R E F OR T R E S U M E S

ED 019 631 AL ODD 413

THE CASE FOR CASE.
BY- FILLMORE, CHARLES J.
PUB DATE APR 67
ECRS PRICE MF-$0.75 HC-$5.48 135P.

DESCRIPTORS- *LANGUAGE UNIVERSALS; *GRAMMAR, *DEEP STRUCTURE)

*SURFACE STRUCTURE, LINGUISTIC THEORY, LANGUAGE TYPOLOGY:
PHRASE STRUCTURE, TRANSFORMATION THEORY (LANGUAGE),
TRANSFORMATION GENERATIVE GRAMMAR, NOMINALS, CASE, BASE
COMPONENT,

IN THIS PAPER, PREPARED FOR THE APRIL 1967 TEXAS

SYMPOSIUM ON LINGUISTIC UNIVERSALS, IT IS PROPOSED THAT THE
GRAMMATICAL NOTION "CASE" DESERVES A PLACE IN THE BASE
COMPONENT OF THE GRAMMAR OF EVERY LANGUAGE. IT IS ARGUED THAT
PAST RESEARCH HAS NOT LED TO VALID INSIGHTS ON CASE
RELATIONSHIPS AND THAT WHAT IS NEEDED IS A CONCEPTION OF BASE
STRUCTURE IN WHICH CASE RELATIONSHIPS ARE PRIMITIVE TERMS OF
THE THEORY AND IN WHICH SUCH CONCEPTS AS "SUBJECT" AND
"DIRECT OBJECT" ARE MISSING. THE LATTER ARE REGARDED AS
PROPER ONLY TO THE SURFACE STRUCTURE OF SOME (BUT POSSIBLY
NOT ALL). LANGUAGES. BASIC TO THESE ARGUMENTS ARE TWO
ASSUMPTIONS NOW TAKEN FOR GRANTED BY GENERATIVE
GRAMMARIANS--(1) THE CENTRALITY OF SYNTAX, AND (2) THE
IMPORTANCE OF COVERT CATEGORIES. THE AUTHOR PROPOSES "THAT
THERE ARE MANY SEMANTICALLY RELEVANT SYNTACTIC REALTIONSHIFS
INVOLVING NOUNS AND THE STRUCTURES THAT CONTAIN THEM, THAT
THESE RELATIONSHIPS ARE IN LARGE FART COVERT BUT ARE
NEVERTHELESS EMPIRICALLY DISCOVERABLE, THAT THEY FORM A
SPECIFIC FINITE SET, AND THAT OBSERVATIONS MADE ABOUT THEM
WILL TURN OUT TO HAVE CONSIDERABLE CROSS-LINGUISTIC
VALIDITY." HE REFERS TO THESE AS "CASE" RELATIONSHIPS AND THE
GRAMMAR PROPOSED AND OUTLINED HERE IS "CASE GRAMMAR." THE
SUBSTANTIVE MODIFICATION TO THE THEORY OF TRANSFORMATIONAL
GRAMMAR WHICH HE PROPOSES AMOUNTS TO A REINTRODUCTION OF THE
"CONCEPTUAL FRAMEWORK" INTERPRETATION OF CASE SYSTEMS BUT
WITH A CLEARER UNDERSTANDING OF THE DIFFERENCE BETWEEN DEEP
AND SURFACE STRUCTURE. THE SENTENCE IN ITS BASIC STRUCTURE
CONSISTS OF A VERB AND ONE OR MORE NOUN PHRASES, EACH
ASSOCIATED WITH THE VERB IN A PARTICULAR CASE RELATIONSHIP.
THE "EXPLANATORY" USE OF THIS FRAMEWORK RESIDES IN THE
NECESSARY CLAIM THAT, ALTHOUGH THERE CAN BE COMPOUND
INSTANCES OF A SINGLE CASE (THROUGH NOUN PHRASE CONJUNCTION),
EACH CASE RELATIONSHIP OCCURS ONLY ONCE.IN A SIMPLE SENTENCE.
THIS DOCUMENT WILL APPEAR IN "PROCEEDINGS OF THE TEXAS
SYMPOSIUM ON LANGUAGE UNIVERSALS, APRIL 13 -15, 1967," EMMON
BACH AND ROBERT HARMS, EDITORS, HOLT, RINEHART AND WINSTON.
(JD)
 U.S. DEPARTMENT OF HEALTH, EDUCATION & WELFARE
OFFICE OF EDUCATION

THIS DOCUMENT iIAS BEEN REPRODUCED EXACTLY AS RECEIVED FROM THE

PERSON OR ORGANIZATION ORIGINATING IT. POINTS OF VIEW OR OPINIONS
STATED DO NOT NECESSARILY REPRESENT OFFICIAL OFFICE OF EDUCATION

CZ) POSITION OR POLICY.

The Case for Case*

1-44

"PERMISSION TO REPRODUCE THIS

by 111111111111 MATERIAL HAS BEEN GRANTED

BY Cif ARLE 5 J :Pt LLMORE

TO ERIC AND ORGANIZATIONS OPERATING

Charles J. Fillmore UNDER AGREEMENTS WITH THE U.S. OFFICE OF
The Ohio State University EDUCATION. FURTHER REPRODUCTION OUTSIDE
THE ERIC SYSTEM REQUIRES PERMISSION OF
THE MUM OWNER."

To appear in Emmon Bach and Robert Harms, eds., Proceedings

of the Texas Symposium, on Language Universals April 13-15,
1967 (Holt, Rinehart and Winston)

*The time I needed for preparing this study was created by an

act of the College of Arts and Sciences of The Ohio State
University: I was released from teaching duties during the
winter quarter of 1967. I gratefully acknowledge this boost
and hope that I may be forgiven for changing topics mid-
quarter. Many co-workers have kept me supplied with suggest-
ions, criticisms and challenges during the past few months.
I am particularly indebted to D. Terence Langendoen (Ohio
State), George Lakoff (Harvard) and Paul M. Postal (Queens
College). I shall probably soon regret that I have not
always been wise enough to follow their counsel.

AL 000 413
 The Case for Case

Charles J. Fillmore
The Ohio State University

Prepared for the 1967 Texas Symposium on Linguistic Universals,

April 13-15.

Speculation on language universals has not always and

everywhere been viewed as a fully respectable pastime for the

scientific linguist. The writer recalls a Linguistic Institute

lecture of not many summers ago in which it was announced that

the only really secure generalization on language that linguists

are prepared to make is that "some members of some human communi-

ties have been observed to interact by means of vocal noises."

Times have changed, it is a pleasure to report, and this is partly

because we now have clearer ideas about what linguistic theories

are theories of, and partly because some linguists are willing to

risk the danger of being dead wrong.

Scholars who have striven to uncover syntactic features common

to all of the world's languages have generally addressed themselves

to three intimately related but distinguishable orders of questions,

and they are: (a) what are the formal and substantive universals

of syntactic structure? (b) is there a universal base, and, if so,

what are its properties? and (c) are there any universally valid

constraints on the ways in which deep structure representations of

sentences are given expression in the surface structure?

Concerning formal universals we find such proposals as Chomsky's,

that each grammar has a base component capable of characterizing

the underlying syntactic structure of just the sentences in the

 ,paim=e

language at hand and which contains at least a set of trans-

formation rules the function of which is to map the underlying

structures provided by the base component into structures more

closely identifiable with phonetic descriptions of utterances

1
in that language. A representative statement on substantive

1Noam Chomsky, ASPECTS OF THE THEORY OF SYNTAX, MIT Press (1965),

pp. 251.27-30.

syntactic universals is Lyons' assertion that every grammar

reequires such categories as Noun, Predicator and Sentence, but

that other grammatical categories and features may be differently

2
arranged in different languages. And Bach has given reasons to

2
John Lyons, "Towards a 'notional' theory of the 'parts of speech',"
JOURNAL OF LINGUISTICS, 11(1966) 209-236; 211, 223.

believe that there is a universal set of transformations which

each language draws from in its own way, and he has shown what

such transformations might look like in the case of relative clause

modification. 3

3Emmon Bach, "On some recurrent types of transformations,"

GEORGETOWN UNIVERSITY MONOGRAPH NO. 18 ON LANGUAGES AND LINGUISTICS
(1965) 3-18.

Discussions on the possibility of a universal base (as distinct

from claims about universal constraints on the form of the base

component) have mainly been concerned with whether the elements

specified in the rules of a universal base--if there is one-- are

sequential or not. A common assumption is that the universal

base specifies the needed syntactic relations, but the assign-

ment of sequential order to the constituents of base structures

is language specific. Appeals for sequence-free representations

4
of the universal deep structure have been made by Halliday ,

M. A. K. Halliday, "Some notes on 'deep' grammar," JOURNAL OF

LINGUISTICS, II (1966) 55-67.

6
Tesniere 5 , and others. Lyons recommends leaving for empirical

Lucian Tesniere, ELEMENTS DE SYNTAXE STRUCTURALE, Paris- Klincksieck

(1959).

6
Lyons (1966) 227.

investigation the question of the relationship between the under-

lying representation and sequential order, an Bach 7 has suggested

'Bach (1965) 13.

that continued investigation of the syntactic rules of the world's

languages may eventually provide reasons for assuming specific

ordering relations in the rules of a universal base.

Greenberg's statistical studies of sequence patterns in

selected groups of languages do not, it seems to me, shed any

direct light on the issue at hand.8 They may be regarded as pro-

Joseph Greenberg, "Some universals of grammar," in J. Greenberg,

ed., LANGUAGE UNIVERSALS, MIT Press (1963) 58-90.

viding data which, when accompanied by an understanding of the

nature of syntactic processes in the specific languages, may even-

 tually lend comfort to some proposal other on either the
sequential properties of the base component or the universal

constraints which govern the surface ordering of syntactically

crganized objects.

Findings which may be interpreted as suggesting answers

to our third question are found in the 'markedness' studies of

Greenberg 9 and in the so-called 'implicational

universals' of

9
Joseph Greenberg, "Language universals," in T. A. Sebeok, ed.,
CURRENT TRENDS IN LINGUISTICS, III, Mouton (1966) 61-112.
111
10
Jakobson. If such studies can be interpreted as making empirical

10
Roman Jakobson, "Typological studies and their contribution to
historical comparative linguistics," in E. Sivertsen, ed.,
PROCEEDINGS OF THE VIIITH INTERNATIONAL CONGRESS OF LINGUISTS,
Oslo University Press (1958) 17-25.

assertions about the mapping of deep structures into surface struc-

tures, they may point to universal constraints of the following

form: While the grammatical feature 'dual' is made use of in one

way or another in all languages, only those languages which have

some overt morpheme indicating 'plural' will have overt morphemes

indicating 'dual'. The theory of implicational universals does not

need to be interpreted, in other words, as a set of assertions on

the character of possible deep structures in human languages and

the ways in which they differ from one another.

The present essay is intended as a contribution to the study

of formal and substantive syntactic universals. Questions of

linear ordering are left untouched, or at least unresolved, and
 -5 IOW

questions of markedness are viewed as presupposing structures

having properties of the kind to be developed in these pagea.

My paper will plead that the grammatical notion 'case'

deserves a place in the base component of the grammar of every

language. In the past, research on 'case' has amounted to an exam-

ination of the variety of semantic relationships which can hold

between nouns and other portions of sentences; it has been consid-

ered equi,ralent to the study of semantic functions of inflectional

affixes on nouns or the formal dependency relations which hold

between specific nominal affixes and lexical-grammatical proper-

ties of neighboring elements; or it has been reduced to a statement

of the morphophonemic reflexes of a set of underlying 'syntactic

relations' which themselves are conceived independently of the

notion of 'case'. I shall argue that valid insights on case rela-

tionships are missed in all these studies, and that what is needed

is a conception of base structure in which case relationships are

11
primitive terms of the theory and in which such concepts as 'subject'

11
Notational difficulties make it impossible to introduce 'case'
as a true primitive as long as the phrase-structure model deter-
mines the form of the base rules. My claim is, then, that a
designated set of case categories is provided for every language,
with more or less specific syntactic, lexical and semantic conse-
quences, and that the attempt to restrict the notion of 'case' to
the surface structure must fail.

and 'direct object' are missing. The latter are regarded as proper

only to the surface structure of some (but possibly not all)

languages.
 Two assumptions are essential to the development of the

argument, assumptions that are, in fact, taken for granted by workers

in the generative grammar tradition. The first of, these is the

centrality of syntax. There was a time when a typical linguistic

grammar was a long; and detailed account of the morphological struc-

ture of various classes of words, followed by a two or three page

appendit called 'Syntax' which offered a handful of rules of thumb

on how to 'use' the words described in the preceding sectiois--

how to combine them into sentences.

In grammars where syntax is central, the forms of words are

specified with respect to syntactic concepts, not the other way

around. The modern grammarian, in other words, will describe the

'comparative construction' of a given language in the most global

terms possible, and will then add to that a description of the

morphophonemic consequences of choosing particular adjectives or

quantifiers within this construction. This'is altogether different

from first describing the morphology of words like taller and

more and then adding random observations on how these words show

up in larger constructions. 12

12
John R. Ross pointed out, during the symposium, that there exist
some syntactic processes which seem to depend on (and therefore
'follow') particular lexical realizations of just such entities as
the comparative forms of adjectives. Compared adjectives, in short,
may be iterated, just as long as they have all been given identical
surface realizations. Witness
(i) she became friendlier and friendlier
(ii) she became sore and more friendly
but not
(iii) *she became friendlier and more friendly
 The second assumption I wish to make explicit is thee,

importance of covert categories. Many recent and not-so-recent

studies have convinced us of the relevance of grammatical properties

which lack obvious 'morphemic' realizations but whose reality can

be observed on the basis of selectional constraints and transforma-

tional possibilities. We are constantly finding that grammatical

features found in one language show up in some form or other in

other languages as well, if we have the subtlety it takes to discover

covert categories. Incidentally, I find it interesting that the

concept 'covert category'--a concept which is making it possible

to believe that at bottom all languages are essentially alike--was

introduced most convincingly in the writings of Whorf, the man

whose name is most directly associated with the doctrine that deep-

seated structural differences between languages determine the

essentially non-comparable ways in which speakers of different

13
languages deal with reality.

Alin
1
3Benjamin Lee Whorf, "A linguistic consideration of thinking in
primitive communities" (c. 1936), in J. B. Carroll, ed., LANGUAGE,
THOUGHT AND REALITY: SELECTED WRITINGS OF BENJAMIN LEE WHORF, MIT
Press (1965) 65-86; 69ff.

One example of a 'covert' grammatical distinction is the one

to which traditional grammarians have attached the labels 'affectum'

vs. 'effectum', in German 'affiziertes Objekt' vs. ' effiziertes

Objekt.' The distinction, which is reportedly made overt in some

languages, can be seen in sentences (1) and (2) below.

(1) John ruined the table

(2) John built the table

Note that in one case the object is understood as existing

antecedently to John's activities, while in the other case its

existence resulted from John's activities.

Having depended so far on only 'introspective evidence', we

might be inclined to say that the distinction is purely a semantic

one, one which the grammar of English does not force us to deal with.

Our ability to give distinct interpretations to the verb-object

relation in these two sentences has no connection, we might feel,

with a correct description of the specifically syntactical skills

of a speaker of English.

The distinction does have syntactic relevance, however. The

'effectum' object, for example, does not permit interrogation of

the verb with do to, while 'Wu, affectum object does. Thus one might

relate sentence (1), but not sentence (a), to the question given

in (3).

(3) what did John do to the table?

Furthermore, while sentence (1) has sentence (4) as a paraphrase,

sentence (5) is not a paraphrase of sentence (2).

(4) what John did to the table wcs ruin it

(5) *what John did to the table was build it14

1
This observation is due to Paul M. Postal.

To give another example, note that both of the relationships in

question may be seen in sentence (6) but that only in one of the

two senses is sentence (6) a paraphrase of sentence (7).

(6) John paints nudes

(7) what John does to nudes is paint them

 - 9 -

There is polysemy in the direct olject of (6), true, but the

difference also lies in whether the objects John painted existed

before or after he did the painting.

I am going to suggest below that there are many semantically

relevant syntactic relationships involving nouns and the structures

that contain them, that these relationships--like those seen in (1)

and (2)--are in large part covert but are nevertheless empirically

discoverable, that they form a specific finite sets and that

observations made about them will turn out to have considerable

cross-linguistic validity. I shall refer to these as 'case' rela-

tionships.
1. Earlier approaches to the study of case

Books written to introduce students to our discipline

seldom fail to acquaint their readers with the 'wrong' ways of

using particular case systems as universal models for language

structure. Grammarians who accepted the case system of Latin or

Greek as a valid framework for the linguistic expression of all

human experience were very likely, we have been told, to spend a

long time asking the wrong kinds of questions when they attempted

to learn and describe Aleut or Thai. We have probably all

enjoyed sneering, with Jespersen, at his favorite 'bad guy',

Sonnenschein, who, unable to decide between'Latin and Old English,

allowed modern English teach to be described as either taking a

Dative and an Accusative, because that was the pattern for Old

English tocsin, or as taking two Accgsatives, in the manner of

Latin doceo and German lehren. 15

15J. Otto Jespersen, THE PHILOSOPHY OF GRAMMAR (1924), Norton

Paperback Reprint (1965), Pp. 357: 175.

Looking for one man's case system in another man's language

is not, of course, a good example of the study of case.. The

apprellMhes to the study of case that do need to be taken seriously

are of several varieties. Many traditional studies have examined,

in somewhat semantic terms, the various uses of case. More recent

work has been directed toward the analysis of the case systems of

given languages, under the assumptions suggested by the word 'system.'

A great deal of research, early and late, has been devoted to an

understanding of the history or evolution of case notions or of

case morphemes. And lastly, the generative grammarians have for

the most part viewed case markers as surface structure reflexes,

introduced by rules, of various kinds of deep and surface

syntactic relations.

1.1 Case uses

The standard handbooks of Greek and Latin typically devote much

of their bulk to the classification and illustration of semantically

different relationships representable by given case forms. The

sub-headings of these classifications are most commonly of the form

'X of where 010 is the name of a particular case and 'Y' is the

name for a particular 'use' of X. The reader will recall such terms
16
as 'Dative of Separation', 'Dative of possession', and so on.

Milli=111ENW
1
For an extensive description of this type, see Charles Bennett,
SYNTAX OF EARLY LATIN, II: THE CASES, Boston (1914) Pp. 409.

Apart from the fact that such studies do not start out from

the point of view of the centrality of syntax, the major defects

of these studies were (a) that the Nominative was largely ignored

and (b) that classificatory criteria were often confused which

ought to have been kept distinct.

The neglect of the nominative in studies of case uses probably

has several sources, one being the etymological meaning ("deviation")

of the Greek term for case, ptOsis, which predisposed grammarians

to limit the term only to the non-nominative cases. The most impor-

tant reason for omitting the nominative in these studies, however,

is the wrongly fssumed clarity of the concept 'subject of the

sentence.' Mi1111... published a study of nominative and accusative

case uses in Lat4a, in 1908, in which he devoted 170 some pages

to the accusative, somewhat less than one page to the nominative,

explaining that "Die beiden casus recta., der Nominativ and der

Vokativ, sind bei dem Streite fiber die Kasustheorie nicht beteiligt.

Im Nominativ steht das Subjekt, von dem der Satz etwas aussagt."17

411111111M10, 111=111111111IMIP.MliVir AE113.11

17C. F. W. Willer, SYNTAX DES NOMINATIVS UND AKKUSATIVS IM LATEINI-
SCHEN, Leipzig (1908), Pp. 175; 1.

The role of the subject was regarded as so clear to Sweet that

he went so far as to claim that the nominative is the only case where

one can speak properly of a 'noun'. A sentence is viewed as a kind

of predication on a given noun, and every noun-like element in a

sentence other than the subject is really a kind of derived adverb,

a part of the predication.18

Quoted in Jespersen (1924) 107.

