B Lexical & Conceptual Semantics B Cognition Special Issues The titles in this series are paperbacked, readily accessible, in some cases expanded and updated editions of the special issues of COGNITION: An International Journal of Cognitive Science, edited by Jacques Mehler and produced by special agreement with Elsevier Science Publishers B.V. The first six are available from M.l.T. Press. VISUAL COGNITION, Steven Pinker, guest editor THE ONSET OF LITERACY: Cognitive Processes in Reading Acquisition, Paul Bertelson, guest editor SPOKEN WORD RECOGNITION, Uli H. Frauenfelder and Lorraine Komisarjevsky Tyler, guest editors CONNECTIONS AND SYMBOLS, Steven Pinker and Jacques Mehler, guest editors NEUROBIOLOGY OF COGNITION, Peter D. Eimas and Albert M. Galaburda, guest editors ANIMAL COGNITION, C. R. Gallistel, guest editor LEXICAL AND CONCEPTUAL SEMANTICS, Beth Levin and Steven Pinker, guest editors Lexical & Conceptual Semantics Edited by Beth Levin and Steven Pinker BLACKWELL Cambridge MA & Oxford UK Copyright© 1991 Elsevier Science Publishers, B.V., Amsterdam, The Netherlands. This edition published by Blackwell Publishers, 1992 238 Main Street Cambridge, MA 02142 108 Cowley Road Oxford, 0X4 1JF, UK Reprinted from Cognition: International Journal of Cognitive Science, Volume 41, Numbers 1-3, 1991. Blackwell Publishers have exclusive licence to sell this English- language book edition throughout the world. All rights reserved. Except for the quotation of short pasages for the purposes of criticism and review, no part may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the publisher. Library of Congress Cataloging-in-Publication Data A C1P catalog record for this book is available from the Library of Congress, isbn: 1-55786-354-7 British Library Cataloging in Publication Data A C1P catalog record for this book is available from the British Library. Printed and bound in the United States of America. This book is printed on acid-free paper. Contents 1 INTRODUCTION ו Beth Levin and Steven Pinker 2 PARTS AND BOUNDARIES 9 Ray Jackendoff 3 THE SYNTAX OF EVENT STRUCTURE 47 James Pustejovsky 4 LEARNING TO EXPRESS MOTION EVENTS IN ENGLISH AND KOREAN: The influence of language specific lexicalization patterns 83 Soonja Choi and Melissa Bowerman 5 WIPING THE SLATE CLEAN: A lexical semantic exploration ו 23 Beth Levin and Malka Rappaport Hovav 6 AFFECTEDNESS AND DIRECT OBJECTS: The role of lexical semantics in the acquisition of verb argument structure 153 Jess Gropen, Steven Pinker, Michelle Hollander and Richard Goldberg 7 SEMANTIC NETWORKS OF ENGLISH 197 George A. Miller and Christiane Fellbaum LANGUAGE INDEX 231 NAME INDEX 232 SUBJECT INDEX 237 1 Introduction* Beth Levin Department of Linguistics, Northwestern University. Steven Pinker Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology. It is the fate of those who dwell at the lower employments of life, to be rather driven by the fear of evil, than attracted by the prospect of good; to be exposed to censure, without hope of praise; to be disgraced by miscarriage, or punished for neglect, where success would have been without applause, and diligence without reward. Among these unhappy mortals is the writer of dictionaries . . . (Preface, Samuel Johnson’s Dictionary, 1755). Samuel Johnson’s characterization of the lexicographer might apply equally well to the writer of mental dictionaries, those cognitive scientists who attempt to specify the mental representations underlying people’s knowledge of word mean- ings. Research in lexical semantics, though enjoying waves of enthusiasm during the past 30 years, is often regarded as having met with limited success. Although a sense of excitement and progress accompanied the research efforts of the 1960s and early 1970s, including Katz and Fodor’s (1963) semantic feature theory, Fillmore’s (1968) case grammar, and the theory of generative semantics proposed by Lakoff (1971), McCawley (1973, 1979) and Ross (1972), shortly thereafter the research area fell on hard times, meeting a series of rebuffs both from linguists and psycholinguists. Efforts to constrain syntactic theories led some theoretical linguists to condemn the efforts of generative semanticists to construct a syntactic *Supported by NSF Grant BNS-8919884 to the first author, and NIH Grant HD 18381 to the second author. 