188 68 2MB
English Pages [502] Year 2010
MEANING AND THE LEXICON
Books by Ray Jackendoff 1972 Semantic Interpretation in Generative Grammar 1977 X-Bar Syntax 1983 A Generative Theory of Tonal Music (co-authored with Fred Lerdahl) Semantics and Cognition 1987 Consciousness and the Computational Mind 1990 Semantic Structures 1992 Languages of the Mind 1994 Patterns in the Mind: Language and Human Nature 1997 The Architecture of the Language Faculty 2002 Foundations of Language: Brain, Meaning, Grammar, Evolution 2005 Simpler Syntax (co-authored with Peter W. Culicover) 2007 Language, Consciousness, Culture: Essays on Mental Structure (The 2003 Jean Nicod Lectures)
MEANING AND THE LEXICON T H E PA R A L L E L ARCHITECTURE 1 9 7 5– 2 0 1 0 Ray Jackendoff
1
3
Great Clarendon Street, Oxford ox2 6dp Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York ß Ray Jackendoff 2010 The moral rights of the authors have been asserted Database right Oxford University Press (maker) First published 2010 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Data available Library of Congress Control Number: 2009934139 Typeset by SPI Publisher Services, Pondicherry, India Printed in Great Britain on acid-free paper by CPI Antony Rowe, Chippenham, Wiltshire ISBN 978-0-19-956888-8 (Pbk.) 978-0-19-956887-1 (Hbk.) 1 3 5 7 9 10 8 6 4 2
Contents Detailed Contents
vii
Abbreviations
xiv
Preface
xv
1
2
3
Prologue: The Parallel Architecture and its Components (2010)
1
Remarks on Chapter 2
35
Morphological and Semantic Regularities in the Lexicon (1975)
40
Remarks on Chapters 3 and 4
85
On Beyond Zebra: The Relation of Linguistic and Visual Information (1987)
88
The Architecture of the Linguistic–Spatial Interface (1996)
112
Remarks on Chapter 5
135
Parts and Boundaries (1991)
138
Remarks on Chapter 6
174
The Proper Treatment of Measuring Out, Telicity, and Perhaps Even Quantification in English (1996)
175
Remarks on Chapters 7–13
222
English Particle Constructions, the Lexicon, and the Autonomy of Syntax (2002)
226
8
Twistin’ the Night Away (1997)
250
9
The English Resultative as a Family of Constructions (2004) Co-authored with Adele E. Goldberg
278
10
On the Phrase The Phrase ‘The Phrase’ (1984)
327
11
Contrastive Focus Reduplication in English (The Salad-Salad Paper) (2004) Co-authored with Jila Ghomeshi, Nicole Rosen, and Kevin Russell
342
4
5
6
7
vi
contents 12
13
Construction after Construction and its Theoretical Challenges (2008)
385
The Ecology of English Noun-Noun Compounds (2009)
413
References
452
Index
473
Detailed Contents Abbreviations
xiv
Preface
xv
1 Prologue: The Parallel Architecture and its Components (2010)
1
The Parallel Architecture The character of interface mappings Outline of Conceptual Semantics The lexicon The syntax–semantics interface Simpler Syntax The issue of semiproductivity
2 5 6 14 20 25 28
1.1 1.2 1.3 1.4 1.5 1.6 1.7
Remarks on Chapter 2 2 Morphological and Semantic Regularities in the Lexicon (1975) Levels of adequacy in description Formulation of two preliminary theories Which theory? Separate morphological and semantic rules Other applications 2.5.1 Prefix-stem verbs 2.5.2 Compound nouns 2.5.3 Causative verbs 2.5.4 Idioms 2.6 The cost of referring to redundancy rules 2.7 Creativity in the lexicon and its implications 2.8 Summary
2.1 2.2 2.3 2.4 2.5
35
40 41 42 49 55 59 59 63 67 72 75 81 82
viii
d et a i l ed c o n te n ts
Remarks on Chapters 3 and 4 3 On Beyond Zebra: The Relation of Linguistic and Visual Information (1987) 3.1 3.2 3.3 3.4 3.5 3.6 3.7
Introduction Properties of the 3D model Outline of Conceptual Semantics Preliminary points of correspondence The use of 3D models in word meanings Enriching the conceptual-3D connection Summary and methodological remarks
4 The Architecture of the Linguistic–Spatial Interface (1996) 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10
Introduction Representational modularity The character of interface mappings Conceptual structure Spatial representation The interface between CS and SR A simple case: the count–mass distinction Axes and frames of reference The lexical encoding of axial vocabulary Final thoughts
Remarks on Chapter 5 5 Parts and Boundaries (1991) The framework The technology of Conceptual Semantics The puzzle and a preliminary solution The features b(ounded) and i(nternal structure) Functions that map between values of b and i 5.5.1 PL (plural) 5.5.2 ELT (element of) 5.5.3 COMP (composed of) 5.5.4 GR (the grinder) 5.5.5 PART (part of) 5.5.6 CONT (containing) 5.6 Dimensionality and directionality 5.7 Boundaries
5.1 5.2 5.3 5.4 5.5
85
88 88 92 96 96 99 101 107
112 112 113 114 115 116 119 122 123 128 134
135 138 138 139 139 143 145 145 147 148 151 153 154 155 159
d e t a il e d c ont e nts
ix
5.8 Using the formalism 5.8.1 Paths 5.8.2 Aspectual functions 5.8.3 The ‘Vendler classes’ 5.8.4 Until and since 5.9 Final remarks
163 163 165 166 169 171
Remarks on Chapter 6
174
6 The Proper Treatment of Measuring Out, Telicity, and Perhaps Even Quantification in English (1996) 6.1 Introduction 6.2 More data and intuitions on measuring out 6.3 Problems with Tenny’s and Dowty’s theories 6.3.1 Tenny 6.3.2 Dowty 6.3.3 Verkuyl and Krifka 6.4 From snapshots to motion 6.5 A formalism for measuring out 6.5.1 Decomposing objects into axis plus cross-section 6.5.2 The cross-section of an event of motion 6.5.3 Projecting the cross-section of an event onto axes 6.5.4 GO reduces to BE plus sp-bound axes 6.6 Telicity with motion verbs 6.7 More applications of sp-binding 6.7.1 Events vs. States 6.7.2 Staying 6.7.3 Extending, covering, and filling 6.7.4 Verbs of change of possession 6.7.5 Changes of properties 6.7.6 Verbs of performance 6.7.7 Verbs of creation and consumption 6.7.8 Conclusions 6.8 Sp-binding on discrete axes; distributive quantification 6.9 Multidimensional measuring out 6.10 More on spray/load verbs 6.11 Conclusions
175 175 176 180 180 183 184 185 187 187 188 190 191 192 195 196 196 197 198 199 200 201 203 204 208 213 217
x
d et a i l ed c o n te n ts 6.A Appendix: A more formal treatment of structure-preserving binding 6.A.1 Strong and weak sp-binding 6.A.2 Continuity 6.A.3 Generalization of continuity to covering and filling 6.A.4 Definition of bounding on an axis
218 218 219 220 220
Remarks on Chapters 7–13
222
7 English Particle Constructions, the Lexicon, and the Autonomy of Syntax (2002) The agenda Linguistics 101 facts about English particles Idiomatic verb-particle combinations Directional particles Aspectual particles The Time-Away Construction Idioms consisting of particles plus something other than the Verb 7.7.1 V-d out 7.7.2 His head off 7.8 The syntactic autonomy of the particle construction 7.1 7.2 7.3 7.4 7.5 7.6 7.7
8 Twistin’ the Night Away (1997) Syntactic properties Semantic properties The aspectual particle away More on aspectual particles Review of the resultative and way constructions Comparison of the three constructions 8.6.1 Resultative versus way construction 8.6.2 Time-away versus way construction 8.6.3 Time-away versus resultative construction 8.7 Still another family of related constructions/idioms 8.8 Two accounts
8.1 8.2 8.3 8.4 8.5 8.6
9 The English Resultative as a Family of Constructions (2004) Co-authored with Adele E. Goldberg 9.1 A constructional view of grammar 9.2 Dimensions of variation in resultatives
226 226 228 232 234 236 240 242 242 243 245
250 251 254 257 258 261 264 264 265 268 270 270
278 278 283
de t ail e d c ont e nts 9.3 The semantics of the resultative 9.3.1 Property vs. Path resultatives 9.3.2 Noncausative vs. causative resultatives 9.3.3 Soundþmotion and disappearance resultatives 9.4 The meaning of resultatives explains their aspectual properties 9.4.1 Telic, atelic, and stative resultatives 9.4.2 Temporal relation of the two subevents 9.5 World knowledge relating to the semantics of the resultative explains additional distributional facts 9.6 How arguments are shared between the two subevents 9.6.1 Full Argument Realization 9.6.2 The Semantic Coherence Principle 9.7 Extending the analysis to additional types of resultatives 9.7.1 Follow cases 9.7.2 Dancing mazurkas 9.7.3 Spit cases 9.7.4 Rappaport Hovav and Levin’s approach 9.8 On the productivity of resultative constructions 9.9 Conclusions
xi 286 288 289 290 293 293 296 298 300 300 304 308 308 313 314 315 316 324
10 On the Phrase The Phrase ‘The Phrase’ (1984)
327
Introducing the construction Semantic and pragmatic constraints E is not an appositive Attachment of E to N’ or N’’ E is not the head Conclusion
327 329 330 334 337 339
10.1 10.2 10.3 10.4 10.5 10.6
11 Contrastive Focus Reduplication in English (The Salad-Salad Paper) (2004) Co-authored with Jila Ghomeshi, Nicole Rosen, and Kevin Russell 11.1 Introduction 11.2 The semantics of CR 11.2.1 Specifying the interpretation 11.2.2 CR in other languages 11.3 The scope of CR 11.3.1 The problem 11.3.2 Scope of CR smaller than a word 11.3.3 CR and object pronouns
342
342 346 346 352 355 355 356 359
xii
d et a i l ed c o n te n ts 11.3.4 The generalization 11.3.5 Prosodic constraints on CR 11.4 An analysis of CR in the parallel architecture framework 11.4.1 Basics of the parallel architecture 11.4.2 A treatment of reduplication in the parallel architecture 11.4.3 Formulating CR 11.5 A Minimalist Program approach to CR 11.6 Final remarks
12 Construction after Construction and its Theoretical Challenges (2008) 12.1 Basic facts 12.2 The productive subconstructions of NPN 12.2.1 N by N 12.2.2 N for N 12.2.3 N to N 12.2.3.1 Close contact or juxtaposition 12.2.3.2 Succession 12.2.3.3 Single transition 12.2.3.4 Comparison/juxtaposition with differing nouns 12.2.4 N after N and N upon N 12.3 The place of NPN in the lexicon and grammar 12.4 Syntactic puzzles posed by NPN 12.4.1 What syntactic category is NPN? 12.4.2 What is the head of NPN, and what is the rest? 12.4.3 Prenominal adjectives 12.4.4 Triplication 12.4.5 Complements and postnominal modifiers to NPN in NP position 12.4.6 Comparison with one N after another and one N at a time 12.5 N after N as a quantifier 12.6 A lexical entry for the NPN construction 12.7 Inconclusion
362 366 368 369 372 373 378 382
385 385 387 388 388 389 389 391 392 393 393 394 399 399 401 402 403 404 405 407 410 411
13 The Ecology of English Noun-Noun Compounds (2009)
413
13.1 Compounds: On the cusp between grammar and lexicon 13.2 Compounds as an evolutionary throwback
413 421
de t ail e d c ont e nts 13.3 Preliminaries to semantic analysis of English N-N compounds 13.4 Aspects of compound meaning that come from semantics of nominals 13.4.1 Profiling 13.4.2 Action modality 13.4.3 Cocomposition 13.5 Semantic structure of (relatively) simple compounds 13.5.1 The compounding schemata 13.5.2 Reversibility of basic functions 13.5.3 Fourteen basic functions 13.6 Using material from the meanings of N1 and N2 13.7 Generative schemata for F 13.8 Closing remarks
xiii
425 429 429 430 432 434 434 435 436 442 445 448
References
452
Index
473
Abbreviations ASL
American Sign Language
CR
contrastive reduplication
CS
conceptual structure
GB
Government and Binding Theory
GF
grammatical function
GPSG
Generalized Phrase Structure Grammar
HPSG
Head-Driven Phrase Structure Grammar
LCS
lexical conceptual structure
LF
Logical Form
LFG
Lexical Functional Grammar
MP
Minimalist Program
PF
Phonetic Form
RP
resultative phrase
S&C
Semantics and Cognition
SPE
Chomsky and Halle, The Sound Pattern of English
SR
spatial representation
UG
Universal Grammar
Preface Curating a retrospective of one’s own papers is inevitably a bit of a selfindulgence. I’m doing so at the suggestion of a number of friends and colleagues who thought that such a collection, amplified by commentary, might serve as a useful journey through the evolution of my thinking over the past thirty-five years. The papers offered here focus on the theory of meaning and on the theory of the lexicon—two issues which, as the years go on, have come to be more and more intertwined. How are the meanings of words formulated, such that they are sufficiently rich to do the work of inference and reference that they are supposed to? How does linguistic meaning interface with visual meaning, such that we can talk about what we see? How are those parts of meaning formulated that don’t come from words? How can it be that the same meanings come sometimes from words and sometimes from other sources? The latter two questions have led to my growing interest in coercions and constructions, in which grammar, lexicon, and pragmatic factors intersect. A further recurring theme among the papers is productivity versus semiproductivity. Standard generative grammars expect us to find complete productivity; yet language is full of semiproductive phenomena, in both morphology (as is well known) and syntax (as is less widely acknowledged). Semiproductivity is still problematic, and it too leads into the structure of the lexicon. Many other papers of mine could have been included in this volume, except that they had already been folded into my books Semantics and Cognition (1983), Semantic Structures (1990), Languages of the Mind (1992), The Architecture of the Language Faculty (1997), Foundations of Language (2002), and Simpler Syntax (2005). (I’ve decided to include ‘On Beyond Zebra’, despite its appearing in Consciousness and the Computational Mind (1987), because linguists rarely engage with that book.) So the picture presented by this volume alone is necessarily somewhat fragmentary. In order to help tie the papers together, I have provided a Prologue that contextualizes the papers in terms of my current framework—the Parallel Architecture and its major components, Conceptual Semantics and Simpler Syntax. Chapter 2 is about the structure of the lexicon. Chapters 3 and 4 concern the relation between language and vision. Chapters 5 and 6 deal with
xvi
preface
the intertwined issues of mereology (parts) and aktionsarten (event structure). Chapters 7–9 turn to special constructions in the English VP: verb-particle constructions, the time-away construction, and the resultative. Chapters 10–12 deal with three less ruly phenomena of English: a quotative NP construction, a reduplicative construction with curious semantics and morphosyntax, and the ubiquitous NPN construction (day after day). Finally Chapter 13 is about the semantics of noun-noun compounds. Many of the papers are preceded by remarks that place them in their time and/or set out unifying themes. Throughout, but especially in the earlier papers, I have also inserted commentaries as ‘subfootnotes’, keyed by superscript letters and printed in sans serif font, which discuss what I think was wrong, how I might now fix it, further formalization, subsequent reactions in the literature, and so on. I have tried to leave the original texts of the articles intact, insofar as it made sense to. I have deleted, abridged, or moved a few sections in the interests of reducing duplication, and I have thoroughly rewritten a couple of very thorny (or muddled?) sections in the interests of comprehensibility. In both cases I give notice in the text. I have kept section and example numbering intact from the originals, even at the expense of some gaps. Short comments have been inserted in square brackets; I have clarified a few passages and updated some references without notice. Each article contains its own acknowledgments. But in addition, I must express my gratitude to Peter Culicover, Maria Pin˜ango, Lila Gleitman, Andrew Winnard, Gina Kuperberg, and my editor John Davey for their encouragement and good advice on the project as a whole. Many thanks to Adele Goldberg, Jila Ghomeshi, Nicole Rosen, and Kevin Russell, my collaborators in Chapters 9 and 11. I deeply appreciate their willingness to cross theoretical divides; I’m sure there were times when I tried their patience. Teresa Salvato has been of great help in converting the older papers into electronic format, as well as in the more general task of making my life at the Tufts Center for Cognitive Studies go smoothly. I’m indebted to Neil Cohn for redrawing the illustrations, which in some of the original papers were embarrassingly crude drawings of my own. I’m grateful to Dan Dennett, Jamshed Bharucha, Kevin Dunn, and Ed Merrin, who made my move to Tufts University in 2005 possible, and to them and Maryanne Wolf, Phil Holcomb, Gina Kuperberg, Mark Richard, Nancy Bauer, Diane Souvaine, and John McDonald, among many others, who have made it such a pleasure to be there. Thanks to the generous efforts of Chris Wood, I was able to complete this project in residence at the fabulous Santa Fe Institute. And, at the end of the day, there’s Hildy to thank profusely, as always. R.J.
chapter 1
Prologue The Parallel Architecture and its Components (2010)
Each of the chapters in this volume has played a significant role in developing the framework I have called the Parallel Architecture. The basic premise of the theory is that phonology, syntax, and semantics are independent generative components in language, each with its own primitives and principles of combination. It contrasts with the mainstream assumption that the syntactic component is the sole generative component of language, responsible for the combinatorial properties of phonology and semantics. In the earlier papers represented here, many divergences from mainstream assumptions are proposed rather casually; I thought of them as minor fixes without great import. However, during the middle 1990s it slowly became clear to me that I had come to think about the architecture of language in a fashion very much at odds with the mainstream conception, and I began to articulate and explore the differences systematically in The Architecture of the Language Faculty (1997) and especially Foundations of Language (2002). The Parallel Architecture has three important subcomponents. The first, Conceptual Semantics, is an approach to meaning that I have been developing steadily since Semantic Interpretation in Generative Grammar (1972), which was a much expanded version of my 1969 dissertation. The first full working out of this component is in Semantics and Cognition (1983), and it is further amplified in Semantic Structures (1990). The second subcomponent of the Parallel Architecture is the lexicon, which comes to include not just words and morphemes, but also idioms, memorized fixed expressions such as cliche´s, constructions, and even the most general rules
2
pr olo g ue
of grammar. This treatment began to emerge in Semantic Structures, but Foundations of Language really fleshed it out and showed how it leads to a considerable reframing of the problem of linguistic knowledge and language acquisition. The third subcomponent of the Parallel Architecture is a greatly simplified theory of syntax and a somewhat more complex theory of the syntax–semantics interface. Again, this was developing implicitly in my work for many years, but a first explicit statement was in Semantic Structures, and a second, far more ambitious statement appeared in Simpler Syntax (2005), co-authored with Peter Culicover. This prologue offers a brief discussion of these theoretical developments, in part consolidating discussions from the original versions of the chapters in this volume, and in part bringing the discussion a bit more up to date.
1.1 The Parallel Architecture The Parallel Architecture builds on insights about phonology and semantics that arose in the middle and late 1970s. Phonology was demonstrated to have highly articulated structure that cannot be derived directly from syntax. In particular, structured units such as syllables and prosodic constituents do not correspond one to one with syntactic units. Moreover, phonological structure includes several independent substructures or tiers, each with its own type of generative structure: segmental-syllabic structure, the metrical grid, intonation contour, and (in tone languages) the tone tier. The tiers are correlated with each other by interface rules: principles that establish optimal correspondence between structures of two independent types. For instance, stress rules are treated not as rules that derive stress patterns de novo, but as interface rules that create the best match between segmental-syllabic patterns and an optimal metrical grid. In turn, since these multitier phonological structures cannot be derived from syntactic structures, the connection between syntax and phonology has to be mediated not by derivations, but by a component of interface rules. The 1970s also witnessed the development of approaches to meaning as different as formal semantics and Cognitive Grammar, with my Conceptual Semantics somewhere in between. Despite their differences, all of these approaches see meaning structures as combinatorial and potentially unlimited in complexity. These structures are not built out of or derived from syntactic units such as NPs and VPs (as they were in Generative Semantics). Rather, they are built of characteristic semantic units such as conceptualized objects, events, times, properties, and quantifiers, which do not correspond one to one with syntactic
prologue
3
categories. As a consequence, semantics too becomes a generative component of language that cannot be derived from syntax, but rather is correlated with syntax by a component of interface rules. Moreover, semantic structure, like phonology, demonstrably has an articulation into tiers, including at least propositional structure (who did what to whom) and an orthogonal dimension of information structure (topic/focus/common ground; old vs. new information). Syntax too may be articulated into independent tiers. LFG (Bresnan 1982b, 2001) is well known for proposing an articulation of syntax into constituent structure (the standard tree) and functional structure (grammatical functions); a stripped-down version of the latter is the grammatical function tier of Culicover and Jackendoff 2005. Autolexical Syntax (Sadock 1991) and Role and Reference Grammar (Van Valin and LaPolla 1997) propose an articulation into phrasal and morphosyntactic tiers—essentially principles that operate above and below the word level respectively. Again, the tiers of syntax must be brought into registration by interface rules. The result is an architecture of the form in Figure 1.1, where phonology, syntax, and semantics also have further internal articulation of a similar sort. Figure 1.1 offers a model in terms of which various other theories can be compared. For instance, in mainstream generative grammar (Chomsky 1965, 1981, 1995), the combinatorial properties of phonology and semantics are derived from syntax; hence there are no independent formation rules for phonology and semantics, only derivational interfaces from syntax into these two components. By contrast, Cognitive Grammar (e.g. Langacker 1987a) claims that all (or at least most) syntactic structure is semantically motivated, so it eliminates or minimizes the syntactic formation rules. Within the Parallel
Phonological Formation Rules
Syntactic Formation Rules
Semantic Formation Rules
Phonological Structures
Syntactic Structures
Semantic Structures
Interface
Interface Interface
Fig. 1.1 The Parallel Architecture
4
pr olo g ue
Architecture, the empirical issue is the proper balance between these two extremes.1 An important constraint on the balance comes from the fact that semantic structure ultimately has to be rich enough to support inference. In the mainstream architecture, where semantic structure is derived from syntactic structure, all this richness has to come from syntax. This puts syntactic theory under constant pressure for greater articulation and complexity. By contrast, the Parallel Architecture grants semantics its own generative capacity, and therefore syntax has to be only rich enough to modulate the mapping between semantics and phonology—a Simpler Syntax (Culicover and Jackendoff 2005). A crucial advantage of the Parallel Architecture is that it extends naturally to the relation of language to other capacities such as visual perception. It ought to be self-evident that the combinatorial structure of the visual field cannot be derived from syntactic NPs and VPs. Nevertheless, speakers refer freely to the structure of the world as understood through vision, e.g. in reporting that The cat is on the mat. Following the overall conception of the Parallel Architecture, the relation between language and vision can be treated as yet another interface component, this time linking the semantic structure of language to the combinatorial structures responsible for visual understanding. This interface is the topic of Chapters 3 and 4. The conclusion is that the relationships among faculties such as language and vision are mediated by exactly the same sorts of formal components—interface rules—as the relationships among the components of language and, within them, the subcomponents or tiers. In other words, the Parallel Architecture is applicable to all scales of mental organization.2 1
Notice that Figure 1.1 contains an interface going directly from phonology to meaning, bypassing syntax. Such an interface is of course impossible under mainstream assumptions, where everything is derived from syntax. However, it provides a way to directly encode the influence of stress and intonation on meaning, without the involvement of syntax, and it is possible that other linguistic phenomena make use of this route as well. See Jackendoff 2002, chapter 8, and Chapter 13 of the present volume. 2 Detailed comparisons of the Parallel Architecture with mainstream generative grammar appear in Jackendoff 1997b, 2002 (especially chapters 5–8), 2007b, 2008; Culicover and Jackendoff 2005. To my knowledge, discussion of the Parallel Architecture by proponents of the Minimalist Program (often without citation, but see, e.g., Phillips and Lau 2004, Marantz 2005) has been limited to the charge that the parallel architecture requires three or more ‘generative engines’ plus the interface components, and so it is a priori less economical and elegant than the traditional architecture, which requires only syntax as its ‘generative engine’. A first reply to this criticism is that the Minimalist Program offers no formal account of either phonology or semantics, nor of the so-called ‘sensorimotor and conceptual-intentional interfaces’, so it is no wonder that it needs only a syntactic engine. A second reply, echoing Chomsky (1972b) (who is attacking Postal’s (1972) ‘The Best Theory’ without actually citing it), is that formal elegance is not the only criterion for an empirical theory: it must first of all account for the facts and generalizations of language. As shown by Culicover and Jackendoff 2005 (chapters 2 and 3), the pursuit of theoretical elegance in mainstream generative grammar has led to continual shrinkage in its descriptive coverage.
prologue
5
1.2 The character of interface mappings To see what interface mappings are like, let us focus for a moment on the interface between phonology and syntax, the two best-understood levels of mental representation. It is obvious that there cannot be a complete translation between phonology and syntax. Many details of phonology, most notably the segmental content of words, play no role at all in syntax. Conversely, many details of syntax, for instance the layering of specifiers and of arguments and adjuncts, are not reflected in phonology. In fact, a complete, information-preserving translation between the two representations would be pointless, since it would in effect make them notational variants—which they clearly are not. The relation between phonology and syntax is actually something more like a partial homomorphism. The two representations share the notion of word (and perhaps morpheme), and they share the linear order of words and morphemes.3 But segmental and stress information in phonology has no direct counterpart in syntax; and syntactic category (N, V, PP, etc.) and case, number, gender, and person features have no direct phonological counterparts.4 Moreover, syntactic and phonological constituent structures often fail to match. A classic example is given in (1). (1) Phonological: [This is the cat] [that ate the rat] [that ate the cheese] Syntactic: [This is [the cat [that ate [the rat [that ate [the cheese] ] ] ] ] ] The phonological bracketing, a flat tripartite structure, contrasts with the relentlessly right-embedded syntactic structure. At a smaller scale, English articles cliticize phonologically to the following word, resulting in bracketing mismatches such as (2). (2) Phonological: [the [big] ] [house] Syntactic: [the [big [house] ] ] Thus in general, the phonology–syntax interface module creates only partial correspondences between these two levels. A similar situation obtains with the interface between auditory information and phonological structure. The complex mapping between wave-forms and 3
Caveats are necessary concerning nonconcatenative morphology such as reduplication and Semitic inflection, where the relation between linear order in phonology and syntax is unclear, to say the least. See Jackendoff 1997b, chapter 6 on nonconcatenative morphology and Chapter 11 of the present volume for reduplication. 4 To be sure, syntactic features are frequently realized phonologically as affixes with segmental content. But the phonology itself has no knowledge of what syntactic features these affixes express. See Jackendoff 1997, chapters 5 and 6.
6
pr olo g ue
phonetic segmentation in a sense preserves the relative order of information: a particular auditory cue may provide evidence for a number of adjacent phonetic segments, and a particular phonetic segment may be signaled by a number of adjacent auditory cues, but the overlapping ‘bands’ of correspondence progress through the speech stream in an orderly linear fashion. On the other hand, boundaries between words, omnipresent in phonological structure, are not reliably detectable in the auditory signal; contrariwise, the auditory signal contains acoustic information about the identity of the speaker that is disregarded by phonology. So again the interface module takes only certain information from each representation into account in establishing a correspondence between them. In addition to general mapping principles such as order preservation, an interface can also make use of specialized learned mappings. The clearest instances of such mappings are lexical items. For instance, the lexical item cat stipulates that the phonological structure /kæt/ can be mapped simultaneously into a syntactic noun and into a conceptual structure that encodes the word’s meaning. In other words, the Parallel Architecture leads us to regard the lexicon as a learned component of the interface modules within the language faculty (see Jackendoff 1997b, 2002). We return to the lexicon in section 1.4. Both these examples—syntax to prosody and segmental phonology to acoustics—show that each level of representation has its own proprietary information, and that an interface communicates only certain aspects of this information to the next level up- or downstream. As will be seen, the same general characteristics obtain in other interfaces, most notably the syntax– semantics interface, which is so crucial to explicating the nature of human language. I put off discussion of this interface to section 1.5. These same characteristics extend to the interface between language and vision, discussed in Chapters 3 and 4, and to the interfaces within the musical faculty (Lerdahl and Jackendoff 1983). To me, the fact that similar characteristics emerge in case after case over multiple faculties of mind is strong evidence in favor of this conception of language.
1.3 Outline of Conceptual Semantics One factor that made it possible to work out a detailed statement of the Parallel Architecture and the lexicon was the theory of Conceptual Semantics, which, unlike other extant semantic theories, was developed in full cognizance of the overall goals of generative grammar. Receiving its first full statement in Semantics and Cognition (S&C), Conceptual Semantics answers to a number of
prologue
7
constraints not normally considered central by theories within philosophy, psychology, and AI. Among the leading points of this theory are the following: 1. Meanings are mentally encoded. This is the foundational assumption of Conceptual Semantics, necessary to embed the theory within the psychologically based framework of generative grammar. The theory is concerned not just with meaning tout court, but with the speaker’s grasp of meaning. This premise decisively parts company with standard model-theoretic approaches, in which meanings are taken to be sets of individuals in possible worlds, extrinsic to the language user rather than encoded in the user’s mind. To adapt the terms of Chomsky 1986, Conceptual Semantics is a theory of I-semantics (‘internal(-ized) semantics’) rather than E-semantics (‘external(-ized)’). Conceptual Semantics encodes meanings in terms of a level of representation called conceptual structure (CS). Conceptual structure is taken to be the form (or one of the forms) in which human thought is couched—it serves as the ‘syntax of thought’.5 It is by hypothesis common to all natural languages: translation preserves conceptual structure. Moreover, CS is to some degree independent of language, and some aspects of it are shared with nonlinguistic organisms such as apes and babies. The language faculty—phonology, syntax, and the interfaces of Figure 1.1—can then be seen as serving the purpose of making thoughts communicable (including to oneself, through verbal imagery).6 2. The reference of linguistic expressions is a mental construct. Standard formal semantics and philosophy of language seek to explicate the relation of reference between language and the external world, usually modeling the world set-theoretically and often using the notion of possible worlds. In Conceptual Semantics, the relation between language and the external world is taken to be mediated by the way the mind understands the world, as encoded in mental representations. Thus this theory contains no notion of reference in the standard sense. Rather, the corresponding construct is reference to the world as the mind construes it, or how the speaker is at the moment inviting the hearer to view the
5
I use this term to distinguish my notion of conceptual structure from Fodor’s (1975) ‘Language of Thought’; the latter carries with it the property of intentionality, from which I wish to distance myself. See Jackendoff (1990, 1991b, 2002) for discussion. 6 In recent writings, Chomsky (e.g. Hauser, Chomsky, and Fitch 2002) has put forward the position that the basic function of language (if one may use the term) is to structure thought, and that ‘externalization’ via phonology is secondary. This fails to explain why the experience of one’s own thought is through verbal imagery, complete with phonological structure. It also fails to explain how it is possible for nonlinguistic organisms such as chimpanzees and baboons to think (Köhler 1927, Cheney and Seyfarth 2007). See Pinker and Jackendoff 2005 for discussion.
8
pr olo g ue
world. This difference plays an important role in many of the chapters in this volume, where semantic analyses prove to depend crucially on what construals of objects and events are possible and salient. 3. Meanings are decompositional. That is, word and phrase meanings have internal structure built up ‘algebraically’ from a finite, innate stock of primitives and principles of combination. The argument for this point (Jackendoff 1983, chapters 5 and 7; 1990; 2002) parallels Chomsky’s (1957) arguments concerning the generativity of syntax: we are capable of grasping an unlimited number of meanings, both for words and for phrases. Since they cannot all be memorized or hard-wired, they must be the product of a finite generative system. This position concurs with the main traditions in lexical semantics (Miller and Johnson-Laird 1976; Lehrer and Kittay 2002; Pinker 1989; Pustejovsky 1995, to cite only a few parochial examples). It rules out, among other things, most approaches to model-theoretic semantics and semantic network theories, in which word meanings are unstructured nodes in a finite network of inferences or associations; it also rules out the monad-meaning postulate theory of Fodor et al. (1980).7 4. Meanings do not, however, decompose into necessary and sufficient conditions. The theory acknowledges the numerous arguments that concepts have fuzzy borderlines and family resemblance properties. S&C, chapter 8, develops and defends in detail appropriate formal devices that incorporate such phenomena integrally into the possible decompositions of word meanings. In particular, some conceptual features and some interactions among features have continuous (i.e. analog) characteristics that permit stereotype and family resemblance effects to be formulated. This aspect of the theory is shared with Cognitive Grammar (e.g. Lakoff 1987). 5. There is no formal distinction of level between semantics and pragmatics. S&C, chapter 6, shows that once one has the primitives and principles of combination appropriate for nonlinguistic tasks such as object categorization, the same machinery will account for many ‘purely semantic’ tasks such as linguistic inference. It is concluded that there is no reason to posit a separate ‘semantic’ component that deals exclusively with inference. Rather, one may view ‘semantics’ as providing the part of the conceptual structure of an utterance that is related directly to linguistic expression, and ‘pragmatics’ as providing the part that arises through inference, heuristics, world knowledge, and understanding of the context. In other words, the same sorts of structures can arise through multiple sources, often overlapping and intertwining in the 7 See also Fodor 1970; Fodor, Fodor, and Garrett 1975; Fodor 1981. For replies, see Jackendoff 1983, chapter 7; 1990, chapter 1; 2002, chapter 9; Pinker 2007.
prologue
9
complete structure. This property of CS will become central at many points on the chapters in this volume, but especially in chapters 5 and 6. I part company here with Bierwisch 1986, Partee 1993, and to a certain extent Pinker 1989, who posit a separate level of ‘linguistic semantics’. Within the present approach, a separate semantic level is unnecessary and to some extent undesirable, given the interdigitation of conceptual structure information coming from language and that coming from nonlinguistic sources (Bolinger 1965 is still a touchstone on this issue). To the extent that there is a domain of ‘linguistic semantics’—the part of meaning that is ‘grammatically relevant’— this consists of the part of conceptual structure which is visible to the interface with syntactic structure. In other words, the domain is defined by the character of the interface, not by the character of conceptual structure itself. However, the issues are at this point far from resolved. The main point, on which Bierwisch, Pinker, and I agree (I am unclear about Partee), is that there is a languageindependent and universal level of CS, whether directly interfacing with syntax or mediated by an intervening level. Since CS serves as the form of linguistic meaning, there must also be a set of interface rules that permit meanings to be expressed, as shown in Figure 1.1. Since different languages express the same meaning in different ways, the interfaces must be language-particular, though no doubt constrained by UG. In addition, as mentioned in section 1.1, the conceptualization of the world must be related to perception and action. Thus CS must be linked by further interfaces to the mental representations proprietary to the perceptual systems and to the production of action. Accordingly, an important goal of Conceptual Semantics is to articulate three systems of principles: .
The formation rules for CS, that is, the primitives and principles of combination that collectively generate the infinite class of possible concepts—both lexical concepts (word meanings) and phrasal concepts (including sentential concepts or propositions)
.
The rules of inference, pragmatics, and heuristics
.
The interface rules between CS and the other representations with which it interacts
Within this inquiry, the leading questions of lexical semantics then come to be framed as: .
What fragments of conceptual structure can be encoded in lexical items?
.
When lexical items are combined syntactically, how are they correspondingly combined in CS, and what principles license these correspondences?
10
pr olo g ue
None of these goals, of course, can be pursued in isolation; they are intimately interdependent. Each chapter in this volume touches on all of them to varying degrees. The formation rules for semantic/conceptual structure include, among other things, a vocabulary of primitive conceptual categories or ‘semantic parts of speech’. Among these categories are such entities as Thing (or Object), Event, State, Action, Place, Path, Property, and Amount (S&C, chapter 3). They are combined according to (at least) three structural principles: function-argument structure, modifier-head relations, and binding. The general form of function-argument structure is notated as in predicate calculus: F(X) for a one-place function and F(X, Y) for a two-place function. (3) shows a number of possible realizations of primitive conceptual categories involved in understanding spatial states and events in terms of function-argument structure.
(3) a. PLACE
[Place PLACE-FUNCTION (THING)]
b. PATH Path
c. EVENT
TO FROM TOWARD AWAY-FROM VIA
THING PLACE
[Event GO (THING, PATH)] [Event STAY (THING, PATH)] [State BE (THING, PLACE)]
d. STATE
[State ORIENT (THING, PATH)] [State EXTEND (THING, PATH)]
Let me briefly explain these expressions. (3a) says that a conceptual constituent of the basic category Place can consist of a Place-function plus an argument of the category Thing. The argument serves as a spatial reference point, in terms of which the Place-function defines a region. For example, in the expression under the table, the table designates a reference object, and under expresses a Place-function that maps the table into the region beneath it. (3b) similarly shows how a Path, or trajectory, can consist of one of five functions that map a reference Thing or Place into a related trajectory. An example of a Path with a reference Thing is to the house; a Path with reference Place is from under the table, where the trajectory begins at the Place ‘under the table’.
prologue
11
(3c) says that a constituent of the category Event can consist of either of the two Event-functions GO or STAY, each of which takes two arguments. The arguments of GO, which denotes motion along a path, are the Thing in motion (or Theme) and the Path it traverses. This structure is seen most transparently in a sentence like Bill went to New York, where the Theme is expressed as the subject and the Path as the PP. The arguments of STAY, which denotes stasis over a period of time, are the Thing standing still (again called Theme) and its Location, as seen in Bill stayed in the kitchen. (3d) gives three realizations of State. The first is used for specifying the location of objects (the dog is in the park), the second for specifying the orientation of objects (the sign points toward New York), and the third for specifying the extension of linear objects along a Path (the road goes along the beach). Beyond the functions in (3) that encode spatial relations, the system includes a family of basic causative functions (CAUSE, LET, HELP, and ENABLE), and a set of mereological functions that encode part–whole relations (PART-OF, BOUNDARY-OF, etc.), the latter developed in Chapter 5. A founding insight of Conceptual Semantics (due to Gruber 1965) is that all of these functions can be applied to semantic fields other than physical space. For instance, an object’s being owned by someone (a social relation) is often expressed crosslinguistically as the object ‘being at’ the owner, and changes of possession are often expressed as the object ‘going’ ‘from’ one owner ‘to’ another. This suggests that these conceptual functions can be decoupled from their physical context so as to apply to more abstract domains as well, such as time (The meeting is at 6), ascription of properties (The light changed from red to green), and (in the case of causation) social coercion (Bill persuaded Alice to go) and logical entailment (P entails Q ¼ ‘P causes Q to be true’).8 Further functions that have been investigated (Jackendoff 2007b) involve the personal and social domains. They include theory-of-mind predicates (e.g. perception, belief, intention, and commitment to norms); value predicates in various domains (e.g. affective, utility, quality, moral); predicates of exchange (actions done in exchange, return, or retaliation for someone else’s action); and the system encompassing obligations, rights, and authority. In addition, of course, function-argument structure appears in the conceptual structure of relational nouns such as kinship terms and of argument-taking adjectives such as proud of and angry at.
8
This insight is treated somewhat differently in Cognitive Grammar (Lakoff 1987), where it is taken to show that linguistic expression and thought are pervaded by an extensive and powerful system of conceptual metaphor.
12
pr olo g ue
There are two other means of conceptual combination besides functionargument structure. Modifier-head relations are exemplified by the relation between a place modifier and an NP or VP that it modifies: a house in the woods, Pat visited Sue in Chicago. This is notated in two different ways in the chapters in this volume:
(4)
HOUSE
a. Thing
[Place IN ([Thing WOODS])]
b. [ Thing HOUSE; [Place IN ([Thing WOODS])]] Other cases of modification include prenominal adjectives modifying NPs (big house); degree phrases modifying APs (extremely noisy); and manner, instrumental, and time adverbials modifying VPs (run slowly, cut with a knife, come at 6:00). The third means of conceptual combination is binding. This is used to define a constituent in conceptual structure in terms of another, more fully fleshed out constituent, the antecedent. Two classic cases are the representations of reflexive pronouns and control. Binding is notated in terms of a Greek superscript on the antecedent and a corresponding Greek letter for the bound constituent, as in (5). (5)
a. Fred likes himself: [State LIKE ([Person FRED]a, [Person a; MALE SG])] b. Al likes to swim: [State LIKE ([Person AL]a, [Event SWIM ([a])])]
Binding is also implicated in the conceptual structure of relative clauses, where the gap in the relative clause is bound to the head, so that the entire event described by the relative clause becomes a modifier of the noun (Culicover and Jackendoff 2005). (6)
the man who Sam saw: [Person MANa; DEF; [Event PAST [Event SEE ([SAM], [a])] ]] ‘man a the (such that) Sam saw a’
One further important aspect of the Conceptual Semantics formalism needs to be mentioned. The type–token distinction is formalized as a simple binary feature TYPE vs. TOKEN, applied to concepts that are otherwise formally similar.9 In this respect the theory differs from, e.g., predicate calculus, in which tokens are encoded as constants and types as predicates—formal 9 An alternative formulation might treat TOKEN as a privative feature, so that tokens are CS constituents with the feature TOKEN and types are constituents lacking this feature. Still another alternative is that tokens are CS constituents linked to indices on the referential tier (see next paragraph), and types are constituents lacking such linkage.
prologue
13
expressions of entirely different sorts. Within the Conceptual Semantics formalism, three basic predication relations emerge as fundamentally similar: (7) a. Token identity (e.g. Clark Kent is Superman) [State IS-TOKEN-IDENTICAL-TO ([x; TOKEN], [y; TOKEN])] b. Categorization (e.g. Rover is a dog) [State IS-INSTANCE-OF ([x; TOKEN], [y; TYPE])] c. Category inclusion (e.g. A dog is an animal) [State IS-INCLUDED-IN ([x; TYPE], [y; TYPE])] The similarity in form among these conceptual structures parallels their similarity in linguistic expression, namely NP is NP. Moreover, it is shown that the three functions, IS-TOKEN-IDENTICAL-TO, IS-AN-INSTANCE-OF, and ISINCLUDED-IN, are at least closely related if not virtually identical—as revealed in the use of the same verb be to express all of them. (This treatment of predication is developed in detail in S&C, chapters 5 and 6.) The structures described so far constitute only one of the tiers of conceptual structure—where tier is taken in exactly the same sense as in phonology: an independent subcomponent with its own primitives and principles of combination, linked to other subcomponents by interfaces. Orthogonal to the specification of who did what to whom, the content of a sentence is also partitioned into topic, focus, and common ground, so-called information structure. There is also some evidence pushing in the direction of a tier of referential structure, which encodes referential claims and referential relations such as specificity, opacity, and scope of quantification, to some degree independent of propositional organization. These aspects of meaning are not addressed in the chapters in the present volume; see Foundations of Language, chapter 12. When one claims that conceptual structure can be described in terms of primitives and principles of combination, and in particular that lexical items can be conceptually decomposed into primitives, it is natural and proper to ask how one justifies proposed primitives. This question in turn falls into two parts. First, how does one tell in general whether one putative primitive is better than another? Of course, an isolated primitive can never be justified; a primitive makes sense only in the context of the overall system of primitives in which it is embedded. With this proviso, however, I think a particular choice of primitives should be justified on the grounds of its capacity for expressing generalizations and explaining the distribution of the data. That is, a proposed system of primitives should be subject to the usual scientific standards of evaluation. The second part of the question of justifying primitives is: How does one tell whether the primitives one has proposed are really primitive? Often this question is raised with the insinuation that if one can’t tell whether one is all the way
14
pr olo g ue
at the bottom, the enterprise is hopeless. My answer is that one probably can’t tell, but that should not be a matter for worry. The decomposition of all substances into ninety-two primitive elements was a major breakthrough at the end of the nineteenth century. But over the course of the twentieth century these primitives in turn were further decomposed, first into electrons plus nucleus, then into electrons plus protons and neutrons, then the latter two of these into quarks, which in turn are combinations of more primitive features such as spin, color, up/down, etc. Each level of decomposition explained more about the nature of matter and raised new questions of its own; and each step was cause for excitement, not discouragement. We will see parts of a similar progression in Chapters 5 and 6, when the categories treated as primitive in Semantics and Cognition (and in (3) ) undergo further decomposition in terms of a more explanatory set of primitives. Among other things, it will develop that there are probably no words that directly express conceptual primitives; all words are composite. This should not be cause for alarm. In chemistry/physics, after all, none of the quarks (not to mention quark-features) appear in isolation. Closer to home, no phonological primitives appear in isolation either; one of the major points of Prague School phonology, preserved in the generative tradition, is that the phoneme is always divisible into features which themselves never occur in isolation.
1.4 The lexicon In the Parallel Architecture, a well-formed sentence is a triplet consisting of well-formed phonological, syntactic, and conceptual structures, plus the links between them established by the interface components. For example, the sentence Sue goes into the room has the structure in (8); the subscripts denote parts of the three structures that correspond. (8)
a. Phonological structure: [Utterance [Wd suw]1 [Wd gow2 [Aff -z3] ]4 [Wd intuw]5 [Wd [Cl ðә6] [Wd ruwm]7]8 ]10 b. Syntactic structure: [S [NP N1] [VP [V V2þ[presþ3sg]3 ]4 [PP P5 [NP Det6 N7 ]8 ]9 ]]10 c. Conceptual structure: [Event PRES3 [Event GO2 ([Person SUE1], [Path INTO5 ([Thing ROOM7; DEF6]8)]9)] ]10
Let me unpack (8) by working through its conceptual structure. The entire Event (subscript 10) corresponds to the phonological utterance and the
prologue
15
syntactic S. This Event is set in a time frame (subscript 3) expressed by tense in syntax and pronounced as the affix -z. The Event is characterized by the Eventfunction GO; that is, this is an event of motion. This function is expressed by the verb and the sound /gow/ (subscript 2). The first argument of GO (subscript 1) corresponds to the subject in syntax, pronounced Sue. The second argument of GO (the Path, subscript 9), corresponds to the PP in syntax. This Path argument is composite: the Path function INTO, expressed by the preposition into (subscript 5), takes a Thing as its argument, expressed by an NP (subscript 8). The Thing in turn decomposes into the category ROOM, expressed by a noun pronounced /ruwm/ (subscript 7), and the feature of definiteness, corresponding to the determiner, pronounced /ðә/ (subscript 6).10 (8) encodes the standard thematic roles of Theme and Goal structurally. Theme, the thing in motion, is the conceptual structure constituent that serves as the first argument of GO; Goal, the point at which motion terminates, is the constituent that serves as the argument of TO or INTO. Thus [SUE] is Theme of (8) and [HOUSE] is Goal of (8). This is an essential feature of Conceptual Semantics: thematic roles are treated as structural positions in conceptual structure, not as an independent system of diacritics (or case-markers). In order to see how (8) is put together from its parts, it is necessary to look at the lexical entries for the two words in (8) that have argument structure, namely go and into. In the entries in (9), the first line is the phonological structure; the second is the syntactic structure; the third is the conceptual structure. The cosubscripting indicates a long-term memory association between these structures. Thus each lexical entry can be though of as a small-scale interface rule; the semantic portion is the word’s lexical conceptual structure (LCS). (9) a. Phonology: Syntax: Semantics: b. Phonology: Syntax: Semantics:
gow2 V2 PP9 [Event GO2 ( [Thing x], [Path y]9)] intuw5 P5 NP8 [Path INTO5 ([Thing z]8 )]
The items’ contextual features are notated in italics in (9): subcategorization appears as part of syntactic structure, and selectional restriction as part of conceptual structure. These are interpreted as typed variables that must be 10
Note that phonological information such as /suw/ is not notated as part of syntactic structure, as is customary. The notation in (8) reflects the division of labor between the three components. The fact that this person’s name is pronounced Sue rather than Wolfgang or Armadillo is a fact of phonology, not syntax. The only thing that is significant for syntax is that this is a proper noun. The syntax is however linked to the pronunciation by the subscript.
16
pr olo g ue
instantiated in a well-formed sentence. Into requires an NP object that is coindexed with the argument position in conceptual structure. Go (in this context at least) requires two arguments, a Thing in motion and a Path that specifies the trajectory of motion. The first carries no subscript, as it is mapped into subject position by independent principles. The second argument is filled in with the reading of the PP following the verb. It is sometimes useful to align the LCSs of the words so as to show each one’s contribution to the overall conceptual structure of the sentence. (10) does this for sentence (8). The important part to see here is how the variables of one item are instantiated by other items. (10) [Event PRES3 [Event GO2 ([Person SUE1], [Path INTO ([Thing ROOM7; DEF6]8)]9)] ]10 -s [Event PRES3 [
]]
[Event GO2 ([Thing x
go: Sue:
]9)]
], [Path y
[PersonSUE1 ]
into:
]8)]9
[Path INTO ([Thing z
the:
DEF6
room:
[Thing ROOM7
]
A conceptual structure similar to (8c) can be expressed in different phonological and syntactic form, for example by a sentence like (11). (11) Sue enters the room. Here enter is an optionally transitive verb, with a lexical entry like (12).11 (12) Phonology: entr11 Syntax: V11 (NP12) Semantics: [Event GO ([Thing x], [Path INTO ([Thing y]12)]11 This verb incorporates into its meaning the conceptual material expressed separately by the preposition into in (8). It leaves the Goal as an open variable, which can be expressed by an NP object. However, this verb also allows the Goal to be left unspecified. The intransitive version, Sue entered, means not just ‘Sue traversed some path,’ but ‘Sue went into something.’ That is, the sense of into appears even when the second argument is unexpressed. We can see from these examples that the Parallel Architecture treats the ‘argument structure’ or ‘u-grid’ of a lexical item not as a separate level of lexical representation (as in Stowell 1981, Williams 1984, Higginbotham 1985, 11
I slide over some tricks of subscripting involved here. See Jackendoff 2002, section 12.1.
prologue
17
Rappaport and Levin 1985, 1988, Grimshaw 1990, and many others), but simply as the collection of typed variables in the item’s lexical structure. The structural positions of these variables in turn determine the u-roles of the syntactic arguments in the sentence. Hence the process of ‘u-marking’ amounts essentially to linking syntax to conceptual structure.12 Within this conception, then, a word is thought of not as a passive unit to be pushed around in a derivation, but as a part of the interface components. It is a long-term memory linkage of a piece of phonology, a piece of syntax, and a piece of semantics, stipulating that these three pieces can be correlated as part of a well-formed sentence. In large part, this conception of lexical items can be adapted easily into a mainstream architecture (and I assumed such an integration for many years). However, a further issue can be raised about the nature of the lexicon. Three different criteria might be assumed as defining a lexical item: (13)
Lexical items are a. The words (and morphemes) of the language. b. The exceptional, idiosyncratic parts of the language that cannot be predicted by rule. c. The parts of language that are listed in long-term memory.
(13a) and (13b) are traditional views of the lexicon; (13c) is a more cognitively based definition. For a stereotypical lexical item such as dog, they coincide: dog is a word of English, not predictable by rule, and obviously stored in speakers’ long-term memory. However, in less stereotypical circumstance, the definitions diverge. The pronunciation, syntax, and meaning of the word unhappy are completely predictable from the word happy plus the regular principle for forming unadjectives. Thus, although unhappy is a word of English and therefore satisfies (13a), it does not satisfy (13b); and its status vis-a`-vis (13c) is an empirical question (my guess is yes for unhappy, though perhaps not for, say, unprovocative). For a different case, the idioms kick the bucket and bite the dust are not words or morphemes of the language, so they do not satisfy (13a). Their phonology and syntax are predictable from the phonology and syntax of the
12
Even though syntactic and semantic argument structures are normally closely correlated, it is important to keep them distinct. There is a double dissociation between them. On one hand, as seen in (12), some semantic arguments need not be expressed syntactically. On the other hand, there are verbs with supernumerary syntactic arguments that add nothing to the semantics, such as the reflexive arguments of perjure oneself, avail oneself of X. Moreover, there are pairs of verbs that are essentially synonymous but realize their arguments in different syntactic configurations, e.g. replace X with Y, substitute Y for X.
18
pr olo g ue
constituent words, just like throw the shovel and chew the gum, but their (identical) meaning is not. So something must be present in the grammar to list their idiosyncrasy (13b), and this something must be stored in speakers’ longterm memory (13c). Finally, cliche´s and fixed expressions such as money can’t buy me love and I don’t care satisfy neither (13a) nor (13b), as they consist of multiple words with their literal predictable meanings. Yet they do satisfy (13c): what makes them cliche´s is precisely that they are phrases everyone knows. DiSciullo and Williams (1987) show there are important grammatical distinctions between words and morphemes on one hand and phrases on the other, a position that the Parallel Architecture adopts. However, they also assert that the issue of what is listed in long-term memory is ‘of no interest to the grammarian’. In contrast, the Parallel Architecture takes this distinction to be crucial to defining the lexicon. That is, the notion of word and the notion of lexical item are not coextensive. This move leaves the way open for including in the lexicon all manner of units larger than single words, such as idioms and cliche´s. After all, they have to be listed somewhere. (Idioms are incorporated into the lexicon in my 1975 paper reproduced here as Chapter 2; cliche´s first appear in The Architecture of the Language Faculty.) This approach further permits the lexicon to contain meaningful constructions— the stock-in-trade of Construction Grammar—such as the soundþmotion construction illustrated in (14). (14) The trolley squealed around the corner. The bullet whistled by my ear. The verbs themselves describe sound emission, and therefore do not license the path PPs as part of their semantic argument structure. Moreover, no word in the sentence expresses the understood motion of the subject. The meaning and syntax are instead provided by a construction that imposes an idiomatic interpretation on the VP and that takes both the verb and the PP as arguments. (15a) states its syntactic structure; (15b) states its semantics. (15) a. Syntax: [VP V1 PP2] b. Semantics: [GO ([Thing x], [Path y]2); WHILE [PRODUCE-SOUND1 ([Thing x])]] ‘go PP while V[produce sound]-ing’ Construction (15) can be thought of as a lexical item, one that happens to lack phonology. It is however more ‘rule-like’ than words and idioms, because its syntax consists entirely of variables that must be filled by other items. VP constructions like this play a major role in the Parallel Architecture beginning in Semantic Structures, and are represented here by Chapters 7, 8, and 9.
prologue
19
In turn, regular VPs, idiomatic VPs, and soundþmotion VPs are all instantiations of the more general piece of structure (16), which therefore can also be thought of as a lexical item, this time consisting only of syntactic variables. (16)
[VP V (NP) (PP)]
A language may also contain constructions with irregular syntax, which Peter Culicover (1999) has termed ‘syntactic nuts’. (17)
a. b. c. d.
