Variation

The Handbook of Second Language Acquisition EDITED BY Catherine J. Doughty and Michael H. Long THE HANDBOOK OF SECOND LANGUAGE ACQUISITION Blackwell Handbooks in Linguistics This outstanding multi-volume series covers all the major subdisciplines within linguistics today and, when complete, will offer a comprehensive survey of linguistics as a whole. Already published: The Handbook of Child Language Edited by Paul Fletcher and Brian MacWhinney The Handbook of Phonological Theory Edited by John A. Goldsmith The Handbook of Contemporary Semantic Theory Edited by Shalom Lappin The Handbook of Sociolinguistics Edited by Florian Coulmas The Handbook of Phonetic Sciences Edited by William J. Hardcastle and John Laver The Handbook of Morphology Edited by Andrew Spencer and Arnold Zwicky The Handbook of Japanese Linguistics Edited by Natsuko Tsujimura The Handbook of Linguistics Edited by Mark Aronoff and Janie Rees-Miller The Handbook of Contemporary Syntactic Theory Edited by Mark Baltin and Chris Collins The Handbook of Discourse Analysis Edited by Deborah Schiffrin, Deborah Tannen, and Heidi E. Hamilton The Handbook of Language Variation and Change Edited by J. K. Chambers, Peter Trudgill, and Natalie Schilling-Estes The Handbook of Historical Linguistics Edited by Brian D. Joseph and Richard D. Janda The Handbook of Language and Gender Edited by Janet Holmes and Miriam Meyerhoff The Handbook of Second Language Acquisition Edited by Catherine Doughty and Michael H. Long The Handbook of Second Language Acquisition EDITED BY Catherine J. Doughty and Michael H. Long © 2003 by Blackwell Publishing Ltd 350 Main Street, Malden, MA 02148-5018, USA 108 Cowley Road, Oxford OX4 1JF, UK 550 Swanston Street, Carlton South, Melbourne, Victoria 3053, Australia Kurfürstendamm 57, 10707 Berlin, Germany The right of Catherine J. Doughty and Michael H. Long to be identified as the Authors of the Editorial Material in this Work has been asserted in accordance with the UK Copyright, Designs, and Patents Act 1988. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs, and Patents Act 1988, without the prior permission of the publisher. First published 2003 by Blackwell Publishing Ltd Library of Congress Cataloging-in-Publication Data The handbook of second language acquisition / edited by Catherine J. Doughty and Michael H. Long. p. cm. – (Blackwell handbooks in linguistics ; 14) Includes bibliographical references and index. ISBN 0-631-21754-1 (hardcover : alk. paper) 1. Second language acquisition. I. Doughty, Catherine. II. Long, Michael H. III. Series. P118.2 .H363 2003 418–dc21 2002154756 A catalogue record for this title is available from the British Library. Set in 10/12pt Palatino by Graphicraft Limited, Hong Kong Printed and bound in the United Kingdom by TJ International, Padstow, Cornwall For further information on Blackwell Publishing, visit our website: http://www.blackwellpublishing.com Contents List of Contributors viii Acknowledgments x I Overview 1 1 The Scope of Inquiry and Goals of SLA Catherine J. Doughty and Michael H. Long II Capacity and Representation 2 On the Nature of Interlanguage Representation: Universal Grammar in the Second Language Lydia White 3 The Radical Middle: Nativism without Universal Grammar William O’Grady 4 Constructions, Chunking, and Connectionism: The Emergence of Second Language Structure Nick C. Ellis 5 Cognitive Processes in Second Language Learners and Bilinguals: The Development of Lexical and Conceptual Representations Judith F. Kroll and Gretchen Sunderman 6 Near-Nativeness Antonella Sorace 3 17 19 43 63 104 130 vi Contents III Environments for SLA 153 7 Language Socialization in SLA Karen Ann Watson-Gegeo and Sarah Nielsen 155 8 Social Context Jeff Siegel 178 9 Input and Interaction Susan M. Gass 224 10 Instructed SLA: Constraints, Compensation, and Enhancement Catherine J. Doughty 256 IV 311 Processes in SLA 11 Implicit and Explicit Learning Robert DeKeyser 313 12 Incidental and Intentional Learning Jan H. Hulstijn 349 13 Automaticity and Second Languages Norman Segalowitz 382 14 Variation Suzanne Romaine 409 15 Cross-Linguistic Influence Terence Odlin 436 16 Stabilization and Fossilization in Interlanguage Development Michael H. Long 487 V Biological and Psychological Constraints 537 17 Maturational Constraints in SLA Kenneth Hyltenstam and Niclas Abrahamsson 539 18 Individual Differences in Second Language Learning Zoltán Dörnyei and Peter Skehan 589 Contents vii 19 Attention and Memory during SLA Peter Robinson 631 20 Language Processing Capacity Manfred Pienemann 679 VI 715 Research Methods 21 Defining and Measuring SLA John Norris and Lourdes Ortega 717 22 Data Collection in SLA Research Craig Chaudron 762 VII The State of SLA 829 23 SLA Theory: Construction and Assessment Kevin R. Gregg 831 24 SLA and Cognitive Science Michael H. Long and Catherine J. Doughty 866 Index 871 Contributors Niclas Abrahamsson Stockholm University Craig Chaudron University of Hawai’i Robert M. DeKeyser University of Pittsburg Zoltán Dörnyei University of Nottingham Catherine J. Doughty University of Hawai’i Nick C. Ellis Bangor University of Wales Susan M. Gass Michigan State University Kevin Gregg Momoyama Gakuin/St Andrew’s University Jan H. Hulstijn University of Amsterdam Kenneth Hyltenstam Stockholm University Judith F. Kroll Pennsylvania State University Michael H. Long University of Hawai’i Contributors Sarah Nielsen Las Positas College John Norris Northern Arizona University Terence Odlin Ohio State University William O’Grady University of Hawai’i Lourdes Ortega Northern Arizona University Manfred Pienemann Paderborn University Peter Robinson Aoyama Gakuin University Suzanne Romaine Merton College, University of Oxford Norman Segalowitz Concordia University Jeff Siegel University of New England, Armadale, and University of Hawai’i Peter Skehan King’s College, London Antonella Sorace University of Edinburgh Gretchen Sunderman University of Illinois at Urbana-Champaign Karen Ann Watson-Gegeo University of California, Davis Lydia White McGill University ix Acknowledgments The editors gratefully acknowledge the following, who provided valuable reviews of one or more of the chapters: Alan Beretta, Craig Chaudron, Richard Cameron, Robert DeKeyser, Susan Gass, Kevin Gregg, Jan Hulstijn, Georgette Ioup, Peter Robinson, Dick Schmidt, Bonnie Schwartz, Larry Selinker, Mary Tiles, Michael Ullman, Jessica Williams, Lydia White, Kate Wolfe-Quintero, and several individuals who prefer to remain anonymous. The support and efficiency of Steve Smith, Sarah Coleman, and Fiona Sewell at Blackwell Publishing were greatly appreciated. The Scope of Inquiry and Goals of SLA 1 I Overview The Scope of Inquiry and Goals of SLA 3 1 The Scope of Inquiry and Goals of SLA CATHERINE J. DOUGHTY AND MICHAEL H. LONG 1 The Scope of Inquiry The scope of second language acquisition (SLA) is broad. It encompasses basic and applied work on the acquisition and loss of second (third, etc.) languages and dialects by children and adults, learning naturalistically and/or with the aid of formal instruction, as individuals or in groups, in foreign, second language, and lingua franca settings (see, e.g., R. Ellis, 1994; Gass and Selinker, 2001; Gregg, 1994; Jordens and Lalleman, 1988; W. Klein, 1986; Larsen-Freeman, 1991; Larsen-Freeman and Long, 1991; Ritchie and Bhatia, 1996; Towell and Hawkins, 1994). Research methods employed run the gamut from naturalistic observation in field settings, through descriptive and quasi-experimental studies of language learning in classrooms or via distance education, to experimental laboratory work and computer simulations. Researchers enter SLA with graduate training in a variety of fields, including linguistics, applied linguistics, psychology, communication, foreign language education, educational psychology, and anthropology, as well as, increasingly, in SLA per se, and bring with them a wide range of theoretical and methodological allegiances. The 1980s and 1990s witnessed a steady increase in sophistication in the choice of data-collection procedures and analyses employed, some of them original to SLA researchers (see, e.g., Birdsong, 1989; Chaudron, this volume; Doughty and Long, 2000; Faerch and Kasper, 1987; Sorace, 1996; Tarone, Gass, and Cohen, 1994), and also in the ways SLA is measured (Bachman and Cohen, 1998; Norris and Ortega, this volume). However, longitudinal studies of children (e.g., Huebner, 1983a, 1983b; F. Klein, 1981; Sato, 1990; Watson-Gegeo, 1992) and adults (e.g., Iwashita, 2001; Liceras, Maxwell, Laguardia, Fernandez, Fernandez, and Diaz, 1997; Schmidt, 1983) are distressingly rare; the vast majority of SLA studies are cross-sectional, with serious resulting limitations on the conclusions that can be drawn on some important issues. Theory proliferation remains a weakness, too, but the experience of 4 Catherine J. Doughty and Michael H. Long more mature disciplines in overcoming this and related teething problems is gradually being brought to bear (see, e.g., Beretta, 1991; Beretta and Crookes, 1993; Crookes, 1992; Gregg, 1993, 1996, 2000, this volume; Gregg, Long, Jordan, and Beretta, 1997; Jordan, 2002; Long, 1990a, 1993, forthcoming a).1 As reflected in the contributions to this volume (see also Robinson, 2001), much current SLA research and theorizing shares a strongly cognitive orientation, while varying from nativist, both special (linguistic) and general, to various kinds of functional, emergentist, and connectionist positions. The focus is firmly on identifying the nature and sources of the underlying L2 knowledge system, and on explaining developmental success and failure. Performance data are inevitably the researchers’ mainstay, but understanding underlying competence, not the external verbal behavior that depends on that competence, is the ultimate goal. Researchers recognize that SLA takes place in a social context, of course, and accept that it can be influenced by that context, both micro and macro. However, they also recognize that language learning, like any other learning, is ultimately a matter of change in an individual’s internal mental state. As such, research on SLA is increasingly viewed as a branch of cognitive science. 2 The Goals: Why Study SLA? Second language acquisition – naturalistic, instructed, or both – has long been a common activity for a majority of the human species and is becoming ever more vital as second languages themselves increase in importance. In many parts of the world, monolingualism, not bilingualism or multilingualism, is the marked case. The 300–400 million people whose native language is English, for example, are greatly outnumbered by the 1–2 billion people for whom it is an official second language. Countless children grow up in societies where they are exposed to one language in the home, sometimes two, another when they travel to a nearby town to attend primary or secondary school, and a third or fourth if they move to a larger city or another province for tertiary education or for work. Where literacy training or even education altogether is simply unavailable in a group’s native language, or where there are just too many languages to make it economically viable to offer either in all of them, as is the case in Papua New Guinea and elsewhere in the Pacific (Siegel, 1996, 1997, 1999, this volume), some federal and state governments and departments of education mandate use of a regional lingua franca or of an official national language as the medium of instruction. Such situations are sometimes recognized in state constitutions, and occasionally even in an official federal language policy, as in Australia (Lo Bianco, 1987); all mean that SLA is required of students, and often of their teachers, as well. Elsewhere, a local variety of a language may be actively suppressed or stigmatized, sometimes even by people who speak it natively themselves, resulting The Scope of Inquiry and Goals of SLA 5 in a need for widespread second dialect acquisition (SDA) for educational, employment, and other purposes. Examples include Hawai’i Creole English (Reynolds, 1999; Sato, 1985, 1989; Wong, 1999), Aboriginal English in Australia (Eades, 1992; Haig, 2001; Malcolm, 1994), and African-American Vernacular English in the USA (Long, 1999; Morgan, 1999; Rickford, 2000). In such cases, a supposedly “standard” variety may be prescribed in educational settings, despite the difficulty of defining a spoken standard objectively, and despite the notorious track record of attempts to legislate language change. The prescribed varieties are second languages or dialects for the students, and as in part of the Solomon Islands (Watson-Gegeo, 1992; Watson-Gegeo and Nielsen, this volume), once again, sometimes for their teachers, too, with a predictably negative effect on educational achievement. In a more positive development, while language death throughout the world continues at an alarming pace, increasing numbers of children in some countries attend various kinds of additive bilingual, additive bidialectal, or immersion programs designed to promote first language maintenance, SLA, or cultural revitalization (see, e.g., Fishman, 2001; Huebner and Davis, 1999; Philipson, 2000; Sato, 1989; Warner, 2001). SLA and SDA are not just common experiences for the world’s children, of course. More and more adults are becoming second language or second dialect learners voluntarily for the purposes of international travel, higher education, and marriage. For increasing numbers of others, the experience is thrust upon them. Involuntary SLA may take the fairly harmless form of satisfying a school or university foreign language requirement, but regrettably often it has more sinister causes. Each year, tens of millions of people are obliged to learn a second language or another variety of their own language because they are members of an oppressed ethnolinguistic minority, because forced to migrate across linguistic borders in a desperate search for work, or worse, due to war, drought, famine, religious persecution, or ethnic cleansing. Whatever they are seeking or fleeing, almost all refugees and migrants need to reach at least a basic threshold proficiency level in a second language simply to survive in their new environment. Most require far more than that, however, if they wish to succeed in their new environment or to become members of the new culture. States and citizens, scholars and laypersons alike recognize that learning a society’s language is a key part of both acculturation and socialization. Finally, less visibly, economic globalization and progressively more insidious cultural homogenization affect most people, knowingly or not, and each is transmitted through national languages within countries and through just a few languages, especially English at present, at the international level. Any experience that touches so many people is worthy of serious study, especially when success or failure can so fundamentally affect life chances. However, the obvious social importance of second language acquisition (SLA) is by no means the only reason for researchers’ interest, and for many, not the primary reason or not a reason at all. As a widespread, highly complex, uniquely human, cognitive process, language learning of all kinds merits careful study for what it can reveal about the nature of the human mind and intelligence. Thus, a 6 Catherine J. Doughty and Michael H. Long good deal of what might be termed “basic research” goes on in SLA without regard for its potential applications or social utility. In linguistics and psychology, for example, data on SLA are potentially useful for testing theories as different from one another as grammatical nativism (see, e.g., Eubank, 1991; Gregg, 1989; Liceras, 1986; Pankhurst, Sharwood-Smith, and Van Buren, 1988; Schwartz, 1992; White, 1989; and chapters by Gregg, Sorace, and White, this volume), general nativism (see, e.g., Eckman, 1996a; O’Grady, 2001a, 2001b, this volume; Wolfe-Quintero, 1996), various types of functionalism (see, e.g., Andersen, 1984; Eckman, 1996b; Mitchell and Miles, 1998, pp. 100–20; Rutherford, 1984; Sato, 1988, 1990; Tomlin, 1990), and emergentism and connectionism (see, e.g., Ellis, this volume; Gasser, 1990; MacWhinney, 2001). Research on basic processes in SLA draws upon and contributes to work on such core topics in cognitive psychology and linguistics as implicit and explicit learning (e.g., DeKeyser, this volume; N. Ellis, 1993, 1994; Robinson, 1997), incidental and intentional learning (e.g., Hulstijn, 2001, this volume; Robinson, 1996), automaticity (e.g., DeKeyser, 2001; Segalowitz, this volume), attention and memory (e.g., N. Ellis, 2001; Robinson, this volume; Schmidt, 1995; Tomlin and Villa, 1994), individual differences (e.g., Segalowitz, 1997; Dörnyei and Skehan, this volume), variation (e.g., Bayley and Preston, 1996; R. Ellis, 1999; Johnston, 1999; Preston, 1989, 1996; Romaine, this volume; Tarone, 1988; Williams, 1988; Young, 1990; Zobl, 1984), language processing (e.g., Clahsen, 1987; Doughty, this volume; Harrington, 2001; Pienemann, 1998, this volume), and the linguistic environment for language learning (e.g., Doughty, 2000; Gass, this volume; Hatch, 1978; Long, 1996; Pica, 1992), as well as at least two putative psychological processes claimed to distinguish first from second language acquisition, that is, cross-linguistic influence (see, e.g., Andersen, 1983a; Gass, 1996; Gass and Selinker, 1983; Jordens, 1994; Kasper, 1992; Kellerman, 1984; Kellerman and Sharwood-Smith, 1986; Odlin, 1989, this volume; Ringbom, 1987; Selinker, 1969) and fossilization (see, e.g., Kellerman, 1989; Long, this volume; Selinker, 1972; Selinker and Lakshmanan, 1992). SLA data are also potentially useful for explicating relationships between language and thought; for example, through exploring claims concerning semantic and cultural universals (see, e.g., Dietrich, Klein, and Noyau, 1995), or relationships between language development and cognitive development (Curtiss, 1982) – confounded in children, but not in SLA by adults. There is also a rich tradition of comparisons among SLA, pidginization, and creolization (see, e.g., Adamson, 1988; Andersen, 1983b; Andersen and Shirai, 1996; Bickerton, 1984; Meisel, 1983; Schumann, 1978; Valdman and Phillips, 1975). In neuroscience, SLA data can help show where and how the brain stores and retrieves linguistic knowledge (see, e.g., Green, 2002; Obler and Hannigan, 1996; Ullman, 2002); which areas are implicated in acquisition (see, e.g., Schumann, 1998); how the brain adapts to additional burdens, such as bilingualism (see, e.g., Albert and Obler, 1978; Jacobs, 1988; Kroll, Michael, and Sankaranarayanan, 1998; Kroll and Sunderman, this volume), or trauma resulting in bilingual or multilingual aphasia (see, e.g., Galloway, 1981; Paradis, The Scope of Inquiry and Goals of SLA 7 1990); and whether the brain is progressively more limited in handling any of those tasks. In what has become one of the most active areas of work in recent years, SLA researchers seek to determine whether observed differences in the success of children and adults with second languages is because the brain is subject to maturational constraints in the form of sensitive periods for language learning (see, e.g., Birdsong, 1999; Bongaerts, Mennen, and van der Slik, 2000; DeKeyser, 2000; Flege, Yeni-Komshian, and Liu, 1999; Hyltenstam and Abrahamsson, this volume; Ioup, Boustagui, El Tigi, and Moselle, 1994; Long, 1990b, forthcoming b; Schachter, 1996). Basic research sometimes yields unexpected practical applications, and that may turn out to be true of basic SLA research, too. Much work in SLA, however, has clear applications or potential applications from the start. The most obvious of these is second (including foreign) language teaching (see, e.g., Doughty, 1991, this volume; Doughty and Williams, 1998; N. Ellis and Laporte, 1997; R. Ellis, 1989; de Graaff, 1997; Lightbown and Spada, 1999; Long, 1988; Norris and Ortega, 2000; Pica, 1983; Pienemann, 1989; Sharwood-Smith, 1993), since SLA researchers study the process language teaching is designed to facilitate.2 For bilingual, immersion, and second dialect education, second language literacy programs, and whole educational systems delivered through the medium of a second language, SLA research findings offer guidance on numerous issues. Examples include the optimal timing of L1 maintenance and L2 development programs, the linguistic modification of teaching materials, the role of implicit and explicit negative feedback on language error, and language and content achievement testing. SLA research findings are also potentially very relevant for populations with special language-learning needs. These include certain abnormal populations, such as Alzheimer’s patients (see, e.g., Hyltenstam and Stroud, 1993) and Down syndrome children, where research questions concerning socalled (first) “language intervention” programs are often quite similar to those of interest for (second) “language teaching” (see, e.g., Mahoney, 1975; Rosenberg, 1982). Other examples are groups, such as immigrant children, for whom it is crucial that educators not confuse second language problems with learning disabilities (see, e.g., Cummins, 1984); bilinguals undergoing primary language loss (Seliger, 1996; Seliger and Vago, 1991; Weltens, De Bot, and van Els, 1986); and deaf and hearing individuals learning a sign language, such as American Sign Language (ASL), as a first or second language, respectively (see, e.g., Berent, 1996; Mayberry, 1993; Strong, 1988). In all these cases, as Bley-Vroman (1990) pointed out, researchers are interested in explaining not only how success is achieved, but why – in stark contrast with almost uniformly successful child first language acquisition – at least partial failure is so common in SLA. 8 Catherine J. Doughty and Michael H. Long NOTES 1 A seminar on theory change in SLA, with readings from the history, philosophy, and sociology of science and the sociology of knowledge, is now regularly offered as an elective for M.A. and Ph.D. students in the University of Hawai’i’s Department of Second Language Studies. The importance of such a “big picture” methodology course in basic training for SLA researchers – arguably at least as great as that of the potentially endless series of “grassroots” courses in quantitative and qualitative research methods and statistics that are now routine – will likely become more widely recognized over time. 2 The utility of some work in SLA for this purpose does not mean that SLA is the only important source of information, and certainly not that a theory of SLA should be passed off as a theory of language teaching. Nor, conversely, does it mean, as has occasionally been suggested, that SLA theories should be evaluated by their relevance to the classroom. REFERENCES Adamson, H. D. 1988: Variation Theory and Second Language Acquisition. Washington, DC: Georgetown University Press. Albert, M. L. and Obler, L. 1978: The Bilingual Brain: Neuropsychological and Neurolinguistic Aspects of Bilingualism. San Diego: Academic Press. Andersen, R. W. 1983a: Transfer to somewhere. In S. M. Gass and L. Selinker (eds), Language Transfer in Language Learning. Rowley, MA: Newbury House, 177–201. Andersen, R. W. 1983b: Pidginization and Creolization as Language Acquisition. Rowley, MA: Newbury House. Andersen, R. W. 1984: The one to one principle of interlanguage construction. Language Learning, 34 (4), 77–95. Andersen, R. W. and Shirai, Y. 1996: The primacy of aspect in first and second language acquisition: the pidgin– creole connection. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 527–70. Bachman, L. and Cohen, A. D. 1998: Interfaces between Second Language Acquisition and Language Testing Research. Cambridge: Cambridge University Press. Bayley, R. and Preston, D. R. (eds) 1996: Second Language Acquisition and Linguistic Variation. Philadelphia: John Benjamins. Berent, G. P. 1996: The acquisition of English syntax by deaf learners. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 469–506. Beretta, A. 1991: Theory construction in SLA. Complementarity and opposition. Studies in Second Language Acquisition, 13 (4), 493–512. Beretta, A. and Crookes, G. 1993: Cognitive and social determinants in the context of discovery in SLA. Applied Linguistics, 14 (3), 250–75. Bickerton, D. 1984: The language bioprogram hypothesis and second language acquisition. In W. E. Rutherford (ed.), Language The Scope of Inquiry and Goals of SLA 9 Universals and Second Language Acquisition. Amsterdam and Philadelphia: John Benjamins, 141–61. Birdsong, D. 1989: Metalinguistic Performance and Interlinguistic Competence. Berlin and New York: Springer Verlag. Birdsong, D. (ed.) 1999: Second Language Acquisition and the Critical Period Hypothesis. Mahwah, NJ: Lawrence Erlbaum Associates. Bley-Vroman, R. 1990: The logical problem of foreign language learning. Linguistic Analysis, 20 (1–2), 3– 49. Bongaerts, T., Mennen, S., and van der Slik, F. 2000: Authenticity of pronunciation in naturalistic second language acquisition. The case of very advanced late learners of Dutch as a second language. Studia Linguistica, 54, 298–308. Clahsen, H. 1987: Connecting theories of language processing and (second) language acquisition. In C. Pfaff (ed.), First and Second Language Acquisition Processes. Cambridge, MA: Newbury House, 103–16. Crookes, G. 1992: Theory format and SLA theory. Studies in Second Language Acquisition, 14 (4), 425–49. Cummins, J. 1984: Bilingualism and Special Education: Issues on Assessment and Pedagogy. Clevedon: Multilingual Matters. Curtiss, S. 1982: Developmental dissociation of language and cognition. In L. K. Obler and L. Menn (eds), Exceptional Language and Linguistics. New York: Academic Press, 285–312. DeKeyser, R. 2000: The robustness of critical period effects in second language acquisition. Studies in Second Language Acquisition, 22 (4), 493–533. DeKeyser, R. 2001: Automaticity and automatization. In P. Robinson (ed.), Cognition and Second Language Instruction. Cambridge: Cambridge University Press, 125–51. Dietrich, R., Klein, W., and Noyau, C. 1995: The Acquisition of Temporality in a Second Language. Amsterdam and Philadelphia: John Benjamins. Doughty, C. J. 1991: Second language instruction does make a difference: evidence from an empirical study of SL relativization. Studies in Second Language Acquisition, 13 (4), 431–69. Doughty, C. J. 2000: Negotiating the L2 linguistic environment. University of Hawai’i Working Papers in ESL, 18 (2), 47–83. Doughty, C. J. and Long, M. H. 2000: Eliciting second language speech data. In L. Menn and N. Bernstein Ratner (eds), Methods for Studying Language Production. Mahwah, NJ: Lawrence Erlbaum Associates, 149–77. Doughty, C. J. and Williams, J. 1998: Focus on Form in Classroom Second Language Acquisition. Cambridge: Cambridge University Press. Eades, D. 1992: Aboriginal English and the Law: Communicating with Aboriginal English-Speaking Clients: A Handbook for Legal Practitioners. Brisbane: Queensland Law Society. Eckman, F. R. 1996a: On evaluating arguments for special nativism in second language acquisition theory. Second Language Research, 12 (4), 335–73. Eckman, F. R. 1996b: A functionaltypological approach to second language acquisition theory. In W. C. Ritchie and T. K. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 195–211. Ellis, N. 1993: Rules and instances in foreign language learning: interactions of explicit and implicit knowledge. European Journal of Cognitive Psychology, 5, 289–318. Ellis, N. 1994: Implicit and Explicit Learning of Languages. New York: Academic Press. 10 Catherine J. Doughty and Michael H. Long Ellis, N. 2001: Memory for language. In P. Robinson (ed.), Cognition and Second Language Instruction. Cambridge: Cambridge University Press, 33–68. Ellis, N. and Laporte, N. 1997: Contexts of acquisition: effects of formal instruction and naturalistic exposure on second language acquisition. In A. M. de Groot and J. F. Kroll (eds), Tutorials in Bilingualism: Psycholinguistic Perspectives. Mahwah, NJ: Lawrence Erlbaum Associates, 53–83. Ellis, R. 1989: Are classroom and naturalistic acquisition the same? A study of classroom acquisition of German word order rules. Studies in Second Language Acquisition, 11 (3), 305–28. Ellis, R. 1994: The Study of Second Language Acquisition. Oxford: Oxford University Press. Ellis, R. 1999: Item versus system learning: explaining free variation. Applied Linguistics, 20 (4), 460 –80. Eubank, L. 1991: Introduction: Universal Grammar in the second language. In L. Eubank (ed.), Point Counterpoint: Universal Grammar in the Second Language. Amsterdam and Philadelphia: John Benjamins, 1– 48. Faerch, C. and Kasper, G. (ed.) 1987: Introspection in Second Language Research. Clevedon: Multilingual Matters. Fishman, J. A. 2001: Can Threatened Languages be Saved? Clevedon: Multilingual Matters. Flege, J. E., Yeni-Komshian, G. H., and Liu, S. 1999: Age constraints on second-language acquisition. Journal of Memory and Language, 41, 78–104. Galloway, L. M. 1981: The convolutions of second language: a theoretical article with a critical review and some new hypotheses towards a neuropsychological model of bilingualism and second language performance. Language Learning, 31 (2), 439–64. Gass, S. M. 1996: Second language acquisition and linguistic theory: the role of language transfer. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 317– 45. Gass, S. M. and Selinker, L. (eds) 1983: Language Transfer in Language Learning. Rowley, MA: Newbury House. Gass, S. M. and Selinker, L. 2001: Second Language Acquisition: An Introductory Course. Second edition. Mahwah, NJ: Lawrence Erlbaum Associates. Gasser, M. 1990: Connectionism and universals of second language acquisition. Studies in Second Language Acquisition, 12 (2), 179–99. Graaff, R. de 1997: The eXperanto experiment: effects of explicit instruction on second language acquisition. Studies in Second Language Acquisition, 19 (2), 249–76. Green, D. W. (ed.) 2002: The cognitive neuroscience of bilingualism. Bilingualism: Language and Cognition, 4 (2), 101–201. Gregg, K. R. 1989: Second language acquisition theory: the case for a generative perspective. In S. M. Gass and J. Schachter (eds), Linguistic Perspectives on Second Language Acquisition. Cambridge: Cambridge University Press, 15–40. Gregg, K. R. 1993: Taking explanation seriously; or, Let a couple of flowers bloom. Applied Linguistics, 14 (3), 276–94. Gregg, K. R. 1994: Second language acquisition: history and theory. Encyclopedia of Language and Linguistics. Second edition. Oxford: Pergamon, 3720–6. Gregg, K. R. 1996: The logical and developmental problems of second language acquisition. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of The Scope of Inquiry and Goals of SLA 11 Second Language Acquisition. San Diego: Academic Press, 49–81. Gregg, K. R. 2000: A theory for every occasion: postmodernism and SLA. Second Language Research, 16 (4), 343–59. Gregg, K. R., Long, M. H., Jordan, G., and Beretta, A. 1997: Rationality and its discontents in SLA. Applied Linguistics, 17 (1), 63–83. Haig, Y. 2001: Teacher perceptions of student speech. Ph.D. dissertation. Edith Cowan University. Harrington, M. 2001: Sentence processing. In P. Robinson (ed.), Cognition and Second Language Instruction. Cambridge: Cambridge University Press, 91–124. Hatch, E. M. 1978: Discourse analysis and second language acquisition. In E. M. Hatch (ed.), Second Language Acquisition: A Book of Readings. Rowley, MA: Newbury House, 401–35. Huebner, T. 1983a: A Longitudinal Analysis of the Acquisition of English. Ann Arbor, MI: Karoma. Huebner, T. 1983b: Linguistic systems and linguistic change in an interlanguage. Studies in Second Language Acquisition, 6 (1), 33–53. Huebner, T. and Davis, K. A. 1999: Sociopolitical Perspectives on Language Policy and Planning in the USA. Amsterdam and Philadelphia: John Benjamins. Hulstijn, J. H. 2001: Intentional and incidental second language learning: a reappraisal of elaboration, rehearsal and automaticity. In P. Robinson (ed.), Cognition and Second Language Instruction. Cambridge: Cambridge University Press, 258–86. Hyltenstam, K. and Stroud, C. 1993: Second language regression in Alzheimer’s dementia. In K. Hyltenstam and A. Viberg (eds), Progression and Regression in Language: Sociocultural, Neuropsychological and Linguistic Perspectives. Cambridge: Cambridge University Press, 222– 42. Ioup, G., Boustagui, E., El Tigi, M., and Moselle, M. 1994: Reexamining the critical period hypothesis: a case study of successful adult SLA in a naturalistic environment. Studies in Second Language Acquisition, 16 (1), 73–98. Iwashita, N. 2001: The role of task-based conversation in the acquisition of Japanese grammar and vocabulary. Ph.D. thesis. University of Melbourne, Department of Linguistics and Applied Linguistics. Jacobs, B. 1988: Neurobiological differentiation in primary and secondary language acquisition. Studies in Second Language Acquisition, 10 (3), 303–37. Johnston, M. 1999: System and variation in interlanguage development. Unpublished Ph.D. dissertation. Canberra: Australian National University. Jordan, G. 2002: Theory construction in SLA. Ph.D. dissertation. London University, Institute of Education. Jordens, P. 1994: The cognitive function of case marking in German as a native and a foreign language. In S. M. Gass and L. Selinker (eds), Language Transfer in Language Learning. Second edition. Amsterdam and Philadelphia: John Benjamins, 138–75. Jordens, P. and Lalleman, J. (eds) 1988: Language Development. Dordrecht: Foris. Kasper, G. 1992: Pragmatic transfer. Second Language Research, 8 (3), 203–31. Kellerman, E. 1984: The empirical evidence for the influence of the L1 in interlanguage. In A. Davies, C. Criper, and A. Howatt (eds), Interlanguage. Edinburgh: Edinburgh University Press, 98–122. Kellerman, E. 1989: The imperfect conditional. In K. Hyltenstam and L. K. Obler (eds), Bilingualism Across 12 Catherine J. Doughty and Michael H. Long the Lifespan: Aspects of Acquisition, Maturity, and Loss. Cambridge: Cambridge University Press, 87–115. Kellerman, E. and Sharwood-Smith, M. (eds) 1986: Cross-Linguistic Influence in Second Language Acquisition. New York: Pergamon. Klein, F. 1981: The acquisition of English in Hawai’i by Korean adolescent immigrants: a longitudinal study of verbal auxiliary agreement. Ph.D. dissertation. University of Hawai’i, Department of Linguistics. Klein, W. 1986: Second Language Acquisition. Cambridge: Cambridge University Press. Kroll, J. F., Michael, E., and Sankaranarayanan, A. 1998: A model of bilingual representation and its implications for second language acquisition. In A. F. Healy and L. E. Bourne, Jr (eds), Foreign Language Learning: Psycholinguistic Studies on Training and Retention. Mahwah, NJ: Lawrence Erlbaum Associates, 365–95. Larsen-Freeman, D. 1991: Second language acquisition research: staking out the territory. TESOL Quarterly, 25 (2), 315–50. Larsen-Freeman, D. and Long, M. H. 1991: An Introduction to Second Language Acquisition Research. London: Longman. Liceras, J. 1986: Linguistic Theory and Second Language Acquisition. Tubingen: Gubter Narr. Liceras, J. M., Maxwell, D., Laguardia, B., Fernández, Z., Fernández, R., and Diaz, L. 1997: A longitudinal study of Spanish non-native grammars: beyond parameters. In A. T. Pérez-Leroux and W. Glass (eds), Contemporary Perspectives on the Acquisition of Spanish. Vol. 1: Developing Grammars. Somerville, MA: Cascadilla Press, 99–132. Lightbown, P. M. and Spada, N. 1999: How Languages are Learned. Revised edition. Oxford: Oxford University Press. Lo Bianco, J. 1987: National Policy on Languages. Canberra: Australian Government Publishing Service. Long, M. H. 1988: Instructed interlanguage development. In L. M. Beebe (ed.), Issues in Second Language Acquisition: Multiple Perspectives. Cambridge, MA: Newbury House, 115– 41. Long, M. H. 1990a: The least a second language acquisition theory needs to explain. TESOL Quarterly, 24 (4), 649–66. Long, M. H. 1990b: Maturational constraints on language development. Studies in Second Language Acquisition, 12 (3), 251–85. Long, M. H. 1993: Assessment strategies for second language acquisition theories. Applied Linguistics, 14 (3), 225– 49. Long, M. H. 1996: The role of the linguistic environment in second language acquisition. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 413 –68. Long, M. H. 1998: SLA: breaking the siege. University of Hawai’i Working Papers in ESL, 17 (1), 79–129. Also to appear in M. H. Long, Problems in SLA. Mahwah, NJ: Lawrence Erlbaum Associates. Long, M. H. 1999: Ebonics, language and power. In F. L. Pincus and H. J. Ehrlich (eds), Race and Ethnic Conflict: Contending Views on Prejudice, Discrimination, and Ethnoviolence. Second edition. Westview/ HarperCollins, 331– 45. Long, M. H. forthcoming a: Theory change in SLA. In M. H. Long, Problems in SLA. Mahwah, NJ: Lawrence Erlbaum Associates. Long, M. H. forthcoming b: Age differences and the sensitive periods controversy in SLA. In M. H. Long, Problems in SLA. Mahwah, NJ: Lawrence Erlbaum Associates. The Scope of Inquiry and Goals of SLA 13 Long, M. H. and Robinson, P. 1998: Focus on form: theory, research and practice. In C. J. Doughty and J. Williams (eds), Focus on Form in Classroom Second Language Acquisition. Cambridge: Cambridge University Press, 15–41. MacWhinney, B. 2001: The Competition Model: the input, the context and the brain. In P. Robinson (ed.), Cognition and Second Language Instruction. Cambridge: Cambridge University Press, 69–90. Mahoney, G. 1975: Ethnological approach to delayed language acquisition. American Journal of Mental Deficiency, 80, 139– 48. Malcolm, I. 1994: Aboriginal English inside and outside the classroom. Australian Review of Applied Linguistics, 17 (1), 147–80. Mayberry, R. 1993: First-language acquisition after childhood differs from second-language acquisition: the case of American Sign Language. Journal of Speech and Hearing Research, 36, 1258–70. Meisel, J. M. 1983: Strategies of second language acquisition: more than one kind of simplification. In R. W. Andersen (ed.), Pidginization and Creolization as Second Language Acquisition. Rowley, MA: Newbury House, 120–57. Mitchell, R. and Miles, F. 1998: Functional/pragmatic perspectives on second language learning. In R. Mitchell and F. Miles (eds), Second Language Learning Theories. London: Arnold, 100–20. Morgan, M. 1999: US language planning and policies for social dialect speakers. In T. Huebner and K. A. Davies (eds), Sociopolitical Perspectives on Language Policy and Planning in the USA. Amsterdam and Philadelphia: John Benjamins, 173–91. Norris, J. and Ortega, L. 2000: Effectiveness of instruction: a research synthesis and quantitative metaanalysis. Language Learning, 50 (3), 417–528. Obler, L. and Hannigan, S. 1996: Neurolinguistics of second language acquisition and use. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 509–23. Odlin, T. 1989: Language Transfer. Cambridge: Cambridge University Press. O’Grady, W. 2001a: Language acquisition and language loss. Ms. University of Hawai’i, Department of Linguistics. O’Grady, W. 2001b: An emergentist approach to syntax. Ms. University of Hawai’i, Department of Linguistics. Pankhurst, J., Sharwood-Smith, M., and Van Buren, P. 1988: Learnability and Second Languages: A Book of Readings. Dordrecht: Foris. Paradis, M. 1990: Bilingual and polyglot aphasia. In F. Boller and J. Grafman (eds), Handbook of Neuropsychology. Vol. 2. New York: Elsevier, 117– 40. Philipson, R. (ed.) 2000: Rights to Language: Equity, Power, and Education. Mahwah, NJ: Lawrence Erlbaum Associates. Pica, T. 1983: Adult acquisition of English as a second language under different conditions of exposure. Language Learning, 33 (4), 465–97. Pica, T. 1992: The textual outcomes of native speaker/non-native speaker negotiation: what do they reveal about second language learning? In C. Kramsch and S. McConnell-Ginet (eds), Text and Context: CrossDisciplinary Perspectives on Language Study. Lexington, MA: D. C. Heath, 198–237. Pienemann, M. 1989: Is language teachable? Applied Linguistics, 10 (1), 52–79. Pienemann, M. 1998: Language Processing and Second Language Development: 14 Catherine J. Doughty and Michael H. Long Processability Theory. Amsterdam and Philadelphia: John Benjamins. Preston, D. R. 1989: Sociolinguistics and Second Language Acquisition. Oxford: Blackwell. Preston, D. R. 1996: Variationist linguistics and second language acquisition. In W. C. Ritchie and T. K. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 229–65. Reynolds, S. B. 1999: Mutual intelligibility? Comprehension problems between American Standard English and Hawai’i Creole English in Hawai’i’s public schools. In J. R. Rickford and S. Romaine (eds), Creole Genesis, Attitudes and Discourse. Amsterdam and Philadelphia: John Benjamins, 303–19. Rickford, J. R. 2000: African American Vernacular English: Features, Evolution, Educational Implications. Oxford: Blackwell. Ringbom, H. 1987: The Role of the First Language in Foreign Language Learning. Clevedon: Multilingual Matters. Ritchie, W. R. and Bhatia, T. J. (eds) 1996: Handbook of Second Language Acquisition. San Diego: Academic Press. Robinson, P. 1996: Learning simple and complex second language rules under implicit, incidental, rule-search and instructed conditions. Studies in Second Language Acquisition, 18 (1), 27– 67. Robinson, P. 1997: Individual differences and the fundamental similarity of implicit and explicit adult second language learning. Language Learning, 47 (1), 45–99. Robinson, P. (ed.) 2001: Cognition and Second Language Instruction. Cambridge: Cambridge University Press. Rosenberg, S. 1982: The language of the mentally retarded: development processes and intervention. In S. Rosenberg (ed.), Handbook of Applied Psycholinguistics: Major Thrusts of Research and Theory. Hillsdale, NJ: Lawrence Erlbaum Associates, 329–92. Rutherford, W. E. (ed.) 1984: Language Universals and Second Language Acquisition. Amsterdam and Philadelphia: John Benjamins. Sato, C. J. 1985: Linguistic inequality in Hawai’i: the post-creole dilemma. In N. Wolfson and J. Manes (eds), Language of Inequality. Berlin: Mouton, 255–72. Sato, C. J. 1988: Origins of complex syntax in interlanguage development. Studies in Second Language Acquisition, 10 (3), 371–95. Sato, C. J. 1989: A non-standard approach to Standard English. TESOL Quarterly, 23 (2), 259–82. Sato, C. J. 1990: The Syntax of Conversation in Interlanguage Development. Tubingen: Gunter Narr. Schachter, J. 1996: Maturation and the issue of UG in L2 acquisition. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 159–93. Schmidt, R. W. 1983: Interaction, acculturation and the acquisition of communicative competence. In N. Wolfson and E. Judd (eds), Sociolinguistics and Second Language Acquisition. Rowley, MA: Newbury House, 137–74. Schmidt, R. W. (ed.) 1995: Attention and Awareness in Foreign Language Learning. Honolulu: University of Hawai’i Press. Schumann, J. H. 1978: The Pidginization Process: A Model for Second Language Acquisition. Rowley, MA: Newbury House. Schumann, J. H. 1998: The neurobiology of affect in language. Language Learning, 48: Supplement 1. Schwartz, B. D. 1992: Testing between UG-based and problem-solving The Scope of Inquiry and Goals of SLA 15 models of L2A: developmental sequence data. Language Acquisition, 2 (1), 1–19. Segalowitz, N. 1997: Individual differences in second language acquisition. In A. M. de Groot and J. F. Kroll (eds), Tutorials in Bilingualism: Psycholinguistic Perspectives. Mahwah, NJ: Lawrence Erlbaum Associates, 85–112. Seliger, H. W. 1996: Primary language attrition in the context of bilingualism. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 605–26. Seliger, H. W. and Vago, R. M. (eds) 1991: First Language Attrition. Cambridge: Cambridge University Press. Selinker, L. 1969: Language transfer. General Linguistics, 9, 67–92. Selinker, L. 1972: Interlanguage. International Review of Applied Linguistics, 10 (3), 209–31. Selinker, L. and Lakshmanan, U. 1992: Language transfer and fossilization: the multiple effects principle. In S. M. Gass and L. Selinker (eds), Language Transfer in Language Learning. Amsterdam and Philadelphia: John Benjamins, 97–116. Sharwood-Smith, M. 1993: Input enhancement in instructed SLA: theoretical bases. Studies in Second Language Acquisition, 15 (2), 165–79. Siegel, J. 1996: Vernacular Education in the South Pacific. International Development Issues No. 45. Canberra: Australian Agency for International Development. Siegel, J. 1997: Using a pidgin language in formal education: help or hindrance? Applied Linguistics, 18, 86–100. Siegel, J. 1999: Creole and minority dialects in education: an overview. Journal of Multilingual and Multicultural Development, 20, 508–31. Sorace, A. 1996: The use of acceptability judgments in second language acquisition research. In W. R. Ritchie and T. J. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 375– 409. Strong, M. (ed.) 1988: Language Learning and Deafness. Cambridge: Cambridge University Press. Tarone, E. E. 1988: Variation and Second Language Acquisition. London: Edward Arnold. Tarone, E. E., Gass, S. M., and Cohen, A. D. 1994: Research Methodology in Second-Language Acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates. Tomlin, R. S. 1990: Functionalism in second language acquisition. Studies in Second Language Acquisition, 12 (2), 155–77. Tomlin, R. and Villa, V. 1994: Attention in cognitive science and second language acquisition. Studies in Second Language Acquisition, 16 (2), 183–203. Towell, R. and Hawkins, R. 1994: Approaches to Second Language Acquisition. Clevedon: Multilingual Matters. Ullman, M. T. 2002: The neural basis of lexicon and grammar in first and second language: the declarative/ procedural model. Bilingualism: Language and Cognition, 4 (2), 105–22. Valdman, A. and Phillips, J. 1975: Pidginization, creolization and the elaboration of learner systems. Studies in Second Language Acquisition, 1 (1), 21– 40. Warner, N. 2001: Küi ka mäna‘ai: children acquire traits of those who raise them. Plenary: Pacific Second Language Research Forum. University of Hawai’i, October 6. Watson-Gegeo, K. A. 1992: Thick explanation in the ethnographic study of child socialization: a longitudinal study of the problem of schooling for 16 Catherine J. Doughty and Michael H. Long Kwara’ae (Solomon Islands) children. In W. A. Corsaro and P. J. Miller (eds), Interpretative Approaches to Children’s Socialization. San Francisco: Jossey-Bass, 51– 66. Weltens, B., De Bot, K., and van Els, T. (eds) 1986: Language Attrition in Progress. Dordrecht: Foris. White, L. 1989: Universal Grammar and Second Language Acquisition. Amsterdam and Philadelphia: John Benjamins. Williams, J. 1988: Zero anaphora in second language acquisition: a comparison among three varieties of English. Studies in Second Language Acquisition, 10 (3), 339–70. Wolfe-Quintero, K. 1996: Nativism does not equal Universal Grammar. Second Language Research, 12 (4), 335–73. Wong, L. 1999: Language varieties and language policy: the appreciation of Pidgin. In T. Huebner and K. A. Davies (eds), Sociopolitical Perspectives on Language Policy and Planning in the USA. Amsterdam and Philadelphia: John Benjamins, 205–22. Young, R. 1990: Variation in Interlanguage Morphology. Amsterdam and Philadelphia: John Benjamins. Zobl, H. 1984: The Wave Model of linguistic change and the naturalness of interlanguage. Studies in Second Language Acquisition, 6 (2), 160–85. On the Nature of Interlanguage Representation 17 II Capacity and Representation 2 On the Nature of Interlanguage Representation: Universal Grammar in the Second Language LYDIA WHITE 1 Introduction In the late 1960s and early 1970s, several researchers pointed out that the language of second language (L2) learners is systematic and that learner errors are not random mistakes but evidence of rule-governed behavior (Adjémian, 1976; Corder, 1967; Nemser, 1971; Selinker, 1972). From this developed the conception of “interlanguage,” the proposal that L2 learners have internalized a mental grammar, a natural language system that can be described in terms of linguistic rules and principles. The current generative linguistic focus on interlanguage representation can be seen as a direct descendent of the original interlanguage hypothesis. Explicit claims are made about the nature of interlanguage competence, the issues being the extent to which interlanguage grammars are like other grammars, as well as the role of Universal Grammar (UG). The question of whether UG mediates L2 acquisition, and to what extent, has been much debated since the early 1980s. This question stems from a particular perspective on linguistic universals and from particular assumptions about the nature of linguistic competence. In the generative tradition, it is assumed that grammars are mental representations, and that universal principles constrain these representations. Linguistic universals are as they are because of properties of the human mind, and grammars (hence, languages) are as they are because of these universal principles. The first decade of research on the role of UG in L2 acquisition concentrated on so-called “access,” exploring whether UG remains available in non-primary acquisition. The issue of UG access relates to fundamental questions such as: what are natural language grammars like? What is the nature of linguistic competence? How is it acquired? UG is proposed as a partial answer, at least in the case of the first language (L1) grammar, the assumption being that 20 Lydia White language acquisition is impossible in the absence of specific innate linguistic principles which place constraints on grammars, restricting the “hypothesis space,” or, in other words, severely limiting the range of possibilities that the language acquirer has to entertain. In L2 acquisition research, then, the issue is whether interlanguage representations are also constrained by UG. 2 UG and the Logical Problem of Language Acquisition UG is proposed as part of an innate biologically endowed language faculty (e.g., Chomsky, 1965, 1981; Pinker, 1994). It places limitations on grammars, constraining their form (the inventory of possible grammatical categories in the broadest sense, i.e., syntactic, semantic, phonological), as well as how they operate (the computational system, principles that the grammar is subject to). UG includes invariant principles, as well as parameters which allow for variation. While theories like Government-Binding (GB) (Chomsky, 1981), Minimalism (Chomsky, 1995), or Optimality Theory (Archangeli and Langendoen, 1997) differ as to how universal principles and parameters are formalized, within these approaches there is a consensus that certain properties of language are too abstract, subtle, and complex to be acquired in the absence of innate and specifically linguistic constraints on grammars. UG is postulated as an explanation of how it is that learners come to know properties of grammar that go far beyond the input, how they know that certain things are not possible, why grammars are of one sort rather than another. The claim is that such properties do not have to be learned. Proposals for an innate UG are motivated by the observation that, at least in the case of L1 acquisition, there is a mismatch between the primary linguistic data (PLD), namely the utterances a child is exposed to, and the abstract, subtle, and complex knowledge that the child acquires. In other words, the input (the PLD) underdetermines the output (the grammar). This is known as the problem of the poverty of the stimulus or the logical problem of language acquisition. As an example of a proposed principle of UG which accounts for knowledge too subtle to be learned solely from input, we will consider the Overt Pronoun Constraint (OPC) (Montalbetti, 1983), a constraint which has recently received attention in L2 acquisition research. The OPC states that in null argument languages (languages allowing both null and overt pronouns), an overt pronoun cannot receive a bound variable interpretation, that is, it cannot have a quantified expression (such as everyone, someone, no one) or a wh-phrase (who, which) as its antecedent.1 This constraint holds true of null argument languages in general, including languages unrelated to each other, such as Spanish and Japanese. Consider the sentences in (1) from English, a language requiring overt subjects. In particular, we are concerned with the coreference possibilities (indicated by subscripts) between the pronominal subject of the lower clause and its potential antecedent in the main clause: On the Nature of Interlanguage Representation 21 (1) a. Everyonei thought [hei would win] b. Whoi thought [hei would win]? c. Johni thought [hei would be late] In (1a), the pronoun he can be bound to the quantifier everyone. On this interpretation, every person in the room thinks himself or herself a likely winner: he, then, does not refer to a particular individual. This is known as a bound variable interpretation. Similarly, in (1b) the pronoun can be bound to the wh-phrase who without referring to a particular individual. In (1c), on the other hand, the pronoun refers to a particular person in the main clause, namely John. (In addition, in all three cases, disjoint reference is possible, with the pronoun in the lower clause referring to some other person in the discourse – this interpretation is not of concern here.) In null argument languages, the situation regarding quantified antecedents is somewhat different. On the one hand, an embedded null subject can take either a quantified or a referential antecedent (or it can be disjoint in reference from other NPs in the sentence), just like overt pronouns in English. This is illustrated in (2) for Japanese:2 (2) a. Darei ga [∅i kuruma o katta to] itta no? Who NOM car ACC bought that said Q Whoi said that (hei) bought a car? de itiban da to] itte-iru b. Tanaka-sani wa [∅i kaisya Tanaka-Mr TOP company in best is that saying-is Mr Tanakai is saying that (hei) is the best in the company On the other hand, overt pronouns are more restricted than either null pronouns in null argument languages or overt pronouns in languages requiring overt arguments. In particular, an overt pronoun may not have a quantified antecedent, as in (3a), whereas it can have a sentence-internal referential antecedent, as in (3b): [karei ga kuruma o katta to] itta no? (3) a. *Darei ga Who NOM he NOM car ACC bought that said Q Whoi said that hei bought a car? kaisya de itiban da to] itte-iru b. Tanaka-sani wa [karei ga Tanaka-Mr TOP he NOM company in best is that saying-is Mr Tanakai is saying that hei is the best in the company The differences between null argument languages like Japanese and languages that do not permit