2 B. Levin and S. Pinker theory in which decompositional representations of word meaning served as the underlying syntactic representation. Meanwhile, Jerry Fodor and his collaborators (Fodor, 1981; Fodor, Fodor, & Garrett, 1975; Fodor, Garrett, Walker, & Parks, 1980) argued that evidence from the psychological laboratory showed that the mental representations of word meaning had no internal structure. Theories of how word meanings are represented in general must be built on research on how particular word meanings are represented. But it is not easy to define a given word, so any attempt to do so becomes an easy target for by now familiar criticisms. If bachelor means “unmarried man”, why is the Pope not a bachelor? If we amend the definition to “unmarried man legally eligible for marriage”, what about a man who has been happily living for 7 years with a woman he has never officially married, or an illegal immigrant who expediently marries a native platonic friend, or a 17-year-old successful entrepreneur living in a penthouse apartment (examples from Winograd, 1976)? If to paint means “cause to be covered with paint”, why isn’t it painting when a paint factory explodes or when Michelangelo dips his brush into the can (Fodor, 1981)? These particular definitions can be patched up, but skeptics foresee a never-ending need for such patching with no real increase in watertightness. The whole enterprise then might seem to be at best tedious and at worst post hoc. Is it really scientifically fruitful to write a 50-page paper on the verb bake? And could there even be an answer to such seemingly academic questions as whether the verb means “to create a cake by cooking in dry heat in an oven” or “to cook by dry heat in an oven, resulting in the creation of a cake?” Inevitably one thinks of Johnson’s entry for lexicographer, which defines the term as “. . . a harmless drudge, that busies himself in . . . detailing the signification of words”, perhaps with doubts about the “harmless” part. As Johnson put it, It appeared that the province allotted me was of all the regions of learning generally confessed to be the least delightful, that it was believed to produce neither fruits nor flowers, and that after a long and laborious cultivation, not even the barren laurel had been found upon it. (Johnson, 1747: 2). Despite the early pessimism, there has been a resurgence of interest in lexical semantics over the last few years in both linguistics and psychology. The new blossoming was caused by several developments, both theoretical and practical. Within theoretical linguistics, it is a response to the increased importance of the lexicon in many current linguistic frameworks (e.g., government-binding, lexical-functional grammar, head-driven phrase structure grammar; see Wasow, 1985). As part of the effort to constrain the power of syntactic rules, more and more facets of syntactic constructions were considered to reflect the properties of the lexical items in these constructions. This shift meant that many linguistic phenomena had to be explained in terms of argument structure - the representa- tion of argument-taking properties of lexical items. And once argument structure began to be used to explain facts of sentence syntax, it became necessary in turn to explain properties of argument structure, leading inexorably to the detailed Introduction 3 examination of the meanings of predicates. The study of lexical semantics no longer divides the field, as it did during the interpretive semantics versus generative semantics debates of the 1970s, but is becoming a unifying focus. Insights regarding word meaning are being compiled eclectically from a variety of linguistic frameworks, current and past, and are incorporated in not too dissimilar ways in most modern linguistic theories. The assumption underlying much of this current linguistic research - that syntactic properties of phrases reflect, in large part, the meanings of the words that head them - also provides a powerful new methodology for studying word meaning. Rather than relying exclusively on intuitions and judgments about aspects of verb meaning, researchers can exploit the fact that subtle differences in word meaning correlate with otherwise puzzling differences in the syntactic structures that the word can appear in. Why can you say Chris cut at the bread but not Chris broke at the bread? The answer, it turns out, depends on the fact that cut is a verb of motion, contact, and causation, while break is a verb of pure causation (Guerssel, Hale, Laughren, Levin, & White Eagle, 1985; Levin, 1985). This implies that motion, contact, and causation must be represented in the meanings of verbs in a format that the syntax can be sensitive to. When the technique of searching for syntax-relevant distinctions is applied to many words and many constructions, a small set of semantic elements tends to recur. Thus evidence from syntactic judgments provides us with a characterization of the scaffolding of semantic structures that verb meanings are built on. Interestingly, the set of elements picked out by this