantic rationale for gender
assignment of nouns from different semantic groups in German. In agree-
ment with the natural sex principle, masculine and feminine genders mark
the terms for male and female adults of each species of domestic and game
animals, and neuter is assigned to non-sex-specific generic and juvenile
terms. Masculine gender is used for types of cloth, of precipitation and
wind, and of minerals. Types of knowledge and discipline have feminine
gender, and games and types of metal—with the exception of alloys—have
neuter gender.
Almost all nouns with vague generic reference, such as das Ding 'thing',
das Gerat 'implement, apparatus', das Gut 'goods' have neuter gender (with
one exception: die Sache 'thing' is feminine). Superordinate terms usually
belong to neuter gender, and items of a more basic level are feminine or
masculine, more rarely neuter—see Table 11.2 (Zubin and Kopcke 1986:
147 ff.). There are a few exceptions. Two superordinate terms are feminine:
Pflanze 'plant, herb' and Farbe 'colour'.

TABLE 11.2. Semantic basis of gender choice in German: an illustration

Superordinate Basic level
Instrument (N) 'musical instrument' Guitarre (F) 'guitar', Trompete (F) 'trumpet'
Obst (N) 'fruit' Apfel (M) 'apple', Pflaume (F) 'plum'
Spielzeug (N) 'vehicle' Bauklotz (M) 'block', Puppe (F) 'doll'
Land (N) 'land' Wald (M) 'woods', Sumpf'(M) 'swamp',
Wiese (F) 'meadow'

Noun class assignment is usually more opaque for inanimates and for
non-human animates. In Mayali (Evans forthcoming: 104-6) masculine

12
These semantic features are found in correlation with others. Noun classes may contain
reference to hierarchically higher, or more powerful, humans: see §12.3.1, on noun classes in
Akan.
13
'Lustre', or 'visibility', is reported to be one of the classificatory parameters in assigning
noun classes in Anindilyakwa (Australian: Leeding 1989; 1996). The use of visibility or lustre
as a parameter for noun categorization could be associated with the cultural importance of
'lustre' and 'light' in Northern Australian Aboriginal culture (Mark Harvey, p.c.); however,
more study is needed to understand how these parameters are employed.
 Semantics of Noun Categorization 281
class includes male humans, the names of certain malevolent beings mostly
associated with the sky, items associated with painting (a male activity),
and also some mammals, some snakes, and some birds and fish. Feminine
class includes female humans, and also some reptiles, fish, and birds. Vege-
table class includes all terms for non-flesh foods, but also a few bird names.
Finally, the 'neuter', or 'residue' class is the most semantically hetero-
geneous—it includes items which do not 'fit' into other classes, e.g. most
body parts, generic terms for plants, and terms for various inanimate
objects.
In Dyirbal water, fire and items associated with fighting belong to Class
2. In contrast, in Gurr-goni and Mayali fire and related things are
'neuter'; 'water' belongs to the neuter class in Gurr-goni, and to vegetable
class in Mayali. In Gurr-goni (R. Green 1995) males and most animals,
fish, and birds belong to the masculine Class I; human females, turtles,
and a few other animals belong to feminine Class II; most living plants
and non-flesh foods belong to 'vegetable' Class HI, together with tradi-
tional canoes, corroborees, and songs. Terms which belong to Class IV
('neuter', or 'residue') cover the following semantic domains: the natural
environment; water, rain and other liquids; fire and associated things; a
few plants; a few body parts; time, structure, and buildings; and language
and ceremonies.
In Jingulu (Pensalfini 1997: 254) nouns divide into four classes, only
some of which are more or less semantically transparent. The vegetable
class includes mostly objects which are long, thin, or pointed; this class
happens to include most vegetables, as well as body parts such as colon,
penis, and neck, instruments such as spears, fire-drills and barbed wire,
natural phenomena such as lightning and rainbows, and also roads and
trenches. The feminine class includes female humans and higher animates,
and also words for axes, the sun, and most smaller songbirds. The semantic
content of the remaining two classes, masculine and neuter, is harder to
define: masculine is mostly used for the rest of animates and neuter for the
rest of inanimates, except that flat and/or rounded inanimates—such as
most trees and eggs, and body parts such as liver and brow—are masculine.
The choice of the four genders in Wambaya (Nordlinger 1998: 63–4), which
is closely related to Jingulu, is somewhat different: for instance, the 'vege-
table' (or 'non-flesh food') class is associated with round shape rather than
with being long or thin. There may be culture-specific explanations for
these varying classifications.
Noun classes in Bantu languages are another example of a semantically
opaque system. Table 11.3 summarizes 'a basic semantic grid common to
Bantu noun class systems' (Spitulnik 1989: 207) based on an interaction of
shape, size, and humanness. However, these parameters provide only a
partial semantic motivation for the Bantu noun classes.
282 Classifiers
TABLE 11.3. Noun classes in Bantu14
Class Semantics
1/2 Humans, a few other animates
3/4 Plants, plant parts, foods, non-paired body parts, miscellaneous
5/6 Fruits, paired body parts, miscellaneous inanimates
7/8 Miscellaneous inanimates
9/10 Animals, miscellaneous inanimates, a few humans
11/10 Long objects, abstract entities, miscellaneous inanimates
12/13 Small objects, birds
6 Masses
14 Abstract qualities, states, masses, collectives
15 Infinitives

Only the Class 1/2 'human' is semantically homogeneous. Shape and size
are also employed as semantic parameters of noun categorization. Classes
associated with shape in ChiBemba are shown in Table 11.4, and classes
associated with size in Table 11.5 (Spitulnik 1989: 210, 212).15 Diminutive
and augmentative classes carry affective overtones (diminutives are endear-
ing and augmentatives are pejorative).

TABLE 11.4. Shape-based classes in ChiBemba

Class Meaning
11 ulu- 'long', 'extending over space'
14 ubu- 'flat', 'surface'
9 iN- 'bounded space', 'orifice-edge', 'round'

TABLE 11.5. Size-based classes in ChiBemba

Class Meanings Associated affective notions
7/8 ici-lifi- 'large' 'gross', 'despised'
12/13 aka-lutu- 'small' 'charming', 'quick', 'cunning'
6 ama- 'mass', 'collective' 'exaggerated'

Denny (1979a: 109-10) suggested the following scheme for some Bantu
noun classes based on configurations involving extendedness (also cf.
Denny 1976 and Denny and Creider 1986).
14
This grid differs somewhat from the semantics reconstructed by Denny and Creider
(1986) and Creider (1975): see Table 2.1.
15
A test of the semantic reality of noun classes in Kikuyu conducted by Burton and Kirk
(1976) showed a correlation between the size of animals and birds and their noun class
assignment. Class 11 was associated with large size, and class 9 with small size.
 Semantics of Noun Categorization 283
Configuration

Solid Outline

Extended Non-extended Extended Non-extended

CL3/4 CL5/6 CL11/10 CL9/10
DIAGRAM 11.2. Extendedness in Proto-Bantu noun classes

Class 9/10, 'non-extended outline', includes rings, holes, and also con-
tainers (e.g. clay pots and gourd bottles), together with houses and geo-
graphical spaces. Class 11/10 involves an outline figure of a different sort:
an extended curve with an interior, e.g. hill, rib, palm of the hand. Ropes
and cords are also placed into this class due to the fact that the parts joined
together by ropes and cords 'must be inside the curve of the cord' (Denny
1979a: 110). The use of the parameters 'extendedness' and 'outline' resem-
bles the way shape is used in numeral classifiers (cf. Table 9.5, for Baniwa).
In the Bantu noun class systems, the only semantically transparent and
straightforward principle of assignment is animacy and/or sex, unlike
Baniwa, where the assignment of almost all the classes is transparent.
(E) Semantic parameters in split noun class systems
In languages with split noun class systems, pronominal noun classes tend
to be based on animacy and sex distinctions, while physical (extendedness,
shape, and size) parameters are found in the non-pronominal classes: see
§2.7.2, for Paumari, and Table 9.5, for Baniwa. Markers of non-pronom-
inal noun classes are often used in other classifier environments: see
Chapter 9.

11.2.2. Semantics of noun classifiers

Noun classifiers as independent lexemes are 'superordinate' terms in a
generic-specific relation with a noun (see §11.1.3). The semantics of generic
noun classifiers relates to the inherent nature of the object and to its
functions. There is usually a classifier for humans, or animates,16 one or
more classifier(s) for non-human animates, and a number of classifiers for
inanimates based on their inherent properties or functions.
The content of HUMAN classification differs from system to system.
Western Austronesian languages, such as Minangkabau, Acehnese, or
Indonesian, have just one general term for person, e.g. Minangkabau urang
padusi (NOUN.CL:PERSON woman) 'a woman'.
16
The Daly languages of Northern Australia have a human/non-human opposition in noun
classes, but not in noun classifiers (Reid 1990). There may be a historical explanation to this:
noun class markers in these languages go back to grammaticalized noun classifiers (see §13.4).
284 Classifiers
Human nouns can be further divided according to SEX. Yidiny (Australian)
has a generic classifier bama 'person', and also sex-related classifiers waguja
'man' and bunya 'woman'.
In Mayan languages of the Kanjobalan branch humans are classified
according to their SOCIAL STATUS, KINSHIP RELATION, or AGE. One of the most
complicated systems is found in Jacaltec. Humans are classified in terms of
sex, respect, kin versus non-kin, and age. There are also two classifiers for
deities: male and female—see Table 11.6 (Craig 1986c: 266–7).

TABLE 11.6. Noun classifiers for humans and

deities in Jacaltec
cuman Male deity
cumi7 Female deity
ya7 Respected human
naj Male non-kin
ix Female non-kin
naj ni7an Young male non-kin
ix ni7 an Young female non-kin
ho7 Male kin
xo7 Young male kin
ho7 nil an Young male kin
xo7 ni7 an Young female kin
unin Infant

Noun classifiers for humans in Akatek distinguish 'known human',

'human, appreciative', and 'man' and 'woman' (see Table 8.4; Zavala
1992: 152). Mam (Table 13.3) has classifiers for men and women; for young
and old men and women; for old men and women to whom respect is due,
and for someone of the same status as the speaker. There is also a classifier
for babies, and just one non-human classifier.
In Australian languages, noun classifiers which refer to social status
include such distinctions as 'initiated man' (§3.2.1). Murrinhpatha (Walsh
1997: 256) has a classifier for Aboriginal people (which also covers human
spirits) and another one for non-Aboriginal people, which also covers all
other animates and their products.17
Nouns with non-human, or inanimate, referents are classified in terms
of inherent nature-based properties ((g) in §11.1.1). They belong to
natural domains of human interaction: animals, birds, fish, plants, water,
fire, minerals, and artefacts. Individual systems may vary. There is often

17
A system of honorific classifiers, as documented by DeLancey (1998: 121) for Tibetan,
can be considered a subtype of functional categorization of nouns based on humans and the
world of their experience.
 Semantics of Noun Categorization 285
a general term for birds and fish, as in Minangkabau (Table 13.4);
Ngan'gityemerri and Akatek have a noun classifier for animals (see Tables
8.1 and 8.4). Classifiers in Murrinhpatha cover fresh water and associated
concepts; flowers and fruits of plants; spears; offensive weapons; fire and
things associated with fire; time and space; speech and language; and there
is a residue classifier (see §12.1.4).
There is usually a noun classifier for culturally important concepts.
Mayan languages have a noun classifier for corn which is important for
traditional agriculture, and for domesticated dogs, while Daly languages
have classifiers for spears, diggings sticks, and woomeras. Table 11.7 shows
the set of classifiers for non-human nouns in Jacaltec (Craig 1986c: 267).

TABLE 11.7. Noun classifiers for non-humans

in Jacaltec
no7 Animal
metx' Dog
te7 Plant
ixim Corn
tx'al Thread
tx'an Twine
k'ap Cloth
tx'otx' Soil/dirt
ch'en Rock
atz'am Salt
ha7 Water
k'a7 Fire

Noun classifiers can distinguish nouns in terms of their functions, e.g.

'habitable place', or 'drinkable liquid' (cf. Dixon 1977, on Yidiny). A
distinction between flesh and non-flesh food is typical for Australian lan-
guages with noun classifiers, cf. Yidiny minja, Emmi awa 'flesh food',
Yidiny mayi, Emmi miya 'non-flesh food' (Dixon 1977: 483; Ford 1998).
Functional distinctions may include 'movable', e.g. Yidiny wirra.ls
Physical properties of referents, arrangement, and quanta ((a–f)> (h–i) in
§11.1.1) are not used for classifying nouns with noun classifiers which are in
a generic-specific relation with the accompanying noun.19

18
Inherent nature- and function-based classifiers are relatively independent systems. We
mentioned in §3.2.1 that function classifiers can cooccur with inherent nature classifiers in
Yidiny.
19
It can be argued, however, that constitution or state, such as liquid, is a concomitant
semantic feature used with function-based noun classifiers. However, constitution is never used
by itself as a parameter.
286 Classifiers
11.2.3. Semantics of numeral classifiers
Numeral classifiers divide into SORTAL and MENSURAL types. Sortal numeral
classifiers describe inherent properties of referents, and mensural classifiers
describe the ways they can be measured (see §4.4). Consequently, sortal
classifiers tend to use inherent properties (a-g) more, while mensural
classifiers prefer temporary ones (h and i).
The choice of SORTAL classifiers is often based on animacy, humanness,
or, more rarely, sex (A). Further classification of humans is often based on
social status or age (B). Additional properties employed in numeral classi-
fiers are (C) physical properties, (D) functional properties, and (E) arrange-
ment and quanta. Mensural classifiers (F) combine these.
(A) Animacy, humanness, and sex in numeral classifiers
Numeral classifiers often provide a two-way division of nouns into HUMAN
and NON-HUMAN, as in numerous small classifier systems in Dravidian
languages, e.g. Telugu and Kannada (Emeneau 1964: 649), and in Ainu.
Alternatively, nouns can be divided into ANIMATE and INANIMATE. This is
the case throughout languages of the Tai family (see Conklin 1981: 130-6;
DeLancey 1986: 447). There may be a separate class for humans, and a
number of classes for non-humans, as in Toba Batak, Makassar, Bugis,
Mori, and Gorontalo (Austroasiatic: Conklin 1981: 241). Acehnese (Durie
1985: 139) has a special classifier for humans (droe); all the other classifiers
are confined to non-hurnans (which include animals).
A three-way classification may divide nouns into HUMANS, NON-HUMAN
ANIMATES, and INANIMATES. Jacaltec (Kanjobalan Mayan) and Squamish
have three numeral classifiers: human, animal, inanimate (Craig 1986b:
265; Table 4.6 above). There may be a special classifier for humans, one
for animals, and a number for inanimates,20 as in Malay, Indonesian,
Minangkabau, and Balinese (Conklin 1981: 240 ff.; Marnita 1996), and
in the Dravidian language Malto (Mahapatra 1979: 126), and in Nivkh
(Table 4.4).
Numeral classifiers can involve reference to ANIMACY and SEX.21 There are
several possibilities. There may be a male animate, a female animate, and

20
Animals can be categorized in different ways. Malay and Minangkabau have just one
classifier for animals. In Achagua (Arawak: Wilson 1992: 62) the animal classifier is applied
only to mammals. Bahwana (Arawak: Ramirez 1992) has several classifiers for different
species of mammals and lizards. Different classificatory techniques can be used for animates
and for inanimates. Movima (Bolivian isolate: Colette Grinevald, p.c.) uses repeaters as
numeral classifiers to classify inanimates; there are special classifiers just for humans and
for animates.
21
The existence of sex- and humanness/animacy-based systems in numeral classifiers goes
against predictions formulated by Adams and Conklin (1973) on the basis of Austroasiatic
languages, and repeated by Croft (1994).
 Semantics of Noun Categorization 287
an inanimate classifier, as in Bora (Bora-Witoto: Thiesen 1996: 43), East
Tucano, and Warekena, Baniwa, and Tariana.
There are hardly any examples of numeral classification based just on
sex. However, in Kolami (Dravidian) and Marathi (Indo-Aryan) numeral
classifiers are used only with animates, and they distinguish masculine and
feminine (Emeneau 1964: 648).
Animate/inanimate or human/non-human are distinguished in most sys-
tems of numeral classifiers (cf. Croft 1994). There are three groups of
exceptions.
(i) Languages with two sets of numeral classifiers may have animacy dis-
tinctions in one, but not in the other. Akatek (Zavala 1992: 130-1, 140–1)
distinguishes humans, animals and inanimates in the suffixed set of
numeral classifiers. 'Independent' numeral classifiers are assigned accord-
ing to shape, position and size (see Table 8.3 and §4.3.2; for Jacaltec, see
Craig 1986b: 265).
(ii) Animacy-based oppositions may be absent from numeral classifier
systems if they are found elsewhere in the language. Sochiapan Chinantec
distinguishes animate and inanimate forms of numerals and some modi-
fiers (Foris forthcoming: 305, Table 6.14 therein), but there are no animacy-
based distinctions in numeral classifiers (Foris forthcoming: 317 ff.). A
similar system is found in the closely related language, Comaltepec Chi-
nantec (Anderson 1989: 57).
(iii) Some recently developed systems of numeral classifiers do not have
any special classifier for animates, or humans. Classification is achieved
according to the physical and functional properties of entities. Kana
(Kegboid, Cross River: Ikoro 1996a: 90-1) has nineteen numeral classifiers
the assignment of which is based predominantly on the shape of a referent.
Most animate nouns are classified with the default classifier ka; some do
not take any classifier (e.g. nee'person'). The word bee 'child' is classified
with bee 'seed-like small objects'. Unlike Akatek and Chinantec, there are
no animacy-based distinctions.22 Wantoat and Award, two closely related
languages from the Morobe province of Papua New Guinea, also have no
special classifier for humans, or for animates (Davis n.d.; Susan Quigley,
p.c.). Classifiers are used with numbers one and two, with demonstratives
and as derivational markers with nouns. Orientation appears to be one of
22
Nouns with human and with non-human referents differ in some ways (see E in Appendix
1). For instance, only proper nouns with a human referent can be pronominalized; in a topic-
comment construction a moved O can be pronominalized only if human; emphatic personal
pronouns usually refer to humans (Ikoro 1996b). Similarly, no animacy distinctions have been
observed in the numeral classifier system in Ngyembccn, a Grasslands language with numeral
classifiers (Viktor Vinogradov, p.c). This is by no means the case in all African languages with
numeral classifiers. Manessy (1961: 158) reports human/non-human and animate/inanimate
distinctions for some Bwamu dialects.
288 Classifiers
the important semantic parameters used in classifier assignment; humans
are assigned the same classifier as 'upright' objects (e.g. tree, pole) (Davis
n.d.: 4).
(B) Further classification of humans
If a language has more than one human classifier,23 there is further classi-
fication of humans according to their social function and status. In
Austroasiatic, some Tibeto-Burman languages, and Korean, humans are
classified by social rank or according to kinship; the choice is culturally
determined. Other variables may include age: children are often classified
as 'underhumans'. Age may be associated with social status in a more
complicated way, e.g. Dioi uses paou1 for respected males and mai5 for
little-respected groups of younger females (Conklin 1981: 132).
Lisu has different classifiers for female kin one generation away, male kin
one generation away, all lateral kin, and all kin two generations away
(Adams and Conklin 1973: 4). Pre-revolutionary Khmer had separate
classifiers for clergy and monks, high persons, dignitaries, and superior
and inferior honourable people (Adams 1989: 63; 1992).
(C) Physical properties
Numeral classifiers used with inanimates, or non-humans, employ many
more values for physical properties than do noun classes (see (B) in
§11.2.1).
SHAPE and DIMENSIONALITY are widely used in numeral classifier systems.
Languages differ in how many dimensions they use and how differentiated
the shapes are in each dimension/Languages tend to encode one and two
dimensions more often than three dimensions (Frawley 1992: 123). In Thai,
flat shapes are less differentiated than long ones and three-dimensional
ones hardly at all (Hundius and Kolver 1983: 206). In Japanese (Downing
1984; 1986) the three-dimensional classifier ko 'roundish small objects' is
less frequent than classifiers for flat, and for long objects (Frawley 1992:
123, 134).
Semantic parameters of EXTENDEDNESS, INTERIORICITY, and BOUNDEDNESS
((a) and (b) in §11.1.1) usually overlap in numeral classifier systems. The
interaction between boundedness and dimensionality in Totonac (Totonac-
Tepehuan: Levy 1994) is shown in Diagram 11.3.

23
Many languages have just one human numeral classifier, e.g. Achagua (Arawak), Tzeltal
and Tzotzil (Mayan), Indonesian, and Japanese (Kiyomi 1992: 20). Japanese has just one
classifier for humans, and several classifiers for non-humans. The latter is based on divisions in
species, e.g. hiki 'animals, insects, fish', too 'large animals (horses, cows)', wa 'birds', bi 'fish',
and hai 'squids'.
 Semantics of Noun Categorization 289
Dimensionality

One-dimensional Two-dimensional
--1
Bounded Unbounded Bounded- Unbounded
qi:- qan- mak-
Long, bounded Long, unbounded Flat, bounded laka-
e.g. banana e.g. road e.g. tortilla Flat surface Flat unbounded
e.g. field e.g. material, cloth
DIAGRAM 11.3. Numeral classifiers in Totonac

DIRECTIONALITY, or ORIENTATION, goes together with dimensionality and

shape in numeral classifiers. Unlike verbal and deictic classifiers (§11.2.5,
§11.2.6; cf. Croft 1994; Frawley 1992: 124-5), it never appears by itself, but
often gets expressed with the same item as dimensionality. Flat objects are
often horizontally spread (see also Palikur, in Diagram 11.4), and long
objects tend to be vertical (e.g. batang 'long vertical objects' in Minangka-
bau: Diagram 8.1).
INTERIORICITY (Denny 1979a: 108) refers to the distinction between rings
and holes. While rings 'focus' on the outside outline, holes 'focus' on the
interior of the outline. Tzeltal (Mayan: Berlin 1968: 123; Denny 1979a: 111)
has four 'hole' classifiers which differ with respect to complete or non-
complete perforation, and with respect to size. The parameter of interioricity
in Tzeltal correlates with material makeup, or consistency; the classifiers
listed in Table 11.8 are applicable to non-flexible objects alone.

TABLE 11.8. Hole classifiers in Tzeltal

Complete perforation Incomplete perforation

Large horn, e.g. complete large perforation putt, e.g. incomplete large perforation
in a board in a board
Small huht, e.g. complete small perforation cub', e.g. complete small perforation
in a board in a board

SIZE is rarely employed as an independent parameter; it usually clusters

with dimensionality and shape, as in Minangkabau alai 'flat, flexible, sheet-
like', bidang 'extended surfaces, e.g. field, farm, garden', petak 'square
objects', or with interioricity, as in Minangkabau bantuak 'circular objects,
e.g. rings and fishhooks' (Marnita 1996: 107).
All languages where numeral classifiers categorize inanimates with
respect to their physical properties employ dimensionality and shape. In
contrast, CONSISTENCY (flexible, rigid) and 'CONSTITUTION (liquid, solid) are
physical properties never encoded by themselves in numeral classifiers.
290 Classifiers
They usually correlate with dimensionality. For instance, flat objects are often
classified as thin and flexible (for example, the classifier on in Acehnese: Durie
1985: 139). These properties may correlate with size, or with form and
arrangement. Sochiapan Chinantec has a special classifier hmaih32 for tall
and flexible referents, such as maize plants and vines; and maih32 for fine and
long objects, e.g. wire, thread, vine (Foris forthcoming: 322, Table 6.21). The
classifier zianh13 is used for a bundle of rigid stick-like objects (Foris forth-
coming: 322, Table 6.20). The unifying rationale for the frequent combina-
tion of these parameters could lie in their 'handling ability'. In other numeral
classifier languages, such properties as rigid, flexible, or liquid are not used at
all, e.g. Akatek Mayan (Table 8.3), or Achagua (Wilson 1992).
Numeral classifiers for inanimates can be based on a complex combination
of DIMENSIONALITY, SHAPE, BOUNDEDNESs, and CONSISTENCY. Diagram 11.4
shows how these parameters interact in the system of Palikur sortal numeral
classifiers for inanimates. Inanimates are classified according to their shape,
dimensionality, and boundedness; there is also one specific classifier, for
plants. Consistency goes together with shape and dimensionality.
Shape Linear - Straight, curved,flexible:river,
(+consistency) rope (-tral-tahr-)
Equal dimensions (2/3)- -Round or square: orange,
box (circle) (-ul-so-)
Irregular dimensions (2/3) - - Other shapes (oval, rectangular) and
irregular: egg, basket, house, land (-al-sa-)
Concave -
Three-dimensional - - Bowl, canoe, ring (-mkul-muk-)
Dimensionality Two-dimensional - - Flat, nonextended: hammock,
mat (-kl-kal-bu)
One-dimensional -Rigid, vertical: arrow, cigar (-ll-ta-)
(-(-consistency)
Boundedness - - Extended, with boundaries Perimeter, height: fire, field, hole,
waterfall (-ikul-rik-)
Specific - - 'Nature'- - Plants (-ktil-kat-)
DIAGRAM 11.4. Numeral classifiers for inanimates in Palikur

Numeral classifier systems often contain a number of semantically highly

specific classifiers based on the MATERIAL and other inherent properties of
objects. Tai languages have specific classifiers for wooden objects (Conklin
1981: 144). Nivkh has classifiers for sledges and fish-nets (Table 4.4), while
Baniwa has a special classifier for excrement (Table 9.5), because of the
importance of identifying animal droppings when hunting game. Type of
material may correlate with other physical properties, most frequently with
 Semantics of Noun Categorization 291
shape and dimensionality. Thus, in Minangkabau, batang 'long vertical
objects' is most often used to categorize wooden things.24
(D) Functional properties
Functional properties are not as frequent in numeral classifier systems as
shape and dimensionality. A frequently cited example of a function-based
classifier in Thai is khan, originally used to refer to things with long handles
which now includes cars, bicycle, motorcycles, buses, and other vehicles (see
§12.3.4). In Hmong (Walter Bisang, p.c.) the classifier rab is used for tools
and instruments in general, and especially for objects with a handle. In
Austroasiatic languages a large number of classifiers have to do with written
and oral speech; classifiers used for counting written materials can be
organized according to the type of material (books, newspapers, or maga-
zines), or varying literary forms (play versus poem); function as a semantic
parameter of noun categorization may interact with shape—units of speech
may also be classified as long or flat (Adams and Conklin 1973: 7-8).
Table 11.9 presents a set of function-based numeral classifiers from
Burmese (Burling 1965; Denny 1976: 127-8).

TABLE 11.9. Function-based classifiers in Burmese25

Classifier Gloss
hte 'clothing for the body (not headgear or footwear)'
sin 'cutting tools'
si 'vehicles'
saun 'written materials (literary pieces, documents)'
le' 'hand implements' (also eyeglasses)
koun 'loop-shaped objects that are worn: garlands, necklaces'
hsaun 'houses, monasteries, royal buildings'

Function-based classifiers are often highly culture-specific. Gilbertese, a

Micronesian language (Silverman 1962), has a classifier kai for means of
subsistence, wa for transport (especially canoes), and kora for baskets.

24
Large systems of numeral classifiers can have even more complicated interactions of
shape, dimensionality, and consistency. Tzeltal has two classifiers for non-flexible objects
which correlate with other parameters. One, lehc, is unspecified for dimension and is used
for thin non-flexible objects of variable shape, e.g. basket, hat, gourd; the other one is used for
two-dimensional objects: pehc for 'thick, non-flexible flat objects where width is greater than
thickness', e.g. bricks, tortillas (Berlin 1968: 99). Classifiers for flexible objects correlate with
the way objects are handled, e.g. lihklhil for 'slender, flexible object in natural extended state',
t'im for slender, flexible objects in stretched position, like a clothes line; and tim for slender,
flexible objects stretched between two points but not taut like a clothes line (Berlin 1968: 109).
25
Transcription changed in agreement with the standard one used for transcribing
Burmese, supplied by Randy LaPolla.
292 Classifiers
Function-based classes in Western Austronesian and Austroasiatic lan-
guages include weapons, tools, and implements, and also various types of
fields relevant for agricultural functions (Adams and Conklin 1973: 8). The
degree to which a language uses functional interaction as a semantic basis
'shows how selective classifiers are in underlining a few key concepts to do
with cultural ecology' (Denny 1976: 128); see §12.3.
The way objects can be used often depends on their physical properties;
this is why functional and physical properties are often expressed with one
classifier. For instance, Burmese koun describes loop-shaped objects which
can be worn (Table 11.9).
The interaction between physical properties (shape, or dimensionality;
consistency; and material make-up) and function in numeral classifiers for
inanimate referents in Minangkabau is exemplified in Diagram 11.5 (adapted
from Marnita 1996: 104). It is impossible to decide whether shape and
dimensionality and material or function are primary in this interaction. In
the overall structure of numeral classifier systems function can be considered
as a secondary semantic feature simply because functional characteristics are
not obligatorily present in numeral classifier systems, unlike physical proper-
ties (though function can be a primary parameter for some classifiers in
some languages, e.g. rab 'tools and instruments' in Hmong). No numeral
classifier systems are based just on functional properties, and there are some
which do not use them at all (e.g. Palikur, Achagua, Chinantec).

Shape and dimension Material make-up

buah 'small round objects, general' uleh 'orange pip'; kalupak 'sheath'
alai 'fiat flexible'; batang 'stick-like vertical' butia 'bulky'; kalupah 'clumps of plants'
seed-like, seed' putiak 'flower bud'
bidang 'extended surfaces' incek 'seed-like, fruit with seeds'
guluang 'rolled objects' cariak 'piece of object which can be torn'
lempeang 'flat pieces of gold, silver' tandan 'banana, big bunch'
uteh 'piece of rope' tangkai 'flowers, leaves with stalk'
sikek 'banana, form of comb' bungkah 'clump of dirt, clay, cooked rice'
petak 'square objects', untai 'chain-like'
bantuak 'circular objects: rings, hooks'
bilah 'knives, cutlery' gagang 'tools with handle,
pucuak letters, long weapons' cloves with stalk'
lareh 'long bamboo weapons and guns' kaki 'umbrellas, long-stem flowers'
bangkawan 'roof kali 'events'
pintu 'shops' lubang 'doors and windows of houses'
palah 'word' piriang 'rice field'
tumpak 'farms, lands, rice fields'
Function

DIAGRAM 11.5. Interaction of semantic domains in numeral classifiers in Minangkabau

 Semantics of Noun Categorization 293
There may be further distinctions made in large classifier systems. Japa-
nese has a few specific classifiers used with abstract nouns ken 'incident',
toori 'method', han 'crime', denwa 'phone call'. (These classifiers go back to
repeaters.) Abstract nouns can also be used with the 'default' classifier tsu,
or without any classifier (Downing 1996: 73); see §12.1.4.
(E) Arrangement and quanta
Arrangement and quanta are frequently used in sortal and in mensural
classifiers. They typically interrelate with physical properties of objects and
their material makeup. Korean has special sortal classifiers for counting
pairs of chopsticks, and pairs of clothing (gloves, shoes). The ways in which
inherent properties correlate with arrangement are often culture specific
and unpredictable. Minangkabau has different mensural classifiers for
bunches of vegetable and fruit, parts of sugar cane, for pieces of meat on
a skewer, or for small objects (fish, fruit) on a skewer.
(F) Semantic properties of mensural classifiers
Mensural classifiers operate with animacy distinctions to a lesser extent
than sortal classifiers. In some languages, e.g. Chinantepec Chinantec,
mensural classifiers are used with inanimate referents only. In other lan-
guages, e.g. Sochiapan Chinantec, Akatek or Tzotzil, there are a few
mensural classifiers for animate referents. Tzeltal tends to use 'temporary
state' mensural classifiers only with inanimate referents. Physical properties
used in mensural classifiers are dimensionality, shape, and consistency, and
other nature- or function-related properties (e.g. a number of special men-
sural classifiers for tobacco in Minangkabau).

11.2.4. Semantics of classifiers in possessive constructions

In this section we will discuss the semantics of relational and possessed
classifiers. Possessor classifiers which refer to the properties of the posses-
sor are very rare, so that no generalizations can be made.26
(A) Semantics of relational classifiers
RELATIONAL classifiers characterize the relationship between the possessor
and the possessed depending on how the possessed object can be handled,
or used. Therefore, categorization of the possessive relationship is based on
functional interaction.
(i) The primary semantic division of nouns is often into CONSUMABLE and
NON-CONSUMABLE, as in Boumaa Fijian (Dixon 1988: 136), or general and
alimentary, as in Manam or Kaliai-Kove (Lichtenberk 1983a: 151-2).
26
Possessor classifiers discussed in Chapter 5 are based on the animate/inanimate distinction.
294 Classifiers
CONSUMABLE objects can be further classified according to the ways in
which they can be consumed (i.e. eaten or chewed, or drunk, sucked, or
licked), or how they can be prepared. For instance, Kosraean (Micronesian:
Lee 1975; Lichtenberk 1983a: 154–5) has classifiers for chewable, drinkable,
and uncooked food. In Kipea, an extinct language from Northeastern
Brazil (Rodrigues 1995, 1997), foodstuff was classified depending on
whether they were gathered or cultivated, or how they were prepared (by
boiling, or roasting).
NON-CONSUMABLE objects can be classified depending on how they were
acquired (i.e. found, or received as a gift, as in Kipea), or transported (as in
Kosraean).
(ii) A number of other, culture-specific, functional distinctions can be
mapped onto relational classifiers, e.g. clothing, vehicles, dwelling-places,
as in Kilivila, or Marshallese.
Tamambo, an Austronesian language from Vanuatu (Jauncey 1997:
§7.4.2), has four relational classifiers: no- 'personal property, general
term for belongings', ha- 'edible items', ma- 'drinkable items', bula- 'plants
and animals one owns' (with the exception of pigs which are the main
commodity, almost equal to money, and which take no-).
VALUE is a semantic parameter associated with function and widely used
in relational classifiers. Oceanic languages from Vanuatu have special clas-
sifiers for valued possessions, e.g. 11.1, from Raga (Lichtenberk 1983a:
154).
11.1. qoe pila-ma
pig CL:VALUED.POSSESSION-your
'your (sg) (valued) pig'
(iii) Human classification found in some systems of relational classifiers
can be described as relating to social function, i.e. the kinship relationship.
In Ponapean, six of the twenty-one relational classifiers refer to kinship. A
noun can be 'reclassified' depending on the choice of a relational classifier,
e.g. pwutak 'boy': in nah pwutak 'CL:DOMINANCE.OF.THE.POSSESSOR boy' 'his
son', rie pwutak 'CL:SIBLING boy' 'his brother', kiseh pwutak 'CURELATIVE
boy' 'his boy relative', wahwah pwutak 'CL:NEPHEW boy' 'his nephew', sawi
pwutak 'CL:CLAN.MEMBER boy' 'his boy clansmember', pelie pwutak 'CL:PEER
boy' 'his peer-group boy' (Rehg 1981: 183).
Relational classifiers can also reflect the closeness of the relationship
between possessor and possessed.27 In Baniwa (Arawak), two relational
classifiers correlate with the degree of closeness of the possessed to the

27
On similarities between classification in possessive constructions, alienable and inalien-
able distinctions, and other types of noun classifications, see Chappell and McGregor (1989).
 Semantics of Noun Categorization 295
possessor: 'my dog which I found' as opposed to 'my dog which I raised'
(examples 5.51 and 5.52). In Kilivila (Senft 1986: 49-54) nouns are divided
into edible, consumable, and closely or more distantly associated with the
possessor.
(B) Semantics of possessed classifiers
Unlike relational classifiers, POSSESSED classifiers characterize the possessed
noun in terms of its (a) animacy and (b) physical properties: shape, size, and
consistency. Possessed classifiers can also be in a generic-specific relation
with the noun they classify (see 5.7, from Apalai); similarly to relational
classifiers, they can categorize the noun in terms of its function.
Generic possessed classifiers describe the possessed noun in terms of
animacy and/or humanness. In this, they differ from relational classifiers.
Cora (Uto-Aztecan: Casad 1982: 236) has classifiers for human, animate
and inanimate (see other examples in Chapter 5). Generic possessed clas-
sifiers are often function-based. Luiseno has a classifier 'aac for pets.
Chemehuevi (Southern Numic branch of Uto-Aztecan) has two possessed
classifiers: -puqku 'pet' and -tgapi 'domesticated plant' (Press 1979: 60-1).
In South American languages function-based possessed classifiers usually
include terms for food and pets.
Large systems of possessed classifiers can combine several semantic
features. Classifiers in Apalai (Carib: Koehn 1994) categorize the pos-
sessed noun in terms of its function (field produce, drink, killed game);
there are a number of other specific classifiers which refer to inherent
properties of culturally significant objects (e.g. nut, corn, seed, firewood,
manioc cake).
Classifiers which combine the properties of relational and possessed
classifiers are based on inherent properties of a noun and its function—
for example, relating to the way it can be handled (see Table 5.4 for a
sample of classifiers in Puluwat).

11.2.5. Semantics of verbal classifiers

Verbal classifiers show several distinct tendencies in their semantics
depending on the type of relationship between a noun and its classifier.
(A) There can be a generic-specific relationship between the verbal classi-
fier and the nominal argument, or (B) the classifier can be chosen accord-
ing to the properties of the referent.
(A) Generic-specific relation in verbal classifiers
Generic verbal classifiers come from incorporated generic nouns. They
have the same semantics as generic noun classifiers (§11.2.2). There is
usually a classifier for animates, or humans, e.g. walga 'body', a classifier
296 Classifiers
for humans in Ngandi (Australian: Heath 1978b), or guk 'body' as a
classifier for animates in Mayali (Evans forthcoming: 269–71). Some
languages have a classifier which refers to an aspect of human classification,
e.g. Mayali yaw 'CL:BABY' (Evans 1996: 77). Nouns can be classified in terms
of their inherent properties, such as function or constitution, e.g. Mayali bo
'CL:LIQUID', or more specific properties. Mayali has generic classifiers for
fruit, meat, or pelt. A generic 'tree' is used to classify species of trees, e.g.
11.2 (= 6.2).
11.2. ga-rrulk-di an-dubang
3NP-vcL:TREE-stand(Np) NCLIII(vegetable)-ironwood.tree
'An ironwood tree is there.'
(B) Properties of verbal classifiers and classificatory verbs
Verbal classifiers and classificatory verbs chosen according to the proper-
ties of the noun may operate in terms of animacy, physical properties, and
sometimes function and arrangement.28
(B1) Semantic restrictions on verbal classifiers and classificatory verbs In
most languages verbal classifiers can be used with all or almost all verbs.
Not all verbs, however, are classificatory.
In North American Indian languages—notably Athabaskan and
Iroquoian, and also some languages of Central and South America (e.g.
Ika, from Colombia: Frank 1990)—classificatory verbs have to do with
handling physical objects ('put', 'carry', 'drop', etc.) or with their position
and location, or existence ('lie', 'hang', 'fall', etc.). According to Carter
(1976: 24), suppletive classificatory verbs in Athabaskan languages refer to
concrete objects, and they describe 'objects at rest, in motion, being
handled, being dropped, or falling' (Carter 1976: 24). In Cora (Casad
1996: 246) classificatory verbs comprise a group of stems whose basic
meaning is 'carry'. In Papuan languages classificatory verbs are mostly
existentials; they may also refer to location, posture, sometimes motion
(e.g. 'carry', as in Enga).29
In a few languages, the use of verbal classifiers follows similar restrictions.
In Palikur, only stative verbs which refer to dimension, physical propensity
and colour take classifiers. Among transitive verbs, classifiers are used only
with those which imply direct physical contact with the object. Verbal
classifiers are used to focus on the shape of an object completely involved

28
Possibly, different semantic parameters are used in verbal classifiers in sign languages,
e.g. a classifier for 'shiny substance' in American sign language (Suppalla 1986: 213). This is a
topic for a separate study.
See Goddard (1996) on difficulties of distinguishing existential from locational meanings
of Enga classificatory verbs, since locative verbs tend to grammaticalize into existentials.
 Semantics of Noun Categorization 297
in an activity; apparently properties related to form and dimensionality are
important only when the direct physical contact is involved.
(B2) Animacy in verbal classifiers and classificatory verbs Animacy dis-
tinctions are only sometimes present in classificatory verbs; they are in the
Athabaskan languages Chipewyan (Carter 1976) and Slave (Rice 1989:
779), in Cherokee (Iroquoian: Blankenship 1996), and in Nevome (Uto-
Aztecan); Cora has distinct classificatory verbs for domesticated animals
and for humans.
Unlike noun classes, numeral and possessed classifiers, verbal classifiers,
and classificatory verbs often do not classify nouns in terms of their
animacy, or human/non-human distinction (pace Croft 1994: 156 ff).
Some languages, e.g. Iroquoian (Mithun 1986: 386), do have a verbal
classifier for animals, or for humans. Alternatively, there may be some
other strategy for dealing with humans. In Imonda (Papuan: Seiler 1986)
human nouns are not classified at all. Imonda does have a few classifiers
which refer to animate nouns, e.g. u(e)- 'small animals' such as fish and
frogs. Athabaskan verbal classifiers—d for round things and n for long
things—do not include any term for people, or for animals: these remain
unclassified.
(B3) Physical properties in verbal classifiers and classificatory verbs Verbal
classifiers and classificatory verbs categorize nouns with respect to their
extendedness, i.e. shape/dimensionality and position; consistency (flexible,
rigid); and constitution (liquid or solid). In Palikur, verbal classifiers are
based on the form and dimensionality of objects (see Diagram 6.1).
SHAPE/DIMENSIONALITY usually go together with CONSISTENCY. This is
illustrated for two Algonquian languages, Ojibway and Cree, in Table
11.10 (Denny 1979a: 99).

TABLE 11.10. Verbal classifiers in Ojibway and Cree

Verbal classifier Translation
kinw-apek-an 'it is long' (e.g. rope)
long-one.dimensional.and.flexible-it.is
kinw-ek-an 'it is long' (e.g. cloth)
long-two.dimensional-it.is
napak-aplk-at 'it is flat' (e.g. ribbon)
flat-one.dimensional and flexible-it is
napak- (i)minak-isi 'it is a flat "roundish" thing'
flat-three dimensional-it.is
wawi-(y)ek-an 'it is round' (e.g. cloth)
round-two.dimensional-it.is
298 Classifiers
Dimensionality and orientation can be employed alongside consistency
and other inherent nature properties, e.g. 'granular masses'. Table 11.11
features Chipewyan classificatory verbs (Carter 1976: 25, 27).30

TABLE 11.11. Chipewyan (Athabaskan) classificatory verbs

Verb stem Object the verb classifies Example
('be in')
-?q 'inanimate solid objects' Lake, axe, stone, hat, body parts, ball
lti 'dead beings' Dead person, bear carcass, dead dog, raw
fish
-ti 'people, sleeping beings' Person, girl, sleeping baby, sleeping bear
-kq 'liquids' Water, blood, boiling water, milk
-dzdy 'granular masses' A pile of sand, a pile of sugar, loose
tobacco, powdered milk
-la 'rope-like objects; objects Rope, veins, several arrows, eyeglasses, two
in sets, or plurality of or more fish, three dogs, two oranges
objects'
-tq 'stick-like objects or Airplane, bow, empty box, canoe, chair,
empty containers' firewood, spear, cigarette, pen
-Lta 'containers with Box with stuff in it, can of beer, pack of
contents' cigarettes, cup of coffee, bottle of whisky
-Itsu 'fabric-like objects' Calendar, parka, pants, sheet of paper,
writing pad, book, one glove, a tree's leaf

How the consistency correlates with dimensionality and extendedness

can be illustrated in terms of the binary oppositions for classificatory verbs
in Western Apache (Denny 1979a: 107). Dimensionality is relatively inde-
pendent from extendedness.
Quality (singular, count, inanimate)

Extended' Non-extended

Rigid Flexible

One-dimensional One-dimensional Two-dimensional

le' tsoos
DIAGRAM 11.6. Classificatory verbs in Western Apache

30
Among verb stems listed by Carter (1976), -da 'sit' used for 'awake beings' (frog, spider,
sitting bear, sitting person, beaver, sitting baby) is usually not included in the set of classifi-
catory verbs (Chad Thompson, Keren Rice, p.c.).
 Semantics of Noun Categorization 299
Classificatory verbs and verbal classifiers often categorize nouns in terms
of their orientation in space. Nevome (Uto-Aztecan: Shaul 1986: 12; Table
6.6 above) has different sets of classificatory verbs for animate and inan-
imate referents depending on their position: horizontal ('lying') or vertical
('standing').
ORIENTATION may correlate with SHAPE/DIMENSIONALITY. Dakota (Siouan:
Boas and Deloria 1941: 126) uses positional verbs to classify nouns in terms
of their dimensionality, e.g. nazi 'stand', for people and animals; hq 'stand'
for long, upright inanimate objects'; lipaya 'lie', for other inanimate objects.
The use of POSTURE verbs as classificatory verbs in Enga (see Table 6.7)
and in a number of other Papuan languages correlates with the shape of the
object and its position in space. Table 11.12 illustrates classificatory exis-
tential verbs in Kamoro, a language from the Asmat family (Drabbe 1955:
39; Lang 1975: 116).31

TABLE 11.12. Classificatory existential verbs in Kamoro (Asmat)

Classificatory verb Semantics Nouns used with
ame Stand Vertical, high and tall or slender things,
people, houses, trees (alone, singular)
epe Sit Pots, dishes, pans, boats on land, plants,
mountains, clouds and celestial bodies
kai Lie Lands, rivers, lakes, fallen trees or wood
mariki Float Fish, people in canoes, anything
floating on water
naa Be there, be above Hanging objects, small things, big
masses such as a heap of rice or a pile of
sago, things lying on top of something
else

Different semantic parameters may interact to yield semantically com-

plicated systems. DIMENSIONALITY and SHAPE may correlate with CONSIS-
TENCY, FORM, and DIRECTIONALITY. A system of classificatory verbs in Ika

31
Classifcatory existential verbs are also found in Kewa (Franklin 1981), Asmat (Asmat
family), Kiwai (Kiwaian family), Huli (Engan), Melpa (Chimbu), Banz, Sinasina (Chimbu),
and Kate (Huon) (see Lang 1975: 115-20, for an overview). A similar system was described for
Southeastern Porno by Moshinsky (1974: 85-7), e.g. sca- 'non-long object rests on a surface',
'sit' (for humans); kto- 'long object rests on a surface in a vertical position', 'stand'; mti- 'long
object rests on a surface in a horizontal position', 'lie'; tla- 'long object turns on a horizontal
axis' (singular forms are given). In many languages positional verbs describe a noun in terms
of its position and even shape. Emmi (Australian) uses the verb 'lie' to refer to long and flat
objects and to geographically extended localities, while 'stand' is used with reference to trees
and mountains (Lys Ford, p.c.). However, this correlation is not systematic, as in Enga and
other Papuan languages; this is why these languages cannot be said to have verbal classifiers:
see (B) in §6.2.3.
300 Classifiers
(Frank 1990: 55) combines these properties: see Table 6.3 and 6.16 and
6.17. Classificatory verbs are found only for location and handling.
(B4) Arrangement and quanta The choice of a classificatory verb may
depend on the quantity of objects. Hani (Tibeto-Burman: LaPolla 1994:
75) has a classificatory existential verb ky31 'existence within a group' (the
full system of classificatory existential verbs in this language is given under
(A) in §6.2.3).
(B5) Function Function is also sometimes used as a semantic basis for
verbal classifiers. Imonda (Papuan: Seiler 1986: 190-3) has verbal classi-
fiers for 'fruit to be picked' (pot), or 'objects which are normally broken',
e.g. biscuits (put-) (see Table 13.6).
(B6) Semantics of different systems of verbal classifiers in one language
Languages with several different systems of verbal classifiers are important
for exploring the semantic possibilities of verbal classifiers. Two systems
cooccur in Waris, a language from the same family as Imonda. There is a
system of verbal classifiers which refer to the direct object (Brown 1981:
101-3); the categorization is based on consistency, shape, function,
arrangement and measures (including two specific classifiers): see Table
6.9. Classificatory verbs (shown in Table 6.10) refer to the orientation of
the S argument, and to its shape. The semantic parameters employed in the
two systems complement each other. In contrast, the two sets of verbal
classifiers in Palikur (Diagram 6.1) overlap; classifiers used with stative
verbs distinguish more parts of objects than classifiers employed with
transitive verbs. There are also a few further, subtler distinctions: the
classifier -kig 'pointed objects' is used with transitive verbs for -kiya 'sharp
edge' (on stative verbs); and the classifier -min used with transitive verbs
covers one-dimensional thin and rigid objects, e.g. tree trunks (classified
with -kat on stative verbs).

11.2.6. Semantics of locative and of deictic classifiers

Locative classifiers are rather rare (found only in South American lan-
guages: see §7.2). In their semantics, they are similar to non-animacy-based
verbal classifier systems. Categorization is based on physical properties, e.g.
shape and dimensionality, as in Palikur (Diagram 7.1).
The choice of a locative classifier may correlate with consistency. Daw
(Maku) has different locative adpositions depending on whether the refer-
ent of the argument is solid or flexible (see examples 7.5. and 7.6). The
choice of a locative classifier in Carib languages depends on dimensionality
and consistency of the referent of a noun (see Table 7.1). Animacy as a
parameter for locative classifiers has been found only in Lokono (see §7.2).
 Semantics of Noun Categorization 301
Deictic classifiers combine reference to shape and position, or to the
directionality of the item classified, similarly to non-animacy based verbal
classifiers (historically, they often come from grammaticalized posture verbs;
see §13.1.3). There is no special classifier for animates or for humans. The
deictic classifiers in Siouan languages indicate the orientation of the referent:
one-dimensional (long, vertical, or 'standing'), two-dimensional (horizontal,
or 'lying'), or three-dimensional (round, or 'sitting') (see §7.3; Barren and
Serzisko 1982: 99 ff.). Deictic classifiers fused with demonstratives can com-
bine reference to other properties of the item, such as extendedness, as in
Eskimo (Denny 1979a: 101; Table 7.4). Deictic classifiers in the Guaicuruan
languages Toba and Pilaga combine reference to visibility, extendedness and
position in space of the object (see Diagram 7.2, and (B) in §7.3).
There are not enough examples of locative or deictic classifiers in the
languages of the world for anything but very preliminary generalizations
concerning their semantics.

11.2.7. Semantic parameters in languages with several different types of

classifier
In languages with several different types of classifier these may semanti-
cally overlap to some extent. However, in languages with noun classes and
noun classifiers, the former tend to be animacy, or humanness-based, while
the latter tend to be function-based.
In Ngan'gityemerri, noun classes distinguish male, female, groups of
humans and a culturally important animal class (canine) (Table 8.1);
noun classifiers categorize inanimate objects in terms of their function
and nature (strikers, fire, liquid, digging sticks, large woomeras, and cane-
grass spears). Table 8.1 also shows that the two systems exhibit a semantic
overlap (one is historically the source for the other: see Chapter 13).
Numeral classifiers tend to be based on animacy and physical properties,
while relational classifiers are based on function, and on the inherent nature
of objects. Ponapean (Rehg 1981: 125, 179) has at least twenty-one rela-
tional classifiers and twenty-nine numeral classifiers. The choice of a
numeral classifier depends on dimensionality, shape and size (round,
long), arrangement (bundle, oblong piece of, strip, heap, etc.). There are a
few specific classifiers, e.g. for yams and banana. Relational classifiers are
function-based. Inanimate nouns are classified into edible and drinkable;
there are special classifiers for things to sleep on and for things to cover
with, as well as for buildings, vehicles, pillow, and catch (at sea or land);
human nouns are classified according to their kin relationship. There are
also a few specific classifiers which combine the properties of relational, and
of possessed classifiers (e.g. classifiers for names and for earrings: Rehg
1981: 180-1). A similar point is illustrated with Mokilese in Table 8.2.
302 Classifiers
Numeral classifiers involve distinctions in terms of physical properties
(shape/dimensionality, boundedness, directionality), and noun classifiers
are more function and inherent-nature based. An example from Akatek
is given in Diagrams 11.7 and 11.8 (see Tables 8.4 and 8.5).

Human

Non-human
Inanimate: dimensionality, shape, directionality and size
(vertical, curved, round, big flat, extended, etc.)
DIAGRAM 11.7. Semantics of fourteen numeral classifiers in Akatek

Human: sex (man, woman), value (appreciative),

familiarity (known)
Animacy Non-human: animal
Inanimate: nature-based: wood, stone, rope, salt;
consistency: water (liquid); function: plant
DIAGRAM 11.8. Semantics of fourteen noun classifiers in Akatek

Noun classifiers are usually based on a generic-specific relation, while the

choice of a numeral classifier depends on properties of the object. The same
noun can be grammaticalized in the function of a generic noun classifier,
and in that of a numeral classifier. The polygrammaticalization (for this
term, see Craig 1991) within Minangkabau of batang 'tree' as a noun
classifier, and as a numeral classifier is shown in Diagram 11.9.

^ 'Tree as a generic'
Generic noun classifier

Numeral classifier ^~~~'Vertical long object, often made of wood'

DIAGRAM 11.9. Polygrammaticalization of batang 'tree' in Minangkabau

In Palikur, animacy and humanness is distinguished only in noun classes

(genders), and in numeral classifiers. Numeral classifiers, verbal classifiers,
locative classifiers and noun classes operate with shape, dimensionality and
other physical properties in different ways. Numeral classifiers distinguish
three dimensions (with three values for two-dimensional objects, two values
for three-dimensional objects, and one value for one-dimensional objects),
boundedness and consistency, and a specific nature-based classifier (tree).
Verbal classifiers distinguish less values for three-dimensional objects, and
more values for one-dimensionals; they distinguish parts of objects (side,
 Semantics of Noun Categorization 303
edge, inside, parts of trees), and form (pointed). This may be explained by
the semantics of the verbs which take classifiers. Transitive verbs which take
classifiers imply direct physical contact with the object, and stative verbs
refer to dimensions, physical properties of objects, and their colours. Loca-
tive classifiers are similar to verbal classifiers; unlike these, they distinguish
constitution (liquid), but do not distinguish different cylindrical or flat
shapes. Gender assignment is based on animacy, humanness and consis-
tency, which goes with shape and material.
Languages with partly overlapping classifiers in multiple functions show
that numeral classifiers tend to have more animacy—or humanness-
based—distinctions. Baniwa (Table 9.5) has a human/non-human distinc-
tion just in numeral classifiers. Verbal classifiers, noun classifiers, and noun
classes distinguish just non-feminine humans, as opposed to non-feminine
animates.

11.2.8. Semantic parameters in multiple classifier systems

Multiple classifier systems (discussed in Chapter 9) are always based on
animacy, humanness, or sex. Inanimates—and, in some systems, also non-
human animates—are classified according to their shape/dimensionality,
boundedness, interioricity, consistency, constitution, and other 'inherent
nature' properties, as well as function. That is, they combine most semantic
parameters found in classifier systems in general, except for directionality.
In Nambiquara (Nambiquara family, Southern Amazonia, Brazil: Lowe
1999) classifiers divide into the semantic groups shown in Diagram 11.10.
Nouns are divided into human and the rest. Humans are further classified
by sex, and the rest by shape, interioricity, constitution, and other specific
(inherent) properties, and by function. There is no classifier for higher
animates—these are unclassified (Ivan Lowe, p.c.). Classifiers are prefixed
to numbers, as in 9.68; they are suffixed to adjectives as agreement
markers, as in 9.69. They are also used as derivational suffixes on nouns,
e.g huk-en'-su2 (shooter-cuHOLE.LiKE-iNDEFiNiTE) 'shotgun' (see (E) in §9.1).
Motuna (Papuan: Onishi 1994) has fifty-one classifiers which are used
with numbers, articles, and demonstratives, possessives, and as derivational
affixes on nouns. There are three general classifiers based on ANIMACY and
HUMANNESS—uru 'human', -wori 'animate', -mah 'thing'. Nine classifiers
refer to SHAPE and DIMENSIONALITY, e.g. round, concave, slender, stick-like,
fine, string-shaped, lengthwise objects. One classifier refers to ARRANGE-
MENT and SHAPE (-jaa 'object which is wrapped lengthwise'). Five mor-
phemes classify artefacts by FUNCTION. There are also classifiers by
QUANTA, and nature-based properties (plants and their parts, human hab-
itation). A few classifiers refer to time (day, month, year/season).
In Kilivila (Austronesian), classifier 'particles' are used with numerals, as
304 Classifiers
- Male jah3lo2
Female ka?3lu3
- SHAPE One-dimensional thin straight 'stick-like' kat3
One-dimensional bone-like su
Two-dimensional bounded leaf-like n?an3
Two-dimensional unbounded flat sheet like ka3lo3
Two- or three-dimensional, round ki3,i.e. 'ball', sphere'
INTERIORICITY - Hole-like, hollow en
CONSTITUTION Powder-like riun3
Liquid you3
SPECIFIC (inherent nature)- 'Country' ko?3
'House-like' thin3
"Work'yuy3
'Situation'jut3
'Word, thought'y'aw
DIAGRAM 11.10. Classifiers in Nambiquara

agreement markers on adjectives, and with demonstratives and other deic-

tic type markers (e.g. interrogatives). The three environments of classifiers
in Kilivila were illustrated in example 9.1. The classifier ke 'wooden objects'
is used on a number, and also on a demonstrative and on an adjective
(Senft 1996: 18).
The semantic domains covered by Kilvila classifiers are shown in
Diagram 11.11 (Senft 1991: 139; 1996: 237-90). Nouns are classified with
respect to their animacy. Inanimate nouns can be further classified accord-
ing to their functional properties, shape, dimensionality, etc. There are also
classifiers which refer to the quantity of animate and inanimate referents.

Animate: animals, persons, body parts

Animacy
Inanimate: function (utensils, roads and journey, texts, ritual items, dress);
nature and extendedness: place, shape, dimensionality, consistency (rigid and
long, row, wooden things, flexible things; flat, thin; holes); nature (yams,
entrances, fire); abstract: time; residue classifier.

Animate beings (team, group on the move, shoal, batch of fish, fish on strings)
Quantity
Inanimate beings (heap, grove, bundles (of taro), part, sheaf, handful),
measures (e.g. span of two extended arms)
DIAGRAM 11.11. Semantics of classifiers in Kilivila
 Semantics of Noun Categorization 305
11.2.9. Conclusions
The possibility of the same classifier morpheme being used in up to six
environments (see Chapter 9) goes against establishing strict correlations
between the morphosyntactic environment for a classifier and its semantics.
We can only establish a number of tendencies which emerged from the data
discussed in this chapter. These are shown in Table 11.13.
ANIMACY or HUMANNESS or SEX are basic for noun classes, numeral clas-
sifiers, and possessed classifiers. Verbal, locative, deictic, and relational
classifiers do not have to be animacy-based. (See Appendix 1, on how
animacy and human/non-human distinctions can be expressed through
other categories in languages with neither noun classes nor classifiers.)
Further subdivisions of humans, or animates, according to their SOCIAL
STATUS and KINSHIP RELATIONSHIP are found only for numeral classifiers
and noun classifiers. Physical properties such as SHAPE, FORM, SIZE, BOUND-
EDNESS, INTERIORICITY, and CONSISTENCY are less often found in noun classes.
They are usually found in all other classifier types except noun classifiers
and relational classifiers. Size, boundedness, interioricity, and consistency
tend to be concomitant with shape.
DIRECTIONALITY or ORIENTATION is found in numeral classifiers, verbal
classifiers (especially classificatory verbs), and deictic classifiers as a feature
concomitant to shape. FUNCTION is almost always a semantic dimension of
relational and possessed classifiers, and it is often found with numeral,
noun and verbal classifiers. It is rare, however, in noun class systems.
NATURE-based specific classifiers are found in all the classifier types but
only rarely in noun class systems. QUANTA and ARRANGEMENT are found
with numeral classifiers and with verbal classifiers. GENERIC-SPECIFIC
relations are used only in noun classifiers, possessed classifiers, and verbal
classifiers (and rarely in numeral classifiers).
These semantic preferences are universal.32 However, they are not strict
rules. In reality, classifiers of different types show a large degree of seman-
tic overlap, and the same morpheme can be used in several classifier
environments.
32
We will see in Chapter 13 how they can be important for semantic extenstions in the
course of the historical development of classifiers.
TABLE 11.13. Preferred semantic parameters in classifiers
Classifier Grammatical function Scope Typical semantics Generic-specific
relation
Noun classes Determination, head-modifier NP Animacy, physical properties, rarely No
agreement nature or function
predicate-argument agreement clause
Numeral classifiers Quantification, enumeration NP Animacy, social status and kinship Rare
relationship, directionality and
orientation, physical properties, nature,
quanta, arrangement, functional
properties
Noun classifiers Determination NP Social status, functional properties, Yes
nature
Verbal classifiers O/S agreement spatial location clause Physical properties, directionality and Yes
orientation, nature, function, quanta,
arrangement, rarely animacy
Relational classifiers Possession NP Functional properties, nature No
Possessed classifiers Possession NP Physical properties, nature, animacy, Yes
functional properties
Locative classifiers Spatial location NP Physical properties, nature, rarely No
animacy
Deictic classifiers Spatial location, determination NP Directionality and orientation, physical No
properties, nature
 12 Semantic Organization and
Functions of Noun Categorization

This chapter considers various aspects of the semantics of noun categor-

ization devices and their implications for the study of human cognition and
culture.
Semantic features encoded in noun categorization reflect principles of
human cognition and world perception. This is why it has often been
argued that classifiers offer 'a unique window' into studying how humans
construct representations of the world and encode them into their lan-
guages (Lakoff 1986; Craig forthcoming). The semantics of classifiers is
often conditioned by the socio-cultural and physical environment in which
a language is spoken and is often affected by social changes, language
planning, and cultural obsolescence.
In §12.1 I discuss the semantic organization and functions of classifiers.
Cognitive and perceptual mechanisms which underlie the structuring of the
world seen in terms of classifiers are analysed in §12.2. How noun categor-
ization could depend on social and cultural environment—and change if
the society changes—will be dealt with in §12.3; conclusions are given in
§12.4.

12.1. Semantic organization and functions of classifier systems

The semantic complexity of an individual classifier varies, and so does its
scope. Some classes are semantically straightforward, while others are
heterogeneous. In this section we consider the principles of overall seman-
tic organization of individual classes, and of systems at large. First, in
§12.1.1, we look at semantic complexity in classifier systems, including
principles of semantic extensions and opacity in classifier choice.
Classifiers can have different functions. They can be employed to delimit
the scope of reference of a noun, or of a modifier within a noun phrase, or
of the predicate within a clause. Classifiers are seldom semantically redun-
dant, because they highlight some relevant aspect of the noun referent (e.g.
its shape, function, or value). The semantic roles of classifiers are discussed
in §12.1.2, and their discourse-pragmantic functions in 12.1.3. The applic-
ability of classifiers, default classes, and residue classes are analysed in
§12.1.4.
308 Classifiers
12.1.1. Semantic complexity in classifier systems
The choice of noun classes and classifiers can be semantically transparent.
In Malay and Minangkabau the classifier 'person' is used with all humans;
the animal classifier (with a literal meaning 'tail') is used for all animals
(Marnita 1996: 106).
On the other hand, the semantic structure of a class can be extremely
heterogeneous. The semantic structure of the classifier -hon in Japanese
(Lakoff 1986: 25-6; Matsumoto 1993: 676-81) is complex. In its most
common use it covers saliently one-dimensional objects, e.g. long, thin,
rigid objects such as sticks, canes, pencils, candles, trees, dead snakes, and
dried fish. It also covers martial arts contests with swords (which are long
and rigid), hits in baseball, shots in basketball, Judo matches, rolls of
tape, telephone calls, radio and TV programmes, letters, movies, medical
injections, bananas, carrots, pants, guitars, and teeth.
This heterogeneity results from various processes of semantic exten-
sion. Different classes are heterogeneous in varying ways, because of
different extensions and relations within the semantic fields of classi-
fiers. While defining a category by simply listing its members is possible
for any classifier, we search for a general characterization which will
accord with the idea of internal homogeneity of cognitive and linguistic
processes. A prototype model accounts for semantic extensions of
classifiers.1
Thus, classifiers do not have to be homogeneous lexical categories
bounded by a set of features shared by all the members (see Sweetser
1986, for different mechanisms behind polysemy and abstraction).
(A) Prototype approach
Some members of a given class are perceived by native speakers as more
salient than others—these members are cognitively more central.2 Other,

1
Downing (1996: 96-8) obtained four sorts of definition for Japanese numeral classifiers
with respect to their membership: specification of kind by means of reference to a roughly
equivalent noun (inductive rule); listing of members; citation of a representative member of a
category; specification of characteristics of members of the category (e.g. shape for hon
'classifier for long, thin things').
2
Diagnostic criteria for prototypes used in the literature, summarized by Downing (1996:
103-4), include: (i) frequent and/or early appearance on lists of category members; (ii)
appearance as the model for an analogical change or as the source for metaphor- or meto-
nymy-based additions to the category; (iii) appearance as the source for the meaning of
morphological derivatives of the category label; (iv) high acceptability rating as a category
member; (v) consistency with the etymological source of the category label; (vi) loss of
category members unrelated to this one; (vii) frequent usage of the category label with respect
to this category member (e.g. frequent citation in dictionary entries for the category label);
(viii) ability of subjects to quickly judge statements about the category affiliation of this
member; (ix) early acquisition as category member by children. One should keep in mind
 Semantic Organization and Functions 309
more peripheral members are associated with a particular class because
they share at least one feature with one, or more, prototypical members (on
'prototype effects' see Rosch 1973; 1975a; 1975b; 1987; Barsalou 1987;
Lakoff 1987).
For instance, prototypical members covered by the classifier hon in
Japanese would be sticks, or canes; hits in baseball (see below) or phone
calls are more marginal. Idiolectal variation is typical of peripheral
membership of classifier category. Matsumoto (1993: 676) points out indi-
vidual variation as to the use of the Japanese classifier hon for non-central
members of the 'one-dimensional' class—i.e. one-dimensional objects
which are circular (e.g. rubber bands or tubes). With respect to child
language acquisition of non-central members of classifier categories,
Matsumoto (1985) noted that his five- to seven-year-old Japanese-speaking
subjects had no difficulty in applying hon to novel objects with a percep-
tually salient thin, long shape; but it was difficult for them to apply it
spontaneously to non-central, conventionalized members of the hon cat-
egory obtained by metonymical extension. His explanation for the acquisi-
tion of these conventionalized uses was that children have to 'learn the use
of hon for atypical cases in an item-by-item fashion, for the acquisition of
knowledge of the membership of atypical members of a classifier category
seem to depend heavily on the actual exposure to these uses in the input'
(Matsumoto 1985: 168).3 Extensions from central to less central members
are discussed in (B).
(B) Extension principles: chaining and metonymy
The prototype and extension model presupposes the existense of more and
less central members. Complex categories are structured by chaining, i.e. by
common properties.
Classifier categories often contain members which have been taken into
this category because of some semantic relationship they bear with its pre-
existing members. Japanese hon used for long, thin things has been
'extended' to hits in baseball, since they refer to 'straight trajectories,
formed by the forceful motion of a solid object, associated with a baseball
bat, which is long, thin and rigid' (Lakoff 1986: 25; but see criticism in

that diachronic semantic extensions (see Chapter 13) have to be distinguished from the
synchronic organization of a class.
3
A recent application of the prototype theory to a selection of classifiers in Mandarin
Chinese (Chang-Smith 1996) confirmed the existence of prototype effects in linguistic cat-
egorization with classifiers. The prototypical member of the category covered by a classifier
proved to be the preferred one, e.g. 'car' for Hang4 'CL:LAND.VEHICLES', 'paper' for zhang(
'CL.FLAT.THIN.OBJECTS', and 'pen' for zhi3 'CL.:LONG.RIGID.OBJECTS'. There were no 'favoured
nouns' for classifier tiao2 'CL:LONG.FLEXIBLE.OBJECTS' which might play the role of prototypes;
this may be due to the wide range of salient perceptual properties associated with this classifier
(Chang-Smith 1996: 52).
310 Classifiers
Matsumoto 1993: 677, and Downing 1996: 100). Then it was 'extended' to
shots and serves in other sports, and even to phone calls. This is what Craig
(forthcoming: 13) describes as a 'chaining model'.
As a result, classifiers and their semantic representations can differ in
complexity, and may be difficult to characterize in terms of a set of
abstract features, or of the prototypicality of their referents. For
instance, the human non-feminine classifier -ite in Baniwa applies to
some human attributes; this marker is employed to refer to shirts and
other items of clothing, eyeglasses, and musical tunes. Thus, it is only
mildly heterogeneous. In contrast, Japanese hon is extremely heteroge-
neous because of various chaining extensions from the prototype. Most of
these extensions become conventionalized; those which do not display
idiolectal variability.
A new member can be taken into a category either because it shares
certain characteristics with most members of the category or because it
is judged to have some similarity with only some of them (see the
discussion of these cognitive principles by Langacker 1987). In Garo
(Adams and Conklin 1973; Conklin 1981) the classifier which refers to
round objects (stones, balls, eyes, coins and round fruit, e.g. oranges and
mangoes) can be extended to just a few other fruit of a different shape
(e.g. bananas).
Extensions can be based on certain rules for transferring class member-
ship. Dyirbal employs the following principle (Dixon 1982: 179):
(i) If some noun has characteristic X (on the basis of which its class membership
would be expected to be decided) but is, through belief or myth, connected with
characteristic Y, then generally it will belong to the class corresponding to Y and
not that corresponding to X. (ii) If a subset of nouns has some particular
important property that the rest of the set do not have, then the members of
the subset may be assigned to a different class from the rest of the set, to 'mark'
this property.

According to these principles 'idealized models of the world'—for

instance, myths and beliefs—can account for other chaining links within
the structure of a class. In Dyirbal, birds belong to 'feminine' Class II
because they are believed to be the spirits of dead human females. In
Palikur, heavenly bodies—sun, moon, stars, and also thunder and light-
ning—are the only inanimate objects with masculine gender; this is because
they are human males in myth. Other extensions are linked to similarities in
shape, or directionality. 'Image-schema' transformation involves various
extensions, e.g. the ones based on analogy between shape and trajectory of
an object ('conduit-metaphor': Matsumoto 1993: 677). These extensions
for the Japanese classifier hon employed for one-dimensional objects and
their paths—e.g. hits and home runs in baseball, as well as serial stories,
 Semantic Organization and Functions 311
movies and play scripts—are based on 'experiential one-dimensionality'
(cf. Lakoff 1986: 26-7).
A further type of extension is the 'Domain of Experience Principle'
which links together members considered as associated with the same
experience domain. Thus, fish in Dyirbal belong to Class 1 since they are
animate, and so do fishing implements because they are associated with the
same activity. These domains are often culture-specific, and subject to
change with socio-cultural changes. The numeral classifier lay in Korean
was originally used with reference to traditional vehicles, and then got
extended to introduced European artefacts with wheels. There was a
further extension, to any electric machinery, and to other kinds of
machines or instruments, including even the piano (Lee 1997).
Metonymical models of different sorts are also applied when linking
objects together in one class. A metonymical model presupposes that 'a
word or expression normally or strictly used of one thing is used of some-
thing physically or otherwise associated with it' (Matthews 1997: 224;
Lakoff 1986: 33, for their properties). Extensions and metonymic transfers
are important for explaining synchronic and diachronic semantic changes
in classifiers. In Austroasiatic languages, shape parameters in inanimate
categorization can be described as 'geometrical generalizations from natu-
rally occurring forms'. In Tai (Conklin 1981: 136), these generalizations are
derived from plants and their component parts. Some typical generaliza-
tions are: small and roundish (from seed), round (from fruit), bulky (from
tuber), flat and sheet-like (from flower, leaf, fibre), long (from stalk, stick,
sprout) (Conklin 1981: 341).
(C) Metaphors in classifier systems
Metaphors and metaphorical transfers (see the definition in Sapir 1977: 6)
are important for the semantic organization of classes and the ways in
which they get extended. This can be illustrated from the semantic or-
ganization of numeral classifiers in Burmese described by Becker (1975;
1986). The structure of underlying Burmese categorization is based on an
interpretation of self—see Table 12.1. As Becker (1975: 118) puts it, 'the
structure underlying classification starts with the self at the centre, divides
the self into head and body, and then ranges objects at four distances from
the self, associating them either with the head (metaphorically top, round)
or with the body (metaphorically, bottom, straight)'. The system of cat-
egorization operates on 'applied metaphors'; thus 'head to the body' is the
same as 'cup to saucer', and so on; the relationship above/below is basic for
categorization. Moreover, since several of the classifiers are words for parts
of a tree, one can say that 'the tree is the metaphor for the person'.
312 Classifiers
TABLE 12.1. Burmese numeral classifiers for inanimate objects
Centre 1st Orbit 2nd Orbit 3rd Orbit 4th Orbit
Self: part of self On self Near self Far from self
(inalienable) (alienable)
Head ywc': pain: loun: sin:
hair on head head-dress round, upper things: upper things which
leaf posts have circular orbit:
furniture sun
cup rivers, sea
script arrows
needles
Body chaun: kwin: cha': si:
hair on body body dress flat, lower things: lower things which
fingers and toes body ornaments boards move in straight
teeth the: mats lines:
pin\ folded clothes saucer vehicles
sticks palmleaf for writing hunted animals
twigs le': horses
pens instruments used in dupes
the hand; swords; thwc
musical instruments; rivers
puppets roads

Understanding metaphorical transfers in classifier systems requires

knowledge of historical and cultural background. As Becker (1986: 337)
puts it: The important thing here is to see that the classifier is selected from
a kind of conceptual space which has been historically shaped.' Thus, in
Burmese, 'unless one knows that the traditional Burmese pictorial map of
the cosmos has man located on an island, from the centre of which flows a
river in a spiral course to the sea, one may question why rivers and oceans
are classified here along with arrows and needles, which move in circular
orbits' (Becker 1975: 118). Rhetorical uses of classifiers are further exten-
sions of metaphors, and 'the use of classifiers in Burmese ... is in part an
art and not just a grammatical convention' (Becker 1975: 113).
Personification is a typical metaphorical extension. It can be realized
with different means. In Standard English, personal pronouns tend to be
used with humans; but they can also be used with non-human animates—
say, pets—to express their particular closeness or dearness to the speaker;
then they become almost human-like. The Tasmanian variety of English
(described in Pawley forthcoming) has strict rules which govern the choice
of pronouns he and she with non-human referents. For instance, when the
sex of a higher animal is not known, the animal is referred to as 'he'; when
 Semantic Organization and Functions 313
referring to an inanimate entity, he is used with plants or parts of living
plants, and any item of goods or portable property (other than vehicles)
that is viewed as trade goods rather than a personal possession.
Gender languages often use gender metaphorically to describe unusual
situations (see (C) in §2.4.3). In Manambu, a small, effeminate man can be
jokingly referred to as feminine. Gender distinctions can be used meta-
phorically in languages with no, or almost no, agreement gender. Ilocano
(Austronesian: Rubino 1997: 75-6) does not have genders; the natural
gender of humans may be distinguished lexically, e.g. lalaki 'boy', babai
'girl'. These lexemes are also used to distinguish natural gender of hybrid
nouns, e.g. kabsat a lalaki 'brother', kabsat a babai 'sister'. They can also
cooccur with inanimates, highlighting semantic features associated with a
metaphorical extension of 'masculinity' or 'femininity'. For instance, basi
a lalaki 'strong sugar cane wine' is associated with masculinity; and basi a
babai 'sweet sugar cane wine' with femininity.
Gender languages often manipulate gender in legends. Personification of
nouns with inanimate or abstract referents follows the gender which is
assigned by non-semantic rules. For instance, Russian smert' 'death' is
feminine and so is personified as a woman in folk tales and cartoons
(Rothstein 1973: 464).4 Similarly, morte 'death', a feminine noun in
Portuguese, is depicted as Dona Morte (Lady Death) in cartoons.
Change of genders can have a stylistic effect. Some nouns of feminine
gender, essentially hostile epithets, can be applied to human beings.
According to Rothstein (1973: 464), in Russian it is more insulting to
call a man dura (fool: feminine) than durak (fool: masculine) (cf. (C) in
§11.2.1; on the use of masculine diminutives of feminine names to express
affective jocularity in Polish, see Wierzbicka 1996: 398).
Some systems accept metaphorical extensions more easily than others—
this happens because systems differ in their productivity. The productivity,
or vitality, of a system is measured by its ability to accept and classify new
members, and reanalyse and extend the semantic range of a noun categor-
ization device over time. Thai has an old classifier system which has great
vitality and productivity; the more productive a system is, the more meta-
phorical and other types of extensions it accepts. In contrast, the Jacaltec
system of noun classifiers can be looked upon as frozen. New loanwords
are simply left unclassified; the system does not accept semantic extensions
either (Craig forthcoming). Frozen or non-productive systems do not have
to be in decay, though they sometimes are, as is the case of many classifier
systems changing due to language obsolescence and contact.5
4
In this and similar cases we cannot tell whether 'death' is depicted as a woman 'because' it
is feminine, or the other way round.
5
There are two ways of expanding membership of a noun class: intensional, i.e. based on
the characteristics shared by most members, and extensional, i.e. based on chaining analogy
314 Classifiers
An instructive example of prototype-and-extension in a multiple classi-
fier system comes from the classifier tua in Thai. The structure of the
category is schematized in Diagram 12.1 (Carpenter 1987: 46). Lines
indicate extensions from a prototypical member to less prototypical ones
(Carpenter 1987: 45-6; cf. Downing 1996: 101-2).

Underwear, bathing suit Cabinets, dressers, beds

Skirts

Trousers, shirts Tables, chairs

.ANIMALS'

Ghosts Mannequins Letters

Numbers
DIAGRAM 12.1. Structure of the tua category in Thai

Carpenter (1987: 45-6) gives the following explanation of the extension

and chaining principles which apply to tua (semantic features which could
have served as basis for the extensions are provided by me in small caps):
This covers a wide, but not incoherent, range of things. The articles of clothing used
with /tua/ are trousers, shirts, jackets, skirts, and, less often, dresses, underwear and
bathing suits. The items of furniture most likely to be used with /tua/ are tables and
chairs, and less often, dressers and beds. These assignments suggests that it is THE
PRESENCE OF LIMBS, giving these objects a body-like shape, that causes speakers to
classify them with /tua/. Originally, the classifier was used with animals, and the
PROTOTYPE is probably some good QUADRUPEDAL ANIMAL, such as a dog or a water
buffalo. Tables and chairs were included on the basis of SHAPE, either because of
their general quadrupedal outline or perhaps because of the specific presence of
limbs. Other kinds of furniture were then added because of their shared FUNCTION
with tables and chairs. Trousers and shirts were also included because of their
SHAPE, again either because generally they follow the contours of the body of

with few members. Downing (1996: 117) hypothesizes, for Japanese, that when the classifier is
associated with a large referent class, 'the sheer number of members meeting the intensional
requirements for membership will work against member-focused expansion. No such obstacle
would exist in the case of categories composed of but a single member, or of a few members
bearing no obvious intensional relationship to each other, as in the case of sao "chests, flags,
poles, samisens, stick-shaped sweets" where it is only the extension-based assimilation that is
possible.' The overall preference for intensional rather than extensional expansion of classi-
fiers is corroborated by a tendency to 'optimization' of prototype-centred categories in
achieving structural stability and semantic flexibility and adaptability. Cf. Geeraerts (1988:
223): 'the categorial system can only work efficiently if it can maintain its overall organisation
for some time ... At the same time, however, it should be flexible enough to be easily
adaptable to changing circumstances. To prevent it from becoming chaotic, it should have a
built-in tendency towards structural stability, but this stability should not become rigidity, lest
the system stops being able to adapt itself to new and unforeseen circumstances.'
 Semantic Organization and Functions 315
their wearer, or specifically because they have limbs. Other kinds of clothing
were included because of their SHAPES. A letter (of the alphabet) in Thai is a
compound, /tua nangseu/ 'body book', so a combination of shape and repetition
of the generic compound head caused letters to be classified with /tua/. Numbers
were included either on the basis of shape or by their shared function with
letters.
Similarly, ghosts were included because of their similarity with the shape
of a human body. Carpenter concludes: 'The internal structure of this
category, then, clearly, mixes prototypes and chains, with strongest mem-
bers being those closest to an animate quadruped, but some chaining based
on similarities to non-prototypical members.'
Opacity in classifier assignment can be due to semantic extensions that
are explainable but unpredictable. Table 12.2 illustrates human classifica-
tion in Burmese based on social status and age (Becker 1975: 116) and its
extensions to human attributes.