null arguments like English are summarized in table 2.1. At issue, then, is how the L1 acquirer of a language like Japanese discovers the restriction on overt pronouns with respect to quantified antecedents. This case constitutes a clear poverty-of-the-stimulus situation. The phenomenon in question is very subtle. In many cases, overt and null pronouns will appear in the same syntactic contexts (although sometimes under different pragmatic 22 Lydia White Table 2.1 Antecedents for pronouns in null and overt argument languages Overt argument languages Null argument languages Referential antecedents Quantified antecedents Null subjects Overt subjects Overt subjects Yes Yes Yes No Yes Yes and discourse conditions), so it is unlikely that the absence of overt pronouns with quantified antecedents would be detected. It is also highly unlikely that L1 acquirers produce utterances incorrectly using overt pronouns with quantified antecedents and are then provided with negative evidence on this point. How, then, could an L1 acquirer of a language like Japanese discover this property? The argument is that the knowledge is built in, in the form of a principle of UG, the OPC; it does not have to be learned at all. 3 UG and the Logical Problem of L2 Acquisition Assuming a logical problem of L1 acquisition, hence motivating UG, people have asked whether the same holds true of L2; that is, whether there is a mismatch between the input that L2 learners are exposed to and the unconscious knowledge that they attain (Bley-Vroman, 1990; Schwartz and Sprouse, 2000; White, 1985). In the case of L2 acquisition, it is important to distinguish between (i) the logical problem and (ii) UG availability. The first issue is whether L2 learners attain unconscious knowledge (a mental representation) that goes beyond the L2 input. (There would be no logical problem at all, if L2 learners turned out not to achieve knowledge that goes beyond the input.) The second issue is whether such knowledge (if found) is achieved by means of UG. These are not in fact the same question, although they are often collapsed, since the way to determine whether UG principles and parameters constrain interlanguage representations is similar to the way to assess whether there is a logical problem of L2 acquisition. However, it is conceivable that there is a logical problem of L2 acquisition, with L2 learners achieving far more than could have come from the input alone, and that their achievement is to be explained by postulating a reliance on the L1 grammar rather than a still-functioning UG (Bley-Vroman, 1990; Schachter, 1988). The strongest case for the operation of UG in L2 acquisition, then, is if learners demonstrate knowledge of subtle and abstract properties which could On the Nature of Interlanguage Representation 23 not have been learned from L2 input alone or from input plus general learning principles (not specifically linguistic) or on the basis of explicit instruction or from the L1 grammar. In such cases, not only is there a logical problem of L2 acquisition but also UG remains the only way to account for the knowledge in question. To demonstrate an L2 logical problem, hence the likelihood of involvement of UG, researchers have sought out genuine L2 poverty of the stimulus cases, in which both of the following hold (White, 1989b, 1990): i ii The phenomenon in question is underdetermined by the L2 input. That is, it must not be something that could have been acquired by simple observation of the L2 input, as an effect of input frequency, or on the basis of instruction, analogical reasoning, etc. The phenomenon in question works differently in the L1 and the L2. If L2 learners show evidence of subtle and abstract knowledge, we want to exclude the possibility that such knowledge is obtained solely via the L1 grammar. However, the requirement that L1 and L2 differ in the relevant respects becomes harder and harder to achieve, in that many properties of UG will of necessity manifest themselves in the L1 in some form (Dekydtspotter, Sprouse, and Anderson, 1998; Hale, 1996). Nevertheless, if the L1 and L2 differ in terms of surface properties, then transfer can be ruled out, at least at this level, as an explanation of successful acquisition. In the first decade of work on SLA from a UG perspective (starting in the early 1980s), research focused mainly on whether or not UG is available to L2 learners, and in what form. The UG question seemed relatively straightforward (and relatively global): is UG available (or accessible) to L2 learners? The assumption was that if you can show that a particular UG principle operates/does not operate then this generalizes to other principles, hence to UG availability/nonavailability in general. Researchers looked for evidence that L2 learners could (or could not) apply principles of UG, and set or reset parameters, as well as investigating the extent to which the L1 was involved, in the form of L1 parameter settings in interlanguage grammars. Hypotheses varied as to whether learners had no access, partial (indirect) access, or full (direct) access to UG, and there were differing views on the role of the L1 grammar. But although the issues were phrased in terms of access to UG, the question was then, and remains, whether interlanguage representations show evidence of being constrained by principles of UG; that is, whether interlanguage grammars are restricted in the same way as the grammars of native speakers are restricted. As a recent example of research which takes into account the logical problem of L2 acquisition and looks for evidence as to whether a principle of UG constrains the interlanguage representation, consider Kanno’s (1997) investigation of the operation of the OPC in the grammars of L2 learners of Japanese (see box 2.1). Using a coreference judgment task, Kanno shows that L2 learners demonstrate subtle knowledge of the restriction on overt pronouns, correctly 24 Lydia White Box 2.1 The Overt Pronoun Constraint (OPC) (Kanno, 1997) Research question: Do adult L2 learners observe principles of UG which are not operative in their L1? In particular, do English-speaking learners of Japanese observe the OPC? Overt Pronoun Constraint (OPC) (Montalbetti, 1983): In null argument languages, an overt pronoun cannot receive a bound variable interpretation. L2 logical problem: i There appears to be nothing in the L2 input to signal the difference between overt and null pronominals with respect to quantified antecedents. It is unlikely that the absence of overt pronouns with quantified antecedents would be detected. This issue is not explicitly taught and not discussed in L2 textbooks. ii Knowledge of the restriction on overt pronouns in Japanese is not available from the L1 English. In English, overt pronouns can receive a bound variable interpretation, contrary to Japanese. Methodology: Subjects: 28 intermediate-level English-speaking adult learners of Japanese. Control group of 20 adult native speakers of Japanese. Task: Coreference judgment task, involving 20 biclausal sentences (4 sentence types, 5 tokens of each). Each sentence had a pronoun subject (overt or null) in the lower clause, and a potential antecedent (quantified or referential) in the main clause. Participants had to indicate whether the subject of the embedded clause could refer to the same person as the subject of the main clause or whether it referred to someone else. Results: Native speakers and L2 learners differentiated in their treatment of overt pronouns depending on the type of antecedent involved (quantified or referential), as well as differentiating between overt and null pronominals in these contexts (see table 2.2), supporting the claim that the OPC is being observed. Native speakers overwhelmingly rejected quantified antecedents for overt pronouns (2 percent), while accepting them in the case of null subjects (83 percent). They indicated that null subjects can always take a sentence-internal referential antecedent (100 percent), whereas for overt pronouns an internal referential antecedent was accepted at about 50 percent (both an internal and an external referent are possible). The L2 learners showed a remarkably similar pattern of results and their responses did not differ significantly from the controls. Conclusion: Adult L2 acquirers of Japanese observe the OPC, suggesting that interlanguage grammars are constrained by UG. Table 2.2 Acceptances of antecedents by subject type (percentages) Null subject kare (“he”) Native speakers (n = 20) L2 learners (n = 28) Quantified antecedent Referential antecedent Quantified antecedent Referential antecedent 83.0 2.0 100.0 47.0 78.5 13.0 81.5 42.0 On the Nature of Interlanguage Representation 25 disallowing quantified antecedents in cases like (3a). Kanno’s test sentences are carefully constructed to control for use of both types of pronoun (overt and null) in the context of both kinds of antecedent (referential and quantified). This allows her to eliminate the possibility that L2 learners simply prohibit overt pronouns from taking sentence-internal antecedents in general, as well as the possibility that they reject quantified antecedents altogether. In addition to considering group results, Kanno shows that subjects largely behave consistently with respect to the OPC when analyzed individually. Such individual analyses are crucial, since the hypothesis is that UG constrains the grammars of individuals, and group results may conceal individual variation. The knowledge demonstrated by these L2 learners of Japanese could not have come from the L1 English, where overt pronouns do take quantified antecedents; it is knowledge that is underdetermined by the L2 input, where null and overt pronouns allow similar antecedents in many cases. The distinction between permissible antecedents for overt and null pronouns is not taught in L2 Japanese textbooks or classes. It seems unlikely that there are relevant surface patterns in the L2 input that could be noticed by the learner, leading to this result. Nevertheless, L2 learners demonstrate knowledge of the restriction, suggesting that L2 representations must be constrained by UG. Similar results have been reported for L2 Spanish by Pérez-Leroux and Glass (1997); that is, adult English-speaking learners of Spanish also observe the OPC. 4 The Comparative Fallacy So far, we have considered the case of learners who acquire subtle knowledge of the constraint on antecedents for pronouns (the OPC). Here, then, properties of the L2 assumed to stem from UG are manifested in the interlanguage grammar. The interlanguage grammar and the L2 grammar converge in this respect, as suggested by Kanno’s results. But what if interlanguage representations fail to demonstrate certain L2 properties? What if the interlanguage and the L2 diverge? Does this necessarily imply lack of UG? This was, in fact, the interpretation taken (implicitly or explicitly) by a number of researchers in the 1980s. Some researchers were quite explicit in their assumption that one should compare L2 learners and native speakers with respect to UG properties, the native speaker of the L2 providing a reference point for assessing UG availability. If L2 learners rendered judgments (or otherwise behaved) like native speakers with respect to some principle or parameter of UG, then they were deemed to have access to UG; on the other hand, if they differed in their judgments from native speakers, then their grammars were assumed not to be constrained by UG. For example, in Schachter’s (1989, 1990) investigations of constraints on wh-movement, this was the underlying rationale for claiming the non-operation of UG. Schachter found that, compared to native speakers, L2 learners of English of certain L1 backgrounds were very inaccurate in their 26 Lydia White judgments on illicit wh-movement out of structures such as embedded questions and relative clauses; hence, Schachter argued, L2 learners do not have access to UG principles independently of the L1. The problem with this kind of approach to UG in L2 acquisition is that it presupposes that the interlanguage representation must converge on the grammar of native speakers of the L2, that the endstate grammar of a second language learner must be identical to that of a native speaker. But this is a misconception (Cook, 1997; Schwartz, 1993, 1998b; White, 1996). An interlanguage grammar which diverges from the L2 grammar can nevertheless fall within the bounds laid down by UG. If we are going to take the issue of representation seriously, we need to consider Bley-Vroman’s comparative fallacy. Bley-Vroman (1983) warned that “work on the linguistic description of learners’ languages can be seriously hindered or sidetracked by a concern with the target language” (p. 2) and argued that “the learner’s system is worthy of study in its own right, not just as a degenerate form of the target system” (p. 4). A number of researchers pointed out quite early on the need to consider interlanguage grammars in their own right with respect to principles and parameters of UG, arguing that one should not compare L2 learners to native speakers of the L2 but instead consider whether interlanguage grammars are natural language systems (e.g., duPlessis et al., 1987; Finer and Broselow, 1986; Liceras, 1983; Martohardjono and Gair, 1993; Schwartz and Sprouse, 1994; White, 1992b). These authors have shown that L2 learners may arrive at representations which indeed account for the L2 input, though not in the same way as the grammar of a native speaker. The issue, then, is whether the interlanguage representation is a possible grammar, not whether it is identical to the L2 grammar. For example, with respect to the violations of constraints on whmovement that Schachter (1989, 1990) reports, Martohardjono and Gair (1993), White (1992b), and, more recently, Hawkins and Chan (1997) argue that L2 learners have a different analysis for the phenomenon in question, whereby structures involving a fronted wh-phrase are derived without movement (based on properties of the L1 grammar), explaining the apparent lack of movement constraints. A related kind of misleading comparison involves the use of control groups in experimental tasks. There is often an (implicit) expectation that L2 speakers should not differ significantly from native speakers with respect to performance on sentences testing for UG properties. Suppose that on a grammaticality judgment task native speakers accept sentences violating some principle of UG at less than 5 percent and accept corresponding grammatical sentences at over 95 percent. In order to demonstrate “access” to this principle, it is not necessary for L2 speakers to perform at the same level. Rather, the issue is whether the interlanguage grammar shows evidence of certain distinctions: does learners’ performance on grammatical sentences differ significantly from their performance on ungrammatical sentences (cf. Grimshaw and Rosen, 1990, for related comments on L1 acquisition)? Do L2 learners distinguish between different kinds of ungrammatical sentences (see Martohardjono, 1993)? If certain On the Nature of Interlanguage Representation 27 sentence types are treated significantly differently from other sentence types, this suggests that the interlanguage grammar represents the relevant distinction (whatever it may be), even if the degree to which L2 learners observe it in performance differs from that of native speakers. To return to Kanno’s study on the OPC, the importance of her results lies not in the fact that the L2 learners did not differ significantly from the native speakers, but rather in the fact that the L2 learners showed a significant difference in their acceptances of quantified antecedents depending on pronoun type, suggesting that their grammars make the relevant distinction between licit and illicit antecedents. It is not the case, however, that one should never compare L2 speakers to native speakers of the L2 as far as properties of the grammar are concerned.3 There are legitimate reasons for asking whether the L2 learner has in fact acquired properties of the L2. After all, the learner is exposed to L2 input in some form, and the L2 is a natural language. What is problematic is when certain conclusions are drawn based on failure to perform exactly like native speakers. Failure to acquire L2 properties may nevertheless involve acquiring properties different from the L1, properties of other natural languages, properties that are underdetermined by the L2 input. Such failure does not necessarily entail lack of UG. 5 UG “Access” and Terminological Confusions Earlier approaches to UG in L2 acquisition revealed a somewhat ambivalent attitude to the L1. Perhaps because the strongest case for UG can be made if one can eliminate the L1 as a potential source of UG-like knowledge, some researchers felt that evidence of the influence of the L1 grammar on the interlanguage representation would somehow weaken the case for UG. Nowhere is this more evident than in the terminological confusions and disagreements that arose over terms like direct access to UG. Direct access for some researchers was taken to mean that L2 learners arrive at UG properties independently of their L1 (e.g., Cook, 1988). For others (e.g., Thomas, 1991b), it meant the instantiation of any legitimate parameter setting (L1, L2, Ln). Similar problems have arisen with the term full access, which at some point replaced direct access. Epstein, Flynn, and Martohardjono (1996) restrict the term full access to the position that UG operates independently of the L1 representation, whereas Schwartz and Sprouse (1996) do not so restrict it. Part of the problem is that terms like direct/full or indirect/partial access are too global. In addition, in some cases at least, an overly simplistic and misleading dichotomy between UG and the L1 is adopted. Since the L1 is a natural language, there is no a priori justification for assuming that a representation based on the L1 implies lack of UG constraints on the interlanguage grammar. What is required is a greater focus on the nature of the representations that L2 learners achieve. It may not always be appropriate to dwell explicitly on 28 Lydia White the UG access question. But by looking in detail at the nature of interlanguage representation, we in fact remain committed to this issue, since evidence of an interlanguage grammar that does not fall within the hypothesis space sanctioned by UG is evidence that UG does not fully constrain interlanguage grammars. 6 Interlanguage Representation: Convergence, Divergence, or Impairment In the 1990s, the UG debate shifted from a consideration of the broad access question to a detailed consideration of the nature of interlanguage representation. Specific grammatical properties have been investigated and claims have been made as to how they are represented. It is largely presupposed that the interlanguage grammar and the grammars of native speakers of the L2 will diverge in some respects, at least initially and possibly also finally (see Flynn, 1996, for a contrary view). Of interest, then, is the nature of that divergence: is it indicative of a representation that is nevertheless constrained by UG (cf. Sorace, 1993) or is it suggestive of some kind of impairment to the grammar, such that the interlanguage representation is in some sense defective? If interlanguage representations were to show properties not found elsewhere in natural languages, this would suggest that they are not UG-constrained, at least in some domains (see Thomas, 1991a, and Klein, 1995). The focus on representation manifests itself particularly clearly in proposals relating to the L2 initial state. Theories about the initial state are theories about the representation that L2 learners start out with, the representations that they initially use to make sense of the L2 input. 6.1 Example: strong features and verb movement Since proposals regarding initial and subsequent interlanguage grammars often dwell, in one way or another, on functional categories, we will consider an example here to illustrate the kinds of properties that researchers have investigated in recent years. Functional categories, such as inflection (I), complementizer (C), and determiner (D), have certain formal features associated with them (tense, agreement, case, number, person, gender, etc.). These features vary as to strength (strong vs. weak). Functional categories are seen as the locus of parametric variation (e.g., Borer, 1984; Chomsky, 1995), which can be found at the level of the categories themselves (not all categories are realized in all languages), at the level of formal features (the features of a particular functional category may vary from language to language), and at the level of feature strength (a particular feature can be strong in one language and weak in another). Here we will consider properties relating to functional projections above the verb phrase (VP). Finite verbs have features (tense, agreement) which have to be checked against corresponding features in I (Chomsky, 1995).4 If features in I are strong, the finite verb raises overtly to check its features, as in the French On the Nature of Interlanguage Representation 29 (4a). If features are weak, overt movement does not take place, as in the English (4b):5 CP (4) Spec C′ C IP Spec I′ I (a) Jean sorti NegP pas VP ti (b) John (does) not leave Feature strength results in a number of syntactic consequences related to word order. In languages such as French, where features in I are strong, there are alternations between the positions of finite and non-finite verbs, since non-finite verbs have no features to check, hence do not raise.6 Comparing French to a language with weak features, like English, there are word order differences between the two with respect to where the finite verb is found (Emonds, 1978; Pollock, 1989). The difference between finite and non-finite verbs in French is illustrated in (5); the differences between finite verbs in French and English are illustrated in (6) and (7). In these examples, we consider only the position of the verb with respect to negation and adverbs, but there is a variety of other verb placement facts which are subsumed under this analysis (see Pollock, 1989): (5) a. ne sortez pas (ne) leave-2PP not b. pas sortir not leave-INF ‘don’t go out’ (6) a. Marie n’aime pas Jean Mary likes not John b. Marie voit rarement Jean Mary sees rarely John (7) a. b. c. d. Mary does not like John *Mary likes not John Mary rarely sees John *Mary sees rarely John 30 Lydia White In French, finite lexical verbs appear to the left of the negative pas while nonfinite verbs appear to the right (compare (5a) and (5b) ). English and French contrast with respect to the position of the finite verb in relation to negation and adverbs (compare (6) and (7) ). In English, lexical verbs appear to the right of negation (7a) and adverbs (7c) and cannot precede them (7b, 7d), in contrast to French (6a, 6b). A range of word order differences between the two languages are thus accounted for by one parametric difference between them, namely the strength of features in I. In the next section, we will use the example of verb movement to illustrate some of the representational issues that are currently being pursued. It should be noted, however, that not all of the theories to be discussed in fact have made claims specifically about verb placement. 