1111111111Nalt
On a little reflection, however, it becomes obvious that seman-

tic differences in the relationships between subjects and verbs are

of exactly the same order and exhibit the same extent if variety

as can be found in the other cases. There is in principle no reason

why the traditional studies of case uses fail to contain such classi-

fications as 'Nominative of Personal Agent', 'Nominative of Patient',

'Nominative of Beneficiary', 'Nominative of Affected Persou', and

'Nominative of Interested Person' (or, possibly, 'Ethical Nominative')

for such sentences as (8) to (12) respectively.

 (8) he hit the ball

(9) he received a blow

(10) he received a gift

(11) he loves her

(12) he has black hair

The confusion of criteria in treatments of the uses of cases

19
has been documented by de Groot in his study of the Latin genitive.

19
A. Willem de Groot, "C1assification of uses of a case illustrated
on the genitive in Latin," LINGUA, VI (1956) 8-66.

Uses of cases are classified on syntactic grounds, as illustrated

by the division of uses of the genitive according to whether the

genitive noun is in construction with a noun, an adjective or a

verb; on historical grounds, as when the uses of the syncretistic

Latin ablative case are divided into three classes, separative,

locative and instrumental; and on semantic grounds, in which there

is a great deal of confusion between meanings that can properly be

thought of as associated with the case forms of nouns, on the one

hand, and meanings that properly reside in neighboring words.

De Groot's dritical treatment of the traditional classification

of Latin genitive case uses is particularly interesting from the

point of view taken here, because in his 'simplification' of the

picture he rejects as irrelevant certain phenomena which generative

grammarians would insist definitely are of syntactic importance.

He claims, for example, that the traditional studies confuse differ-

ence of referents with differences of case uses. Thus, to de Groot

the traditional three senses of statua Mvronis (the statue possessed

 - 14 -

by Myro--genitivus possessivus; statue sculpted by Myro--

genitivus subjectivus; statue depicting Myro--genitive of

represented subject) as well as the subjective and objective

senses of amor patris, are differences in practical, not in

linguistic facts. From arguments :such as this he is able to combine

twelve of the classical 'uses' into one, which he then labels the

'proper genitive', asserting that "the proper genitive denotes, and

"20
consequently can be used to refer to, ,am thing-to-thing relation.

20
de Groot (1956) 35.
1111111MP,
He ends by reducing the thirty traditional 'uses of the genitive'
21
to eight , of which two are rare enougit to be left out of consider-

2
From de Groot (1956) 30:
I. Adjunct to a noun
A. Proper genitive, eloquentia hominis
B. Genitive of quality, homo magnae eloquentiae
II. Adjunct to a substantivalr
C. Genitive of the set of persons, ,reliqui ,editum
III. Conjunct ('complement') of a copula
D. Genitive of the type of person, Ltspientis est aperte
odisse
IV. Adjunct to a verb
E. Genitive of purpose, Aegyptum profiscitue cognoscende
antiquitatis
F. Genitive of locality, Romae consules creabantur
IVa. Adjunct to a present participle
G. Genitive with a present participle, laboris fugiens
V. Genitive of exclamation, mercimoni lepidi

ation, and a third, 'genitive of locality', is really limited to

specific place names.

Benveniste's reply to de Groot's analysis appeared in the issue

22
of LINGUA dedicated to de Groot . There he proposes still further

22 Emile Benveniste, "Pour l'analyse des fonctions casuelles: le

genitif latin," LINGUA, XI (1962) 10-18.
simplifications of the classification. Noting that de Groot's

'genitive of locality' applies only to proper place names, that

is, that it occurs only with place-names having -o- and stems,

in complementary distribution with the ablative, Benveniste wisely

suggests that this is something that should be catalogued as a fact

about place names, not as a fact about uses of the genitive case.

Benveniste's conclusions on the remaining genitive constructions

is quite congenial to the generative grammarian's position. He

proposes that the so-called 'proper genitive' basically results

from the process of converting a sentence into a nominal. The

meaning distinction between 'genitivus subjectivus' and 'genitivus

objectivus' constructions merely reflects the difference between

situations in which the genitive noun is an original subject or

and those where it is an original object, the genitive representing

a kind of neutralization of the nominative/accusative distinction

23
found in the underlying sentences.

23It must be said, however, that Benveniste's desentential inter-

pretation is diachronic rather than synchronic, for he goes on to
explain that it is on analogy from these basic verbal sources that
new genitive relations are created. From ludus pueri and risus
pueri, where the relation to ludit and ridet is fairly transparent,
the pattern was extended to include somnus pueri, mos pueri, and
finally liber pueri. The genitive grammarian may be inclined to
seek synchronic verbal connections--possibly through positing
abstract entities never realized as verbs--for these other geni-
tives too. (See Benveniste (1962) 17.)

At least from the two mentioned studies of uses of the Latin

genitive, it would appear (a) that some case- uses are purely irre-

gular, requiring as their explanation a statement of the idio-

syncratic grammatical requirements of specific lexical items, and

 - 16 -

(b) that some semantic differences are accounted for independently

of assigning 'meanings' to cases: either by recognizing meaning

differences in 'governing' words, or by noting leaning differences

in different underlying sentences. The suggestion that one can

find clear special meanings associated with surface cases fails

to receive strong support from these studies.

1.2 Case systems

There are reasonable objections to approaching the case system

of one language from the point of view of the surface case system

of another (say, Classical Latin) by merely checking off the ways

in which a given case-relation in the chosen standard is given

expression in the language under observation. An acceptable alter-

native, apparently, is the inverse of this process: one identifies

case morphemes in the new language within the system of noun

inflection, and then relates each of these to traditional or

24
'standard' case notions. To take just one recent example, Redden

2
James E. Redden, "Walapai II: morphology," INTERNATIONAL JOURNAL
OF AMERICAN LINGUISTICS, XXXII (1966), 141-163.

finds five case indices in Walapai (four suffixes and zero) and

identifies each of these with terms taken from the tradition of

case studies: -6 is nominative, .1 is accusative, -k is allative/

adessive, -1 is illative/incssive, and -m is ablative/abessive.

Under each of these headings the author adds information about

the uses of the case-forms that may not be deducible from the

labels themselves. Nominative, for example, occurs only once in

 - 17 -

a simple sentence--coordinate conjunction of subject nouns

requires use of the -m suffix on all the extra nouns introduced;

accusative is used with some noun tokens which would not be

considered direct objects in English; allative/adessive has a

partitive function; and ablative/abessive combines ablative,

instrumental, and comitative functions.

In a study of this type,slace what is at hand is the surface

structure of the inflection system of Walapai nouns, the descrip-

tive task is to identify the surface case forms that are distinct

from each other in the language and to associate 'case functions'

with each of these. What needs to be emphasized is (a) that such

a study does not present directly available answers to such questions

as "how is the indirect object expressed in this language?" (i.e.,

the system of possible case functions is not called on to provide

a descriptive framework), and (b) that the functions or uses them-

selves are not taken as primary terms in the description (i.e., the

various 'functions' of the 'ablative/abessive' suffix -m are not

interpreted as giving evidence that several distinct cases merely

happen to be homophonous).25

25-These remarks are not intended to be critical of Redden's study.

Indeed, in the aosence of a universal theory of case relationships
there is no theoretically justified alternative to this approach.

One approach to the study. of case systems, then, is to restrict

oneself to a morphological description of nouns and to impose no

constraints on the ways in which the case morphemes can be identi-

fied with their meanings or functions. This is distinct from studies

of case systems which attempt to find a unified meaning for each

 -18

of the cases discovered in the language being studied. An

example of the latter approach is found in the now discredited

'localistic' view of the cases in Indo-Eurcpean, by which

dative is 'the case of rest,' accusative 'the case of movement

26
to,' and genitive 'the case of movement from.' And recent

interpretation, discussed briefly in Jespersen (1924) p.

-276This
186, appears to date back to the Byzantine grammarian Maxime Planude.
INI110.0111MIO

attempts to capture single comprehensive 'meanings' of the cases

have suffered from the vagueness and circularity expected of any

attempt to find semantic characterizations of surface-structure

phenomena.27

27
As an illustration of this last point, take Gonda's claim that
the Vedic dative is called for whenever a noun is used to refer to
the 'object in view.' The vacuity of this statement is seen in his
interpretation of
(i) vataya kaptla vidyut (Patanjali)
'a reddish lightning signifies wind'
as "the lightning has, so to say, wind in view". J. Gonda, "The
unity of the Vedic dative," LINGUA XI (1962) 141-150; 147.

28 29
The well-known studies of Hjelmslev and Jakobson are

28
Louis Hjelmslev, "La cat6gorie des cas," ACTA JUTLANDICA, Aarhus,
VII.1 (1935) i-xii, 1-184; and IX.2 (1937) i-vii, 1-78.

29
Roman Jakobson, "Beitrag zur allgemeinen Kasuslehre," TRAVAUX
DU CERCLE LINGUISTIQUE DE PRAGUE, VI (1936) 240-288; reprinted in
E. P. Hamp, F. W. Householder and R. Austerlitz, eds., READINGS
IN LINGUISTICS, II, University of Chicago Press (1966) 51-89.

attempts not only to unliover unified meanings of each of the cases,

but also to show that these meanings themselves form a coherent

 -19-

system by their decomposability into distinctive oppositions.

The possibility of vagueness is of course the more increased inas-

much as the number of oppositions is less than the number of

30
cases.

30 See, in this regard, the brief critical remarks of A. H. Kuipers,

in "The Circassian nominal paradigm: a contribution to case theory,"
LINGUA, XI (1962) 231-248; 231.

The difficulties in discovering a unified meaning for each of

the cases in a case system has led to alternative view: that all

but one of the cases can be given more or less specific meanings,

the meaning of the residual case being left open. This residual

case can either have whatever relation to the rest of the sentence

is required by the meanings:,of the neighboring words, or it can

serve any purely case-like function not pre-empted by the other

cases. Goedicke is said by Bennett to have explained the accusa-

tive as "the case used for those functions not fulfilled by the

other cases." The fact that Bennett, following Whitney, ridiculed

this view on the grounds that case could be so described,

suggests that Goedicke's remark could not have been very clearly
31 32
expressed. A different approach is taken by Diver , who does

31
Bennett (1914) 195, fn. 1. At the time of writing I have not
yet had access to the Goedicke original.

3 2William Direr, "The system of agency of the Latin noun," WORD,

XX (1964) 178-196.

not assign the 'left-over' function to a particular case as such,

but to whatever case or cases are not required for a given realiza-
 -20-

tion of what he calls the 'agency system'. Briefly* and ignoring

his treatment of passive sentences, Diver's analysis is this: A

verb can have one, two, or three nouns (or noun phrases) associated

with it, corresponding generally to the intransitive, normal tran-

sitive, and transitive indirect object sentence types respectively.

In a three-noun sentence, the nouns are nominative, dative and

accusative, the nominative being the case of the Agent, the accusa-

tive the case of the Patient, with the dative, as the 'residue' case,

capable of expressing any notion compatible with the meaning of the

remainder of the sentence. The function of the dative in a three-

noun sentence, in other words, is 'deduced' from the context, not

present as one of a number of possible 'meanings' of the dative case. 33

33The following is from Diver (1964) 181: "In the sentence senatus
imperium mihi dedit 'the senate gave me supreme power', therm7M-
tive, with the syntactic meaning of Agent, indicates the giver;
the Accusative, with the syntactic meaning of Patient, indicates
the gift. The question is: Does the Dative itself indicate the
recipient or merely that the attached word is neither the giver
nor the gift?" Diver makes the latter choice. In particular, he
states that "knowing that mihi, in the Dative, can be neither the
Agent (the giver) nor the TaTent (the gift), we deduce that it is
the recipient."

In two-noun sentences, one of the nouns is nominative, the other

either dative or accusative, but typically accusative. The nomina-

tive here is the case of the agent, but this time the accusative

(or the dative, whichever occurs) is the residue case. In a two-

noun sentence, in other words, the accusative is not limited to

the meaning of Patient but can express any number of other meanings

as well. And, since it no longer contrasts with dative, it can be

replaced by a dative. The choice between dative and accusative in

 - 21 -

two-noun sentences, since it is not semantically relevant, is

subject to random kinds of free and conditioned variation.

Carrying the argument through, the noun found in a one-noun

sentence can express any meaning relationship with the verb. The

noun, though most frequently nominative, may be accusative or

dative, but the choice is not due to meanings associated with

these cases. When the noun is nominative its 'syntactic meaning'

may be that of Agent, Patient, or anything else.

The inadequacy of Diver's treatment is clear. In the first

place, it seems unlikely that, as used in Diver ^ paper, the

notions Agent and Patient are in any sense satisfactory semantic

primitives. To agree that imperium in senatus imperium mihi dedit

is the Patient is nothing more than to agree to say the word 'Patient'

on seeing an accusative form in a three-noun sentence. For many

of Diver's examples, his argument would have been every bit as con-

vincing if he had said that there is an unvarying function performed

by the dative, but the role of the accusative depends on such matters

as the lexical meaning of the verb. Furthermore, the "couple of

dozen verbs" which appear in two-noun sentences and which exhibit

some kind of semantic correlation involving the supposedly non-

significant choice of accusative or dative should probably not be

set aside as unimportant exceptions.

Diver's proposal may be thought of as an attempt to identify

the semantic contribution of cases seen as syntagmatically identi-

fied entities, while the positing of distinctive oppositions, in

the manner of Hjelmslev and Jakobson, is an attempt to see the

functioning of cases from the point of view of the concept of

 - 22-

paradigmatic contrast. The latter view has been criticized by

Kurylowicz . The apparent contrast seen in Polish and Russian

3 Jerzy Kury owicz, "Le problme du classement des cas" (1949), in

his collection ESQUISSES LINGUISTIQUES, WrocZaw-Krakov (1960) 131-
150; 134, 141.

between accusative and genitive (partitive) direct object, as

between (13) and (14)

(13) daj nam chleb 'give us the bread'

(14) daj nam chleba 'give us some bread'

is not a difference in the syntactic function of the object nouns

relative to the verb, but is rather a difference which falls into

that area of syntax that deals with the effect of the choice of

article, in languages having articles, on the semantic content of

the associated noun. The fact that in Russian the difference is

reflected as a difference in noun inflection does not along deter-

mine its character as a part of the case system proper of the

language.

The vertical contrast between locative and accusative nouns

after locative/directional prepositions, as in (15) and (16)

(15) on prygajet na stole 'he jumps (up and down) on

the table'

(16) on prygajet na stol 'he jumps onto the table'

is a difference that would be discussed in transformational-grammar

terms as involving a distinction between preposition-phrases which

are inside and those which are outside the verb phrase constituent.

That is, a Locative preposition-phrase which occurs outside the

 -23-

constituent VP is one which indicates the place where the

action described by the VP takes place. A Locative preposition-

phrase inside the VP is a complement to the verb. Inside a VP

the difference between the locative and directional senses is

entirely dependent on the associated verb; outside the VP the

sense is always locative.

Kurylowicz discussed (15) and (16) in essentially the same

terms. To him the directional phrase na stol is 'more central'

to the verb than the locative phrase na stole. An apparent con-

trast appears just in case the same verb may appear sometimes with

and sometimes without a locative (or directional) complement.

There is thus no genuine paradigmatic contrast found in such pairs

as (13)-(14) or (15)-(16).

Kurylowicz's own approach to the study of case systems brings

another order of grammatical fact into consideration: sentence

relatedness. Cases, in his view, form a network of relationships

mediated by such grammatical processes as the passive transformation.

The distinction between nominative and accusative, for example, is

a reflection in the system of cases of the more basic distinc-

tion between passive and active sentences. In his terms, hostis

occiditur becomes the predicate hostem occidit, the primary change

from occiditur to occidit bringing with it the concomitant change

from hostis to hostem.

Nominalizations of sentences have the effect of relating both

accusative and nominative to the genitive, for the former two are

neutralized under conversion to genitive, as illustrated by the

change from plebs secedit to secessio plebis (genitivus subjectivus)

 - 24 -

as opposed to the change from hostem occidere to occisio hostie

(genitivus objectivus).

The relationship between nominative and accusative, then, is

a reflex of diathesis; the relationship of these two to genitive

is mediated through the process of constructing deverbal nouns.

The remaining cases--dative, ablative, instrumental and locative- -

enter the network of relationships in that, secondarily to their

functions as adverbials, they each provide variants of the accusa-

tive with certain verbs. That is, there are verbs that 'govern'

the ablative (e.g., utor), say, rather than the accusative for
35
their 'direct objects'.

1
35 Kurylowicz (1960) 138-139; 144-147; 150. Also se, his THE
INFLECTIONAL CATEGORIES OF INDO-EUROPEAN, Heidelberg-Carl Winter
(1964) Pp. 246; 179-181. Somewhat similar interpretations of the
connections between case and diathesis are found in Klaus Heger,
"Valenz, Diathese and Kasus," ZEITSCHRIFT FUR ROMANISCHE PHILOLOGIE
(1966) 138-170; 161.

1.3 Case histories

In addition to studies of case uses and interpretations of

the cases in a given language as elements of a coherent system,

the literature also contains many historical studies of cases;

and these, too, are of various kinds. Some workers have sought

to discover the original meanings of the cases of a language or

family of languages, while others have sought to trace case morphemes

back to other kinds of morphemes--either syntactic function words

or some kind of derivational morphemes. Still others have seen

 -25-

in the history of one case system a case system of a different

type--with or without assumptions concerning the 'essential

primitivity' of the earlier type.

A very common assumption among linguistic historians has

been that case affixes are traceable back to non-case notions.

The form which eventually became the Indo-European case ending

representing nominative singular masculine, i.e. *-s, has been

interpreted the demonstrative *so which had been converted into

a suffix indicating a definite subject; and the *so in turn is

believed by some to have originated as a Proto-Indo-Hittite sen-

tence connective.36

'George s . Lane, review of Y. M. Biese, SOME NOTES ON THE ORIGIN

OF THE INDO-EUROPEAN NOMINATIVE SINGULAR, Helsinki (1960) Pp. 15;
in LANGUAGE XXVII (1951) 372-374.
=111031

The same form has also been interpreted as a derivational morpheme

indicating a specific individual directly involved in an activity,

contrasting with a different derivational affix *-m indicating a

37 Scholars who can
non-active object or the product of an action.

37See, for example, the statement by Winfred P. Lehmann in his

"On earlier stages of the Indo-European nominal inflection,"
LANGUAGE, XXXIV (1958) 179-202; 190.

rest with the latter view are those who do not require of them-

selves the belief that 'synthetic' languages necessarily have

38
antecedent 'analytic' stages.

36The impression is sometimes given that the identification of the

etymon of a case affix brings with it an account of the intellectual
evolution of the speakers of the language in question. If the
 - 26 -

interpretation of *-m and *-s as derivational morphemes is correct,

it does not follow that one has discovered, in the transition from
the earliest function of these elements to their later clear case-
like uses any kind of 'abstraction' process, or tendency to pass
from 'concrete' to 'relational' modes of thought. Our methods of
reconstruction should certainly make it possible to detect basic
(i.e., deep structure) linguistic evolutionsif it is there to
discover, but the etymology of surface-structure morphemes should
not lead to assumptions about deep typological differences. What
I mean is that the underlying case structures of Proto -Indo-
European may have been just as precisely organized as those of any
of the daughter languages, and that the changes that have occurred
may have been entirely a matter of morphophonemic detail. From
the preponderance of (derived) active nouns in subject position,
one generation may have 're-interpreted' the suffix as a marker
of human subject; a later generation may have re-interpreted it
as merely a marker for the subject use of a particular set of
words--to state the possibilities in the most simple-minded way.
The change, in short, may well have been entirely in the economies
of bringing to the surface underlying structural features which
themselves underwent no change whatever.
SEN.MIMMIC,_

A second kind of speculation on historical changes within

case systems traces case systems of one kind back to case systems

of another kind. Of particular interest here is the suggestion

that the Indo-European case systems point back to an original

'ergative' system. Case typologies will be discussed in slightly

greater detail below, but briefly we can characterize an 'ergative'

system as one which assigns one case (the Ergative) to the subject

of a transitive verb, another to both the subject of an intransi-

tive verb and the object of a transitive verb. An 'accusative'

system, on the other hand, is one which assigns one case to the

subject of either transitive or intransitive verbs, another (the

Accusative) to the object of a transitive verb. A common feature

of ergative systems is that the 'genitive' form is the same as

the Ergative (or, put differently, that the Ergative case has a

'genitive' function).
 - 27 -

The connection of Indo-European *-s with animateness (the

subject of a transitive verb is typically animate), the original

identity of the nominative singular *-s with the genitive ending,

and the identity of the neuter ending *-m with the masculine

accusative form, have led many investigators to the conclusion

that our linguistic ancestors were speakers of an 'ergative'

language. 39 It will be suggested below that, if such a change

39See particularly C. C. Uhlenbeck, "Agens and Patiens im

Kasussystem der indogermanischen Sprachen," INDOGERMANISCHE
FORSCHUNGEN, XII (1901), 170-171, where the *-m ending was identi-
fied as a subject marker, the *-s as the agent marker in passive
sentences ( a common interpretation of 'ergative' systems); and
Andre Valliant, "L'ergatif indo-europeen," BULLETIN DE LA SOCIETE
LINGUISTIQUE DE PARIS, XXVII (1936) 93-108. Lehmann finds the
arguments unconvincing, noting for example that evidence of an
'ergative' ending cannot be found in plural nouns or in feminines
in a. (Lehmann (1958) 190.)

has taken place, it is a change which involves the notion 'subject'.