TABLE 12.2. Animate classifiers in Burmese

hsu pa u: jau kaun
Buddha and Deities, saints, People of status, Ordinary Animals, ghosts,
his attributes: monks, royalty teachers, scholars humans dead bodies,
relics, images, depraved people,
the Law children

The classifier hsu used for Buddha got extended to Buddha's words, and
thus to Buddhist law. This classifier has undergone further analogical and
metaphorical extensions. Hsu can apply to the whole field of human existence,
and this term was extended to items 'conceptually similar to the system with
its centre and measured distances, e.g. concentric networks like mosquito nets
and fish nets (both of which in traditional Burma were conical in shape),
gardens (which were laid out as a wheel), and staircases' (Becker 1975: 116).
These extensions, ultimately based on extendedness and shape, are reminis-
cent of shape extensions of genders and noun classes (see §11.1.1). They
contribute to the increase of opacity in classifier assignment.
Similar metaphorical extensions are found in other languages. In Bugis,
tau 'human classifier' is usually employed for counting people; however,
classifier lisa 'small spherical objects' can be used for counting people who
died in a war, presumably, because the heads of dead enemies used to be cut
off and people were counted by their heads (Sirk 1983: 63).
Thus, the composition of a classifier category and its expansion may be
hard to predict. Though it is often possible for an analyst to suggest a
rationale for the inclusion of most members of a given category, this may
have not been the actual rationale employed by speakers. There is also a
316 Classifiers
certain degree of unpredictability associated with social and cultural
conventions.6
(D) Semantic structure and hierarchical organization of classifiers
Classifier categories are semantically heterogeneous. There have been a few
attempts to present the semantic organization of classifiers in terms of
taxonomic trees and binary oppositions (Denny 1979a; 1979b; see criticism
by Downing 1996: 125); however a superordinate-subordinate approach
has limited applicability to classifier systems. The reason why superordi-
nacy relations are limited and are unlikely to involve all the classifiers lies
in the coexistence of different and cross-cutting semantic rationales (kind
and quality classifiers), and on different types of extension. Some referents
are included in more than one class (see §12.1.2 on variability in classifier
assignment and the issue of semantic roles and functions of classifiers).
Thus, a sword in Japanese can be referred to by a shape-based classifier hon
'long, slender object', by a kind classifier furl 'sword', or by a function-
based quality classifier ten 'items in an inventory, works of art'. It is also
impossible to make all shape-based classifiers superordinate to kind classi-
fiers which possess appropriate shape, because kind classifiers may unite
referents of different shapes (e.g. the verbal classifier -pit used for any
irregular shape object in Palikur). Other problems with creating distinct
taxonomies result from a generic classifier, such as Japanese tsu, which
participates in all of them.
There are hardly any hierarchical relations in the assignment of noun
classes (genders), deictic or locative classifiers; hence the discrepancies
between scientific and folk taxonomies and noun categorization devices.
Assignment of relational classifiers implies a functional categorization of
objects, with no clear-cut hierarchical relations.
The few Japanese numeral classifiers which have superordinate-subordinate
relation are given in Table 12.3. In all these cases 'the more general term
[can] be used for all members of the category denoted by a more specific
term' (Downing 1996: 126).
6
A problem may arise in how to distinguish metaphors from polysemy. For instance, in
Kilivila (Senft 1996: 19) a 'dinghy' can be referred to with a classifier 'child', as well as
'wooden', e.g. ma-gudi-na waga ke-kekita (this-CL:CHiLD-this canoe CL:WOODEN-small) 'this
small dinghy'. This can be interpreted in two ways: either as a metaphorical extension (child
> small object), or a polysemy: child, any small creature, or object. We think that a decision
should be made in terms of language-internal criteria on what is literal and what is meta-
phorical meaning, before there is a universally accepted 'serviceable' 'clarification of the
distinction . . . between conventional metaphor and systematic polysemy', to 'separate (even
provisionally) the literal from the metaphorical... we need to accommodate the fact that over
time metaphorical expression-systems may lose their metaphorical duality and assume the
status of literal meanings, as when body-part terms become locational adpositions' (Goddard
1996: 150). This change is somewhat similar to semantic processes which take place in
grammaticalization.
 Semantic Organization and Functions 317
TABLE 12.3. Verified superordinate-subordinate pairs in Japanese numeral classifiers
Superordinate Subordinate
1. dai 'vehicles, furniture, machines' ki 'air vehicles', ryo 'train cars', taku 'tables,
desks'
2. hen 'literary work' shu 'poems', ku 'poems'7
3. hiki 'animals' too 'large animals'
4. heya 'rooms' shitsu 'rooms', ma 'Japanese style rooms'
5. nin 'human beings' mei 'human beings' (honorific)
6. hon 'long, thin objects' furi 'swords'
7. tsu 'inanimates' most kind-classifiers for inanimates

The situation is different with generics. The generic-specific taxonomy

can be nature-based, or functionally defined (produce, pets, food, prey).
Evidence from Yidiny shows a hierarchical relation between the two (cooc-
currence of the two, i.e. 'person' plus 'man', or 'person' plus 'woman'),
which is by no means universal (see §3.2.1 and 3.4 above).
Thus, a taxonomic approach to classifiers may be useful, but only in
limited circumstances.8 On the other hand, prototype-extension and chain-
ing models are more applicable in heterogeneous non-hierarchical systems
such as noun classes, numeral classifiers, some verbal, and all locative and
deictic classifiers. These models do not apply to those instances in which
there is a generic-specific relationship between classifier and noun.

12.1.2. Semantic roles of classifiers

We saw in §11.1 that classifiers differ as to the degree of 'abstractness' of the
features involved. This is particularly relevant for classifiers of inanimates
because these involve many more shape- and function-based features. We
saw in §11.2 that there are fewer 'specific' classifiers in the domain of
animates and especially of humans (though there are examples, e.g. 'guest'
in Kana, 'dog' in Cahuilla, or 'canine' in Ngan'gityemerri). Classifiers
which centre round a narrow array of representative members are 'induc-
tive' or 'kind classifiers'; also called taxonomy-specific (Downing 1996:
90 ff., 118), and those which are organized on the basis of a feature are
'deductive', or 'quality classifiers'. The two differ in how restricted they are.
Kind classifiers are more culture-specific, they are acquired later, and they
are more semantically redundant. Quality classifiers lack these properties.
Classifiers can have several semantic functions which are linked together.
7
The classifier ku refers to haiku (17-syllable poems) and other short poems, while shu
refers to other poems (Downing 1996: 20 and 22).
8
See Lobel (forthcoming) for a convincing application of this model to Vietnamese; a
somewhat different analysis is suggested by Goral (1978).
318 Classifiers
(A) Quantifying and individuating functions
The quantifying function of numeral classifiers is connected to the idea
that 'the noun refers to some kind of mass and the classifier gives a unit of
this mass' (Denny 1986: 298). A numeral classifier is then viewed as a unit
of collectivity (Greenberg 1978).9 Classifiers are used when reference to
particular individuals is required (see (B) below). This function explains
the anaphoric use of classifiers. Nouns are more often deleted from
numeral phrases than classifiers, since classifiers refer to the type of indi-
vidual being enumerated, and nouns only specify some of their properties
(Denny 1986: 301). In the following example from Minangkabau (Marnita
1996: 93), both the noun and the classifier appear in the question, but only
the classifier is retained in the answer:
12.1. bara lamang sa-batang
how.much lamang10 one-NUM.CL:LONG.RiGiD
'How much is one lamang?'
12.2. saribu sa-batang
one.thousand one-NUM.CL:LONG.RIGID
'One thousand (rupees) for one.'
Once the property, or domain of reference (traditional food 'lamang'), is
established, the noun can be omitted; but reference to a general class of
long-shaped objects is essential.'!
Individuation is essential for the use of classifiers with deictics (see also
Lobel forthcoming: 51-2, examples 6.28-30 for an illustration of the sin-
gulativizing function of classifiers in Vietnamese). With Thai demonstra-
tives, classifiers are more likely to be used when reference to particular
individuals has to be stressed, for instance, in the case of contrastive focus.
In 12.3, from Thai (Conklin 1981: 86), the classifier lang5 'building, roof,
cannot be omitted.
12.3. baan3 lang5 nii4 mii1 (pra 2 tuu') saam5 pra2tuu1
this CLtROOF this have (door) three CL:DOOR
suan2 lang5 nan 4 mii1 soong5 pra 2 tuu ]
but ct:ROOF that have two CLIDOOR
This building has three doors, but that building has two doors.'
9
T'sou (1976) gives a different explanation for quantifying functions of numeral classifiers
in terms of quantity and entity.
10
Lamang is traditional Minangkabau food made of glutinous rice which is cooked inside a
medium-size piece of bamboo.
1
' A separate question is to what extent numeral classifiers have to express actual counting,
enumeration, or just specification. Also, is enumeration universal? Hale (1975) showed that so-
called numbers in Australian languages are specifiers which traditionally did not have any
counting functions. In some South American languages numbers establish quantity; but
contexts of actual counting are almost nonexistent (see §4.1).
 Semantic Organization and Functions 319
(B) Classifying and categorizing functions of classifiers
Classifiers provide information about sorts, or classes, of units. Consequently,
one of the functions of classifiers with nouns is to provide expectations about
the verb predicates; conversely, a function of classifiers with verbs is to
provide information about a nominal argument. For instance, only some
predicates are appropriate with human or with inanimate subjects, or objects
(Denny 1986: 302-3). Classifiers serve to organize human knowledge into
classes according to the principles of human perception and human function-
ing; they 'have a specialized role separate from that of nouns—they establish
expectations about the verb predicate which the speaker will use, and about
other verb predicates which are likely to be relevant as thought and conversa-
tion continue . . . nouns have the quite different role of helping to identify
the thing being referred to' (Denny 1986: 303). Classifiers also represent
semantic categorization, i.e. the organization of knowledge encoded in a
language. This knowledge correlates with socio-cultural variables, and with
universal principles of cognition. It may reflect the cognitive and functional
categorization of objects within a particular culture (see §§12.2 and 12.3).
However, classifiers encode different information from that carried by
nouns. There are then two possibilitites—both show that classifiers are not
semantically redundant.
One is ADDING INFORMATION to the nominal. This is especially salient in
cases when different classifiers are used with the same noun. This 'vari-
ability' in categorization is attested for almost any classifier type.
The choice of a classifier for humans may depend on their social status.
A student can be considered an ordinary person, or else a member of a
higher class, as in Korean. When inanimate nouns appear with different
classifiers, these highlight different aspects of their meaning.
A well-known example from Burmese (Becker 1975: 113) illustrates this
point. 'River' can be spoken of in at least eight contexts, shown in Table
12.4. Note that here numeral classifiers are most certainly not semantically
redundant.

TABLE 12.4. Redassification of an inanimate noun in Burmese

Noun Numeral Classifier Translation
myi? 19 ya? 'river one place' (e.g. destination for a picnic)
myi? 13 tan 'river one line' (e.g. on a map)
myi? hmwa 'river one section' (e.g. a fishing area)
myi? te 'sin 'river one distant arc' (e.g. a path to the sea)
myi? te thwc 'river one connection' (e.g. tying two villages)
myi? te 'pa 'river one sacred object' (e.g. in mythology)
myi? te khu 'river one conceptual unit' (e.g. in a discussion of
rivers in general)
myi? te myi? 'river one river' (the unmarked case)
320 Classifiers
Specific classifiers ('kind classifiers' in Downing 1996: 93) can also add
information about the referent, since they allow speakers to distinguish one
sense of the referent from all the others. The repeater myi? in Table 12.4
indicates that a river is looked upon just as a river, and helps discard other
senses (see further examples in Downing 1996: 92-3).
Classifiers can add unusual information about the referent. The use of ten,
a classifier for works of art, with stamps in Japanese may be done to convey
the idea that some particular stamps are of artistic value (Downing 1996: 94;
ex. (38)). The classifier dai 'small countable objects' can be used with nouns
normally describing large objects (e.g. houses or boats) to refer to toys.
Examples of variable categorization can be given from other types as
well. In Yidiny, different noun classifiers can be used with a word for 'cave'
depending on whether it is looked upon as a possible habitation, or as
something made of stone. 'Plant', or 'banana' can be either planted or
eaten, and these functions require the use of different possessed classifiers,
as in Palikur or Apalai (see Chapter 5).
In closed noun class systems the variability of categorization is often
limited. Variable noun class assignment and its semantic effects were
discussed in §2.4.3.
Classifiers also SUPPLEMENT the information carried by nominals. They
provide speakers with the means of accommodating new referents into the
pre-existing system of categorization. In Japanese, for instance, kokki
'national flag' can be used with different classifiers depending on its posi-
tion, e.g. whether it is flying or folded up. If a language has several noun
categorization devices, these may supplement the information carried by
nouns in varying ways. Taxonomies reflected in two classifier systems differ
from one another, and supplement the information contained in nouns in
different ways.12

12.1.3. Discourse-pragmatic functions of classifiers

Semantic roles of classifiers correlate with discourse. The use of noun
categorization devices often depends on the role of the nominal argument
12
Ben ton (1968: 142-3) illustrates the point about how classifiers supplement the lexical
contents of nouns: The judicious use of classifiers . . . makes possible the extension of
meaning of a particular base with a minimum of ambiguity. Nouns thus often have a highly
generalised meaning, different segments of which are expressed with the aid of different
classifiers. Within the classifier systems themselves different patterns emerge. There are points
of overlap, and points of contrast. Where the use of different classifiers within a system reveals
different shades of meaning, the juxtaposition of numeral and possessive classifiers may
extend the process further. The classifiers in Truquese thus at the same time provide a
means for ordering the universe, and a method for structuring concepts without multiplying
vocabulary.' Classifiers may have additional functions. In Palikur, verbal classifiers are used
depending on the completeness of involvement of a referent (see §6.4.1, and see also
Aikhenvald and Green 1998).
 Semantic Organization and Functions 321
in the discourse, its topic continuity, or the specificity of its reference,
especially if classifiers are not obligatory in a given morphosyntactic
context. Classifiers of different types show certain preferences for different
discourse parameters (see (A-C) below). All classifiers can be used
anaphorically; they help to track participants in discourse (see (D) below).
This is associated with the individuating role of classifiers; they are found
with almost any type.

(A) Discourse functions of noun classes

Agreeement in NOUN CLASSES, and overt noun class marking often depends
on the definiteness, topicality, and givenness of the referent noun in dis-
course. This means that noun class agreement or overt noun class marking
may depend on whether the noun is newly introduced into the text, or
refers to an already known participant. These functions are also known as
'reference management';13 they are explained in (Al-3) below. The choice
of a noun class marker may also correlate with the specificity of the referent
(discussed in (A4)). These correlations are typically found in cases where
the overt marking of noun class, or noun class agreement, is not gramma-
tically obligatory.
(Al) Agreement in noun class often correlates with definiteness or topicality
of the noun Noun class agreement in NPs can depend on the referential
properties of the noun, such as its definiteness. Noun class agreement often
occurs only if the noun is topical, or definite. In Motuna (Papuan) gender
agreement of the predicate-argument type is obligatory only with a topical
subject constituent (see (C) in §2.4.2). In 'Oro Nao (Chapacuran: Everett
and Kern 1997) the object-marking enclitic which distinguishes three
genders is used only if the direct object is definite.
(A2) Overt noun class marking on a noun can signal its definiteness In
Gola (West Atlantic: Westermann 1947: 17; Heine 1982a: 193), class-mark-
ing prefixes and suffixes act similarly to definite articles; an indefinite noun
is unmarked for noun class, e.g. kul 'a tree', ke-kul 'the tree'; gbalia 'a dwarf
antelope', o-gbalia-a 'the dwarf antelope' (note that here noun class is
marked by a combination of a prefix and a suffix). 14
(A3) Overt noun class marking can correlate with the function of a noun in
discourse In the Australian languages Nunggubuyu (Heath 1984: 169-70)
and Warray (Harvey 1987: 53) the presence of a noun class prefix on a
noun is correlated with definiteness or givenness, and its absence indicates

13
For discussion of general mechanisms of reference management and reference tracking,
see Chafe (1994).
14
The correlation between gender marking and specificity, or individuation, is the basis for
tracing gender markers to article-like elements (see Greenberg 1978: 61 ff).
322 Classifiers
focus and foregrounding. In Wardaman noun class prefixes are used when
introducing a new participant and for maintaining reference to the noun
throughout the discourse (Merlan 1983). This is typical for a number of
northern Australian prefixing languages with noun classes; see Merlan
et al. (1997: 85). In Abau, an isolate from the East Sepik of Papua New
Guinea, the choice between masculine and feminine gender may be deter-
mined by the discourse prominence of a noun: a foregrounded noun is
assigned masculine gender and a backgrounded one is assigned feminine
(Lock forthcoming). Overt noun class marking can also signal focusing of
a particular noun-class-related property. In Alamblak (Papuan) overt
gender marking on nouns correlates with focusing a particular shape-
related property of the noun. For example, the use of kun-r 'house' with
a masculine rather than the usual feminine suffix indicates that the house is
unusually long (Bruce 1984: 97).
(A4) Overt noun class marking can depend on specificity of a referent The
presence of an overt marker can correlate with a specific individuated
referent, and its absence with a more generic referent. In Turkana (Eastern
Nilotic), the overt gender prefix on nouns can be omitted from the names
of animals in folk tales when the names are used in a generic sense
(Dimmendaal 1983: 221).
(B) Discourse functions of other classifier types
Among other classifier types, the use of NOUN CLASSIFIERS and of NUMERAL
CLASSIFIERS—provided they are not obligatory—frequently depends on the
definiteness and pragmatic properties of the referent noun: whether it has
just been introduced into the discourse; whether it is topically continuous;
or whether it is PRAGMATICALLY SALIENT—i.e. either is in focus or is impor-
tant in the discourse.
NOUN CLASSIFIERS in Minangkabau have to be used with a specific noun
when a new referent is being introduced (Marnita 1996: 85-6). After the
referent has been introduced in 12.4, it is referred to just with a noun
classifier in 12.5.
12.4. Mak ado buruang merpati
mother have NOUN.CL:BIRD pigeon
datang ka rumah awak
come to house 1PL
'Mother, there is a pigeon coming to our house.' (a child speaking)
12.5. Buruang sia tu garan
NOUN.CL:BIRD who that probably
'Whose bird is that?' (mother answering)
Noun classifiers are often omitted if the referent has already been
 Semantic Organization and Functions 323
established in discourse or is known from the context (Marnita 1996:
84-5).
NOUN CLASSIFIERS in Jacaltec (Craig 1986b) signal thematically salient
NPs. Classifiers typically accompany referential nouns, and non-referential
nouns in attributive functions may be unclassified. In 12.6a a non-
referential noun, a proper name 'Gabriel Mateo', is used as a nominal
predicate without a classifier (note that the classifier naj is used in a
pronominal anaphoric function: see §3.2.4); while in 12.6b the same proper
name used as a referential subject noun takes a classifier (Craig 1986b:
267).
12.6a. Kap Mat s-b'ih
Gabriel Mateo poss-name
naj
NOUN.CL:MALE.NON.KIN
'Gabriel Mateo was his name.'

12.6b. caw can ye naj Kap Mat

very smart is NOUN.CL:MALE.NON.KIN Gabriel Mateo
'Gabriel Mateo is very smart.'
Noun classifiers function similarly to markers of definiteness. A noun is
usually introduced with the indefinite marker (homophonous with the
numeral 'one'), or without a classifier, and noun classifiers are found
with second mentions of it (Craig 1986b: 269-70). In 12.7a, the word
'pigs' is used without a classifier when first mentioned; in 12.7b, at its
next mention, the classifier no? 'animal' is used.
12.7a. xto pax ix k'opo a7o7
went again NOUN.CL:FEMALE.NON.KIN girl give
yet txitam
food.of pigs
'The girl went back to feed pigs.'
12.7b. chin tit pax a7o7 yet no7
I come back give food.of NOUN.CL:ANIMAL
txitam an
pig PL
'I will come back to feed the pigs.'
Classifiers are also used as discourse markers of 'importance': there is a
clear tendency for an indefinite NP with a classifier to refer to an important
participant (Craig 1986b: 272-3). Noun classifiers (which are also used as
numeral classifiers) also function as specificity markers in Dulong-
Rawang, a Tibeto-Burman language (Randy LaPolla, p.c.).
324 Classifiers
The use of NUMERAL CLASSIFIERS is often regulated by the status of the
noun in discourse. Classifiers often occur with nouns in NPs for initial
mentions of referents (Japanese: Downing 1986; Malay: Hopper 1986;
Burmese: Becker 1986). Discourse factors which influence the use of
classifiers in Malay are shown in Diagram 12.2 (Hopper 1986: 313-14).
-« MORE CONDUCIVE TO USE OF CLASSIFIER LESS »-
(a) Specific Non-specific
(b) Persistent Not persistent
(c) Presentative Anaphoric
DIAGRAM 12.2. Factors regulating the use of classifiers in Malay

The parameter specific/non-specific refers to the opposition between a

single intended referent as opposed to a generic class of referents. Specific
indefinite nouns are new in discourse, and they are usually accompanied by
a classifier; indefinite nouns which are not specific refer to a class of entities,
a role or a function, and they are left unclassified (Hopper 1986: 314).
Persistence in discourse relates to the topical continuity of a noun in dis-
course, and its importance; unclassified nouns in Malay discourse tend to be
mentioned less than those accompanied by classifiers. This is linked to the
'presentative' function of a noun: classifiers are typically used to accompany
a newly introduced noun 'which is being presented for deployment' (Hopper
1986: 320) as subjects of predicates such as datang 'come, arrive', kelihatan
'be seen, be sighted', or the existential particle ada 'there is/are'. In contrast,
nouns which are not new in discourse are almost never classified. This
contrasts with the use of noun classifiers as definiteness markers in Jacaltec.
A similar correlation with definiteness and classifier use has been
observed for Mandarin Chinese (Erbaugh 1986: 408); unlike Malay, a
distinction is made between specific classifiers and the general classifier
ge. Specific classifiers typically mark the first mention of a new item; they
occur with indefinite nouns rather than definite ones; 'once reference is
established, subsequent mentions take the general classifier or constructions
where no classifier is required' (Erbaugh 1986: 408). In 12.8 the first
mention of 'bicycle' is accompanied by a specific classifier ('vehicle'); after
that 'bicycle' is used with the general classifier ge (classifiers are underlined).
12.8. cong nei.bian guolai yi-ge xiao hai-zi,
from there over come one.GEN.CL small child
uh, . . . ql, ql, qi.zhe yi-liang jiaotache uh
uh ride ride ride.PROG one-CL:VEHICLE bicycle uh
shi yi.ge hen keai.de xiao.de jiaotache
be one.GEN CL very cute.MoD little.MOD bicycle
'From over there comes a child, uh, ride, ride, riding a-vehicle bicycle,
uh, (it) is a-general classifier very cute little bicycle.'
 Semantic Organization and Functions 325
The obligatory use of a classifier in Malay, or of a specific classifier in
Mandarin, is reminiscent of the role overt noun class markers play in
discourse organization in Wardaman and Nunggubuyu (see (A3) above).15
MULTIPLE CLASSIFIERS in the classifier languages of East and Southeast
Asia are used depending on the discourse prominence of a referent. In
White Hmong, classifiers serve two functions: they increase the precision of
reference and degree of discourse prominence and they individualize the
referent (Daley 1996, Riddle 1989, Bisang 1993). A newly introduced noun
has to be used with a classifier, as in 12.9 (Bisang 1993: 25).
12.9. ua ciav nws pom ib lub
suddenly he see one CL:HOLLOW, ROUND16
nkau npuas
pigpen
'Suddenly, he discovered a pigpen.'
Indefinite nouns with a generic referent typically occur without a
classifier.17 Unlike Malay, a definite NP takes a classifier, as shown in
12.10 (Bisang 1993: 26):
12.10. ces nyob nyob tus poj ntsuag txawm yug
then one day CL:ANIM widow then give.birth
ta ib tug me tub
TAM one CL:ANIM son
'Finally, one day the widow gave birth to a son.' (The widow has
been introduced in the preceding sentence.)
Within a narrative discourse, further occurrences of the same participant
can be left without a classifier. Classifiers may then be used by speakers to
indicate higher salience of one referent relative to the other referents in the
local context. In 12.1 la, teb chaws 'country' (underlined) is used without a
classifier; the same item, teb chaws 'country', appears with a classifier in
12.lib. Here, the classifier is used to emphasize the fact of the speaker's
possible stay in Laos after the war broke out and everyone was leaving for
Thailand (Daley 1996: 105-6; Riddle 1989).

15
This also agrees with the Categoriality Hypothesis developed by Hopper and Thompson
(1984), according to which the prototypical function of a noun is to 'introduce a significant
new participant into the discourse" (Hopper 1986: 323). Since classifiers in Malay come from
nouns, and 'classification is as good an example of purely nominal morphology as it is possible
to have in Malay', this prototypical property of a noun agrees with the concreteness, indivi-
duation, and discourse-persistence of classified nouns.
16
Gloss provided in accordance with Bisang (1993: 31).
17
Compare the absence of classifiers in VO constructions such as nuv ntses 'to catch fish', or
caij nees 'to ride a horse' (Bisang 1993: 26).
326 Classifiers
12.11 a. neeg kuj txhua leej txhua tus tsuas
person then all person all CL:ANIM only
nrhiav kev tawm teb chaws los tsuas
look.for way leave country Laos
tuaj mus rau sab thaib teb
come go to side Thailand
'Everyone looked for a way to leave Laos and go to the Thai side.'
12.lib. Kuv tau peb hnub
tomqab xav hais tias
I get three day
after think that
yog tsis khiav ces
yuav nyob
if not run then
will stay
lub teb chaws los tsuas ntawd
CL:HOLLOW.ROUND Laos there
'Three days after that I thought that if I didn't escape, then I would
live in Laos . . .'
Thus, classifiers are not just definiteness markers.18 Along similar lines,
classifiers correlate with a definite and singularized interpretation of a
referent, and are widely used for referent tracking in Vietnamese (Lobel
forthcoming). In Thai, classifiers in NPs which contain demonstratives
serve to foreground the noun, and tend to be used when 'a contrast or
emphasis is expressed". They can substitute for a noun 'where their pres-
ence indicates that the nominal has been previously mentioned (or is
understood)' (Conklin 1981: 88). In Newari, a numeral classifier can be
repeated on the noun itself 'to bring in emphasis', or 'for rhetorical pur-
poses' (Bhaskararao and Joshi 1985: 22).
DEICTIC CLASSIFIERS often function similarly to definite articles, as is the
case in the Siouan languages Yuchi and Mandan (Barren and Serzisko
1982). In Pilaga (Guaicuruan) the use of deictic classifiers also correlates
with individuation of nouns (Vidal 1997).
Classifiers which combine NUMERAL and DEICTIC uses are often used as
specificity markers in other environments (see Chapter 9). This is the case
in Cantonese (Pacioni 1997; forthcoming).
CLASSIFIERS IN POSSESSIVE CONSTRUCTIONS and LOCATIVE CLASSIFIERS usually
have no discourse functions (see (C) below). VERBAL CLASSIFIERS are often
obligatory, and their use does not then depend on discourse parameters
18
As Riddle (1989) put it (quoted in Bisang 1993: 29): 'the classifier is not a marker of
definiteness like le in French or the in English per se, although its use partially overlaps with
these articles. In each case, the classifier is used, if needed, to clarify reference, or to emphasise
individuation or individual characteristics. In other words, a major function of the use of a
classifier is to mark an NP as referentially salient if this would not be obvious from the context.
This may occur because of some inherent potential ambiguity in the context or because the
speaker wishes to put forward a particular point of view.' Similarly, in Vietnamese classifiers are
used to identify thematically significant referents in the narrative (see Daley 1996: 106-7).
 Semantic Organization and Functions 327
(see (D) below for their use as anaphoric markers). However, in some
multiple classifier systems in Northern Amazonian languages the use of
non-obligatory verbal classifiers may depend on the pragmatic salience and
discourse-persistence of the S/O constitutent. In these cases classifiers are
not used to relate to newly introduced nouns; rather, they are employed to
emphasize an important, or an unusual, participant. For instance, in
Baniwa of Icana (Aikhenvald 1996c) classifiers are used on the predicate
of a relative clause, or of a purposive clause to refer to the O of the
predicate of the main clause if it is definite or focal. The classifier is used
to mark agreement with a definite referent in focus in 12.12. In 12.13 the
referent of the object constituent is indefinite, and the verbal classifier is
not used with the purposive verbal form. (Purposive forms are underlined.)
12.12. wa-tua wa-takha puapua
IpL-go-FUT 1 PL-cut aruma
wa-dzekata-kasu-pa
1 PL-make-PURP-CL:STICK.LIKE
'We shall go and cut aruma (palm tree) to use (it).'
12.13. pe3i ri-uhwa ri-kapa awakada-riku
hawk 3sGNF-sit 3SGNF-see bush-LOC
ri-kapa-ka3u kwaka i-nu-ri i-kahre
3sGNF-See-PURP What INDEF-COME-REL INDEF-to
'The hawk was sitting in the bush in order to look in the direction of
whoever was coming.'19
In Palikur verbal classifiers are used to refer to an S/O argument to
indicate its complete involvement in the action, or state (see §6.4.1). Clas-
sifiers are also used — as a kind of focus marker — if the noun in S/O
function is unusual. The verb 'cook' is rarely used with verbal classifiers
(because it presumably always implies complete involvement of the object;
cooking a little is not cooking). However, the classifier is used in 12.14 in
which the serpent is cooking a person which is an unusual object to cook
(Aikhenvald and Green 1998).
12.14. eg iw-e-gi ay-ta-re nikwe-ni eg
3 F take-coMPt-3M there-DIR-ANA thus-PAUSAL
seated-VB-CAUS water hot-?-DUR.MASC water
a-daha-ni sakah-pjta-e-gi
3N-for-POSS COOk-V.CL:IRREG-COMPL-3M
'She (serpent) took him (man) there [and] put hot water on to cook
him.'
19
A similar principle operates in relative clauses (Aikhenvald 1996c); similar examples for
Tariana are discussed in Aikhenvald (1994a: 427).
328 Classifiers
(C) Obligatory classifier systems and discourse
Obligatory noun categorization devices show many fewer correlations with
discourse properties and the definiteness of nouns than do optional
devices. The variability of classifier assignment may be accounted for by
the discourse properties of a noun. This variability may involve (i) choice
between a specific and a general classifier; (ii) choice between different
specific classifiers; (iii) different agreement choices; and (iv) choice between
different classificatory techniques.
(C1) Choice between a specific and a general classifier In Mandarin
Chinese the choice of a specific or general classifier depends on whether
the noun is newly introduced into discourse, or is already established (see
12.8). An in-depth experimental study of general classifiers tsu and ko in
Japanese (Zubin and Shimojo 1993: 496) showed 'a shift away from
specific classifiers and toward tsu and ko as more focus is placed on
the numeral itself, and 'a pragmatic substitution of a general for a
specific classifier when attention is shifted away from the nature of the
referent'.20
(C2) Choice between different specific classifiers depending on the focused
property In Tariana, a female can be referred to with a numeral, verbal,
or possessed classifier -ita 'human non-feminine' if her femininity does not
have to be focused on, or is clear from the context; otherwise a human
feminine classifier, -ma, is preferred. This is similar to 'variable categoriza-
tion' of nouns, where different classifiers are used to highlight different
aspects of the same referent, to focus on its particular property; this applies
to large classifier systems, and sometimes also to small systems of noun
classes. In Oromo (East Cushitic) some nouns can be assigned masculine
gender instead of their usual feminine gender, and this indicates an unu-
sually big size of the object, e.g. ablee tun (knife this.FEM) 'this knife', ablee
xun (knife this.MASc) 'this (big) knife' (Table 12.4 shows the possibilities of
'reclassifying' one noun, 'river', in Burmese). Variable noun categorization
can have semantic effect, and pragmatic effect.
(C3) Different agreement choices Variable categorization and variable
agreement can result in focusing different properties of a referent. In
Australian languages, concordial superclassing (which means that one
functionally unmarked class is used with modifiers instead of the expected
agreement classes) often depends on the function of the noun in discourse.
In Warray, masculine agreement appears on modifying adjectives when a
comment is being made rather than new information being provided by the

20
For similar phenomena in some multiple classifier systems, see Senft (1996: 239 ff.) and
Aikhenvald (1999a).
 Semantic Organization and Functions 329
modifier (Harvey 1987; Sands 1995: 264). New information is associated
with functionally marked classes.
(C4) Choice between different classificatory techniques In Tariana, a
multiple classifier language, different classifier techniques are used depend-
ing on the status of a noun in discourse. If an inanimate noun is not in
focus, or does not have to be emphasized, a regular classifier will be used
(Aikhenvald 1994a). Otherwise, the noun itself must be used as a 'repeater'
classifier—see 9.73 and 9.74.
(D) Classifiers as anaphoric and participant tracking devices
Classifiers of any type can be used as anaphoric and participant-tracking
devices. NOUN CLASS markers are used as anaphoric elements for participant
tracking in Australian and Papuan languages, e.g. Ungarinjin (Rumsey
1982: 37) (cf. Merlan et al. 1997). In Yimas, with its highly elliptical
discourse, concordial system (i.e. noun class) marking on verbs, is funda-
mental in the tracking of referents in discourse (Foley 1986: 88) (see also
Dixon 1972: 71-2, for anaphoric uses of noun-class marked determiners in
Dyirbal).
NUMERAL CLASSIFIERS are used anaphorically in Japanese, Burmese,
Vietnamese, and Malay (see Downing 1986; 1996; Becker 1986; Daley
1996; Hopper 1986), as well as in Vietnamese, Hmong, and Cantonese
(see Bisang forthcoming; Lobel forthcoming). In Minangkabau numeral
classifiers can be used anaphorically21 without the noun to refer to a
previously mentioned object, or if the referent is clear from the context
(Marnita 1996: 93). If a person is buying bananas, they can take a bunch
and ask for the price (12.15a); 12.15b is the continuation of the dialogue.
12.15a. bara pisang sa-sikek Mak
how.much banana one-NUM.CL:BUNCH 2nd.FEM.HON
'How much does a bunch of banana cost?'

12.15b. agiah duo sikek

give two NUM.CL: BUNCH
'Give (me) two (bunches).'
In Japanese, numeral classifiers can be considered an extremely advanta-
geous mechanism for anaphoric reference, since 'they provide a means of
anaphorically representing a group of referents in an evenhanded way,
without focusing the reference on one as opposed to others of the indivi-
duals involved in the grouping' (Downing 1986: 369). For instance, the
21
A similar usage is frequent in Tzotzil (Mayan), e.g. a question: jay-p'ej alaxa (INTER-
NUM.ct:ROUND orange) 'How many oranges?', and an answer: j-p'ej no'ox (one-NUM.CL:ROUND
only) 'One (round) only' (De Leon 1987: 76).
330 Classifiers
choice of an anaphorically used numeral classifier hu-tari (two-CL:PERSON)
'two people' can be preferred to other ways of referring to two people—e.g.
with a collective noun—in order to 'avoid their destruction as individuals',
or 'avoiding the nuance that the couple is centered round one member or
the other' (Downing 1986: 370).
VERBAL CLASSIFIERS can be used anaphorically to maintain reference to
participants in discourse. In the following example from Imonda (from
Seiler 1985: 220, discussed in Merlan et al. 1997: 96-7), a woman, ago, is
an established participant; she is reintroduced in the first line of 12.16 (she
has not been mentioned for several lines in the narrative), and then referred
to anaphorically with the classificatory prefix fa- in the following line
(Rushforth 1991 exemplifies the anaphoric uses of classificatory verbs in
Mescalero Apache).
12.16. ed-nei ne-n-b ago-iane-m
PX-SOURCE eat-PAST-DUR Woman-NON.PL-GOAL
ainam fa-i-koho fa-eha kse
quickly CL-LiNK-go v.CL-put fuck
'He ate this and then quickly grabbed the woman, laid her down and
fucked her.'
Anaphoric functions of classifiers often go together with their deictic
uses. 12.17, from Tzotzil (De Leon 1987: 54), illustrates the use of numeral
classifiers with anaphoric and deictic functions.22 This example involves
buying candles in a store. The referent, candles, is not overtly mentioned,
but it is being pointed at. Both merchant and customer are looking at a
collection of multicoloured candles hanging on a line; the classifier ch'ix
'longish objects' is used to refer to them.
12.17a. jay ch'ix cha-k'an
INTER NUM.CL:LONGISH 2p-Want
'How many (longish ones) do you want?'
12.17b. ta j-kan ox ch'ix
preposition 1 ERG-want NUM.CL:LONGISH three
'I want three (longish ones).'
Deictic and anaphoric functions are attested with classifiers of other
types.23 NOUN CLASSIFIERS are used in both functions in a number of Mayan
22
Similar examples are found in Minangkabau (Marnita 1996: 94); cf. a conversation in a
tobacconist's shop between a customer and a shopkeeper: bara (ko) salai Pak (how.much
(DEM) one+NUM.CL:FLAT 2nd.MASC.HON) 'How much does this (tobacco leaf) cost, sir?' A
customer uses a classifier without the accompanying noun, pointing at the object. The
demonstrative, ko, is optional.
23
Deixis is usually distinguished from anaphora. As Lyons (1977: 673) puts it, 'anaphora
presupposes that the referent should already have its place in the universe of discourse. Deixis
 Semantic Organization and Functions 331
languages (Jacaltec: Craig 1986a; 1986b; 1986c; or Mam: England 1983).
Example 12.18 shows an anaphoric use of noun classifiers in Jacaltec which
is functionally equivalent to pronominalization. A referent mentioned in
12.18 is subsequently referred to just by a classifier in 12.19 (underlined)
(Craig 1986c: 264). Anaphoric noun classifiers are extensively used in
possessive constructions in Jacaltec; see ex. 3.12 and 3.13, for an illustra-
tion of their syntactic roles (Craig 1986c: 275-6).
12.18. xil naj xuwan no7
Saw NOUN.CL:MALE.NON.KIN John NOUN.CL:ANIMAL
lab'a
snake
'John saw the snake.'
12.19. xil naj no.7
Saw NOUN.CL:MALE.NON.KIN NOUN.CL:ANIMAL
'He saw it.'
DEICTIC CLASSIFIERS may also be used anaphorically (see 7.18, from
Pilaga, and the discussion in §7.3).
Anaphoric uses are widespread in multiple classifier languages. They are
crucial for 'securing coherence in discourse' in Kilivila (Senft 1996: 21). In
12.20a, a noun, 'tataba-board', is introduced without a classifier; then it is
referred to with the classifier ke 'CUWOODEN' in 12.20b, and the overt noun
is ellipsed.
12.20a. a-tatai tataba
Iso-carve tataba. board
'I carve a tataba-board.'
12.20b. tauwau tabalu m-to-si-na
men Tabalu.subclan this-CL:MALE-PL-THIS
ma-ke-na si koni
this-CL:WOODEN-this their sign.of.honour
'These men belonging to the Tabalu-subclan, this wooden one [i.e.
tataba-board] is their sign of honour.'
In Tariana, classifiers can be used anaphorically in two discourse situa-
tions: (a) a participant is mentioned once, in the beginning of the narrative,
and then it is consistently referred to with the help of a classifier; (b) a

does not; indeed deixis is one of the principal means open to us of putting entities into the
universe of discourse so that we can refer to them subsequently.' It has been claimed that
anaphora depends on deixis, and that pronominalization involves both anaphoric and deictic
reference (De Leon 1987: 53-4). Frequent combination of both anaphoric and deictic uses of
classifiers, and their use for pronominalization, confirms this (see further examples in De Leon
1987: 40).
332 Classifiers
participant whose identity is clear from the context is only referred to with
the help of a classifier.
Example 12.21 illustrates a situation of type (a). It is taken from a story
about how two men met an evil spirit on a river where they came to fish; the
evil spirit took off his shirt, and then the whole story evolves round this
shirt which has all the power of the evil spirit itself. The magic shirt
(yarumakasi) is introduced at the beginning of the narrative (12.2la).
The case-marking (-nuku 'topical non-subject': see Aikhenvald 1994b)
indicates that the shirt is going to be the future topic of discourse. Later
on, the shirt is consistently referred to either as ha-ne-maka 'that (distant)
one made of cloth' (12.21b) or as diha-maka 'it-made, of.cloth' (12.21c).
Examples 12.21a-c are consecutive in the text they come from.
12.21a. diha yaru-maka-si-nuku di-sole(. . .)
he thing-CL:CLOTH-NONPOSS-TOP.O 3SGNF-take Off(. . .)
'He [the evil spirit] took off the shirt [lit. thing made of cloth].'

12.21b. ira-mha pa-pe-niki

need-FRES.NON.vis iMP-throw-CMPL
ha-ne-maka-nuku
DEM:INAN-DIST-CL:CLOTH-TOP.O
na-pidana nu-kesini hau piha pi-a
3pl+say-REM.P.I.NFR IsG-friend yes 2SG 2SG-go
pi-pe-niki ha-ne-maka-nuku
2sG-thrOW-CMPL DEM:INAN-DISTAL-CL:CLOTH-TOP.O
ha-ne-maka-naka karuna-naka wa-na
DEM-DIST-CL:CLOTH-EYEW.PRES be.afraid-EYEW.PRES 1PL-OBJ
'It is necessary to throw that [shirt] away, they [the men] said, my
friend, yes, you go and throw that [shirt] away. That [shirt] is
dangerous for us.'
12.21c. haiku-na dhita di-na-tha-pidana
tree-CL:VERT 3SGNF+take SSGNF-OBJ-FRUST-REM.P.INFR
di-ni-thepi di-pe-niki di-na-pidana
3sGNF-make-TO.WATER 3sGNF-thrOW-CMPL 3SGNF-OBJ-REM.P.INFR
diha-maka dhe-kha di-a-hna
he-CL:CLOTH 3sGNF+enter-AWAY 3SGNF-gO-PAUS
'He (one of the men) took a tree-trunk and tried to throw it away,
in vain. It [shirt] came upon the man.'
A situation of type (b) is illustrated with 12.22. The story is about how a
deer and a jaguar decided to live together. It is important that they chose to
live in the same house, one in one room, and the other in the other. Neither
the house, nor the rooms are introduced with the help of full lexemes, since
 Semantic Organization and Functions 333
the meaning is clear from the context. Numeral classifiers are used. The
lexical item panisi 'house' is used only in the last sentence of the text,
12.22b, to summarize what happened to the house.
12.22a. pa-piu-pidana ijiri na-siwa-kaka na-ni
one-CL:TiME-REM.P.INFR animals 3pL-self-REC 3pl-do
pa-dapana-pidana na-ya yawi neri
one-CL:HAB-REM.P.INFR 3pL-live jaguar deer
neri pa-dawa-se yawi pa-dawa
deer one-CL:ROOM-LOC jaguar one-CL:ROOM
na-ya-pidana pa-dapana-ya
3PL-Hve-REM.P.INFR One-CL:HAB-EMPHATIC
'Once the animals decided between themselves to live in one
[house], jaguar and deer, deer in one [room], jaguar in another
[room], they lived in one [house].'
12.22b. ne-pidana naha panisi-nuku na-peni-kha
SO-REM.P.INFR they house-TOP.O 3pL-leave-AWAY
'So they abandoned the house.'
In East and Central Tucano languages a full noun is almost always
omitted from a noun phrase; classifiers are used in anaphoric function
(Barnes 1990: 289). Similar functions are attested for classifiers in other
languages, e.g. Yagua (Payne 1990; Payne and Payne 1990), Munduruku
(Goncalves 1987), Palikur (Aikhenvald and Green 1998), Resigaro (Allin
1975), Bora (Thiesen 1996: 42), and Nambiquara (Lowe 1999). As a result,
classifiers are more frequent in discourse than full nouns (and this may be
one of the reasons why Resigaro borrowed classifiers rather than full nouns
from Bora: see §13.7.1).
(E) Conclusions
All non-obligatory classifier systems have some discourse-pragmatic func-
tions. The use of classifiers correlates with referentiality, specificity, defi-
niteness, topical continuity, and the salience in discourse of the noun.
Generic noun classifiers tend to correlate with definite referents, while
numeral classifiers are often used to introduce a new referent.
Obligatory classifier systems tend to display similar properties, but to a
lesser extent. All classifier types are used as anaphoric pronouns and as
participant tracking devices. Further grammaticalization of classifiers in
these functions leads to their syntacticization as relative clause markers.
The way classifiers can participate in the organization of discourse relates
to their other functions (discussed in §12.2). The discourse-pragmatic and
anaphoric functions of classifiers described in this chapter provide a counter-
argument against the idea of 'redundancy' for noun categorization devices.
334 Classifiers
12.1.4. Applicability of classifiers and default classes
Noun categorization devices differ in their applicability. Noun class sys-
tems usually involve an obligatory choice: every noun in a language must
be assigned to a noun class.24 Classifiers of other types vary from one
language to another. In languages with numeral classifiers, every countable
noun generally has to be classified; there may be a few abstract nouns
which are used without a classifier (Downing 1996). Noun classifiers,
verbal classifiers, relational and possessed classifiers are often used only
with nouns of particular semantic groups. In Athabaskan languages only
nouns with concrete reference are classified using classificatory verbs.
Possessed classifiers in South American languages are used only with nouns
whose referents can be handled in certain culturally significant ways
(consumed, domesticated, planted, etc.).
The applicability of classifiers depends on how productive the system is.
In a fully productive system, each newly introduced object must be classi-
fied in some way or ways. If the system is frozen, new items may be
unclassified. It does not necessarily mean that the system is in decay. Craig
(1986a) reports that in Modern Jacaltec nouns which refer to newly intro-
duced artefacts (e.g. made of nylon or plastic) are not used with a classi-
fier.25 This is a matter in which noun classes behave differently: in a noun
class system, every noun has to be assigned to a class.
The applicability of a classifier system correlates with the function and
semantics of classifiers. If classifiers are bound morphemes, they tend to
form an obligatory system in which every noun in a language has to be
assigned a classifier.26
The lack of a classifier can be considered a special noun categorization
device. Abstract nouns can also be used without a classifier in Japanese
(Downing 1996: 73) and Korean (Lee 1997). In Thai, classifiers are also
omitted in arithmetical statements (e.g. 'three plus three is six'), and in
cases when the numeral does not denote any quantity, e.g. 'page 23', or
'room 23' (Hundius and Kolver 1983: 182-3).
The absence of a classifier may also be due to the lack of salience of a
referent; this appears to be the case with nouns left 'unclassified' in
Vietnamese (Daley 1996: 136). The use of repeaters instead of classifier
affixes in some multiple classifier systems (see (Fl) and (F2) in §9.1) is an
alternate technique used for classifying otherwise unclassifiable nouns and
24
Occasional exceptions are exemplified in §2.3.
The productivity of a system can change historically. Noun classifiers in Jacaltec must
have been more productive at the time of the conquest and colonization than they are now:
noun categorization then absorbed the names of new artefacts made of metal and glass
introduced by Spaniards into the 'rock' class (Craig forthcoming).
26
Even then some nouns may be left unclassified; e.g. in Nambiquara (Ivan Lowe, p.c.)
large mammals are not classified.
 Semantic Organization and Functions 335
for marking the discourse salience of nouns which do have a 'regular'
classifier.
Classifiers in Vietnamese can be omitted if they denote something which
is inherent to the meaning of the enumerated noun, as in may bay (air-
plane) bon (four) dong co (engine) 'four-engined airplane', as opposed to
may bay (airplane) voi (with) bon (four) chiec (classifier) dong co (engine)
Ion (big) 'airplane with four big engines', where the classifier has to be
used. The omission of classifiers also correlates with the lack of referenti-
ality of a noun: if a classifier is omitted a noun cannot be modified (see
further examples in Lobel forthcoming: 37-40).
If noun categorization is based on universal parameters which can be
assigned to any object—e.g. animacy or physical properties—every noun
will be classified. In contrast, if noun categorization is based on culture-
specific features which cannot be assigned to every entity—e.g. function, or
social hierarchy—some nouns will be left unclassified.
The multiple classifier system in Palikur is such an example. The seman-
tics of genders, and of numeral, verbal, and locative classifiers are based on
animacy and/or physical properties (the choice depends on the type of
classifier), and every noun has to be assigned a classifier. In contrast,
possessed classifiers classify objects depending on their function and use
(e.g. plants can be cultivated or eaten); so nouns get classified only if they
refer to objects for which these particular functional uses are appropriate.
Human nouns (except for 'child') do not take any possessed classifier (see
§5.5.1).
If every noun in a language has to be classified, some nouns may turn out
to be semantically incompatible with the categories used for classifier
assignment. To give the inventory of classifiers its systemic structure, a
'general', or 'residue' or 'default' member is employed to encompass other-
wise unclassifiable nouns. This is how neuter gender is used in German, or
the classifier ge is used in Mandarin.
However, the notion of a 'general' classifier appears to be semantically
and functionally complex. At least three distinct phenomena can be dis-
tinguished under the rubric of 'general', or 'default' classifier (Zubin and
Shimojo 1993: 491). A general classifier can be in a 'RESIDUE' (or 'comple-
ment': Zubin and Shimojo 1993) function if it is a remainder category for
referents outside the domains covered by other classifiers. It is in DEFAULT
function if it can be substituted for other classifiers under specialized
pragmatic conditions. The third function is UNSPECIFIED REFERENT FUNCTION;
then a classifier is used to refer to an unknown entity. In addition, general
classifiers often have a core meaning of their own.
Japanese has a few specific classifiers used with abstract nouns—ken
'incident', toori 'method', han 'crime', denwa 'phone call'. (These classifiers
go back to repeaters.) Abstract nouns can also be used with the 'default'
336 Classifiers
classifier tsu, or without any classifier (Downing 1996: 73). Similar tenden-
cies have been observed for many languages with large classifier systems. In
Vietnamese and Burmese, abstract nouns appear without a classifier
(Nguyen 1957: 131-2; Pe 1965: 181), and in Tai languages they tend to
take a classifier of the repeater type, or appear without any classifier
(Conklin 1981: 364). In Palikur, they take the 'vertical' numeral classifier
-t (as a default, or residue term).
Residue classifiers are found in noun categorization systems of most
kinds. 'Neuter' gender often plays this role in gender systems (e.g. in Latin,
or Palikur). In Diyari (Austin 1981) it is masculine gender, and in Kalau
Lagaw Ya (Bani 1987) it is feminine (both languages have just two genders,
masculine and feminine). Most numeral classifier systems have a residue
classifier, e.g. Mandarin ge, Korean kay. In Athabaskan languages a verb
stem which refers to three-dimensional objects is used for reference to
objects to which no other classifier can be applied (e.g. Koyukon, as
described by Landar 1967; Henry and Henry 1965; see Denny 1979a:
99). A residue member is also found in systems of relational or possessed
classifiers, e.g. ah in Ponapean (Rehg 1981: 180), and Apalai -kyry- 'thing'
(Koehn 1994: 42). Pilaga has a residue deictic classifier hen used with mass
nouns with no specific reference, and with otherwise 'unclassifiable nouns'
(e.g. sky, land/earth, moon or sun) (Vidal 1997: 82).
Residue classifiers may also be used in default function. In Baniwa, the
classifier 'round' is used for otherwise unclassifiable items; it can also
substitute for other classifiers if no particular shape property of the item
is focused on. The same residue classifier can be used to refer to unspecified
objects.
Some languages use different classifiers in residue function, and to refer
to an unspecified referent. In Navajo a classificatory verb for three-
dimensional objects is used for otherwise unclassifiable things; and a verb
for flat flexible objects is employed for unspecified referents (Carter 1976).
Residue classifiers can have a core meaning of their own. The residue
numeral classifiers in Ponapean and in Burmese are associated with round-
ness (cf. Denny 1979a: 100). The residue and default classifier ge in
Mandarin derives from a noun meaning 'bamboo stalk'; later it came to
be used with humans and as a general classifier. In Korean, the residue and
default classifier kay means 'small countable inanimates'.27 Japanese has
two residue classifiers, tsu, of Japanese origin, and ko, a Sino-Japanese
term. It was shown by Zubin and Shimojo (1993) that tsu does not have
any meaning of its own, and is used only as a residue and default classifier
21
It can also be used as an alternative classifier for most small countable items (Lee 1997),
e.g. yenphil twu kay (pencil two CL : RESEDUE) 'two pencils'. If it is used with large items, e.g.
boats (instead of classifier chek), or houses (instead of classifier chay), the noun is automati-
cally understood as referring to a toy boat, or a toy house.
 Semantic Organization and Functions 337
for non-animate concepts. In contrast, ko seems to have a core meaning of
its own: it refers to smallish concrete inanimates.28
A residue category may be more complex. In German, all three genders
are used as residue terms depending on the semantic domain. In the 'bev-
erage' domain, the residue category is masculine, within the 'ship' domain it
is feminine, and within the 'building' domain it is neuter (Zubin and
Shimojo 1993: 492). Neuter gender is also used for unspecified referents.
Languages with multiple classifier systems can have several residue
classifiers, one for each function. In Palikur, neuter gender is an 'everything
else' category for inanimate nouns; the numeral classifier for vertical
objects is used as a residue term in this system. And for verbal and locative
classifiers the default classifier is the one which refers to 'irregular shape'.

12.2. Human cognition and classifiers

Languages use noun categorization devices in different ways to map basic
cognitive categories onto their systems. This section discusses how cat-
egories reflected in classifiers correlate with what we know of mechanisms
in terms of which human cognition operates. As Allan (1977: 308) put it:
'That languages should classify entities along similar lines is not surprising
if one takes the view that human perceptions are generally similar, and that
they stimulate a cognitive categorization of the world which is reflected by
linguistic categories and classes.'
In §12.2.1 we consider the perceptual correlates of semantic parameters
used in classifiers. §12.2.2 surveys the cognitive mechanisms at work in
noun categorization.

12.2.1. Perceptual correlates of noun categorization

Noun categorization devices in all languages reflect the same physical
world. We have seen that all languages use a number of properties of
referent nouns for their categorization; these properties are based on their
nature, and function. Nature-based properties relate to animacy and
humanness; for inanimates they are based first and foremost on shape.
Reasons for the importance of vision related parameters in noun categor-
ization have been suggested by Adams and Conklin (1973: 8): 'One of the
most fascinating facts of numeral classification is its dependence on the
visual feature of form. There are no metaphors based on sound, feel, taste,
or smell'; these might be 'less useful because the impressions gained from
28
These classifiers possibly differ in other functions, too. The classifier tsu can serve as a
dummy inanimate slot filler if no particular referent is being denoted (rather like it in English).
338 Classifiers
them are more time based and transitory'. Note also Allan (1977: 308): To
say that a classifier has meaning is to say that it indicates the perceived
characteristics of the entities which it classifies; in other words, classifiers
are the linguistic correlates to perception.'
The evidence for the universality of parameters such as animacy and
dimensionality comes from child language acquisition. In a seminal paper,
Clark (1977) showed that the patterns of overextensions of lexical items by
English-speaking children are based on parameters very similar to the ones
used in classifier systems. These universal natural categories include ani-
macy, shape, size, texture (or material), and function. The most frequent
categories of overextension are ROUND and LONG. (Thus, for instance, the
children's lexical item moot 'moon' is overextended to such round objects as
cakes, round marks on windows, round shapes in books, round postmarks,
letter 'O'; and the children's item tee 'stick' was used for canes, umbrellas,
rulers, and other stick-like objects.) Relative size tends to be less important
than shape. Other properties, like colour, are never used as the basis for
overextensions; neither is this property used in classifier systems. She then
arrives at the conclusion that
both classifier systems and children's over-extensions reflect a basic categorisation
process that goes on first at the non-linguistic level. . . One way that people seem to
organize entities is to group them on the basis of their perceptible properties, with
shape playing a very important role . . . The data from children suggest that some
properties of shape may be more salient than others and thus more likely to be used
in categorisation . . . Within classifier systems, then, one might expect to see a
progression from systems that only distinguish animates from inanimates, to sys-
tems with more and more complex subdivisions using several dimensions at once to
produce a large number of classifier-categories. (Clark 1977: 460-1)29
Children prefer to group basic level objects by perceptual features rather
than by functional features because perceptual features are readily avail-
able; this explains the predominance of perceptual features over functional
ones in categorization via classifiers (Tversky 1986: 72).
Other studies confirm the 'anchoring' of categories encoded in classifiers
in the mentally projected world (Frawley 1992: 134). More specific connec-
tions between various parameters of perception have been established by
the current theory of vision (Marr 1982; Baron 1987; Frawley 1992: 134-5).
Visual processes involve the following sequence of computations: one
dimension > two dimensions (flat surfaces) > two dimensions (curved
surfaces) > three dimensions (normal objects). This explains why more
shape distinctions tend to be made in classifiers which apply to one- and
two-dimensional objects than to three-dimensional ones. One- and two-
29
Also see Rosch (1975a; 1975b), on the importance of shape in human perception. For
some criticism of Clark (1977), see Senft (1996: 13).
 Semantic Organization and Functions 339
dimensionality are perceived first, before three-dimensionality. The promi-
nence of curvature in visual perception also explains why curved shape
appears to be important in classifier semantics.
Salient physical properties encoded in classifiers are integrated into the
domain of physical interaction of humans with their environment, and this
is where functional properties come in. Functional properties reflected in
classifier systems correlate with social interaction and socio-cultural envir-
onment (§12.3.2).30

12.2.2. Cognitive mechanisms and noun categorization

What cognitive mechanisms underlie human categorization of the world
encoded in classifiers of different morphosyntactic types? The mechanism
of human cognition reflected in noun categorization devices is from basic
level to subordinate level of categorization.
Classic work by Rosch and other psychologists has confirmed the exis-
tence of a basic level of categorization: 'In taxonomies of concrete objects,
there is one level of abstraction at which the most basic category cuts are
made. Basic categories are those which carry the most information, possess
the highest category cue validity, and are, thus, the most differentiated
from one another' (Rosch et al. 1976). The cognitive importance of basic
level categories lies in their predictive power, due to clustering of mutually
independent properties of entities. Lee (1988: 232) exemplifies this with the
following:
It is a fact about the world that animals which have wings are almost invariably
birds and have other properties of birds (feathers, two short legs, beak, etc.). In this
sense, the property 'has wings' has a high 'cue validity', that is, it is a good predictor
of other properties. . . . Because of these . . . correlations, we need only identify one
of these properties when we want to know what kind of animal we are dealing with.
Bird, for the urban English speaker, is therefore a basic level category.
Basic-level categories also show a high degree of internal coherence, and
their members share many more properties with each other than with
members of other categories. Generic-specific relations in noun categoriza-
tion systems are the result of setting up classes of objects, or persons, on
the principle of basic-level categorization. Generic classifiers (like the ones
found in Mayan, Austronesian, or Australian languages) are the direct
correlates of the basic-specific (or subordinate) categorization level.
The basic level of categorization is associated with salient properties of
objects, such as shape and other physical properties (e.g. consistency) via
extension of classes to new nouns. Since 'cognitively salient properties tend
30
For further issues concerning the psycholinguistic reality of noun categorization in
cognition, see Carroll and Casagrande (1958) and the discussion in Lucy (1992b: 201-7).
340 Classifiers
to be those with high cue validity' (Lee 1988: 236), it is natural to suggest
that initial members of classes serve as prototypes for further extensions
based on these properties. 'Shape' is generally considered the most impor-
tant of these properties, since 'the function of an object may be unknown,
or variable over time' (Erbaugh 1984).31 However, function extensions may
have higher cue validity than shape extensions, and this is what happens
with respect to such domains as human categorization where social status
is a kind of functional categorization. One expects more functional exten-
sions in the realm of possessive constructions which are more directly
linked to handling of objects and to the ways objects relate to their
possessor. This is indeed the case. What functional categories have most
validity in a system is necessarily linked to the ways categories are con-
ventionalized and formed through socio-cultural environment.32

12.3. Social and cultural issues in noun categorization

Universal semantic features encoded in noun categorization devices are the
reflection of perceptual and cognitive mechanisms shared by humans.
Different parameters are exploited to different extents; and the specific
choice of parameters is related to cultural, environmental, and social vari-
ables. These variables are employed to different extents in different noun
categorization devices. They often play a role in the way semantic exten-
sions operate in languages; and they are frequently used to explain the
choice of semantically opaque classifiers.
At first sight, the semantic organization of noun categorization devices is
a confirmation of a version of the Sapir-Whorf hypothesis—that 'through
the analysis of a language one can show the viewpoint of a people' (Denny
1979a: 117, n. 1). Indeed, of all nominal and verbal grammatical categories,
classifiers are the easiest to immediately connect with extralinguistic
phenomena—either of physical environment, or of culture.33 For instance,
31
According to Rosch (1975a), colour is not predictive of other attributes, and is thus a
relatively useless attribute for categorization of objects.
32
Cf. Rosch (1975a: 28): '[when] we speak of the formation of categories we mean their
formation in the culture.'
33
Attempts have been made to provide extralinguistic motivation for other categories. For
example, numerous failed attempts have been made to relate ergativity to the psychology of
people (Dixon 1994: 214-15). Spatial distinctions may obviously connect to physical environ-
ment: a language is likely to distinguish 'up hill' and 'down hill' in a deictic system only if it is
spoken in mountaneous regions, as are Caucasian languages. If a language distinguishes
alienable and inalienable possession, the attribution of an object to the inalienable class
may depend on how valuable it is, e.g. houses in North Arawak culture. However, even if
there are correlations between these grammatical categories and some extralinguistic para-
meters, more studies are needed to investigate whether they will change in the case of cultural
change.
 Semantic Organization and Functions 341
if a language has a classifier for 'domesticated animals', people must be
familiar with domestication of animals, and, consequently, are unlikely to
be nomadic hunters and gatherers; and one would not expect people in a
desert to have a classifier for 'canoes'. Moreover, it will be shown that a
cultural change (e.g. introduction of new artefacts, or rapid social changes)
often results in a change—or in restructuring—of a classifier system.
We saw in §2.4.3 that variable noun class, or gender assignment may
reflect metaphorical extensions which create a jocular effect. In Machi-
guenga (Peruvian Arawak: Shepard 1997), a language with masculine and
non-masculine noun classes and a large set of numeral and verbal classi-
fiers, non-masculine noun class (which covers females and inanimates) can
be used by men in jocular reference to a 'third man', as a 'part of typical
male joking behaviour in which men feminise one another with sexual, and
especially homosexual, comments and jokes' (Shepard 1997: 53).
In establishing correlations between linguistic and non-linguistic para-
meters one should avoid circularity. Does Palikur have an elaborate system
of numeral classifiers based on shape and dimensionality because it has an
elaborate system of ethnogeometrical distinctions, or is the explanation the
other way round? The danger of circular explanation has been pointed out
by Craig (1986c: 290):
It is difficult to avoid a certain amount of circularity when trying to argue that the
set of classifiers of social interaction reflects the social and religious organization of
the Jacaltec culture. Although one may find ethnographic evidence based on obser-
vation of the social organization which is relatively independent of language,
whatever is known of the organization of the Jacaltec pantheon is based on
language data . . . The existence of ample language-independent ethnographic
evidence of the material culture will greatly lessen the problem of circularity
encountered here.34
In §12.3.1 we show how sex (also known as 'gender') roles and social
structure can be reflected in noun categorization devices relating to
humans. The relevance of physical environment and of various cultural
parameters for noun categorization of inanimates is discussed in §12.3.2.
The culture-specific character of metaphorical extensions is dealt with in
§12.3.3. Evidence from language planning and language change, especially
in the context of language obsolescence, provides further support for a
close link between noun categorization and extralinguistic parameters
(§12.3.4). Not all semantic parameters employed in noun categorization
depend equally on extralinguistic parameters. This, and the problem of the
predictability of social and cultural parameters employed in noun categor-
ization, are considered in §12.4.
a34
An interesting discussion of the importance of sex (or natural gender) distinctions in
Turkish—which has no gender or classifiers—can be found in Braun (forthcoming a and b).
342 Classifiers
12.3.1. Social structure in noun categorization
Hierarchical social structure and kinship relations are frequently reflected
in a system of numeral and noun classifiers relating to humans (see §11.2.2,
11.2.3). Austroasiatic languages typically have numeral classifiers for
deities, humans of high status (e.g. governors), and monks (Adams 1989:
59-61).
Noun classifiers used with humans in Mayan languages reflect 'the
organization of the powers that be ... in the midst of, and beyond the
world of humans' (Craig 1986b: 272). Social and kinship status can be
encoded in relational classifiers, as is the case for Ponapean (see (A) in
§11.2.4).
In a recent study on possessive classifiers in Ponapean, Keating (1997)
showed how social status of the speaker and of the addressee is reflected in
the choice of a possessive classifier in Ponapean. Possessive classifiers used
in honorific speech register differ semantically from those used in the
common register. For instance, the common register distinguishes the
classifiers kene 'edible things' and nime 'drinkable things'; in the honorific
register (used when addressing a chief or speaking in the chief's presence)
three classifiers—koanoat, pwenieu, and sak—refer to all comestibles
(food and drink); they distinguish the rank of possessor: paramount chief,
the paramount chieftess, and the secondary chief, respectively (Keating
1997: 262). Thus, if one is invited to share a chief's food, this share would
be referred to with a classifier corresponding to the status of the owner.
Keating (1997: 262) illustrates this in the following way: 'a plate of food
sent to me by the paramount chieftess, as I stood by the video camera
filming a feast, was announced to the gathering as Elizabet, kepin pwenieul'
(lit. portion Poss.CL:paramount chieftess). In contrast, 'humiliative' or
status lowering speech is characterized by neutralization of all the semantic
oppositions found in common speech (see §10.5). One would not expect to
find classifiers related to social hierarchies in egalitarian societies, e.g. in
Australian languages; and indeed, they are absent there.
Social status and the power associated with it may be encoded in differ-
ent ways. Various attempts have been made to establish a link between the
semantics of noun classes and its cultural motivation in Niger-Congo
languages. Osam (1994: 127-32) showed how Akan ontology is reflected
in the semantic organization of its noun class system. The hierarchical
ordering of the Akan structure of being reflects a 'decreasing order of
power where one entity is of higher power than the one below' (Osam
1994: 127). The issue of power is central to Akan ontology; and the Proto-
Akan noun categorization system of animates apparently also involves a
hierarchy of social interaction, from the physical sphere to spiritual. Class
1 contains nouns which refer to powerful beings. These beings may have
 Semantic Organization and Functions 343
spiritual power, e.g. 'devil', 'god', or more culturally specific persons, e.g.
'master drummer' (because of his ability to play and invoke the spirits).
Some also have social power, e.g. 'priest', 'elder', and a number of names
for professions. Animals which are in Class 1 can also be considered
powerful, e.g. 'elephant', or 'eagle' (who is 'the king of the air': Osam
1994: 131). Class 2 contains humans, animals, instruments, and other
miscellaneous items. Humans which belong to this class have no power,
either physical or social (e.g. 'child', 'slave', 'servant', 'orphan'). Animals in
this class are less dangerous, or less powerful.
Palmer and Arin (1995a; 1995b) investigated the semantic basis of Shona
noun classes from the point of view of correlations with Shona culture.
Classes 1/2, 5/6, and 9/10 consist of participants, instruments, and distinc-
tions which are intrinsically linked to scenarios of possession by ancestral
spirits and associated rituals. Thus, persons and ancestors belong to Class
1/2; at least some central participants which fit into the schema of ritual
danger, and mediums of ancestral chiefs, belong to 5/6. The central parti-
cipants of Class 9/10 can be described as governed by a scenario of ritual
protection, dominance and fertility. There is an abstract semantic opposi-
tion of Class 5/6 to Class 9/10: 'danger and evil versus protective dom-
inance' (Palmer and Arin 1995a: 25). All the elements in these classes relate
to central participants and instruments by similarity, metaphor, or meto-
nymy (see §12.3.3).

12.3.2. Environment and culture in noun categorization

Functional parameters encoded in noun categorization devices are directly
related to the culture of the speakers of a language. Consequently, an
elaborate system of relational classifiers presupposes certain kinds of activ-
ities. Speakers of languages which have relational classifiers for vehicles are
bound to have vehicles; speakers of languages which have classifiers for
'fruit to be picked', as does Cahuilla, must undertake this sort of activity.
In Cahuilla, 'temporary' classifiers describe the way in which an item can
be handled, e.g. the ways in which food can be cooked. These classifiers also
correlate with the semantics of the possessed nouns. For instance, only certain
fruits or blossoms can be picked from trees; they are classifed with ?ay?a
'classifier: picked items'. This classifier cannot be used with such objects as
the acorn (which is employed with another classifier, ci?a, used for edible
items after they have fallen off the tree (or plant) and when they are dry and
picked from the ground—Seiler 1977: 302). One classifier, roughly translated
as 'partner', is used with a limited subclass of animals (which refer to a
totemic moiety) (Seiler 1977: 305-6). The use of relational classifiers depends
on socio-cultural conventions. In Cahuilla, certain animals, e.g. 'feral cat',
cannot be pets, due to socio-cultural restrictions (Seiler 1977: 306).
344 Classifiers
Typical functions appear important in the choice of unique or specific
classifiers which relate to the function or nature of objects. It is no wonder
North Amazonian languages spoken by people who live along big rivers
have a special classifier for 'canoes'. That longhouses are divided into
culturally important compartments, or 'rooms', also gets expressed
through the classifier system; see, for instance, the classifiers in Baniwa
in Table 9.5. The existence of a special class used to refer to cows in
Fulfulde (see Breedveld 1995: 69) can be considered a direct consequence
of the importance of cattle in the culture.
Typical functions and objects which relate to them can be found in
numeral classifier systems. Japanese and Korean have classifiers for books
and written documents, while vernacular Amazonian languages do not
have such classifiers, since they lacked these artefacts.
An overall study of classifiers in Jacaltec, and Jacaltec culture, by Craig
(1986c: 287) showed that this 'linguistic system overtly marked the main
features of the culture at some point back in time', and that it 'encompasses
all aspects of traditional Jacaltec life'. The classifiers are 'few enough to
isolate very selectively certain objects of the Jacaltec culture (corn but not
beans, weaving but not carpentry), and at the same time, they are in enough
number to produce together a very realistic picture of the Jacaltec culture
as a whole (human beings and powers of nature, men's work, women's
work, and how basic subsistence needs are met)' (p. 287).
Some semantic oppositions in classificatory verbs in Athabaskan lan-
guages relate to culturally important notions. Classificatory verbs in Slave
distinguish two manners in which different objects can be handled: one is
'careful, controlled, respectful, polite, humble, reserved, gentle, concerned
with human behavior', and the other implies 'lack of care, reservation and
control', but 'not negative in connotation'. As Rice (1989: 784) points out,
this distinction in manner of handling between the two sets of verbs is
'relevant throughout the entire culture'.
Possessive classifiers in Ponapean show an interesting correlation with
traditional beliefs and with the kinship system, more precisely, with a
culturally important notion of mana which describes 'the sacred and
dangerous power which flows from the deities through the chiefs to the
people'; 'mana flows matrilineally to descendants within chiefly clans',
and consequently the 'belief that mana extends to possessions makes
possessive constructions a meaningful category' in distinguishing honori-
fic, status-lowering, and common speech registers (Keating 1997: 249).35
Consequently, in Ponapean maternal and not paternal relatives have
specific classifiers (e.g. ullap 'paternal uncle', wahwah 'man's sister's child')

35
We have seen in §10.5 how Ponapean employs different systems of possessive classifiers
depending on the speech style.
 Semantic Organization and Functions 345
(Keating 1997: 253). That the honorific general classifier, sapwellime, is
composed of sapwe 'land' and lime 'hand, arm' can be explained by a
strong cultural link between high status and land ownership; in contrast,
the all-purpose possessive classifier in humiliative speech, tungoal, means
'food, eating and this correlates with the link between low status and food,
or nourishment as the product of the land' (Keating 1997: 264-5).
Function is rarely encoded in noun class systems (see (C) in §11.2.1).
This is why one hardly ever finds correlations between functional aspects of
material and spiritual culture and the semantics of noun classes assigned to
inanimates. Australian languages, with a noun class of 'non-flesh food', are
a notable exception to this.36
The predictive power of these correlations is, however, rather limited. It
is quite understandable that classifier languages spoken in hunter-and-
gatherer societies (like in Aboriginal Australia) with their heavy reliance
on vegetable food growing in the bush, would have a special classifier for
non-flesh food. However, classifier languages such as Daw or Nambiquara
spoken within other hunter-and-gatherer societies in other parts of the
world do not have such a class (see Dixon and Aikhenvald 1999).
Correlations between physical properties encoded in classifier systems
and physical environment or other cultural parameters are much more
diffucult to establish. An attempt to establish such correlations was under-
taken by Denny (1979a: 108, 112-15). He explains the existence of deictic
classifiers in Eskimo and Toba which are portmanteau with visible and
non-visible members—he calls this a 'distal' style—by the fact that they
'hunt in open treeless environment';37 and 'the Athabaskans and Algon-
quins who hunt in closed forest environment, employ a proximal style in
which classification is embedded in verbs of handling, proximal variables
such as hardness and flexibility occur, and the extendedness variable has
proximal values' (p. 108). The presence of interioricity (holes vs. rings) as a
variable in noun categorization is said to relate to the way land used to be
settled, the presence of fixed domicile, and the division of landscape into
privately owned areas. The distinction between rigid and flexible gets
encoded in classifier systems due to technology systems. Thus, because of
different cultures and dwelling types of societies, the physical properties—
equally accessible to all—get encoded in one system but not in the other.
Attractive and convincing as this may seem, there is a danger of circularity
(especially since not all peoples who live in similar environments have
exactly the same classifier systems).
36
We will see in Chapter 13 that this noun class developed as the result of grammaticaliza-
tion of noun classifiers, where 'function' is a frequently employed semantic parameter.
37
However, this hypothesis can easily be falsified. Dyirbal, traditionally spoken in a thick
rainforest, distinguishes visible and invisible deictics, while coastal Yidiny does not (see Dixon
1977: 181).
346 Classifiers
12.3.3. Culture-specific metaphorical extensions
Metaphoric extensions of sex- and animacy-based noun classes and classi-
fiers to inanimates are often linked to the socio-cultural stereotypes asso-
ciated with sex. In a fascinating study of sex roles as revealed through gender
reference, Mathiot (1979) showed how role images of males and females are
realized in the use of personal pronouns. The use of the pronouns he and she
observed with inanimate referents in American English was found to cor-
relate with a number of stereotyped features—part of the inherent image
and role image American men and women have of themselves, and of each
other. The semantic opposition BEAUTIFUL/UGLY manifests men's conception
of women's vs. men's appearance; and the semantic opposition manifesting
men's conception of women's and men's achievement potential is INCOMPE-
TENT/COMPETENT. Thus a beautiful flower is referred to as 'she', and an ugly
cactus as 'he' (Mathiot 1979: 18-19). In contrast, the inherent image and
role image American women have of themselves and of men can be for-
mulated in one semantic opposition: MATURE/INFANTILE (Mathiot 1979: 25).
Regular polysemy of feminine and diminutive in Afroasiatic languages is
often accounted for by the low and subdued status of women in traditional
Afroasiatic speaking societies (Diakonoff 1988). This is another example of
a correlation between gender and social status.
In a few New Guinea languages with masculine and feminine genders,
masculine is associated with culturally important roles, and feminine with
insignificant things. This is the case in Angave, an Angan language (Speece
n.d.: 1ll), and in Abu' Arapesh (Otto Nekitel, p.c.). Robert Conrad (p.c.)
reports that in Felefita, an Arapesh language (Torricelli phylum), mascu-
line can replace any other gender provided the object is sufficiently impor-
tant. In coastal Arapesh languages, this results in the massive expansion of
masculine gender, which has become the unmarked one.
We saw under (B) in §12.1.2 how metaphorical extensions based on the
linguistic map of the world in Burmese culture explain why rivers and
oceans are put into the same class as arrows and needles. Along similar
lines, the complicated semantic structure of the nge-cla.ss—the specific class
for cows—in the Maasina dialect of Fulfulde can be explained by Diagram
12.3, based on two metaphorical extensions: from cows to sources of light,
and from cows to ceremonies (Breedveld 1995: 70-1).
Sources of light (fire, light, sun) Ceremony

Which influence the Occasion of transfer of

movement of cows property rights to cows

Cows
DIAGRAM 12.3. Semantic network of the nge-class in Maasina Fulfulde
 Semantic Organization and Functions 347
Semantic extension principles, such as the Myth-and-Belief principle and
Important Property principles in Dyirbal (see Dixon 1982), are based on
cultural intricacies. Once they become obsolescent, the classifier assign-
ment becomes opaque.