The more you eat, the worse you feel. How about some lunch? One more beer and I’m leaving. day after day
These too can be formalized as lexical entries; Chapters 10–12 deal with three cases. Finally, the formalism for lexical entries generalizes readily to regular affixes. For instance, the regular 3rd singular present tense verb affix can be written as (18). (18)
Phonology: [Wd [Wd x]2 [Aff -z] ]1 Syntax: [V V2 þ [presþ3sg] ]1 Semantics: [Event/State PRES1 [Event/State y]2]
One consequence is that regular morphology is to be treated analytically (or declaratively) rather than procedurally. For instance, the English regular past tense is not a rule that says ‘add -d to a verb to form its past tense’. Rather, the past tense is a lexical entry parallel to (18), which combines with verbs the same way transitive verbs combine with their direct objects, namely through the instantiation of a variable. In principle, this approach extends to all regular affixation; Chapter 11 deals with an intriguing regular reduplicative affix of English, illustrated in (19). (19)
I’ll make the tuna salad, and you make the SALAD-salad.
Thus we arrive at the position (Foundations of Language, chapter 6) that words, idioms, rules of grammar, and regular affixes are all stated in a common format, namely as pieces of stored structure. This continuum between idiosyncrasy and regularity is a feature of Cognitive Grammar, HPSG, and Construction Grammar as well as the Parallel Architecture. Such a treatment goes strongly against the traditional assumption that a language can be partitioned cleanly into a lexicon and a grammar. Mainstream generative grammar has adopted this central assumption uncritically. However, open-minded examination of a fuller range of phenomena reveals instead that words are in one corner of a multidimensional continuum of stored structures, maximally general
20
pr olo g ue
rules are in another corner, and in between are all sorts of phenomena of varying degrees of regularity.
1.5 The syntax–semantics interface As mentioned in section 1.1, mainstream generative grammar derives all semantic combinatoriality from syntactic combinatoriality. Following traditional grammar and philosophers of language such as Frege and Carnap, this approach takes it for granted that the syntax–semantics interface is essentially an isomorphism: phrase and sentence meanings are built up strictly from the meanings of their words, following the constituency dictated by syntactic structure. Culicover and Jackendoff 2005 call this assumption interface uniformity. A consequence of this assumption is that syntax is forced to be at least as combinatorially complex as semantics—if not more so, since it also has to answer to its own internal imperatives such as word order and agreement. And indeed this consequence has been achieved twice in the history of generative grammar: in the Generative Semantics movement of the late 1960s and early 1970s (Lakoff 1971), and in Government-Binding Theory of the 1990s and the Minimalist Program. In both cases, the burgeoning complexity of derivations and the proliferation of hidden structure have been hailed as a great advance in understanding, but (in my judgement at least) rigorous empirical analysis and psychological plausibility have been sacrificed (for detailed discussion, see Culicover and Jackendoff 2005, chapters 2 and 3).13 In the Parallel Architecture, the combinatorial properties of meaning are not derived from syntax; they arise from autonomous conceptual combinatoriality. Syntax functions in the grammar not as the fundamental generative mechanism, but rather as an intermediate stage in the mapping between meaning and sound (in either direction). Part of the mapping is provided by the words, which are interface rules between small-scale pieces of meaning and sound. The remaining part of the mapping is the encoding of the semantic relations among the words: the function-argument, head-modifier, and binding relations. This is the role of syntax: it encodes the semantic relationships among the words in a sentence in terms that are visible to phonology, such as linear order, inflectional morphology, and anaphoric elements—as well as coding the overall semantic force of a clause, such as declarative vs. interrogative. But syntax need not encode any more of semantic structure than is necessary to map between phonology and meaning. 13
The Minimalist Program often presents itself with a rhetorical façade of rigor (e.g. Chomsky 2000). On the other hand, when attacked for lack of precision, proponents typically protest that this is not a theory, it is only a program. You can’t have it both ways.
prologue
21
In Semantics and Cognition (1983) I adopted a version of interface uniformity called the Grammatical Constraint, and used it to argue for conceptual structures that are more like syntactic structure than like quantificational logic. This canonical or default correspondence accounts for a vast range of facts about semantic compositionality, particularly in the domain of argument structure. However, by the time of Semantic Structures (1990), I had discovered numerous syntactic structures that did not map in canonical fashion into conceptual structure, and this paved the way for seeing the syntax–semantics interface as a ‘dirty’ correspondence, similar to the relation of phonology to syntax—hence providing stronger evidence for the Parallel Architecture. Consider first the canonical mapping. Here the interface between CS and syntax preserves embedding relations among constituents: If a syntactic constituent XP expresses the CS constituent X’, and if another syntactic constituent YP expresses the CS constituent Y’, and if XP contains YP, then, as a rule, X’ contains Y’.
.
A verb (or other argument-taking item) in syntax corresponds to a function in CS, and the subject and object of the verb normally correspond to CS arguments of the function.
.
.
Adjuncts in syntax normally correspond to modifiers in CS.
Hence much of the overall structure of syntax corresponds to CS structure. This can be seen in the correspondences illustrated in (8)–(12) above. But even in the canonical mapping between syntax and CS, each representation has properties not shared by the other. Unlike syntax, CS has no notion of linear order. For instance, CS can be expressed syntactically in English, where the verb precedes the direct object, or in Japanese, where the verb follows the direct object. Furthermore, CS is indifferent to grammatical gender and grammatical case features (nominative/accusative), which are purely syntactic in character. Conversely, there are aspects of CS to which syntax is indifferent. Most prominently, other than argument structure, much of the conceptual material bundled up inside a lexical item is invisible to syntax, just as phonological features are. As far as syntax is concerned, the meanings of cat and dog (which have no argument structure) are identical, as are the meanings of eat and drink (which have the same argument structure): the syntactic reflexes of lexical meaning differences are extremely coarse.14 14
Taylor 1996 and Wierzbicka 2007 attack this position, on the grounds that if two words are syntactically indistinguishable, then every collocation involving these words is expected to be the same. But this confuses syntactic structure with idiosyncratic lexical behavior. For example, just
22
pr olo g ue
And even within the canonical mapping, it is characteristic for the syntax– semantics correspondence to be many-to-many. For instance, (20a,b) are noun complement constructions and (20c,d) are noun-noun compounds. But the semantic relationships cut across this pattern: in (20a,c) the head denotes a part of the dependent, and in (20b,d) the dependent denotes an ingredient of the head (see Chapters 5 and 13). (20) a. b. c. d.
leg of a chicken house of stone chicken leg stone house
Under mainstream assumptions, the syntactic structure must reflect the semantics precisely; thus the superficially different (20a) and (20c) must be derived from a common underlying syntactic form, and the superficially parallel (20a) and (20b) must be derived from distinct syntactic forms. By contrast, the Parallel Architecture proposes that no hidden syntax is involved. Rather, English syntax offers complementation and compounding as alternative ways to combine words, and the syntax–semantics interface is flexible enough to map either meaning into either form. Going beyond the canonical mapping, many constructions do not preserve relative embedding between syntax and semantics. A first case is the relation of tense to sentence meaning: in (8)–(10), for instance, the tense, one of the most deeply embedded elements in syntactic structure, expresses the outermost operator in the semantics. This has typically been mitigated in syntactic theory by making tense the uppermost underlying predicate in the sentence, then through a sequence of deformations attaching it to the verb. Such operations are typical of what proves necessary if interface uniformity is to be preserved. Another case of mismatch is the soundþmotion construction illustrated in (14)–(15) (The bullet whistled by my ear). The semantic head of the construction is the function GO, which appears nowhere in the syntax; and the syntactic head, the verb, corresponds to a manner modifier in semantics. This construction belongs to a larger class of constructions discussed in detail in Chapters 7–9. Another syntax–semantics mismatch is illustrated in (21).
because cat and dog are syntactically identical does not mean that alongside the compound hot dog we should find hot cat; just because cellar and basement are synonymous (in my speech) does not imply that alongside wine cellar we should find wine basement; and just because eat and drink are syntactically identical does not imply that alongside the idioms eat NP out of house and home and drink NP under the table we should find drink NP out of house and home and eat NP under the table.
prologue (21)
23
that disaster of a theory (¼ ‘that theory, which is a disaster’) a nightmare of a discussion (¼ ‘a discussion that was a nightmare’) a hell of a guy (¼ ‘a remarkable guy’)
In these expressions, the of-complement functions as the semantic head: that disaster of a theory is referring to a particular theory, not a particular disaster. Conversely, the syntactic head functions semantically as a modifier: the theory is being incidentally characterized as disastrous. Thus this construction reverses the canonical correspondence of heads and complements (Asaka 2002, and for the cognate construction in Dutch, Booij 2002).15 The soundþmotion construction illustrates another ‘dirty’ characteristic of the syntax–semantics interface: as noted a moment ago, the notion of motion, notated as GO is CS, is not expressed in syntax at all. Another example of unexpressed meaning, due to Talmy (1978a), is (22). (22)
The light flashed until dawn.
The interpretation of (22) contains the notion of repeated flashes. But this repetition is not coded in the verb flash, since The light flashed normally denotes only a single flash. Nor is it encoded in until dawn, since, for instance, Bill slept until dawn does not imply repeated acts of sleeping. Rather, the notion of repetition comes from the fact that (a) until dawn specifies the temporal bound of an otherwise unbounded process; (b) the light flashed is a pointevent and therefore temporally bounded; (c) in order to make these compatible, a principle of coercion interprets the flashing as stretched out in time by repetition. This notion of repetition, then, appears in the CS of (22) but not in the LCS of any of its words. Chapters 5 and 6 (from 1991 and 1996 respectively) discuss this particular type of coercion (‘aspectual coercion’) in detail. Another phenomenon with similar properties is reference transfer (Nunberg 1979, Fauconnier 1985): (23)
15
a. The ham sandwich over in the corner wants more coffee. (¼ ‘guy with ham sandwich’) b. Plato is on the top shelf, next to Chomsky. (¼ ‘book by/bust of Plato/ Chomsky’)
Collins, Moody, and Postal 2008 discuss a construction of African-American Vernacular which appears to have the same property: Bill ’s ass can be used to mean ‘Bill’, and all its selectional properties correspond to the semantics. Collins et al., assuming interface uniformity, conclude that this construction is produced from a hitherto unattested underlying form by hitherto unattested principles of raising. The implausibility of their analysis reflects the difficulties of mainstream theory in microcosm. In a Parallel Architecture treatment, it is altogether parallel to a hell of a guy (a construction which they do not address).
24
pr olo g ue
Jackendoff 1992 (incorporated into Culicover and Jackendoff 2005) shows that these cannot be derived from a semantically transparent underlying syntactic form, yet they interact with grammatical phenomena such as binding, so they cannot be purely ‘pragmatic’, either. Based on numerous phenomena of this sort, I concluded in The Architecture of the Language Faculty that the correspondence between syntax and CS is much like the correspondence between syntax and phonology. Certain parts of the two structures are in fairly regular correspondence, but many parts of each structure are invisible to the other, and there are numerous noncanonical correspondences as well. Thus the mainstream assumption of interface uniformity must be replaced with an approach in which simple ‘Fregean’ compositionality is still at the core of the syntax–semantics interface, but there are also numerous cases of ‘enriched composition’ that violate uniformity. Another angle on the interface comes from considering crosslinguistic differences. Despite the fact that CS is universal, languages can have different strategies in how they typically bundle up conceptual elements into lexical items. For example, Talmy 1980 documents how English builds verbs of motion primarily by bundling up motion with accompanying manner, while Romance languages bundle up motion primarily with path of motion, and Atsugewi bundles up motion primarily with the type of object or substance undergoing motion. Levinson 1996 shows how the Guggu Yimithirr lexicon restricts the choice of spatial frames of reference to cardinal directions (see section 4.8). These strategies of lexical choice affect the overall grain of semantic notions available in the language. In addition, of course, there are differences in meaning among apparently synonymous lexical items across languages, such as the differences among prepositions discussed by Bowerman 1996. Moreover, each language has its own repertoire of special meaningful syntactic constructions such as those illustrated above in (15), (17), (19), and (21). Languages also differ in what elements of conceptual structure they require the speaker to express in syntax. For example, French and Japanese require the speaker always to differentiate his or her social relation to the addressee, a factor largely absent from English. Finnish and Hungarian require the speaker to express the multiplicity (or repetition) of events, using iterative aspect, a factor absent from English, as seen in (22). On the other hand, English requires the speaker to express the multiplicity of objects by using the plural suffix, a requirement absent in Chinese. Again these differences suggest that the syntax– semantics interface is not a simple isomorphism.
prologue
25
1.6. Simpler Syntax From this perspective emerges the basic stance of Simpler Syntax (Culicover and Jackendoff 2005). Semantic structure requires a certain degree of complexity in order to explain inference and the relation of language and thought to perception. Given that this complexity is independently necessary, it should play as large a role as possible in constraining the grammatical structure of sentences, and syntax should play as small a role as possible. Hence the theory should cut syntactic structure down to the bare minimum necessary to accomplish the sound–meaning mapping.16 However, the theory must still acknowledge that the ‘generative engines’ of syntax and morphosyntax cannot be eliminated altogether: they are necessary to account for differences among languages in word order, case marking, agreement, handling of long-distance dependencies, and the existence of special constructions. The resulting syntactic theory is by no means simple, but it is far simpler than mainstream models (whatever ‘simple’ means in this context). A consequence of this stance is that syntactic structures can be kept relatively flat (i.e. undifferentiated). Aside from linear order, there is no syntactic distinction between specifiers, arguments, and adjuncts, as this is already provided for in the semantics. The result is X-bar skeleta for NP, AP, and PP with only two layers, as in (24a–c). The only exception to the two-layer schema is S, which is a three-layer projection of V, as in (24d).
(24) (a)
NP … N …
AP
(b) …
A …
PP
(c) …
P
S
(d) …
... VP … …
V
…
One price of this structural simplification is the need for trees with multiply branching nodes, rather than the strictly binary branching in contemporary mainstream structure. However, Culicover and Jackendoff 2005 argue that strictly binary branching is often not an advantage but a disadvantage. For instance, a putative advantage of strictly binary branching is that rules of grammar need not be sensitive to linear order, only to dominance. But from a larger perspective, linear order comes for free in the signal, and hierarchical 16 This is thus a ‘minimalist’ approach to syntax, but with different premises from the Minimalist Program about what is to be minimized. See also Jackendoff 2008 for a discussion of how this approach represents an ‘alternative minimalist vision’.
26
pr olo g ue
structure does not. So rules that depend in part on linear order actually should be easier for the child to learn. Hence the alleged advantage of strictly binary branching evaporates on closer inspection. Simpler Syntax makes use of almost no empty nodes in syntactic structure. This is desirable in principle, because empty nodes make heavier demands both on the learner (who has to figure out where they are) and on processing (which also has to figure out where they are). Most empty nodes in the classical theory are posited either for semantic reasons or to promote syntactic uniformity. For instance, it is standard to posit a phonologically empty element PRO that serves as subject of an infinitival VP where there is none at the surface, thereby giving all verbs a syntactic subject. Simpler Syntax instead allows for infinitival VPs without syntactic subjects, and it uses the interface to identify their ‘understood’ subjects in Conceptual Structure. Return to the example Al likes to swim (5b), which illustrated the conceptual structure associated with control of missing subjects of infinitivals. The mapping to phonology and syntax can be encoded as in (25), with no empty subject in the syntax of the complement clause. (25) Phonology: æl1 layk2-s3 tuw4 swim5 Syntax: [S [NP N]1 [VP [V V2þ[presþ3sg]3] [VP to4 V5] ]] Conceptual structure: [State PRES3 [State LIKE2 ([Person AL]1a, [Event SWIM5 ([a])])] ] Likewise, ellipsis is not derived through empty nodes or deletion. Rather, elliptical configurations, especially when they are syntactically unusual (as in Gapping), are treated as meaningful constructions listed in the lexicon. The interpretation of an elliptical construction is derived from the conceptual structure of its antecedent—or from the conceptual structure of the context—not from a deleted syntactic structure. Culicover and I cite many cases of ellipsis for which there is no plausible syntactic antecedent, such as the underlined constituents in (26). (26) a. It seems we stood and talked like this before. We looked at each other in the same way then. But I can’t remember where or when. (¼ ‘where or when we stood and talked like this before and looked at each other in the same way as we’re looking at each other now’) [Rodgers and Hart] b. Something Alan said attracted Bea’s attention—or vice versa—but I can’t remember which. (¼ ‘whether something Alan said attracted Bea’s attention, or whether something Bea said attracted Alan’s attention’) c. [Spoken to someone about to jump off a building] Don’t!!! (¼ ‘Don’t jump!’)
prologue
27
Similarly, constructional meaning, such as that of the soundþmotion construction illustrated in (15), has no syntactic reflex. Nor do cases of coercion such as those in (22) and (23). Some of these constructions are treated in mainstream theory in terms of syntactic (or PF) deletion of unexpressed elements; others are not treated in mainstream theory at all. Culicover and I show that they are all best treated in terms of elements of semantics that have no syntactic realization. Like other constraint-based theories such as HPSG and LFG, Simpler Syntax has no movement and no covert level of syntactic structure such as Logical Form. The effects ascribed to movement in mainstream theory are accounted for with a variety of mechanisms, most of them shared with other constraintbased theories, especially HPSG. These mechanisms include: Free phrase order (e.g. among adjuncts in VP, where the order is constrained only by prosody and focus)
.
.
Alternative argument realizations (e.g. dative alternation)
For long-distance dependencies, operator–trace relations along the lines of HPSG (where trace is the only kind of empty node in Simpler Syntax). The wellknown constraints on long-distance dependencies arise from multiple sources, only some of which are syntactic. Others arise from processing complexity and from semantics.
.
As illustrated above, binding and control are relations over Conceptual Structure, not over syntactic structure, though they may involve syntactic conditions on the relation between anaphoric elements and antecedents.
.
In order to account for so-called A-movements, in particular passive and raising, it is unfortunately necessary to introduce extra machinery. Simpler Syntax proposes a grammatical function (GF-) tier that modulates the syntactic realization of semantic arguments expressed as NPs, that is, subjects, objects, and indirect objects. We are not too dismayed by this extra mechanism, as the principles behind it appear in every substantive syntactic theory: as f-structure in LFG, as essentially all of Relational Grammar, as the complement hierarchy in HPSG, and as abstract case in GB/MP. Given that this tier plays no role in the analyses in the present volume, the reader is referred to Culicover and Jackendoff 2005 for motivation and details. In a way, Simpler Syntax is a realization of the Interpretive Semantics of the late 1960s and early 1970s, developed as an alternative to Generative Semantics. Back then it looked as though the expressive power of the syntactic component could be considerably reduced through appropriate use of interface principles (then called ‘projection rules’). But the theory could not be worked
28
pr olo g ue
out in detail, in large part because there was no independent characterization of semantics on which the interface could be anchored. The development of Conceptual Semantics over the intervening three decades—along with the field’s deepened understanding of a wide range of phenomena—has made it possible to develop a rigorous theory of independent syntactic and semantic components and their connection.
1.7 The issue of semiproductivity I also want to bring up an issue that is not directly related to the Parallel Architecture per se, but which does concern the structure of the lexicon. This issue is a theme running through the chapters in this volume (especially Chapters 2, 7, 8, 9, 12, and 13): the problem of semiproductivity. The basic observation behind generative grammar is that language cannot consist simply of a list of memorized utterances: speakers can say and understand an unlimited number of things they’ve never heard before. Thus generative grammar has always concentrated on characterizing the productive rules of language. A stereotypical productive rule is the phrase structure of the English VP: one can learn new transitive verbs and compose them with novel direct objects and novel adjuncts without limit. Within morphology, a stereotypical productive rule is the English regular plural. This can be applied to new nouns one has never heard before, and there are no surprises. In fact, the classic illustration of speakers’ knowledge of its productivity is the ‘wugs test’ (Berko 1958), in which 6-year-old children reliably produce plurals for nonsense nouns (and 4- and 5-year-old children do so somewhat less comprehensively). Another well-known productive rule is English expletive infixation (McCarthy 1982), which, if properly constrained prosodically, is perfectly reliable in licensing forms such as manu-fuckin-facturer. In contrast, semiproductive phenomena have always evoked some embarrassment or avoidance in generative theory. Such phenomena display a regularity that can be stated as a rule. Nevertheless, acceptable instances of the rule must be learned individually. Often speakers have the intuition that a new form sounds unusual; they may observe that they have never heard it before, or they may find it amusing. A classic case of semiproductivity is the English irregular past tense with ablaut. Although one can state phonological conditions that are necessary for a verb to have an ablaut past tense, speakers must still learn which verbs undergo ablaut, and what the vowel changes to. For instance, alongside sing-sang, ring-rang, and stink-stank, we find that the phonologically similar swing has the past swung, the homophone of ring has wring-wrung, stink has
prologue
29
the alternate past form stunk, and think and bring fall into an entirely different family with seek, fight, teach, and buy. Evidently, then, the appropriate past tense form must be listed in the lexicon. Yet simple listing of the forms in the lexicon apparently precludes capturing the family resemblances among particular cases. Two cases discussed in Chapter 2 are the English deverbal nouns such as contribution, destruction, appraisal, and harassment, and English denominal verbs such as butter (the bread), pocket (the money), weed (the garden), de-claw (the cat), hammer (a nail), father (a child), mother (one’s students), waitress (in a restaurant), and so on. To some extent the forms are predictable; for example verbs ending in -ate invariably have a nominal in -ation. But not every case is predictable. Recite has the two nominals recital and recitation, with different meanings. The phonologically parallel incite has incitement rather than incital or incitation, and excite has excitation and excitement but not excital. Alongside butter (the bread), we might expect to find mustard (the bread), but we do not (except, if Google can be trusted, very rarely in the context of recipes); alongside pocket the money we might expect envelope the letter (where envelop means something rather different synchronically). As with the ablaut past tenses, the issue is how the grammar can both list the existing forms and capture the subregularities among them. An additional complication in these cases, which involve derivational morphology, is that the meanings of the existing forms are not entirely predictable either. For instance, a recitation is only vaguely related to reciting, and excitation, unlike excite, can pertain to electrons but not to people (this point goes back to Chomsky 1970). And despite the semantic similarity of the nouns father and mother, the corresponding verbs have radically different meanings. Within a semiproductive pattern, moreover, there may be little pockets of true regularity. For instance, among the English denominal verbs, there is a subclass that means ‘fasten with N’, such as nail, screw, tape, and glue. This class seems altogether productive, in that the new fastener velcro, unlike mustard, occasions no hesitation in forming a related verb. There is no evident reason, morphological or semantic, for one of these classes to be productive and the other semiproductive. Semiproductivity is not confined to morphology. Chapters 9 and 12 deal with two phrasal constructions that unexpectedly prove to be semiproductive: the English resultative construction with the structure V-NP-AP (e.g. burn the pot black) and the NPN construction (e.g. day after day). Similarly, Chapter 7 deals with the verb-particle construction, where, as is well known, there is productive use of particles as directional phrases (e.g. send the letters out), overlaid with a plethora of verb-particle idioms (e.g. look the answer up). In each of these cases,
30
pr olo g ue
as in derivational morphology, we find an interweaving of productive cases, semiproductive cases, and singleton idioms. The overall issue, then, is how a grammar/lexicon encodes semiproductive phenomena, at once listing the instances and capturing the generalizations among them. Chapter 2, originally from 1975, develops a theory of lexical redundancy rules which is intended precisely to enable the grammar to express such generalizations. Some of the background of this approach—and its drawbacks—are discussed in the remarks preceding the chapter and in the commentaries interspersed throughout. A decade later, the productive/semiproductive distinction reappeared as a focus of the ‘past tense debate’ between connectionists (Rumelhart and McClelland 1986) and Pinker and associates (e.g. Pinker and Prince 1988, Pinker 1999). The connectionists claimed that all items are listed, and the cases that we call productive are just more frequent. There are no rules per se in language; regularities are simply a result of analogical extension. Such an approach makes some sense for semiproductive phenomena: because we’ve never heard mustard the bread, we don’t say it, but we still understand it by analogy with butter the bread. The approach makes less sense, however, for truly productive phenomena such as wugs or Bachs or manu-fuckin-facturer. Pinker and his associates defend a mixed position. They agree with the connectionists and disagree with traditional generative grammar (e.g. Chomsky and Halle 1968) about semiproductive phenomena: there are no semiproductive rules, only networks of associations among listed items. However, they disagree with the connectionists and agree with traditional generative grammar about the status of productive phenomena: these are the product of derivational rules along the lines of ‘To form plural nouns in English, add -z.’ Foundations of Language essentially adopts Pinker’s position on the split between semiproductive and productive morphology. Items related semiproductively are taken to be stored in the lexicon and to be related only by association; but there is no separate statement of a rule per se that captures the regularity among them. On the other hand, productive phenomena are expressed by rules. Unlike Pinker’s approach, though, rules are themselves taken to be lexical items. For instance, the regular third person singular present tense inflection in English is the lexical item encoded in (18) above: a phonological suffix -z that attaches to a word, linked to a morphosyntactic suffix [3sgþpres] that attaches to a verb stem, linked to the meaning PRESENT pertaining to a situation expressed by the verb. On further reflection, though, the position that semiproductive phenomena are merely associative is suspicious. It implies, for instance, that the grammar
prologue
31
nowhere says explicitly that there is a V-NP-AP resultative construction or an NPN construction: there is only a collection of similar-looking idiomatic constructions, some of which are productive. Similarly, consider the situation in morphology. If there are no semiproductive rules, the grammar will also nowhere say that there is an affix -al for forming deverbal nouns: there is just a collection of deverbal nouns that happen to end in -al. And, considering a case discussed in section 2.5.1, the grammar will nowhere say that there is a semiproductive system for forming Latinate verbs such as transfer, infer, confer, transmit, commit, transport, import, comport that involves a specific family of prefixes and stems—there will just be a collection of phonologically associated verbs. All of these conclusions seem odd (at least to a linguist). An alternative possibility is that the grammar does contain explicit semiproductive rules, but that there is some formal difference between them and productive rules. Let’s ask what this formal difference might be. It cannot be anything about the form of the rule. For instance, the rule (or lexical item) that encodes V-NP-AP resultatives is of exactly the same general form as the completely productive rule (15) for the soundþmotion construction. The only alternative that springs to mind is some diacritic feature on the rule, stipulating whether it is productive or semiproductive. Some approaches (e.g. HPSG) arrange the lexicon in terms of inheritance hierarchies, so that related items inherit properties from more abstract items in a taxonomy. If the lexicon is arranged in terms of inheritance hierarchies, a semiproductive rule can be thought of as a node in the hierarchy whose daughters are all the instances. For instance, the rule for -al nominals will be an expression consisting of a variable followed by -al, and this expression will dominate all the known instances of -al nominals in the inheritance hierarchy. This expression would in essence say that there is an -al affix in English with certain stipulated properties. Inheritance hierarchies are not confined to the linguistic lexicon. The literature on semantic memory for concepts has explored parallel notions in considerable detail—in particular, taxonomies with default values and exceptions (Raphael 1968, Smith and Medin 1981, Murphy 2002). Consider the taxonomy for, say, animals. One is familiar with a finite number of kinds of animals. Upon encountering a novel kind of animal, one may recognize it as an animal because of its similarity to animals one knows. But one also recognizes it as a new kind and stores it away as an instance of another member of the taxonomy. If this parallel between lexical and conceptual taxonomies is genuine, it provides a solid argument against the position that there are no semiproductive
32
pr olo g ue
rules, only associations among instances. For this position would entail that, similarly, there is no category of animal in mental representation, and there are no categories of cat and dog and so on that are particular cases of animals— there are only various collections of individuals one happens to have experienced. In other words, the notion of semantic taxonomies would collapse entirely to the crudest sort of exemplar theory, a conclusion rather difficult to stomach. The argument, then, is that if there is a category of animals in mental representation, there should be no problem with specific semiproductive rules for forming -al nominals, V-NP-AP resultatives, NPN constructions, and so on. (And Pinker and I were wrong about semiproductivity.) Suppose then that the proper way to think about semiproductive rules is as nodes in an inheritance hierarchy. Where do productive rules fit in? The problem is that they potentially have exactly the same form as semiproductive rules. They too are expressions consisting of some combination of variables and constants. What makes them productive is only that their instances cannot all be listed. But it is not that they have no listed instances. Experimental research has shown that even within productive phenomena, high-frequency derived instances may be stored (Baayen, Levelt, and Haveman 1993; Nooteboom, Weerman, and Wijnen 2002), and these would necessarily appear in an inheritance hierarchy. Moreover, since the learner has to induce the existence and form of a productive rule from heard and stored instances, a number of stored instances of such rules are likely to remain after acquisition, at least for some period of time. A case for which experimental results are unnecessary is English compounding, discussed in Chapter 13. On one hand, we store thousands and thousands of compounds with specialized meanings, so the phenomenon is at least semiproductive. But on the other hand, we constantly encounter novel compounds, have no difficulty interpreting them, and do not notice them as unusual (unlike mustard the bread). So this is a productive rule that clearly also has huge numbers of listed instances. Hence, apparently the only difference between a productive and a semiproductive rule is that productive rules license one to go beyond the listed instances without any special effort. This suggests that the formal distinction between the two sorts of rule is specifically localized in a diacritic on the variable: those in productive rules are marked [þproductive] and those in semiproductive rules are not. This is still less of a difference than before. Is this a Good Thing or a Bad Thing? I hope it’s a Good Thing, because that’s where the facts seem to take us. A child acquiring a language has to pick out the regularities in the input and construct rules. But how does the child discover whether a particular regularity
prologue
33
is productive or semiproductive? It’s not given in the data. How is it that everyone comes up with essentially the same answer? I find this a serious puzzle for linguistic theory. In morphology, sometimes there is competition between regularities, for instance between regular past tense, ablaut, zero marking, and the tiny family of verbs whose past ends in -ought. Presumably only one of these can be the regular default, so the others must be admitted as semiproductive. But on the other hand, there isn’t always a competing form. For instance, zero denominal verbs such as butter are not in competition with anything. One might guess that productive phenomena are more frequent in the data than semiproductive phenomena. For instance, this is presumably a clue that the -ed past is the regular form. But this can’t always be the case. Consider the morphological form illustrated in (27). (27)
(all) X-ed out ¼ ‘exhausted from experiencing X to excess’ I’m Olympic’d out. [recorded in conversation with someone who worked at the Olympics] You must be entertained out. [recorded in conversation after weeks of guests] He’s all knitted out. [after knitting for three days solid] I’m Edward G. Robinsoned out. [after watching an Edward G. Robinson marathon]
This morphological form, mentioned again in section 7.7, is completely productive, with either nouns or verbs in the variable position. Yet it is exceedingly infrequent—I would guess one hears it perhaps three times a year in conversation. How do language learners come to the conclusion that it is productive on the basis of so little evidence? It is plausible that children observing a regularity initially encode it as semiproductive, and later, if evidence warrants, upgrade it to productive. This would comport with the evidence that on the whole children are cautious about extending generalizations to new items (Tomasello 2003), yet they do eventually ‘go productive’. On the other hand, children are known to overgeneralize semiproductive processes (e.g. Bowerman’s example Mommy, giggle me, overgeneralizing the English causative), so it is not clear that there is a uniform oneway move from semiproductive to productive. Ideally, we would like an account in which children do not have to ‘decide’ whether a phenomenon is productive or not. Rather, we would hope that the course of acquisition would follow from the distribution of the data in speech and independent properties of how the brain learns.
34
pr olo g ue
The material presented in this volume (Chapters 2, 7, 8, 9, 12, and 13) does not solve these problems, it only makes them more pointed. Insofar as semiproductivity is an issue not only for phonology and morphology but also syntax, in my opinion it must take a place as one of the central issues of linguistic theory for the coming years.
Remarks on Chapter 2 I began work on ‘Morphological and semantic regularities in the lexicon’ in 1968–9, when still a graduate student at MIT. I presented a version of it at the LSA’s 1969 summer meeting at the University of Illinois (back when the LSA still had annual summer meetings). After the publication of Semantic Interpretation in Generative Grammar in 1972, I resumed work on it, and it was finally published in Language in 1975. One colleague, a prominent phonologist and morphologist, wrote to me that this was the worst paper he had ever read in his life—even worse than Chomsky!—and the best thing I could do for the field was get out of it immediately. I treasured this letter and had it taped to my office door for many years. Another colleague wrote that each member of her seminar had applied the approach to a different language, and it worked like a charm for all of them. My reaction was that if it was that easy, there must be something wrong with it! Between these two responses and my growing interest in other issues, I let the topic drop for many years. Nevertheless, some of the ideas persisted in the background of my work, especially as the role of the lexicon became more prominent in the Parallel Architecture. Around the time I started thinking about assembling the present volume, two good friends independently mentioned having reread this paper with pleasure. So I’ve decided to resurrect it, partly as a relic of the old days, but also partly because its issues turn out to be pertinent to my current vision of grammar. At the same time, because of its age, it calls for more extensive commentary than other papers reprinted here. The chapter addresses a hot topic of the so-called Linguistics Wars (Harris 1993): how to specify the relations among lexical items while preserving each item’s idiosyncrasies. Let me contextualize this a little bit for contemporary readers. Chomsky 1957 and especially Lees 1960 derived NPs such as the enemy’s destruction of the city from full sentences such as The enemy destroyed the city. But this treatment left open how it is that some verbs can undergo the nominalization transformation and others cannot. Lakoff 1970 (written originally in 1965) developed a theory of ‘exception features’ that allowed lexical items to undergo or avoid undergoing particular transformations. Exception features made it possible to stipulate which verbs undergo the nominalization transformation.
36
remarks on chapter 2
They also made possible much of the elaborate machinery of Abstract Syntax and eventually Generative Semantics (e.g. Ross 1967, McCawley 1968, Lakoff 1971). Chomsky’s opening salvo against these developments was his series of Thursday afternoon lectures in the fall of 1967, published in 1970 as ‘Remarks on Nominalization’. People who have only read the paper have remarked that they don’t see what was so special about it; this may be partly because we take everything in it so much for granted now. But Chomsky’s lectures were electrifying, especially because Lakoff and Ross were in attendance, constantly challenging him. I took from the lectures at least four major advances on the Aspects framework (which was not yet called the Standard Theory). First, a careful look at the idiosyncrasies of the syntax and semantics of deverbal nouns shows that one cannot state sufficiently general transformations that derive NPs headed by these nouns from sentences. This undermined Lees’s approach to nominals, which had been taken over wholeheartedly by Lakoff. (Connoisseurs may notice that the footnotes of the published version mention Lees only briefly but critique Lakoff extensively.) The rejection of a transformational derivation for complex NPs led to the other three advances. First, the Aspects theory retained a relic of the old generalized transformations of Syntactic Structures, which derived subordinate clauses by inserting kernel sentences into other clauses: S was taken to be the only recursive node in phrase structure. But once the enemy’s destruction of the city is not derived from an underlying sentence, it is necessary to admit phrase structure recursion on nodes other than S, in particular on NP. To those of us who had not been raised on generalized transformations, this move seemed like a no-brainer, but to the older folks (in their thirties!) it was a bit startling. Second, since complex NPs were no longer derived from Ss, it became necessary to explain the structural parallelisms between Ss and complex NPs, especially those such as the city’s destruction by the enemy that look something like passives. Chomsky’s stunning move was to treat syntactic category labels in the same way as phonological segments: not as primitives but as collections of features. He proposed a small collection of features that define the lexical categories, plus features that define ‘bar levels’ for phrasal categories projected upward from the head. This of course was the beginning of X-bar theory; although it had been anticipated by Zellig Harris, it was an important innovation within transformational grammar. X-bar theory made it possible to state syntactic rules—both phrase structure rules and transformations—that apply to more than one category at a time, thereby solving at one stroke (in principle at least) the problem of generalizing syntactic rules across categories. This move undermined a great deal of crucial evidence for Abstract Syntax/
remarks on chapter 2
37
Generative Semantics, for example arguments by Lakoff that adjectives and quantifiers are underlying verbs, and arguments by Ross that English modals are underlying verbs and that sentences that serve as the antecedents of pronouns are dominated by NP. There remained the question of the relationships among lexical items. How is destruction related to destroy, if not by transformational derivation? Chomsky proposed that a lexical entry can be left unspecified as to category, but it still can stipulate that certain of its semantic (and phonological) features are in part dependent on which category it is realized as. This line of thought was never really worked out (perhaps until Marantz 1997). My recollection is that what took hold in common currency instead was the notion of lexical redundancy rules from The Sound Pattern of English (Chomsky and Halle 1968). Lexical redundancy rules were intended to capture the regularities among lexical items, leaving lexical entries that contained only idiosyncratic features. Thus, for example, given the lexical rule for deverbal nouns, the entry for destruction would not have to contain any material that could be predicted from the entry for destroy. This idea eventually developed into the notion of lexical rules in LFG and HPSG, which organize lexical entries into inheritance hierarchies (see especially Pollard and Sag 1987, chapter 8). Although suddenly everyone was talking about lexical redundancy rules, nobody was stating them or working out a theory of them: they remained only a vague promissory note. The paper reprinted here was an attempt to figure out how they work. In the course of it I was able to amplify many of Chomsky’s arguments against Lakoff’s exception features. But I was also surprised to discover that the notion of lexical redundancy rules in common currency— here called the ‘Impoverished Entry Theory’—simply doesn’t work. It turns out to be impossible to extract all the redundancies among lexical items into rules, and then measure the complexity of the lexicon by just counting features. In retrospect, I see that the notion of the lexicon that I was attacking was not Chomsky and Halle’s invention. Rather, it was their adaptation of a deep-seated common-practice tradition found not only in earlier linguists such as Bloomfield, but also in philosophers of language such as Frege and Russell. (It is debatable, though, whether knowing this at the time would have helped me make my case.) My alternative, the ‘Full Entry Theory’, proposes that lexical entries are fully specified in all their details, but the parts of them that are predictable by lexical redundancy rules don’t count for purposes of evaluating the ‘information content’ of the lexicon. I was able to show that the formal properties of this solution account for an interesting range of phenomena, so it captures something important about the texture of the lexicon. Nevertheless, it remained mysterious to me (and still does) how features can be present but ‘not cost
38
remarks on chapter 2
anything’. This is one reason why I dropped the idea for many years. (It was however picked up and elaborated by Bochner 1993.) What was right about this approach was that it recognized the importance of the issue of semiproductivity and tried to account for it in terms of the cost of the lexicon. One thing that was wrong about it was that it treated productive morphology the same way as semiproductive morphology: like Halle 1973, it lists fully productive paradigms in the lexicon along with semiproductive ones. As Hankamer 1989 points out, this means that in the case of a language such as Turkish with massive morphology, tens of thousands of forms have to be listed for every verb, which makes a mockery of the idea that the lexicon is a learned list of items. This is another reason I dropped the idea. A solution started to emerge in the context of the debate in the late 1980s about the status of the English past tense. As discussed in section 1.7, Pinker and colleagues (e.g. Pinker and Prince 1988, 1991; Pinker 1999) made a crucial distinction: items with productive morphology are produced by rule and need not be listed in the lexicon, but items with semiproductive morphology have to be listed. The result is a heterogeneous treatment of morphology, in principle a major ‘imperfection’ in grammar, but in practice evidently unavoidable, and confirmed by psycholinguistic and neurolinguistic evidence (Jackendoff 2002, sections 6.2–3). This move breaks the logjam in the present chapter’s treatment of productive morphology: members of productive paradigms do not have to be listed, and do not count in the evaluation of the lexicon’s information content. Meanwhile, my own view of the lexicon (section 1.4) was emerging, in which not only words, but also productive rules of grammar and meaningful constructions are formalized as lexical items. This solved another peculiar result of the present chapter. The problem was how idioms can possibly ‘cost less’ by virtue of adhering to standard phrase structure. My conclusion was that the phrase structure rules are in this case acting as lexical redundancy rules instead of generative mechanisms. This made little sense in 1975. But in the Parallel Architecture, where phrase structure rules are part of the lexicon too, there is no reason they cannot fulfill both generative and redundancy functions. This becomes clearer in Chapters 7, 8, and 9, which discuss meaningful VP constructions and relate them to the phrase structure rules for VP, and especially in Chapters 12 and 13, where the phrase structure rules for NPN constructions (day after day) and noun-noun compounds have to serve both as productive rules and as templates for a plethora of listed items. Finally, recent developments have begun to offer a new foothold on the notion of information content in the lexicon. Moscoso del Prado Martı´n, Kostic´, and Baayen 2004 in particular introduce the idea of the ‘entropy’ of a family of morphological paradigms—a measure of its information content. This
remarks on chapter 2
39
is related to the number of members in the family, their relative frequency, and the frequency of the members of the family as a whole. The authors use this measure to predict the latency for lexical access of words that belong to morphological families. Their treatment does not immediately extend to the phenomena in the present chapter. But it has the right flavor, relates to the now extensive psycholinguistic literature, and introduces a number of refinements that I couldn’t have dreamed of in my probabilistic/statistical innocence. My hope is that understanding this measure of information content will tell us something important about the character of memory storage in the brain. The upshot is that many of the theoretical features of the present chapter that were decidedly odd back in 1975 start to have some resonance in a contemporary framework. So perhaps it’s worth another look.
chapter 2
Morphological and Semantic Regularities in the Lexicon (1975)
The starting point of the Lexicalist Hypothesis, proposed in Chomsky’s ‘Remarks on Nominalization’ (1970), is the rejection of the position that a nominal such as Bill’s decision to go is derived transformationally from a sentence such as Bill decided to go. Rather, Chomsky proposes that the nominal is generated by the base rules as an NP, no S node appearing in its derivation. His paper is concerned with the consequences of this position for the syntactic component of the grammar. The present chapter will develop a more highly articulated theory of the lexical treatment of nominals, show that it is independently necessary, and extend it to a wide range of cases other than nominalizations. The goal of this chapter is very similar to that of Halle 1973: the presentation of a framework in which discussion of lexical relations can be made more meaningful. I will not present any new and unusual facts about the lexicon; rather, I will try to formulate a theory which accommodates a rather disparate range of well-known examples of lexical relations. The theory presented here, which was developed independently of Halle’s, has many points of correspondence with it; I have, however, attempted a more elaborate working out of
[This chapter appeared originally in Language 51.3 (1975), 639–71, and is reprinted here by permission of the Linguistic Society of America.] My thanks go to John Bowers, François Dell, Noam Chomsky, Morris Halle, and to classes at the 1969 Linguistic Institute and Brandeis University for valuable discussion. Earlier versions of this chapter were presented to the 1969 summer LSA meeting at the University of Illinois and to the 1970 La Jolla Syntax Conference. Thanks also to Dana Schaul for many useful examples scattered throughout the chapter.
morphological and semantic regularities
41
numerous details. I will mention important differences between the theories as they arise.
2.1 Levels of adequacy in description In a theory of the lexicon, we can distinguish three levels of adequacy in description, parallel to those discussed by Chomsky 1965 for grammatical theory. The first level consists in providing each lexical item with sufficient information to describe its behavior in the language. This corresponds to Chomsky’s level of observational adequacy, in which the grammar is required to enumerate correctly the set of sentences in the language. A theory of the lexicon meeting the second level of adequacy expresses the relationships, subregularities, and generalizations among lexical items of the language, e.g. the fact that decide and decision are related in a systematic fashion. This level corresponds to Chomsky’s level of descriptive adequacy, which requires the grammar to express correctly relationships between sentences, such as the active–passive relation. A theory of the lexicon meeting the third level of adequacy describes how the particular relationships and sub-regularities in the lexicon are chosen—why the observed relationships, and not other imaginable ones, form part of the description of the lexicon in question. One of the questions that must be answered at this level is, e.g., why decide rather than decision is chosen as the more ‘basic’ of the two related items. This element of the theory takes the form of an ‘evaluation measure’ which assigns relative values to competing lexical descriptions available within the theory. This is the level of explanatory adequacy. As Chomsky emphasizes, the evaluation measure does not decide between competing theories of the lexicon, but between competing descriptions within the same theory. Each theory must provide its own evaluation measure, and a comparison of competing theories must be based on their success in meeting all three levels of adequacy. Evaluation measures have typically been built into linguistic theories implicitly as measures of length of the grammar, i.e. its number of symbols. One place where such a measure is made explicit is in Chomsky and Halle 1968, chapter 8. The abbreviatory conventions of the theory—parentheses, braces, etc.—are designed so as to represent linguistically significant generalizations in terms of reduced length of grammatical description. Similarly, Chomsky and Halle develop the concept of marking conventions in order to be able to distinguish more ‘natural’ (i.e. explanatory) rules from less ‘natural’ ones, in terms of the number of symbols needed to write the rules.
42
morphological and semantic regularities
In section 2.2, I will present two theories of the lexicon compatible with the Lexicalist Hypothesis. One has a traditional evaluation measure which is applied to the number of symbols in the lexicon; the other has a more unusual measure of complexity, referring to ‘independent information content’. In section 2.3 I will show that the latter theory is preferable. It is hoped that such an example of a nontraditional evaluation measure will lead to greater understanding of the issue of explanatory adequacy, which has been a source of great confusion in the field.
2.2 Formulation of two preliminary theories The fundamental linguistic generalization that must be captured by any analysis of English is that words like decision are related to words like decide in their morphology, semantics, and syntactic patterning. For Lees 1960, it seemed very logical to express this relationship by assuming that only the verb decide appears in the lexicon, and by creating the noun decision as part of a transformational process which derives the NP John’s decision to go from the S John decided to go. However, for reasons detailed in Chomsky 1970, this approach cannot be carried out consistently without expanding the descriptive power of transformations to the point where their explanatory power is virtually nil. Without transformations to relate decide and decision, we need to develop some other formalism. Chomsky takes the position that decide and decision constitute a single lexical entry, unmarked for the syntactic feature that distinguishes verbs from nouns. The phonological form decision is inserted into base trees under the node N; decide is inserted under V. Since Chomsky gives no arguments for this particular formulation, I feel free to adopt here the alternative theory that decide and decision have distinct but related lexical entries. In regard to Chomsky’s further discussion, the theories are equivalent; the one to be used here extends more naturally to the treatment of other kinds of lexical relations (cf. section 2.5). Our problem then is to develop a formalism which can express the relations between lexical entries in accord with a native speaker’s intuition.1 1
Advocates of the theory of generative semantics might at this point be tempted to claim that a formalism for separate but related lexical items is yet another frill required by lexicalist syntax, and that generative semantics has no need for this type of rule. I hasten to observe that this claim would be false. In the generative semantic theory of lexical insertion developed in McCawley 1968 and adopted by Lakoff 1971, lexical items such as kill and die have separate lexical entries, and are inserted onto distinct derived syntactic/semantic structures. For a consistent treatment of lexical insertion, then, break in The window broke must be inserted onto a tree of the form [v BREAK], while break in John broke the window must be inserted onto [v CAUSE BREAK], which has undergone Predicate Raising; in other words, break has two distinct lexical entries. Semantically,
morphological and semantic regularities
43
It is important to ask what it means to capture a native speaker’s intuition of lexical relatedness. It makes sense to say that two lexical items are related if knowing one of them makes it easier to learn the other—i.e. if the two items contain less independent information than two unrelated lexical items do. A grammar that expresses this fact should be more highly valued than one that does not. The advocate of a transformational relationship between decide and decision claims that this intuitive sense of relatedness is expressed by his transformation, in that it is unnecessary to state the shared properties of the words twice. In fact, it is unnecessary to state the properties of decision at all, since they are predictable from the lexical entry of decide and the nominalization transformation.2 Hence a grammar containing the nominalization transformation contains less independent information than one without it—since instead of listing a large number of nominalizations, we can state a single transformation. Within such a grammar, the pair decide–decision contains fewer symbols than a random pair such as decide–jelly: given decide, there need be no lexical entry at all for decision, but jelly needs a lexical entry whether or not decide is listed. Furthermore, the regularity of decide–decision means that many pairs will be related by the transformation, so a net reduction in symbols in the grammar is accomplished, and the evaluation measure will choose a grammar including the transformation over one without it. Since the Lexicalist Hypothesis denies a transformational relationship between decide and decision, their relationship must be expressed by a rule within the lexical component. Transformational grammar has for many years had a name for the kind of rule that expresses generalizations within the lexicon—it is called a lexical redundancy rule; but little work has been done until now toward a formalization of such rules. The first question we must ask is: By what means does the existence of a lexical redundancy rule reduce the independent information content of the lexicon? There are two possibilities. The first, which is more obvious and also more akin to the transformational approach, gives decide a fully specified entry; but the entry for decision is either non-existent or, more likely, not fully specified. The redundancy rule fills in the missing information from the entry of decide at some point in the derivation of a sentence containing decision, the two breaks are related in exactly the same way as die and kill; but clearly break and break must be related in the lexicon in a way that die and kill are not. A similar argument holds for rule and ruler vs. rule and king. Thus generative semantics requires rules expressing lexical relations for exactly the same reasons that the Lexicalist Hypothesis needs them. Only in the earlier ‘abstract syntax’ of Lees 1960 and Lakoff 1970 are such rules superfluous. 2 Of course, it also is difficult to express the numerous idiosyncrasies of nominalizations, as Chomsky 1970 points out at some length.
44
morphological and semantic regularities
perhaps at the stage of lexical insertion. As in the transformational approach, the independent information content of decide–decision is reduced, because the entry for decision does not have to be filled in. The evaluation measure again can simply count symbols in the grammar. We may call this theory the impoverished-entry theory. Within such a theory, a typical lexical entry will be of the form given below. All aspects of this form are traditional except for the ‘entry number’, which is simply an index permitting reference to a lexical entry independent of its content:
(1)
entry number /phonological representation/ syntactic features SEMANTIC REPESENTATION
For example, decide will have the form (2). The entry number is arbitrary, and the semantic representation is a fudge standing for some complex of semantic markers. The NP indices correlate the syntactic arguments of the verb to the semantic arguments (cf. Jackendoff 1972, chapter 2, for discussion of this [2010: and for a less stipulative version that invokes general linking principles, see Jackendoff 1990, Culicover and Jackendoff 2005]):
(2)
784 ı /dec¯d/ +V +[NP1 —— on NP2] NP1 DECIDE ON NP2
We now introduce a redundancy rule, (3), in which the two-way arrow may be read as the symmetric relation ‘is lexically related to’. The rule thus can be read: ‘A lexical entry x having such-and-such properties is related to a lexical entry w having such-and-such properties.’a
a In the Parallel Architecture formalism, in which rules are formalized in the same format as words, the formalization of redundancy rules as relations between two lexical entries is problematic. Section 1.4 offered a formalization of the English regular 3rd singular present tense verb suffix in the format of Parallel Architecture lexical entries. Following that model, a counterpart of (3) would be:
(3’) Phonology: [Wd [x]2 þ sˇәn]1 Syntax: [N V2 þ Af]1 Semantics: [ABSTRACT RESULT OF [F(X, Y)]2]1 The parts in italics are variables to be filled by the root. Similar adaptations are possible with most of the redundancy rules throughout the chapter.
morphological and semantic regularities
(3)
x /y + ion/ +N +[NP1’s —— (P) NP2] ABSTRACT RESULT OF ACT OF
45
w /y/ +V +[NP1 —— (P) NP2] NP1 Z NP2
NP1’S Z-ING NP2
Given the existence of (3), decision needs only the following lexical entry:
(4)
375 derived from 784 by rule (3)
This theory thus reduces the lexical entry for decision to a cross-reference to the related verb plus a reference to the redundancy rule. The entries of many other nouns will be simplified similarly by the use of a reference to (3). The independent information content of the lexicon can be determined straightforwardly by adding up the information in lexical entries plus that in redundancy rules; hence the evaluation measure can be stated so as to favor grammars with fewer symbols.b A second possible approach to lexical redundancy rules, the full-entry theory, assumes that both decide and decision have fully specified lexical entries, and that the redundancy rule plays no part in the derivation of
The expression in (3’) contains no counterpart of the subcategorization features of the verb and noun, given in (3). My hope is that they will follow from the semantic argument structure of the verb, though it’s not clear to me that this works out straightforwardly. After all this time, we still don’t know a lot about the relation between syntactic and semantic argument structure of nouns. Although Grimshaw 1990 was a start, major compilations on argument structure such as Levin and Rappaport Hovav 2005 are almost exclusively confined to verbs. b An immediate issue arises if we take the psychological reality of the lexicon seriously. If the entry for decision is as in (4), lexical access to the word decision must involve reconstituting it from decide plus the redundancy rule—a process that Jackendoff 1997b calls ‘Lexical Assembly’. Words such as transformationalist with recursive morphology would require multiple steps of reconstitution, each of which requires an additional call to the lexicon. Although no doubt transformationalist takes more time to access than transform, it does not seem likely that the difference is a factor of four. Many of the arguments later on in the chapter against the impoverished entry theory also create parallel problems for lexical access (an issue that of course was not on the radar in the early 1970s). At the time this passage was written, the impoverished-entry theory was a carefully constructed straw man, meant to elaborate a notion that was ‘in the air’. However, some years later, Pollard and Sag (1987, chapter 8) proposed a remarkably similar formal theory of the lexicon, intended to eliminate rampant redundancy from lexical entries. They even have lexical items that include references to the redundancy rules that apply to them (e.g. example (405), p. 215)—though they list applicable redundancy rules within the base form rather than the derived form, rather parallel to Lakoff 1970. As a consequence, alas, all the arguments against the impoverished-entry theory apply to this proposal.