1.4 Case in current generative grammar

A hitherto largely unquestioned assumption about case in the

writings of generative grammarians has been made explicit by Lyons:

"'case' (in the languages in which the category is to be found) is

not present in 'deep structure' at all, but is merely the inflex-

ional 'realization' of particular syntactic relationships." 40

40
Lyons (1966) 218.

The syntactic relationships in question may in fact be relation-

ships that are defined only in the surface structure, as when the

surface subject of a sentence (destined to assume, say, a

 .28

'nominative' form) has appeared as the result of the application

of the passive transformation, or when the 'genitive' marker is

introduced as an accompaniment to a nominalization transformation.

One of Chomsky's few remarks on case is in a discussion of the

peripheral nature of stylistic inversions; although case forms

are assigned to English pronouns relatively late in the grammar,

determined largely on surface structure position, the stylistic

inversion rules are later still. In this way it becomes possible

to account for such forms as him I like, etc.; the shift of him

to the front of the sentence must follow the assignment of case

forms to the pronouns. 41

Chomsky (1965) 221f.

It seems to me that the discussion of case could be seen in

a somewhat better perspective if the assignment of case forms

were viewed as exactly analogous to the rules for assigning pre-

positions in English, or postpositions in Japanese. 42 There are

42
The suggestion is of course not novel. According to Hjelmslev,
the first scholar to show a connection between prepositions and
cases was A.-F. Bernhardi, in ANFANGSGRUNDE DER SPRACHWISSENSCHAFT,
Berlin (1805); see Hjelmslev (1934) 24. I have not yet seen the
Bernhardi work.

languages which use case forms quite extensively, and the assump-

tion that the case forms of nouns can be assigned in straightforward

ways on the basis of simply defined syntactic relations seems to

be based too much on the situation with English pronouns.

Prepositions in English - -or the absence of a preposition

before a noun phrase, which may be treated as corresponding to a

zero or unmarked case affix--are selected on the basis of several

types of structural features, and in ways that are exactly analogous

toth0.56 which determine particular case forms in a language like

Latin: identity as (surface) subject or object, occurrence after

particular verbs, occurrence in construction with particular nouns,

occurrence in particular constructions, and so on. The only diffi-

culties in thinking of these two processes as analogous are that

even the most elaborate case languages may also have combinations

of, say, prepositions with case forms, and that some prepositions

have independent semantic content. The first of these difficulties

disappears if, after accepting the fact that the conditions for

choosing prepositions are basically of the same type as those for

choosing case forms, we merely agree that the determining condi-

tions may simultaneously determine a preposition and a case form.

The second difficulty means merely that a correct account will

allow certain options in the choice of prepositions in some con-

texts, and that these choices have semantic consequences. Analogous

devices are provided by the 'true' case languages, too, for example

by having alternative case choices in otherwise identical construc-

tions, or by having semantically functioning prepositions or post-

positions.

The syntactic relations that are involved in the selection of

case forms (prepositions, affixes, etc.) are?in practice, of two

types, and we may call these 'pure' or 'configurational' relations,

on the one hand, and 'labeled' or 'mediated' relations on the

43
other hand. 'Pure' relations are relations between grammatical

3
The distinction would be more accurately represented by the
opposition 'relations' vs. 'categories', because when a
 -30-

phrase-structure rule introduces a symbol like Manner or Extent,

symbols which dominate manner adverbials and extent phrases, these
symbols function, as far as the rest of the grammar is concerned,
in exactly the same ways as such 'intentional' category symbols
as S or NP. This fact has much more to do with the requirements
of the phrase-structure model than with the ' categorial' character
of the grammatical concepts involved. In an earlier paper I
discussed the impossibility of capturing in a base component of
a grammar of the type presented in Chomsky (1965) both such infor-
mation that in a clumsy wavy is a manner adverbial (and as such
represents an instance of highly constrained lexical selection
as well as a quite specific positional and co-occurrence potential
which it shares with other manner adverbials) and that it is a
preposition phrase. See my "Toward a modern theory of case,"
THE OHIO STATE UNIVERSITY PROJECT ON LINGUISTIC ANALYSIS REPORT
NO. 13 (1966) 1-24.
The intention on the part of grammarians who have introduced
such terms as Temp, Extent, etc., into their rules is to let'
these terms represent relations between the phrases which they
dominate and some other element of the sentence (e.g., the VP as
a whole); nobody, as far as I can tell, has actually wished these
terms to be considered as representing distinct types of grammati-
cal categories on the order of NP or Preposition Phrase.

constituents expressible in terms of (immediate) domination.

Thus, the notion 'subject' can be identified as the relation

between a NP and an immediately dominating S, while the notion

'direct object' can be equated with the relation that holds between

a NP and an immediately dominating VP. Where the relation 'subject

of is understood to hold between elements of the deep structure,

one speaks of the deep-structure subject; where it is understood

to hold between elements of the (pre-stylistic) surface structure,

one speaks of the surface-structure subject. This distinction

appears to correspo2A. to the traditional one between 'logical

subject' and 'grammatical subject'.

By 'labeled' relation I mean the relation of a NP to a

sentence, or to a VP, which is mediated by a pseudo-cateogry label

such as Manner, Extent, Location, Agent, etc.

 -31-

It is clear that if all transformations which create

surface subjects have the effect of attaching a NP directly

to a S, under conditions which guarantee that no other NP is

also directly subjoined to the same S and if it always turns out

that only one NP is subjoined to a VP in the pre-stylistic surface

structure, then these two 'pure' relations are exactly what deter-

mine the most typical occurrences of the case categories 'nomina-

tive' and 'accusative' in languages of a certain type. For remain-

ing case forms, the determination is either on the basis of idiosyn-

cratic properties of specific governing words, or on the basis of

a 'labeled' relation, as when the choice of la is determined by

reference to the dominating category Extent in the extent phrase

of sentences like (17).

(17) he missed the target by two miles

In my earlier paper (see fn. 43) I pointed out that there is

no semantically constant value associated with the notion 'subject

oft (unless it is possible to make sense of the expression 'the

thing being talked about', and, if that can be done, to determine

whether such a concept has any connection with the relation 'subject'),

and that there are no semantically relevant relations residing in

the surface subject relation which are not somewhere also expressible

by 'labeled' relations. The conclusion I have drawn from this is

that all semantically relevant syntactic relations between NPi

and the structures which contain them must be of the 'labeled' type.

The consequences of this decision include (a) the elimination of

the category VP, and (b) the addition to some grammars of a rule,

or system of rules, for creating 'subjects'. The relation 'subject',

 - 32 -

in other words, is now seen as exclusively a surface-structure

phenomenon.
 -33-

2. Some preliminary conclusions

I have suggested that there are reasons for questioning

the deep-structure validity of the traditional division between

subject and predicate, a division which is assumed by some to

underlie the basic form of all sentences in all languages. The

position I take seems to be in agreement with that of Tesniere,

who holds that the subject/predicate division is an importation

into linguistic theory from formal logic of a concert which is

not supported by the facts of language, and, furthermore, that

the division actually obscures the many structural parallels between

44
'subjects' and 'objects'. The kinds of observations that some

4
Tesniere (1959) 103-105.

scholars have made about surface differences between 'predicative'

45
and 'determinative syntagms1 may be accepted without in any way

45
See, for example, C. E. Bazell, "Syntactic relations and ling-
uistic typology," CAHIERS FERDINAND DE SAUSSURE, VIII (1949) 5-20;
esp. p. 8 where the difference is expressed in such terms as
'degrees of cohesion', 'liaison features' found within the predi-
cate but not between subject and predicate.

requiring one to believe that the subject/predicate division plays

a part in the deep structure syntactic relations among the consti-

tuents of sentences.

Once we have interpreted 'subject' as an aspect of the surface

structure, claims about so-called 'subjectless' sentences in lang-

uages which have superficial subjects in some sentences, or

reports about languages which appear to lack entirely entities

corresponding to the 'subjects' of our grammatical tradition, no

longer need to be regarded as particularly disturbing. Unfor-

tunately, there are both good and bad reasons for asserting that

particular languages or particular sentences are 'subjectless',

and it may be necessary to make it clear just what it is that I

am claiming. A distinction must be drawn between not havin% a

constituent which could properly be called 'subject', on the

one hand, and losin such a constituent by anaphoric deletion,

46
on the other hand. Robins, in his review of Tesniere (1959),

The tagmemicists in particular, because of their notation for

'optional' constituents, have had to come to grips with this
distinction. A 'tagmemic formula' may be thought of as an attempt
to present in a single statement a quasi-generative rule for produ-
cing a set of related sentences and the surface structure (short
of free variation in word order)of these sentences. If the
formulas for transitive and intransitive clauses are expressed
as (i) and (ii) respectively
(i) +Subj +Pred +Obj +Loc +Time
(ii) 7subj +Pred +Loc Time
it is clear (0 that any clause containingAa Pred can satisfy either
of these formulas, and (b) that the potential appearance of such
constituents as Loc and Time are less relevant to the description
of these clauses than is that of the constituent Obj. Pike draws
a distinction, which cross-cuts the optional/obligatory distinction,
between 'diagnostic' and 'non-diagnostic' elements of clauses, as,
e.g., in his recent H.E.W. report, TAGMEMIC AND MATRIX LINGUIST1C$
APPLIED TO SELECTED AFRICAN LANGUAGES (1966), esp. Chapter 1:
Clauses. Grimes, on the other hand, seems to suggest introducing
the 'diagnostic' constituents obligatorily, allowing for their
deletion under certain contextual or anaphoric conditions. See
Joseph E. Grimes, HUICHOL SYNTAX, Mouton (1964) Pp. 105, esp. 16f.

accuses Tesniere of failing to isolate the subject from the rest

of the sentence. To Robins, Tesnik,e's decision to allow the

subject to be treated as being merely a complement to the verb

must be related to the fact that the subject is omissible in

such languages as Latin.47 Ifihitrue that the omissibility of

 -35-

7R H. Robins, "Syntactic analysis," ARCHIVUM LINGUISTICUM, XIII

(1961) 78-89. (Reprinted in Hamp, Householder and Austerlitz,
READINGS IN LINGUISTICS II, Chicago (1966) 386-395.)

'subjects' is what convinced Tesniere that they are subordinated

to verbs, and if the non-omissibility in any language of the sub-

ject constituent would have persuaded him that there is a special

status for 'subject' vis a vis 'predicate' in the underlying

structure of sentences in all languages, then that, it seems to

me, is a bad reason for coming up with what might be a correct

analysis.

It seems best to have a place in linguistic theory for the

operation of anaphoric processes, processes which have the effect

of shortening, simplifying, de-stressing sentences which are partly

identical to their neighbors (or which are partly 'understood').

It happens that English anaphoric processes make use of pronominali-

zation, stress reduction, as well as deletion, under conditions

where other languages might get along exclusively with deletion.48

For an extremely informative description of these processes in

English, see Lila Gleitman, "Coordinating conjunctions in English,"
LANGUAGE XLI (1965) 260-293; also see Zellig S. Harris, "Co-
occurrence and transformation in linguistic structure," LANGUAGE
XXXIII (1957) 283-340, esp. section 16.

tinder some conditions, in languages of the latter type, the deleted

element happens to be the 'subject'. The non - occurrence of subject

nouns in some utterances in some languages is not by itself, in

other words, a good argument against the universality of the

subject/predicate division. There are better ones. Some of these

have already been suggested, others are to appear

shortly.

By distinguishing between surface and deep structure case

relationships, by interpreting the 'subject' and 'object' as

aspects of the surface structure, and by viewing the specific

phonetic shapes of nouns in actual utterances as determinable

by many factors, vastly variable in space and time, we have

eliminated reasons for being surprised at the non-comparability

of (surface) case systems. We find it partly possible to agree

with Bennett when, after surveying a few representative nine-

teenth century case theories, he stated that they erred in

sharing the "doubtful assumption that all the cases must

belong to a single scheme, as though parts of some consistent insti-

49
tution". We need not follow him, however, in concluding that

Bennett (1914) 3.

the only valid type of research into the cases is to determine

the earliest value of each case, taken one at a time.

Greenberg has remarked that cases themselves cannot be compared

across languages--two case syktems may have different numbers of

cases, the names of the cases may conceal functional differences- -

but that case uses may be expected to be comparable. He predicts,

for instance, that the uses of cases will be "substantially

similar in frequency but differently combined in different lang-

"50
uages. Greenberg's recommendations on the cross-linguistic

50
Greenberg (1966) 80, 98; the quotation is from p. 98.
 -37-

study of case uses were presented in connection with the 'true'

case languages, but it seems clear that if a 'Dative of Personal

Agent' in one language can be identified with an 'Ablative of Per-

sonal Agent' in another language, then the 'Personal Agent'

relationship between a noun and a verb ought also to be recogni-

zable in the so-called 'caseless' languages on exactly the same

grounds. If, furthermore, it turns out that other grammatical

facts can be associated with sentences containing the Personal

Agent relationship, it would appear that the concepts underlying

the study of case uses may have a greater linguistic significance

than those involved in the description of surface case systems.

These additional facts might include the identification of a

limited set of nouns and a limited set of verbs capable of enter-

ing into this relationship, and whatever additinnal generalizations

prove to be statable in terms of this classification. Higher level

dependencies may be discovered, such as the limitation of Bene-

factive phrases to sentences containing a Personal Agent relation-

ship in their deep structure.

The question should now be asked, of course, whether we are

justified in using the term 'case' for the kind of remote

syntactic-semantic relations that are at issue. There is among

many scholars a strong feeling that the term should be used only

where clear case morphemes are discoverable in the inflection of

nouns. To Jespersen, it is wrong to speak of 'analytic' cases,

even when there is no 'local' meaning in the preposition phrases,

because cases are one thing and preposition-plus-object constructions

 -38-
51
are another. Jespersen's position is colored a little by his

51
Jespersen (1924) 186.

belief that the caselessness of English represents a state of

52
progress for which we ought to be grateful.

52
Jespersen (1924) 179: "However far back we go, we nowhere find
a case with only one well-defined function: in every language
every case served different purposes, and the boundaries between
these are far from being clear-cut. This, in connection with
irregularities and inconsistencies in the formal elements character-
izing the cases, serve to explain the numerous coalescences we
witness in linguistic history ("syncretism") and the chaotic rules
which even thus are to a great extent historically inexplicable.
isIIftheEnliWlmouLtaEs2n2farther than the others in grate fu
simplifying these rules, we should be devoutl
our wa to force it back into the disorder and com lexit of
centuries ago." talics added.
1111111m.

Cassidy, in his 1937 appeal to rescue the word 'case' from

abuse, wrote: "'Case' will be properly used and will continue to

have some meaning only if the association with inflection be

fully recognized, and if stretching of the term to include other

sorts of 'formal' distinction be abandoned."53 In a similar vein,

53Frederick G. Cassidy, "'Case' in modern English," LANGUAGE, XIII

(1937) 240-245; 244.

Lehmann chides Hirt for suggesting that an awareness of cases had

to precede the development of case endings--that there was, in

other words, "among the speakers of pre-Indo-European and Proto-

Indo-European a disposition for cases." Lehmann continues: "We

can account for Hirt's statement by the assumption that to him

a case was a notional category, whether or not it was exemplified

 -39-

in a form. To us a particular case is non-existent unless it

54
is represented by forms which contrast in a system with others."

54
Lehmann (1958) 185.

The claim that syntactic relations of various types must exist

before case endings could be introduced to give them expression

would su-'ely have gone unchallenged; what was offensive, apparently,

was the use of the word 'case'.

It seems to me that if there are recognizable intra-sentence

relationships of the types discussed in studies of case systems- -

whether these are reflected in case affixes or not--and if these

same relationships can be shown to be comparable across languages;

and if there is some predictive or explanatory use to which assump-

tions concerning the universality of these relations can be put,

then surely there can be no meaningful objection to using the

word 'case', in a clearly-understood deep-structure sense, to

identify these relationships. The dispute on the term 'case'

loses its force in a linguistics which accepts the centrality of

syntax. 55

55The universality of case as a grammatical category is affirmed

in Hjelmslev (1934) 1. A recent study of Velten's, from a
Jakobsonian point of view, reveals enough of the historical
continuity of 'synthetic' and 'analytic' cases to suggest that
the linguist has no right to assign cases and prepositions to
different 'chapters' of the study of grammar. (H. V. Velten,
"On the functions of French de and a," LINGUA, XI (1962) 449-
452.) The deep structure notion of cases may be thought of as
involving an extension of the synchronic concept of 'syncretism'.
The asual synchronic sense of case syncretism assume the form of
a decision to posit a case contrast that may not be expressed
overtly in most contexts as long as it appears overtly in "one
 -4o-

part of the system." (See Leonard Newmark's statement in "An

Albanian case system," LINGUA, XI (1962) 312-321; 313.) Deep
structure cases may simply be nowhere overtly reflected as
affixes or function words. The notion we are after probably
corresponds to Meinhof's ' Kasusbeziehungen'. (C. Meinhof, "Der
Ausdruck der Kasusbeziehungen in afrikanischen Sprachen,"
SCRITTI IN ONORE DI A. TROMBETTI, Milano (1938) 71-85; 71. The
Meinhof reference, which I have not seen, was quoted in Henri Frei,
"Cas et deses en francais," CAHIERS FERDINAND DE SAUSSURE, XII
(1954) 29-47; fn. p. 31.)

We may agree, then, for our present purposes, with Hjelmslev,

who suggests that the study of cases can be pursued most fruit-

fully if we abandon the assumption that an essential characteristic

of the grammatical category of case is expression in the form of

affixes on substantives. I shall adopt the usage first proposed,

56
as far as I can tell, by Blake of using the term 'case' to

56
Frank Blake, "A semantic analysis of case," in CURME VOLUME OF
LINGUISTIC STUDIES (LANGUAGE Monograph No. 7) (1930) 34-49.

identify the underlying syntactic-semantic relationship, and the

term 'case form' to mean the expression of a case relationship in

a particular language--whether through affixation, suppletion, the

use of clitic particles, or constraints on word order.

 - 41 -

3. Case grammar

The substantive modification to the theory of transforma-

tional grammar which I wish to propose amounts to a re-introduction

of the 'conceptual framework' interpretation of case systems,

but this time with a clear understanding of the difference between

deep and surface structure. The sentence in its basic structure

consists of a verb and one or more noun phrases, each associated

with the verb in a particular case relationship. The 'explanatory'

use of this framework resides in the necessary claim that, although

there can be compound instances of a single case (through noun

phrase conjunction), each case relationship occurs only once in

57
a simple sentence.