12.3.4. Socio-cultural motivations for change in noun categorization

Further evidence for the socio-cultural motivation of noun categorization
devices comes from language change, language obsolescence, and language
planning. Noun categorization devices are probably the only grammatical
category which directly reflects these social phenomena and shifts; they are
freely reflected in the lexicon.
These processes affect classifiers which relate to the function of objects,
and also classifier assignment based on cultural knowledge.
(A) Change in noun categorization can be motivated by social changes
Social changes which affect the social organization of a society may affect
the set of classifiers related to human categorization. Pre-revolutionary
Khmer had a complicated system of classifiers for humans and deities,
typical for an Austroasiatic language. This system had terms for monks, the
royalty, the image of Buddha; people were divided into 'high persons',
'dignitaries', 'superior persons', and 'inferior persons' (Adams 1989: 63).
The Khmer Rouge revolution affected the social order, terms of address,
and honorific classifiers.38 (Note that after the defeat of the Khmer Rouge
the classifier system was at least partly restored: Tony Diller, p.c.) This is
very similar to the way the lexicon can get affected by social changes (e.g.
Comrie and Stone 1977; Selischev 1928).
Social changes may affect principles of assignment of noun class, or
gender—at least to some extent. Russian has a number of hybrid nouns
for which gender agreement is determined by the sex of the referent, and
not by their form (see §2.3.2). Rothstein (1973: 464) reports that the
number of 'hybrid' nouns referring to professions in Russian increased
with the liberation of women after the 1917 revolution. More professions
became open to women, and consequently, nouns like vrac 'doctor (mascu-
line)', or sudja 'judge', began to allow variable agreement according to sex.
Noun class assignment can change with a shift in social attitudes. Lak
(Northeast Caucasian) has four noun classes. Class 1 consists of human
males, Class 2 of human females, Class 3 of other animates and some
inanimates, and Class 4 of inanimates only. According to Khaidakov
38
As Shawcross (1979: 376) puts it, in post-revolutionary Cambodia, 'all forms of address
that betoken social or family relationships were abolished and names were simplified. Father,
mother, doctor were all replaced by comrade.'
348 Classifiers
(1963: 49-50), modern etiquette requires that in addressing women outside
one's own family the markers of Class 3 be employed (no further details are
given).
Changes and variation in noun categorization can be motivated by
socio-cultural factors. Wolof (West Atlantic: Irvine 1978) has eight singular
and two plural classes. Noun class assignment is based partly on semantics
and partly on phonological principles. The society is highly stratified.
Verbal fluency, correctness, and elaboration are associated with low social
rank, for high ranks 'incorrect' speech is the norm. Deliberate 'errors' in
noun classs assignment are associated with high rank of the speaker (Irvine
1978: 41-3). One class (hi- class) is expanding at the expense of other
classes among middle-aged noblemen for whom 'incorrect' speech is an
indicator of their social status, and this change in a prestige group is now
being imitated by lower classes (Irvine 1978: 60-1; also see discussion in
McLaughlin 1997).
The composition of a classifier can change due to technological innova-
tions. In Thai, khan, a classifier for objects with handles, once applied to
bicycles, was later extended to all vehicles, according to their similarity of
function (see §13.9.2).
(B) Set of classifiers, or classifier assignment, can be affected by language
obsolescence
Numerous examples show how cultural obsolescence leads to the reduction
of largeish classifier systems, e.g. numeral or relational classifiers.
Cahuilla (Uto-Aztecan: Seiler 1977: 306) had a special possessive classi-
fier for 'moiety animals'. In the early days Cahuilla society was divided into
two moieties, one associated with the Coyote, and the other with the
Wildcat. The two moieties were in an exogamic relationship. However,
the obsolescence of the moiety system led to the loss of appropriate
classifiers.
A drastic reduction in the system of numeral classifiers has been observed
for Minangkabau (Marnita 1996: 163-5). This reduction goes together with
a narrowing of the spheres in which the language is used, under the pressure
of Indonesian, the national language. The increase in use of loan measure
terms (metre, litre, and kilogram) contributed to the loss of traditional
mensural classifiers. Young people appear to be unfamiliar with some
culture specific classifiers, e.g. sumpik 'blowing bamboo weapon', simply
because they do not use the object. Obsolescence of traditional practices
among young city-dwellers has resulted in the loss of appropriate classifiers,
e.g. cabiek 'piece of betel leaf. Younger speakers of Korean and Japanese,
especially those born overseas and not exposed to traditional culture, tend
to know relatively few specific classifiers (Lee 1997; Sanches 1977; Masa
Onishi, p.c.). This is comparable to the loss of lexical knowledge.
 Semantic Organization and Functions 349
The principles of noun class assignment can be affected by cultural
obsolescence.39 Schmidt (1985: 156-7) describes a series of changes in
rules for noun class assignment in Young People's Dyirbal. Mythological
association as a basis for class membership is lost. In traditional times
birds were believed to be spirits of dead human females, and consequently
assigned to Class II, 'feminine'. With the loss of this belief, speakers of
Young People's Dyirbal treat birds as members of the 'animate' Class 1.40
(C) Language planning can affect the use of classifiers
Thai has a number of rules concerning the use of classifiers in the 'royal
linguistic register'.41 Juntanamalaga (1988: 319) gives an example of an
order of King Mongkut issued in 1854 whereby 'noble' animals such as
elephants and horses should be counted without any classifier; the classifier
tua could only be used for animals of a 'lower' status. Royal vocabulary
replaced the classifier kha 'egg'—which developed associations with
testes—with/oVng 'water bubbles' (Juntanamalaga 1988: 320-1).
In Setswana (Bantu) all human beings and many names of tribes and
peoples belong to the human Class 1/2. A number of ethnonyms which
refer to strange and unusual groups such as Chinese or Bushmen belong to
Class 5/6, together with substances, e.g. dirt or clay, and abstract nouns,
e.g. foreign institutions (Anderson and Janson 1997: 34-5; Joe Tsonope,
p.c.). In modern prescriptive Setswana it is, however, not recommended to
use this class to refer to Bushmen (human class should be used then), since
it is felt that they are being treated 'like dirt' when referred to with a non-
human class (Joe Tsonope, p.c.).
Until recently, the masculine pronoun he in English was used as a generic
term for human reference and also as one subordinate term, for male
reference (see the discussion in Alpher 1987). During recent years, it has
become the norm that the generic unmarked pronoun 'they' be used (to
avoid linguistic chauvinism). This is illustrated in Diagram 12.4.

39
Though the number of closed noun classes does not necessarily get reduced; see §13.7.3
on language obsolescence and reduction of noun class and gender systems.
40
Other changes may be due to overall simplification of the system to make it more similar
to gender in English. This may explain why concept association is also abandoned (Schmidt
1985: 157). Traditional Dyirbal assigned yarra 'fishing line' and barrban 'fish spear' to Class I,
by association with_jahu 'fish'; speakers of Young People's Dyirbal place these two words into
Class IV, with other inanimates.
41
The use of classifiers in Thai is regulated by stylistic rules. Omission of classifiers is
characteristic for informal Thai (Juntanamalaga 1988: 316). The choice of a particular classi-
fier can depend on speech register. Classifiers an (a general inanimate classifier) and bay (small
leaf-like or roundish objects) are replaced by repeater constructions (Juntanamalaga 1988:
320). The use of classifiers in Thai is also directly linked to speech styles. Classifiers are widely
used anaphorically by common people, and also as a sign of familiarity with each other; but
not in 'royal' speech styles. Some classifiers, such as tua, are totally unacceptable when
speaking to members of the royal family (Tony Diller, p.c.).
350 Classifiers
Traditional English usage Modern usage

he (referent unspecified for gender) they (unspecified)

he (masculine) she (feminine) he (masculine) she (feminine)

DIAGRAM 12.4. Gender pronouns in former and contemporary prescribed English
usage

12A. Conclusions

We have seen that, semantically, classifiers are heterogeneous, non-

hierarchically organized systems which employ both universal and culture-
specific parameters. The ways these parameters work are conditioned and
restricted by cognitive mechanisms and the socio-cultural environment.
Among the universal parameters are animacy, humanness, and physical
properties, e.g. shape, dimensionality, consistency. Culture-specific param-
eters can cover certain functional properties and social organization.
Classifiers are the grammatical means languages use for structuring and
organizing concepts without multiplying lexical vocabulary. This organiza-
tion is founded on basic-level categorization which involves perceptually
salient parameters also important for physical, functional, and social
interaction of human beings with their environment. Not all noun categor-
ization systems, and not all semantic parameters, correlate equally with
non-linguistic factors. Large systems of classifiers show more dependency
on non-linguistic parameters than do grammaticalized noun classes.
Classificatory parameters associated with function rather than physical
properties are more sensitive to cultural and other non-linguistic factors.
Human categorization, as a sort of 'social' function, depends entirely on
social structure. Functional categorization of inanimate and non-human
objects is directly related to cultural notions. Animacy and sex, when
extended metaphorically, are influenced by social stereotypes and beliefs.
Correlations between the choice of physical properties encoded in
classifiers and non-linguistic parameters are much less obvious. They
may relate to the cultural salience of certain shapes or forms; and they
may be ultimately based on typical metaphorical extensions.
A further problem with finding cultural correlates for noun categoriza-
tion is their inductive character. In quite a few cases we can explain what
social, cultural, or even environmental parameter a classifier correlates
with in a given society. But we will never be able to predict the ways in
which non-linguistic parameters would be reflected in the grammar of a
language. The example of the lack of vegetable food class outside Australia
 Semantic Organization and Functions 351
was mentioned above. Another example along similar lines is human
categorization in Japanese (Downing 1996: 157). One would expect Japa-
nese, a language with a well-developed system of honorifics, to have a well-
developed system of classifiers reflecting different social statuses—just like
Korean. However, unlike Korean, Japanese has just one honorific classifier
(mei).
 13 Origin and Development of Noun
Categorization Devices

This chapter deals with the origin, evolution, and decay of classifiers of
different kinds which may have different language-internal—lexical or
grammatical—sources and may involve different grammaticalization
paths. I also discuss the paths of evolution and restructuring of noun
categorization devices, and the development and loss of agreement on
different targets. Language-internal and language-external motivation for
the emergence, restructuring and decay of classifiers are considered next.
Further on, I analyse the semantic changes within classifier systems which
take place in the course of grammaticalization—from a lexical source to a
classifier—as well as within an already established classifier system.
Appendix 2 provides further examples of semantic changes in classifiers.
Different kinds of classifiers tend to have different language-internal
sources—some have lexical sources (see §13.1), while others come from
closed grammatical classes; for instance, a demonstrative can become a
gender agreement affix, or an adposition can become a locative classifier
(see §13.2).
The problem of the relative 'age' of classifier systems in a given language,
with a special focus on languages with several classifier types and on the
relative age of noun categorization devices, is discussed in §13.3.
In the course of the history of a language, one noun categorization device
can develop into another; the principles underlying this internal evolution
of noun categorization devices are discussed in §13.4. Grammaticalization
and reanalysis in the evolution of noun categorization devices are consid-
ered in §13.5. Processes of reduction and loss of noun categorization are
analysed in §13.6. Language-external motivations for the emergence,
restructuring, and decay of noun categorization are considered in §13.7.
The genesis, evolution, and decay of agreement are dealt with in §13.8. The
development and internal evolution of noun categorization devices go
together with semantic changes which occur in the passage from lexical
sources to classifiers or within an already established system; these issues
are considered in §13.9. Finally, §13.10 presents a set of conclusions.1

' I am fully aware that prehistories of many of the language families considered here—
especially of those without a documented history and a fully reconstructed prehistory—are far
from clear.
 Development of Noun Categorization 353
13.1. Lexical sources for classifiers

Classifiers often come directly from open classes of lexical items, or from a
subclass of an open class (Craig forthcoming: 61). This involves gramma-
ticalization—from a lexical item to a grammatical marker—or polygram-
maticalization, whereby one lexical item gives rise to more than one
grammatical marker. The development of nouns into classifiers also
involves the evolution of a classifier construction out of syntactic construc-
tions of other types.
The most common lexical source for classifiers are nouns; verbs are used
less frequently. In languages which have adjectives or adverbs as open
lexical classes these tend not to develop into classifiers. Different semantic
groups of nouns and verbs give rise to different classifier types. In §13.1.1,1
show which semantically defined subgroups of nouns tend to become
classifiers. The difference between nouns and verbs as potential sources
for classifiers is that verbs which grammaticalize to classifiers usually come
from a semantically defined subgroup. In contrast, there are cases where
almost any noun can be used as a classifier of the 'repeater' variety: see
§13.1.2. Classifiers can come from verbs (§13.1.3) or from deverbal nomi-
nalizations (§13.1.4). All classifiers of a given type in a language may have
one kind of source (either nouns or verbs); classifier systems of mixed
origins—some from nouns, and some from verbs—are discussed in §13.1.5.

13.1.1. From a noun to a classifier

The choice of which set of nouns becomes classifiers is typically language-,
family-, or area-specific (why certain nouns become grammaticalized as clas-
sifiers in some languages but not in others can to some extent be accounted
for by socio-cultural conventions and traditions: see §12.3). Australian lan-
guages typically use generic nouns such as 'vegetable food', 'meat' (or 'edible
animal'), and various human divisions (e.g. 'man', 'woman', 'person') as
generic classifiers. Mayan languages typically have a number of classifiers
which refer to the domain of social interaction, culture and beliefs (e.g. 'male
kin', 'respected male', 'deity'). Classifiers can come from words for 'animal',
'dog', 'corn', 'rock', 'water'. Words referring to food, drink, or pets may
become relational classifiers, or possessed classifiers.
Semantic subgroups of nouns which frequently grammaticalize as clas-
sifiers are: (A) body parts; (B) kinship nouns, and nouns denoting humans
(man, woman) and higher animates; (C) generic (or superordinate) nouns;
(D) unit counters.2 This list is by no means exhaustive. Nouns of other
2
Greenberg (1990: 189) mentions the following semantic groups: '(a) superordinate terms
such as a "person" as a classifier for humans, and "tree" for individual species; (b) items in
354 Classifiers
semantic groups which refer to culturally important objects, or notions—
such as a house or a canoe—can also become classifiers (E).
Many of these nouns—notably, body parts, kinship nouns and generic
nouns—can give rise to more than one classifier type, both cross-
linguistically and within a single language; then they are said to undergo
polygrammaticalization. The kinds of semantic changes which typically
occur when nouns from these semantic groups become classifiers are dis-
cussed in §13.9.
Table 13.1 shows the lexical groups of nouns which tend to grammati-
calize and the different types of noun categorization devices that they
develop into.

TABLE 13.1. Groups of nouns which tend to develop to classifiers

Lexical groups of nouns Types of classifier Examples of languages
A. Body parts Verbal classifiers Munduruku, Yanomami,
Australian (e.g. Mayali),
Palikur, Southeast Asian
languages (Thai, Chinese)
Numeral classifiers Totonac, Mayan,
Minangkabau, Hmong,
Palikur, Kana, Southeast
Asian languages
Locative classifiers Palikur
B. Nouns referring to Noun classifiers Mayan, Australian,
kinship, humans and Austroasiatic,
higher animates Austronesian, Daw (Makii)
Noun classes Eastern Nilotic, Zande
Possessed classifiers Ponapean
C. Generic nouns Noun classifiers Minangkabau, Mayan,
Mixtec, Australian
Verbal classifiers Australian
Possessed classifiers Carib languages, Nade'b
(Makii), Yuman languages;
Palikur, Island Carib,
Bahwana (Arawak),
Tariana, Tupi-Guarani
Numeral classifiers Viet-Mwong, Thai, Lao
D. Unit counters Numeral classifiers Kana, Minangkabau
E. Culturally important Any type Arawak languages,
items: e.g. house, canoe Ngan'gityemerri, Ponapean

one-to-one relation to the objects being counted, among the most common of which are
"head" for animates and "trunk" or "stalk" for trees; (c) words which themselves designate
arbitrary or insignificant units like "piece" and "grain", etc.'
 Development of Noun Categorization 355
(A) Body parts as a source for classifiers
Of the groups mentioned in Table 13.1, body parts are the most frequent
source of VERBAL classifiers—see the examples in Chapter 6. For instance,
in Munduruku (Tupi), of about 120 classifiers, at least 96 originate from
body parts (Goncalves 1987: 24-9).
The connection between the lexical source and verbal classifier is often
transparent, as in Baniwa (Table 9.5). Some Australian languages use
special semi-suppletive forms of body parts as classifiers (e.g. Tiwi:
§6.2.4); the connection between an incorporated body part and a classifier
can sometimes only be reconstructed (see §6.4.2 on the origin of shape-
related classifiers in Athabaskan languages).
Verbal classifiers often originate in grammaticalized constructions with
body part incorporation. Incorporated body parts can be syntactically
and semantical ly distinct from the same items grammaticalized as verbal
classifiers (see §6.2.1, for these distinctions in Mayali based on Evans
1996: 76-8; also see §6.4.1, and Aikhenvald and Green 1998, for
criteria distinguishing incorporated body parts and verbal classifiers in
Palikur). Sometimes, if the same morpheme functions as a verbal classi-
fier and as an incorporated noun, ambiguities may arise, as in 13.1 from
Xamatauteri (Yanomami: Ramirez 1994: 131). The use of verbal classi-
fiers in Yanomami is obligatory for nouns in S/O function, and in
copula clauses. The morpheme ko has two meanings and two functions:
it means 'heart' as a body part, and 'round object' as a verb-incorporated
classifier.
13.1. korekoremi-ko
type, of.parro t-heart/vcL: ROUND
'This is a parrot's heart', or 'This is one parrot.'
Body parts and parts of plants often become NUMERAL classifiers. They
often tend to become mensural classifiers; more rarely, they may become
sortal classifiers. This development is typical for Mesoamerican languages.
In Totonac (Totonacan-Tepehuan family: Levy 1993), many numeral clas-
sifiers, SORTAL as well as MENSURAL, come from body parts—see examples in
Table 13.2.
Numeral classifiers derived from body parts can combine 'sortal' and
'mensural' functions. Two of the eleven numeral classifiers in Palikur are
derived from body parts, ukulwok 'hand' and biyulbiy 'mouth' (Aikhenvald
and Green 1998). They can be used to classify the noun they come from, as
in 13.2.
13.2. pi-wok-na i-wak-ti
tWO-NUM.CL:HAND-two INDEFINITE.PERSON-hand-NON.POSSESSED
'two hands, or fingers'
356 Classifiers
TABLE 13.2. Numeral classifers from body parts in Totonac
Gloss Meaning as body part
Sortal classifier
cha':- Human cha':n 'leg'
tan- Animal tan 'buttocks'
laka- Flat, surface lakan 'face'
paq- Flat, unbounded paqan 'arm, wing'
Mensural classifier
ak- Units of length, e.g. 'metres' akdn 'head'
qalh- Sips, bites of food qalh 'mouth'

Unlike all other numeral classifiers they can be also used as measure
terms, as shown in 13.3.
13.3. paha-uku-wa kumat
One-NUM.CL:HAND-EMPHATIC beans
'one handful of beans'
Body part terms are widely used as MENSURAL classifiers in Tzotzil
(De Leon 1987: 93). The most frequently used mensural classifiers are
associated with the hand and its parts, e.g. k'et 'handful', used to refer
to a pile of grains such as corn, beans and coffee, torn 'two hands (hank) of
thatch'; poch: 'palm', and ch'utub: 'span between thumb and forefinger'. In
Minangkabau (Marnita 1996: 131) and in Hmong (Bisang 1993: 35), body
parts tend to become mensural classifiers, e.g. Hmong jan 'finger', and also
'a term used for measuring cloth'.3
Less frequently, body parts become just SORTAL classifiers; then they cat-
egorize nouns in terms of their shape. There are typically just a few body parts
which get grammaticalized that way. In Kana, a Cross River Benue-Congo
language with a system of numeral classifiers (see Chapter 4), only two
classifiers out of nineteen are derived from body parts: dkpa 'flatly shaped
objects', is derived from 'skin', and 'spot-like objects' is derived from 'eye'
(Ikoro 1996a: 90-6). Of over twenty numeral classifiers in Ngyemboon, a
Niger-Congo language from Cameroon, only two come from body parts
('head' and 'hand': Viktor Vinogradov, p.c.). Body parts used as superordi-
nate class nouns are a source for sortal numeral classifiers in languages from
east and mainland Southeast Asian languages (Bisang 1996). They are used
in classification of animates, and more rarely of inanimates. In Bahnar
(Central Bahnaric) kal 'head' is used for living beings, people and bonded
souls (i.e. slaves), as well as to count animals, boats and valuables, and as a
3
This is reminiscent of systems with incipient numeral classifiers (§4.4); these often come
from body parts such as 'head' and 'horn'.
 Development of Noun Categorization 357
denigrating term for humans; mat 'eye, pupil of the eye' is one of the
honorific classifiers for humans (Adams 1989: 71). In Chrau, Mnong
Gar, and Rolo'm (South Bahnaric) voq 'head' is used as a generic classifier
for animals (Adams 1989: 77) (see Appendix 2 for further examples).
Body parts and parts of plants may grammaticalize as LOCATIVE CLASSI-
FIERS. Three of the twelve locative classifiers in Palikur originated from
body parts: -kigsa 'on.POINTED' is related to -kig 'nose'; -vigku 'on.ROAD,
RIVER' derives from -vigik 'bone, marrow'; peru 'on.BRANCH.LIKE' comes
from pew 'branch' (see Diagram 7.1).
(B) Nouns referring to kinship, humans, and higher animates as a source
for classifiers
Some members from the large category of animates, humans and kinship
terms frequently develop into NOUN CLASSIFIERS. This is typical for Mayan
languages (Kanjobalan branch), Australian, Austroasiatic, Austronesian,
and a few South American languages, e.g. Daw (Maku).
In Jacaltec (Kanjobal subgroup of Mayan: Craig 1986c: 266-7), twelve
noun classifiers which belong to the subsystem of social interaction are
derived from nouns with human referents, and from kinship nouns (noun
classifiers for humans are derived from nouns meaning 'man' and 'woman'
in Akatek, a language from the same subgroup—see Table 8.4; Zavala
1992: 152). The majority of noun classifiers in Mam, another language
of the same family, are derived from nouns denoting humans and kinship
terms (the only exception is a classifier for 'non-human') (England 1983:
158-9). Table 13.3 (England 1983: 158) shows noun classifiers in Mam;
those derived from kinship terms and nouns referring to humans are in
bold. In some cases, the grammaticalization of nouns as classifiers involves
phonological reduction and semantic changes (see §13.5.1).

TABLE 13.3. Noun classifiers derived from common nouns in Mam

Classifers Semantics Common nouns Semantics
jal Non-human Jiil Wild animal
nu7xh Baby nu7xh Child
b'ixh Person of the same status, fondly ? 9
q'a Young man q'aa Young man
txin Young woman txiin Young woman
ma Man matiij Big
xu7j Woman xu7j Woman
swe7j Old man swe7j Old man
xhyaa7 Old woman yaab'aj Old woman
xnuq Old man, respectfully xiinaq Man
xuj Old woman, respectfully xu7j Woman
358 Classifiers
Animate or human nouns are frequently grammaticalized as DERIVATIONAL
devices which overtly mark the gender of a noun on the noun itself. Many
languages use words for 'man' or 'male', and 'woman' or 'female', with a
classifier-like function as a technique to distinguish genders of hybrid
nouns, e.g. English female crocodile, male crocodile* In other languages,
these terms have taken on the formal status of noun class markers. The
development of nouns with a human referent and kinship nouns to overt
NOUN CLASS markers on nouns, which then got extended to be agreement
markers within noun and verb phrases, was suggested for Eastern Nilotic
by Heine and Vossen (1983) (cf. Heine and Reh 1984: 219-20). Two gender
markers can be reconstructed for Pro to-Eastern Nilotic—masculine *lo
and feminine *na—which are 'likely to be derived from the relational
nouns, *IV 'member/person of and *nyaa- 'girl, daughter'. These new
gender markers emerged as the result of the reanalysis of lexical items,
initially used as heads of genitive constructions. Later, they got gramma-
ticalized as prenominal modifiers, combining with the existing demonstra-
tive roots, and then replacing them. Along similar lines, Mupun, a West
Chadic language, inherited the Proto-Chadic system of two genders
realized through agreement on nominal modifiers (Frajzyngier 1993); this
language is also developing an overt gender marking on proper names,
some common nouns, and even pronominal elements, via grammaticaliza-
tion of naa 'woman' which has given rise to the prefix na-, and daa 'man',
which has given rise to the prefix da-, e.g. na-komtak 'such and such a
female', da-komtak 'such and such a male'.5
Lexical nouns can develop into personal pronouns, and subsequently get
reanalysed as NOUN CLASS markers. Such a scenario was reconstructed for
Zande (East Adamawa, or Ubangian: Claudi 1985; Heine and Reh 1984:
220-5). Synchronically, Zande has four genders in the singular (masculine,
feminine, animal, and inanimate), and three in the plural (human, animal,
and inanimate). Gender is distinguished in personal, possessive, and reflex-
ive pronouns, and also realized in agreement between the subject and
adjectival predicates. It is not marked on head nouns. Diachronic evidence
shows the following lexical origins for the gender system in Zande (Heine

4
In Tucano, a language with three genders and a large number of classifiers-repeaters, the
lexemes 'man' and 'woman' are used to differentiate the sexes of animals, e.g. seme tmi 'male
paca (a large rodent)' (lit. paca man), seme mania 'female paca" (lit. paca woman) (Ramirez
1997: 208). In Manambu, where masculine and feminine gender are usually not marked on the
noun, the lexemes takw 'woman' and du 'man' are sometimes used as parts of compounded
structures to provide disambiguation, e.g. yanan 'grandchild', yanan-takw 'granddaughter',
yanan-du 'grandson'. Lexical expression of gender distinction sometimes cooccurs with overt
gender marking on nouns, e.g. colloquial Portuguese filho homem 'son man', for 'son', and
filha mulher 'daughter woman', for 'daughter'.
5
The forms themselves may be loans into Chadic, cf. Proto-Benue-Congo *na 'mother' and
*da 'father': Frajzyngier (1993: 44).
 Development of Noun Categorization 359
and Reh 1984: 224; Claudi 1985: 127-35): 'man, male' > masculine gender
marker; 'person' > human; 'thing, prey' > animal (sg); 'animal, meat' >
animal (pi); 'thing' > inanimate. See §13.5.1 on the grammaticalization of
nouns to pronouns, and then to agreement markers.
Grammaticalization of lexical nouns may account for just a few noun
classes out of a system. For instance, the diminutive gender marker pi in the
Bantu languages of Cameroon conies from Niger-Congo *bi 'child'; the
locative gender marker ku comes from the noun *ku 'the outside' (Heine
1982a: 214). The augmentative marker ka- (a vestigial noun class prefix)
appears to be derived from ka 'mother' in Kana (Ikoro 1996a: 60). See
§13.3, on how this may correlate with the relative age of classifier systems.
Kinship nouns can also get grammaticalized as POSSESSED classifiers.
Ponapean (Rehg 1981; Keating 1997: 253) has twenty-two possessed
classifiers, three of which developed out of kinship nouns, e.g. kiseh 'POSS.
^RELATIVE', lit. 'relative'; ullepe 'POSS.CUMATERNAL.UNCLE', ullap 'maternal
uncle'; wahwah 'POSS.CL:NIECE,NEPHEW', lit. 'man's sister's child'. This is not
attested outside Micronesia.
(C) Generic nouns as classifiers
Generic nouns which develop into different classifier types usually include a
term for humans (see B), for non-human animates, and for plants and
various types of food. Generic or superordinate nouns often develop into
NOUN CLASSIFIERS. A sample of generic classifiers in Minangkabau (Marnita
1996: 82) is given in Table 13.4. Each of these is also used as an independent
noun. (This is often not the case; for example, not all classifiers in Mam
have an identifiable corresponding independent noun; and many of those
which do underwent considerable phonological changes: see Table 13.3.)

TABLE 13.4. Generic classifiers in Minangkabau

Classifier Semantics

urang Person, human

ikan Fish
buruang Bird
batang Tree, species of plants
kayu Wood, tree
daun Leaf
incek Seed
anak Seedling (lit. child)
aia Water, drink

Noun classifiers in Mixtec developed via grammaticalization of syntactic

compounds which contained generic nouns, e.g. 'man', 'woman', 'animal',
360 Classifiers
'tree' as their first component (De Leon 1987: 148-50). Examples of noun
classifiers from generic nouns in Australian languages were discussed in
Chapter 3.
Incorporated generic nouns can develop into VERBAL classifiers. This is a
typical development for some northern Australian languages. In Mayali
(§6.2; Evans 1996: 76-8) incorporated generic noun classifiers are a closed
class of about forty items, which include 'tree', 'baby', 'ground', 'grass',
'fire', 'liquid'.
Generic nouns frequently develop into POSSESSED classifiers. These often
include various types of food, firewood, and pets (see examples in (A) of
§5.2). Systems vary as to what proportion of possessed classifiers main-
tain their connection with the lexical nouns. In Apalai, a Carib language
(Koehn 1994), the connection between possessed classifiers and their
lexical sources is transparent. So is it in a number of Arawak languages
(Palikur, Island Carib, and Bahwana). These languages have a small
number of possessed classifiers (e.g. 'food', 'pet', 'catch', 'plantation')
apparently acquired recently under the possible influence of Carib. In
other Carib languages, such as Panare (Carlson and Payne 1989; see also
Table 5.1 above), the origin of some items can only be recovered if
compared to lexical items in other languages (cf. the comparison in
Koehn 1994: 42-3); not all possessed classifiers originate in generic
nouns.
POSSESSED classifiers, NOUN classifiers, and NUMERAL classifiers often
come from a word meaning 'thing'. In Tariana yarupe 'thing' is a generic
possessed classifier for non-human nouns (Aikhenvald 1994a). In a number
of Tupi-Guarani languages of South America, mba'e 'thing' tends to
develop into a kind of noun classifier for non-humans. In Apalai (Carib),
kyry 'thing' is developing into an unmarked generic possessed classifier,
alongside other generics. In some Viet-M«ong languages cai 'thing' is used
as a generic NUMERAL classifier (Adams 1989: 182-3). In Thai and Lao ?an
'thing' can be used as a numeral classifier for any inanimate, or abstract
noun (Adam Chapman, p.c.).
(D) Unit counters as a source for classifiers
Unit counters and measure words often give rise to NUMERAL CLASSIFIERS of
the mensural type (I have found no examples of unit counters becoming
sortal classifiers). They may include abstract words, such as Kana akpo, a
classifier used to count 'inanimate objects with a trunk', e.g. tree, which
comes from akpo 'length'; and akpo (note different tones), a classifier 'used
to count objects which have either irregular or heap-like shapes', from kpb
'heap' (Ikoro 1996a: 90, 94-5). They are often anthropomorphic measure
terms, as in Minangkabau kapa, a classifier used for 'objects that can be
held in the palm of one's hand', e.g. cooked rice, sand, dirt from kapa
 Development of Noun Categorization 361
'handful', or ujuang jari, a classifier used for 'things which one can take
with a tip of one's finger', lit. 'finger tip' (Marnita 1996: 134).
(E) Nouns from other semantic groups as sources for classifiers
Specific classifiers used in any function often arise from culturally impor-
tant lexical items, e.g. a canoe, or a house, as in Arawak languages of South
America; or large woomeras, canegrass spears, or digging sticks, as in
Ngan'gityemerri (Australian). In Ponapean (Rehg 1981, Keating 1997:
253), a number of possessed classifiers for items of personal use come
from names for these items, e.g. mware 'POSS.CL:TITLE/GARLAND' from
mwar 'title', or ede 'POSS.CLINAME', from ahd 'name'. A noun which gram-
maticalizes as a classifier may undergo some semantic changes, e.g. were
'POSS.CL:CANOE,VEHICLES', from wahr 'canoe', or pelie 'POSS.CL:SIBLING', from
pelie 'member of a matched pair' (see §13.9).

13.1.2. Repeater phenomena and the origin of classifier constructions

In some languages any—or almost any—noun can be used in the
classifier slot, to classify the same noun, or semantically related nouns.
This 'repeater' phenomenon is widespread in Southeast Asian languages,
in some Micronesian languages, and in some South American languages
(see §9.1). Repeaters and partial repeaters as numeral classifiers are
reported in Thai, Lao, Burmese, and Lahu (Goral 1978: 33-4; see
Chapter 4 above). In Truquese repeaters are used both as numeral
classifiers and as possessive classifiers (Chapter 8; Benton 1968). Repea-
ters are developing into ad hoc classifiers for various inanimate objects
in Tariana (Aikhenvald 1994a: 447-8); the repeater technique in this
language is the result of areal diffusion from. East Tucano. For instance,
the form ehkwapi 'day, daylight, weather' is used as a repeater to 'classify'
itself, and can occasionally be employed as a classifier for any natural
phenomenon.
Repeaters can be shown to be the source for more grammaticalized
classifier systems. Classifier constructions in Chinese and other Sino-
Tibetan languages originated from constructions with 'repeaters'.
Hashimoto (1977) suggests the following structure for the erstwhile
classifier construction:
NOUN1 + NUMERAL + NOUN1

The 'repeated' part (i.e. the second occurrence of the noun) then under-
went 'certain generalization of meaning', and a number of classifiers
appeared (Hashimoto 1977). For instance, a general term for man, jen2,
is used for humans, instead of repeating their specific names. If the head
noun is a compound, or a disyllabic word, only part of it may be repeated.
362 Classifiers
In Nakhi (Sino-Tibetan: Hashimoto 1977) it is usually the 'categorial noun
stem', as in 13.4.
13.4. zer-ndzer tg'i ndzsr
willow-tree this tree
'this tree'
In Tariana and East Tucano languages, it is usually the last derivational
suffix; 13.5 is from Tariana (also see (F) in §9.1).
13.5. diha-kema pasole-kema
this-siDE left.hand-SIDE
'this left hand side'
Repeater constructions can be the source for large systems of noun
categorization devices which typically include numeral, deictic, possessed,
and relational classifiers. Verb-incorporated classifiers and noun classes
can originate in repeaters if the same set of classifier morphemes is used
in different morphosyntactic environments (as in Tariana, or East Tucano
languages). There is no evidence that closed grammaticalized noun class
systems can be created in this way.

13.1.3. From a verb to a classifier

Verbs become classifiers more rarely than nouns. Table 13.5 shows the
semantic subgroups of verbs which can become classifiers.

TABLE 13.5. Semantic groups of verbs which develop into classifiers

Lexical groups of verbs Classifiers Examples
Posture and motion verbs Classificatory verbs Engan, Ku Waru
Deictic classifiers Guaicuruan, Siouan
Verbs of handling Verbal classifiers Imonda
(involving typical activities)
Possessed classifiers Ponapean
Relational classifiers Mam
Numeral classifiers Tzeltal

POSTURE and MOTION verbs tend to become CLASSIFICATORY verbs. Posture

verbs are used as classificatory existential verbs in Ku Waru (Papuan: Table
6.8). In Engan and Waris classificatory verbs include posture verbs
('stand', 'sit', 'hang', 'lie inside', 'lie'), one motion verb ('come') and one
transitive verb ('carry') (Tables 6.7 and 6.10 respectively). A tendency for
POSTURE verbs to develop classificatory overtones, i.e. to be used to classify
their S argument in terms of its form, is found in a number of Indo-
 Development of Noun Categorization 363
European languages. The use of posture verbs in Russian correlates with
the position and physical form of its subject (Rakhilina forthcoming); the
same tendency has been described for German by Borneto (1996).
POSTURE verbs also develop into DEICTIC classifiers, as in Siouan and in
Guaicuruan languages. Deictic classifiers in Guaicuruan languages go
back to combinations of deictic elements plus posture verbs, e.g. Proto-
Guaicuruan 'singular masculine absent vertical' *(e)-d:a, from Proto-
Guaicuruan *d:a 'stand' (Ceria and Sandalo 1995: 181). In Siouan
languages, such as Mandan, deictic classifiers which refer to the position
of animate objects, and to the form of inanimate objects, often function as
main or auxiliary verbs (Barren and Serzisko 1982: 100). Thus, classifiers
retain a synchronic link with the source verb. In 7.8 and 13.6 -mak 'lie' is
used as a deictic classifier for long horizontal objects (e.g., a river):
13.6. ma:ta mak-omakoc
river lie-NARRATIVE.PAST
'The river was there (lying).'
This semantic development is due to the fact that deixis by its nature is
connected to position and location in space. Systems of spatial deixis often
include reference to parameters such as visibility, dimensions such as
height, and reference to up and down: see Anderson and Keenan (1985:
290-2; Himmelmann 1997).6
VERBS OF HANDLING—as a part of serial verb constructions—gave rise to
verbal classifiers in Imonda (W. Seiler 1983; 1986), and a number of other
languages from the Waris family (W. Seiler 1986: 194; Foley 1986: 91; see
§6.2.2 above); a sample of these verbs is given in Table 13.6 (W. Seiler 1986:
190-3; see also §13.5.2 below). These classifiers refer to an O or S argument.

TABLE 13.6. A sample of verbal classifiers in Imonda

Classifier Semantics Source verb Semantics
pot- 'fruit to be picked' pot 'pick fruit'
pui- 'objects which are normally broken' puiual 'break in two'
i- 'scooping water out of an area i 'scoop water
dammed up for the purpose of out'
collecting fish'

Many objects in Imonda are classified depending on how they can be

handled; hence multiple class membership: for instance, a fish, tobto, is

6
Posture verbs in Siouan languages undergo polygrammaticalization (Barren and Serzisko
1982: 102): besides becoming classifiers, they grammaticalize as auxiliary verbs; they also
grammaticalize as demonstratives.
364 Classifiers
typically referred to with different classifiers depending on the state it is
in—still to be caught and killed, or already cooked and ready to be eaten
(W. Seiler 1986: 202-3). An example like 13.7 (repeated from 6.7) could
have meant at an earlier stage 'pick (up) and give me coconut'. The first
verb was just specifying a way of manipulating an object in this state, and it
gradually lost its verbal meaning in this context (but not in others, in the
case of pot- 'pick fruit') becoming reanalysed as a classifier (cf. Foley 1986:
91).7
13.7. sa ka-m pot-ai-h-u
COCOnut ISG-GOAL CL:FRUIT-give-RECIPIENT-IMPERATIVE
'Give me the coconut.'
POSSESSED classifiers can be derived from verbs of handling which involve
typical activities. A number of possessed classifiers for specific inanimate
objects in Ponapean (Keating 1997: 253) are derived from verbs, e.g. tie
'poss.CL:EARRINGs', from tie 'to wear earrings', ullunge 'POSS.CL:PILLOWS',
from ulung 'to use a pillow'. A number of relational classifiers which
describe the way the possessed noun can be handled by the possessor in
Mam (Mayan: Craig forthcoming; Roberto Zavala, p.c.) could derive from
verbs of manipulation which describe typical activities involving the object
to be classified; the relational classifier chi' 'cooked food' could have come
from the verb chi' 'to eat cooked food', and the relational classifier lo'
'fruit' could have come from the verb lo' 'to eat fruit' (e.g. n-chi'-ye' kyix
(lsg-POSS.CL:COOKED-POSS fish) 'my (cooked) fish').8
Verbs and verbal roots are never the only source for RELATIONAL,
POSSESSED, or NUMERAL classifiers. If a numeral classifier system is of mixed
origins (discussed in §13.1.5) some classifiers can come from verbs. These
verbal roots refer to handling done with body parts (e.g. in Tzeltal, 'strike
with an open hand', or 'take large mouthfuls', or 'break between fingers':
Berlin 1968: 213). They can also include other verbs such as 'sleep', 'press',
'break', or 'trap' (Berlin 1968: 214). Hopkins (1970) provides further
examples of numeral classifiers derived from transitive and positional verbs
in Tzeltal, Jacaltec, and Chuj (Mayan). Usually, all these verbs contain
information about the shape, size, material, or position of the O or S
argument. Verbs which describe mental processes, or perception, or modal
verbs, do not become classifiers. This is reminiscent of two phenomena.

7
Which verbs will become classifiers in Imonda is scarcely predictable. Many of them seem
to describe typical activities and ultimately relate to cultural and social issues (cf. §12.3). The
etymological link between the source verb and a classifier can be easily established in some
cases; in other cases, however, the process of grammaticalization of a verb to a classifier has
resulted in phonological reduction and semantic bleaching; further examples are given by
W. Seiler (1986: 202-3).
8
According to Lyle Campbell (p.c.), they may alternatively have derived from nouns.
 Development of Noun Categorization 365
First, classificatory verbs (see §6.2) always describe manipulations which
imply contact with the S or O (not the A). Second, in many languages
verbal classifiers can only be used with verbs which presuppose physical
contact and manipulation of the S or O argument.

13.1.4. Classifiers from deverbal nominalizations

Deverbal nominalizations are often the source for RELATIONAL and
POSSESSED classifiers. In Cahuilla (Uto-Aztecan: H. Seiler 1983: 36-8;
1977: 299-305; §5.5.2 above) the classifiers in possessive constructions
are derived from nominalized verbs. These verbs describe the ways in which
objects can be handled, e.g. ?ay?a, the classifier used for 'fresh fruit or
blossoms plucked from the tree', from the transitive verb stem ?dy 'pluck'
(ex. 5.53); ci'a, the classifier used for edible items fallen off the tree and
picked from the ground, from ci- 'pick from the ground'; wes'a, the classi-
fier used for planted trees and fruits, from wes 'plant'; and ?as, the classifier
for pets, from the transitive verb stem -as 'own' (Seiler 1977: 305). The
generic classifier used for inanimates, -exan, is derived from the verb -exan
'do' (p. 300).
In these examples the semantic connection between the source verb, and
the classifier is transparent. A more opaque connection may only be
explainable by reference to the cultural background. Thus, ki?iw?a (Seiler
1977: 300), a classifier used for trees, plants and fruits in a group, growing
wild, comes from the transitive verb stem 'wait'. Trees and their fruits were
not privately owned; the places where they grew were instead assigned to a
lineage. The members of a lineage were allowed to harvest these places
when the appropriate time came. So, 'waiting' used to be essential for the
appropriate handling of these objects (see §12.3).
RELATIONAL classifiers in Micronesian languages go back to deverbal
nominalizations, as suggested by Harrison (1988: 74-5). The classifier
for edible things is 'obviously related to an extant, or reconstructable
verb "to eat"' (Harrison 1988: 67); see also Manam 'ana- 'edible possessive
classifier', from 'ani 'eat' (Harrison 1988: 77); and see Lichtenberk (1985:
119) for a possible connection between the Proto-Oceanic classifier *ma
'drinkable items', and the verb *inum 'drink'.
In Apalai (Carib: Koehn 1994: 42-3) just a few POSSESSED classifiers
can also be shown to come from deverbal nouns, e.g. -napy 'vegetable/
fruit', -enapy 'eat vegetable or fruit'. Other classifiers derive from nouns.
Similarly, in a number of Austroasiatic languages, some deverbal nom-
inalizations are used as NUMERAL classifiers. The form vanong, derived
from a verb stem vong 'fall down' with the nominalizing infix -an-, is
used as a classifier for animals in general in Chrau (Mon-Khmer; Adams
1989: 77).
366 Classifiers
13.1.5. Classifiers of mixed origin
In many cases the whole system of classifiers comes from just one source.
For instance, in Kilivila (Senft 1993; 1996) all classifiers derive from nouns.
In some languages, however, classifiers of one type can have mixed origins:
some may come from nouns, and some from verbs.
Dixon (forthcoming) suggests a mixed origin for most noun class
systems in Australian languages. The 'vegetable', or non-flesh-food class
typically comes from a generic noun classifier (see §13.4), while markers for
masculine and feminine class tend to come from masculine and feminine
forms of third person singular free pronoun.
NUMERAL classifiers come from verbs as well as from nouns in numer-
ous Austronesian and Austroasiatic languages; but statistically, more
classifiers appear to come from nouns. Objects are classified according
to the ways they can be handled, or consumed; or in agreement with their
inherent properties ('drink' can be used as a classifier for liquids). About
one-fifth of the numeral classifiers in Minangkabau are derived from
verbs. Most of these are mensural classifiers, and they derive from verbs
which imply physical manipulation, such as atua 'to pierce' (for fish and
small fruits on a skewer), balah 'to cut in two parts' (for objects cut in
halves), kapiang 'cut with an axe' (for big pieces of wood). A number of
classifiers come from ingestive verbs, e.g. taguak 'swallow', for drinks;
suok 'feed', for food eaten with a hand; and from verbs which refer to an
action performed with a body part, e.g. pikua 'to carry on the shoulder',
for objects carried on the shoulder (Marnita 1996: 121 ff.). Only a few
sortal shape-based classifiers are derived from verbs, and they have a very
specific meaning, e.g. guluang 'to roll' for a rolled object, such as a roll of
paper; lempeang 'to slice', a classifier for 'flat piece of silver, gold and
bronze'.
In Austroasiatic languages the majority of numeral classifiers come from
nouns; but there are a few instances of numeral classifiers for inanimate
nouns derived from verbs, e.g. Gold Palaung kqr-vyang 'the classifier for
circles or coils', from vyang 'go round' (Adams 1989: 103).
Ponapean (Micronesian: Rehg 1981; Keating 1997: 253) is an example of
a system of POSSESSED classifiers of mixed origins. Of twenty-two classifiers
used in the common speech register (see §10.5), five classifiers referring to
humans come from nouns; and four classifiers for food, drink, catch (e.g. of
fish) and a share (e.g. of food) come from verbs; of ten classifiers for
specific items (titles, names, buildings, etc.) and locations three come from
verbs, and six from nouns (the etymology of two generic classifiers and the
classifier kie 'sleeping pads' is unknown). One classifier, rie 'siblings',
comes from the word for 'two'. Classifiers used in humiliative and honorific
registers come from nouns, e.g. tungoal, a humiliative possessed classifier
 Development of Noun Categorization 367
(used for all possessions), comes from the word for food, or eating; moa-
toare, a possessed classifier in honorific register used for sleeping gear,
comes from the word for mat, or sleeping, used in the honorific speech.
A few classifiers in the honorific register appear to be derived from com-
pounds, e.g. sapwellime 'possessed classifier in honorific register for general
possession' comes from sahpw 'land' and lime 'hand' (honorific); Keating
(1997: 263^4) provides socio-cultural reasons for this—high status is linked
with ownership of land.
Classifiers can be derived from roots which belong to more than one
class. In Mayan languages NUMERAL classifiers are typically related to a
class of roots called 'positional' which indicate position or shape; they
may be nouns in one language and verbs in another. In Tzotzil, classifiers
typically derive from positional roots, or transitive roots which encode
semantic dimensions of shape, size, material, position, etc. (De Leon
1987: 60-1), e.g. jav 'cut, sever' (positional), jav- 'to be cut' (intransitive
verb), -jav 'to cut' (transitive verb) > jov 'piece' (numeral classifier); vA':
'standing up in bipedal position' (positional root), va*i- 'to stand up'
(intransitive verb), -va'an 'to stand something' (transitive verb) > -vo'
'numeral classifier for persons'. In Tzeltal (Berlin 1968: 20-2) numeral
classifiers form a subclass of nouns derived from transitive verb roots,
positional verb roots, and from a few noun roots.

13.2. From a closed class to a noun categorization system

Items from closed grammatical classes, such as demonstratives or locatives,
usually give rise to a relatively small closed system of noun classes, and not
to large systems of classifiers. Only in Loven (West Bahnaric) is there a
form yur used for classifying animals in general which comes from a
demonstrative, or interrogative 'that' or 'what' (Adams 1989: 77), and is
one term in a large system of numeral classifiers.
Greenberg (1978) suggested the following chain for the evolution of
overt gender marking from a demonstrative, or from an anaphoric element:
Stage 0 *- Stage 1 »- Stage 2 >- Stage 3
(1) Demonstrative (2) Definite article (3) Non-generic marker on (4) Marker of nominality
nouns, or specific article
DIAGRAM 13.1. Evolution of gender markers (1)

This development was reconstructed by Greenberg (1978: 61 ff.) for

Voltaic languages. His scheme involves semantic change from demonstra-
tive to definite article, and from definite article to specific article, plus a
passage from article to noun marker. It also involves formal changes, e.g. a
368 Classifiers
shift from free word to clitic (such as demonstrative to article) and from
clitic to affix.
A similar development for overt gender markers on nouns can be sug-
gested for Berber languages. In Berber, masculine gender is usually marked
with the prefix a-, and feminine with t-a-; these can be shown to go back to
fossilized and desemanticized definite articles (see Vycichl 1957; Aikhen-
vald 1984). Turkana (Eastern Nilotic: Dimmendaal 1983) is a particularly
instructive example of how overt gender markers on nouns are used as
markers of nominality (Stage 3), and at the same time sometimes maintain
their function as specific 'articles' (see §12.1.3, on how Turkana gender
prefixes correlate with discourse organization: Dimmendaal 1983: 222);
also see Childs (1983), on the application of Greenberg's schema to the
development of noun classes in the Southern branch of West Atlantic
languages.
Another path of evolution, suggested by Heine (1997b) for Khoe
languages, implies that, besides demonstratives, erstwhile 3rd person pro-
nouns can trigger the spread of gender marking. In Hiechware, person,
gender, and number are marked on personal pronouns only; in Kxoe there
is optional gender marking on demonstratives (and in one language, Naro, it
has also spread to adjectives in a noun phrase). Heine hypothesizes that
personal pronouns were used as specifiers and markers of emphasis; later
they combined with demonstratives and other modifiers and lost their prag-
matic overtones in some languages. The structure of personal pronouns in
Kxoe shows that person/number/gender markers in this language have to do
with emphasis: the demonstrative stem is used to emphasize personal pro-
nouns. Third person pronouns which distinguish two genders consist of a
demonstrative stem xa- and the appropriate person, number, and gender
marker, e.g. xa-md '3rd person singular masculine', xa-hs '3rd person sin-
gular feminine' (Heine 1997b). Emphatic 1st and 2nd person pronouns can
also be formed by adding the same demonstrative stem onto the personal
pronoun, e.g. ti '1st person singular non-emphatic'; xa-ti'lst person singular
emphatic'. In the 3rd person, non-emphatic pronouns are no longer used.9
These considerations allow us to suggest the alternative scenario—see
Diagram 13.2.

Stage 1 >• Stage 2 >- Stage 3

3rd person pronoun (also Article-like marker Marker of nominality
used as a specificity marker)
DIAGRAM 13.2. Evolution of gender markers (2)

9
A similar scenario can be reconstructed for the Arawak languages of South America.
There, gender markers (masc. or non-feminine *-ri, feminine *-ru) on nouns go back to 3rd
person pronouns and cross-referencing affixes.
 Development of Noun Categorization 369
The chains represented in Diagram 13.1 and in Diagram 13.2 are not so
different: cross-linguistically, demonstratives and 3rd person pronouns
often overlap in their functions.
The development of gender, animacy, and a human/non-human distinc-
tion in pronouns may have to do with their reference. Personal pronouns are
often restricted to humans, while demonstratives can be used for non-
human referents. This may give rise to a new animate/inanimate, or perso-
nal/non-personal, distinction. In formal Finnish, the 3rd person pronoun
hdn is used for human referents only, while the erstwhile proximate demon-
strative se is used for non-humans (Tiit-Rein Viitso, Liisa Jarvinen p.c.; also
see Appendix 1). In Bengali, an Indo-Aryan language with numeral classi-
fiers but no grammatical genders, 3rd person pronouns refer only to
humans, and demonstratives refer to non-humans (Masa Onishi, p.c.).10
In languages which already have noun classes this may lead to creation of a
different system of classes (or genders) for different word classes and
targets. Warray (Australian: Harvey 1987) has four noun classes, generic
noun classifiers, and a feminine/non-feminine distinction in 3rd person
singular pronouns. Independent pronouns can have only human referents,
and this may result in the creation of an additional, human/non-human
distinction (see (C) under §2.4.4). Genders, or noun classes can also develop
via reanalysis of other closed grammatical systems (e.g. Kiowa-Tanoan
languages: §13.5.2)."
A closed class of locative adpositions can be a source for NOUN CLASSES.
The rise of new 'locative' genders then relates to the reanalysis of locative
expressions. These may contain a relational noun which refers to a place;
for instance, the neuter gender prefix i- in Turkana arose from a gramma-
ticalized anaphoric noun meaning 'place just referred to'. Locative adposi-
tions can be clitics which ultimately go back to a subclass of nouns with
locative meanings. Locative genders in languages of the Congo branch of
Bantu go back to locative and adverbial expressions (Greenberg 1990: 260;
also see Givon 1976: 173-5). According to Heine (1982a: 214), at least
some of these derive from locational nouns (e.g. *ku- 'the outside1, which is
still used as a preposition in a number of Bantu languages). Classifiers of
other types can also come from locational adpositions which are used as
derivational suffixes (some of these probably go back to independent
locational nouns).
The morpheme -(V)ku is used as a classifier 'inside of; cavity; extended'
in many Arawak languages (Payne 199la: 384). In Palikur (Aikhenvald and
10
See also Corbett (1991: 311) for how such a distinction is developing into a new gender
system in Persian, and for a similar phenomenon in Latvian.
11
According to Nichols (1992: 141), 'all that is needed in order for noun classes to arise is a
covert animacy system, a potentially recruitable formal distinction, and pre-existent agreement
patterns'.
370 Classifiers
Green 1998) it is used as a numeral classifier, as a verbal classifier, and as a
locative classifier. Synchronically, however, there are three independent
morphemes, since they display different allomorphy. This form can also
be used in multiple classifier systems—in Tariana and Baniwa it appears as
an adjectival agreement marker, as a numeral, and as a verbal classifier. In
other languages of the family the same morpheme is used as a locative case
marker (e.g. Machiguenga -ku, Bare -uku 'locative'); or as a derivational
affix marking location (Tariana -iku 'place of). Palikur also has a locative
classifier -hakwa 'OH.LIQUID' which is used as a derivational formative in
locational nouns such as paraw-hakwa (water-on.LIQUID) 'ocean'.

13.3. Languages with several classifier types, and the relative

age of noun categorization devices

Multiple classifier systems which use different morphemes in different

functions often come from different lexical sources. The emergence of
such systems from different and sometimes from the same lexical source
can be considered as the result of polygrammaticalization processes (see
Craig 1991; Heine 1992; §13.5.1 below). The basic assumption is that if a
system is semantically transparent, and connects easily with lexical sources,
it is relatively new. The old noun classes in Bantu are difficult to trace back
to lexical sources (Traugott and Heine 1991b: 10). In Palikur, different
nouns from the same semantic group—e.g. body and plant parts—give rise
to different classifier types: 'hand' and 'mouth' become numeral classifiers,
while 'branch' and 'nose' became verbal classifiers and locative classifiers,
and 'trunk' gave rise to both numeral and verbal classifiers.
In Truquese repeaters which become classifiers have the same meaning
independently of whether they are used as numeral or as possessed classi-
fiers (Benton 1968). Example 8.3 illustrates the same item batang 'tree' as a
numeral classifier, and as a noun classifier in Minangkabau; as a numeral
classifier it describes the noun in terms of its shape (long and vertical), and
as a noun classifier it is a superordinate item.
The same items used as numeral classifiers and as noun classifiers can
also be said to undergo polygrammaticalization (cf. DeLancey 1986: 440-1,
on the categorization of nouns with numeral classifiers and with noun
classifiers in Tai languages). Classifier morphemes of more than one type
in a language can correspond to different stages of grammaticalization of
the same set (see the discussion of Ngan'gityemerri in §13.5.1 and in
§13.8.1).
Sometimes, of a large classifier system only a few specific classifiers can
be shown to have a lexical origin. Motuna, a Papuan language, has fifty-
one classifiers; of these, only a few specific classifiers—those which refer to
 Development of Noun Categorization 371
time spans—can be related to nouns, e.g. -ru 'day' from ruu 'day', -mu 'day
(in the past)' from muu 'night', -mori 'year' from moi 'canarium almond
season/year' (Onishi 1994: 169), with all the rest having no known etymol-
ogy. This system must be old. In contrast, of nineteen numeral classifiers in
Kana (Kegboid: Ikoro 1996a), only two have no etymology. We conclude
that this system must be relatively new; this is confirmed by other historical
and comparative considerations of the origin of numeral classifiers in
Kegboid languages (Suanu Ikoro and Gerrit Dimmendaal, p.c.).
In Daw (Makii) noun classifiers come from superordinate generic nouns
which are still used in the language in a non-classifier function (e.g. daw
'CLASSIFIER:PEOPLE', 'person', also used as a 3rd person pronoun: Martins
1994). Some locative classifiers can be shown to be historically derived
from nouns; however, a lexical source can only be reconstituted by com-
parison with a genetically related language. For instance, the locative
classifier -mi' LOC.CL:LIQUID' is cognate to the noun mi 'water' in Nadeb, a
language from the same family. This term for water is not used in Daw.
Thus, since locative classifiers lost their lexical source in Daw, and noun
classifiers did not, one may assume that locative classifiers are older than
noun classifiers.
Another indication of the age of a system is the degree of phonological
reduction of a morpheme. Chinantec (Foris forthcoming: 257, 302-3) has
nine noun classifiers, and several dozen numeral classifiers. In just one case
the same lexeme is used in two classifier functions, i.e. undergoes poly-
grammaticalization. Mai3 'sphere' gave rise to a numeral classifier for
round objects (mat3), and a noun classifier mi3 also used with round
objects and spheres (example 8.6). The noun classifier underwent phono-
logical reduction while the numeral classifier did not. The different degree
of phonological depletion shows the difference in the relative age of the two
systems.
In languages with a small closed system of two or three genders and a
large system of classifiers in several functions these two may have different
origins. In North Arawak languages with multiple classifier systems, clas-
sifiers have lexical origins, and gender markers come from a closed class of
cross-referencing markers. Gender markers go back to Proto-Arawak.
Given the areal distribution of classifiers in the Amazon, we may hypothe-
size that the classifiers were acquired more recently, at the level of indivi-
dual subgroups (Aikhenvald 1996c; further evidence is given in Aikhenvald
forthcoming c).
In other cases the relative age of different noun categorization systems just
cannot be established. In some Arawa languages of Southern Amazonia,
both gender and the ka- noun class (see §2.7.2) are semantically non-
transparent and do not easily connect to any lexical sources. Consequently,
it is impossible to determine their relative ages.
372 Classifiers
13.4. Internal evolution of noun categorization

The internal evolution of noun categorization devices often involves a

passage from one type of noun categorization to another. A lexical noun
categorization device can develop into a closed grammaticalized system;
then this system may become less semantically transparent. Noun classi-
fiers grammaticalize in this way, and become noun class or 'gender'
markers (the hypothesis that noun classifiers, or class terms, could have
been reinterpreted as numeral classifiers in Thai has been put forward by
DeLancey 1986: 444—5; see also Bisang 1993; forthcoming). This develop-
ment is also linked to the emergence of agreement (see §13.8). Following
the general principle of unidirectionality in grammaticalization, the reverse
development never occurs.
This scenario has been suggested for Proto-Australian (Dixon 1982: 171;
Sands 1995: 253, 285-6). Australian languages show a continuum of gram-
maticalization: from generic nouns in noun classifier function to noun class
agreement markers on modifiers and verbs. Thus, some noun class markers
are less advanced on the grammaticalization continuum from semi-lexical
noun classifiers to noun classes than others (I return to this issue in
Chapter 15).
The most instructive example is the behaviour of the class of 'vegetable
food' which ranges from a generic noun to a grammaticalized noun class
marker (Sands 1995: 285-6). This passage from an optional generic in a
noun phrase to an obligatory and sometimes semantically opaque prefix or
suffix can be shown as a grammaticalization chain with three stages.
Grammaticalization chains of this sort can be recovered if related
languages are compared (cf. Heine 1992; 1997b).
Stage 1. Generic-specific pairings as a discourse device. In Guugu Yimidhirr
and many other languages generic mayi may be used in generic-specific
constructions for vegetable foods (Haviland 1979).
Stage 2. The generic noun comes to be used as an obligatory component of
noun phrases as a noun classifier omissible under specific conditions. For
instance, in Yidiny (Dixon 1977; 1982) mayi is one of the set of noun
classifiers used with specific terms; it can be omitted under certain conditions
(see Chapter 3).
Stage 3. The noun classifier is grammaticalized as a noun class affix; then
it undergoes phonological reduction. There are two possibilities.
(a) In Dyirbal the suffix -m 'vegetable food' attaches to deictics and
determiners as an agreement marker.
(b) In north Australian languages of the prefixing type, a prefix ma- (or
mi-, or m-) 'vegetable food class' is attached to the head noun, and is also
 Development of Noun Categorization 373
used as an agreement marker, e.g. Ngandi (Heath 1978b). The semantics of
this noun class may be widened, as in Yanyuwa, where it covers non-flesh
food, firesticks, articles made of fibrous materials, etc. In Wardaman the
ma- class includes all flora (Merlan 1994: 60; see (B) in §13.9.2). In Olgolo
phonologically reduced generic noun classifiers are being transformed into
overt prefixed markers on nouns. They are optional, e.g. y- is based on uyu-
'fish' and n- on ina- animal'. This evolution of noun-class prefixes from free
classifier forms has a phonological motivation: the language is eliminating
vowel-initial words (which appeared as the result of consonant-initial
dropping) (Dixon 1982: 207-10). There are no indications that noun
classes tend to develop first as derivational markers (head classes), or
as agreement markers (see §13.8). In Zande (§13.1.1) and Dyirbal gender
markers never developed a derivational function; they did in Ngandi and
in Wardaman.
Changes in noun categorization can involve evolution from covert to
overt noun class marking. In African languages an additional noun class
marking affix is sometimes added to the noun to 'strengthen' the overt
class marking (see Greenberg 1978). This results in a renewal of overtly
marked head classes.12 In the Southern branch of West Atlantic lan-
guages noun class marking suffixes on the head nouns are used to
reinforce the prefixes of the old noun classes. In these languages (Childs
1983: 25-6) the 'prefix erosion' was followed by introduction of noun
class-marking suffixes which, at least in some languages, develop from
pronouns (following the scenario outlined in §13.2 above; see also
Greenberg 1978).
The development of a new agreement target can result in the creation
of a new classifier type, through extension of existing classifier mor-
phemes to another function. This can be triggered by areal diffusion
(§13.7). The development of a new agreement class may be achieved by
substitution of independent pronominal forms for the affixed pronominal
forms. If new 3rd person pronouns have an animacy distinction this is a
way of setting up a gender agreement system (cf. Nichols 1992: 141). A
noun categorization system can undergo expansion via grammaticaliza-
tion of repeaters (see §13.5.1) or reanalysis of other nominal grammatical
categories (§13.5.2) (further changes in noun categorization systems are
discussed in §13.9.2).

12
This phenomenon is also attested in Bantu languages with pre-prefixes (Greenberg 1978:
66-7). In many Bantu languages, the nasal prefix of the human noun class became fused with
the stem, and another prefix was added, e.g. ba-n-tu (NCL2:pt-NCL-person) 'people'.
374 Classifiers
13.5. Grammaticalization13 and reanalysis in noun
categorization systems

The development and reanalysis of noun categorization devices involves

grammaticalization, and often polygrammaticalization, of lexical and
grammatical items. Grammatical closed classes such as demonstratives or
anaphoric specificity markers tend to grammaticalize into markers of
closed agreement classes. Lexical classes tend to grammaticalize into larger
classes of classifiers. These paths conform to the principle of unidirection-
ality of grammaticalization. Grammaticalization in the development of
noun categorization is discussed in §13.5.1, and in §13.5.2 I deal with
reanalysis in noun categorization.

13.5.1. Grammaticalization in the development of noun categorization

Grammaticalization focuses on how grammatical forms and constructions
arise, how they are used, and what the most probable paths for their devel-
opment are (cf. Hopper and Traugott 1993: 1-2). The development of noun
categorization involves grammaticalization of members of open lexical
classes (nouns and verbs, discussed in §13.1), and of closed classes, e.g.
demonstratives or locative adpositions (§13.2). Grammaticalization follows
the principle of unidirectionality (Bybee et al. 1994: 17; Heine et al. 1991):
once an item gets grammaticalized the process usually cannot be reversed.
Grammaticalization of members of open classes as classifiers often
involves a grammaticalization chain (Heine 1992; 1997b), from a lexical
item to an element of a closed class, and then to a grammatical marker.
When lexical nouns get grammaticalized as markers of noun class they
assume a pronominal (anaphoric) or a deictic function. In Zande (East
Adamawa: Claudi 1985; Heine and Reh 1984: 222-4; §13.1 above) full
gender markers arose via pronominalization of nouns. The masculine
pronoun ko in Zande comes from a noun *ko 'man, male' (which is still
used as an independent noun in related languages, e.g. Banda ka, Ngala
kwa). Zande is developing gender agreement on adjectives in predicative
use, as shown in 13.8.
13.8. k5-ni baksre-ko
he-be big-he
'He is big.'
13
I use the term 'grammaticalization' with the sense of'the attribution of a grammatical
character to a formerly independent word' following Meillet (1912: 132). See also the defini-
tion by Heine (1997a: 76), who considers grammaticalization as a process 'whereby a linguistic
expression E, in addition to its conventional meaning M,, receives a more abstract and more
grammatical meaning M2'.
 Development of Noun Categorization 375
In Zande, the same items are used as pronouns, and as gender markers.
In some Mayan languages noun classifiers derived from nouns are also
used anaphorically as pronouns (see §3.2.4; and see England 1983: 155, for
Mam; Craig 1977: 177, for Jacaltec). Gender markers *lo 'masculine' and
*na 'feminine', reconstructable for Proto-Eastern Nilotic 'are likely to be
derived from relational nouns (*IV 'member/person of, and *nyaa- 'girl,
daughter' which formed the head of genitive constructions before they
developed into pre-nominal modifiers *lo and *na (Heine and Reh 1984:
219)); they became agreement markers on adjectives in some languages,
e.g. Ongamo. When a noun class system is established, additional genders
can be acquired through grammaticalization of other anaphoric elements.
Maasai has an incipient third gender which originates in a reanalyzed
locative phrase (Tucker and Mpaayei 1955: 15—33; Heine and Claudi
1986: 43-51; Corbett 1991: 313-14).
Different stages of grammaticalization processes may (but do not have
to) coexist in a single language. We have seen in §13.4 that in some
Australian languages noun classes derived from noun classifiers—used as
anaphoric pronouns—coexist with them, but in others it is not the case.
If a set of lexical items becomes classifiers of different kinds, this will
involve polygrammaticalization (Craig 1993; Heine 1992).14 Body parts can
undergo polygrammaticalization in two ways. They can be used simulta-
neously as numeral classifiers (both sortal and mensural), and as quanti-
fiers. Mensural classifiers are often derived from body parts in
Minangkabau (see (A) under §13.1.1), e.g.jari 'finger', a mensural classifier
used for measuring cloth (Marnita 1996: 131). There is also a sortal
classifier derived from a body part: ikua 'tail' is a numeral classifier used
for all animals (Marnita 1996: 107). A few other body parts are used as
specific sortal classifiers, e.g. kaki 'leg', for umbrellas; tangkai 'stem, stalk'
for flowers with a stalk. The development of sortal classifiers out of body
parts involves metonymy in most cases. In languages with multiple classi-
fier systems different body parts can get grammaticalized in different
functions. For instance, in Palikur 'hand' and 'mouth' are used as numeral
classifiers (see examples 13.2 and 13.3 and discussion of their ambiguous
status as sortal and mensural classifiers). Kig 'nose' is grammaticalized as a
verbal classifier, and as a locative classifier. One item, kat 'vertical', is
derived from a word for 'trunk, stem of a plant', and is used in two
functions: as a numeral classifier and as a verbal classifier (with stative
verbs only). See §13.3, on the polygrammaticalization of mat3 'sphere' in
Chinantec (Foris forthcoming).

14
Polygrammaticalization implies the passage of lexical item 1 to grammatical functions A
and B; e.g. in Rama (Chibchan: Craig 1991: 487), the verb 'go' gets grammaticalized as a
purposive marker and as a marker for a tense-aspect choice.
376 Classifiers
GENERIC NOUNS can undergo polygrammaticalization; in this case they
give rise to noun classifiers and to noun class markers. A generic noun
which undergoes polygrammaticalization can become a bound morpheme
and at the same time can still be used as a noun classifier. In Ngan'gitye-
merri miyi is used as a noun classifier; there is also a prefix mi- used for
vegetable food (see further discussion in §13.8 on how this correlates with
the development of agreement).
Another grammaticalization path in the development of classifiers is
from repeaters, as an almost open class, to a more closed class of classifiers.
In Tariana (North Arawak: Aikhenvald 1994a) repeaters become extended
to items other than themselves; for instance, turapa 'cone' can be used to
classifiy not only the same noun, but also any other object which is cone-
like in shape, e.g. turapa hanu-turapa (cone big-NCL.REPEATER:CONE) 'a big
cone'; yen hanu-turapa (basket big-NCL:CONE) 'a big cone-shaped basket'.
In Kilivila (Senft 1993a; 1996: 353) some repeaters undergo phonological
depletion and are extended to items other than themselves.
Grammaticalization of nouns as classifiers often goes together with their
phonological reduction (in harmony with the Parallel Path hypothesis,
according to which semantic change in grammaticalization goes together
with phonological change: By bee et al. 1994: 19). The development of noun
classifiers from independent generic nouns in Mixtec involved 'abbrevia-
tion' of generic nouns (Pike 1949: 129-31; De Leon 1987: 148-50), in the
course of change from a syntactic compound to procliticized classifier-
plus-nouns. This is shown in Diagram 13.3.

Syntactic compound »- Proclitic classifier + noun »- Meaning

kiti-kwaa ti-kwaa
ANIMAL, ROUND.OBJECT-yellow NOUN.CL:ANIMAL, ROUND.OBJECT-yellow 'orange'
zilo-tichi to-tichi
TREE-avocado NOUN.CL:TREE-avocado 'avocado tree'
DIAGRAM 13.3. Phonological reduction in the development of noun classifiers in
Mixtec

In Daw a generic classifier plus specific noun constitute one phonological

word, with the classifier losing its independent tone (a similar phenomenon
is attested in Tucano: Ramirez 1997, and Tariana: Aikhenvald in prep.,
spoken in the same linguistic area). Different noun classifiers can show
different degrees of phonological depletion. In Mam (Kanjobalan Mayan:
England 1983; Table 13.3 above) some noun classifiers preserve the same
form as nouns (e.g. nu7xh 'NOUN.CLIBABY', nu7xh 'baby'); and some undergo
reduction (e.g. xnuq 'NOUN.CL:OLD.MAN:RESPECTFULLY', xiinaq 'old man').
The tendency appears for noun classifiers to become monosyllabic. In
Chinantec (Foris forthcoming) phonological reduction is obligatory for
 Development of Noun Categorization 377
some noun classifiers, but not for others: the derivation of mi3
'NOUN.CLtSPHERICAL.OBJECTs' from mat3 'sphere' involves vowel elision; while
vowel and tone change in mu 2 I /mt1 'NOUN.CL:FLAT' from mu21 'leaf is
optional. In contrast, numeral classifiers do not undergo phonological
depletion, e.g. mat3 'NUM.CL:SPHERICAL.OBJECTS', from mat3 'sphere'. These
examples show that phonological depletion is a tendency rather than the
rule in grammaticalization.

13.5.2. Reanalysis in noun categorization

New noun classes often arise as the result of reanalysis of another gram-
matical category of the noun. These categories are typically the ones with
which noun classes tend to interact (see Chapter 10). Noun classes often
interrelate with number; the choice of number frequently depends on the
animacy of the noun. In the Kiowa-Tanoan languages nouns have an
inherent number which conditions the assignment of noun classes (Watkins
1984: 78, p.c.; §10.1; this is called 'spontaneous generation of categories' by
Nichols 1992: 141). And see Corbett (1991: 198-9) on how new singular-
plural pairings gave rise to new genders in Andi dialects (Northeast
Caucasian).
The creation of a new agreement noun class via reinterpretation of
number may go together with the reduction of a proto-system. Grebo, a
Kru language from Africa, underwent a drastic reduction of the Proto-Kru
system. Of four original singular classes, only two remain. The proto-
human class *c now also includes humans and important and valuable
items, while proto-non-human *e is restricted to small, unimportant, or
worthless things (Marchese 1988: 329). In the plural there is an opposition
of human (o-) and non-humans (e-). The disintegration of the Proto-Kru
class system resulted in the creation of a 'new' singular-plural pairing and,
consequently, of a new set of agreement classes—see Diagram 13.4.

Singular Plural
Humans
Important valuable non-humans
Small worthless things
DIAGRAM 13.4. Singular and plural noun classes in Grebo

CASE forms can be reanalysed so that new noun classes emerge. Slavic
languages innovated a new subgender based on animacy. The development
of a new declension pattern was due to the necessity to distinguish subject
from object in those languages where constituent order was not used for
marking grammatical relations (Comrie 1978). For most masculine nouns,
the distinction between nominative and accusative was lost in the singular,
378 Classifiers
and then the extension of genitive to accusative function allowed the
distinction to be reintroduced. This happened first with masculine human
nouns, and was then extended to all animate nouns (see Corbett 1991: 99).
Other devices used for marking grammatical relations can be reinterpreted
as gender markers. In Pre-Proto-Northern Iroquoian cross-referencing
prefixes developed a human/non-human opposition (Chafe 1977: 505).
New gender oppositions can be created as the result of the reanalysis of
DERIVATIONAL morphology. Indo-European languages originally had two
genders: animate and inanimate (as preserved in Hittite). The feminine
gender appeared later within animate through reanalysis of feminine
derivational affixes -a: and *-i: (Meillet 1964; Brosman 1979, Melchert
1994).
There may be an expansion of already existing derivational suffixes. The
South Dravidian subgroup of Proto-Dravidian (Krishnamurti forth-
coming) innovated a gender distinction in personal pronouns by extending
the feminine derivation suffix *-al which already existed in nominal
morphology to the underlying pronominal root *aw, then the originally
non-masculine human and non-human *atu was restricted just to non-
human. Diagram 13.5 shows the Proto-Dravidian gender system (in
singular number), and Diagram 13.6 shows the modified system in
Proto-South-Dravidian.

Proto-form *awantu *atu

Meaning masculine human non-masculine human, non-human
DIAGRAM 13.5. Gender in Proto-Dravidian (singular)

Proto-form *awantu *awal *atu

Meaning masculine human feminine human non-human
DIAGRAM 13.6. Gender in Proto-South-Dravidian (singular)

In Spanish, the reanalysis of a dialectal masculine form of the article lo

may have given rise to the new neuter gender which is used for abstract
nouns, e.g. lo hermoso 'the beautiful' (Posner 1966: 134). Reanalysis and
subsequent grammaticalization can follow more complicated routes. In
Yanyuwa (Australian), a noun classifier for edible vegetable class became
a noun class prefix, as in other Australian languages. However, it involved
an unusual case of reanalysis and subsequent grammaticalization.
Yanyuwa is an ergative language; nouns in the function of the transitive
subject take an ergative case suffix -ngu. When the 'food' class prefix ma- is
added to a noun in ergative form it has the form murjgu- (with vowel
assimilation). We can hypothesize that originally a transitive subject NP
consisted of generic mayi 'vegetable food' (the source for class prefix
ma-) + ergative case rjgu, and a specific noun + ergative case, and then
 Development of Noun Categorization 379
mayi+-ngu was reinterpreted as a class marker, and underwent phono-
logical reduction (Dixon forthcoming).
Verbal classifiers emerged as the result of reanalysis in Imonda (Papuan:
W. Seiler 1983; 1986). This language developed verbal classifiers out of
desemanticized and often phonologically depleted first components of
serial verb constructions (see §13.1.3). The classifier fa which is used as a
generic classifier, arose as the result of reanalysis of the first syllable of the
verb fa-ka 'give', and then became generalized as a classifier (W. Seiler
1986: 194-5).
When one kind of noun categorization device is lost in the evolution of a
language it may be replaced by another, via reanalysis. Bengali lost its
grammatical gender and number marking due to phonological coalescence
of paradigms (Kolver 1982b; §13.6) and acquired numeral classifiers using,
in part, the same morphological 'material'. The feminine gender marker -i
in Old Indo-Aryan and New Indo-Aryan (see Chatterji 1926: 672 ff.;
Kolver 1982b) became reanalysed as a diminutive used with numeral
classifiers.15

13.6. Reduction and loss of noun categorization devices

Reduction and loss of noun categorization can be due to language-internal

reasons. Noun classes can get reduced, and even lost—due to the coales-
cence of paradigms—for morphological and/or phonological reasons. This
is typically found in fusional languages. Loss of noun classes is often
accompanied by loss of number or case distinctions with which noun
classes interacted.
In Indie and Iranian languages the masculine and feminine declensional
paradigms merged. This resulted in a complete loss of gender oppositions
in Assamese, Bengali, Nepali, Oriya, Persian, Beludzhi, and Ossete (see
Kolver 1982b, for the history of Bengali). In the history of French, mascu-
line and neuter endings merged in the period between the end of the fifth
century and the end of the eleventh (Pope 1952); as the result, neuter
gender was lost.16

15
In other instances, the material used may be completely different, but the traces of the
original system could still be felt. Kana (Kegboid, Cross River) lost a Proto-Cross-River noun
class system, but acquired a typologically unusual set of numeral classifiers (see §4.2.3). We are
not aware of a whole system of genders, or noun classes, developing into a classifier system. We
predict that this is not likely to happen, since such a development would contradict the
principle of unidirectionality in grammaticalization.
16
The majority of neuter nouns became masculine (Posner 1966: 134-5). Certain Latin
neuter nouns became masculine in some Romance languages and feminine in others. So, mare
'sea' became feminine in French la mer but masculine in Italian il mare, and Portuguese o mar
(Posner 1966: 135). Even in Latin these nouns are sometimes treated as neuter, and sometimes
380 Classifiers
Morphological causes may contribute to the restructuring and loss of
gender classes. Old English lost gender classes partly due to the conver-
gence of inflections (genitive singular -es, originally only masculine and
neuter, spread to feminine; and nominative-accusative plural -as, originally
only masculine, spread to feminine and neuter) (see Dixon 1982: 171, and
references therein). The loss of noun classes in Jukun (Central Jukunoid
branch of Platoid; Benue-Congo, Niger-Congo: Storch 1997) was
conditioned by word stem apocope and morphological reduction.
Coalescence of paradigms and the ensuing collapse of genders in
Indo-European languages could be speeded up by the semantic opacity
of gender. In most cases, the first gender to be lost appears to be neuter.
The Sele Fara dialect of Slovene completely lost its third gender, neuter, in
favour of masculine (Priestly 1983). The main mechanisms were phono-
logical and analogical. The reduction of final unstressed -o, a nominative
singular marker on neuter nouns, to 3 followed by the regular loss of this
short vowel accelerated their reinterpretation as masculine singular, e.g.
*mesto > masts > mast 'town'. The nominative-accusative plural neuter
ending in -a was reinterpreted as belonging to the feminine gender; later it
was 'masculinized', too, by analogy with the masculine accusative plural.
The neuter gender disappeared completely within three decades (Priestly
1983: 353-4).
Loss of genders can be due to the loss of formal markers. A number of
North Arawak languages lost gender-sensitive pronominal cross-referen-
cing suffixes, and this must have contributed to the overall loss of gender
distinctions in Bare and Bahwana (Aikhenvald 1995a; Ramirez 1992).
Gender reduction may go together with the expansion of a functionally
unmarked gender. In Kore, an Eastern Nilotic language, feminine is the
unmarked gender. The feminine demonstrative has been generalized to
refer to both masculine and feminine nouns (Gerrit Dimmendaal, p.c.).
A similar expansion of the functionally unmarked masculine class in
Mayali (Evans 1997 and forthcoming) has led to the loss of agreement in
one dialect of this language.
In languages with a distinction between noun classes on nouns and on
pronouns, pronominal noun classes tend to be more conservative (Heine

as masculine. In Spanish 'sea' can be either masculine or feminine, with a subtle change in
meaning: el mar means 'the sea as something difficult or foreboding', and la mar describes the
sea as something pleasant and peaceful (Lyle Campbell, p.c.). In the history of Old French the
neuter gender merged with masculine very early; some nouns became feminine (e.g. some
plural neuter forms ending in -a were 'mistaken' for feminine, e.g. Latin arma 'arms' which
became feminine in all Romance languages: Posner 1966: 134). Most changes occurred as the
result of morphological and semantic analogy; thus, abstract nouns in -orem became feminine
under the influence of other abstract nouns. Tree names in -us became masculine following
masculine gender assignment of the ending -us; later arbor 'tree' also became masculine (Pope
1952: 304).
 Development of Noun Categorization 381
1982a: 214). Animere, a Togo Remnant language from Eastern Ghana,
distinguishes two genders in personal pronouns; nominal genders have
been lost (Heine 1982a: 207-8). Some Arawa languages of Southern
Amazonia have lost the shape-based ka- noun class (see §2.7.2, for the
discussion of Paumari); but all of them still distinguish masculine and
feminine genders. Pronominal genders tend to be more semantically trans-
parent (see Chapter 2), possibly because pronouns frequently refer only to
humans (this may be one of the factors which favour the retention of
gender distinctions). (See §13.8.2, on the different ways in which noun
class and gender agreement undergo loss depending on the target.)
Traces of old gender distinctions may 'survive' somewhere else in the
grammar. Bengali reinterpreted the old feminine form in terms of its
system of numeral classifiers (also see Barz and Diller 1985; Masa Onishi,
p.c.). Kegboid languages, a subgroup of the Benue-Congo family spoken in
Nigeria, lost Proto-Benue-Congo noun classes but acquired a peculiar
system of numeral classifiers, an unusual phenomenon for African
languages (Ikoro 1996a). A gender system can survive in the form of
derivational-type affixes, as in the Australian languages Gunbarlang,
Limilngan, Rembarrnga, and Nungali (Sands 1995: 269 ff.).
When classifier systems other than noun classes get reduced, classifiers
for animates and humans appear to be more stable than others. Oceanic
languages have a tendency to lose numeral classifiers. Fijian has only one
remainder of a former system of numeral classifiers: lewe, used for counting
humans (Churchward 1941: 44; Dixon 1988: 148). A comparison of tradi-
tional Malay classifiers and classifiers in Bahasa Indonesia (Conklin 1981:
211) shows that the human classifier (orang) and the animal classifier (ekor)
are the most stable. Specific classifiers tend to disappear first. Classifiers for
inanimates based on shape, function, and material tend to be replaced by
the generic classifier buah previously used for round objects (Marnita 1996:
148–55; 165-6); specific classifiers, such as lempeang 'flat piece, tradition-
ally used for pieces of tobacco, gold and dirt', are replaced by shape-based
classifiers, such as alai 'flat'.
The system of noun classifiers became reduced in Acehnese (Durie 1985:
134–5) (especially if compared to the closely related Minangkabau:
Marnita 1996); however, classifiers for person, plants and plant parts, birds
and fishes are retained. (This is reminiscent of some recently acquired
systems of noun classifiers, as in the case of Daw, or Awa Pit. Noun
classifiers are obligatory with persons and plants, and optional with other
semantic groups.)
Loss of classifiers of other types has also been observed. However, it is
not clear whether any of these were lost for phonological or morphological
reasons. Many Athabaskan languages lost the verbal classifiers (tradition-
ally referred to as 'genders') d 'round objects' and n 'long objects'. Only
382 Classifiers
Koyukon, Dena'ina, Tanana, Ahtna and Carrier retain productive verbal
classifiers (Thompson 1993; Kari 1990; Bill Poser, p.c.; see above n. 10 to
Chapter 6).
Many Oceanic languages have reduced the sphere of use of relational
classifiers; some have eventually lost them. In Longgu, an Oceanic lan-
guage spoken in the Solomon islands (Hill 1992), the relational classifiers,
'edible' and 'drinkable', are used only for predicative possession; even in
this context they are frequently omitted (for an overview of systems of
relational classifiers, see Lichtenberk 1983a). Many Oceanic languages lost
*ma 'classifier for drinks', with *ka 'the classifier for food' replacing it. In
some languages it happened the other way round: *ma replaced *ka as a
marker of alimentary possession in general, as in Tabar, an Oceanic
language spoken off New Ireland (Lichtenberk 1985: 118). These processes
resulted in restructuring the composition of classes; some relational classi-
fiers got lost as a result (Lynch 1992: 19-20).
According to Lynch (1992; 1993), one of the factors favouring simplifi-
cation and, ultimately, even loss of relational classifiers could have been
contacts with Papuan languages, especially in the case of the Austronesian
languages spoken in the Solomons and in New Ireland (see (A) under
§13.7.1). Takia (Karkar island: Bruce Waters, p.c.) has two sets of posses-
sive affixes which are used interchangeably, thus having lost a distinction
between relational classifiers.