46
morphological and semantic regularities
sentences, as it does in both the transformational theory and the impoverishedentry theory. Rather, the redundancy rule plays a role in the information measure for the lexicon. It designates as redundant that information in a lexical entry which is predictable by the existence of a related lexical item; redundant information will not be counted as independent. In the full-entry theory, lexical entries again have the form of (1), except that an entry number is unnecessary. Decide has the form of (2), minus the entry number. Decision, however, will have the following entry:
(5)
¯ + ion/ /decıd +N +[NP1’s —— on NP2] ABSTRACT RESULT OF ACT OF
NP1’S DECIDING NP2 We evaluate the lexicon as follows: first, we must determine the amount of independent information added to the lexicon by introducing a single new lexical entry; then, by adding up all the entries, we can determine the information content of the whole lexicon. For a first approximation, the information added by a new lexical item, given a lexicon, can be measured by the following convention: (6)
(Information measure) Given a fully specified lexical entry W to be introduced into the lexicon, the independent information it adds to the lexicon is (a) the information that W exists in the lexicon, i.e. that W is a word of the language; plus (b) all the information in W which cannot be predicted by the existence of some redundancy rule R which permits W to be partially described in terms of information already in the lexicon; plus (c) the cost of referring to the redundancy rule R.
Here (6a) is meant to reflect one’s knowledge that a word exists. I have no clear notion of how important a provision it is (it may well have the value zero), but I include it for the sake of completeness. [2010: It is, as it were, the cost of the brackets necessary for coding a new lexical item.] The heart of the rule is (6b); this reflects one’s knowledge of lexical relations. Finally, (6c) represents one’s knowledge of which regularities hold in a particular lexical item; I will discuss this provision in more detail in section 2.6. To determine the independent information content of the pair decide–decision, let us assume that the lexicon contains neither, and that we are adding them
morphological and semantic regularities
47
one by one into the lexicon. The cost of adding (2), since it is related to nothing yet in the lexicon, is the information that a word exists, plus the complete information content of the entry (2). Given (2) in the lexicon, now let us add (5). Since its lexical entry is completely predictable from (2) and redundancy rule (3), its cost is the information that a word exists plus the cost of referring to (3), which is presumably less than the cost of all the information in (5). Thus the cost of adding the pair decide–decision is the information that two words exist, plus the total information of the entry (2), plus the cost of referring to redundancy rule (3). Now note the asymmetry here: if we add decision first, then decide, we arrive at a different sum: the information that two words exist, plus the information contained in (5), plus the cost of referring to redundancy rule (3) (operating in the opposite direction). This is more than the previous sum, since (5) contains more information than (2): the extra phonological segments þion and the extra semantic information represented by abstract result of act of. To establish the independent information content for the entire lexicon, we must choose an order of introducing the lexical items which minimizes the sum given by successive applications of (6). In general, the more complex derived items must be introduced after the items from which they are derived. The information content of the lexicon is thus measured as follows: (7) (Information content of the lexicon) Given a lexicon L containing n entries, W1 , . . . , Wn, each permutation P of the integers 1, . . . , n determines an order AP in which W1 , . . . , Wn can be introduced into L. For each ordering AP, introduce the words one by one and add up the information specified piecemeal by procedure (6), to get a sum SP. The independent information content of the lexicon L is the least of the n! sums SP, plus the information content of the redundancy rules.c Now consider how an evaluation measure can be defined for the full-entry theory. Minimizing the number of symbols in the lexicon will no longer work, because a grammar containing decide and decision, but not redundancy rule (3), contains fewer symbols than a grammar incorporating the redundancy rule, by
c (7) and (8) are hideously formalistic and hardly brain-friendly: in particular, they lead to a combinatorial explosion in evaluating the information content of the lexicon. One possible way to mitigate this problem is to accept (i) that the lexicon stored in the brain is probably not optimal; and (ii) that if the brain is actually applying such an information measure in organizing the lexicon, it’s probably doing so locally, i.e. ‘reordering’ only individual clusters of morphologically related items. Local evaluation is computationally far more tractable, and it’s all that I ever use in the course of this chapter.
48
morphological and semantic regularities
exactly the number of symbols in the redundancy rule. Since we would like the evaluation measure to favor the grammar incorporating the redundancy rule, we will state the evaluation measure as follows: (8)
(Full-entry theory evaluation measure) Of two lexicons describing the same data, that with a lower information content is more highly valued.
The details of the full-entry theory as just presented are somewhat more complex than those of either the transformational theory or the impoverishedentry theory. However, its basic principle is in fact the same: the evaluation measure is set up so as to minimize the amount of unpredictable information the speaker knows (or must have learned). However, the measure of unpredictable information is no longer the number of symbols in the lexicon, but the output of information measure (7): this expresses the fact that, when one knows two lexical items related by redundancy rules, one knows less than when one knows two unrelated items of commensurate complexity. I will argue that the full-entry theory, in spite of its apparent complexity, is preferable to the impoverished-entry theory. As a prelude to this argument, I will mention two other discussions of redundancy rules. The formulation of morpheme-structure rules [2010: now called phonotactic rules]—those redundancy rules which predict possible phonological combinations within the words of a language—is also open to an impoverished-entry theory and a full-entry theory. The former approach is used in Halle 1959 and in the main presentation of Chomsky and Halle 1968, where the redundancy rules are treated as part of the readjustment rules. However, chapter 8 of Chomsky and Halle 1968 describes some difficulties in this theory pointed out by Stanley 1967. The alternative theory presented is (I believe) a notational variant of the full-entry theory: the redundancy rules do not play an active role in a derivation, but rather function as part of the evaluation measure for the lexicon.3 If the fullentry theory turns out to be correct for the closely related area of morphemestructure rules, we should be inclined to prefer it for the rules relating lexical items. Halle 1973 proposes a variant of the full-entry theory for the processes of word formation we are concerned with here. In his theory, the redundancy rules generate a set of ‘potential lexical items’ of the language. He then uses the
3 Chomsky and Halle retain impoverished lexical entries, but only for the purpose of counting up features not predicted by redundancy rules and listing what potential words actually exist. Paired with each impoverished entry, however, is a fully specified entry, which is what actually takes part in the derivation.
morphological and semantic regularities
49
feature [+Lexical Insertion] to distinguish actual words from non-existent but possible words. A ‘special filter’ supplies unpredictable information, including the value of [Lexical Insertion]. The filter thus contains all the information of (6a) and (6b), but has nothing that I can relate to (6c). Consider the contents of Halle’s filter, an unordered list of idiosyncratic information. This list must include reference to every lexical item, including all potential but non-existent ones. It is not rule-governed—rather, it is intended to state precisely what is not rule-governed. It is clear why Halle sets up the lexicon in this way: he is trying to retain a portion of the lexicon where the independent information can be measured simply by counting features, and the filter is just such a place. Our formulation of the information measure in the full-entry theory has freed us of the necessity of listing the independent information separately, or of distinguishing it extrinsically from the redundant information. Instead we have a lexicon containing merely a set of fully specified lexical entries (giving exactly those words that exist), plus the set of redundancy rules. (I will mention Halle’s theory again briefly at the end of section 2.5.1.)
2.3 Which theory? The argument for fully specified entries comes from consideration of words whose affixation is predictable by a redundancy rule, but whose putative derivational ancestors are not lexical items of English. Examples are aggression, retribution, and fission, which have the morphological and semantic properties of the nouns described in redundancy rule (3), but for which there are no corresponding verbs aggress, retribute, or fiss.d Our intuition about these items is that they contain less independent information than comparable items which cannot be partially described by a redundancy rule (e.g. demise and soliloquy), but that they contain more than comparable items which are related to genuine lexical items (e.g. decision, attribution). How can the three theories we have discussed describe these verbs? The transformational theory must propose a hypothetical lexical item marked
d Chomsky 1970 (n. 36) offers a large list of -able/-ible adjectives that have the characteristic meaning ‘affordance for activity’ that goes with this suffix, but for which there is no verb such that they mean ‘able to be V-d’. Many of these are like aggression in that they have no lexical item at all as root: probable, feasible, (im)practicable, formidable, sociable, amiable, amenable, inexorable, tractable, delectable, ineluctable, irascible, incredible, audible, legible, eligible, intelligible, indelible, possible, plausible, compatible. Notice that a few of these have the characteristic negative prefix in- as well, which only exacerbates the problems they raise.
50
morphological and semantic regularities
obligatorily to undergo the nominalization transformation (cf. Lakoff 1970). Thus the lexicon must be populated with lexical items such as fiss which are positive absolute exceptions to various word-formation transformations. The positive absolute exception is of course a very powerful device to include in grammatical theory (see discussion in Jackendoff 1972). Furthermore, the use of an exception feature to prevent a lexical item from appearing in its ‘basic’ form is counter-intuitive: it claims that English would be simpler if fiss were a word, since one would not have to learn that it is exceptional. Lakoff in fact claims that there must be a hypothetical verb king, corresponding to the noun king as the verb rule corresponds to the noun ruler. Under his theory, the introduction of a real verb king would make English simpler, in that it would eliminate an absolute exception feature from the lexicon. In other words, the evaluation measure for the transformational theory seems to favor a lexicon in which every noun with functional semantic information has a related verb. Since there is little evidence for such a preference, and since it is strongly counter-intuitive in the case of king, the transformational account—besides requiring a very powerful mechanism, the absolute exception—is incorrect at the level of explanatory adequacy. Next consider the impoverished-entry theory. There are two possible solutions to the problem of non-existent derivational ancestors. In the first, the entry of retribution is as unspecified as that of decision (4); and it is related by redundancy rule (3) to an entry retribute, which however is marked [Lexical Insertion]. The cost of adding retribution to the lexicon is the sum of the information in the entry retribute, plus the cost of retribution’s references to the redundancy rule and to the (hypothetical) lexical item, plus the information that one word exists (or, more likely, two—and the information that one of those is non-lexical). Under the reasonable assumption that the cost of the crossreferences is less than the cost of the phonological and semantic affixes, this arrangement accurately reflects our initial intuition about the information content of retribution. Furthermore, it eliminates the use of positive absolute exceptions to transformations, replacing them with the more restricted device [Lexical Insertion]. Still, it would be nice to dispense with this device as well, since it is rather suspicious to have entries which have all the properties of words except that of being words. The objections to hypothetical lexical items in the transformational theory at the level of explanatory adequacy in fact apply here to [Lexical Insertion] as well: the language is always simpler if this feature is removed. We might propose eliminating the hypothetical lexical entries by building them into the entries of the derived items:
morphological and semantic regularities
(9)
51
511 derived by rule (3) from ¯ /retribut/ +V +[NP1 —— for NP2] NP1 RETRIBUTE NP2
The cost of (9) is thus the information that there is a word retribution, plus the information within the inner brackets, plus the cost of referring to the redundancy rule. Again, the assumption that the cross-reference costs less than the additional information /ion/ and abstract result of act of gives the correct description of our intuitions. This time we have avoided hypothetical lexical items, at the expense of using rather artificial entries like (9). This artificiality betrays itself when we try to describe the relation between sets like aggression–aggressive–aggressor, aviation–aviator, and retribution– retributive. If there are hypothetical roots aggress, aviate, and retribute, each of the members of these sets can be related to its root by the appropriate redundancy rules (3), (10a), or (10b), where (10a) and (10b) respectively describe pairs like predict–predictive and protect–protector (I omit the semantic portion of the rules at this point for convenience—in any case, section 2.4 will justify separating the morphological and semantic rules):
(10)
x a. /y + ive/ +A
w /y/ +V
x b. /y + or/ +N
w /y/ +V
Suppose we eliminate hypothetical lexical items in favor of entries like (9) for retribution. What will the entry for retributive look like? One possibility is:
(11)
65 derived by rule 10a from ¯ /retribut/ +V +[NP1 — for NP2] NP1 RETRIBUTE NP2
52
morphological and semantic regularities
But this solution requires us to list the information in the inner brackets twice, in retribution and retributive: such an entry incorrectly denies the relationship between the two words. Alternatively, the entry for retributive might be (12) (I use (3’) here to denote the inverse of (3), i.e. a rule that derives verbs from -ion nouns; presumably the presence of (3) in the lexical component allows us to use its inverse as well):
(12)
65 derived by 3¢ and 10a from 511
Thus retributive is related to retribution by a sequence of redundancy rules, and the independent information content of the pair retribution–retributive is the information that there are two words, plus the information within the inner brackets of (9), plus the cost of referring to (3’) once and (10a) twice. This is closer to the intuitively correct solution, in that it relates the two words. However, it is still suspicious, because it claims retribution is more basic than retributive. Clearly the entries could just as easily have been set up with no difference in cost by making retributive basic. The same situation will arise with a triplet like aggression–aggressor–aggressive, where the choice of one of the three as basic must be purely arbitrary. Intuitively, none of the three should be chosen as basic, and the formalization of the lexicon should reflect this. The impoverished-entry theory thus faces a choice: either it incorporates hypothetical lexical items, or it describes in an unnatural fashion those related lexical items which are related through a non-lexical root.e Consider now how the full-entry theory accounts for these sets of words, beginning with the case of a singleton like perdition (or conflagration), which has no relatives like perdite, perditive, etc., but which obviously contains the -ion ending of rule (3). We would like the independent information content of this item to be less than that of a completely idiosyncratic word like orchestra—but
e A further difficulty with the impoverished-entry theory: As Pollard and Sag 1987 and Lakoff 1987 recognize (see also section 1.7), the relation between a redundancy rule and the memorized items that are its subcases can be taken as not unlike that between any sort of cognitive category and its memorized instances. And similar arguments can be mounted against an impoverished-entry theory in this more general domain. For example, my knowledge of my cat Peanut could not be stripped down to just the bare idiosyncratic nonredundant information necessary to discriminate Peanut from other cats—this information would be so fragmentary as to be unusable. Rather, my knowledge of Peanut is in many respects redundant with my knowledge of the general category CAT, but also contains various specializations, elaborations, and exceptions (for instance in the case of my family’s late lamented three-legged cat Cindy). In other words, the relation of a redundancy rule to its lexically listed instances falls under a much more generally attested mental phenomenon, always a desideratum in linguistic theory when achievable.
morphological and semantic regularities
53
more than that of, say, damnation, which is based on the lexical verb damn. The impoverished-entry theory resorts either to a hypothetical lexical item perdite or to an entry containing another entry, like (9), which we have seen to be problematic. The full-entry theory, on the other hand, captures the generalization without extra devices. Note that (6b), the measure of non-redundant information in the lexical entry, is cleverly worded so as to depend on the existence of redundant information somewhere in the lexicon, but not necessarily on the existence of related lexical entries. In the case of perdition, the only part of the entry which represents a regularity in the lexicon is in fact the -ion ending, which appears as part of the redundancy rule (3). What remains irregular is the residue described in the right-hand side of (3), i.e. that part of perdition which corresponds to the nonlexical root perdite. Hence the independent information content of perdition is the information that there is a word, plus the cost of the root, plus the cost of referring to rule (3). Perdition adds more information than damnation, then, because it has a root which is not contained in the lexicon; it contains less information than orchestra because the ending -ion and the corresponding part of the semantic content are predictable by (3) (presumably the cost of referring to (3) is less than the information contained in the ending itself; see section 2.6). We see then that the full-entry theory captures our intuitions about perdition without using a hypothetical lexical item. The root perdite plays only an indirect role, in that its cost appears in the evaluation of perdition as the difference between the full cost of perdition and that of the suffix; nowhere in the lexicon does the root appear as an independent lexical entry. Now turn to the rootless pair retribution–retributive. Both words will have fully specified lexical entries. To determine the independent information content of the pair, suppose that retribution is added to the lexicon first. Its independent information, calculated as for perdition above, is the information that there is a word, plus the cost of the root retribute, plus the cost of referring to (3). Note again that retribute does not appear anywhere in the lexicon. Now we add to the lexicon the entry for retributive, which is entirely predictable from retribution plus redundancy rules (3) and (10a). According to information measure (6), retributive adds the information that it is a word, plus the cost of referring to the two redundancy rules. The cost of the pair for this order of introduction is therefore the information that there are two words, plus the information in the root retribute, plus the cost of referring to redundancy rules three times. Alternatively, if retributive is added to the lexicon first, followed by retribution, the independent information content of the pair comes out the same, though this time the cost of the root appears in the evaluation of retributive.
54
morphological and semantic regularities
Since the costs of these two orders are commensurate, there is no optimal order of introduction, and thus no reason to consider either item basic. Similarly, the triplet aggression–aggressor–aggressive will have, on any order of introduction, an independent information content consisting of the information that there are three words, plus the information content of the root aggress, plus the cost of referring to redundancy rules five times (once for the first entry introduced, and twice for each of the others). Since no single order yields a significantly lower information content, none of the three can be considered basic to the others.f Thus the full-entry theory provides a description of rootless pairs and triplets which avoids either a root in the lexicon or a claim that one member of the group is basic, the two alternatives encountered by the impoverished entry theory. The full-entry theory looks still more appealing when contrasted with the transformational theory’s account of these items. The theory of Lakoff 1970 introduces a positive absolute exception on perdite, requiring it to nominalize; but aggress may undergo either -ion nominalization, -or nominalization, or -ive adjectivalization, and it must undergo one of the three. Thus this theory would be forced to introduce Boolean combinations of exception features, together marked as an absolute exception, in order to describe this distribution—patently a brute force analysis. In the full-entry theory, then, the lexicon is simply a repository of all information about all the existing words; the information measure expresses all the relationships. Since the full-entry theory escapes the pitfalls of the impoverishedentry theory, without giving up adequacy of description, we have strong reason to prefer the former, with its non-standard evaluation measure. From here on, the term ‘lexicalist theory’ will be used to refer only to the full-entry theory.
f An interesting question now arises: Does the lexicon contain explicit links among related items? The theory being worked out here does not assume such links: relations are only implicit, in that they contribute to the measure of information content in the lexicon. By contrast, in a theory that incorporates inheritance hierarchies in the lexicon, relations are explicit: decision is subordinate to decide—and perhaps to þion as well. A strict inheritance hierarchy appears to run into trouble with rootless pairs and triples like retribution– retributive and aggression–aggressor–aggressive, as none of the words is subordinate to any of the others. If one wishes to maintain subordination as the defining relation in the hierarchy, it is hard to see how to avoid ‘virtual’ roots such as retribute and aggress in the lexicon. An intermediate possibility is that there are explicit links among related words, but the relation involved in the links is looser than subordination, perhaps just something like ‘is related to’. This would allow rootless pairs still to be related without listing a virtual root. Yet another possibility would be to establish links among the relevant parts of words, i.e. linking the underlined parts of retribution and retributive to each other, and linking the affixes to their individual entries. This offers the possibility of making a link cost more when there are phonological disparities among related items, as in disparate–disparity, hierarchy–hierarchical, torrent–torrential, and so on. In turn, this leads to a consideration of how phonological regularities might mitigate this extra cost, getting us into territory far beyond the scope of this chapter, but verging on issues explored by e.g. Burzio 2002 and Steriade 2001.
morphological and semantic regularities
55
Before concluding this section, let us consider a question which frequently arises in connection with rootless pairs and triplets: What is the effect on the lexicon if a back-formation takes place, so that a formerly non-existent root (say retribute) enters the language? In the transformational theory, the rule feature on the hypothetical root is simply erased, and the lexicon becomes simpler, i.e. more regular. In the lexicalist theory, the account is a bit more complex, but also more sophisticated. If retribute were simply added without disturbing the previous order for measuring information content, it would add to the cost of the lexicon the information that there is a new word plus the cost of referring to one of the redundancy rules. Thus the total cost of retribution– retributive–retribute would be the information that there are three words, plus the information in the root retribute, plus the cost of four uses of redundancy rules. But now that retribute is in the lexicon, a restructuring is possible, in which retribute is taken as basic. Under this order of evaluation, the information content of the three is the information that there are three words, plus the information in retribute, plus only two uses of redundancy rules. This restructuring, then, makes the language simpler than it was before retribute was introduced, except that there is now one more word to learn than before. What this account captures is that a back-formation ceases to be recognized as such by speakers precisely when they restructure the evaluation of the lexicon, taking the back-formation rather than the morphological derivatives as basic. I speculate that the verb aggress, which seems to have only marginal status in English, is still evaluated as a back-formation, i.e. as a derivative of aggression–aggressor–aggressive, and not as their underlying root. Thus the lexicalist theory of nominalizations provides a description of the diachronic process of back-formation which does more than simply erase a rule feature on a hypothetical lexical item: it can describe the crucial step of restructuring as well.
2.4 Separate morphological and semantic rules At the outset of the discussion, I stated redundancy rule (3) so as to relate lexical items both at the morphological and semantic levels. In fact, this formulation will not do. It claims that there is a particular meaning, ABSTRACT RESULT OF ACT OF V-ING, associated with the ending -ion. However, several different semantic relations obtain between -ion nominals and their related verbs, and several nominalizing endings can express the same range of meanings. Some of the morphological rules are stated in M1 (the morphological part of (3) ), M2, and M3:
56
(13)
morphological and semantic regularities
/y/ +V
M1:
/y + ion/ +N
M2:
/y + ment/ +N
M3:
/y + al/ +N
/y/ +V /y/ +V
Some of the semantic rules are S1 (the semantic part of (3) ), S2, and S3:
(14) S1:
+N +[NP1’s —— ((P)NP2)] ABSTRACT RESULT OF ACT OF
S2:
NP1’S Z-ING NP2
+N +[—–– (NP2)] GROUP THAT Z-S (NP2)
+N +[(NP1’s)—— ((P) NP2)] S3:
(NP1’s)
ACT PROCESS
OF
+V +[NP1 —— ((P) NP2)] NP1 Z NP2
+V +[NP1 —— (NP2)] NP1 Z (NP2)
+V +[NP1 —— ((P) NP2)] NP1 Z NP2
Z-ING NP2 An example of the cross-classification of the morphological and semantic relations is the following table of nouns, where each row contains nouns of the same semantic category, and each column contains nouns of the same morphological category. (15)
M1 M2 M3 S1: discussion argument rebuttal S2: congregation government S3: copulation establishment refusal
morphological and semantic regularities
57
That is, John’s discussion of the claim, John’s argument against the claim, and John’s rebuttal of the claim are semantically related to John discussed the claim, John argued against the claim, and John rebutted the claim respectively in a way expressed by the subcategorization conditions and semantic interpretations of S1; the congregation and the government of Fredonia are related by S2 to they congregate and they govern Fredonia; and John’s copulation with Mary, John’s establishment of a new order, and John’s refusal of the offer are related by S3 to John copulated with Mary, John established a new order, and John refused the offer.4 There are further nominalizing endings such as -ition (supposition), -ing (writing), and zero (offer); and further semantic relations, such as one who Z’s (writer, occupant) and thing in which one Z’s (residence, entrance). The picture that emerges is of a family of nominalizing affixes and an associated family of noun–verb semantic relationships. To a certain extent, the particular members of each family that are actually utilized in forming nominalizations from a verb are chosen randomly. Insofar as the choice is random, the information measure must measure independently the cost of referring to morphological and semantic redundancy rules (cf. section 2.6 for further discussion).g How do we formalize the information measure, in light of the separation of morphological rules (M-rules) and semantic rules (S-rules)? An obvious first point is that a semantic relation between two words without a morphological relationship cannot be counted as redundancy: thus the existence of the verb rule should render the semantic content of the noun ruler redundant, but the 4 I assume, with Chomsky 1970, that the of in nominalizations is transformationally inserted. Note also that the semantic relations are only approximate—the usual idiosyncrasies appear in these examples. g This separation of M-rules and S-rules dissolves the lockstep of phonology, syntax, and semantics and quietly begins to lay the groundwork for the Parallel Architecture. Restated in terms of the Parallel Architecture, the M-rules are (morpho)phonology-to-(morpho)syntax interface rules, and the S-rules are (morpho)-syntax-to-semantics interface rules. In these terms, (3’) now separates into (13’) and (14’).
(13’) M1: Phonology: [Wd [x]2 þ sˇәn] Syntax: [N V2 þ Af]1 (14’) S1: Syntax: [N V2 þ Af]1 Semantics: [ABSTRACT RESULT OF [F(X, Y)]2]1 Two further thoughts: First, I am left wondering whether the separation into M-rules and S-rules is too separate. It does not explicitly recognize that the three M-rules in (13) and the three S-rules in (14) form a family, for example distinct from the -er/or, -ant/ent, and -ist endings that form agentive nominals. A further complication is that S2 is related to the agentive endings, and -ist applies to things other than verbs, as in violinist. Second, reformulating redundancy rules as lexical entries, as in (13’)–(14’), means that (as suggested by Bochner 1993) the evaluation measure for the lexicon can apply to these entries just as it does to normal words, capturing the regularities among them. In fact, one could argue that it is precisely in the interests of optimizing the information content of the lexicon that the rules are split into M-rules and S-rules when the evidence warrants.
58
morphological and semantic regularities
semantic content of king must count as independent information. Hence we must require a morphological relationship before semantic redundancy can be considered.h A more delicate question is whether a morphological relationship alone should be counted as redundant. For example, professor and commission (as in ‘a salesman’s commission’) are morphologically related to profess and commit; but the existence of a semantic connection is far from obvious, and I doubt that many English speakers other than philologists ever make the association. What should the information content of these items include? A permissive approach would allow the phonology of the root to be counted as redundant information; the only non-redundant part of professor, then, would be some semantic information like TEACH. A restrictive approach would require a semantic connection before morphology could be counted as redundant; professor then would be treated like perdition, as a derived word with a non-lexical root, and both the phonology /profess/ and the semantics TEACH would count as independent information. In section 2.5.1 I will present two cases in which only morphological rules play a role, because there are no semantic regularities. However, where semantic rules exist, it has not yet been established whether their use should be required in conjunction with morphological rules. Therefore the following restatement of information measure (6) has alternative versions: (16) (Information measure) Given a fully specified lexical entry W to be introduced into the lexicon, the independent information it adds to the lexicon is (a) the information that W exists in the lexicon; plus (b) (permissive form) all the information in W which cannot be predicted by the existence of an M-rule which permits W to be partially described in terms of information already in the lexicon, including other lexical items and S-rules; or (b’) (restrictive form) all the information in W which cannot be predicted by the existence of an M-rule and an associated S-rule (if there is one) which together permit W to be partially described in terms of information already in the lexicon; plus (c) the cost of referring to the redundancy rules. Examples below will show that the permissive form of (16) is preferable.
h Is it actually the case that a morphological relation is necessary before semantic redundancy can be considered? Semantic priming might suggest otherwise (assuming kill primes die, for instance). I leave the question open.
morphological and semantic regularities
59
2.5 Other applications The redundancy rules developed so far describe the relation between verbs and nominalizations. It is clear that similar rules can describe de-adjectival nouns (e.g. redness, entirety, width), deverbal adjectives (predictive, explanatory), denominal adjectives (boyish, national, fearless), de-adjectival verbs (thicken, neutralize, yellow), and denominal verbs (befriend, originate, smoke). Likewise, it is easy to formulate rules for nouns with noun roots like boyhood,5 adjectives with adjective roots like unlikely, and verbs with verb roots like re-enter and outlast.6 Note, by the way, that phonological and syntactic conditions such as choice of boundary and existence of internal constituent structure can be spelled out in the redundancy rules. Note also that a complex form such as transformationalist can be accounted for with no difficulty: each of the steps in its derivation is described by a regular redundancy rule. No question of ordering rules or of stating rule features need ever arise (imagine, by contrast, the complexity of Boolean conditions on exceptions which would be needed in the entry for transform, if we were to generate this word in a transformational theory of the lexicon). Transformationalist is fully specified in the lexicon, as are transform, transformation, and transformational. The total information content of the four words is the information that there are four words, plus the information in the word transform, plus idiosyncratic information added by successive steps in derivation (e.g. that transformation in this sense refers to a component of a theory of syntax and not just any change of state, and that a transformationalist in the sense used in this chapter is one who believes in a particular form of transformational theory), plus the cost of referring to the three necessary redundancy rules. Note that the information measure allows morphologically derived lexical items to contain more semantic information than the rule predicts; we use this fact crucially in describing the information content of transformationalist. More striking examples will be given below. With these preliminary observations, I will now present some more diverse applications of the redundancy rules.
2.5.1
Prefix-stem verbs
Many verbs in English can be analyzed into one of the prefixes in-, de-, sub-, ad-, con-, per-, trans-, etc. followed by one of the stems -sist, -mit, -fer, -cede, -cur, 5 The abstractness of nouns in -hood, mentioned by Halle 1973 as an example, is guaranteed by the fact that the associated semantic rule yields the meaning state (or period) of being a Z. 6 Cf. Fraser 1965 [published as Fraser 1974] and Bresnan 1982a for an interesting discussion of this last class of verbs.
60
morphological and semantic regularities
etc. Chomsky and Halle 1968 argue, for phonological reasons, that the prefix and stem are joined by a special boundary ¼. Whether a particular prefix and a particular stem together form an actual word of English seems to be an idiosyncratic fact: (17) transist persist consist assist subsist desist insist
transmit permit commit admit submit demit immit
transfer prefer confer affer suffer defer infer
transcede precede concede accede succeed decede incede
transcur precur concur accur succur decur incur
We would like the information measure of the lexicon to take into account the construction of these words in computing their information content. There are two possible solutions. In the first, the lexicon will contain, in addition to the fully specified lexical entries for each actually occurring prefix-stem verb, a list of the prefixes and stems from which the verbs are formed. The redundancy rules will contain the following morphological rule, which relates three terms:i
(18) /x = y/ +V
/x/ +Prefix /y/ +Stem
The information content of a particular prefix-stem verb will thus be the information that there is a word, plus the semantic content of the verb (since there is no semantic rule to go with (18), at least in most cases), plus the cost of referring to morphological rule (18). The cost of each individual prefix and stem will be counted only once for the entire lexicon. Since we have up to this point been unyielding on the subject of hypothetical lexical items, we might feel somewhat uncomfortable about introducing prefixes and stems into the lexicon. However, this case is somewhat different from
i
In Parallel Architecture notation, this would look something like (18’):
(18’) Phonology: /x2 ¼ y3 /1 Syntax: [V Prefix2 þ Stem3]1
morphological and semantic regularities
61
earlier ones. In the case of perdition, the presumed root is the verb perdite. If perdite were entered in the lexicon, we would have every reason to believe that lexical insertion transformations would insert perdite into deep structures, and that the syntax would then produce well-formed sentences containing the verb perdite. In order to prevent this, we would have to put a feature in the entry for perdite to block the lexical insertion transformations. It is this rule feature [Lexical Insertion] which we wish to exclude from the theory. Consider now the lexical entry for the prefix trans-:
(19)
/trans/ +Prefix
Trans- has no (or little) semantic information, and as syntactic information has only the marker [ þPrefix]. Since the syntactic category Prefix is not generated by the base rules of English, there is no way for trans- alone to be inserted into a deep structure. It can be inserted only when combined with a stem to form a verb, since the category Verb does appear in the base rules. Hence there is no need to use the offending rule feature [Lexical Insertion] in the entry for trans-, and no need to compromise our earlier position on perdite.j However, there is another possible solution which eliminates even entries like (19), by introducing the prefixes and stems in the redundancy rule itself. In this case, the redundancy rule consists of a single term, and may be thought of as the simplest type of ‘word-formation’ rule:
(20)
trans per con aD suB de . . .
=
sist mit fer cede tain . . ..
+V
Since the Parallel Architecture lists morphological affixes such as þion in the lexicon (see (3’) and (13’) ), this argument applies to them as well. j
62
morphological and semantic regularities
The information content of prefix-stem verbs is the same as before, but the cost of the individual prefixes and stems is counted as part of the redundancy rule, not of the list of lexical items. The two solutions appear at this level of investigation to be equivalent, and I know as yet of no empirical evidence to decide which should be permitted by the theory or favored by the evaluation measure. This is the case of morphological redundancy without semantic redundancy promised in section 2.4. Since, for the most part, prefixes and stems do not carry semantic information, it is not possible to pair (18) or (20) with a semantic rule.7 Obviously the information measure must permit the morphological redundancy anyway. Besides complete redundancy, we now have three cases to consider: those in which a semantic redundancy rule relates a word to a nonlexical root (e.g. perdition), those in which the semantic rule relates a word incorrectly to a lexical root (e.g. professor), and those in which there is no semantic rule at all. The three cases are independent, and a decision on one of them need not affect the others. Thus the decision to allow morphological redundancy for prefix-stem verbs still leaves open the question raised in section 2.4 of how to treat professor. It should be pointed out that word-formation rules like (20) are very similar to Halle’s (1973) word-formation rules. The major difference here between his theory and mine is that his lexicon includes, in addition to the dictionary, a list of all morphemes in the language, productive and unproductive. The present theory lists only words in the lexicon.k Productive affixes are introduced as part of lexical redundancy rules, and non-productive non-lexical morphemes (such as perdite) do not appear independently anywhere in the lexical component. Other than the arguments already stated concerning the feature [Lexical Insertion], I know of little evidence to distinguish the two solutions. However, since Halle has not formulated the filter, which plays a crucial role in the evaluation measure for his theory of the lexicon, it is hard to compare the theories on the level where the present theory makes its most interesting claims.
7
If they did carry semantic information, it would be more difficult, but not necessarily impossible, to state the rule in the form of (20). This is a potential difference between the solutions, concerning which I have no evidence at present. [2010: My guess is that (18)–(19) probably generalizes better to a wide range of morphological phenomena.] k
However, in the Parallel Architecture lexicon, like Halle and unlike the present chapter, all the affixes are listed, whether productive or semiproductive!
morphological and semantic regularities
2.5.2
63
Compound nouns
The compound nouns in (21) are all formed by concatenating two nouns: (21)
a. garbage man, iceman, milkman, breadbasket, oil drum b. snowman, gingerbread man, bread crumb, sand castle c. bulldog, kettledrum, sandstone, tissue paper
Although the meaning of each compound is formed from the meanings of the two constituent nouns, the way in which the meaning is formed differs from line to line. Part of a speaker’s knowledge of the English lexicon is the way in which the meanings of compounds are related to the meanings of their constituents: thus we would say that someone did not know English if he (seriously) used garbage man to mean ‘a man made out of garbage’, by analogy with snowman. If one brought Lees 1960 up to date, one would get an approach to compounds which uses transformations to combine nouns randomly, controlled by exception features so as to produce only the existing compounds with the correct meanings. But how can such exception features be formulated? Either noun in a compound can be changed, with a corresponding change in acceptability: we have garbage man, garbage truck, but not garbage gingerbread, garbage tree; we also have garbage man, gingerbread man, but not ant man, tissue man. Thus the use of exception features will require each noun in the lexicon to be cross-listed with every other noun for the compounding transformations. Furthermore, since gingerbread is itself a compound, ginger, bread, and man will all somehow have to be related by the exception features. In the end, the exception features appear to be equivalent to a listing of all the existing compounds along with their meanings. In the lexicalist theory, we can dispense with exception features in the description of compounds. We simply give each actually occurring compound a fully specified lexical entry, and in the list of redundancy rules we enter morphological rule (22) and semantic rules (23a,b,c), describing the data of (21a,b,c) respectively. Of course, there are a great number of additional semantic rules (cf. Lees, chapter 4); I list only these three as a sample:
(22) /[x] [y]/ +N
/x/ +N /y/ +N
64
morphological and semantic regularities
(23) a.
b.
c.
+N Z
+N Z THAT CARRIES W
+N W +N Z
+N Z MADE OF W
+N Z LIKE A W
+N W +N Z +N W
The redundancy rules thus define the set of possible compounds of English, and the lexicon lists the actually occurring compounds.l The information measure (16) gives an intuitively correct result for the independent information in compounds. For example, since the nouns garbage and man are in the lexicon, all their information will be counted as redundant in evaluating the entry for garbage man. Thus the independent information content of garbage man will be the information that such a word exists, plus any idiosyncratic facts about the meaning (e.g. that he picks up rather than delivers garbage), plus the cost of referring to (22) and (23a). The information of a complex compound like gingerbread man is measured in exactly the same way; but the independent information in its constituent gingerbread is reduced because of its relation to ginger and bread. Gingerbread man is thus parallel in its evaluation to the case of transformationalist cited earlier. Now consider the problem of evaluating the following nouns: (24) a. blueberry, blackberry b. cranberry, huckleberry c. gooseberry, strawberry
l As pointed out by Levi 1978, this account is insufficient, as it does not extend to the creation of novel compounds. This difficulty is related to the more general problem of how to deal with regularity, such that not every regular form has to be listed. It is also connected with the duality of phrase structure rules that comes up in section 2.5.4, where phrase structure rules are used not only for the productive generation of sentences, but also as redundancy rules for the syntax of idioms. See Chapter 13 for a more developed account of the syntax and semantics of compounding within the Parallel Architecture.
morphological and semantic regularities
65
Blueberry and blackberry are obviously formed along the lines of morphological rule (25) and semantic rule (26). This combination of rules also forms flatiron, highchair, madman, drydock, and many others.
(25) /[x] [y]/ +N (26) +N Z WHICH IS W
/x/ +A /y/ +N +N Z +A W
Thus blueberry and blackberry are evaluated in exactly the same way as garbage man. Cranberry and huckleberry contain one lexical morpheme and one nonlexical morpheme. The second part (-berry) and its associated semantics should be redundant, but the phonological segments /kræn/ and /hçkl/, and the semantic characteristics distinguishing cranberries and huckleberries from other kinds of berries, must be non-redundant. Hence this case is just like perdition, where a non-lexical root is involved, and the information measure formulated for the case of perdition will yield the intuitively correct result. One problem is that the lexical categories of cran- and huckle- are indeterminate, so it is unclear which morphological rule applies. Likewise, it is unclear which semantic rule applies. However, I see nothing against arbitrarily applying the rules which cost least; this convention will minimize the information in the lexicon without jeopardizing the generality of the evaluation procedure. We observe next that gooseberry and strawberry contain two lexical morphemes and are both berries, but gooseberries have nothing to do with geese and strawberries have nothing to do with straw. This case is thus like professor, which has nothing to do semantically with the verb profess, and exactly the same question arises in their evaluation: should they be intermediate in cost between the previous two cases, or should they be evaluated like cranberry, with straw- and goose- counted as non-redundant? The fact that there is pressure towards phonological similarity even without a semantic basis (e.g. gooseberry was once groseberry) is some evidence in favor of the permissive form of (16), in which morphological similarity alone is sufficient for redundancy.
66
morphological and semantic regularities
Another semantic class of compound nouns (exocentric compounds) differs from those mentioned so far in that neither constituent describes what kind of object the compound is. For example, there is no way for a non-speaker of English to know that a redhead is a kind of person, but that a blackhead is a kind of pimple.8 Other examples are redwing (a bird), yellow jacket (a bee), redcoat (a soldier), greenback (a bill), bigmouth (a person), and big top (a tent). The morphological rule involved is still (25); the semantic rule must be (27).
(27)
+N Z
+N THING WITH A
Z WHICH IS W
+A W
This expresses the generalization inherent in these compounds, but it leaves open what kind of object the compound refers to. The information measure gives as the cost of redhead, for example, the information that there is a word, plus the information that a redhead is a person (a more fully specified form of thing in (27) ), plus the cost of referring to (27). This evaluation reflects precisely what a speaker must learn about the word. A transformational theory of compound formation, on the other hand, encounters severe complication with this class of compounds. Since a compounding transformation must preserve functional semantic content, the underlying form of redhead must contain the information that a redhead is a person and not a pimple, and this information must be captured somehow in rule features (or derivational constraints) which are idiosyncratic to the word redhead. I am sure that such constraints can be formulated, but it is not of much interest to do so. The need for these elaborate rule features stems from the nature of transformations. Any phrase-marker taken as input to a particular transformation corresponds to a fully specified output phrase-marker. In the case of exocentric compounds, the combination of the two constituent words by rule (27) does not fully specify the output, since the nature of thing in (27) is inherently indeterminate. We thus see an important empirical difference between lexical redundancy rules and transformations: it is quite natural and typical for lexical redundancy rules to relate items only partially, whereas transformations cannot express partial relations. Several illustrations of this point have appeared already, in the morphological treatment of perdition and cranberry and in the semantic treatment of transformationalist. However, the case of exocentric compounds is 8
I am grateful to Phyllis Pacin for this example.
morphological and semantic regularities
67
perhaps the most striking example, since no combination of exception features and hypothetical lexical items can make the transformational treatment appear natural. The lexicalist treatment, since it allows rules to relate exactly as much as necessary, handles exocentric compounds without any remarkable extensions of the machinery.m
2.5.3
Causative verbs
There is a large class of verbs which have both transitive and intransitive forms:9 (28) a. b. (29) a. b. (30) a. b.
The door opened. Bill opened the door. The window broke. John broke the window. The coach changed into a pumpkin. Mombi the witch changed the coach from a handsome young man into a pumpkin.
It has long been a concern of transformational grammarians to express the fact that the semantic relations of door to open, of window to break, and of coach to change are the same in the transitive and intransitive cases. There have been two widely accepted approaches, both transformational in character. The first, that of Lakoff 1970, claims that the underlying form of the transitive sentence contains the intransitive sentence as a complement to a verb of causation—i.e. that the underlying form of (28b) is revealed more accurately in the sentence Bill caused the door to open. [2010: This approach has been resurrected under new auspices in Hale and Keyser 1993, 2002.] The other approach, case grammar, is that of Fillmore 1968. It claims that the semantic relation of door to open is expressed syntactically in the deep structures of (28a) and (28b), and that the choice of subject is a purely surface fact. The deep structures are taken to be (31a) and (31b) respectively: (31)
a. past open [Objective the door] b. past open [Objective the door] [Agentive by Bill]
9 This class may also include the two forms of begin proposed by Perlmutter 1970. [2010: In retrospect, no, it doesn’t. See Culicover and Jackendoff 2005, p. 201.] m Here is a premonition of the break with transformations and with a syntax-based theory of grammar. In retrospect, we might see the proposal for lexical redundancy rules here as a new mechanism for capturing grammatical phenomena, which over the years gradually encroaches on the hegemony of transformations, culminating (in my thought at least) in the Parallel Architecture and Simpler Syntax. Of course, I was not alone in growing skeptical about the efficacy of transformations around this time: currents were also beginning to stir that led to LFG and GPSG, thoroughly transformationless approaches to syntax.
68
morphological and semantic regularities
These proposals and their consequences have been criticized on diverse syntactic and semantic grounds (cf., e.g., Chomsky 1972, Fodor 1970, and Jackendoff 1972, chapter 2); I do not intend to repeat those criticisms here. It is of interest to note, however, that Lakoff’s analysis of causatives is the opening wedge into the generative semanticists’ theory of lexicalization: if the causative verb break is the result of a transformation, we would miss a generalization about the nature of agentive verbs by failing to derive the causative verb kill by the same transformation. But since kill has as intransitive parallel not kill but die, and since there are many such causative verbs without morphologically related intransitives, the only way to avoid an embarrassing number of exceptions in the lexicon is to perform lexical insertion after the causative transformation, as proposed by McCawley 1968. Again, the difficulty in this solution lies in the nature of transformations. There are two cross-classifying generalizations which a satisfactory theory must express: all causative verbs must share a semantic element in their representation; and the class of verbs which have both a transitive causative form and an intransitive noncausative form must be described in a general fashion. Expressing the second generalization with a transformation implies a complete regularity, which in turn loses the first generalization; McCawley’s solution is to make a radical move to recapture the first generalization. There remains the alternative of expressing the second generalization in a way that does not disturb the first. Fillmore’s solution is along these lines; but he still requires a radical change in the syntactic component, namely the introduction of case markers. The lexicalist theory can leave the syntactic component unchanged by using the power of the lexicon to express the partial regularity of the second generalization. The two forms of break are assigned separate lexical entries:
(32)
/breyk/ +V a. +[NP1—— ] NP1 BREAK /breyk/ +V b. +[NP2 —— NP1] NP2 CAUSE (NP1 BREAK)
morphological and semantic regularities
69
The two forms are related by the following morphological and semantic rules:10
(33)
/x/ +V
a. /x/ +V b. +V +[NP1—— ] NP1 W
+V +[NP2 —— NP1] NP2 CAUSE (NP1W)
Thus the independent information contained in the two entries for break is the fact that there are two words,11 plus the independent information in the intransitive form (32a), plus the cost of referring to the redundancy rules. Hence the relation between the (a) and (b) sentences in (28)–(30) is expressed in the lexicon and not in the transformational component.n This solution permits us still to capture the semantic similarity of all causative verbs in their lexical entries; thus die and kill will have entries (34a) and (34b) respectively:
10 Since (33a) is an identity rule, it is possibly dispensable. I have included it here for the sake of explicitness, and also in order to leave the form of the information measure unchanged. 11 Perhaps the use of the identity rule (33a) could make the two words count as one, if this were desirable. I have no intuitions on the matter, so I will not bother with the modification. n This case is interesting in the Parallel Architecture notation. If the causative verbs are thought of as carrying a zero affix, (33a, b) come out as follows:
(33’) a. Phonology: Syntax: b. Syntax: Semantics:
/x2/1 [V V2 þ Af]1 [V V2 þ Af]1 [Event CAUSE (X, [F (Y)]2 )]1
If causative verbs are thought of as the same verb as their noncausatives, phonology need not even be invoked. In this case something like (33’) will do the job. Here the same syntactic form is coindexed to two different semantic possibilities. (33’) Syntax: V1 or 2 Semantics: [Event CAUSE (X, [F (Y)]2 )]1 Jackendoff 1990 (section 4.1) notates this possibility with a dotted line under the optional outer function and arguments, here CAUSE (X, . . . ), and presents a few other cases where this notation proves useful.
70
(34)
morphological and semantic regularities
¯ /dı/ +V a. +[NP1—— ] NP1 DIE /kil/ +V b. +[NP2 —— NP1] NP2 CAUSE (NP1 DIE)
Die and kill are related semantically in exactly the way as the two entries of break: one is a causative in which the event caused is the event described by the other. However, since there is no morphological rule relating (34a–b), the information measure does not relate them; the independent information contained in the two entries is the fact that there are two words, plus all the information in both entries. Thus the lexicalist theory successfully expresses the relation between the two breaks and their relation to kill and die, without in any sense requiring kill and die to be exceptional, and without making any radical changes in the nature of the syntactic component. A further possibility suggested by this account of causative verbs is that the partial regularities of the following examples from Fillmore are also expressed in the lexicon: (35) a. Bees swarmed in the garden. We sprayed paint on the wall. b. The garden swarmed with bees. We sprayed the wall with paint. Fillmore seeks to express these relationships transformationally, but he encounters the uncomfortable fact that the (a) and (b) sentences are not synonymous: the (b) sentences imply that the garden was full of bees and that the wall was covered with paint, but the (a) sentences do not carry this implication. Anderson 1971 shows that this semantic difference argues against Fillmore’s analysis, and in favor of one with a deep-structure difference between the (a) and (b) sentences. A lexical treatment of the relationship between the two forms of swarm and spray could express the difference in meaning, and would be undisturbed by the fact that some verbs, such as put, have only the (a) form and meaning, while others, such as fill, have only the (b) form and meaning. This is precisely parallel to the break–break vs. die–kill case just discussed. [2010: The last thirty years have seen much more discussion of these cases, especially spray. See Chapter 6.]
morphological and semantic regularities
71
Consider also some of the examples mentioned in Chomsky 1970. The relation of He was amused at the stories and The stories amused him can be expressed in the lexicon, and no causative transformation of the form Chomsky proposes need be invoked. [2010: See Jackendoff 2007a, chapter 7, for discussion of the semiproductivity of this lexical relation.] The nominalization his amusement at the stories contrasts with the stories’ amusement of him because amusement happens to be most directly related to the adjectival amused at rather than to the verb amuse. [2010: I now find this explanation rather fishy. Sorry.] Other causatives do have nominalizations, e.g. the excitation of the protons by gamma rays. I take it then that the existence of only one of the possible forms of amusement is an ad hoc fact, expressed in the lexicon. Chomsky also cites the fact that the transitive use of grow, as in John grows tomatoes, does not form the nominalization the growth of tomatoes by John. Rather the growth of tomatoes is related to the intransitive tomatoes grow. Again we can express this fact by means of lexical relations. This time, the relation is perhaps more systematic than with amusement, since nouns in -th, such as width and length, are generally related to intransitive predicates. Thus the meaning of growth can be predicted by the syntactic properties of the redundancy rule which introduces the affix -th. The transitive grow does in fact have its own nominalization: the growing of tomatoes by John. Thus Chomsky’s use of causatives as evidence for the Lexicalist Hypothesis seems incorrect—in that causatives do have nominalizations, contrary to his claim. But we can account for the unsystematicity of the nominalizations, as well as for what regularities do exist, within the present framework. Note also that our account of causatives extends easily to Lakoff’s class of inchoative verbs (1970). For example, the relation of the adjective open to the intransitive verb open (‘become open’) is easily expressed in a redundancy rule similar to that proposed for causatives. As further evidence for the lexicalist theory, consider two forms of the verb smoke: (36)
a. The cigar/The chimney smoked. b. John smoked the cigar/the chimney.
The intransitive verb smoke means ‘give off smoke’; it is related to the noun smoke by a redundancy rule that applies also to the appropriate senses of steam, smell, piss, flower, and signal. The transitive form of smoke in the sense of (36b) is partially related to the intransitive form by (33) in that it means ‘cause to give off smoke’, but it contains additional information—something like ‘by holding in the mouth and puffing’. This information is not predictable from the redundancy rule, but it provides the clue to the anomaly of John smoked the chimney
72
morphological and semantic regularities
(so, if John were a giant, he might well use a chimney like a pipe, and then the sentence might be acceptable). A transformational theory has no way to capture this partial generalization without artificiality. The lexicalist theory simply counts the unpredictable information as non-redundant, and the predictable information as redundant. While we are on the subject of smoke, it may be interesting to point out some other senses of smoke as illustration. Call the noun smoke1, and the intransitive and transitive senses just discussed smoke2 and smoke3 respectively. There is another transitive verb smoke4, which means ‘permeate or cover with smoke’ as in John smoked the ham. The redundancy rule relating smoke4 to smoke1 is also seen in verbs like paint, another sense of steam, water (in water the garden), and powder (as in powder your nose), flour, and cover. There is another intransitive smoke5, meaning ‘smoke3 something’. The ambiguity in John is smoking is between smoke2 and smoke5. Smoke5 is related to smoke3 by a redundancy rule that also handles two forms of eat, drink, draw, read, cook, and sing. From smoke3 we also get the nominalization smoke6, ‘something that is smoked3’ (e.g. A cigar is a good smoke) by the redundancy rule that also gives the nouns drink, desire, wish, dream, find, and experience. The verb milk (as in milk a cow) is related to the noun as smoke1 and smoke3 are related, but without an intermediate The cow milked (‘The cow gave off milk’); the relation between the two milks requires two sets of redundancy rules used together. We thus see the rich variety of partial regularities in lexical relations: their expression in a transformational theory becomes hard to conceive, but they can be expressed quite straightforwardly in the lexicalist framework.
2.5.4
Idioms
Idioms are fixed syntactic constructions which are made up of words already in the lexicon, but which carry meanings independent of the meanings of their constituents. Since the meanings are unpredictable, the grammar must represent a speaker’s knowledge of what constructions are idioms and what they mean. The logical place to list idioms is of course in the lexicon, though it is not obvious that the usual lexical machinery will suffice. Fraser 1970 discusses three points of interest in the formalization of idioms. First, they are constructed from known lexical items; the information measure, which measures how much the speaker must learn, should reflect this. Second, they are for the most part constructed in accordance with known syntactic rules (with a few exceptions such as by and large), and in accordance with the
morphological and semantic regularities
73
syntactic restrictions of their constituents. Third, they are often resistant to normally applicable transformations; e.g., The bucket was kicked by John has only the non-idiomatic reading. I have nothing to say about this third consideration, but the first two can be expressed in the present framework without serious difficulty.o Let us deal first with the question of the internal structure of idioms. Since we have given internal structure to items like compensation and permit, there seems to be nothing against listing idioms too, complete with their structure. The only difference in the lexical entries is that the structure of idioms goes beyond the word level. We can thus assign the lexical entries in (37) to kick the bucket, give hell to, and take to task.12
(37)
a.
NP1 [VP [V kik] [NP[Art D´][N b k´t]]] NP1 DIE
b.
NP1 [VP [V giv] [NP[N hel]][PP [P tuw] NP2]] NP1 YELL AT NP2
c.
NP1 [VP [V teyk] NP2[PP [P t´][NP [N tœsk]]]] NP1 CRITICIZE NP2
The lexical insertion rule will operate in the usual way, inserting the lexical entries onto deep phrase markers that conform to the syntactic structure of the lexical entries. Since the structure of the entries goes beyond the word level, the idiom must be inserted onto a complex of deep-structure nodes, in contrast to ordinary words which are inserted onto a single node.p
12
The normal notation for strict subcategorization restrictions is difficult to apply in this case, so I have for convenience adopted a notation in which the strict subcategorization conditions are combined with the phonological and syntactic representations, in an obvious fashion. No particular theoretical significance is intended by the change in notation. This proposal, which appears to be much like that of Katz 1973, was arrived at independently. o Jackendoff 1997b extends the argument for lexical storage of idioms to cliche´s and other fixed expressions, including things like memorized poems. Nunberg, Sag, and Wasow 1994 observe that some idioms, rather than being resistant to transformations, have structures that a transformational theory would have to derive, for instance fit to be tied, a tough nut to crack, X is not what it/he/she’s cracked up to be. p This treatment of subcategorization and of lexical insertion—including the possibility of discontinuous lexical items—becomes the standard treatment of subcategorization in Semantic Structures and in the Parallel Architecture, where lexical entries are unified with branches of phrasal nodes, rather than being inserted. Thus this is another quiet break with the unspoken tradition of algorithmic generation of syntactic structure. The differences between the two approaches did not become clear until much later: the algorithmic approach to lexical insertion morphed over the years into Merge; the approach adopted here morphed into Unification. See Jackendoff 2002, chapter 6, and especially Jackendoff 2008.