6.2 Initial state Proposals concerning the initial interlanguage representation can broadly be classified into two types: (i) the interlanguage representation conforms to properties of natural language (though not necessarily the L2); or (ii) the interlanguage representation differs from adult natural languages in fundamental respects (which, however, may not be permanent). Into the first category falls the Full Transfer/Full Access (FTFA) Hypothesis of Schwartz and Sprouse (1994, 1996). I will also consider Epstein et al.’s (1996) Full Access Hypothesis in this category. Although the Full Access Hypothesis is not, strictly speaking, a hypothesis about the initial state (Epstein et al., 1996, p. 750), it nevertheless has clear implications for the nature of the earliest grammar. The second category includes the Minimal Trees Hypothesis of Vainikka and YoungScholten (1994, 1996), as well as Eubank’s (1993/4, 1994) claim that initially features are neither strong nor weak but rather “inert” or “valueless.” Schwartz and Sprouse (1994, 1996) propose that the L1 grammar constitutes the interlanguage initial state. In other words, faced with L2 input that must be accounted for, learners adopt the representation that they already have. Schwartz and Sprouse (1994) originally presented this proposal in the context of an analysis of the acquisition of German word order by a native speaker of Turkish. Schwartz and Sprouse (1996) and Schwartz (1998a) extend the analysis to French-speaking learners of English, arguing, following White (1991a, 1991c, 1992a), that the initial interlanguage grammar includes strong features, because this is the case in the L1 French. In consequence, verbs are incorrectly placed with respect to adverbs, as White found. However, a potential problem for FTFA is that while White’s (1992a) subjects had considerable problems with adverb placement, producing and accepting forms like (7d), they did not have equivalent problems with negation, correctly recognizing the impossibility of (7b).7 According to FTFA, the interlanguage representation is necessarily different from the grammar of native speakers of the L2, at least initially; it is nevertheless On the Nature of Interlanguage Representation 31 UG constrained, exemplifying functional categories and features, as well as syntactic properties that derive from feature strength. The interlanguage representation may or may not converge on the L2 grammar in later stages of development. When the L1 representation is unable to accommodate the L2 input, the learner has recourse to options made available through UG. Once the L2 input reveals an analysis to be inappropriate, there is restructuring of the interlanguage representation. For example, in the case of verb raising, there are properties of the L2 input that could signal the need to change from strong to weak feature values: the presence of do-support in negatives (7a) shows that finite lexical verbs in English do not raise (Schwartz, 1987; White, 1992a). Thus, convergence might be expected in this case. In contrast to FTFA, Epstein et al. (1996, p. 751) and Flynn (1996) claim the L1 grammar is not implicated in the initial interlanguage representation. The implicit logic of their argumentation suggests that UG must be the initial state8 and that the early grammar in principle has available all functional categories, features, and feature values, from UG, so that an appropriate representation for the L2 can be constructed without recourse to categories or features from the L1. As far as representation of functional categories is concerned, there is no development on such an account: the L2 categories are in place from early on; because they are appropriate, there is no need for subsequent restructuring of the grammar. In terms of our example, this would mean that a French-speaking learner of English should assume weak features initially, hence would make no word order errors, contrary to fact, at least as far as adverb placement is concerned (White, 1991a, 1991c). Similarly, an English-speaking learner of French should assume strong features, hence exhibiting verb raising. Again, there is research that suggests that this is not inevitable. White (1989a, 1991b) reports that Englishspeaking children learning French fail to consistently accept verb raising in a variety of tasks. Hawkins, Towell, and Bazergui (1993) suggest that intermediate proficiency adult English-speaking learners of French fail to reset from the weak L1 feature strength to the strong value required by the L2. Although Schwartz and Sprouse (1996) and Epstein et al. (1996) differ radically in their claims about the involvement of the L1 grammar, they share the assumption that the interlanguage representation shows a full complement of functional categories, drawn either from the L1 or from UG. In other words, the interlanguage representation is a grammar sanctioned by UG, both in the initial state and subsequently. Other theories posit a greater degree of divergence between what is found in the interlanguage grammar and what is found in the grammars of adult native speakers. Vainikka and Young-Scholten (1994, 1996) propose the Minimal Trees Hypothesis, whereby the initial state lacks functional categories altogether, only lexical categories (N, V, P, etc.) being found. Lexical categories are assumed to be drawn from the L1 grammar, hence to exhibit the same properties as the L1 with respect to headedness, for example. Thus, this theory 32 Lydia White shares with FTFA the assumption that L1 properties are found in the initial representation. However, as far as functional categories are concerned, Vainikka and Young-Scholten (1994, 1996) assume no transfer at all. Vainikka and Young-Scholten’s (1994) proposals are based on an examination of spontaneous production data from adult learners of German whose L1s are Turkish and Korean. The evidence that they adduce is largely morphological: in early production data from adult learners of German, inflectional morphology is lacking. This leads them to conclude that the corresponding abstract categories are lacking in the interlanguage grammar. (See Sprouse, 1998, and Lardiere, 2000, for arguments against assuming such a close relationship between surface morphology and abstract syntactic categories.) In addition, Vainikka and YoungScholten (1994) claim that the early grammar lacks word orders that would be the result of movement of the finite verb to a functional projection. In terms of our example, the prediction of Minimal Trees is that French-speaking learners of English should not produce errors like (7d), since these are the result of verb movement from V to I (motivated by strong features) (Schwartz, 1998b; Schwartz and Sprouse, 1996). If the functional category I is altogether absent and there is only a VP projection, there is nowhere for the verb to move to. Hence, the only interlanguage word order should be the order that is in fact correct for English, namely (7c), contrary to fact. (See Vainikka and Young-Scholten, 1996, 1998, for discussion.) Further evidence against Minimal Trees is provided by Grondin and White (1996), who examine spontaneous production data from two English-speaking children learning French. Grondin and White show that there is both morphological and syntactic evidence in favor of an IP projection in early stages. For example, the children show an alternation in verb placement with respect to negation: finite verbs precede pas whereas non-finite verbs follow it, suggesting movement of the finite verb to I; this is inconsistent with Minimal Trees, which postulates no I in the early grammar. However, as Vainikka and Young-Scholten (1996) point out, these data may not be truly representative of the initial state, since the children had several months of exposure to the L2 prior to beginning to speak. In some sense, the Minimal Trees Hypothesis might be seen as implying a defective interlanguage grammar (Lardiere, 2000), since it postulates a period during which the representation lacks functional categories, which are otherwise presumed to be a necessary characteristic of natural language grammars. However, this impairment is assumed to be temporary, with functional categories developing gradually until, eventually, all functional categories appropriate for the L2 are acquired. Furthermore, Vainikka and Young-Scholten (1994, 1996) take the position that gradual emergence of functional categories is also characteristic of L1 acquisition (Clahsen, Eisenbeiss, and Vainikka, 1994); thus, for them, L2 acquisition in this domain is similar to L1. The final initial state proposal to be considered here also implies that interlanguage grammars are in some sense defective. Eubank (1993/4, 1994) shares with Schwartz and Sprouse (1994, 1996) the assumption that the L1 grammar On the Nature of Interlanguage Representation 33 constitutes a major part of the initial state: L1 lexical categories and functional categories are assumed to be present. However, Eubank maintains that the initial representation lacks fully specified feature values, at least some interlanguage features being unspecified or “inert.” In Eubank (1993/4) and subsequently (e.g., Eubank and Grace, 1998) the focus is specifically on feature strength: while features are strong or weak in natural language grammars, they are argued to be neither in the interlanguage, suggesting an impairment in this domain. According to Eubank, a consequence of inertness is that finite verbs will vary optionally between raised and unraised positions; this will be true regardless of what language is being acquired as the L2 and regardless of the situation in the L1. In the case of French-speaking learners of English, then, variable word orders are expected, that is, both (7c) and (7d). The same would be expected of English-speaking learners of French. In support, Eubank (1993/4) points to White’s (1991a, 1991c) results on the position of the verb with respect to the adverb, where there was some evidence of variability, with francophone subjects allowing word orders like not only (7d) but also (7c). However, Yuan (2000) shows that French-speaking and English-speaking learners of Chinese (a language with weak features, hence lacking verb movement) are very accurate in positioning verbs in Chinese, even at the beginner level, showing no evidence of optional verb placement. In fact, Eubank’s assumption that raising of finite verbs will be optional appears to be a stipulation which does not follow from any particular theory of feature strength: if features have no strength, there is nothing to motivate verb raising, since this requires a strong feature value (Robertson and Sorace, 1999; Schwartz, 1998b). Prévost and White (2000) provide evidence that finite verbs in adult L2 French and German fail to appear in non-finite positions (i.e., unraised); instead, they occur almost exclusively in positions appropriate for finite verbs, suggesting that inertness cannot be involved. In its early instantiation, Eubank’s proposal was not unlike (indeed, was modeled on) similar proposals that features in L1 acquisition are initially underspecified (e.g., Hyams, 1996; Wexler, 1994). Although a grammar with underspecified features is in some sense defective, underspecification in L1 is assumed to be a temporary property. Similarly, Eubank originally assumed inertness to be a passing phase in the interlanguage representation, with L2 feature strength ultimately attainable. 6.3 Beyond the initial state Initial state theories necessarily have implications for the nature of representation during the course of development, as well as for endstate representation (that is, the steady state interlanguage grammar). According to FTFA, while the L1 grammar forms the interlanguage initial state, restructuring takes place in response to L2 input; hence, convergence on the relevant L2 properties is possible, though not guaranteed, since in some cases the L1 grammar may appear to accommodate the L2 input adequately and thus change will not 34 Lydia White be triggered. Divergent outcomes, then, would not be surprising, but the interlanguage representation is nevertheless assumed to be UG-constrained. There are researchers who agree with Schwartz and Sprouse that the L1 grammar is the initial state but who maintain that at least some (and possibly all) L1 features and feature values remain in the interlanguage representation, L2 features or feature values not being acquirable (Hawkins, 1998; Hawkins and Chan, 1997; Liceras, Maxwell, Laguardia, Fernández, and Fernández, 1997; Smith and Tsimpli, 1995). This means that development in the form of restructuring toward a more appropriate functional structure for the L2 is not expected. On Epstein et al.’s proposal, there is no reason to expect change or development in the domain of functional categories for a different reason, since all categories (including L2 categories) are present from early stages. Convergence on the L2 grammar, then, is guaranteed (Flynn, 1996, p. 150). The only kind of development to be expected is in the surface instantiation of abstract categories in the language-particular morphology of the L2. The Minimal Trees Hypothesis also appears to predict eventual convergence on the L2 functional properties, as L2 functional categories are gradually added, in response to the L2 input. Whether predicting ultimate divergence from or convergence on the L2 grammar, the above researchers agree that the interlanguage representation does not suffer from any essential long-term impairment, that it ends up with characteristics of a natural language, be it the L1, the L2, or some other language. This contrasts with recent proposals that the interlanguage representation suffers from a permanent deficit, rendering it unlike natural languages, hence not fully UG-constrained. In recent work, Beck (1998) has suggested that inert feature values are a permanent phenomenon, a proposal also adopted by Eubank in later work (e.g., Eubank and Grace, 1998). In other words, the interlanguage representation is assumed to be defective not just initially and temporarily but permanently. In terms of our example, this means that variable word orders in the case of English-speaking learners of French or French-speaking learners of English are predicted to be found even in the endstate. The results of Yuan (2000), mentioned above, argue against this claim: Yuan demonstrates that L2 learners can indeed reset feature strength to the value appropriate for the L2, even when the L1 value is different (as is the case for the French-speaking learners of Chinese), and that there is no variability in word order at any level of proficiency. Meisel (1997) proposes more global impairment to functional (and other) properties. He argues that interlanguage grammars are of an essentially different nature from those found in L1 acquisition. He points to differences between L1 and L2 acquisition: in L1 acquisition, the position of the verb is determined by finiteness (compare (5a) and (5b) ), whereas, according to Meisel, in L2 acquisition it is not. Prévost and White (2000) provide counter-arguments and data that show that verb placement is not as free as Meisel suggests. In order to investigate the nature of the interlanguage representation in the functional domain, some of the researchers discussed above have considered both morphological properties (namely whether inflection is present or absent, On the Nature of Interlanguage Representation 35 accurate or faulty) and syntactic ones (whether there are alternations suggestive of verb movement to higher functional projections). Thus, Vainikka and Young-Scholten (1994) argued that the early interlanguage exhibits both a lack of verbal morphology and a lack of word orders indicating movement; Eubank (1993/4) argued that syntactic optionality is associated with absence of inflection; Meisel (1997) argued that both interlanguage morphology and interlanguage verb placement are variable. But what is one to conclude if syntactic reflexes of feature strength are demonstrably present and morphological ones are lacking or not robustly present? If the interlanguage contains a full complement of functional categories, it might seem somewhat mysterious that L2 learners reveal problems in the domain of morphology associated with functional categories, such as verb inflection. If functional categories are in place, and in place early, why should L2 learners have problems with morphology? Yet it is well known that they exhibit variability in their use of inflection, with tense and agreement morphology sometimes present and sometimes absent in L2 production. This issue is addressed by Lardiere (1998a, 1998b), who provides a case study of an adult L2 English speaker, Patty, whose L1 is Chinese and whose interlanguage grammar is clearly at its endstate. Patty reveals a lack of consistency in her use of English inflectional morphology: tense marking on verbs in spontaneous production is at about 35 percent, while 3rd person singular agreement is less than 17 percent. At the same time, Patty shows full command of a variety of syntactic phenomena which suggest that tense and agreement are represented in her grammar, with appropriate weak values. For example, Patty shows 100 percent correct incidence of nominative case assignment (nominative case being checked in I, hence implicating this functional category) and complete knowledge of the fact that English verbs do not raise. In other words, she shows no variability in verb placement with respect to adverbs or negation. Word orders like (7b) and (7d) are never found; rather she consistently produces orders like (7a) and (7c), suggesting that verbal features are appropriately weak. According to Eubank and Grace (1998), if interlanguage grammars have permanently inert features, then learners with an L1 with weak features, such as Chinese, learning an L2 also with weak features, like English, should allow optional verb movement. However, Lardiere shows that Patty’s interlanguage grammar disallows verb movement and that her problems are not due to any deficit in functional features as such. Even in the absence of appropriate inflectional morphology, functional categories and their feature specifications are present in the grammar and function in ways appropriate for the L2. In this case, then, the underlying grammar does in fact converge on the native grammar, though the surface morphology is divergent, in the sense that it is often absent. Lardiere argues that this divergence reflects a problem in mapping from abstract categories to their particular surface morphological manifestations. This problem in surface mapping is very different from the impairment to the grammar implied by inert features. In the former case, abstract properties are 36 Lydia White present and the grammar shows reflexes of feature strength, such as appropriate case marking and word order. There is nothing in UG that says that past tense in English must be realized by a morpheme /-ed/ or that agreement must manifest itself as /-s/ in the 3rd person singular. Yet it is this realization that is problematic, rather than the syntactic consequences of tense or agreement. To conclude this section, while the issues are by no means resolved, it seems clear that we have left behind the more general, global question (is there access to UG?) and are now probing quite intricate properties of the interlanguage representation, in order to understand the nature of the grammar that the learner creates to account for the L2. (Of course, the issue of UG involvement is still central, since a grammar constrained by UG will be different in nature from one that is not.) Interesting conceptual questions are being raised: does it make sense to think of an interlanguage representation as being defective in one domain (morphological mapping) but not another (syntax); does it make sense to think of some features being impaired but not others? If the interlanguage representation indeed draws on a variety of knowledge sources (UG, the L1, etc.), how do these come together? 7 Beyond Representation UG is a theory relevant to the issue of linguistic competence, a theory as to the nature of grammatical representation. Although UG provides constraints on possible grammars in the course of acquisition, it is not, of itself, a theory of acquisition. This point is often misunderstood, perhaps because of terms like “Language Acquisition Device” (LAD) (Chomsky, 1965), which many people in the past equated with UG. It would be more accurate to think of UG as a component within an LAD or as part of a language faculty. A theory of language acquisition will also have to include learning principles, processing principles, triggering algorithms, etc. In other words, in addition to a theory of constraints on interlanguage representation, we need a theory of how that representation is acquired, a theory of development (whether we are talking about L1 or L2 acquisition). A number of researchers have pointed out that theories of acquisition must explain both the representational problem (what L2 learners come to know) and the developmental problem (how they attain this knowledge) (e.g., Carroll, 1996; Felix, 1987; Gregg, 1996; Klein and Martohardjono, 1999). Most research looking at the operation of UG in second language acquisition has focused on the nature of the L2 learner’s grammar, looking for evidence for or against the involvement of principles and parameters of UG, and exploring the nature of the initial state and subsequent grammars. These are representational issues, as we have seen. Even if one looks for UG-based properties in learner grammars at various points in time, this is a question of representation rather than development. A representational theory is not the same as a developmental one; there is clearly a need for both and room for both. A representational theory makes claims On the Nature of Interlanguage Representation 37 about what learner grammars are like (a grammar at time X conforms to property X and at time Y to property Y) but does not seek to explain how or why grammars develop in a particular way. We should bear in mind that UG itself is not a learning theory; it can only interact with other theories that try to explain development. To account for grammar change (i.e., development), one needs a theory of how the L2 input interacts with the existing grammar, what properties of the input act as triggers for change, what properties force changes to the current representation, what might drive stages of acquisition. Some L2 learnability work has looked into these kinds of questions (the role of positive and negative evidence, learning principles, proposals that grammar change is failure driven, possible triggers in the input, etc.) (e.g., Schwartz and Sprouse, 1994; Trahey and White, 1993; White, 1991a). However, this is an area where much remains to be done. Another issue is relevant in this context. In the field of second language acquisition, there is often a confusion between competence (in the sense of underlying linguistic representation) and performance (use of that representation to understand and produce language). People often look at L2 performance, note that it differs from that of native speakers, and argue that this demonstrates essential defects in competence, or lack of UG (the comparative fallacy again). But it is in fact possible that L2 learners’ underlying competence is to some extent hidden by performance factors, such as the demands of processing or parsing. Knowledge and use of knowledge do not always coincide. In recent years, there has been an increase in research which investigates how the interlanguage mental representation is accessed during processing, seeking to determine how the representation is used on-line and off-line and the extent to which processing pressures may mask competence (e.g., Juffs and Harrington, 1995; Schachter and Yip, 1990). Again, this is an area where more research is needed. 8 Conclusion It is not the aim of UG-based theories of second language acquisition to account for all aspects of L2 development. These theories concentrate largely on the nature of unconscious interlanguage knowledge. I have argued that it is not necessary to show that the interlanguage representation is identical to the grammars of native speakers of the L2 in order to demonstrate that the representation is constrained by UG. The pursuit of interlanguage representation has led to a number of interesting and competing proposals: that interlanguage grammars are natural language grammars, constrained by UG (on some accounts, restricted to L1 properties, on other accounts not), versus that interlanguage grammars suffer from impairments (permanent, according to some researchers). The local impairment position contrasts with earlier views which assumed a more global deficit, in the form of a total inability to reset parameters (e.g., Clahsen and Muysken, 1989). 38 Lydia White In conclusion, it is important to bear in mind that claims for UG operation in L2 acquisition are simply claims that interlanguage grammars will fall within a limited range, that the “hypothesis space” is specified by UG. As Dekydtspotter et al. (1998, p. 341, n. 1) point out: “Given that the sole ‘role’ of UG is to restrict the hypothesis space available to the language acquirer, Full Restriction might be a more perspicuous name than the standard Full Access.” If we have to use such terms at all, this one has many advantages, since it focuses our attention on properties of the learner’s representation, while at the same time reminding us that the restrictions come from UG. NOTES 1 For a more recent treatment of this phenomenon, see Noguchi (1997). 2 The examples are drawn from Kanno (1997). The following abbreviations are used: NOM = nominative; ACC = accusative; TOP = topic. 3 Of course native speaker control groups should be included in experiments in order to make sure that the test instrument achieves what it is meant to test. This is a different matter. 4 For purposes of exposition, I ignore analyses that have tense (T) and agreement (Agr) heading their own projections (e.g., Pollock, 1989). 5 Where features are weak, feature checking is achieved by the mechanism of covert movement (Chomsky, 1995). 6 This is an oversimplification, which I will adopt for the sake of the argument. See Pollock (1989). 7 See White (1992a) and Schwartz and Sprouse (2000) for analyses that account for these data in a full transfer framework. 8 In fact, Epstein et al. (1996, p. 751) reject this possibility as well, so that it is impossible to determine their precise position on the initial state. REFERENCES Adjémian, C. 1976: On the nature of interlanguage systems. Language Learning, 26, 297–320. Archangeli, D. and Langendoen, T. (eds) 1997: Optimality Theory: An Overview. Oxford: Blackwell. Beck, M. 1998: L2 acquisition and obligatory head movement: Englishspeaking learners of German and the local impairment hypothesis. Studies in Second Language Acquisition, 20, 311– 48. Bley-Vroman, R. 1983: The comparative fallacy in interlanguage studies: the case of systematicity. Language Learning, 33, 1–17. Bley-Vroman, R. 1990: The logical problem of foreign language learning. Linguistic Analysis, 20, 3– 49. Borer, H. 1984: Parametric Syntax. Dordrecht: Foris. Carroll, S. 1996: Parameter-setting in second language acquisition: explanans and explanandum. On the Nature of Interlanguage Representation 39 Behavioral and Brain Sciences, 19, 720–1. Chomsky, N. 1965: Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, N. 1981: Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1995: The Minimalist Program. Cambridge, MA: MIT Press. Clahsen, H. and Muysken, P. 1989: The UG paradox in L2 acquisition. Second Language Research, 5, 1–29. Clahsen, H., Eisenbeiss, S., and Vainikka, A. 1994: The seeds of structure: a syntactic analysis of the acquisition of Case marking. In T. Hoekstra and B. D. Schwartz (eds), Language Acquisition Studies in Generative Grammar. Amsterdam: John Benjamins, 85–118. Cook, V. 1988: Chomsky’s Universal Grammar: An Introduction. Oxford: Blackwell. Cook, V. 1997: Monolingual bias in second language acquisition research. Revista Canaria de Estudios Ingleses, 34, 35–49. Corder, S. P. 1967: The significance of learners’ errors. International Review of Applied Linguistics, 5, 161–70. Dekydtspotter, L., Sprouse, R., and Anderson, B. 1998: Interlanguage A-bar dependencies: binding construals, null prepositions and Universal Grammar. Second Language Research, 14, 341–58. duPlessis, J., Solin, D., Travis, L., and White, L. 1987: UG or not UG, that is the question: a reply to Clahsen and Muysken. Second Language Research, 3, 56–75. Emonds, J. 1978: The verbal complex V′–V in French. Linguistic Inquiry, 9, 151–75. Epstein, S., Flynn, S., and Martohardjono, G. 1996: Second language acquisition: theoretical and experimental issues in contemporary research. Brain and Behavioral Sciences, 19, 677–758. Eubank, L. 1993/4: On the transfer of parametric values in L2 development. Language Acquisition, 3, 183–208. Eubank, L. 1994: Optionality and the initial state in L2 development. In T. Hoekstra and B. D. Schwartz (eds), Language Acquisition Studies in Generative Grammar. Amsterdam: John Benjamins, 369–88. Eubank, L. and Grace, S. 1998: V-to-I and inflection in non-native grammars. In M. Beck (ed.), Morphology and its Interface in L2 Knowledge. Amsterdam: John Benjamins, 69–88. Felix, S. 1987: Cognition and Language Growth. Dordrecht: Foris. Finer, D. and Broselow, E. 1986: Second language acquisition of reflexivebinding. In S. Berman, J.-W. Choe, and J. McDonough (eds), Proceedings of NELS 16. Amherst, MA: Graduate Linguistics Students Association, 154 –68. Flynn, S. 1996: A parameter-setting approach to second language acquisition. In W. Ritchie and T. Bhatia (eds), Handbook of Language Acquisition. San Diego: Academic Press, 121–58. Gregg, K. R. 1996: The logical and developmental problems of second language acquisition. In W. Ritchie and T. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 49–81. Grimshaw, J. and Rosen, S. T. 1990: Knowledge and obedience: the developmental status of the binding theory. Linguistic Inquiry, 21, 187–222. Grondin, N. and White, L. 1996: Functional categories in child L2 acquisition of French. Language Acquisition, 5, 1–34. Hale, K. 1996: Can UG and the L1 be distinguished in L2 acquisition? Behavioral and Brain Sciences, 19, 728–30. Hawkins, R. 1998: The inaccessibility of formal features of functional categories 40 Lydia White in second language acquisition. Paper presented at the Pacific Second Language Research Forum. Tokyo, March. Hawkins, R. and Chan, Y.