13.7. Language-external motivations for the development and

decay of noun categorization

Language-external motivations for the emergence of noun categorization

devices include language contact, pidginization, creolization, and language
obsolescence. The creation and restructuring of noun categorization often
result from language contact (§13.7.1), while creolization and language
obsolescence more often lead to their loss and reduction (§13.7.2 and
§13.7.3). The impact different language external motivations can have on
noun categorization is considered in §13.7.4.
Areal diffusion—in addition to direct borrowing—is often explicable by
language contact in a bi- or multi-lingual situation. Borrowing usually
takes place from a prestige language into languages with lesser prestige.
This is an explanation offered for the penetration of Spanish into Quechua;
Tucano into Tariana, or Malinke (Mande) loans into the minority
language, Badyaranke (Northern subgroup of West Atlantic, spoken in
Guinea Bissau: Ducos 1979). Or the linguistic group which provides the
source of borrowing may just be more numerous, without having more
prestige; this is the case of Baniwa and Kubeo discussed under A2 in (A),
 Development of Noun Categorization 383
§13.7.1. Direct diffusion is often limited because of prohibitions against
lexical loans, as in the Vaupes region (see Aikhenvald 1996a). This is not
the case in Australian languages.
Areal diffusion in Australian languages (as described by Heath 1978a;
Evans forthcoming, 1997; Harvey 1997; Dixon 1997; forthcoming) could
be of an entirely different nature. There, the linguistic systems have been in
close contact for a long time, in a situation of equilibrium, with no special
prestige or other social pressures. A different kind of reason for the
borrowing of Bantu class prefixes into Ndunga-le—their phonological
salience—was suggested by Pasch (1988: 59).n

13.7.1. Language contact and noun categorization

Structural changes induced by language contact may result in indirect
diffusion of patterns and categories. Another possibility is straight borrow-
ing of forms (see the notion of indirect diffusion vs. direct diffusion, or
borrowing, in Heath 1978a; and the distinction between contact as a
situation, and borrowing as a process in Harris and Campbell 1995:
122). Direct diffusion is more rarely the principal reason for the emergence
of new noun categorization devices—or for the restructuring of the already
existing ones—than is indirect diffusion. Indirect diffusion will be consid-
ered under (A), and direct diffusion, or borrowing, under (B) below.
(A) Indirect diffusion and noun categorization
Noun categorization devices can emerge under areal influence. Or an
already existing system of noun categorization devices can change in a
certain way, i.e., undergo reduction, augmentation, or restructuring under
areal influence (cf. Nichols 1992: 139; Bisang 1996).
(Al) Creation of noun categorization systems The more lexico-syntactic
the noun categorization is, the easier it is to diffuse. That is, the most
frequently diffused new patterns involve syntactic constructions with clas-
sifiers which have a clear lexical source rather than with more grammati-
calized noun class agreement systems. Possessed classifier constructions of
the type generic noun plus possessed noun were borrowed from Carib
languages into a number of North Arawak languages, e.g. Palikur, Island
Carib, and Bahwana (see Aikhenvald and Green 1998). In the South Asian
17
Recently a number of universal claims have been proposed concerning grammatical
borrowing (see the exposition and criticism in Campbell 1993; Harris and Campbell 1995:
120-41). The data on language external sources for noun categorization systems confirm some
of these principles, e.g. the structural compatibility requirement. However, some data dis-
cussed below provide a counterexample to these universal tendencies, e.g. the example from
Ayacucho Quechua where gender agreement has been borrowed goes against the statement
that 'free-standing grammatical forms are more easily borrowed than bound morphemes'.
384 Classifiers
linguistic area numeral classifiers have been shown to have spread from the
Indo-Aryan languages of the region to Dravidian languages (Emeneau
1964: 644-50).
Indirect diffusion of closed grammatical systems is rarer. Kakua, a
Makii language spoken in the Vaupes region, is the only language in the
family to have a feminine/non-feminine gender distinction in 3rd person
independent pronouns and subject prefixes. This language underwent
strong areal influence from neighbouring East Tucano languages; it is likely
that Kakua acquired gender under areal influence from East Tucano
(Martins and Martins 1999). The Oceanic languages spoken in Southwest
New Britain are likely to have acquired incipient gender distinctions under
the influence of surrounding Papuan languages (Malcolm Ross, p.c.).18
Noun categorization devices are often so much a feature of a linguis-
tic area that it is difficult to establish the direction of diffusion—i.e.
which languages got them first. This is the case in Mesoamerica (Campbell
et al. 1986: 550), in the Caribbean area (Costenla Umana 1991), and in
the linguistic area of East and Mainland Southeast Asia (Bisang 1996).
Shape-based genders in particular, and other kinds of gender marking,
are likely to be an areal feature of the Sepik river basin in New
Guinea (Foley 1986: 265); however, the source of diffusion is impossible
to establish.19
(A2) Restructuring of noun categorization Indirect areal diffusion may
result in the partial restructuring of classifier systems. This involves intro-
ducing new classifier types into a system which already has classifiers.
Kubeo, a Central Tucano language spoken in a community bilingual in
an Arawak language, Baniwa, uses classifiers with verbs following the
pattern found in Baniwa (Gomez-Imbert 1996: 451).
Tariana extended the use of classifiers to demonstratives under the
areal influence of East Tucano languages in the context of obligatory
multilingualism in the Vaupes river basin. In Arawak languages (for
example, the closely related Baniwa) spoken outside the Vaupes area,
demonstratives distinguish only feminine and non-feminine forms. Example
13.9 is from Tariana. A classifier of a repeater type panisi 'house' is
attached to the demonstrative. The same technique can be observed in
13.10, from Tucano.

18
It is possible that Eastern Nilotic languages (Turkana, Teso, Lotuko, Maa) acquired their
gender structures due to areal influence of neighbouring Niger-Congo languages (Heine
1982a: 215).
19
In these cases it is hard to decide whether shared classifiers or genders are due to areal
diffusion or are inherited. This throws doubt on the suggestion by Nichols and Peterson (1998:
612) that 'the presence of numeral classifiers is among the genetically most stable of features'
(cf. also Nichols 1997).
 Development of Noun Categorization 385
13.9. ha-panisi
DEMtlNAN-HOUSE
'this (house)'
13.10. ati-wi'i
DEM:INAN-HOUSE
'this (house)'
In contrast, Baniwa uses a non-feminine form of a demonstrative, as in
hliehe panfi (DEM:NF house) 'this house.' A similar expansion of classifiers
onto demonstratives took place in Resigaro, another North Arawak lan-
guage, which is in close contact with Bora and Ocaina (Bora-Witoto
family). Like Tucano languages, but unlike Arawak, Bora also has classi-
fiers used with demonstratives (Thiesen 1996; Allin 1975; Aikhenvald
forthcoming c).
Reduction and augmentation of gender and noun class systems often
happen through diffusion in Australian languages. In Mayali (Arnhem
Land, Australia) the southwestern dialect has become more like its neigh-
bour Jawoyn, with only three instead of four agreement classes, and the
easternmost dialect has become more like their neighbours Rembarrnga
and Dangbon, both of which lack noun class agreement (Evans 1997; cf.
Harvey 1997).
Changes in the composition of noun classes, or in classifier assignment
can be accounted for by areal diffusion. In Tariana all animals are classi-
fied as 'animate non-feminine', under the influence of the Tucano lan-
guages. In Baniwa, Tariana's close relative, animals are classified
according to their shape (e.g. a jaguar is classified as vertical, and a bird
is classified as concave, or banana-shaped). Kubeo is a Tucano language
which has been in close contact with Baniwa for a long time; in this
language animals have come to be classified according to their shape,
just like in Baniwa (Gomez-Imbert 1996).
Language contact can affect principles of gender assignment, without
changing the number of genders or their composition. For example, in
Western Oromo dialects the system of gender assignment became semanti-
cally transparent (with feminine as the gender for diminutives and mascu-
line as the unmarked one) under the influence of other Lowland East
Cushitic languages, and of Arabic (Clamons 1995).
A language can acquire a system of noun classes via grammaticalization
of classifiers under areal pressure. Yanyuwa (Australian) is spoken on the
edge of the region whose languages have prefixes and noun classes; both
features have recently diffused into Yanyuwa (Dixon forthcoming). It has
acquired a system of seven noun classes marked by prefixes (Kirton 1971:
20, 38). At least some of these are the result of typical Australian
386 Classifiers
grammaticalization of noun classifiers into noun class markers, e.g. ma-
'food', from mayi 'edible vegetable' (see §13.4).
(A3) Reduction of noun categorization Language contact can result in the
reduction of noun categorization. The Minangkabau spoken by the
younger generation tends to reduce the system of numeral classifiers to
make it similar to Bahasa Indonesia in which virtually all Minangkabau
speakers are bilingual. The number of numeral classifiers commonly used
in Bahasa Indonesia (as compared to traditional Malay) has been drasti-
cally reduced during the course of this century, after it became the national
language, possibly, due to the influence of Javanese and English (Conklin
1981: 201; also see §12.3).
It has also been observed that mensural classifiers tend to undergo more
dramatic reduction in the speech of younger people than sortal classifiers.
This is due to the fact that mensural classifiers often refer to culture-
specific arrangements and measures, and the obsolescence of cultural
knowledge inevitably leads to their loss (Marnita 1996, for Minangkabau;
Lee 1997, for Korean; also see Sanches 1977, for Japanese).
An example of the loss of a multiple classifier system as the result of
areal diffusion comes from Retuara, a Central Tucano language which is in
contact with Yucuna, a North Arawak language. Yucuna has the typical
Arawak opposition of two genders, and a small system of numeral classi-
fiers. Under pressure from Yucuna, Retuara lost the typical Tucano mul-
tiple classifier system, preserving just a gender system (inanimate,
masculine and feminine) (Strom 1992: 45-50; Gomez-Imbert 1996: 445).20
(B) Direct diffusion and noun categorization
We have no example of a complete system of noun categorization being
borrowed. However, parts of a system have been borrowed in different
parts of the world.
According to Heath (1978a: 88), 'the spread of noun-class systems over
much of north-central and north-western Australia may well have been
largely accomplished through direct diffusion of the actual affixes, rather
than by independent development in each language group'. This is espe-
cially conspicuous in the cases where there is no internal etymology avail-
able for noun class affixes in a language; but these are available in a
neighbouring language. As Heath (1978a: 88) puts it, 'while in some
languages a correlation can be made between some noun-class affixes
and particular noun stems (for example, ma- as a non-human noun-class
prefix and a stem like mayi meaning "vegetable food"), this is not possible
20
Reduction of large classifier sets as the result of language attrition has been reported by
Wurm (1992a, for Reef-Santa Cruzan languages). These languages are in the process of losing
verbal classifiers but not classifiers in other contexts (Stephen Wurm, p.c.).
 Development of Noun Categorization 387
in Nunggubuyu, Ngandi and Warndarang. There is consequently a strong
probability that the noun-class systems in these languages were borrowed
from languages to the west.' Furthermore, Warndarang borrowed non-
human prefixes for three classes from Ngandi or from pre-Nunggubuyu
(Heath 1978a: 90-1).
Another example of a massive borrowing of noun class markers which
resulted in the restructuring of a system comes from Africa (Pasch 1988).
Ndunga-le, a Mba language from Ubangi, an Eastern subgroup of Niger-
Congo, has an original system of six singular and three plural noun classes.
There are over fifty non-borrowed nouns which also contain singular and
plural noun class prefixes borrowed from the neighbouring Bantu langu-
ages; in some cases the first syllable of a borrowed noun was reanalysed as a
noun class prefix (Pasch 1988: 58).
Resigaro, a North Arawak language which underwent drastic restructur-
ing under the influence of the genetically unrelated Bora, borrowed a large
number of grammatical morphemes (Aikhenvald forthcoming c). Bora
influence on the Resigaro classifiers involves borrowing of bound mor-
phemes, and grammaticalization of borrowed free morphemes as classifiers.
Resigaro has around fifty-six classifiers (Allin 1975: 154ff), of which
thirty-six have been borrowed from Bora. Some classifiers which
correspond to bound morphemes in Bora, e.g. Resigaro -git 'long and
flat', Bora -kwaa (classifier which appears in words for 'finger', 'toe');
Proto-Bora-Muinane -gat (Aschmann 1993: 131). Other classifiers in
Resigaro correspond to bound morphemes and to free morphemes in
Bora. In Resigaro, a classifier of a Bora origin may be attached to the
noun of Arawak origin, e.g. Resigaro classifier -mi 'canoe' in hiita-mi21
'canoe' (also used as a classifier in Bora: -mi 'canoe; other transport'—
Thiesen 1996: 102—and as a root in Bora mii-ne 'canoe', cf. Proto-Bora-
Muinane *mii-ne—Aschmann 1993: 136). Classifiers which correspond to
free morphemes in Bora include some for body parts, e.g. Resigaro -?osi 'cl:
hand', singulative -ke-?osi 'hand' (Resigaro -ke from Proto-Arawak *kapi:
Payne 1991a); Bora hojtsii 'hand', Proto-Bora-Muinane -?6xtsii 'hand'.
The extensive borrowing of classifiers in almost all the semantic fields can
be explained by the important role classifiers play in discourse: once the
referent is established it is referred to with a classifier, so that classifiers
appear to be more frequent in discourse than nouns themselves.22

21
Cf. the cognates of Resigaro hiitu/hiita in Arawak languages: Bare isa, Achagua iida,
Tariana ita(-whya), Baniwa ita, Piapoco Ida, Yucuna hiita.
22
Languages with no gender or noun class distinction may acquire a gender-marking
derivational suffix via borrowing and reanalysis. Finnish and Estonian acquired a feminine
derivational suffix, Finnish -ttarel-ttare, Estonian -tar from the reanalysis of the word for
'daughter' (tytar), itself reportedly borrowed from Baltic (Lithuanian/Latvian); this gramma-
ticalization may have been motivated by matching the feminine suffix in German (Lyle
388 Classifiers
Borrowing of an agreement system is extremely rare. One such example
comes from Ayacucho Quechua (see §2.4.4). A similar example comes from
Ilocano (Austronesian: Rubino 1997: 138-9). Similarly to many other
Austronesian languages, Ilocano does not have grammatical gender;
however, a large number of loan adjectives from Spanish resulted in the
creation of masculine and feminine distinctions in loan adjectives, e.g.
tsismoso 'gossipy' (masc.), tsismosa 'gossipy' (fem.), from Spanish chis-
moso, chismosa 'a gossip, gossipy'. Nouns borrowed from Spanish often
have two gender forms to distinguish the sexes, e.g. kosinero 'cook (masc.)',
kosinera 'cook (fern.)', sugalero 'gambler (masc.)', sugalera 'gambler (fern.)'
(Rubino 1997: 75).
Reduction of an agreement system and of noun class marking may be
due to massive borrowing. Ducos (1979) describes the loss of the Proto-
Gur noun class system in Badyaranke. This loss may be due to massive
lexical borrowing from Malinke, a Mande language which has been the
language of prestige among Badyaranke for 500 or 600 years. Lexical
borrowing of prefixless nouns led to the loss of derivational—and then
of agreement—functions of noun classes.

13.7.2. Creolization and noun categorization

The REDUCTION and LOSS of class/gender distinctions is a universal feature
of the pidginization and creolization of languages (Heine and Reh 1984:
42). Indo-European-based Creole languages do not have any gender dis-
tinctions; and there are no genders in the Nubi language of Kibera, an
Arabic-based Creole (Heine 1982b). In pidginized Hausa the loss of gender
led to the replacement of feminine gender markers by masculine ones
(which are functionally unmarked). Similar examples have been observed
in Kenya Pidgin Swahili (Heine and Reh 1984: 42; see also Alexandre 1968:
280). Noun class systems in Lingala and Zairian Swahili were simplified
due to the influence of non-Bantu languages and creolization processes (see
Bokamba 1977; Helma Pasch, p.c.); this simplification has to do with the
reduction of agreement domains (see §13.8.2). Kinshasa Lingala, which
originally developed as a trade language, is moving towards a unified plural
marking, whereby the plural marked with ba- (Class la) is extended to all
nouns, even those that belong to other classes. Kituba, a Creole spoken in
Campbell, p.c.). Limited borrowing of gender-marking derivational morphemes is also found
in languages with agreement gender systems (see Pasch 1988: 60 for more examples of
borrowing noun class affixes whereby they usually lose their primary meaning). The Berber
languages of Central Morocco have gender-differentiated compounds with the first component
borrowed from Arabic; the second component does not have to be borrowed, e.g. bu (from
Arabic bu 'father, man of. . .') in bu uham 'master of the house' (Laoust 1928: 14-15). All
gender-indicating suffixes in Turkish are borrowed from languages with grammatical gender,
such as Arabic (-e) and Slavic (-ife) (Braun forthcoming b).
 Development of Noun Categorization 389
southern Zaire and derived from Kikongo, is evolving just one plural
marking thus eliminating the traditional one-to-one singular-plural noun
class correspondence and contributing to the overall disintegration of the
noun class system (Stucky 1978). A similar situation is found in Fanagalo, a
Bantu-based creole spoken in Botswana (Anderson and Janson 1997: 183).
Most Creoles and pidgins have no numeral classifiers (even if they are
surrounded by classifier languages).23 The Russian-Chinese pidgin spoken
in Harbin (northeastern China) in the early twentieth century had no
numeral classifiers (Johanna Nichols, p.c.).
An instructive example of language change in the context of partial
creolization and language shift comes from Modern Tiwi spoken by about
1500 people on Melville and Bathurst Islands in the far north of Australia.
The Tiwi community has undergone rapid changes during the last few
decades through contact with European-Australian culture (Lee 1987: 9).
Traditional Tiwi, which is still one of the four language codes used in the
community, underwent a restructuring which is described as 'creolization'
as it became Modern Tiwi, or 'anglicised' Tiwi (Lee 1987: 10).24 Traditional
Tiwi distinguishes two genders, masculine and feminine. Gender is assigned
on the basis of sex and physical size (masculine referents are considered
narrow and small, feminine ones wide and large: see Table 11.1; gender
assignment is, however, not transparent, e.g. all trees are feminine whatever
their size) (Osborne 1974: 51). Young people and children tend to assign
gender on the basis of natural sex, and are inconsistent in gender assign-
ment to nouns other than those denoting humans and dogs (Lee 1987: 84).
The feminine form of an adjective and of possessive pronouns is used with
any referent, as the unmarked choice in a noun phrase (Lee 1987: 92-3,
109, 243).25 Modern Tiwi also tends to systematically distinguish just
human/non-human, instead of the erstwhile masculine/feminine.

13.7.3. Language obsolescence and noun categorization

Loss or REDUCTION of noun classes is often observed in language obsoles-
cence. These processes are often accelerated by patterns which exist in a
dominant language, due to bilingualism.
23
The only example of a Creole language with numeral classifiers appears to be Chacabano
Zamboangueno, a Spanish-based Creole spoken in the Philippines (Carl Rubino, p.c.).
24
Whether Modern Tiwi is an example of a creole or not is still an open question—see the
excellent analysis of this problem in Lee (1987: 355-7). She arrives at a preliminary conclusion
that Modern Tiwi appears to be 'a creolised form of an anglicised Baby (Talk) Tiwi'. 'In the
development of M(odern) T(iwi), children (and adults to a certain extent) have drawn upon
Tiwi, English, and Pidgin English, to produce a hybrid or amalgam' (Lee 1987: 357).
25
Another factor which might have contributed to some loss of gender agreement in Tiwi
could have been phonological reduction, as compared to traditional Tiwi (see Lee 1987: 118,
145).
390 Classifiers
Traditional Dyirbal had four semantically assigned noun classes (§2.4.1).
Young People's Dyirbal has gradually simplified its noun class system so
that it became similar to the way he, she, it are used in English. The noun
class referring to 'non-flesh food' was lost. The scope of noun Class 2 was
reduced and came to be reserved only for females (it used to include water,
fire, and things associated with fighting). Gender assignment by mythical
association was lost; exceptions became regularized; and the use of the
residue class was expanded (Schmidt 1985). Dahalo (Cushitic) is losing
ground to Swahili, and is accordingly losing the common Cushitic distinc-
tion of masculine and feminine gender (Dimmendaal p.c.).
Loss and restructuring of gender distinctions among semi-speakers in a
situation of language attrition has been observed for Scottish Gaelic
(Dorian 1981: 124-9, 147-8). Gaelic has masculine and feminine genders
marked by initial mutation of the noun after the definite article by agree-
ment of an attributive adjective (by initial mutation), by the form of the
definite article, and by the form of pronouns; there is also a gender-marked
diminutive suffix. In personal pronouns, there is a tendency to extend the
masculine pronoun a to all inanimate nouns (Dorian 1981: 125). Adjective
agreement via consonantal lenition shows considerable variation. Gender
marking on nouns remains very strong. However, since only a subclass of
nouns can be diminutivized, and there are phonological limitations of head-
marking gender lenition, it appears that even for fluent speakers the gender
system has become more limited in its application (Dorian 1981: 129).
Dorian (1981: 148) predicts that if Eastern Sutherland Gaelic had 'future
generations of fluent speakers', gender marking in pronouns and agree-
ment on adjectives 'would surely disappear', but head-marking gender on
nouns would survive, 'producing a kind of lexicalization of gender speci-
fication linked to specific nouns'. That head-marking noun classes appear
to be more persistent than noun class agreement, may be related to the
differences in 'speed' of obsolescence of grammar and of the lexicon; see
§13.8.2.
Language attrition does not necessarily result in loss of noun class.
Rather, an endangered language may try and assimilate its gender patterns
to the dominant language. In Arvanitika, an endangered dialect spoken by
Albanians who immigrated to Greece in the eleventh and fifteenth cent-
uries, the three genders remain distinct. This may be due to the fact that the
three-way gender distinctions in Arvanitika and Greek are structurally
similar (Trudgill 1977: 35; Sasse 1985). In contrast, the gender systems
of Gaelic and English, or Dyirbal and English, are quite different; the
result is a drastic reduction of the system in the obsolescent language.
Language obsolescence often results in the obsolescence of numeral
classifiers if the dominant language does not have them. Warekena (North
Arawak: Aikhenvald 1998b) is in the process of being replaced by Lingua
 Development of Noun Categorization 391
Geral (Tupi-Guarani) and by Portuguese; it has almost lost its numeral
classifiers. Some northern Australian languages display loss and obsoles-
cence of verb-incorporated and numeral classifiers due to language con-
tact. Modern Tiwi, as compared to Traditional Tiwi (Lee 1987: 318), has
lost 'virtually all incorporated forms', and, consequently, verbal classifiers,
as the result of an overall simplification of the verbal structure.
Partial language obsolescence may result in the restructuring of a
classifier system (see §13.7.1, on the reduction of classifier inventory in
Minangkabau to make it more similar to the dominant Indonesian
language).

13.7.4. Language-external motivations and their impact on noun

categorization
Language contact, creolization, and language obsolescence show a few
notable differences in their impact on noun categorization. Creolization
typically leads to the LOSS or drastic REDUCTION of noun categorization.
Language contact may lead to the CREATION of a new system, or to its
RESTRUCTURING. Language obsolescence results in LOSS or RESTRUCTURING of
noun categorization according to the pattern present in the dominant
language. While 'dying languages . . . show much the same sorts of change
we are familiar with from perfectly ordinary change in "healthy languages"'
(Dorian 1981: 151), they tend to restructure their noun categorization
devices according to the patterns found in the dominant language.26
Another distinguishing feature of change in language obsolescence and
creolization is its speed: drastic grammatical restructuring sometimes
happens within a short timespan (see Schmidt 1985: 213, on Young
People's Dyirbal); areal diffusion within linguistic areas is usually slower
(Aikhenvald 1996a).27

13.8. Development and loss of agreement

13.8.1. The genesis and development of agreement
The two types of agreement considered in §2.4, (a) agreement of the
predicate with its arguments, and (b) head-modifier agreement within an
NP, develop in different ways.
26
Development in language obsolescence also differs from normal historical development
for a 'healthy' language in that it is modelled on a dominant pattern while normal diachronic
change is not (cf. Campbell and Muntzel 1989; Sasse 1992).
27
In all the examples so far, the dominant language has had a smaller system of classifiers.
We have no clear examples in which it is the other way round; see discussion in Aikhenvald
(forthcoming c).
392 Classifiers
(A) Predicate-argument agreement
VERBAL agreement usually arises from an anaphorically-used independent
pronoun which is subsequently cliticized to the verb; it then becomes
incorporated as an agreement marker. The mechanism is basically prag-
matic, and it involves reanalysis and grammaticalization of independent
members of a closed class. This happened in Manambu (Ndu); see also the
examples in Lehmann (1982) and Corbett (1991: 137-9).28
Properties associated with discourse-topic NPs have also played an
important role in the development of subject-verb agreement in Arabic
(Russell 1984). There is no regular etymological relationship between the
verb agreement affixes and anaphoric pronouns in Semitic languages.
Person-, number-, and gender-sensitive agreement suffixes (used in the
perfect) seem to have originated from nominal and adjectival inflections.
Verb-argument agreement emerged as the result of the expansion of agree-
ment markers from head-modifier type constructions to other construction
types.29
(B) Head-modifier agreement within a noun phrase
For the development of gender agreement within a noun phrase two
mechanisms have been proposed by Heine and Reh (1984: 230-1). Agree-
ment may arise if demonstratives develop into subordination markers, e.g.
relative clause markers, as the result of their grammaticalization. Given
that relative constructions are an important strategy for forming nominal
qualifiers, they may become the main source for gender agreement. Example
13.11, from Maasai, an Eastern Nilotic language, shows how relative clause
markers are used to 'introduce nominal qualifiers' (Heine and Reh 1984:
231).
13.11. in-kishu na-a-pishana na-a-ibor
Pi-cattle REL-PL-seven REL-pt-be.white
'seven white cows'
In Eastern Nilotic languages gender marking did not affect personal

28
Grammaticalization of agreement goes together with a conflation of topic marking and
of grammatical marking of a constituent. Givon (1976: 151) remarks: 'when a language
reanalyses the topic constituent as the normal subject or object of the neutral, non-topicalised
sentence pattern, it perforce also has reanalyzed subject-topic agreement as subject agreement
and object-topic agreement as object agreement.' Bresnan and McChombo (1986) offer a
theoretical discussion of the relation between anaphora and agreement and suggest possible
steps by which such a historical sequence might take place.
Lehmann (1988: 59-60) suggests that agreement of the verb-argument type (external
agreement), and agreement of the head-modifier type (internal agreement) originate in differ-
ent sorts of pronoun: the former comes from anaphoric pronouns, and the latter from personal
pronouns. In §13.2, we have seen that noun class agreement markers can originate in classifiers
or pronouns which were also used anaphorically.
 Development of Noun Categorization 393
pronouns, and did not lead to the emergence of verbal agreement.
However, it is important to note that, since there is a widely used
grammaticalization mechanism involving development from demonstra-
tives to 3rd person pronouns, the demonstrative gender distinction may
well spread to personal pronouns as well, and subsequently give rise to
gender agreement by the mechanism referred to above (Givon 1976).30
Another path for creating gender agreement involves the pragmatic use
of 3rd person pronouns31 as a kind of 'afterthought' to specify additional
information. This development is described for Khoe, a Central Khoisan
language, by Heine and Reh (1984: 232-4; see also Heine 1997b). The
original 3rd person pronouns which distinguish three genders (masculine,
feminine, and common) and three numbers (singular, dual, and plural) gave
rise to gender and number agreement. The interrogative matma- may be
used on its own, e.g. ma hl-nye-ta (who do-JUNCTURE-ASPECT) 'Who has
done it?'. However, it is frequently followed by gender-sensitive personal
pronouns to narrow the range of possible referents, e.g. ma-ma (who-
3sg.masc) 'who?' (male referent), ma-he (who-3sg.fem) 'who?' (female
referent). This kind of structure, in which a gender sensitive pronoun is
used to emphasize the gender of the referent, is used with other word
classes, too. This has resulted in the creation of agreement constructions,
illustrated in 13.12.
13.12. xa-ma ma-ma a-ma
DEM-3SG.MASC who-3SG.MASC DEM-3SG.MASC
'Who (male) is this?'
(C) Further evidence in favour of the independent development of
predicate-argument and head-modifier agreement
The thesis of independent development of agreement on verbs and within
noun phrases is corroborated by the existence of 'split' agreement systems
(see §2.7). In these systems an argument agrees with its predicate, and the
head noun agrees with its modifiers, for different categories. Since verbal
agreement markers often originate in anaphorically used demonstratives
or 3rd person pronouns, and these are more prone to have a small
number of animacy-related gender categories, it is then understandable
why in split systems the gender agreement is preferred on verbs, and
largeish noun classes and classifiers are used to mark agreement with other
targets (see Table 2.10 above). In addition, predicate-argument and head-
modifier agreement markers behave differently in language evolution and

3(1
This process seems to have just started in some Eastern Nilotic languages, e.g. Bari and
Toposa—see Heine and Reh (1984: 231, n. 1).
Note that the noun class agreement markers can also develop from lexical sources, e.g.
nouns, or noun classifiers via their pronominalization (see §13.1).
394 Classifiers
in obsolescence. In Warekena and Bare, two endangered North Arawak
languages from the North Amazon, gender agreement is maintained in
predicate-argument contexts but lost from adjectival modifiers and from
some demonstratives.32
(D) Head-class or agreement class?
A generic noun classifier can first develop to be an overt noun class marker
on the head noun, via cliticization and phonological reduction, and then to
be an agreement marker. An alternative scenario would involve cliticization
of a generic noun to a modifier which immediately results in the creation of
an agreement system. This scenario has been suggested for the develop-
ment of the Ngan'gityemerri noun class system by Reid (1997: 215-17).
Stage 1: The precursor to the development of agreement: generic-specific
'pairing' of nouns as a common construction type, as in 13.13.
13.13. gagu wamanggal kerre ngeben-da
animal wallaby big ISGS+AUX-shoot
'I shot a big wallaby.'
Stage 2: the generics are independent words, often favoured over specifics
especially to maintain reference; as a result, one can get a noun phrase
made up of a generic noun plus a modifier, as in 13.14.
13.14. gagu kerre ngeben-da
animal big 1SGS+AUX-shoot
'I shot a big [wallaby].'
Stage 3: where specific nouns are included, both the specific noun and
modifiers tend to 'attract' generics. This 'repetition' of generics is the
predecessor of agreement, as in 13.15.
13.15. gagu wamanggal gagu kerre ngeben-da
animal wallaby animal big ISGS+Aux-shoot
'I shot a big wallaby.'
Stage 4: Repeated generic nouns cliticize to the following specific nouns,
and undergo phonological reduction. They are reduced to proclitics and
develop into agreement noun class markers, as in 13.16.
13.16. wa=ngurmumba wa=ngayi darany-fipal-nyine
male=youth male=big 3soS+AUX-return-FOCUS
'My initiand son has just returned.'

32
The genesis of 'agreement'-type phenomena in verb-incorporated classifiers has been
explained by Mithun (1986: 384). Incorporation resulted in lexicalization of compounds, and
then external arguments were added to specify the meaning.
 Development of Noun Categorization 395
Stage 5: Noun class marking proclitics become obligatory prefixes in head
class and agreement class functions. They become more fused with the stem
(e.g. a series of morphological processes start applying on the boundaries).
At this stage, they become fully grammaticalized (note that - stands for an
affix boundary, and = for a clitic boundary).
13.17. wu-pidirri wu=makarri
CANINE-dingo CANINE-bad
'a bad dog'
Stage 6: There is now 'prefix absorption' which implies the lexicalization of
head class prefixes. Some noun class prefixed roots may be interpreted as
stems which can take further noun class marking (Reid 1997). This may
lead to double class marking, e.g. wa-mumu (male-police) 'policeman'; wur-
wa-mumu (female-male-police) 'policewoman'. (Historically, this may lead
to loss of head-class marking prefixes once they become fused with the
stem. This may, or may not affect agreement; see below.)
This scenario is not applicable to other prefixing Australian languages
with noun classes, where class markers have typically come from a variety
of sources—such as generics and 3rd singular pronouns (see Dixon forth-
coming). The typical developmental scenario in Australian languages is for
pronominal prefixes to first develop on verbs, and then, at a slightly later
stage (and in just some languages) for the prefixing profile to be extended
to nouns through development of noun class prefixes. Indirectly, this con-
firms the idea of relatively independent development of agreement on
verbs, and within a noun phrase.
A different scenario applies to Dyirbal, an Australian language with no
prefixes, where some noun classes developed out of generics. Generic nouns
have become suffixes to modifiers, i.e. deictics and determiners.33 The
following stages in the development can be suggested:
Stage 1: Generic-specific pairings.
Stage 2: Constituent order being 'Demonstrative Generic Specific' generics
become cliticized to prehead modifiers (they cannot become prefixes
because there is no prefix slot available).
Stage 3: Cliticized generics become suffixes to prehead demonstratives,
and undergo phonological reduction. This is how agreement arises.
In this case, the development of agreement classes does not presup-
pose previous existence of head classes. This is corroborated by further
observations.
Firstly, there is often a certain mismatch between 'head' classes and
33
A similar scenario could apply to Bandjalang (Crowley 1978; Dixon forthcoming).
396 Classifiers
'agreement' classes in Australian languages (e.g. examples in §2.4.3). This
may indicate their different relative age. In Mayali, nouns with animate
reference trigger agreement according to the sex of the referent, and it
may be different from the noun class prefix which appears on the noun
itself (see §2.4.3). !X66, a Southern Khoisan language, has five noun
classes (Traill 1994: 20-2). They are marked by suffixes on nouns and
also realized in agreement on various targets (adjectives, relativizers,
object markers). Noun suffixes bear a strong phonological similarity to
agreement markers; an example of such an 'alliterative' concord is given
in 13.18 (= 2.14). Noun class markers are underlined (only nouns from
classes one to four may be head-marked; if they are not, the concord is
not alliterative).
13.18. ft a |na-i !xa-i t-i
I past see-NCLl:O lion-NCLl big-NCLl which-NCLl
|'aa |li k-i
dead is which-NCLl
'I saw a large dead lion'
Second, some languages have no head classes, just agreement classes.
This is the case with gender marking in many Papuan and Amazonian
languages, especially those with multiple classifier systems, and in some
Australian languages (e.g. Gaagudju and Ungarinjin). We saw in §9.1 that
in some languages of northwest Amazonia which mark noun class on the
head noun, this marking has a different semantic effect than agreement
class marking: it is associated with the countability and individuation of
the noun, e.g. Tariana deri 'banana: collective', den-pi (banana-CL:LONG)
'banana tree', deritfi (banana+CL:BUNDLE) 'banana bundle'. Head classes
and agreement classes are acquired by children at different rates (see
Tsonope 1988, and Chapter 14 below).
The loss of overt gender marking on nouns does not necessarily precede
the loss of gender agreement. It can be argued that coalescence of nominal
paradigms with their overt gender marking triggered the overall gender loss
in some Indo-European languages. However, in Old Iranian the overt
marking of masculine and neuter gender on nouns was lost, but agreement
on modifiers remained intact.
Lower Cross languages largely lost head-class marking, but there are
remnants of agreement on adjectives and verbs (Faraclas 1989: 390-1). In
Albanian neuter gender seems to have been lost from adjectives, and not
from nouns (Priestly 1983; Hamp 1958).
We conclude that there is not enough general evidence in favour of the
primacy of development either of head classes or of agreement classes. At
least in some cases, the two could have developed independently.
 Development of Noun Categorization 397
(E) Expansion of agreement
Once an agreement system has been established, it can be expanded to
other targets. The expansion can occur under areal influence.34 A dialect of
German, spoken in Canton Wallis in Switzerland, close to French-speaking
regions, acquired agreement between nouns and predicative adjectives,
under French influence (Edgar Suter, p.c.).
A possible source for the expansion of agreement could be alliterative
agreement, or repetition. Repeaters are a frequently-used source for ad hoc
classifiers which then get grammaticalized and restricted to usage with a
particular group of nouns. In Tariana, ehkwapi 'day; something that
appears during the day', was first used as a repeater, to classify the same
noun, e.g. ehkwapi matfa-ehkwapi (day nice-NCL:REPEATER:DAY) 'a nice day'.
Then it was extended to mark agreement with several nouns referring to
natural phenomena, e.g. enukwa matfa-ehkwapi (sky nice-NCL:DAY) 'nice,
clear sky', de:pi matfa-ehkwapi (night nice-NCL:DAY) 'nice, clear night'.35
Noun class agreement on different targets may have different origins. In
some Mindi languages (Australian: Nordlinger 1998: 262-3; I. Green 1995)
noun class is marked on case suffixes, while in another Mindi language
noun class is marked on case prefixes; demonstratives have noun class
prefixes in all the languages. Dixon (forthcoming) reconstructs the follow-
ing scenario for the development of noun class agreement in these
languages.
(i) There would have first evolved markers—separate words or clitics
attached to some word in a noun phrase—that combined information
about noun class and about case (possibly, from an earlier classifier-plus-
case).
(ii) These markers then attached to demonstratives as prefixes; this is why
all the Mindi languages have noun class prefixes on demonstratives.
(iii) At a later stage, in the Eastern Mindi branch the case-class markers
became enclitics to a noun phrase and then suffixes; after that they became
obligatory suffixes to both nouns and adjectives. In one West Mindi lan-
guage, Nungali, the case-class markers became proclitics to a noun phrase

34
In multiple classifier systems, it is often hard to show which target developed first. In
some cases development of new agreement targets was triggered by areal diffusional patterns.
This is the case for classifiers used on demonstratives in Resigaro (under Bora and Ocaina
influence), and in Tariana (under Tucano influence), mentioned in §13.7.
35
In many languages of the world repetition of a constituent is a syntactic and pragmatic
device for emphasis, e.g. English this table is a big table. The difference between these syntactic
constructions, and 'repetition' as an agreement device in Tariana is that repeaters used as
classifiers in Tariana form a part of a corresponding grammatical word; also, unlike the
syntactic constructions in English repeater phenomena represent obligatory agreement. Repe-
tition as an agreement device may come from grammaticalization of constructions like the
English one above (see §13.1.2, on repeaters and the origin of noun categorization).
398 Classifiers
and then prefixes to nominal modifiers such as adjectives (being later
generalized to also apply to some, but not all, nouns).

13.8.2. Decline and loss of agreement

The decline and loss of agreement is in no way the mirror image of its rise
and evolution since it happens for different reasons. As with the loss of
noun categorization devices, (A) it can take place due to language-internal
reasons, and (B) it may happen differently on different targets. Agreement
can be affected by language-external phenomena—language contact, cre-
olization and language obsolescence (C).
(A) Decline and loss of agreement for language-internal reasons
Agreeement may be lost due to a merger of paradigms because of a sound
change. For instance, in Lower Cross languages (Benue-Congo: Faraclas
1989: 390-1) overt marking of noun classes was lost due to the assimilation
of prefix vowels, and the loss of CV prefixes; this contributed to the loss of
agreement classes.
It can also happen via morphological analogy, as in many Indo-
European languages (Priestly 1983). Loss of concord is one of the reasons
for—or concomitant effects of—gender loss. The loss of gender in Old
English is partly due to the loss of agreement with adjectives and demon-
stratives (Dixon 1982: 171 and references therein).
(B) Decline and loss of agreement on different targets
Gender and gender agreement are most persistent in personal pronouns
(see Demuth et al. 1986: 459, for Cross River and Kru languages; Priestly
1983, for Indo-European; Heine 1982a: 212, for Daju dialects (Eastern
Sudanic), Aikhenvald 1995a, for Bare). Reconstruction of the development
of gender and number in Dravidian languages (Krishnamurti 1975) shows
that neutralization in gender takes place in verbs before it spreads into
pronouns; hence pronominal gender is more stable.
Similar evidence comes from Akan, a Kwa language of Ghana. Akan is
in the process of losing a noun class system which it inherited from the
proto-language (see Williamson 1989: 11-12, on noun classes in Kwa
languages). There are six classes with a partial semantic basis and vestiges
of singular-plural pairings of the Bantu type (Osam 1994: 120-1; see also
§12.3.1 above on semantics of noun classes in Akan). Noun class agreement
in noun phrases has been completely lost. A few nominalized adjectives
contain Class 1 (human) class prefixes (Osam 1994: 123); they are used
headlessly and reflect a 'derivational' use of noun class prefixes, e.g. o-
kzsee (cLl-fat) 'a fat person, a fat one'. Subject-verb agreement has been
lost in the majority of dialects. Only Twi dialects preserve an animate/
 Development of Noun Categorization 399
inanimate distinction in cross-referencing subject prefixes (Osam 1994:
124-5, 132, 138), e.g. o-be-yera (3sg.Subj:Animate-FUT-be.lost) 'S/he will
get lost', and c-be-yera (3sgSubj:Inanimate-FUT-be.lost) 'It will get lost'.36
In Kru languages overt noun class marking has been largely reanalysed
as a part of the noun stem (Marchese 1988: 332), but the agreement system
is still active. Class agreement is often lost from adjectives; agreement in
demonstratives tends to be more resistant (Marchese 1988: 335).37
The loss of agreement often involves lexicalization and fossilization of
gender markers, as is the case in the Australian languages of Arnhem
Land (Sands 1995: 255-6). For instance, in Warray (Harvey 1987: 55 ff.)
noun class prefixes become lexicalized with nouns; the pronominal agree-
ment is regular, but the system of adjective agreement with nouns is being
lost.
This agrees with the hierarchy of retention of gender oppositions put
forward by Priestly (1983: 340): NOUN > ADJECTIVE > PRONOUN. For
instance, many Indo-European languages lose neuter gender and neuter
agreement; but there are sometimes remainders of neuter in anaphoric
pronouns, e.g. French ce, Italian do < ecce hoc. Thus, agreement appears
to be more stable on pronouns and demonstratives than on other targets.
But note that this is at odds with the data on the obsolescence of gender in
Scottish Gaelic (see §13.7.3); this shows that language change in obsoles-
cence may be distinct from language-internally motivated change.
(C) Decline and loss of agreement in language contact, creolization, and
obsolescence
Agreement patterns can undergo changes as the result of indirect areal
diffusion (some examples were given in §13.7.1). In creolization and
language obsolescence, agreement undergoes more changes than head-
marking of noun classes or genders.
Noun class agreement is being gradually lost from creolized Bantu
languages. In the Mankandza dialect of Lingala, a trade language
(Bokamba 1977), noun class agreement with the subject is obligatory; in
other dialects, such as Kinshasa Lingala, agreement still exists, but is
reduced to a notional opposition animate/inanimate. Thus, the reduction
of agreement involves semantic change in the composition of classes, into
just animate and inanimate. It can be seen from the following examples
that noun Class 7 has been reinterpreted as a general 'inanimate' marker.

36
There is no singular/plural distinction in these old noun class prefixes.
37
There may be some semantic hierarchy of the adjectives involved; 'good' and 'big' seem
most resistant; but this deserves a special analysis (Marchese 1988: 355). Further study is
needed to provide a hierarchy of agreement in terms of reduction and loss of agreement.
Evidence from Lower Cross languages suggests that agreement on numerals is the first to
disappear (Faraclas 1989). (Note that Kru languages do not have agreement on verbs.)
400 Classifiers
Example 13.19 comes from Mankandza Lingala, and 13.20 from Kinshasa
Lingala.
13.19. mu-nkanda mu-ko-kweya
NCL3-book/letter NCL3-TA-fall
'A/the book will fall down.'
13.20. mu-nkanda e-ko-kweya
NCL3-book/letter NCL7-TA-fall
'A/the book will fall down.'
The same phenomenon is discussed for Swahili by Bokamba (1977; see
also Alexandre 1968; Polome 1968). In various Lingala dialects, and in
some varieties of Swahili, adjectival agreement is lost, e.g. Mankandza
Lingala mo-soni mu-ye (NCL3-pencil NCL3-this) and Kinshasa Lingala mo-
soni oyo (NCL3-pencil this) 'this pencil'. The process of agreement loss in
a creolized language can go even further. In Kituba (a Creole based on
Kikongo: Stucky 1978: 227-8) agreement has been lost with all targets.
Reduction of agreement in language contact and creolization can go
together with a tendency towards the semantic restructuring of classes.
In modern 'anglicized' Tiwi, agreement with demonstratives is regular
with humans, or high animate head nouns (Lee 1987: 118). This is part
of the tendency to restructure the Traditional Tiwi masculine and feminine
opposition into human and non-human (see §13.7.2).
Different agreement types can be lost at different speeds in language
obsolescence. Such evidence comes from Bare and Warekena, obsolescent
Arawak languages from Brazil. Both have almost lost gender agreement on
adjectives; however, gender marking is still obligatory in cross-referencing
on verbs (Aikhenvald 1995a; 1998b).
Similar processes occur in the obsolescence of languages with multiple
noun class systems. In Paumari (Arawa) the shape-based ka- noun class is
being lost by younger speakers whose main language is Portuguese
(Aikhenvald MS). The feminine/masculine gender distinction is more
persistent. This may be due to the influence of Portuguese, with its two
genders.

13.9. Semantic changes in noun categorization devices

When lexical items become grammaticalized as noun categorization
devices, they undergo semantic changes. The general principles underlying
these changes are outlined in §13.9.1. I show in §13.9.2 how the changes
affect the composition of classes and their semantic opacity.
 Development of Noun Categorization 401
13.9.1. From lexical item to classifier: principles of semantic change
Grammaticalization of noun categorization devices involves semantic
changes. These semantic changes fit in well with the general tendencies of
semantic change discussed in grammaticalization theory. They involve
'bleaching', 'abstraction', and metaphorical extensions (cf. Heine et al.
1991: 39-45), or a meaning change from a more concrete notion (e.g. a
word for 'fruit') to a more abstract notion (e.g. a classifier for 'round objects').
Lexical items which get grammaticalized as classifiers can have generic
meanings, e.g. 'person', or 'thing'. They can also have more specific mean-
ings, and belong to a 'subordinate' level of categorization. Semantic exten-
sions and principles of semantic changes38 depend on the type of classifier,
its preferred semantics, and the type of lexical source. In (A) we consider
semantic change from a noun to a classifier. Changes which take place in
the passage from a verb to a classifier are analysed in (B).
(A) From a noun to a classifier: semantic changes
When nouns become classifiers they may undergo a number of semantic
changes: (Al) a noun with generic reference may become a generic classi-
fier; (A2) a noun with generic reference may become a classifier restricted
to a specific class of referents; (A3) a noun with specific reference may
become a classifier for a more general class of referents; or there can be
semantic changes by extension, which may be due to metaphor and meto-
nymy (cf. §12.1.1).39 Little or no semantic change is involved in the creation
of highly specific classifiers; for instance, in Baniwa specific classifiers for a
room in the house, and for excrement come from the words for room and
for excrement respectively. These terms may be then extended to further
items, by metaphor, metonymy, or just salient properties (A4).
It is generally accepted that 'the source concepts' in grammaticalization
are of frequent and general use, and their frequent use is due to their being

38
Natural tendencies of semantic change in lexical items are still a largely unexplored field.
Semantic change between lexical items may be conceived as the addition of an element of
meaning to the semantic system, or the loss of an element of meaning 'while the form remains
constant' (e.g. Wilkins 1981; 1996: 269). Semantic changes 'within a speech community involve
polysemy at their beginning point or at their endpoint' (Wilkins 1996: 269). For the semantic
change from a lexical item to classifier there is often no evidence of synchronic polysemy at the
endpoint; that at the beginning can only be reconstructed. The form itself often does not remain
constant; the more grammaticalized the classifier, the likelier it is to undergo phonological
reduction.
39
A noun which comes to be used as a classifier can shift its meaning as a free noun. In
Northern Iroquoian languages, the liquid classifier -hnek- assumed a specialized meaning of
'whisky, liquor' when used as a free noun (Mithun 1986: 391). Thus, classifiers and free nouns
of the same origin can undergo independent semantic change (which may have corresponded
to polysemy at an earlier stage). Synonyms, when used as classifiers, can have slightly different
meanings which are often unpredictable. Minangkabau butia 'seed' is used for inorganic seed-
like and bulky objects; and incek 'seed' is used for organic seed-like objects (Marnita 1996).
402 Classifiers
'fundamental' elements. They can often be denned as belonging to the
basic level of categorization, like basic-level verbs of physical state such
as 'sit', 'stand', and 'lie' (Heine et al. 1991: 33). Another generalization is
that 'categories of the subordinate level are unlikely to serve as source
concepts' (Heine et al. 1991: 33; Sweetser 1988: 402). Exceptions to these
generalizations can be found with classifiers, since there are frequent
examples of changes from a subordinate (more specific) level of categor-
ization to a superordinate (more 'abstract') level (A3). However, these
changes reflect universal cognitive mechanisms which underlie the structure
of noun categorization (see §12.2.2).
(Al) Noun with generic reference becomes a generic classifier Superordi-
nate nouns meaning 'man' or 'woman' often give rise to NOUN CLASSES; a
noun meaning 'man' becomes a marker for masculine, and one meaning
'woman' is used for feminine (see the examples from Mupun, Zande, and
some Australian languages in §13.1.1). A generic term for 'person' may
develop into a gender marker if there is a gender/sex-based distinction
already in the system. 'Person' developed into the non-feminine marker
in Eastern Nilotic, after feminine had already evolved (Heine and Reh
1984: 219).
Generic nouns used as NOUN CLASSIFIERS preserve their generic meaning.
The generic term 'person' is often used as a source for noun classifiers, e.g.
in Australian, and as a source for numeral classifiers meaning 'human in
general'. In Minangkabau, batang 'tree, trunk' (a reflex of Austronesian
'trunk; tree; timber; platform': Conklin 1981: 259) is used as a generic
classifier for all trees, and the generic 'flower' is used as a generic for
flowers (cf. Minangabau bungo, Acehnese bungong: Durie 1985: 135).
The same can be observed for VERBAL classifiers. A generic noun, 'domes-
tic animal', covers all animals, including pigs, in Cayuga (Iroquoian);
'water' becomes a classifier for all drinkable liquids in Ngandi (Mithun
1986: 387, 389; Heath 1978b).
In the languages of Southeast Asia, default classifiers are often based on
a generic term for 'body', or 'person'. Default human numeral classifier pui
in the Southern Lawa group of the Palaungic branch of Mon-Khmer
derives from Proto-Waic *bxu 'person' (Adams 1992: 110). 'Fruit' is
another frequent source of a default classifier, e.g. Indonesian and
Minangkabau buah 'fruit, generic classifier'; White Tai xang3 'fruit', 'clas-
sifier for objects in general' (Conklin 1981: 150). A classifier for 'round
things' is frequently used as a default classifier (§12.1.4), the semantic
change being then fruit > round > generic. A similar change is found in
Totonac aq 'a default classifier' which comes from aq ~ akan 'head' (Levy
1994). General classifiers may also come from an item meaning 'thing', e.g.
the possessive classifier in Tariana yarupe 'thing'.
 Development of Noun Categorization 403
This semantic change fully agrees with current ideas on lexical sources
for grammaticalization. When a noun meaning 'man' or 'woman' becomes
a noun classifier, very little, if any semantic change is involved, e.g. Akatek
(Zavala 1993) woman > woman; man > man; Jacaltec woman > female
non-kin classifier.
(A2) Noun with generic reference becomes a classifier for a specific class of
referents A change from a generic to a more specific referent takes place
with noun classifiers. In Mam (Table 13.3) a generic noun 'man' acquired
the more specialized meaning of 'old man, respectfully', and 'woman' that
of 'old woman, respectfully'. In Kana, a generic noun came to be used as a
specific classifier: the numeral classifier nee derived from the word for
'person', is used with only one noun 'guest' (Ikoro 1996a: 93).
(A3) Noun with specific reference becomes a classifier for a more general
class of referents In Eastern Nilotic, kinship terms at a more subordinate
level of categorization became gender markers (e.g. a noun meaning
'daughter, girl' became a feminine gender marker: Heine and Reh 1984:
219). Kinship terms can be used to classify humans according to their age
and social status. Ba 'grandmother' in Vietnamese is used as a numeral
classifier to classify women over 40; ong 'grandfather' is used to classify
'personified or deified animals', e.g. tigers, elephants, and whales in pro-
verbs (Adams 1989: 86-90). It is unusual to derive generic classifiers from
kinship terms. This is, however, the case in Kana. The 'default' numeral
classifier comes from the word ka 'mother' (Ikoro 1996a: 90-1).
In Khmu (Austroasiatic), ta 'grandfather' is used as a masculine class
term with names of boys (as opposed to i, with unknown etymology, used
with names of girls) (Adams 1989: 57). In West Bahnaric *raa?, the 'numeral
classifier for humans', comes from a more specific noun 'adult human'
(Adams 1992: 110). A noun meaning 'child' often appears as a source for
noun classifiers, and numeral classifiers for any young animate. In Kana,
nwii 'child', or 'offspring' is the source for a classifier for young beings,
human or non-human. In the noun classifier system in Mam, it is used as a
classifier for babies. In noun class systems, 'child' can get grammaticalized
as a diminutive gender marker (as in Bantu languages: Heine 1982a: 214).
The way in which a noun with specific reference can become a generic
classifier for the whole species is similar to the development from a proto-
type to its extensions. In Cayuga, the stem for 'car' is used as a classifier for
all vehicles. In Mohawk the classifier for fruit (-ahy-) is also the word for
'berry' (Mithun 1986: 391).40

40
Synchronically, this does not mean that the source-noun retains its prototypicality. For
instance, modern speakers of Mohawk do not consider 'berries' such as strawberries and
blackberries as prototypical 'fruit' any more (Mithun 1986: 391).
404 Classifiers
(A4) Semantic extensions The preferred semantics of different classifier
types can explain different patterns of semantic change. Shape is more
important as a semantic basis in numeral classifier systems than for
noun classifiers. In contrast, material make-up and function are more
important in noun classifier systems (see Chapter 11).
Thus, for noun classifiers with inanimate reference, the most common
semantic changes from noun to noun classifier are as now listed.

(a) Extension by material makeup

The specific noun kuru 'water' in Ngan'gityemerri (Reid 1997) and bana
'fresh water' in Yidiny (Dixon 1982: 199) are used to classify all drinkable
liquids. In the languages of the Kanjobalan branch of Mayan the classifier
for corn is used for corn, and also for products made of corn. Similarly,
the classifier for 'stone' is used with objects made of stone (Zavala 1992:
158-9).
Material makeup can involve different extensions by common properties,
e.g. in Ngan'gityemerri yenggi 'fire' is used to classify all things associated
with fire, such as firewood, charcoal, smoke, firestick.

(b) Extension by function

In Ngan'gityemerri syiri 'weapon' became extended to all things which
involve striking, e.g. lightning; and yawurr 'tree' was extended to all
wooden artefacts, and then to all modern manufactured products, even if
they are not made of wood (Reid 1997).
The most common extensions from noun to numeral classifiers are the
following.
Extension by shape, e.g. batang 'tree, trunk' used to classify vertical things
in Minangkabau (Conklin 1981: 259; Marnita 1996; see also Appendix 2
below);
Extension by material makeup, e.g. batang 'tree, trunk' used for categor-
ization of wooden things and long parts of plants in Bugis and Banggais
(Conklin 1981: 259);
Extension by shape and material, e.g. batang 'tree, trunk' used to classify
vertical things fabricated from wood and other long inflexible objects in
Indonesian (Conklin 1981: 259); ton 'tree' is used to classify both plants
and vertical things in Thai (Carpenter 1992);
More rarely, extension by material makeup and function, e.g. in Kana te 'tree'
is used to classify wooden objects used as means of transport (Ikoro 1996a:
91-2). In Thai, the classifier for objects with handles, initially used only for
bicycles, was extended to all other vehicles (with the exception of carts)
(Carpenter 1986). In Tariana the classifier for 'canoe', -whya, is used for any
vehicle. 'Leaf is often used as classifier for containers in Austroasiatic
 Development of Noun Categorization 405
1
languages, e.g. Siamese bai 'leaf. Interestingly, fruits are also covered by
this classifier—they are then conceived as 'flesh-covered, meat-holding
vessels' (Conklin 1981: 152). In Ponapean (Micronesian), wahr 'canoe' was
grammaticalized as were 'POSS.CL:CANOE,VEHICLES' (§13.1.1).
These extensions are typical for nouns with inanimate referents. 'Child',
unlike other nouns with a human referent, can be extended to small and
round things. In Vietnamese, con 'child, be small' is used to classify people
'less than humans' (i.e. of low social status), monsters, animals in general,
some inanimate items such as kites, and images of animate beings. This
classifier also covers body parts which are considered more 'lively' (eye,
pupil of the eye, heart, penis), and by semantic extension, a number of
other things (figure, number, cards which may be related to the images of
human beings) (Adams 1989: 89-91). In Siamese luuk3 'child' is used as a
classifier for round things (Conklin 1981: 152).
In each case the semantic extensions may be language-specific and
almost unpredictable (cf. Matisoff 1991, on the unpredictability of
semantic change in grammaticalization). One lexical item, 'leaf, gets
extended to various semantic domains in a number of Tai languages
(Conklin 1981: 159); in each case it involves different shape properties.
In Wu-Ming, Nung, and White Tai 'leaf is used to cover FLAT EXPANDED
objects (paper, fabric, board), CONTAINERS (plates, hats, baskets, pails,
bottles, fruits), ROUND FLAT objects such as rice cakes and paper, FLAT
surfaces, such as paper and documents, and also blades. In Dioi, the
same classifier is used only with paper, fabric, and board, and is not
extended to either containers or blades. In Black Tai, it has the same
meanings as in Nung but for the 'container' extension. In Western
Austronesian languages, seed-based classifier meanings extend from small
round objects to teeth, eggs, fruits, and even to mid-sized spherical objects
and bulky objects (Conklin 1981: 246-8). 'Fruit/stone'-based classifiers—as
generic classifiers—range from fruit and stones to medium range objects,
bulky objects, lumps/loaves, inanimates and even animals (Conklin 1981:
261-2).
Numeral classifiers frequently come from body parts. Semantic changes
often involve extensions by shape and size (e.g. 'eye' becomes a classifier
for small round objects); however, extensions are often fairly idiosyncratic.
Semantic changes from a lexical noun to a classifier for a noun with an
inanimate referent often involve metonymy (description of the whole by its
part). This happens with nouns which refer to parts of plants, or of
inanimate objects. 'Bay of a house' became a classifier for houses in
Khmer; and 'roof is used to classify houses in Waic languages (Adams
1989: 120). Handles are often used to classify objects with handles (see (C)
in §13.9.2, on classifier khan in Thai).
406 Classifiers
(B) From a verb to a classifier: semantic changes
When a verb becomes a classifier it will specify some of the properties of
its original argument in S or O function, but never in the function of an
A (this agrees with the predominantly absolutive basis on which verbal
classifiers operate: cf. Keenan 1984). Typical semantic changes in verbs
are:
(i) Posture verbs become deictic classifiers and verbal classifiers which refer
to the SHAPE and ORIENTATION of their S arguments. For instance, the verb
'stand' becomes a classifier for vertical or one-dimensional referents which
may also be large and strong. 'Lie' becomes a classifier for two-dimensional
or horizontal objects, which can be associated with the ground, e.g.
locations or crawling animals in Enga. 'Sit' tends to be associated with
three-dimensionality, or roundness (see Table 6.7 above).
(ii) Stance and posture verbs become deictic and verbal classifiers referring
to POSITION. 'Hang' is used to refer to objects which have this functional
property—i.e. hanging, or protruding, as in Enga (Table 6.7).
(iii) Verbs of handling become classifiers which refer to the FUNCTION of the
object or the way it has to be handled. In Imonda, 'pick fruit' became a
verbal classifier for 'fruit to be picked', and 'break in two' for 'objects
which are normally broken' (Table 13.6). (Examples from Cahuilla were
given in §13.1.4.)
(iv) Verbs from other semantic groups may become specific numeral clas-
sifiers for their typical S or O arguments, e.g. the development of 'spread
out' to 'clothing' in Tai (Conklin 1981: 167), and to 'mats' in Makassar
(Conklin 1981: 442). Other examples include 'rinse, bath' to 'storms' in
Dioi (Conklin 1981: 380); 'be long and straight as bamboo' to 'flutes and
guns' as in Shan, a Tai language (Adams 1989: 127); note also 'hang down'
as a classifier for clothing; 'pluck' for 'flowers', and 'slice' for slices as of
bread in Bugis (Conklin 1981: 435), 'tie together, bind' for bundles in Shan
(Conklin 1981: 406); and 'roll up' for cigars, rolled leaves, and other sorts
of rolls in Siamese (Conklin 1981: 426).
Mensural classifiers often come from body-oriented verbs. The classifier
will then refer to the result of the action, e.g. Minangkabau gigik 'bite', for
a bite of food, or (h) isok 'inhale', for traditional cigarettes, and palik 'to
take with a tip of a finger', for creamy things in a small amount (see also
examples in De Leon 1987).
(v) Verbs can also become verbal classifiers which describe the location or
instruments typical of the activity, e.g. 'carry' in Enga used for referents of
sexual activities, such as vagina or penis (see Table 6.7).
 Development of Noun Categorization 407
13.9.2. Further changes in noun categorization devices
The composition of an established system of noun categorization may
change as the result of losing or acquiring new members. Semantic changes
may result in the increasing semantic opacity of assignment of a noun class,
or of a classifier, or lead to a more semantically transparent system.
(A) Change in composition of a noun class
In numerous Bantu languages, the human class (1/2) tends to absorb non-
human animates. This semantic extension results in the change of the
composition of the class. For instance, in Kuria this class includes per-
sonified animals (Aksenova and Toporova 1994: 76), and in Pogoro and
Tonga all animals are included in this class (Aksenova and Toporova
1990: 28). Consequently, in these languages, the 'human' noun class
became simply animate. (Note that here noun class assignment does not
become more opaque.) In Shona, semantic changes in the composition of
classes affected the class assignment of humans and animates which
became distributed between several different classes. Class 1/2 is the
main class for ordinary and public persons; it also includes chiefs, evil
spirits, and thunder. Class 5/6 is organized around the notion of ritual
danger; it includes persons who inspire fear (e.g. ancestors, wild persons,
foreigners). Class 9/10 centres around 'protection' against the potential
danger; it includes protective spirits and medicine men. Palmer and Arin
(1995a) hypothesize that the reason for this restructuring lies in the social
organization of the Shona (see §12.3.1 above).
Semantic changes within a classifier system can occur as the result of
introducing a new member into a system of noun categorization devices.
This happens most often in languages with classifiers as independent lexical
items. In Thai, the influx of new items of material culture resulted in the
creation of new classifiers (Carpenter 1986: 18-19). Some of these new
classifiers are loanwords themselves, e.g. chut (from English suit), which is
used to refer to dresses and Western style suits, as well as to pyjamas and
bathing suits. Others are of Thai origin. For instance, lawt 'tube' came to
be used as a classifier for test tubes, light bulbs, and drinking-straws, and
rian 'coin' is used colloquially for counting dollars.
(B) Increasing opacity in noun class assignment
In Australian languages the class which originally referred to 'edible
vegetables' (typically marked with initial ma-, mi-, or m-) often underwent
changes in its composition. As a result, its assignment can become ex-
tremely opaque. In Ngan'gityemerri, it includes exclusively edible plants
(the only exception being ngikin 'dried faeces'; this extension is explained
by Reid (1997) in terms of the vegetable appearance of dried faeces of
408 Classifiers
herbivores). In Dyirbal, the corresponding class includes all vegetable foods
and honey. In Wardaman (Merlan 1994: 61) this class underwent a number
of extensions: it includes all flora, especially vegetables and vegetable parts.
The 'vegetable food' class assignment is very opaque in Gurr-goni
(R. Green 1995: 58–9): its central meaning is still 'edible vegetables', cover-
ing most living plants and vegetable foods and most items made from plant
materials, most body parts, dead bodies, and faeces; and also corroborees
and songs, traditional canoes, paper, bombs, and directions. Some of these
changes in the composition of noun classes are explainable by (i) semantic
extension by important property, for instance, from edible vegetables to all
flora, as in Wardaman; (ii) metaphorical extensions; for instance, the
inclusion of traditional canoes and other means of transport into the
vegetable class in Gurr-goni may be accounted for by the fact they used
to be made out of wood; then canoes were extended as a means of trans-
port, and this is how a loanword, erriplen 'plane', got assigned to the
vegetable class (R. Green 1995: 58), and (iii) metonymy: from edible
vegetables, or flora to parts of plants (cf. §13.4).41
Loss of formal gender markers can make gender assignment rules more
complicated. In Colloquial French, the loss of final -e, the feminine marker,
has led to the creation of complicated phonological and morphological
rules of assignment (cf. Corbett 1991).
(C) Increasing opacity in classifier choice
Changes in composition may affect numeral classifiers. A few classes in
Thai have been reorganized under the pressure of new concepts, or new
items from Western culture (Carpenter 1986: 19). The classifier khan, with
its lexical meaning 'long handle' or 'dyke between rice paddies', was
originally used to refer to things with long handles. Nowadays it includes
cars, bicycles, motorcycles, buses, and names for other vehicles (with the
exception of ox-cart), and words like spoon, fork, and umbrella, as well as
traditional musical instruments and other utensils. This semantic extension
appears to have started from bicycles, an object with handlebars, that are
like handles, which was the first foreign vehicle to be introduced to
Thailand. Subsequently, automobiles and other vehicles were assigned
the same classifier, as a 'class extension based on the functional category
of vehicle'.

41
Mayali (Evans 1997) provides another example of increasing semantic opacity of the
'vegetable food' class. Besides all non-meat foods, it includes a variety of other nouns, such as
(a) anatomical terms pertaining to genitalia, sexually produced fluids or excretion, (b) some
geographical terms, (c) a few bird names, (d) some wooden implements, e.g. weapons, (e)
words for rain and rain-water, and (f) manner adverbials.
 Development of Noun Categorization 409
(D) Increasing semanticity of noun class assignment
Gender systems can shift from more grammatical to more semantic, as in
Cantabrian Spanish (§2.3.4). If a language loses gender agreement and
preserves the gender distinctions in personal pronouns only, the assignment
tends to become predominantly semantic even if it was not so before; this is
the case in English with its three genders just in third person pronouns, he/
she/it. Along the similar lines, Chechen speakers 'sometimes point out that
words of the B-gender refer to round objects, as some of them in fact do,
although many do not ... If there is a speaker expectation that gender
classes will be shape-based, then analogical reclassifications over time will
probably lead to just such a situation' (Nichols 1992: 140).42
(E) Semantic changes due to introduction of new semantic features
Classes 7/8 and 12/13 in ChiBemba (Bantu) acquired additional values of
'large' and 'small' respectively which got extended to affective notions
('despised' and 'endeared') (Spitulnik 1989: 212); for instance, in Kuria,
Classes 5/6 and 20 have augmentative and pejorative values (Aksenova and
Toporova 1994: 81, 102).
(F) Historically attested semantic changes
Languages with a long-standing written tradition provide evidence for the
development of classifier systems over time. In Chinese, classifiers were
used sporadically with culturally salient objects during the classical period
(500-206 BC); the use of classifiers increased from about 100 AD (Bisang
1996: 540). Semantic development of Chinese classifiers from highly spe-
cific to generalized reference is discussed by Erbaugh (1986: 428-31). Table
13.7 illustrates the historically attested semantic changes from individual to
general reference of three Mandarin Chinese classifiers.
Semantic extensions are usually made by shape, which is the preferred
semantic property of numeral classifiers (see Chapter 11; and note the
development from 'small branch' to 'long things in general' in the history
of the classifier tido, in Table 13.7). Extralinguistic factors, such as the
cultural importance of an object, may influence the rate of change. In
Mandarin, culturally important classifiers, e.g. pi 'horses' and ben 'books'
have retained their exclusive reference. Other classifiers underwent seman-
tic changes, e.g. gen 'threads, hairs, strings' first referred to stalks of grass;

42
This tendency often appears in poetry and metaphors. A famous example of direct
association between masculine gender and masculinity and feminine gender and femininity
in German comes from Goethe's poem 'Der Tannenbaum' (DEF.ART:MASC pine.tree), 'the pine
tree' who is dreaming of encountering a she-palm tree (die Palme) (H. Seiler 1987). This effect
was lost in Lermontov's translation of the poem into Russian, since both tree-names are
feminine.
410 Classifiers
TABLE 13.7. Historical changes in Mandarin Chinese classifiers
Dynasty, time Classifier tiao Classifier mei Classifier ge
Shang, c.l 400 BC Small branch Trunk of bamboo Bamboo
tree

Zhou, 1100 BC Trees

{
Qin, 255 BC Sticks Wooden objects

Han, 200 BC Lengths of bamboo

Post-Han, 25 AD Snakes, lengths of Flutes, swords, Arrows

cloth, strings of birds, fish, jewels,
gold ingots dishes

Tang, 600-900 AD String, clothing General classifier Arrows, candles,

dogs, chickens,
horses

Song, 960-1117 AD Long things in Fruit, birds, people

general

Ming, 1368-1643 AD General

Qing, 1644-1912 AD General classifier for

people, and
unclassified objects

Modern, 1912- Virtually dies out

except for needles,
badges

ke was first used to classify peaches, then all fruits, then all small, round
objects in general (Erbaugh 1986: 430).
An overview of the historical development of Japanese numeral classi-
fiers is given by Downing (1996: 39-46). She notes the importance of
animacy distinctions in the eighth-century Japanese system; round objects
lacked their own classifier, and the gap was filled by the use of the general
classifier tsu. There were a number of classifiers used for small functionally-
defined culturally-salient categories, e.g. hata for counting looms, or
hashira for counting gods and exalted persons (these items do not appear
in the classifier inventories of Modern Japanese: Downing 1996: 19-22).
The default, or general, classifier tsu was used to refer to any entity except
 Development of Noun Categorization 411
human beings and, probably, gods (Downing 1996: 46). In the early docu-
ments in Thai, tua was used to refer exclusively to four-legged animals;
later it became extended to articles of clothing, ghosts, and letters (cf.
Diagram 12.1).
The analysis of classifier production by representatives of varying social
groups and generations in Kilivila, a multiple classifier language, by Senft
(1996: 202-27) shows several tendencies illustrating the language change in
progress. Speakers tend to replace specific classifiers with shape-based
classifiers. The greatest number of innovations have been observed among
school children aged from 8 to 14 'because of their readiness for a playful
exploration of the possibilities the C(lassifier) P(articles) system offers and
because of their increased linguistic awareness' (Senft 1996: 235). Innova-
tions and language change patterns in classifier usage have also been
observed among adults between 21 and 35. This tendency has a socio-
linguistic explanation: it appears that consultants with low social status try
to overcome language barriers that mark intrasocietal status by changing
patterns of classifier use.

13.10. Sources of noun categorization devices: a summary

Noun categorization devices of different kinds tend to develop from dif-
ferent lexical sources—open classes of nouns or verbs, or closed classes of
pronouns or adpositions. Only numeral, relational, and possessed classi-
fiers develop from deverbal nominalizations; these classifiers can also have
mixed origins in some language groups: see Table 13.8.
The internal evolution of highly grammaticalized noun categorization
devices is often based on the reanalysis of other categories, and internally
motivated by phonological and morphological factors. The evolution of
noun categorization devices is often due to language external reasons, e.g.
areal diffusion, creolization, and language obsolescence. If a noun categor-
ization system tends to be lost, animate and human classes are more stable
than others.
The general principles of semantic change in noun categorization sys-
tems correspond to the main parameters of cognitive extension—metaphor
and metonymy, conditioned and bounded by cultural and social con-
straints. There is a great deal of similarity between the diachronic semantic
change from lexical nouns to classifiers, and the synchronic semantic
extensions which correlate with preferred semantic parameters for each
classifier type. Thus, numeral and verbal classifiers tend to get extended
according to the physical properties of their referent, such as shape or
dimensionality.
Table 13.8 summarizes the information contained in Tables 13.1 and
412 Classifiers
13.5, and the accompanying discussion. We have also seen that culturally
important items can give rise to most classifier types (see Table 13.1); they
are especially well represented in large multiple classifier systems.

TABLE 13.8. Typical sources for noun categorization devices

Noun categorization devices Sources
Noun classes Nouns (kinship, humans and higher animates)
Demonstratives
Locatives
Derivational suffixes
Mixed origin
Numeral classifiers Nouns (unit counters, body parts, generics, etc.)
Generic nouns
Deverbal nominalizations
Verbs of handling (involving typical activities)
Mixed origin
Noun classifiers Generic nouns
Nouns referring to kinship, humans and higher
animates
Locative classifiers Nouns referring to body parts
Deictic classifiers Posture verbs
Possessed classifiers Nouns referring to kinship and humans
Generic nouns
Verbs of handling (involving typical activities)
Deverbal nominalizations
Mixed origin
Relational classifiers Verbs of handling (involving typical activities)
Deverbal nominalizations
Mixed origin
Verbal classifiers Incorporated nouns—body parts and generics
Reanalysed serial verbs
Classificatory verbs Incorporated nouns (body parts)
Posture and motion verbs
 14 Noun Categorization Devices in
Language Acquisition and
Dissolution

The behaviour of noun categorization devices in child language acquisition

and language development, as well as in language dissolution, provides
important clues as to how such devices are assigned, what the categorical
interrelations between them are, and which of them is the most stable. It
also provides us with information on the psycholinguistic reality of the
assignment of noun categorization devices.
In this chapter, we consider language acquisition, development, and
dissolution of two kinds of noun categorization device, noun classes and
numeral classifiers. We have to restrict ourselves to the discussion of just
these two because of the limitations of actual psycholinguistic research
undertaken thus far. No experiments have been undertaken on language
acquisition or dissolution of other classifier types, or of how these pro-
cesses take place in multiple classifier systems. These are topics for future
study.
In §14.1 we look at child language acquisition of gender and noun
classes, and agreement patterns. Section 14.2 considers the acquisition of
numeral classifiers. Processes of language dissolution in noun class and
classifier systems are described in §14.3. Conclusions are given in §14.4.