74
morphological and semantic regularities
As with ordinary lexical entries, the strictly subcategorized NPs must have a specific grammatical relation with respect to the entry, and this is indicated in the entries of (37). In the case of take NP to task, the strictly subcategorized direct object is in fact surrounded by parts of the idiom; i.e., the idiom is discontinuous. But in the present theory, this appears not to be cause for despair, as our formalisms seem adequate to accommodate a discontinuous lexical item. This last observation enables us to solve a puzzle in syntax: which is the underlying form in verb-particle constructions, look up the answer or look the answer up? The standard assumption (cf. Fraser 1965) is that the particle has to form a deep-structure constituent with the verb in order to formulate a lexical entry; hence look up the answer is underlying, and the particle movement transformation is a rightward movement. But Emonds 1972 gives strong syntactic evidence that the particle movement rule must be a leftward movement. He feels uncomfortable about this result because it requires that look . . . up be discontinuous in deep structure; he consoles himself by saying that the same problem exists for take . . . to task, but does not provide any interesting solution. Having given a viable entry for take . . . to task, we can now equally well assign discontinuous entries to idiomatic verb-particle constructions, vindicating Emonds’s syntactic solution. [See more extensive discussion of verb-particle constructions in Chapter 7.] By claiming that the normal lexical insertion process deals with the insertion of idioms, we accomplish two ends. First, we need not complicate the grammar in order to accommodate idioms. Second, we can explain why idioms have the syntactic structure of ordinary sentences: if they did not, the lexical insertion rules could not insert them onto deep phrase markers. Our account of idioms thus has the important virtue of explaining a restriction in terms of already existing conventions in the theory of grammar—good evidence for its correctness.q Now that we have provided a way of listing idioms, how can we capture the speaker’s knowledge that idioms are made up of already existing words? To relate the words in the lexicon to the constituents of idioms, we need morphological redundancy rules. The appropriate rules for kick the bucket must say that a verb followed by a noun phrase forms a verb phrase, and that an article followed by a noun forms a noun phrase. But these rules already exist as phrase-structure rules for VP and NP. Thus, in the evaluation of idioms, we must use the phrase-structure rules as morphological redundancy rules. If this is
q This statement disregards the idioms with anomalous syntactic structures such as by and large, for the most part, day in day out, how about X?, and so on. Presumably these cost more because their syntactic structure is not redundant with normal syntactic rules, as described in the next paragraph of the text.
morphological and semantic regularities
75
possible, the independent information in kick the bucket will be the information that it is a lexical entry, plus the semantic information die, plus the cost of referring to the phrase-structure rules for VP and NP. Though mechanically this appears to be a reasonable solution, it raises the disturbing question of why the base rules should play a role in the information measure for the lexical component. Some discussion of this question will appear in section 2.7. At this point I will simply note that this solution does not have very drastic consequences for grammatical theory. Since the base rules can be used as redundancy rules only if lexical entries go beyond the word level, no descriptive power is added to the grammar outside the description of idioms. Therefore the proposal is very limited in scope, despite its initially outrageous appearance.r If the base rules are used as morphological redundancy rules for idioms, we might correspondingly expect the semantic projection rules to be used as semantic redundancy rules. But of course this cannot be the case, since then an idiom would have exactly its literal meaning, and cease to be an idiom. So we must assume that the permissive version of the information measure is being used: both morphological and semantic redundancy rules exist, but only the morphological rules apply in reducing the independent information in the idiom. This is further evidence that the permissive version of the information measure must be correct. Note, by the way, that a transformational theory of nominalization contains absolutely no generalization of the approach that accounts for idioms. Thus the lexicalist hypothesis proves itself superior to the transformational hypothesis in a way totally unrelated to the original arguments deciding between them.
2.6 The cost of referring to redundancy rules In evaluating the independent information of lexical entries, we have continually included the cost of referring to redundancy rules. We have not so far specified how to calculate this cost, or how to relate it quantitatively to other costs in the lexicon. In this section I will propose some preliminary answers to those questions. In the discussion of the full-entry theory in section 2.2, I said that the cost of referring to a redundancy rule in evaluating a lexical entry represents one’s knowledge of which regularities hold in that particular lexical entry. In order to be more specific, let us reconsider the meaning of the information measure in the
r
Again, this is presented as a harmless change, but in retrospect it is drastic!
76
morphological and semantic regularities
full-entry theory. In measuring the independent information contained in a lexical entry, we are in effect measuring how much new information one needs in order to learn that lexical item. If the lexical item is totally unrelated to anything else in the lexicon, one must learn it from scratch. But if there is other lexical information which helps one know in advance some of the properties of the new word, there is less to learn; this is captured in clause (b) of the information measure.s In learning that a new lexical item can be formed on the basis of an old lexical item and a redundancy rule, however, something must be learned besides the identity of the old lexical item: namely, which redundancy rule to apply. For example, part of one’s knowledge of the lexicon of English is the fact that the nominalizations of refuse and confuse are refusal and confusion, not refusion and confusal, although in principle the latter forms could exist. That is, in learning the words refusal and confusion, one must learn the arbitrary fact that, of the choice of possible nominal affixes, refuse uses -al and confuse uses -ion. Clause (c) of the information measure, the cost of referring to the redundancy rule, is meant to represent this knowledge. I am claiming therefore that the evaluation of refusal must take into account the fact that it, and not refusion, is the proper nominalization of refuse. For a clear case of the use of clause (c), let us turn to another example. Botha 1968 discusses the process of nominal compounding in Afrikaans, which contains many compounds which are morphologically simple concatenations of two nouns, as in English. But there are also many compounds in which the two nouns are joined by a ‘link phoneme’ s or @. Botha demonstrates at great length that there is no phonological, morphological, syntactic, or semantic regularity in the use of link phonemes; i.e. the link phoneme must be learned as an idiosyncrasy of each individual compound. In the present theory, the Afrikaans lexicon contains three morphological rules for noun compounds:t
s Other possible ways of thinking about the information measure: (1) the ‘carrying cost’ of ‘stocking’ one’s lexicon with this item; (2) the ease of accessing the item. The latter of these is the criterion of Moscoso del Prado Martı´n, Kostic´, and Baayen 2004, where the theory is tested against speed of lexical access. I’m not sure this can be the only criterion, and I’m also not sure how to combine it with the other two. t In the Parallel Architecture format, these redundancy rules, being themselves lexical entries, would in turn fall under the information metric. That is, the three Afrikaans forms should collectively cost less than a collection of totally dissimilar forms such as [N1-N2], [N2-fa-N1], and [mis-N1-ple-N2]. The generalization could be captured by a higher-level redundancy rule of the form (38’).
(38’) Phonology: Syntax:
/x1 – (seg) – x2 / N1 – N2
morphological and semantic regularities
(38) a.
b.
c.
/[x] [y]/ +N
/[x] s [y]/ +N
/[x] [y]/ +N
77
/x/ +N /y/ +N /x/ +N /y/ +N /x/ +N /y/ +N
Since all the morphological information of a particular compound is predicted by one of the three rules in (38), clause (b) of the information measure contributes nothing to the information content of the compound. But since the speaker must learn which of the three is appropriate, clause (c) must contribute the cost of the information involved in making this choice. A third example involves inflectional morphology. Halle 1973 argues that paradigmatic information should be represented in the dictionary, and in fact that only and all fully inflected forms should be entered. As a consequence, the lexical insertion rules must enter partial or complete paradigms into deep structures, and the rules of concord must have the function of filtering out all but the correct forms, rather than that of inserting inflectional affixes.13 Under Halle’s proposal, part of the task of the lexical component of English is to list the correspondences between the present and past tense forms of verbs. Accordingly, we can state a few morphological redundancy rules relating present to past tense forms in English:
13 This of course requires rules of concord to be of a different formal nature than ordinary transformations. But perhaps this is not such a bad result, considering that the most convincing cases for Lakoff’s global rules seem to be in this area. An independent argument that concord rules differ formally from transformations could serve as evidence that transformations need not be global: only the very limited class of concord rules, which are no longer transformations at all, need information from various levels of derivation. This more highly structured theory reduces the class of possible grammars.
78
(39)
morphological and semantic regularities
a.
/x/ +[V+pres]
/x#d/ +[V+past]
b.
/C0VC0/ +[V+pres]
/C0VC0 + t/ +[V+past]
c.
d.
V /C0 · back C0/ · round
V /C0 ñ· back C0/ ñ· round
+[V+pres]
+[V+past]
/C0VC0/ +[V+pres]
/C0ox + d/ +[V+past]
Here (39a) is the regular rule for forming past tenses, and the other three represent various irregular forms: (39b) relates keep–kept, dream–dreamt, lose– lost, feel–felt, etc.; (39c) relates cling–clung, hold–held, break–broke, etc.; the very marginal and strange (39d) relates just the six pairs buy–bought, bring– brought, catch–caught, fight–fought, seek–sought, and think–thought. Note that (39b–c) take over the function of the ‘pre-cyclic readjustment rules’ described by Chomsky and Halle (1968, pp. 209–10).14,u
14 I have not considered the question of how to extend the phonological generalization of (39c) to other alternations such as mouse–mice, long–length. Perhaps the only way to do this is to retain the rule in the phonology, and simply let the lexical redundancy rule supply a rule feature. But a more sophisticated account of the interaction of the morphological rules might capture this generalization without a rule feature; e.g. one could consider factoring morphological rules into phonological and syntactic parts, as we factored out separate morphological and semantic rules in section 2.4. In any event, I am including all the phonology in (39) because many people have been dissatisfied with the notion of readjustment rules: I hope that bringing up an alternative may stimulate someone to clarify the notion. u The idea that the lexicon contains full paradigms, whether regular or not, I now take to be a mistake (see Remarks preceding this chapter). In the Parallel Architecture, regular affixation is accomplished by free combination of stems with affixes (see section 1.4). The ‘filtering function’ of rules of concord eventually led to the idea of a ‘second call to the lexicon’ or alternatively ‘late lexical insertion’ at some point in the derivation after agreement applies. This notion eventually matured into Distributed Morphology (Halle and Marantz 1993). An alternative approach, in which all rules of grammar have a ‘filtering’ rather than an ‘algorithmic’ function, turned into constraint-based grammar formalisms, which encompasses practically every other approach to generative grammar besides the narrow Chomskyan tradition. Rules (39b, c, d), which involve ablaut rather than affixation, cannot be stated in the standard Parallel Architecture format. This remains an important challenge to filling out the framework so as to encompass morphophonology.
morphological and semantic regularities
79
A final preliminary point in this example: in the evaluation of a paradigm by the information measure, I assume that the information that a word exists is counted only once for the entire paradigm. Although one does have to learn whether a verb has a nominalization, one knows for certain that it has a past tense, participles, and a conjugation. Therefore the information measure should not count knowledge that inflections exist as anything to be learned. Now let us return to the problem of measuring the cost of referring to a redundancy rule. Intuitively, the overwhelmingly productive rule (39a) should cost virtually nothing to refer to;v the overwhelmingly marginal rules (39b–d) should cost a great deal to refer to, but less than the information they render predictable. The disparity in cost reflects the fact that, in choosing a past tense form, (39a) is ordinary and unremarkable, so one must learn very little to use it; but the others are unusual or ‘marked’ choices, and must be learned. We might further guess that (39b–c), which each account for a fair number of verbs, cost less to refer to than (39d), which applies to only six forms (but which is nevertheless perceived as a minor regularity). Still, the pair buy–bought contains less independent information than the totally irregular pair go–went, which must be counted as two independent entries. These considerations lead to a formulation of the cost of reference something like (40). (40)
The cost of referring to redundancy rule R in evaluating a lexical entry W is IR,W PR,W, where IR,W is the amount of information in W predicted by R, and PR,W is a number between 0 and 1 measuring the regularity of R in applying to the derivation of W.
For an altogether regular rule application, such as the use of (39a) with polysyllabic verbs, PR,W will be zero. With monosyllabic verbs and (39a), PR,W will be almost but not quite zero; the existence of alternatives means that something must be learned. For (39b–d), PR,W will be close to 1; their being irregular means that their use does not reduce the independent information content of entries nearly as much as (39a). In particular, (39d) will reduce the independent information content hardly at all. In fact, it is quite possible that the total information saved by (39d) in the evaluation of the six relevant pairs of lexical entries is less than the cost of stating the rule. Our evaluation measure
v In fact, if regular forms are not listed at all, they shouldn’t add anything to the information measure for the lexicon. They do however exact a cost in lexical access, namely the cost of composing them from forms that are listed.
80
morphological and semantic regularities
thus reflects the extremely marginal status of this rule. In other cases, perhaps the nominalizing affixes and Afrikaans compounds, the various possible derived forms are in more equal competition, and PR,W will have a value of, say, 0.3. I will not suggest a precise method of calculating PR,W, as I believe it would be premature. However, the general concept of how it should be formulated is fairly clear. Count a lexical pair related by R as an actual use of R. Count a lexical entry which meets one term of the structural description of R, but in whose evaluation R plays no role, as a non-use of R. For example, confuse counts as a non-use of the rule introducing the -al nominal affix, since it meets the structural description of the verbal term of the rule, but there is no noun confusal. The sum of the actual uses and the non-uses is the number of potential uses of R. PR,W should be near zero when the number of actual uses of R is close to the number of potential uses; PR,W should be near 1 when the number of actual uses is much smaller than the number of potential uses; and it should rise monotonically from the former extreme to the latter. If phonological conditions can be placed on the applicability of a redundancy rule, PR,W decreases; i.e. the rule becomes more regular. For example, if the actual uses of (39b) all contain mid vowels (as I believe to be the case), then this specification can be added to the vowel in (39b), reducing the potential uses of the rule from the number of monosyllabic verbs to the number of such verbs with mid vowels. Since the number of actual uses of the rule remains the same, PR,W is reduced; and, proportionately, so is the cost of referring to (39b) in the derivations where it is involved.w It is obvious that this concept of PR,W must be refined to account for derivations such as perdition with non-lexical sources; for compounding, where the number of potential uses is infinite because compounds can form parts of compounds; and for prefix-stem verbs, where the lexical redundancy rule does not relate pairs of items. Furthermore, I have no idea how to extend the proposal to the evaluation of idioms, where the base rules are used as lexical redundancy rules. Nevertheless, I believe the notion of regularity of a lexical rule and its role in the evaluation measure for the lexicon is by this point coherent enough to satisfy the degree of approximation of the present theory.
w Many of these same factors go into the measure of information content calculated by Moscoso del Prado Martı´n et al. 2004. Their measure is based largely on token frequency rather than type frequency, which I believe is what I had in mind, trying to think back to a time when no one was thinking about frequency at all. It is still unclear how to extend this measure to the cases mentioned in the next paragraph of the text.
morphological and semantic regularities
81
2.7 Creativity in the lexicon and its implicationsx The accepted view of the lexicon is that it is simply a repository of learned information. Creativity is taken to be a product of the phrase-structure rules and transformations. That is, the ability of a speaker to produce and understand new sentences is ascribed to his knowledge of a productive set of rules which enable him to combine a fixed set of memorized words in infinitely many ways. If we were to adhere to this view strictly, it would be difficult to accept the treatment of the lexicon proposed here. For example, it is quite common for someone to invent a new compound noun spontaneously and to be perfectly understood. This creative use of the compound rule, we would have to argue, is evidence that compounding must be a transformational process rather than a lexical one. This conclusion would fly in the face of all the evidence in section 2.5.2 against a transformational account of compounds. The way out of the dilemma must be to follow the empirical evidence, rather than our preconceived notions of what the grammar should be like. We must accept the lexicalist account of compounds, and change our notion of how creativity is embodied in the grammar. The nature of the revision is clear. Lexical redundancy rules are learned from generalizations observed in already known lexical items. Once learned, they make it easier to learn new lexical items: we have designed them specifically to represent what new independent information must be learned. However, after a redundancy rule is learned, it can be used generatively, producing a class of partially specified possible lexical entries. For example, the compound rule says that any two nouns N1 and N2 can be combined to form a possible compound N1N2. The semantic redundancy rules associated with the compound rule provide a finite range of possible readings for N1N2. [See Chapter 13 for how I would now modify this judgment.] If the context is such as to disambiguate N1N2, any speaker of English who knows N1 and N2 can understand N1N2 whether he has heard it before or not, and whether it is an entry in his lexicon or not. Hence the lexical rules can be used creatively, although this is not their usual role. In section 2.5.4, I proposed that the description of idioms uses the phrasestructure rules as lexical redundancy rules. In broader terms, the rules normally used creatively are being used for the passive description of memorized items. Perhaps this change in function makes more sense in light of the discussion
x In retrospect, this entire section is eerily prescient of the Parallel Architecture, in particular its dissolution of the lexicon–grammar distinction.
82
morphological and semantic regularities
here: it is a mirror image to the creative use of the normally passive lexical redundancy rules. We have thus abandoned the standard view that the lexicon is memorized and only the syntax is creative. In its place we have a somewhat more flexible theory of linguistic creativity. Both creativity and memorization take place in both the syntactic and the lexical component. When the rules of either component are used creatively, no new lexical entries need be learned. When memorization of new lexical entries is taking place, the rules of either component can serve as an aid to learning. However, the normal mode for syntactic rules is creative, and the normal mode for lexical rules is passive. Is there, then, a strict formal division between phrase-structure rules and morphological redundancy rules, or between the semantic projection rules of deep structure and the semantic redundancy rules? I suggest that perhaps there is not, and that they seem so different simply because of the differences in their normal mode of operation. These differences in turn arise basically because lexical rules operate inside words, where things are normally memorized, while phrase-structure rules operate outside words, where things are normally created spontaneously. One might expect the division to be less clear-cut in a highly agglutinative language, where syntax and morphology are less separable than in English. To show that the only difference between the two types of rules is indeed in their normal modes of operation, one would of course need to reconcile their somewhat disparate notations and to show that they make similar claims. Though I will not carry out this project here [but see Chapter 1!], it is important to note, in the present scheme, that the syntactic analogue of a morphological redundancy rule is a phrase-structure rule, not a transformation. This result supports the lexicalist theory’s general trend toward enriching the base component at the expense of the transformational component.15
2.8 Summary This chapter set out to provide a theory of the lexicon that would accommodate Chomsky’s theory of the syntax of nominalizations. This required a formalization of the notion ‘separate but related lexical entries’. The formalization developed uses redundancy rules not for part of the derivation of lexical entries, but for part of their evaluation. I take this use of redundancy rules to be a major theoretical innovation of the present approach. 15 Halle 1973 argues for a view of lexical creativity very similar to that proposed here, on similar grounds.
morphological and semantic regularities
83
In turn, this use of redundancy rules entails the formulation of a new type of evaluation measure. Previous theories have used abbreviatory notations to reduce the evaluation measure on the grammar to a simple count of symbols. But we have seen that the usual notational conventions cannot capture the full range of generalizations in the lexicon. Accordingly I have formulated the evaluation measure as a minimization of independent information, measured by the rather complex function (16) and its refinement in (40). The abandonment of the traditional type of evaluation measure is a second very crucial theoretical innovation required for an adequate treatment of the lexicon.16 The concept of lexical rules that emerges from the present theory is that they are separated into morphological and semantic redundancy rules. The M-rules must play a role, and the S-rules may, in every lexical evaluation in which entries are related. Typically, the redundancy rules do not completely specify the contents of one entry in terms of another, but leave some aspects open. This partial specification of output is a special characteristic of lexical redundancy rules not shared by other types of rules; I have used this characteristic frequently in arguing against transformational solutions. In the discussion of nominalizations, I have taken great pains to tailor the information measure to our intuitions about the nature of generality in the lexicon. In particular, attention has been paid to various kinds of lexical derivatives with non-lexical sources, since these form an important part of the lexicon which is not accounted for satisfactorily in other theories. While our solutions were developed specifically with nominalizations in mind, there is little trouble in extending them to several disparate areas in the lexicon. I have shown that parallel problems occur in these other areas, and that the solution for nominalizations turns out to be applicable. Insofar as the success of a theory is measured by how easily it generalizes to other problems, this theory thus seems quite successful for English. A more stringent test would be its applicability to languages where morphology plays a much more central role. Another measure of a theory’s success is its salutary effect on other sectors of the theory of grammar. The most important effect of the present theory is to eliminate a major part of the evidence for Lakoff’s theory of exceptions to transformations (1970): the lexicon has been set up to accommodate comfortably 16 One might well ask whether the traditional evaluation measure has inhibited progress in other areas of the grammar as well. I conjecture that the approach to marking conventions in SPE (Chapter 9) suffers for this very reason: Chomsky and Halle set up marking conventions so that more ‘natural’ rules save symbols. If, instead, the marking conventions were used as part of an evaluation measure on a set of fully specified rules, a great deal of their mechanical difficulty might well be circumvented in expressing the same insights.
84
morphological and semantic regularities
both regular and ad hoc facts, with no sense of absolute exceptionality; and transformations are not involved in any event. Since (in Jackendoff 1972) I have eliminated another great part of Lakoff’s evidence, virtually all of Lakoff’s socalled exceptions are now accounted for in a much more systematic and restricted fashion. We also no longer need hypothetical lexical entries, a powerful device used extensively by Lakoff. With practically all of Lakoff’s evidence dissolved, we see that the theory of exceptions plays a relatively insignificant role in lexicalist grammar. A small dent has also been made in the highly controversial area of idiosyncratic phonological readjustment rules, though much further work is needed before we know whether they are eliminable. There are three favorable results in syntax as well. First and most important, the analysis of causative verbs, which supposedly provides crucial evidence for the generative semantics theory of lexicalization, can be disposed of quietly and without fuss, leaving the standard theory of lexical insertion intact. Second, idioms can be listed in the lexicon and can undergo normal lexical insertion; some of their syntactic properties emerge as an automatic consequence of this position. Third, the direction of the English particle movement transformation can finally be settled in favor of leftward movement. Thus a relatively straightforward class of intuitions about lexical relations has been used to justify a theory of the lexicon which has quite a number of significant properties for linguistic theory. Obviously, many questions remain in the area of morphology. I would hope, however, that this study has provided a more congenial framework in which to pose these questions.
Remarks on Chapters 3 and 4 During the period that ‘On Beyond Zebra’ was written, one of the hottest topics in cognitive science was visual imagery, thanks to groundbreaking work by Roger Shepard and Lynn Cooper (1982) and especially by Stephen Kosslyn (1980). At the very same time, David Marr was developing a detailed computational theory of the mental representations involved in visual perception, culminating in his posthumously published 1982 book Vision. Marr drew a strong parallel between his approach and that of generative linguistics: the first goal of his theory was to specify the function the brain is computing, parallel to the problem of linguistic competence. Regrettably, Marr’s early death more or less ended this line of investigation. For one thing, difficulties were found with some of his solutions, and there was no one to pick up the banner and carry on the work; it is not part of the culture of psychology and neuroscience to think in terms of formal mental representations. Moreover, by the middle 1980s, the connectionist revolution called into question the entire notion of mental representation, and the allure of brain imaging techniques came quickly to dominate the study of vision. Both these trends worked to the detriment of more abstract theoretical approaches, a situation that to my knowledge still persists today, and that has contributed to the distancing of linguistic theory from the rest of cognitive science. I was attracted to this work on vision for several reasons. First, I had been studying research on visual perception in connection with my work on music perception and cognition (Lerdahl and Jackendoff 1983). Second, Steve Kosslyn and Lynn Cooper were both colleagues of mine for brief periods of time. Third, John Macnamara’s (1978) challenge, ‘How can we talk about what we see?’, led me to consider more specifically the connection between vision and language. To me, it seemed very natural to answer Macnamara’s question by connecting Marr’s 3D model with my notion of conceptual structure. This idea turned out to be a major breakthrough in my thinking. The metaphor of a breakthrough is apt: it felt like my work in semantics had been tunneling deeper and deeper into the mind, and suddenly I had opened this little passageway into a vast new region, well explored, though not by linguists. Four important advances immediately emerged. First, a constant nagging problem in lexical semantics was the problem of ‘distinguishers’: the nasty little
86
remarks on chapters 3 and 4
parts of lexical meanings that cannot be characterized in terms of simple features—an example in this chapter is the difference in appearance between ducks and geese. At the same time, there were proposals in the literature that words contain as part of their meaning an image of a stereotypical instance (Putnam 1975, Rosch and Mervis 1975). But this raised the problem of how one codes an image and compares it to perceived objects. Marr’s 3D model solved both problems at once: it was intended as a computationally tractable abstract representation of object shape. In these terms, (part of) the solution to the problem of distinguishers is to augment the coding of lexical items to include a fourth kind of representation besides the customary phonology, syntax, and conceptual structure: something like a 3D model—which Marr argued is necessary for object categorization in any event. A second advance was that it became possible to think of a full computational path all the way from hearing a sentence to checking it out visually. No longer did the process of understanding a sentence have to back up indefinitely into ill-understood levels of abstraction. Rather, there was a point where understanding started to emerge out the other side into the real world. This provided an answer to the predominant philosophical view that language has to relate to the world: it does, through the minds of language users. Thus many issues concerning the reference of linguistic expressions turn out to be issues about how humans perceive and comprehend the world through their perceptual systems. The philosophers may not have considered this an advance, but to a linguist of the mentalistic persuasion this result is paramount. Third, the division of labor between conceptual structure and the 3D model called into question Jerry Fodor’s assertion in Modularity of Mind (1983) that the central systems of the mind are homogeneous and unstructured. Finally, the connection between language and vision by necessity cannot be implemented by a derivation from syntactic structure. On formal grounds, the units of vision are far too distant from those of syntax. But more importantly, visual cognition has existed in the animal kingdom far longer than language, so on evolutionary grounds a derivation from syntactic structures is worse than implausible. As a consequence, it became necessary to express the conceptual structure-to-3D model connection in terms of interface rules. This treatment was an important motivation toward the Parallel Architecture within language. I wrote three major papers on the topic of the language–vision connection, two of which are reprinted here in this chapter and the next. The third paper, coauthored with Barbara Landau, appeared as Landau and Jackendoff 1991 and was reprinted in slightly abridged form in my Languages of the Mind (1992). Along with contemporaneous work by Leonard Talmy (1983), this work was, I believe, on the vanguard of an explosion of research on the relation of language
remarks on chapters 3 and 4
87
to vision and its ramifications crosslinguistically and in language acquisition (some randomly chosen major works are Bloom et al. 1996, Coventry and Garrod 2004, Herskovits 1986, Levinson 2003, van der Zee and Slack 2003, plus the journal Spatial Computation and Cognition). During this period, I was privileged to attend two formative conferences on the topic, one at the Max Planck Institute for Psycholinguistics in Nijmegen, and one at the University of Arizona; the latter led to the Bloom et al. volume, in which the paper presented here as Chapter 4 appeared. Although the language–vision connection has been taken up avidly by linguists, it seems to have had little impact in the vision community. However, Macnamara’s question has a converse side: How do we see what we talk about? The richness of spatial language suggests that the visual system delivers far more complex information than is usually considered. Some of the challenges posed by spatial language for visual cognition appear in sections 3.6 and 4.8.
chapter 3
On Beyond Zebra The Relation of Linguistic and Visual Information (1987) And I said ‘You can stop, if you want, with the Z Because most people stop with the Z But not me! In the places I go there are things that I see That I never could spell if I stopped with the Z. I’m telling you this ’cause you’re one of my friends: My alphabet starts where your alphabet ends! (Dr. Seuss, On Beyond Zebra)
3.1 Introduction One of the fundamental problems for a theory of natural language was stated most succinctly by John Macnamara in the title of his 1978 (unpublished) paper: ‘How Can We Talk About What We See?’ In order to approach this problem, a point of connection must be found between the theory of language and the theory of vision. The present chapter develops such a point of connection, and shows that it not only helps answer Macnamara’s question, but also solves certain outstanding problems in both theories. [This chapter appeared originally in Cognition 26 (1987), 89–114, and is reprinted here by the permission of the editor and the publisher, Elsevier.] Much of this chapter was drafted while the author was a Fellow at the Center for Advanced Studies in the Behavioral Sciences in 1983–4, a year for which I will be forever grateful. I was encouraged in this work by discussions with Leonard Talmy, Lynn Cooper, Steven Pinker, Lucia Vaina, David Olson, Carol Krumhansl, James Cutting, Donald Hoffman, and Roger Shepard, among others. This research was supported in part by NSF Grants IST-8120403 and IST-8420073 to Brandeis University and by NSF Grant BNS-7622943 to the Center for Advanced Studies in the Behavioral Sciences.
on bey ond zeb ra
89
I am interested in particular in psychological answers to Macnamara’s question. So the following sort of solution, caricatured on Tarskian or Davidsonian semantics, will not do: Certain sentences are true of the world; speakers grasp the truth of these sentences by using their eyes, and thereby can warrantably assert them. In this sort of theory, the major problem is taken to be the explication of truth, a speaker-independent relation between language and the world. However, the psychological issue of what it means to grasp the relation is largely unaddressed. Barwise and Perry (1983) purport to address it by adverting to a primitive version of Gibsonian psychology of perception. But given that the possibilities look tremendously dim for constructing phonological and syntactic theories along Gibsonian lines, there seems little prospect of such an answer fitting into a unified theory of the language capacity. Entering the realm of information-processing theories of mind, it is not enough to say, for example, that we talk about what we see by transferring information (or representations) from visual memory into linguistic memory. The heart of the problem, as Macnamara recognized, is one of translation: in order to talk about what we see, information provided by the visual system must he translated into a form compatible with the information used by the language system.a So the essential questions are: (1) What form(s) of information does the visual system derive? (2) What form of information serves as input to speech? (3) How can the former be translated into the latter? Let me be more specific about what I mean by a ‘form of information’. One might think of each form as a discrete ‘language of the mind’ (one of which is the ‘language of thought’ in Fodor’s (1975) sense). More specifically, each form of information is a structured system of distinctions or oppositions, built up out
a Nowadays I would not use the term ‘translation’ any more. This term implies that the mapping between vision and language is meaning-preserving. But it is not. There are elements of language that do not correspond to anything in vision, and vice versa. A better characterization of the relation is that it is a partial homology. See discussion in section 1.2. An important point is that correspondences between two levels need not be correspondences between primitives of the two levels. In fact, in general they had better not be, because then the levels would be essentially notational variants of each other and incapable of doing different work in the mind. Often the correspondences will be between a composite in one level and a composite in another. This is clear in the syntax–phonology correspondence, where for instance the phonological complex /went/ corresponds to the syntactic complex [V þ past], and the subparts of each bear no correspondence whatsoever. Another terminological point: I have become reluctant to use the terms ‘information’, ‘symbol’, and ‘representation’ any more, because of their objectionable overtones of a homunculus that the information is informing and to which the representations are representing something. In recent work I have attempted to avoid any whiff of this sort of intentionality, and have replaced these terms with the more neutral ‘mental structure’, which does the same work within the theory quite effectively. (See Jackendoff 2002, 20, 279; 2007b, 5–7.)
90
on beyond zebra
of a finite set of primitives and principles of combination. More standard terms with essentially the same sense are level of representation as used in linguistics, Kosslyn’s (1980) format, and Marr’s (1982) stage of representation. The language faculty, for instance, involves three distinct forms of information— phonological, syntactic, and semantic/conceptual. The phonological primitives include the phonological distinctive features, the notions of segment, syllable, and word, the elements of stress and timing systems, and the elements of intonational patterns. The principles of combination include the combination of a set of phonological distinctive features into a segment (or speech sound), the concatenation of segments into syllables and words, and the principles for associating stress, timing, and intonation with strings of segments. Similarly, the syntactic primitives include the syntactic categories (Noun, Verb, etc.), the phrasal categories (S, NP, etc.), and the elements of inflectional systems such as case, gender, number, tense, and aspect. The principles of combination include the domination of one node by another, the linear ordering of nodes, and the elements of syntactic transformations (or alternative devices for capturing long-distance dependencies). [Semantic/conceptual primitives and principles of combination were discussed in section 1.3.] The overall logical organization of information used by the language faculty is shown in (1).b (1) Organization of language phonological formation rules
syntactic formation rules
conceptual formation rules
syntactic structures
semantic/ conceptual structures
auditory information motor information
phonological structures
The primitives and principles of combination for each form of information are collectively labeled as ‘formation rules’ for that level. In addition, though, there must be a set of principles for translating from acoustic information to meaning in order to accomplish language understanding, and similarly from meaning to vocal tract instructions in order to accomplish speech. These principles of
b Here is the Parallel Architecture already in 1987, though not identified as substantively different from mainstream linguistic theory. In more recent work, as seen in Chapter 1, I tend to use the term ‘interface rules’ instead of ‘correspondence rules’.
on bey ond zeb ra
91
translation are encoded as sets of ‘correspondence rules’ that link adjacent levels; they are notated in (1) as arrows and double arrows between structures. I should stress that (1) is the logical organization of the information, not a model of language processing. It is consistent with many different theories of the time-course of processing, as long as they respect the formal autonomy of the three kinds of linguistic information. For instance, it is consistent with a parallel processing system in which translation from acoustic form to phonological structure is taking place for one part of a sentence, while an earlier part of the sentence is already undergoing translation from syntactic to semantic form. (Jackendoff 1987b, chapter 7 goes over various possibilities in detail, demonstrating the distinction between the theory of the logical organization of structure and the theory of the time-course of processing. The point is not new, but it is still widely misunderstood.) The need to explain the function of the visual system in terms of the forms of information it derives is most explicitly articulated in the work of Marr (1982). (Two good introductions are Marr and Nishihara 1978 and Pinker 1984.) Marr calls such a description a computational theory, a theoretical prerequisite to the descriptions of processing (algorithmic theory) and neurological implementation. According to Marr’s theory, information transduced by the retinas must be translated successively into three distinct forms before object recognition can be achieved: the primal sketch, which encodes the local organization of boundary elements in the visual field; the 2½D sketch, which encodes the geometry of the surfaces visible to the viewer; and the 3D model, which encodes the spatial configuration of viewed objects in viewer-independent terms.c The overall logical organization of Marr’s theory of visual information is given in (2).
(2) Organization of vision (Marr)
retinal array
primal sketch formation rules
2½D sketch formation rules
3D model formation rules
primal sketch
2½D sketch
3D model
c As mentioned in the Remarks preceding this chapter, Marr’s approach to vision has largely fallen into neglect. Readers who reject Marr’s approach are encouraged to replace it with something else that has the requisite properties, such as the capacity for viewpoint-invariant object categorization, for decomposing objects into parts, for assigning abstract axes, for characterizing degrees of freedom in object-internal shape change, and so on. Marr’s approach extends naturally to situations richer than single objects, as observed in this chapter and the next; a substitute theory should be able to speak to these phenomena as well. Biederman’s (1987) approach in terms of ‘geons’ satisfies some but not all of these criteria.
92
on beyond zebra
Again, the formation rules characterize the formal elements out of which levels are built (for instance, ‘blobs’, ‘bars’, ‘terminations’ in the primal sketch, ‘regions’ and ‘contours’ in the 2½D sketch); the correspondence rules characterize the principles by which information of one form is translated into the next. The problem of how we talk about what we see can be understood more clearly in terms of diagrams (1) and (2). What is necessary for this task to be possible at all is a set of correspondence rules linking forms of information in the two faculties. Through these correspondence rules, visual information can be translated into a form suitable to be couched as a linguistic expression. The hypothesis to he pursued here is that the desired correspondence rules are to be stated between semantic/conceptual structure and the 3D model, i.e. between linguistic meaning and ‘visual meaning’—the mind’s reconstruction of ‘what is out there in the world’. From the point of view of the semantics of natural language, this possibility is tremendously exciting. For if a connection can be forged between semantic/ conceptual structure and the information derived by the visual system, semantics will not have to back up into ever murkier levels of ‘interpretation’ to explain our ability to talk about the visually perceived world. Rather, there will be a complete succession of levels of representation from the retina all the way to the vocal tract—a totally computational theory of how we talk about what we see. Information can pass up through the forms encoded by the visual system, be translated from 3D model organization into semantic/conceptual organization, and proceed down the linguistic forms into speech.d The next two sections sketch some properties of the 3D model and semantic/ conceptual structure; we then turn to the relation between them.
3.2 Properties of the 3D model Marr explicitly designs the 3D model to be appropriate for encoding an object’s shape in long-term memory, so that it may be recognized on subsequent occasions. He develops representations only for single objects, not for the complete configuration in the visual field; we will propose some extensions in section 3.6. There are three salient characteristics of the 3D model. First, it is volumetric: objects are represented as occupying volume in space, by contrast with the
d As mentioned in the Remarks preceding this chapter, this result is fundamental to semantic theory, in that it leads to a genuinely psychological theory of reference to the ‘real world’. See Jackendoff 2002, chapter 10.
on bey ond zeb ra
93
surface representations of the lower levels. Second, it is object-centered: it makes explicit the shape and size constancy of objects, regardless of the viewer’s position, by contrast with the viewer-centered representations of the lower levels. Third, it is hierarchical: it represents the three-dimensional structure of objects not just in terms of holistic shape (i.e. it is not a ‘statue in the head’), but rather in terms of a hierarchical decomposition of the object into parts and parts of parts. For example, consider diagram (3), from Marr and Nishihara (1978), which suggests the organization of the 3D structure for a human figure.
(3)
A sample 3D model structure
At the coarsest layer of description, the figure is represented simply by a cylinder, itself defined by a vertical axis down the middle and a cross-section. At the next layer of description, the cylinder is elaborated into a torso, a head, and four limbs, each of which is a cylinder defined by its own axis. The points of attachment of the head and limbs to the torso, and the angles that they form, are specified in terms of the coordinate system defined by the axis of the torso. In turn, each of the parts is subject to elaboration at finer levels of detail. For instance, the arm cylinder can be elaborated as two connected cylinders corresponding to the upper and lower arm; the lower arm can be elaborated into lower arm and hand; the hand can be elaborated into palm and fingers, the fingers into joints. Thus the 3D model is a sort of tree structure. In each case of elaboration, the configuration of parts is specified in terms of the coordinate system defined by the axis of the main part in the configuration. Thus the positions of the fingers, for instance, are specified most directly in terms of their configurations within the hand. Their position with respect to the body as a whole is specified only indirectly, through the hand’s position on the arm and the arm’s on the torso. This, then, is what Marr means by saying the description is object-centered: through recursive elaboration, the positions of the parts of an object are specified
94
on beyond zebra
ultimately with respect to the main axes of the object as a whole, and can therefore be specified without mention of the viewer’s position. Of course, in integrating the object as a whole into the scene, its position with respect to the viewer must be specified. But this can be determined entirely by specifying the position of the object’s main axis, and the positions of the parts will fall into place automatically.1 Though the formal characteristics of the 3D model representation are at present far from well understood, the general outlines are fairly clear. The primitive units of the representation are the coordinate axes and the means for generating simple shapes (such as the cylinders of (3)) around them. The principles of combination provide the means to combine simple shapes into more complex shapes. From (3) we can see that the principles of combination essentially must provide a way to elaborate a description of an object from a coarse description to the next higher degree of specificity, for example from the single cylinder to the torso plus head and limbs. Further layers of description will be possible simply by applying the principles of combination recursively. Note that the principles are not of the form ‘such-and-such a part is connected to such-and-such a part.’ [2010: This is more or less Biederman’s (1987) approach.] Rather, the dispositions of the parts are specified in relation to the coordinate axes of the next coarser layer of description: ‘(the axis of) such-andsuch a part at layer of description Li is disposed in such-and-such a configuration with respect to the axes of such-and-such a element of layer Li1.’ Principles of this form are a great deal like the phrase-structure rules of syntax, in which, for example, a relative clause is understood as an expansion not of the head noun, but of the next larger unit, the noun phrase. It is moreover possible to discern a notion of ‘head’ not unlike that in syntax. Just as one talks of the head noun—the principal element of a noun phrase, the one that determines the noun phrase’s categorical status—one can in many cases speak of the head constituent of a 3D model elaboration: that subsidiary element in an elaboration whose axes and volumetric form are identical to (or most similar to) the coarser-layer element being elaborated. In the human figure, 1
There seems to be no bar to locating the parts directly with respect to the viewer, but in the optimally hierarchical description, the one that permits identification of the object as a whole, this will be redundant. However, such description in the 3D model may well be used on occasion, either for particular tasks or for dealing with unfamiliar objects. For example, this appears to be a plausible account of Jolicoeur and Kosslyn’s (1981) experiments, in which subjects seemed sometimes to use viewer-centered descriptions for identification of rather bizarre objects. Jolicoeur and Kosslyn imply that subjects are storing a 2½D sketch. Closer attention to the formal details of Marr’s theory of the 2½D sketch would show that this is unlikely: recall that there is no notion of object in the 2½D sketch, through which identification could take place.
on bey ond zeb ra
95
for instance, the torso is the head of the initial elaboration; the palm is the head of the hand. (In other cases, though, such as the elaboration of the arm into upper and lower arm, the presence of a head constituent is less obvious. Perhaps there is an analogy here to coordination [2010: or parataxis] in language.) Now consider the task of identifying an object that can change its shape, for example a human or animal. The long-term memory 3D model of such an object, to which its presented appearance can be compared, will not represent just a single pose of this object. Rather, the parameters specifying angle of attachment of the various parts in a decomposition like (3) will give a range of possible values, corresponding to the possible configurations of the joints. On the other hand, certain aspects will be kept rigid, for example the point of attachment. (If somehow the shoulder joint could slide up and down the torso, the way the carriage slides across a typewriter, that too could be specified by a variable.) Thus the object-centered decomposition of objects makes possible a description of an object’s possible variations in shape. Similarly, the 3D model permits an intuitively attractive account of the representation of the action of moving figures, as Marr and Vaina (1982) point out. For instance, in describing the action of walking, the 3D model can specify the angular changes in each of the joints, and relate these changes in a temporal framework. Moreover, a rather natural segmentation of actions arises at points in time when joints start or stop changing in angle, or when angular change reverses direction. For instance, the swing of a leg in walking is bounded at its beginning and end by points of stasis in the change of angle at the hip joint. (This corresponds nicely to observations of Cutting 1981 on the segmentation of events.) Notice that such specifications can be made only in an objectcentered framework, since it is the angular change measured ‘objectively’ at the hip that is significant, not the angular change in the visual field. Thus not only for object identification but for the description of action, an objectcentered description along the lines of Marr’s seems a necessity. A further move of abstraction permits the 3D level to encode variations in form among individuals of a common type. For instance, consider how a longterm representation might encode what horses look like. By allowing a range of permissible sizes, both overall and constituent by constituent, and by specifying ranges of proportions of size among parts, one can arrive at a geometrically parameterized encoding that can be matched with a class of presented objects more variable than a specific individual. Notice how this conception differs from the widespread hypothesis that one stores a mental image of a stereotypical instance of a category. As often conceived, such an image is of a particular individual—a comparatively rigid template against which putative instances are matched by some unspecified
96
on beyond zebra
computational process of ‘judging similarity’. The 3D model, by comparison, is decompositional and hierarchical, and thus capable of greater generality than a rigid template, in that the elaboration of the object into parts can specify proportions and tolerances rather than a single fixed value for the size, shape, and disposition of the parts. (The phenomenon of an experienced image of a prototypical instance probably arises in part through fixing typical or focal values for the variable parameters.) Moreover, the 3D model representation is not an arbitrary stipulated representation, but one motivated and developed on the grounds of its general adequacy for visual form recognition tasks.
3.3 Outline of Conceptual Semantics [This section of the original text has been incorporated into section 1.3.]
3.4 Preliminary points of correspondence We are now in a position to begin to work out some principles of correspondence between the 3D model and conceptual structure. It is important to bear in mind that the former was motivated by the demands of the visual system, and the latter by the demands of language. One might not anticipate that any relationship between the two need exist. Any relationship that is found, therefore, is an unexpected windfall, and a vindication of the claims of both representations to psychological reality. The first point of contact, of course, is the notion of physical object as a significant unit of both levels, though encoded in quite different fashions, to be sure. Thus a basic principle of correspondence maps constituents of the 3D level that encode objects into constituents of the conceptual level that encode objects; there is no other pair of levels between which this fundamental correspondence could be stated. Another important correspondence that can be established is in the encoding of the relation of part–whole or inalienable possession. The relation has always been one of the staples of linguistic semantics; Marr’s hierarchical theory of the 3D model allows it to be spelled out in structural terms. Basically, for physical objects X and Y, X IS A PART OF Y obtains in conceptual structure just in case the 3D object representation corresponding to X is an elaboration within the 3D object representation corresponding to Y. Next, recall that Marr designed the 3D model level to encode longterm memory information suitable for either object identification or object
on bey ond zeb ra
97
categorization. But now let us ask: how is the long-term memory for a known individual distinguished from that for a known category? So far the two are not distinguished in formal structure—they’re both 3D models—so what makes the difference? One’s first impulse is to claim that memories of individuals and memories of categories differ in vagueness or generality, individuals being much more specific. But this will not do. One may be somewhat vague about the appearance of a slightly known individual—say the car that hit mine and sped off into the night—and therefore encode it rather imprecisely in memory; while one may on the other hand be very knowledgeable about the appearance of a very precisely delimited category (say IBM PC keyboards) and therefore encode it in great detail and specificity. Further reflection suggests that in fact there are no features of the 3D model, which is purely geometric in conception, that can distinguish representations of individuals from representations of categories. For example, the 3D models for the individual ‘my dog Rover’ and for the category ‘dogs that look just like Rover’ are necessarily identical, because of the way the category is defined. What is needed to distinguish the two kinds of representations is in fact the binary TYPE/TOKEN feature of conceptual structure, an algebraic form of representation. Only an algebraic structure can provide the proper sort of distinct two-way opposition.e Let me be a little more precise. The claim is that visual memory contains not just 3D representations but matched pairs of representations: a 3D model for how the thing looks, and a conceptual structure that ‘annotates’ the visual representation, specifying at least whether this is taken as a representation of a token or a type. The visual forms given by perception are automatically linked to the TOKEN feature in conceptual structure: that is, what one directly sees consists of particular individuals. On the other hand, what one learns and stores in memory can be linked either with TOKEN (if one is remembering an individual) or with TYPE (if one has learned a category). Now consider the relation between an individual being perceived and a remembered individual or category. The two 3D model representations must be juxtaposed and compared, and the outcome of the matching process must be recorded. (This is the ‘Ho¨ffding step’ of classical perception theory—see Neisser 1967.) But the outcome of a match cannot be represented visually: it is basically of the form ‘successful match’ or ‘unsuccessful match’. It can however be encoded in conceptual structure. A successful match in object identification is e
Note that this argument and the next apply to any theory of visual recognition, not just to Marr’s.
98
on beyond zebra
encoded conceptually by the relation IS-TOKEN-IDENTICAL-TO, an algebraic relation between two TOKEN concepts. Object categorization, however, is encoded by the relation IS-AN-INSTANCE-OF, a relation between a TOKEN and a TYPE [see section 1.3]. The overall forms of the two relations are given in (10); the vertical lines indicate associations or linkages between representations at the two levels. (10) a. Object identification conceptual level: IS-TOKEN-IDENTICAL-TO ([TOKEN]i, [TOKEN]j) j j 3D level: visually 3D model derived from 3D model memory b. Object categorization conceptual level: IS-AN-INSTANCE-OF ([TOKEN]i, [TYPE]j) j j 3D level: visually 3D model derived from 3D model memory Note that the 3D model part of the judgment is exactly the same in both cases: the comparison of a structure derived from vision with one from memory. The only difference between identification and categorization, then, lies in the conceptual level. The notion of paired 3D and conceptual structures helps solve another, more often recognized problem concerning the visual encoding of categories of actions. The need for such categories has cropped up occasionally in the literature. For instance, Marr and Vaina (1982) discuss how a few ‘basic action types’ such as throwing, saluting, walking, etc. can be defined in terms of sequences of motions of body parts in the 3D model. Peterson (1985) suggests that there is a class of ‘natural actions’ described by verbs like throw and push, analogous to ‘natural kinds’ like dog and banana. Like natural kinds, natural actions are learned by ostension (‘This is what it looks like’) more than by definition. How are action categories to be encoded? Here is the problem, which in its essence goes back at least to the British empiricists. A visual representation of the action of walking, for example, requires by its very nature a walking figure, say a generalized human. But then, what is to make it clear that this is a representation of walking rather than of human? The requisite distinction is
on bey ond zeb ra
99
not available in the geometric representation—but it is available in conceptual structure, where we have the algebraically structured features that distinguish primitive conceptual categories. By linking the 3D figure in motion to an ACTION TYPE concept rather than to a THING TYPE concept, we can encode the fact that the motion of the figure rather than its shape is taken as the significant information in the 3D model. Thus again a linkage of 3D and conceptual structures provides the right range of distinctions.
3.5 The use of 3D models in word meanings We have just seen that the visual system must make use of the level of conceptual structure to help encode long-term memories of individuals and categories. In this section we will see that language probably makes use of the 3D model in encoding distinctions among word meanings. First, there are distinctions of meaning among words that appear to be spelled out far more naturally in terms of spatial structure than in terms of conceptual structure. A good example (brought to my attention by Thomas Kuhn) is distinguishing among ducks, geese, and swans. In conceptual structure it is quite natural to make a taxonomy of these types, such that they are distinct from one another and together form a larger type ‘waterfowl’, itself a subtype of birds. But how are the differences among these types to be expressed? Clearly, one of the most salient differences, and the one by which a perceived individual is normally classified into one or the other of these categories, is how ducks, geese, and swans look—their relative sizes and the proportions and shapes of their respective parts. Now the idea that these differences are represented in conceptual structure by features like [+LONG NECK] [2010: or as distinguishers, a` la Katz and Fodor 1963] is implausible, because the features seem so ad hoc. Yet these have been the sorts of features to which descriptive semanticists have had to resort, for lack of anything better. (One suspects, in fact, that the evident need for such bizarre features is one of the major factors contributing to the suspicion with which lexical semantics has often been regarded.) However, notice that descriptions of size, proportion, and shape of parts, being purely geometric notions, are quite naturally expressed in the 3D model, which must include them in any event in order to accomplish object recognition. This suggests that conceptual structure may be divested of a large family of ad hoc descriptive features by encoding such distinctions in 3D model format, where they are not ad hoc at all, but precisely what this level of representation is designed to express.
100
on beyond zebra
An immediate implication is that the representation of a word in long-term memory need not be just a triple of partial phonological, syntactic, and conceptual structures, but may contain a partial 3D model structure as well. This conclusion reflects the intuition that knowing the meaning of a word that denotes a physical object involves in part knowing what such an object looks like. It is the present theory’s counterpart of the view that one’s lexical entry may contain an image of a stereotypical instance. However, as observed in section 3.2, the 3D model provides a much more coherent account of what lies behind such an intuition than does a rigid ‘picture-in-the-head’ notion of stereotype, allowing it to play a more interesting role in a formal lexical semantics. Not only nouns benefit from 3D model representations. For instance, verbs such as walk, run, jog, lope, and sprint differ from each other in much the same way as duck, goose, and swan. It is embarrassing even to consider a set of binary algebraic features that will distinguish them. However, since the 3D model level can encode actions, it can naturally provide the relevant distinctions in gait and speed as a part of the verbs’ lexical entries. Going further afield, consider functional definitions, which pick out objects that one can use in a certain way. For instance, an important component of the concept of chair is that it is something to sit in.f How is this to be encoded? Sitting is a ‘natural action’, specifiable by an ACTION TYPE linked to a 3D model of what sitting looks like. The chair, in turn, can be specified as an auxiliary character in the action: it is the surface upon which the acting figure comes to rest. In the 3D model, its appearance can be specified very coarsely, giving only its approximate size and the crucial horizontal surface that the figure makes contact with. Thus, as Vaina (1983) points out, a functional definition can be encoded by linking a particular object in a 3D action description with an OBJECT TYPE in conceptual structure—the 3D model encodes what one does with the object, plus only enough of the object’s appearance to show how one does it.g While the formal niceties of such word meanings are yet to be worked out, I think it is possible to see the germ of an important descriptive advance here. By using linkages of 3D models with conceptual structure, one can begin to circumvent the limitations of the purely algebraic systems to which semantics
f The notion of characteristic action that an object performs or that one performs on an object is part of what Millikan 1984 calls the object’s proper function and what Pustejovsky 1995 calls the word’s telic quale. Pustejovsky shows how characteristic actions play a role in understanding sentences, whether one wants to call such understanding semantic or pragmatic. (See section 13.4 for further discussion.) g More specifically: the visual encoding of action of sitting has to encode a schema of a person (but no particular person) configured in a particular position on a support (but no particular support). Something in memory has to stipulate that what’s significant in this encoding is the spatial configuration of the figure, not
on bey ond zeb ra
101
has been largely confined, and at the same time begin to see how language can make contact with the world as perceived. This is not to say that all elements of linguistic meaning are conveyed by 3D models. Far from it. Some of the essential algebraic features of conceptual structure were shown in the previous section to be necessary even for visual memory, not to mention language. Moreover, such aspects of meaning as negation and quantification are fundamentally conceptual, and cannot be translated into a 3D model. And of course there are plenty of words that express auditory, social, and theoretical concepts, for which no 3D counterpart should be expected. [See Jackendoff 2007b, chapters 6–11, for treatments of theory-of-mind and social concepts.] The point is only that when language has an opportunity to exploit the expressive power of the 3D model, it does so, and hence that one should expect words for spatial concepts to exhibit characteristically geometric as well as algebraic distinctions.
3.6. Enriching the conceptual-3D connection If the 3D model is to be the component of the visual system most directly responsible for our ability to talk about what we see, it must be rich enough in expressive power to provide all the visual distinctions that we can express in language. (It may of course be even richer—there may be further 3D model distinctions that are not expressible in language but which play a demonstrable role in spatial cognition or the capacity for action. [2010: That would be why a picture is worth ten thousand words.]) the figure itself. In fact, the requisite distinction is not statable as part of the geometry. However, it is statable in the algebraic encoding of CS: [Situation SIT (X, Y)] is associated with the configuration, X with the sitter and Y with the support. This is drawn very crudely in the figure below.