-C. 1997: The partial availability of Universal Grammar in second language acquisition: the “failed features” hypothesis. Second Language Research, 13, 187–226. Hawkins, R., Towell, R., and Bazergui, N. 1993: Universal Grammar and the acquisition of French verb movement by native speakers of English. Second Language Research, 9, 189–233. Hyams, N. 1996: The underspecification of functional categories in early grammar. In H. Clahsen (ed.), Generative Perspectives on Language Acquisition: Empirical Findings, Theoretical Considerations, Crosslinguistic Comparisons. Amsterdam: John Benjamins, 91–127. Juffs, A. and Harrington, M. 1995: Parsing effects in second language sentence processing: subject and object asymmetries in wh-extraction. Studies in Second Language Acquisition, 17, 483–516. Kanno, K. 1997: The acquisition of null and overt pronominals in Japanese by English speakers. Second Language Research, 13, 265–87. Klein, E. 1995: Evidence for a “wild” L2 grammar: when PPs rear their empty heads. Applied Linguistics, 16, 87–117. Klein, E. and Martohardjono, G. 1999: Investigating second language grammars: some conceptual and methodological issues in generative SLA research. In E. Klein and G. Martohardjono (eds), The Development of Second Language Grammars: A Generative Perspective. Amsterdam: John Benjamins, 3–34. Lardiere, D. 1998a: Case and tense in the “fossilized” steady state. Second Language Research, 14, 1–26. Lardiere, D. 1998b: Dissociating syntax from morphology in a divergent end- state grammar. Second Language Research, 14, 359–75. Lardiere, D. 2000: Mapping features to forms in second language acquisition. In J. Archibald (ed.), Second Language Acquisition and Linguistic Theory. Oxford: Blackwell, 102–29. Liceras, J. 1983: Markedness, contrastive analysis and the acquisition of Spanish as a second language. Ph.D. thesis. University of Toronto. Liceras, J., Maxwell, D., Laguardia, B., Fernández, Z., and Fernández, R. 1997: A longitudinal study of Spanish nonnative grammars: beyond parameters. In A. T. Pérez-Leroux and W. Glass (eds), Contemporary Perspectives on the Acquisition of Spanish. Vol. 1: Developing Grammars. Somerville, MA: Cascadilla Press, 99–132. Martohardjono, G. 1993: Wh-movement in the acquisition of a second language: a crosslinguistic study of three languages with and without movement. Ph.D. thesis. Cornell University. Martohardjono, G. and Gair, J. 1993: Apparent UG inaccessibility in second language acquisition: misapplied principles or principled misapplications? In F. Eckman (ed.), Confluence: Linguistics, L2 Acquisition and Speech Pathology. Amsterdam: John Benjamins, 79–103. Meisel, J. 1997: The acquisition of the syntax of negation in French and German: contrasting first and second language acquisition. Second Language Research, 13, 227– 63. Montalbetti, M. 1983: After binding: on the interpretation of pronouns. Ph.D. dissertation. MIT. Nemser, W. 1971: Approximative systems of foreign language learners. International Review of Applied Linguistics, 9, 115–23. Noguchi, T. 1997: Two types of pronouns and variable binding. Language, 73, 770–97. On the Nature of Interlanguage Representation 41 Pérez-Leroux, A. T. and Glass, W. 1997: OPC effects in the L2 acquisition of Spanish. In A. T. Pérez-Leroux and W. Glass (eds), Contemporary Perspectives on the Acquisition of Spanish. Vol. 1: Developing Grammars. Somerville, MA: Cascadilla Press, 149–65. Pinker, S. 1994: The Language Instinct. New York: William Morrow. Pollock, J.-Y. 1989: Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry, 20, 365–424. Prévost, P. and White, L. 2000: Missing surface inflection or impairment in second language acquisition? Evidence from tense and agreement. Second Language Research, 16, 103–33. Robertson, D. and Sorace, A. 1999: Losing the V2 constraint. In E. Klein and G. Martohardjono (eds), The Development of Second Language Grammars: A Generative Approach. Amsterdam: John Benjamins, 317– 61. Schachter, J. 1988: Second language acquisition and its relationship to Universal Grammar. Applied Linguistics, 9, 219–35. Schachter, J. 1989: Testing a proposed universal. In S. Gass and J. Schachter (eds), Linguistic Perspectives on Second Language Acquisition. Cambridge: Cambridge University Press, 73–88. Schachter, J. 1990: On the issue of completeness in second language acquisition. Second Language Research, 6, 93–124. Schachter, J. and Yip, V. 1990: Grammaticality judgments: why does anyone object to subject extraction? Studies in Second Language Acquisition, 12, 379–92. Schwartz, B. D. 1987: The modular basis of second language acquisition. Ph.D. dissertation. University of Southern California. Schwartz, B. D. 1993: On explicit and negative data effecting and affecting competence and “linguistic behavior.” Studies in Second Language Acquisition, 15, 147–63. Schwartz, B. D. 1998a: On two hypotheses of “Transfer” in L2A: minimal trees and absolute L1 influence. In S. Flynn, G. Martohardjono, and W. O’Neil (eds), The Generative Study of Second Language Acquisition. Mahwah, NJ: Lawrence Erlbaum Associates, 35–59. Schwartz, B. D. 1998b: The second language instinct. Lingua, 106, 133–60. Schwartz, B. D. and Sprouse, R. 1994: Word order and nominative case in nonnative language acquisition: a longitudinal study of (L1 Turkish) German interlanguage. In T. Hoekstra and B. D. Schwartz (eds), Language Acquisition Studies in Generative Grammar. Amsterdam: John Benjamins, 317–68. Schwartz, B. D. and Sprouse, R. 1996: L2 cognitive states and the full transfer/ full access model. Second Language Research, 12, 40–72. Schwartz, B. D. and Sprouse, R. 2000: When syntactic theories evolve: consequences for L2 acquisition research. In J. Archibald (ed.), Second Language Acquisition and Linguistic Theory. Oxford: Blackwell, 156–86. Selinker, L. 1972: Interlanguage. International Review of Applied Linguistics, 10, 209–31. Smith, N. and Tsimpli, I.-M. 1995: The Mind of a Savant. Oxford: Blackwell. Sorace, A. 1993: Incomplete and divergent representations of unaccusativity in non-native grammars of Italian. Second Language Research, 9, 22– 48. Sprouse, R. 1998: Some notes on the relationship between inflectional morphology and parameter setting in first and second language acquisition. In M. Beck (ed.), Morphology and the Interfaces in Second Language Knowledge. Amsterdam: John Benjamins, 41–67. 42 Lydia White Thomas, M. 1991a: Do second language learners have “rogue” grammars of anaphora? In L. Eubank (ed.), Point Counterpoint: Universal Grammar in the Second Language. Amsterdam: John Benjamins, 375–88. Thomas, M. 1991b: Universal Grammar and the interpretation of reflexives in a second language. Language, 67, 211–39. Trahey, M. and White, L. 1993: Positive evidence and preemption in the second language classroom. Studies in Second Language Acquisition, 15, 181–204. Vainikka, A. and Young-Scholten, M. 1994: Direct access to X′-theory: evidence from Korean and Turkish adults learning German. In T. Hoekstra and B. D. Schwartz (eds), Language Acquisition Studies in Generative Grammar. Amsterdam: John Benjamins, 265–316. Vainikka, A. and Young-Scholten, M. 1996: Gradual development of L2 phrase structure. Second Language Research, 12, 7–39. Vainikka, A. and Young-Scholten, M. 1998: The initial state in the L2 acquisition of phrase structure. In S. Flynn, G. Martohardjono, and W. O’Neil (eds), The Generative Study of Second Language Acquisition. Mahwah, NJ: Lawrence Erlbaum Associates, 17–34. Wexler, K. 1994: Optional infinitives, head movement and the economy of derivations. In D. Lightfoot and N. Hornstein (eds), Verb Movement. Cambridge: Cambridge University Press, 305–50. White, L. 1985: Is there a logical problem of second language acquisition? TESL Canada, 2, 29–41. White, L. 1989a: The principle of adjacency in second language acquisition: do L2 learners observe the subset principle? In S. Gass and J. Schachter (eds), Linguistic Perspectives on Second Language Acquisition. Cambridge: Cambridge University Press, 134–58. White, L. 1989b: Universal Grammar and Second Language Acquisition. Amsterdam: John Benjamins. White, L. 1990: Second language acquisition and universal grammar. Studies in Second Language Acquisition, 12, 121–33. White, L. 1991a: Adverb placement in second language acquisition: some effects of positive and negative evidence in the classroom. Second Language Research, 7, 133–61. White, L. 1991b: Argument structure in second language acquisition. Journal of French Language Studies, 1, 189–207. White, L. 1991c: The verb-movement parameter in second language acquisition. Language Acquisition, 1, 337–60. White, L. 1992a: Long and short verb movement in second language acquisition. Canadian Journal of Linguistics, 37, 273–86. White, L. 1992b: Subjacency violations and empty categories in L2 acquisition. In H. Goodluck and M. Rochemont (eds), Island Constraints. Dordrecht: Kluwer, 445–64. White, L. 1996: Universal grammar and second language acquisition: current trends and new directions. In W. Ritchie and T. Bhatia (eds), Handbook of Language Acquisition. New York: Academic Press, 85–120. Yuan, B. 2000: Is thematic verb raising inevitable in the acquisition of a nonnative language? In C. Howell, S. Fish, and T. Keith-Lucas (eds), Proceedings of the 24th Annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press, 797–807. The Radical Middle 43 3 The Radical Middle: Nativism without Universal Grammar WILLIAM O’GRADY 1 Introduction A phenomenon as puzzling and complex as language acquisition is no doubt worthy of the controversy that its study has engendered. Indeed, it would be unreasonable to expect a broad consensus on such a profoundly mysterious phenomenon after a mere 30 or 40 years of investigation, much of it focused on the acquisition of a single language. Under these circumstances, the most that can perhaps be hoped for in the near term is some agreement on the research questions that need to be addressed and on the merits and shortcoming of the various explanatory ideas that are currently being pursued. In the longer term, of course, one hopes for a convergence of views, and even now there is some indication that this has begun in a limited way, as I will explain below. Nonetheless, for the time being at least, there is still ample room for disagreement on many important points. The purpose of this chapter is to outline a view of language acquisition – both first and second – that is sometimes referred to as “general nativism.” I will begin in the next section by offering an overview of this approach, including its principal claims and the major challenges that it faces. Section 3 outlines a general nativist theory of syntactic representations with respect to a wellestablished asymmetry in the development of relative clauses in the course of first and second language acquisition. Section 4 addresses the possible advantages of general nativism compared to other theories of language acquisition. 2 Defining General Nativism There is a near-consensus within contemporary linguistics (which I will not question here) that language should be seen as a system of knowledge – a sort of “mental grammar” consisting of a lexicon that provides information about 44 William O’Grady the linguistically relevant properties of words and a computational system that is responsible for the formation and interpretation of sentences. The details of the computational system and even of the lexicon are the subject of ongoing dispute, of course, but there is substantial agreement on a number of points. For instance, it seems clear that the grammar for any human language must assign words to categories of the appropriate type (noun, verb, etc.), that it must provide a set of mechanisms for combining words into phrases and sentences with a particular internal architecture, and that it must impose constraints on phenomena such as “movement” and pronoun interpretation. What makes matters especially interesting for theories of language acquisition is that grammars that include even these basic and relatively uncontroversial mechanisms are underdetermined by experience in significant ways. As far as we can tell, for instance, the input to the acquisition process (i.e., the speech of others) includes no direct information about the criteria for category membership, the architecture of syntactic representations, or the content of constraints on movement and pronoun interpretation. (For a general review, see O’Grady, 1997, pp. 249 ff.) How then can a language be acquired? Theories of linguistic development typically address this problem by assuming that children are endowed with an “acquisition device” – an innate system that both guides and supplements the learner’s interaction with experience. This much is accepted by a broad spectrum of researchers ranging from Slobin (e.g., 1985, p. 1158) to Chomsky (e.g., 1975, p. 13), but differences arise on one important point. In one class of acquisition theories, a significant portion of the grammar is taken to be “given in advance” by the acquisition device. This grammatical component of the inborn acquisition device is known as Universal Grammar, or UG – a system of categories and principles that is taken to determine many of the core properties of human language (see figure 3.1). Such theories are instances of what might be called “grammatical nativism,” since they adopt the view that the innate endowment for language includes actual grammatical categories and principles. Elsewhere, I have referred to this view as “special nativism” (O’Grady, 1997, p. 307), because of its commitment to the existence of innate mechanisms with a specifically grammatical character (see also White, this volume). Grammatical nativism contrasts with “general nativism,” which posits an innate acquisition device but denies that it includes grammatical categories or principles per se. According to this view (which might also be labeled “cognitive nativism” or “emergentism,” as is more common these days), the entire grammar is the product of the interaction of the acquisition device with experience; no grammatical knowledge is inborn (see figure 3.2) (see Ellis, this volume). Acquisition device Experience UG Grammar Figure 3.1 The UG-based acquisition device The Radical Middle 45 Acquisition device Experience Grammar Figure 3.2 The general nativist acquisition device Later in this chapter, I will suggest that there are some signs of convergence between general nativism and recent versions of grammatical nativism. For now, though, I would like to emphasize the profound historical difference between the two views. UG is not simply the name for whatever mechanisms happen to be involved in grammatical development. As I interpret the literature on grammatical nativism, proponents of the view that UG is part of the acquisition device subscribe to a very strong claim about its content and character – namely, that it is an autonomous system of grammatical categories and principles – autonomous in the sense that it is not reducible to non-linguistic notions and grammatical in the sense that it is primarily concerned with matters of wellformedness, not parsing or processing or other types of language-related cognition. (For detailed discussion, see Newmeyer, 1998.) All varieties of general nativism reject these assumptions, however much they may disagree on what the acquisition device actually does comprise. Skepticism concerning UG is widespread in the field of language acquisition research. Relatively little of the literature on first language acquisition is couched within a UG framework, and the same seems to be true of the literature on second language acquisition as well. In addition to the huge amount of work that simply ignores UG, there is also a substantial and varied literature that explicitly rejects it in one form or another. This includes work by Martin Braine (1987), Dan Slobin (1985), Melissa Bowerman (1990), and Michael Tomasello (1995) (among many others) on first language acquisition and work by Eric Kellerman (Kellerman and Yoshioka, 1999), Fred Eckman (1996), Kate Wolfe-Quintero (1992, 1996), and others on second language acquisition. It should be noted, though, that there is no unified general nativist approach to language acquisition and certainly no agreement on the particular views that I outline in the remainder of this chapter. As I see it, the principal limitation of most work on general nativism lies in its failure to develop a theory of learnability and development that is tied to an explicit and comprehensive theory of grammar (see also Gregg, 1996). Most non-UG work is quite casual in its approach to syntax: the phenomena whose acquisition is being investigated are typically analyzed informally and on a caseby-case basis, without reference to an overarching syntactic theory. By contrast, work in the special nativist tradition has not only put forward a theory of learnability (built around an inborn UG) but linked it to a far-reaching and explicit theory of grammar (transformational grammar in its various incarnations). For reasons that I will discuss further below, the most promising theories of language posit explanatory principles that make reference to phonological, 46 William O’Grady syntactic, and semantic representations of various sorts. Yet the vast majority of work on general nativism either makes no reference to such representations or adopts a very casual view as to their properties, typically avoiding any explicit proposal about their architecture or ontogeny. A good illustration of this point comes from an important body of research on the acquisition of relative clauses by second language learners (e.g., Doughty, 1991; Eckman, Bell, and Nelson, 1988; Gass, 1979, 1980). This work has yielded a robust and interesting finding: subject relative clauses such as (1) are easier than direct object relatives such as (2) for second language learners. (The same seems to be true for first language acquisition, all other things being equal; see O’Grady, 1997, p. 179 for discussion.) (1) Subject relative: the truck that [_ pushed the car] (2) Object relative: the truck that [the car pushed _] Further, it has been observed that this finding parallels an important generalization in syntactic typology dating back at least to Keenan and Comrie (1977): direct object relatives are more marked than subject relatives. (That is, some languages have only subject relatives, but any language with direct object relatives must also permit subject relatives.) The developmental pattern and its relationship to Keenan and Comrie’s typological generalization raise questions that force us to address the two principal explanatory challenges confronting contemporary linguistics: i Why is language the way it is (e.g., why do all languages with direct object relatives also have subject relatives, but not vice versa)? ii How is it acquired (e.g., why are subject relatives easier for language learners than direct object relatives)? It is my position that neither of these questions can be answered without reference to hierarchically structured symbolic representations. On this view, then, the first priority for general nativism must be a theory of syntactic representations that includes a proposal about their composition and architecture. 3 A General Nativist Theory of Representations In a number of recent publications (e.g., O’Grady, 1996, 1997, 1998), I have put forward the outlines of a general nativist theory of syntactic representations. As I see it, the key to such a theory lies in two propositions. First, syntactic categories, which are treated as purely formal elements in special nativism, must be reducible to a semantic base. I have made one proposal about precisely how The Radical Middle 47 Step 1: Combination of the subject and verb N V Mary speaks Figure 3.3 First step in the formation of the sentence Mary speaks French Targeting the verb Step 2: Combination with the second argument N V N V N Mary speaks Mary speaks French Figure 3.4 Second step in the formation of the sentence Mary speaks French this might be achieved (O’Grady, 1997, 1998), and other ideas can be found in the literature on grammatical categories (e.g., Croft, 1991; Langacker, 1987). Second, contra the view adopted within UG-based approaches to language acquisition, the computational principles that combine and arrange words to form phrases and sentences cannot be specifically grammatical in character (that is, there is no X-bar Schema, no Empty Category Principle, and so forth). How then do we account for the sorts of grammatical phenomena that have been the focus of so much linguistic research since the early 1960s? In recent work on this matter (e.g., O’Grady, 2001b), I have proposed that the theory of sentence structure can and should be unified with the theory of sentence processing. As I see it, the processor itself has no specifically grammatical properties. Rather, its design reflects two more general computational features – a propensity to operate on pairs of elements (a characteristic of the arithmetical faculty as well)1 and a propensity to combine functors with their arguments at the first opportunity (a storage-reducing strategy that I refer to simply as “efficiency”). The system operates in a linear manner (i.e., “from left to right”), giving the result depicted in figure 3.3 in the case of a simple transitive sentence such as Mary speaks French. In the next step, the verb combines directly with its second argument, an operation that requires splitting the previously formed phrase in the manner depicted in figure 3.4. (Such an operation has long been assumed, at least implicitly, in the literature on sentence processing; see, e.g., Frazier, 1987, p. 561; Levelt, 1989, p. 242; Marcus, 1980, pp. 79–80.) Syntactic representations in this type of efficiency-driven computational system have the familiar binary-branching design, with the subject higher than the direct object – but not as the result of an a priori grammatical blueprint such as the X-bar schema. Rather, their properties are in a sense epiphenomenal – the by-product of a sentence formation process that proceeds from left to 48 William O’Grady right, combining a verb with its arguments one at a time at the first opportunity. Syntactic representations are thus nothing more than a residual record of how the computational system goes about combining words to form sentences. The architecture of the proposed syntactic representations offers a promising account of why subject relatives are easier than direct object relatives. The key idea is that the relative difficulty (and, by extension, the developmental order) of structures that contain gaps is determined by the distance (calculated in terms of intervening nodes) between the gap and its filler (e.g., the nominal modified by the relative clause). As illustrated in (3) and (4), there is one such node in the case of subject relatives (i.e., S) and two in the case of object relatives (i.e., S and VP):2 (3) Subject relative: the truck that [S _ pushed the car] (4) Direct Object relative: the truck that [S the car [VP pushed _]] A problematic feature of English is that structural distance is confounded with linear distance: subject gaps are not only less deeply embedded than object gaps, they are also linearly closer to the head noun. In order to ensure that structural distance rather than linear distance is responsible for the contrast in the difficulty of relative clauses, it is necessary to consider the acquisition of languages such as Korean, in which the relative clause precedes the head. (The verbal suffixes in Korean simultaneously indicate both tense and clause type. RC = relative clause.) Subject relative: [S _ namca-lul cohaha-nun] yeca man-Acc like-RC.Prs woman “the woman who likes the man” structural distance: one node (S) linear distance: two words b. Direct object relative: [S Namca-ka [VP _ cohaha-nun]] yeca man-Nom like-RC.Prs woman “the woman who the man likes” structural distance: two nodes (VP and S) linear distance: one word (5) a. If structural distance is the key factor, then the subject relative should be easier; on the other hand, if linear distance is the key factor, the direct object relative should be easier. O’Grady, Lee, and Choo (forthcoming) investigated this matter with the help of a comprehension task (see box 3.1), uncovering a strong and statistically significant preference for subject relative clauses. The Radical Middle 49 Box 3.1 The acquisition of relative clauses in Korean as a second language (O’Grady et al., forthcoming) Research questions: Is there a subject–object asymmetry in the acquisition of Korean relative clauses? If so, does it reflect a contrast in linear distance or in structural distance? Methodology: Subjects: 53 native English speakers studying Korean as a second language – 25 second-semester students at the University of Texas at Austin, 20 fourth-semester students at the same institution, and 8 fourth-semester students at the University of Hawai’i at Manoa. Task: Picture selection, in accordance with the following instructions: Each page of this booklet contains a series of three pictures. As you go to each page, you will hear a tape-recorded voice describing a person or animal in one of the three pictures. Your job is simply to put a circle around the person or animal described in the sentence. (Do NOT put the circle around the entire box.) Figure 3.5 presents a sample page from the questionnaire. Figure 3.5 Sample test items 50 William O’Grady Subjects who correctly understand relative clauses should circle the right-hand figure in the third panel in response to a subject relative clause such as (ia) and the left-hand figure in the second panel in response to a direct object relative such as (ib): Subject relative clause: [_ namca-lul cohaha-nun] yeca man-Acc like-RC.Prs woman ‘the woman who likes the man’ b. Direct object relative clause: [namca-ka _ cohaha-nun] yeca man-Nom like-RC.Prs woman ‘the woman who the man likes’ (i) a. Results: The subjects did far better on subject relative clauses than on direct object relatives, with scores of 73.2 percent correct on the former pattern compared to only 22.7 percent for the latter. This contrast is highly significant (F 30.59, p = .0001). Equally revealing is an asymmetry in reversal errors (i.e., the number of times a pattern of one type was misanalyzed as a pattern of the other type): direct object relatives were misunderstood as subject relatives 115 times while subject relatives were misanalyzed as direct object relatives only 26 times – a clear indication that subject relatives are the easier pattern. Conclusion: Learners of Korean as a second language find subject relatives far easier than direct object relatives, which supports the claim that structural distance between a gap and its filler is the key factor in determining the relative difficulty of these patterns. If the structural distance account is correct, we expect to find comparable asymmetries in the development of other gap-containing structures as well. Wh-questions are a case in point. As illustrated in (6) and (7), subject and object wh-questions exhibit a contrast that parallels the asymmetry found in relative clauses: (6) Subject wh-question: Who [S _ met Mary]? (7) Object wh-question: Who did [S Mary [VP meet _]]? The relative difficulty of these two patterns has been studied for both first language acquisition (Yoshinaga, 1996) and second language acquisition (Kim, 1999) with the help of an elicited production task. Both studies revealed significantly better performance on subject wh-questions and a strong tendency for these patterns to be used in place of their direct object counterparts, but not vice versa. The Radical Middle 51 By adopting a particular theory of syntactic representations, then, we are able to uncover a plausible computational explanation for why object relatives are more difficult than subject relatives for language learners and for why object wh-questions are harder than subject wh-questions. This is a potential step forward, not only because it helps explain the developmental facts, but also because it sheds light on the typological facts as well. In particular, it