571t follows that whenever more than one case form appears in the
surface structure of the same sentence (on different noun phrases),
either more than one deep-structure case is involved or the sen-
tence is complex. If, for example, German le` hren is described as
a verb which ',takes two accusatives', we have reason to believe
that in the deep structure, the two object nouns are distinct as
to case. Often enough the language will provide evidence for the
distinction, as in the occurrence of such passive sentences as
das wurde mir gelehrt.

It is important to realize that the explanatory value of a

universal system of deep-structure cases is of a syntactic and not

(merely) a morphological nature. The various permitted arrays of

distinct cases occurring in simple sentences expresses a notion of

'sentence type' that may be expected to have universal validity,

independently of such superficial differences as subject selection.

The arrays of cases defining the sentence types of a language

have the effect of imposing a classification of the verbs in the

language (according to th( sentence type into which they may be

 -42-

inserted), and it is very likely that many aspects of this

classification will be universally valid.

Case elements which are optionally associated with specific

verbs, together with the rules for forming subjects, will serve

to explain various co-occurrence restrictions. For example, in

(18) the subject is in an Agent relation to the verb, in (19)

the subject is an Instrument; and in (20) both Agent and Instru-

ment appear in the same sentence, but in this case it is the

Agent which appears as the subject, not the Instrument.

(18) John broke the window

(19) a hammer broke the window

(20) John broke the window with a hammer

That the subjects of (18) and (19) are grammatically differ-

ent explains the fact that the combined meanings of the two

sentences is not produced by conjoining their subjects. Thus (21)

is unacceptable.

(21) *John and a hammer broke the window

Only noun phrases representing the same case may be conjoined.

Similarly, the fact that only one representative of a given case

relationship may appear in the same simple sentence, together with

the generalizations on subject selection and the redundancies

which hold between cases and lexical features (e.g., between Agent

and animateness), explains the unacceptability of sentence (22).

(22) *a hammer broke the glass with a chisel

It is unacceptable, in particular, on the interpretation that both

hammer and chisel are understood instrumentally. It cannot repre-

sent a sentence containing an Agent and an Instrument, since the

 - 43 -

58
noun hammer is inanimate.

58The author is aware that in sentence (18) one might be talking

about what John's body did as it was tossed through the window
and that in sentence (19) one might be speaking metaphorically,
personifying hammer. Under either interpretation sentence (21)
turns out to be acceptable, and under the personification interpre-
tation, sentence (22) becomes acceptable. What is important to
realize is that these interpretations, too, are explainable by ref-
erence to exactly the same assumptions appealed to in explaining
their 'face value' interpretations.

The dependencies that can be accounted for by making these

assumptions are that the subject of an active transitive sentence

must be interpretable as a Personal Agent just in case the

sentence contains a with-phrase of Instrumental import. Apparent

exceptions to this generalization can be seen to have different

underlying structures. Sentence (23) looks like an exception, but

by attending to the effect of the word its, the essential differ-

ence between (23) and sentences (22) and (24) becomes apparent.

(23) the car broke the window with its fender

(24) *the car broke the window with a fender

Sentence (24) violates the conditions that have been discussed,

but sentence (23) is a paraphrase of sentence (25) and may be

interpreted 's having the same deep structure as (25).

(25) the car's fender broke the window

What is suggested here is that sentences (23) and (25) are

Agentless sentences containing a possessed noun as the Instrument

(the car's fender). The rules for choosing a subject allow an

option in this case: either the entire Instrument phrase may appear

as the subject (as in (25)), or the 'possessor' alone may be made

the subject, the remainder of the Instrument phrase appearing

with the preposition with (as in (23)). The second option

requires that a 'trace' be left behind in the Instrument phrase,

in the form of the appropriate possessive pronoun. A similar

explanation is suggested for such sentences as (26) and (27), which

are also interpretable as deep-structurally identical.

(26) your speech impressed us with its brevity

(27) the brevity of your speech impressed us

The superficial nature of the notion 'subject of a sentence' is made

apparent by these examples in a particularly persuasive way, because

in the possessor-as-subject cases, the 'subject' is not even a major

constituent of the sentence; it is taken from the modifier of one

of the major constituents.

In the basic structure of sentences, then, we find what might

be called the 'proposition', a tenseless set of relationships invol-

ving verbs and nouns (and embedded sentences, if there are any),

separated from what might be called the 'modality' constituent.

This latter will include such modalities on the sentence-as-a-whole

as negation, tense, mood and aspect. 59

II
ihe exact nature of the

59There are probably good reasons for regarding negation, tense and
modd as associated directly with the sentence as a whole, and the
perfect and progressive 'aspects' as features on the V. See for
a statement of this position Lyons (1966) 218, 223.

modality constituent may be ignored for our purposes. It is likely,

however, that certain 'cases' will be directly related to the

modality constituent as others are related to the proposition itself,

60
as for example certain temporal adverbs.
 -45-

VIIIINIIIMMO.N1=11111Imanumme
o
In my earlier paper I suggested that sentence adverbials in
general are assigned to the modality constituent. I now believe
that many sentence adverbs are introduced from superordinate sen-
tences (by transformations of a type we may wish to call 'infra-
jections1). This possibility has long been clear for unmistakable
sentence adverbs like unfortunate a, but there are also quite con-
vIncing reasons for extending the infrajection interpretation to
adverbs like willingly, easiLE, carefully, etc.

The first base rule, then, is (28), abbreviated to (28').

(28) Sentence - Modality + Proposition

p61
(28') S -- M

61
The arrow notation is used throughout, but this should not be
interpreted as meaning that the proposal for a case grammar
requires an assumption of a left-to-right orientation of the
constituent symbols of the rewriting rules.

The P constituent is 'expanded' as a verb and one or more case

categories. A later rule will automatically provide for each of

the cases the categorial realization as NP (except for one which

may be an embedded S). In effect the case relations are represented

by means of dominating category symbols.

The expansion of P may be thought of as a list of formulas

of the form (29), where at least one case category must be chosen

and in which no case category appears more than once.

(29) P + V + C1 + + C
n

Whether these formulas can IDE; collapsed according to the familiar

abbreviatory conventions is not at present clear. For our pur-

poses we may simply think of P as representable by any of a set

of formulas including V + A, V + 0 + A, V + D, V + 0 + I + A, etc.

(The letter symbols are interpreted below.)

 -46-

The case notions comprise a set of universal, presumably

innate, concepts which identify certain types of judgments which

human beings are capable of making on the events that are going

on around them, judgments on such matters as who did it, who it

happened to, what got changed, etc. The cases that appear to be

needed include:

Agentive (A), the case of the perceived instigator of

62
the action identified by the verb, typically animate.

2
The escape qualification 'typically' expresses my awareness that
contexts which I will say require Agents are sometimes occupied by
'inanimate' nouns like robot or 'human institution' nouns like
nation. Since I know of no way of dealing with these matters at
the moment, I shall just assume for all Agents that they are 'animate.'

Instrumental (I), the case of the inanimate force or

object causally involved in the action or state

63
identified by the verb.

3 Pau1 Postal has reminded me of the existence of sentences like

(i)
(i) I rapped him on the head with a snake.
The requirement that Instrumental NPs are 'inanimate' is the require-
ment to interpret (i) as having in its underlying structure some-
thing equivalent to with the body of a snake. The fact that there
are languages which would require mention of a stem meaning 'body'
in this context may be considered as support for this position;
and so may the absence, pointed out by Lakoff, of sentences like (ii)
(ii) *John broke the window with himself
(See George Lakoff, "Instrumental adverbs and the concept of deep
structure," unpublished dittographed manuscript, 1967.)

Dative (D), the case of the animate being affected by

the state or action identified by the verb.

Factitive (F), the case of the object or being resulting

from the action or state identified by the verb, or

1116_
 - 47

understood as a part of the meaning of the verb.

Locative (L), the case which identifies the location

or spatial orientation of the state or action

identified by the verb.

Objective (0), the semantically most neutral case, the

case of anything representable by a noun whose role

in the action or state identified by the verb is

identified by the semantic interpretation of the

verb itself; conceivably the concept should be limited

to things which are affected by the action or state

64 The term is not to be
identified by the verb.

64 In Fillmore (1966) the neutral case was unwisely and misleadingly

labeled 'ergative'.

confused with the notion Direct Object, nor with the

name of the surface case synonymous with Accusative.

Additional cases will surely be needed. Suggestions for

adding to this list will appear in various places below.

It is important to notice that none of these cases can be

interpreted as matched by the surface-structure relations Subject

or Object in any particular language. Thus, John is A in (29) as

much as in (30); the key, is I in (31) as well as in (32) or (33);

John is D in (34) as well as in (35) and (36); and Chicago. is L

in both (37) and (38).

(29) John opened the door

(30) the door was opened by John

 (31) the key opened the door

(32) John opened the door with the key

(33) John used the key to open the door

(34) John believed that he mould win

(35) we persuaded John that he would win

(36) it was apparent to John that he would win

(37) Chicago is windy

(38) it is windy in Chicago

The list of cases includes L, but nothing corresponding to

what might be called Directional. There is a certain amount of

evidence, as was mentioned above, that locational and directional

elements do not contrast but are superficial differences determined

either by the constituent structure or by the character of the

associated verb. An example provided by Hall (39) suggests, by

the occurrence of the pro-replacement word there, that to the

store and at the store are variants of the same entitly, determined

by the movement or non-movement character of the associated verb.65

(39) she took him to the store and left him there

5Barbara Hall (now Barbara Hall Partee), SUBJECT AND OBJECT IN

ENGLISH, M.I.T. Ph. D. Dissertation (1965) 77.
The putative contrast between locational and directional
expressions as well as the distinction between 'optional' and
'obligatory' locative expressions, as exemplified in Hall's
examples (i) and (ii) seem to point to the difference between
elements which are 'inside the VP' and elements which are 'out-
side the VP'.
(i) John keeps his car in the garage
(ii) John washes his car in the garage
In our terms this would be equivalent either to determining whether
there is a difference between a L as a constituent of P and a L as
a constituent of M, or whether there can be two L elements within
P, distinguished in terms of degree of selectivity of verbs. The
highly restricting L selects verbs like 1422, but and leave, but
not p'lish, wash and build; the weakly restricting L selects
verbs like 2.21192, wash and build, but not believe, know or want.
However this distinction is interpreted, the second or
'outer'L is in some respects similar in its 'selectional' proper-
ties to what might be called the Benefactive case (B). B, too,
is involved in the selection of verbs in the sense that some verbs
do not accept B modification (*he is tall for you); but the restric-
tion here may have more to do with dependent relations between .

cases than with dependencies directly connected with the verb. It

appears, in fact, that those verbs which allow 'outer L' and B
modification are precisely those which take Agents. I have no
ideas on how these dependencies can be stated, but it would appear
that the second L and the B can appear only in sentences contain-
ing A's.
Thus the ime direct vs. regime indirect interpretation of
the difference between (iii) and (iv)
(iii) it demeure a Paris
(iv) it travaille a Paris
may have simply to do with the fact that the subject of (iv) is
actually an A. Both the specific verb and the occurrence of an
'outer L! are determined by the presence of an A. (See, in this
connection, Bazell!s discussion of Gougenheim's review of de Boer's
French syntax in C. E. Bazell, "Syntactic relations and linguistic
typology," CARTERS FERDINAND DE SAUSSURE, VIII (1949) 5-20; 10.)
,AM

I have stated that A and D are 'animate' participants in the

activity of the associated verbs, and I have also suggested that

verbs are selected according to the case environments which the

sentence provides--what I &-all refer to as the 'case frame'.

There are, then, the two problems of lexical selection, that of

the nouns and that of the verbs. Those features of nouns required

by a particular case are to be specified by obligatory rules of the

type illustrated by (39), a rule which specifies that any N in an

A,D
(39) N [i.Animate] / [ X ]

A or D phrase must contain the feature C Animate]. (Recall the

qualification of fn. 62.)

To take care quite generally of lexical features associated

with specific cases, we may appeal to a rule which associates

with each noun a label which identifies the case relation which
it holds with the rest of the sentence. Such a rule might

associate with every noun under L the feature [+Locative], for

example. Since abstract nouns such as idea cannot serve as heads

of L expressions, they will be marked [- Locative] .66

By allowing highly restricting lexical features to be associated

with given case units we have returned to that extension of
'cases' to 'adverb forms' proposed by Bopp, Wainer and Hartun7.
Some adverbs, on this view, are really nouns capable of 'takikig'
only one case form. Since deep structure cases are in fact all
'defective' to some extent, with respect to the nouns which they
accept, such a concept as inflectional scope no longer provides
a clear demarcation between 'case forms proper' and 'adverbs'.
See the discussion of this question in Hjelmslev (1935) 40.
Ammummommosr

The insertion of verbs, on the other hand, depends on the

particular array of cases, the 'case frame', provided by the

67
sentence. The verb run, for example, may be inserted into the

671 am adhering, in this discussion, to the Postal-Lakoff doctrine,

which I find thoroughly convincing, that adjectives constitute a
subset of verbs. See George Lakoff, ON THE NATURE OF SYNTACTIC
IRREGULARITY, Harvard Computation Laboratory, Report No. NSF-16,
Mathematical Linguistics and Automatic Translation (1965),
Appendices.

frame [ A ], the verb sad into the frame [ D ], verbs like

remove and epen into [ 0 + A ], verbs like murder and terrorize

(i.e., verbs requiring 'animate subject' and 'animate object') into

[ D + A ], verbs like give into E 0 + D + A ], and so on.

In lexical entries for verbs, abbreviated statements called

'frame features' will indicate the set of case frames into which

the given verbs may be inserted. These frame features have the

effect of imposing a classification of the verbs in the language.

 -51-

The classification has the complexity it does not only because

of the variety of case environments possible within P, but also

because many verbs are capable of occurring in more than one

distinct case environment. This last fact can be represented

most directly by allowing facultative representation of cases in

the frame feature expressions.

The word open, to take a familiar example, can occur in

E 0 3, as in (40) ; in C 0 + A 3, as in (41) ; in [ 0 + I ] ,

as in (42); and in C 0 + I + A 3, as in (43).

(40) the door opened

(41) John opened the door

(42) the wind opened the door

(43) John opened the door with a chisel

The simplest representation of this set of possibilities makes use

of parentheses to indicate the 'optional' elements. The frame

feature for open may thus be represented as (44).

(44) -1-E 0 (I) (A) 368

Case frames are represented in square brackets, with 'underline'

indicating the position of the element with respect to which the
expression is an environmental frame. A frame feature is repre-
sented in square brackets with '+' or '-' in front, indicating
that the set of case frames represented by the expression within
the brackets are those which will (if the feature is marked '+')
or which will not (if the feature is marked '-') accept the
lexical item with which the feature is associated.

Other verbs having this same feature are turn, move, rotate, bend,
 -52-

For a verb like kill it is necessary to indicate, express-

ing it in familiar terms, that it takes an animate object and

either an animate or an inanimate subject; and that if there is

an animate subject, an instrument phrase may also co-occur. The

frame feature for kill, in other words, will have to specify that

either an Instrument or an Agent must be specified, and both may.

If the linked parentheses notation can be introduced to indicate

that at least one of the linked elements must be chosen, the

frame feature for kill can be given as (45).

(45) +[ D (IA) ]

The verb murder, on the other hand, is one which requires an Agent.

Its frame feature differs from that of (44) and (45) because the

element A is obligatorily present. It is given as (46).

(46) +[ D (I) A ]

The environmental subclassification of verbs is sensitive to

more than the mere array of cases in P. Since one of the cases may

be represented by S (an embedded sentence), verbs are also sub-

classified in terms of whether the 0 element is a sentence. By

convention we shall interpret the symbol 0 in frame features as

indicating NPs, and the symbol S as indicating an 0 to which an

S has been embedded.

The frame feature +[ S ] characterizes such verbs as true,

interesting, etc.; the feature +[ S + D ] is common to such

verbs as want and expect; verbs predict and cause

appear in the frame [ S + A ]; and verbs like force and ersuade

69
are insertable into the frame [ S + D + A ].

IL
69
It should be pointed out that descriptions of embedded sentences
as it + 8 realizations of the category NP in 'subject/object'
grammars must somehow guarantee that this particular expansion of
NP is limited to the subjects of intransitive sentences and the
objects (direct or oblique) of transitive sentences. All such
restrictions are rendered unnecessary by the decision to limit
complement S to the case element 0.

are
VerbsAdistinguished from each other not only by specification

of the case frames into which they can be inserted, but also by

their transformational properties. The most important variables

here include (a) the choice of a particular NP to become the surface

subject, or the surface object, wherever these choices are not

determined by a general rule; (b) the choice of prepositions to go

with each case element, where these are determined by idiosyncratic

properties of the verb rather than by a general rule; and (c)

other special transformational features, such as, for verbs taking

S complements, the choice of specific complementizers (that, iaa,

for to, etc.) and the later transformational treatment of these

elements.

The use of parentheses in expressing the frame features,

together with the transformational introduction of subjects, makes

it possible to reduce the number of semantic descriptions in the

lexicon. The semantic interpretation of a P will introduce all

information provided by specific case relationships represented in

the P, allowing such information to be omitted from the semantic

descriptions of verbs. In the case of verbs .having the feature (44),

as we have seen, certain related transitive and intransitive verbs

need not be given separate semantic description. This point may

be further demonstrated with the English verb cook. The frame

feature of cook is presumably something like (47)

(47) 4-C 4 (A) ]

and an idiosyncratic transformational feature of the verb is

that just in case the A is present and the 0 is some NP repre-

senting a typical NP for the verb (i.e., something like food or

a meal), the 0 element may be deleted. The semantic description

of the verb will do no more than identify a particular activity

having a result of a particular ki4d on the object identified by

the 0 element. The same semantic entry, in other words, will

account for the use of cook in all of the sentences (48) - (50).

(48) Mother is cooking the potatoes

(49) the potatoes are cooking

(50) Mother is cooking

Instead of saying that the verb has three different meanings, we

can be satisfied to say that there is a certain variety in the

case frames which accept it, and that is is one of the 'deletable

object' verbs. The fact that A is obligatorily animate and that

0 is unspecified for animateness accounts for the fact that if we

can read sentence (49) as ambiguous, it is because we can accept

certain violations of grammatical requirements in 'personifications'

of the type we have learned in nursery school, whereas if we

accept (50) as in fact ambiguous, it is because we are acquainted

with the range of activities found in human societies.

The example with cook shows that the lexicon need not contain

as many semantic entry tokens under the present proposal as it

70
would in a subject/object grammar. It will now be shown that
70
It may appear that facultative representation of cases in frame
features has the advantages it does in English because there are
so many verbs which can be used transitively or intransitively in
the same form. It is a language-particular coincidence that Eng-
lish uses the same form in these words. The identification of
transitive and intransitive 222/2, or transitive and intransitive
cook, is justified because the semantic characterization of the
verb is the same in all of the uses discussed. (We must distinguish
between the semantic characterization of a verb and the semantic
interpretation of sentences containing the verb. In the latter
case, all of the co-constituents and the semantic role they play
as determined by their cases are taken into account.) Wherever
that condition can be satisfied, facultative representation is
called for. It will turn out that for some languages the occur-
rence or non-occurrence of one of the 'optional' cases will have
an effect on the verb. If, for verbs of the type [ 0 (A) 3,
the appearance of the A determines a variant of the verb differ-
ent from that when A is missing (distinguishing the 'transitive'
from the 'intransitive' use of the 'same' verb), or if the
absence of the A requires some additive element (e.g., a 'reflex-
ive' morpheme) not needed when the A is expressed, these facts
can be provided transformationally. See Mantaro J. Hashimoto,
"The internal structure of basic strings and a generative treat-
ment of transitive and intransitive verbs," paper read before the
1966 Tokyo International Seminar in Linguistic Theory. (By
extending the range of acceptable surface variants of verbs under
these conditions to suppletion, it may even be possible to inter-
pret the contrasts exemplified in (51) - (53) below as surface
lexical variation.)

this same flexibility makes it possible to ra:%duce the number of

semantic entry IzEIE, for now it is feasible to show that some

syntactically different words are in fact semantically identical

(with respect to that aspect of their meanings which are indepen-

dent of the contribution of the associated cases). This may be

true for verbs like like and please, to give the example that

comes most quickly to mind. These words may be described as

being synonymous and each having the frame feature +[ 0 + D ]

differing only in their subject selection features. The verb

 - 56 -

like, in fact, has in its history the subject selection feature

possessed by please.