14.1. Acquisition and development of noun classes

Acquisition and development of noun classes (which include genders) have
been studied for Indo-European, Bantu, and Semitic languages. Studies of
gender acquisition overlap with acquisition of agreement since noun
classes are realized through agreement. In mixed (i.e. morphologically,
semantically, and phonologically determined) systems of gender assign-
ment, the formal characteristics of genders and noun classes are mastered
first.
For instance, in Hebrew, every noun is assigned to masculine or feminine
gender; and the two genders are distinguished in the singular and in the
plural. The assignment is mixed; as was shown in §2.3.4, most feminine
nouns have their plural in -ot, and masculine in -im; but there are exceptions.
414 Classifiers
A longitudinal and cross-sectional study by Levy (1983a; 1983b) showed the
importance of phonological features for acquisition of portmanteau gender
and number marking, and the comparative irrelevance of semantics or
syntax. In particular, sex-gender correspondence proved irrelevant to the
acquisition of inflected animate nouns. In the acquisition of Hebrew, young
children make most errors in assigning correct gender to nouns whose
gender cannot be predicted by their form (Berman 1985: 299-301).
Working out formal correspondences between nouns to assign genders
and to mark agreement appears to be important in morphologically com-
plex languages, especially in the cases where gender-sensitive morphemes
are portmanteau with number (see §10.1). Similar results have been achieved
for Russian (Popova 1973), German, and French (Karmiloff-Smith 1979).
Children can also operate with a number of different intralinguistic cues
at a time. The study of the acquisition of genders in Spanish by Perez-
Pereira (1991) showed that the greater the number of converging cues the
more easily the children acquire the gender assignment. In his experiment,
children aged from four to eleven pay far more attention to morphopho-
nological and syntactic than to extralinguistic clues. The feature animacy/
inanimacy did not produce any substantial difference in the way Spanish
children handled nouns. In assigning genders, children rely on intralinguis-
tic regularities. The relative markedness of genders is another important
clue which follows from the dominance of formal, rather than semantic,
principles of gender assignment in early gender acquisition. Spanish chil-
dren tended to attribute masculine gender to nouns most often because it is
unmarked, and therefore easier to acquire (Perez-Pereira 1991: 584). The
same tendency was observed for French (Karmiloff-Smith 1979), and
Hebrew (Levy 1983a; 1983b; Berman 1985). A different result was achieved
for German where the feminine forms of the definite article die and of the
indefinite eine get overgeneralized at first.1 This can be explained by their
frequency of use, and greater salience than other forms.
Salience of actual forms appears to be an important factor in the acqui-
sition of genders in Spanish and French (Perez-Pereira 1991: 587). Neu-
tralization of gender forms and syncretism can be an obstacle in gender
acquisition. Smoczynska (1985) noted that Polish children master gender
much earlier than Russian children in spite of considerable similarities
between the two languages. The suggested reasons have to do with Russian
phonology. One reason is reduction of both a (typical feminine singular
ending) and o (typical neuter singular ending) in post-tonic syllables to [A]
or [a], so that the actual forms are neutralized.2 This is not the case in
1
The article die could have overgeneralized also due to the fact that it is identical to the
plural definite article (with no gender distinction).
2
A study of child acquisition of genders in Russian dialects where there is no vowel
reduction in post-tonic syllables would be helpful to prove this point.
 Language Acquisition and Dissolution 415
Polish. Another reason for delayed gender acquisition by Russian children
is the existence of numerous hypocoristic forms which refer to boys, or men,
but end in -a and have a feminine declension, e.g. Kolja 'diminutive for
Nicholas'. There is also a number of frequently used common nouns with
masculine reference which end in -a and belong to a 'feminine' declension;
for these, however, agreement is masculine, e.g. djadja 'uncle'. In children's
speech, these nouns often take feminine agreement, e.g. djadja sidel-a 'uncle
sit+PAST-FEM', instead of djadja side! 'uncle was sitting' (Popova 1958: 109).
Morphological complexity may slow down the process of gender acquisi-
tion. Children have more difficulties in mastering gender systems of
languages with three genders where gender markers interact with other
categories, such as number and case, and where genders are 'ambiguous,
barely transparent and scarcely predictable morphophonological markings'
(Perez-Pereira 1991: 587). This is the case in German, Czech, Serbo-
Croatian, and Russian (see Perez-Pereira 1991: 585, 587, and references
therein). In contrast, children learning languages with clearly differentiated
and regular marking of just two genders—e.g. Egyptian Arabic, Hebrew,
French, or Spanish—master these systems more easily and more quickly
(Levy 1983b; Perez-Pereira 1991: 588).
Another important feature in gender acquisition is its formal opacity
and regularity. It has been shown that children rely more on formal than on
semantic cues when identifying antecedents of gender-marked genitive
pronouns in German (Bohme and Levelt 1979). Slightly different results
were achieved for Icelandic (Mulford 1983). To determine reference based
on pronominal gender Icelandic children needed semantic as well as formal
information. These differences may be due to different degrees of 'opacity'
in gender marking in these two closely related languages. In German,
formal correlates of grammatical gender are relatively stable and are easier
to 'discover', while in Icelandic the formal gender correlates are more
opaque, and so children have to rely much more on natural gender dis-
tinctions (Mulford 1983: 89).
It has been noticed that English children acquire the right use of gender-
sensitive 3rd person pronouns later than, say, German children. This result
shows that 'the more extensive and productive the system of gender mark-
ing in a language, the easier is its learning' (Perez-Pereira 1991: 588). Note
also Mills (1986: 107): 'It would rather appear that the extensive system of
grammatical gender marked on many different parts of speech gives the
German child more opportunity to seek regularities in the system and to
learn to produce these forms accurately.' (See Corbett 1991: 82-8, for
further examples of acquisition of genders in Indo-European languages.)3

3
Another aspect of language acquisition is how children assign gender, or gender and
number portmanteau markers, to nonce words. Levy (1983a: 119) reports that children did
416 Classifiers
Considerable work has been done on child language acquisition of noun
class systems in a number of Bantu languages (Demuth 1988; Demuth et al.
1986; Kunene 1986; Suzman 1980; Herbert 1991; Tsonope 1988). As we
saw in Chapters 2 and 10, noun class prefixes are portmanteau with
number. Noun class assignment is semantically opaque (though a certain
semantic core is associated with almost every class: see Chapter 2 and
Spitulnik 1989). Agreement is heavily linked to phonology, i.e. there is a
large amount of alliterative agreement.
The data on Sesotho child acquisition of noun classes described by
Demuth (1988) showed that the system of gender agreement was estab-
lished before nouns were divided into gender classes. In Sesotho, a noun
and its modifier are attributed a class feature specification and are treated
as 'some kind of prosodic, cognitive, or grammatical unit' (Demuth 1988:
316). Young language learners then start using agreement productively,
gradually working out the appropriate phonological marking for nouns.
Sesotho nouns are learned in conjunction with their inherent noun class
features realized in agreement; and the head-marked noun class appears
later. Similar results have been documented by Tsonope (1988) for
Setswana. That noun class is first acquired as a feature of a phrase, rather
than a feature of an individual noun, is accounted for by the fact that the
Bantu noun class agreement system is phonologically transparent and
pervasive.4 As we saw above for Indo-European gender acquisition, formal
regularity helps acquisition of gender agreement systems. The fact that
there are very few examples of 'overgeneralization' of a particular set of
noun class prefixes in the studies of Bantu noun class acquisition indicates
that the noun prefix is first learned as a part of the noun, and it is not
segmented till later (Tsonope 1988: 148; Herbert 1991: 11).
The semantics of noun classes does not seem to play any important
role in noun class acquisition for Bantu languages: the acquisition of
formal regularities precedes the acquisition of semantics. Tsonope (1988)
showed that children begin to consolidate semantically based noun
classes, such as human nouns, only later, after the agreement system
has been acquired. In some cases semantic clues may have played a
limited role. For instance, the acquisition of demonstratives by Setswana
children seemed to suggest that they are able to make a distinction
between human and non-human nouns very early. However, other factors

better in inflecting nonce words denoting animates than in inflecting nonce words denoting
inanimates; however, younger children had difficulties with inflecting nonce words. Kunene
(1986), who worked on getting children to recognize and assign class to novel words, arrived at
slightly different conclusions as to noun class assignment with SiSwati-speaking children from
those reached by Demuth (1988).
4
Lehmann (1982) also argues that noun class is prototypically realized through agreement
within a phrase.
 Language Acquisition and Dissolution 417
could have contributed to this (e.g. the frequency of human nouns in
discussions, or phonological restrictions).5
In the acquisition of the semantics of the Bantu noun class system, there
is a tendency to acquire the human/non-human distinction before other
underlying semantic contrasts. This explains the tendency for all nouns
denoting humans to be occasionally treated as members of Class 1/2
(predominantly human) (cf. Suzman 1980, for Zulu). Overgeneralization
of human classes took place in experiments of noun class assignment to
novel nouns. When given a choice of uncontextualized noun class assign-
ment, children tended to classify plural nouns into just human/non-human.
However, in spontaneous speech production no Overgeneralization
occurred (Demuth et al. 1986: 466).
The main conclusion, for child acquisition of the Bantu noun class
system, is that agreement principles are acquired before the actual marking
of nouns. This may also be linked to the phonological reduction of nouns
in children's speech (Herbert 1991: 111). Demonstratives and possessives
have been shown to be the first targets which show agreement (Demuth
1988: 319; Herbert 1991: 111). In Zulu the first concordial subsystem
consistently employed was for subject marking on verbs (Suzman 1982: 57).
The primacy in acquisition of concordial noun classes over head mark-
ing agrees with certain developments in the history of Bantu languages. In
Grebo (Kru: see Diagram 13.4 above; also Demuth et al. 1986: 467) the
overt noun class marking is lost, but the concord system remains. The fact
that the first signs of subject/verb agreement to appear in children's speech
are anaphoric (Suzman 1982; Herbert 1991: 111) confirms the hypothesis
of an anaphoric origin for the genesis of agreement (see §13.5 and §13.8).

14.2. Acquisition of numeral classifiers

Similar to the studies of gender and noun class acquisition, existing studies
on the acquisition of numeral classifiers are limited to a few languages—
Chinese (Erbaugh 1986), Thai (Carpenter 1991; 1992; Gandour et al.
1984), Japanese (Sanches 1977; Matsumoto 1985), Garo (Burling 1973),
and Kilivila (Senft 1996).6
Classifier acquisition is reported as much slower than acquisition of
noun classes. For instance, even ten-year old Thai children got only 89
5
The acquisition of noun classes can be influenced by sociolinguistic parameters: subjects
with urban backgrounds may acquire these noun classes which include the most borrowings
faster (Suzman 1980: 52).
6
Both Chinese and Thai use classifiers with deictics and with numerals; however, there are no
data on how classifiers are acquired with deictics. Kilivila (Senft 1996) also employs classifiers
in several different environments; there are no data on which environment is acquired first.
418 Classifiers
per cent of their classifiers right (Carpenter 1991: 98). Classifier production
in Kilivila starts at the age of about four; the frequency of classifier
production increases dramatically at the age of nine (Senft 1996: 180). In
contrast, the noun class agreement system is acquired by Sesotho children
at three years of age (Demuth 1988).7
Studies on the acquisition of numeral classifiers shows that, unlike noun
classes, extralinguistic regularities and semantics are very important at the
first stages of classifier acquisition. In the acquisition of Garo classifiers,
the animacy distinction was acquired before other categories (Burling
1973). The same results were obtained for Thai classifiers (Candour et
al. 1984). Overuse of the general classifier was reported for Japanese
children (Sanches 1977; Matsumoto 1985). Children tend to overextend
the use of classifiers based on shape and function. At age seven, the general
classifier is still overused.
In her seminal study on how children acquire classifiers in Mandarin
Chinese, Erbaugh (1986: 431) established the order in which classifiers of
different types are acquired. In principle, classifier use by children was
similar to that by adults. Children, however, used classifiers more rarely,
and showed a tendency to overuse the general classifier ge, especially at an
early age, starting from two years (Erbaugh 1986: 415). Specific classifiers
developed slowly and were rarely used between the age of one year and ten
months and three years and ten months. They started being used lexically,
specific to a single referent, and were then extended to mark a prototypical
member of a class, and so on to further extensions.8 The most frequently
generalized feature used for semantic extensions was shape (predominantly
vertical extension and small size, similar to some patterns of historical
development of classifiers considered in §13.9).
The following scheme shows the relative order of acquisition of Chinese
classifiers (< stands for 'acquired earlier': Erbaugh 1986: 431):
classifiers for discrete, countable, portable concrete objects < classifiers for
large immovable ones < classifiers for actions < classifiers for abstractions
and honorifics;
measures < special classifiers;
valued items < common ones < conventionalized sets.
Items with unique reference were acquired before the prototypical
members. Abstraction by extension, and small size, occurred before shape.

7
Susan Quigley (p.c.) also reports that young Award-speaking children in New Guinea do
not use some specific classifiers.
8
This order of acquisition may also relate to the order of development of conversation
about different objects, in which case it could be an artefact of communication strategies with
children.
 Language Acquisition and Dissolution 419
This last statement is also confirmed by results of child acquisition of
Japanese numeral classifiers. Sanches (1977: 61) arrives at the conclusion
that most numeral classifiers in Japanese 'are learned in relation to the
items they classify, in lists, rather than as representative of categories of
criterial attributes that are potentially generalizable to an infinite variety
of items'.
Animate classifiers are acquired first in Kilivila. They are followed by
general classifiers for inanimates, and the classifiers which are acquired
next denote concrete, specific objects, and salient features of objects (Senft
1996: 192). These conclusions have also been corroborated by investiga-
tions on Hokkien (Ng 1989: 123), and Cantonese (Luke and Harrison
1986).
A recent study of acquisition of classifiers by young Mandarin-speaking
children showed the dependency between frequency of a classifier and its
order of acquisition (Hu 1993). This study showed that the classifier zhl
'animate' is acquired very early, due to its frequency. Other classifiers which
are acquired early are those which relate closely to the child's life. In
contrast, the honorific classifier wei and the classifier tdi used for machines
do not occur in children's speech until the age of six. This may be due to the
irrelevance of such distinctions in the children's life (Hu 1993: 125).
As we saw in Chapter 9, Chinese uses classifier morphemes in two
environments: with numerals and with deictics. These uses were acquired
differently; classifiers with numbers were acquired before those with
demonstratives. Classifiers with near-demonstratives were acquired before
the far ones.
Errors in the acquisition of Thai classifiers showed the early overuse of
general classifier ?an, and repetition of the head noun in the classifier slot
instead of making reference to its semantic content. In learning correspon-
dences between perceptual properties of nouns and classifiers children have
to figure out which perceptible characteristics are relevant to semantic
categorization, since historical changes in classifier systems sometimes
make the classifier assignment opaque. The first semantic opposition to
be acquired is animate vs. inanimate (Juntanamalaga 1988: 322). Children
then start making semantic overextensions based on extralinguistic resem-
blances among category members (shape, consistency, and function being
frequent bases for extensions), to compensate for semantic limitations of
classifier assignment (Candour et al. 1984: 471–2). One strategy is over-
extension of more 'comprehensive', or 'general' classifiers and their overuse
in place of more specific classifiers (e.g. general classifier fan). The opposite
strategy consists in the overuse of 'repeaters', i.e. overspecialization—the
use of every noun as its own classifier. Both strategies make the choice of
classifiers less dependent on the inherent, or conventionalized semantic
features of an object. This 'lightens' the semantic load of a classifier, and
420 Classifiers
makes it easier to use classifiers without having access to the full range of
extralinguistic information.
Language-specific properties of the semantic organization of classifier
systems play a role in the order of their acquisition.
The shape-based classifiers appear to be acquired relatively early by
speakers of Mandarin Chinese; classifiers which refer to non-extended
round objects are acquired earlier than classifiers which refer to extended
objects.
The results obtained by Ng (1991) concerning classifier use by Hokkien
speakers of different ages, showed the overuse of the general classifier
among children aged five to eight, indicating that 'Hokkien children learn
the conventional use of classifiers after going through a long period of only
producing the general classifiers' (p. 81). There was a certain number of
overextensions based on perceptual clues (shape and form) and on
animacy. In contrast, Gandour et al. (1984) demonstrated that in Thai,
classifiers which refer to arrangement and quanta are acquired before the
shape-based classifiers. Ng (1991: 81) hypothesizes that Hokkien children
acquire the shape classifiers more easily than Thai children due to a larger
degree of semantic complexity encoded in Thai classifiers, and also to the
fact that Thai has more classifiers referring to shape than does Hokkien;
Thai has at least six classifiers for one-dimensional objects, four classifiers
for two-dimensional objects, and four more classifiers for three-dimensional
objects, while Hokkien only has five shape-based classifiers.
Thus, the rate of acquisition of classifiers may be determined by
their semantic complexity, that is, the number of contrasts encoded
in them. The one-dimensional classifier hon and the two-dimensional
classifier mat in Japanese do not encode concomitant features of flexi-
bility or rigidness unlike their Hokkien counterparts; according to the
results obtained by Matsumoto (1985), they are acquired earlier than
the corresponding classifiers in Hokkien because they are semantically
'simpler'.
Stages in the acquisition of classifiers are best characterized as 'stages of
organizing knowledge' (Carpenter 1991: 108-9). Children make almost no
word order mistakes with classifiers. Once they have learnt that something
'belongs in the post-numeral position, they figure out its identity bit by bit,
starting with the information that it must come from a closed set of words
that conventionally appear in the post-numeral position' (Carpenter 1991:
109). This also explains why children tend not to be innovative with
classifiers (Carpenter 1992).
These results agree with the results on acquisition of genders and noun
classes—namely that children are sensitive to formal regularities and syn-
tagmatic patterns. The relationship between a classifier and a number is
acquired before the relationship between the classifier and the head noun.
 Language Acquisition and Dissolution 421
However, to learn the full system of numeral classifiers, syntagmatic
intralinguistic patterns are not sufficient. Children must then learn the
correspondences between extralinguistic and intralinguistic categories, and
this is why acquisition of classifiers is slow. The acquisition of classifiers by
children consists in learning a conventional, pre-existing system which
involves a large amount of cultural extralinguistic knowledge.
The order of development of classifiers in Mandarin Chinese parallels
the historic processes of changes in classifier systems (Erbaugh 1986: 426-
30). The results of Thai classifier acquisition as compared to the history of
Thai classifiers are not as clear-cut (Carpenter 1992). The main difference
between the two is that child language development is teleological, i.e. the
end-product is the acquisition of a pre-existing system, while historical
change does not have to be this way. We have seen in Chapter 13 that
classifier choice can change in response to external factors, e.g. cultural
change. Conventionality is more important in child language acquisition.
This is why there are differences between the kinds of semantic primes
important for classifying objects, for adults and for children. For instance,
children seem to use material as an organizing feature much more than
adults do. Carpenter (1992: 142) explains this as a consequence of the
world of children being a special subculture in which the importance of
material features (e.g. breakability) can override other perceptual features,
e.g. shape, or size, before the children actually have the linguistic and the
extralinguistic experience to know what are the 'right', i.e. the conven-
tionalized attributes of linguistic categories.
In spite of these differences, both developmental and diachronic patterns
reveal certain tendencies in common. Among such tendencies are the
similarity between shape classifiers and containers, and an historical and
a developmental relation between these. Another tendency is for the clas-
sifiers which are first learned to be predominantly nouns. We have seen
above (Chapter 13), that, in the history of different types of classifier
system, classifiers may come from verbs; however, these instances are
much rarer, and they are, presumably, acquired later.
The acquisition of classifiers parallels the general tendencies of cognitive
categorization, e.g. unique reference before prototypic members and exten-
sions, and more concrete items and categories before more abstract ones
(Erbaugh 1986; Lee 1988).
Clark (1977: 460) showed that cognitive paths of categorization in chil-
dren's acquisition of word meaning resemble categorization patterns in
classifier patterns. Both in child language acquisition and in classifiers,
shape is the primary basis for categorization. 'Both classifier systems and
children's overextensions reflect a basic categorization process that goes on
first at the non-linguistic level' (see §12.2).
422 Classifiers
14.3. Dissolution of noun classes and of numeral classifiers
The expectation that patterns of dissolution in aphasia might be the mirror
image of acquisition by children comes from the seminal work by Jakobson
(1941). Earlier acquired features are expected to be most resistant to loss in
aphasia, and in other situations of language dissolution (Herbert 1991:
125). However, acquisition and dissolution of noun classes show some
striking differences.
Herbert's study (1991) is a pioneering one in showing how noun class
prefixes and agreement behave in language dissolution in Bantu, namely,
Broca-type aphasia, and how this is different from child language acquisi-
tion. While children start with acquiring agreement and end up with
acquiring overt noun class marking for nouns, the production of nouns
with an overt prefix is regular in aphasic speech. There is also no systematic
reduction in the inventory of prefixes, and the agreement system is pre-
served. The most common errors are errors when the noun prefix on the
noun is correct but the generated concords are not, i.e. nouns are assigned
to 'wrong' agreement classes in subject-verb agreement. Wrong concords
appear more rarely in head-modifier Noun Phrases.
Semantically, what child language acquisition shares with the aphasic
data is the prominence of a human/non-human distinction. In aphasics,
many concord errors involved transfer of nouns with human referents to
Class 1/2 (Herbert 1991: 126).
The data on language dissolution agree with data on child language
acquisition in that both show the primacy of concord, singular and plural
distinctions, and the feature human/non-human. The difference lies in the
head-marking of nouns which is stable with aphasics but appears to be
acquired later by children (Herbert 1991: 128).
Recent work on dissolution of numeral classifiers in Thai (Candour et al.
1985) agrees with the mirror image hypothesis for language acquisition and
language dissolution. Classifiers based on inherent perceptual character-
istics of objects are less resistant to aphasic disruption than those which are
not. Errors in animate classifiers typically involved substitution of honori-
fics by classifiers referring to ordinary persons, or by generic tua 'animal,
thing'. Classifiers based on configuration and quanta, e.g. groups,
appeared to be the least stable in language dissolution. Errors made by
adult aphasics are here similar to errors made by young children. Aphasics
tended to overuse the general classifier ?an, and to use repeaters instead of
appropriate classifiers (Gandour et al. 1985: 552).9

9
Other factors could have contributed to these errors. In informal spontaneous conversa-
tion in Thai, the general classifier is an appropriate substitute for almost any special classifier.
This is not true for other styles and registers. The overuse of the general classifier could have
 Language Acquisition and Dissolution 423
Tseng et al. (1991) compared two types of aphasic speaker of Mandarin
Chinese and Taiwanese: those with Broca's aphasia, resulting in agramma-
tism (defined as 'dropping out of connective words, auxiliaries, and gen-
eral loss of obligatory grammar') and those with Wernicke's aphasia,
resulting in paragrammatism (defined as 'the substitution of an inappropri-
ate grammatical form for the correct target': p. 185). The general tendency
in both types was to substitute a more specific classifier with a general
('neutral') one. It has also been observed that Broca's aphasics tended to
avoid classifier constructions, while the Wernicke's patients showed more
instances of substitution of a 'correct' classifier with an incorrect one, more
often the general one, and even more often the neutral one.10
That is, like children, aphasics were trying to 'lighten' the semantic load
of each classifier, making the classifier assignment less conventionalized
and less dependent on extralinguistic information they did not possess to
the same extent as before.

14.4. Conclusions

Children's acquisition of noun classes, and of numeral classifiers, shows a

number of fundamental differences. In both cases syntagmatic rules (and
agreement, in the case of genders and noun classes) are acquired first.
Overt marking of gender and noun class is acquired later. The first seman-
tic division to be acquired relates to human/non-human. The tendency to
overgeneralize a particular noun class over others is slight. Children have
been shown to rely much more on intralinguistic information than on
extralinguistic (Perez-Pereira 1991).
In contrast, extralinguistic information is more important for lexical-like
systems of classifiers. Animacy and shape, material and function, and
knowledge of socio-cultural categories are among the most important fac-
tors in classifier acquisition. The tendency to 'lighten' the semantic load of
classifiers goes in two directions—overgeneralization, i.e. reliance on the
general classifier, or overspecialization, i.e. overuse of the repeater technique.
In mastering a classifier system, children have to rely on extralinguistic
also been due to the fact that aphasics lost their capacity of adequately understanding and
manipulating varying styles and registers of discourse (Candour et al. 1985: 552). The overuse
of repeaters could have been partly due to echolalia (i.e. mere repetition as a part of the
disease). Differences found in classifier responses could have been partly due to frequency of
use of the associated noun.
10
However, according to Ahrens (1994), Tseng et al. (1991) underestimated the role of
code-switching from Mandarin to Taiwanese among their speakers. Her experimental studies
showed that the use of the general classifier ('classifier neutralization') followed the same lines
in aphasics and in normal speakers. More studies are needed.
424 Classifiers
information rather than on intralinguistic. This is why they acquire classifier
systems later than agreement noun classes.
Existing studies of language dissolution—however few—also point in
the same direction. Language dissolution of numeral classifiers in Thai
mirrors the acquisition of classifiers by children. Disintegration of extra-
linguistic knowledge in aphasics results in overgeneralization (overuse of
the general classifier) or overspecialization (overuse of repeaters). However,
dissolution of noun classes differs from child language acquisition in that
head-marked noun classes which are acquired late by children are not lost
in aphasia. The principles of concord also persist; and the main errors in
language dissolution concern wrong assignment of agreement classes, and
sometimes overgeneralization of classes.
Acquisition of classifiers has much in common with acquisition of the
lexicon, unlike the acquisition of noun classes and genders. Acquisition of
classifiers has many features in common with principles of their historical
development; the same would be arguable for noun classes and genders.
In sum, as Carpenter (1991: 111) put it, 'it is possible that numeral
classifiers are acquired differently from other, previously described kinds
of noun classes largely because they are organized and represented in
speakers' minds in a fundamentally different way'.
More investigation on acquisition and dissolution of other classifier
types, especially for multiple classifier systems, is needed, before we can
fully understand the underlying cognitive and acquisitional principles.
 15 Conclusions

This concluding chapter recapitulates and summarizes the general themes

which have emerged from a cross-linguistic study of noun categorization
devices. An overview of established classifier types used in different
morphosyntactic environments and of their synchronic and diachronic
properties—which cover their morphological realization and role in
agreement, principles of assignment, interaction with other grammatical
categories, semantics, and functions, as well as their origin and role in
language acquisition and dissolution—are summarized in §15.1. The co-
occurrence of different classifiers in one language, multiple classifier lan-
guages, and the problem of prototypes and continua in noun categorization
are discussed in §15.2. The last section, §15.3, considers further prospects
for the study of noun categorization devices.

15.1. Properties of classifier types

In this section I present an overview of the established types of classifiers.
First, I discuss the morphosyntactic environments (loci) in which classifiers
appear and constituents which they characterize—(A) morphosyntactic
locus of coding and (B) scope, or domain of categorization (as introduced
in §1.5). For each type, I outline its prototypical characteristics. Classifiers
are then characterized with respect to their other properties outlined in §1.5:
(C) principles of assignment; (D) kinds of surface realization; (E) role in
agreement; (F) markedness relations; (G) degree of grammaticalization and
lexicalization; (H) interaction with other grammatical categories; (I) seman-
tic organization of the system; (J) evolution and decay; and (K) language
acquisition and dissolution. Classifier systems also differ as to how obliga-
tory they are, and in their size. Smallish sets of noun classes are more likely
to be obligatory than large sets of classifiers of other types. Possessor
classifiers are a very rare kind, and as a consequence no generalizations
can be made concerning them. Locative and deictic classifiers are quite rare,
and the generalizations which can be made concerning them are limited.
(A) Morphosyntactic locus of coding; (B) Scope, or domain of
categorization
NOUN CLASS SYSTEMS and GENDERS are closed, highly grammaticalized,
obligatory systems, realized through obligatory affixal agreement within
426 Classifiers
and sometimes outside a noun phrase. Their inventory is often smallish
(from two to ten). Different subtypes of noun class system can cooccur in
one language. Noun classes can have an NP or a clause as their scope. If
noun classes have a clause as their scope this signifies 'noun class' agree-
ment of an argument with a predicate. This argument may be A or S, O or
S, or an oblique. Noun classes can be marked on the noun itself.
NOUN CLASSIFIERS characterize the head noun itself; they appear inde-
pendently of any other element in an NP. They can be realized as indepen-
dent lexical items (cf. traditional terms such as 'generics', or 'sortal
classifiers'), or as affixes to the noun. Noun classifiers may develop into
noun classes by first fusing with the noun they refer to, and later develop-
ing into agreement markers.
NUMERAL CLASSIFIERS characterize nouns in numerical noun phrases and
expressions of quantity. They can be realized as independent lexical items,
affixes to numbers, or quantifying terms; or they may, very rarely, form one
constituent with a noun. Their inventory is often large.
Classifiers in possessive NPs fall into three groups. RELATIONAL CLASSIFIERS
characterize the nature of the possessive relation; they are usually in-
dependent lexemes, but may be affixed to the possessed noun or possessor
pronoun. POSSESSED CLASSIFIERS characterize the possessed noun; they are
usually affixed to it, but may alternatively be expressed through an in-
dependent word. Some 'borderline' systems combine the properties of
the possessed and of the relational type. A rather rare type, POSSESSOR
CLASSIFIERS, characterizes the possessor.
VERBAL CLASSIFIERS mark agreement in the verb with an extra-predicate
NP argument, almost always in S or O function, but sometimes in an
oblique function. Verbal classifiers come in three subtypes: incorporated
classifiers, affixes, and suppletive classificatory stems. Different subtypes of
verbal classifiers can cooccur in one language.
Two further, rather rare classifier types operate within an NP. LOCATIVE
CLASSIFIERS occur in locative expressions, marking agreement with the head
noun. They can be affixed to a locative, or fused with it. DEICTIC CLASSIFIERS
occur with deictic elements (articles and demonstratives). They can be
affixed to or fused with a deictic.
Table 15.1 summarizes the scope of classifier types, their grammatical
function and constituents they refer to.
(C) Principles of assignment
When compared to other types of classifiers, noun class systems differ in
the principles of assignment. Noun class assignment can be governed by
the semantic, morphological, or phonological properties of a noun, or a
combination of these. In a sense all systems of noun class assignment are
mixed, since there is always a semantic core which involves universal
TABLE 15.1. Scope of classifier types
Classifier type Scope Grammatical function Constituent referred to
1. Noun class/gender Inside a head-modifier NP Determination Head noun
Head-modifier agreement
Outside NP Predicate-argument agreement A/S or S/O oblique
2. Noun classifiers Noun Determination Head noun
3. Numeral classifiers Numeral/quantifier NP Quantification, enumeration Head noun
4. Relational classifiers Possessive NP Possession Possessive relation
5. Possessed classifiers Possessive NP Possession Possessed noun
6. Possessor classifiers Possessive NP Possession Possessor
7. Verbal classifiers Clause S/O agreement S/O or oblique
8. Locative classifiers Adpositional NP Spatial location Argument of adposition
9. Deictic classifiers Attributive NP Spatial location, determination Head noun
428 Classifiers
semantic parameters, such as sex, humanness, animacy (see Chapter 2).
The assignment of all other classifier types is always semantic, even though
it may sometimes be a little opaque; it is never based on the morphological
or phonological properties of the noun. This characteristic of noun class
systems follows from their higher degree of grammaticalization in compar-
ison to other classifier types. See Table 15.2.
(D) Surface realization; (E) Role in agreement
Noun classes and verbal classifiers can never be expressed with free lex-
emes, unlike classifiers of other types. Verbal classifiers, locative and deictic
classifiers never appear attached to nouns. Classifiers of any type can
appear attached to an item in or outside an NP, as a suffix or a clitic.
Further details on the morphological devices used for marking each of
the classifier types are summarized in Table 15.3. Noun classes, deictic
classifiers, possessor classifiers, and verbal classifiers participate in agree-
ment; numeral and possessed classifiers participate in agreement only if
expressed with bound morphemes. Relational and noun classifiers belong
to non-agreeing noun categorization devices; and so do numeral and
possessed classifiers expressed with free morphemes.
(F) Markedness relations
Classifier types differ as to the relations of markedness. There may be a
functionally and/or formally unmarked member of a noun class system;
markedness relations can seldom be established in other types. However,
default classifiers present in some systems may be considered analogous to
functionally unmarked noun classes.
(G) Degree of grammaticalization and lexicalization
Noun classes are the most grammaticalized noun categorization devices.
Noun classifiers, which are more lexical, may be grammaticalized into
noun classes. The choice of other classifier types involves a quasi-lexical
selection.
(H) Interaction with other grammatical categories
The ways in which different classifier types show interactions with other
categories are shown in Table 10.17. Noun classes tend to interact with
noun categories (e.g. number) and verbal categories (e.g. tense), while
verbal classifiers show interdependencies with clausal categories (such as
grammatical relations) and with verbal categories. Classifiers in possessive
constructions may be restricted to one type of possession (e.g. alienable or
inalienable). Interaction of any type of classifier with other grammatical
categories depends on the scope of the classifier (Table 15.1) and their
realization as free or as bound morphemes.
TABLE 15.2. Assignment of classifiers
Noun class Numeral Relational Possessed Locative Deictic Noun Verbal
classifiers classifiers classifiers classifiers classifiers classifiers classifiers
1. Semantic assignment May be only partially Fully
2. Every noun is Yes Not necessary
associated with one
class(ifier) or belongs
to a class
TABLE 15.3. Morphological realization of classifiers
Prefix/ Suffix/ Apophony Suppletion Stress Reduplication Noun Repeaters
proclitic enclitic incorporation
Noun classes + (only for yes
classes marked
on the noun
itself, or head
classes)
Numeral classifiers yes
Noun classifiers (only in multiple
classifier systems)
Relational classifiers rare
Possessed classifiers yes
Verbal classifiers
Locative classifiers +
Deictic classifiers 4 (only in multiple
classifier systems)
 Conclusions 431
(I) Semantic organization of the system
Classifiers of established types differ in their preferred semantics (Table
11.13); however, preferences for different semantic parameters represent
only tendencies. Animacy and physical properties—and, rarely, functional
properties—tend to be encoded into noun classes and numeral classifiers.
Noun classifiers tend to encode functional properties and social status, and
more rarely properties referring to inherent nature. Verbal classifiers tend
to encode physical properties, other inherent nature characteristics, and
directionality, but not necessarily animacy. Relational classifiers encode
functional properties, while possessed classifiers encode physical proper-
ties, nature, animacy, and sometimes functional properties. Locative and
deictic classifiers tend to encode physical properties; deictic classifiers tend
to encode directionality as well. Generic-specific relations are characteristic
of noun classifiers, verbal classifiers and sometimes possessed classifiers,
but not of other types (they are rare in numeral classifiers).
Classifiers may also differ in the organization of the system. In noun
class systems, every noun has to be assigned to a class; this is not necessa-
rily so with other types.
Noun classes in general tend to be obligatory, while the realization of
other types may be optional. All classifier types are used anaphorically and
as participant tracking devices. The use of optional noun categorization
devices correlates more with referentiality, specificity, definiteness, topical
continuity, and the discourse salience of a noun than that of the obligatory
ones (see Chapter 12).
(J) Evolution and decay
Noun categorization devices differ in their etymological source, and, con-
sequently, in the grammaticalization processes which apply to them. In
particular, closed word classes give rise only to noun classes, not to other
categorization devices. Nouns are a typical source for all classifier types;
however, different lexical groups of nouns tend to grammaticalize as dif-
ferent categorization devices (see Table 13.1). Verbs can give rise only to
numeral classifiers, deictic classifiers, relational classifiers, and some kinds
of verbal classifier (Table 13.5). A less grammaticalized and less obligatory
device may develop into a more obligatory one (e.g. development of noun
classifiers into noun classes in Australian languages: see §13.4). Classifiers
can get reanalysed and drift into a different type within the same scope (e.g.
reinterpretation of an old noun class system as numeral classifiers in
Bengali). The development and structure of a classifier system is often
affected by sociolinguistic parameters, language planning, and language
obsolescence factors; these changes tend to affect large classifier systems
more than closed, grammaticalized noun classes.
432 Classifiers
(K) Language acquisition and dissolution
Acquisition of noun classes and of numeral classifiers show a number of
fundamental differences. In mastering a numeral classifier system, children
appear to rely more on extralinguistic information than they do in acquir-
ing a more closed and more grammaticalised system of noun classes.
Studies in language dissolution appear to be a mirror-image of their
acquisition (Chapter 14). More work is needed to substantiate these pre-
liminary generalizations.
In spite of the morphosyntactic differences between classifiers of distinct
types, they represent a unitary phenomenon of noun categorization. This
was corroborated by the fact that different varieties of classifiers share the
same semantic features (see §11.1), and most of the same functions (§12.1);
they share etymological sources and paths of grammaticalization, evolu-
tion and decay. In addition, one kind of noun categorization device can be
reanalysed as another. Languages can have more than one kind of noun
categorization device; and one noun categorization device can be realized
in multiple environments. But how discrete are the types of noun categor-
izations, and are there any possibilities of having more types of classifier
than those described in this book? This issue is dealt with in the following
section.

15.2. Cooccurrence of classifier types and multiple classifier

languages; prototypes and continua

We have seen that a number of languages have more than one system of
classifiers; in most cases one of these systems is noun classes, while the
other one can be of almost any other kind. Typical combinations are: noun
classes and numeral classifiers; noun classes and noun classifiers; noun
classes and verbal classifiers; numeral classifiers and relational classifiers;
and numeral classifiers and noun classifiers.
The possibility of the coexistence of several classifier types within one
language discussed in Chapter 8 is a strong argument in favour of the
proposed typology, alongside the varying properties of the classifiers as
summarized in §15.1. Languages can have up to six formally distinct sets of
classifiers, which differ both in function and in semantics.
It is unusual for languages to distinguish all the three kinds of classifier
in possessive constructions; however, there are examples of languages with
relational and possessed classifiers. The reason for the rarity of deictic
classifiers as a separate type may be due to their overlap with noun classes
in scope and in the constituent referred to. Locative classifiers may be
relatively rare because adpositional NPs are problematic as a separate
 Conclusions 433
type of NP (they often overlap with possessive NPs). The extreme rarity of
possessor classifiers may be due to the fact that the nature of the possessor
allows little variation in languages, a prototypical possessor being human,
or animate; there is thus hardly any need to categorize the possessor (see
Chapter 5).
Some languages—mostly from Southeast Asia and South America,
together with a few Austronesian and Papuan languages—have multiple
classifier systems, i.e. morphemes which appear in all the environments
listed in Table 15.1 but generally have some behaviourial differences. Mul-
tiple classifier languages employ the same—or almost the same—set of
morphemes in up to six different environments; see Chapter 9. Multiple
classifier systems are distinct from noun classes realized on different targets
(see §9.2).
Besides combining several classifier types in one language, languages
can have several subtypes of one type of noun categorization. The most
widespread instance is the possession of different noun class systems (see
§2.7), usually, a larger system of 'nominal', and a smaller system of 'pro-
nominal' noun classes (§2.7.1), which may be in complementary distribu-
tion with respect to morphosyntactic environments. I have also described
rarer instances of languages with more than one morphological type of
numeral classifier (see §4.3), and of languages with two coexisting kinds of
verbal classifiers (see §6.4). No languages have been found with several
distinct subtypes of noun classifiers, or of any of the classifiers in possessive
constructions, or of locative or deictic classifiers.
The existence of different subtypes of established types suggests the
possibility of an 'open-endedness' for the proposed typology. I have
mentioned several times throughout this book how the established types
are better seen as referring to 'focal points' on a continuum of noun
categorization devices rather than to discrete types (cf. also Craig forth-
coming: 43).
These continuum phenomena are of the following kinds. First, classifiers
could be just emerging, having not yet acquired the full status of gramma-
tical morphemes: they may be employed with only a minority of nouns and
be easily omissible (cf. the discussion of incipient numeral classifiers in
§4.4.2). Second, classifiers may be difficult to distinguish from other
morphological mechanisms; for instance, there are often difficulties in
distinguishing noun incorporation and verbal classifiers (see Chapter 6,
and especially n. 2 there). Third, there are borderline cases where it is far
from clear how many classifier types a language has: see the discussion in
§9.3.
Continua can also be established within particular types. For instance, in
many cases the distinction between mensural and sortal numeral classifiers
can be seen as that of a continuum (see Chapter 4, and the examples from
434 Classifiers
Palikur body parts in Chapter 13 which show that the same morpheme can
be used in both ways). It may be difficult to distinguish one type of
classifier from another; for instance, if a language has classifiers used on
demonstratives, it may be unclear whether this implies the presence of
deictic classifiers, or just noun classes which are realized on this target
(cf. §7.3). In the case of poorly known languages (such as Kipea, an extinct
Macro-Je language from Brazil: Rodrigues 1997) it is almost impossible to
tell a multiple classifier system from a system with agreement on multiple
targets (cf. §9.2).
Thus, the classifier types outlined here correspond to 'focal points' on
the cline of possibilities for grammaticalized noun categorization devices
realized in distinct morphosyntactic environments. The 'focal' points at the
edge of the continuum have more prototypical properties than the points
along the continuum. Further morphosyntactic contexts may develop into
new contexts in which classifiers may be used—indeed, in some cases it
seems to be almost impossible to tell whether we are dealing with new
emergent classifier types or extensions of already existing ones; see the
discussion in §9.3. In particular, subtypes within certain types—noun
classes, numeral, and verbal classifiers—are the most likely candidates to
give rise to new kinds of noun categorization devices.
The whole typology of noun categorization devices might be thus best
presented in terms of focal points (with their prototypical properties) along
a continuum of noun categorization. This conclusion is significant for a
descriptive approach to noun categorization devices; instead of trying to fit
noun categorization devices in a particular language into a certain type, the
important thing is to place them onto a continuum, and then decide which
prototypes they most resemble.

15.3. Prospects for future studies

A number of areas require further work in order to refine our understand-
ing of noun categorization systems. In spite of growing interest in noun
categorization devices all over the world (see §1.4), there is an urgent need
to provide more good descriptive studies of individual systems, especially
in areas of little-known 'exotic' and highly endangered languages, such as
Amazonia and New Guinea. In many cases, for languages with some
documentation the analysis of noun categorization devices requires further
work. For instance, gender systems in languages such as Deni and Kulina
(Arawa) and Chapacuran, and animacy systems in Baining languages (East
New Britain), require urgent further investigation. More studies are
required on numeral classifier languages in West Africa, in particular Cross
River languages and Grasslands languages. 'Repeater' phenomena as
 Conclusions 435
agreement devices are in an urgent need of a systematic study, in particular
in West Atlantic languages such as Bainouk and Landuma.
Multiple classifier languages are most in need of detailed study. These
include numerous South Arawak languages (e.g. Pareci, Waura, Yawala-
piti, and various Campa languages), Tupi, Bora-Witoto, Tucano and Maku
languages from the Amazon, and Southeast Asian languages such as Miao-
Yao. Multiple classifier systems in languages of the South Pacific are in
need of urgent systematic study; these include Reef-Santa Cruzan lan-
guages from the Solomons, Papuan languages from Central and South
Bougainville, and multiple classifier languages from New Guinea, such as
Awara and Wantoat, from Morobe province, and languages of the Angan
family. Even for well-described languages there is often a need for more
detailed study, especially a text-based one, in order to establish relation-
ships between discourse and noun categorization devices.
The acquisition and dissolution of classifiers of different types, especially
in multiple classifier systems, is also in urgent need of study—this is
probably the most notable gap in classifier studies.
Noun categorization devices offer a rich ground for collaborative
research by descriptive linguists, typologists, sociolinguists, and psycho-
linguists, together with sociologists, philosophers, and psychologists. Noun
categorization devices offer numerous possibilities for important projects,
in core areas of linguistics, as well as in a wide range of cross-disciplinary
fields. The most important task, however, is first to pursue descriptive
studies, in order to collect additional materials which may then assist us
in rethinking the whole framework.
It is also important to note that in spite of differences along many
parameters all classifiers reflect (in different ways) a single phenomenon—
the categorization of nominals by humans, through human language. They
reflect common cognitive mechanisms, and common semantic features,
such as humanness and animacy.
This book includes material from about 500 languages, but these cover
no more than one-tenth of the world's languages; further studies of noun
categorization devices in previously undocumented or scarcely documented
languages will help deepen our understanding of the mechanisms of human
cognition.
 Appendix 1
Noun Categorization by Means Other than
Classifiers

Languages without any noun categorization devices—noun classes or clas-

sifiers of other types—tend to express some of the meanings characteristic
of classifiers through other categories, e.g. number, case, declensions or
derivational affixes. Typically, these categories are then sensitive to ani-
macy and humanness distinctions. In languages with noun classification
the choice made in these categories may also depend on animacy or
humanness of the referent, alongside with—but independently of—noun
classification devices. How these categories can be employed to categorize
nouns is discussed here.
(A) In many languages with no noun classification, NUMBER depends on
the animacy, or humanness, of the noun. Typically, there are more number
distinctions made for animate or human nouns than for inanimates or non-
humans. For instance, in Comanche (Uto-Aztecan) dual and plural mark-
ing is obligatory for human nouns, optional for animate nouns, and seldom
used for inanimates (Charney 1993: 49-50). In Koasati (Muskogean:
Kimball 1991: 447) human nouns may be optionally marked with a suffix
-ha; other nouns have no plural marker. The distinction of animate/
inanimate is combined with number marking in Basque (Iturrioz et al.
1986a). In Epena Pedee, a Choco language from Colombia (Harms 1994:
55), plural marking is obligatory on nouns with human referents. In Kobon
(Papuan), when a generic noun with an inherently plural referent functions
as subject, singular or plural marking on the verb depends on the animacy
of the noun—plural is marked only if the noun has a human referent
(Onishi 1997b: 18; Davies 1981).
This tendency also holds in languages with classifiers but no agreement
noun classes. In Ainu, which has two numeral classifiers (-nl-iw 'human'
and -pel-p 'non-human'), the plural suffix -utar is used just with human and
animate nouns (Onishi 1996a). In Japanese, number choices correlate with
humanness and animacy on the one hand and politeness on the other:
polite plural suffixes -tati (neutral), -ra (intimate), or -domo (humble,
 Appendix 1 437
derogatory) may be attached to a noun with an animate referent; -gata
'plural: polite' can only be attached to a noun with a human referent
(Onishi 1996b). In Slave and other Athabaskan languages (Rice 1989)
the expression of number is obligatory only with human referents.
If a language has optional number marking and/or agreement, fewer
choices are available for inanimates and non-humans. In Motuna, a
Papuan language from Bougainville with an elaborate multiple classifier
system and five noun classes, inanimate nouns do not distinguish dual and
paucal, but animate nouns do (Onishi 1994). See also Smith-Stark (1974).
(B) Animate/inanimate and human/non-human distinctions can be realized
in the way a language marks GRAMMATICAL RELATIONS. The choice of case
marker often depends on animacy, or humanness. In Awa Pit (or Cuaiquer,
Barbacoan: Curnow 1997: 66), accusative marking is obligatory only for
referential human NPs (Awa Pit has no other noun classification devices,
except, possibly, for a small set of noun classifiers: Curnow 1997: 121). In
Amele (Papuan) indirect objects (cross-referenced with an enclitic on the
verb) must have a human referent; and one class of verbs takes an optional
O clitic when a 3rd person singular O has a human or an animate referent
(Roberts 1987).
The number of case distinctions—indicating the grammatical function
of a noun—correlates with its animacy in agreement with the Nominal
Hierarchy proposed by Silverstein (1976); see also Dixon (1994); Frawley
(1992: 93-4) (and see Diagram 10.1 above). The choice between ergative
and nominative/accusative marking in languages where split ergativity is
conditioned by the semantics of nouns is controlled by the degree of
animacy of the subject: the more animate it is, the likelier it is to receive
nominative/accusative marking. The likelihood of ergative marking is
inversely proportional to the degree of animacy of the subject (Dixon
1994: 85).
One often gets more case distinctions for nouns with human or animate
reference than for those with non-human or inanimate reference, indepen-
dently of whether a language has agreement noun classes or not. In Spanish,
the preposition a is used to mark just human objects. In Yessan-Mayo, a
language with two genders from the East Sepik region of New Guinea,
the dative case marker -ni can also be used to mark O function just
with animate nouns (Foley 1986: 101); inanimate transitive objects are
unmarked.
Similarly, in Central Pomo a formal distinction between Sa and S0 is
made only if they refer to human referents (Mithun 1991: 521-3): mu-tu is
the form for 3rd person human S0, while another form, mu-l, is used to
cover human Sa and non-human Sa and S0. In Wara (Risto Sarsa, p.c.) and
Wipi (Dondorp and Shim 1997: 25), languages of the Fly River of New
438 Classifiers
Guinea, masculine and feminine genders are distinguished for absolutive
cross-referencing markers on the verb, but not for the ergative.
In Manambu, only proper nouns and personal pronouns are obliga-
torily marked for accusative when used as direct objects. In Bengali accu-
sative noun phrases are obligatorily marked by the suffix -ke when they
have definite human reference (Onishi 1996c). In both Dyirbal, Australian
(Dixon 1972: 43), and Dogon, from the Niger-Congo family (Plungian
1995: 12), there is an accusative marker, used for O function only with
proper names of people and some nouns with human reference such as the
kin terms 'mother' and 'father'.
Some Mayan languages—which have numeral classifiers, but not noun
classes—show an interesting dependency between the relative placement
of transitive subject and object on the nominal hierarchy and their con-
stituent order. When A and O are equal on the animacy hierarchy (e.g.
both are human, or animate), the order is VAO; when the A is higher on
the hierarchy than the O, the order is VOA (Campbell 2000; England
1991).
The existence of non-animacy-based classifier systems does not go
against the universality of animacy as a parameter for human categoriza-
tion. Even if animacy or humanness is not found in a classifier system, it is
likely to be present somewhere else in the language. Waris has animacy in
classificatory verbs, and verbal classifiers have no animacy distinctions;
there are also complicated dependencies between humanness or animacy
of the referent and the choice of markers for locative and allative cases
(Brown 1981: 3). In Ika the verbal prefix an- is used with transitive verbs to
indicate a non-human object (Frank 1990: 73); however, humanness or
animacy are not distinguished in classificatory verbs in this language.
(C) Animacy may be realized through the ways in which POSSESSION is
marked. The scope of -komo 'collective' in Hixkaryana (Carib) depends
on whether the possessor or the possessed are human: it refers to the
possessor, if it is human (ki-kanawa-rt-komo 'liNCL-canoe-poss-coLL' 'our
canoe'); and to the possessed if it is human (i-ha-ri-komo '3p-grandchild-
POSS-COLL' 'his grandchildren'). If both possessor and possessed are human
there is a potential ambiguity: ki-kuku-ru-komo can mean both 'our rela-
tives', or 'our (collective) relative' (Derbyshire 1985: 200). In many lan-
guages possessors must be animate, or human, as in Hua (Haiman 1980;
Onishi 1997a).
(D) The division of nouns into DECLENSIONAL PARADIGMS can reflect
animacy or humanness distinctions in languages with no other noun
classification devices. These can be remnants of a lost noun class system—
see the example of Lezgian in §10.6. Chukchee also has no genders, or
 Appendix 1 439
classifiers; but nouns are divided into declensional paradigms according to
the animacy of the referent (Michael Dunn, p.c.).
In languages with an independent system of noun classes or classifiers,
animacy reflected in the organization of paradigms may provide a basis for
an additional classification of nouns. So-called subgenders in Slavic
languages are based on animacy distinctions; the choice of accusative or
genitive case for O function and the choice of agreement forms depends on
the animacy of the head noun: see Corbett (1991: 165-7) and (C) in §2.4.4,
for an analysis of subgenders in Russian. For just a few words, animate/
inanimate declensions are used to distinguish homonyms. Koshki, plural of
'cat', belongs to the animate declension if it refers to the animals; koshki in
the inanimate declension refers to boots used by alpinists for climbing
mountains. The two words are obviously connected historically. Sometimes
shifting a noun from inanimate to animate declension type can be used to
achieve the stylistic effect of personifying an important referent. In the
famous poem by Gumilev, 'Captains', ships are personified through the use
of an animate accusative form (which is the same as genitive) instead of
the normal inanimate accusative (the same as nominative) (Rothstein
1973: 465).
(E) There are other ways in which animacy and humanness can be encoded
in grammatical systems outside noun classification devices. Personal pro-
nouns tend to be restricted to human referents. Kana (Kegboid: Ikoro
1996a) has almost no animacy distinctions in numeral classifiers; humans
are distinguished from non-humans in the ways they can be pronominal-
ized (Ikoro 1996b). In Hua (Papuan: Haiman 1980), a language without
any classifiers, personal pronouns must have human referents, and deictics
always have inanimate referents. Similarly, in Bengali 3rd person 'ordinary'
pronouns are used to refer to humans, and the demonstrative is used with
non-humans (Onishi 1996c). In Korafe (Binanderre family, New Guinea:
Cindi Farr, p.c.) personal pronouns are used only to refer to human beings
and personified non-human animates. In Jarawara, only animates (includ-
ing the personified sun, moon, and stars) can be referred to by the 3rd
person plural pronoun mee. In Lao, 3rd person pronoun man is normally
used for non-human referents, animate or inanimate; with human refer-
ents, it is only used for small children (Chapman 1996). In Chinese, the
personal pronoun ta can only be used with human referents (Walter
Bisang, p.c.). Imonda (Papuan) has no verbal classifiers based on animacy
or humanness; there are different coordinating clitics for non-humans
(-no), and for humans (-/) (Seiler 1985: 68-9). See also see the discussion of
Finnish in §13.2.
Japanese has two existential verbs: aru requires an S with an inanimate
referent; iru requires an S with an animate referent (LaPolla 1994). In
440 Classifiers
Hua, there are two copulas: one, bai-, is used with animate subjects, and
the other, no-, is used with inanimate subjects (Haiman 1980; Onishi
1997a).
Another technique languages use to distinguish animates from inanimates,
or humans from non-humans, is through different interrogative words.
Uralic languages typically have no noun classes, or classifiers of other
types,1 the distinction between the human interrogative 'who' (*ki) and
the non-human 'what' (*mika, *mi) goes back to Proto-Uralic (Collinder
1965: 138).
Animacy distinctions2 in interrogatives appear to be independent of
whether there are any other animacy-based noun classification devices.
For instance, Igbo has different forms for 'who' and 'what'; there are no
other animacy distinctions (but see above for derivation); and Tamazight
Berber languages (e.g. Zemmur, Seghrouchen, Izayan: Laoust 1928: 197),
which distinguish two genders, have just one form, mat, for both 'who' and
'what'.
(F) Animacy, humanness, and gender distinctions are often expressed
through derivation (see §10.9). Balto-Finnic languages have no noun
classes, or other classifiers; however, there are a few feminine derivational
suffixes, e.g. in Estonian: kuningas 'king', kuninganna 'queen', opilane 'stu-
dent (masculine or feminine)', opilanna 'female student'; laulaja 'singer
(masculine or feminine)', laula-tar 'female singer' (further examples are
given in n. 22 to Chapter 13).
Igbo has an animate/inanimate distinction in the way nouns are derived
from verbs. If the derived noun is inanimate the prefix is a homorganic
nasal; but if it is animate, the derivational prefix is a harmonizing vowel, 6-
or o-, e.g. ggwuu d3i 'yam digger' (an instrument), ogwuu d3i 'yam digger'
(a person) (from gwu d3i 'dig up yams') (Ikoro 1997).
Gender distinctions can also be realized in different discourse tech-
niques, proverbs, and terms of verbal abuse (see Braun forthcoming a,
for a case study of Turkish). Gender differences can also be reflected in
the ways different terms relating to professions are used; in Turkish,
'goldsmith' or 'taxi driver' most often has a male referent, while 'secretary'
or 'cleaning person' typically implies a female.
Thus, distinctions between animate/inanimate or human/non-human are
made on a grammatical level in the great majority of languages. There are
very few languages in which similar distinctions are made on the lexical
level only. One such example is Ewe (Kwa: Felix Ameka 1991, and p.c.).
1
However, Hungarian does have numeral classifiers, according to Beckwith (1992).
2
The semantics of 'who' and 'what' may differ in terms of scope even between related
languages; for example, 'who' refers to humans and higher animates in Russian, but only to
humans in Polish (Anna Wierzbicka, p.c.).
 Appendix 1 441
Ewe lacks a human/non-human or animate/inanimate distinction anywhere
in its grammar, even in interrogatives. A question marker ka can cooccur
with any noun, e.g nu ka (thing INTERROGATIVE) 'which thing, what', ame ka
(person INTERROGATIVE) 'which person, who', afi ka (place INTERROGATIVE)
'which place, where'.
 Appendix 2
From Nouns to Classifiers: Further Examples of
Semantic Change

(A) Body parts as sources for classifiers: semantic extensions

Body and body part terms can be used as classifiers for humans and/or
non-human animates, and for inanimates. In the two cases they undergo
different semantic changes.
Only the generic 'body' is used to classify humans and animates as a
verbal classifier, e.g. in Ngandi, or Mayali (Chapter 6). Both 'body' and
various body parts occur as numeral classifiers to classify humans and
animates. 'Body' may be used as a general classifier for 'person', cf. Katu
(Katuic, Mon-Khmer: Adams 1992: 109) chanak 'general human classifier',
from Proto-East Katuic *cak 'body'. Proto-Waic (Mon-Khmer) *ki? 'man,
body' is found as a human classifier in Northern Palaungic languages
(Adams 1992: 109).
'Body' is a numeral classifier for animals and sometimes for all animates
in Tai languages (with the form based on tua: Conklin 1981: 135-6). In
Thai, tua 'body' is becoming a generic classifier (Carpenter 1986). In
Burmese, kaun 'body' is used to classify animals, ghosts, dead bodies,
depraved people and children (Becker 1975).
Body parts may be used as classifiers for humans, and animates in
general. The use of body parts involves metonymy (a human, or an animate
in general is defined by its salient body part). Semantic change is from a
more specific to a more general referent. The body parts most frequently
used for classifying humans are 'head' and 'eye'.1 Some examples are given
below.

1
This semantic change corresponds to the following natural tendencies of semantic change
discovered by Wilkins (1981; 1996: 273): 'It is a natural tendency for a term for a visible
person-part to shift to refer to the visible whole of which it is a part . . . e.g. "navel" >
"belly" > "trunk" > "body" > "person"', and also 'it is natural tendency for a person-part
term to shift to refer to a spatially contiguous person part within the same whole ( e.g. "belly"
<-> "chest"; "skull" <-> "brain")'.
 Appendix 2 443
2
(Al) 'Top, head' > classifier for humans
This is widespread in Southeast Asian languages. In Nicobarese, koi 'head'
is used to classify humans (Adams 1989: 93). In Burmese, the body part
noun 'u 'head', which also has a more abstract meaning 'beginning, origin,
top', is used for people of status, scholars, and teachers (Becker 1975).
(A2) Head > classifier for animals
Khmer (Mon-Khmer) uses kba:l 'head' for animals, and also—pejoratively—
for humans (Adams 1989: 67); the same usage was observed for Bahnar
(Central Bahnaric) ko-l (Adams 1989: 71). In various South Bahnaric
languages, 'head' is also used for animals, e.g. Chrau voq (Adams 1989:
77) (note also English head of cattle (Lehrer 1986); and see §4.4.2 on
incipient classifier systems using 'head' for animal classification in Omani
Arabic).
(A3) 'Eye' > classifier for humans
This is widely attested in Austroasiatic languages. The word for 'pupil of
the eye'—probably the most salient part of the eye—can be used as a
human classifier (in agreement with the part-to-whole tendency formulated
by Wilkins). In North Bahnaric *ngaay 'human classifier' comes from an
item with the reconstructed meaning 'pupil of the eye' (Adams 1992: 110).
Vietnamese ngu'O'i 'person; human classifier' also comes from this lexeme
(Adams 1989: 84). Items meaning 'eye, pupil of the eye' may acquire a
more restricted reference. They can become classifiers which refer to the
social status of a person (honorifics), e.g. Bahnar (Central Bahnaric) mat
'eye, pupil of the eye' used as an honorific (the same morpheme means
'precious stone' in Mon) (Adams 1989: 71).
Rarer possibilities include:
(A4) Tail > classifier for humans
This is found in Aslian languages (Austroasiatic: Adams 1989: 94).
(AS) Tail > classifier for animals
Indonesian and Minangkabau use 'tail' as a classifier for non-human
animates (ekor and ikua respectively). Some Aslian languages also use
'tail' for classifying animals (Adams 1989: 94). Body parts may be used
to refer to subclasses of animals. In Khmer, kontuy 'tail' is used as a
classifier for fish (Adams 1989: 67).
(A6) Leg > classifier for humans
Totonac (Totonac-Tepehuan: Levy 1994) has a numeral classifier cha:
'human', from cha':n 'leg'.
2
I will not discuss the semantic change top <-> head, and the possibility of polysemy for
corresponding items; cf. however Wilkins (1981: 172): 'it is a natural tendency for a word for
"top" to take on the meaning of "head" and vice versa.'
444 Classifiers
A function-based semantic extension of a body part can yield classifiers
for humans. The word for hand or arm may become a classifier for an
artist, or artisan, i.e. a 'handy' person, e.g. Vietnamese tay (Adams 1989:
89).
(A7) Skin > non-human animates
This has been suggested as a source for Mai 'nang 'non-human animates',
cf. Tai nang 'skin' (Adams 1989: 61).
Names of specific body parts can be used as classifiers for species, or
genuses. The changes shell, skin, bark > crabs, and breasts and wings >
birds are attested in Sedang (North Bahnaric) (Adams 1989: 79).
When used to classify inanimates, body parts become shape- and size-
based classifiers, both when employed as numeral classifiers, and as verbal
classifiers (round head, or eye; long arm; short flexible finger). In a few
cases, the extension is achieved by function. The selection of an appro-
priate body part as a shape- or size-based classifier depends on the choice a
language makes. The shape and size parameters involved are ROUND, SMALL,
FLAT, VERTICAL, LONG FLEXIBLE, and THIN.
'Head' often becomes a classifier for ROUND objects. In Munduruku a2
'head' is a classifier for round objects (Gon9alves 1987: 24). Athabaskan
verbal classifier d which refers to roundish objects (see Chapter 6) is etymo-
logically related to *da 'head' in Proto-Eyak-Athabaskan (Jeff Leer, p.c.).
Another kind of extension, also found in numeral classifiers, concerns
extension by SHAPE and ARRANGEMENT, e.g. Thai Ma 'head', also used as a
classifier for clustered and root vegetables (Carpenter 1986).
'Eye' is a frequent source for classifiers for ROUND and SMALL objects. In
Kana, dEE 'numeral classifier for spot-like objects' derives from 'eye'. In
Caddo (Caddoan) the verbal classifier for small round objects, ic'ah-, is
also the noun stem for 'eye' (Mithun 1986: 390).
Other body parts can be used with a similar meaning, e.g. Yanomami
ko 'heart, kidney', employed as a classifier for round things (Ramirez
1994: 131). In Tarascan, the classifier for roundish objects comes from
'buttocks'.
A frequent source for a classifier for FLAT objects or objects with
extended surface is the word for skin. In Totonac mak-, a classifier for
flat and bounded objects, comes from makni 'body, skin'. In Baniwa and
Tariana -ya, a classifier for spread out, flat objects—such as skins—comes
from -ya 'skin', and in Kana dkpa, a classifier for flat objects, also comes
from the word for skin (Ikoro 1996a: 96). Yanomami (Ramirez 1994: 131)
si 'skin, bark' is a classifier for objects with an extended surface.
Another source for the classification of FLAT objects is the word for 'arm,
or wing'. In Totonacpaq- 'flat, unbounded' comes frompaqdn 'arm, wing'.
Another classifier for flat surfaces comes from 'face': Totonac laka- 'flat,
 Appendix 2 445
surface' and lakan 'face'. In Tarascan, the classifier for flat surfaces comes
from 'back'.
Classifiers for VERTICAL objects may come from the word for 'arm', e.g.
Munduruku ba4 'arm; classifier for long, roundish and rigid objects'
(Goncalves 1987: 25). In Tarascan 'neck' is the classifier for longish
objects.
'Finger' is a frequent source for classifiers for LONG FLEXIBLE or THIN
objects, e.g. Munduruku bi/ 'hand, finger' used as a classifier for long
flexible objects. Baniwa and Tariana -whi 'needle, thin thing' became a
classifier for thin longish objects.
Body parts may be used as sources for shape- and size-based classifiers of
particular species. In Yanomami u is a classifier for long larvae and fruits;
and is also the word for bone (Ramirez 1994). In Wu-Ming, klaau2 'head' is
used to classify particular round objects, e.g. onions or scallions.
There are also FUNCTIONAL extensions; for instance, Burmese IE? 'hand' is
used to classify handtools (Adams 1989: 139). Other functional extensions
may be less semantically transparent. According to Conklin (1981: 239,
437), mata 'eye' is the source of a classifier which refers to the working
parts of various instruments, such as knives, and other bladed things, in
Toba Batak. This classifier got extended to fishhooks, shovels, and brooms
in Gorontalo.
Body parts can be used as specific classifiers for inanimate objects as the
result of metaphoric extensions. These extensions often involve metonymy:
an object is classified after its salient part. In Vietnamese, khau 'mouth' is
used to classify firearms which have an opening similar to a mouth (Adams
1989: 144). In Minangkabau, kaki 'leg' is a specific classifier for umbrellas,
with their long bodies (Marnita 1996: 115).
Body parts can be used to classify more abstract items, via metonymy,
e.g. 'eye' > 'visible aspect, trait, feature', for example, in Dioi (Conklin
1981: 380).
Unexpected semantic extensions may be due to pre-existing polysemy. A
polysemous body part 'head, front, beginning' became the classifier for
'roots, tubers', as well as for round vegetables in Lii (Conklin 1981: 165).
Otherwise, 'head' is never used as a classifier for non-round objects.
Body parts which are associated with handling things (mouth for eating
or drinking, or hand/arm for grabbing or measuring) often give rise to
MENSURAL classifiers. The word for 'mouth' may become a mensural classi-
fier meaning 'bite, sip, mouthful', as in Totonac (Table 13.2). In Hmong
linear measures are associated with the hands, or their parts (Bisang 1993:
35), e.g. ntiv 'finger' used for a linear measure, a finger-length, or xib 'the
palm of the hand' used for an equivalent measure; cf. Kilivila yuma, yam,
yuma 'hand; measure: the span of two extended arms' (Senft 1993a);
Palikur -uku-l-wok- 'hand', classifier for hands or handfuls. See §13.1, for
446 Classifiers
more examples. The semantics of mensural classifiers can be quite idiosyn-
cratic. In Minangkabau, jari 'finger' is used only for measuring cloth, and
kapalo 'head' is used just for measuring the distance between horses in a
race (Marnita 1996: 131-2).
Mensural classifiers are only rarely derived from other body parts. Two
examples are Totonac ok 'units of length, e.g. meters' which comes from
akdn 'head', and pa:- 'units of volume, e.g. pails' which comes from pa:n
'belly' (Levy 1994).
(B) Sources for shape-based numeral classifiers
A typical source for a classifier for ROUND objects is the word for 'fruit', as in
Kam-Muang and White Tai (Conklin 1981: 154), and Western Austronesian
(Conklin 1981: 234-5; Marnita 1996). Another frequent source is 'stone',
e.g. Gorontalo botu 'stone' and 'classifier for spheres and fruits-like objects'
(Conklin 1981: 233); reflexes of the same stem are used to classify round
things in a number of Micronesian languages (Conklin 1981: 301); note also
Khmer krcep 'stone, kernel', 'classifier for round things' (Adams 1989: 108).
'Seed' and 'grain' are typical sources for SMALL ROUND things (Adams
1989: 104-5). Some examples are Thai med 'grain', used as a classifier for
small things like jewels and buttons (Carpenter 1986); Mon-Khmer me?
'seed, kernel' used as classifier for round things (Adams 1989: 104); and
Minangkabau incek 'seed', used as a classifier for seeds, grains, and small
round objects, e.g. eyes (Marnita 1996: 111). Another source for a classifier
for small and round things is 'egg', cf. Baniwa (Table 9.5).
For LONG and VERTICAL things, typical sources are 'tree' or 'tree trunk', or
'stick', e.g. Indonesian and Minangkabau batang 'tree, trunk', used as a
classifier for long vertical things, e.g. a stick, and Palikur ah 'tree trunk'
used as a verbal classifier for vertical things.
An additional semantic feature may be FLEXIBILITY and LENGTH; in
Gorontalo agu 'wood' is used to classify trees, and also flexible objects,
such as snakes, and 'anything that has the quality of length, physical or
metaphorical' (Conklin 1981: 260, 438).
'Leaf is a typical source for a classifier for FLAT FLEXIBLE objects, e.g.
Proto-Arawak *pana 'leaf, -pana 'classifier: leaf-like, flat'; this is also
found in numerous Mon-Khmer languages (Adams 1989: 155), and in
Dioi and numerous Tai languages (Conklin 1981: 158-60).
'Threads', 'strings', and 'veins' can be used to classify LONG FLEXIBLE items
(examples from Mon-Khmer languages are given in Adams 1989: 143).
The choice of item to be used as the source for a shape classifier is often
associated with material culture. Thus, Proto-Arawak -maka is the noun
for 'hammock', and also a classifier for 'extended cloth like objects'; it is
well known that the hammock is a central item in Amazonian culture (see
§12.3).
 Appendix 3
Fieldworker's Guide to Classifier Languages

The aim of this guide is to provide field linguists working on a previously

undescribed language with orientation as to the questions which should be
asked in order to establish a complete picture of how noun categorization
operates in a language; and, if the language has no classifiers, what other
categories can be used to express such universal categories as human/non-
human and animate/inanimate.1
PRELIMINARY INFORMATION is needed as a starting-point. After that,
questions relevant for establishing classifier systems and noun categoriza-
tion devices are divided into seven broad areas. After each question, a brief
explanation is given; relevant chapters of this book are indicated in
parentheses.
Preliminary information on the language includes:
(i) morphological type: e.g. isolating, agglutinating, fusional; synthetic,
analytic, polysynthetic; head-marking or dependent-marking;
(ii) word classes: open classes (e.g. nouns, verbs, adjectives) and closed
classes;
(iii) grammatical categories for open classes (e.g. number for nouns, tense
for verbs);
(iv) transitivity classes of verbs;
(v) marking of grammatical relations;
Nature of sources:
(vi) mostly based on texts with corroborative grammatical elicitation and
extensive lexical elicitation; or mostly based on elicitation.
One should concentrate on gathering and analysing texts in a language
near the beginning of any linguistic fieldwork. The classifier patterns
found in texts should then be confirmed and systematically studied through

' It is based on the author's own field experience in different parts of the world, student
supervision in Brazil and Australia, reading of grammars, and talking to other linguists about
their field experiences.
448 Classifiers
carefully directed elicitation. Lexical elicitation is essential to work out the
semantics of classifiers and their assignment principles. If at all possible,
the researcher should take into account—but not entirely rely on—the
intuitions of native speakers when trying to account for seemingly opaque
semantics of genders or classifiers. Extensive work with texts is recom-
mended to understand the discourse use of classifiers and agreement var-
iation.
(A) Establishing types of classifier and their inventories in given morpho-
syntactic contexts (Chapters 2-9)
(Al) Are any noun categorization devices used within an NP? Test the
appearance of classifier-like markers (i) on adjectival modifiers within a
noun phrase; (ii) on demonstratives, articles, interrogatives (test for noun
classes: Chapter 2, and/or deictic classifiers: Chapter 7); (iii) within an NP
to characterize a noun by itself, generic-specific type (test for noun
classifiers: Chapter 3); (iv) on numerals and quantifiers (test for numeral
classifiers: Chapter 4).
(A2) If there is agreement in noun class in NP or clause, is it obligatory? If
not, does it depend on (a) discourse saliency of a noun; (b) its topicality or
topical continuity; (c) contrastive focus; (d) first mention of a noun in
discourse; (e) definiteness of a noun? If it is, are there any variations in
the choice of agreement form depending on (a-e) or other factors?
In a language with classifiers, can two or more nouns belonging to
different classes be coordinated? What classifier is then chosen (test for
classifier resolution)? Are there any formal or functional markedness
relations in the classifier system? Is there any zero-agreement, or neutral/
default agreement? (See §2.5.1.)
(A3) Is there any overt noun class marking (§2.6 in Chapter 2)? If there
are classifiers, are they bound or free morphemes (suffixes, prefixes, etc.)?
(A4) Are there any noun categorization devices used within a possessive
NP referring to (i) the ways a referent of the noun can be possessed, or
handled, e.g. eaten, or drunk, or sold; (ii) intrinsic properties of the referent
of the possessed or of the possessor (e.g. shape, animacy)? (Tests for
relational, possessed and possessor classifiers respectively: Chapter 5.)
(A5) Are there any noun categorization devices marked on adpositions (if
there are any)? (Test for locative classifiers: Chapter 7.)
(A6) On the verb: does the form of a verb (transitive or intransitive)
change depending on (i) the physical orientation of its argument; (ii) other
properties—animacy, shape, etc.—of its argument? (Test for verbal classi-
fiers and classificatory verbs: Chapter 6.)
 Appendix 3 449
(A 7) If a classifier can appear on the head noun and on a modifier, are
there any differences in its (i) form; (ii) meaning; (iii) syntactic behaviour?
(A8) How many classifiers are there?
(A9) Are there different choices of classifiers available associated with
different types of modifier? For instance, different classifier choices may
be made for demonstratives and for adjectives in head-modifier NPs ('split
agreement' system). Is there any double marking?
(A10) If there is more than one set of classifier morphemes, used in
different morphosyntactic contexts, do they overlap? How are they differ-
ent? (Chapter 8.)
(A11) If the single set of classifier morphemes is used in several different
morphosyntactic environments, are there any differences in form or in
meaning? (Test for a multiple classifier language: Chapter 9.)
(B) Correlations with other categories (Chapter 10)
(B1) Do classifier distinctions differ depending on number?
(B2) Do they differ depending on grammatical relations (e.g. genders may
be distinguished for O, but not for the subject (A/S))?
(B3) Does the choice of possessive classifier depend on an alienable/
inalienable distinction?
(B4) Are any classifier distinctions neutralized in certain verbal tenses, or
voices?
(B5) Are there fewer classifier distinctions in some deictic categories?
(B6) Can the use of different classifiers distinguish different meanings of a
word?
(B7) Does the presence of classifiers affect the structure of the lexicon?
That is, in languages with large systems of classifiers the assignment of
which is based on shape, there may be fewer lexical items which refer to
the size of the object (e.g. big, large, round). Is this true for your
language?
(C) Semantics of noun categorization devices (Chapter 11)
(C1) How are classifiers assigned to humans; non-human animates; inan-
imate concrete objects? How are higher animates (mammals) distinguished
from lower animates?
(C2) How are classifiers assigned to body parts?
450 Classifiers
(C3) How are classifiers assigned to abstract nouns and natural phenom-
ena? How are classifiers assigned to loans?
(C4) Is there any default or general classifier? If so, how is it used: (i) for
otherwise unclassifiable items ('residue' classifier), (ii) for unspecified or
unknown referent ('unspecified referent' classifier); (iii) can it be substi-
tuted for other classifiers under special pragmatic conditions ('default'
classifier)?
(C5) Are there any semantic extensions to classifiers (e.g. 'male/female' to
'big/little'; 'female' to 'cute and small')?
(D) Functions of classifiers (Chapter 12)
(D1) Are classifiers obligatory?
(D2) If not, does their use correlate with any of the parameters in (A2)?
Can a classifier be 'dropped', or can a noun be 'dropped', and what is the
effect of this?
(D3) If yes, does the choice of a classifier depend on any of (a-c) in (A2)?
Are any noun categorization devices marked on the noun itself? If there are
numeral or other classifiers are they also used as derivational morphemes
(cf. (E) in §9.1)?
(D4) Can a classifier be used without a head noun (i.e. anaphorically)?
(D5) What are the functions of classifiers in narratives? Does the use of
classifiers depend on textual genre?
(D6) Does the semantic organization of classifiers correlate with extra-
linguistic factors, e.g. the world-view and/or mythological concepts of the
speakers? Do social factors affect the composition of each class? Is there
any correlation between classifiers and culture, or language planning?
(E) Origin and acquisition of classifiers (Chapters 13 and 14)
(E1) Is there any information about the origin of classifiers (e.g. devel-
oped from nouns, or from verbs)?
(E2) Is there any information on the semantic evolution of classifiers?
(E3) Are there any data on the behaviour of classifiers in (i) language
obsolescence; (ii) the speech of bilinguals and semilinguals; (iii) language
diffusion; (iv) language reduction or expansion?
(E4) Is there any information on generational differences in the use and
production of classifiers? Are there any data on how children employ
classifiers? (Chapter 14)
 Appendix 3 451
(F) Animacy and humanness outside noun categorization devices (Appendix
1)
(F1) Does number marking or number agreement depend on whether the
referent of a noun is human, or animate?
(F2) Does the way in which grammatical relations are marked (by case or
cross-referencing) depend on the animacy or humanness of the argument
(A, S, or O)?
(F3) Are there any derivational devices used for animate or for human
nouns?
(F4) Are there any differences in the ways personal pronouns are used
with nouns with human and with non-human referents?
(F5) Are there any humanness or animacy distinctions in interrogatives?
 References

Abbott, M. (1991). 'Macushi', in Derbyshire and Pullum (1991: 23-160).