makes sense to think that the cut-off points that languages adopt in defining the limits for relative clause formation are determined by the same measure of computational complexity that defines developmental difficulty. Thus, subject relatives – the computationally simplest structure – will be the most widespread typologically.3 Moreover, any language that allows the computationally more difficult direct object relatives will also permit the simpler subject relatives. And so on. This cannot be all there is to it, of course. Syntactic representations have properties other than just binarity, and syntactic principles make reference to more than just structural distance. The illustration given here omits many details in order to make the key point – which is that the best prospects for an explanatory general nativist theory of language lie in an approach that takes syntactic representations as its starting point. As we have just seen, reference to such representations allows us to make a proposal not only about how language is acquired (e.g., why subject relatives are acquired first) but also about why language is the way it is (e.g., why any language that allows object relatives must also allow subject relatives). The parallels between first and second language acquisition that are manifested in the emergence of relative clauses lend credence to the idea that the two phenomena are fundamentally alike, at least in some respects. I believe that this is right, at least insofar as computational operations are concerned. The matter is hardly clear, though. Indeed, the facts are somewhat difficult to interpret: as Bley-Vroman (1994, p. 4) has observed, experimental work on computational principles in second language acquisition has yielded indecisive results – “better than chance, [but] far from perfect.” Although this seems to suggest diminished access to the computational mechanisms underlying sentence formation, a less pessimistic view is adopted by Uziel (1993), who follows Grimshaw and Rosen (1990) in arguing that any indication that learners perform above the level of chance on contrasts involving computational principles should be interpreted as evidence for access to those principles – a not unreasonable proposal in light of the many extraneous factors (e.g., inattention, processing limitations, vocabulary deficits, nervousness, and so forth) that can interfere with performance in experimental settings. (See also White, this volume.) If this is right, then performance on computational principles should improve as the effect of extraneous factors diminishes. There is already some indication that this is right: Kanno (1996) investigates the status of a computational principle that is responsible for the asymmetry in the admissibility of case drop in subject and direct object positions in Japanese (see section 4 for details). Because the contrast is manifested in very simple sentences, Kanno was able to elicit 52 William O’Grady grammaticality judgments for sentences that were just two and three words long, thereby dramatically diminishing the potential effect of extraneous factors. Interestingly, she reports that adult learners of Japanese as a second language do not perform significantly differently from native speakers in assessing the relative acceptability of the two patterns. Why then are adults such poor language learners? There are a number of possibilities, of course, two of which I find particularly interesting. First, it is evident that some parts of the language faculty fare less well than the computational system with the passage of time. For instance, the ability to distinguish among phonemic contrasts apparently begins to diminish by the age of 12 months (Werker, Lloyd, Pegg, and Polka, 1996), with the result that language acquisition after age six or so typically results in a foreign accent (Long, 1990, p. 266). There also appears to be a significant decline in learners’ ability to exploit subtle semantic contrasts, including those underlying such familiar phenomena as the the/a contrast in English (Larsen-Freeman and Long, 1991, p. 89) or the wa/ga (topic/nominative) contrast in Japanese (Kuno, 1973, p. 37; Russel, 1985, p. 197). This suggests that the acquisition device comprises several autonomous components (at least a computational module, a perceptual module, and a conceptual module), each with its own maturational prospects and its own role to play in shaping the outcome of second language learning. A second possibility, which focuses just on syntactic deficits (see, e.g., O’Grady, 2001a), is that the computational system, while intact, is underpowered in the case of adult language learners. The effects of this deficit are manifested in patterns which, for one reason or another, place extra demands on the computational system. One such pattern involves object relative clauses, which require the establishment of a link between a direct object gap and a structurally distant filler. As we have seen, both children and adults have trouble with these patterns compared to subject relative clauses. Interestingly, similar problems have been observed in agrammatic aphasics (e.g., Grodzinsky, 2000). Another sort of pattern that may place an extra burden on the computational system involves double object datives such as (8), compared to their prepositional dative counterparts as in (9): (8) Double object dative: agent goal theme The boy sent the donkey the horse. (9) Prepositional dative: agent theme goal The boys sent the horse to the donkey. As observed by Dik (1989), Langacker (1995, pp. 18–20), and Talmy (1988), among others, the word order employed in the prepositional pattern (agent– theme-goal) is iconic with the structure of the event, which involves the agent The Radical Middle 53 acting on the theme and then transferring it to the goal, giving the “action chain” (to employ Langacker’s term) depicted in (10): (10) agent → theme → goal Interestingly, the double object dative, with its non-iconic agent–goal–theme order, is harder to comprehend, both for children in the early stages of language acquisition (Osgood and Zehler, 1981; Roeper, Lapointer, Bing, and Tavakolian, 1981; Waryas and Stremel, 1974) and for adult second language learners (Hawkins, 1987; Mazurkewich, 1984; White, 1987). And here again, agrammatic aphasics have been found to have difficulty with this pattern too (Caplan and Futter, 1986; Kolk and Weijts, 1996, p. 111; O’Grady and Lee, 2001). All of this suggests that in the early stages of language acquisition (and perhaps in the case of agrammatism as well) the computational system may be too underpowered to reliably execute the more demanding tasks involved in natural language processing, including dealing with long-distance dependencies and non-iconic word order. Whereas children routinely overcome this deficit, its effects in the case of adults may be longer lasting, contributing to the pattern of partial attainment that is typical of second language learning. 4 The Advantages of General Nativism In evaluating general nativism, it is useful to compare it with two well-known alternatives – UG-based special nativism, which posits inborn grammatical categories and principles, and connectionism, certain varieties of which deny the existence of traditional symbolic representations and principles altogether (e.g., Elman, Bates, Johnson, Karmiloff-Smith, Parisi, and Plunkett, 1996). Each approach has its own merits, of course, but it is nonetheless possible to identify considerations that justify continued pursuit of the general nativist research program. The potential advantage of general nativism with respect to special nativism is obvious. All scientific work, including the special nativist research program, seeks the most general properties and principles possible. One does not posit a grammatical rule specifically for passivization if the properties of passive structures can be derived from a more general grammatical principle. And one does not posit a grammatical constraint if the phenomena that it accounts for can be derived from principles that are not specific to the language faculty. (For an identical view within grammatical nativism, see Lightfoot, 1982, p. 45.) Interestingly, the pursuit of this very goal within the special nativist research program has led to a partial convergence of views with general nativism in recent years. As observed in O’Grady (1999), work within the “Minimalist Program” that has grown out of Government and Binding theory (e.g., Chomsky, 1995) suggests that UG as it was conventionally understood is being abandoned even by those traditionally committed to grammatical nativism in 54 William O’Grady its strongest form. The latest generation of explanatory principles focuses on the notion of economy, demanding “short moves” (the “Minimal Link Condition”) that take place only if necessary (“Last Resort”) and are postponed for as long as possible (“Procrastinate”) – in short, the sort of principles that one would expect to find in almost any computational system. (In fact, Fukui, 1996, has gone so far as to suggest that the economy principles of the Minimalist Program follow from the laws of physics!)4 A concrete example of this convergence of views can be seen in the treatment of gap-containing structures in the two varieties of nativism, where one can find parallel proposals for calculating relative complexity and markedness. As explained above, I have suggested that the relative ease of subject gaps compared to object gaps can be explained with reference to their distance from the “filler” (the head in the case of relative clauses, the wh-word in the case of questions). Working within the minimalist program, Collins (1994, p. 56) has put forward a virtually identical proposal: the cost of “movement operations” is determined by the number of nodes traversed. In the final analysis, then, general and special varieties of nativism agree on the existence of an inborn acquisition device, of hierarchically structured symbolic representations, and of explanatory principles that refer to these representations. The principal difference between the two approaches revolves around the precise nature of these constructs, with disagreement centered on the question of whether the language faculty includes inborn categories and mechanisms that are narrowly grammatical in character. But even here, there is agreement that we should seek out the most general constructs that are consistent with a viable account of the properties of language and the facts of development. What remains to be determined is whether some of these constructs have the status necessary to justify continued adherence to the traditional conception of Universal Grammar. At first glance at least, the type of general nativism advocated here shares much less common ground with connectionism. This is somewhat ironic since, in a sense, connectionism is an extreme form of general nativism. Indeed, some of its current proponents (e.g., Elizabeth Bates and Brian MacWhinney) were earlier associated with a more traditional general nativist perspective (e.g., Bates and MacWhinney, 1988), and Elman et al. (1996, p. 114) note that connectionism embodies aspects of Piaget’s (general nativist) theory of the mind. As I see it, the attractiveness of connectionism stems in large part from the fact that it takes the pursuit of generality so seriously, ultimately arriving at the strongest possible conclusion concerning the nature of the human language faculty – namely that it has no special properties of its own, grammatical or otherwise. This idea deserves to be taken seriously. Ultimately, though, the connectionist program must be evaluated in terms of the same criteria as apply to all theories of language: it must account both for how language is acquired and for why it is the way it is. To date, connectionist work seems to have concentrated almost exclusively on the former question. There have been The Radical Middle 55 impressive results in this area, but, for me at least, the challenge of explaining why language is the way it is has yet to be satisfactorily addressed. A simple example will help illustrate this point. As is well known, many languages exhibit so-called “subject–verb” agreement: affixation on the verb records person and number features of the subject. For example: (11) 3rd person, singular subject: English That man works hard. Spanish Ese hombre trabaja mucho. 3rd person, plural subject: Those men work∅ ∅ hard. Esos hombres trabajan mucho. We know from the intriguing work of Elman (1993) and others that it is possible to build a connectionist net that can “learn” subject–verb agreement without reference to hierarchical syntactic representations per se. Moreover, on the face of it, it appears that such a proposal could count as an explanation for how at least this feature of language is acquired. But there is another challenge here. This is because the same connectionist net could almost certainly “learn” a language – call it Lisheng – in which agreement is triggered by the direct object rather than the subject: (12) Lisheng 3rd person, singular object: I visited-a that city. 3rd person, plural object: I visited-an those cities. The problem is that there is apparently no such language: there are languages such as English and Spanish in which the verb agrees only with the subject and languages such as Swahili in which the verb agrees with both the subject and the direct object, but no languages in which the verb agrees only with the direct object (e.g., Croft, 1990, p. 106). Why should this be? This asymmetry has a straightforward explanation in theories of language that make use of hierarchically structured syntactic representations: the need for agreement to mark a head–argument relation increases with the computational distance between the two elements. Since verbs are structurally closer to their direct objects than to their subjects in the sort of representation that I posit, it follows that the need for agreement is greater in the latter case. This is true not only for SOV languages such as Tamil, in which the subject is linearly more distant from the verb, but also for SVO languages such as English, in which the subject and direct object are both adjacent to the verb, and for VSO languages such as Irish, in which the subject is linearly closer to the verb than is the direct object (see figure 3.6).5 56 William O’Grady SOV: SVO: VSO: NP NP NP V NP V NP V NP Figure 3.6 The subject–object asymmetry Syntactic representations such as these shed light on other phenomena as well. For instance, it is surely no accident that in languages such as Japanese, case can be dropped from the direct object but not from the subject (Fukuda, 1993): the need for case presumably is greater on the more distant of the verb’s arguments: (13) a. Case drop on the subject: *Dare gakusei-o nagutta-no? who student-ac hit -Ques ‘Who hit the student?’ b. Case drop on the direct object: Gakusei-ga dare nagutta-no? student-Nom who hit -Ques ‘Who did the student hit?’ Explanations such as these are plainly based on processing considerations. As such, they are perfectly compatible with Elman et al.’s hint (1996, p. 386) that linguistic universals are perhaps attributable to processing mechanisms – an idea that they do not develop. Crucially, however, the specific processing factors that underlie agreement and case drop asymmetries come to light only when we consider symbolic representations with the defining properties of traditional syntactic structure – binary branching and a subject–object asymmetry. (Recall, though, that these architectural features are derived from general computational properties, not UG, in the approach that I adopt.) It remains to be seen how and whether the connectionist program deals with these issues. In the course of proposing an account for why language is the way it is with respect to phenomena such as agreement and case drop, a theory based on traditional symbolic representations also takes us a good deal of the way toward understanding how language is acquired. In the case of agreement, for instance, it seems reasonable to suppose that the computational demands associated with keeping track of the structurally more distant verb–subject relation create a place in syntactic representations where agreement would be especially welcome. Confounding factors make it difficult to test this prediction against developmental data, since subject agreement morphemes are more frequent than their object agreement counterparts and may occur in the more salient word-initial or word-final position (vs. word-medial position). Nonetheless, the developmental facts are at least suggestive. The Radical Middle 57 In languages with both subject and object agreement, there seem to be only two developmental patterns: either subject agreement is learned before object agreement (the case in Sesotho, according to Demuth, 1992, p. 600), or the two types of agreement emerge simultaneously (this is apparently what happens in West Greenlandic (Fortescue and Olsen, 1992), K’iche’ Maya (Pye, 1992), Walpiri (Bavin, 1992), and Georgian (Imedadze and Tuite, 1992). There appear to be no languages in which object agreement is acquired before subject agreement. Turning now to case drop, if in fact the computational demands associated with keeping track of the more distant verb–subject relation make it worthwhile to retain case on the subject while permitting its suppression on the direct object, we would expect this contrast to be evident in the course of linguistic development. This seems to be right: Suzuki (1999) reports that children learning Japanese exhibit an overwhelming greater tendency to have a case marker on the subject than on the direct object, even though they sometimes use the wrong case form (see also Lakshmanan and Ozeki, 1996; Miyata, 1993). Moreover, as noted in the preceding section, Kanno (1996) reports that the same tendency is strongly manifested in adult second language learners, even when there is no relevant experience or instruction. 5 Conclusion Reduced to its essentials, the study of language is centered on the investigation of two very fundamental questions – why language is the way it is, and how it is acquired. To date, the most detailed answer to these questions has come from proponents of grammatical nativism, who have put forward a theory that simultaneously addresses both questions: Universal Grammar determines the properties that any human language must have and, by virtue of being inborn, it helps explain the success and rapidity of the language acquisition process. A defining feature of UG-based theories is their commitment to hierarchically structured symbolic representations. Not only are the key properties of language defined in terms of these representations, but the mechanisms determining a sentence’s pronunciation and interpretation are thought to make crucial reference to them as well. On this view, then, the end point of the language acquisition process can be seen, in part at least, as the ability to associate such representations with the sentences of one’s language. At the other extreme, recent work in connectionism denies the existence of conventional syntactic representations, of Universal Grammar, and of an inborn acquisition device specifically for language. Language acquisition, it is claimed, is not fundamentally different from any other type of learning and can be accounted for by the same mechanisms as are required for interaction with the environment in general. My own work has been exploring a radical idea of a different sort. As I have characterized it, general (or cognitive) nativism differs from connectionism in being committed to the existence of hierarchically structured symbolic 58 William O’Grady representations as part of a theory of why language is the way it is and to the existence of an inborn acquisition device as part of a theory of how language is acquired. At the same time, it differs from grammatical nativism in not positing inborn categories or principles that are exclusively grammatical in character. Differences as deep as these are unlikely to be resolved immediately, but the challenge is at least clear – we need a viable account both of the properties that define human language and of the acquisition of individual languages on the basis of very limited types of input. There is surely a place for the study of second language acquisition in all of this. At the very least, research on second language learning provides opportunities to observe the acquisition device functioning under conditions of duress – either because of extreme limitations on the available input (as in the case of classroom learning) or because one or more of its component modules have been compromised, or both. It is perhaps not too optimistic to think that the further study of this phenomenon will provide opportunities to extend and deepen our understanding of the acquisition device for human language. NOTES 1 When we add three or more numbers (e.g., 7 + 4 + 8), we always proceed in a pair-wise fashion; no one is able to compute all the numbers in a single step. 2 As predicted, direct object relatives are known to be easier than indirect object relatives, in both first language acquisition (de Villiers, Tager Flusberb, Hakuata, and Cohen, 1979; Hildebrand, 1987) and second language acquisition (Gass, 1979; Wolfe-Quintero, 1992). However, depth of embedding cannot account for the relative preference for preposition stranding over “pied-piping” found in children learning English as a first language (e.g., McDaniel, McKee, and Bernstein, 1998) and, possibly, in second language learners too (White, 1989, pp. 122ff): i Preposition stranding: three intervening nodes: the man who [S you [VP talked [PP to _]]] ii Pied-piping: two intervening nodes: the man to whom [S you [VP talked _]] The obvious explanation for this contrast is simply that the pied-piped structure is all but non-existent in the input. But this raises the question of why English is this way, given the general tendency in human language to avoid preposition stranding. J. Hawkins (1999) makes an interesting proposal in this regard, but space does not permit further discussion of this matter here. 3 The same should be true of whquestions as well, and there do in fact appear to be some languages in which only subjects undergo whmovement (Cheng, 1991). 4 The Minimalist Program still falls well short of being general nativist, however. Chomsky (1995) makes a number of proposals with a strong special nativist character, including a property “P” that permits multiple The Radical Middle 59 nominative patterns in Japanese by allowing a feature to remain active even after being checked and deleted (p. 286) and a parameter that licenses multiple subject constructions in Icelandic by permitting an unforced violation of Procrastinate (p. 375). 5 As illustrated in the syntactic representation for VSO languages, the computational system I adopt permits discontinuous constituents. For extensive discussion, see O’Grady (2001b). REFERENCES Bates, E. and MacWhinney, B. 1988: What is functionalism? Papers and Reports on Child Language Development, 27, 137–52. Bavin, E. 1992: The acquisition of Walpiri. In D. Slobin (ed.), The Crosslinguistic Study of Language Acquisition. Vol. 3. Hillsdale, NJ: Lawrence Erlbaum Associates, 309–71. Bley-Vroman, R. 1994: Updating the Fundamental Difference Hypothesis. Talk presented at the EuroSLA Convention, Aix-en-Provence. Bowerman, M. 1990: Mapping thematic roles onto syntactic functions: are children helped by innate linking rules? Linguistics, 28, 1253–89. Braine, M. 1987: What is learned in acquiring word classes – a step toward an acquisition theory. In B. MacWhinney (ed.), Mechanisms of Language Acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates, 65–87. Caplan, D. and Futter, C. 1986: Assignment of thematic roles to nouns in sentence comprehension by an agrammatic patient. Brain and Language, 27, 117–34. Cheng, L. 1991: On the typology of wh-questions. Unpublished Ph.D. dissertation. MIT. Chomsky, N. 1975: Reflections on Language. New York: Pantheon. Chomsky, N. 1995: The Minimalist Program. Cambridge, MA: MIT. Collins, C. 1994: Economy of derivation and the Generalized Proper Binding Condition. Linguistic Inquiry, 25, 45–61. Croft, W. 1990: Typology and Universals. New York: Cambridge University Press. Croft, W. 1991: Syntactic Categories and Grammatical Relations: The Cognitive Organization of Information. Chicago: University of Chicago Press. Demuth, C. 1992: The acquisition of Sesotho. In D. Slobin (ed.), The Crosslinguistic Study of Language Acquisition. Vol. 3. Hillsdale, NJ: Lawrence Erlbaum Associates, 557–638. Dik, S. 1989: The Theory of Functional Grammar. Part I: The Structure of the Clause. Dordrecht: Foris. Doughty, C. 1991: Second language instruction does make a difference. Studies in Second Language Acquisition, 13, 431–69. Eckman, F. 1996: On evaluating arguments for special nativism in second language acquisition theory. Second Language Research, 12, 398–419. Eckman, F., Bell, L., and Nelson, D. 1988: On the generalization of relative clause instruction in the acquisition of English as a second language. Applied Linguistics, 9, 1–13. Ellis, N. 1996: Sequencing in SLA: phonological memory, chunking and points of order. Studies in Second Language Acquisition, 18, 91–126. 