The verb show, to give another kind of example, might well

have the same semantic representation as see, differing from it

only in that the frame feature for show contains an A where that

for see does not. The verbs kill and die appear to be related

in a similar way.

(51) see (+E 0 + D ]) vs. show GE 0 + D + A ])

(52) die ( +C D ]) vs. kill (+ D (I A) ])

We have seen, then, instances of synonymy where there are

identical frame features but different subject selection features,

and instances of synonymy where there are frame feature differences

depending on whether a particular case category was present or

absent. We may now turn to examples of synonymy where the

difference is in the choice of one case or another.

It will be recalled that both A and D are animate. The

semantic descriptions of certain verbs may refer to the animate-

ness of the associated noun, independently of whether the 'source'

of the animateness is A or D. That is, the semantic representation

of certain verbs may specify a relationship or a process associated

with the necessarily-animate participant in the state or activity

identified by the verb. The relation of hear and listen to the

necessarily-animate NP is the same in both cases, the difference

in the semantic interpretation of the Ps containing them determined

by the semantic contribution of the associated cases, and by the

fact that the frames that contain hear are E 0 + D ] and those

that contain listen are C 0 + A ]. The fact that in the case

 57 -

of listen the relationship is understood as involving the active

participation of the person identified as A is due to the presence

of A, not to a special meaning of listen. The same distinction

can be seen between see and know, on the one hand, and look and

learn, on the other.

(53) see, know (+E 0 + D ]) vs. look, learn

( +E 0 + A ])

This latest point leads one to those properties of English

verbs with which Lakoff associates the terms 'stative' and 'non-

stative'. The question we need to ask is whether Lakoff's

71
George Lakoff, "Stative adjectives and verbs in English,"
Harvard Computation Laboratory, Report No. NSF-17, Mathematical
Linguistics and Machine Translation (1966) 1.1-16.

features are primitives in the lexical entries for verbs, or

whether they permit reduction to concepts of the type I have been

outlining. Lakoff has noticed that the 'true imperative', the

progressive aspect, the occurrence of Benefactive (B) phrases,

and do so substitution occur only with 'non - stative' verbs.

Lakoff's discussion suggests that one must assign 'stative' and

'non-stative' as features on verbs and then guarantee that B

phrases are permitted only with 'non - statives' (put the other way

around, one must guarantee that the presence of a B expression allows

only for the selection of 'non-statives'), that the imperative

transformation can be applied only if the verb is 'non-stative'

and so on. The treatment that I prefer is implicit in what I

have already presented. The transformation which accounts for the

'true imperatives' an apply only to sentences containing As, and

the occurrence of B expressions (and 'Outer Ls') is dependent on

the presence of an A. The progressive aspect can only be chosen

in association with particular case frames, for example, those

containing As. No special features indicating stativity need be

added to verbs, because (if this suggestion is correct) only those

verbs which occur in Ps containing As will show up in these

72
sentences anyway.

72The do-so evidence is not so easy to interpret in this way.

Still, the connection between 'non- stative' verbs and verbs that
can 'take' A is too compblling to be simply wrong.

3.5 Surface phenomena

To recapitulate, our discussion so far has suggested that

the deep structure of (the propositional component of) every

simple sentence is an array consisting of a V plus a number of

NPs holding special labeled relations (cases) to the sentence,

where these relations are provided for categorially and include

such concepts as Agentive, Instrumental, Objective, Factitive,

Locative, Benefactive, and perhaps several others. Complex

sentences involve recursion through the category Sentence under

the case category Objective. Verbs are subclassified according

to the case environments which accept them, and the semantic

characterizations of verbs relate them either to specific case

elements in the environment or to elements containing features

(such as animateness) introduced as obligatory accompaniments of

particular cases.
 -59-

The present section will deal with some of the ways in

which deep structures of the type proposed in this essay are converted

into surface representations of sentences. The various mechanisms

involve selection of overt case forms (by suppletion, affixation,

addition of prepositions or postpositions), 'registration' of part-

icular elements in the verb, subjectivalization, objectivalization,

sequential ordering, and nominalizations.

A surface case system may be related to the set of underlying

cases in a variety of ways. Two deep cases may be represented in

the same way in the surface structure, as when D and 0 direct

objects are both represented with the 'accusative' case in many

languages (where the determining factor may be occurrence immediately

after the verb at some stage of derivation). A and D may be rep-

resented by the same overt form, where the determining factor may

be the case-linked animateness. Or the superficial form of a case

element may be determined by an idiosyncratic property of some

governing word.

The rules for English prepositions may look something like

this: the A preposition islE; the I preposition is 111 if there is

no A, otherwise it is with; the 0 and F prepositions are typically

zero; the B preposition is for; the D preposition is typically to;

the L and T (for Time) prepositions are either semantically non-

empty (in which case they are introduced as optional choices from

the lexicon) or they are selected by the particular associated noun

(on the street, at the corner (=intersection of two streets), in the

corner (of a room); on Monday, at noon, in the afternoon). Specific

 -6o

verbs may have associated with them certain preposition-choice

requirements, introduced as exceptions to the above generaliza-

tion.73

73The verb blame, for example, chooses ('governs') for for 0 and on
for D. The 0 preposition is at for look meaning 'examine', for for
look meaning 'seek', to for listen, etc. Changes in the original
prerowtion assignment may be brought about by transformations: the
rule:, which provide surface subjects and direct objects delete
prepositions (replace them by zero), and the rules which form
deverbal (=desentential) nominals convert some of the original case
forms into 'genitive', either by replacing the assigned preposition
with of, or, in some cases, removing the original preposition and
affixing the 'geaitive' suffix.

The position of prepositions can be guaranteed either by

having the case categories re-written as Prep + NP, or by having

12E22 be one of the obligatory constituents of NP. I shall make the

former choice, though the grounds for making the decision one way

or the other are not particularly clear. The 'universal' charge:ter

of the base rules is kept intact by the assumption that prepositions,

postpositions9 and case affixes--semantically relevant or not--are

all in fact realizations of the same underlying element, say K

(for Kasus). We may regard all of the case categories as therefore

rewritten as K + NP.

Every English sentence has a surface subject, if only formally

so. For most combinations of cases there is a 'preferred' or

'unmarked' subject choice; for some there is no actual choice--the

subject is uniquely determined. In general the 'unmarked' subject

choice seems to follow the following rule:

(54) If there is an A, it becomes the subject; otherwise,

if there is an I, it becomes the subject; other-

wise, the subject is the 0.

 Suppose, for example, that the base representation of a

particular sentence is item (55):

(55)
,000101."

V
/o
oN
K NP

d N
1 I

past open 0 the door

Since the sentence contains only one case category, it is obliga-

torily moved to the front (and hence directly subjoined to the

category S) where it will later undergo subject-preposition deletion.

There is a stage, in other words, where the form of the sentence in

question is that represented in (56).

(56)

K NP

0 the door past open

The subject-preposition deletion rule removes the preposition and

deletes the case label. After application of the subject-preposi-

tion deletion rule, the form of the sentence is that represented

in (57).
 -62-

(57) s
.....-----i----___
m
.,...---:
N
i 1

the door past open

The final surface form, shown in (58), results from incorporation

of the tense into the verb.

(58)

NP P

d N V
I 1 1

the door opened

For a base configuration containing an A a distinction must be

74
made between the 'normal' and the 'non-normal' choice of subjects.

The choice of terms is not to be taken seriously.

The choice of the A as the subject, in accordance with the rule

proposed in (54) above, requires no modification of the verb. The

changes from (59) to (60) represent subject-fronting, those from

(60) to (61) show subject-preposition deletion, and those from

(61) to (62) indicate the effect of a third rule, object-preposition

deletion.75 The eventual surface structure of the sentence whose deep

75 Verbs are categorized according to whether theydblete the prep-

osition of the following case category, i.e., whether they 'take on'
a direct object. The object-preposition deleting property of a
verb may be modified by a transformation.
 -63-

structure is (59) is (63).

(59)

NP NP
9\
past give 0 the books to my brother by John

(6o)

NP

NP K NP

I
I

John past give re; the books to my brother

by

(61)

NP

NP tit P

1
John past give 0 the books to my brother
 - 64 -

(62) S

NP M

John past give the books my brother

(63)

NP

V NP D

K NP

1 1
John gave the books to my brother

If it is noted that the verb ial is one which, with A as subject,

allows either 0 or D to become the direct object, an alternative

surface form for (59) is (64) (assuming that case-label deletion

occurs when zero K elements are 'deleted').

(64)
NP

NP

d N d N
I I
John gave my brother the books

The Inormal9 choice of subject for sentences containing an A,

as stated in generalization (54) (which is a generalization for

 -65-

English) is the A. The verb 'ye also allows either 0 or D to

appear as subject as long as this 'non - normal' choice is 'registered'

in the V. This"registeringl of a 'non-normal' subject takes place

via the association of the feature [ +Phssit.e] with the V. This

feature has three effects; the V loses its object-preposition

deletion property, it loses its ability to absorb the tense (requir-

ing the automatic insertion of a be in the M constituent), and it

must now be filled by a special 'passive' form (i.e., given). The

sequence (65) to (68) develops the choice of 0 as subject; the sequence

(69) to (73) shows the result of choosing D as subject.

(65)

7,7/`_
Ei-Pasj
K
K NP

K1d iNP by John

0 the books past give to my brother

(66)

d N
[44
V A

K NP

d
y 1

the books past give to my brother by John

 -66-

(67)

[PassV

K NP
///'N
NP

the books give my brother by John

(68)

_,...),1' 1glp
d
I
I 7 I
the books were given my brother by John

(69) S
1
M

K NP
[PassV

.............0 A
.....
K I[ NP

N
I I
I I 1
to my brother past give 0 the books by John

(70)

d N

NP K NP

my brother past give 0 the books by John

 -67-

(71)

NP M

d N
:fq K NP NP

I d N
1

brother past be give 0 the books by John

(72) S

A
+Pass
d N K NP

brother past be give the books by John

my

(73) S

NP
[Pass'
Mane.
K

given
1
the books by
T
John
my brother was

We have seen that where there is only one case category, its

NP must serve as the surface subject. Elramples (59) to (73) have

shown ways of dealing with sentences containing more than one case

category where one designated case could provide the subject with-

out effecting any change in the V, or others could do so as long

as a 'record' of this decision was attached to the V.

For many of the verbs which 'take more than one case category,
by the verb
the one which contributes the subject is indicated
 -68-
0 + D
itself. Of the verbs which are accepted into the frame C

please, belong, interesting, etc., choose 0 as subject, like, want,

76
think, etc., choose D.

represent-
714,s mentioned above, by regarding the differences here as
ing no more than idiosyncratic facts about the syntactic properties
of these verbs, we can accept historical changes like those with like,
want and think, from verbs of the type which chme 0 to verbs of the
type which choose D, to be merely a matter of detail in the subject-
selection processes in our language. In other words, we do not need
to agree with Jespersen when he describes the change in English from
the use of ekpressions of the type him like oysters to those of the
'meaning' of the
type he likes oysters as reflecting a change in the
verb like from something like "to be agreeable to" to something like
"to take pleasure in" (Jespersen 1924) 160). The change seems merely
to be a result of the inter-influencing of the two surface processes
of choosing the first word and establishing verbal concord.

Sometimes subjects are created not by moving one of the case

elements into the 'subject' position, but by copying a particular

element into that position. This seems to be a consequence of the

positional treatment of subjects in English and seems to be related

77
to the use of purely formal subjects.

77 From the fact that there may only be one case in a simple sentence,
it becomes possible to allow all subjects to be formed by a copying
NP in the same
transformation. Sentences with two copies of the same
either
case undergo one of a number of changes: the second copy is
deleted or replaced by a pro-form, or the first copy is replaced by
a pro-form.

Copying for pro-replacement can be illustrated with that-clauses.

The 'verb' true occurs in the frame [ S], i.e., in the configuration

(74).
 -69-

Since there is only one case element, it is obligatorily the sub-

ject. The context requires that the complementizer that be provided

for the embedded sentence. By subject copying, (76) is derived from

(75).

V 0

pres true that John likes Mary

that John likes Mary pres true that John likes Mary

The structure of (76) undergoes either second-copy deletion, yield-

ing (77), or first -copy pro-replacement, giving us (78).

S V
I

that John likes Mary pres true

 -70-

(78)

0 M

0
t

it
1\
pres be true that John likes Mary

Verbs expressing meteorological conditions have the frame

feature +[ L 1. Choosing hot in that frame, we can construct

the sentence wh -se deep structure is represented by (79). From (79)

we get, by subject copying, item (80). By second copy deletion (and

subject preposition deletion) item (80) becomes (81); on the other

hand, if the first copy is replaced by its pro-form (in this context,
78
it), the resulting sentence is (82).

781t is likely that the correct analysis of subject copying is a little

different from this. There is considerable evidence that when the
first copy is replaced by its pro-form, the second copy is actually
outside of P, that is, that it is 'extraposed' in the sense of Rosenbaum.
If this Is true, then since the sentences having undergone extraposition
must be created in two steps anyway, it is likely that the sentential
subjects are formed in the usual way--not by copying--and that they
are later extraposed, leaving behind, in the subject position, a
'trace' in the form of expletive it.
The examples and the analysis of meteorological vabs are adapted
from D. Terence Langendoen, "Some problews concerning the English
expletive 'it'," THE OHIO STATE UNIVERSITY RESEARCH FOUNDATION, PROJECT
ON LINGUISTIC ANALYSIS, REPORT NO. 13 (1966), 104-134.

(79)

M P

K NP

pres hot in the studio

 - 71 -

(8o)

111

pres hot in the studio

(81)

NP M

the studio is hot

(82)

K NP

it is hot in the studio

Under certain conditions, a first-copy L may be replaced by

an expletive there. The case frame [ 0 + L ] may be filled

by a blank verb (i.e., zero). This situation (of verbless sentences)

may call for the introduction of the element be into the M constitu-

ent, which is a process we have already seen to be necessary for

verbs which are adjectives as well as for verbs which have been

modified through addition of the feature [I-Passive]. For verbless

 - 7 2 -

sentences of the type r_

0 + L ], the 'normal' subject choice

is 0. Thus, from (83) we get (84), and eventually (85).

(83)

K NP K NP

pres 0 many toys in the box

(84)

K V L
,-----s,
K NP

d N
I I

0 many toys pres 0 in the box

(85) S

M L
,./...1.--NTP4**,.
d N K--------------.6ITP

d N
I
1 1 I I

many toys are in the box

An alternative subject choice, through subject-copying, is the

L. Thus, from (83) we might get (86).

 (86) 13

gP

d N

1 I

0 0 many toys in the box

in the box pres

(unstressed)
The pro-form for L in verbless sentences is expletive
the
there. The result of modifying (86) by pro-replacement of
second-copy L, as suggested
subject L is (87); extraposition of the

in fn. 78, has been carried out in (87).

(87)

ct

are many toys in the box

there

expletive there is
Alternatively to replacing the first-copy L by

to retain the L NP as subject. This decision requires the regular

pronominalization of the repeated NP. It further requires modifi-

with the
cation of the verb: the hitherto empty V position is filled

function verb have.79 Since have is a V, it is capable of absorbing

79 For a recent argument on the transformational introduction of be

Emmon Bach, "Have and be,"
and have in all of their occurrences, see
LANdrrfft, forthcoming. For a more adequate treatment of existential
"Some
sentences than I have presented, see especially P. Gregory Lee,
properties of be sentences," 1967 forthcoming.
 - 74 -

the tense, making the addition of be to M no longer necessary. The

result of choosing the first L as subject results, through subject-

preposition deletion, have-insertion, object-preposition deletion,

repeated NP pronominalization, and tense affixation, in (88).

(88)

NP

d N V NP L

d . N K NP

the box has many toys in it

The general position I am taking on the verb have is that in

verbless sentences (i.e., when the V constituent is present but

lexically empty), have is obligatorily inserted just in case the

subject is a NP which is not from the case O. The most obvious case

is that of the empty verb in the frame [ 0 + D ] , a context

which in English requires D to be the subject, resulting in the

typical have sentences. Other languages, e.g., French, seem to have

contexts in which the subject choice is optional--situations where

X a Y is in a paraphrase relation with Y est a X. Other languages,

80
e.g. Estonian, do nc1t have anything equivalent to the verb have.

0Another situation for introducing have, accounting for connections

between such pairs of sentences as (177;nd (ii), are discussed below
in the section on inalienable possession.
(i) my knee is sore
(ii) I have a sore knee

Some languages have subjectivalization processes; and, as I have

suggested for English, there seems to be an analogous objectivalization

 75 IND

process which has the superficial effect of bringing a particular nom-

inal element into closer association with the verb.

The formal rather than purely notional character of the direct

object was noticed by Jespersen. His examples show intra-language

paraphrase relations like that between (89) and (90), and cross-

language differences like that between (91) and (92).

(89) present something to a person

(90) present a person with something

(91) furnish someone with something

lun81
(92) fournir quelque cho se a quelqu

81
Jespersen (1924) 162.

When such phenomena were examined by Hall, she took one form as

basic, the other as derived. 'Derived subjects', in her analysis,

are possible just in case there is no 'deep subject'; 'derived

objects', on the other hand, have the effect of displacing the ori-

ginal deep structure object and attaching a with to it. Her

examples include (93)-(94) and (95)-(96).

(93) John smeared paint on the wall

(94) John smeared the wall with paint

(95) John planted peas and corn in his garden

(96) John planted his garden with peas and corn

Hall provides rules which move the locative element (the wall or his

garden of (93) and (95) respectively) into the direct object posi-
Va

tion by a transformation which also attaches with to the former

direct object.
 - A
7-

From the point of view taken here, it would be just as easy to

say that both on the wall and with paint were initially provided

with prepositions (as L and I case elements), the verb smear

having the property that whichever of these elements is chosen as

'direct object' is drawn next to it and loses its preposition.

(In other languages, the process might be expressed as converting

82
an original case specification to laccusative'.)

2
There are semantic difficulties in treating subject and object
transformitionally, in the sense that different choices are often
accompanied by semantic differences of one sort or another. These
differences are more on tne order of 'focusing' - -to be as vague as
possible--than anything else, and do not seem to require positing
'subjects' and 'objects' in the deep structure. The 'focusing'
difference may be extremely slight, as in the pairs (i)-(ii) and
(iii)-(iv), or it may have somewhat more 'cognitive content' as in
the pairs (v)-(vi) and (vii)-(viii).
(i) Mary has the children with her
(ii) the children are with Mary
(iii) he blamed the accident on John
(iv) he blamed John for the accident
(v) bees are swarming in the garden
(vi) tht garden is swarming with bees
(vii) he sprayed paint on the wall
(viii) he sprayed the wall with paint
Sentence (vi) seems to suggest, while (v) does not, that the whole
garden has bees in it everywhere; and (viii) suggests, while (vii) does
not, that the entire wall got covered with paint.
To the extent that other grammars make use of derived subjects
and derived objects--which is the only alternative, within subject/
object grammars, to treating verbs like luma, blame open, break,
etc. as involving elaborate and unexplained examples of homonymy- -
the semantic difficulties are just as great for them as they are
for case grammar. Since the 'semantic effect' of the transforma-
tions in question is so different in kind from the semantic role of
the case relations themselves, and since the latter are not affected
by these processes, I am inclined to tolerate the re-introduction into
grammatical theory of transformations which have semantic import (of
this highly restricted kind).

Subjectivalization, where it occurs, results in a neutralization

of underlying case distinctions to a single form, usually called the

 -77-

'nominative'. Objectivalization, where it occurs, neutralizes case

distinctions to a single form which, where it is distinct from the

form assigned to subjects, is traditionally termed 'a:cusativel.

A third process which has the effect of effacing deep-structure

case distinctions is the formation of nominals from sentences. The

case modifications under nominalization transformations usually involve

what is called the 'genitive'.