technique is in many instances similar to the set of elements that can be marked overtly by the morphology of some languages, that define the common thread between literal and quasi-metaphorical uses of a given verb, and that are needed to specify the meanings of hundreds or thousands of verbs in English and other languages (Jackendoff, 1990; Miller & Johnson- Laird, 1976; Pinker, 1989; Talmy, 1985). Such convergences increase confidence that the core content of semantic representations is beginning to be identified, and that researchers are not just indulging their intuitions about the best way to define a word. The development within computer science of computational and statistical techniques that can be applied to on-line text corpora and machine-readable dictionaries adds powerful new tools to the toolkit available for the study of lexical representation (e.g., Boguraev, 1991; Boguraev & Briscoe, 1989; Byrd, Calzolari, Chodorow, Klavans, & Neff, 1987; Church, Gale, Hanks, Hindle, & Moon, to appear; Church & Hanks, 1989; Zernik, 1991; among many others). These technologies, by providing access to large amounts of data and allowing for the semi-automatic verification of hypotheses, are already showing great promise, and may soon lead to even more striking results. The study of lexical semantics might also repay the favor to computer science. The development of natural language-understanding systems depends on the availability of large-scale com- prehensive lexicons. Current systems face what has sometimes been called a 4 B. Levin and S. Pinker “lexical bottleneck” (Byrd, 1989) - limitations in system performance attributable to the inadequacy of their lexicons. In the past, the lexicons of natural language- processing systems were created with the technological requirements of a system in mind (especially in terms of the ability to support inference), regardless of their fidelity to the human mental lexicon. But it is hard to believe that such systems would not profit from insights about how the human mind represents word meaning and maps it onto grammar (Levin, 1991; Pustejovsky & Boguraev, to appear). After all, that’s where the words and grammar come from. Psychology, too, cannot afford to do without a theory of lexical semantics. Fodor (1975, 1981; Fodor et al., 1980) points out the harsh but inexorable logic. According to the computational theory of mind, the primitive (nondecomposed) mental symbols are the innate ones. If people know 50,000 word meanings, and if most of these cannot be decomposed into finer-grained elements, then people must have close to 50,000 primitive concepts, and they must be innate. And Fodor, after assessing the contemporary relevant evidence, concluded that most word meanings are not decomposable - therefore, he suggested, we must start living with the implications of this fact for the richness of the innate human conceptual repertoire, including such counterintuitive corollaries as that the concept car is innate. Whether or not one agrees with Fodor’s assessment of the evidence, the importance of understanding the extent to which word meanings decompose cannot be denied, for such investigation provides crucial evidence about the innate stuff out of which concepts are made. Current evidence that there is some linguistically relevant internal structure to verb meaning has provided an intriguing set of candidates for basic conceptual elements, reviewed in Jackendoff (1990) and Pinker (1989). How much of a speaker’s vocabulary can be exhaustively captured in terms of these elements is, of course, an open question. Lexical semantics has also come to play an increasingly central role in the study of language acquisition. Infants do not know the grammar of the particular language community they are born into, but they do have some understanding of the conceptual world that the surrounding language users are expressing. Since concepts are in turn intimately tied to the meanings of words, the child’s semantic machinery might play an important role in allowing him or her to break into the rest of the language system, a hypothesis sometimes called “semantic bootstrap- ping” (see Pinker, 1984). At the same time the semantic representations of particular words, especially verbs, vary from language to language and must themselves by acquired, and the acquisition of verb meaning has become a lively topic in developmental psycholinguistic research (Bowerman, 1989; Clark, 1982; Gentner, 1982; Gleitman, 1990). The impetus for this special issue of Cognition is the revival of interest and research on lexical and conceptual semantics. The issue presents a range of representative recent studies that approach lexical and conceptual semantics from