Adams, K. L. (1989). Systems of Numeral Classification in the Mon-Khmer,
Nicobarese and Aslian Subfamilies of Austroasiatic. Canberra: Pacific Linguistics.
—(1992). 'A Comparison of the Numeral Classification of Humans in Mon-
Khmer', Mon-Khmer Studies 21: 107-29.
and Conklin, N. F. (1973). 'Towards a Theory of Natural Classification',
Papers from the Annual Regional Meeting of the Chicago Linguistic Society 9:
1-10.
Ahrens, K. (1994). 'Classifier Production in Normals and Aphasics', Journal of
Chinese Linguistics 22: 203-46.
Aikhenvald, A. Y. (1984). 'A Structural and Typological Classification of Berber
Languages'. PhD thesis, Institute of Oriental Studies, Moscow.
—(1986). 'On the reconstruction of syntactic system in Berber-Lybic', Zeitschrift
fur Phonetik, Sprachwissenschaft und Kommunikationsforschung 39: 527-39.
—(1990). Sovremennyj ivrit [Modern Hebrew]. Moscow: Nauka.
—(1994a). 'Classifiers in Tariana', Anthropological Linguistics 34: 407-65.
—(1994b). 'Grammatical Relations in Tariana', Nordic Journal of Linguistics 7:
201-18.
—(1995a). Bare. Languages of the World/Materials 100. Munich: Lincom
Europa Materials 100.
—(1995b). 'Person-marking and Discourse in North-Arawak Languages', Studia
Linguistica 49: 152-95.
—(1996a). 'Areal Diffusion in Northwest Amazonia: The Case of Tariana',
Anthropological Linguistics 38: 73-116.
——(1996b). 'Classe nominal e genero nas h'nguas Aruak', Boletim do museu
Goeldi 10: 137-259.
—(1996c). 'Classifiers in Baniwa', Moscow Linguistic Journal 3: 7-33.
—(1998a). 'Physical Properties in a Gender System: An Example from Manambu,
a Ndu Language of New Guinea', Language and Linguistics in Melanesia 27:
175-87.
—(1998b). 'Warekena', in Derbyshire and Pullum (1998: 215–39).
—(1999a). 'Arawak Languages', in Dixon and Aikhenvald (1999: 65-106).
—(1999b). 'Double Marking of Syntactic Function in Tariana', in E. Rakhilina
and Y. Testelets (eds.), Typology and Linguistic Theory: From Description to
Explanation. For the 60th Birthday of Aleksandr E. Kibrik. Moscow: Languages
of Russian Culture, 114–22.
(forthcoming a). 'Gender', in C. Lehmann and J. Mugdan (eds.), Handbuch
der Morfologie. Berlin: Mouton de Gruyter, article 98.
— (forthcoming b). 'Classifiers with Demonstratives and Articles in a Multiple
 References 453
Classifier System: A Case Study from North West Amazonia', in G. Senft (ed.),
Systems of Noun Classification. New York: Oxford University Press.
— (forthcoming c). 'Areal Diffusion, Genetic Inheritance and Problems of Sub-
grouping: A North Arawak Case Study', in A. Y. Aikhenvald and R. M. W.
Dixon (eds.), Areal Diffusion and Genetic Inheritance: Problems in Comparative
Linguistics. Oxford: Oxford University Press.
— (forthcoming d). 'Typological Distinctions in Word Formation', in T. Shopen
(ed.), Language Typology and Syntactic Description. Cambridge: Cambridge
University Press.
—(in preparation). A Grammar of Tariana.
— MS. 'Noun Phrase Structure in Paumari: An Interaction of Gender and Noun
Class'.
— and Dixon, R. M. W. (1998). 'Dependencies between Grammatical Systems',
Language 74: 56–80.
(1999). 'Other Small Families and Isolates', in Dixon and Aikhenvald
(1999: 341-83).
—and Green, D. (1998). 'Palikur and the Typology of Classifiers', Anthropo-
logical Linguistics 40: 429-80.
Aksenova, I., and Toporova, I. (1990). Vvedenije v Bantuistiku [Introduction to
Bantu Linguistics]. Moscow: Nauka.
—— —— (1994). Jazyk Kuria [The Kuria Language]. Moscow: Nauka.
Alexandre, P. (1968). 'Note sur la reduction du systeme des classes dans les
languues vehiculaires a fonds bantu', La Classification nominale, 277-90.
Alexeyev, M. (1985). Voprosy sravniteljno-istoricheskoj grammatiki lezginskih
jazykov. Morfologija. Sintaksis. [Problems of comparative and historical grammar
of Lezghic languages. Morphology. Syntax]. Moscow: Nauka.
Allan, K. (1977). 'Classifiers', Language 53: 284–310.
—(1980). 'Nouns and Countability', Language 56: 341-567.
Allin, T. (1975). A Grammar of Resigaro'. PhD thesis, University of St Andrews.
Alpher, B. (1987). 'Feminine as the Unmarked Grammatical Gender: Buffalo Girls
Are No Fools', Australian Journal of Linguistics 7: 169-87.
Amaya, M. T. (1997). Categorias gramaticales del ette taara. Lengua de los chimilas.
Bogota: Centro Ediciones CCELA, Uniandes.
Ameka, F. (1991). 'Ewe: Its Grammatical Constructions and Illocutionary
Devices'. PhD thesis, Australian National University.
Anderson, J. L. (1989). Comaltepec Chinantec Syntax. Dallas: Summer Institute of
Linguistics and University of Texas at Arlington.
Anderson, L.-G., and Janson, T. (1997). Languages in Botswana: Language Ecology
in Southern Africa. Botswana: Longman.
Anderson, M., and Anderson, T. (1991). Sudest Grammar Essentials. Ukarumpa:
Summer Institute of Linguistics.
Anderson, N., and Anderson, C. (1976). Podopa Grammar Essentials. Ukarumpa:
Summer Institute of Linguistics.
Anderson, S. R. (1992). A-Morphous Morphology. Cambridge: Cambridge Univer-
sity Press.
—and Keenan, E. (1985). 'Deixis', in T. Shopen (ed.), Language Typology and
454 Classifiers
Syntactic Description, iii: Grammatical Categories and the Lexicon. Cambridge:
Cambridge University Press, 259–308.
Andrews, E. (1990). Markedness Theory: The Union of Asymmetry and Semiosis in
Language. Durham, NC: Duke University Press.
Aronoff, M. (1991). 'Noun Classes in Arapesh', Yearbook of Morphology. Dordrecht:
Foris, 21-30.
—(1994). Morphology by Itself. Cambridge: MIT Press.
Aschmann, R. P. (1993). Proto-Witotoan. Summer Institute of Linguistics and
University of Texas at Arlington Publications in Linguistics. Dallas: Summer
Institute of Linguistics and University of Texas at Arlington.
Asher, R. E. (1985). Tamil. London: Croom Helm.
Austin, P. (1981). A Grammar of Diyari, South Australia. Cambridge: Cambridge
University Press.
Axelrod, M. (forthcoming). 'Gender and Aspect in Koyukon'.
Baarle, P. van, and Sabajo, M. (1997). Manuel de la langue Arawak. Paris: Editions
du Saule.
Bani, E. (1987). 'Masculine and Feminine Grammatical Gender in Kala Lagaw Ya',
Australian Journal of Linguistics 7: 189-201.
Bantia, T. K. (1993). Punjabi. London: Routledge.
Barlow, M. (1992). A Situated Theory of Agreement. New York: Garland.
—and Ferguson, C. A. (1988a). 'Introduction', in Barlow and Ferguson (1988b:
1-22).
(eds.) (1988b). Agreement in Natural Language: Approaches, Theories,
Descriptions. Stanford: Center for the Study of Language and Literature.
Barnes, J. (1990). 'Classifiers in Tuyuca', in Payne (1990: 273-92).
Baron, N. S. (1971). 'A Reanalysis of English Grammatical Gender', Lingua 27:
113-40.
Baron, R. (1987). The Cerebral Computer. Hillsdale, NJ: Lawrence Erlbaum.
Barron, R. (1982). 'Das Phanomen klassifikatorischer Verben', in Seiler and
Lehmann (1982: 133–46).
—and Serzisko, F. (1982). 'Noun Classifiers in the Siouan Languages', in Seiler
and Stachowiak (1982: 85-105).
Barsalou, L. W. (1987). 'The instability of graded structure: implications for the
nature of concepts', in U. Neisser (ed.), Concepts and Conceptual Development.
Cambridge: Cambridge University Press, 101–40.
Barz, R. K., and Diller, A. V. N. (1985). 'Classifiers and Standardisation: Some
South and South-East Asian Comparisons', in D. Bradley (ed.), Language Policy,
Language Planning and Sociolinguistics in South-East Asia. Canberra: Pacific
Linguistics, 155-84.
Bascom, B. (1982). 'Northern Tepehuan', in R. W. Langacker (ed.), Studies in
Uto-Aztecan Grammar, iii: Uto-Aztecan Grammatical Sketches. Dallas: Summer
Institute of Linguistics and University of Texas at Arlington, 267-315.
Basso, K. H. (1968). The Western Apache Classificatory Verb System: A Formal
Analysis', Southwestern Journal of Anthropology 24: 252-66.
Bechert, J. (1982). 'Grammatical Gender in Europe: An Areal Study of a Linguistic
Category', Papiere zur Linguistik 26: 23-34.
 References 455
Becker, A. J. (1975). 'A Linguistic Image of Nature: The Burmese numerative
classifier system', Linguistics 165: 109-21.
—(1986). 'The Figure a Classifier Makes: Describing a Particular Burmese
Classifier', in Craig (1986a: 327–43).
Beckwith, C. I. (1992). 'Classifiers in Hungarian', in I. Kenesei and Cs. Pleh (eds.),
Approaches to Hungarian, iv: The Structure of Hungarian. Szeged: JATE, 197-206.
—(1998). 'Noun Specification and Classification in Uzbek', Anthropological
Linguistics 40: 124–40.
Benton, R. A. (1968). 'Numeral and Attributive Classifiers in Truquese', Oceanic
Linguistics 7: 104–46.
Berlin, B. (1968). Tzeltal Numeral Classifiers: A Study in Ethnographic Semantics.
The Hague: Mouton.
Berman, R. (1978). Modern Hebrew Structure. Tel Aviv: University Publishing
Projects.
—(1985). 'The Acquisition of Hebrew', in D. I. Slobin (ed.), The Crosslinguistic
Study of Language Acquisition. Hillsdale, NJ: Lawrence Erlbaum, 255-371.
Bhaskararao, P., and Joshi, S. K. (1985). 'A Study of Newari Classifiers', Bulletin of
the Deccan College Research Institute 44: 17-31.
Birk, D. B. W. (1976). The Malak Malak Language, Daly River, Western Arnhem
Land. Canberra: Pacific Linguistics.
Bisang, W. (1993). 'Classifiers, Quantifiers and Class Nouns in Hmong', Studies in
Language 17: 1-51.
—(1996). 'Areal Typology and Grammaticalization: Processes of Grammatic-
alization Based on Nouns and Verbs in East and Mainland South East Asian
Languages', Studies in Language 20: 519-98.
(forthcoming). 'Classifiers in East and Southeast Asian languages: Counting
and Beyond', in J. Gvozdanovic (ed.), Changes in Numeral Systems. Berlin:
Mouton de Gruyter.
Blankenship, B. (1996). 'Classificatory Verbs in Cherokee', in P. Munroe (ed.),
Cherokee Papers from UCLA. California: UCLA, 61-74.
—(1997). 'Classificatory Verbs in Cherokee', Anthropological Linguistics 39:
92–110.
Blewett, K. (forthcoming). Kunua Grammar Essentials. Ukarumpa: Summer Institute
of Linguistics.
Bloomfield, L. (1933). Language. New York: Holt, Rinehart & Winston.
Boas, F, and Deloria, E. (1941). Dakota Grammar. Washington: US Government
Printing Office.
Bohme, R., and Levelt, W. J. M. (1979). 'Children's Use and Awareness of Natural
and Syntactic Gender in Possessive Pronouns', paper presented to the Conference
on Linguistic Awareness and Learning to Read, Victoria.
Bokamba, E. G. (1977). 'The Impact of Multilingualism on Language Structures:
The Case of Central Africa', Anthropological Linguistics 19: 181-202.
—(1985). 'Verbal Agreement as a Noncyclic rule in Bantu', in D. L. Goyvaerts
(ed.), African Linguistics: Essays in Memory of M. W. K. Semikenke. Amsterdam:
John Benjamins, 9-54.
Bolt, J. E., Hoddinott, W. G., and Kofod, F. M. (1971). An Elementary Grammar of
the Nungali Language of the Northern Territory. Armidale: Mimeo.
456 Classifiers
Borgman, D. M. (1990). 'Sanuma', in Derbyshire and Pullum (1990: 15-248).
Borneto, S. C. (1996). 'Liegen and stehen in German: A Study in Horizontality and
Verticality', in E. Casad (ed.), Cognitive Linguistics in the Redwoods: The Expan-
sion of a New Paradigm in Linguistics. Berlin: Mouton de Gruyter, 459-506.
Braun, F. (forthcoming a). 'Genderless = Gender-neutral? Empirical evidence from
Turkish'.
—(forthcoming b). 'The Communication of Gender in Turkish'.
Breedveld, J. O. (1995). 'The Semantic Basis of Noun Class Systems: The Case of
the KE and NGE Classes in Fulfulde', Journal of West African Languages 25: 63-
74.
Breen, G. (1976a). 'Wangkumara', in R. M. W. Dixon (ed.), Grammatical Categories
in Australian Languages. Canberra: Australian Institute of Aboriginal Studies,
336–9.
—(1976b). 'Wagaya', in R. M. W. Dixon (ed.), Grammatical Categories in Aus-
tralian Languages. Canberra: Australian Institute of Aboriginal Studies, 340-2,
590–4.
Bresnan, J., and McChombo, S. A. (1986). 'Grammatical and Anaphoric Agree-
ment', Papers from the Annual Regional Meeting of the Chicago Linguistic Society
22: 278-97.
Broschart, J. (1997). 'Locative classifiers in Tongan', in G. Senft (ed.), Referring to
Space: Studies in Austronesian and Papuan Languages. Oxford: Clarendon Press,
287-315.
Brosman, P. W. Jr. (1979). 'The Semantics of the Hittite Gender System', Journal of
Indo-European Studies 7: 227-36.
Brown, R. (1981). 'Semantic Aspects of some Waris Predications', in K. J. Franklin
(ed.), Syntax and Semantics in Papua New Guinea Languages. Ukarumpa:
Summer Institute of Linguistics, 93-123.
Bruce, L. (1984). The Alamblak Language of Papua New Guinea East Sepik.
Canberra: Pacific Linguistics.
Bulygina, T. V., and Shmelev, A. D. (1996). 'Nespecifirovannyj pol i soglasovanie pri
anafore' ['Unspecified Gender and Anaphoric Agreement'], Moscow Linguistic
Journal 2: 98-103.
Burling, R. (1965). 'How to Choose a Burmese Numeral Classifier', in M. E. Spiro
(ed.), Context and Meaning in Cultural Anthropology, in Honor of A. Irving
Hallowell. New York: Free Press, 243-64.
— (1973). 'Language Development of a Garo and English-speaking Child', in
C. F. Ferguson and D. I. Slobin (eds.), Studies of Child Language Development.
San Francisco: Holt, Rinehart & Winston, 69-90.
Burton, M., and Kirk, L. (1976). 'Semantic Reality of Bantu Noun Classes: The
Kikuyu Case', African Linguistics 7: 157-74.
Bybee J., Perkins, R., and Pagliuca, W. (1994). The Evolution of Grammar. Chicago:
University of Chicago Press.
Campbell, L. (1985). 'Review of J. A. Suarez, The Mesoamerican Indian Languages'.
Journal of Linguistics 21: 216-21.
—(1993). 'On Proposed Universals of Grammatical Borrowing', in H. Aertsen and
 References 457
R. J. Jeffers (eds.), Historical Linguistics (1989): Papers from the 9th International
Conference on Historical Linguistics. Amsterdam: John Benjamins, 91-109.
— (1997). American Indian Languages. New York: Oxford University Press.
— (2000). 'Valency Changing Derivations in K'iche', in R. M. W. Dixon and
A. Y. Aikhenvald (eds.), Changing Valency: Case Studies in Transitivity.
Cambridge: Cambridge University Press, 236–81.
— Kaufman, T., and Smith-Stark, T. (1986). 'Meso-America as a Linguistic
Area', Language 62: 530–70.
and Muntzel, M. (1989). 'The Structural Consequences of Language Death',
in Dorian (1988: 181-96).
Capell, A., and Hinch, H. E. (1970). Maung Grammar, Texts and Vocabulary. The
Hague: Mouton.
Carlson, R., and Payne, D. (1989). 'Genitive Classifiers', in Proceedings of the 4th
Annual Pacific Linguistics Conference. Eugene: University of Oregon, 89–119.
Carlson, T. (1991). Taenae Grammar Essentials. Ukarumpa: Summer Institute of
Linguistics.
Carpenter, K. (1986). 'Productivity and Pragmatics of Thai Classifiers', Berkeley
Linguistics Society: Proceedings of the Annual Meeting 12: 14-25.
—(1987). 'How Children Learn to Classify Nouns in Thai'. PhD thesis, Stanford
University.
(1991). 'Later rather than Sooner: Children's Use of Extralinguistic Inform-
ation in the Acquisition of Thai Classifiers', Journal of Child Language 18:
93-113.
(1992). Two Dynamic Views of Classifier Systems: Diachronic Change and
Individual Development', Cognitive Linguistics 3: 129-50.
Carroll, J. B., and Casagrande, J. B. (1958). 'The Function of Language Classifica-
tion in Behavior', in E. Maccoby, T. Newcomb, and E. L. Hartley (eds.), Readings
in Social Psychology. New York: Holt, Rinehart & Winston, 8-31.
Carter, R. M. (1976). 'Chipewyan Classificatory Verbs', International Journal of
American Linguistics 42: 24–30.
Casad, E. (1982). 'Cora', in R. W. Langacker (ed.), Studies in Uto-Aztecan Gram-
mar, iv: Southern Uto-Aztecan Grammatical Sketches. Dallas: Summer Institute
of Linguistics and University of Texas at Arlington, 151–459.
—(1996). 'What Good Are Locationals, Anyway?' in M. Putz and R. Dirven
(eds.), The Construal of Space in Language and Thought. Berlin: Mouton de
Gruyter, 239-67.
Ceria, V. G., and Sandalo, F. (1995). 'A Preliminary Reconstruction of Proto-
Waikuruan with Special Reference to Pronominals and Demonstratives',
Anthropological Linguistics 37: 169-91.
Chafe, W. L. (1967). Seneca Morphology and Dictionary. Washington, DC:
Smithsonian Press.
—(1977). 'The Evolution of Third Person Verb Agreement in the Iroquoian
Languages', in C. N. Li (ed.), Mechanisms of Syntactic Change. Austin: University
of Texas Press, 493-524.
—(1994). Discourse, Consciousness, and Time: The Flow and Displacement of
458 Classifiers
Conscious Experience in Speaking and Writing. Chicago: University of Chicago
Press.
Chang-Smith, M. (1996). 'Chinese Noun Classification in Relation to Prototype
Theory'. MA thesis, University of Queensland.
Chapman, A. (1996). 'A Grammatical Summary of Lao', materials for the research
project 'Universals of Human Languages', Research Centre for Linguistic
Typology, Australian National University.
Chapman, S. MS. Dicionario Paumari. Summer Institute of Linguistics, Porto
Velho, Brazil.
—and Derbyshire, D. (1991). 'Paumari', in Derbyshire and Pullurn (1991: 161-
354).
Chappell, H., and McGregor, W. (1989). 'Alienability, Inalienability and Nominal
Classification', Proceedings of the Berkeley Linguistic Society 15: 24-36.
— — (eds.). (1996). The Grammar of Inalienability: A Typological Perspective
on Body Part Terms and the Part–Whole Relation. Berlin: Mouton de Gruyter.
Charney, J. O. (1993). A Grammar of Comanche. Lincoln: University of Nebraska
Press.
Charters, H. (1995). 'What Are the Consequences of Classifier Languages for
Semantic Theory?' MS, Department of Linguistics, Australian National
University.
Chatterji, S. K. (1926/1970–2). The Origin and the Development of the Bengali
Language. 2 vols., London: Allen & Unwin.
Childs, T. (1983). 'Noun Class Affix Renewal in Southern West Atlantic', in J. Kaye,
H. Koopman, D. Sportiche, and A. Dugas (eds.), Current Approaches to African
Linguistics, ii. Dordrecht: Foris, 17-30.
—(1993). Grammar of Kisi. Berlin: Mouton de Gruyter.
Christensen, S. (1995). Yonggom Grammar Essentials. Ukarumpa: Summer Institute
of Linguistics.
Churchward, C. M. (1941). A New Fijian Grammar. Australasian Medical Publishing
Company.
—(1953). A Grammar of Tongan. Oxford: Oxford University Press.
Clamons, C. R. (1993). 'Gender Assignment in Oromo', in M. Eid and G. Iverson
(eds.), Principles and Prediction: The Analysis of Natural Language. Amsterdam:
John Benjamins, 269–84.
— (1995). 'How Recent Contact Erased Ancient Traces in the Gender Systems of
the Oromo Dialects', Berkeley Linguistic Society: Parasession: 389–400.
Clark, E. V. (1977). 'Universal Categories: on the Semantics of Classifiers and
Children's Early Word Meaning', in A. Juilland (ed.), Linguistic Studies Offered
to Joseph Greenberg. Saratoga: Alma Libri, 449-62.
Classification nominale (1967). La Classification nominale dans les langues negro-
africaines. Paris: CNRS.
Claudi, U. (1985). Zur Entstehung von Genussystemen. Hamburg: Buske.
Codrington, R. (1885). The Melanesian Languages. Oxford: Clarendon Press; repr.
1974, Amsterdam, Philo Press.
Cohen, P. (1976). 'The Noun Phrase in Jeh', Mon-Khmer Studies 5: 139-52.
 References 459
Coleman, C. (1982). 'A Grammar of Gunbarlang with Special Reference to
Grammatical Relations'. Honours subthesis, Australian National University.
Collinder, B. (1965). An Introduction to the Uralic Languages. Berkeley: University
of California Press.
Comrie, B. (1978). 'Genitive-Accusatives in Slavic: The Rules and Their Motiva-
tion', International Review of Slavic Linguistics 3: 27–42.
—(1981). The Languages of the Soviet Union. Cambridge: Cambridge University
Press.
and Stone, G. (1977). The Russian Language since the Revolution. Oxford:
Clarendon Press.
Conklin, N. F. (1981). The Semantics and Syntax in Numeral Classification in Tai
and Austronesian'. PhD thesis, University of Michigan.
Connelly, M. J. (1984). 'Basotho Children's Acquisition of Noun Morphology'.
PhD thesis, University of Essex.
Conrad, R. (1978). 'Some Muhiang Grammatical Notes', in Miscellaneous Papers
on Dobu and Arapesh. Workpapers in Papua New Guinea Languages 25: 89-130.
Ukarumpa: Summer Institute of Linguistics.
—(1996). 'More on Arapesh Noun Classes', MS.
—and Conrad, J. (n.d.). Iwam Essentials for Translation. Ukarumpa: Summer
Institute of Linguistics.
Conrad, R. I, with Wogiga, K. (1991). An Outline of Bukiyip Grammar. Canberra:
Pacific Linguistics.
Cooke, J. R. (1986). Pronominal Reference in Thai, Burmese, and Vietnamese. Los
Angeles: University of California Press.
Corbett, G. (1979). 'The Agreement Hierarchy', Journal of Linguistics 15: 203-24.
—(1983a). 'Resolution Rules: Agreement in Person, Number, and Gender', in
G. Gazdar, E. Klein, and G. K. Pullum (eds.), Order, Concord, and Constituency.
Dordrecht: Foris, 176–206.
—(1983b). Hierarchies, Targets and Controllers: Agreement Patterns in Slavic.
London: Croom Helm.
—(1988). 'Agreement: A Partial Specification Based on Slavonic Data', in
Barlow and Ferguson (1988b: 23-54).
—(1991). Gender. Cambridge: Cambridge University Press.
—(1994a). 'Agreement', in R. E. Asher (ed.), The Encyclopedia of Language and
Linguistics, i. Oxford: Pergamon Press, 54–60.
(1994b). 'Types of Typology, Illustrated from Gender Systems', in Agreement,
gender, number, genitive. Eurotyp Working Papers: Team 7: Noun Phrase Struc-
ture. Working Paper 23: 1-38. Konstanz: Eurotyp Programme in Language
Typology, European Science Foundation.
Corominas, J. (1954). Diccionario critico etimologico de la lengua castellana, iii.
Berna: Francke.
Costenla Umana, A. (1991). Las lenguas del area intermedia: introducion a su
estudio areal. San Jose: Editorial de la Universidad de Costa Rica.
Counts, D. L. (1969). A Grammar of Kaliai-Kove. Honolulu: University of Hawaii
Press.
Craig, C. G. (1977). The Structure of Jacaltec. Austin: University of Texas Press.
460 Classifiers
Craig, C. G. (ed.) (1986a). Noun Classes and Categorization: Proceedings of a
Symposium on Categorization and Noun Classification, Eugene, Oregon, October
1983. Amsterdam: John Benjamins.
—(1986b). 'Jacaltec Noun Classifiers', Lingua 70: 241–84.
—(1986c). 'Jacaltec Noun Classifiers: A Study in Language and Culture', in
Craig (1986a: 263-94).
(1991). 'Ways to Go in Rama: A Case Study in Polygrammaticalization', in
Traugott and Heine (1991a: ii, 455-92).
— (1992). 'Classifiers in a Functional Perspective', in M. Fortescue, P. Harder,
and L. Kristoffersen (eds.), Layered Structure and Reference in a Functional
Perspective. Amsterdam: John Benjamins, 277-301.
— (1993). 'A Morphosyntactic Typology of Classifiers', contribution to the
workshop 'Back to Basic Issues in Nominal Classification', Cognitive Anthro-
pology Research Group, Max-Planck Institute for Psycholinguistics, Nijmegen.
— (1996). 'Nominal Classification', plenary talk at the 3rd Australian Linguistic
Institute, Canberra.
(forthcoming). 'Classifiers', in C. Lehmann and J. Mugdan (eds.), Handbuch
der Morfologie, Berlin: Mouton de Gruyter, article 97.
Creider, C. A. (1975). 'The Semantic System of Noun Classes in Proto-Bantu',
Anthropological Linguistics 17: 127-38.
Croft, W. (1990). Typology and Universals. Cambridge: Cambridge University Press.
—(1994). 'Semantic Universals in Classifier Systems', Word 45: 145-71.
—(1995). 'Autonomy and Functionalist Linguistics', Language 71: 490-532.
—(1996). '"Markedness" and "Universals": From the Prague School to
Typology', in K. R. Jankowsky (ed.), Multiple Perspectives on the Historical
Dimensions of Language. Munster: Nodus, 15-21.
Crofts, M. (1973). Gramatica Munduruku. Brasilia: Summer Institute of Linguistics.
—(1985). Aspectos da lingua Munduruku. Brasilia: Summer Institute of Linguistics.
Crowley, T. (1978). The Middle Clarence Dialects of Banjalang. Canberra: Australian
Institute of Aboriginal Studies.
—(1979). 'Yaigin', in R. M. W. Dixon and B. J. Blake (eds.), Handbook of
Australian Languages, i. Canberra: Australian National University Press, 363-84.
Curnow, T. (1997). 'A Grammar of Awa Pit'. PhD thesis, Australian National
University.
Daley, K. A. C. (1996). The Use of Classifiers in Vietnamese Narrative Texts'. MA
thesis, University of Texas at Arlington.
Davidson, W, Elford, L. W., and Hoijer, H. (1963). Athabascan Classificatory
Verbs', in H. Hoijer (ed.), Studies in the Athabaskan Languages. Berkeley:
University of California Press.
Davies, J. (1981). Kobon. Amsterdam: North-Holland.
Davis,.D. R. (n.d.). 'Noun Class Markers in Wantoat'. Ukarumpa: Summer Institute
of Linguistics.
De Leon, M. de L. P. (1987). 'Noun and Numeral Classifiers in Mixtec and Tzotzil:
A Referential View'. PhD thesis, University of Sussex.
DeBray, R. G. A. (1951). Guide to the Slavonic Languages. London: J. M. Dent.
 References 461
DeLancey, S. (1986). 'Towards a history of Thai classifier system', in Craig (1986a:
437-52).
—(1998). 'Semantic Categorization in Tibetan Honorific Nouns', Anthropologi-
cal Linguistics 40: 109-23.
Demuth, K. A. (1988). 'Noun Classes and Agreement in Sesotho Acquisition', in
Barlow and Ferguson (1988b: 305-22).
—Faraclas, N., and Marchese, L. (1986). 'Niger-Congo Noun Class and Agree-
ment Systems in Language Acquisition and Historical Change', in Craig (1986a:
453-71).
Denny, J. P. (1976). 'What Are Noun Classifiers Good For?' Papers from the Annual
Regional Meeting of the Chicago Linguistic Society 12: 122-32.
—(1979a). 'The "Extendedness" Variable in Classifier Semantics: Universal
Semantic Features and Cultural Variation', in Mathiot (1979: 97-119).
—(1979b). 'Semantic Analysis of Selected Japanese Numeral Classifiers for
Units', Linguistics 17: 317-35.
—(1986). 'The Semantic Role of Classifiers', in Craig (1986a: 297-308).
and Creider, C. A. (1986). 'The semantics of noun classes in Proto-Bantu', in
Craig (1986a: 217–40).
Derbyshire, D. C. (1985). Hixkaryana and Linguistic Typology. Dallas: Summer
Institute of Linguistics and the University of Texas at Arlington.
—(1999). 'Carib Languages', in Dixon and Aikhenvald (1999: 23-64).
—and Payne, D. L. (1990). 'Noun Classification Systems of Amazonian
Languages', in Payne (1990: 243-72).
—and Pullum, G. K. (eds.) (1986). Handbook of Amazonian Languages, i. Berlin:
Mouton de Gruyter.
(eds.) (1990). Handbook of Amazonian Languages, ii. Berlin: Mouton de
Gruyter.
(eds.) (1991). Handbook of Amazonian Languages, iii. Berlin: Mouton de
Gruyter.
(eds.) (1998). Handbook of Amazonian Languages, iv. Berlin: Mouton de
Gruyter.
Diakonoff, I. M. (1988). Afrasian Languages. Moscow: Nauka.
Dimmendaal, G. I. (1983). The Turkana Language. Dordrecht: Foris.
—(forthcoming). 'Noun Classification in Baale', in A. Mietzner and R. Vossen
(eds.), Festschrift for F. Rottland. Cologne: Ruhiger Koppe.
Dixon, R. M. W. (1968). 'Noun Classes', Lingua 21: 104–25.
—(1972). The Dyirbal Language of North Queensland. Cambridge: Cambridge
University Press.
—(ed.) (1976). Grammatical Categories in Australian Languages. Canberra:
Australian Institute of Aboriginal Studies.
—(1977). A Grammar of Yidin. Cambridge: Cambridge University Press.
—(1980). The Languages of Australia. Cambridge University Press: Cambridge.
(1982). Where Have All the Adjectives Gone? and other essays in semantics and
syntax. Berlin: Mouton.
— (1986). 'Noun Classes and Noun Classification in Typological Perspective', in
Craig (1986a: 105-12).
462 Classifiers
Dixon, R. M. W. (1988). A Grammar of Boumaa Fijian. Chicago: University of
Chicago Press.
—(1994). Ergativity. Cambridge: Cambridge University Press.
—(1995). 'Fusional Development of Gender Marking in Jarawara Possessed
Nouns', International Journal of American Linguistics 61: 263-94.
—(1997). The Rise and Fall of Languages. Cambridge: Cambridge University
Press.
—(forthcoming). Australian Languages: Their Origin and Development.
Cambridge: Cambridge University Press.
—and Aikhenvald, A. Y. (eds.) (1999). The Amazonian Languages. Cambridge:
Cambridge University Press.
and Vogel, A. (forthcoming). The Jarawara Language of Southern Amazonia.
Donaldson, T. (1980). Ngiyambaa: The Language of the Wangaaybuwan. Cam-
bridge: Cambridge University Press.
Dondorp, A., and Shim, J.-W. (1997). Wipi. Ukarumpa: Summer Institute of
Linguistics.
Doneux, J. (1967). 'Discussion for S. Sauvageot 1967', in La Classification nominale,
234–5.
Dorian, N. (1978). 'The Fate of Morphological Complexity in Language Death',
Language 54: 590–609.
—(1981). Language Death: The Life Cycle of a Scottish Gaelic Dialect.
Philadelphia: University of Pennsylvania Press.
(ed.) (1989). Investigating Obsolescence: Studies in Language Contraction and
Death. Cambridge: Cambridge University Press.
Downing, P. (1984). 'Japanese Numeral Classifiers: A Semantic, Syntactic, and
Functional Profile'. PhD thesis, University of California (Berkeley).
—(1986). "The Anaphoric Use of Classifiers in Japanese', in Craig (1986a: 345-75).
—(1996). Numeral Classifier Systems: The Case of Japanese. Amsterdam: John
Benjamins.
Drabbe, P. (1955). Spraakkunst van het Marind zuidkust Nederlands Nieuw-Guinea.
Vienna: Missiehuis St. Gabriel.
Drossard, W. (1982). 'Nominalklassifikation in ostkaukasischen Sprachen', in Seiler
and Stachowiak (1982: 155–78).
Dryer, M. S. (1992). 'The Greenbergian Word Order Correlations', Language 68:
81-138.
Ducos, G. (1979). 'Evolution d'une langue a classes nominales', La Linguistique 15:
43-54.
Dul'son, A. P. (1968). Ketskij jazyk [The Ket Language]. Tomsk: Izdateljstvo
Tomskogo Universiteta.
Durie, M. (1985). A Grammar of Acehnese: On the Basis of a Dialect of North Aceh.
Dordrecht: Foris.
—(1986). 'The Grammaticization of Number as a Verbal Category', Berkeley
Linguistics Society: Proceedings of the Annual Meeting 12: 355-70.
Eades, D. (1979). 'Gumbaynggir', in R. M. W. Dixon and B. Blake (eds.), Handbook
of Australian Languages, i. Canberra: Australian National University Press and
Amsterdam: John Benjamins, 245-361.
 References 463
Eather, A. (1990). 'A Grammar of Nakkara (Central Arnhem Land Coast)'. PhD
thesis, Australian National University, Canberra.
Edel'man, J. (1980). 'K substratnomu naslediju centraljnoasiatskogo jazykovogo
sojuza' ['Towards the Substatum Inheritance of the Central Asian Linguistic
Area'], Voprosy Jazykoznania 5: 21-32.
Ekdahl, M., and Butler, N. (1979). Aprenda Terena. Brasilia: Summer Institute of
Linguistics.
Elbert, S. H. (1974). Puluwat Grammar, Canberra: Pacific Linguistics.
Emeneau, M. B. (1964). 'India as a Linguistic Area', in Hymes (1966: 642-53).
England, N. (1983). A Grammar of Mam, a Mayan Language. Austin: University of
Texas Press.
—(1991). 'Changes in Basic Word Order in Mayan Languages', International
Journal of American Linguistics 57: 446-86.
Erbaugh, M. (1984). 'Scissors, Paper, Stone: Perceptual Foundations for Noun
Classifier Systems', Papers and Reports on Child Language Development 23: 41-9.
—(1986). 'Taking Stock: The Development of Chinese Noun Classifiers
Historically and in Young Children', in Craig (1986a: 399–436).
Evans, N. (1994). 'The Problem of Body Parts and Noun Class Membership in
Australian Languages', University of Melbourne Working Papers in Linguistics 14:
1-8.
—(1996). 'The Syntax and Semantics of Body Part Incorporation in Mayali', in
Chappell and McGregor (1996: 65-109).
— (1997). 'Head Classes and Agreement Classes in the Mayali Dialect Chain', in
Harvey and Reid (1997: 105–6).
(forthcoming). A Draft Grammar of Mayali.
Everett, D. L., and Kern, B. (1997). A Grammar of 'Oro Nao. London: Croom
Helm.
Facundes, S. (1994). 'Noun Categorization in Apurina'. MA thesis, University of
Oregon.
Faraclas, N. G. (1989). 'Cross River', in J. Bendor-Samuel (ed.), The Niger-Congo
Languages. New York and Lanham: University Press of America, 377-99.
Fleischmann, L., and Turpeinen, S. (1975). Bine Grammar Essentials. Ukarumpa:
Summer Institute of Linguistics.
Foley, W. A. (1986). The Papuan Languages of New Guinea. Cambridge: Cambridge
University Press.
— (1991). The Yimas Language of New Guinea. Stanford: Stanford University
Press.
Ford, K., and Ober, D. (1991). 'A Sketch of Kalaw Kawaw Ya', in S. Romaine (ed.),
Language in Australia. Cambridge: Cambridge University Press, 118–42.
Ford, L. J. (1990). 'The Phonology and Morphology of Bachamal Wogait'. MA
thesis, Australian National University.
—(1998). 'A Grammar of Emmi'. PhD thesis, Australian National University.
Foreman, V. (1974). Grammar of Yessan-Mayo. Summer Institute of Linguistics,
Santa Ana, California.
Foris, D. (forthcoming). A Grammar of Sochiapan Chinantec.
Fortune, E. F. (1942). Arapesh. New York: J. J. Augustin.
464 Classifiers
Frajzyngier, Z. (1993). A Grammar of Mupun. Berlin: Dietrich Reimer.
Frank, P. (1990). Ika Syntax. Dallas: Summer Institute of Linguistics and the
University of Texas at Arlington.
Franklin, K. (1981). 'Existential and Pro-verbs in Kewa', in K. J. Franklin (ed.),
Syntax and Semantics in Papua New Guinea Languages. Ukarumpa: Summer
Institute of Linguistics, 151–72.
Fraser, N. M., and Corbett, G. G. (1997). 'Defaults in Arapesh', Lingua 103: 25-57.
Frawley, W. (1992). Linguistic Semantics. London: Lawrence Erlbaum.
Friedrich, P. (1970). 'Shape in Grammar', Language 46: 379–407.
—(1971). The Tarascan Suffixes of Locative Space: Meaning and Morphotactics.
Bloomington: Indiana University.
Gagne, R. C. (1966). Eskimo Language Course. Ottawa: Department of Indian
Affairs and Northern Development.
Gandour, J., Buckingham, H., and Dardarananda, R. (1985). 'The Dissolution of
Numeral Classifiers in Thai', Linguistics 23: 547–66.
—Petty, S. H., Dardarananda, R., Dechongkit, S., and Mukngoen, S. (1984).
'The Acquisition of Numeral Classifiers in Thai', Linguistics 22: 455-79.
Garibian, A. S. (1976). 'Armjanskij Jazyk' [Armenian Language], in Jazyki Azii i
Afriki: Indo-evropejskie jazyki. Moscow: Nauka, 94–109.
Geeraerts, D. (1988). 'Where Does Prototypicality Come From?', in B. Rudzka-
Ostyn (ed.), Topics in Cognitive Linguistics. Amsterdam: John Benjamins, 207-29.
Gerzenstein, A. (1994). Lengua maka: Estudio descriptivo. Instituto de Linguistica,
Universidad de Buenos Aires.
Givon, T. (1969). Studies in Chibemba and Bantu Grammar. Department of Lin-
guistics, University of California, Los Angeles.
—(1970). 'The Resolution of Gender Conflicts in Bantu Conjunction: When
Syntax and Semantics Clash', Papers from the Sixth Regional Meeting, Chicago
Linguistic Society, April 16-18, 1970. Chicago: Chicago Linguistic Society, 250-
61.
—(1971). 'Some Historical Changes in the Noun Class System of Bantu, Their
Possible Causes and Wider Implications', in C.-W. Kim and H. Stahlke (eds.),
Papers in African Linguistics. Edmonton: Linguistic Research, 33-54.
—(1976). 'Topic, Pronoun and Grammatical Agreement', in Li (1976: 149-89).
—(1981). 'On the Development of Numeral "one" as an Indefinite Marker',
Folia Linguistica Historica 2: 35-53.
(1991). 'Some Substantive Issues Concerning Verb Serialization: Grammatical
vs Cognitive Packaging', in C. Lefebre (ed.), Serial Verbs: Grammatical, Compara-
tive and Cognitive Approaches. Amsterdam: John Benjamins, 137–84.
Goddard, C. (1985). A Grammar of Yankunytjatjara. Alice Springs: Institute for
Aboriginal Development.
—(1996). 'Cross-linguistic Research on Metaphor', Language and Communication
16: 145-51.
Gomez-Imbert, E. (1986). 'De la forme et du sens dans la classification nominale
en Tatuyo, Langue Tukano orientale d'Amazonie colombienne', Universite
Paris IV.
— (1996). 'When Animals Become "Rounded" and "Feminine": Conceptual
 References 465
Categories and Linguistic Classification in a Multilingual Setting', in J. J.
Gumpertz and S. C. Levinson (eds.), Rethinking Linguistic Relativity. Cambridge:
Cambridge University Press, 438–69.
Goncalves, C. H. R. C. (1987). Concordancia em Munduruku. Campinas: Editora da
Unicamp.
Gonzalez Nanez, O. (1985). 'Los numerales en un dialecto curripaco', Boletin de
Linguistica 5: 15–28.
Goral, D. R. (1978). 'Numeral Classifier Systems: A Southeast Asian Cross Linguistic
Analysis', Linguistics of the Tibeto-Burman Area 4: 1-72.
Gordon, L. (1986). Maricopa Morphology and Syntax. Berkeley: University of
California Press.
Gralow, F. L. (1993). Un bosquejo del idioma Koreguaje. Bogota: Asociacion Instituto
Linguistico de Verano, Bogota.
Green, D. (1996). 'O sistema numerico na lingua Palikur', Boletim do museu Goeldi
10: 261-303.
—and Green, H. (1972). Surface Grammar of Palikur. Brasilia: Summer Institute
of Linguistics.
Green, I. (1989). 'Marrithiyel'. PhD thesis, Australian National University.
—(1995). 'The Death of "Prefixing": Contact Induced Typological Change in
Northern Australia', Proceedings of the Annual Meeting, Berkeley Linguistics
Society 21: 414–25.
(1997). 'Nominal Classification in Marrithiyel', in Harvey and Reid (1997:
229-53).
Green, M., and Igwe, G. E. (1963). A Descriptive Grammar of Igbo. Berlin: Akademie.
Green, R. (1987). 'A Sketch Grammar of Burarra'. Fourth year honours thesis,
Australian National University, Canberra.
—(1995). 'A Grammar of Gurr-goni'. PhD thesis, Australian National University.
Greenberg, J. H. (1963). 'Some Universals of Grammar with Particular Reference
to the Order of Meaningful Elements', in J. H. Greenberg (ed.), Universals of
Language. Cambridge, Mass.: MIT Press, 58-90.
—(1966). Language Universals, with Special Reference to Feature Hierarchies.
The Hague: Mouton.
—(1972). 'Numeral Classifiers and Substantival Number: Problems in the Gen-
esis Type', Working Papers in Language Universals, reprinted in Greenberg (1990:
16-93).
— (1978). 'How Does Language Acquire Gender Markers?' in J. H. Greenberg,
C. A. Ferguson, and E. A. Moravcsik (eds.), Universals of Human Languages, iii:
Word Structure. Stanford: Stanford University Press, 241-70.
(1990). On Language: Selected Writings of Joseph H. Greenberg, K. Denning
and S. Kemmer (eds.). Stanford, Stanford University Press.
Griffiths, G., and Griffiths, C. (1976). Aspectos da lingua Kadiweu. Brasilia: Summer
Institute of Linguistics.
Haas, M. R. (1942). The Use of Numeral Classifiers in Thai', Language 18: 201-5.
(1978). 'Classificatory verbs in Muskogee', International Journal of American
Linguistics 14: 244–6; repr. in A. S. Dil (ed.) Language, Culture, and History:
Essays by Mary R. Haas. Stanford University Press: Stanford, 302-7.
466 Classifiers
Haas, M. R. (1967). 'Language and Taxonomy in Northwestern California',
American Anthropologist 69: 358-62.
Hagman, R. S. (1977). Nama Hottentot Grammar. Bloomington: Indiana University.
Haiman, J. (1980). Hua: A Papuan Language of the Eastern Highlands of New
Guinea. Amsterdam: John Benjamins.
Hale, K. L. (1975). 'Gaps in Grammar and Culture', in M. D. Kinkade et al. (eds.),
Linguistics and Anthropology: In Honor of C. F. Voegelin. Lisse: Peter de Ridder,
295-315.
Hamel, P. J. (1994). A Grammar and Lexicon of Loniu, Papua New Guinea. Depart-
ment of Linguistics, Research School of Pacific and Asian Studies, Australian
National University, Canberra.
Hamp, E. (1958). 'Gender Shift in Albanian Plurals', Romance Philology 12: 147-
55.
Harms, P. L. (1994). Epena Pedee Syntax. Dallas: Summer Institute of Linguistics
and the University of Texas at Arlington.
Harris, A., and Campbell, L. (1995). Historical Syntax in Cross-linguistic Perspective.
Cambridge: Cambridge University Press.
Harrison, S. P. (1976). Mokilese Reference Grammar. Honolulu: Hawaii University
Press.
(1988). 'A Plausible History for Micronesian Possessive Classifiers', Oceanic
Linguistics 27: 63-78.
Harvey, M. (1987). 'The Warray Language from Adelaide River'. MA thesis,
Australian National University.
(1992). The Gaagudju People and Their Language'. PhD thesis, University of
Sydney.
(1997). 'Nominal Classification and Gender in Aboriginal Australia', in
Harvey and Reid (1997: 17-62).
— (forthcoming). 'Limilngan'.
and Reid, N. (eds.) (1997). Nominal Classification in Aboriginal Australia.
Amsterdam: John Benjamins.
Hasada, R. (1995). 'Number System in Japanese', paper presented at the Workshop
on Grammatical Categories, Australian National University, Canberra.
Hashimoto, M. J. (1977). 'The Genealogy of the Classifier in Sino-Tibetan',
Computational Analysis of Asian and African Languages 1: 69-78.
Haspelmath, M. (1993). A Grammar of Lezgian. Berlin: Mouton de Gruyter.
(1997). Indefinite Pronouns. Oxford: Oxford University Press.
Haviland, J. (1979). 'Guugu Yimidhirr', in R. M. W. Dixon and B. Blake (eds.),
Handbook of Australian Languages, i. Canberra: Australian National University
Press, 27-180.
Hayward, R. J. (1989). 'The Notion of "Default Gender": A Key to Interpreting the
Evolution of Certain Verb Paradigms in East Ometo, and Its Implications for
Omotic', Afrika und Ubersee 12: 17-32.
and Corbett, G. G. (1988). 'Resolution Rules in Qafar', Linguistics 26: 259-79.
Heath, J. (1978a). Linguistic Diffusion in Arnhem Land. Canberra: Australian
Institute of Aboriginal Studies.
 References 467
— (1978b). Ngandi Grammar, Texts and Dictionary. Canberra: Australian Institute
of Aboriginal Studies.
— (1983). 'Referential tracking in Nunggubuyu', in P. Munro and J. Haiman
(eds.), Switch-Reference and Universal Grammar, Amsterdam/Philadelphia: John
Benjamins, 129–49.
(1984). Functional Grammar of Nunggubuyu. Canberra: Australian Institute of
Aboriginal Studies.
Heine, B. (1982a). 'African Noun Class Systems', in Seiler and Lehmann (1982:
189–216).
(1982b). The Nubi Language of Kibera: An Arabic Creole. Berlin: Dietrich
Reimer.
(1992). 'Grammaticalization Chains', Studies in Language 16: 335-68.
(1997a). Possession: Cognitive Sources, Forces, and Grammaticalization.
Cambridge: Cambridge University Press.
(1997b). 'Grammaticalization Chains Across Languages: An Example from
Khoisan', paper presented at the Symposium on Grammaticalization, Houston,
March 1997.
and Claudi, U. (1986). On the Rise of Grammatical Categories: Some Examples
from Maa. Berlin: Reimer.
and Hunnemeyer, F. (1991). Grammaticalization: A Conceptual Frame-
work. Chicago: University of Chicago Press.
— and Reh, M. (1984). Grammaticalization and Reanalysis in African Languages.
Hamburg: Helmut Buske.
and Vossen, R. (1983). 'On the Origin of Gender in Eastern Nilotic', in R.
Vossen and M. Bechchaus-Gerst (eds.), Nilotic Studies: Proceedings of the
International Symposium on Languages and History of the Nilotic Peoples. Berlin:
Reimer, 255-68.
Henry, D., and Henry, K. (1965). 'Koyukon Classificatory Verbs', Anthropological
Linguistics 7: 110–16.
Herbert, R. K. (1991). 'Patterns in Language Change, Acquisition and Dissolution:
Noun Prefixes and Concords in Bantu', Anthropological Linguistics 33: 103-34.
Hewitt, G. (1979). Abkhaz. Amsterdam: North-Holland.
Hill, D. (1992). 'Longgu Grammar'. PhD thesis, Australian National University.
Himmelmann, N. P. (1997). Deiktikon, Artikel, Nominalphrase: zur Emergenz
syntaktischer Struktur. Tubingen: Niemeyer.
Hiranburana, S. (1979). 'A Classification of Thai Classifiers', in N. D. Liem (ed.),
Southeast Asian Linguistic Studies, iv. Canberra: Pacific Linguistics, 39-54.
Ho, J. (1997). 'Socio-semantic Aspects of Human Measure Words in Cantonese', in
M. Clark (ed.), Papers in Southeast Asian Linguistics 16. Canberra: Pacific
Linguistics, 67-77.
Hock, H. H. (1991). Principles of Historical Linguistics. Berlin: Mouton de Gruyter.
Hoijer, H. (1945). 'Classificatory Verb Stems in the Apachean Languages',
International Journal of American Linguistics 11: 13-23.
Holmquist, J. C. (1991). 'Semantic Features and Gender Dynamics in Cantabrian
Spanish', Anthropological Linguistics 33: 57-81.
Hopkins, N. A. (1970). 'Numeral Classifiers in Tzeltal, Jacaltec and Chuj (Mayan)',
468 Classifiers
Papers from the Annual Regional Meeting of the Chicago Linguistic Society 6: 23-
35.
Hopper, P. J. (1986). 'Some Discourse Functions of Classifiers in Malay', in Craig
(1986a: 309-25).
and Thompson, S. (1984). 'The Discourse Basis for Lexical Categories in
Universal Grammar', Language 60: 703-52.
Hopper, P. I, and Traugott, E. C. (1993). Grammaticalization. Cambridge:
Cambridge University Press.
Hu, Q. (1993). The Acquisition of Chinese Classifiers by Young Mandarin Speaking
Children'. PhD thesis, Boston University.
Huffman, F. E. (1970). Modern Spoken Cambodian. New Haven: Yale University
Press.
Hundius, H., and Kolver, U. (1983). 'Syntax and Semantics of Numeral Classifiers
in Thai', Studies in Language 7: 165-214.
Hurd, C. (1977). 'Nasioi Projectives', Oceanic Linguistics 16: 111-78.
Hyde, V. (1971). An Introduction to the Luizeno Language. Banning, Calif.: Malki
Museum Press.
Hyman, L. M. (ed.) (1980). Noun Classes in the Grassfields Bantu Borderland. Los
Angeles: University of Southern California, Department of Linguistics.
Hymes, D. (ed.) (1964). Language in Culture and Society: A Reader in Linguistics
and Anthropology. New York: Harper & Row.
Ikoro, S. M. (1994). 'Numeral Classifiers in Kana', Journal of African Language and
Linguistics 15: 7-28.
(1996a). The Kana Language. Leiden: University of Leiden.
(1996b). 'A Grammatical Summary of Kana', Materials for the Research
Project 'Universals of Human Languages', Australian National University.
(1997). 'A Grammatical Summary of Igbo', Materials for the Research Project
'Universals of Human Languages', Australian National University.
Irvine, J. (1978). 'Wolof Noun Classification: The Social Setting of Divergent
Change', Language in Society 7: 37-64.
Iturrioz Leza, J. L., Gomez Lopez, P., Leal Carretero, S., and Ramirez de la Cruz, R.
(1986). 'Individuacion en huichol. II: Aspectos morfologicos y sintacticos de las
clases nominales. III: Las series sufijales', Funcion 1: 422–62; 2: 154–63.
Jackson, E. (1966). Waura: formulario dos vocabularios padroes do Museu Nacional.
Brasilia: Summer Institute of Linguistics.
and Richards, J. (1966). Tentative Phonemics Statement of Waura. Brasilia:
Summer Institute of Linguistics.
Jaisser, A. (1987). 'Hmong Classifiers', Linguistics of the Tibeto-Burman Area 10:
169-75.
Jakobson, R. O. (1941). Kindersprache, Aphasie und allgemeine Lautgesetze.
Uppsala: Almquist & Wiksell.
(1984). 'The Gender Pattern of Russian', in L. R. Waugh and M. Halle (eds.),
Russian and Slavic Grammar: Studies 1931–1981. Berlin: Mouton, 141-3.
Jauncey, D. (1997). 'A Grammar of Tamambo'. PhD thesis, Australian National
University.
 References 469
Jensen, J. T. (1977). Yapese Reference Grammar. Honolulu: University Press of
Hawaii.
Jones, R. B. Jr. (1970). 'Classifier Constructions in Southeast Asia', Journal of
American Oriental Society 90: 1-12.
Jones, W., and Jones, P. (1991). Barasano Syntax. Dallas: Summer Institute of
Linguistics and the University of Texas at Arlington.
Jun, W., and Guoqiao, Z. (1993). An Outline Grammar of Mulao. National Thai
Studies Centre, Australian National University, Canberra.
Juntanamalaga, P. (1988). 'Social Issues in Thai Classifier Usage', Language
Sciences 10: 313–30.
Kakumasu, J. (1986). 'Urubu-Kaapor', in Derbyshire and Pullum (1986: 326–406).
Kari, J. (1989). 'Affix Positions and Zones in the Athabaskan Verb Complex: Ahtna
and Navajo', International Journal of American Linguistics 55: 424—54.
(1990). Ahtna Athabaskan Dictionary. Alaska Native Language Center,
University of Alaska, Fairbanks.
Karmiloff-Smith, A. (1979). A Functional Approach to Child Language: A Study of
Determiners and Reference. Cambridge: Cambridge University Press.
Kayser, A. (ed. K. H. Rensch) (1993). Nauru Grammar. Yarralumla, Canberra:
Embassy of the Federal Republic of Germany.
Keating, E. (1997). 'Honorific Possession: Power and Language in Pohnpei,
Micronesia', Language and Society 26: 247-68.
Keenan, E. L. (1978). 'On Surface Form and Logical Form', in B. B. Kachru (ed.),
Linguistics in the Seventies: Directions and Prospects. Urbana: University of
Illinois, 163-203.
(1984). 'Semantic Correlates of the Ergative/Absolutive Distinction', Linguistics
22: 197–223.
Kerr, I. (1995). Gramatica pedagogica del cuiba-waimonae. Bogota: Asociacion
Instituto Linguistico de Verano.
Key, M. R. (1972). 'Linguistic Behavior of Male and Female', Linguistics 88: 15-31.
(1979). The Grouping of South American Indian Languages. Tubingen: Gunter
Narr.
Khaidakov, S. M. (1963). 'The Principles of Non-class Division in Lak' (in Russian,
with a summary in English), Studia Caucasica 1: 48-55.
(1980). Principy imennoj klassifikacii v dagestanskih jazykah [The Principles of
Nominal Classification in Daghestanian Languages]. Moscow: Nauka.
Kibrik, A. E., Kodzasov, S. V., and Olovjannikova, I. P. (1972). Fragmenty gram-
matiki hinalugskogo jazyka [Fragments of a Grammar of Khinalug]. Moscow:
Izdateljstvo Moskovskogo Universiteta.
Kimball, G. D. (1991). Koasati Grammari. Lincoln: University of Nebraska Press.
Kirton, J. F. (1971). Papers in Australian Linguistics. Australian National Univer-
sity, Canberra: Pacific Linguistics.
Kiyomi, S. (1992). 'Animateness and Shape in Classifiers', Word 43: 15-36.
Klein, H. E. M. (1978). Una gramatica de la lengua Toba: morfologia verbal y
nominal. Montevideo: Universidad de la Republica.
(1979). 'Noun Classifiers in Toba', in Mathiot (1979: 85-95).
Koehn, E., and Koehn, S. (1986). 'Apalai', in Derbyshire and Pullum (1986: 33-127).
470 Classifiers
Koehn, S. (1994). The Use of Generic Terms in Apalai Genitive Constructions',
Revista Latinoamericana de estudios etnolinguisticos 8: 39–48.
Kohler, O. (1962). 'Studien zur Genussystem und Verbalbau der zentral Khoisan-
Sprachen', Anthropos 57: 529–46.
(1971). 'Noun Classes and Grammatical Agreement in !Xu (3u-/hoa dialect)',
Actes du Huitieme Congres International de Linguistique Africaine, Abidjan 24-28
Mars 1969. Abidjan, 489–522.
Kolver, U. (1982a). 'Klassifikatorkonstruktionen in Thai, Vietnamesisch und
Chinesisch', in Seiler and Lehmann (1982: 160-85).
(1982b). 'Interaktion von nominalen Kategorien am Beispiel der Entwicklung
des modernen Bengali', in Seiler and Lehmann (1982: 244–51).
Kooyers, O. (n.d.). Washkuk Grammar Sketch. Ukarumpa: Summer Institute of
Linguistics.
Kopcke, K. M., and Zubin, D. A. (1984). 'Sechs Prinzipien fur die Genuszuweisung
im Deutschen: Ein Beitrag zur naturlichen Klassifikation', Linguistische Berichte
93: 26–50.
Krauss, M. (1968). 'Noun Classification Systems in Athabaskan, Eyak, Tlingit and
Haida Verbs', International Journal of American Linguistics 34: 194–203.
Krejnovic, E. A. (1961). 'Imennye klassy i sredstva ih vyrazhenija v ketskom jazyke'
['Noun Classes and Means of Their Expression in Ket']', Voprosy jazykoznanija
2: 106-16.
Krishnamurti, Bh. (1975). 'Gender and Number in Proto-Dravidian', International
Journal of Dravidian Linguistics 4: 328-50.
(forthcoming). 'Gender and Number in Dravidian'.
and Gwynn, J. P .L. (1985). A Grammar of Modern Telugu. Delhi: Oxford
University Press.
Kuipers, A. H. (1967). The Squamish Language. The Hague: Mouton.
(1974). The Shuswap Language. The Hague: Mouton.
Kunene, E. (1986). 'Acquisition of SiSwati noun classes', South African Journal of
African Languages 6: 34–7.
Lakoff, G. (1986). 'Classifiers as a Reflection of Mind', in Craig (1986a: 13-52).
(1987). Women, Fire and Other Dangerous Things: What Categories Reveal
about the Mind. Chicago: University of Chicago Press.
Landar, H. (1967). Ten'a Classificatory Verbs', International Journal of American
Linguistics 33: 263-8.
Lang, A. (1975). The Semantics of Classificatory Verbs in Enga and Other Papua
New Guinea Languages. Canberra: Pacific Linguistics.
Langacker, R. W. (1977). Studies in Uto-Aztecan Grammar, i: An Overview of Uto-
Aztecan Grammar. Dallas: Summer Institute of Linguistics and the University of
Texas at Arlington.
(1987). Foundations of Cognitive Grammar, i: Theoretical Prerequisites.
Stanford: Stanford University Press.
Langdon, M. (1970). A Grammar of Diegueno. Berkeley: University of California
Press.
Laoust, E. (1918). Ntifa: son parler. Paris: Hachette.
(1928). Cows de berbere marocain. Paris: Hachette.
 References 471
— (1931). Siwa. v. 1. Son parler. Paris: Hachette.
Lapointe, S. (1985). A Theory of Grammatical Agreement, New York: Garland.
LaPolla, R. (1994). 'Parallel Grammaticalizations in Tibeto-Burman Languages:
Evidence of Sapir's "Drift"', Linguistics of the Tibeto-Burman Area 17: 61-80.
(with Chenglong Huang) (forthcoming). Grammatical Sketch of the Qiang
Language with Texts and Annotated Glossary.
Laskowski, R. (1988). The Systemic Prerequisites of the Development of the
Declensional Patterns of the Slavic Languages: The Category of Gender',
Scando-Slavica 34: 111–25.
Lawton, R. S. (1980). 'The Kiriwinan Classifiers'. MA thesis, Australian National
University.
(1993). Topics in the Description of Kiriwina. Canberra: Pacific Linguistics.
Laycock, D. C. (1969). Three LamaLamic Languages of North Queensland',
Papers in Australian Linguistics 4: 71-97.
(1975). The Torricelli Phylum', in S. Wurm (ed.), New Guinea Area Languages
and Language Study, i: Papuan Languages and the New Guinea Linguistic Scene.
Canberra: Pacific Linguistics, 767-80.
and Z'graggen, J. (1975). The Sepik-Ramu Phylum', in S. Wurm (ed.), New
Guinea Area Languages and Language Study, i: Papuan Languages and the New
Guinea Linguistic Scene, Canberra: Pacific Linguistics, 731-63.
Lee, J. (1987). Tiwi Today. A Study of Language Change in a Contact Situation.
Canberra: Pacific Linguistics.
Lee, K.-D. (1975). Kusaiean Reference Grammar. Honolulu: University Press of
Hawaii.
Lee, M. (1988). 'Language, Perception and the World', in J. Hawkins (ed.), Explain-
ing Language Universals. Oxford: Blackwell, 211–46.
Lee, Y. (1997). 'Classifiers in Korean'. Honours thesis, Australian National Uni-
versity, Canberra.
Lee Kwok Loong, J. (1997). 'Nominal Classification in Yapese'. Honours thesis,
University of Western Australia.
Leeding, V. (1989). 'Anindilyakwa Phonology and Morphology'. PhD thesis,
University of Sydney.
(1996). 'Body Parts and Possession in Anindilyakwa', in Chappell and
McGregor (1996: 193-250).
Lehmann, C. (1982). 'Universal and Typological Aspects of Agreement', in Seiler
and Stachowiak (1982: 201-67).
(1988). 'On the Function of Agreement', in Barlow and Ferguson (1986b:
55-66).
Lehrer, A. (1986). 'English Classifier Constructions', Lingua 68: 109–8.
Levy, P. (1993). Totonac Body-parts: Are They Classifiers?', paper presented at the
workshop 'Back to Basic Issues in Classification', Nijmegen, May 1993.
(1994). How Shape Becomes Grammar: On the Semantics of Part Morphemes in
Totonac. Cognitive Anthropology Research Group. Max-Planck Institute for
Psycholinguistics, Nijmegen.
Levy, Y. (1983a). The Acquisition of Hebrew Plurals: The Case of the Missing
Gender Category', Journal of Child Language 10: 107–21.
472 Classifiers
Levy, Y. (1983b). 'It's Frogs All the Way Down', Cognition 15: 75-93.
(1988). 'On the Early Learning of Formal Grammatical Systems: Evidence
from Studies of the Acquisition of Gender and Countability', Journal of Child
Language 15: 179-87.
Li, C. N. (ed.) (1976). Subject and Topic. New York: Academic Press.
and Thompson, S. A. (1981). Mandarin Chinese: A Functional Reference
Grammar. Berkeley: University of California Press.
Lichtenberk, F. (1983a). 'Relational Classifiers', Lingua 60: 147–76.
(1983b). A grammar of Manam. Honolulu: University of Hawaii Press.
(1985). 'Possessive Constructions in Oceanic Languages and in Proto-Oceanic',
in A. Pawley and L. Carrington (eds.), Austronesian Linguistics at the 15th Pacific
Science Congress. Canberra: Pacific Linguistics, 93-140.
(1995). 'Number Marking in To'aba'ita', paper presented at the workshop on
'Grammatical Categories', Australian National University, Canberra.
Lindrud, S., and Nicholson, R. (n.d.) 'Notes on the Taulil language of East New
Britain'. Ukarumpa: Summer Institute of Linguistics.
Lock, A. (forthcoming). Abau Grammar Essentials. London: Longmans.
Lobel. E. (forthcoming). 'Classifiers vs. Genders and Noun Classes: A Case Study
in Vietnamese', in B. Unterbeck (ed.), Gender in Grammar and Cognition. Berlin:
Mouton de Gruyter.
Lowe, I. (1999). 'Nambiquara family', in Dixon and Aikhenvald (1999: 269-92).
Lucy, J. A. (1992a). Grammatical Categories and Cognition. A Study of the Linguis-
tic Relativity Hypothesis. Cambridge: Cambridge University Press.
(1992b). Language Diversity and Thought: A Reformulation of the Linguistic
Relativity Hypothesis. Cambridge: Cambridge University Press.
Luke, K. K., and Harrison, G. (1986). 'Young Children's Use of Chinese (Cantonese
and Mandarin) Sortal Classifiers', in S. R. Henry Kao and R. Hoosain (eds.),
Linguistics, Psychology and the Chinese Language. Hong Kong: University of
Hong Kong, 125–47.
Lynch, J. (1992). '"For My Part . . .": The Grammar and Semantics of Part
Possession in the Languages of Tanna', Australian Journal of Linguistics 12:
249-70.
(1993). 'Proto-Oceanic Possessive Marking', paper presented at the 1st Inter-
national Conference on Oceanic Linguistics, 4–9 July 1993, Port Vila, Vanuatu.
Ross, M., and Crowley, T. (forthcoming). The Oceanic Languages. London:
Curzon Press.
Lyons, J. (1968). Introduction to Theoretical Linguistics. Cambridge: Cambridge
University Press.
(1975). 'Deixis and Anaphora', in T. Myers (ed.), The Development of Con-
versation and Discourse. Edinburgh: Edinburgh University Press, 88-103.
(1977). Semantics. 2 vols. Cambridge: Cambridge University Press.
Magomedbekova, Z. M. (1967). 'Chamalinskij jazyk' ['Chamalal language'], in E.
A. Bokarev and K. V. Lomtatidze (eds.) Jazyki narodov SSSR, IV: Iberijsko-
kavkazskie jazyki. Moscow: Nauka, 384-99.
Mahapatra, B. P. (1979). Malto: An Ethnosemantic Study. Manasagangotro,
Mysore: Central Institute of Indian Languages.
 References 473
Malinowski, B. (1920). 'Classificatory Particles in the Language of Kiriwina',
Bulletin of the School of Oriental Studies 1: 33–78.
Manessy, G. (1961). Le bwamu et ses dialectes. Dakar.
Marchese, L. (1988). 'Noun Classes and Agreement Systems in Kru: A Historical
Approach', in Barlow and Ferguson (1988b: 323–42).
Marnita, R. (1996). 'Classifiers in Minangkabau'. MA thesis, Australian National
University, Canberra.
Marr, D. (1982). Vision. San Francisco: Freeman.
Martin, S. E. (1975). A Reference Grammar of Japanese. New Haven: Yale
University Press.
Martins, S. A. (1994). 'Analise da morfosintaxe da lingua Daw (Maku-Kama) e sua
classificacao tipologica'. MA thesis, Universidade Federal de Santa Catarina,
Florianopolis, Brazil.
and Martins, V. (1999). The Maku language family', in Dixon and Aikhenvald
(1999: 251–67).
Mathiot, M. (ed.) (1979). Ethnology: Boas, Sapir and Whorf Revisited. The Hague:
Mouton.
and Roberts, M. (1979). 'Sex Roles as Revealed through Referential Gender in
American English', in Mathiot (1979: 1–47).
Matthews, P. H. (1997). The Concise Oxford Dictionary of Linguistics. Oxford:
Oxford University Press.
Matisoff, J. A. (1991). 'Areal and Universal Dimensions of Grammatization in
Lahu', in Traugott and Heine (1991a: ii. 383–453).
Matsumoto, Y. (1985). Acquisition of Some Japanese Numeral Classifiers: The
Search for Convention', Stanford Papers and Reports in Child Language Devel-
opment 24: 79-86.
(1993). 'Japanese Numeral Classifiers: A Study on Semantic Categories and
Lexical Organization', Linguistics 31: 667-713.
Mattei-Muller, M.-C. (1974). 'El sistema de posesion en la lengua panare', Antro-
pologica 38: 3-14.
McGregor, D. E., and McGregor, A. R. F. (1982). Olo Language Materials.
Canberra: Pacific Linguistics.
McKay, G. R. (1975). 'Rembarrnga, a Language of Central Arnhem Land'. PhD
thesis, Australian National University, Canberra.
McLendon, S. (1975). A Grammar of Eastern Pomo. Berkerley: University of
California Publications in Linguistics 74.
McLaughlin, F. (1997). 'Noun Classification in Wolof: When Affixes Are Not
Renewed', Studies in African Linguistics 26: 1-28.
Meillet, A. (1912). 'L'Evolution des formes grammaticales', Scientia 12/26 (Milan);
repr. 1951, in Linguistique historique et linguistique generale. Paris: Klincksieck,
130–8.
Meillet, A. (1964). 'The Feminine Gender in the Indo-European Languages', in
Hymes (1964: 124).
Melchert, H. C. (1994). 'The Feminine Gender in Anatolian', Fruh-, Mittel-,
Spatindogermanisch: Akten der IX Fachtagung der indogermanischen Gesellschaft.
Wiesbaden: Ludwig Reichert, 231–44.
474 Classifiers
Merlan, F. (1983). Ngalakan Grammar, Texts and Vocabulary. Canberra: Pacific
Linguistics.
(1994). A Grammar of Wardaman: A Language of the Northern Territory of
Australia. Berlin: Mouton de Gruyter.
and Rumsey, A. (1991). Ku Waru: Language and Segmentary Politics in the
Western Nebilyer Valley. Papua New Guinea. Cambridge: Cambridge University
Press.
Roberts, S. P., and Rumsey, A. (1997). 'New Guinea "Classificatory Verbs"
and Australian Noun Classification: A Typological Comparison', in Harvey and
Reid (1997: 63-103).
Mills, A. E. (1986). The Acquisition of Gender: A Study of English and German.
Berlin: Springer.
Minor, E., and Minor, L. (1971). Resumen de la gramatica [Huitoto]. Lomalinda:
Instituto Linguistico de Verano.
Mithun, M. (1984). 'The Evolution of Noun Incorporation', Language 60: 847-94.
(1986). 'The Convergence of Noun Classification Systems', in Craig (1986a:
379-98).
(1991). 'Active/Agentive Case-marking and Its Motivations', Language 67:

Moore, B., and Franklin, G. (1979). Breves noticias da lingua Maku-Hupda. Brasi-
lia: Summer Institute of Linguistics.
Moore, D. L. (1984). 'Syntax of the Language of the Gaviao Indians of Rondonia'.
PhD thesis, City University of New York.
Moravcsik, E. A. (1971). 'Agreement', in J. H. Greenberg (ed.), Universals of Human
Language, iv: Syntax. Stanford: Stanford University Press, 331-74.
Morice, A. G. (1932). The Carrier Language (Dene family): A Grammar and
Dictionary Combined. 2 vols. Vienna: Collection Internationale de Monographies
Linguistiques. Anthropos, vol. 9.
Mosel, U. (1982). 'Number, Collection and Mass in Tolai', in Seiler and Stachowiak
(1982: 123–54).
(1983). Adnominal and Predicative Possessive Constructions in Melanesian
Languages. Arbeiten des Kolner Universalien-Projekts, No. 50. Cologne:
University of Cologne.
and Spriggs, R. (forthcoming). 'A Grammar of Teop'.
Moshinsky, J. (1974). A Grammar of Southeastern Pomo. Berkeley: University of
California Press.
Moussay, G. (1981). La Langue Minangkabau. Paris: Association Archipel.
Mufwene, S. S. (1980). 'Bantu Class Prefixes: Inflectional or Derivational?', Papers
from the Annual Regional Meeting of the Chicago Linguistic Society 16: 246–58.
Mujica, M. I. O. (1992). 'Aspectos fonologicos e gramaticais da lingua Yawalapiti
(Aruak)'. MA thesis, Universidade de Campinas.
Mulford, R. (1983). 'Semantic and Formal Factors in the Comprehension of
Icelandic Pronoun Gender', Papers and Reports on Child Language Development
22: 83-91.
Munroe, P. MS. 'The Garifuna Gender System'.
 References 475
Nekitel, O. (1985). 'Sociolinguistic Aspects of Abu', a Papuan Language of the
Sepik Area, Papua New Guinea'. PhD thesis, Australian National University.
(1986). 'A Sketch of Nominal Concord in Abu' (an Arapesh language)', in
D. C. Laycock et al. (eds.), Papers in New Guinea Linguistics 24. Canberra: Pacific
Linguistics, 177-205.
(forthcoming). 'Gender in Abu' Arapesh'.
Ng, B. C. (1989). 'The Acquisition of Numeral Classifiers in Hokkien, a Southern
Min Language'. PhD, La Trobe University.
(1991). 'Word Meaning Acquisition and Numeral Classifiers', La Trobe Work-
ing Papers in Linguistics 4: 73-83.
Nguyen, D. H. (1957). 'Classifiers in Vietnamese', Word 13: 124–52.
Nichols, J. (1986). 'Head-marking and Dependent-marking Grammar', Language
62: 56–119.
(1989a). 'The Nakh Evidence for the History of Gender in Nakh-Daghestanian',
in H. I. Aronson (ed.), The Non-Slavic Languages of the USSR: Linguistic
Studies. Chicago: Chicago Linguistic Society, 158-75.
(1989b). 'The Origin of Nominal Classification', in K. Hall, M. Meacham,
and R. Shapiro (eds.), Proceedings of the Fifteenth Annual Meeting of the Berkeley
Linguistics Society, February 18–20 1989: 409-20.
(1992). Linguistic Diversity in Space and Time. Chicago: University of Chicago
Press.
(1997). 'Modelling Ancient Population Structures and Movement in Linguis-
tics', Annual Review of Anthropology 26: 359-84.
and Peterson, D. A. (1998). 'A Reply to Campbell', Language 74: 605-14.
Nordlinger, R. (1998). A Grammar of Wambaya, Northern Territory (Australia).
Canberra: Pacific Linguistics.
6 Siadhail, M. (1989). Modern Irish: Grammatical Structure and Dialectal Varia-
tion. Cambridge: Cambridge University Press.
Onishi, M. (1994). 'A Grammar of Motuna (Bougainville, Papua New Guinea)'.
PhD thesis, Australian National University.
(1996a). 'A Grammatical Summary of Ainu', materials for the research project
'Universals of Human Languages', Australian National University.
(1996b). 'A Grammatical Summary of Japanese', materials for the research
project 'Universals of Human Languages', Australian National University.
(1996c). 'A Grammatical Summary of Bengali', materials for the research
project 'Universals of Human Languages', Australian National University.
(1997a). 'A Grammatical Summary of Kobon', materials for the research
project 'Universals of Human Languages', Australian National University.
(1997b). 'A Grammatical Summary of Amele', materials for the research
project 'Universals of Human Languages', Australian National University.
Osam, E. K. A. (1994). 'Aspects of Akan Grammar: A Functional Perspective'.
PhD thesis, University of Oregon, Eugene.
Osumi, M. (1996). 'Body Parts in Tinrin', in Chappell and McGregor (1996:
344–62).
Osborne, C. R. (1974). The Tiwi Language: Grammar, Myths and Dictionary of the
476 Classifiers
Tiwi Language Spoken on Melville and Bathurst Islands, Northern Australia.
Canberra: Australian Institute of Aboriginal Studies.
Ott, W., and Ott, R. (1983). Diccionario ignaciano y castellano con apuntes grama-
ticales. Cochabamba: Summer Institute of Linguistics.
Pacioni, P. (1997). 'Possessive Constructions, Classifiers and Plurality in Cantonese
and Some Other Chinese Dialects', in T. Hayasi and P. Bhaskararao (eds.), A
Report of the Joint Research Project: Analysis and Description of Individual
Languages and Linguistic Typology. Tokyo: Institute for the Study of Languages
and Cultures of Asia and Africa, 199-222.
(forthcoming). 'Classifiers, Specificity and Typology in Asian Languages'.
Palmer, G., and Arin, D. N. (1995a). Ancestral Spirit Scenarios in Bantu Noun
Classification: The Shape of a Heuristic System', paper presented to the 9th
Annual Meeting of the AAA, Washington DC, 15-19 November 1995.
(1995b). 'The Domain of Ancestral Spirits in Bantu Noun Classifica-
tion', paper presented at the 4th International Cognitive Linguistics Conference,
Albuquerque, New Mexico, 16–21 July 1995.
Panfilov, V. Z. (1968). Nivxskij Jazyk [The Nivkh Language]. Moscow: Nauka.
Parker, G. J. (1969). Ayacucho Quechua Grammar and Dictionary. The Hague:
Mouton.
Parker, J., and Parker, D. (1977). Baining Grammar Essentials. Ukarumpa: Summer
Institute of Linguistics.
Pasch, H. (1985). 'Possession and possessive Classifiers in 'Dongo-ko', Afrika und
Ubersee 68: 69-85.
(1986). Die Mba-Sprachen. Die Nominalklassensysteme und die genetische
Gliederung einer Gruppe von Ubangi Sprachen. Hamburg: Helmut Buske.
(1988). 'Die Entlehnung von Bantu-Prafixen in eine Nichtbantu-Sprache',
Zeitschrift fur Phonetik, Sprachwissenschaft und Kommunikationsforschung 41:
48–63.
Paul, P. (1972). 'How Do Loan Words from English Get Their German Gender?'
Queensland Studies in German Language and Literature, 1: 47-60.
Pawley, A. (1973). 'Some Problems in Proto-Oceanic Grammar', Oceanic Linguistics
12: 103-88.
(forthcoming). 'Using "He" and "She" for Inanimate Things in English:
Questions of Grammar and Worldview', in N. Enfield (ed.), Ethnosyntax.
and Sayaba, T. (1990). 'Possessive-marking in Wayan, a Western Fijian
Language: Noun Class or Relational System', in J. H. C. S. Davidson (ed.),
Pacific Island Languages: Essays in Honour of G. B.Milner. Honolulu: University
of London and University of Hawaii Press, 147-71.
Payne, David. L. (1991a). A Classification of Maipuran Arawakan Languages
Based on Shared Lexical Retentions', in Derbyshire and Pullum (1991: 355-500).
(1991b). 'La interaccion de la fonologia, la gramatica y el lexico en la
investigacion comparativa del maipuran', Revista latinoamericana de estudios
etnolinguisticos. Linguistica Arawaka 6: 241-58.
Payne, Doris L. (1986). 'Noun classification in Yagua', in Craig (1986a: 113–31).
(1987). 'Noun Classification in the Western Amazon', Linguistic Sciences 9:
 References 477
— (1990). The Pragmatics of Word Order: Typological Dimensions of Verb Initial
Languages. Berlin: Mouton de Gruyter.
— (ed.) (1990). Amazonian Linguistics: Studies in Lowland South American
Indian Languages. Austin: University of Texas Press.
and Payne, T. E. (1990). 'Yagua', in Derbyshire and Pullum (1990: 249–74).
Payne, John (1989). 'Pamir languages', in R. Schmitt (ed.), Compendium Linguarum
Iranicarum. Wiesbaden: Reichert, 417–44.
Payne, Judith (1989). Lecciones para el aprendizaje del idioma Asheninca. Pucallpa:
Instituto Linguistico de Verano.
Pe, H. (1965). 'A Re-examination of Burmese Classifiers', Lingua 15: 163-86.
Peeke, M. C. (1968). 'Preliminary Grammar of Auca Ecuador'. PhD thesis, Indiana
University, Bloomington.
Pensalfini, R. (1997). 'Jingulu Grammar, Dictionary, and Texts'. PhD thesis,
Massachusetts Institute of Technology.
Perez-Pereira, M. (1991). 'The Acquisition of Gender: What Spanish Children Tell
Us', Journal of Child Language 18: 571-90.
Pet, W. J. A. (1987). 'Lokono Dian, the Arawak Language of Suriname: A
Grammatical Sketch of Its Grammatical Structure and Lexicon'. PhD thesis,
Cornell University.
Peterson, M. N. (1955). Ocherk litovskogo jazyka [An Outline of Lithuanian].
Moscow: Izdateljstvo Akademii Nauk SSSR.
Pike, K. (1949). 'A Problem in Morphology-Syntax Division', Acta Linguistica 5:
125-38.
Plank, F. (1986). 'Das Genus der deutschen Ge-Substantive und Verwandtes
(Beitrage zur Vererbungslehre 1)', Zeitschrift fur Phonetik, Sprachwissenschaft
und Kommunikationsforschung 39: 44–60.
and Schellinger, W. (1997). 'The Uneven Distribution of Genders over
Numbers: Greenberg Nos. 37 and 45', Linguistic Typology 1: 53-101.
Plungian, V. A. (1995) Dogon. Munich: Lincom Europa.
and Romanova, O. I. (1990). 'Imennaja klassifikacija: Grammaticheskij
aspekt' [Nominal Classification: Grammatical Aspect], Izvestija Akademii
Nauk, Serija literatury i jazyka 493: 231-6.
Polome, E. (1968). 'Lumumbashi Swahili', Journal of African Languages 7: 15-25.
Pope, M. K. (1952). From Latin to Modern French with Especial Consideration of
Anglo-Norman: Phonology and Morphology. Manchester: Manchester University
Press.
Popova, M. I. (1958). 'Grammaticeskie elementy jazyka v reci detej preddos-
kol'nogo vozrasta', Voprosy psixologii 4: 106-17. (Russian version of Popova
1973.)
(1973). 'Grammatical Elements of Language in the Speech of Pre-school
Children', in C. A. Ferguson and D. I. Slobin (eds.), Studies of Child Language
Development. New York: Holt, Rinehart and Winston, 269-80.
Posner, R. (1966). The Romance Languages. Cambridge: Cambridge University
Press.
Pozdniakov, K. I. (1995). Sravniteljnaja grammatika atlanticheskih jazykov [A Com-
parative Grammar of Atlantic Languages]. Moscow: Nauka.
478 Classifiers
Prado, M. (1979). 'Markedness and the Gender Feature in Spanish', in D. L.
Malsch et al. (eds.), Proceedings of the Eighth Annual Meeting of the Western
Conference on Linguistics. Carbondale, Edmonton: Linguistic Research Current
Inquiry into Language and Linguistics 26: 113–22.
Prasse, K.-G. (1972). Manuel de gramaire touaregue (Tahaggart). 3 vols., Copenhagen:
Akademisk Forlag.
and agg-albostan ag-Sidiyan, E. (1985). Tableaux morphologiques: dialects
Touareg de I'Adrar du Mali berbere. Copenhagen: Akademisk Forlag.
Press, M. L. (1979). Chemehuevi: A Grammar and Lexicon. Berkeley: University of
California Press.
Priestly, T. M. S. (1983). 'On "Drift" in Indo-European Gender Systems', Journal of
Indo-European Studies 11: 339–63.
(1993). 'Slovene', in B. Comrie and G. G. Corbett (eds.), The Slavonic Lan-
guages. London: Routledge, 388–51.
Pullum, G. K. and Zwicky, A. M. (1986). 'Phonological Resolution of Syntactic
Feature Conflict', Language 62: 751-73.
Pym, N., and Larrimore, B. M. (1979). 'The Iwaidja Verb System: A Description',
Papers on Iwaidja Phonology and Grammar. Darwin: Summer Institute of
Linguistics. Australian Aborigines Branch, 65–151.
Quigley, S. (forthcoming). Award Grammar Essentials. Ukarumpa: Summer
Institute of Linguistics.
Quinn, H. (forthcoming). Systems of Verbal Classification, iv: Agent and Patient
Oriented Differentiations.
Rakhilina, E. (forthcoming). 'Lokativnye predikaty v russkom jazyke stojatj, sidetj,
lezhatj' [Locative predicates in Russian 'stand', 'sit', 'lie'], Voprosy jazykozna-
nija.
Ramirez, H. (1992). Bahuana: une nouvelle langue de la famille Arawak. Paris:
Amerindia.
(1994). Le parler Yanomami des Xamatauteri. Paris.
(1997). A fala Tukano dos Yepa-masa, i: Gramatica. Manaus: Inspetoria
Salesiana.
Rankin, R. L. (1976). 'From Verb to Auxiliary to Noun Classifier and Definite
Article: Grammaticalization of the Siouan Verbs "sit", "stand", "lie"', in R. L.
Brown, K. Houlihan, and A. MacLeish (eds.), Proceedings of the 1976 Mid-
America Linguistics Conference. Minnesota: University of Minnesota, 273-83.
Rastorgueva, V. S., Efimov, V. A., and Kerimova, A. A. (1978). 'Iranskie jazyki'
[Iranian languages], in N. I. Konrad (ed.), Jazyki Azii i Afriki, ii: Indoevropejskie
jazyki. Moscow: Nauka, 7-253.
Rehg, K. (with D. C. Sohl) (1981). Ponapean Reference Grammar. Honolulu: Uni-
versity of Hawaii Press.
Reid, N. (1990). 'Ngan'gityemerri: A Language of the Daly River Region, Northern
Territory of Australia'. PhD thesis, Australian National University, Canberra.
(1997). 'Class and Classifier in Ngan'gityemerri', in Harvey and Reid (1997:
165-228).
Rice, K. (1989). A Grammar of Slave. Berlin: Mouton de Gruyter.
 References 479
Richards, J. (1973). 'Dificuldades na analise de possessao nominal em Waura', Serie
Linguistica 1: 11–29.
(1988). 'A estrutura verbal Waura', Serie Linguistica 9: 197-218.
Riddle, R. (1989). 'White Hmong Noun Classifiers and Referential Salience', paper
given at the 22nd International Conference on Sino-Tibetan Languages and
Linguistics. University of Hawaii, Manoa.
Rigsby, B., and Rude, N. (1996). 'Sketch of Sahaptin, a Sahaptian Language', in
Handbook of North American Indians. Languages, xvii. Washington, DC: Smith-
sonian Institution, 666-92.
Rijkhoff, J. (1990). 'Toward a Unified Analysis of Terms and Predications', in
J. Nuyts and A. M. Bolkenstein (eds.), Layers and Levels of Representation in
Language Theory: A Functional View. Amsterdam: John Benjamins, 165-92.
Roberts, J. R. (1987). Amele. London: Croom Helm.
Rodrigues, A. D. (1986). Linguas brasileiras: para o conhecimento das linguas
indigenas. Sao Paulo: Loyola.
(1995). 'Some Morphological and Syntactic Aspects of Kariri', paper pre-
sented at the SSILA meeting, Albuquerque, NM.
(1997). 'Nominal Classification in Kariri', Opcion 13: 65-79.
(1999). 'Macro-Je Languages', in Dixon and Aikhenvald (1999: 164-206).
Rosch, E. (1973). 'On the Internal Structure of Perceptual and Semantic Categories,
in T. E. Moore (ed.), Cognitive Development and the Acquisition of Language.
New York: Academic Press, 111-44.
(1975a). 'Cognitive Reference Points', Cognitive Psychology 1: 532–47.
(1975b). 'Cognitive Representations of Semantic Categories', Journal of
Experimental Psychology 104: 192-233.
(1987). 'Principles of Categorization', in E. Rosch and B. B. Lloyd (eds.),
Cognition and Categorization. Hillsdale, NJ: Erlbaum, 27-48.
Mervis, C. B., Gray, W. D., Johnson, D. M., and Boyes-Braem, P. (1976).
'Basic Objects in Natural Categories', Cognitive Psychology 8: 382–439.
Rothstein, R. A. (1973). 'Sex, Gender, and the October Revolution', in S. R.
Anderson and P. Kiparsky (eds.), A Festschrift for Morris Halle. New York:
Holt, Rinehart and Winston, 460–6.
Rowan, O., and Burgess, E. (1979). 'Parecis Grammar', Arquivo Linguistico 149.
Brasilia: SIL.
Royen, G. (1929). Die nominalen Klassifikations-Systeme in den Sprachen der Erde:
Historisch-kritische Studie, mit besonderer Berticksichtigung des Indogerman-
ischen. Vienna: Anthropos.
Rubino, C. R. G. (1997). 'A Reference Grammar of Ilocano'. PhD thesis, University
of California, Santa Barbara.
Rude, N. (1986). 'Graphemic Classifiers in Egyptian Hieroglyphics and Mesopota-
mian Cuneiform', in Craig (1986a: 133-8).
Rumsey, A. (1982). An Intrasentence Grammar of Ungarinjin, North-Western Aus-
tralia. Canberra: Pacific Linguistics.
Rushforth, S. (1991). 'Uses of Bearlake and Mescalero (Athapaskan) Classificatory
Verbs', International Journal of American Linguistics 57: 251–66.
Russell, R. A. (1984). 'Historical Aspects of Subject–Verb Agreement in Arabic', in
480 Classifiers
G. Alvarez, B. Brodie, and T. McCoy (eds.), Proceedings of the First Eastern
States Conference on Linguistics. Columbus: Ohio State University, 116–27.
Sanches, M. (1977). 'Language Acquisition and Language Change: Japanese
Numeral Classifiers', in B. G. Blount and M. Sanches (eds.), Sociocultural
Dimensions of Language Change. New York: Academic Press, 51–62.
and Slobin, L. (1973). 'Numeral Classifiers and Plural Marking: an Implica-
tional Universal', Working Papers in Language Universals 11: 1-22.
Sandalo, F. (1996). 'A Grammar of Kadiweu'. PhD thesis, University of Pittsburgh.
Sands, A. K. (1995). 'Nominal Classification in Australia', Anthropological Linguis-
tics 37: 247-346.
Sapir, D. (1977). The Anatomy of Metaphor', in J. D. Sapir and J. C. Crocker (eds.),
The Social Use of Metaphor: Essays on the Anthropology of Rhetoric. Philadelphia:
University of Pennsylvania Press, 3-32.
Sapir, E. (1932). 'Two Navajo Puns', Language 8: 217-19.
Sasse, H.-J. (1985). 'Sprachkontakt und Sprachwandel: die Grazisierung der
albanischen Mundarten Griechenlands', Papiere zur Linguistik 32: 37-95.
(1992). 'Language Decay and Contact-induced Change: Similarities and
Differences', in M. Brenzinger (ed.), Language Death: Factual and Theoretical
Explorations with Special Reference to East Africa. Berlin: Mouton de Gruyter,
59-80.
Saul, J. E., and Wilson, N. F. (1980). Nung Grammar. Dallas: Summer Institute of
Linguistics and the University of Texas at Arlington.
Sauvageot, S. (1967). 'Note sur la classification nominale en bainouk', in La
Classification nominale, 225-36.
Saxton, D. (1982). 'Papago', in R. W. Langacker (ed.), Studies in Uto-Aztecan
Grammar, iii: Uto-Aztecan Grammatical Sketches. Dallas: Summer Institute of
Linguistics and the University of Texas at Arlington, 93-266.
Schane, S. A. (1970). 'Phonological and Morphological Markedness', in M. Bierwisch
and K. E. Heidolph (eds.), Progress in Linguistics. The Hague: Mouton, 286–94.
Schaub, W. (1985). Babungo. London: Croom Helm.
Schauer, S., and Schauer, J. (1978). 'Una gramatica del Yucuna', Artigos en linguis-
tica e campos afines 5: 1-52.
Scherbak, A. M. (1977). Ocherki po sravniteljnoy morfologii tyurkskih jazykov: imya
[Essays on comparative morphology of Turkic languages: noun]. Leningrad:
Nauka.
Schmidt, A. (1985). Young People's Dyirbal: An Example of Language Death from
Australia. Cambridge: Cambridge University Press.
Schwartz, L., Newman, P., and Sani, S. (1988). 'Agreement and Scope of Modifica-
tion in Hausa Coordinate Structures', Papers from the Annual Regional Meeting
of the Chicago Linguistic Society 24: 278-90.
Seiler, H. (1977). Cahuilla Grammar. Banning, Calif: Malki Museum Press.
(1983). Possession as an Operational Dimension of Language. Tubingen:
Gunter Narr.
(1985). 'Zum Verhaltnis von Genus und Numerus', in H. M. Olberg,
G. Schmidt, and H. Bothien (eds.), Sprachwissenschaftliche Forschungen: Fest-
schrift fur Johann Knobloch. Zum 65. Geburtstag am 5. Januar 1984 dargebracht
 References 481
von Freunden und Kollegen. Innsbruck: Institut fur Sprachwissenschaft
Innsbrucker Beitrage zur Kulturwissenschaft, 453-7.
(1986). Apprehension: Language, Object and Order, iii: The Universal Dimen-
sion of Apprehension. Tubingen: Gunter Narr.
— (1987). 'Genus und Pragmatizitat', Cahiers Ferdinand de Saussure 41: 205-18.
— and Lehmann, C. (eds.). (1982). Apprehension: Das sprachliche Erfassen von
Gegenstanden, i: Bereich und Ordnung der Phanomene. Tubingen: Narr.
and Stachowiak, F. J. (eds.). (1982). Apprehension: Das sprachliche Erfassen
von Gegenstanden, iii: Die Techniken und ihr Zusammenhang in Einzelsprachen.
Tubingen: Narr.
Seiler, W. (1985). Imonda, a Papuan Language. Canberra: Pacific Linguistics.
(1986). 'From Verb Serialisation to Noun Classification', in Papers in Pidgin
and Creole Linguistics 24. Canberra: Pacific Linguistics, 11–19.
(1989). 'Noun-classificatory Verbal Prefixes as Reanalysed Serial Verbs', Lin-
gua 68: 189-206.
Seki, L., and Aikhenvald, A. Y. (forthcoming). 'A Reconstruction of the Proto-
Xinguan Arawak'.
Selischev, A. M. (1928). Jazyk revoliutsionnoi epokhi: iz nabliudenij nad russkim iazy-
kom poslednikh let (1917-1926) [The Language of the Revolutionary Age: From
Notes on Russian of the Recent Years]. Moscow: Rabotnik prosveshcheniya.
Senft, G. (1985). 'Klassifikationspartikel im Kilivila: Glosses zu ihrer morpho-
logischen Rolle, ihrem Inventar und ihrer Funktion in Satz und Diskurs',
Linguistische Berichte 99: 373-93.
(1986). Kilivila: The Language of Trobriand Islanders. Berlin: Mouton de
Gruyter.
(1987). 'The System of Classificatory Particles in Kilivila Reconsidered: First
Results on Its Inventory, Its Acquisition, and Its Usage', Language and Linguis-
tics in Melanesia 16: 100-25.
(1991). 'Network Models to Describe the Kilivila Classifier System', Oceanic
Linguistics 30: 131-55.
(1993). 'What Do We Really Know about Nominal Classification Systems?',
contribution to workshop 'Back to Basic Issues in Nominal Classification', Cog-
nitive Anthropology Research Group, Max-Planck Institute for Psycholinguistics.
(1994). 'Grammaticalization of Body-parts Terms in Kilivila', Language and
Linguistics in Melanesia 25: 98-9.
(1996). Classificatory Particles in Kilivila. New York: Oxford University Press.
(ed.) (1997). Referring to Space: Studies in Austronesian and Papuan
Languages. Oxford: Clarendon Press.
Serzisko, F. (1982). 'Gender, Noun Class and Numeral Classification: A Scale of
Classificatory Techniques', in R. Dirven and G. Radden (eds.), Issues in the
Theory of Universal Grammar. Tubingen: Gunter Narr, 95-123.
Shaul, D. L. (1986). Topics in Nevome Syntax. Berkeley: University of California
Press.
Shawcross, W. (1979). Sideshow: Kissinger, Nixon and the Destruction of Cambodia.
New York: Simon & Schuster.
Shepard, G. Jr. (1997). 'Noun Classification and Ethnozoological Classification in
482 Classifiers
Machiguenga, an Arawakan Language of the Peruvian Amazon', Journal of
Amazonian Languages 1: 29-57.
Shepardson, K. W. (1982). An Integrated Analysis of Swahili Augmentative-
Diminutives', Studies in African Linguistics 13: 53-76.
Sherzer, J. (1976). An Areal-Typological Study of American Indian Languages North
of Mexico. Amsterdam: North-Holland.
Shields, K. Jr. (1978). 'English Gender: Some New Approaches to an Old Problem',
Linguistics, special issue: 205-25.
Silverman, M. G. (1962). 'Numeral Classifiers in the Gilbertese Language', Anthro-
pology Tomorrow 8: 41-56.
Silverstein, M. (1976). 'Hierarchy of Features and Ergativity', in Dixon (1976:
112-71).
(1986). 'Classifiers, Verb Classifiers, and Verbal Categories', Berkeley Linguis-
tics Society: Proceedings of the Annual Meeting 12: 497-514.
Sirk, U. (1983). The Buginese Language. Moscow: Nauka.
Slobin, Dan I. (1977). 'Language Change in Childhood and in History', in
J. Macnamara (ed.), Language Learning and Thought. New York: Academic
Press, 185-214.
Smith, I., and Johnson, S. (1999). 'Kugu-Nganhcara', in R. M. W. Dixon and
B. Blake (eds.), Handbook of Australian Languages, v. Melbourne: Cambridge
University Press.
Smith-Stark, S. (1974). 'The Plurality Split', Papers from the Annual Regional
Meeting of the Chicago Linguistic Society 10: 657-71.
Smoczynska, M. (1985). 'The Acquisition of Polish', in D. I. Slobin (ed.), The
Crosslinguistic Study of Language Acquisition, i: The Data. Hillsdale, NJ:
Erlbaum, 595-686.
Sohn, H.-M. (1994). Korean. London: Routledge.
Sokolova, V. S. (1966). 'Shugnano-rushanskaya gruppa' [Shugnan-Rushan Group],
in V. V. Vinogradov (ed.), Jazyki narodov SSSP, i: Indoevropejskie jazyki.
Moscow: Nauka, 362-97.
Speece, R. (n.d.). Angave Grammar. Ukarumpa: Summer Institute of Linguistics.
Speirs, A. (1974). 'Classificatory Verb Stems in Tewa', Studies in Linguistics 24:
45–74.
Spitulnik, D. (1989). 'Levels of Semantic Restructuring in Bantu Noun Classifica-
tion', in P. Newman and R. D. Botne (eds.), Current Approaches to African
Linguistics, v. Dordrecht: Foris Publications, 207-20.
Sridhar, S. N. (1990). Kannada. London: Routledge.
Stebbins, T. (1997). 'Asymmetrical Nominal Number Marking: a Functional
Account', Sprachtypologie und Universalienforschung 50: 5–47.
Steele, S. (1,978). 'Word Order Variation: A Typology Study', in J. H. Greenberg,
C. A. Ferguson, and E. A. Moravcsik (eds.), Universals of Human Language, iv:
Syntax. Stanford: Stanford University Press, 585-623.
Steinberg, E., and Caskey, A. F. (1988). 'The Syntax and Semantics of Gender
Disagreement: An Autolexical Approach', Papers from the Annual Regional
Meeting of the Chicago Linguistic Society 24: 291-303.
Steinhauer, H. (1986). 'Number in Biak: Counterevidence to Two Alleged Language
 References 483
Universals: A Summary', in P. Geraghty, L. Carrington, and S. A. Wurm (eds.),
FOCAL I: Papers from the Fourth International Conference on Austronesian
Linguistics. Canberra: Pacific Linguistics, 171–3.
Storch, A. (1997). 'Where Have All the Noun Classes Gone? A Case Study in
Jukun', Journal of African Languages and Linguistics 18: 157-70.
Strom, C. (1992). Retuara Syntax. Dallas: Summer Institute of Linguistics and the
University of Texas at Arlington.
Stucky. S. U. (1978). 'How a Noun Class System May Be Lost: Evidence from
Kituba Lingua Franca Kikongo', Studies in the Linguistic Sciences 8: 217-33.
Stumme, H. 1899. Handbuch des Schilhischen von Tazerwalt. Leipzig: J. C.
Hinrichs'sche Buchhandlung.
Stump, G. T. (1993). 'Reconstituting Morphology: The Case of Bantu Preprefixa-
tion', Linguistic Analysis 23: 169-204.
Subrahmanyam, P. S. (1968). A Descriptive Grammar of Gondi. Annamalainagar:
Annamalai University.
Suarez, J. A. (1983). The Mesoamerican Indian Languages. Cambridge: Cambridge
University Press.
Suppalla, T. (1986). 'The Classifier System in American Sign Language', in Craig
(1986a: 181–214).
Suzman, S. M. (1980). 'Acquisition of the Noun Class System in Zulu', Papers and
Reports on Child Language Development 19: 45-52.
(1982). 'Strategies of Acquiring Zulu Concord', South African Journal of
African Languages 2: 53-67.
Sweetser, E. (1986). 'Polysemy vs. Abstraction: Mutually Exclusive or Complemen-
tary?', Berkeley Linguistics Society: Proceedings of the Annual Meeting 12: 528-
38.
(1988). 'Grammaticalization and Semantic Bleaching', Berkeley Linguistics
Society: Proceedings of the Annual Meeting 14: 389–405.
T'sou, B. K. (1976). 'The Structure of Numeral Classifier Systems', in P. N. Jenner,
L. C. Thompson, and S. Starosta (eds.), Austroasiatic Studies, ii. Honolulu:
University of Hawaii Press, 1215–47.
Tai, J. H.-Y. (1992). 'Variation in Classifier Systems Across Chinese Dialects:
Towards a Cognition-Based Semantic Approach', in Symposium Series of the
Institute of History and Philology. Academia Sinica. Number 2. Chinese Lan-
guages and Linguistics. Chinese Dialects. Taipei, Republic of China, 587-608.
Talmy, L. (1985). 'Lexicalization Patterns: Semantic Structure in Lexical Forms', in
T. Shopen (ed.) Language Typology and Syntactic Description, iii: Grammatical
Categories and the Lexicon. Cambridge: Cambridge University Press, 57-149.
Taylor, D. M. (1952). 'Sameness and Difference in Two Island Carib Dialects',
International Journal of American Linguistics 18: 223-30.
Taylor, J. R. (1989). 'Possessive Genitives in English', Linguistics 21: 663-86.
(1995). Linguistic Categorization: Prototypes in Linguistic Theory. 2nd edn.
Oxford: Clarendon Press.
Thiesen, W. (1996). Gramatica del idioma Bora. Pucallpa: Instituto Linguistico de
Verano.
Thomason, S. G., and Kaufman, T. (1988). Language Contact, Creolization and
Genetic Linguistics. Berkeley: University of California Press.
484 Classifiers
Thompson, C. (1993). 'The Areal Prefix Hu- in Koyukon Athabaskan', Inter-
national Journal of American Linguistics 59: 315-33.
Tiersma, P. M. (1982). 'Local and General Markedness', Language 58: 832–49.
Todd, E. (1975). The Solomon Language Family', in S. Wurm (ed.), New Guinea
Area Languages and Language Study, i: Papuan Languages and the New Guinea
Linguistic Scene. Canberra: Pacific Linguistics, 805–46.
Traill, A. (1994). A Xoo! Dictionary. Cologne: Rudiger Koppe.
Trask, L. (1993). A Dictionary of Grammatical Terms in Linguistics. London:
Routledge.
Traugott, E. C. (1988). 'Pragmatic Strengthening and Grammaticalization', Pro-
ceedings of the Fourteenth Annual Meeting of the Berkeley Linguistics Society.
General Session and Parasession on Grammaticalization, 406–16.
and Heine, B. (eds.) (1991a). Approaches to Grammaticalization. 2 vols.
Amsterdam: John Benjamins.
(1991b). Introduction to Traugott and Heine (1991a: 1-14).
Trudgill, P. (1977). 'Creolization in Reverse: Reduction and Simplification in the
Albanian Dialects of Greece', Transactions of the Philological Society 77: 32-50.
Tseng, O., Chen, S., and Hung, D. (1991). 'The Classifier Problem in Chinese
Aphasia', Brain and Language 41: 184-202.
Tsonope, J. (1988). 'The Acquisition of Setswana Noun Class and Agreement
Morphology—with Special Reference to Demonstratives and Possessives'. PhD
thesis, State University of New York at Buffalo.
Tucker, A. N., and Mpaayei, J. T. O. (1955). A Maasai Grammar with Vocabulary.
London: Longmans.
Tversky, B. (1986). 'Components and Categorization', in Craig (1986a: 63–76).
Vail, L. (1974). 'The Noun Classes in Ndali', Journal of African Languages 11: 21-
47.
Vapnarsky, V. (1993). De quelques procedes de classification en Maya Itza: les
dassificateurs numeraux. Chantiers Amerindia, 11.2. Paris: Association d'Ethno-
linguistique Amerindienne.
Vasmer, M. (1953). Russisches etymologisches Worterbuch, i. Heidelberg: Carl
Winter.
Vidal, A. (1994). 'Noun Classifiers in Pilaga: A Study on Grammaticalization and
Deixis', paper presented at the SSILA/CAIL conference at the 96th AAA Meet-
ing, Atlanta, Nov.–Dec. 1994.
(1995). 'Noun Classification in Pilaga (Guaykuruan)'. MA thesis, University
of Oregon, Eugene.
(1997). 'Noun Classification in Pilaga: Guaykuruan', Journal of Amazonian
Languages 1: 60-111.
Vietze, H. P (1979). 'Nominalklassen in altaischen Sprachen', Zeitschrift zur
Phonetik, Sprachwissenschaft und Kommunikationsforschung 32: 745-51.
Vincennes, L., and Dallet, J. (1960). Initiation a la langue berbere. Kabylie. Fort
National: Fichier de Documentation Berbere.
Vinogradov, V. A. (1996). 'Coexistent Classificative Systems in Ngyembccn', paper
presented at the Department of African Linguistics, Leiden.
 References 485
Voeltz, E. (1971). 'Surface Constraints and Agreement Resolution: Some Evidence
from Xhosa', Studies in African Linguistics 2: 37–60.
Vycichl, W. (1957). 'L'Article defini du berbere', in Memorial Andre Basset, Paris,
pp. 139–46.
Wajanarat, S. (1979). 'Classifiers in Mal (Thin)', Mon-Khmer Studies 8: 295–303.
Wald, B. (1975). 'Animate Concord in Northeast Coastal Bantu: Its Linguistic and
Social Implications as a Case of Grammatical Convergence', African Linguistics
6: 267-314.
Walsh, M. (1976). 'The Murinypata Language of North-west Australia'. PhD
thesis, Australian National University, Canberra.
(1997). 'Nominal Classification and Generics in Murrihnpatha', in Harvey
and Reid (1997: 255-92).
Wang, Fu-shih (1972). The Classifier in the Wei Ning Dialect of the Miaio Lan-
guage in Kweichou', in H. C. Purnell (ed.), Miao and Yao Linguistic Studies:
Selected Articles in Chinese, Translated by Chang Yu-Hung and Cho Kwo-ray.
Southeast Asia Program, Cornell University, 111-85.
Waterson, N. (1966). 'Numeratives in Uzbek: A Study in Colligation and Colloca-
tion', in C. E. Bazell (ed.), In Memory of J. R. Firth. London: Longmans, 454–74.
Watkins, L. (1976). 'Position in Grammar: Sit, Stand, Lie'. Graduate Student
Association, University of Kansas..
(1984). A Grammar of Kiowa. Lincoln: University of Nebraska Press.
(1995). 'Noun Classes in Kiowa-Tanoan Languages', paper presented at the
Annual Meeting of the SSILA, Albuquerque, NM, 7-9 July.
Watters, J. (1981). 'A Phonology and Morphology of Ejagham—With Notes on
Dialect Variation'. PhD thesis, University of California, Los Angeles.
Welmers, W. E. (1973). African Language Structures. Berkeley: University of
California Press.
West, B. (1980). Gramatica popular del Tucano. Colombia: Instituto Linguistico de
Verano.
Westermann, D. (1947). 'Pluralbildung und Nominalklassen in einigen afrikan-
ischen Sprachen'. Abhandlungen der Deutschen Akademie der Wissenschaften zu
Berlin 1945/6. Phil.-hist. Klasse No. 1. Berlin.
Westley, D. O. (1991). Tepetotutla Chinantec Syntax. Dallas: Summer Institute of
Linguistics and the University of Texas at Arlington.
Whitney, H. (n.d.). Akoye Grammar Essentials. Ukarumpa: Summer Institute of
Linguistics.
Wierzbicka, A. (1985). 'Oats and Wheat: The Fallacy of Arbitrariness', in J.
Haiman (ed.), Iconicity in Syntax. Amsterdam: John Benjamins, 311—42.
(1996). Semantics: Primes and Universals. Oxford: Oxford University Press.
Wiesemann, U. (1972). Die phonologische und grammatische Struktur der Kaingang-
Sprache. The Hague: Mouton.
(ed.) (1986). Pronominal Systems. Tubingen: G. Narr.
Wilkins, D. (1981). 'Towards a Theory of Semantic Change'. Honours thesis,
Australian National University, Canberra.
(1989). 'Mparntwe Arrernte (Aranda). Studies in the Structure and Semantics
of Grammar'. PhD thesis, Australian National University, Canberra.
486 Classifiers
Wilkins, D. (1996). 'Natural Tendencies of Semantic Change and the Search for
Cognates', in M. Durie and M. Ross (eds.), The Comparative Method Reviewed.
Oxford: Oxford University Press, 264-304.
Williamson, K. (1989). 'Niger-Congo Overview', in J. Bendor-Samuel (ed.), The
Niger-Congo Languages. Lanham, MD: University Press of America, 3–45.
Wilson, P. J. (1992). Una descripcion preliminar de la gramatica del Achagua (Ara-
wak). Bogota: Summer Institute of Linguistics.
Wilson, W. H. (1982). Proto-Polynesian Possessive Marking. Canberra: Pacific
Linguistics.
Wise, M. R. (1986). 'Grammatical Characteristics of Preandine Languages:
Arawakan Languages of Peru', in Derbyshire and Pullum (1986: 567-642).
(1999). 'Small Language Families and Isolates in Peru', in Dixon and Aikhen-
vald (1999: 307–0).
Witherspoon, G. (1972). Language and Art in the Navajo Universe. Ann Arbor:
University of Michigan Press.
Woodbury, A. C. (1981). 'Study of the Chevak Dialect of Central Yupi'k Eskimo'.
PhD thesis, University of California, Berkeley.
Worsley, P. M. (1954). 'Noun-classification in Australian and Bantu: Formal or
Semantic?', Oceania 24: 275-88.
Wurm, S. A. (1981). 'Notes on Nominal Classification Systems in Ayiwo, Reef
Island, Solomon Islands', in A. Gonzales and D. Thomas (eds.), Linguistics
across Continents: Studies in Honor of Richard S. Pittman. Manila: Summer
Institute of Linguistics and The Linguistic Society of the Philippines, 123–42.
(1987). 'Semantics and World View in Languages of the Santa Cruz Archi-
pelago, Solomon Islands', in R. Steel and T. Threadgold (eds.), Language Topics:
Essays in Honor of Michael Halliday. Amsterdam: John Benjamins, 439-51.
(1992a). 'Change of Language Structure and Typology in a Pacific Language
as a Result of Culture Change', in T. Dutton (ed.), Culture Change, Language
Change: Case Studies from Melanesia. Canberra: Pacific Linguistics, 149-65.
(1992b). 'Some Features of the Verb Complex in Northern Santa Cruzan,
Solomon Islands', in T. Dutton, M. Ross, and D. Tryon (eds.), The Language
Game: Papers in Memory of Donald C. Laycock. Canberra: Pacific Linguistics,
527-51.
Yu, E. O. (1988). 'Agreement in Left Dislocation of Coordinate Structures', Papers
from the Annual Regional Meeting of the Chicago Linguistic Society 24: 322-37.
Zalizniak, A. A. (1967). Russkoje Imennoje Slovoizmenenie [Russian Nominal
Inflection]. Moscow: Russkij Jazyk.
and Paducheva, E. V. (1976). 'K tipologii otnositeljnogo predlozhenija'
[Towards a Typology of Relative Clauses], Semiotika i informatika 6: 51-101.
Zavala, R. (1992). El Kanjobal de San Miguel Acatan. Mexico: Universidad Auton-
oma de Mexico.
(1993). 'Multiple Classifier Systems in Akatek Mayan', contribution to the
workshop 'Back to Basic Issues in Nominal Classification', Cognitive Anthro-
pology Research Group, Max-Planck Institute for Psycholinguistics, Nijmegen.
(forthcoming). 'Multiple Classifier Systems in Akatek Mayan', in G. Senft
(ed.), Systems of Noun Classification. Cambridge: Cambridge University Press.
 References 487
Zograf, G. A. (1976). 'Indoarijskije jazyki' [Indoaryan Languages], in N. I. Konrad
et al. (eds.), Jazyki Azii i Afriki, i: Indoevropejskie jazyki. Moscow: Nauka, 110-
271.
Zubin, D. A., and Kopcke, K. M. (1981). 'Gender: A Less than Arbitrary Gram-
matical Category', Papers from the Annual Regional Meeting of the Chicago
Linguistic Society 17: 439–49.
(1986). 'Gender and Folk Taxonomy: The Indexical Relation between
Grammatical and Lexical Categorization', in Craig (1986a: 139-80).
and Shimojo, M. (1993). 'How "General" Are General Classifiers? With
Special Reference to Ko and Tsu in Japanese', in J. S. Guenter, B. A. Kaiser,
and C. C. Zoll (eds.), Proceedings of the Berkeley Linguistics Society: Proceedings
of the Annual Meeting 19: 490–502.
Zwicky, A. M. (1977). 'Hierarchies of person', in W. A. Beach, S. A. Fox, and
S. Philosoph (eds.), Papers from the Thirteenth Regional Meeting, Chicago
Linguistic Society, 714–33. Chicago: Chicago Linguistic Society.
This page intentionally left blank
 List of Languages

Abau (isolate: East Sepik Province, Papua New Guinea)

Abaza (Northwest Caucasian)
Abelam (Papuan)
Abkhaz (Northwest Caucasian)
Abu' (Arapesh)
Acehnese (Western Austronesian)
Achagua (North Arawak, Arawak)
Ahaggar (Tuareg, Berber, Afroasiatic)
Ahtna (Athabaskan)
Aikana (isolate)
Ainu (isolate)
Akan (Kwa)
Akatek (Kanjobalan branch of Mayan)
Akhvakh (Northeast Caucasian)
Akoye (Angan, Papuan)
Alamblak (Sepik, Papuan)
Alawa (Australian)
Albanian (Indo-European)
Amele (Papuan)
Amharic (Semitic)
Amuesha (Arawak)
Ancient Egyptian (Afroasiatic)
Andi (Northeast Caucasian)
Angave (Angan, Papuan)
Animere (Togo Remnant)
Anindilyakwa (Australian)
Apalai (Carib)
Apurina (Arawak)
Arabic (Semitic, Afroasiatic)
Arapesh (Torricelli, Papuan)
Archi (Northeast Caucasian)
Armenian (Indo-European)
Arvanitika (dialect of Albanian, Indo-European)
Asmat (Asmat, Papuan)
Assamese (Indo-Aryan, Indo-European)
Atakapa (isolate)
Awa Pit (or Cuaiquier; Barbacoan)
490 Classifiers
Awabakal (Australian)
Awara (Wantoat family, Morobe province, Papuan)
Ayacucho Quechua (Quechuan)
Aymara (Aymara)
Aztec (Uto-Aztecan), see Nahuatl
Babungo (Bantu)
Badyaranke (Gur)
Bagval (Northeast Caucasian)
Bahnar (Central Bahnaric, Mon-Khmer)
Bahwana (North Arawak, Arawak)
Baining (Papuan, East New Britain)
Bai'nouk (West Atlantic)
Balinese (Western Austronesian)
Banda (Eastern Nilotic-Adamawa, Nilo-Saharan)
Banggais (Loinang-Banggai; Central and South Celebes, Austronesian)
Baniwa (of Icana) (North Arawak, Arawak)
Banjalang (Australian)
Banz (Chimbu, Papuan)
Barasano (East Tucano, Tucano)
Bare (North Arawak, Arawak)
Bari (Eastern Nilotic, Nilo-Saharan)
Basque (isolate)
Beludzhi (Indo-Iranian)
Bengali (Indie)
Biak (South Halmahera-West New Guinea subgroup; Austronesian)
Bihari (Indo-Aryan, Indo-European)
Biloxi (isolate)
Bilua (Papuan, Solomon Islands)
Bine (Fly River, Papuan)
Black Tai (Tai)
Blackfoot (Algonquian)
Bodo (Sino-Tibetan)
Bora (Bora-Witoto)
Bororo (Macro-Je)
Boumaa Fijian (Oceanic, Austronesian)
Bowili (Togo Remnant)
Brazilian Portuguese (Romance, Indo-European)
Budukh (Lezgian, Daghestanian)
Bugis (South Celebes)
Bukiyip (Arapesh, Torricelli, Papuan)
Burmese (Tibeto-Burman)
Burushaski (isolate)
Bwamu (Gur)
 List of Languages 491
Caddo (Caddoan)
Cahuilla (Uto-Aztecan)
Camus (Eastern Nilotic, Nilo-Saharan)
Canela (Je)
Cantabrian Spanish (Romance, Indo-European)
Carrier (Athabaskan)
Cayuga (Iroquoian)
Central Khoisan (Khoe)
Central Pomo (Pomoan)
Central Yup'ik Eskimo (Eskimo-Aleut)
Chacabano Zamboangueno (creole)
Chamalal (Northeast Caucasian)
Chambri (Lower Sepik, Papuan)
Chamicuro (Arawak)
Chayahuita (Cahuapanan)
Chechen (Nakh-Daghestanian)
Chemehuevi (Uto-Aztecan)
Chenapian (Papuan)
Cherokee (Iroquoian)
Chevak (Central Yup'ik Eskimo)
ChiBemba (Bantu)
Chimila (Chibchan)
Chinese (Mandarin, Cantonese: Sino-Tibetan)
Chinantepec Chinantec (Chinantec, Otomanguean)
Chipewyan (Athabaskan)
Chiricahua Apache (Athabaskan)
Chitimacha (isolate)
Chrau (Mon-Khmer)
Chuj (Mayan)
Chukchee (Chukotka-Kamchatkan)
Classical Arabic, see Arabic
Colville (Salishan)
Comaltepec Chinantec (Chinantec, Otomanguean)
Comanche (Uto-Aztecan)
Cora (Uto-Aztecan)
Cree (Algonquian)
Creek (Muskogean)
Cuaiquer (Barbacoan)
Cuiba (Guahibo)
Czech (Slavic, Indo-European)
Dahalo (East Cushitic, Afroasiatic)
Daju (East Sudanic)
Dakota (Siouan)
492 Classifiers
Dangbon (Australian)
Danish (Germanic, Indo-European)
Dasenech (East Cushitic)
Daw (Maku)
Dena'ina (Athabaskan)
Deni (Arawa)
Dhegiha (isolate)
Dioi (Austroasiatic)
Diyari (Australian)
Dogon (Gur)
Dogrib (Athabaskan)
'Dongo-ko (Mba, Benue Congo)
Dulong-Rawang (Tibeto-Burman)
Dutch (Germanic, Indo-European)
Dyirbal (Australian)
Eastern Sutherland Gaelic (Celtic, Indo-European)
Egyptian Arabic (Semitic, Afroasiatic)
Ejagham (Cross River, Benue-Congo)
Emmi (Australian)
Enga (Engan, Papuan)
English, American English (Germanic, Indo-European)
Epena Pedee (Choco)
Eskimo, Chevak dialect of (Eskimo-Aleut)
Estonian (Balto-Finnic, Uralic)
Ewe (Kwa)
Eyak (Eyak-Athabaskan)
Eywo (Papuan, Central Bougainville)
Fanagalo (Bantu-based Creole)
Felefita (Arapesh, Papuan)
Fijian (Oceanic, Austronesian)
Finnish (Balto-Finnic, Uralic)
Folopa, or Podopa (Teberan)
French (Romance, Indo-European)
Ful (West Atlantic)
Fulfulde, see Ful
Gaagudju (Australian)
Gadjang (Australian)
Gaelic (Celtic, Indo-European)
Gapapaiwa (Oceanic, Austronesian)
Garifuna, or Black Carib (North Arawak, Arawak)
Garo (Austroasiatic)
Gaviao (Tupi)
German (Germanic, Indo-European)
 List of Languages 493
Gilbertese (Micronesian, Austronesian)
Godoberi (Northeast Caucasian)
Gola (West Atlantic, Niger-Congo)
Gold Palaung (Austroasiatic)
Gondi (Dravidian)
Gorontalo (Austroasiatic)
Grebo (Kru)
Greek (Indo-European)
Guahibo (Guahibo)
Guajiro (North Arawak, Arawak)
Gunbarlang (Australian)
Gunwinjgu (Australian)
Gur (Voltaic)
Gurr-goni (Australian)
Guugu Yimidhirr (Australian)
Gwich'in (Athabaskan)
Haida (isolate?)
Hakka (Sinitic, Sino-Tibetan)
Halia (Oceanic, Austronesian)
Hani (Tibeto-Burman)
Harar (Oromo, East Cushitic, Afroasiatic)
Hausa (Chadic, Afroasiatic)
Hawaiian (Polynesian, Austronesian)
Hebrew (Semitic, Afroasiatic)
Hiechware (Khoisan)
Hindi (Indo-Aryan, Indo-European)
Hittite (Anatolian, Indo-European)
Hixkaryana (Carib)
Hmong (Miao-Yao)
Hohodene (dialect of Baniwa of Icana, North Arawak, Arawak)
Hokkien (Sinitic, Sino-Tibetan)
Holikachuk (Athabaskan)
Hopi (Uto-Aztecan)
Hua (Papuan)
Hualapai (Uto-Aztecan)
Huave (isolate)
Huli (Engan, Papuan)
Hungarian (Ugric, Uralic)
Hupa (Athabaskan)
Hupda (Maku)
latmul (Ndu, Papuan)
Ibibio (Niger-Congo)
Icelandic (Germanic, Indo-European)
494 Classifiers
Idu (Tibeto-Burman)
Igbo (Kwa, Niger-Congo)
Ignaciano (South Arawak, Arawak)
Ika (Chibchan)
Ilocano (Western Austronesian)
Imonda (Waris, Papuan)
Indonesian, Bahasa Indonesia (Western Austronesian)
Ingush (Nakh-Daghestanian)
Iraqw (South Cushitic, Afroasiatic)
Irish (Celtic, Indo-European)
Island Carib (North Arawak, Arawak)
Italian (Romance, Indo-European)
Itonama (isolate)
Iwaidja (Australian)
Iwam (isolate: Sepik Province, Papua New Guinea)
Izayan (North Berber, Afroasiatic)
Jabuti (isolate)
Jacaltec (Kanjobalan branch of Mayan)
Japanese (isolate)
Jarawara (Arawa)
Javanese (Western Austronesian)
Jawoyn (Australian)
Jeh (Mon-Khmer)
Jingulu (Australian)
Jukun (Jukunoid, Niger-Congo)
Kabyle (North Berber, Afroasiatic)
Kadiweu (Guaicuruan)
Kaingang (Je)
Kakua (Maku)
Kalau Lagaw Ya, Kalaw Kawaw Ya, Western Torres Strait (Australian)
Kaliai-Kove (Oceanic, Austronesian)
Kam-Muang (Austroasiatic)
Kamoro (Asmat, Papuan)
Kana (Cross River, Benue-Congo)
Kannada (Dravidian)
Kariri (Kipea-Kariri, Macro-Je)
Karo (Tupi)
Karok (isolate)
Katcha (Kordofanian)
Kate (Huon, Papuan)
Katu (Katuic, Austroasiatic)
Kayapo (Je)
Kenya Pidgin Swahili (creole)
 List of Languages 495
Keriaka (Non-Austronesian, Central Bougaineville)
Ket (isolate)
Kewa (Engan)
Khinalug (Northeast Caucasian)
Khmer (Mon-Khmer)
Khmu (Khmuic, Austroasiatic)
Khoe (Khoisan)
Kikongo (Bantu)
Kikuyu (Bantu)
Kilivila (Austronesian, Trobriand Islands)
Kiowa (Kiowa-Tanoan)
Kipea (Kipea-Kariri, Macro-Je)
Kituba (Bantu)
Kiwai (Kiwaian, Papuan)
Koaia (isolate)
Koasati (Muskogean)
Kobon (Kalam)
Kolami (Dravidian)
Korafe (Binanderre, Papuan)
Kore (Eastern Nilotic, Nilo-Saharan)
Korean (isolate)
Kosraean (Micronesian, Austronesian)
Koyukon (Athabaskan)
Kraho (Je)
Ku Waru (Waris, Papuan)
Kubeo (Central Tucano, Tucano)
Kugu-Ngancara (Australian)
Kulina (Arawa)
Kuot (Papuan, New Ireland)
Kuria (Bantu)
Kurripaco (dialect of Baniwa of Icana, North Arawak, Arawak)
Kusaiean (Micronesian, Austronesian)
Kwoma, or Washkuk (distantly related to Ndu family, Papuan)
Kxoe (Khoisan)
Lahu (Tibeto-Burman)
Lak (Northeast Caucasian)
Lama (Gur, Niger-Congo)
Lama Lama (Australian)
Landuma (West Atlantic)
Lao (Tai)
Latin (Italic, Indo-European)
Latvian (Baltic, Indo-European)
Lavukaleve (Papuan, Solomon Islands)
496 Classifiers
Lezgian (Northeast Caucasian)
Lii (Austroasiatic)
Limilngan (Australian)
Lingala, Mankandza Lingala, Kinshasa Lingala (Bantu)
Lingua Geral, or Tupinamba, Tupi-Guarani
Lisu (Tibeto-Burman)
Lithuanian (Baltic, Indo-European)
Lokono (North Arawak, Arawak)
Longgu (Oceanic, Austronesian)
Loniu (Oceanic, Austronesian)
Lotuko (Eastern Nilotic, Nilo-Saharan)
Loven (West Bahnaric, Mon-Khmer)
Lower Chinook (Chinookan)
Luganda (Bantu)
Luiseno (Uto-Aztecan)
Ma (Ubangi, Niger-Congo)
Maa (Eastern Nilotic, Nilo-Saharan)
Maasai (Eastern Nilotic, Nilo-Saharan)
Maasina (Fulfulde, West Atlantic, Niger-Congo)
Machiguenga (Campa, Peruvian Arawak, Arawak)
Macushi (Carib)
Maka (Mataguayo)
Makassar (South Celebes)
Mal (or Thin) (Mon-Khmer)
Malak-Malak (Australian)
Malay (Western Austronesian)
Malinke (Mande)
Malto (Dravidian)
Mam (Mayan)
Manam (Oceanic, Austronesian)
Manambu (Ndu, Papuan)
Mandan (Siouan)
Marathi (Indo-Aryan, Indo-European)
Maricopa (Yuman)
Marind (Marind, Papuan)
Marrithiyel (Australian)
Marshallese (Micronesian, Austronesian)
Maung (Australian)
Mayali (Gunwinjgu) (Australian)
Mba (Ubangi, Niger-Congo)
Melpa (Chimbu, Papuan)
Menomini (Algonquian)
Mescalero Apache (Athabaskan)
 List of Languages 497
Miao of Wei Ning (Miao Yao, Austroasiatic)
Minangkabau (Western Austronesian)
Missima (Oceanic, Austronesian)
Miwok (Miwok-Costanoan)
Mixtec (Mixtecan, Otomanguean)
Mnong Gar (Austroasiatic)
Mocovi (Guaicuruan)
Mohawk (Iroquoian)
Mokilese (Micronesian, Austronesian)
Mon (Mon-Khmer)
Monumbo (Torrichelli phylum, Papuan)
Mori (Chapacuran)
Motuna (Papuan of Bougainville)
Movima (isolate)
Mparntwe Arrente (Australian)
Mufian (Arapesh, Papuan)
Muhiang (Arapesh)
Mulao (Tai)
Munduruku (Tupi)
Mupun (West Chadic, Afroasiatic)
Murrinhpatha (Australian)
Murui Witoto (Bora-Witoto)
Nadeb (Maku)
Nahuatl (Uto-Aztecan)
Naiki (Dravidian)
Nakhi (Sino-Tibetan)
Nama (Khoisan)
Nambiquara (Nambiquara)
Napues, or Kunua (Non-Austronesian, Central Bougainville)
Naro (Khoisan)
Nasioi (Papuan of Bougainville)
Natchez (isolate)
Nauru (Micronesian, Austronesian)
Navajo (Athabaskan)
Ndali (Bantu, Niger-Congo)
Ndunga-le (Mba, Ubangi, Niger-Congo)
Nepali (Indo-Aryan, Indo-European)
Nevome (Uto-Aztecan)
Newari (Tibeto-Burman)
Nez Perce (Sahaptian)
Ngala (Eastern Nilotic-Adamawa, Nilo-Saharan)
Ngala (Ndu, Papuan)
Ngalakan (Australian)
498 Classifiers
Ngandi (Australian)
Ngan'gityemerri (Australian)
Ngiyambaa (Australian)
Ngyembccn (Grasslands, Niger-Congo)
Nias (Western Austronesian)
Nicobarese (Austroasiatic)
Nivkh, or Gilyak (isolate)
Nootka (Wakashan)
Northern Tepehuan (Uto-Aztecan)
Ntifa (North Berber, Afroasiatic)
Nubi (Arabic-based creole)
Nung (Tai)
Nungali (Australian)
Nunggubuyu (Australian)
Nyanja (Bantu)
Ocaina (Bora-Witoto)
Ojibway (Algonquian)
Old Church Slavonic (Slavic, Indo-European)
Old English (Germanic, Indo-European)
Olgolo (Australian)
Ollari (Dravidian)
Olo (Torricelli, Papuan)
Omani Arabic (Semitic, Afroasiatic)
Omani-Zanzibar dialect of Arabic (Semitic, Afroasiatic)
Ongamo (Eastern Nilotic, Nilo-Saharan)
Oriya (Indo-Aryan, Indo-European)
'Oro Nao (Chapacuran)
Oromo (East Cushitic, Afroasiatic)
Ossete (Iranian, Indo-European)
Palaungic (Austroasiatic)
Palikur (North Arawak, Arawak)
Panara (Je)
Panare (Carib)
Papago (Uto-Aztecan)
Pareci (South Arawak, Arawak)
Parji (Dravidian)
Patjtjamalh, or Bachamal (Australian)
Paumari (Arawa)
Persian (Iranian, Indo-European)
Petats (Oceanic, Austronesian)
Piapoco (North Arawak, Arawak)
Pilaga (Guaicuruan)
Piraha (Mura-Piraha)
 List of Languages 499
Piratapuya (East Tucano, Tucano)
Pittapitta (Australian)
Pogoro (Bantu)
Polish (Slavic, Indo-European)
Pomoan (Pomo)
Ponapean (Micronesian, Austronesian)
Ponca (Siouan)
Portuguese, Brazilian (Romance, Indo-European)
Potawatomi (Algonquian)
Puluwat (Micronesian, Austronesian)
Punjabi (Indo-Aryan, Indo-European)
Qafar (Cushitic, Afroasiatic)
Qiang (Tibeto-Burman)
Queyu (Tibeto-Burman)
Raga (Oceanic, Austronesian)
Rama (Chibchan)
Rembarrnga (Australian)
Rendille (East-Cushitic, Afroasiatic)
Resigaro (North Arawak, Arawak)
Retuara (Central Tucano, Tucano)
Rikbaktsa (Macro-Je)
Ro·lo·m (South Bahnaric, Austroasiatic)
Romansch (Romance, Indo-European)
Roshani (Pamir, Indo-European)
Rotokas (Papuan, Central Bougainville)
Rumanian (Romance, Indo-European)
Russian (Slavic, Indo-European)
Sahaptin (Sahaptian)
Saliba (Saliba-Piaroa)
Sanskrit (Indo-Aryan, Indo-European)
Savosavo (isolate)
Scottish Gaelic (Celtic, Indo-European)
Sedang (North Bahnaric, Mon-Khmer)
Seghrouchen (North Berber, Afroasiatic)
Sele Fara, dialect of Slovene (Slavic, Indo-European)
Seneca (Iroquoian)
Serbo-Croatian (Slavic, Indo-European)
Sesotho (Bantu)
Setswana (Bantu)
Shan (Tai, Austroasiatic)
Shilh (North Berber, Afroasiatic)
Shona (Bantu)
Siamese (Austroasiatic)
500 Classifiers
Sierra Popoluca (Mixe-Zoquean)
Sinasina (Chimbu, Papuan)
SiSwati (Bantu)
Siuci (dialect of Baniwa of Icana, North Arawak, Arawak)
Siwa (East Berber, Afroasiatic)
Slave (Athabaskan)
Slovene (Slavic, Indo-European)
Sochiapan Chinantec (Chinantec, Otomanguean)
Southeastern Pomo (Pomoan)
Southern Paiute (Uto-Aztecan)
Spanish (Romance, Indo-European)
Squamish (Salishan)
Standard Fijian (Oceanic, Austronesian)
Swahili (Bantu)
Swedish (Germanic, Indo-European)
Tabar (Oceanic, Austronesian)
Tacana (Pano-Tacana)
Taenae (Angan, Papuan)
Tagalog (Western Austronesian)
Takia (Oceanic, Austronesian)
Tamachek (Tuareg, Berber, Afroasiatic)
Tamambo (Oceanic, Austronesian)
Tamazight (North-Berber, Afroasiatic)
Tamil (Dravidian)
Tanaina, Dena'ina (Athabaskan)
Tanana (Athabaskan)
Tarascan (isolate)
Tariana (North Arawak, Arawak)
Taulil (Papuan language of East New Britain)
Tazerwalt Shilh (North Berber, Afroasiatic)
Telugu (Dravidian)
Teop (Austronesian of Bougainville)
Tepetotutla Chinantec (Chinantec, Otomanguean)
Tequislatec (Tequislatecan)
Terena (South Arawak, Arawak)
Teso (Eastern Nilotic, Nilo-Saharan)
Tewa (Kiowa-Tanoan)
Thai (Tai)
Tibetan (Tibeto-Burman)
Timbira (Je)
Tinrin (New Caledonia, Austronesian)
Tiwi (Australian)
Tlingit (Eyak-Athabaskan)
 List of Languages 501
To'aba'ita (Oceanic, Austronesian)
Toba (Guaicuruan)
Toba Batak (Austroasiatic)
Tolai (West New Britain, Austronesian)
Tonga (Bantu)
Tongan (Polynesian, Austronesian)
Toposa (Eastern Nilotic, Nilo-Saharan)
Totonac (Totonacan)
Trumai (isolate)
Truquese (Micronesian, Austronesian)
Tsafiki(Barbacoan)
Tsez (Northeast Caucasian)
Tsimshian (Penutian)
Tsova-Tush (Nakh, Northeast Caucasian)
Tucano (East Tucano, Tucano)
Tunica (isolate)
Turkana (Eastern Nilotic, Nilo-Saharan)
Turkish (Turkic)
Tuyuca (Central Tucano, Tucano)
Twi (dialects of Akan)
Tzeltal (Mayan)
Tzotzil (Mayan)
Ungarinjin (Australian)
Upper Chinook (Chinookan)
Upper Sorbian (Slavic, Indo-European)
Urubu-Kaapor (Tupi-Guarani, Tupi)
Uzbek (Turkic)
Vietnamese (Viet-Muong, Austroasiatic)
Wagaya (Australian)
Wambaya (Australian)
Wangkumara (Australian)
Wantoat (Wantoat family, Morobe province, Papuan)
Waorani (isolate)
Wara (Fly river, Papuan)
Warao (isolate)
Wardaman (Australian)
Warekena (North Arawak, Arawak)
Wari' (Chapacura)
Waris (Waris, Papuan)
Warndarang (Australian)
Warray (Australian)
Washo (isolate)
Waura (Xinguan Arawak, Arawak)
502 Classifiers
Wayan (Western Fijian, Oceanic)
Wei Ning (Miao Yao, Austroasiatic)
West Flemish (Germanic, Indo-European)
Western Apache (Athabaskan)
Western Oromo (East Cushitic, Afroasiatic)
Western Torres Strait (see Kalau Lagaw Ya)
White Hmong (Miao-Yao)
White Tai (Tai)
Wipi (Fly river, Papuan)
Wiyot (Ritwan, Algic)
Wogamusin (Papuan)
Wolof (West Atlantic)
Worrorra (Australian)
Wu-Ming (Austroasiatic)
Wunambal (Australian)
Xamatauteri (Yanomami)
Xhosa (Bantu)
!X66 (Khoisan)
!Xu (Khoisan)
Yagua (Peba-Yagua)
Yaigin (Australian)
Yana (isolate)
Yandruwanhtha (Australian)
Yankuntjatjara (Australian)
Yanomami (Yanomami)
Yanyuwa (Australian)
Yapese (Oceanic, Austronesian)
Yaruro (isolate)
Yavapai (Yuman)
Yawalapiti (Xinguan Arawak)
Yessan-Mayo (Papuan)
Yidiny (Australian)
Yimas (Lower Sepik, Papuan)
Yonggom (Ok)
Yoruba (Kwa)
Young People's Dyirbal (Australian)
Yuchi (isolate)
Yucuna (North Arawak, Arawak)
Yue dialects (Sinitic, Sino-Tibetan)
Yuki (Yukian)
Yurok (Ritwan, Algic)
Zairean Swahili, see Swahili
Zande (Eastern Nilotic-Adamawa)
 List of Languages 503
Zapotec (Zapotecan, Otomangean)
Zayse (Omotic, Afroasiatic)
Zemmur (North Berber, Afroasiatic)
Zezuru (dialects of Shona, Bantu)
Zulu (Bantu)
 List of Language Families, Linguistic Areas, and
Proto-languages

African Central Celebes

Afroasiatic Central Jukunoid
Algonquian Central Khoisan
Amazonian Central Tucano
Anatolian Chadic
Angan Chapacuran
Arapesh Chapahuan
Arawa Chemakuan
Arawak Chibchan
Aslian Chimbu
Asmat Chinantec (Otomanguean)
Athabaskan Chinookan
Australian Choco
Austroasiatic Chukotka-Kamchatkan
Austronesian Costanoan
Aymara Cross River
Baltic Cushitic
Balto-Finnic Daghestanian
Bantu Daly
Barbacoan Dravidian
Benue-Congo (Cross River) East Bahnaric
Berber East Berber
Binanderre East Cushitic
Bolivian East New Britain
Bora-Witoto East Sepik
Caddoan East Slavic
Cahuapanan East Tucano
Campa Eastern Adamawa
Carib Eastern Mindi
Caucasian Eastern Nilotic
Celtic Eastern Nilotic-Adamawa
Central American Eastern Sudanic
Central Bahnaric Engan
Central Bougainville Eskimo-Aleut
 List of Language Families 505
Eyak-Athabaskan Lowland East Cushitic
Finno-Ugric Macro-Je
Fly River Magadhan
Germanic Maku
Gran Choco Malayo-Polynesian
Grassfields Bantu Mande
Grasslands Bantu Marind
Great Basin Mataguayo
Guahibo Mayan
Guaicuruan Mba
Gulf Melanesian
Gur Mesoamerican
Harakmbet Miao
Hokan Miao-Yao
Huon Micronesian
Indie Milne Bay Province
Indo-Aryan Mindi
Indo-European Miwok-Costanoan (or Utian)
Indo-Iranian Mixe-Zoquean
Iranian Mixtecan (Otomanguean)
Iroquoian Mon-Khmer
Italic Morobe Province
Je Munda
Jivaro Mura-Piraha
Kadugli-Kongo Muskogean
Kalam Nakh-Daghestanian
Kanjobalan Mayan Nambiquara
Katuic Ndu
Kegboid New Caledonia
Khmuic New Indo-Aryan
Khoe Niger-Congo
Khoisan Nilo-Saharan
Kiowa-Tanoan Non-Austronesian (or Papuan)
Kipea-Kariri North Arawak
Kiwaian North Bahnaric
Kordofanian North Berber
Kru North Khoisan
Kwa North Kimberley
Lezgian Northeast Caucasian
Loinang- Banggai Northern Athabaskan
Lower Cross Northern Australian
Lower Sepik Northern Iroquoian
Lowland Amazonia Northern Je
506 Classifiers
Northern Palaungic Proto-Eyak-Athabaskan
Northern Tai Proto-Guaicuruan
Northwest Caucasian Proto-Gur
Northwest Coast Proto-Kru
Numic (Uto-Aztecan) Proto-Oceanic
Oceanic Proto-South Dravidian
Ok Proto-Uralic
Old Indo-Aryan Proto-Waic
Old Iranian Pueblo
Omotic Quechua
Otomanguean Reef-Santa Cruzan
Pacific Coast Athabaskan Ritwan (Algic)
Palaungic Romance
Paleosiberian Sahaptian
Pamir Saho-Afar
Pano Saliba-Piaroa
Pano-Tacana Salish
Papuan Samoyed
Papuan of Bougainville Semitic
Peba-Yagua Sepik
Penutian Sepik Hill
Peruvian Arawak Shugnan-Rushan
Piman Sino-Japanese
Platoid Sino-Tibetan
Polynesian Siouan
Pre-Andine Arawak Slavic
Pre-Proto-Northern-Iroquoian Solomon Islands
Proto-Afroasiatic South Amazon
Proto-Akan South American Indian
Proto-Arawa South Arawak
Proto-Arawak South Bahnaric
Proto-Athabaskan South Berber
Proto-Australian South Celebes
Proto-Bantu South Cushitic
Proto-Benue-Congo South Dravidian
Proto-Berber South Halmahera-West New
Proto-Bora-Muinane Guinea
Proto-Bora-Witoto South Slavic
Proto-Chadic South Slavonic
Proto-Cross-River Southeast Asian
Proto-Dravidian Southeast Solominic
Proto-East-Katuic Southeastern Iranian
Proto-Eastern Nilotic Southern Khoisan
 List of Language Families 507
Southern Lawa Voltaic
Southern Numic (Uto-Aztecan) Waic
Southern West Atlantic Wakashan
Tacana Wantoat
Tai Waris
Takic (Uto-Aztecan) West Atlantic
Teberan West Bahnaric
Tequislatecan West Chadic
Tibeto-Burman West Mindi
Ticuna West New Britain
Togo Remnant West Slavonic
Torricelli West Tucano
Totonacan Western Austronesian
Trobriand Islands Western Fijian
Tuareg, Berber Witoto
Tucano Xingu-Pareci
Tungus-Manchurian Xinguan Arawak
Tupi Yanomami
Tupi-Guarani Yuchian
Turkic Yukian
Ubangi Yuman
Ugric Zaparoan
Uralic Zapotecan (Otomanguean)
Uto-Aztecan Zoquean (Mixe-Zoquean)
Viet-Mutong
This page intentionally left blank
 Index of Languages, Linguistic Areas, and
Language Families

Abau 67, 77, 123, 277, 322 Arawa 4, 32, 34, 54, 58-9, 70-1, 76, 80, 100,
Abaza 40, 48 138, 246, 251-2, 257-9, 272, 371, 381,
Abelam 253 400, 434
Abkhaz 32-3, 40, 48 Arawak 2-4, 12,14, 39, 50, 54, 57,69, 76, 80,
Abu' Arapesh 23, 26, 262, 346 93, 100, 106, 123, 143, 150, 174, 183,
Acehnese 84, 91, 97, 190, 283, 286, 290, 381, 198, 201-2, 204-5, 210, 226-7, 230, 244,
402 246, 248, 256, 259, 262, 272, 279, 286,
Achagua 185, 198, 200, 202, 259, 286, 288, 288, 294, 360-1, 368-9, 384-7, 400
290, 292, 387 Archi 43, 47, 176
African 10, 19, 67-8, 70, 76-7, 101, 243,287, Armenian 77, 263
373, 381 Arvanitika 390
Afroasiatic 19, 54, 59, 77, 92, 121, 248, 252, Aslian 443
276-7, 288, 346 Asmat 171, 299
Ahaggar 254 Assamese 102, 105, 117, 260, 379
Ahtna 167, 169, 382 Atakapa 169
Aikana 80 Athabaskan 9, 11, 123, 152–4, 156, 165,
Ainu 121, 286, 436 167-9, 176, 185, 205, 250, 265, 296, 334,
Akan 280, 342, 398 336, 344, 355, 381, 437, 444
Akatek 88, 90-1, 98, 113-14, 117-19, 187-9, Australian 2, 9-10, 12, 23, 25, 31, 33-5,
252, 284-5, 287, 290, 293, 302, 357, 403 37-8, 41, 43, 49, 53-7, 59, 65-7, 70, 76,
Akhvakh 23 79, 81–4, 86, 88-92, 95, 97, 99-101, 147,
Akoye 79 150-1, 160-1, 191, 200, 228-9, 246-7,
Alamblak 27, 42, 56, 58, 276-7, 322 252, 256-7, 277-8, 280, 284-5, 296-9,
Alawa 257 318, 321, 328-9, 339, 342, 345, 353-5,
Albanian 396 357, 360-1, 366, 369, 372, 275, 378, 381,
Algonquian 79-80, 100, 121, 123, 154, 276, 383, 385, 391, 395-7, 399, 402, 407, 431,
297 438
Algonquian-Ritwan 123 Austroasiatic 97, 116, 121, 227, 260, 272,
Amazonian 7, 10, 51, 82, 95, 193, 266, 396 286, 291-2, 311, 342, 347, 354, 357,
Amele 437 365-6, 403-4, 443
American Indian 79, 125 Austronesian 28, 85, 100, 110, 113, 124-5,
American sign language 296 133, 142, 147, 171, 183, 204-5, 210, 244,
Amharic 262, 277, 279 267, 294, 303, 313, 339, 354, 357, 366,
Amuesha 80 382, 388, 402, 433
Ancient Egyptian 82 Awa Pit 86, 97, 381, 437
Andi 278, 377 Awabakal 228
Angan 79, 123, 218, 244, 346, 435 Awara 124, 207, 211, 218, 287, 418, 435
Angave 79, 123, 218, 244, 346 Ayacucho Quechua 48, 383, 388
Animere 381 Aymara 80
Anindilyakwa 5, 12, 35, 43, 57, 66, 101, 151, Aztec 123
160-1, 169, 191, 200, 202, 207, 247, 256,
280 Babungo 24, 57
Apalai 128-9, 142, 175, 295, 320, 336, 360, Badyaranke 61, 382, 388
365 Bagval 23
Apurina 57, 143 Bahnar 356, 443
Arabic 40, 121, 385, 388, 392 Bahwana 80, 107, 229, 259, 286, 354, 360,
Arapesh 26, 59, 62, 79, 262, 276, 346 380, 383
510 Index of Languages, Linguistic Areas, and Language Families
Baining 33, 79, 218, 244, 434 Caddo 444
Bainouk 59, 61-2, 435 Caddoan 169, 201-2, 444
Balinese 286 Cahuapanan 208
Baltic 245, 387 Cahuilla 143-4, 317, 343, 348, 365, 406
Balto-Finnic 440 Campa 191, 435
Banda 374 Camus 277
Banggais 404 Canela 129
Baniwa (of Icana) vii, 4, 10, 31, 37, 39, 58, Cantabrian Spanish 27, 45, 276-7, 409
69, 77, 100, 116, 131, 133–4, 136, 142-3, Cantonese 207, 214-15, 226-7, 326, 329, 419
145, 147, 185, 192, 201-2, 219, 227, Carib 11-12, 80, 127-9, 147, 172, 198, 259,
230-2, 234-5, 240-1, 246, 265, 272, 283, 295, 300, 354, 360, 365, 383, 438
287, 290, 294, 303, 310, 327, 336, 344, Carrier 167-8, 382
355, 370, 382, 384-5, 387, 401, 444-6 Caucasian 244, 252, 340
Banjalang 79, 395 Cayuga 402-3
Bantu 2, 5, 9, 22, 24, 30-3, 35-6, 38, 44, 50, Central American 80, 123
53, 57, 63, 65-6, 95, 100, 176, 24,8, 266, Central Bahnaric 356, 443
279, 281-3, 349, 359, 369-70, 373, 383, Central Bougainville 79
387-9, 398-9, 403, 407, 409, 413, Central Jukunoid 380
416-17, 422 Central Khoisan 246-7, 393
Banz 299 Central Porno 437
Barasano 70 Central Tucano 93, 219, 227, 333, 384, 386
Barbacoan 86, 97, 123, 437 Central Yup'ik Eskimo 181
Bare 4, 247, 266, 268, 370, 380, 387, 394, 400 Chacabano Zamboangueno 389
Bari 393 Chadic 358
Basque 436 Chamalal 32
Beludzhi 379 Chambri 123
Bengali 77, 102, 105, 249, 369, 379, 381, 431, Chamicuro 80
438-9 Chapacuran 257-9, 321, 434
Benue-Congo 24, 57, 59, 99, 101, 110, 124, Chapahuan 123
249, 356, 380-1, 398 Chayahuita 207-8, 227
Berber 41, 51, 59, 92, 252-3, 255-6, 279, 368, Chechen 27, 41, 49, 57, 263, 409
388 Chemakuan 79, 121
Biak 244 Chemehuevi 127, 295
Bihari 102 Chenapian 67, 123
Biloxi 169 Cherokee 11, 80, 161-3, 175, 297
Bilua 79 Chevak (Eskimo) 181
Binanderre 439 Chibchan 97, 123, 154, 156, 375
Bine 33, 79 ChiBemba 52, 279, 282, 409
Black Tai 227, 405 Chimbu 171, 299
Blackfoot 123 Chimila 123
Bodo 105 Chinantec 189, 191, 202, 257, 287, 292, 371,
Bolivian 286 376
Bora 221, 246-7, 287, 333, 385, 387, 397 Chinantepec Chinantec 293
Bora-Witoto 69, 80, 123, 220-2, 226, 246-7, Chinese 104, 118, 121, 205, 212, 226-7, 273,
287, 385, 435 354, 361, 389, 409, 417-18, 439
Bororo 129 Chipewyan 154, 249, 297-8
Boumaa Fijian 133-4, 136, 138, 140, 293 Chiricahua Apache 154, 250
Bowili 27 Chitimacha 169
Brazilian Portuguese 42, 266 Choco 436
Budukh 68 Chrau 357, 365, 443
Bugis 286, 315, 404, 406 Chuj 364
Bukiyip 62, 276 Chukchee 438
Burmese 100, 103, 260, 262, 267, 291-2, Classical Arabic 120
311-12, 315, 319, 324, 328-9, 336, 346, Colville 123
361, 442-3, 445 Comaltepec Chinantec 117-18, 257, 287
Burushaski 35, 77 Comanche 80, 436
Bwamu 287 Cora 127, 154, 191, 202, 295-7
 Index of Languages, Linguistic Areas, and Language Families 511
Costanoan 80 Enga 158-9, 166, 250, 296, 299, 406
Cree 297 Engan 158-9, 171, 299, 362
Creek 158, 251 English 21, 24, 26, 29, 36, 54, 86-7, 100,
Cross River 101, 110, 124, 287, 356, 379, 115-16, 140, 153, 176, 228, 326, 337-9,
398, 434 349-50, 358, 386, 389-90, 397, 407, 409,
Cuaiquer 437 415, 443
Cuiba 50, 221, 224 American English 346
Cushitic 28, 92, 276, 390 Old 380, 398
Czech 415 Pidgin 389
Standard 312
Daghestanian 60, 263 Tasmanian 312
Dahalo 390 Epena Pedee 436
Daju 398 Eskimo 177, 181-2, 266, 301, 345
Dakota 158, 169, 299 Estonian 387, 440
Daly 97, 283, 285 Ewe 440
Dangbon 385 Eyak 176, 206-7, 209, 225
Danish 243 Eyak-Athabaskan 121, 169, 206, 209
Dasenech 277 Eywo 79
Daw 12, 85, 139, 147, 172, 174-5, 192, 202,
259, 300, 345, 354, 357, 371, 376, 381 Fanagalo 389
Dena'ina, or Tanaina 167, 382 Felefita 346
Deni 70, 434 Fijian 2, 11,113, 141, 381
Dhegiha 169 Old 113
Dioi 215, 288, 405-6, 445-6 Standard 133
Diyari 23, 336 Finnish 369, 387, 439
Dogon 438 Finno-Ugric 102
Dogrib 250 Fly River 33, 79, 277. 437
'Dongo-ko 129-30, 147, 192, 202 Folopa, or Podopa 123
Dravidian 19, 22-3, 41, 46, 77, 99, 103, 121, French 19-20. 40, 53–4, 62, 326, 379, 399,
185, 202, 249, 276, 286-7, 384, 398 414-15
Dulong-Rawang 207, 215, 323 Colloquial 408
Dutch 26 Old 380
Dyirbal 23, 35-6, 41, 43, 45, 56, 79, 228, 281, Ful 59, 243
310-11, 329, 345, 347, 372-3, 390, 395, Fulfulde 344, 346
408, 438
Traditional 349, 390 Gaagudju 49, 70, 396
Young People's 349, 390-1 Gadjang 228
Gaelic 390
East Adamawa 358, 374 Gapapaiwa 147
East Asian 11, 82, 99, 101, 103, 121, 205-6, Garifuna, or Black Carib 34, 40, 44
325 Garo 310, 417-18
East Berber 243 Gaviao 86
East Cushitic 25, 51, 60, 328 German 24-5, 27, 39, 153, 243, 280, 335,
East New Britain 33, 79, 244, 434 337, 363, 387, 397, 409, 414-15
East Sepik 4, 24, 42, 54, 58, 62, 123, 253, Gilbertese 291
276-7, 322, 437 Godoberi 23, 276
East Slavic 244 Gola 58, 321
East Tucano 4, 50, 70, 110, 131, 218-19, Gold Palaung 366
221-2, 259, 287, 333, 361-2, 384 Gondi 266
Eastern Mindi 397 Gorontalo 286, 445-6
Eastern Nilotic 42, 58, 59, 77, 276-7, 354, Gran Choco 80
358, 368, 380, 392-3, 402, 403 Grassfields Bantu 24, 57, 124
Eastern Sudanic 21, 398 Grasslands Bantu 10, 287, 434
Eastern Sutherland Gaelic 390 Great Basin 123, 169
Egyptian Arabic 120, 415 Grebo 377, 417
Ejagham 99-100, 105, 111, 124, 249 Greek 19, 390
Emmi 84, 86, 97, 161, 257, 285, 299 Guahibo 36, 50, 80, 94, 123, 204, 220-2, 224
512 Index of Languages, Linguistic Areas, and Language Families
Guaicuruan 12, 80, 130, 147, 175, 177-8, 363, 378, 380, 388, 396, 398-9, 413, 415-
180-3, 191, 198, 251, 301, 326, 362-3 16
Guajiro 54 Indonesian, Bahasa Indonesia 283, 286, 288,
Gulf 80, 123 381, 386, 402, 404, 443, 46
Gunbarlang 200, 202, 381 Ingush 27, 41, 49, 57, 263
Gunwinjgu 44, 169 Iranian 77, 121, 185, 379
Gur 53, 59 Old 396
Gurr-goni 33, 38, 53-5, 66, 70, 100, 281, 408 Iraqw 26, 70
Guugu Yimidhirr 372 Irish, Modern 262
Gwich'in 169 Iroquoian 161, 169, 201-2, 243, 296–7, 402
Island Carib 259, 354, 360, 383
Haida 121, 169 Italian 19, 379, 399
Hakka 215, 226 Itonama 80
Halia 124 Iwaidja 37
Hani 156, 300 Iwam 123
Harakmbet 69, 80, 123, 171 Izayan 243, 253, 256, 440
Harar 25
Hausa 25, 388 Jabuti 80
Hawaiian 136 Jacaltec 82-3, 88, 225, 252, 284-7, 313, 323,
Hebrew 44, 52, 92, 138, 243, 254, 264, 331, 334, 341, 344, 357, 364, 375, 403
413-15 Modern 334
Hiechware 368 Japanese 2, 8, 21, 99, 106, 113, 121, 155,
Hindi 53, 102, 121 261-2, 288, 293, 308-10, 314, 316-17,
Hittite 82, 378 320, 324, 328-9, 334-6, 344, 348, 351,
Hixkaryana 175, 438 410, 417—20, 436, 439
Hmong 12, 104, 131-2, 144-5, 147, 183, Modern 410
205-7, 215-16, 225, 291-2, 329, 354, Jarawara 54, 59, 100, 251, 258, 262, 272, 439
356, 445 Javanese 386
Hohodene 143 Jawoyn 385
Hokan 169 Je 11, 23, 36, 80, 97, 147,228,244
Hokkien 215, 226, 419-20 Jeh 104
Holikachuk 169 Jingulu 34, 281
Hopi 80 Jivaro 80
Hua 438–40 Jukun 380
Hualapai 127
Huave 123 Kabyle 39, 256
Huli 299 Kadiweu 130, 175, 181, 198
Hungarian 102-4, 115-16, 121, 440 Kadugli-Kongo 277
Huon 171, 299 Kaingang 23, 36, 228
Hupa 123, 249 Kakua 80, 228, 384
Hupda 139, 147 Kalau Lagaw Ya, Kalaw Kawaw Ya 253, 336
Kaliai-Kove 134, 293
latmul 253 Kam-Muang 446
Ibibio 104 Kamoro 166, 299
Icelandic 415 Kana 99, 101, 110-11, 287, 317, 354, 356,
Idu 155 359-60, 371, 379, 403-4, 439, 444
Igbo 77, 440 Kanjobal of San Miguel Acatan 113
Ignaciano 69, 207, 210, 213, 227 Kanjobalan Mayan 82, 88, 90-1, 97-8, 123,
Ika 154, 156, 165, 171, 296, 299, 438 187, 202, 284, 286, 357, 376,404
Ilocano 313, 388 Kannada 286
Imonda 152, 166, 297, 300, 330, 362-3, 379, Kariri 11, 229
406, 439 Karo 171
Indie 77, 99, 103, 105, 121, 185, 202, 379 Karok 80, 123
Indo-Aryan 27, 102, 260, 287, 369, 384 Katcha 25, 277
Old 379 Kate 299
Indo-European 3, 19, 24, 31, 54, 59, 77, 86, Katu 442
121, 243, 248, 252, 255-6, 262-3, 280, Katuic 442
 Index of Languages, Linguistic Areas, and Language Families 513
Kayapo 129 Landuma 61, 435
Kegboid 101, 110-11, 124, 287, 371, 379, Lao 89, 103, 222, 225, 268, 354, 360-1, 439
381, 439 Latin 19, 40, 47, 53, 255-6, 262, 336, 379-80
Kenya Pidgin Swahili 388 Latvian 369, 387
Keriaka 79 Lavukaleve 79, 244, 257a
Ket 23, 42, 77 Lezgian 60, 263, 438
Kewa 299 Lezgic 263
Khinalug 47, 248 Lii 445
Khmer 104, 117, 288, 347, 405, 443, 446 Limilngan 25, 59, 381
Khmu 403 Lingala 388, 400
Khmuic 272 Kinshasa 388, 399-400
Khoe 368, 393 Mankandza 399-400
Khoisan 42, 59, 77, 254, 276 Lingua Geral 390
Kibera 388 Lisu 288
Kikongo 30, 389, 400 Lithuanian 245, 255, 262, 387
Kikuyu 57, 63-4, 282 Lokono 50, 54, 172, 174-5, 201-2, 262, 279,
Kilivila 8, 104, 110, 136-7, 141, 191, 204, 300
207, 210, 223, 225, 260, 294-5, 303–4, Longgu 382
316, 331, 366, 376, 411, 417-19, 445 Loniu 124
Kiowa 80, 248, 250 Lotuko 384
Kiowa-Tanoan 11, 80, 248, 250, 369, 377 Loven 367
Kipea 11, 133, 136, 147, 294, 434 Lower Chinook 79
Kipea-Kariri 133, 135, 147, 229 Lower Cross 396, 398-9
Kituba 388, 400 Lower Sepik 27, 58, 79, 123
Kiwai 299 Lowland Amazonia 123, 171, 216, 222
Kiwaian 171, 299 Lowland Amazonian 7, 36, 93, 99, 201
Koaia 80 Lowland East Cushitic 385
Koasati 158, 169, 436 Luganda 52
Kobon 436 Luiseno 127, 295
Kolami 23, 41, 67, 287
Korafe 439 Ma 276
Kordofanian 25 Maa 384
Kore 380 Maasai 375
Korean 106-7, 113, 115, 120-1, 260-1, 273, Maasina 346, 392
288, 293, 311, 319, 334, 336, 348, 351, Machiguenga 12, 191, 199, 202, 207, 213,
386 341, 370
Kosraean 294 Macro-Je 11, 36, 129, 228, 259, 434
Koyukon 157, 167, 176, 209, 250, 265, 336, Macushi 128-9, 140, 142, 175
382 Magadhan 102
Kraho 129 Maka 80, 253
Kru 377, 398-9, 417 Makassar 286, 406
Kulina 434 Maku 12, 80, 85, 97, 139, 147, 172, 174, 192,
Ku Waru 159, 166, 362 228, 259, 300, 354, 357, 371, 384, 435
Kubeo 207, 219, 226, 382, 384-5 Mai 101, 104, 444
Kugu-Ngancara 90 Malak-Malak 70
Kuot 32-3, 79 Malay 249, 286, 308, 324-5, 329, 381, 386
Kuria 279, 407, 409 Malinke 382, 388
Kurripaco 143 Malto 23, 102, 104, 112, 185, 206, 248, 286
Kusaiean 108 Mam 87, 140, 252, 284, 331, 357, 362, 364,
Kwa 398, 440 375-6, 403
Kwoma/Washkuk 77 Manam 134, 137, 293, 365
Kxoe 42, 368 Manambu 4, 32, 37, 42, 45, 54, 58, 243-4,
253, 263–4, 276-8, 313, 358, 392, 438
Lahu 361 Mandan 3, 177, 326, 363
Lak 34, 278, 347 Mandarin (Chinese) 98-9, 183, 206-7, 309,
Lama 53 324-5, 328, 335-6, 409-10, 418-23
Lama Lama 92 Mande 77, 382, 388
514 Index of Languages, Linguistic Areas, and Language Families
Marathi 102, 106, 121, 287 Mimduruku 12, 123, 152, 160-1, 171, 205,
Maricopa 127 207, 216-18, 224-5, 227, 333, 354-5,
Marind 60, 77 444-5
Marrithiyel 37, 97 Mupun 358, 402
Marshallese 294 Murrinhpatha 37, 84, 86, 89-90, 97, 284-5
Mataguayo 253 Murui Witoto 246
Maung 65, 267 Muskogean 158, 169, 251, 436
Mayali 31, 38, 41, 53, 55, 92, 150–1, 169,
266, 281-1, 296, 354-5, 360, 380, 385, Nadeb 147, 259, 354, 371
396, 408, 442 Nahuatl 123
Mayan 84, 87, 90-1, 97, 103, 113, 115, 123, Naiki 41, 67
140, 252, 285, 288-9, 329-30, 339, 342, Nakh 41, 47
353–4, 357, 364, 367, 375, 438 Nakh-Daghestanian 27
Mba 33, 67, 75-6, 129-30, 147, 192, 243, Nakhi 362
259, 387 Nama 254
Melanesian 110 Nambiquara 93, 123, 204, 221, 224-5,
Melpa 171, 299 303-4, 333-4, 345
Menomini 123 Napues, or Kunua 79
Mescalero Apache 154-5, 157, 330 Naro 368
Mesoamerican 12, 81-2, 97, 101, 153, 169, Nasioi 12, 79, 131-2, 139, 162, 176, 207,
335 219-20, 259
Mexican 123 Natchez 169
Miao 132, 144 Nauru 99, 110, 205
Miao of Wei Ning 104 Navajo 167, 250, 336
Miao-Yao 12, 131, 147, 212, 215, 435 Ndali 63-5
Micronesian 108, 110, 125, 136, 140-1, 147, Ndu 4, 24, 32, 42, 54, 58, 77, 243-4, 253,
186-7, 205, 260, 291, 294, 361, 365-6, 278, 392
405, 446 Ndunga-le 383, 387
Minangkabau 10, 84-5, 90, 97-8, 100, 117, Nepali 102, 379
120, 187, 189, 202, 253, 262, 272, 283, Nevome 158, 169, 297, 299
285-6, 289, 291-3, 302, 308, 318, 322, New Caledonia 142
329-30, 348, 354, 356, 359-60, 366, 370, New Guinea Highlands 158
375, 381, 386, 391, 401-2, 404, 406, 443, New Indo-Aryan 379
445-6 Newari 9, 207, 212, 265, 326
Mindi 397 Nez Perce 123
Missima 147 Ngala (Nilo-Saharan) 374
Miwok 80 Ngala (Papuan) 253
Mixtec 252, 260, 354, 359, 376 Ngalakan 33, 49, 56
Mnong Gar 357 Ngan'gityemerri 84, 89, 92-3, 95, 97, 185-6,
Mocovi 130 201-2, 285, 301, 317, 354, 361, 370, 376,
Modern Hebrew 22, 28, 53, 248, 264 394, 404, 407
Modern Irish 262 Ngandi 169, 246, 296, 373, 387, 402, 442
Mohawk 403 Ngiyambaa 9
Mokilese 11, 133, 186-7, 202, 301 Ngyembccn 124, 287, 356
Mon 272, 443 Nias 112
Mon-Khmer 101, 104, 365, 402, 442-3, 446 Nicobarese 443
Monumbo 59 Niger-Congo 33, 58-9, 67, 75-7, 99, 104,
Mori 286 129, 147, 192, 259, 276, 342, 356, 359,
Morobe Province 79, 124, 211, 287 380, 384, 387, 438
Motuna 12, 34, 40, 69, 79, 131-2, 162, 176, Nilo-Saharan 51, 77
207, 219-21, 248, 279, 303, 321, 370, 437 Nivkh, or Gilyak 100, 108-9, 121, 206, 274,
Movima 80, 110, 286 286, 290
Mparntwe Arrente 82, 84, 88 Non-Austronesian, also see Papuan 32-4,
Mufian 62 79, 218
Muhiang Arapesh 26 Non-Bantu 38
Mulao 9 Nootka 249
Munda 77 North Amazonian 77, 220, 226, 327, 344
 Index of Languages, Linguistic Areas, and Language Families 515
North American Indian 11, 97, 105, 147, 247, 257, 259, 264-5, 268, 278-9,
154, 158, 177, 182, 296 289-90, 292, 296-7, 300, 302, 310, 316,
North Arawak vii, 10, 23, 27, 31, 33-4, 36, 320, 327, 333, 335-7, 341, 354-5, 357,
39–40, 44, 50, 58, 60, 69, 92-3, 100, 360, 369-70, 375, 383, 434, 445-6
106-7, 109-10, 116, 131, 133, 142-3, Pamir 60
147, 152, 163, 171-2, 192, 198, 203-4, Panara 129
220-3, 235, 246-7, 253, 259, 265-8, 340, Panare 127-8, 360
371, 376, 380, 383, 385-7, 390, 394 Pano 80
North Bahnaric 443–4 Papago 127
North Berber 39, 243, 254 Papuan 3, 10-12, 23, 38, 59, 67, 69, 77, 79,
North Khoisan 58 97, 123, 131-2, 139, 144, 147, 149, 152,
North Kimberley 34-5 158-9, 162, 166, 169, 205, 211, 218-19,
Northeast Caucasian 23, 32, 34, 40, 43, 47, 243–1, 248, 257, 259, 276-7, 279,
60, 77, 176, 263, 276, 278, 347, 377 296-300, 303, 321-2, 329, 362, 370, 379,
Northern Athabaskan 157, 167, 169, 205 382, 384, 396, 433, 435-7, 439
Northern Australian 44, 53-4, 95, 161, 169, Pareci 226, 435
185, 280, 283, 322, 372 Parji 23, 41, 67
Northern Iroquoian 401 Patjtjamalh 84, 97
Northern Je 129 Paumari 4, 32, 34, 70-6, 100, 138, 146, 184,
Northern Palaungic 442 246, 252, 257, 283, 381, 400
Northern Tai 227 Peba-Yagua 80, 123, 221
Northern Tepehuan 127 Persian 77, 121, 369, 379
Northwest Caucasian 32-3, 48, 77 Peruvian Arawak 191, 199, 213, 341
Northwest Coast 79, 121 Petals 124
Ntifa 254 Piapoco 387
Nubi 388 Pilaga 12, 86, 178, 180-2, 251, 266, 301, 326,
Numic (Uto-Aztecan) 80 331, 336
Nung 100, 103-4, 118, 207, 211, 227, 405 Piman 169
Nungali 65-6, 381, 397 Piraha 80
Nunggubuyu 33, 59, 66, 92, 150, 169, 200, Piratapuya 4
202, 321, 325, 387 Pittapitta 228
Nyanja 65 Platoid 380
Pogoro 407
Ocaina 385, 397 Polish 244, 279, 313, 414-15, 440
Oceanic 6, 11, 97, 121, 124, 133-7, 143, 145, Polynesian 135-6
147, 294, 381-2, 384 Pomoan 79, 169
Ojibway 121, 154-5, 297 Ponapean 11, 186, 202, 260-1, 294, 301,
Ok 77, 171, 277 336, 342, 344, 354, 359, 361-2, 364, 366,
Old Church Slavonic 244 405
Olgolo 91, 373 Ponca 177-8, 182
Ollari 23 Portuguese 2, 25, 28, 31, 30, 43, 45, 47, 52-3,
Olo 59, 277 57, 62, 75, 100, 109, 243, 313, 358, 379,
Omani Arabic 120, 443 391, 400
Omani-Zanzibar dialect of Arabic 120 Potawatomi 123
Omotic 52 Pre-Andine Arawak 143
Ongamo 375 Pre-Proto-Northern-Iroquoian 378
Oriya 102, 379 Proto-Afroasiatic 92
Oromo 25, 28, 92, 276-7, 279, 328 Proto-Akan 342
'Oro Nao 257-8, 321 Proto-Arawa 60
Ossete 379 Proto-Arawak 70, 143, 371, 387, 446
Otomanguean 80, 117-18, 123 Proto-Athabaskan 250
Proto-Australian 372
Pacific Coast Athabaskan 80 Proto-Bantu 24, 283
Palaungic 82, 402 Proto-Benue-Congo 111, 358, 381
Paleosiberian 77, 100, 108, 121, 206 Proto-Berber 256
Palikur 3, 8, 12, 27, 39, 69, 100, 112, 142, Proto-Bora-Muinane 387
163-5, 172-6, 192-203, 225-6, 228-9, Proto-Bora-Witoto 246
516 Index of Languages, Linguistic Areas, and Language Families
Proto-Chadic 358 Sepik Hill 77
Proto-Cross-River 379 Serbo-Croatian 415
Proto-Dravidian 378 Sesotho 57, 416, 418
Proto-East-Katuic 442 Setswana 50, 349, 416
Proto-Eastern Nilotic 358, 375 Shan 406
Proto-Eyak-Athabaskan 154, 169, 444 Shilh 256
Proto-Guaicuruan 181, 363 Shona 32, 65, 343, 407
Proto-Gur 388 Siamese 404-6
Proto-Kru 377 Sierra Popoluca 123
Proto-Oceanic 124, 134, 365 Sinasina 299
Proto-South-Dravidian 378 Sino-Japanese 113, 336
Proto-Uralic 440 Sino-Korean 113
Proto-Waic 402, 442 Sino-Tibetan 105, 361-2
Pueblo 80, 123 Siouan 3, 7, 12, 80, 158, 169, 177, 181-3,
Puluwat 140-1, 145, 295 276, 299, 301, 326, 362-3
Punjabi 27 SiSwati 416
Siuci 143
Qafar 25, 51 Siwa 243, 253
Qiang 155, 175 Slave 154, 167, 169, 250, 297, 437
Quechua 80, 382 Slavic 48, 56, 245, 377, 388, 439
Queyu 155 Slovene 244-6, 380
Sochiapan Chinantec 12, 191, 201, 287, 290,
Raga 135, 294 293
Rama 375 Solomon Islands 79, 257
Reef-Santa Cruzan 79, 147, 171, 218, 386, South Amazon 58
435 South American vii, 2, 10-12, 67, 76, 100,
Rembarrnga 381, 385 123, 129, 144-5, 147, 204-5, 244, 247,
Rendille 60 274, 295, 300, 318, 334, 357, 361, 433
Resigaro 50, 69, 226, 246-7, 333, 385, 387, South American Indian 6, 11-12, 59, 99,
397 105, 133, 147, 149, 172, 213
Retuara 386 South Arawak 152, 160, 209-10, 213, 435
Rikbaktsa 36, 228 South Bahnaric 357, 443
Ritwan 123 South Berber 254
Ro-lo-m 357 South Cushitic 26, 70
Romance 57, 379-80 South Dravidian 67, 99-100, 102, 104, 112,
Romansch 47 185, 206, 248, 266, 378
Roshani 60 South Halmahera 244
Rotokas 79 South Slavic 244
Rumanian 45-6 Southeast Asian 11, 82, 90, 99, 101, 103,
Russian 22, 25-6, 28, 38, 41, 43-4, 48, 51, 54, 121, 205-6, 211, 325, 354, 356, 361, 402,
57, 115, 120, 138, 140, 153, 243, 256, 433, 435, 443
262–4, 272, 313, 347, 363, 389, 409, Southeast Solomonic 244
414-15, 439–0 Southeastern Iranian 60
Southeastern Pomo 299
Sahaptin 123 Southern Khoisan 35, 396
Saho-Afar 25 Southern Lawa 402
Saliba 80, 123 Southern Numic (Uto-Aztecan) 127, 295
Salish 79, 109, 114, 212, 123, 169 Southern Paiute 80
Sanskrit 256 Southern West Atlantic 368
Savosavo 79 Southwest Mexico 163, 169
Scottish Gaelic 390, 399 Spanish 27, 45, 47-8, 52-3, 110, 249, 253–4,
Sedang 444 378, 380, 382, 388-9, 414-15, 437
Seghrouchen 440 Squamish 109, 114, 205, 286
Sele Fara, dialect of Slovene 380 Swahili 19, 31, 34-5, 38, 44, 53, 63, 228, 266,
Semitic 51, 252, 392, 413 390, 400
Seneca 243 Zairian 388
Sepik 26, 384 Swedish 40
 Index of Languages, Linguistic Areas, and Language Families 517
Tabar 382 Totonac 123, 288-9, 354-6, 402, 443-6
Tacana 80 Totonacan-Tepehuan 288, 355, 443
Taenae 79, 123, 218 Trobriand Islands 204
Tagalog 48 Trumai 80
Tai 9,97, 100,103–4,211,286,290,311,336, Truquese 110, 136, 141, 145, 187, 202, 267,
370, 405-6, 442, 444, 446 320, 361, 370
Taiwanese 423 Tsafiki 123
Takia 147, 382 Tsez 40, 176
Takic (Uto-Aztecan) 143 Tsimshian 121
Tamachek 245, 254, 256 Tsova-Tush 41, 47
Tamambo 294 Tuareg 245, 254
Tamazight 39, 440 Tucano 4, 58, 69., 80, 93, 100, 110, 123,
Tamil 22, 56, 248 204-5, 207, 218-22, 224-6, 235, 248-9,
Tanana 167, 382 255, 259, 264, 358, 376, 382, 384-6, 397,
Tarascan 163, 169, 444-5 435
Tariana vii, 2, 4, 8, 33, 36, 50, 58, 69, 77, Tunica 80, 169
92-4, 100, 110, 1310, 138, 142, 163, 183, Tupi 12, 80, 86, 123, 152, 160, 171, 204, 216,
185, 201, 204, 220-6, 230, 235-6, 355, 435
239-41, 246-7, 249, 253–4, 259, 264-5, Tupi-Guarani 11, 129, 147, 259, 360, 391
268, 287, 327-9, 331, 354, 360-2, 370, Turkana 42-3, 58, 277, 322, 368-9, 384
376, 382, 384-5, 387, 396-7, 402, 404, Turkic 99, 102-3, 121
444-5 Turkish 341, 388, 440
Tashkent Uzbek 102 Tuyuca 93, 207, 218-19, 221
Taulil 79, 218 Twi 398
Tazerwalt Shilh 254 Tzeltal 103, 115, 272, 274, 288-9, 291, 293,
Teberan 123 362, 364, 367
Telugu 46, 100, 108, 112, 286 Tzotzil 249, 288, 293, 329–30, 356, 367
Teop 124, 183
Tepetotutla Chinantec 257 Ubangi 33, 75-6, 129, 147, 192, 243, 259,
Tequislatec 80 276, 387
Terena 80, 152, 160-1, 207, 209-10, 220 Ubangian 358
Teso 384 Ungarinjin 9, 34-5, 56, 257, 329, 396
Tewa 250-1 Upper Chinook 79, 123
Thai 100, 103-5, 118, 205, 207, 212, 214, Upper Sorbian 244
226-7, 260, 262, 268, 313-15, 318, 326, Uralic 121, 440
334, 348-9, 354, 360-1, 372, 404-5, Urubu-Kaapor 129
407-8, 411, 417-22, 442, 444, 446 Uto-Aztecan 11, 80, 123, 126-7, 129, 143-4,
Tibetan 82, 97, 284 147, 154, 158, 169, 259, 295, 297, 299,
Tibeto-Burman 9, 121, 154-6, 175, 212, 215, 348, 365, 436
265, 288, 300, 323 Uzbek 102, 104
Ticuna 80
Timbira 129 Viet-Muong 354, 360
Tinrin 142 Vietnamese 103-4, 117-18, 120, 205, 207,
Tiwi 33, 56, 59, 150, 160, 169, 200, 257, 211-12, 262, 317-18, 326, 329, 334-6,
276-7, 355, 389, 400 403, 405, 443-5
Modern 389, 391 Voltaic 59, 367
Traditional 389, 391, 400
Tlingit 121, 249 Wagaya 54
To'aba'ita 113 Waic 405
Toba 130, 178-81, 191, 202, 266, 301, 345 Wakashan 121, 169
Toba Batak 286, 445 Wambaya 281
Togo Remnant 27, 59, 381 Wangkumara 54
Tolai 136 Wantoat 124, 218, 287, 435
Tonga 407 Waorani 80, 123, 171, 207, 217-18, 221,
Tongan 136, 176 224-5, 227
Toposa 393 Wara 79, 277, 437
Torricelli 26, 59, 79, 277, 346 Warao 80
518 Index of Languages, Linguistic Areas, and Language Families
Wardaman 57, 92, 201-2, 322, 325, 373, 408 Xingu-Pareci 226
Warekena 4, 60, 100, 109, 143, 185, 247, 266, Xinguan Arawak 76, 182, 209-10, 213, 220,
268, 287, 390, 394, 400 228
Wari' 32 !Xoo 35, 396
Wans 3, 149, 152, 159, 166-7, 171, 184, 250, !Xu 58
300, 362-3,438
Warndarang 387
Yagua 69, 80, 107, 171, 204-5, 207, 217-18,
Warray 70, 92, 321, 328, 369, 399
220-2, 224-5, 333
Washo 123 Yaigin 252
Waura 76, 182-3, 207, 210, 213, 220, 226-8,
Yana 80
435
Yandruwanhtha 228
Wayan 11, 133
Yankuntjatjara 82, 88
Wei Ning 104, 208
Yanomami 80, 354-5, 444-5
West Atlantic 35, 58-9, 61, 321, 348, 373,
Yanyuwa 33, 65-6, 373, 378, 385
382, 435
Yaruro 80
West Bahnaric 367, 403
Yavapai 127
West Chadic 358 Yawalapiti 207, 209-10, 228, 435
West Flemish 34
Yessan-Mayo 437
West Mindi 397 Yidiny 2, 83-5, 87, 89, 91, 93, 97, 150, 284-5,
West New Guinea 244
317, 320, 345, 372, 404
West Slavic 244
Yimas 26, 59, 62, 79, 276, 329
West Tucano 222 Yonggom 77, 277
Western Apache 250, 298
Yoruba 77
Western Austronesian 82, 84-5, 91, 97, 112,
Yuchi 169, 177-8, 181, 183, 326
124, 187, 189, 253, 283, 292, 405, 446
Yuchian 80
Western Fijian 11, 133
Yucuna 106, 185, 386-7
Western Oromo 385
Yue 215
Western Torres Strait 23, 56, 253
Yuki 80, 249
White Hmong 325
Yukian 80
White Tai 215, 402, 405, 446 Yuman 11, 126, 129, 147, 259, 354
Wipi 79, 437 Yurok 123
Witoto 94
Wiyot 123
Wogamusin 67, 123 Zande 58, 276, 354, 358, 373-5, 402
Wolof 62, 348 Zaparoan 80
Worrorra 34-5 Zapotec 123
Wu-Ming 405, 445 Zayse 52
Wunambal 34-5 Zemmur 243, 256, 440
Zezuru 65
Xamatauteri 355 Zoquean 123
Xhosa 53 Zulu 36, 417
 Index of Authors