The brief mention above of situations in which there was an S

embedded to the case category 0 suggested the ways in which case

grammar must deal with verb and adjective complementation. A

second source of embedded sentences is within the NP itself. The

rule for NP may be stated as (97).

(97) NP--- N (S)

Where the N is an ordinary lexical item and the adjunct S contains

a co-referential copy of the same N, the result is a NP consisting of

a noun modified by a relative clause. One of the most obvious sources

of 'genitive' is from relative clauses built on sentences which, by

themselves, would have assumed the form X has Y. The N in the modi-

fied NP is the same as the N contained in the D of the adjunct sen-

tence, and the V is empty. Thus, from (98) we get (99) by deleting

the repeated noun, the tense, and the 'empty' verb, and reattaching

the D to the dominating NP.

 - 78 -

(98) NP

N P

7
books pres 0 0 books to John

(99) NP

to John books

A D subjoined to a NP has its case marker modified--in this case to

the sibilant suffix. Note (100).

(100) NP

D N

NP
I
John 's books

The 'true possessive' construction--resulting either in a

NP of the form X's Y or Y of X in English--has as its source a

sentence which by itself would have the form X has Y. The fact

that in some languages there are instances of adnominal D not modified

to the 'genitive' (dery Vater sein Haus, 'Dative of Possession')

supports the view that conversion to genitive is a matter of the

surface structure.
 -79-

The treatment of de7erbal nouns which seems most satisfactory

to me is that, except for the purely productive cases, the deriva-

tion of a noun from a verb is a matter of historical not synchronic

fact. The synchronic reality is expressed by indicating that a

given noun has a particular kind of relationship to a specific verb

(or set of verbs), and that some of these nouns may, others must,

appear in the NP frame [ 2 ] .

That is, instead of having a synchronic process for producing

such words as Latin amor from its associated verb, wnat is needed

is the classification of such a word as an abstract noun having a

83
particular kind of relationship with the verb amo. Nouns having

3This treatment allows for the inclusion of nouns which lack

etymological connections with their related verbs. We might wish
to indicate for book a connection of the intended kind with the verb
write, thus accounting for the ambiguity of your book between the
book which you own (ordinary relative clause modification) and the
book which You wrote.

this kind of special relationship to specific verbs can take part in

a process which introduces into the NP elements which 'originally'

depended on the associated verb. The processes in question fre-

quently have the effect of converting the form of the subsidiary NPs
84 Thus the noun amor when qualified by a sentence
to the genitive.

Tr
Exactly what universal constraints there are, if any, on the
element to be converted to genitive is not at all clear. It appears
that if there is only one element that shows up in the NP, it
frequently takes the genitive form. Compare the ambiguous sentence
(i) with sentences (ii) and (iii).
(i) my instructions were impossible to carry out
(a) so I quit
(b) so he quit
(ii) my instructions to you are to go there
 (iii) *my your instructions are to go there
In English it appears that if the conditions which allow the form
of the of-genitive and the s-genitive are satisfied by two different
NPs in the associated sentence, multiple genitive constructions
become possible, as in the following example borrowed from Jespersen.
(iv) Gainsborough's portrait of the duchess of Devonshire
Japanese allows conversion to genitive in true relative clauses,
as well as in the reduced relative clauses. A paraphrase of (v) is
(vi); no is the postposition most closely associated with functions
which we would call 'genitive'.
(v) boku ga yonda zasshi 1I+subject+read-past+magazine 1

"the magaiines I read"

(vi) boku no yonda zasshi

of the form deus amat . . . yields amor dei; when qualified by a

sentence of the form deum amat . . . the result is again amor dei.

The D and 0 forms, in other words, are equally reduced to the geni-

tive, resulting when only one noun is involved, in potential

85
ambiguities.

Jespersen's suggestion that the ambiguity of amor dei is in the

verb rather than the noun--the noun unambiguously identifying the sub-
ject, the verb being ambiguously either active or passive--must be
understood as the hypothesis that only those NP constituents which are
capable of conversion to surface subjects (with a given verb) may
appear under genitive modification as modifiers of the deverbal
noun. For English this may well be true. (Jespersen (1924) 170).
4.0., Some remarks on language typology

The view of universal grammar which is emerging is something

like this: In their deep structure, the propositional nucleus of

sentences in all languages consists of a V and one or more NPs,

each having a separate case relationship to the P (and hence to

the V). The most straightforward deep structure commonalities

between languages are to be sought at this 'deepest' level.

The lexical insertion rule for verbs is sensitive to the

particular array of cases in the P. Since no distinction is needed

between 'strict subcategorization features' and at least the highest

level of 'selectional features' (because redundancy relations exist

between cases and some lexical features, and because there is no

'subject' outside of a 'VP' whose features need to be dealt' with

separately), the lexical insertion rule for verbs can be a strictly

local transformation which responds to nothing more than the cases

which are co-constituents of V (with the exception, as noted above,

that it must be known whether the 0 element is a NP or a S).

The criteria for typological classification that have suggested

themselves so far in this study are these:

I. the presence or absence of modifications on the NPs as

determined by the deep case categories

A. the nature of such modification (prepositional,

affixal, etc.)

B. the conditions for the choice of particular case forms

(which, when stated in their simplest form, constitute

 what is usually formulated as the 'case system' of

the language)

the presence or absence of concordial modifications

of the verb

A. the nature of the concord (number agreement, incor-

poration of 'traces' of case categories, feature

changes on V, etc.)

B. the relation to sublect selection (topicalization)

III. the nature of anaphoric processes

A. type of process (replacement by pro-form, deletion,

g

destressing, replacement by unstressed variants, etc.)

B. conditions of application

IV. topicalization processes (where 'subject selection' may

be thought of as a special case of topicalization)

A. formal processes (fronting, modifying the case form,

etc.)

B. the variety of topicalization processes in the same

language

V. word order possibilities

A. fa0:4-;ors determining 'neutral' word order (nature of

case categories, 'ranking' of noun classes, topic

selection, etc.)

B. conditions determining or constraining stylistic

variations on word order.

It is important to realize that all of these typological criteria

are based on superficial processes, and that there are no particularly

Lb,good reasons for believing a priori that there will be much coincidence
 - 83 -

in the ways in which the different criteria sort out the world's

languages.

4.1 The bases for determining case forms

The forms of the NPs in a P are determined on the basis of a

variety of factors, one of which is the case category of the NP.

Thus a NP under I (i.e., an instrumental noun) is assigned a parti-

cular form depending in whole or in part on the fact that it is

under I.

Surface case forms of NPs are most elaborately developed in

the personal pronouns. The study of the 'case' aspects of pronoun

systems reveals a great deal about the variety of relationships

that can hold between deep and surface cases.

Sapir's typological distinctions for Amerindian pronominal

86
systems can be expressed in case - grammar terms quite simply.

86 Edward Sapir, review of C. C. Uhlenbeck's "Het passieve karakter

van het verbum transitivum of van het verbum actionist in talen van
Noord-Amerika," in INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS, I
(1917) 82-86.
INV

If we ignore whatever complications may exist in 'passive' construc-

tions, and if we ignore all deep structure cases except A and 0, we

can imagine sentences of the following three types given in their

underlying propositional form):

(a) V + A intransitive sentences with active 'subjects'

(b) + 0 + A transitive sentences with agents

(c) V + 0 intransitive sentences with inactive

'subjects'
 - 84 -

Since the V element is constant to the formulas, we can represent

these three sentence types by presenting the case frames in three

lines, as follows:

(101) A
0 A
0

According to Sapir, then, there are languages which, like

Yana, have only one form for pronouns in all four of these

positions.

(102).

There are languages like Paiute that have a separate form for

the 0 element in the transitive sentence, all others being the

same. The two forms are traditionally called 'nominative' and

'accusative'.

(103)

accusative 0 A 4.- nominative

There are languages like Chinook which give one form to the A

of transitive sentences, another to the remaining cases. The

terms 'ergative' and 'nominative' are often given to a distinction

made in this way.

(104)
nominative mo ergative
 -85-

There are languages like Dakota which have separate forms for

A and 0; here the terms are usually 'active' and 'inactive'.

(105 )
active

And, lastly, there is the situation found in Takelma, which has

one form for the pronominal NP of intransitive sentences, and two

separate forms for the A and 0 of transitive sentences. Thus:

(106)

What these observations are intended to suggest is merely

that if I correctly understand Sapir's analysis of the pronominal

systems of these languages, then the case concepts I have been

discussing, together with the notion of clause types which various

arrays of them define, provide the categorial and configurational

information for determining the surface distinctions that are found

in these languages.

4.2 Verbal concord

There are these various ways, and possibly more, in which

cases and case environments are involved in determining the case

 -86-

forms of NPs within a P. An additional factor is whether the

given NP has been chosen as subject in languages having subjecti-

valization processes. Choosing subjects or topics is related to

another aspect of the superficial structure of sentences, and that

is verbal concord.

The choice of subject in English always has the effect of

determining number concord (on those verbal and auxiliary elements

capable of reflecting number concord). Apart from number concord,

the choice of subject might involve modification of the verb to its

passive form, or introduction of the verb have.

The information 'registered' in the V may have only to do with

the choice of subject, as in English, or it may be more elaborate.

Languages which 'incorporate' pronominal affixes into the V may do

so for more than one NP at a time; or noun stems themselves from

87
particular cases may be incorporated into the verbal expression.

7Grammatical devices for providing concord of this type have been

worked out for subject and object incorporated pronominal prefixes
in Mohawk by Paul M. Postal. See his "Mohawk prefix generation,"
PROCEEDINGS OF THE NINTH INTERNATIONAL CONGRESS OF LINGUISTS,
Mouton (1963), 346-355.

The subject selection rules discussed for English may be

compared with the topicalization processes that are described for

%IN

Philippine languages. The situation for Maranao has recently been

described by licKaughan. One NP is chosen as topic for every sentence,

and this choice is recorded in the following way: its original case

preposition is replaced by so, and an affix is inserted into the V

which indicates the case category of the chosen NP. There is

apparently considerable freedom in the choice of topic. To take the

verb meaning 'to butcher' (/sombali9/), we find that when the topic

noun is an original I, the verb takes on the prefix /i-/, as shown in

(108); and when the topic is an original B, the suffix /-an/ is

added to the V, as seen in (109).

(107) somombali9 so mama, sa karabao

'the man butchers the carabao'

(108) isombali9 o mama9 so gelat ko karabao

'it is with the knife that the man butchers the carabao'

(109) sombal9an o mama9 so majoir sa karabao

'it is for the mayor that the man butchers the

carabao188

Howard McKaughari, "Overt relation markers in Maranao, " LANGUAGE,

XXXVIII (1962) 47-51. The examples and the description of the
relationships are from McKaughan, but a great deal of guessing lies
behlnd my interpretation.

The choice of sentence subjects, or 'topics', from particular

casas appears to be the most satisfactory way of accounting for the

many types of voice modifications of verbs such as those described

as middle, pseudo-reflexive, etc., in the Indo-European languages.

4.3 Anaphoric processes

Anaphoric processes are best understood from the point of view

of an extended concept of sentence conjunction. That is, every

language has ways of simplifying sentences connected by conjunction

or subjunction) and the processes used under these conditions seem

to be exactly the same as those used in sentences connected in

discourse. The grammarian's job, therefore, is to describe these

processes as they work in fully independently intelligible

sentences, and then to assume that the utterances that occur in

connected texts or conversations can best be understood from the

point of view of a shared knowledge of the language's anaphoric

89
processes on the part of speaker and hearer. The fact that in

In other words, the grammarian will describe the process by which

(i) in converted to (ii) by noting the conditions under which repeated
elements in conjoined sentences may undergo deletion and pro-replace-
ment and under which conjoined sentences can have words like too and
either added to them.
(i) Mary didn't want any candy and Mary didn't take any
candy
(ii) Mary didn't want any candy and she didn't take any,
either
In contexts in which the information contained in the first conjunct
of (i) is already understood by the addressee (by having just been
spoken by him, for example), a speaker of English feels free to use
the reduced form (iii).
(iii) she didn't take any, either
There is no reason, it seems to me, to expect the grammar of a lang-
uage to generate sentences like (iii) directly.

these anaphoric or reduced forms English uses pro-replacement under

conditions that would call for deletion in some other language may

thus be seen as a superficial difference between the two languages.

The point is important--and it was mentioned above in connec-

tion with 'bad' reasons for rejecting the universality of the

subject/predicate division--because the absence of subjects in the

final surface forms of sentences in some languages is seen by many

scholars as having great typological relevance. The optional absence

of NP constituents in languages with person-marker incorporation

(e.g. Chinook) has led scholars to claim that such languages lack

the nexus relations that Europeans understand as 'subject' and 'object'

 relations
but have instead what are described as 'appositional'
90 In languages without pronominal incorporation,
between NPs and Vs.

90 See, e.g., the statement in Alf Sommerfelt, "Sur la notion du sujet

en georgian," MELANGES DE LINGUISTIQUE ET DE PHILOLOGIE OFFERTS A
JACQUES VAN GINNEKEN, Paris (1937) 183-185.

a distinction is made by some scholars between true subject/predicate

languages and those in which the so-called 'subject' is as much a

'complement' to the V as is the direct object or any of the various

adverbial elements. To Martinet, a subject is different from a

complement only if it is "constitutive of the minimal utterance"- -

that is, only if it is obligatorily present in both full and anaphori-

91
cally reduced utterances. In Japanese, the 'minimal utterance'

=1,11,..11M
9
'Andre Martinet, A FUNCTIONAL VIEW OF LANGUAGE, Oxford-Clarendon
(1962), 61-62.

lacks a subject, and hence, the argument goes, Japanese sentences lack

the subject/predicate structure of sentences in our more familiar

languages. To Martinet's disciple Saint-Jacques, this typological

'fact' about Japanese is regarded as excessively important. It is

only by dint of considerable intellectual effort that the Westerner

can achieve that liberation from familiar ways of thinking about

language which ic required for an understanding of the true character

92
of Japanese. Or so Saint-Jacques tells us. It seems to me that

92 Bernard Saint-Jacques, ANALYSE STRUCTURALE DE LA SYNTAXE DU

JAPONAIS MODERNE, Paris-Klincksieck (1966) Pp 121; 36.
language typology offers enough genuine excitements to make it

possible for us to give this one up. The intellectual achievement

of which M. Saint-Jacques speaks is that of knowing that when there

is an 'understood' NP to deal with, some people replace it by a

pronoun, others get rid of it.

4.4 Topicalization

The fourth criterion has to do with topicalization processes,

devices for isolating one constituent of a. sentence as 'topic', of

bringing one particular constituent of a sentence into some kind of

'focus'. Where topicalization is distinct from processes for

'emphasizing' a constituent, we have much the same thing as what

have been calling 'subjectivalizationl, but which I shall now begin

calling 'primary topicalization'. Primary topicalization for English

involves position and number concord; stylistic changes involving

stress assignment, late word order changes, and possibly the 'cleft-

sentence construction', fall into what might be called 'secondary

topicalization?. From what I understand of McKaughan's account,

primary topicalization in Maranao involves replacement of the original

preposition associated with a noun by so and introduction into the V

of an associated case indicator, while secondary topicalization

involves moving an NP to which so has been added to the front of the

sentence.93 One might refer to Oertel's study of the disjunct use

a
93McKaughan (1962) 47f.

94
of cases in Brahmanic prose as a study of secondary topicalization.
 - 91 -

Hanns Oertel, THE SYNTAX OF CASES IN THE NARRATIVE AND DESCRIPTIVE

PROSE OF THE BRAHMANAS9 Heidelberg (1936) Pp. 364. Oertel distinguishes
'pendent' uses of edisjunet case, where the 'topic' is in the 'nomin-
ative' even if its original role in the sentence was not that of
subject (comparable, I assume, to he in he, I like him), and "pro-
leptic' uses, where the topic retains the original case form, is moved
to the front of the sentence, and may or may not be resumed (in the
form of a demonstrative) in the remainder of the sentence (comparable
to him in him, I like (him)).

I would imagine that all languages possess some means of carrying out

'secondary topicalization', but it may be the case that some lack

the process of 'primary topicalization' Osubjecti.valization0.95

95Jeffrey Gruber's recent study of topicalization in child language

suggests that ontogenetically, motivated (what I am calling 'secon-
dary') topicalization precedes the use of formal subjects in English.
It may be that when one device for topicalization becomes 'habitual',
it freezes into a formal requirement and the language must then call
on other processes for motivated topicalization. (See Jeffrey Gruber,
"Topicalization in child language," FOUNDATIONS OF LANGUAGE, III
(1967) .) Kenneth Hale reports that for Walbiri, an
'ergative' language of aboriginal Australia, there is apparently no
'subjectivalization' process, but any constituent may be repeated
to the right of the proposition, the element inside the proposition
being replaced by a pro-form. (Correspondence, 1967).

The notion 'subjectivalization' is useful only if there are

sentences in a language which offer a choice of subject. Languages

described as not having passives, or languages described as only

capable of expressing transitive sentences passively, apparently

lack the grammatical process of primary topicalization.

This question leads naturally to the problem of the so-called

'ergative' languages. Recall that in the accusative type of pronominal

system, the pattern was

 -92-

in it
and thatAthe ergative typepas

Now when languages of the accusative type have passive versions

of sentences whose propositional form is [V 0 A], the case forms

associated with the elements in the passive version are generally

'nominative' for the 0 and 'agentive' (realized as ablative, instru-

mental, or what have you, depending on the language) for the A. If

passive sentences were introduced into our three-line diagrams and

their active counterparts removed, we would get the pattern

nominative agentive

which is exactly like that for the regular assignment of cases in

the ergative languages. This fact, plus the use of the term

' nominative' for subject-of-intransitive-cum-object-of-transitive

in these languages, has led many scholars to identify the ergative

case in ergative languages with the agentive case form found in

passive sentences in accusative languages, and to conclude that the

ergative languages are really 'passive' languages, languages in which

96
transitive sentences can only be expressed passively. For both of
 -93-

96
Note that even if there is a different form for the verb in
0 ] and [ 0 A ] case frames, this cannot be interpreted
as evidence of 'passivity'. As mentioned earlier, in languages not
of the ergative type there may still be systematic variation of the
same verb root depending on whether it is used transitively or
intransitively.

these systems, the case that has been given the name 'nominative' is

frequently described as the 'subject' in a subject/predicate construc-

tion, and the 'ergative' element in the one instance and the

'accusative' element in the other are treated as verbal complements. 97

97See, e.g., the statement in Nikolai S. Trubetzkoy, "Le rapport entre

le determine, le determinant et le defini," MELANGES DE LINGUISTIQUE,
OFFERTS A CHARLES BALLY, Geneva (1939) 75-82.

The difficulty of determining the !subject' in ergative languages has

been described by Martinet. Some scholars identify as subject the

word which would be the subject in a translation of the sentence

into French--i.e., the nominative in intransitive sentences and the

ergative in transitive sentences. Others regard the nominative as the

subject in all sentences, thus giving transitive sentences a 'passive'

interpretation. Lafon gives up on transitive sentences--he uses the

term 'subject' only for intransitive sentences, saying of transitive

98
sentences that they have no subject.

.,111
9 Andre Martinet, Le sujet comme fonction linguistique et l'analyse
syntaxique du Basque," BULLETIN DE LA SOCIETE DE LINGUISTIQUE DE
PARIS (1962) 73-82; 78f.

Valliant, on the other hand, spoke of the northern Caucasian

languages as having three types of verbs (i) true intransitives,

 - 94 -

with subjects in the 'nominative'; (ii) 'operative pseudo -

transitives', with 'pseudo-subject' in the 'ergative'; and (iii)

'affective pseudo- transitives', with 'pseudo-subjects' in the

'dative'. 99 It seems quite clear that what he is dealing with are

9
Vaillant (1936) 93.

sentences having Ps of the three types CV 0], [V 0 A] and [V 0 D],

where the surface cases for 0, A and D are 'nominative', 'ergative'

and 'dative' respectively. It looks very much as-if that is all there

is to say. For my part I would much rathar say of the ergative lang-

uages that they lack subjectivalizatio5 that all transitive sentences

undergo obligatory passivization, or that some of t!leir sentences

contain true subjects while others do not.