60 William O’Grady Elman, J. 1993: Learning and development in neural networks: the importance of starting small. Cognition, 48, 71–99. Elman, J., Bates, E., Johnson, M., Karmiloff-Smith, A., Parisi, D., and Plunkett, K. 1996: Rethinking Innateness: A Connectionist Perspective on Development. Cambridge, MA: MIT Press. Fortescue, M. and Olsen, L. 1992: The acquisition of West Greenlandic. In D. Slobin (ed.), The Crosslinguistic Study of Language Acquisition. Vol. 3. Hillsdale, NJ: Lawrence Erlbaum Associates, 111–219. Frazier, L. 1987: Sentence processing: a tutorial review. In M. Coltheart (ed.), Attention and Performance XII: The Psychology of Reading. Hillsdale, NJ: Lawrence Erlbaum Associates, 559–86. Fukuda, M. 1993: Head government and case marker drop in Japanese. Linguistic Inquiry, 24, 168–72. Fukui, N. 1996: On the nature of economy in language. Ninti Kagaku [Cognitive Studies], 3, 51–71. Gass, S. 1979: Language transfer and universal grammatical relations. Language Learning, 29, 327– 44. Gass, S. 1980: An investigation of syntactic transfer in adult second language learners. In R. Scarcella and S. Krashen (eds), Research in Second Language Acquisition. Rowley, MA: Newbury House, 132– 41. Gregg, K. 1996: The logical and developmental problems of second language acquisition. In W. Ritchie and T. Bhatia (eds), Handbook of Second Language Acquisition. San Diego: Academic Press, 49–81. Grimshaw, J. and Rosen, S. 1990: Knowledge and obedience: the developmental status of the binding theory. Linguistic Inquiry, 21, 187–222. Grodzinsky, Y. 2000: The neurology of syntax: language use without Broca’s area. Behavioral and Brain Sciences, 23, 1–71. Hawkins, J. 1999: Processing complexity and filler-gap dependencies across grammars. Language, 75, 244–85. Hawkins, R. 1987: Markedness and the acquisition of the English dative alternation by L2 speakers. Second Language Research, 3, 20–55. Hildebrand, J. 1987: The acquisition of preposition stranding. Canadian Journal of Linguistics, 32, 65–85. Imedadze, N. and Tuite, K. 1992: The acquisition of Georgian. In D. Slobin (ed.), The Crosslinguistic Study of Language Acquisition. Vol. 3. Hillsdale, NJ: Lawrence Erlbaum Associates, 39–109. Kanno, K. 1996: The status of a nonparametrized principle in the L2 initial state. Language Acquisition: A Journal of Developmental Linguistics, 5, 317–34. Keenan, E. and Comrie, B. 1977: Noun phrase accessibility and Universal Grammar. Linguistic Inquiry, 8, 63–100. Kellerman, E. and Yoshioka, K. 1999: Inter- and intra-population consistency: a comment on Kanno (1998). Second Language Research, 15, 101–9. Kim, S. 1999: The subject–object asymmetry in the acquisition of wh-questions by Korean learners of English. Paper presented at the Hawaii Language Acquisition Workshop. Honolulu, Hawaii. Kolk, H. and Weijts, M. 1996: Judgments of semantic anomaly in agrammatic patients: argument movement, syntactic complexity, and the use of heuristics. Brain and Language, 54, 86–135. Kuno, S. 1973: The Structure of the Japanese Language. Cambridge, MA: MIT Press. Lakshmanan, U. and Ozeki, M. 1996: The case of the missing particle: objective case assignment and scrambling in the early grammar of Japanese. In The Radical Middle 61 A. Stringfellow, D. Cahana-Amitay, E. Hughes, and A. Zukowski (eds), Proceedings of the 20th Annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press, 431– 42. Langacker, R. 1987: Nouns and verbs. Language, 63, 53–95. Langacker, R. 1995: Raising and transparency. Language, 71, 1–62. Larsen-Freeman, D. and Long, M. 1991: An Introduction to Second Language Acquisition Research. New York: Longman. Levelt, W. 1989: Speaking: From Intention to Articulation. Cambridge, MA: MIT Press. Lightfoot, D. 1982: The Language Lottery. Cambridge, MA: MIT Press. Long, M. 1990: Maturational constraints on language development. Studies in Second Language Acquisition, 12, 251–85. Marcus, M. 1980: A Theory of Syntactic Recognition for Natural Language. Cambridge, MA: MIT Press. Mazurkewich, I. 1984: The acquisition of the dative alternation: unlearning overgeneralizations. Cognition, 16, 261–83. McDaniel, D., McKee, C., and Bernstein, J. 1998: How children’s relatives solve a problem for minimalism. Language, 74, 308–34. Miyata, H. 1993: The performance of the Japanese case particles in children’s speech: with special reference to “ga” and “o.” MITA Working Papers in Psycholinguistics, 3, 117–36. Newmeyer, F. 1998: Language Form and Language Function. Cambridge, MA: MIT Press. O’Grady, W. 1996: Language acquisition without Universal Grammar: a proposal for L2 learning. Second Language Research, 12, 374–97. O’Grady, W. 1997: Syntactic Development. Chicago: University of Chicago Press. O’Grady, W. 1998: The acquisition of syntactic representations: a general nativist approach. In W. Ritchie and T. Bhatia (eds), Handbook of Language Acquisition. San Diego: Academic Press, 157–93. O’Grady, W. 1999: Toward a new nativism. Studies in Second Language Acquisition, 21, 621–33. O’Grady, W. 2001a: Language acquisition and language deficits. Invited keynote address presented to the Japan Society for Language Sciences. O’Grady, W. 2001b: Syntactic computation. Ms. University of Hawai’i, Department of Linguistics. O’Grady, W. and Lee, W. 2001: The Isomorphic Mapping Hypothesis: evidence from Korean. To appear. Brain and Cognition, 46, 226–30. O’Grady, W., Lee, M., and Choo, M. forthcoming: The acquisition of relative clauses in Korean as a second language. Studies in Second Language Acquisition. Osgood, C. and Zehler, A. 1981: Acquisition of bi-transitive sentences: pre-linguistic determinants of language acquisition. Journal of Child Language, 8, 367–84. Pye, C. 1992: The acquisition of K’iche’ Maya. In D. Slobin (ed.), The Crosslinguistic Study of Language Acquisition. Vol. 3. Hillsdale, NJ: Lawrence Erlbaum Associates, 221–309. Roeper, T., Lapointe, S., Bing, J., and Tavakolian, S. 1981: A lexical approach to language acquisition. In S. Tavakolian (ed.), Language Acquisition and Linguistic Theory. Cambridge, MA: MIT Press, 35–58. Russel, R. 1985: An analysis of student errors in the use of Japanese -wa and -ga. Papers in Linguistics, 18, 197–221. Slobin, D. 1985: Crosslinguistic evidence for the language-making capacity. In D. Slobin (ed.), The Crosslinguistic Study of Language Acquisition. Vol. 2. 62 William O’Grady Hillsdale, NJ: Lawrence Erlbaum Associates, 1157–256. Suzuki, T. 1999: Two aspects of Japanese case in acquisition. Unpublished Ph.D. dissertation. University of Hawai’i at Manoa. Talmy, L. 1988: Force dynamics in language and cognition. Cognitive Science, 12, 49–100. Tomasello, M. 1995: Language is not an instinct. Cognitive Development, 10, 131–56. Uziel, S. 1993: Resetting Universal Grammar parameters: evidence from second language acquisition of Subjacency and the Empty Category Principle. Second Language Research, 9, 49–83. de Villiers, J., Tager Flusberb, H., Hakuata, K., and Cohen, M. 1979: Children’s comprehension of relative clauses. Journal of Psycholinguistic Research, 8, 499–528. Waryas, C. and Stremel, K. 1974: On the preferred form of the double object construction. Journal of Psycholinguistic Research, 3, 271–79. Werker, J., Lloyd, V., Pegg, J., and Polka, L. 1996: Putting the baby in the bootstraps: toward a more complete understanding of the role of the input in infant speech processing. In J. Morgan and K. Demuth (eds), Signal to Syntax. Mahwah, NJ: Lawrence Erlbaum Associates, 427–47. White, L. 1987: Markedness and second language acquisition: the question of transfer. Studies in Second Language Acquisition, 9, 261–85. White, L. 1989: Universal Grammar and Second Language Acquisition. Philadelphia: John Benjamins. Wolfe-Quintero, K. 1992: Learnability and the acquisition of extraction in relative clauses and wh questions. Studies in Second Language Acquisition, 14, 39–70. Wolfe-Quintero, K. 1996: Nativism does not equal Universal Grammar. Second Language Research, 12, 335–73. Yoshinaga, N. 1996: Wh-questions: a comparative study of their form and acquisition in English and Japanese. Ph.D. dissertation. University of Hawai’i at Manoa. Constructions, Chunking, and Connectionism 63 4 Constructions, Chunking, and Connectionism: The Emergence of Second Language Structure NICK C. ELLIS 1 Introduction and Overview Constructivist views of language acquisition hold that simple learning mechanisms operating in and across human systems for perception, motor action, and cognition while exposed to language data in a communicatively rich human social environment navigated by an organism eager to exploit the functionality of language are sufficient to drive the emergence of complex language representations. The various tribes of constructivism – that is, connectionists (Christiansen and Chater, 2001; Christiansen, Chater, and Seidenberg, 1999; Levy, Bairaktaris, Bullinaria, and Cairns, 1995; McClelland, Rumelhart, and the PDP Research Group, 1986; Plunkett, 1998), functional linguists (Bates and MacWhinney, 1981; MacWhinney and Bates, 1989), emergentists (Elman, Bates, Johnson, Karmiloff-Smith, Parisi, and Plunkett, 1996; MacWhinney, 1999a), cognitive linguists (Croft and Cruse, 1999; Lakoff, 1987; Langacker, 1987, 1991; Ungerer and Schmid, 1996), constructivist child language researchers (Slobin, 1997; Tomasello, 1992, 1995, 1998a, 2000), applied linguists influenced by chaos/ complexity theory (Larsen-Freeman, 1997), and computational linguists who explore statistical approaches to grammar (Bod, 1998; Jurafsky, 1996) – all share a functional-developmental, usage-based perspective on language. They emphasize the linguistic sign as a set of mappings between phonological forms and conceptual meanings or communicative intentions; thus, their theories of language function, acquisition, and neurobiology attempt to unite speakers, syntax, and semantics, the signifiers and the signifieds. They hold that structural regularities of language emerge from learners’ lifetime analysis of the distributional characteristics of the language input and, thus, that the knowledge 64 Nick C. Ellis of a speaker/hearer cannot be understood as an innate grammar, but rather as a statistical ensemble of language experiences that changes slightly every time a new utterance is processed. Consequently, they analyze language acquisition processes rather than the final state or the language acquisition device (see Sorace, this volume; White, this volume). They work within the broad remit of cognitive science, seeking functional and neurobiological descriptions of the learning processes which, through exposure to representative experience, result in change, development, and the emergence of linguistic representations. Section 2 of this review describes cognitive linguistic theories of construction grammar. These focus on constructions as recurrent patterns of linguistic elements that serve some well-defined linguistic function. These may be at sentence level (such as the imperative, the ditransitive, the yes-no question) or below (the noun phrase, the prepositional phrase, etc.). Whereas GovernmentBinding Theory denied constructions, viewing them as epiphenomena resulting from the interaction of higher-level principles-and-parameters and lower-level lexicon, cognitive linguistics – construction grammar in particular (Croft, 2001; Goldberg, 1995) – has brought them back to the fore, suspecting instead that it is the higher-level systematicities that emerge from the interactions of constructions large and small. Section 3 concerns the development of constructions as complex chunks, as high-level schemata for abstract relations such as transitives, locatives, datives, or passives. An acquisition sequence – from formula, through low-scope pattern, to construction – is proposed as a useful starting point to investigate the emergence of constructions and the ways in which type and token frequency affect the productivity of patterns. Section 4 presents the psychological learning mechanisms which underpin this acquisition sequence. It describes generic associative learning mechanisms such as chunking which, when applied to the stream of language, provide a rich source of knowledge of sequential dependencies ranging from low-level binary chunks like bigrams, through phonotactics, lexis, and collocations, up to formulae and idioms. Although a very basic learning mechanism, chunking results in hierarchical representations and structure dependency. Emergentists believe that many of the rule-like regularities that we see in language emerge from the mutual interactions of the billions of associations that are acquired during language usage. But such hypotheses require testing and formal analysis. Section 5 describes how connectionism provides a means of evaluating the effectiveness of the implementations of these ideas as simulations of language acquisition which are run using computer models consisting of many artificial neurons connected in parallel. Two models of the emergence of linguistic regularity are presented for detailed illustration. Other simulations show how analysis of sequential dependencies results in grammatically useful abstract linguistic representations. The broad scope of connectionist and other distributional approaches to language acquisition is briefly outlined. The review concludes by discussing some limitations of work to date and provides some suggestions for future progress. Constructions, Chunking, and Connectionism 65 2 Construction Grammar This section outlines cognitive linguistic analyses of the interactions between human language, perception, and cognition, and then focuses on construction grammar (Croft, 2001; Fillmore and Kay, 1993; Goldberg, 1995; Langacker, 1987; Tomasello, 1998a, 1998b) as an approach for analyzing the ways in which particular language patterns cue particular processes of interpretation. If words are the atoms of language function, then construction grammar provides the molecular level of analysis. 2.1 Cognitive linguistics Cognitive linguistics (Barlow and Kemmer, 2000; Croft and Cruse, 1999; Goldberg, 1995; Lakoff, 1987; Lakoff and Johnson, 1980; Langacker, 1987, 1991; Talmy, 1988; Ungerer and Schmid, 1996) provides detailed qualitative analyses of the ways in which language is grounded in human experience and in human embodiment, which represents the world in a very particular way. The meaning of the words of a given language, and how they can be used in combination, depends on the perception and categorization of the real world around us. Since we constantly observe and play an active role in this world, we know a great deal about the entities of which it consists, and this experience and familiarity is reflected in the nature of language. Ultimately, everything we know is organized and related in some meaningful way or other, and everything we perceive is affected by our perceptual apparatus and our perceptual history. Language reflects this embodiment and this experience. The different degrees of salience or prominence of elements involved in situations that we wish to describe affect the selection of subject, object, adverbials, and other clause arrangement. Figure/ground segregation and perspective taking, processes of vision and attention, are mirrored in language and have systematic relations with syntactic structure. Thus, paradoxically, a theory of language must properly reflect the ways in which human vision and spatial representations are explored, manipulated, cropped and zoomed, and run in time like movies under attentional and scripted control (Kosslyn, 1983; Talmy, 1996a). In language production, what we express reflects which parts of an event attract our attention; depending on how we direct our attention, we can select and highlight different aspects of the frame, thus arriving at different linguistic expressions. The prominence of particular aspects of the scene and the perspective of the internal observer (i.e., the attentional focus of the speaker and the intended attentional focus of the listener) are key elements in determining regularities of association between elements of visuo-spatial experience and elements of phonological form. In language comprehension, abstract linguistic constructions (like simple locatives, datives, and passives) serve as a “zoom lens” for the listener, guiding their attention to a particular perspective on a scene while backgrounding other aspects (Goldberg, 1995). 66 Nick C. Ellis Thus, cognitive linguistics describes the regularities of syntax as emergent from the cross-modal evidence that is collated during the learner’s lifetime of using and comprehending language. Cognitive linguistics was founded on the principle that language cognition cannot be separated from semantics and the rest of cognition. The next section shows how it similarly denies clear boundaries between the traditional linguistic separations of syntax, lexicon, phonology, and pragmatics. 2.2 Constructions Traditional descriptive grammars focus on constructions, that is, recurrent patterns of linguistic elements that serve some well-defined linguistic function. As noted earlier, these may be at sentence level (such as the imperative, the ditransitive, the yes-no question) or below (the noun phrase, the prepositional phrase, etc.). The following summary of construction grammar, heavily influenced by Langacker (1987) and Croft and Cruse (1999), illustrates the key tenets. A construction is a conventional linguistic unit, that is, part of the linguistic system, accepted as a convention in the speech community, and entrenched as grammatical knowledge in the speaker’s mind. Constructions may (i) be complex, as in [Det Noun], or be simple, as in [Noun] (traditionally viewed as “syntax”); (ii) represent complex structure above the word level, as in [Adj Noun], or below the word level, as in [NounStem-PL] (traditionally viewed as “morphology”); or (c) be schematic, as in [Det Noun], or specific, as in [the United Kingdom], traditionally viewed as “lexicon.” Hence, “morphology,” “syntax,” and “lexicon” are uniformly represented in a construction grammar, unlike both traditional grammar and generative grammar. Constructions are symbolic. In addition to specifying the properties of an utterance’s defining morphological, syntactic, and lexical form, a construction also specifies the semantic, pragmatic, and/or discourse functions that are associated with it. Constructions form a structured inventory of speakers’ knowledge of the conventions of their language (Langacker, 1987, pp. 63–6), usually described by construction grammarians in terms of a semantic network, where schematic constructions can be abstracted over the less schematic ones which are inferred inductively by the speaker in acquisition. This non-modular semantic network representation of grammar is shared by other theories such as Word Grammar (Hudson, 1984, 1990). A construction may provide a partial specification of the structure of an utterance. Hence, an utterance’s structure is specified by a number of distinct constructions. Constructions are independently represented units in a speaker’s mind. Any construction with unique, idiosyncratic formal or functional properties must be represented independently in order to capture speakers’ knowledge of their language. However, absence of any unique property of a construction does not entail that it is not represented independently and simply derived from other, more general or schematic constructions. Frequency of occurrence may lead to independent representation of Constructions, Chunking, and Connectionism 67 even “regular” constructional patterns. This usage-based perspective implies that the acquisition of grammar is the piecemeal learning of many thousands of constructions and the frequency-biased abstraction of regularities within them. Many constructions are based on particular lexical items, ranging from simple (Howzat! in cricket) to complex (Beauty is in the eye of the beholder). The importance of such lexical units or idiomatic phrases is widely acknowledged in SLA research when discussing holophrases (Corder, 1973), prefabricated routines and patterns (Hakuta, 1974), formulaic speech (Wong Fillmore, 1976), memorized sentences and lexicalized stems (Pawley and Syder, 1983), formulae (R. Ellis, 1994), sequences in SLA (N. Ellis, 1996, 2002), discourse management (Dörnyei and Kormos, 1998; Tannen, 1987), register (Biber and Finegan, 1994), style (Brewster, 1999), and lexical patterns and collocational knowledge (Carter, 1998; Hoey, 1991; Lewis, 1993; Schmitt, 2000). According to Nattinger (1980, p. 341), “for a great deal of the time anyway, language production consists of piecing together the ready-made units appropriate for a particular situation and . . . comprehension relies on knowing which of these patterns to predict in these situations.” As Pawley and Syder (1983, p. 192) put it: In the store of familiar collocations there are expressions for a wide range of familiar concepts and speech acts, and the speaker is able to retrieve these as wholes or as automatic chains from the long-term memory; by doing this he minimizes the amount of clause-internal encoding work to be done and frees himself to attend to other tasks in talk-exchange, including the planning of larger units of discourse. But other constructions are more abstract. Goldberg (1995) focuses on complex argument structure constructions such as the ditransitive (Pat faxed Bill the letter), the caused motion (Pat pushed the napkin off the table), and the conative (Sam kicked at Bill). She holds that these abstract and complex constructions themselves carry meaning, independently of the particular words in the sentence. For example, even though the verb kick does not typically imply transfer of possession, it works in the ditransitive Pat kicked Bill the football, and even though one is hard pressed to interpret anything but an intransitive sneeze, the caused motion Pat sneezed the napkin off the table is equally good. These abstract argument structure constructions thus create an important top-down component to the process of linguistic communication. Such influences are powerful mechanisms for the creativity of language, possibly even as manifest in derivational phenomena such as denominal verbs (They tabled the motion) and deverbal nouns (Drinking killed him) (Tomasello, 1998b). Constructions show prototype effects. For example, for ditransitive constructions there is the central sense of agent-successfully-causes-recipient-to-receivepatient (Bill gave/handed/passed/threw/took her a book), and various more peripheral meanings such as future-transfer (Bill bequeathed/allocated/granted/ reserved her a book) and enabling-transfer (Bill allowed/permitted her one book). Prototype effects are fundamental characteristics of category formation, again 68 Nick C. Ellis blurring the boundaries between syntax and lexicon and other cognitive domains (N. Ellis, 2002). 3 Learning Constructions If linguistic systems comprise a conspiracy of constructions, then language acquisition, L1 or L2, is the acquisition of constructions. There is nothing revolutionary in these ideas. Descriptive grammars (e.g., Biber, Johansson, Leech, Conrad, and Finegan, 1999; Quirk, Greenbaum, Leech, and Svartvik, 1985) are traditionally organized around form–function patterns; so are grammars which are designed to inform pedagogy (e.g., Celce-Murcia and Larsen-Freeman, 1983). But what about the processes of acquisition? To date, construction grammar has primarily concerned descriptions of adult competence, although language acquisition researchers, particularly those involved in child language, are now beginning to sketch out theories of the acquisition of constructions which involve a developmental sequence from formula, through low-scope pattern, to construction. 3.1 Formulae and idioms Formulae are lexical chunks which result from memorizing the sequence of frequent collocations. Large stretches of language are adequately described by finite-state grammars, as collocational streams where patterns flow into each other. Sinclair (1991, p. 110), then director of the Cobuild project, the largest lexicographic analysis of the English language to date, summarized this in the principle of idiom: A language user has available to him or her a large number of semi-preconstructed phrases that constitute single choices, even though they might appear to be analyzable into segments. To some extent this may reflect the recurrence of similar situations in human affairs; it may illustrate a natural tendency to economy of effort; or it may be motivated in part by the exigencies of real-time conversation. Rather than its being a somewhat minor feature compared with grammar, Sinclair suggests that, for normal texts, the first mode of analysis to be applied is the idiom principle, as most text is interpretable by this principle. Whereas most of the material that Sinclair was analyzing in the Bank of English was written text, comparisons of written and spoken corpora demonstrate that collocations are even more frequent in spoken language (Biber et al., 1999; Brazil, 1995; Leech, 2000). Parole is flat and Markovian because it is constructed “off the top of one’s head,” and there is no time to work it over. Utterances are constructed as intonation units which have the grammatical form of single clauses, although many others are parts of clauses, and they are often highly predictable in terms of their lexical concordance (Hopper, 1998). Language Constructions, Chunking, and Connectionism 69 reception and production are mediated by learners’ representations of chunks of language: “Suppose that, instead of shaping discourse according to rules, one really pulls old language from memory (particularly old language, with all its words in and everything), and then reshapes it to the current context: “ ‘Context shaping’, as Bateson puts it, ‘is just another term for grammar’ ” (Becker, 1983, p. 218). Even for simple concrete lexis or formulae, acquisition is no unitary phenomenon. It involves the (typically) implicit learning of the sequence of sounds or letters in the word along with separable processes of explicit learning of perceptual reference (N. Ellis, 1994c, 2001). Yet however multifaceted and fascinating is the learning of words (Aitchison, 1987; Bloom, 2000; N. Ellis and Beaton, 1993a, 1993b; Miller, 1991; Ungerer and Schmid, 1996), lexical learning has generally been viewed as a phenomenon that can readily be understood in terms of basic processes of human cognition. Learning the form of formulae is simply the associative learning of sequences. It can readily be understood in terms of the process of chunking which will be described in section 4. The mechanism of learning might be simple, but the product is a rich and diverse population of hundreds of thousands of lexical items and phrases. The store of familiar collocations of the native language speaker is very large indeed. The sheer number of words and their patterns variously explains why language learning takes so long, why it requires exposure to authentic sources, and why there is so much current interest in corpus linguistics in SLA (Biber, Conrad, and Reppen, 1998; Collins Cobuild, 1996; Hunston and Francis, 1996; McEnery and Wilson, 1996). Native-like competence and fluency demand such idiomaticity. 