[Situation SIT3 (X1, Y2)] ]
102
on beyond zebra
This section will use evidence from language to suggest some natural enrichments of Marr’s theory. These will help extend the 3D model beyond the description of individual objects and actions to a description of the full spatial array. The evidence comes from sentences concerning spatial location and motion. It has often been noted (Clark and Chase 1972; Gruber 1965; Jackendoff 1976; Langacker 1983; Miller and Johnson-Laird 1976; Talmy 1983) that spatial relationships between two objects are pretty much never expressed symmetrically in language. Rather, the language usually distinguishes two roles: a landmark or reference object, which often appears as the object of a preposition, and a figural object or theme, which often appears as grammatical subject. For instance, in (11a), the table is the reference object and the book is figural. And it is intuitively clear that there is a distinct difference between (11a) and (11b), where the roles of reference object and figure have been exchanged. (11) a. The book is on the table. b. The table is under the book. The usual examples illustrating sentences of spatial relation, like (11), use only the neutral verb be. However, the asymmetry between the figural and reference objects is clearer in sentences of location that use more specific verbs, as in (12). (12) The book is standing/lying/leaning/resting on the table. Here the lexical distinctions among the verbs encode object-internal information about the figural object, the book, in particular the spatial disposition of its major coordinate axis. In other words, while the neutral verb of location be gives only the very coarsest description of the subject, more specific verbs of motion and location elaborate some internal details of its 3D model. In turn, this supports the asymmetry of the relation between the figural object and the reference object. The proper treatment of this asymmetry in conceptual structure (Herskovits 1985; Jackendoff 1978, 1983, chapter 9; Talmy 1983) is to make use of the conceptual category Place (sections 1.3 and 1.5). As seen there, location sentences like (11) and (12) assert not a spatial relation between two objects, but the Place at which the figural object is located. In turn, the Place is specified as a function of the reference object, each choice of preposition specifying a different function and hence determining a different Place. The conceptual structure of such sentences is therefore organized as in (13). (13) [State BE ([Thing BOOK], [Place ON ([Thing TABLE])])]
on bey ond zeb ra
103
If indeed conceptual structure is set in correspondence with a 3D model structure, the notion of Place ought to play a role in the latter level of representation. And in fact it seems quite natural to encode in the 3D model the notion of regions of space related to an object, determined in terms of the object’s 3D representation. [Much more detail appears in Jackendoff and Landau 1991, Landau and Jackendoff 1993.] For example, in expresses a function that (for a first approximation) maps an object into the region consisting of its interior. On maps an object into the region consisting of its surface (in many cases, its upper surface). Near maps an object into a region exterior to but contiguous with the object. Beside is like near, but restricts the region to roughly horizontal contiguity (a cloud above a mountaintop may be near it; it cannot be beside it). More interesting are the prepositions that make use of the reference object’s own intrinsic coordinate axes. For instance, one can be either beside or along a road, but one can only be beside, not along, a tree (unless the tree has been felled). Evidently the domain of the function expressed by along is (roughly) objects of significant extension in one horizontal dimension only; this function maps such an object into an exterior contiguous region. Another well-known class of examples of this sort consists of prepositions such as in front of, behind, on top of, and to the right of. These get used in two ways. Suppose the reference object is a person or a house. Then it has its own intrinsic axes, used in the 3D model to determine the internal layout of its parts. For instance, the head of a person and the roof of a house go on top, as specified by the directed up–down axis; the face and front door go on the front, as specified by the directed front–back axis. These axes, independently necessary to establish the form of the object, may simply be extended beyond the surface of the object to determine regions that can be referred to by prepositional phrases such as in front of the house, behind the man, and so on.h On the other hand, some objects such as featureless spheres have no intrinsic axes. In such cases, the position of the speaker (or hearer) extrinsically imposes a set of coordinate axes on the reference object: the front is the side facing me (or you), so that behind the sphere picks out a region contiguous to the sphere and on the side of it not facing us.
h Notice that some of these axes go beyond Marr’s central axis, which simply defines a generalized cone. For example, the front-to-back and side-to-side axes of a human figure are necessary in order to define the position of the face with respect to the shoulders, to make sure the nose is oriented in the same direction as the navel, and to get all the parts of the face properly aligned with each other. See sections 4.8–9 for more discussion of multiple axes in spatial structure, and section 6.5 for formalization of multiple axes in conceptual structure.
104
on beyond zebra
As has been frequently noted (Clark 1973, Olson and Bialystok 1983, Talmy 1983), ambiguities arise in the use of such prepositions in cases where two possible sets of coordinate axes are available and in conflict. For instance, in the situation depicted in (14), the ball is behind the house may be read as describing the ball at either position X (house’s intrinsic axes) or position Y (speakerimposed axes).
(14)
Similarly, if Fred is lying down, Fred’s hat is on top of his head can describe a configuration where the hat is in its normal position relative to his head (on top of is in terms of Fred’s intrinsic axes) or one where it is covering his face (on top of is in terms of gravitational axes). One of the points of Olson and Bialystok (1983) is that such conflict between coordinate systems may occur even in purely spatial tasks, where language is not involved. This suggests, at the very least, that extending object-internal coordinate axes to the space exterior to an object plays a role in spatial understanding, and that linguistic expressions of location are simply encoding information that is present for independent purposes.i Let us turn next to the linguistic description of the motion of objects. This often divides rather nicely into two parts, which may be called object-internal and object-external aspects. (Lasher 1981 makes a similar distinction between ‘contour motion’ and ‘locomotion’.) Object-internal motion is in many cases
i
An important discovery has been that languages differ in their preferences for coordinate systems. In particular, the Australian language Guugu Yimithirr and the Mayan language Tzeltal express horizontal spatial relations among objects almost exclusively in terms of absolute coordinates (the counterpart of north, south, etc.) rather than intrinsic coordinates (right of, left of, in front of, behind) (Levinson 2003). This has led to a controversy over whether the linguistic difference has an effect on nonlinguistic thought. My reading of the experimental evidence (Levinson 2003; Li and Gleitman 2002; Li, Abarbanell, and Papafragou 2005) is that linguistic differences can exert a biasing effect under very special conditions, but they do not create fundamental differences in spatial thought.
on bey ond zeb ra
105
expressed by lexical distinctions among verbs of motion; object-external motion by the structure of accompanying prepositional phrase arguments. Consider the possibilities in (15), for instance.
(15)
walked ran John squirmed crawled soared
under the bridge. into the room. through the tunnel. over the rug. along the road
As in the location sentences (12), the differences among the verbs reflect the internal dispositions and motions of the parts of John’s body, that is, they express the object-centered description of John himself. As observed in section 3.5, these differences are not easily characterized in terms of conceptual features; they are, however, rather naturally differentiated in terms of 3D action descriptions.j On the other hand, the differences among the prepositional phrases in (15) reflect the motion of John as a whole. For this part of the description, John can be regarded as a very coarsely described object (a point or an undifferentiated lump) traveling along some externally specified trajectory. Thus the total description can be seen as hierarchical, the outer layer being the external trajectory of the object, the inner layer being an object-centered elaboration of the object-internal motion. Talmy (1980) points out that in some languages, for instance French and Spanish, this bifurcation of motion description is even clearer, in that one cannot say ‘John squirmed through the tunnel’, but must say, literally, ‘John went through the tunnel squirming.’ Here the main clause expresses the objectexternal motion, while the dependent clause squirming expresses the objectinternal motion. A similar constraint exists in Japanese (Mitsuaki Yoneyama, personal communication).2, k 2
On the other hand, there are verbs in English that express object-external motion, as in (i).
(i) John circled the tree. The fly spiraled down to the table.
Thus the division is not grammatically rigid; nonetheless, it manifests itself in fairly clear semantic intuitions that in turn are often reflected grammatically. j The differences among the verbs in (15) are usually attributed to ‘manner modifiers’ which are then left uncharacterized. If these differences are encoded in 3D model structures, the only conceptual structure feature necessary is something like MANNER-OF-MOTION, which is then coindexed with the appropriate 3D action model. k Talmy also observed that French and Spanish are much richer than English in their vocabulary of verbs that incorporate expressions of Path, such as enter (‘go into’) and cross (‘go across’), and that there are few
106
on beyond zebra
The Marr theory as it stands does not include any notion of ‘trajectory traversed by an object’. The object-external part of the motion appears in conceptual structure as an Event constituent and a Path of the sort in (16): (16) John goes into/enters the room: [Event GO ([Thing JOHN], [Path TO ([Place IN ([Thing ROOM])])])] In particular, the Path constituent corresponds to the trajectory of motion, and the preposition expresses the function that maps the reference object (here, the room) into the Path (here, a trajectory that begins outside the room and ends inside it). Some of the prepositions that express such functions treat the reference object at the coarsest layer of description. To, for instance, treats its object essentially as a point (which may be elaborated, of course, by the object’s own description). Many other prepositions of path, however, exploit the reference object’s geometry to some degree or another. For instance, into and through describe paths traversing the interior of the reference object, and onto encodes a path that terminates on the reference object’s surface. One of the more complex cases is across. As pointed out by Talmy (1983), the kinds of objects one can go across usually have one horizontal axis longer than the other, and edges roughly parallel to the long axis. Going across such an object involves traversing a path from one of these parallel edges to the other. For instance, one can go across a road or a river, and across the short dimension of a swimming pool, but not across the long dimension of a swimming pool—in fact there happens to be no appropriate English preposition for this case. (There are exceptions to these conditions, but I think the basic principles stand. This description may be thought of as ‘stereotypical’ across.)l
verbs that incorporate both (e.g. a verb that would mean ‘go over something by crawling’). Talmy took this to be a major typological split among languages. However, it has turned out to be more of a probabilistic or proportional difference; for instance, English, which Talmy classifies as a manner-incorporating language, has plenty of verbs like enter and cross. It has also turned out that there are languages which disfavor both manner and path incorporation, for instance Hindi (Narasimhan 1998). The crosslinguistic differences in what kind of material appears in verbs has become a major area of psycholinguistic research, both in adult processing and in child language acquisition (Slobin 1996). l This is not quite the right characterization, as it does not include the case when the reference object is a line rather than two-dimensional, as in across the border. This denotes a trajectory perpendicular to the line, going from one side of the line to the other. In the stereotypical two-dimensional case, the line in question is the major linear axis of the object. However, other cases of across, such as across the field, do not require a long narrow reference object; here across denotes horizontal traversal of a surface. Stereotypical across, then, seems to be the union of these two cases, each functioning as a preference rule. (For preference rules, see Jackendoff 1983, chapter 8.)
on bey ond zeb ra
107
It is not difficult to imagine extending Marr’s notion of the 3D model to include such information as Paths. Just as with Places, in the cases where Pathfunctions make use of the geometry of the reference object, this seems most often to be information that is independently necessary for a Marr-type description of the object itself, for instance its major coordinate axes. Hence the extension of the 3D model to include Paths seems quite natural, and yet another correspondence can be established between constituents of conceptual structure and 3D model structure. There is nonlinguistic motivation as well for a notion of Path in spatial understanding. If an organism is to use something like a 3D model derived from vision to direct its action—to find its way around in the world—it will necessarily have to compute trajectories, both those of moving objects, to see where they will end up, and those that it plans to traverse in its own future action. Thus the enrichment proposed in order to adequately account for motion expressions in language in fact serves an independent purpose. The language is just capitalizing on what is already present in spatial understanding.m
3.7 Summary and methodological remarks From this very preliminary investigation of the correspondence between conceptual structure and 3D model structure, then, a number of points emerge. First is that the two structures can indeed be brought into contact, and that much of spatial thinking depends on aspects of both. Thus these two levels of m
In addition to objects, actions, places, and paths, various other sorts of entities also can be derived from the visual field. Semantics and Cognition presents the following argument. Consider how deictic anaphora as in (i) comes to be understood. (i) Would you pick THAT up, please? [pointing] The hearer determines the reference of THAT by following the speaker’s point, to discover some relevantly salient object in the visual environment. Understanding the sentence (and therefore picking up the right thing) requires integrating information derived from language and information derived from vision, so that appropriate action can be formulated. However, reference to objects is not the only kind of deictic anaphora in English. The part of meaning coming from language can direct attention to other sorts of entities. For instance we can refer to substances, as in (iia), and to kinds of objects, as in (iib). (ii)
a. Where do you think THAT came from? [pointing to muddy water pouring out of a pipe in the basement] b. I’d sure like one of THOSE. [pointing to a (single) Porsche]
Places and paths have characteristic deictic anaphors: (iii)
a. Would you put your hat THERE, please? [pointing] b. There was a fly buzzing around right HERE. [pointing to empty space] c. The fly went THATAWAY! [pointing] [continued]
108
on beyond zebra
representation constitute a central core accessed by a number of different peripheral faculties, including visual perception, language, and action. Second is that the theory of Conceptual Semantics developed in S&C, summarized in section 1.3, contains many of the right sorts of elements to interface with spatial understanding. This is evidence that it is on the right track towards a properly mentalistic theory of the semantics of natural language, and in particular towards an answer to the question of how we can talk about what we see. Third is that language can provide evidence for the constitution of the 3D model level, in that it motivates elements such as Places and Paths that are not immediately apparent in intuitions about the theory of vision per se. Such enrichment is expressed naturally in terms of elements which are already made available by the Marr theory, and which turn out on reflection to be supported by the organism’s performance at nonlinguistic tasks. These enrichments
Actions do as well. (iv)
a. Can you do THIS? [demonstrating] b. THAT had better never happen in MY house! [pointing to, say, some random kids smoking pot]
Notice how these actions must be independent of who is performing them (as in sitting, above), because in (iv.a) the hearer is invited to evaluate his or her own ability to perform the same action as the speaker is currently performing, and because in (iv.b) what the speaker is sanctioning is an action, not by these unknown individuals who are carrying out the action, but by anyone at all, especially the hearer. You can’t perform another person’s action! In (v), the hearer is invited to pick out not just the action, but some distinctive character or manner of action – an entity even more abstract than the action itself. (v) Can you walk like THIS? [demonstrating] In (vi.a), there is no fish in the visual environment. The hearer is invited to pick out the distance between speaker’s hands and apply that to the imagined fish. In (vi.b), the individuals currently in the room may or may not have been at the party. This is irrelevant. The hearer is invited to pick out the numerosity of individuals in the room, considered as an aggregate, and apply that to the aggregate of imagined individuals at the imagined party. (vi) a. The fish that got away was THIS/YAY long. [demonstrating] b. There were about THIS many people at the party last night. [gesturing around the room] To sum up, deictic anaphora shows us that language talks about objects, substances, kinds, locations, paths, actions, manners of action, distances, and numerosities—as though they are entities in the real world detectable through visual cognition. These types of entities are among the basic ontological categories of Conceptual Semantics (see Jackendoff 1983 and section 1.4). This variety of entities presents a challenge to theories of visual cognition, in that the visual system must provide structures appropriate to match up with these deictic anaphors. How does the visual system do this? The varieties of deictic anaphora also present a challenge to standard model-theoretic semantics, where reference to individuals and perhaps to events is deemed essential. The deictic anaphor in (i) is an indexical to an individual in the model. These examples show that reference to all these other entities works similarly, and so the model must contain them. Conceptual Semantics admits them as fundamental building blocks of our understanding of the world.
on bey ond zeb ra
109
moreover lead the theory toward an adequate description of the full spatial field, in that they encode relations among objects. Fourth, the 3D model level, thus enriched, does not conform to one’s intuitive stereotype of what information the visual system delivers. The ‘visual world’ is not simply a collection of holistic objects—‘statues in the head’. Rather, the 3D representation is teeming with elements that one does not ‘see’, such as the hierarchical part–whole structure and the coordinate axis systems proposed by Marr, and now regions determined by the axes of objects and trajectories being traversed by objects in motion. In other words, the information necessary to encode spatial understanding includes a great deal that, while still purely geometric rather than conceptual, is not part of visual appearance as such. Some (though I hope not all) readers may question this idea: how can a theory of perception countenance the presence of abstract visual information that one cannot see? From the vantage point of linguistic theory, though, such a situation seems entirely as it should be. A major thrust of generative linguistic theory is that there are elements of hierarchical and abstract structure which one cannot hear and that one does not speak, but which must play a role in explicating one’s understanding of language. Moreover, there is nothing inherently suspect in investigating such entities: one can ask meaningful empirical questions about them and choose intelligently between alternative hypotheses. This is exactly the situation we have arrived at here in visual theory, paralleling language. If it calls for new methodology and new forms of argumentation, so be it; the flowering of linguistic theory in the last quarter-century has been a direct result of giving up the expectation of overly concrete solutions. I want to conclude by replying to two methodological objections raised by Zenon Pylyshyn in response to the present proposal.3 His first point is that it may be too premature in the development of linguistic and visual theory to responsibly attempt to connect them. While I agree with Pylyshyn that the studies of conceptual structure and the 3D model are both still in their infancy, I do not see this as reason to abstain from approaching their relationship. The necessity to translate back and forth between the two forms of structure is an important boundary condition on the adequacy of both theories, and thus each can provide a source of stimulation for the other—as we have already seen in the present study. It seems to me, in fact, that the sooner this boundary condition can be brought to bear, the better, for it can help prevent false
3 These points were raised at the conference on Mental Representation in Vancouver in February 1986, sponsored by Simon Fraser University and the University of British Columbia.
110
on beyond zebra
moves at fundamental choice points in the development of both semantic and visual theories. Pylyshyn’s second point stems from his long-standing (1973, 1980) opposition to theories that require image-like representations. (It should be noted that some of his standard objections do not pertain to the 3D model. For instance, the 3D model is highly structured, and well-defined computations can be carried out over it, so it is not the ‘holistic’ sort of representation to which he rightly objects.) The basic issue is one of parsimony: if all concepts can be expressed in algebraic (or propositional) form, why posit an additional geometrical level of representation? The use of 3D model structures for encoding long-term memory would seem to be redundant at best. There are various levels of reply. Most superficially, it has not to my knowledge been demonstrated that all visual concepts can indeed be encoded propositionally. Moreover, by now it should be unquestionable that information cannot be mapped directly from retinal configuration to full object constancy, expressed propositionally. Rather, it seems inevitable that visual perception employs some intermediate stage(s) of encoding with geometric properties. Thus it would appear difficult, if not impossible, to dispense with geometric levels of representation altogether. In particular, Pylyshyn offers not a hint of how to replace the geometric representations of Marr’s theory with something he finds more palatable. This line of reply to Pylyshyn is rather brute force, in that it simply denies that propositional representations alone are adequate for accounting for how we talk about what we see. But the issue of parsimony itself deserves a closer look is well. Interestingly, the argument closely parallels the debate over Generative Semantics in the early 1970s. Postal (1972), like Pylyshyn, argued that the best theory is the one with the fewest levels in it. In particular, a theory of language that coalesced syntactic and semantic information into a single level (Generative Semantics) was to be preferred to a theory that had separate syntactic and semantic levels (Interpretive Semantics). Chomsky’s (1972) reply to Postal was that one does not measure the value of theories in terms of raw numbers of components, but in terms of their explanatory power. For instance, suppose that in order to describe some phenomenon in a one-component theory one must introduce ad hoc devices that perforce miss generalizations, whereas a two-component theory expresses these generalizations naturally—and suppose this situation arises time after time. In such a case, the two-component theory is clearly preferable. Chomsky cites numerous linguistic phenomena that cannot be described in their full generality unless one adopts a theory that separates syntactic from semantic structure. It was arguments of this sort that led to the demise of Generative Semantics.
on bey ond zeb ra
111
A reply much like Chomsky’s can be raised in response to Pylyshyn. We have found here a natural division of labor between conceptual structure and the 3D model. There are algebraic characteristics of concepts that are totally inexpressible in the 3D model, for example the type-token distinction and the distinction between objects and actions. On the other hand, there are geometric characteristics of concepts that are at best blatantly ad hoc in conceptual structure, for example the distinction between the necks of ducks and geese and the distinction between jogging and loping. By sharing the encoding of spatial concepts between the two levels of representation, each level only has to do what it is good at; neither need be cluttered with ad hoc descriptive devices. Moreover, we have found substantial points of correspondence through which the two levels can be coordinated. This seems to me a promising beginning for an explanatory theory of spatial thought and its linguistic expression. The message of the present chapter, then, is that in pursuit of such a theory it is necessary to find a synthesis that transcends the boundaries of pure linguistic theory and pure visual theory—in short, to go On Beyond Zebra.
chapter 4
The Architecture of the Linguistic–Spatial Interface (1996)
4.1 Introduction How do we talk about what we see? More specifically, how does the mind/brain encode spatial information (visual or otherwise), how does it encode linguistic information, and how does it communicate between the two? This chapter lays out some of the boundary conditions on a satisfactory answer to these questions, and illustrates the approach with some sample problems. The skeleton of an answer appears in Figure 4.1. At the language end, speech perception converts auditory information into linguistic information, and speech production converts linguistic information into motor instructions to the vocal tract. Linguistic information includes at
[This chapter appeared originally in P. Bloom, M. Peterson, L. Nadel, and M. Garrett (eds.), Language and Space (Cambridge, Mass.: MIT Press, 1996), 1–30. It is reprinted here by permission of MIT Press.] I am grateful to Barbara Landau, Manfred Bierwisch, Paul Bloom, Lynn Nadel, Bhuvana Narasimhan, and Emile van der Zee for extensive discussion, in person and in correspondence, surrounding the ideas in this paper. Further important suggestions came from participants in the Conference on Space and Language sponsored by the Cognitive Anthropology Group at the Max-Planck Institute for Psycholinguistics in Nijmegen in December 1993 and of course from the participants in the Arizona workshop responsible for the present volume (Bloom et al. 1996). This research was supported in part by NSF Grant IRI-92-13849 to Brandeis University, by a Keck Foundation grant to the Brandeis University Center for Complex Systems, and by a Fellowship to the author from the John Simon Guggenheim Foundation.
t h e li n g u is t i c – s p a t i a l in t e r f a c e
113
auditory signals linguistic information
visual information
eye
motor signals LANGUAGE
VISION
Fig. 4.1 Coarse sketch of the relation between language and vision
least some sort of phonetic/phonological encoding of speech.1 At the visual end, the processes of visual perception convert retinal information into visual information, which includes at least some sort of retinotopic mapping. The connection between language and vision is symbolized by the central double-headed arrow in Figure 4.1. Since it is clear that there cannot be a direct relation between a retinotopic map and a phonological encoding, the solution to our problem lies in elaborating the structure of this double-headed arrow.
4.2 Representational modularitya Filling Figure 4.1 out a bit, each faculty ramifies into a number of linked levels of representation. Figure 4.2 sketches the levels of representation I will be working with here and the interfaces among them. The Parallel Architecture for language (see chapter 1) can be recognized as embedded in this figure. Each label in Figure 4.2 stands for a level of representation. The arrows stand for interfaces. Double arrows can be thought of either as interfaces that process bidirectionally, or as pairs of complementary unidirectional interfaces (the correct choice is an empirical question). For instance, the phonology–syntax interface functions from left to right in speech perception and from right to left in speech production. Figure 4.2 expands the ‘linguistic representation’ of Figure 4.1 into three levels involved with language: the familiar levels of phonology and syntax, plus
1
This is an oversimplification, because of the existence of languages that make use of the visual/ gestural modalities. See Emmorey 1996. a Excised from this section is a statement of representational modularity (Jackendoff 1987c, chapter 12; 1992, chapter 1), also called structure-constrained modularity (Jackendoff 2002, chapter 7). Representational modularity is a processing notion that is strictly speaking not germane to the present discussion. I think its role in the original version of this chapter was as a stand-in for the notion of the Parallel Architecture, which was at that time about a year away from being formulated explicitly (though it was implicit in earlier work such as Chapter 3). Accordingly, in the present version I have made minor adjustments in terminology throughout.
114
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e g-p audition, smell, emotion, . . . auditory
phonology
syntax
conceptual structure
imagistic
spatial representation
motor
eye
retinotopic
auditory localization, haptic, proprioception, action, . . .
Fig. 4.2 Slightly less coarse sketch of the relation between language and vision
conceptual structure, a central level of representation that interfaces with many other faculties. Similarly, ‘visual representation’ in Figure 4.1 is expanded into levels of retinotopic, imagistic, and spatial representation, corresponding roughly to Marr’s (1982) primal sketch, 2½D sketch, and 3D model respectively; the last of these again is a central representation that interfaces with other faculties.b In this picture, the effect of Fodorian faculty-sized modules emerges through the link-up of a series of levels of representation and interfaces; communication among faculties is accomplished by interfaces of exactly the same general character as the interfaces within faculties. The crucial interface for our purposes here is that between the most central levels of the linguistic and visual faculties, conceptual structure and spatial representation. Before examining this interface, we have to discuss two things: (1) the general character of interfaces between representations (section 4.3), and the general character of conceptual structure and spatial representation themselves (sections 4.4 and 4.5).
4.3 The character of interface mappings [The material originally in this section has been transposed to sections 1.2 and 1.5. Reviewing these sections, we should expect the interface between two levels of representation to be a ‘dirty’ mapping—not a one-to-one isomorphism but a partial homomorphism. Typically, only some aspects of each representation b This exposition of the visual faculty steps away from Chapter 3’s explicit dependence on Marr’s theory, which by the time this was written had already lapsed from general favor. The passage here is therefore trying to phrase the approach in more general terms. In particular, the idea of spatial representation, which serves the function of Marr’s 3D model, is no longer vision-specific, insofar as it also interacts with auditory localization, haptic perception (sense of touch), proprioception (body sense), and the formulation of action. The amodal character of this level was pointed out but not emphasized in Jackendoff 1987c, chapter 12; the term spatial representation originates in Jackendoff and Landau 1991.
t h e li n g u is t i c – s p a t i a l in t e r f a c e
115
are mapped into the other, and the mappings are not between primitives of the two levels, but between composite constituents. Some aspects of the interface are governed by general rules, but others may well be particular learned associations. Within language, the words are just such specialized interface rules.]
4.4 Conceptual structure Let us now turn to the crucial modules for the connection of language and spatial cognition, conceptual structure (CS) and spatial representation (SR). The idea that these two levels share the work of cognition is in a sense a more abstract version of Paivio’s (1971) dual coding hypothesis. To use the terms of Mandler 1996, Tversky 1996, and Johnson-Laird 1996, CS encodes ‘propositional’ representations, and SR is the locus of ‘image-schema’ or ‘mental model’ representations. [As discussed in section 1.3,] conceptual structure is an encoding of linguistic meaning that is independent of the particular language whose meaning it encodes. It is an ‘algebraic’ representation, in the sense that conceptual structures are built up out of discrete primitive features and functions. Although CS supports formal rules of inference, it is not ‘propositional’ in the standard logical sense, in that (a) propositional truth and falsity are not the only issue it is designed to address, and (b) unlike propositions of standard truth-conditional logic, its expressions refer not to the real world or to possible worlds, but rather to the world as humans conceptualize it. Conceptual structure is also not entirely digital, in that some conceptual features and some interactions among features have continuous (i.e. analogue) characteristics which permit stereotype and family resemblance effects to be formulated. [A minimal theory of conceptual structure must at the very least encode the residue of meaning that cannot be encoded in any sensory/perceptual format. Here is a sample of nonsensory distinctions of meaning made by natural language, as suggested in sections 3.4 and 3.6:] 1. CS must contain pointers to all the sensory modalities, so that sensory encodings may be accessed and correlated (see next section). 2. CS must contain the distinction between tokens and types, so that the concept of an individual (say, a particular dog) can be distinguished from the concept of the type to which that individual belongs (all dogs, or dogs of its breed, or dogs that it lives with, or all animals, or whatever). 3. CS must contain the encoding of quantification and quantifier scope.
116
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
4. CS must be able to abstract actions (say running) away from the individual performing the action (say Harry or Harriet running). 5. CS must encode taxonomic relations (e.g. a bird is a kind of animal). 6. CS must encode social predicates such as ‘is uncle of’, ‘is a friend of’, ‘is fair’, and ‘is obligated to’. 7. CS must encode modal predicates, such as the distinction between ‘is flying’, ‘isn’t flying’, ‘can fly’, and ‘can’t fly’. I leave it to readers to convince themselves that none of these aspects of meaning can be represented in sensory encodings without using special annotations (such as pointers, legends, or footnotes); CS is, at the very least, the systematic form in which such annotations are couched. [Further material on CS has been excised here and incorporated into sections 1.3 and 1.4.]
4.5 Spatial representation For the theory of spatial representation—the encoding of objects and their configurations in space—we are on far shakier ground. The best articulated (partial) theory of spatial representation I know of is Marr’s (1982) 3D model, with Biederman’s (1987) ‘geonic’ constructions as a particular variant. [See subfootnote b, p. 114.] Here are some criteria that a spatial representation (SR) must satisfy. 1. SR must encode the shape of objects in a form that is suitable for recognizing an object at different distances and from different perspectives, i.e. it must solve the classic problem of object constancy.2 2. SR must be capable of encoding spatial knowledge of parts of objects that cannot be seen, for instance the hollowness of a balloon (Landau and Jackendoff 1993). 3. SR must be capable of encoding the degrees of freedom in objects that can change their shape, for instance human and animal bodies. 4. SR must be capable of encoding shape variations among objects of similar visual type, for example making explicit the range of shape variations characteristic of different cups. That is, it must support visual object categorization as well as visual object identification. 2 As a corollary, SR must support the generation of mentally rotated objects whose perspective with respect to the viewer changes during rotation, in particular for rotations on an axis parallel to the picture plane, in which different parts of the object are visible at different times during rotation—a fact noted by Kosslyn (1980).
t h e li n g u is t i c – s p a t i a l in t e r f a c e
117
5. SR must be suitable for encoding the full spatial layout of a scene, and for mediating among alternative perspectives (‘What would this scene look like from over there?’), so that it can be used to support reaching, navigating, and giving instructions (Tversky 1996). 6. SR must be independent of spatial modality, so that haptic information, information from auditory localization, and felt body position (proprioception) can all be brought into registration with one another. It is important to know by looking at an object where you expect to find it when you reach for it and what it should feel like when you handle it. Strictly speaking, factors 5–6 here go beyond the Marr and Biederman theories of object shape. But there is nothing in principle to prevent these theories from serving as a component of a fuller theory of spatial understanding, rather than strictly as theories of high-level visual shape recognition. By the time visual information is converted into shape information, its strictly visual character is lost—it is no longer retinotopic, for example; nor, as Marr stresses, is it confined to the observer’s point of view.3,c SR contrasts with CS in that it is geometric (or even quasi-topological) in character, rather than algebraic. But on the other hand, it is not ‘imagistic’—it is not to be thought of as encoding ‘statues in the head’. An image is restricted to a particular point of view, whereas SR is not. An image is restricted to a particular instance of a category (recall Berkeley’s objection to images as the vehicle of thought: how can an image of a particular triangle stand for all possible triangles?4), whereas SR is not. An image cannot represent the unseen parts of
3 Some colleagues have objected to Marr’s characterizing the 3D sketch as ‘object-centered’, arguing that objects are always seen from some point of view or other—at the very least the observer’s. However, I interpret ‘object-centered’ as meaning that the encoding of the object is independent of point of view. This neutrality permits the appearance of the object to be computed as necessary to fit the object into the visual scene as a whole, viewed from any arbitrary vantage point. Marr, who is not concerned with spatial layout, but only with identifying the object, does not deal with this further step of reinjecting the object into the scene. But I see such a step as altogether within the spirit of his approach. 4 A different sort of example comes from Christopher Habel (paper presented at Nijmegen space conference): the ‘image-schema’ for along, as in the road is along the river, must include the possibility of the road being on either side of the river. An imagistic representation must represent the road being specifically on one side or the other. c Seeing spatial understanding as amodal (or multimodal) rather than predominantly visual is an advance on the view in Chapter 3. As far as I can tell from a distance, current neuroscience approaches to vision (e.g. Koch 2004) don’t come close to satisfying any of these criteria other than possibly object constancy. Thus to some extent they are a retreat from Marr’s program of inquiry; my impression is that few people think in a broad way about the task that visual cognition has to accomplish. As observed in the text, I’m establishing here a task even more demanding than Marr’s.
118
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
an object—its back and inside, and the parts of it occluded from the observer’s view by other objects—whereas SR does. An image is restricted to the visual modality, whereas SR can equally well encode information received haptically or through proprioception. Nevertheless, even though SRs are not themselves imagistic, it makes sense to think of them as encoding image-schemas: abstract representations from which a variety of images can be generated. Figure 4.2 postulates a separate module of imagistic (or pictorial) representation one level toward the eye from SR. This corresponds roughly to Marr’s 2½D sketch. It is specifically visual; it encodes what is consciously present in the field of vision or visual imagery (Jackendoff 1987c, chapter 14). The visual imagistic representation is restricted to a particular point of view at any one time; it does not represent the backs and insides of objects explicitly. At the same time, it is not a retinotopic representation, since it is normalized for eye movements and it incorporates information from both eyes into a single field, including stereopsis. (There is doubtless a parallel imagistic representation for the haptic faculty, encoding the way objects feel, but I am not aware of any research on it.) It is perhaps useful to think of the imagistic representation as ‘perceptual’ and SR as ‘cognitive’. The two are related through an interface of the general sort found in the language faculty: they share certain aspects, but each has certain aspects invisible to the other. Each can drive the other through the interface: in visual perception, an imagistic representation gives rise to a spatial representation that encodes one’s understanding of the visual scene; in visual imagery, SRs give rise to imagistic representations. In other words, the relation of images to image-schemas (SRs) in the present theory is much like the relation of sentences to thoughts. Image-schemas are not skeletal images, but rather structures in a more abstract and more central form of representation.5 This layout of the visual and spatial levels of representation is of course highly oversimplified. For instance, I have not addressed the well-known division of visual labor between the ‘what-system’ and the ‘where-system’, which deal, roughly speaking, with object identification and object location respectively (O’Keefe and Nadel 1978, Ungerleider and Mishkin 1982, Farah et al. 1988, Jeannerod 1994, Landau and Jackendoff 1993).d My assumption, 5 It is unclear to me at the moment what relationship this notion of image-schema bears to that of Mandler (1992, 1996), although there is certainly a family resemblance. Mandler’s formulation derives from work such as Lakoff (1987) and Langacker (1987a), in which the notion of level of representation is not well developed, and in which no explicit connection is made to research in visual perception. I leave open for future research the question of whether the present conception can help sharpen the issues with which Mandler is concerned. d More contemporary thinking considers the dorsal or ‘where’ system as interfacing with motor representations and as involved in the formulation of action (Jeannerod 1994).
t h e li n g u is t i c – s p a t i a l in t e r f a c e
119
perhaps unduly optimistic, is that such division of labor can be captured in the present approach by further articulation of the visual–spatial modules in Figure 4.2 into smaller modules and their interfaces, much as Figure 4.2 is a further articulation of Figure 4.1.
4.6 The interface between CS and SR We come at last to the mapping between CS and SR, the crucial link between the visual system and the linguistic system. What do these two levels share, such that it is possible for an interface module to communicate between them? [Reviewing Chapter 3], the most basic unit these levels share is the notion of a physical object, which appears as a geometrical unit in SR and as a fundamental algebraic constituent type in CS.6 In addition, the Marr–Biederman theory of object shape proposes that object shapes are decomposed into geometric parts in SR. This relation maps straightforwardly into the part–whole relation, a basic function in CS that of course generalizes far beyond object parts. The notions of place (or location) and path (or trajectory) play a basic role in CS (Talmy 1983, Jackendoff 1983, Langacker 1987a); they are invoked, for instance in locational sentences such as The book is lying on the table (place) and The arrow flew through the air past my head (path). Because these sentences can be checked against visual input, and because locations and paths can be given obvious geometric counterparts, it is a good bet that these constituents are shared between CS and SR.7 (The Marr–Biederman theory does not contain places and paths, since they arise only in encoding the behavior of objects in the full spatial field, an aspect of visual cognition not addressed by these theories.) The notion of physical motion is also central to CS, and obviously it must be represented in spatial cognition so we can track moving objects. More speculatively, the notion of force appears prominently in CS (Talmy 1985, Jackendoff 1990), and to the extent that we have the impression of directly perceiving 6
Fundamental, but not necessarily primitive: Jackendoff 1991 [Chapter 5 of the present volume] decomposes the notion of object into the more primitive feature complex [Material, þbounded, inherent structure]. The feature [Material] is shared by substances and aggregrates; it distinguishes them all from Situations (Events and States), Spaces, Times, and various sorts of abstract entities. The feature [þbounded] distinguishes objects from substances, and also closed events (or accomplishments) from processes. The feature [inherent structure] distinguishes objects from groups of individuals, but also substances from aggregates and homogeneous processes from repeated events. 7 On the other hand, it is not so obvious that places and paths are encoded in imagistic representation, since we don’t literally see them except when dotted lines are drawn in cartoons. So this may be another part of SR that is invisible to imagistic representation. That is, places and paths as independent entities may be a higher-level cognitive (non-perceptual) aspect of spatial understanding, as also argued by Talmy 1996.
120
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
forces in the visual field (Michotte 1954), these too might well be shared between the two representations.8 Our discussion of interfaces in [sections 1.2 and 1.5] leads us to expect some aspects of each representation to be invisible to the other. What might some of these aspects be? Section 4.4 noted that CS encodes the token vs. type distinction (a particular dog vs. the category of dogs), quantificational relations, and taxonomic relations (a bird is a kind of animal), but that these are invisible to SR. On the other hand, SR encodes all the details of object shapes, for instance the shape of a violin or a butter knife or a German shepherd’s ears. These geometric features do not lend themselves at all to the sort of algebraic coding found in CS; they are absolutely natural to (at least the spirit of) SR. In addition to general mappings between constituent types in CS and SR, individual matchings can be learned and stored. Lexical entries for physical object words can contain a spatial representation of the object in question, in addition to their phonological, syntactic, and conceptual structure. For instance, the entry for cat might look something like (4).
(4)
Phonology: /kæt/ Syntax: +N, –V, +count, +sing CS: Individual, Type of animal, Type of carnivore Function: (often Type of Pet) SR: [3D model with motion affordances] Auditory: [sound of meowing]
In (4), the SR takes the place of what in many approaches (e.g. Rosch and Mervis 1975, Putnam 1975) has been informally termed an ‘image of a prototypical instance of the category’. The difficulty with an image of a prototype is that it is computationally nonefficacious: it does not meet the demands of object shape identification laid out as criteria 1–4 in the previous section. A more abstract spatial representation, along the lines of a Marr 3D model, meets these criteria much better; it is therefore a more satisfactory candidate for encoding one’s knowledge of what the object looks like. As suggested by the inclusion of 8 Paul Bloom has asked (p.c.) why I would consider force but not, say anger to be encoded in SR, since we have the impression of directly perceiving anger as well. The difference is that physical force has clear geometric components—direction of force and often contact between objects— which are independently necessary to encode other spatial entities such as trajectories and orientations. Thus force seems a natural extension of the family of spatial concepts. By contrast, anger has no such geometrical characteristics; its parameters belong to the domain of emotions and interpersonal relations. Extending SR to anger, therefore, would not yield any generalizations in terms of shared components.
t h e li n g u is t i c – s p a t i a l in t e r f a c e
121
a. One way to view (4) Phonology + Syntax + CS LANGUAGE
+ “SR” ???
b. Another way to view (4) Phonology + Syntax LANGUAGE
+
CS + SR ‘CONCEPT’
Fig. 4.3 Two ways to view the integration of spatial structures into lexical entries
‘auditory structure’ in (4), a lexical entry should encode (pointers to) other sensory characteristics as well. The idea, then, is that the ‘meaning’ of a word goes beyond the features and functions available in CS, in particular permitting detailed shape information in a lexical SR. (A word must have a lexical CS; it may have an SR as well.) Such an approach might be seen as threatening the linguistic integrity of lexical items: as suggested by Figure 4.3a above, it breaks out of the purely linguistic system. But an alternative view of entries like (4) places them in a different light. Suppose one deletes the phonological and syntactic structures from (4). What is left is the nonlinguistic knowledge one has of cats—the ‘concept’ of a cat, much of which could be shared by a nonlinguistic organism. Phonological and syntactic structures can then be viewed as further structures tacked onto this knowledge to make it linguistically expressible, as suggested in Figure 4.3b. With or without language, the mind has to have a way to unify multimodal representations and store them as units (that is, to establish long-term memory ‘binding’ in the neuroscience sense); (4) represents just such a unit. The structures that make this a ‘lexical item’ rather than just a ‘concept’ simply represent an additional modality into which this concept extends: the linguistic modality.e Having established general properties of the CS–SR interface, we must raise the question of exactly what information is on either side of it. How do we decide? The overall premise behind the Parallel Architecture, of course, is that each module is a specialist, and each particular kind of information belongs in a particular representation. For instance, details of shape are not duplicated in CS, and taxonomic relations are not duplicated in SR. For the general case, we can state a criterion of economy: all other things being equal, if a certain kind of
e The reconceptualization in Figure 4.3b represents a quiet but radical breaking of bonds with formal semantics and a joining of hands with psychology. Semantics is no longer about words: it is about the structure of thoughts and about how language expresses them.
122
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
distinction is encoded in SR, it should not also be encoded in CS, and vice versa. I take this maximal segregation to be the default assumption. Of course, all other things are not equal. The two modules must share enough structure that they can communicate with each other—for instance, they must share at least the notions mentioned at the beginning of this section. So we do not expect as a baseline that the information encoded by CS and SR is entirely incommensurate. Let us call this the criterion of interfacing. What evidence would help decide whether a certain kind of information is in CS as well as SR? One line of argument comes from interaction with syntax. Recall that CS is by hypothesis the form of central representation that most directly interacts with syntactic structure. Therefore, if a semantic distinction is communicated to syntax, so that it makes a syntactic difference, that distinction must be present in CS and not just SR. (Note that this criterion applies only to syntactic and not lexical differences. As pointed out in section 1.5, dog and cat look exactly the same to syntax.) Let us call this the criterion of grammatical effect. A second line of argument concerns nonspatial domains of CS. As is well known (Gruber 1965; Jackendoff 1976, 1983; Talmy 1978a; Lakoff and Johnson 1980; Langacker 1987a), the semantics of many nonspatial conceptual domains show strong parallels to the semantics of spatial concepts. Now if a particular semantic distinction appears in nonspatial domains as well as in the spatial domain, it cannot be encoded in SR alone, which by definition pertains only to spatial cognition. Rather, similarities between spatial and nonspatial domains must be captured in the algebraic structure of CS. I will call this the criterion of nonspatial extension.
4.7 A simple case: the count–mass distinction A familiar example will make these criteria clearer. Consider the count–mass distinction. SR obviously must make a distinction between single individuals (a cow), multiple individuals (a herd of cows), and substances (milk)—these have radically different appearances and spatial behavior over time. (Marr and Biederman, of course, have little or nothing to say about what substances look like.) The criterion of economy leads us to prefer that, all else being equal, SR is the only level that encodes these differences. But all else is not equal. The count–mass distinction has repercussions in the marking of grammatical number and in the choice of possible determiners (count nouns use many and few, mass nouns use much and little, for example). Hence the criterion of grammatical effect suggests that the count–mass distinction is encoded in CS also.
t h e li n g u is t i c – s p a t i a l in t e r f a c e
123
Furthermore, the count–mass distinction appears in abstract domains. For example, threat is grammatically a count noun (many threats/ much threat), but the semantically very similar advice is a mass noun (much advice/ many advices). Since the distinction between threats and advice cannot be encoded spatially—it doesn’t ‘look like anything’—the only place to put it is in CS. That is, the criterion of nonspatial extension applies to this case. In addition, the count–mass distinction is closely interwoven with features of temporal event structure such as the event–process distinction (as will be seen in Chapter 5; see also Verkuyl 1972, 1993; Dowty 1979; Hinrichs 1985; Pustejovsky 1991a). To the extent that events have a spatial appearance, it is qualitatively different from that of objects. And distinctions of temporal event structure have a multitude of grammatical reflexes. So the criteria of nonspatial extension and grammatical effect both apply again, to argue for the count–mass distinction being encoded in CS. A further piece of evidence comes from lexical discrepancies in the grammar of count and mass nouns. An example is the contrast between noodles (count) and spaghetti (mass)—nouns that pick out essentially the same sorts of entities in the world. A single one of these objects can be described as a singular noodle, but the mass noun forces one to use the phrasal form stick (or strand) of spaghetti. (In Italian, spaghetti is a plural count noun, and one can refer to a single spaghetto.) Since noodles and spaghetti pick out similar entities in the world, there is no reason to believe that they have different lexical SRs. Hence there must be a mismatch somewhere between SR and syntax. A standard strategy (e.g. Bloom 1994) is to treat them as alike in CS as well, and to localize the mismatch somewhere in the CS–syntax interface. Alternatively, the mismatch might be between CS and SR. In this scenario, CS has the option of encoding a collection of smallish objects (or even largish objects such as furniture) as either an aggregate or a substance, then syntax follows suit by treating the concepts in question as grammatically count or mass respectively.9 Whichever solution is chosen, it is clear that SR and syntax alone cannot make sense of the discrepancy. Rather, CS is necessary as an intermediary between them.
4.8 Axes and frames of reference We now turn to a more complex case with a different outcome. Three subsets of the vocabulary invoke the spatial axes of an object. I will call them collectively the axial vocabulary. 9 This leaves open the possibility of CS–syntax discrepancies in the more grammatically problematic cases like scissors and trousers. I leave the issue open.
124
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
1. The ‘axial parts’ of an object—its top, bottom, front, back, sides, and ends— behave grammatically like parts of the object, but, unlike standard parts such as a handle or a leg, they have no distinctive shape. Rather, they are regions of the object (or its boundary) determined by their relation to the object’s axes. The up– down axis determines top and bottom, the front–back axis determines front and back, and a complex set of criteria distinguishing horizontal axes determines sides and ends (Miller and Johnson-Laird 1976, Landau and Jackendoff 1993). 2. The ‘dimensional adjectives’ high, wide, long, thick, deep, etc. and their nominalizations height, width, length, thickness, and depth refer to dimensions of objects measured along principal, secondary, and tertiary axes, sometimes with reference to the horizontality or verticality of these axes (Bierwisch 1967, Bierwisch and Lang 1989). 3. Certain spatial prepositions, for instance above, below, next to, in front of, behind, alongside, left of, and right of, pick out a region determined by extending the reference object’s axes out into the surrounding space. For instance, in front of X denotes a region of space in proximity to the projection of X’s front–back axis beyond the boundary of X in the frontward direction (Miller and Johnson-Laird 1976; Landau and Jackendoff 1993; Landau 1996). By contrast, inside X makes reference only to the region subtended by X, not to any of its axes; near X denotes a region in proximity to X in any direction at all. Notice that many of the ‘axial prepositions’ are morphologically related to nouns that denote axial parts. It has been frequently noted (among many others, Miller and Johnson-Laird 1976; Olson and Bialystok 1983; practically every paper in Bloom et al. 1996) that the axial vocabulary is always used in the context of an assumed frame of reference. Moreover, the choice of frame of reference is often ambiguous; and since the frame determines the axes in terms of which the axial vocabulary receives its denotation, the axial vocabulary too is ambiguous. The literature usually invokes two frames of reference: an intrinsic or objectcentered frame, and a deictic or observer-centered frame. Actually the situation is more complex. Viewing a frame of reference as a way of determining the axes of an object, it is possible to distinguish at least eight different available frames of reference (many of these appear as special cases in Miller and Johnson-Laird 1976, who in turn cite Bierwisch 1967, Teller 1969, and Fillmore 1971, among others). A. Four intrinsic frames all make reference to properties of the object. 1. The geometric frame uses the geometry of the object itself to determine the axes. For instance, the dimension of greatest extension can determine its length (Figure 4.4a). Symmetrical geometry often implies a top-tobottom axis dividing the symmetrical halves and a side-to-side axis
t h e li n g u is t i c – s p a t i a l in t e r f a c e
125
passing from one half to the other (Figure 4.4b). A special case concerns animals, whose front is intrinsically marked by the position of the eyes. 2. In the motion frame, the front of a moving object is determined by the direction of motion. For instance, the front of an otherwise symmetrical double-ended tram is the end facing toward its current direction of motion (Figure 4.4c). 3. Two intrinsic frames depend on functional properties of the object. The canonical orientation frame designates as the top (or bottom) of an object the part which in the object’s normal orientation is uppermost (or lowermost), even if it does not happen to be at the moment. For instance, the canonical orientation of the car in Figure 4.4d has the wheels lowermost,
(a)
(b)
(c)
(d)
(e)
Fig. 4.4
126
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e so the part the wheels are attached to is the canonical bottom, even though it is pointing obliquely upward in this picture.
4. Intrinsic parts of an object can be picked out by the canonical encounter with it. For instance, the part of a house where the public enters is functionally the front (Figure 4.4e). (Inside a building such as a theater, the front is the side that the public normally faces, so that the front from the inside may be a different wall of the building than the front from the outside.) B. Four environmental frames project axes onto the object based on properties of the environment. 1. The gravitational frame is determined by the direction of gravity, regardless of the orientation of the object. In this frame, for instance, the hat in Figure 4.5a is on top of the car. 2. The geographical frame is the horizontal counterpart of the gravitational frame, imposing axes on the object based on the cardinal directions north, south, east, and west, or a similar system (Levinson 1996). 3. The contextual frame is available when the object is viewed in relation to another object, whose own axes are imposed on the first object. For instance, Figure 4.5b pictures a page on which is drawn a geometric figure. The page has an intrinsic side-to-side axis that determines its width, regardless of orientation. The figure on the page inherits this axis, and therefore its width is measured in the same direction. 4. The observer frame may be projected onto the object from a real or hypothetical observer. This frame establishes the front of the object as the side facing the observer, as in Figure 4.5c. We might call this the ‘orientation-mirroring observer frame’. Alternatively, in some languages, such as Hausa, the front of the object is the side facing the same way as the observer’s front, as in Figure 4.5d. We might call this the ‘orientationpreserving observer frame’. It should be noticed further that axes in the canonical orientation frame (Figure 4.4d) are derived from gravitational axes in an imagined normal orientation of the object. Similarly, axes in the canonical encounter frame (Figure 4.4e) are derived from a hypothetical observer’s position in the canonical encounter. So in fact only two of the eight frames, the geometric and motion frames, are entirely free of direct or indirect environmental influence.f
f In the stereotypical case, many of these frames of reference coincide. For instance, if an object is in its canonical orientation, its top is the same in the gravitational and canonical orientation frames. An object that has an intrinsic front (say an animal but not a symmetrical tram) normally moves in the direction its eyes are
t h e li n g u is t i c – s p a t i a l in t e r f a c e
(a)
(b)
(c)
(d)
127
Front Fig. 4.5
One of the reasons the axial vocabulary has attracted so much attention in the literature is its multiple ambiguity among frames of reference. In the preceding examples alone, for instance, three different uses of front appear. Only the geographical frame (in English, at least) has its own unambiguous vocabulary. Why should this be? And what does it tell us about the distribution of information between CS and SR? This will be the subject of the next section. But before going on, let us take a moment to look at how frames of reference are used in giving route directions (Levelt 1996, Tversky 1996). Levelt presents subjects with diagrams such as Figure 4.6, and asks them to tell him how to get, say, from point 1 to point 5. The route from circle 1 to circle 5 can be described in two different ways: (5) a. ‘Geographic’ frame: From 1, go up/forward to 2, right to 3, right to 4, down to 5. b. ‘Observer’ frame: From 1, go up/forward to 2, right to 3, straight/ forward to 4, right to 5.
pointed, so its front is the same in the geometric and motion frames. In (what we are told is) crab motion, the two frames do not coincide; we typically choose the geometric frame as primary and speak of the crab walking sideways—rather than saying the crab’s eyes are on the side of its head, which in turn is on the side of the animal. Of course, the classic case of frame conflict is a canonical encounter with a person. Here the geometric frame puts left and right exactly the reverse of the orientation-mirroring observer frame, so we have to be careful to talk about my left versus your left. It is worth re-emphasizing that little of this frame-of-reference issue has found its way into the neuroscience of vision.
128
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
Fig. 4.6 One of Levelt’s ‘maps’
The problem is highlighted by the step from 3 to 4, which is described as ‘right’ in (5a) and ‘straight’ in (5b). The proper way to think of this seems to be to keep track of the hypothetical traveler’s orientation. In the ‘geographic’ frame, the traveler maintains a constant orientation, so that up always means up on the page; that is, the traveler’s axes are set contextually by the page (frame B3). The puzzling case is the ‘observer’ frame, where the direction from 2 to 3 is ‘right’ and the same direction, from 3 to 4, is ‘straight’ or ‘forward’. Intuitively, as Levelt and Tversky point out, one pictures oneself traveling through the diagram. From this the solution follows immediately: ‘forward’ is determined by the observer’s last move, i.e. using the motion frame (A2). The circles, which have no intrinsic orientation, play no role in determining the frame. If they are replaced by landmarks that do have intrinsic axes, as in Tversky’s examples, say a building, a third possibility emerges, that of setting the traveler’s axes contextually by the landmarks (B3 again). And of course geographical axes (B1) are available as well if the cardinal directions are known.
4.9 The lexical encoding of axial vocabulary Narasimhan 1993 reports an experiment that has revealing implications for the semantics of the axial vocabulary. Subjects were shown irregular shapes (‘Narasimhan figures’) of the sort shown in Figure 4.7, and asked to mark on them their length, width, height, or some combination of the three. Because length, width, and height depend on choice of axes, responses revealed subjects’ judgments about axis placement. This experiment is unusual in its use of irregular shapes. Previous experimental research on axial vocabulary with which I am familiar (e.g. Bierwisch and Lang 1989, Levelt 1984) has dealt only with rectilinear figures or familiar objects, often only in rectilinear orientations. In Narasimhan’s experiment, the subjects have to compute axes online, based on visual input; they cannot simply call up intrinsic axes stored in long-term memory as part of the canonical representation of a familiar object.
t h e li n g u is t i c – s p a t i a l in t e r f a c e
129
Fig. 4.7
But of course linguistic information is also involved in the subjects’ responses. In particular, the dimension that the subject is asked to mark influences the choice of axis, as might be expected from the work of Bierwisch and Lang. Length biases the subject in favor of intrinsic geometric axes (longest dimension), while height biases the subject toward environmental axes (gravitational or page-based frame). So, confronted with a shape such as Figure 4.8a, whose longest dimension is oblique to the contextual vertical, subjects tended to mark its length as an oblique, and its height as an environmental vertical. Sometimes subjects even marked these axes on the very same figure: subjects did not insist by any means on orthogonal axes! The linguistic input, however, was not the only influence on the choice of axes. Details in the shape of the Narasimhan figure also exerted an influence. For example, Figure 4.8b has a flattish surface near the (contextual) bottom.