It is very frequently claiued that the ergative languages are

DOO
more primitive than the accusative langiiages, and with the

10
()See, e.g., Tesniere (1959) 112.

assumption that the ergative construction is really a passive con-

struction, this has led such scholars as KurOowicz, Schuchardt, and

Uhlenbeck, to assume that the passive construction represents a more

primitive concept in the evolution of language than that of the

active transitive construction. Evidence mustered for this position

includes the signs that pre-Indo-European was of the ergative type,

and the fact that some languages have 'invented' have-like verbs in

relative recent times. The invention of have made it possible to

give active expression to certain tense or aspectual forms which had

 - 95

remained unaffected by the general change from passive to active

expression (as is seen, for example, in the c. 3rd century shift

from expressions of the type inimicus mihi occisus est and mihi illud

factum est to transitive expressions using habeo: inimicum occisum

101
habeo and habeo illud factum.

101 Jacques van Ginneken, "Avoir et fitre du point de vue de la

linguistique generale," MELANGES DE LINGUISTIQUE, OFFERTS A CHARLES
BALLY, Geneva (1939) 83-92; 86.

It seems very unlikely to me that syntactic changes of the type

known from the present state of our knowledge are really eapable of

showing an intellectual evolution of a type as potentially significant

as whatever might be understood as the transition from an essentially

passive to an essentially active point of view. The connection

claimed by van Ginneken between ergativity and the 'feminine' character

of the cultures that have ergative languages is one which also deserves
102
to be brought into question.

102 The following seems worth quoting in full: "Nous sommes tous des
hommes, et tous nous avons deux talents: les facultes plus actives
de l'appetit et de la volontel et les facultes plus passives des
sensations et de l'apprehension; mais it est evident que les deux
sexes de l'humanite montrent sous ce rapport une difference sensible.
"Wethnologie moderne, qui a ecarte definitivement comme insuf-
fisante la doctrine du developpement uniforme, nous apprend cependant
que le progres de l'humanite a balance presque toujours entre les
cultures plus feminines ou plus masculines, dites cultures matriarcales
et patriarcales. Ce sont toujours les cultures matriarcales tries
prononcees qui, comme le basque, ont un verbe transitif de nature
passive avec comme casus rectus un patiens et comme casus obliquus un
agens; mais les cultures patriarcales, comme l'indo-europeenne ont un
verb transitif de nature active, animiste et magique, avec un sujet
au casus rectus et un object au casus obliquus. Chaque peuple a
donc le verbe qu'il merite." (van Ginneken (1939) 91Z)
 - 96 -

4.5 Word order differences

The fifth criterion suggested for a language typology is

that of word order. The variables that determine or constrain

the freedom of word order in the languages of the world are very

likely to have many important connections with the case structure

of sentences; but this is an area which I have not examined at all.

 - 97 -

5. The grammar of inalienable possession

The preceding sections have amounted to an informal descrip-

tion of a syntactic model for language and a few demonstrations

of the operations of the model of the sort that has come to be

called 'restatement linguistics'. In the present section I

shall attempt to show how a particular substantive modification

of the rules will permit a uniform way of describing the interes-

ting collection of grammatical facts associated with what is

called 'inalienable possession'.

Every language, one can be sure, has nouns which express

concepts that are inherently relational. Examples of inherently

relational nouns in English are side, daughter and face. One

doesn't speak of a side, but of a side of something; one doesn't

say of someone that she is a daughter, only that she is somebody's

daughter; and although it is possible of speak of having seen

a face, the word is typically used when referring to 'his face'

or 'your face' or the like. The relational nouns most frequently

discussed in the linguistic literature are names of body parts

and names of kinsmen. My discussion here will concentrate on body

parts.

5.1 The data

5.1.1 Significant syntactic relationships exist between the

dative and the genitive cases in all of the Indo-European lang-

uages; and in all but Armenian, according to Havers, the dative

 -98-

and the genitive case forms figure in paraphrase relationships

103
of kinds that are highly comparable from language to language.

103
Wilhelm Havers, UNTERSUCHUNGEN ZUR KASUSSYNTAX DER INDOGERMAN-
ISCHEN SPRACHEN, Strassbourg (1911) Pp. 335; 317.

The relationship is observed only when the associated noun is of

a particular type. To take some of the modern German examples

given by Havers, we observe that a paraphrase relation exists

between (111) and (112) as well as between (113) and (114); but

that of the two sentences (115) and (116), the latter is ungram-

matical (as a paraphrase of (115)).

(111) die Kugel durchbohrte dem Feind das Herz

(112) die Kugel durchbohrte das Herz des Feindes

(113) er hat mir die Hand verwundet

(114) er hat meine Hand verwundet

(115) der Vater baute seinem Sohn ein Haus

(116) *der Vater baute ein Haus seines Sohnes

It whould be noted that Herz and Hand are the names of body parts,

while Haus is not.

5.1.2 There are cases like the above where a given language exhibits

in itself the paraphrase relationship, and there are also cases

where it appears that one language has chosen the dative expression,

another the genitive. Notice the following sentences, also from

Havers.

(117) my heartaches; mir blutet das Herz

 (118) Tom's cheeks burned; Tom brannten die Wangen

(119) she fell on her mother's neck; aie fiel ihrer

104
Mutter um den Hale

Havers (1911) 1

5.1.3 There are adnominal (possessive) uses of dative construc-

tions, particular, it appears, when the possessive pronoun is

also used with the possessed element. Here the most readily

available examples are with kinship terms.

(120) dem Kerl seine Mutter

(121) sa mere a 1ui105

105
Examples are from Havers (1911) 283

5.1.4 Many languages have separate possessive affixes for nouns

that are obligatorily possessed (j.nalienables) and nouns that are

optionally possessed (alienables). The difference in Fijian is

apparently expressed by pre-posing the possessive morpheme to

indicate alienable possession, suffixing it to indicate inalien-

able possession. Since the category 'inalienable' is a category

of grammar rather than a property of real world objects (since,

in other words, some objects grammatically classed as inalienable

can in fact be separated from their !owners'), the distinction

can be seen most clearly if both methods of expression can be

used with the same noun stem. 'IAvy-Bruhl gives a persuasive

example of this situation: Fijian uluqu means the head which is

 -100 -

now firmly attached to my neck, while kequ ulu, also translatable

as 'my head', would refer to the head which, say, I am about to

106
eat.

106
L. Levy-Bruhl, "L'expression de la possession dans les langues
melanesiennes," BULLETIN DE LA SOCIETE LINGUISTIQUE DE PARIS,
XIX (1916) 96-104; 99.

Languages may have separate morphemes for indicating alienable

and inalienable possession, and they may have further distinctions

among these morphemes depending on the type of inalienable poss-

ession (as Nootka, for example, suffixes -?at- to nouns representing

physically inseparable entities, e.g. body parts, but uses other

means for kinship terms), or they may merely have a class of nouns

incapable of occurring as free forms--noun stems requiring affix-

107
ation of possession indicators.

107
This last situation is sometimes described by saying that
nouns are 'inflected for person'. See Gabriel Manessy,
relation genitive dans quelques langues mande," PROCEEDINGS OF
THE NINTH INTERNATIONAL CONGRESS OF LINGUISTS, Mouton (1964)
467-475; 468.
The full variety of the treatment of inalienable possession
in different Amerindian languages is catalogued in Edward Sapir's
review of C. C. Uhlenbeck, "Het identificeerend karakter der
possessieve flexie in talen van Noord-Amerika," in INTERNATIONAL
JOURNAL OF AMERICAN LINGUISTICS, I (1917) 86-90.

In all of these cases, it appears, the features in question

are 'grammatical' rather than purely 'notional'. Discussions of

inalienable possession almost always contains lists of nouns

whose grammatical classification is the opposite to what one

would notionally expect. Levy-Bruhl mentions a case where the

 - 101 -

worf for 'left hand' functions as a body...Tart word grammatically,

108 And Arapaho classifies
but the word for 'hand' does not.

lO8L6vy
-Bruhl (1916) 96.

109
'louse' (or 'flea') among the inalienables, a situation that

10
9Zdenik Salzmann, "Arapaho VI: noun," INTERNATIONAL JOURNAL OF
AMERICAN LINGUISTICS, XXXI (1965) 136-151; 139.

invites people who like to speculate on these things to propose

something or other on the Arapaho conception of 'self'.

5.1.5 Mika Ivi6 has recently discussed many instaaces of what

she calls 'non-omissible determiners'. Among the examples she

cites are many that involve nouns of the type frequently

included among the inalienables. The adjective cannot be deleted,

for example, in the Serbo-Croatian expression (122), for (123)

110
is ungrammatical.

110
Milka Ivi6, The grammatical category of non-omissible
determiners," LINGUA, XI (1962); and "Non-omissible determiners
in Slavic languages," PROCEEDINGS OF THE NINTH INTERNATIONAL
CONGRESS OF LINGUISTS, Mouton (1964) 476-479; the example is
from p. 477.

(122) devojka crnih oaiju "the girl with black eyes"

(123) *devojka aiju

What is misleading about her discussion, it seems to me, is the

decision to associate with the adjective the 'category of non-

omissibility'. It is as if we wished to say, for the English

sentence (124), that there is amething grammatically significant

about the word missing, since its deletion results in a

sentence (125) which is somewhat different in type from the

original; put differently, sentence (124/ does not say the

same thing that (126) does. What is genuinely important about

(124) is its paraphrasability as (127) Lor (128)) and the fact

that the construction exhibited by (124) is restricted to cer-

tain kinds of nouns. Note the ungrammaticality of (129):

(124) I have a missing tooth

(125) I have a tooth

(126) I have a tooth and it is missing

(127) my tooth is missing

(128) one of my teeth is missing

(129) *I have a missing five-dollar bill

5.1.6 Note that in sentences (124) and (127), three things are

involved: (a) .a Possessor (an 'Interested Person', to use the

traditional term), (b) a Body Part, and (c) an Attribute --(a)

me, (b) tooth, and (c) missing respectively--and that the

sentences provide alternate ways of ascribing the Attribute to

the Possessor's Body Part. They are two distinct superficial

ways of expressing the same relationship among these three con-

cepts.

Using P, B and A for (a), (b) and (c) above, we mny represent

the expression as seen in (124) as (130) and that as seen in (127)

as (131).

(130) pnom have CA -0 Bacc)

(131) Dgen -0 B] be A
 -103-

The same element, in other words, which in some of the para-

phrases mentioned above appeared in either the dative or the

genitive case forms, is now also seen appearing as the subject

of the verb have. Bally, in fact, speaks of the invention of

the word have as having precisely the function of allowing the

'personne interesseel, which otherwise would have to appear

either in dative or genitive form, to become the subject of a

sentence. Examples of all three surface appearances of a first

person Possessor are given by Bally as (132)-(134). (133) and

(134) correspond to expression types (130) and (131) respectively;

the expression type exemplified by (132) is given as (135).

(132) mihi Bunt capilli nigri

(133) j'ai les cheveux noirs

(134) mes cheveux sont noirsill

pdat [Boom A]

111 Charles Bally, "L'expression des idees de sphere personnelle

et de solidarite dans les langues indo-europeennes," FESTSCHRIFT
LOUIS GAUCHAT, Aarau (1926) 68-78; 75.

5.1.74 Henri Frei surveyed this variety of surface representa-

tionsAthe 'same' sentences, and added a fourth type, a type inter-

mediate, in a sentence, between that suggested by formula (135)

and that of (130). His example was sentence (136) (which also

provided the title of his paper), which , xemplifies the expression

type we may wish to represent as (137).

(136) Sylvie est jolie des yeux

(137) pnom [A Boblique].
 - 104 -

Frei points out that the construction seen in (136) is related

to the category of inalienable possession, since while (138)-

(139) are acceptable sentences, (140)-(141) are not.

(138) elle est fine de doigts

(139) elle est bien faite des jambes

(140) *elle est fine d'etoffe

(141) *elle est bien faite des vitements112

112
Henri Frei, "Sylvie est jolie des yeux," MELANGES DE LINGUISTIQUE
OFFERTS A CHARLES BALLY, Geneva (1939) 185-192; 188. The express-
ions are limited to clear relational nouns, not only to body parts.
Frei notes such phrases as "des couloirs spacieux et bas de plafond"
and "fibre de moeurs." He beautifully demonstrates the distinct-
ness of the sentences involving inalienable possession from
overtly similar sentences of different grammatical structures with
the contrast between (i) and (ii) below. (p. 186)
(i) la salle est pleine de visages
(ii) la femme est pleine de visage

5.1.8 'Since Frei sees this diversity as resulting from the

attempt to 'condense' two judgments into one sentence--the two

judgements that P has B and that B is A (in our terms)--he

relates the constructions in question to the much-discussed

edoubleeubjecte constructions of Japanese. In one type of this

latter construction, two nouns appear before a verb or adjective,

the first followed by the particle bra (indicating what I have

called 'secondary topicalization'), the second by the particle

a (the particle of 'primary topicalization'). (Variations on

the order and in the choice of particles do not change the status

of the construction; the form described is the one most stylisti-

cally neutral.) The second of these nouns is of the inalienable

type, the first identifies the object with respect to which the

object identified by the second noun is 'inalienable'. The

hackneyed example of the double-subject construction is item

(142), a sentence which has (143) as a sort of forced paraphrase.

In (143), the particle no is the particle whose functions are

closest to those we would be inclined to label 'genitive'.

(142) zoo wa hana ga nagai "elephant wa nose La long"

(143) zoo no hana ga nagai.

5.1.9 That expressions involving entities viewed as being

closely associated with an 'interested person' have unique

grammatical properties has also been observed in certain seman-

tically unmotivated uses of 'reflexive pronouns' and the parallels

one finds between these and various uses of the 'middle voice'.

The connection with dative forms is seen in the fact that in

some languages a kind of 'dative reflexive' is used in these

special situations. Note items (144) and (145).

(144) se laver les mains

(145) ich wasche mir die Hande

The connection between this use of the 'reflexive' and the

category of inalienable possession is indicated by BalXy where

he points out that in item (146), jambe is the inalienable

entity, while in (147) the word jambe can only (or, depending on

my informants, can also) be understood as some independently

possessed object, such as the leg of a table.

(146) je me suis cease la jambe

(147) jai casa6 ma jambe

Notice that the jambe which does not have the possessive adjective
 - 106 -

is the one which is grammatically characterized as 'obligatorily

113
possessed':

113
Bally (1936) 68f.
aMIMOMIIMP

5.2 Adnominal datives

One way of introducing a possessive modifier of a noun has

already been suggested: a sentence which could on its own

assume the form 'X has Y' is embedded to NP. Since it is desir-

able for an embedded sentence to have a semantic interpretation

that contributes to the meaning of the whole sentence, the sentence-

embedding source of possessives is needed as an explanation for

alienable possession. In other words, one is satisfied to have

the meaning of (148) represented as a part of the meaning of

(149), though we may reject such a zelationship between (150)

and (151).

(148) I have a dog

(149) my dog

(150) I have a head

(151) my head

A distinct method is required for introducing the possessive

element in the case of inalienable possession, a method which

reflects the fact that the relationship between the two nouns

in 'inalienable possession' is not (Else Frei) a sentential

relationship.
 For the types of inalienable possession that we have

considered so far--in which the relationship haE zaways been

to an animate or 'personal' entity--thb solution is to say

that some nouns obligatorily take D complements. This can be

managed by adding to the grammar another way of writing NB,

namely rule (152).

(152) NP --m N (D)

In the way that frame-features for Vs relate to environments

of Vs provided by the constituent P, frame-features for Ns

relate to environments provided by the constituent NP. It was

suggested above that Ns which obligatorily take S complements

are assigned the feature 4 S ]. We may now add that Ns

which obligatorily take D complements are characterized as

having the feature 4 D ]; and these are the inalienably

possessed nouns. The notation imposes a subclassification of

nouns into those which require adnominal D (such as son, child

in the meaning 'offspring', German Mann in the meaning 'husband')

and those which reject adnominal D (such as person, child in

the meaning 'very young person', Miatin in the meaning 'man').

The two sources of possessive modifiers which the grammar

now makes possible (adnominal D and adnominal S of a certain

type) provide the deep-structure differences needed for deter-

mining the difference in the form of the possessive modifiers in

those languages which make the distinction overt in that way..

Where further distinctions are made (as between body parts and

kinship terms), the information on which such distinctions need

to be based may be included as lexical features of the Ns themselves.

 - 108-

The general configuration of NPs containing Ds, then, is

that shown in (153).

(153) NP

K
Nr
N

In some cases the adnominal D remains in the NP and in fact

retains the surface features associated with D, as in item (154);

typically, however, a D inside a NP is changed to a genitive form,

as in (155).

(154) secretary to the president

(155) the president's secretary

114
If determiners are universal, then the expansion of NP

1177--
I am
inclined to think that they are. See my "The syntax of
English preverbs," to appear in GLOSSA, II (1967).

must make provisions for them; but if they are not, then those

languages which have them will need 'segmentalization' rules of the

type described by Postal115. At any rate, the determiners (which

115
Paul M. Postal, "On so-called 'pronouns' in English," GEORGETOWN
UNIVERSITY MONOGRAPH NO. 19, LANGUAGES AND LINGUISTICS (1966),
177-206.

I represent as 'd') will figure in the various things that can

happen to adnominal D. Sometimes, for example, when a D remains

in the NPvithout undergoing genitive modification, certain of

its features are copied onto the determiner so that the determiner
 - 109 -

may eventually assume the form of the appropriate 'possessive

adjective'. This seems to account for such expressions as the

possessive dative with kinship terms seen in some German dialects

116
(recall example (120)) in Ossetic, etc.

116 V, I. Abaev, A GRAMMATICAL SKETCH OF OSSETIC, Indiana Univer-

sity Pr )ss (1964), 18.

5.3.Some illustrations

The D constituent often need not remain the NP: under some

conditions it may be 'promoted', so to speaks from the status

of a modifier of a N (which it is in the deep structure) to the

status of a major constituent on the next higher level of the

syntactic structure. This can be seen in sentences having the

base configuration [ V + L + A ]: just in case the N under L is

a body part, the D which in the deep structure is subjoined to L

is 'promoted' to become a constituent of P, yielding a sentence

superficially of the type + D + L + A].

The verb pinch is accepted into the case frame [ L + A],

and, except when it has taken on the feature E+Passive], it is a

verb which deletes the preposition of the following constituent.

Let us consider sentences derivable from the deep structure seen

in (156).
 - 110-

(156)

K NP

d N

K'13
N

past pinch on the nose to John Mary

We shall see what happens to the sentence under four conditions:

when the D remains inside L and A becomes the subject; when the D

remains inside L and L becomes the subject; when the D is pro-

moted and A becomes the subject; and when the D is promoted and

D becomes the subject.

Whenever D remains inside NP (in this sentence), it is pre-

posed to the N and converted to its genitive form, displacing the

original determiner. Since it is a personal noun, the K element

assumes the form of a genitive suffix. With non-promoted D, in

other words, (156) eventually becomes (157).

(157)

V L

K NP K NP

NP K
I

past pinch on John el

s nose Mary
 -111-

Diagrams (158) - (161) show the development from (157) when A

is made the subject: the subject-preposition is deleted and its case

category is erased; the preposition after pinch is deleted and the

case category L is erased; and the tense is absorbed into the V.

(158)

K NP V
1

K NP

D N
/%6.
NP
I I

by Mary past pinch on John ,s nose

(159)

NP

N V

D N

NP
I I

Mary past pinch on John !s nose

(16o)

NP

11

NP
I I

Mary pant pinch John nose

 - 112 -

(161)

NP
I

N V NP

NP
Is
May pinched John nose

If the L of (157) is chosen as subject instead of the A, the

result is (162). This choice of subject requires the V to assume

the feature [ +Passive], which causes it to lose its ability to

delete following prepositions and its ability to take tense affixes.

The surface structure eventually resulting from (162) is (163).