3.2 Limited scope patterns The learning of abstract constructions is more intriguing. It begins with chunking and committing formulae to memory. But there is more. Synthesis precedes analysis. Once a collection of like examples is available in long-term memory, there is scope for implicit processes of analysis of their shared features and for the development of a more abstract summary schema, in the same way as prototypes emerge as the central tendency of other cognitive categories. Consider first the development of slot-and-frame patterns. Braine (1976) proposed that the beginnings of L1 grammar acquisition involve the learning of the position of words in utterances (e.g., More car, More truck, etc. allow induction of the pattern “more + recurring element”). Maratsos (1982) extended this argument to show that adult-like knowledge of syntactic constructions (including both syntactic relations and part-of-speech categories like verb and noun) can also result from positional analysis without the influence of semantic categories like agent and action. He proposed that this learning takes place through the amassing of detailed information about the syntactic handling of particular lexical items, followed by discovery of how distributional privileges 70 Nick C. Ellis transfer among them. The productivity of distributional analyses resultant from connectionist learning of text corpora will be described in section 5. It is important to acknowledge the emphases of such accounts on piecemeal learning of concrete exemplars. Longitudinal child-language acquisition data suggest that, to begin with, each word is treated as a semantic isolate in the sense that the ability to combine it with other words is not accompanied by a parallel ability with semantically related words. An early example was that of Bowerman (1976), who demonstrated that her daughter Eva acquired the more + X construction long before other semantically similar relational words like again and all-gone came to be used in the similar pivot position in two-word utterances. Pine and Lieven (Lieven, Pine, and Dresner Barnes, 1992; Pine and Lieven, 1993, 1997; Pine, Lieven, and Rowland, 1998) have since demonstrated widespread lexical specificity in L1 grammar development. Children’s language between the ages of 2 and 3 years is much more “low-scope” than theories of generative grammar have argued. A high proportion of children’s early multiword speech is produced from a developing set of slot-and-frame patterns. These patterns are often based on chunks of one or two words or phrases and they have “slots” into which the child can place a variety of words, for instance subgroups of nouns or verbs (e.g., I can’t + Verb; where’s + Noun + gone?). Children are very productive with these patterns and both the number of patterns and their structure develop over time. But they are lexically specific. Pine and Lieven’s analyses of recordings of 2–3-year-old children and their mothers measure the overlap between the words used in different slots in different utterances. For example, if a child has two patterns, I can’t + X and I don’t + X, Pine and Lieven measure whether the verbs used in the X slots come from the same group and whether they can use any other CAN- or DOauxiliaries. There is typically very little or no overlap, an observation which supports the conclusion that (i) the patterns are not related through an underlying grammar (i.e., the child does not “know” that can’t and don’t are both auxiliaries or that the words that appear in the patterns all belong to a category of Verb); (ii) there is no evidence for abstract grammatical patterns in the 2–3year-old child’s speech; and (iii) that, in contrast, the children are picking up frequent patterns from what they hear around them, and only slowly making more abstract generalizations as the database of related utterances grows. Tomasello (1992) proposed the Verb Island hypothesis, in which it is the early verbs and relational terms that are the individual islands of organization in young children’s otherwise unorganized grammatical system – in the early stages the child learns about arguments and syntactic markings on a verbby-verb basis, and ordering patterns and morphological markers learned for one verb do not immediately generalize to other verbs. Positional analysis of each verb island requires long-term representations of that verb’s collocations, and, thus, this account of grammar acquisition implies vast amounts of longterm knowledge of word sequences. Only later are syntagmatic categories formed from abstracting regularities from this large dataset in conjunction with morphological marker cues (at least in case-marking languages). Goldberg (1995) Constructions, Chunking, and Connectionism 71 argues that certain patterns are more likely to be made more salient in the input because they relate to certain fundamental perceptual primitives, and, thus, that the child’s construction of grammar involves both the distributional analysis of the language stream and the analysis of contingent perceptual activity: Constructions which correspond to basic sentence types encode as their central senses event types that are basic to human experience . . . that of someone causing something, something moving, something being in a state, someone possessing something, something causing a change of state or location, something undergoing a change of state or location, and something having an effect on someone. (Goldberg, 1995, p. 39) Goldberg and Sethuraman (1999) show how individual “pathbreaking” semantically prototypic verbs form the seed of verb-centered argument structure patterns. Generalizations of the verb-centered instances emerge gradually as the verb-centered categories themselves are analyzed into more abstract argument structure constructions. The verb is a better predictor of sentence meaning than any other word in the sentence. Nevertheless, children ultimately generalize to the level of constructions, because constructions are much better predictors of overall meaning. Although verbs thus predominate in seeding low-scope patterns and eventually more abstract generalizations, Pine et al. (1998) have shown that such islands are not exclusive to verbs, and that the theory should be extended to include limited patterns based on other lexical types such as bound morphemes, auxiliary verbs, and case-marking pronouns. 3.3 Exemplar frequency and construction productivity The research reviewed thus far has focused on piecemeal learning, the emergence of syntactic generalizations, and the elements of language which seed such generalizations. There is another important strand in L1 constructionlearning research that concerns how the frequency of patterns in the input affects acquisition. Usage-based linguistics holds that language use shapes grammar through frequent repetitions of usage, but there are separable effects of token frequency and type frequency. Token frequency is how often in the input particular words or specific phrases appear; type frequency, on the other hand, counts how many different lexical items a certain pattern or construction is applicable to. Type frequency refers to the number of distinct lexical items that can be substituted in a given slot in a construction, whether it is a wordlevel construction for inflection or a syntactic construction specifying the relation among words. The “regular” English past tense -ed has a very high type frequency because it applies to thousands of different types of verbs, whereas the vowel change exemplified in swam and rang has a much lower type frequency. Bybee (Bybee, 1995; Bybee and Thompson, 2000) shows how the productivity of a pattern (phonological, morphological, or syntactic) is a function of its type rather than its token frequency. In contrast, high token 72 Nick C. Ellis frequency promotes the entrenchment or conservation of irregular forms and idioms – the irregular forms only survive because they are very frequent. Type frequency determines productivity because: (i) the more lexical items that are heard in a certain position in a construction, the less likely it is that the construction is associated with a particular lexical item, and the more likely it is that a general category is formed over the items that occur in that position; (ii) the more items the category must cover, the more general are its criterial features, and the more likely it is to extend to new items; and (iii) high type frequency ensures that a construction is used frequently, thus strengthening its representational schema and making it more accessible for further use with new items (Bybee and Thompson, 2000). 3.4 The same sequence for SLA? To what degree might this proposed developmental sequence of syntactic acquisition apply in SLA? SLA is different from L1A in numerous respects, particularly with regard to: i mature conceptual development: a in child language acquisition knowledge of the world and knowledge of language are developing simultaneously whereas adult SLA builds upon pre-existing conceptual knowledge; b adult learners have sophisticated formal operational means of thinking and can treat language as an object of explicit learning, that is, of conscious problem-solving and deduction, to a much greater degree than can children (N. Ellis, 1994a); ii language input: the typical L1 pattern of acquisition results from naturalistic exposure in situations where caregivers naturally scaffold development (Tomasello and Brooks, 1999), whereas classroom environments for second or foreign language teaching can distort the patterns of exposure, of function, of medium, and of social interaction (N. Ellis and Laporte, 1997); iii transfer from L1: adult SLA builds on pre-existing L1 knowledge (MacWhinney, 1992; Odlin, this volume), and, thus, for example, whereas a young child has lexically specific patterns and only later develops knowledge of abstract syntactic categories which guide more creative combinations and insertions into the slots of frames, adults have already acquired knowledge of these categories and their lexical membership for L1, and this knowledge may guide creative combination in their L2 interlanguage to variously good and bad effects. Nevertheless, unless there is evidence to the contrary, it is a reasonable default expectation that naturalistic SLA develops in broadly the same fashion as does L1 – from formulae, through low-scope patterns, to constructions – and that this development similarly reflects the influences of type and token frequencies in the input. (But see Doughty, this volume, for a discussion of how L1 and L2 processing procedures differ.) Constructions, Chunking, and Connectionism 73 There are lamentably few longitudinal acquisition data for SLA that are of sufficient detail to allow the charting of construction growth. Filling this lacuna and performing analyses of SLA which parallel those for L1A described in section 3.2 is an important research priority. But the available evidence does provide support for the assumption that constructions grow from formulae through low-scope patterns to more abstract schema. For a general summary, there are normative descriptions of stages of L2 proficiency that were drawn up in as atheoretical a way as possible by the American Council on the Teaching of Foreign Languages (ACTFL) (Higgs, 1984). These Oral Proficiency Guidelines include the following descriptions of novice and intermediate levels that emphasize the contributions of patterns and formulae to the development of later creativity: Novice Low: Oral production consists of isolated words and perhaps a few highfrequency phrases . . . Novice High: Able to satisfy partially the requirements of basic communicative exchanges by relying heavily on learned utterances but occasionally expanding these through simple recombinations of their elements . . . Intermediate: The intermediate level is characterized by an ability to create with the language by combining and recombining learned elements, though primarily in a reactive mode. (ACTFL, 1986, p. 18) Thus, the ACTFL repeatedly stresses the constructive potential of collocations and chunks of language. This is impressive because the ACTFL guidelines were simply trying to describe SLA as objectively as possible – there was no initial theoretical focus on formulae – yet nonetheless the role of formulae became readily apparent in the acquisition process. There are several relevant case studies of child SLA. Wong Fillmore (1976) presented the first extensive longitudinal study that focused on formulaic language in L2 acquisition. Her subject, Nora, acquired and overused a few formulaic expressions of a new structural type during one period, and then amassed a variety of similar forms during the next. Previously unanalyzed chunks became the foundations for creative construction (see also Vihman’s, 1982, analyses of her young son Virve’s SLA). Such observations of the formulaic beginnings of child L2 acquisition closely parallel those of Pine and Lieven for L1. There are a few studies which focus on these processes in classroom-based SLA. R. Ellis (1984) described how three classroom learners acquired formulae which allowed them to meet their basic communicative needs in an ESL classroom, and how the particular formulae they acquired reflected input frequency – they were those which more often occurred in the social and organizational contexts that arose in the classroom environment. Weinert (1994) showed how English learners’ early production of complex target-like German foreign language negation patterns came through the memorization of complex forms in confined linguistic contexts, and that some of these forms were used as a basis for extension of patterns. Myles, Hooper, and Mitchell (1998; Myles, Mitchell, 74 Nick C. Ellis and Hooper, 1999) describe the first two years of development of interrogatives in a classroom of anglophone French L2 beginners, longitudinally tracking the breakdown of formulaic chunks such as comment t’appelles-tu? (what’s your name?), comment s’appelle-t-il? (what’s his name?), and où habites-tu? (where do you live?), in particular the creative construction of new interrogatives by recombination of their parts, and the ways in which formulae fed the constructive process. Bolander (1989) analyzed the role of chunks in the acquisition of inversion in Swedish by Polish, Finnish, and Spanish immigrants enrolled in a 4-month intensive course in Swedish. In Swedish, the inversion of subject– verb after a sentence-initial non-subject is an obligatory rule. Bolander identified the majority of the inversion cases in her data as being of a chunk-like nature with a stereotyped reading such as det kan man säga (that can one say) and det tycker jag (so think I). Inversion in these sort of clauses is also frequent when the object is omitted as in kan man säga (can one say) and tycker jag (think I), and this pattern was also well integrated in the interlanguage of most of these learners. Bolander showed that the high accuracy on these stereotyped initial-object clauses generalized to produce a higher rate of correctness on clauses with non-stereotyped initial objects than was usual for other types of inversion clause in her data, and took this as evidence that creative language was developing out of familiar formulae. Although there are many reviews which discuss the important role of formula use in SLA (e.g., Hakuta, 1974; Nattinger and DeCarrico, 1992; Towell and Hawkins, 1994; Weinert, 1995; Wray, 1992), there is clearly further need for larger-sampled SLA corpora which will allow detailed analysis of acquisition sequences. De Cock (1998) presents analyses of corpora of language-learner productions using automatic recurrent sequence extractions. These show that second language learners use formulae at least as much as native speakers and at times at significantly higher rates. There is much promise of such computerbased learner corpus studies (Granger, 1998), providing that sufficient care is taken to gather the necessarily intensive longitudinal learner data. There is also need to test the predictions of usage-based theories regarding the influences of type frequency and token frequency as they apply in SLA. 4 Psychological Accounts of Associative Learning This section concerns the psychological learning mechanisms which underpin the acquisition of constructions. Constructivists believe that language is cut of the same cloth as other forms of learning. Although it differs importantly from other knowledge in its specific content and problem space, it is acquired using generic learning mechanisms. The Law of Contiguity, the most basic principle of association, pervades all aspects of the mental representation of language: “Objects once experienced together tend to become associated in the imagination, Constructions, Chunking, and Connectionism 75 so that when any one of them is thought of, the others are likely to be thought of also, in the same order of sequence or coexistence as before” ( James, 1890, p. 561). 4.1 Chunking What’s the next letter in a sentence beginning T . . . ? Native English speakers know it is much more likely to be h or a vowel than it is z or other consonants, and that it could not be q. But they are never taught this. What is the first word in that sentence? We are likely to opt for the, or that, rather than thinks or theosophy. If The . . . begins the sentence, how does it continue? “With an adjective or noun,” might be the reply. And, if the sentences starts with The cat . . . , then what? And then again, how should we complete The cat sat on the . . . ? Fluent native speakers know a tremendous amount about the sequences of language at all grains. We know how letters tend to co-occur (common bigrams, trigrams, and other orthographic regularities). Likewise, we know the phonotactics of our tongue and its phrase structure regularities. We know thousands of concrete collocations, and we know abstract generalizations that derive from them. We have learned to chunk letters, sounds, morphemes, words, phrases, clauses, bits of co-occurring language at all levels. Psycholinguistic experiments show that we are tuned to these regularities in that we process faster and most easily language which accords with the expectations that have come from our unconscious analysis of the serial probabilities in our lifelong history of input (N. Ellis, 2002). Furthermore, we learn these chunks from the very beginnings of learning a second language. N. Ellis, Lee, and Reber (1999) observed people reading their first 64 sentences of a foreign language. While they read, they saw the referent of each sentence, a simple action sequence involving colored geometrical shapes. For example, the sentence miu-ra ko-gi pye-ri lon-da was accompanied by a cartoon showing a square moving onto red circles. A linguistic description of this language might include the following facts: (i) that it is an SOV language; (ii) it has adjective–noun word order; (iii) grammatical number (singular/ plural) agreement is obligatory, and in the form of matching suffix endings of a verb and its subject and of a noun and the adjective that modifies it; (iv) that the 64 sentences are all of the type: [N]Subject [A N]Object V; and (v) that lexis was selected from a very small set of eight words. But such explicit metalinguistic knowledge is not the stuff of early language acquisition. What did the learners make of it? To assess their intake, immediately after seeing each sentence, learners had to repeat as much as they could of it. How did their intake change over time? It gradually improved in all respects. With increasing exposure, performance incremented on diverse measures: the proportion of lexis correctly recalled, correct expression of the adjective–noun agreement, correct subject–verb agreement, totally correct sentence production, correct bigrams and trigrams, and, overall, conformity to the sequential probabilities of the language at letter, word, and phrase level. With other measures it was similarly 76 Nick C. Ellis apparent that there was steady acquisition of form–meaning links and of generalizable grammatical knowledge that allowed success on grammaticality judgment tests which were administered later (Ellis et al., 1999). To greater or lesser degree, these patterns, large and small, were being acquired simultaneously and collaboratively. Acquisition of these sequential patterns is amenable to explanation in terms of psychological theories of chunking. The notion of chunking has been at the core of short-term memory research since Miller (1956) first proposed the term. While the chunk capacity of short-term memory (STM) is fairly constant at 7 ± 2 units, its information capacity can be increased by chunking, a useful representational process in that low-level features that co-occur can be organized together and thence referred to as an individual entity. Chunking underlies superior short-term memory for patterned phone numbers (e.g., 0800-123777) or letter strings (e.g., AGREEMENTS, FAMONUBITY) than for more random sequences (e.g., 4957-632518, CXZDKLWQPM), even though all strings contain the same number of items. We chunk chunks too, so Ellis is wittering on about chunking again is better recalled than again wittering on is about Ellis chunking, and, as shown by Epstein (1967) in a more rigorous but dreary fashion than Lewis Carroll’s, A vapy koobs desaked the citar molently um glox nerfs is more readily read and remembered than koobs vapy the desaked um glox citar nerfs a molently: A chunk is a unit of memory organization, formed by bringing together a set of already formed elements (which, themselves, may be chunks) in memory and welding them together into a larger unit. Chunking implies the ability to build up such structures recursively, thus leading to a hierarchical organization of memory. Chunking appears to be a ubiquitous feature of human memory. (Newell, 1990, p. 7) It operates at concrete and abstract levels, as we shall now see. Sequences that are repeated across learning experiences become better remembered. Hebb (1961) demonstrated that, when people were asked to report back random nine-digit sequences in short-term memory task, if, unbeknownst to the participants, every third list of digits was repeated, memory for the repeated list improved over trials faster than memory for non-repeated lists. This pattern whereby repetitions of particular items in short-term memory result in permanent structural traces has since become known as the Hebb effect. It pervades learning in adulthood and infancy alike. Saffran, Aslin, and Newport (1996) demonstrated that 8-month-old infants exposed for only 2 minutes to unbroken strings of nonsense syllables (for example, bidakupado) are able to detect the difference between three-syllable sequences that appeared as a unit and sequences that also appeared in their learning set but in random order. Chunks that are repeated across learning experiences also become better remembered. In early Project Grammarama experiments, Miller (1958) showed Constructions, Chunking, and Connectionism 77 that learners’ free recall of redundant (grammatical) items was superior to that of random items, and hypothesized that this was because they were “recoding” individual symbols into larger chunks which decreased the absolute number of units. Structural patterns that are repeated across learning experiences as well become better remembered. Reber (1967) showed that memory for grammatical “sentences” generated by a finite-state grammar improved across learning sets. More recent work reviewed by Manza and Reber (1997), Mathews and Roussel (1997), and others in Berry (1997) shows that learners can transfer knowledge from one instantiation to another, that is, learn an artificial grammar instantiated with one letter set (GFBQT) and transfer to strings instantiated in another (HMVRZ), so that if there are many letter strings which illustrate patterned sequences (e.g., GFTQ, GGFTQ, GFQ) in the learning set, the participants show faster learning of a second transfer grammar which mirrors these patterns (HMZR, HHMZR, HMR) than one which does not (HMZR, VMHZZ, VZH). Learners can also demonstrate cross-modal transfer, where the training set might be letters, as above, but the testing set comprises sequences of colors which, unbeknownst to the participant, follow the same underlying grammar. These effects argue for more abstract representations of tacit knowledge. Hebb effects, Miller effects, and Reber effects all reflect the reciprocal interactions between short-term memory and long-term memory (LTM) which allow us to bootstrap our way into language. The “cycle of perception” (Neisser, 1976) is also the “cycle of learning,” such that bottom-up and top-down processes are in constant interaction. Repetition of sequences in phonological STM results in their consolidation in phonological LTM as chunks. The cognitive system that stores long-term memories of phonological sequences is the same system responsible for perception of phonological sequences. Thus, the tuning of phonological LTM to regular sequences allows more ready perception of input which contains regular sequences. Regular sequences are thus perceived as chunks, and, as a result, language- (L1 or L2) experienced individuals’ phonological STM for regular sequences is greater than for irregular ones. This common learning mechanism underpins language acquisition in phonological, orthographic, lexical, and syntactic domains. But this analysis is limited to language form. What about language function? Learning to understand a language involves parsing the speech stream into chunks which reliably mark meaning. The learner does not care about theoretical analyses of language. From a functional perspective, the role of language is to communicate meanings, and the learner wants to acquire the label–meaning relations. Learners’ attention to the evidence to which they are exposed soon demonstrates the recurring chunks of language (to use written examples, in English e follows th more often than x does, the is a common sequence, the [space] is frequent, dog follows the [space] more often than it does book, how do you do? occurs quite often, etc.). At some level of analysis, the patterns refer to meaning. It does not happen at the lower levels: t does not mean anything, nor does th, but the does, and the dog does better, and how do you do? does very well, thank you. In these cases the learner’s goal is satisfied, and the fact that 78 Nick C. Ellis this chunk activates some meaning representations makes this sequence itself more salient in the input stream. When the learner comes upon these chunks again, they tend to stand out as units, and adjacent material is parsed accordingly (see Doughty, this volume, for a detailed discussion of this). What is “meaning” in such an associative analysis? At its most concrete, it is the perceptual memories which underpin the conscious experience which a speaker wishes to describe and which, with luck, will be associated with sufficient strength in the hearer to activate a similar set of perceptual representations. These are the perceptual groundings from which abstract semantics emerge (Barsalou, 1999; Lakoff, 1987). Perceptual representations worth talking about are complex structural descriptions in their own right, with a qualifying hierarchical schematic structure (e.g., a room schema which nests within it a desk schema which in turn nests within it a drawer schema, and so on). These visuostructural descriptions are also acquired by associative chunking mechanisms, operating in a neural system for representing the visual domain. When we describe the structural properties of objects and their interactions we do so from particular perspectives, attending to certain aspects and foregrounding them, sequencing events in particular orders, etc., and so we need procedures for spotlighting and sequencing perceptual memories with language. The most frequent and reliable cross-modal chunks, which structure regular associations between perception and language, are the constructions described in sections 2 and 3. Chunking, the bringing together of a set of already formed chunks in memory and welding them into a larger unit, is a basic associative learning process which can occur in and between all representational systems. 4.2 Generic learning mechanisms Constructivists believe that generic, associative-learning mechanisms underpin all aspects of language acquisition. This is clearly a parsimonious assumption. But additionally, there are good reasons to be skeptical of theories of learning mechanisms specific to the domain of language, first because innate linguistic representations are neurologically implausible, and second because of the logical problem of how any such universals might come into play: i Current theories of brain function, process and development, with their acknowledgement of plasticity and input-determined organization, do not readily allow for the inheritance of structures which might serve, for instance, as principles or parameters of UG (Elman et al., 1996; Quartz and Sejnowski, 1997). ii Whether there are innate linguistic universals or not, there is still a logical problem of syntactic acquisition. Identifying the syntactic category of words must primarily be a matter of learning because the phonological strings associated with words of a language are clearly not universal. Once some identifications have been successfully made, it may be possible to use prior grammatical knowledge to facilitate further identifications. But the