Fig. 4.8
130
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
Some subjects (8 percent) apparently interpreted this surface as a base that had been rotated from its canonical orientation, and drew the height of the figure as an axis orthogonal to this base, that is, as a ‘canonical vertical’. Nothing in the linguistic input created this new possibility: it had to be computed online from the visual input. As a result of this extra possibility, this shape presented three different choices for its axis system, as shown in the figure. We see, then, that linguistic and visual input interact intimately in determining subjects’ responses in this experiment. However, the Parallel Architecture does not allow us to just leave it at that. We must answer the question: At what level of representation does this interaction take place? The obvious choices are CS and SR. The fact that the subjects actually draw in axes shows that the computation of axes must involve SR.g The angle and positioning of a drawn axis is continuously variable, in a way expected in the geometric SR but not expected in the algebraic feature complexes of CS. How does the linguistic input get to SR, so it can influence the subjects’ response? That is, at what levels of representation do the words length, width, and height specify the axes and frames of reference they can pick out? There are two possibilities: 1. The CS hypothesis. The axes could be specified in the lexical entries of length, width, and height by features in CS such as [+ maximal], [+ vertical], [+ secondary]; the frames of reference could be specified by CS features such as [+ contextual], [+ observer]. General correspondences in the CS–SR interface would then map features into the geometry of SR. According to this story, when subjects judge the axes of Narasimhan figures, the lexical items influence SR indirectly, via these general interpretations of the dimensional features of CS. (This is, I believe, the approach advocated by Bierwisch and Lang.) 2. The SR hypothesis. Alternatively, we know that lexical items may contain elements of SR such as the shape of a dog. Hence it is possible that the lexical entries of length, width, and height also contain SR components that specify axes and frames of reference directly in the geometric format of SR. This would allow the axes and reference frames to be unspecified (or largely so) in the CS of these words. In this story, when subjects judge the axes of Narasimhan figures, the SR of the lexical items interacts directly with SR from visual input. I propose that the SR hypothesis is closer to correct. The first argument comes from the criterion of economy. Marr (1982) demonstrates, and this experiment g It also shows that axes can be computed online: they do not always arise as part of the memorized shape description of familiar objects. This is a challenge for theories of object identification based on memorized exemplars.
t h e li n g u is t i c – s p a t i a l in t e r f a c e
131
confirms, that people use SR to pick out axes and frames of reference in novel figures. In addition, people freely switch frames of reference in visuomotor tasks. For example, we normally adopt an egocentric (or observer) frame for reaching, but an environmental frame for navigating—in the latter, we see ourselves moving through a stationary environment, not an environment rushing past.10 These are SR functions, not CS functions. Consequently, axes and frames of reference cannot be eliminated from SR. This means that a CS feature system for these distinctions at best duplicates information in SR—it cannot take the place of information in SR. Next consider the criterion of grammatical effect. If axes and frames of reference can be shown to have grammatical effects, it is necessary to encode them in CS. But in this domain, unlike the count–mass system, there seem to be few grammatical effects. The only thing special about the syntax of the English axial vocabulary is that dimensional adjectives and axial prepositions can be preceded by measure phrases, as in three inches long, two miles wide (with dimensional adjectives), and four feet behind the wall, seven blocks up the street (with axial prepositions). Other than dimensional adjectives, the only English adjective that can occur with a measure phrase is old; such pragmatically plausible cases as eighty degrees hot and twelve pounds heavy are ungrammatical. Similarly, many prepositions do not occur with measure phrases (ten inches near the box); and those that do are for the most part axial (though away, as in a mile away from the house, is not).11 Thus the fact that a word pertains to an axis does seem to make a grammatical difference. But that is about as far as it goes. No grammatical effects seem to depend on which axis a word refers to, much less which frame of reference the axis is computed in, at least in English.12 So the criterion of grammatical effect 10 For a recent discussion of the psychophysics and neuropsychology of the distinction between environmental motion and self-motion, see Wertheim 1994 and its commentaries. Wertheim, however, does not appear to address the issue, crucial to the present enterprise, of how this distinction is encoded so that further inferences can be drawn from it—i.e. the cognitive consequences of distinguishing reference frames. 11 Measure phrases also occur in English adjective phrases as a specifier of the comparatives more/-er than and as . . . as, for instance ten pounds heavier (than X), three feet shorter (than X), six times more beautiful (than X), fifty times as funny (as X). Here they are licensed not by the adjective itself, but by the comparative morpheme. 12 Bickel 1994 however, points out that the Nepalese language Belhare makes distinctions of grammatical case based on frame of reference. In a ‘personmorphic’ frame for right and left, the visual field is divided into two halves, with the division line running through the observer and the reference object; this frame requires the genitive case for the reference object. In a ‘physiomorphic’ frame for right and left, the reference object projects four quadrants whose centers are focal front, back, left, and right; this frame requires the ablative case for the reference object. I leave it for future research to ascertain how widespread such grammatical distinctions are and to what extent they might require a weakening of my hypothesis.
132
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
dictates at most that CS needs only a feature that distinguishes axes of objects from other sorts of object parts; the axial vocabulary will contain this feature. Distinguishing axes from each other and frames of reference from each other appears unnecessary on grammatical grounds. Turning to the criterion of nonspatial extension, think about the use of axis systems and frames of reference in nonspatial domains. It is well known that analogues of spatial axes occur in other semantic fields, and that axial vocabulary generalizes to these domains (Gruber 1965, Jackendoff 1976, Talmy 1978a, Langacker 1987a, Lakoff 1987). But all other axis systems I know of are only one-dimensional, for example numbers, temperatures, weights, ranks, and comparative adjectives (more/less beautiful/salty/exciting/etc.). A cognitive system with more than one dimension is the familiar three-dimensional color space, but one does not describe differences in color using any sort of axial vocabulary. Kinship systems might be another multidimensional case, and again the axial vocabulary is not employed. In English, when a nonspatial axis is invoked, the axis is almost always up/down (higher number, lower rank, of higher beauty, lower temperature, my mood is up, etc.). Is there a reference frame? One’s first impulse is to say that the reference frame is gravitational—perhaps because we speak of the temperature rising and falling and of rising in the ranks of the army, and because rise and fall in the spatial domain pertain most specifically to the gravitational frame. But on second thought, we really wouldn’t know how to distinguish among reference frames in these spaces—what would it mean to distinguish an intrinsic upward from a gravitational upward, for example? About the only exception to the use of the vertical axis in nonspatial domains is time, a one-dimensional system that goes front to back.13 Time is also exceptional in that it does display reference frame distinctions. For instance, one speaks of the times before now, where before means ‘prior to’, as though the observer (or the ‘front’ of an event) is facing the past. But one also speaks of the hard times before us, where before means ‘subsequent to’, as though the observer is facing the future. A notion of frame of reference also appears in social cognition, where we speak of adopting another’s point of view in evaluating their knowledge or 13 A number of people have also pointed out the political spectrum, which goes from right to left. According to the description of Bickel 1997, the Nepalese language Belhare is a counterexample to the generalization about time going front to back: a transverse axis is used for measuring time, and an up–down axis is used for the conception of time as an opposition of past and future. [Mandarin is reported also to have a system of time words that invokes an up–down axis (Boroditsky 2001; January and Kako 2007).]
t h e li n g u is t i c – s p a t i a l in t e r f a c e
133
attitudes. But compared to spatial frames of reference, this notion is quite limited: it is analogous to adopting an observer frame for a different (real or hypothetical) observer; there is no parallel to any of the other seven varieties of reference frames. Moreover, in the social domain there is no notion of axis that is built from these frames of reference. So again an apparent parallel proves to be relatively impoverished. In short, very little of the organization of spatial axes and frames of reference is recruited for nonspatial concepts. Thus the criterion of nonspatial extension also gives us scant reason to encode in CS all the spatial distinctions among three-dimensional axes and frames of reference. All we need for most purposes is the distinction between the vertical and other axes, plus some special machinery for time and perhaps for social point of view. Certainly nothing outside the spatial domain calls for the richness of detail needed for the spatial axial vocabulary. Our tentative conclusion is that most of this detail is encoded only in the SR component of the axial vocabulary, not in the CS component; it thus parallels such lexical SR components as the shape of a dog. Let me call this the ‘Mostly SR’ hypothesis. A skeptic committed to the CS hypothesis might raise a ‘functional’ argument against this conclusion: Perhaps multiple axes and frames of reference are available in CS, but we don’t recruit them for nonspatial concepts because we have no need for them in our nonspatial thought. Or perhaps the nature of the real world does not lend itself to such thinking outside of the spatial domain, so such concepts cannot be used sensibly. If one insists on a ‘functional’ view, I would urge quite a different argument. It would often be extremely useful for us to be able to think in terms of detailed variation of two or three nonspatial variables, say the relation of income to educational level to age, but in fact we find it very difficult. For a more ecologically plausible case, why do we inevitably reduce social status to a linear ranking, when it so clearly involves many interacting factors? The best way we have of thinking multidimensionally is to translate the variables in question into a Cartesian graph, so that we can apply our multidimensional spatial intuitions to the variation in question—we can see it as a path or a region in space. This suggests that CS is actually relatively poor in its ability to encode multidimensional variation; we have to turn to SR to help us encode it. This is more or less what would be predicted by the Mostly SR hypothesis. That is, the ‘functional’ argument can be turned around and used as evidence for the Mostly SR hypothesis. The case of axes and frames of reference thus comes out differently from the case of the count–mass distinction. This time we conclude that most of the relevant distinctions are not encoded in CS, but only in SR, one level further removed from syntactic structure.
134
t h e l i n g u i s t i c – s p a t ia l i n t e r f a c e
This conclusion is tentative in part because of the small amount of linguistic evidence adduced for it so far—one would certainly want to check the data out crosslinguistically before making a stronger claim. But it is also tentative because we don’t have enough formal theory of SR to know how it encodes axes and frames of reference. It might turn out, for instance, that the proper way to encode the relevant distinctions is in terms of a set of discrete (or digital) annotations to the geometry of SR. In such a case, it would be hard to distinguish an SR encoding of these distinctions from a CS encoding. But in the absence of a serious theory of SR, it is hard to know how to continue this line of research.h
4.10 Final thoughts In order to sort out empirical issues in the relation of language to spatial cognition, it is useful to think in terms of the Parallel Architecture. This forces us to sort out the levels of representation involved in language, abstract conceptual thought, and spatial cognition, and to take seriously the issue of how these levels communicate with one another. In looking at any particular phenomenon within this framework, the crucial question has proved to be at which level(s) of representation it is to be encoded. We have examined cases where the choice between CS and SR comes out in different ways: this shows that the issue is not a simple prejudged matter, and must be evaluated for each case. However, for the moment, we are at the mercy of the limitations of theory. Compared to the richness of phonological and syntactic theory, the theory of CS is in its infancy; and SR, other than the small bit of work by Marr and Biederman, is hardly even in gestation. This makes it difficult to decide among (or even to formulate) competing hypotheses in any more than sketchy fashion. It is hoped that the present article spurs theorists to remedy the situation.
h Vastly more crosslinguistic data is available now than when this passage was written, but so far as I know, no one has used this evidence to address the present question of how frame-of-reference information is distributed between CS and SR. The challenge still stands.
Remarks on Chapter 5 The problems addressed in this chapter and the next are quite possibly the most difficult I’ve ever worked on. The initial puzzle was how to account for the unspoken sense of repetition in Talmy’s (1978a) example The light flashed until dawn. Up to that point I had been rather successful doing Conceptual Semantics in terms of the spatial primitives mentioned in Chapter 1, plus their extension to other semantic fields such as possession and change of properties. But Talmy’s example posed one of those problems like the four-color theorem in mathematics, where solving an innocent little puzzle turns out to require major theoretical innovations. Work by Renaat Declerck, Christer Platzack, and Henk Verkuyl began to reveal the scope of the problem: the sense of repetition can be induced by interactions among many different parts of the sentence, both lexical and grammatical. Although I wouldn’t have put it this way at the time, this strikes a blow against Fregean composition (see section 1.5), which takes sentence meaning to be constructed by simply pasting together word meanings in accordance with syntactic structure. It also strikes a blow against the assumption in mainstream generative grammar of ‘interface uniformity’, which requires underlying syntactic structure (or syntactic Logical Form) to be isomorphic to semantic structure. In this particular subsystem of language, the lexical and syntactic sources of meaning are extremely heterogeneous and cannot be reduced to uniform structures. It was clear early on that the system cuts across many of the ontological categories of Semantics and Cognition (see section 1.3). It concerns not just repetition of events but its parallel in the plurality of objects, and not just the event–process distinction but its parallel in the mass–count distinction. It so happens that the Conceptual Semantics formalism makes it easy to express these parallels between objects and events—whatever the grammatical differences between nouns and verbs—and in fact takes this parallelism as a central aspect of the analysis (Semantic Structures called it ‘X-Bar semantics’). The parallels carry over into the Place and Path system as well, and again prove to be completely natural in the Conceptual Semantics formalism as compared to other frameworks on the market, which typically do not even recognize categories of Places and Paths.
136
remarks on chapter 5
Even with these tools at my disposal, I had to experiment for over ten years before I got a formal system that worked. ‘Parts and Boundaries’ (1991, here Chapter 5) introduced four major innovations, all of which were essential to the solution. The pair of features +bounded and +internal structure that distinguish singular count, plural count, and mass NPs, as well as their counterparts in the verbal (or aspectual) and prepositional system. A system of two binary features yields four possible values, of which singular, plural, and mass are three. A key innovation in the system is recognizing that the fourth value is groups, which to my knowledge do not play a role in other theories of this domain. . The system of six functions that map entities of one type into another, enabling the theory to capture plurality, partitive, and compositional relations, as well as their inverses. . The system of dimensionality in the spatial domain (developed in Chapters 3 and 4), and its extension into the event and locational systems. This system offers an insightful analysis of boundaries, and, to my knowledge, it is unique to the present approach. . The use of coercion to account for pieces of meaning that are left unexpressed, such as the repetition in Talmy’s example. At the time ‘Parts and Boundaries’ was written, the notion of coercion was just beginning to be explored (at least in the communities I frequented), and I referred to it as ‘rules of construal’. (I have changed the terminology in the text of the chapter.) In the intervening nearly two decades, coercion has become a well-accepted phenomenon (again, in certain communities), and it has even been the subject of considerable psychological experimentation. But at that time, as a violation of Fregean compositionality, it was kind of subversive. .
These innovations led to a great number of substantial descriptive insights, in particular creating a much more highly structured system of semantic primitives to replace the list of Semantics and Cognition (and Chapter 1), and also opening up many new domains of words and constructions for description in terms of Conceptual Semantics. However, all this did not entirely account for Talmy’s example. The remaining problem was how all the boundedness and plurality features of the constituents of a sentence percolate up to produce the telicity of the sentence as a whole. The necessary piece did not work itself out till five years later, in ‘The Proper Treatment of Measuring Out . . . ’ (1996, here Chapter 6). The idea is that the temporal structure of a sentence is the product of ‘structure-preserving binding’ with other constituents; for example, the temporal structure of a
remarks on chapter 5
137
motion event is correlated with the spatial structure of the path of motion. The formalism again makes essential use of the notion of dimensionality, derived from the analysis of spatial structure. The outcome, to my delight, was not only a revealing account of aspectuality that covered far more ground than most approaches in the literature, but also a further unification of the Conceptual Semantics formalism—and a fascinating approach to quantification that treated it as entirely parallel to aspectuality. My impression is that these papers have not had much impact on the subcommunities of linguists most concerned with aspectuality, probably because of the unfamiliar premises and formalisms of Conceptual Semantics. I recently had the occasion to give a talk on this material, and the reaction of a prominent formal semanticist was that this was all well and good, but it didn’t explain anything. My somewhat astonished reply was that, among other things, I had unified mass–count semantics with aspectuality, I had found a surprising relation between Inchoatives and Goals, I had revealingly formalized the Vendler classes, I had solved the ‘imperfective paradox’, and I had come to an understanding of the basic nature of ‘measuring out’ and telicity, with possible extensions to quantification. If that isn’t explanation, I’m not sure what counts.
chapter 5
Parts and Boundaries (1991)
5.1 The framework The issue addressed in this chapter is the set of primitives and principles of combination for the conceptual domain dealing with parts and boundaries. These are to be universal: they define what there is for language to express, and they do not depend on the vehicle of expression. We will also be concerned with the correspondence (or interface) rules that determine the mapping from conceptual structure into syntactic and morphological structure of English. Such rules are of the form ‘Such-and-such a configuration in conceptual structure corresponds to such-and-such a syntactic configuration.’ They thus must contain two structural descriptions, one for each of the levels being placed in correspondence. Since the syntactic side of the correspondence is in part language-particular, it is to be expected that the correspondence rules will also be language-particular, though undoubtedly constrained by principles of Universal Grammar that pertain to the correspondence rule component. [Reviewing the discussion in Chapter 1,] a lexical item can be seen as a correspondence or long-term memory association between well-formed fragments of phonological, syntactic, and conceptual structure. Hence the lexicon is conceived of as part of the correspondence rule component. The leading [This chapter originally appeared in Cognition 41 (1991), 9–45, and is reprinted here by permission of the editor and the publisher, Elsevier.] This research was supported in part by NSF Grants IST 84-20073 and IRI 88-08286 to Brandeis University. I am grateful to Steven Pinker, Paul Bloom, and three anonymous referees for useful comments on earlier versions of this paper.
pa rts and boundar ies
139
questions of lexical semantics then come to be framed as: (a) What fragments of conceptual structure can be encoded in lexical items (of, say, English)? (b) When lexical items are combined syntactically, how are they correspondingly combined in conceptual structure, and what principles license these correspondences? [The rest of the original text of this section has been incorporated into Chapter 1.]
5.2 The technology of Conceptual Semantics [This section has been incorporated into Chapter 1.]
5.3 The puzzle and a preliminary solution The problem that motivates the present investigation might be first illustrated with a sentence discussed by Talmy (1978a): (5) The light flashed until dawn. (5) conveys a sense of the light flashing repetitively. However, the ‘core’ sentence the light flashed suggests not repetition but a single flash. Nor does the sense of repetition come from until dawn: Bill slept until dawn, for instance, does not express repeated acts of sleeping. Hence there is evidently no lexical item in the sentence that can be said to contribute the sense of repetition; it must arise from combining the words into the sentence. Thus three questions must be answered: (a) How is repetition encoded in conceptual structure? (b) What principles of correspondence license its use in (5), despite the absence of a lexical item that contains it? (c) Why is it required in the interpretation of (5)? To get a little more feel for the problem, let us explore some related examples. (6) Until dawn, a. Bill slept. b. the light flashed. [repetition only] c. lights flashed. [each may have flashed only once] d. Bill ate the hot dog. e. Bill ate hot dogs. f. Bill ate some hot dogs. g. Bill was eating the hot dog. h. ?Bill ran into the house. [repetition only] i. people ran into the house. [each may have entered only once] j. ?some people ran into the house. [each person entered repeatedly]
140
parts and b oundaries k. Bill ran toward the house. l. Bill ran into houses. [he may have entered each house once] m. Bill ran into some houses. [he entered each house repeatedly] n. Bill ran down the road. o. Bill ran 5 miles down the road. [OK only on the reading where 5 miles down the road is where Bill was, not on the reading where 5 miles down the road is how far he got]
Some of the phrases in (6) combine with until dawn without changing sense; some add the sense of repetition; some are ungrammatical. Moreover, the possibilities are influenced by the choice of verb (6a vs. 6b vs. 6c); by the choice of singular vs. bare plural vs. some þ plural in subject (6h, i, j), object (6d, e, f), or object of a preposition (6h, l, m); by the choice of aspect (6d vs. 6g); by the choice of preposition (6h vs. 6k vs. 6n); and by the choice of prepositional specifier (6n vs. 6o). We are thus dealing with a semantic system whose effects are felt in practically every part of the syntax. A properly general solution to the sense of repetition in (5) must therefore be an account of this entire system, and it must extend naturally to all the cases in (6).a With this in mind, let’s sketch out the overall form of the solution, beginning with the sense of repetition. As has been pointed out many times (e.g., Gruber 1967, Talmy 1978a, Hinrichs 1985, Langacker 1987a, among others), the semantic value of repetition is identical to that of the plural, that is, it encodes the multiplicity of a number of entities belonging to the same category. In the case of objects, the plural maps an expression denoting an instance of a category (say apple) into an expression denoting a multiplicity of instances of the category (apples). In the case of repetition, an expression denoting a single instance of a particular category of events (the light flashed) is mapped into an expression denoting multiple instances of the same category. In English, the resulting expression does not differ in form; but there are languages such as Hungarian and Finnish that have an iterative verb aspect used for this purpose. Note also that if the event is expressed in English with a noun, for instance a flash, then its plural denotes repeated events, for instance flashes. Thus the identification of repetition with plurality is syntactically justified as well. A consequence of this analysis is that the multiplicity of entities is a feature of conceptualization that is orthogonal to the distinction between objects and
a The generality of the system is missed in most theories of aspectuality I am acquainted with (see references in the text). It is however noticed by Declerck 1979, whose analysis was a major influence on mine.
pa rts and boundar ies
141
events.b Such a result is consistent with the evidence from (6) that the system of conceptual encoding we are looking for cuts across this conceptual distinction. Next consider the rule that permits (5) to be interpreted as repetitive despite the absence of any iterative morpheme. This rule appears to belong to a class of rules that might generally be called ‘rules of construal’ [2010: now called coercions]. Example (7), adapted from Nunberg 1979, is a standard example that invokes such a rule [this case of coercion is now called reference transfer]. (7) [One waitress says to another:] The ham sandwich in the corner wants another cup of coffee. The lexical entry for ham sandwich certainly does not specify a potential reading ‘customer with a ham sandwich’; nor is there any other lexical item in the sentence that licenses such a reading. Rather, there is a general coercion rule that may be stated roughly as: ‘A constituent identifying an individual X may be used/understood to identify an individual contextually associated with X.’ This principle licenses the insertion of nonlexical material into the conceptual structure of a sentence, roughly ‘individual contextually associated with.’ In the process, the lexical material identifying X comes to be subordinated to the role of modifier of the new material, so that the subject of (7), for example, is understood as ‘individual contextually associated with a ham sandwich’.1 Of course, if the rule used in (7) could operate freely, chaos would result. What renders its application appropriate in (7) is the fact that the literal interpretation of (7) is ill formed: a ham sandwich can’t want anything. This seems characteristic of this class of rules: the interpreter avails him- or herself of them to understand otherwise ill-formed or pragmatically inappropriate utterances. (Jackendoff 1990 suggests that rules of this sort fall into the same class as what Levin and Rapoport (1988) have called rules of ‘lexical subordination’ and what Jackendoff (1990) calls ‘superordinate adjuncts’.) The coercion rule responsible for (5) has the effect of substituting ‘multiple events of category X’ for ‘event of category X’. What motivates its application? The basic insight is that the conceptual structure of until dawn places a temporal boundary on an otherwise temporally unbounded process. So, for instance, Bill slept expresses a process that is conceptualized as unbounded: the speaker’s focus for the moment lies within the process, excluding the boundaries 1
This is how the rule looks from the point of view of syntax. From the point of view of semantics, a coercion rule licenses leaving material out of syntax, hence economizing the overt expression of thought. b This parallelism of objects, events, and locations is central to Conceptual Semantics, but not to other approaches to semantics.
142
parts and b oundaries
from view. The full sentence Bill slept until dawn then expresses the termination of this process. However, the light flashed expresses an inherently bounded occurrence: the light goes on, then goes off, and the event is over. Thus it cannot be subject to the extrinsic bounding imposed by until dawn. This is the illformedness, parallel to that of the literal interpretation of (7), that motivates applying a coercion rule. The effect of applying the rule is to map the ‘core’ event into a sequence of flashes that can go on indefinitely; this sequence can then be bounded in time by the expression until dawn. By contrast, Bill ate the hot dog is inherently bounded and cannot be repeated (barring regurgitation), so applying the coercion rule to (6d) does not result in a well-formed reading; the sentence is therefore unacceptable.2 The basic formal shape of this solution appears in (8), a first approximation to the conceptual structure of (5). (8)
LIGHT FLASHED PLURAL
UNTIL
Event Event
BOUNDED
, [Time DAWN]
UNBOUNDED
Event BOUNDED
Unpacking this, UNTIL is a function that bounds an unbounded event (its first argument) with a Time (its second argument), producing a bounded event. PLURAL is a function that maps a bounded entity (its argument) into an unbounded multiplicity of entities of the same type; in the interpretation of (5) this function and the feature UNBOUNDED are contributed by the coercion rule. The idea behind this solution appears in many sources (e.g. Vendler 1957; Verkuyl 1972, 1989; Talmy 1978a; Dowty 1979; Platzack 1979; Declerck 1979; Mourelatos 1978; Hinrichs 1985; Pustejovsky 1991a). In the course of subsequent sections, this solution will be refined and placed in the context of the larger system that is responsible for the facts in (6) and many other phenomena. In particular, my strategy is to make full use of the cross-categorial properties of this system, using the much richer grammatical resources of the noun system to elucidate the properties of the verbal system standardly called aktionsarten or event structure. (There is no space here to compare my proposals at any length with those in the extensive literature, only a small portion 2 Notice that in a language like Chinese which lacks a plural morpheme, this coercion rule will be responsible for the interpretation of plurality in noun phrases as well as sentences.
pa rts and boundar ies
143
of which is cited above and later in the text. I hope to address the differences in future work.c)
5.4 The features b(ounded) and i(nternal structure) To begin approaching a more general solution, we introduce a pair of fundamental conceptual features. Consider first the feature of boundedness. It has frequently been noted (Fiengo 1974, Gruber 1967, Talmy 1978a, Bach 1986, among many others) that the distinction between count and mass nouns strongly parallels that between temporally bounded events and temporally unbounded processes. For example, one hallmark of a count noun, say apple, is that one cannot divide its referent up and still get something named by the same count noun, i.e. another apple. By contrast, with a mass noun such as water, one can divide its referent up and still get something describable as water (as long as one does not divide it up so small as to break up the molecular structure). The same criterion applies to events vs. processes. One cannot divide up the event The light flashed and get smaller events describable as The light flashed, but one can divide up a process described as Bill slept into smaller parts also describable as Bill slept. Accordingly, we will introduce a feature +bounded, or +b, in both the object and the event system. Individual objects (usually described by count nouns) and completed events will be encoded as þb (replacing the notation BOUNDED in (8); unbounded substances (usually described by bare mass nouns) and unbounded processes will be encoded as b (replacing UNBOUNDED in (8) ). Let me be slightly more specific about what is intended by b. As suggested in the previous section, a speaker uses a b constituent to refer to an entity whose boundaries are not in view or not of concern; one can think of the boundaries as outside the current field of view. This does not entail that the entity is absolutely unbounded in space or time; it is just that we can’t see the boundaries from the present vantage point.d A second feature encodes plurality. As is well known, plurals and mass nouns pattern together in various respects, in particular admitting many of the same determiners, including some, all, a lot of, no, any, and, significantly, the zero determiner. Bare mass nouns and bare plurals, but not singulars, can occur in
c
Alas, I never did, with the exception of some brief discussion that appears here in section 6.3. The term ‘unbounded’ led to a certain amount of misinterpretation, which is why I replaced it with ‘nonbounded’ in ‘The Proper Treatment of Measuring Out’; see section 6.5. d
144
parts and b oundaries
expressions of distributive location such as (9). When they serve as direct object of a verb such as eat, the resulting sentence is a process (10a, b), by contrast with singulars, which create closed events (10c). (9)
a. There was water all over the floor. b. There were books all over the floor. c. There was a book all over the floor.3
(10) a. Bill ate custard until dawn. b. Bill ate hot dogs until dawn. c. Bill ate a hot dog until dawn. We will therefore group bare mass nouns and bare plurals together as unbounded (b). Talmy suggests the term medium to encompass them both. The difference between the two kinds of media is that plurals entail a medium comprising a multiplicity of distinguishable individuals, whereas mass nouns carry no such entailment. We will encode this difference featurally; the difference in entailment can then be a consequence of inference rules that refer to the feature in question. I will call the feature +internal structure, or +i. Aggregates—the entities normally expressed by plural nouns—will be þi. Substances—the entities normally expressed by mass nouns—will be i. (Note: the value i does not mean lack of internal structure, but rather lack of necessary entailment about internal structure.) The +i distinction can be applied in the þb domain as well as the b: it provides a way of encoding the difference between individuals and groups. (Here I begin to move away from standard proposals.) A group entity is bounded, but there is a necessary entailment that it is composed of members. An individual may have a decomposition into parts, but that is not a necessary part of its individuality. Thus the feature system, applied to objects and substances, comes out as (11). (11) þb, i: þb, þi: b, i: b, þi:
individuals (a pig) groups (a committee) substances (water) aggregates (buses, cattle)
Individuals correspond to the conceptual category of Thing (or Object) employed in chapter 1, 3, and 4. We therefore need a larger supercategory that contains all of the entities in (11). Let us call it Material Entity (Mat for A reader has observed that There was a copy of the NY Times all over the floor is grammatical. It appears that in this sentence the newspaper is being conceptualized as an unbounded aggregate of sheets of paper. 3
pa rts and boundar ies
145
short). The term Thing, previously regarded as primitive, now becomes composite: it is an abbreviation for [Mat, þb, i]. Note however its privileged status: of the four subcategories of Mat, only Thing has an inherent shape. Therefore it is the only subcategory that has physical boundaries. (Groups are bounded in quantity but do not have an inherent shape.) The features b and i can be applied in the event/process domain as well: [þb, i]: [b, i]: [b, þi]: [þb, þi]:
(11.1)
closed events [John ran to the store] unbounded homogeneous processes [John slept] unbounded iterative processes [The light flashed continually] bounded iterative events [The light flashed until dawn]
Thus the feature system cuts across major conceptual categories, expressing the generality of the phenomena of boundedness and plurality.
5.5 Functions that map between values of b and i 5.5.1 PL (plural) How is the notion of plurality to be represented? There are two possibilities. Suppose (12a) is the conceptual structure of a dog, where the features [Mat, þb, i] set the entity within the major category of individual objects, and DOG is a stand-in for the conceptual information that distinguishes dogs from other categories of individual objects. Then there are two possible ways to encode the term dogs, shown in (12b) and (12c). (12) a. b.
+b, –i = a dog Mat DOG
–b, +i = dogs Mat DOG –b, +i
c. Mat
PL
+b, –i Mat DOG
= dogs
In (12b), the plural has been expressed by changing the b and i features of a dog from those for an individual to those for an aggregate. The plural morpheme is thus conceived of as expressing a feature-changing process. In (12c), by contrast, the lexical entry for dog has been left unchanged; it appears as the argument of a conceptual function PL that maps its argument into an aggregate. The plural morpheme is then taken to express this function.
146
parts and b oundaries
I will adopt the latter solution, because it permits correspondence rules along lines known from the principles of verb argument structure illustrated in section 1.5. In particular, it gives the correspondence rules a property of ‘morphological transparency’: for the most part, addition of syntactic information (including morphology) does not change features of the base element, but rather adds operators around the base. In the present case, the LCS of dog is found directly in (12c), embedded in the operator PL; by contrast, in (12b) the LCS of dog has disappeared. One reason for adopting representation (12c) is what happens when we pluralize a group-noun such as herd or committee. Under the feature-changing treatment, the plural comes out as (13a), which is no longer distinct from the plural of an individual. Under the morphologically transparent treatment, it comes out as (13b), in which one can still discern that the plurality is of groups rather than individuals.
(13)
a.
–b, +i Mat COMMITTEE
= committees
–b, +i b.
PL Mat
Mat
+b, +i COMMITTEE
= committees
The plural morpheme thus has the conceptual structure (14a); the LCS of the noun to which it applies fits into the A-marked argument slot. A lexical plural such as people or cattle has an LCS like (14b).e (14) a.
N + plur =
–b, +i PL ([+b]A) –b, +i
b. people =
PL Mat
+b, –i Mat PERSON
Note that the plural morpheme cares only that the noun to which it applies designates a bounded entity. It does not care whether that entity is a Material
e This analysis of the plural is an opening wedge toward seeing regular affixation as expressed by a lexical item of standard form, undergoing free combination just like other standard lexical items. It therefore became central to the development of the Parallel Architecture’s unification of the lexicon with rules of grammar (section 1.4).
pa rts and boundar ies
147
or an Event (such as earthquakes), nor whether it is an individual or a group. On the other hand, the entity must be bounded, so that mass nouns, which are b, cannot be pluralized. (We return to apparent counterexamples like three coffees in section 5.5.3.) In the verbal system, PL is the function that iterates events. Thus a constituent of The light flashed until dawn is (15). This expression is the unbounded process that will eventually be bounded by until dawn; it replaces the notation for the first argument of UNTIL given in (8). (15)
–b, +i PL Event / Process
+b, –i Event LIGHT FLASHED
In this case, PL is not introduced by a morpheme in the sentence. Rather, as argued in section 5.3, it is introduced by a coercion rule. There are verbs that appear to lexically include PL, parallel to lexically plural nouns such as cattle. For example, pound and hammer normally describe not a single blow (as hit does), but a sequence of blows iterated into a process. PL is one of a class of functions that map an entity with one value of b and i into another entity with different values. Having considered and rejected numerous hypotheses about the constitution of this class, I am not completely confident that the functions about to be proposed are properly characterized. (A different but related set is proposed by Winston, Chaffin, and Herrmann 1987.) However, the discussion to follow shows the range of phenomena for which any competing analysis of this class of functions must be responsible.
5.5.2 ELT (element of) A sort of inverse of PL is evoked in phrases like a grain of rice, a stick of spaghetti. In these phrases, the second noun is grammatically a mass noun, but it happens that it denotes an aggregate rather than a substance. The first noun picks out an individual of the aggregate. Hence, to express the meaning of the phrase, we need a function that maps an aggregate into a single element of the aggregate. I will call the function ELT (element of). A grain of rice then comes out as (16).
(16)
+b, –i
a grain of rice =
ELT Mat
–b, +i Mat RICE
148
parts and b oundaries
A possible extension of ELT to another feature combination might be a drop of water, in which a drop is conceptualized as the natural unit into which the substance water divides itself. +b, –i
(17) a drop of water =
ELT Mat
–b, –i MatWATER
This extension is not, however, a natural inverse of the plural, since it conceptualizes a multiplicity of individuals (drops) combining into a substance rather than an aggregate. I will leave open whether this extension is correct. PL and ELT thus form a pair that can be thought of as approximate inverses. I will call PL an including function: the function maps its argument into an entity that includes the argument as a subentity. By contrast, ELT is an extracting function: the function maps its argument into a subentity of the larger entity denoted by the argument. It is a characteristic of including functions that they transmit existential claims to their arguments. For instance, if there are dogs around, there is a dog around. By contrast, extracting functions do not transmit existential claims to their arguments. For instance, having a grain of rice around does not entail that there is a larger aggregate body of rice around—this single grain may be all we have. The other functions to be introduced form pairs in the same way as PL and ELT: one member of the pair will be an including function and one an extracting function.
5.5.3 COMP (composed of) Consider an expression like a house of wood. The phrase of wood describes the substance of which the house is composed. To encode this relation, let us introduce a function COMP. Preserving the syntactic relations of subordination, this function will take a substance as its argument and map it into an individual.
(18) a house of wood =
+b, –i HOUSE COMP Mat
–b, –i Mat
(‘a house composed of wood’)
WOOD
Substituting an aggregate for the substance in (18), we can create expressions like a house of bricks.
pa rts and boundar ies
(19)
149
+b, –i HOUSE a house of bricks =
–b, +i COMP Mat
+b, –i
PL Mat
Mat
BRICK
In (18) and (19), the noun house contributes only the content [þb, i, HOUSE]; it is presumably not part of the LCS of house that it has to be composed of something. In other words, the COMP function is contributed by the modifying construction of wood or of bricks. However, there are other nouns whose LCS contains COMP as an essential part. For instance, a pile or a stack is an inherently bounded collection of smaller elements, combined to create a particular form. Thus the LCS of these items is something like (20).
(20)
+b, –i PILE COMP ([–b])
a. pile = Mat
b. stack =
+b, –i STACK COMP ([–b, +i]) Mat
The difference between the two expresses the fact that one can have a pile of bricks (aggregate) or a pile of sand (substance), but only a stack of bricks, not a stack of sand.4 In the examples above, the COMP function maps its argument into an individual. But COMP can also provide an analysis for group-nouns such as herd, flock, and group.
(21)
+b, +i –b, +i a flock of birds =
COMP Mat
4
PL Mat
+b, –i Mat BIRD
Notice that a stack of wood implies that the wood is in discrete largish pieces, i.e. an aggregate rather than just a substance. However, one cannot have a stack of wood chips, even though wood chips is plural. That is, stack imposes further selectional restrictions that are not addressed here, probably having to do with orderly geometric arrangement of the elements of the stack.
150
parts and b oundaries
Note the difference between stack and flock: a stack has an inherent shape, which makes it [þb, i], while a flock has no shape of its own, which makes it [þb, þi].5 So far COMP has been introduced by a lexical item or by the of-construction. It can also be introduced by a coercion rule, in which case it serves as what has sometimes been called the ‘Universal Packager’, attributed in the literature to David Lewis:6 (22) I’ll have a coffee/three coffees. Here coffee is construed as ‘bounded individual composed of coffee’. The syntax of the construction is that of count nouns: it uses the indefinite article and the plural, which are conceptually incompatible with the LCS of coffee, a substance. Therefore a coercion rule, inserting the operator COMP, must apply to make the representation well formed.
(23) a.
+b, –i a coffee =
COMP Mat
–b, –i Mat COFFEE
‘a portion of coffee’
–b, +i
b.
+b, –i
coffees =
PL Mat
COMP Mat
–b, –i Mat COFFEE
‘portions of coffee’
(Note: this reading of coffees is distinct from the reading that means ‘varieties of coffee’, as in The store sells seventeen coffees, each from a different country; the latter is due to a separate coercion rule.) A more general situation in which a coercion rule makes use of COMP appears in examples like (24).
5 This extension does not appear to have full generality: an individual can be composed of either a substance or an aggregate, but a group may be composed only of an aggregate—the notion of a group composed of a substance seems anomalous or conceptually ill formed. I leave open how this asymmetry in the COMP function is to be resolved. 6 David Lewis has informed me (personal communication) that he has not used this term in print; he in turn attributes the notion to lectures or writings by Victor Yngve in the 1960s which he is now unable to trace. The same goes for the notion of the ‘Universal Grinder’ in section 5.5.4.
pa rts and boundar ies (24)
151
a. Will you mop up that water, please? b. They loaded the sand into the truck. c. The boys were impressed.
In each of these, the italicized NP expresses a bounded entity composed of a substance or aggregate. One might think that the definite article is the source of COMP. But in fact in other contexts the very same phrases can be unbounded. That water in (25a), for instance, denotes a contextually identifiable medium, not a fixed amount. (25)
a. That water kept spurting out of the broken hose. b. The sand stretched out as far as we could see. c. The boys arrived for hours on end.
Apparently, then, definiteness contributes only the content ‘contextually identifiable’; the determination of boundedness depends on other constraints. The unbounded reading (26a) can be derived directly from the LCSs of that and water; the bounded reading (26b) is the result of applying a coercion rule that inserts COMP. (DEF is the conceptual structure associated with definiteness.)f
(26) a. that water (in (25a)) =
b.
–b, –i WATER DEF
+b, –i that water (in (24a)) = DEF COMP
–b, –i WATER
5.5.4 GR (the grinder) COMP, like PL, is an including function: it maps its argument into a larger entity that includes the argument. It therefore has the existential entailment characteristic of an including function: if there is a house of wood around, there is wood around. The inverse of COMP therefore ought to be an extracting function whose argument is an individual or group, and which maps its
f In other words, the þ mass noun denotes an individual only by virtue of a coercion. From the point of view of Fregean composition, this is a fairly startling result.
152
parts and b oundaries
argument into a substance or aggregate of which the individual or group is composed. Such a function is found in the so-called ‘Universal Grinder’, illustrated in the grisly (27). (27) There was dog all over the street. Here the bare singular and the distributive location force the term dog to be interpreted as a substance. As usual, the relevant coercion rule does not simply change the lexical features of dog to make it into a substance. Rather, it preserves well-formedness by introducing a function GR, whose argument is the LCS of dog.
(28)
–b, –i dog (substance) =
GR Mat
Mat
+b, –i DOG
Given this operator, we can also use it in the lexicon to express the relation between animals and their meat (29a), animal body parts and their meat (29b), and similar paired words like rock/a rock and stone/a stone.
(29) a. pork =
–b, –i MEAT GR
b. liver =
+b, –i PIG
–b, –i MEAT GR
+b, –i LIVER
GR applied to an individual yields a substance. For symmetry, it is useful to stipulate that GR applied to a group yields the aggregate of which the group is composed. This makes it not quite a true inverse of COMP, but it is close. As required, GR is an extracting function: its output is a subentity of its argument. Like our other extracting function ELT, it does not transmit existential claims to its argument: if there is dog all over the place, it does not follow that there is a dog around. In the verbal system, GR appears to be (one of) the reading(s) of the progressive aspect in English. For example, Bill is running to the store can be construed
pa rts and boundar ies
153
as ‘the process out of which the event Bill runs to the store is composed’. This analysis allows us to incorporate Bach’s (1986) solution to the ‘imperfective paradox’—the fact that even if it is true that Bill is writing a novel, there need not yet (or ever) be a novel such that Bill is writing it. Bach, drawing on Link’s (1983) treatment of the semantics of the mass/count distinction, points out that the existence of a part of an object does not entail the existence of the whole object. For instance, one may find (or make) a part of a violin without there being (now or ever) a violin of which this is a part. Similarly, Bach argues, the progressive is extracting a part of an event, and hence carries no entailment that the event is ever carried to completion. Since the existence of the (complete) novel depends on the completion of the event, the novel too carries no existential claim. In the present analysis the same conclusion follows from the claim that the progressive involves applying the extracting function GR to the event, which does not transmit existential claims (or in the case of events, truth claims) to its argument. Before going on to the next function, notice that all the functions discussed so far can be introduced by coercion rules, and that at least two, ELT and COMP, show up in the N of NP construction. This is one of the difficulties of separating these functions clearly—their great degree of syntactic overlap, when they are expressed at all.g
5.5.5 PART (part of) Another N of NP construction occurs in (30). (30)
a. a leg of the table b. the roof of the porch c. a part of the group
This partitive construction takes as its argument a bounded entity and addresses its internal articulation, picking out an identifiable bounded part. It is thus an extracting function like ELT and GR; as noted by Bach, it shares the characteristic entailment of extracting functions. Unlike GR, it extracts a bounded part, not an unarticulated substance. Unlike ELT, it presumes that the entity g
At the time I wrote this, I found it a fascinating result that the syntax–semantics interface is many to many, and that the same meaning can be expressed lexically, constructionally, or implicitly. But I did not see it as more than a minor wrinkle in a more or less classical theory of grammar. It was only some years later that I came to realize how these results challenge Fregean composition and interface uniformity, and how central the latter are to mainstream approaches in both generative grammar and formal semantics. A great deal of the development of the Parallel Architecture consisted of making these largely tacit assumptions explicit, and in finding where these assumptions led classical theory to miss the sorts of generalizations worked out here.
154
parts and b oundaries
extracted from is nonhomogeneous: a house can have different kinds of parts, but rice has only one kind of internal element, a grain. Thus it appears that the partitive is a distinct function from the other two. I will encode (30a), for example, as (31).
(31)
+b, –i LEG PART
+b, –i TABLE
Note that words like leg and roof are lexically partitive—a leg has to be a leg of something. (30c) shows the same operator applied to a group noun, yielding a smaller group.
(32)
+b, +i part of the group = PART
+b, +i GROUP
A word like subcommittee lexically incorporates the whole complex in (32). PART also appears to be able to map its argument into a substance. An example is (the) blood of a pig, which seems altogether parallel to a/the heart of a pig in its structure. A possibly more controversial extension would be to cases with an unbounded argument, as in an ingredient of stew, whose proper analysis I leave open. Another frequent syntactic realization of the PART function is as a nominal compound. So for example, parallel to the examples above we have table leg, porch roof, pig blood, pig heart, and stew ingredients (though not group part).
5.5.6 CONT (containing) Each of the other functions has an approximate inverse. This suggests that PART should too. What would be its properties? It would have to be an including function that mapped its argument into an entity containing the argument as a part. One possible instance of such a function is in compounds like drop-leaf table (‘a table whose identifying part is a drop-leaf’) and beef stew (‘stew whose identifying ingredient is beef’). By extension, it would also be invoked by a coercion rule in synecdoche and exocentric compounds, where a distinguishing part is used to identify the whole (Hey, Big-Nose!). Another possible case is NPs containing a with-modifier, such as table with a drop-leaf and
pa rts and boundar ies
155
house with an orange roof. It is clear that this relation is distinct from the other including functions, PL and COMP; let’s call it CONT (‘containing’). Then beef stew, for instance, would have the structure (33). [See Chapter 2 and especially Chapter 13 for more detailed discussion of compounds of all sorts.]
(33) beef stew =
–b, –i STEW CONT
–b, –i BEEF
Thus we have found six functions that map one combination of the features b and i into another, expressing different relations of parts to wholes. (34) summarizes.h (34)
Including functions: PL COMP Extracting functions: ELT GR
CONT PART
5.6 Dimensionality and directionality We next have to look briefly at the dimensionality and directionality of entities and how they are encoded in conceptual structure.i The intuition behind the analysis of dimensionality comes from Marr’s (1982) notion of a generalized cone, which plays an important role in his theory of the conceptualization of shape (see Chapters 3 and 4). A generalized cone is a 3-dimensional shape that is created by moving a cross section along an axis. The cross section may vary smoothly in size, getting fatter or thinner, but its shape remains the same. Thus a football is a generalized cone and so is a pyramid or, roughly, a leg or an arm, or a snake, or a tree trunk, or a stalagmite. (Marr 1982, 223)
Thus if we take a slice of a generalized cone perpendicular to the axis, we always get the same shape; but the size is a continuous function of the position on the axis.
h The fact that these functions come in pairs suggests that ideally each pair reduces to a single relation that can be ‘profiled’ in either direction; see sections 13.3–4. On the other hand, to the extent that they are not exact inverses, they must be retained as distinct. i The next two paragraphs appeared originally in ‘The Proper Treatment of Measuring Out’ and have been transposed here for clearer exposition. Talmy 1983 also speaks of the idealization of objects to their principal dimensions.
156
parts and b oundaries
a. Tube:
b. H-beam Fig. 5.1
For example, consider a cylindrical tube or an H-beam. A representative slice of a tube is an annulus; of an H-beam, an H shape. This cross-section is projected onto a linear (one-dimensional) axis to form the shape of the object as a whole. The axis itself may be straight or curved. The drawings in Figure 5.1 suggest this decomposition (for the case where the cross-section is constant over the length of the axis). Viewing this analysis more broadly, the basic observation is that a point is conceptualized as zero-dimensional, a line or curve as one-dimensional, a surface as two-dimensional, and a volume as three-dimensional, where dimensionality is essentially the number of orthogonal degrees of freedom within the object. However, dimensionality has more structure than this: following and extending the Marr concept of generalized cones, we can decompose an object’s dimensionality into a hierarchical arrangement of axes and crosssections. Consider for example a road, a river, or a ribbon. These can be schematized as a line (the primary dimension, Marr’s linear axis) elaborated by a linear cross-section (the secondary dimension), yielding a surface. The primary dimension of these objects may be bounded or unbounded; the secondary dimension is bounded. In order to encode dimensionality in conceptual structure, we will introduce a four-valued feature DIM nd, where n varies from 0 to 3. (35) illustrates the use of this feature; the secondary dimension appears in the inner brackets.
(35)
⫾b, –i road, river, ribbon = DIM 1d +b, –i DIM 1d
Contrast these items to a layer or a slab, which are basically thickened surfaces. Here the primary dimension is a bounded or unbounded surface, and the secondary dimension is an orthogonal dimension that is bounded and linear, giving a volume. (36a) shows this representation. A different case arises with a tube or a beam (Figure 5.1), whose primary dimension is linear, and whose cross-section is a two-dimensional shape (36b).
pa rts and boundar ies
(36) a.
157
⫾b, –i layer, slab = DIM 2d +b, –i DIM 1d
b. tube, beam =
⫾b, –i DIM 1d +b, –i DIM 2d
On the other hand, a sphere has no salient decomposition into axes, so its dimensionality is just [DIM 3d]. The dimensionality feature can easily be extended to time and to states and events. Points in time, states at a point in time, and point-events are [DIM 0d], while periods of time and states and events with duration are [DIM 1d]. This of course does not leave much room for distinctions of primary vs. secondary dimensionality, but we will see shortly that such possibilities arise nevertheless. The dimensionality feature is subject to a principle of reconstrual that I will call the zero rule: a bounded object can always be idealized as a point. Under this idealization, the object’s intrinsic dimensionality becomes secondary and the primary dimensionality is 0d. (This is the principle that allows cities to be represented by points on maps.)
(37) (Zero rule – idealizing object as point) X +b DIM nd
X +b DIM 0d [DIM nd]
The use of this principle will become evident in the subsequent sections. A further wrinkle in the dimensionality feature is that any one-dimensional axis can have a direction or orientation. So, for example, a line has no intrinsic direction, but a vector and an arrow do. We can encode this by adding a further distinction to the dimensionality feature, marking vectors and arrows as [DIM 1d DIR] and ordinary lines as just [DIM 1d].7 A surface or volume can acquire directionality only by being decomposed into linear axes. For instance, the human body has a primary directed up–down 7
I am treating DIR as a ‘privative’ feature, that is, one that is either present or absent. Alternatively it could be treated as a binary feature +DIR. However, the only descriptions in the present chapter where the feature –DIR is necessary are Place and State (see (38) ), which may not prove to need independent definitions—they may be just the residue of Spaces and Situations when Paths and Events are removed.
158
parts and b oundaries
dimension, a secondary directed front-to-back dimension, and a tertiary sideto-side dimension that is symmetric rather than directed. However, a sphere and a layer have no inherent directionality. I would like to use the directionality feature to resolve a problem in the set of major conceptual categories in Jackendoff (1983, 1990) (and in chapter 1). This class includes Thing (now expanded to Material), State, Event (now including processes), Place, Path, Time, and others. When these categories were proposed, there was clearly a close relation between States and Events and between Places and Paths, but this relationship found no formal expression. So let us consider the relation between Places and Paths. Places can be regions of any dimensionality: at this point is zero-dimensional, along the line is onedimensional, in the circle is two-dimensional, and in the cup is threedimensional. Thus Places share the dimensionality features of objects. But Paths can be only one-dimensional and must moreover be directed: there is an intrinsic direction in which they are viewed—in the standard case, from Source to Goal. This suggests that Places and Paths can be combined into a supercategory that may be called Space. Paths are the subset of spaces whose dimensionality feature is [DIM 1d DIR] and Places are the rest. In other words, the relation of Places and Paths can be formally expressed by a feature distinction. I would like to extend this, a little speculatively, to the relation between Events and States. States like X is red or X is tall are conceptualized as ‘just sitting there’—they have no inherent temporal structure. (The ‘state’ of being in continuous motion, however, is now encoded as a process, or unbounded Event.) Events, by contrast, do have an inherent temporal structure which proceeds in a definite direction. I would like to suggest therefore that the two categories be combined into a supercategory called Situation, with States as the undirected case and Events as the directed case. (Bach (1986) uses the term eventuality in the same sense.) (38) summarizes this further analysis of the S&C primitives.
(38)
[PLACE] =
SPACE –DIR
SPACE [PATH] = DIM 1d DIR [STATE] =
SITUATION –DIR
[EVENT]=
SITUATION DIR
pa rts and boundar ies
159
There is a slight asymmetry in this reanalysis, forced by the existence of point-events such as The light turned on or The clock ticked once. According to our treatment of dimensionality, these should be zero-dimensional; but as Events, they are directional. Thus we have to admit the possibility of zerodimensional directed Situations, whereas the original intuitions motivating directionality pertained only to one-dimensional entities. One can think of a number of solutions for this asymmetry, but nothing much hangs on it for now, so I will leave it unresolved. The main point, however, is that the dimensionality and directionality features, developed to account for conceptual properties of objects, turn out to permit an insightful unification of completely independent conceptual categories.
5.7 Boundaries What sort of entity is a boundary? It follows from the definition of the feature system that only a [þb, i] category—an individual—can have a boundary. If an entity is conceptualized as [b], this means it is conceptualized without a boundary; in order to discuss its boundary, we have to first reconceptualize it by applying the COMP function. A [þb, þi] entity, a group, is bounded in quantity, but it has no inherent shape—it is just a collection of individuals. Hence it has no discernible entity serving as a boundary. A basic condition on boundaries is that a boundary has one dimension fewer than what it bounds: a line can be bounded by a point, a region by a line, and a volume by a surface. However, this basic condition is an idealization of the actual situation. Consider a stripe that bounds a circle: it is locally twodimensional, not one-dimensional. What makes the stripe a boundary for the circle is its schematization as a line (its primary dimension) elaborated by a cross-section (its secondary dimension). At the schematic level of primary dimensionality it is one-dimensional, as the basic condition stipulates. From this we can see that the actual condition on dimensionality of boundaries is that the schematization of a boundary has one dimension fewer than the schematization of what it bounds. This enables us to make an important generalization in the conceptual structure of words like end and edge. Consider what kinds of things can have ends, and their dimensionality. A line (1d) has a 0d end; a ribbon (2d) has a 1d end; a beam (3d) has a 2d end. This is not very enlightening. However, the proper analysis emerges if we observe that each of these objects has a 1d primary dimensionality, that is, they are all schematized as lines. By descending
160
parts and b oundaries
to the level of the 1d schematization, we can treat the end in each case as a point bounding the line. How then does the end acquire its actual dimensionality? Consider again the beam, whose dimensionality was given in (36b). The 2d secondary dimensionality here represents the cross-section of the beam, say an H shape. The end of the beam inherits its shape from this cross-section. More generally, an end can be schematized as having a 0d primary dimension, elaborated by the same secondary dimension as the object it bounds.