Abbott, M. 129 Bascom, B. 127

Adams, K. L. 6, 11, 82, 97-9, 101, 103, Basso, K. H. 11
116-18, 120, 271-2, 286, 288, 291-2, Becker, A. J. 117, 120, 311-12, 315, 319, 324,
310, 337, 342, 347, 357, 360, 365-7, 329, 442-3
402-3, 405-6, 442-6 Beckwith, C. I. 102, 116, 121, 440
agg-albostan ag-Sidiyan, E. 245, 254 Benton, R. A. 103, 110, 141, 187, 267, 320,
Ahrens, K. 115,423 361, 370
Aikhenvald, A. Y. 8, 11-12, 22, 27-8, 30, 32, Berlin, B. 103, 115, 272, 274, 289, 291, 364,
34, 37, 39-40, 50, 54, 60, 67, 69-70, 74, 367
76-7, 80, 92-4, 109, 112, 116, 123, 131, Berman, R. 138, 414
138, 142-3, 150, 163, 166, 172-4, 182, Bhaskararao, P. 9, 212, 265, 326
192-3, 196, 198-9, 201, 209-10, 213, Birk, D. B. W. 70
225-6, 228-30, 235, 239-40, 242-3, Bisang, W. 11-12, 21, 101, 103-4, 106,
246-7, 255-7, 259, 263-4, 266, 270, 276, 116-17, 120, 131-2, 144, 208, 212,
278-9, 320, 327-9, 332-3, 345, 355, 215-16, 260, 291, 325-6, 356, 372,
360-1, 368-9, 371, 376, 380, 383, 385, 383-4, 409, 439, 445
387, 390-1, 398, 400 Blankenship, B. 11, 161-2, 175, 297
Aksenova, I. 279, 407 Blewett, K. 79, 147, 218
Alexandra, P. 388, 400 Bloomfield, L. 4
Alexeyev, M. 263 Boas, F. 158, 299
Allan, K. 6, 12-13, 117, 172, 271, 274, Bohme, R. 415
337-8 Bokamba, E. G. 53, 388, 399-400
Allin, T. 246, 333, 385, 387 Boley, P. 70
Alpher, B. 54, 349 Bolt, J. E. 66
Amaya, M. T. 123 Borneto, S. C. 153, 363
Amberber, M. 262, 279 Boxall, E. 124
Ameka, F. 440 Braun, F. 341, 388
Anderson, C. 123 Breedveld, J. O. 344, 346
Anderson, J. L. 117-18, 257, 287 Breen, G. 54
Anderson, L. G. 349, 389 Bresnan, J. 29, 34, 176, 392
Anderson, N. 123 Broschart, J. 176
Anderson, S. R. 29-30, 363 Brosman, Jr., P. W. 378
Andrews, E. 50 Brown, L. 112
Arin, D. N. 343, 407 Brown, R. 3, 11, 166, 171, 300, 438
Aronoff, M. 26, 62, 262 Bruce, L. 27, 42, 56, 276, 322
Aschmann, R. P. 246, 387 Bulygina, T. V. 51, 54
Asher, R. E. 22, 248 Burgess, E. 226
Austin, P. 336 Burling, R. 103, 116, 291, 417-18
Axelrod, M. 157, 167, 265 Burton, M. 282
Butler, N. 152, 160, 209
Baarle, P. van 279 Bybee, J. 374, 376
Bani, E. 23, 56, 336
Bantia, T. K. 27 Callister, B. 147
Barlow, M. 29, 39 Campbell, L. 79, 121, 123, 169, 364, 383-4,
Barnes,J. 11, 93, 218-19, 333 388, 391, 438
Baron, R. 338 Capell, A. 267
Barron, R. 3, 7, 12, 176-8, 301, 326, 363 Carlson, R. 11, 86, 125, 127, 133, 140, 186,
Barsalou, L. W. 309 360
Barz, R. K. 11, 102, 117, 381 Carlson, T. 79, 123
520 Index of Authors
Carpenter, K. 103, 314, 404, 407-8, 417-18, De Leon, M. de L. P. 249, 252, 260, 329-31,
420-1, 424, 442, 444, 446 356, 360, 367, 376, 406
Carroll, J. B. 339 DeBray, R. G. A. 244
Carter, R. M. 11, 154, 249-50, 296-8, 336 DeLancey, S. 82, 86, 97, 214, 284, 286, 370,
Casad, E. 127, 154, 295-6 372
Casagrande, J. B. 339 Deloria, E. 158, 299
Caskey, A. F. 53 Demuth, K. A. 57, 398, 416-18
Ceria, V. G. 12, 130, 181, 363 Denny, J. P. 6, 24, 57, 100, 106, 121, 154, 181,
Chafe, W. L. 243, 321, 378 196, 271, 273, 282-3, 289, 291-2, 297-8,
Chang-Smith, M. 309 301, 316, 318-19, 336, 340, 345
Chapman, A. 89, 268, 360, 439 Derbyshire, D. C. vii, 6-7, 11, 34, 71, 80, 99,
Chapman, S. 34, 71-2, 138, 252 107, 123, 171, 175, 199, 208, 217, 227,
Chappell, H. 125, 258, 294 252, 438
Charney, J. O. 436 Diakonoff, I. M. 92, 346
Charters, H. 208 Dickinson, C. 123
Chatterji, S. K. 379 Diller, A. V. N. 11, 102, 117, 347, 349, 381
Childs, T. 368, 373 Dimmendaal, G. I. 22, 42, 58, 77, 100, 111,
Christensen, S. 77, 171, 277 322, 368, 371, 380, 390
Churchward, C. M. 113, 136, 381 Dixon, R. M. W. 2, 4, 6, 9-12, 23, 33-4,
Clamons, C. R. 25, 28, 92, 276, 385 43-1, 50, 54, 56, 59-60, 67, 79-81, 83-5,
Clark, E. V. 338, 421 87, 91, 97, 99-100, 113, 116, 123, 133-4,
Claudi, U. 358-9, 374-5 136, 140, 150, 164, 169, 228, 242-3, 247,
Codrington, R. 11, 133 251, 255, 258, 262-4, 266, 270-2, 285,
Cohen, P. 104 293, 310, 329, 340, 345, 347, 366, 372-3,
Coleman, C. 200 379-1, 383, 385, 395, 397-8, 404, 437-8
Collinder, B. 440 Donaldson, T. 9
Comrie, B. 40, 77, 176, 272, 347, 377 Dondorp, A. 79, 437
Conklin, N. F. 6, 11, 97, 99, 120, 215, 227, Doneux, J. 62
271, 286, 288, 290-2, 310-11, 318, 326, Dorian, N. 390-1
336-7, 381, 386, 402, 404-6, 442, 445-6 Downing, P. 8, 11, 106, 113, 120, 196, 288,
Connelly, M. J. 22 293, 308, 310, 314, 316-17, 320, 324,
Conrad, J. 123 329-30, 334, 336, 351,410-11
Conrad, R. J. 10, 26, 62, 79, 123, 346 Drabbe, P. 60, 299
Cooke, J. R. 262 Dryer, M. S. 104
Corbett, G. 6, 8, 10, 20, 22-3, 25, 28-9, 32-5, Ducos, G. 77, 382, 388
37, 39-41, 45-8, 51-1, 56, 67, 244, 253, Dul'son, A. P. 23, 42
278, 369, 375, 377-8, 392, 408, 415, Dunn, M. 439
439 Durie, M. 29, 91, 97, 190, 286, 290, 381, 402
Corominas, J. 253
Costenla Umana, A. 80, 123, 169, 384 Ekdahl, M. 152, 160, 209
Counts, D. L. 134 Elbert, S. H. 140-1
Craig, C. G. vii, 6-8, 10-14, 67, 81-4, 87-91, Emeneau, M. B. 100, 102, 106, 112, 121, 185,
97, 103, 110, 116, 125-6, 176, 206, 271, 286-7, 384
284-7, 302, 307, 310, 313, 323, 331, 334, England, N. 87, 140, 331, 357, 375-6, 438
341-2, 344, 357, 364, 370, 375, 433 Erbaugh, M. 324, 340, 409-10, 417-18, 421
Creider, C. A. 24, 57, 282 Evans, N. 8, 30-1, 33, 38, 41, 44, 49, 53, 55,
Croft, W. 7-8, 12, 24, 50-1, 125, 158, 180, 65-6, 92, 150-1, 200, 266, 279-80, 296,
189, 271, 274, 278, 286-7, 289, 297 355, 360, 380, 383, 385, 408
Crofts, M. 216 Everett, D. L. 32, 80, 257-8, 321
Crowley, T. 252, 395
Curnow, T. 86, 97, 437 Facundes, S. 57, 143
Faraclas, N. G. 396, 398-9
Daley, K. A. C. 325-6, 329, 334 Farr, C. 439
Dallet, J. 39 Ferguson, C. A. 29
Davidson, W. 11 Fleischmann, L. 33, 79
Davies, J. 436 Foley, W. A. 10, 26, 59-60, 62, 79, 158, 166,
Davis, D. R. 124, 287-8 171, 329, 363-4, 384, 437
 Index of Authors 521
Ford, K. 253 Haviland, J. 372
Ford, L. J. 83-4, 86, 90, 97, 161, 257, 285, Hayward, R. J. 51-2
299 Heath, J. 59, 92, 150, 200, 247, 296, 321, 373,
Foris, D. 12, 191, 287, 290, 371, 375 383, 386-7, 402
Fortune, E. F. 26, 62, 79 Heine, B. 10, 25-7, 37-8, 42, 58-60, 68, 70,
Frajzyngier, Z. 358 75, 77, 125, 134, 146, 243, 276, 321,
Frank, P. 154, 156, 171, 296, 300, 438 358-9, 368-70, 372, 374-5, 381, 384,
Franklin, G. 139 388, 392-3, 398, 401-3
Franklin, K. 299 Henry, D. 336
Fraser, N. M. 52 Henry, K. 336
Frawley, W. 14, 271-4, 288-9, 338, 437 Herbert, R. K. 36, 416-17, 422
Friedrich, P. 163, 169 Hewitt, G. 32-3, 48
Hill, D. 382
Gagne, R. C. 181 Himmelmann, N. P. 363
Gandour, J. 417, 419-20, 422-3 Hinch, H. E. 267
Garibian, A. S. 263 Hoijer, H. 11, 154, 250
Geeraerts, D. 316 Holmquist, J. C. 27, 45
Gerzenstein, A. 253 Hopkins, N. A. 123, 364
Givon, T. 29, 52-3, 176, 369, 392-3 Hopper, P. J. 249, 324-5, 329, 374
Goddard, C. 82, 88, 296, 316 Hu, Q. 419
Goethe, J. W. 409 Huffman, F. E. 116
Gomez-Imbert, E. 219, 227, 384-6 Hundius, H. 103, 213-14, 227, 288, 334
Gonpalves, C. H. R. C. 12, 152, 160, 216, Hurd, C. 12, 132, 139, 162, 176, 219, 259
333, 355, 444-5 Hyman, L. M. 10
Goral, D. R. 11, 118, 211, 317, 361
Green, D. 8, 27, 39, 69, 112, 142, 150, 163, Ikoro, S. M. 101, 110-11, 124, 287, 356,
165-6, 172-3, 192–4, 196, 198-9, 225, 359-60, 371, 381, 403-4, 439-40
229, 247, 257, 320, 327, 333, 355, 370, 383 Irvine, J. 348
Green, H. 165 Iturrioz Leza, J. L. 436
Green, I. 37, 97, 397
Green, R. 33, 38-9, 53-4, 56, 70, 100, 281, Jackson, E. 213, 220
408 Jaisser, A. 215
Greenberg, J. H. 6, 37, 50-1, 54, 100, 104-5, Jakobson, R. O. 51, 422
120, 244, 249, 254, 318, 321, 353, 367-9, Janson, T. 349, 389
373 Jarvinen, L. 369
Griffiths, C. 130, 175, 198 Jauncey, D. 294
Grinevald, C., also see Craig, C. G. 286 Jensen, J. T. 97
Gumilev, N. S. 439 Johnson, S. 90
Guoqiao, Z. 9 Jones, P. 70
Gwynn, J. P. L. 46, 108 Jones, Jr., R. B. 104
Jones, W. 70
Haas, M. R. 9, 158, 161, 169, 249 Joshi, S. K. 9, 212, 265, 326
Hagman, R. S. 254 Jun, W. 9
Haig, G. 28 Juntanamalaga, P. 349, 419
Haiman, J. 438-40
Hale, K. L. 100, 318 Kakumasu, J. 129
Hamel, P. J. 124 Kari, J. 167-9, 382
Hamp, E. 396 Karmiloff-Smith, A. 414
Harms, P. L. 436 Kayser, A. 110
Harris, A. 383 Keating, E. 11, 187, 260, 342, 344-5, 359,
Harrison, G. 419 361, 364, 366-7
Harrison, S. P. 11, 133, 186, 365 Keenan, E. L. 29, 162, 257, 363, 406
Harvey, M. vii, 11, 23, 25, 49, 59, 70, 81, 92, Kern, B. 32, 80, 257-8, 321
280, 321, 329, 369, 383, 385, 399 Kerr, I. 50, 224
Hasada, R. 2 Key, M. R. 110
Hashimoto, M. J. 226-7, 361-2 Khaidakov, S. M. 34, 347
Haspelmath, M. 21, 263 Kibrik, A. E. 47, 176
522 Index of Authors
Kimball, G. D. 158, 169, 436 McChombo, S. A. 29, 34, 176, 392
Kirk, L. 282 McGregor, A. R. F. 277
Kirton, I F. 33, 66, 385 McGregor, D. E. 277
Kiyomi, S. 7, 271-2, 278, 288 McGregor, W. 125, 258, 294
Klein, H. E. M. 12, 179 McGuckin, C. 147
Koehn, E. 128 McLaughlin, F. 62, 348
Koehn, S. 128, 295, 336, 360, 365 Magomedbekova, Z. M. 32
Kohler, O. 77 Mahapatra, B. P. 104, 112, 185, 248, 286
Kolver, U. 103–4, 213-14, 227, 273, 288, 334, Malinowski, B. 210
379 Manessy, G. 287
Kooyers, O. 77 Marchese, L. 377, 399
Kopcke, K. M. 24-5, 280 Marnita, R. 85, 97-8, 100, 117, 120, 189-90,
Koval', A. I. 61 253, 262, 286, 289, 308, 318, 322-3,
Krauss, M. 11, 154, 176, 209 329-30, 348, 356, 359, 361, 366, 375,
Krejnovic, E. A. 23, 42 381, 386, 401, 404, 445-6
Krishnamurti, B. 46, 77, 108, 378, 398 Marr, D. 338
Kuipers, A. H. 109, 114 Martin, J. 158, 251
Kunene, E. 416 Martin, S. E. 106
Martins, S. A. 11, 80, 85, 139, 174, 192, 259,
Lakoff, G. 307-9, 311 384
Landar, H. 336 Martins, V. 80, 384
Lang, A. 11, 158, 171, 299 Mathiot, M. 346
Langacker, R. W. 127, 144, 310 Matisoff, J. A. 405
Laoust, E. 39, 253-4, 388, 440 Matsumoto, Y. 308-10, 417-18, 420
Lapointe, S. 29 Mattei-Muller 34, M.-C. 127
LaPolla, R. 48, 155-6, 175, 215, 291, 300, Matthews, P. H. 28-9, 272, 311
323, 439 Meillet, A. 374, 378
Larrimore, B. M. 37 Melchert, H. C. 378
Laskowski, R. 48 Merlan, F. 11, 49, 56-7, 92, 149, 159, 171,
Lawton, R. S. 210 322, 329-30, 373, 408
Laycock, D. C. 67, 92, 123, 277 Mills, A. E. 22, 415
Lee, J. 150, 276, 389, 391, 400 Minor, E. 246
Lee, K.-D. 108, 294 Minor, L. 246
Lee, M. 339-40, 421 Mithun, M. 11-12, 150-1, 160-2, 201, 297,
Lee, Y.-S. 106-7, 115, 260, 273, 311, 334, 394, 401-3, 437, 444
336, 348, 386 Moore, B. 139
Leeding, V. 35, 58, 66, 151, 160, 191, 200, Moore, D. L. 86
247, 280 Morice, A. G. 168
Leer, J. 169, 444 Mosel, U. 124, 136, 183
Lehmann, C. 7, 15, 21, 29, 392, 416 Moshinsky, J. 299
Lehrer, A. 116, 443 Moussay, G. 9
Lermontov, M. Y. 409 Mpaayei, J. T. O. 375
Levelt, W. J. M. 415 Mufwene, S. S. 31
Levy, P. 288, 355, 502, 443, 446 Mujica, M. I. O. 210
Levy, Y. 414-15 Mulford, R. 415
Li, C. N. 98-9, 206 Munroe, P. 34, 40
Lichtenberk, F. 3, 11-12, 90, 113, 125-6, Muntzel, M. 391
133-7, 293–4, 365, 382
Lindrud, S. 79 Nedjalkov, V. P. 109
Lindstrom, E. 32-3, 79 Nekitel, O. 10, 23, 26, 79, 262, 346
Lobel, E. 117, 119-20, 211, 317-18, 326, 329 Ng, Bee Chin 419-20
Lock, A. 67, 77, 123, 277, 322 Nguyen, D. H. 336
Lowe, I. 93, 221, 303, 333-4 Nichols, J. 7, 27, 31, 49, 57, 60, 137, 255,
Lucy, J. A. 339 258, 263, 369, 373, 377, 383–4, 389, 409
Luke, K. K. 419 Nicholson, R. 79
Lynch, J. 134, 382 Nientao, S. 254
Lyons, J. vii, 115-6, 255, 330 Nordlinger, R. 281, 397
 Index of Authors 523
O Siadhail, M. 262 Rastorgueva, V. S. 60, 185
Obata, K. 79 Reh, M. 358-9, 374-5, 388, 392-3, 402-3
Ober, D. 253 Rehg, K. 11, 186, 294, 301, 336, 359, 361, 366
Onishi, M. 12, 34-5, 69, 79, 105-6, 132, 162, Reid, N. vii, 9, 11, 81, 84, 89, 92-3, 95, 97,
176, 219-21, 279, 303, 348, 369, 371, 185-6, 201, 283, 394-5, 404, 407
381, 436-40 Rice, K. 154, 167, 169, 250, 297-8, 344, 437
Osam, E. K. A. 342-3, 398-9 Richards, J. 76, 182, 213, 220
Osborne, C. R. 33, 56, 59, 160, 200, 276, 389 Riddle, R. 325-6
Osumi, M. 142 Rigsby, B. 123
Ott, R. 213 Roberts, J. R. 437
Ott, W. 213 Rodrigues, A. D. 11, 129, 133, 135, 229, 294,
434
Pacioni, P. 214-15, 226, 326 Rosch, E. 14, 309, 338–40
Paducheva, E. V. 88 Ross, M. 384
Palmer, G. 343, 407 Rothstein, R. A. 313, 347, 439
Panfilov, V. Z. 100, 108 Rowan, O. 226
Parker, D. 33, 79, 244 Royen, G. 5
Parker, G. J. 48 Rubino, C. R. G. 313, 388-9
Parker, J. 33, 79, 244 Rude, N. 82, 123
Pasch, H. 28, 33, 63, 75, 129-30, 192, 383, Rumsey, A. 9, 56, 329
387-8 Rushforth, S. 149, 154, 157-8, 330
Paul, P. 24 Russell, R. A. 40, 392
Pawley, A. 11, 124, 133, 244, 312
Payne, David L. 143, 369, 387 Sabajo, M. 279
Payne, Doris L. vii, 6-7, 11, 30, 69, 80, 86, Sanches, M. 100, 249, 348, 386, 417-19
99, 107, 123, 125, 127, 133, 140, 171, Sandalo, F. 12, 130, 175, 181, 198, 363
186, 199, 208, 217, 220-2, 227, 266, 333, Sands, A. K. 11, 49, 55, 58, 79, 81, 86, 91-2,
360 95, 97, 150, 169, 185, 191, 200, 228, 329,
Payne, T. E. 107, 217, 220, 222, 333 372, 381, 399
Pe, H. 11, 103, 116 Sapir, D. 311
Peeke, M. C. 217 Sapir, E. 158
Pensalfini, R. 281 Sarsa, R. 277, 437
Perez-Pereira, M. 414-15, 423 Sasse, H.-J. 390-1
Pet, W. J. A. 50, 174, 201, 262 Saul, J. E. 100, 103–4, 118, 211-12
Peterson, M. N. 245, 255, 384 Sauvageot, S. 61-2
Pike, K. 376 Saxton, D. 127
Plank, F. 25, 244 Sayaba, T. 11, 133
Plungian, V. A. 61, 438 Schane, S. A. 54
Polome, E. 400 Schaub, W. 24, 57
Pope, M. K. 379-80 Schauer, J. 106
Popova, M. I. 414-15 Schauer, S. 106
Poser, W. 168, 382 Schellinger, W. 244
Posner, R. 378-80 Scherbak, A. M. 102
Pozdniakov, K. I. 77 Schmidt, A. 349, 390-1
Prasse, K.-G. 245, 254 Schwartz, L. 53
Press, M. L. 127, 295 Seiler, H. 7, 11, 125, 136, 143, 146, 171,
Priestly, T. M. S. 245, 380, 396, 398-9 176-7, 182, 343, 348, 365, 409
Pullum, G. K. 53 Seiler, W. 11, 152, 171, 297, 300, 330, 363-4,
Pym, N. 37 379, 439
Seki, L. 210
Quigley, S. 124, 211, 287, 418 Selischev, A. M. 347
Quinn, H. 169 Senft, G. 8, 104, 136-7, 204, 210,223, 260,
295, 304, 316, 328, 331, 338, 366, 376,
Rakhilina, E. 153-363 411, 417-19, 445
Ramirez, H. 107, 229, 286, 355, 358, 376, Serzisko, F. 3, 6-7, 12, 28, 75, 176-8, 248,
380, 444-5 301, 326, 363
Rankin, R. L. 177 Shaul, D. L. 158, 299
524 Index of Authors
Shawcross, W. 347 Tsonope, J. 22, 50, 349, 396, 416
Shepard Jr., G. 12, 191, 199-200, 213, 341 T'sou, B. K. 11, 116,318
Shepardson, K. W. 44, 63 Tucker, A. N. 375
Sherzer, J. 79, 121 Turpeinen, S. 33, 79
Shim, J.-W. 79, 437 Tversky, B. 338
Shimojo, M. 328, 335-7
Shmelev, A. D. 51, 54 Vail, L. 63
Silverman, M. G. 291 Vasmer, M. 26
Silverstein, M. 247, 437 Vidal, A. 12, 86, 180-1, 251, 326, 336
Sirk, U. 315 Vietze, H. P. 102, 121
Slobin, L. 100, 249 Viitso, T.-R. 369
Smith, I. 90 Vincennes, L. 39
Smith-Stark, S. 247, 437 Vinogradov, V. A. 61, 124, 287, 356
Smoczynska, M. 414 Voeltz, E. 53
Sohn, Ho-Min 107, 113 Vogel, A. 59, 258
Sokolova, V. S. 60 Vossen, R. 358
Speece, R. 79, 123, 244, 346 Vycichl, W. 368
Speirs, A. 11,250
Spitulnik, D. 281-2, 409, 416 Waddy, J. 43
Spriggs, R. 124, 183 Wajanarat, S. 101, 104
Stachowiak, F. J. 7 Walsh, M. 37, 84, 86, 89-90, 97, 284
Stebbins, T. 247, 274 Wang, Fu-shih 208
Steele, S. 15, 29 Waters, B. 147, 382
Steinberg, E. 53 Watkins, L. 80, 177-8, 250
Steinhauer, H. 244 Walters, J. 99, 124
Stone, G. 347 Welmers, W. E. 31-2, 35, 77
Storch, A. 380 West, B. 218, 221-2, 264
Strom, C. 386 Westermann, D. 321
Stucky, S. U. 389, 400 Westley, D. O. 257
Stumme, H. 254 Whitney, H. 79
Stump, G. T. 63-5 Wierzbicka, A. 274, 279, 313, 440
Suarez, J. A. 12, 153, 169 Wiesemann, U. 36
Subrahmanyam, P. S. 266 Wilkins, D. 81-2, 84, 401, 442-3
Suppalla, T. 296 Williamson, K. 398
Suter, E. 397 Wilson, N. F. 100, 103–4, 118, 211-12
Suzman, S. M. 416-17 Wilson, P. J. 198, 286, 290
Sweetser, E. 308, 402 Wilson, W. H. 136
Wise, M. R. 199, 222
Talmy, L. 156 Wogiga, K. 62
Taylor, D. M. 44 Woodbury, A. C. 181
Taylor, J. R. 146 Worsley, P. M. 5, 12
Terrill, A. 79, 244, 257 Wurm, S. A. 79, 147, 171, 218, 386
Thiesen, W. 221-2, 246-7, 287, 333, 385, 387
Thompson, C. 9, 11, 167-9, 176, 250, 298, Ye, E. O. 53
382
Thompson, S. A. 98-9, 206, 325 Zabolocky, N. A. 86
Todd, E. 244 Zalizniak, A. A. 88
Toporova, I. 279, 407 Zavala, R. 11, 88, 90-1, 98, 113-14, 118-19,
Traill, A. 35, 396 187, 284, 287, 357, 364, 403–4
Trask, L. 272 Z'graggen, J. 67, 123
Traugott, E. C. 370, 374 Zograf, G. A. 185
Trudgill, P. 390 Zubin, D. A. 24-5, 280, 328, 335-7
Tseng, O. 423 Zwicky, A. M. 52-3
 Subject Index

A (transitive subject) 15, 33-4, 37, 60, see also repeaters

162, 194, 197, 200-1, 256, 365, 378, alternative ordering of morphemes 94
426-7, 438, 449, 451 anaphoric function 21, 50, 70, 329-30, 374,
ablaut 60 392, 417
asolutive 49, 406 of classifiers in multiple classifier
abstract nouns 47, 63, 72, 157, 159, 174, 200, languages 331-3
236, 275, 293, 304, 334-6, 349, 360, of classifiers in possessive
378, 450 constructions 127, 131
acquisition 1, 16, 18, 22, 309, 338, 413-25, of deictic classifiers 173, 331
432, 435, 450 of noun classifiers 81, 87-9, 252, 329-30
of multiple classifier systems 413, 424 of numeral classifiers 98, 118, 329
of noun classes 413-17, 420, 423–4, 432 of verbal classifiers 150, 157, 327, 330
of numeral classifiers 417-21, 423-4 animacy 1-3, 8, 17, 354, 357-9, 363, 366,
adjectival modifiers 15, 35, 69, 207, 209, 369, 373, 377-8, 381, 389-90, 393,
212-13, 215, 217, 226, 448 396, 410, 428, 435, 442-3, 448-50
see also adjective in classifiers in possessive
adjective 17, 19-20, 31, 35, 37-40,47,49, 53, construction 125-7, 130-1, 139, 143,
56, 59-61, 70, 94-5, 100, 104, 112, 146
204, 209-10, 213-21, 224-30, 234-5, in deictic classifiers 173-4
253, 265, 304, 353, 388, 396-400, 447 hierarchy 247
adpositions 18, 33, 172-4, 176, 180, 352, in language acquisition and
369, 427 dissolution 414, 418-19, 423
adverbs 17, 19-20, 34, 180, 353 in languages with more than one classifier
agglutinating languages 6, 10, 20, 205, 248, type 185, 187, 189, 191, 197, 199-200
447 in locative classifiers 177, 181
and numeral classifiers 99, 102-3, 109 in multiple classifier systems 213, 218, 222,
agreement 5, 15, 20-1, 27, 31-45, 94, 193, 227, 239–40
229, 247, 265, 328-9, 358, 425, 428, in noun class systems 21-2, 25-29, 34, 38,
436, 439, 448 41-2, 44, 47-9, 51, 55, 57, 62, 69-70,
anaphoric 21, 29, 47 75-7, 79-80
constraints on 39-41 in numeral classifiers 98, 101-2, 106,
default or neutral 51-2 113-15, 117, 121, 123
double 32, 38 and other categories 242, 244, 246, 248,
genesis of 372, 374-5, 391-8, 417 250-1, 256, 260, 263, 436-40
hierarchy 39, 229 as a semantic parameter in noun
lack of 88, 125-6 categorization 271-2, 275, 279-80,
loss in language dissolution 422 282-3, 286-8, 293, 295, 297, 300-6,
loss and reduction of 388-9, 398-400 312-13, 315, 317, 319, 329, 335-8,
neutralization of 40 342, 349-50
and numeral classifiers 110, 118, 199 in verbal classifiers 149, 152, 155-6, 158,
quirky 38 161
semantic 31, 37, 75 animate referent, see animacy
syntactic 31, 37, 75 aphasia 16, 422-4
target of 36 apophony 58-9, 430
variability in 28, 37, 41-4 areal diffusion 198, 219, 226-7, 313, 353,
agreement class 1-2, 95 371, 450, 383-8, 391, 397-8,
see also noun class, agreement 399-400
alienable possession, see possession type direct diffusion 382-6, 388, 391
alliterative concord 35-6, 396-7 see also loans
526 Subject Index
areal diffusion (cont.): partly analysable 156-7, 171
numeral classifiers, as areal feature 101, posture verbs 159
121, 123-4 stance verbs 166
areal prefix: suppletive 153-6, 161, 169
in locative expressions 176 classifier, deictic 3, 8, 12, 18, 172, 176-83,
as verbal classifier 168-9 306, 425, 448
arrangement as a semantic parameter, or and definiteness 177
configuration 274, 285, 293, 296, and deictic categories 266
299-300, 304, 422 discourse functions of 326
article 2-3, 7-8, 12, 20, 31, 69, 76, 235, 367, emergence as a dictinct type 183
369 fused with demonstratives 177,181
article classifier systems 7, 12, 172, 176-82 and gender 179-81
see also classifier, deictic in languages with more than one classifier
aspect 242, 264-5 type 191, 197, 202
assignment of gender, see gender assignment lexical sources for 362-3, 412
Athabaskan linguistic tradition 9, 167, 169, in locative expressions 178
265 in multiple classifier systems, see multiple
attributive noun phrase, see head-modifier classifier systems
noun phrase and number 251-2
augmentative 63-4, 359 obligatoriness of 179
Australianist linguistic tradition 9 origin of 177, 182
autoclassifiers, see repeaters and other categories 269-70
plural class in 178
Bantuist linguistic tradition 9 reclassification in 178, 180, 182
basic linguistic theory vii, 4 semantics of 300-1
batang, classifier in Minangkabau 90, classifier, locative 2-3, 8, 12, 18, 172-6, 302,
189-90, 289, 291-2, 302, 318, 359, 448
370, 402, 404, 446 discourse functions of 326
body parts 353-7, 405 distinctions comparable to 176
body part possession 130-1 in languages with more than one classifier
and noun classes 66, 72 type 192-3, 196-7, 201-2
as numerals 100 lexical sources for 357, 412
polygrammaticalization of 375 and number 252
as sources for classifiers 442-6 origin of 370, 375
borrowing, see loans and other categories 269-70
boundedness as a semantic parameter 181 residual classifier 174
semantics of 300-1, 306, 335, 425
case 21, 57, 242, 377, 379, 397, 436-7 and verbal classifier 175
case-marking, see case classifier, noun 2, 5, 8, 11, 285, 306, 425-35,
categorization, linguistic of a noun 448
(referent) vii anaphoric function of 81, 87-9
chaining model 309-11, 315 cooccurrence of 83-4
child-language acquisition, see acquisition deletion of 89
class morphemes 161 discourse functions of 322-3
see also classifier, verbal distinguishing from compounds 85-7
class nouns 87 distinguishing from free noun 85-7
classificatory adpositions, see classifier, grammaticalization of 81, 84-8, 90-1, 93,
locative 95
classificatory noun incorporation, see noun in languages with more than one classifier
incorporation type 184, 186, 191-2, 202
classificatory technique 7 lexical sources for 357, 359-60, 412
classificatory verb 11, 153-62, 336, 344, 363, loss of 381
412, 448 in multiple classifier systems 230
correlations with number 153, 158, 249-52 see also multiple classifier systems
existential 155-6, 158-9, 166, 171, 175, 299 and noun class 92-5, 97
origins of 160-1, 163, 165, 166, 169 and number 252
see also classifier, verbal, origin of and numeral classifiers 90
 Subject Index 527
obligatoriness 81, 89, 186 in multiple classifier systems, see multiple
omission of 90, 95 classifier systems
origin of 191, 371-3, 376 and noun classifiers 130
and other categories 268-70 and numeral classifiers 147
properties and realization of 91 origin of 383
as relative clause markers 88-9, 92-3 and possession types 259
semantic change in 402-4 realization of 126, 145
size of inventory 81, 84-7 used independently of possession
syntactic function of 87 type 129-33
classifier, numeral 2-3, 8, 11, 17, 105, see also classifiers in possessive
227-30, 265, 272, 283, 302, 305-6, constructions
426, 435—6, 438, 448 classifier, possessor 2-3, 17, 125, 139-40,
absence of 100 306, 425-7, 448
applicability of 334 categorization of possessor in 140
as areal feature 101, 121, 123-4 in languages with more than one classifier
change in 120, 348, 402-6, 408, 409-12 type 192, 202
constituency of 99, 105, 110-11 in multiple classifier systems, see multiple
dependence on size of number 100, 107, classifier systems
109, 112, 114, 117-29, 190, 212 and possession type 259
functions of 98, 118, 318-20, 324, 326 rarity of 146
genitive or attributive with 106-7, 121 realization of 139
incipient system of 101, 120-1 see also classifiers in possessive
in languages with more than one constructions
classifier type 186-7, 189, 193, 196-8, classifier, relational 2, 11, 17, 125, 133-9,
202 306, 343, 426-35, 448
lexical sources for 355-7, 360, 362, 364, and alienable possession 133-8
366-7, 411-12 in languages with more than one
loss of 381, 390-1 classifier type 184, 191-2, 197,
mensural 114-15, 293, 355-6, 366, 386, 201-2
406, 445-6 lexical sources for 364-5
more than one kind of in one loss of 382
language 101, 112-14 in multiple classifier systems, see multiple
in multiple classifier languages, see classifier systems
multiple classifier systems and possessed classifiers 137—8, 140-2,
and noun class 98, 109, 184 144-6
and noun classifiers 113-14, 123 and possession type 259
and number category 100-1, 249-52 realization of 136, 145
obligatoriness of 100, 106, 114, 117, 190 see also classifiers in possessive
and obligatory plural marking 100 constructions
origin of 109, 111, 113, 121, 370-2, 375, classifier, verb-incorporated, see classifier,
379, 384, 386, 443-6 verbal
and other categories 268-70 classifier, verbal 3, 6, 8, 11-13, 16-17, 227,
pragmatic functions, (uses) of 117 302, 305-6, 394, 426-35, 435, 439
properties and realization of 98, 102-12 anaphoric functions of 150, 157, 330
and quantifier 101, 115-20 applicability of 334
semantics of 186-93, 311-12, 316, 335, difference from lexical selection of
342, 344 verbs 153, 355, 362-5
sortal 114-15, 286-93, 355-6, 366, 386 discourse functions of 149, 326-7
classifier, possessed 2-3, 11, 17, 125, 132-62, generic noun as 150, 403
230 and grammatical function 257
and alienably possessed nouns 126-9, grammaticalization of 160-3
131-2 in languages with more than one classifier
and inalienably possessed nouns 128 type 184, 193, 197, 200-2
in languages with more than one classifier lexical sources for 355, 360, 362-5, 411-2
type 191-2, 202 limitation to some semantic groups of
lexical sources for 358, 360, 362, 364-7, verbs 149, 153, 165, 296-7
411-12 loss of
528 Subject Index
classifier, verbal (cont.): see also human cognition
in multiple classifier systems, see multiple cognitive process 308
classifer systems colour 338, 340
obligatoriness of 149 complementizer 34
origin of 149, 160-1, 163, 165-6, 169, 370, complete involvement as a semantic
375, 379 parameter 165-6, 196-7
see also classificatory verb, origin of compounds and numeral classifiers 103,
and other categories 265-6, 268-70 120
properties and realization of 149-58 concord, see agreement
reclassification 157, 164 concord, alliterative, see alliterative concord
semantic change in 402 concordial class 10, 19
semantics of 295-300, 335 see also noun class
size of inventory 150, 152 concordial systems, see noun class
and verb classes 265 concrete 275
verbal classifiers: conjunction of nouns, see coordination
in Athabaskan tradition 9, 167, 169, 265 consistency as a semantic parameter 3, 251,
in Australianist tradition 9 273, 289-91, 293, 295, 297-8, 300,
in South and Southeast Asian 302, 305-6, 339, 350
tradition 9 in classifiers in possessive
classifiers: constructions 127
article systems of, see article classifier in locative classifiers 172, 174-5
systems in numeral classifiers 115
coexistence of more than one type in one in verbal classifiers 149, 156
language 14 consonant-initial dropping 91
construction 13 constituent order 34, 40, 214-15, 378, 395,
contingent properties of 15-16 420
definitional properties of 13, 14-16 and numeral classifier constructions 104,
interaction with other categories 15, 21, 106-7, 120
41, 57, 425, 428 constituent:
marginal types of 8 peripheral 34, 37
in possessive constructions: topical, 34, 197
anaphoric function of 127, 131 contact 313
change in 348 see also language contact; areal diffusion
discourse functions of 326, 426-35 continuum 3, 7-8, 13-14, 57, 425, 432-3,
generic 126 grammaticalization as a continuum 93, 95,
in languages with more than one 185
classifier type 193, 196-7 noun categorization as a continuum 3,
in multiple classifier systems, see 201, 425, 432-3
multiple classifier systems quantifier and numeral classifier as
and number 252 continuum 120
obligatoriness 126, 142 see also prototype and continuum
omission of 132 approach
origin of 140 contrast 197
and other categories 268-70 controllers 51
and possession types 259 non-prototypical controllers 51
semantics of 293-5, 301, 305-6, 343-5 coordination 52, 56
size of inventory 145 corpus, see sample
specific 126 correlations between classifiers and other
and speech register 342 categories 7
syntactic function 127 countability 249, 274, 396
size of inventory 6 see also countable nouns
specific, see unique classifiers countable nouns, or count nouns 24, 115-20,
typology of 5-12 220-1, 249, 274
closed class 19, 36-7, 103, 241, 367, 374, 392, creolization of languages 382, 388-9, 391,
411, 447 398–400
closed systems 17 see also pidginization
cognitive mechanisms 13 cross-classification, see reclassification
 Subject Index 529
cross-referencing 14, 35, 48, 60, 68, 70, 80, epicenes 41
110, 192, 194, 200-1, 213, 252, 255-6, ergative 263, 378, 437-8
259, 264, 399, 438 evolution of classifiers 16, 18, 352, 367-70,
425,431,450
decay of classifiers 16, 18, 352, 425, 431 existential verbs 155-6, 158-9, 166, 171, 175,
declension 13, 25, 263, 377, 436, 438 299, 439
interaction with classifiers 243, 262-3, 268-9 extendedness as a semantic parameter 178,
declension classes, see declension 420
default: extension, semantic 83, 307, 309, 311, 338,
classifier 141, 274, 279, 281, 293, 304, 307, 346, 403-4, 408-11, 421, 445
335-7, 423, 450
see also residue classifier female, see feminine 23-4, 26-7
or neutral agreement forms 53 feminine 2, 14, 244-7, 252-6, 259, 262-3,
definiteness 39, 40, 117, 211, 215-16, 243, 266, 277-81, 286, 310, 313, 322,
249, 321, 323-4, 326-7, 333 336-7, 346, 350, 358, 368, 378-81,
deictic classifier, see classifier, deictic 384, 386, 388-9, 393, 400, 408, 413-15
deictic modifiers, see demonstratives in languages with more than one classifier
deictics, see demonstratives type 186, 194, 197, 199, 200
demonstratives 8, 14, 17, 19-20, 31, 36, in multiple classifier languages 210, 213,
39, 47, 49, 56, 59, 61, 65, 68-70, 76, 218, 228, 230, 240
172, 192, 194, 197, 199, 210, 213, 226, in noun classes 19-27, 32, 35, 37-42, 44-5,
228, 235, 239, 255, 265, 318, 326, 352, 47-8, 50, 54, 56-60, 66-7, 69-72,
367, 380, 392, 394, 398, 416-18, 439, 75-80, 180, 182
448 in numeral classifiers 106
distinct classifiers and agreement class for focal instances 14
182-3 see also prototype
gender on 183, 213 focusing 40, 58, 107, 223, 318, 327, 328
numeral classifiers with 183, 208 form:
dependencies between derivation and other in classifiers in possessive constructions
categories 242-3, 249, 252-7, 268-70 125-6, 146
dependent-marking languages 82 in noun classifiers 82
derivation 5, 28, 30, 92, 220, 269, 358, 440 in numeral classifiers 102, 106, 113, 123
derivational functions: as a semantic parameter 185, 187, 297
of classifiers 220-2, 225, 235, 252, 266, in verbal classifiers 152-3
287, 303 frequency of classifier types 4-5
of noun class 84, 358, 388 function:
see also noun class, overt in a clause (A, S, O) 3
determiner 35 as a semantic parameter 83, 87, 186,
see also demonstratives 271-3, 278, 287, 292-3, 295-6, 300,
development of classifiers 4-5, 16 305-7, 317, 320-35, 339-40, 344-5,
dimensionality as a semantic parameter 180, 387, 404, 406, 418, 423
268, 272, 288-93, 297-306, 338, 350, fusional languages 6, 10, 20, 205, 248, 256, 447
363, 411 numeral classifier in 99, 105, 108, 121
diminutive 25, 63, 69, 105, 111, 279, 358, fuzzy boundaries between types 12
379 fuzzy types 230-41
direct object, see O; object, direct
directionality as a semantic parameter 266, gender 2, 8, 10, 19-80, 350, 440
272, 289, 305-6 controller 45-6
disambiguating polysemous referent 43 in languages with more than one classifier
discourse-pragmatic properties 30, 37, 39, type 192, 196
62 loss of 379-81, 388-9, 390-1, 396
discourse conditions for generic nouns 86 in multiple classifier languages, 213, 228
dissolution: see also multiple classifier systems
of classifiers 16, 18, 422-423, 425, 432, origin of 367-71, 395-6
435 residual 47-8
of noun classes 422 target 45-6
dual 244-7, 251, 254-5, 393, 437 see also noun class
530 Subject Index
gender assignment 10, 15, 21, 22-8, 275-7, honorifics 63, 284, 317, 422, 443
413, 425 human cognition vii, 2, 3, 13, 18, 307, 319,
mixed 10, 15, 21, 25-8, 275, 348 337, 339, 424, 435
morphological 15, 22-5, 263, 275, 408 human noun vii, 19
phonological 10, 25, 62, 275, 348, 408 see also human referents; humanness
semantic 15, 56, 275, 348, 390 human referents vii, 21
gender: see also human noun; humanness
in Athabaskan linguistic tradition 9, 167, humanness 1, 16, 275, 281-8, 295-7, 301-6,
169, 265 308, 310, 313, 315, 317, 337, 342,
and noun class in Bantuist linguistic 349-51, 354, 357, 359, 369, 373, 378,
tradition 9 381, 388-9, 400, 403, 428, 442-3,
and noun classes, interchangeable use of 449-50
8, 9 and interaction of classifiers with other
gender systems, see noun class categories 242, 244, 246, 248, 259-60,
gender-sensitive affixes 194, 228 262-3, 268, 435-40
generic: and language acquisition and
classifier 130, 328, 426 dissolution 411-12,417
term in agreement 54, 55 in multiple classifier languages 208
generic classifier, or general classifier 212, see also multiple classifier systems
442 as a semantic parameter in classifiers in
see also default classifier possessive constructions 127, 146
generic noun classifier, see generics as a semantic parameter in languages with
generic-specific: more than one classifier type 185,
pairings 86, 395 187, 189, 197
relations 275, 295, 302, 305, 372, 395 as a semantic parameter in locative
generics, or generic nouns 85-6, 128-9, 186, classifiers 180
190, 197, 211, 275, 317, 349, as a semantic parameter in noun classes
394-5, 353-4, 358-9 21-6, 39–40, 42-3, 48-50,
polygrammaticalization of 376, 402-3 53-4, 56-8, 60, 67-8, 72, 77, 79
genitive classifier, see classifiers in possessive as a semantic parameter in noun classifiers
constructions 82-3
grammatical relations 3, 18, 243, 255-7, as a semantic parameter in numeral
268-9, 437, 449 classifiers 105, 109, 112-14, 123, 192
grammaticalization 1, 15,21, 185, 283, 316, hybrid nouns 41, 313, 347, 358
333, 352, 355-9, 361, 370, 372-7, 385,
392, 395, 397, 400-1, 405, 411, 426, inalienable possession, see possession type
428, 432 inalienably possessed nouns 33, 65, 71
of noun classifiers 81, 84-8, 90-1, 93, 95, see also Chapter 5; possession type
186, 345 inanimate 312-13, 329, 336
unidirectionality in 372, 374 in classifiers in possessive
of verbal classifiers 160-3 construction 125-7, 130-1, 139, 146
grammaticalized system 19 in deictic classifiers 173—4
grammaticization, see grammaticalization in languages with more than one classifier
graphic noun classifiers 82 type 185, 189
in locative classifiers 177, 181
head 29, 63 in multiple classifier languages 222-3
in a noun-classifier construction 90 in noun classes 21, 23, 25, 27, 34, 41-3,
in numeral classifier construction 105 47-9, 51, 55, 58, 61, 70, 75, 77, 79-80
in possessive constructions 146 in numeral classifiers 109, 113, 119, 123
head classes 30, 61, 63, 394-6 in verbal classifiers 149, 152, 155-6,
head of a noun phrase 13, 39, 165 158, 161
head-marking languages 82 see also animacy
head-modifier: incipient types of classifiers 185
agreement 29, 37, 59, 61, 70, 76, 193–4, incorporation, see noun incorporation
198, 226, 306, 391–4, 422 indefinites 21, 55
noun phrase 17, 29, 31 see also indefinite pronouns
hon, classifier in Japanese 308-11, 316, 420 indefinite pronouns 54-5, 85
 Subject Index 531
indefinite reference 51, 54 in multiple classifier languages 208, 210,
individualizing functions of classifier 213, 218, 228
morphemes 93 see also multiple classifier systems
individuation 51, 58, 166 and language acquisition and
inductive approach 4 dissolution 413-15
inflection 5, 15, 28, 30, 220, 266 as a semantic parameter in languages with
inflectional languages, see fusional languages more than one classifier type 180,
inflectional properties of nouns 81 185-6, 194, 197, 199-200
inherent properties in noun categorization 2, in noun classes 19-27, 32, 35, 37–42,
17, 272, 303 44-50, 52–4, 56-60, 66, 69-72, 75-80
classifiers in possessive constructions 125, in numeral classifiers 106
144-5 male, see masculine gender
numeral classifiers 115 mass nouns 115-20, 220-1, 249, 278, 282
verbal classifiers 153, 158-9 material, or material make up 273, 289-90,
interioricity as a semantic parameter 273, 292, 302, 305-6, 404, 423
288-9, 302, 305, 345 measure words 115-20, 418
interrogatives 19, 21, 33, 35-6, 54-5, 59, 61, metaphor 311-13, 316, 341, 343, 346, 401,
85, 212, 228, 240, 367, 440-1, 448, 451 408
intralocative classifiers 7, 12 and polysemy 316
see also classifier, locative metaphoric extension 21, 315
isolating languages 6, 10, 20, 82, 205, 447 see also metaphor
noun classifiers in 82 metonymy 308-11, 315, 343, 401, 405
numeral classifiers in 99, 101, 103, 105 modifiers 17, 19, 29, 31, 39
morphological loci 13-14, 20, 31, 36-7, 68,
khan classifier in Thai 291, 348, 405, 408 211, 218-19, 240, 425-6
kinship: motion verbs 362-4
possession 146 multiple classifier systems 12, 16, 36, 77, 118,
as a semantic parameter in noun 124, 131, 133, 139, 142, 144, 160, 183,
categorization 82, 131, 141, 271-2, 204-41, 266, 271, 396-7, 425, 432-5,
280, 284, 305-6, 342, 353-4, 357-9 437
terms 41, 64, 66, 131, 357 adjectives with classifiers in 204, 209-11,
213, 216-18, 220, 222-3, 225, 227,
language: 230, 235
change 347, 389, 399,411 classifiers with deictics/demonstratives
contact 383-8, 398-40 in 204, 206, 208-10, 212-13, 215-18,
see also areal diffusion 220, 222-3, 225-7, 235, 265
obsolescence 5, 307, 313, 347, 348, 382, classifiers in possessive constructions
386, 389-91, 398–400,450 in 222, 225-7
planning 307, 347 constituent order in 214
lexicalization 15, 399, 426, 428 default classifier in 337
lexicon, structure of 242, 243, 347 derivational functions of classifiers in 252
loans 246, 333, 358, 383, 386-8, 450 discourse functions of classifiers in 325-7
noun class 57, 62 locative classifiers in 206-7, 225
numeral classifiers 105, 109-10 noun classifiers in 206, 211, 215-16, 370
see also diffusion, direct numeral classifiers in 370
locative classifier, see classifier, locative numerals with classifiers in 204, 206, 208,
loss of classifiers 1, 16 210-12, 215-19, 222-7, 235, 265
lustre, see visibility as a semantic parameter origin of 370, 386
possessed classifiers in 370
macroclass 49 relational classifiers in 225-6
markedness 15, 50-6, 71, 425, 428 semantics of 303-4
formal 15, 109 verbs with classifiers in 206, 208-9, 213,
functional 15, 38, 49, 71, 262 216, 219, 225
reversal 51
masculine gender 2, 14, 277-81, 286, 313, natural gender, see sex as a semantic
322, 328, 336-7, 346, 350, 358, 368, parameter
377, 379-80, 386, 388-9, 393, 400 negation 264-5
532 Subject Index
neuter 19, 23-5, 40-2, 44-5, 47, 65-6, 199, semantics of 342, 275-83, 345
222, 244, 255, 263, 266, 276, 279, 281, size of 6
335, 337 suprasegmental realization of 60
neutralization of noun class distinctions 51 variable assignment of 28, 68-9, 267
nominalizations 218-19, 221, 225 see also gender
nominative-accusative 437 noun class systems, see noun class
non-feminine 222, 230, 240, 256, 310, 384-5 noun classifier, see classifier, noun
non-human 16, 244, 247, 268 noun incorporation 150-1, 160-1, 165, 295,
in languages with more than one classifier 360, 394, 430
type 185, 189 body part incorporation and classificatory
as a semantic parameter in classifiers in 151, 165, 355
possessive constructions 127 classificatory 150-1, 160, 202, 227
in noun classes 41, 43, 48-9, 53, 75, 77 and lexical compounding 151
in numeral classifiers 105, 113, 123 noun phrase 20, 22
see also humanness number 18-19, 24, 52, 57, 74, 242, 279,
non-sex differentiable 23, 25 368, 377, 379-80, 388-9, 392-3,
noun class 2, 5, 8, 10, 17, 92, 184, 229, 231, 436-7, 449
234-5, 239, 252, 306, 425-35, 436, interaction with classifiers 243, 248-52,
438, 440, 448 254-5, 263-4, 268-70
agreement, decline of 398-400 and numeral classifiers 117
development of 392-8 numeral, numbers 2, 17, 20, 31, 40, 59, 61,
on multiple targets 228 67, 69, 75, 114, 199-200, 399, 418
applicability of 21, 334 numeral classifier, see classifier, numeral
changes in 407-9 numeral system 99-100
covert 56, 57-8, 193, 358, 373
and deictic cagegories 266 O (direct object of a transitive verb) 3, 15,
discourse functions of 321-2 17, 34, 73, 74, 149-50, 152-8, 160-8,
distinct systems in one language 45 184, 196-7, 200-1, 209, 250-1, 256-7,
double marking of 63-6, 95 265, 287, 306, 325, 327, 363-5, 406,
and gender, coexistence of 10 426-7, 437-9, 449, 451
and grammatical function 255-7 object, direct 3, 33-4
in languages with more than one classifier see also O
type 189, 191-2, 198, 200-2 object marker 35
lexical sources for 358-9, 362, 366, 411-12 obsolescence of language, see language
loss of 210, 379-81, 388-9, 390-1, 396 obsolescence
marked on noun 2, 17 omission of classifiers 335
see also noun class, overt open class 37, 411, 447
markedness relations 50-2, 54-6 classifiers in possessive constructions as
marking of 11, 19-78 129, 141
in multiple classifier systems, see multiple numeral classifiers as 98, 101, 103
classifier systems orientation as a semantic parameter 18, 149,
nominal 67-77, 184, 380-1 153, 158-9, 166, 175-6, 178, 287, 289,
and noun classifiers 185 298-9, 305-6, 406, 448.
and number 243-9 see also position in space
number of 28, 45 origin of classifier 1, 18, 196, 203, 226, 352,
and numeral classifiers 184-5 361, 450
opaque semantics of 63
origin of 368-9, 372-5, 377-9, 384, 386, paucal 248, 437
395-6 perception, perceptual features 307, 337-8,
overt 44, 56-9, 61, 63, 84, 91-5, 322, 325, 421-2
358, 373, 422, 448 peripheral arguments 162
and person 252-5 person 21, 52, 392
and possession 258-60 interaction with classifiers 245, 252-5,
productivity 58 263-4, 268-70
pronominal 37, 67-77, 184, 192, 205, 226, personal pronouns 39, 68-9
230, 246, 259, 283, 380-1, 398 phonological reduction of classifiers 357,
resolution 50, 52-4 371,376, 395
 Subject Index 533
pidginization 382, 388-9, 391 numeral classifiers 83, 98, 113
see also creolization of languages pronominal noun class, see noun class,
plural 21-2, 24, 45-6, 48, 50, 74, 77, 100, pronominal
187, 221, 240, 243-9, 251, 253-6, 263, pronominalization 13, 252
377,388-9,413 pronouns 21, 35-7, 39, 70, 79, 194, 210, 228,
in numeral classifier languages 100-1 252-6, 262, 358, 368, 373, 392-3, 398,
plural marking, see plural 415,439,451
polarity, see negation prototype 8, 14, 308-9, 314-15, 317, 325,
politeness 21, 102, 143, 260-2, 268-70, 436 403, 421, 425, 432-3
polygrammaticalization 363, 370, 374-6, and continuum approach 14
402-3
polysemy 271, 316 quanta as a semantic parameter 274, 285,
polysynthetic languages 99, 108, 248, 447 293, 299, 304, 422
position in space 149, 153, 158-9, 166, quantifier 2, 17, 49, 107, 115, 249
175-6, 178 quantifiers and classifiers 115-20, 249
as a semantic parameter 17, 250, 277, 296, quantity:
406 of an entity 115
see also orientation as a semantic expression of 98
parameter
possessed noun 2, 15, 31-2, 59-60, 66, 125, rarity of classifiers 5
294-5 rational gender 22-3
possessee 125 realization of noun classes and classifiers
see also possessed noun 6-7, 15, 17, 20-1
possession: reanalysis in noun categorization 186, 249,
alienable 126-46. 352, 370, 373, 377-9, 392, 399
see also possession type reclassification 267, 320, 328
direct 133 with classifiers in possessive
see also possession, inalienable constructions 137–41, 143
inalienable 33, 126, 146 in languages with more than one classifier
see also inalienably possessed nouns type 196
indirect 133 in multiple classifier systems 223, 267
see also possession, alienable with noun classes 43, 267
possession type 243, 257-9, 268-70,438, 449 with noun classifiers 83, 267
and noun class agreement 258-9 with verbal classifiers 267
possession type and type of classifier reduplication 59, 109, 430
137-9, 126-46 reference management 321
see also possessive construction, type of referent tracking 157, 192, 211, 321, 326
possessive construction, type of 2, 5, 31, 125, referentiality 333, 335
137-9, 126-46, 449 relative clauses 88-9, 221-3, 231, 327
possessive noun phrase 31 relative pronoun 39
possessives 19, 417 relativizer 35, 92, 327, 392, 396
possessor 2, 15, 32, 294-5 classifiers as 219, 221, 225, 333
postposition, see adpositions repeaters 5, 3, 15, 17, 59, 103, 353, 361-2,
posture verbs 153, 159, 182, 299, 362-3, 370, 376, 384, 397, 423, 430, 434-5
406 as classifiers in possessive
predicate 20, 39 constructions 128, 136, 141-2
categories 242 in languages with more than one classifier
predicate classifier systems, see classifier, type 187
verbal in multiple classifier systems 222-7
predicate-argument agreement type 29, 37, see also multiple classifier systems
40, 59, 70, 76, 193–4, 197, 306, 392, as noun class agreement markers 61-3
422 as noun classifiers 91
preposition, see adpositions as numeral classifiers 99, 103–4, 110
productivity of classifier systems 334 residual class 79
proficiency in: residue classifier 274, 279, 281, 293, 304
noun classifiers 83 see also default classifier
534 Subject Index
S (intransitive subject) 3, 15, 17, 33, 34, 37, size as a semantic parameter 185, 273,
60, 71, 74, 149-50, 152–4, 162-3, 165-8, 277-9, 282-3, 289, 295, 405, 418, 449
184, 194, 196-7, 200-1, 209, 250-1, in classifiers in possessive
256-7, 265, 306, 327, 363-5, 405, 426, constructions 126, 131, 146
437, 439, 449, 451 in genders and noun classes 21, 43, 58, 69
sample 4 in languages with more than one classifier
Sapir-Whorf hypothesis 340 type 185, 187
scope 6, 15, 20, 81, 95, 243, 258, 425-6 in numeral classifiers 98, 117
self-classifiers, or autoclassifiers, see in verbal classifiers 149, 152
repeaters social status as a semantic parameter in
semantic basis of categorization 4, 13, 21 noun categorization 260, 271-2,
semantic change 18, 352, 357, 361 280, 284, 288, 305-6, 315, 335, 342,
semantic extension 83, 307, 309, 311, 338, 350-1
346, 403—4, 408-11, 421, 445 in noun classifiers 82, 84
see also extension, semantic in numeral classifiers 98
semantic function 8, 84 socio-cultural changes 311, 347-50
semantic groups of nouns 24 socio-cultural parameters in
semantic opacity 20, 196, 275, 280, 283, categorization 16-17, 307
307-8, 315, 347,400, 407-8 specific classifier, see unique classifier
see also semantic transparency specificity 215, 321–4, 333
semantic residue 28 speech register 187, 260-2, 342, 366-7, 423
semantic transparency 20, 196, 275, 280, split agreement 30, 67, 76
283, 307-8, 315, 347, 407, 409, 416 split systems of noun classes 15, 66, 283
see also semantic opacity structure as a semantic parameter 17, 281
semantically transparent choice of noun in classifiers in possessive constructions,
classifier 82 126-7, 131
semi-repeaters, or partial repeaters 110 in noun classifiers 82
serial verbs as a source for classifiers 363 in numeral classifiers 98
sex as a semantic parameter 1,17, 19-22, 25, in verbal classifiers 149
44, 58, 68, 102, 185, 187, 275, 283-4, style 349, 423
286-8, 304, 341, 347, 350, 358, 396, subgender 48
428 subject 33-4, 37, 39, 194, 200, 422, 440
sex-differentiable 23, 26, 42, 43 superclassing 43, 48-9, 53, 55, 70, 228-9, 248
shape as a semantic parameter 2, 3, 17, superordinate-subordinate relation in
249-50, 256, 266, 268, 272-3, 275, classifier systems 316-17, 349, 370, 401
277-8, 281-3, 288-93, 295-7, suppletion 59, 252
299-307, 315-18, 337-40, 350, 356, suppletive classificatory verbs 11, 205
366, 384-5, 404-6, 409, 411, 418, 423, see also classificatory verb
446, 448-9 syntactic function 39
in classifiers in possessive
constructions 125-7, 131, 146 target of agreement 36
in deictic classifiers 177 target gender, see gender, target
in genders and noun classes 21, 42, 45, 69, tense 242, 263-4
76 topic continuity 321
in languages with more than one classifier topical constituent 34
type 185, 187, 191, 197, 200 topicality 34, 37, 39, 197, 321
in locative classifiers 172 transitive construction 73-4
in multiple classifier systems 200 tua, classifier in Thai 314-15, 349, 411, 422,
see also multiple classifier systems 442
in noun classifiers 82 type, agglutinative, see agglutinating
in numeral classifiers 98, 101-2, 105, languages
113-5, 117, 121, 123 type, fusional, see fusional languages
in verbal classifiers 149, 152-4, 156, 161, type, isolating, see isolating languages
163 type, polysynthetic, see polysynthetic
singular 21, 24, 45-6, 240, 243-52, 255, 263, languages
377, 388-9, 413
singular marking, see singular unclassified nouns 249, 272, 297, 324
 Subject Index 535
unique classifier 98, 166, 185, 273, 275, 293, verbal classifiers:
300-1, 344 in Athabaskan tradition 9, 167, 169, 265
unit counters 353-4, 360-1 in Australianist tradition 9
universal 4, 10, 244, 272, 338, 388. in South and Southeast Asian tradition 9
see also universal parameter in noun verb, noun class agreement on 22, 35, 47, 69,
categorization 72
universal parameter in noun verb-argument agreement, see predicate-
categorization 15, 22, 388 argument agreement type
verbal categories, see predicate categories
value as a semantic parameter 135, 274, 294, verbs, stative, semantic groups of 165
302, 307 verbs, telic 165
variable classification, see reclassification visibility as a semantic parameter 181, 266,
vegetable gender or noun class 66, 79, 278-9, 280, 363
281, 350, 366, 372-3, 408
verbal classifier, see classifier, verbal word order, see constituent order