(162)

+Pass
K NP

NP N
i 1
on John 's nose past pinch by Mary

(163) S
T
xP M P

D N V

NP K
1P
N

John nose was pinched by Mary

 - 113 -

Backing up to (156), we may now see the consequences of

'promoting' adnominal D. When the D is removed from L and

becomes the left-most case constituent in P, the resulting

structure is (164).

(164)

NP K P K NP

1 1 1 I 1

past pinch to John on the nose by Mary

The possible subjects for (164) are the A or the newly promoted

D. When the subject is A, we get (165), a structure which, on

application of the rules we have learned, eventually becomes (166).

(165)

NP

11 I d N
I I

past pinch to John on the nose

(166)

xr

N .
NP
I

1
Mary pinched John on the nose
 - ilk -

When D is made subject, on the other hand, we get (167); on

applying the rules appropriate to a V with the feature [4-Passive],

we eventually get (168).

(167)

D M

NP
INP
// N
1 di

John past pinch on. the nose by Mary

(168) S
1
NP M
V
Ef-Pas]
K NP K NP
1

d/ \N'
1
John was pinched on the nose by Mary
I

We may turn to the problem which interested Bally and Frei

and examine the role of adnominal D in sentences which assign

attributes to obligatorily possessed elements. The basic structure

of such sentences can be illustrated by item (169).

 - 115

K NP

I
pres beautiful eyes to the girl

In languages which allow the D to remain in the NP, the D element

is converted to its genitive form. In English this results in (170).

Since (170) has only the form [ V + 0 ], the 0 is necessarily

chosen as subject; the result for English is item (171).

(170)

NP

NP K

d N

Pres beliutiful 0 the girl 's eyes

 - 116 -

(171)

NP

1
the girl ' eyes are beautiful

Notice that since the V is an adjective, it is incapable of

117
'absorbing' the tense , requiring the provision of be within

117 Stated more accurately, Vs which are adjectives, passives or

progressives are incapable of absorbing the rlghtmost affix in M.

the M constituent. (171) is a rendering of (170) in which the V

is predicated on the 0 and the D is subjoined to the 0. Thus it

is analogous to our earlier sentences (127) and (134) and is of

the type indicated in (131).

Suppose next that the D of (170) does get 'promoted'. The

result of introducing the D in this way as an immediate constituent

of the P is item (172).

(172)

NP

d N
I IN
pres beautiful to the girl 0 eyes
 - 117 -

Some languages allow the 0 of configuration (172) to become the

subject, with the D element remaining in the expected surface form

for D, as in sentence (132). Others allow the D element to undergo

secondary topicalization when the 0 is subject, resulting, for

example, in one case of the 'double-subject' construction in

Japanese (recall (142)). The general expression type for senten-

ces resulting from (172) when 0 becomes subject is suggested by

formula (135) above.

Many languages allow the D to become subject. When this

happens and there are no other changes, the 0 appears in some

oblique case-form. This is so because, since beautiful is not a

true verb, the body-part word cannot be converted into an 'object'.

The initial structure is seen in (173); it is one which is not

typical of English, though it is perhaps seen in such expressions

as those given in (174) and it may represent a stage in the deriva-

tion of phrase of the type given in (175).

(173)

the girl pres beautiful 0 eyes

(174) tall of stature; blue in the face; etc.

(175) broad-chested; fat-legged; etc.

 =1111.01k

- 118 -

It appears to be the structural form underlying (136), whose

expression type is given in fermIlla (137). The construction is

apparently quite rare in French; Frei speaks of it as a 'short-

circuited' version of the sentences with have.

Another possibility, when D is subject, is to attach the

adjective to the NP indicating the body-part. I propose, in an

unhappily quite ad hoc fashion, that this be done without removing

the constituent label V. I believe there are some arguments for

retaining at least an abstract V under P at all times, so this

constraint may turx out to be better motivated than it seems.

The reason for this decision is that this appears to reflect what

actually happened in those languages which adopted a verb like

have.

The structure I have in mind is that shown in (176).

NP

V N
1-----"-----'""'"---...1
the girl pres 0 0 beautiful eyes

With the V under P vacant, the M must have a be added to it in

those languages which allow structures of this type to become

sentences directly. Notice that in this construction the modified

body-part NP as a whole is 'in' some case form. The formula for

this expression type has not yet been given; it would be something
 - 119-

like (177).
oblique
(177) P be EA-0B]

Conceivably this is the structure underlying such predicates as

those shown in (178); the difference between predicates of the

type (137) and those of the type (177) is seen in the Latin

paraphrases (179) and (180) respectively.

(178) of tall stature; di bello aspetto

(179) aequus animo

(180) aequo animo

The last possibility, then, is to insert into the vacated

V position the function word have, a verb which takes the modi-

fied body-part noun as its 'object'. In English, we have seen,

this involves deleting the preposition. The result of modifying

(176) in this way is (181).

(181)

V NP

V N

the girl pres have beautiful eyes

In short, it appears that the considerable surface variety

found in sentences involving attribution of some property to an

inalienable noun is to be accounted for by positing for universal

grammar, in the spirit of Bach (1965), a set of recurrent trans-

formations which each language uses somewhat in its own way. For
 - 120 -

sentences of the general structural type (182)

(182) PE v E D + N

where V is an adjective and N is a body-part noun, the options

are (a) - (d) below:

(a) PROMOTE D

(b) CHOOSE D AS SUBJECT

(c) COPY ADJECTIVE INTO BODY-PART NP

(d) INSERT HAVE INTO THE VACATED V

When (a) is not applied, the D becomes a genitive modifier of the

body;part N and the whole 0 becomes the subject. When (b) is not

applied, the 0 becomes the subject. When (c) is not applied': the

'short-circuited' sentences of Frei are the result. Rule (d) is

available only to those languages which have 'invented' have.

5.4 Further remarks on inalienable possession

If the feature of inalienable possession is to be treated as

a universal property of language, then either vocabulary items

which are translations of each other will be categorized alike

with respect to alienability, or the ways in which languages

separately classify the 'same' things may possibly reflect differ-

ences in the psychic make-up of the speakers of different Lang=

uages. Many scholars have seen in the data on inalienabilia an

opportunity for the science of language to shed light on primitive

mentality and on the possible range of man's concept of 'self'.

Since the differences appear more and more to be differences on

the level of surface strncturb, it may be advisable to wait some

 - 121-

118
time before reaching auy conclusions on these matters.

For representative statements on the sociological relevance

of the study of inalienable possession, see Levy-Bruhl (1916)
103; Bally (1926) 68 et passim; Frei (1939) 192; and van
Ginneken (1939) 90. For a catalogue of noun classifications
based on grammatical differences associated with inalienable
possession, see Haiim Rosen, "Die Ausdrucksform fur 'veriusser-
lichen' and 'unverausserlichen' Besitz im Frilhgriechinchen,"
LINGUA, VIII (1959) 264-293; 268f.

Adnominal D will certainly be needed for more than body-

part nouns and names of relatives. Directional indicators like

right and left are probably nouns of this type too. The reason

that these words appear typically without any personal reference

in English and maly other languages is that they frequently refer

to position or direction with respect to the speaker or addressee

of the utterance, and there are simply many situations in which

an adnominal D does not need to be expressed if it identifies

speaker or hearer.

There are, too, many relational nouns which do not have a

specifically personal reference. We might wish to say that

certain 'locational' nouns take an adnominal L. These nouns

sometimes name parts of the associated objects, as in (183),

and they sometimes identify a location or direction stated with

reference to the associated object but not considered as a part

of it, as seen in (184). 'Nouns' of the second type appear

superficially as prepositions in English.

(183) corner of the table, edge of the cliff, top

of the box

(184) behind the house, ahead of the car, next to

the tower
 - 122-

6. Problems and suggestions

There is a considerable residue of unsolved problems in the

grammatical description of language phenomena, and it is disap-

pointing though not surprising to realize how many of them

remain unsolved under the formulation of grammar l have been

suggesting. Those which come most quickly to mind are coordinate

conjunction, nominal predicates, and 'cognate objects'.

6.1 Coordinate conjunction

There may be a relationship between the ways in which

languages deal with 'comitative' constructions and the phenomenon

of coordinate conjunction of NPs. Put in case terms, there may

might
be a relationship between conjunction of NPs and what one

wish to refer to as a Comitative case. Jespersen noticed the

parallels between with (a preposition which has a Comitative

function) and the conjunctor and, as in such pairs of sentences

as (185) and (186).

(185) he and his wife are coming

(186) he is coming with his wifell9

119 Jespersen (1924) 90

Japanese has separate devices for indicating sentence conjunction

used for NP-conjuncttion
and NP-conjunction, and the posiposition
is identical with the Comitative postposition. In a conjunction

of NPs, all but the last have the postposition to. The last one

has the postposition appropriate for the case role of the whole

NP. Compare (187) and (188).

(187) Tanaka-san to Hashimoto-san ga kimashita

"Mr. Tanaka and Mr. Hashimoto came"

(188) Hashimoto-san ga Tanaka-san to hanashimashita

"Mr. Hashimoto spoke with Mr. Tanaka"

Redden points out that in Walapai a sentence has only one noun

in the 'nominative' case. Noun conjunctions are effected by

having the 'ablative' suffix--the suffix with Comitative function- -

on all but one of the nouns in a conjunction. Thus, in (189),

/4/ is nominative, /-m/ ablative.

(189) /hatAitAaa halm/

"the dog and the boy" (lit. "the dog with the

boy")

It may be that rule (190) is needed as an expansion rule for

NP.

(190) NP NP + C

Using X as a cover term for the various case categories, (190)

will produce such structures as (191).

(191) X

K NP

NP
 - 124 -

The case category C has a very special status, since the

selectional constraints on nouns under C are those of the

superordinate C. What is needed, in other words, is a rule

which imposes on any N under C the same redundant features

that are associated with the dominating non-C case.

A subjoined C under some circumstances must remain in the

large NP. In languages which lack a generalized conjunctor, the

case marker is simply that appropriate to C (the postposition

to in Japanese, the suffix -m in Walapai); in languages which

have a generalized conjunctor, it replaces the case marker, in

the way that and replaces with under certain conditions.

The structure underlying (185) and (186), then, might be

something like (192); we ignore the source of his.

(192)

V
[Pro];

NP

N K NP

d N
I

pres come by he with the wife

If the C remains inside the NP, the entire A becomes the subject,

yielding sentence (185); if the C is promoted, however, as in

 - 125 -

the structure shown in (193), it is left behind when the A

becomes the subject, resulting in sentence (186).

(193) S

M
V------- A
[N] -. /'''*%. /*"4""

h
K NP K NP

N
1 1

Ares come with his wife by he

It is quite unlikely that the numerous problems associated

with NP conjunction can be appreciably simplified through this

approach, but that there is some connection between conjunction

and comitative uses of NPs cannot be doubted. Lakoff and Peters

have recently presented very persuasive arguments that the

'direction' of the relationship is the opposite of eat I have

suggested; that, in other words, Comitative phrases are derived

120
from NP conjunction rather than the other way around.

120 George Lakoff and Stanley Peters, "Phrasal Conjunction and

Symmetric Predicates," HARVARD COMPUTATION LABORATORY REPORT
NO. NSF #17 (1966) VI-1 - VI-49.

6.2 Nominal Predicates

Nothing that has been said so far suggests a way of providing

for sentences of the wwwmff+re

N be N type. It is clear that they represent
 - 126 -

involving any of the case

a distinct sentence type from those

relations discussed above, though there may be more than one

case relationship involved in these sentences. (The terms

Essive and Translative come to mind.)

There are some nouns that appear in predicate position but

which are restricted in their occurrence elsewhere. It might

be possible to treat these nouns as, on one level, Vs which are

restricted to the form A]. Examples are words like idiot,

bastard and fool. The environment contains A because the subject

is always animate and because the constructions exhibit selec-

Vs having
tional and transformational properties associated with

As in their environment. Notice items (194) and (195).

(194) don't be a fool

(195) he's being a bastard again

This interpretation appears to account for the fact that we have

sentences like (196), but not--with idiot used in this same

'evaluative' sense--(197).

(196) John is an idiot

(197) an idiot hit the first homerun

Further evidence that the word is properly treated as a V is

found in the fact that these nouns may accept types of modifi-

cation usually associated with adjectives, as in (198).

(198) John is quite an idiot

The serious problems are (a) with the use of words like

idiot, fool, etc., in other contexts, as in (199), and (b) with

 -127-

the use of non-evaluative Ns in predicate sentence, as in (200).

(199) that rat swiped my lunch

(200) that boy is my nephew

A new case category or two could be invented for the occasion,

of course, but such matters as the number agreement requirement

between subject and predicate NPs remain as serious as they ever

were. Perhaps some solution is forthcoming along the lines of

Prof. Bach's proposals elsewhere in this volume.

6.3 Cognate objects

A difficulty of another sort is presented by the so-called

'cognate- object' constructions. These are constructions in

which, at the very least, there is a high selectivity between

a specific V and an 'object' 14 and in which the V+N combination

in one language might well be matched by a V alone in another

language.

Slightly modifying a recent analysis by Sandra Babcock121,

121 Sandra S. Babcock, "Syntactic dissimilation" (1966) unpublished

paper; compare too the interpretation in terms of 'quasi- trans-
formations.' found in Zellig Harris, "Co-occurrence and transfor-
mation in linguistic structure," LANGUAGE, XXXIII (1957) 283-
240, Sec. 30.

I would propose that there are contexts in which the case category

F (Factitive) may be left lexically empty, and that certain words

classified as Vs may be inserted specifically into frames con-

taining dummy Fs. These words may have associated with them
 - 128 -

special N representatives (like, e.g., bath) and special pro-Vs

(like, e.g., take). The rules that apply to dummy-F sentences

are the following:

(a) COPY THE N-REPRESENTATIVE OF THE V UNDER THE F

(b) REPLACE THE V BY THE DESIGNATED PRO-V

The rules may have separate conditions of optionality for

different Vs. The cognate-object V dream may appear as a V in

its own right, or it may appear in dummy-F sentences. As a

cognate-object verb, it has dream as its N-representative, have

as its pro-V; it is further specified as selecting either the

preposition about or of for the 0 constituent and as not requir-

ing rule (b).

When the N-representative associated with dream is copied

into the F constituent, the result is sentence (201); when the

associated verb have replaces the V, the result is (202).

(201) John dreamed a dream about Mary

(202) John had a dream about Mary

With these devices, we may in fact consider extending the

interpretation of cognate-object constructions in the following

way. Some words may be treated as cognate-object Vs even though

4.

the rule for replacing the pro-V is obligatory. The V nightmare,

e.g., might have nightmare listed as its N-representative and

have as its pro-V. Thus, on applying rule (a), structure (203)

becomes the intermediate structure (204); on applying rule (b),

(204) is converted to (205). Analogous uses of this device

 -129-

could possibly account for the connection between suggest,

and make a suggestion, shove someone and give someone a shove,

and so on, but many serious problems remain. In particular it

is not obvious how sentences like (206) and (207) can be dealt

with in accordance with these proposals.

(203) P

V F
1

nightmare 0

(204)

nightmare nightmare

(205)

have nightmare

(206) she made several ridiculous suggestions

(207) I had a terrible nightmare last night

6.4 Other problems

There are many issues for which I cannot even pretend to

see solutions. The apparent connection between surface cases

and 'partitive' functions; the restriction of 'definiteness' in

some languages to NPs in particular surface case relations

 -130-

(typically, the 'direct object'); the extreme variety of surface

realizations of the same meaning (from the same deep structure?)

that Jespersen illustrates in connection with what he calls

122
'rank shifting' , to name just a few.

122
Jespersen (1924) 91

The difficulties mentioned so far are empirical in nature,

but many formal problems exist as well. One of these is whether

the permitted arrays of cases under P need to be generated via

phrase-structure rules, since one of the most important functions

of the PS-rules has been that of defining grammatical relations- -

that is, that of defining phenomena which are here partly treated

categorially rather than configurationally. Related to this

problem is the apparent dependency relations that exist amolE

cases. It appears, for example, that the occurrence of B (Bene-

factive) phrases in a sentence has more to do with whether the

sentence contains an A than with independent specific properties

of Vs. One is almost willing to allow these facts to be

expressed by a generative process which chooses a verb, then

chooses the cases required by that verb, then chooses the other

cases compatible with the cases originally chosen. The issue is

not whether the permitted sequences can or cannot be generated

by PS-rules--there is no doubt that they can--but whether the

kinds of co-occurrence or dependency relationships that seem to

obtain might not be more efficiently stated in some other way.

 - 131-

(Modifications of transformational grammar of the type

introduced in Chomsky's ASPECTS (see fn. 1) made it no longer

necessary to use PS-rules for subclassification of lexical

categories or for the choice of lexical items. If the pro-

vision of syntactic relations of certain kinds must also be

handled by some device other than PS-rules, there is a chance

that rules of this type may be abandoned altogether.)

Whether the cases should be represented as categories

dominating NPs or in some other way is an issue which seems to

me to be fairly wide open. One advantage of the categorial

treatment is that NPs made subject and object may be said to have

lost their 'original' case relation to the sentence (by the rule

which 'erases' the case category whenever the case marker K has

been deleted, i.e., a 'node-razing' rule) with the result that

their form can only be determined by referring to their 'pure

relational' status. Thus it would appear that the surface dis-

tinction between labeled and configurationally defined relations

on NPs may correspond with the traditional distinction between

the 'concrete' and the !grammatical' cases. (How the genitive

figures in this distinction is not clear under either interpre-

tation.)

Several people have pointed out to me the apparent converti-

bility of case grammar underlying representations into objects

which resemble dependency diagrams and tagmemic formulas. If

the K elements are interpreted as constituents of NPG, then the

case categories unarily dominate NPs. This makes them equivalent

 -132 -

to labels on the branches that like P with the various NP

that are directly related to it. If the only function of the

P is to provide a constituent in terms of which the NPs can

be related to the V, one may just as well represent these

relationships more directly by replacing the node P by the V.

The result is no longer a constituent-structure diagram, since

lexical elements are inserted into 'dominating' nodes; but it

may turn out to be just as possible to represent the needed

constituent organization of sentences from a 'stemmatic' diagram

of the type used by Tesniere or Hays, as from a phrase-structure

tree liagram.

There is an easy conversion from case grammar underlying

representations to 'tagmemic' formulas, too, as long as the case

categories unarily dominate NPs. Or, for that matter, a case-

grammar diagram could simply be read off as a tagmemic formula,

as long as certain symbols were designated as function indica-

tors. One can as easily say "NP filling an A slot" as anything

else. The crucial difference between the modification of

transformational grammar that I have been suggesting and the

typical tagmemic study is in the insistence here on discovering

the 'deepest' level of the 'deep structure'.

 - 133 -

7. Closing words

One criticism of case grammar that has been brought to

my attention is that it is too strongly motivated by semantic

considerations. Many of the analyses have (hopefully) the

result that certain semantic distinctions and inter-language

commonalities are revealed in fairly direct ways in case

grammar deep structures, but, it has been argued, syntactic

analyses should be based on syntactic data alone, and one

language at a time.

The question arises whether there is a 'level' of syntac-

tic description that is discoverable one language at a time

based on purely syntactic criteria. If it is possible to

discover a semantically justified universal syntactic theory

along the lines I have been suggesting; if it is possible by

rules (beginning, perhaps, with those which assign sequential

order to the underlying order-free representations) to make

these 'semantic deep structures' into the surface forme of

sentences; then it is likely that the syntactic deep structure

of the type that has been made familiar from the work of

Chomsky and his students, is going to go the way of the

phoneme. It is an artificial intermediate level between the

empirically discoverable 'semantic deep structure' and the

observationally accessible surface structure, a level the

properties of which have more to do with the methodological

 - 134-

commitments of grammarians than with the nature of human

123
languages.

1231 wrongly
ong y expected these remarks to have a certain shock
value at the symposium. There I was attacked for constructing
deep structure representations that were too hampered by
considerations of surface syntactic facts to get at the under-
lying semantic realities. This is a problem which, as they
say, requires further study.