(39) a. line
= [DIM 1d]
end of line
= [DIM 0d]
b. ribbon =
DIM 1d [DIM 1d]
end of ribbon =
DIM 0d [DIM 1d]
c. beam =
DIM 1d [DIM 2d]
end of beam =
DIM 0d [DIM 2d]
Using the zero rule (37) from right to left, the end of the ribbon and the end of the beam can be reanalyzed as entities in their own right, with one and two dimensions respectively. An end is therefore fundamentally a point that bounds a line. An edge, by contrast, is fundamentally a line that bounds a surface. For instance, the edge of a ribbon (2d) is one-dimensional. A table-top can be conceptualized as a surface (2d) elaborated by a thickness (1d). The edge of a table-top is the boundary of the surface (1d) elaborated by the same thickness, hence a ribbon-like surface.
(40) a. ribbon
= [DIM 2d]
b. table-top =
DIM 2d [DIM 1d]
edge of ribbon
= [DIM 1d]
edge of table-top =
DIM 1d [DIM 1d]
Notice that a ribbon is conceptualized under different schematizations, depending on whether one is identifying its end or its edge. One further very important wrinkle. What do you do when you cut off the end of a ribbon? It would be absurd to think of just cutting off the geometric boundary, as the analysis so far would suggest. Rather, in this context, the end of the ribbon includes the geometric boundary plus some pragmatically determined but relatively small part of the body of the ribbon. Similarly for putting the cup on the end of the table, in which the end includes some part of the top surface. These examples show that the primary dimension of an end, the one that bounds the linear axis of the object, need not be just 0d, but can be expanded a small amount along the axis. I will encode this expansion by the notation 0þed,
pa rts and boundar ies
161
as in (41). This notation may be thought of as designating a dimensionality that is something more than a point but something less than a line.
(41) object = DIM 1d [DIM nd]
end of object =
DIM 0(+e)d [DIM nd]
The expansion of the boundary is optional in the case of end, and one might want to attribute this possibility to a general coercion rule. However, there are other boundary words for which the expansion is obligatory. Consider a crust. This is a surface bounding a volume, plus an expansion of the surface some small pragmatic distance into the volume; it is hard to tell where the crust of a loaf of bread breaks off and the body of the bread begins. We can express the dimensionality of a crust therefore as [DIM 2þed]. Similarly, the border of a rug is liable to extend further into the rug from the geometric boundary than does the edge. Thus, to distinguish surface from crust and border from edge, something like the epsilon notation is necessary in lexical conceptual structure. In turn, the optional expansion of end may be either lexical or supplied by a coercion rule; in either case, though, the formal effect is encoded by the epsilon notation. This treatment of the dimensionality of ends gives us an immediate solution for a well-known puzzle in event structure. If the end of a talk is located at its temporal boundary, it must take place at a point in time. However, it is perfectly acceptable to say Fred is ending/finishing his talk, where the use of progressive implies a process taking place over a period of time. What is going on? The solution lies in the optional expansion of the end some small pragmatically determined distance back into the body of the talk, so that the end has dimensionality [DIM 0þed]. The expanded end takes up a period of time, and the activity within this period can therefore be described as ending the talk. In short, the analysis of end developed to account for obvious geometric intuitions generalizes to the less transparent temporal case, providing a natural explanation.j We complete this section by offering a formalism for the functions that relate boundaries to what they bound. As in the case of the functions introduced in section 5.5, there is a pair of boundary functions that are approximate inverses of each other. (42) gives a first pass.k
j
Despite being hidden in the inner workings of the chapter, this is a major result. It is possible only through the formal parallelism of objects and events, through the innovation of dimensional schematization, and the gimmick of epsilon dimensionality. k Again, ideally the pair of functions BD and BDBY reduces to a single relation that can be ‘profiled’ to pick out either argument.
162
parts and b oundaries
(42) a.
X +b, –i DIM n – 1d BD
b.
Y +b, –i DIM nd BDBY
= ‘an X that bounds Y’
Y +b, –i DIM nd
X +b, –i DIM n – 1d
= ‘a Y that is bounded by X’
For our purposes here, one refinement is necessary in these functions.8 Just in case the entity being bounded (Y in (42) ) is directed, the two boundaries must be distinguished as top and bottom, front and back, or beginning and end. Accordingly, we introduce the notation in (43) as a subcase of (42).
(43) a.
X +b, –i DIM 0(+e)d BD±
b.
X +b, –i DIM 1d DIR
Y +b, –i DIM 1d DIR BDBY±
X DIM 0(+e)d +b, –i
= ‘an X that terminates (+) or originates (–) Y’
= ‘a Y that has X as a terminus (+) or origin (–)’
8 The definitions in (42) and (43) stipulate that the entity being bounded and its boundary both be [þb, i]. This pertains, of course, only to the dimension whose boundary is being determined. A river, for instance, has boundaries for its secondary dimension (its edges), while its primary dimension may be regarded as unbounded.
pa rts and boundar ies
163
Thus, assuming an axis directed from back to front, the front of an object will be its BDþ and the back its BD; the beginning of an event will be its BD and the end its BDþ. Another refinement is necessary to specify that a boundary inherits its secondary dimensionality from the object it bounds, as shown in the cases of end and edge in the previous section. However, this plays no formal role in what is to follow, so I will leave it for another occasion. [Chapter 6 takes this problem up, for certain cases.]
5.8 Using the formalism I have introduced a fair amount of new machinery here, but each piece was motivated by its ability to capture aspects of the conceptualization of objects and substances as well as their linguistic reflexes. We now apply the machinery to a variety of analyses in Path and Event structure.
5.8.1 Paths The first case is the Path-function TO, whose argument position defines the thematic role Goal, and which is treated as a conceptual primitive in section 1.3 and in Jackendoff (1983, 1990) (and in most other sources as well). This function is most directly expressed by to in English but is also incorporated in a wide range of other prepositions and verbs, as seen in section 1.5. We can now decompose TO. It defines a Path that terminates at the Thing or Place that serves as its argument. This is easily encoded in terms of the features and functions introduced here:
(44)
+b, –i DIM 1d DIR BDBY+([Thing/Space X])
TO X = Space
That is, TO specifies a one-dimensional bounded directed space (i.e. a bounded Path), bounded on its positive end by the Goal. FROM, the function whose argument defines Source, differs from TO only in that BDBYþ is replaced by BDBY. That’s all there is to it. VIA is a Path-function that defines Routes, again primitive in Jackendoff 1983 (and section 1.3). It forms part of the LCS of prepositions like through (‘via the interior of’) and past (‘via near’). In the present notation it can be analyzed as (45).
164
parts and b oundaries
(45) VIA[Place X] = Space
–b, –i DIM 1d DIR CONT ([Space X])
This is a directed one-dimensional Space (a Path) that is unbounded—if you tell me you went past my house I have no idea where you started or ended. The only thing I know about your path is that it includes the region near my house as a significant part. That is precisely what CONT was designed to encode in expressions like beef stew (section 5.6). The other two major Path-functions in Jackendoff 1983 are TOWARD and AWAY-FROM, which are like TO and FROM except that they do not include the Goal and Source respectively. In the mass–count test they behave like substances: any part of a Path toward the house is also describable as toward the house, whereas this is not true of to the house. We therefore want to describe TOWARD as unbounded. (46) gives two possible analyses.
⫺b, –i DIM 1d DIR
(46) a. TOWARD X =
GR Space
b. TOWARD X = Space
+b, –i DIM 1d DIR BDBY+ ([X])
–b, –i DIM 1d DIR BDBY+ ([X])
(46a) treats TOWARD X as a ‘ground-up’ version of TO X, that is, roughly as the ‘Path-substance’ of which TO X is made. (46b) treats it by analogy with the notion of an ‘open interval’ in mathematics—a space that is bounded by but does not include the Goal. In this treatment, we have to admit the possibility of [b] entities that have boundaries. In either case, AWAY-FROM is identical except that BDBY replaces BDBYþ. At present I do not know how to decide between these alternatives. The inverse of TO is a function called AT-END-OF in Jackendoff (1990). This appears as part of the reading of a number of prepositions, for instance across in Bill is across the road from here. In this example, across the road expresses a Place that is at the terminus of a Path that begins here and extends across the road. (47) analyzes this function.
pa rts and boundar ies
(47) [Place AT-END-OF ([Path X])] = Space
165
+b, –i DIM 0d BD+ ([X])
5.8.2 Aspectual functions INCH (inchoative) is a function that maps a State into an Event culminating in that State. It is an optional element in the conceptual structure of such verbs as stand, sit, point, cover, extend, and surround. For example, Bill is standing can mean either that Bill is in an erect posture (stative) or that he is in the process of adopting that posture (inchoative); the army surrounded the city can mean either that the army was in a spatial configuration that extended around the city (stative), or that the army moved so as to be in such a configuration (inchoative). In Jackendoff (1990) (and many other sources) INCH is treated as primitive, but again the present analysis permits a decomposition:
(48) INCH ([State X]) (‘State X comes about’) = Sit
+b, –i DIM 1d DIR BDBY+ ([Sit X])
Notice that this is identical to the analysis of TO, except that the major category feature Situation replaces Space! That is, the present analysis formally captures a deep parallelism between the end of a path and the state at the end of an event. [This parallelism is taken up further in Chapter 6.] The last section spoke of beginning and finishing as Events that serve as boundaries of other Events. Here is a formal treatment of finish; begin replaces BDþ with BD.
(49) Situation X finishes/ends =
+b, –i DIM 0(+e)d DIR BD+ [Sit X]
Sit
Section 5.5.4 analyzed the progressive aspect as ‘grinding up’ an action into a process, showing how this solves the ‘imperfective paradox’. This analysis can be extended to express the difference between stop doing X and finish doing X. Both are termini of an action; but you can stop doing something without finishing it. Here is a possible analysis for stop running to the store.
166
(50)
parts and b oundaries
+b, –i DIM 0(+e)d DIR +b BD+
–b
COMP
GR
+b RUN TO STORE
Sit
This unpacks as follows: the bounded Event run to the store is ground up by GR into a process; some of this process is gathered up into a unit by COMP; the end of this unit is picked out by BDþ. It is this boundary Event that is expressed by stop running to the store. In turn, since run to the store has been ground up, there is no inference of completion. (50) has a lot of functions in it. Which ones are lexical? My guess is that BDþ and COMP are due to the verb stop, which can also apply to States and Processes such as Paul stopped being sick and Paul stopped sleeping. GR in (50) is likely inserted by a coercion rule that converts a closed Event into an unbounded entity so that it can be bounded again internally. An alternative coercion rule available in this context inserts our old friend PL, which creates a different kind of unbounded entity. This is the most likely reading of The light stopped flashing, and a secondary reading of Bill stopped running to the store (all the time), namely the termination of a sequence of iterations.
(51)
+b, –i DIM 0(+e)d DIR +b, +i BD+
COMP
–b, +i PL
+b RUN TO STORE
Sit
Notice that in (51) there is no extracting function in the chain, so this time we can infer that Bill did run to the store.
5.8.3 The ‘Vendler classes’ Much of the discussion of event structure has taken as a starting point the socalled Vendler classes of States, Activities, Accomplishments, and Achievements.
pa rts and boundar ies
167
It is by now well known that these classifications pertain to entire sentences, rather than to verbs, as Vendler (1957) thought. There is not space here to discuss the extensive literature. However, I have become convinced, especially by the work of Declerck (1979), that the distinctions have to do with temporal structure, and have nothing to do with causation or volition, as implied by Dowty’s (1979) influential analysis. The present formalism provides a straightforward encoding of the Vendler classes and permits us to set up a couple of other cases that Vendler and many later investigators have missed.l States are simply undirected Situations, of 0 or 1 dimension. They may be bounded or unbounded; but I don’t think they can be intermittent, hence they are [i]. The formal specification is (52).
(52)
State = Sit
–i [–DIR]
Activities correspond to what have been called here processes: unbounded directed Situations.
(53)
–b
Activity = Sit
[DIR]
These can be produced either intrinsically (swim), by grinding bounded Events (be running to the store), or by iterating bounded Events (flash repeatedly). But these latter two cases are just elaborations of the basic case shown in (53), in which the ‘core’ (run to the store, flash) is embedded as the argument of a GR or PL function. (In other words, the ‘conversion’ of an Accomplishment (event) into an Activity (process) is produced by a coercion rule that adds an operator.) Accomplishments (e.g. run to the store, eat an apple) are directed Situations with a final boundary. They intrinsically take place over a period of time, so they have to be one-dimensional.
(54) Accomplishment =
+b [DIM 1d DIR] BDBY+ ([ ]) Sit
l I find it astonishing that in the face of many detailed analyses of the Vendler classes and their inadequacies, they are still standardly taken as a starting point for theories of aktionsart. By now it should be clear that Vendler’s analysis, like, say, the Syntactic Structures analysis of the English passive, was a pioneering and galvanizing achievement, but that it is time to move on.
168
parts and b oundaries
However, an Accomplishment can be subjected to the zero rule (37), which idealizes it as a point. This is what allows us to attribute an Accomplishment to a point in time, as in Bill ate an apple at 6:00. The trickiest case is the Achievements such as reach the top, arrive, die, and, notably, finish. Our analysis of this last verb in the previous section provides the key: they are all Events that mark the culmination of some larger Event. Although they are fundamentally zero-dimensional, the optional expansion with epsilon provides a little temporal window into which we can sneak a progressive.
(55) Achievement =
+b, –i [DIM 0(+e)d DIR] BD+ ([ ]) Sit
Related to the class of Achievements but not distinguished by Vendler are Inceptions such as leave, commence, and start. These are just like Achievement except that BDþ is changed to BD.
(56) Inception =
+b, –i [DIM 0(+e)d DIR] BD– ([ ]) Sit
Another class includes Point-events like flash and click. These are not regarded as having appreciable duration. (57a) gives their schema. A final class is Duratives like stay, keep, and persist, which like Activities are not inherently bounded, but unlike Activities cannot be asserted at a point in time. (57b) gives their schema.
(57)
a. Point-event =
+b [DIM 0d DIR]
Sit
b. Durative = Sit
–b [DIM 1d DIR]
The upshot of this analysis is a general agreement with such writers as Verkuyl (1989) and Pustejovsky (1991a), who regard the Vendler classes not as a basic division of the aspectual system, but rather as various realizations of a set of more fundamental parameters. Here the parameters available are those of dimensionality and bounding, motivated independently for the conceptualization of objects; and therein lies their novelty.
pa rts and boundar ies
169
5.8.4 Until and since We finally return to our initial example, The light flashed until dawn, which we can now formalize. Recall the informal analysis of section 5.3: until places a boundary on an otherwise unbounded Event. This comes out as (58).
+b [DIM 1d DIR]
(58) X until Y =
COMP Sit
X –b
BDBY+ ([Sit/Time Y]) Sit This is a bounded Event, composed of the State or Process X (until doesn’t care which), and ended by the Situation or Time Y.9 The fact that X must be unbounded in order to be the argument of COMP explains why The light flashed, which has event structure (57a), cannot appear unmolested before until. The day is saved by the coercion rule that inserts iteration, to give (59), a more complete version of our original attempt in (5).
9
For simplicity, I have treated until as a function of two arguments: the Event to be bounded and the Time. This actually does not accord too well with the syntactic pattern of the sentence, in which until Y is a modifier of the sentence expressing the Event to be bounded. Improving this analysis would take us deeply into the theory of arguments, adjuncts, and modifiers, a topic beyond the scope of this chapter.m m There are two possible ways to pursue this problem. One is to regard (58) as stipulating a syntax– semantics mismatch: until Y, although a syntactic adjunct, functions semantically as an operator taking the rest of the sentence as argument. Another way to formalize this involves makes use of either lambdaabstraction or binding (section 1.3), which in this context are notational variants. The problem is that in X until Y, X should be a semantic head rather than an argument of until, as it is in (58). The proper structure, if notated in terms of lambda-abstraction, is (ia); in terms of binding, it is (ib).
X
(i) a.
lz
+b [DIM 1d DIR] COMP
z –b
BDBY+ (Y) [continued]
170
parts and b oundaries
(59) The light flashed until dawn = +b [DIM 1d DIR] –b, +i COMP
Sit
PL
+b [DIM 0d DIR] LIGHT FLASH
BDBY+ ([Time DAWN])
For some reason, the coercion rule that inserts GR instead of PL cannot apply with until, so that Bill ran into the room until we stopped him can only mean repeated running into the room, not our stopping him before he had a chance to get all the way in. I don’t know why. (Using the progressive, Bill was running into the room until we stopped him, is cheating—it grinds the Event into a Process before submitting it to until.) However, another variation is available, seen in Bill went away until Tuesday. Here the State that results from or is the culmination of Bill going away persists until Tuesday. I am not sure how to formalize this case.n
X
·
+b [DIM 1d DIR] b.
a
COMP
–b BDBY+ (Y) These conform more closely to the syntactic form, in that UNTIL Y is a modifier. On the other hand, they yield the wrong results in terms of the telicity of the entire sentence: X until Y is bounded, but X itself is unbounded. My inclination is to accept the syntax–semantics mismatch of (58). After all, verbal aspect is also syntactically subordinate (in Simpler Syntax at least), but can affect boundedness in much the same way as until Y. n
The analysis is:
+b [DIM 1d DIR] – + BDBY (BD (BILL WENT AWAY)) + BDBY (TUESDAY) Sit
pa rts and boundar ies
171
Since is approximately the reverse of until. Bill has liked Sue since 1948 expresses a State beginning with (BDBY) 1948, containing (CONT) the discourse reference time (in this sentence, NOW), and composed of (COMP) Bill liking Sue. In The light has flashed since dawn, the most prominent reading iterates the light flashed into a process so that it can be unbounded, as required for it to be the argument of COMP. Another reading, more prominent in The light has flashed just once since dawn, appears to substitute CONT for COMP, so that a single flashing can constitute a significant part of the period since dawn. Note that ever since can be used only in the first reading: (60)
a. Ever since dawn, the light has flashed. (iterative) b. Ever since dawn, the light has flashed just once.
Some of these complications seem to be tied up with the strongly preferred use of perfective aspect with since, a problem beyond the scope of this chapter. We have not dealt with the conceptual structure of measurement and quantity, so we cannot formalize phrases like for 3 hours, in 3 hours, and 3 times, which have been crucial in the study of event structure at least since Vendler (1957). However, the present approach suggests that for 3 hours should be constructed so as to be compatible with expressions in the noun system such as 3 inches of rope, which measures out a quantity of an unbounded substance; by parallelism, X took place for 3 hours measures out a quantity of an unbounded Situation (i.e. State or Process). X took place in 3 hours ought to be parallel to Object X is located within 3 miles of Place Y; both of them require bounded entities for X. 3 times ought to just put a count on iterations of bounded Events, just as 3 cows puts a count on iterations of cow. That is, when counting and measuring can be formalized in the noun and preposition system, the treatment should generalize naturally to the aspectual system along the lines seen here in the formalization of parts, composition, and boundaries. [Such a treatment begins in Chapter 6.]
5.9 Final remarks I want to make four points in closing. First is that I have proposed what may seem like a substantial amount of machinery, including the features +b and +i, the six extracting and including functions (PL, ELT, COMP, GR, PART, and CONT), the dimensionality feature (including the epsilon dimensionality), the directionality feature, and the two boundary functions BD and BDBY. All of these parts have been necessary to get at the proper analysis of our initial puzzle,
172
parts and b oundaries
The light flashed until dawn. This may seem like excessive use of force. However, using this machinery, we have been able to address along the way a wide range of phenomena, including the plural, collective nouns like group and pile, N-of-NP constructions and N-N compounds, boundary nouns like end and crust and prepositions like to and from, the Vendler classes, progressive aspect, and the ‘imperfective paradox’. Thus we see that the true scope of the solution has proven to be an extremely broad one. A cornerstone of the solution has been the ‘X-Bar’ character of the major conceptual categories—the possibility of features and functions that apply equally to Things, Places, and Events. To the extent that the description here has been successful, this vindicates and deepens this aspect of the theory of Conceptual Semantics. Second, despite the fact that this chapter is ostensibly about lexical semantics, the distinction between lexical semantics and phrasal semantics has played only an incidental role. The very same features and functions can appear in conceptual structure by virtue of either lexical entries, morphological affixes, constructional meanings (N of NP and N-N compounds), or coercion rules. In a sense, this supports an even more fundamental tenet of Conceptual Semantics: that conceptual structure is autonomous from language and that there is no intervening level of ‘purely linguistic semantics’ intervening between it and syntax. The conceptual features and functions proposed here are indifferent to how they are expressed syntactically; it just so happens that four different kinds of correspondence rules—lexical entries, morphological affixes, constructional meanings, and coercion rules—are all capable of licensing relations between syntactic and conceptual structure in this domain. Third, let us return to the issue of semantic primitives raised in section 1.3: when we propose a conceptual analysis of a word or phrase, how do we know we have got it all the way down to primitives? The answer is that we don’t know, but this shouldn’t stop us. For instance, the identification in Jackendoff (1983) of a conceptual category Path spelled out by a repertoire of five Pathfunctions was, I believe, an advance that permitted an insightful description of many phenomena. The fact that these putative primitives have now been subjected to further decomposition in order to bring them into a still larger orbit does not negate the earlier treatment. Similarly, the functions proposed here— PL, ELT, COMP, GR, PART, CONT, BD, and BDBY—will no doubt themselves submit to further analysis, as well may the ontological supercategories Material, Situation, Space, and Time. I am not disturbed by this state of affairs. Rather, I am cheered by the analogy to our favorite high-prestige model, physics, where, as pointed out in section 1.3, the decomposition of matter into ever smaller and more general primitives has been one of the major scientific successes of [the last] century, and where the
pa rts and boundar ies
173
prospect of not yet having hit bottom is an exciting spur to further research. For those who are disturbed by semantic decomposition, the phenomena analyzed here present a major challenge for a nondecompositional theory (be it a theory of monads connected by meaning postulates, as in Fodor et al. (1980) or a connectionist theory of meaning). Finally, return to the issue of mentalist semantics, also raised in section 1.3. There has been considerable philosophical dispute (e.g. Fodor 1987, Schiffer 1987, Putnam 1988) over whether a theory of meaning is even possible. Closer examination reveals that the sort of theory in dispute is always a theory of realist semantics, that is, one that asks for a direct connection between language and the real world; it may or may not in addition contain a psychological component. Schiffer concludes that there is no such theory, and that philosophy of language must find a new set of presuppositions under which to pose questions about meaning. I would like to suggest that the proper questions to ask are those of mentalist semantics, namely the characteristics in terms of which speakers construe the reality they experience. These are the presuppositions under which the present study has been conducted, and under which some progress has apparently been made. Let me illustrate with one example. As long as one sticks with unanalyzed sentences like snow is white and tigers have stripes, one can happily remain under the presumption that sentences are connected to the world pure and simple. But consider the word end. What do the end of a table, the end of a trajectory, and the end of a speech have in common, such that we use the word end for them all? Nothing, unless we admit the possibility of schematizing objects, trajectories, and events in terms of a common abstract notion of bounded one-dimensionality. It is hard to regard this schematization as an inherent property of reality; but it makes a great deal of sense in terms of the psychological organization with which one construes reality. What we have seen here is that such psychological organization lies only a very short distance below the surface of everyday lexical items—and that progress can be made in exploring it. This suggests to me that the issue for philosophers of language ought not to be whether it is possible to do realist semantics, but rather how one can make sense of the explanations offered by mentalist semantics within a broader psychological, social, and biological context.
Remarks on Chapter 6 This chapter continues and builds on the treatment of aspectuality begun in Chapter 5. One of the main points established here is that, contrary to influential views on aspectuality, telicity is not connected in any sort of one-to-one fashion with syntactic argument structure. Rather, the telicity of a sentence is influenced by the lexical semantics of the verb and the boundedness of its arguments. In many cases, pragmatics (or world knowledge) also plays an important role. The main insight is that the temporal shape of a motion event tracks the spatial shape of the trajectory of the object in motion. This insight is then extended to other sorts of events, such as an object changing its properties, changing in who possesses it, and coming into and going out of existence. It also permits an account of covering and filling relations, and—particularly unexpectedly—distributive quantification. The analysis of telicity again depends on the notion of dimensionality originally motivated in the spatial domain (Chapters 3 and 4) and extended to events and times in Chapter 5. As a result, it becomes possible to draw a formal relationship between the fundamental spatial functions BE, GO, STAY, and EXTEND, which were treated as independent primitives in my earlier work (see section 1.3). From a larger perspective, the analysis developed here makes even clearer the independence of conceptual structure from syntactic structure. The multidimensional conceptual structures necessary to account for telicity may be expressed lexically, syntactically, or—when they are pragmatic—not at all. Furthermore, they bear little resemblance to canonical syntactic structures. They therefore are an important source of evidence for the Parallel Architecture and against the mainstream syntax-based architecture.
chapter 6
The Proper Treatment of Measuring Out, Telicity, and Perhaps Even Quantification in English (1996)
6.1 Introduction Tenny (1987, 1992) discusses a semantic phenomenon she calls ‘measuring out’. The clearest examples involve verbs of consumption and creation, for example (1): (1) a. Bill ate an apple. [consumption] b. Bill drew a circle. [creation] [This chapter appeared originally in Natural Language and Linguistic Theory 14 (1996), 305–54, and is reprinted here by permission of the publisher, Springer Netherlands.] While written to stand alone, this chapter is best regarded as a companion piece to Chapter 5, which develops the fundamental approach to aspectuality and dimensionality adopted here. My choice of title obviously is indebted to Richard Montague; readers should be aware, however, that my use of the term ‘proper’ is intended to invoke a different set of criteria than Montague is likely to have had in mind. I am grateful to James Pustejovsky, Joan Maling, Piroska Csuri, and Henk Verkuyl for important discussions on this topic, and to Verkuyl, Manfred Krifka, Carol Tenny, and Fritz Newmeyer for detailed remarks on earlier versions of the paper. This is not to say that they endorse the present version. This research was supported in part by NSF Grant IRI 92-13849 to Brandeis University, in part by Keck Foundation funding of the Brandeis Center for Complex Systems, and in part by a Fellowship to the author from the John Simon Guggenheim Memorial Foundation.
176
measuring out, telicity, quantification
Tenny’s (1987, 77–8) characterization of ‘measuring out’ is as follows. The apple referred to by the direct object in [(1a)] ‘measures out’ gradually the event described by the verb phrase. Someone who eats an apple progresses through the event in increments of apple. . . . We may think of an event as a series of snapshots of the objects involved, at points along a time line. The snapshots record the property that is changing in the object. . . . In the case of apple-eating there will eventually be some snapshot in which the apple is gone. . . . It is the existence of this distinctive point of time, provided by some changing property of the object, that makes a delimited accomplishment.
Similar considerations apply to (1b), where successive stages of the event correspond to more and more of the circle coming into existence, until a final stage in which it is complete. Dowty (1991), from similar observations, terms the direct objects in (1) ‘incremental themes’. Although Tenny and Dowty base rather elaborate (and different) theories of syntactic argument selection on the phenomenon of measuring out/incrementality, they do not characterize these intuitions formally. The present chapter develops a formal treatment within the framework of Conceptual Semantics (Jackendoff 1983, 1990). This treatment shares certain features with the related approaches of Krifka (1989, 1992) and Verkuyl (1972, 1993); Hinrichs (1985) is another important source.1
6.2 More data and intuitions on measuring out Tenny, following Vendler (1957), Verkuyl (1972), Dowty (1979), and others, observes that the events described in (1) are telic (alternatively, temporally delimited, bounded, accomplishments); the standard test for telicity in such sentences is their behavior with temporal adverbials such as those in (2): (2)
1
a.
Bill ate the apple/drew the circle in an hour/by 8:00. It took Bill two hours to eat the apple/draw the circle. b. ??Bill ate the apple/drew the circle for an hour/until 8:00.2
Krifka and Verkuyl both state their approach within a model-theoretic framework. Although the present approach is not as formally rigorous as theirs, researchers committed to model-theoretic semantics should find here a more complete statement of the linguistic generalizations that such an account must encompass. 2 These paradigms are partially corrupted by the fact that a temporal adverb can often ‘coerce’ a sentence from a normally telic reading to an atelic (activity) reading. For example, eat the apple for an hour, if acceptable, implies that the apple is not completely consumed. The progressive aspect (Bill was eating an apple/drawing a circle) always has the effect of producing an atelic reading; so it is important to perform these tests with non-progressive examples.
measuring out, telicity, quantification
177
By contrast, if the direct object of (1a) is changed to a bare mass noun, the event is atelic (alternatively, non-delimited, unbounded, activity, process): (3) a. Bill ate custard in an hour/by 8:00.3 b. Bill ate custard for an hour/until 8:00. The idea behind an account of the atelic reading is that custard describes a nonbounded quantity of a substance. By ‘nonbounded’ I mean not necessarily limitless, as would be suggested by ‘unbounded’.4 Rather, following Talmy’s (1978) analysis, the custard may well be ultimately bounded, but its boundaries are outside the context of the scenario presented in the sentence—‘offscreen’, as it were. Hence the presented scenario contains no ‘snapshot’ in which the custard is gone, and in which the event of eating is inherently completed and cannot continue. In other words, if a direct object that measures out an event is nonbounded, then the event is necessarily atelic.5 Nonboundedness can also appear with count nouns. For instance, (4a) specifies a bounded number of sandwiches, and the event is telic. By contrast, (4b) specifies a nonbounded number of sandwiches—not an infinite number, but a collection whose boundaries lie outside the frame described by the sentence. Hence the event is atelic. (4) a. Bill ate fifteen sandwiches in an hour/by 8:00/for an hour/until 8:00. b. Bill ate sandwiches for an hour/until 8:00/in an hour/by 8:00. For the moment we will concentrate here on the mass/singular count contrast ((1a) vs. (3)), returning to plurals in sections 6.8 and 6.9.
3
Again, there is a possible coercion: this sentence can be read as ‘Bill began to eat custard in an hour/by 8:00,’ a telic event. For present purposes we exclude such readings. See [Chapter 5] for how they are derived by coercion rules. 4 ‘Non-bounded’ corresponds to the feature [b] in Chapter 5, there characterized as ‘unbounded’. The change in terminology is only for reasons of clarity. 5 It is sometimes claimed that the definite article makes a mass NP bounded. However, the following example shows that this claim is false: (i) Bill ate the custard for hours/until he was full.
This example does not imply that all the custard was entirely consumed. Rather, the definiteness in (i) pertains to a previously mentioned mass of custard of nonbounded size, e.g. There was custard everywhere he looked. However, the definite article, unlike the bare mass noun, is also consistent with a bounded reading: (ii) Bill ate the custard in an hour/by 8:00.
That is, a definite mass NP is ambiguous between a bounded and a nonbounded reading, whereas a bare mass NP must be nonbounded. See section 5.5 for a bit more discussion.
178
measuring out, telicity, quantification
Creation and consumption verbs are not the only examples of measuring out. A second case appears with verbs of performance. Compare the telicity when a bounded direct object is present (5a) to the atelicity either with a bare NP object (5b) or in the intransitive use (5c):6 (5)
a. Bill read the paper/sang the tune in 2 minutes/by 8:00/?for 2 minutes/ ?until 8:00. b. Bill read fiction/sang opera for 2 minutes/until 8:00/in 2 minutes/by 8:00. c. Bill read/sang for 2 minutes/until 8:00/in 2 minutes/by 8:00.
Again, at each point in time during the events in (5a), Bill is reading a part of the paper or singing part of the tune; the event ends when all parts have been used up. When the direct object is a bare NP, as in (5b), or there is no direct object, as in (5c), nothing specifies a natural endpoint, so the events are understood as atelic. Thus these direct objects too are ‘incremental themes’. A third, somewhat different, case appears with verbs of motion. The standard observation is that (6a), containing a goal expression, and (6b), containing a measure phrase, are telic, and (6c), lacking these complements, is not. (From here on, I will abbreviate the tests for telicity to the two cases with in and for. ‘In/for’ indicates telic sentences, ‘for/in’ indicates atelic, and ‘in/for’ indicates ambiguous telicity.) (6)
a. Bill pushed the cart to NY in/??for two days. b. Bill pushed the cart four miles in/??for two days. c. Bill pushed the cart for/?in two days.
What should be added (as noted in Chapter 5) is that the choice of PP affects telicity: (6)
d. Bill pushed the cart into the house/over the bridge in/??for two minutes. e. Bill pushed the cart along the road/toward the house for/?in two minutes. f. Bill pushed the cart through the grass/down the hill in/for two minutes.
6 The caveats in nn. 2 and 3 apply to these cases. But a further caveat must also be invoked: the atelic Bill sang the tune for hours can imply that he sang the tune repeatedly, an indefinite number of times. This reading is of course impossible with creation and consumption verbs, because one cannot create or consume the same object repeatedly. For present purposes we exclude the repeated reading as well; the judgments indicated on sentences pertain to the ‘single performance’ reading. A test to distinguish the two possible atelic readings is the use of the expressions bit by bit and over and over. The former suggests that a single performance need not have been completed (as in n. 2); the latter suggests that repetition is taking place:
(i) Bill sang the tune/ate the apple bit by bit for ten minutes (and still didn’t finish it). (ii) Bill sang the tune/ate the apple over and over for ten minutes (and still didn’t finish it).
measuring out, telicity, quantification
179
In the previous cases, parts of the apple and the tune measured out the event. But in (6), it is not parts of the cart that measure out the event, but the position of the cart. That is, the cart in these examples is not an incremental theme. Using Tenny’s imagery again, the successive ‘snapshots’ of the event show the cart in different locations. The PPs in (6a,d) have specified endpoints—that is, they describe bounded paths—and the events are telic. By contrast, if the PPs in (6e) have an endpoint, it is outside the scene (notice that going toward the house does not imply getting there); thus they describe nonbounded paths, and the events are atelic. If the PP is ambiguous as to whether it has an endpoint, as in (6f), the sentence can express either a telic or an atelic event. Verkuyl (1993) notes that the Dutch preposition naar, which translates as either ‘to’ or ‘towards’, depending on context, has the same property. Notice further that adding all the way before the prepositions in (6f) disambiguates them and simultaneously makes the event telic: (7) Bill pushed the cart all the way through the grass/all the way down the hill in/for two minutes. A fourth instance of measuring out involves verbs of covering and filling, including the well-known cases spray and load:
(8) a.
Bill sprayed paint on the wall for/ in an hour. Bill loaded dirt into the truck
b.
Bill sprayed all the paint on the wall Bill loaded all the dirt into the truck
c.
Bill sprayed the wall with paint Bill loaded the truck with dirt
in/ for an hour.
in/for an hour.
(8a,b) display the same alternations as the verbs of consumption. However, (8c), even with a bare NP as the object of with, can be telic, in which case the wall is understood as completely covered and the truck as completely filled. In (8c), then, there is a reading in which the direct object measures out the event. It also measures out the object of with: enough paint must be used to cover the wall, enough dirt must be used to fill the truck. All the cases in (1)–(8) have involved direct objects of causative verbs. But similar cases occur with subjects of noncausative (unaccusative) verbs: (9) Verbs of consumption and creation a. The apple dissolved in the acid in/for an hour. A pile of dirt accumulated in/for an hour.
180
measuring out, telicity, quantification b. Custard dissolved in the acid for/in an hour. Dirt accumulated for/in an hour.
(10) Verbs of motion a. The cart rolled to New York in/for two days. b. The cart rolled four miles in/for two days. c. The cart rolled for/in two days. d. The cart rolled into the house/over the bridge in/??for two minutes. e. The cart rolled along the road/toward the house for/?in two minutes. f. The cart rolled through the grass/down the hill in/for two minutes. g. The cart rolled all the way through the grass/all the way down the hill in/for two minutes. In (9), the subject is an incremental theme, its behavior paralleling that of the object in (1)–(4). In (10), the subject is in motion, paralleling the object in (6)–(7); the PP plays exactly the same role in determining telicity as it does in (6)–(7). I do not know of analogous cases for verbs of performance; and, other than fill, I have encountered no intransitive verbs with double forms parallel to spray and load. (Swarm is often cited, but its semantics is not so simple.)
6.3 Problems with Tenny’s and Dowty’s theories 6.3.1 Tenny Appealing to the phenomena discussed above and several other phenomena with similar properties, Tenny (1987, 1992) proposes a theory called the Aspectual Interface Hypothesis, which can be summarized roughly as follows: (11) Aspectual Interface Hypothesis a. An entity that measures out an event is affected by the event. b. Affected entities invariably are realized as direct arguments of verbs. In turn, direct arguments are mapped into objects of transitive verbs and subjects of intransitive (unaccusative) verbs. c. Thus the property of measuring out, marked as a lexical property of a verb, is sufficient to determine (many aspects of) the verb’s semanticsto-syntax mapping. Let us examine claim (11a). First consider verbs of motion. In describing events of motion, Tenny speaks of the direct argument measuring out the event, just as it does with verbs of consumption, creation, and performance. In fact, though, all the syntactic variables affecting telicity in the motion case have to do with the PP
measuring out, telicity, quantification
181
or measure phrase rather than with the direct object. The presence or absence of the PP in (6) affects telicity, comparable to the presence or absence of the NP in (5). The boundedness of the PP in (6) affects telicity, comparable to the boundedness of the NP in (2)–(3) and (5a, b). Thus in the case of motion verbs, the PP rather than the NP measures out the event. However, the PP is not affected by the event: nothing happens to the Path, nor necessarily to anything that marks the Path. Contrary to claim (11a), then, some entities that measure out events are not affected by the event. (The difference between motion verbs and the other cases of measuring out is acknowledged by Tenny (1994), weakening the claims in (11).) Contrariwise, some entities are affected by an event but do not measure it out. Consider verbs like chew, knead, jiggle, and spin. Using a standard test for affectedness (Lakoff 1970, Jackendoff 1990), the objects of these verbs are affected:7 (12)
What John did to the bread was chew/knead/jiggle/spin it.
But the bread does not measure out the event. Even if it is construed as a bounded loaf, the event is not telic: (13)
John chewed/kneaded/jiggled/spun the loaf of bread for/in an hour.
For a different case, the causative verbs point and face take an affected object that clearly changes position: (14)
What Bill did to the sign was point/face it at the house.
However, in this case, unlike the motion verbs, a nonbounded preposition results in a telic event: (15)
Bill pointed/faced the sign toward Harry/along the road in/for two minutes.8
In addition, Dowty 1991 and Verkuyl 1993 point out that many verbs of consumption and creation do not necessarily measure out in the sense suggested 7
Tenny’s (1987, 1992) criterion for affectedness, the ability to form middles, is a bit broader than the do to test. For instance, This book reads easily, which passes Tenny’s test, is much better than What Bill did to the book was read it. More generally, the relation of the middle test to measuring out is obscure. For instance, it is hard to see how the objects of buy and sell can differ with respect to measuring out, yet sell passes the middle test and buy does not: (i) This book sells/buys well.
Tenny (1994) no longer relies on the middle test for affectedness, but as far as I can tell she has not proposed a replacement that is independent of telicity, i.e. that is not circular. 8 Again there is a coercion reading that I ignore: ‘Bill pointed the sign toward Harry, and it stayed that way for 2 minutes before he moved it again.’
182
measuring out, telicity, quantification
by Tenny in the passage quoted above. An apple, because of its size, takes several bites to eat, so one can produce a series of snapshots of partially eaten apple. On the other hand, one can eat a handful of peanuts in a single gulp, and there will be no intermediate snapshots in which some of the peanuts remain intact. Thus the measuring-out property of eat is evidently connected in part to the nature of the objects in question and the pragmatics of how the action can be carried out on them. Similarly, one can imagine a circle being drawn by some sort of computerdriven plotter that sprays the entire circle on the page at once. In such a case there will be no snapshot where part of the circle is drawn and another part is not. The same is true of the spray/load verbs. A single shpritz may suffice to spray your shoes with water; a single scoop of a steamshovel may suffice to load a truck with dirt. Hence, many of the standard examples of measuring out by direct objects turn out to have a partially pragmatic basis. (For other cases such as sing and build it is harder to imagine non-measuring-out circumstances.) But these pragmatic choices make no difference for affectedness or for argument structure, which remain fixed. Conversely, entities in motion normally do not measure out events, as seen in (6)—rather, the Path does. However, under certain pragmatic circumstances, the entity in motion can measure out the event ((16a) is taken from Declerck 1979, and (16b–d) from Dowty 1991): (16) a. b. c. d.
The parade passed the house. John entered the icy water (very slowly). The crowd exited the auditorium (in 21 minutes). Moving slowly but inexorably, the iceberg took several minutes to pierce the ship’s hull to this depth. e. Fog rolled into the city (for hours).
In these cases we can imagine a series of snapshots in which different parts of the parade are moving by the house, different amounts of John’s body are already in the water, and so forth. Again, these changes in measuring out have no impact on affectedness or argument structure. In short, Tenny is correct in perceiving an important link between affectedness (or Patienthood) and the direct argument position (Jackendoff 1990, chapter 11). But a range of relatively straightforward data shows that affectedness and argument structure are not directly connected with measuring out and telicity, as Tenny claims.9 9 Grimshaw (1990) uses what she calls an ‘aspectual tier’ in semantics as a determinant of syntactic argument structure. Her main motivation for this tier (chapter 2) is based on the
measuring out, telicity, quantification
183
6.3.2 Dowty Dowty 1991 posits a less direct correlation between measuring out and argument position. He introduces the term ‘Incremental Theme’ for any argument involved in a measuring out relation. Such an argument need not be a theme in Gruber’s (1965) sense; for instance, Dowty speaks of PPs like those in (6) as ‘Incremental Theme Paths’. That is, he views various traditional thematic roles as permitting an additional overlay of incrementality. In Dowty’s theory, incrementality is one of several factors that contribute weight to the ‘ProtoPatienthood’ of an argument, and hence to its being assigned to the direct object position. But even the partial correlation between incrementality and Proto-Patienthood is suspect. In Dowty’s system, ‘Proto-Patient’ contrasts with ‘Proto-Agent’, which normally is assigned to subject position. The factors that contribute to Proto-Agency are movement, volition, sentience/perception, independent existence, and causation (Dowty 1991, 572–3). Let us examine these factors one by one. We will find that they largely predict whether incrementality is possible in the subject; hence there will be (little or) no need to mention incrementality as a separate factor biasing an argument towards object position. First consider volition and sentience. They are properties of individuals; they cannot be attributed to substances or successive parts of people. Thus if volition or sentience is attributed to an argument, it thereby is biased toward subject position, and at the same time it prohibits incrementality. Second, Dowty illustrates the factor ‘independent existence’ using the stative verb need, which because of its stativity cannot manifest incrementality. Third, consider movement. In cases such as (16) and (17), incrementality does occur with subjects to which motion is attributed.
distinction between the fear class and the frighten class of verbs. But her argument that this distinction is specifically aspectual (pp. 26–8) is (to me at least) obscure: the aspectual analysis of fear is simply stipulated, and that of frighten is (as I read it) evasive. Moreover, even as worked out, Grimshaw’s treatment of the distinction has nothing to do with telicity as such, since both fear and noncausative frighten (the relevant case) are atelic. This suggests that Grimshaw’s ‘aspectual’ dimension in fact has little connection with Tenny’s. Consequently, whatever success Grimshaw’s analysis achieves by including an ‘aspectual’ dimension, it does not constitute evidence for Tenny’s hypothesis. As mentioned above, Tenny 1994 weakens some of the claims of the Aspectual Interface Hypothesis, in particular by introducing a different treatment of motion verbs in which the Theme no longer directly measures out the event. She also explicitly excludes from her analysis verbs whose direct arguments are not affected. But she still maintains that there is a strong connection between measuring-out and argument structure: ‘Only the aspectual part of thematic structure is visible to the universal linking principles’ (Tenny 1994, 116)—a position which is refuted here.
184
measuring out, telicity, quantification
(17) a. Water gradually filled the boat. b. Water descended the mountain for hours. (16) and (17) have direct objects, so the subject cannot have been moved from object position by Unaccusative Movement. Hence incrementality must in fact be possible in subject position. Dowty’s remaining factor that biases an argument toward subject position is causation. Here one does in fact find incremental subjects, but I have been unable to find examples where the subject is incremental and the object is not. For instance, in Dripping water gradually discolored the stone, the water is incremental, but so is the stone’s state of discoloration.a The real generalization, therefore, is not that incrementality biases an argument towards object position, as Dowty claims. Rather, incrementality can occur equally well in subject and object positions (as well as in other arguments such as Path PPs). However, it is excluded by certain kinds of semantic factors such as volition; and a cause cannot be the sole incremental argument in a sentence. In addition, as observed above, pragmatic factors that induce incrementality have no effect whatsoever on argument structure. I conclude that Incremental Theme is not a factor in argument selection.10
6.3.3 Verkuyl and Krifka Verkuyl and Krifka are also concerned with pinpointing those arguments that can induce telicity effects. Verkuyl (1972, 1993), noticing that the choice of verb makes a difference, proposes a lexical feature of the verb called [þADD-TO]. This feature interacts formally with the boundedness feature of an argument (called by Verkuyl [þSpecified Quantity] or [þSQA]) to produce telicity distinctions. Krifka (1992) claims that incrementality is assigned in terms of the 10
Verkuyl (1972, 1993) has also steadfastly maintained a strong asymmetry between incrementality in object and in subject position; indeed, it forms an important part of his theory of compositionality. I suspect that he is responding to the same symptoms as Dowty, especially the fact that volition and sentience, most closely associated with subject position, cannot be incremental. a
Further critique of Dowty’s notion of Proto-Agency appears in Jackendoff 2007a (203 n. 8). Briefly:
.
Dowty’s criterion of ‘independent existence’ probably has something to do with Topicality rather than agency, as Dowty notes. The criterion of ‘movement’ picks out the role of Theme (the object whose motion or location is being specified). Because theme precedes location in the linking hierarchy (Jackendoff 1990, chapter 11), themehood creates a pressure toward subjecthood if there is no agent. Volitionality and causation are genuine signs of agency, and they create a bias toward subject position. Sentience does create a pressure towards subject position, but in ‘psych-predicates’ the sentient argument may appear in object position instead, as in The noise annoys Bill.
.
. .
measuring out, telicity, quantification
185
thematic role of the NP in question; the idea is that certain thematic roles produce telicity effects. However, he does not characterize the requisite thematic roles, nor their interaction with pragmatic factors. Our task here will be to pick out which verbs require telicity effects and which verbs merely allow them—that is, a deeper understanding of Verkuyl’s diacritic [þADD-TO]—and to describe the interaction between boundedness of arguments, choice of verb, incrementality, and telicity.
6.4 From snapshots to motion In order to develop a more rigorous analysis of the semantics of measuring out, I want to focus first on motion verbs, for which the intuitions are clearest.11 Tenny and Verkuyl suggest that an event of motion be viewed as a series of snapshots, each of which depicts the object in motion in a different location. Primitive versions of this ‘snapshot’ view, for instance Schank 1973 and Jackendoff 1972, essentially reduce a motion to its beginning and end states; that is, X goes from Y to Z is given no more inferential structure than X is first at Y and then at Z. On this view, the Source (Y) and Goal (Z) are the only points for which location of the Theme (X) is specified. Such accounts also make it impossible to describe atelic motion, where there is no endpoint. Better articulated versions, for instance Miller and Johnson-Laird 1976, Langacker 1987a, Pustejovsky 1991a, and Verkuyl 1993, formalize motion in terms of a countable succession of points. Here is Pustejovsky’s treatment (1991a, 56); Miller and Johnson-Laird’s, Langacker’s, and Verkuyl’s are in a similar spirit: (18)
Process (P): a sequence of events identifying the same semantic expression Examples: run, push, drag Structural representation: P e1
. . .
en
where e1 . . . en are subevents, in each of which the theme is in a different location I wish to reject this ‘snapshot’ conceptualization, on the grounds that it misrepresents the essential continuity of events of motion. For one thing, aside from the beginning and endpoints, the choice of a finite set of subevents 11 Here I concur with Verkuyl’s emphasis; I diverge from Tenny and Krifka, who treat motion only secondarily.
186
measuring out, telicity, quantification
is altogether arbitrary. How many subevents are there, and how is one to choose them? Notice that to stipulate the subevents as equally spaced, for instance one second or 3.5 milliseconds apart, is as arbitrary and unmotivated as any other choice. Another difficulty with a ‘snapshot’ conceptualization concerns the representation of nonbounded events (activities) such as John ran along the river (for hours). A finite sequence of subevents necessarily has a specified beginning and ending, so it cannot encode the absence of endpoints. And excluding the specified endpoints simply exposes other specified subevents, which thereby become new endpoints. Thus encoding nonbounded events requires major surgery in the semantic representation. For example, Verkuyl 1993 builds a nonbounded interval out of a union of an infinite number of bounded subintervals. If there were such a major semantic distinction between bounded and nonbounded intervals, one might expect some reflex in their syntax. But to the house and toward the house, for example, are altogether parallel syntactically (and even use the same preposition in Dutch). A final difficulty with the ‘snapshot’ conceptualization is that it does not distinguish its chosen sequence of subevents as motion: it just specifies a sequence of momentary states. It does not say that in each of these snapshots the object is moving, rather than suddenly appearing at a new location in the new subevent. Except under strobe lighting and in certain cases of pathology, we do not experience the world like an old-time movie, flickering from one state to another.12 Instead of treating motion as a finite sequence of states, I would like to encode it as continuous change over time. On such an approach, we conceptualize both the path over which motion takes place and the time period during which motion takes place as dense intervals—they map onto the real numbers, not the natural numbers. The position of the theme along the path is encoded as a function of time, so that, for any arbitrary moment of time, there is a corresponding position.13 Moreover, the function relating position and time is continuous in the mathematical sense: there are no instantaneous jumps in 12
Crick 1994 cites the description by Hess, Baker, and Zihl 1989 of a patient lacking awareness of most types of movement. ‘When first examined, the patient was in a very frightened condition. This is not surprising since objects or persons she saw in one place suddenly appeared in another without her being aware they were moving. This was particularly distressing if she wanted to cross a road, since a car that at first seemed far away would suddenly be very close. When she tried to pour tea into a cup she saw only a glistening, frozen arc of liquid’ (Crick 1994, 192). 13 Here I diverge from Verkuyl (even though he claims to be translating my views into modeltheoretic terms) and from Hinrichs 1985, both of whom treat a Path as a spatio-temporal complex—a sequence of locations linked with a sequence of times. As justified in detail in Jackendoff 1983, I treat a Path as a purely spatial entity. The basic reason for doing so is that a Path can play a role not only in an event of motion, as in (i), but also in three kinds of states, none
measuring out, telicity, quantification
187
position. (The appendix to this chapter suggests more precise treatments of the denseness of intervals and the continuity of functions between intervals.) I wish also to relate the path and the time to the structure of the event in which they each play a role. The theme is at the beginning of the path at the beginning of the event and at the end of the path at the end of the event. If the path has distinct segments, then the event can be divided into segments corresponding to when the theme is on the associated parts of the path. This connection between the path and the event appears in Verkuyl’s description of the verb as an ‘odometer’ that keeps track of the progress of the theme along the path; Krifka likewise speaks of the ‘transfer of reference properties’ from the theme to the event. In addition, the beginning and end of the event take place at the beginning and end of the time respectively. If the event has distinct segments, the first segment takes place before the second, and so forth. This connection too is shared by Verkuyl and Krifka. Verkuyl has a ‘B-function’ that maps the theme into the set of spatial-temporal positions it goes through; Krifka has a ‘temporal trace function’ from the event to time. These connections establish the relationship between the boundedness of the path and the telicity of the event. If the path has an explicit endpoint, then the event and the time period have explicit endpoints as well, and the event is telic (an accomplishment). If the path lacks an explicit endpoint, then so do the event and the time period, and the event is atelic (an activity).
6.5 A formalism for measuring out 6.5.1 Decomposing objects into axis plus cross-sectionb To express these intuitive connections among the path, event, and time, I wish to invoke another mathematical analogy, the notion of decomposing an object into a primary linear dimension plus a cross section, based on Marr’s (1982) theory of the conceptualization of shape and developed in section 5.6. of which invoke time at all: the extension of an object along a Path, as in (ii), the orientation of an object along a Path, as in (iii), and the location of an object at the end of a Path, as in (iv). (i) (ii) (iii) (iv)
Bill ran down the road. The line of trees extended down the road. The sign pointed/faced down the road. My house is three miles down the road.
Time plays a role only in the motion case. It is proper therefore to attribute the correlation of time and space to the motion predicate itself rather than to the Path. b
Some paragraphs from the original version of this section have been transposed to Chapter 5.
188
measuring out, telicity, quantification
The formal notation to be used here for dimensionality is shown in (20); it is slightly more elaborate than that in Chapter 5. A tube is a 2d annulus projected along a 1d (linear) axis; an H-beam is a 2d H-shape projected along a 1d axis.
(20) a. Tube
b. H-beam
[1d] 2d Annulus shape
[1d] 2d H-shape
The double vertical connecting the two specifications of dimensionality symbolizes the formal operation of projecting a cross-section (the bottom dimensional specification) onto an axis (the upper dimensional specification). As Marr argues, the notion of projecting a cross-section onto an axis is cognitively necessary in order to capture the way we understand objects. The basic principle of inference for such representations is that any part of the axis has a corresponding cross-section. In particular, the endpoints of the axis correspond to cross-sections of the object, and these cross-sections are what constitute the ends of the object (Chapter 5). Similarly, any arbitrary intermediate point or segment of the axis corresponds to a representative cross-section of the object.
6.5.2 The cross-section of an event of motion Now let us generalize this decomposition of objects to events. Suppose we think of an event of motion as a multidimensional construct spread out along some interval on the time axis [t1–t2]. (Alternatively, the interval may be open on either end.) Then for each point ti on the time axis there will be a ‘cross-section’ or ‘representative slice’ of the event that represents the theme at some location li. Let’s first formalize this slice. Using the notation of Jackendoff (1983, 1990), the form of such a slice would be (21):
(21)
BE([Thing X], [Place li]) State
[Time ti]
(21) encodes a State that consists of a Thing X, the theme, being at the Place li; the Time ti at which this State obtains is encoded as a modifier of the State.14 14 Nothing particular hangs on this notation; the reader is free to substitute another, so long as it makes the same distinctions, in particular separating the notion of a location, notated here as
measuring out, telicity, quantification
189
In order to be suitable for present purposes, this notation must be altered somewhat. One change is substantive: the notation as it stands does not encode the obvious relation between the categories of Place and Path, nor that between the categories of State and Event. Chapter 5 suggests that Places and Paths fall under a supercategory of Space. Paths are those Spaces that are one-dimensional and have a direction; all other Spaces are Places. For instance, to the house and toward the house express pieces of space that are one-dimensional and have a specified direction (and are therefore Paths). By contrast, at the house expresses a zero-dimensional piece of space, and on the table expresses a two-dimensional piece of space; both are therefore Places. Along the river expresses essentially a one-dimensional space, but it has no inherent direction. It can therefore serve as a Place (The house is along the river) or, by imposing a direction, as a Path (Bill ran along the river). Similarly, the distinction between States and Events is that Events are directed— they have an implicit connection with the passage of time—and States are not. So these two categories can be subsumed under a supercategory Situation, also distinguished by the feature of directedness. Consequently, in (21) we will replace the label State by the label Sit(uation). Since situation (21) occurs at a point in time, we will also add the feature [0d] (zero dimensionality). Similarly, we will replace the label Place by Space and add the feature [0d]. This cross-section occurs at a point in time, so Time also receives the feature [0d]. (For simplicity, I will ignore the feature of directedness, which plays no essential role in what is to follow.) Finally, for typographical reasons to become clear in a moment, we will move the Time constituent out to the right of the BE function and its arguments, separating it with a semicolon. This is a simple notational change with no [Place ]), from the predicate of being at a place, notated here as [State BE ( . . . , [Place . . . ])], and from the time at which the Theme occupies this place. See Jackendoff (1983, chapter 4) and n. 13 above for justification of these distinctions; some of the arguments are implicit in what is to follow here. Some readers may wonder why I have treated Time as a modifier of BE rather than as a third argument. Logically, it is of course possible to treat every State and Event as having an extra Time argument. However, with the exception of a few verbs such as last, Time never appears as a syntactic argument of the verb. It is always specified by a time adjunct and/or by Tense (which, following something like a Reichenbachian account, specifies the time of the event only in relation to utterance time and a presupposed reference time). The treatment in (21) makes the distinction between Time and the other participants in the event explicit. However, nothing in the analysis to follow turns on the distinction, and readers who so prefer should feel free to consider Time a third argument of BE. Finally, aficionados of the Davidsonian Event variable should observe that the notation in (21) contains such a variable implicitly, in the form of the [State ] constituent. Existence claims, referentiality claims, and quantification are carried out over this constituent, just as they are carried out over a [Thing ] constituent, and there is no need for a separate variable to do this work.
190
measuring out, telicity, quantification
theoretical significance. We therefore arrive at (22) as our representation of the ‘temporal cross-section’ of an event of motion:
(22)
0d BE ([ Sit
Thing
X],
Space
0d ; li
Time
0d ti
6.5.3 Projecting the cross-section of an event onto axes We next must project this cross-section onto the event of motion as a whole. In the case of objects, shown in (20), the cross-section is projected along a single linear axis. However, in (22) there are three axes to consider at once: the pointsituation is projected onto a durative event (we can think of the axis as the course of the event); the point in space is projected onto a path; and the point in time is projected onto a time interval:
(23)
[1d]
Sit
[1d]
0d BE ([
Thing
X], [
Space
[1d]
0d]); [
Time
0d]
We are now in a position to add the crucial part that specifies measuring out. The intuition behind measuring out is that the three one-dimensional axes in (23) are not independent; rather, they are linked to one another. In particular, if one chooses any particular point on the event axis, the cross-section it determines is linked to particular points on the path axis and on the time axis. The correspondence is smooth and continuous, so that nearby points on the event axis correspond to nearby points on the path and time axes, and continuous intervals on the event axis correspond to continuous intervals on the path and time axes. Finally, the topological property of boundedness is preserved: a bounded interval on one axis maps into bounded intervals on the other two axes, and similarly for a nonbounded axis. The relation between the event and time axes (Krifka’s (1992) ‘temporal trace function’) can be characterized as an isomorphism: the parts of the event map one to one onto the parts of the time period, and vice versa. The story with respect to the path axis is slightly more complex. In the default interpretation, the theme moves smoothly along the path from the beginning to the end, so the event and path axes are also isomorphic. However, in the course of an event of motion, it is also possible for the theme to temporarily pause or even to reverse
measuring out, telicity, quantification
191
direction from time to time (Verkuyl stresses this point). Hence there may be more than one cross-section of the event in which a particular point of the path is occupied by the theme. In the general case, then, the relation between the event and path axes is somewhat weaker than an isomorphism. Leaving aside the details of this relation (spelled out in the appendix), we will say that the event, path, and time axes are connected by a structure-preserving binding relation, or in brief that they are sp-bound. We will notate this relation by superscripting Greek letters to the axes in question. Using this notation, the representation of the event of motion becomes (24):
(24)
[1d]·
Sit
0d BE ([
[1d]·
Thing
X], [
Space
[1d]·
0d]); [
0d]
Time
[sp-bound axes]
[cross-section]
In (24), the position of the theme at a particular time is represented by the crosssection. The theme’s change of position measures out the event, because the path axis along which the theme’s position is projected is sp-bound to the axis that represents the course of the event and to the axis that represents the time interval over which the event takes place. That is, measuring out is a consequence of the sp-binding of the path, event, and time axes. Verkuyl’s account of the dependence among event, path, and time is similar to the one to worked out here. However, Verkuyl uses a finite (or ‘snapshot’) conceptualization of the course of events. The novelty here is in the use of continuous axes, parallel to the move in mathematics to a differential calculus. Krifka’s account, like the present account, allows for continuous change, although, as mentioned in n. 11, he does not deal with motion predicates.
6.5.4 GO reduces to BE plus sp-bound axes Jackendoff (1983, 1990) and many other authors treat an event of motion in terms of a putatively primitive function GO, whose two arguments are a theme and a path. (24) has the effect of decomposing this primitive. The cross-section specifies the momentary location of the theme argument; this cross-section is projected along three axes, one of which is the Path argument of the GO function. The three axes are sp-bound to each other, encoding the correlation of position, time, and event structure.
192
measuring out, telicity, quantification
Such a decomposition accomplishes two things. First, it reduces GO to the more primitive function BE—as is desired by those who wish to treat motion as a series of locational ‘snapshots’. However, the reduction is in terms of a continuous function, thereby meeting the objections raised in section 6.4 to the ‘snapshot’ approach. Second, because of the sp-binding, (24) explicitly represents the Path argument as measuring out the event and the time, providing a direct formal link between the boundedness of the path and the telicity of the event.15 Is the cost of this notation worth it? Wouldn’t we be better off with a simple primitive function GO than with this cumbersome machinery? The answer is no, because all the cumbersome machinery is independently necessary. Let us go through the pieces one by one. First, the machinery of dimensionality is necessary to encode the form of objects in any event—as we have seen in (20). Second, the notion of projection of a cross-section along an axis is independently necessary for the description of objects a` la Marr (or in fact for any intuitive description of bars, beams, and the like). There remains the notion of sp-binding among axes to be justified. The rest of this chapter is devoted to working out what mileage we can get from this notion.
6.6 Telicity with motion verbs Let us see how the sp-binding in (24) accounts for the interaction between measuring out and telicity in motion sentences. Recall that a telic event has a definite termination, a final state; the associated time period is also endbounded. Because of the sp-binding, the final state of the event can be characterized as the cross-section of the event at the final point in time. By contrast, an atelic event has no such definite endpoint, and the time period is likewise nonbounded. In (24), the boundedness of the event and the time are linked by sp-binding to the boundedness of the path. If the path has an endpoint, so do the event and the time period; if the path has no definite endpoint, neither do the event or the time period. In other words, using [+b] as the feature signaling presence (þ) or absence () of an end boundary, only the following two 15 This is not to stop us from continuing to use the notation GO(X,Y). It is just that its theoretical status changes from a primitive to a convenient abbreviation for (24): it is a ‘molecular’ rather than ‘atomic’ conceptual structure. See section 1.3 and Jackendoff (1990, 4) for discussion of what it means to decompose what was previously treated as primitive.
measuring out, telicity, quantification
193
configurations are well formed; any other combination of the boundedness features is not:c
(25) a. End-bounded event, path and time 1d
·
1d
+b
·
1d
+b
·
+b
0d Sit
BE ([
Thing
X], [
Space
0d]); [
Time
0d]
b. Nonbounded event, path and time 1d
·
1d
+b
·
1d
+b
·
+b
0d Sit
BE ([
Thing
X], [
Space
0d]); [
Time
0d]
In a standard expression of motion, the path is expressed but the time is not. Thus the path appears to control the telicity of the event. For instance, the prepositions in (26) specify bounded paths, so the event is telic, and the sentence allows time adverbials that pertain to bounded time periods; we are dealing with configuration (25a). By contrast, the prepositions in (27) specify nonbounded paths, so the opposite is the case, and (25b) is the relevant structure: a. The cart rolled to NY/into the house/over the bridge (in/for an hour). b. Bill pushed the cart to NY/into the house/over the bridge (in/for an hour). (27) a. The cart rolled along the road/toward the house (for/in an hour). b. Bill pushed the cart along the road/toward the house (for/in an hour). (26)
In (28), the path is specified by a measure phrase that expresses the path’s length rather than its spatial configuration. Since a measure phrase specifies a definite length, the path must have an endpoint; hence the structure must be (25a) and the event is telic:
c —except in the case of an object (say a pendulum) undergoing continuous oscillation. In this case the Situation and Time may be –b while the Space is +b. This is accounted for in the formalization of sp-binding in the appendix.
194
measuring out, telicity, quantification
(28) a. The cart rolled four miles (in/for an hour). b. Bill pushed the cart four miles (in/for an hour). The prepositions through and down (and Dutch naar, ‘to/toward’) are ambiguous in boundedness. For example, through the door normally specifies a path from one side of the door to the other, but through the sky normally specifies a path that has no definite endpoints (presumably because in turn the sky is nonbounded). Correspondingly, sentences of motion such as (29a–c) are of ambiguous telicity, depending on which reading of the preposition is most felicitous. However, if a specifier such as all the way is added to such a PP, as in (29d), then an endpoint is invoked and, by the sp-binding, the event becomes unambiguously telic: (29) a. b. c. d.
The cart rolled through the grass/down the hill in/for ten minutes. The marble rolled through the door in/??for a second. The cloud floated through the sky for/in an hour. The cart rolled all the way through the field in/for ten minutes. (similarly with causatives)
In the cases so far, sp-binding has used the boundedness of the path to determine the boundedness of the time period. However, the influence can also pass in the opposite direction, from time to path. In cases of ambiguous telicity such as (29a), the choice of temporal adverbial fixes the boundedness of the Time constituent; this determines whether structure (25a) or (25b) is appropriate, which in turn determines whether the path is construed as bounded or not. (Krifka’s ‘transfer of reference properties’, which performs a function parallel to the sp-binding of path to event, is similarly bidirectional.) On the other hand, if the Path and Time constituents specify opposite values of boundedness, it should be impossible to impose a consistent value across the three axes in (24). And indeed such sentences are unacceptable, for instance John ran toward (b) the house in an hour (þb) or John ran into (þb) the house for an hour (b) (which undergoes a coercion into an iterative reading). In short, the dependence between telicity of motion sentences and boundedness of their PP complements follows clearly from the analysis of motion in (24). What if there is no PP complement (e.g. The cart rolled; Bill pushed the cart)? As we saw in section 6.1, these are always interpreted as atelic. It therefore appears that in the absence of a stipulated path, a nonbounded path is assumed by default.16
16
Alternatively, in the absence of a specified path, these verbs are interpreted as pure manner-of-motion, which is a process (activity), hence atelic. On this alternative, the GO function (motion along a path) is imposed on the manner-of-motion by coercion when a path is
measuring out, telicity, quantification
195
Next consider transitive verbs such as enter, approach, and cross. Jackendoff (1983, 1990), following Gruber (1965), treats these verbs as ‘incorporating’ a path-function into their LCS: enter NP decomposes in part as ‘go into NP’; approach NP as ‘go toward NP’; and cross NP as ‘go across NP.’ More formally, (30a) is the LCS of approach in the 1983 notation. (30b) gives the more elaborate decomposition developed here, in which the [b] feature of the Path-function TOWARD is the source of the atelicity. (In turn, TOWARD may be decomposed as in the analysis in chapter 5, example (46).)
(30) X approaches Y a. [Event GO ([Thing X)], [Path TOWARD ([Thing Y])] b.
1d –b
·
1d –b TOWARD ([Thing Y])
·
1d –b
·
0d Sit
BE ([
Thing
X], [
Space
0d]);
[
Time
0d]
The telicity of the event expressed by these verbs thus depends on the boundedness of the ‘incorporated’ path-function. Since into is bounded, enter is telic (31a); since toward is nonbounded, approach is atelic (31b). Since across is ambiguous (like through), the telicity of cross is ambiguous, and can depend on the choice of object (31c, d). Similarly, the transitive use of walk, as in walk the Appalachian Trail, ‘incorporates’ a path-function of ambiguous boundedness, something like the union of along and all the way along; hence this use too is of ambiguous telicity (31e): (31)
a. b. c. d. e.
Bill entered the room in/for a minute. Bill approached the mountain for/in an hour. Bill crossed the border in/for ten seconds. The cloud crossed the sky in/for a minute. Bill walked the Appalachian Trail in/for six months.
6.7 More applications of sp-binding Having seen in some detail how sp-binding applies to events of motion, we next look at a broader range of applications. expressed. See Jackendoff (1990, chapter 10); Levin and Rapoport (1988), Pustejovsky (1991b) for such treatments of manner-of-motion verbs (though without attention to the aspectual details).
196
measuring out, telicity, quantification
6.7.1 Events vs. States What distinguishes an Event from a State? Intuitively, an Event has a timecourse, whereas a State ‘just sits there’, with no dependence on time—only a location in time. This suggests that Events but not States have an axis that is sp-bound to time:
(32) a. Canonical State: [Sit F(X,Y); [Time T]] b. Canonical Event: 1d
Sit
·
0d F(X,Y);
1d
·
0d T Time
(32a) has no dependence between the time constituent and the structure of the situation. The time interval may be a point or an open or closed interval, but the state remains unchanged. By contrast, (32b) does exhibit a dependence between time and the structure of the situation: precisely what Krifka has in mind with his ‘temporal trace function’. On the present approach, this dependence is the hallmark of all event concepts.
6.7.2 Staying Suppose we create an Event just like (24) except that the Space position is not projected onto an axis sp-bound to the event:
(33) 1d
·
1d
·
0d Sit
BE ([Thing X], [Space 0d]); [Time 0d]
This is an event in which a thing X remains in the same place over a period of time. In other words, (33) decomposes the primitive function STAY of Gruber (1965) and Jackendoff (1983, 1990)—the basic function involved in verbs
measuring out, telicity, quantification
197
such as stay, remain, and their causative keep. Notice that (33) is minimally different from a State: it has event structure, but nothing changes within the event.
6.7.3 Extending, covering, and filling Next consider the possibility of treating the theme as the projection of a crosssection. Suppose, for instance, that the theme is an object like a beam or a tube or a road, so that it has a representation like (20). Then it is possible to sp-bind its linear axis to another linear axis in the situation. (34) is one combination (another appears in section 6.9):
(34)
1d
Sit
·
0d BE ([
1d
Thing
·
1d
2d]), [
Space
·
0d]; [
Time
T]
Since time is not projected in (34), this Situation is a State. However, the State has a cross-section in which a cross-section of the theme is located at a crosssection of the path; these cross-sections are projected over the linear axes of the theme and the path to form the entire State. As a result, the theme extends over the path. Thus (34) decomposes the primitive EXT of Jackendoff (1983, 1990) (‘Extensional GO’ in Gruber (1965)) to the primitive function BE plus projection and axial sp-binding. This function appears in such examples as The road goes from New York to Boston and The tube extends along the edge of the roof. The path can be said to measure out the theme (or vice versa), but spatially rather than temporally—a case not considered by Tenny, Dowty, Krifka, or Verkuyl. A slight modification of (34) also provides a nice way to encode covering and filling relations (ALL ON and ALL IN in Jackendoff 1990). When X covers Y, every point on the surface of Y is occupied by a part of X; when X fills Y, every point in the volume of Y is occupied by a part of X. That is, Y again measures out X spatially, but in two and three dimensions respectively. We can express this by extracting from the cross-section a surface or volume rather than an axis:
198
(35)
measuring out, telicity, quantification
a. X covers Y (every point on Y is occupied by a point of X) 2d
Sit
·
2d
0d BE ([
Thing
·
2d
0d]; [
Space
ON [
Thing
·
0d]]); [
Time
T]
b. X fills Y (every point in Y is occupied by a point of X) 3d
Sit
·
0d BE ([
3d
Thing
·
0d], [
3d
Space
IN [
Thing
·
0d]]); [
Time
T]
Notice that covering and filling relations are the end states of spray and load respectively, on the reading in which the surface is totally covered and the container is totally filled. We return to these cases in section 6.10.
6.7.4 Verbs of change of possession One of the great insights of Gruber (1965) is that change of possession can be treated as parallel to change of location, thereby accounting for the use of such verbs as be, go, and keep and the prepositions from and to in possessional as well as locational expressions. Under Gruber’s analysis as adapted in Jackendoff 1983, change of possession is treated as a case of the conceptual function GO(X, P), where X is the object changing possession and P is a path from the prior owner to the recipient. What has not been pointed out in this analysis is that change of possession is aspectually different from change of location. Unlike spatial motions, events like Bill gave/lent/sold the book to Harriet are basically without duration: the time at which the book ceases to belong to Bill is not before the time at which book comes to belong to Harriet. If we add temporal modifiers to sentences such as those in (36), these pertain to the time it took Bill to make up his mind, or to locate the book, or some other preliminary action. The actual change of possession is instantaneous: (36) Bill gave/lent/sold the book to Harriet quickly/in two hours. It took Bill two hours to give/lend/sell the book to Harriet.
measuring out, telicity, quantification
199
The difference in aspectuality between motion and possession sentences follows from the present theory. Consider the nature of the ‘possessional path’ from Bill to Harriet. Unlike spatial paths, there are no points in between the Source and the Goal. ‘Possessional space’ is not continuous—it consists only of separate points corresponding to possible owners of objects. This fact about the nature of ‘possessional space’ leads directly to the desired aspectuality. On the present analysis, in an event of motion, the event and time axes are sp-bound to the path. Since the path has no points between Source and Goal, the time axis can have no points between the initial point and the endpoint—precisely the desired conclusion. I take this to be an unexpected and highly advantageous consequence of the present analysis.
6.7.5 Changes of properties Consider next Gruber’s Identificational domain, the class of predicates that ascribe properties to objects. Gruber considers predicates of change of state to express motion in the Identificational domain. It happens that some predicates of change of state are telic (37), some atelic (38), and some of indeterminate telicity (39) (see again nn. 2 and 3): (37)
a. The water got hot in/for two minutes. b. The amoeba flattened out in/for two minutes.
(38)
a. The water got hotter for/in two hours. b. The crowd grew for/in two hours.
(39)
a. The water cooled down in/for two minutes. b. The chemical reddened in/(continuously) for half a minute.
When these changes are spread out over time, as in (37)–(39), the property in question is regarded as forming a linear scale. For example, temperature is regarded as scalar in the (a) sentences; degree of flatness, size, and redness are treated as scales in the (b) sentences. Following Gruber, the object ‘moves’ along the scale, so we have an exact parallel to motion sentences. If the ‘path’ has a boundary (reaching hot or flat), the sentence is telic. If the ‘path’ is nonbounded (going on indefinitely in the ‘hotwise’ or ‘bigwise’ direction), the sentence is atelic. If the ‘path’ is ambiguously bounded (parallel to the case of cross), the sentence is of indeterminate telicity. The formal structure of these sentences is (40), paralleling (24); the scale for the property in question is laid out on the projection of the Property constituent, and the time interval and the course of the event are sp-bound to it:
200
measuring out, telicity, quantification
(40) 1d
Sit
·
1d
0d BEident ([
Thing
X], [
Property
·
1d
0d]); [
Time
·
0d]
It should be mentioned that a resultative sentence such as Bill hammered the metal flat means essentially ‘Bill made the metal become flat by hammering (on) it’ (Jackendoff 1990, chapter 10). This sentence is telic because its temporal structure is sp-bound to the Property scale that ends at flat. Notice that if the AP complement is inherently nonbounded, the sentence becomes atelic: Bill stoked the fire hotter and hotter and Bill hammered the metal flatter and flatter have no specified endpoint (although they seem awkward with for two hours). (See further discussion in Chapter 9.)
6.7.6 Verbs of performance Next consider verbs of performance such as read and sing. The entities being performed, such as books and tunes, are conceptualized as objects through which there is a specified linear path that the performer realizes over time. Thus there is an sp-binding between an axis that describes the linear course through the object and the axis that describes the temporal course of performance. What is the temporal cross-section that is projected onto these axes? There is no obvious primitive within the system of Jackendoff (1983, 1990) that can be used for this purpose; but, intuitively, at a particular moment in time the performer is involved with a particular part of the object. So let us introduce a tentative function PERFORM (X,Y), where X is the performer and Y the object being performed. Then the basic structure is something like (41):17
(41) 1d
·
1d
·
1d
·
Y
Sit
0d PERFORM ([
Thing
X], [
Thing
0d]); [
Time
0d]
17 It remains to be seen how the function PERFORM is decomposed into further primitives. Expressions like (i) suggest that the performance object is construed as determining a sort of path (or perhaps linear enclosure) along which the performer is located.
(i) We’re now in/up to chapter 7/the second act. We’ve gotten through three chapters/verses already.
However this eventually is elaborated, the structure of the sp-binding in (41) will be preserved. So nothing in the present account depends on the further decomposition of PERFORM.
measuring out, telicity, quantification
201
The sp-binding of the theme axis to the event and time axes explicitly encodes the fact that the performance object is an ‘incremental theme’. Again, because of the sp-binding, when a bounded object serves as the object of performance, the event automatically becomes telic (42a); when there is a nonbounded object, the event is atelic (42b). And again, if no object is specified, the same thing happens as with the path in verbs of motion: the performance object is by default construed as nonbounded, so the event is understood as atelic (42c): (42)
a. Bill read the paper/sang the tune in/??for five minutes. b. Bill read fiction/sang opera for/in five minutes. c. Bill read/sang for/in five minutes.
6.7.7 Verbs of creation and consumption Tenny observes that when an object being consumed (such as an apple being eaten) measures out the event, we construe the object much like a performance object, in terms of a path through it—except that at each point in time, the corresponding part of the object goes out of existence. Similarly with creation: corresponding to each point in time there is a part of the (eventual) object that comes into existence. However, there is an alternative interpretation of creation and consumption sentences in which the object as a whole comes gradually into existence or goes out of existence, for example in Star Trek ‘beaming’ or the gradual vanishing of the Cheshire Cat. Also, as we noted in section 6.3, it is not necessary for an object to be created or destroyed over time: one can gulp down a handful of popcorn all at once or have a new idea spring ‘fully formed’ into one’s mind. So the analysis of verbs of creation and consumption must allow a number of options that are distinguished only pragmatically. How shall these possibilities be encoded in the present framework? Jackendoff 1983 proposes a primitive BEEx as the conceptual structure predicate encoding existence; it has only two possible locations, in or out of existence. Coming into existence and going out of existence are inchoatives of this predicate; creating and destroying are the causatives of these inchoatives. (43) exhibits the inchoatives; because inchoatives are events, the time period must be sp-bound to the course of the event. (INCH may be further analyzed in the terms of Chapter 5, example (48).)
(43) a. X comes into existence at T: ·
[0/1d]
·
[0/1d]
INCH [BEEx (X, IN EXISTENCE)]; [T]
202
measuring out, telicity, quantification
(43) b. X goes out of existence at T: [0/1d]
·
· [0/1d]
INCH [BEEx (X, OUT OF EXISTENCE)]; [T] If the time period is taken as zero-dimensional, the event is regarded as a pointevent; the whole object appears or disappears instantaneously. However, if the time period is one-dimensional, two options become available. In one case, the theme is extended in one dimension and sp-bound to the other axes; this is Tenny’s case of eating the apple bite by bite.18 (44a) is the structure of coming into existence bit by bit (for example, the way a circle comes into existence when we draw it the usual way). In the other case, existence is treated like a scalar property: there is a gradual route from existence to non-existence or vice versa. This scale is then bound to time, giving us the reading in which the object as a whole appears gradually, shown in (44b):
(44) a. X comes into existence at T: · · 1d 1d X
1d T
·
INCH [BEEx ([0d], IN EXISTENCE)]; [0d] b. X as a whole comes into existence gradually over interval T: · · · 1d 1d 1d IN EXISTENCE T
INCH [BEEx(X, [0d])];
0d
The instantaneous reading plus (44a,b) are the desired three readings for creation and consumption verbs; these readings are differentiated only by pragmatics for verbs like appear and disappear and for causatives such as create and destroy. On the other hand, a verb like build seems to imply piece by piece creation, that is, its LCS specifies a causative of structure (44a); eat allows As Tenny observes, the linear axis of the object is a patent fiction here, as it corresponds only to the linear ordering of the parts created or consumed over time. I take it that a more rigorous account of this particular aspect of the problem is not essential to the general solution for measuring out. 18
measuring out, telicity, quantification
203
instantaneous destruction or piece-by-piece destruction, but gradual holistic destruction seems implausible. (Contrast with digest.)d Structure (44a) makes explicit the relationship discussed in sections 6.1 and 6.2 between parts of the event and parts of the created or destroyed object. In particular, if the object is nonbounded, the event is atelic, as in Bill ate custard for/in an hour; and if no direct object is present, a nonbounded argument is again assumed by default, as in Bill ate for/in an hour.
6.7.8 Conclusions The upshot of all these examples is that, by decomposing an event into a crosssection plus a series of sp-bound axes, the notion of measuring out is explicitly represented, and the measuring out relations between different parameters of the situation can be formally differentiated in terms of which axes are sp-bound. Moreover, a number of functions previously considered as distinct primitives can be reduced to the single function BE. It is important to point out that not all of the primitive situation-functions of Jackendoff (1983, 1990) can be so analyzed. For example, the function ORIENT (X,Y), which appears in verbs such as point and line up, denotes the orientation of the linear axis of X along the Path Y. ORIENT cannot be reduced to simple location. It specifies a relation between two axes, but the relation is not one of sp-binding. As a result, no parameter of this function measures out any other. This is why The sign pointed to New York and The sign pointed toward New York do not differ in telicity, despite the different boundedness in their paths. Similarly, the function that encodes a theme in contact with a location and exerting force on it establishes no sp-binding between the event and either the theme or the location. For example, verbs like chew, nibble, squeeze, and knead imply contact with and action on the Patient, but there is no dependence between the point of contact and the time of action: one can keep chewing at the same place indefinitely. Therefore there are none of the dependencies between telicity and the boundedness of the theme and path arguments that occur with motion verbs. It will be recalled that Krifka differentiates measuring-out verbs from nonmeasuring-out verbs simply by stipulating that certain theta-roles induce a ‘transfer of reference properties’. We can now be more specific. An argument with a particular theta-role measures out time just in case (a) the Situation of
d Typically, unless it matters for some reason, we don’t differentiate between instantaneous action, gradual action one part at a time, or gradual action globally over the whole object. Thus perhaps we don’t want to say these verbs are strictly speaking ambiguous, but rather vague.
204
measuring out, telicity, quantification
which it is an argument is an Event, so that the event and time axes are sp-bound, and (b) this argument projects an axis sp-bound to the event. However, there are other cases of measuring out that do not involve the time axis at all, as seen in section 6.7.3, and still others where the event and time are sp-bound but other arguments are not involved, as seen in section 6.7.2. We can also see that Verkuyl’s characterization of measuring-out verbs as [þADD-TO] is inadequate, in that it does not specify precisely which pairs of arguments are linked by sp-binding. In order to distinguish the different types of measuring out, one needs to say whether the theme, location, and/or time axes are involved. The present notation accomplishes this explicitly. Finally, we can differentiate cases in which measuring out is lexically stipulated by the verb from those in which it is a possibility dictated by the pragmatics of the situation. In the latter cases, optional axes are projected that are not part of the LCS of the verb. Still, once they are projected, they determine telicity in exactly the usual way. This is a natural outcome in a theory like Conceptual Semantics, which makes no sharp distinction in representation between semantics and pragmatics (see Jackendoff 1983, especially chapter 6); it is more problematic for those theories that wish to maintain a strong distinction.
6.8 Sp-binding on discrete axes; distributive quantification We noticed in section 6.2 that atelicity can be created by bare plurals as well as bare mass nouns, for instance (45)–(48): (45) Verbs of creation and consumption: a. Bill ate apples/drew circles for/in an hour. b. Bill ate three apples/drew three circles in/for an hour. (46) Verbs of performance a. Bill read poems/sang songs for/in ten minutes. b. Bill read three poems/sang three songs in/for ten minutes. (47) Verbs of covering a. Bill loaded trucks for/in an hour. b. Bill loaded three trucks in/for an hour. (48) Themes of verbs of motion a. Ants scurried past my door for/in an hour. People ran into the schoolroom for/in an hour. b. Five thousand ants scurried past my door in/for an hour. Five hundred people poured into the schoolroom in/for an hour.
measuring out, telicity, quantification
205
So far, our formalism for measuring out has involved sp-binding only among dense (or continuous) axes: the dense time axis, spatially dense paths, and mass terms as themes. We now must extend the formalism to the case of plurals, which divide up into discrete parts. For Krifka and Verkuyl, plurals represent the core cases of boundedness of NPs affecting telicity. Their treatment of dense axes is taken as an extension of the ‘snapshot’ approach, which presupposes a division into a finite number of discrete parts. In the present approach we will come from the other direction, generalizing from dense axes to axes consisting of discrete individuals. In the formalism of Chapter 5, this proves straightforward, since the difference between bare mass nouns and bare plurals is a single feature: bare mass nouns are [b, i] (nonbounded and having no necessary internal structure) and bare plurals are [b, þi] (nonbounded but having inherent division into discrete members). Otherwise, bare mass nouns and bare plurals are similar, both permitting specifiers such as some, all, a lot of, more, most, and so forth. (Numerals, of course, occur only with [þi] constituents.) Many different approaches to the mass/count distinction have arrived at a similar relationship between mass nouns and plurals (see Chapter 5 for references). Suppose we think of the ‘representative slice’ of Bill ate apples as the sample event of Bill’s eating a single apple. The apple is then projected onto an axis of discrete elements: there is one element on the axis for each apple that Bill ate. Since the number of apples is nonbounded (as indicated by the bare plural), the axis is a nonbounded one, just as in Bill ate custard. This axis of discrete elements may in turn be linked pragmatically to the event axis, along the lines of the pragmatic linking in (44a), thereby dividing the event axis into discrete apple-eatings, one per apple.19 And the time axis, being sp-bound to the event axis, is therefore divided into discrete time periods, one per apple. The result can be notated as (49), where instead of [1d] for the dimensionality, we use [PLUR] to express a discrete axis. (For typographical convenience I have suppressed all the internal details of eat as a verb of destruction.)
(49)
PLUR
Sit
19
·
PLUR
·
PLUR
·
EAT (BILL, [APPLE]); [Time 0d]
I say ‘may be’ rather than ‘is’ because this division is not essential. What is only occasionally noted in the literature is that Bill may be taking bites indiscriminately from random apples, in which case he is not eating one apple after another.
206
measuring out, telicity, quantification
In (49), the [PLUR] above APPLE comes from the morphological plural in apples. The other [PLUR] features are not morphologically marked. They are, however, clearly part of the meaning of the sentence: we understand there to be a multiplicity of eatings and a multiplicity of (not necessarily distinct) times at which these eatings took place. If the plural axis of apples is nonbounded in (49) (Bill ate apples), the time is nonbounded and the event is atelic; if the plural axis of apples is bounded (Bill ate 6 apples), the time too is bounded and the event is telic. This structure obviously can be applied to all of the cases in (45)–(48), projecting the plural in the appropriate constituent and sp-binding it to the event. Structure (49) finds clear antecedents, not only in Verkuyl (1993), but also in Fauconnier’s (1985) theory of ‘mental spaces’, in which he analyzes the plural as potentially creating a series of spaces, one per individual. The simultaneous projection of a sample individual, time period, and event creates Fauconnier’s series of spaces. The advantage of the present notation is that it explicitly schematizes the ‘sample event’ and does not require a definite number of ‘spaces’; just in case the plural axis is nonbounded the number of ‘spaces’ is nonbounded as well. The sharp-eyed reader will have noticed that (49) is well on its way to a novel formalization for scope of distributive quantification. (The relationship between measuring out by a discrete axis and distributive quantification is noted as well by Fauconnier, Verkuyl, and Krifka in their respective formalisms, as well as by Schein 1994, where a Davidsonian event variable comes to be bound by a quantifier.) Consider a standard example such as three boys lifted a box. On the ‘group’ reading, the three boys collectively lift a single box; there is one event of lifting at one particular time. On a second reading, each boy individually lifts the very same box; there are three events of lifting, but only one box. On the third reading, each boy lifts a box, but there is no supposition that the boxes are the same; there are thus three events of lifting, each of which has its own time period. (50) encodes these three possibilities (omitting the internal structure of lift, essentially ‘cause to go upward’):
(50)
a. Three boys lifted a box. Group reading (‘wide scope’ for box) [Sit LIFT ([3 + PLUR [BOY]], [BOX]); [T]] b. One lifting per boy, same box (‘wide scope’ for box) 3 + PLUR
Sit
LIFT
·
3 + PLUR
·
([BOY], [BOX]);
3 + PLUR
[T]
·
measuring out, telicity, quantification
207
c. One lifting per boy, possibly different boxes (‘narrow scope’ for box) 3 + PLUR
Sit
LIFT
·
3 + PLUR
([BOY],
·
3 + PLUR
[BOX]);
·
3 + PLUR
·
[T]
An immediate advantage of conceiving quantifiers this way is that the two ‘wide scope’ readings of a box can be clearly distinguished. One of these, the ‘group reading’, involves no projection of the plural over the event axis, so there is only a single event. The other ‘wide scope’ reading projects the plural over the event axis, but the event axis in turn does not project to the second NP, so although there are three events, there is only one box. Notice also that in the ‘narrow scope’ reading, box is connected to a plural axis even though there is no morphological plural: there are three (not necessarily distinct) boxes. Another immediate advantage is that no particular problem arises with quantifiers over mass nouns, such as All of the dirt fell into a truck. The projection of dirt, instead of being [PLUR], will be continuous, say [DIM 3d], and everything else will follow the rules of projection and sp-binding laid out in previous parts of this chapter. I will not attempt here to characterize the rules that produce the interpretations in (50b, c). These rules will be the present theory’s counterpart of a rule of Quantifier Raising (raising of the quantifer three); perhaps a better term in this approach would be Quantifier Spreading, with ‘spreading’ taken in the phonological sense. Such rules are very clearly within the ambit of Conceptual Structure; there appears to be little need to invoke a syntactic level such as LF. In particular, notice that the relation between (50a) and (50b) is very similar to the pragmatic application of ‘measuring out’ found with verbs of creation and consumption in section 6.7.7—a relation which, so far as I know, has never been proposed to be represented syntactically at LF. The treatment in (50) of how a quantifier binds a variable unifies this phenomenon with the treatment of measuring out. It will clearly take a great deal of work to recast traditional work on quantifiers in these terms, but certain nice points have already emerged.20 20
A speculation: This treatment of quantification suggests an evolutionary source for quantificational concepts. The traditional representation of quantification treats it as an operator that is independent of the system of thematic roles and the encoding of events. In order for the human mind to have come to be able to grasp quantification, then, this independent system must have developed as a complete innovation in the innate structure of concepts. The representation
208
measuring out, telicity, quantification
6.9 Multidimensional measuring out Section 6.7 omitted a further possible measuring out structure. Suppose we start with the same cross-section as that of a motion verb, but sp-bind the event and time axes to the axis of the theme instead of the path:
(51)
1d
·
1d
·
1d
·
0d ([Thing 0d], [Space L]); 0d BE Sit What kind of event is this? It would be one in which different parts of the theme are at Location L over a period of time. Examples might be the verbs flow, pour, and stream. (52) shows that the boundedness of the subject of flow affects the telicity of the event, as is characteristic of arguments that measure out: (52) a. Muddy water flowed over the dam for/in 2 hours. b. 20 million gallons of water flowed over the dam in/for two hours. Thus in this case, noticed by Verkuyl (1972, 101) and Declerck (1979), the theme of a motion verb measures out the Event—a situation not considered by Tenny, Dowty, or Krifka. (In fact this is a genuine case of the theme of a motion verb measuring out, in contrast to the spurious cases adduced by Tenny, in which the Path actually does the measuring out). However, the analysis of flowing is actually a bit more complex. Intuitively, the representative slice of the event described in (52) places us at the dam, watching some particular bit of water; at different times, different bits of water are present at this point. This is why the theme measures out the event. But we can also view the event from the point of view of a particular bit of water traveling over time, in which case flow acts as a normal motion verb whose path affects telicity: (53) a. The crest of the flood flowed along the levee for/in two hours. b. The crest of the flood flowed down to St Louis in/for two weeks.
suggested here, however, is a natural outgrowth of structure independently necessary for the concept of measuring out, a relationship involved in concepts as simple as motion verbs. Thus on the present view, the evolution of quantificational concepts requires a less radical innovation in the innate structure of thought—an advantage of a sort not generally considered in formal semantics, but one that carries a certain amount of weight in a theory that situates semantic competence in the mind of the language user.
measuring out, telicity, quantification
209
These two distinct points of view are possible because flow inherently describes the motion of a substance spread out linearly over space. Flow thus differs from verbs like roll and climb, whose themes are normally bounded objects and therefore can be schematized as points. How is the case of flow to be represented? Returning to the Marr theory of objects, consider an object with two axes such as the rectangular solid in (54). It can be seen in at least two different ways: as a rectangular vertical crosssection that is swept out horizontally by a horizontal axis (54a), or as a rectangular horizontal cross-section that is swept out vertically by the vertical axis (54b).
(54) a.
=
+
b.
+
=
This perspectival ambiguity can be encoded in conceptual structure by piling one axis on top of the other, the main (sweeping-out) axis on top, the crosssection on the bottom. The two analyses are of course extensionally equivalent. (55a,b) are the encodings of (54a,b) respectively:
(55) a.
1d VERTICAL
2d HORIZONTAL
b.
1d HORIZONTAL
2d VERTICAL
We will treat the two perspectives for flowing in a similar fashion. (51) represents a cross-section of a flowing event in which we choose a particular point and see successive parts of the theme go by. The full description of
210
measuring out, telicity, quantification
the event integrates this cross-section over space, adding a second layer of axes:
(56)
[1d]‚
[1d]‚ [1d]·
[1d]· 0d BE ([
0d], ([
Thing
Sit
Space
L]); ([
[1d]‚
(iii)
[1d]·
(ii)
0d]
(i)
Time
Line (i) is a bit of the theme at a particular location in space at a particular time. Line (ii) leaves us at the same place, watching bits of the theme pass over time. Line (iii) integrates over places, at each of which one can see bits of the theme pass by. Time also appears in line (iii), because the same sequence of bits of theme appears at different places at different times. The other perspective on the situation is shown in (57); it inverts the order of elaboration of axes:
(57)
[1d]·
[1d]·
[1d]‚ 0d BE ([ Sit
[1d]· (iii) [1d]‚
Thing
0d], ([
Space
[1d]‚ (ii)
L]); ([
Time
0d]
(i)
Line (i) is the same as before. But here line (ii) shows a bit of the theme moving along the path. Thus lines (i) and (ii) are identical to the standard motion schema. This cross-section is the one we see in sentences like (53). Line (iii) then integrates the motion schema over parts of the theme: each part of the theme is in motion. Again time is implicated in line (iii), because each part of the theme moves through a given space at a different time. The one thing that is curious about these structures is that the Time constituent is treated as having two independent axes. Of course, these have to be collapsed somehow onto a single Time axis. Intuitively, the situation is as depicted in (58), where each arrow depicts the motion of an individual bit of the substance in motion (the lower set of axes), and the
measuring out, telicity, quantification
211
vertical dimension depicts the relative position of successive bits (the upper set of axes): (58)
---------------> ---------------> ---------------> ---------------> ...
It is not clear to me how to formalize the collapsing of these two dimensions of Time into one, but since nothing hangs on it for the moment, I will not essay a solution.21 Verbs such as flow, pour, and stream (and perhaps leak) contain a multidimensional structure like (56)–(57) in their LCS. On the other hand, if an ordinary verb of motion is used with an appropriately linear theme, the motion can be pragmatically elaborated into a ‘flowing’ situation. This is what we see in examples like (59) (repeating in part examples from section 6.3): (59)
a. Fog rolled into the city for hours. b. The snake crossed the crack in the rock in five seconds. c. The soldiers passed my house in/for two hours.
All of these take the perspective illustrated in (52): the observer is fixed at a point, watching parts of the theme go by. But this perspective is possible only because of the pragmatics of the theme. So, as with the creation and consumption verbs, we have alternatives that arise pragmatically. (59c) presents a case where the theme is a plural rather than a continuous object. Its axis is therefore [PLUR] rather than [1d], following the treatment in the previous section. What if there are two such discrete axes? Then we are in a situation with multiple quantifiers, for example Three boys lifted four boxes. Because it is optional to project a plural axis, such a sentence offers many different interpretations. The minimal case is when three boys are involved in a single act of lifting four boxes (say, piled on top of each other); here no plural axes are projected. If we project three, but not four, we get two readings corresponding to (50b, c): (60)
a. Three events of lifting, one by each boy, each event being a lifting of the same four boxes piled up. b. Three events of lifting, one by each boy, each event being a lifting of four boxes, but not necessarily the same four in each case.
21 Part of Tenny’s (1994) Measuring Out Constraint (MOC), which forms the foundation of her revised theory, is the condition ‘There can be no more than one measuring-out for any event described by a verb.’ The case of flow appears to be a counterexample to this condition, since it involves two separate dimensions of measuring out.
212
measuring out, telicity, quantification
Similarly, if we project four but not three (more salient in the passive Four boxes were lifted by three boys), we get the readings in (61): (61) a. Four events of lifting, one for each box, each event being a collective lifting of the box by the same three boys. b. Four events of lifting, one for each box, each event being a collective lifting of the box by three boys, but not necessarily the same three in each case. In the maximal case there are twelve events of lifting. Corresponding to the relative scope of the quantifiers, there are two ways of projecting the axes to get this result:
(62) a. Twelve events of lifting, three for each box, each consisting of a lifting by a boy, none of the twelve boys necessarily the same. [4 + PLUR]·
[4 + PLUR]· [4 + PLUR]· [4 + PLUR]·
[3 + PLUR]‚
[3 + PLUR]‚
LIFT
([BOY],
[3 + PLUR]‚
[BOX]);
[T]
Sit
b. Twelve events of lifting, three for each boy, each consisting of a lifting by a boy, none of the twelve boys necessarily the same. [3 + PLUR]‚
[3 + PLUR]‚ [3 + PLUR]‚ [3 + PLUR]‚
[4 + PLUR]·
LIFT
[4 + PLUR]· [4 + PLUR]·
([BOY],
[BOX]);
[T]
Sit
The multiply quantified reading in (62a) is derived by projecting a second plural axis, for the boxes, above the structure in (50b). That is, the cross-section (or sample event) which is projected is expressed by the reading (50b) of three boys lifted a box. The reading in (62b) is derived from the corresponding reading of a boy lifted four boxes.22 A similar treatment of multiple quantifiers is found in Verkuyl (1993). However, I find Verkuyl overly concerned with all the pragmatic combinations possible, for example realizations of (62a) in which some of the boxes are the same, so there are fewer than twelve altogether. I would 22
measuring out, telicity, quantification
213
Like the optional projection of the theme axis found in motion verbs, the projections due to quantified plurals are all optional, which is why there is a range of possible readings. However, in the case of quantification, the interchanging of axes does not preserve meaning as in it does in the case of flowing. The reason for this difference between quantification and geometric measuring out is left for future research. (However, if the present analysis is correct, it helps explain why people have such trouble with multiple quantification: it does not behave the way a perspective shift should.)
6.10 More on spray/load verbse The spray/load verbs remain to be dealt with. These verbs are complex enough that I am not prepared to formalize them. However, I will deal with them informally and pick out some salient points. The spray/load verbs are distinguished from simple verbs of putting (i.e. causing an object to move to a goal) in two respects. First, they occur in two different syntactic frames: theme in/on location (63a) and location with theme (63b). (63)
a. Bill loaded dirt into the truck. Bill sprayed paint on the wall. b. Bill loaded the truck with dirt. Bill sprayed the wall with paint.
Second, these verbs specify some spatial distribution of the theme. As observed by Pinker 1989, they come in three subclasses, depending on exactly what sort of spatial distribution is stipulated. Verbs like spray and splash stipulate that the theme is spatially distributed while in motion, and in fact these verbs do not have to specify an end location. (64)
Spatial distribution while in motion: The water sprayed/splashed out of the bucket.
Verbs like smear and dab stipulate that the theme ends up spatially distributed over the final location. They also stipulate something about the manner in which the theme got there. For instance, smearing implies a particular sort of motion in applying the theme to the location. If the spray/splash verbs are used
consider such a proliferation of possibilities to indicate simple vagueness in the interpretation of the sentence, a position advocated also by Schein 1994. e The exposition of this section was unclear in the original, and I have taken the liberty of rewriting it for this edition.
214
measuring out, telicity, quantification
with a goal phrase, then the spraying or splashing also describes how the theme is applied to the goal (65b). (65) Spatial distribution of endpoint þ manner of application: a. Bill smeared/dabbed paint on the wall. b. Bill sprayed/splashed paint on the wall. A third class, with verbs like pack, cram, and load, stipulates that in the end state, the theme is constrained by the shape or capacity of the location. (66) a. Theme constrained by shape of location Bill packed/crammed cement into the crack. b. Theme constrained by capacity of location Bill loaded books onto the truck. (64)–(66) present these verbs in the ‘theme in/on location’ frame (63a). It is often claimed that the ‘location-with-theme’ syntax (63b) adds a further entailment, namely that the theme entirely covers or fills the location. However, (67) shows that the ‘location-with-theme’ syntax only favors the covered/filled reading, but does not force it, especially with the spray and smear classes. (67) a. Bill sprayed/splashed the wall with paint (for ten minutes), but it still wasn’t covered. b. Bill smeared/dabbed the wall with paint (for ten minutes), but it still wasn’t covered. c. ?Bill loaded the truck with dirt for an hour, but there was still room for more. d. ?Bill crammed/packed the crack with cement (for five minutes), but it still wasn’t full. A further degree of semantic freedom with many of these verbs is whether the parts of the theme must move sequentially to the location. Dabbing paint on the wall entails a sequence of individual dabs. But not all these verbs require sequential motion. As mentioned in section 6.3, one can load a truck by putting into it either many successive shovelfuls of dirt or just one big scoop of a steamshovel. Likewise, one can spray the wall with paint by covering different areas over time, but one can also spray one’s shoes with water in a single burst. Thus, for load and spray (though not for dab), it is up to pragmatics whether the action of the theme is spread out over time. This exactly parallels the pragmatic temporal patterning with verbs of creation and consumption (section 6.7.7). This pragmatic factor adds an extra dimensionality to the sentence; it sp-binds parts of the theme to time. But it is emphatically not part of the LCS of all spray/ load verbs.
measuring out, telicity, quantification
215
Keeping these distinctions in mind, let us examine Tenny’s (1992, 1994) analysis of telicity in the spray/load verbs. Tenny observes that telicity in sentences with load and spray verbs can be affected by the choice of syntactic frame, as in (68). (68)
Bill loaded dirt into the truck for/in an hour. Bill sprayed water on the wall for/in an hour. b. Bill loaded the truck with dirt in an hour. Bill sprayed the wall with water in an hour. c. ()Bill loaded the truck with dirt for an hour. ()Bill sprayed the wall with water for an hour. a.
In (68a), the nonbounded NPs dirt and water induce atelicity, just like the object in Bill ate apples for/in an hour. By contrast, (68b) is telic, so (according to Tenny) water and dirt no longer measure out the event. Rather, Tenny claims, the truck and the wall measure out the events, which are complete when the truck is full and the wall is covered. Thus, according to Tenny, the choice of NP in direct object position determines the temporal course of the event. However, Tenny’s analysis fails to take account of the full range of facts. The first problem is a disparity in judgments, indicated by the parenthesized stars in (68c). Tenny accepts the standard claim that these sentences are unambiguously telic and that the location ends up completely filled or covered. However, (67) shows that telicity and complete filling or covering are only a default, particularly with spray-type verbs. These semantic properties therefore cannot be entirely determined by the syntactic frame. Second, Tenny’s account predicts that in the ‘location-with-theme’ frame, the theme does not affect telicity, only the location does. But this prediction is false. In all the examples so far, the theme has been a bare mass or plural noun, which normally denotes a nonbounded quantity. But if the bare NP expressing the theme is changed to an unambiguously bounded quantity, the sentences become unambiguously telic: (69)
Bill loaded the truck with three tons of dirt in/for an hour. Bill sprayed the wall with thirty gallons of water in/for five minutes.
So, contrary to Tenny, the boundedness of the theme does affect telicity in the ‘location-with-theme’ construction: atelicity is possible only when the theme is nonbounded. In other words, the theme can always measure out the event, regardless of its syntactic position. Finally, the basic import of Tenny’s aspectual analysis is that syntactic argument structure affects telicity. But we have noticed that, with many spray/load
216
measuring out, telicity, quantification
verbs, the event can be understood as either instantaneous or spread out over time, depending on extraneous pragmatic factors. Yet the argument structure is identical in the two readings. This means that temporal structure is not entirely determined by syntactic frame. In short, the evidence on which Tenny bases her theory does not represent the full range of data, and examination of a broader range of facts undermines her analysis entirely. What is really going on? These cases present one significant difference from the cases discussed so far. Typically, when an argument is sp-bound to time, its boundedness determines telicity. Thus bounded NPs produce telic interpretations, and bare mass and plural NPs produce atelic interpretations. However, in the ‘location-with-theme’ frame (63b), a bare NP theme does not force atelicity, it only permits it. So the real question is how a bare NP theme, which usually forces atelicity, can permit a telic reading in this context. It appears that telicity is connected to the possibility of a covering/filling reading, which, as seen in section 6.7.3, involves sp-binding of the theme to the location. As it happens, all verbs of covering and filling—even statives, with no temporal structure at all—permit a bare NP theme, either in subject position (70a) or in a with-complement (70b): (70) a. Trees lined the road. Grass covered the field. Water filled the tank. b. The road was lined with trees. The field was covered with grass. The tank is filled with/full of water. And indeed, (68b) has a covering/filling reading and (68c) does not. The connection between a covering/filling reading and telicity with a bare NP complement suggests the following approach to telicity with spray/load verbs. (Something similar is suggested by Pinker 1989.) .
.
.
First, if a covering/filling relation is achieved over time, the event is necessarily telic, since there is a natural endpoint to the event. Second, spray/load verbs optionally specify the achievement of a covering/ filling relation in addition to their more specific character (spray–splash class vs. smear–dab class vs. cram–load class). Third, just in case a spray/load verb specifies the achievement of a covering/ filling relation, it takes on a covering/filling verb’s property of permitting a bare NP theme. Therefore, the covering/filling reading is telic even if the theme is a bare NP.
measuring out, telicity, quantification
217
Two independent factors seem to favor a covering/filling reading. .
.
First, verbs of the cram–load class have a bias toward such a reading, because the endstate of the theme is constrained by the shape or capacity of the container. By contrast, the spray–splash verbs and smear–dab verbs are more neutral. Second, the ‘location-with-theme’ construction creates a bias toward such a reading, while the ‘theme-on/in-location’ construction is more neutral.
Thus load the truck with dirt, where both biases appear, most heavily favors a covering/filling reading, and spray paint on the wall, where neither bias appears, is most neutral. What remains of Tenny’s account, then, is the bias toward covering/filling readings in the ‘location-with-theme’ construction. This is a genuine interaction between meaning and an argument structure alternation, and it does have something to do with measuring out. But it has nothing to do with temporal structure, and it is only a bias, not a determinative factor.23
6.11 Conclusions Let me summarize the main conclusions of the investigation. First, the telicity of an event is formalized in terms of the end-boundedness or nonboundedness of its event and time constituents. Second, an NP or PP argument affects the telicity of an event (or ‘measures it out’) either if the verb specifies an sp-binding between this argument and the event and time axes, or if the argument itself pragmatically projects an axis that is spbound to the event and time axes. Third, contrary to Tenny, measuring out is not directly related to affectedness, nor is it confined to the ‘direct argument’. Fourth, this account reduces a whole family of conceptual functions—GO, STAY, EXT, covering/filling, and flowing—into a cross-section containing a BE function, which is projected onto two or more sp-bound axes. It also provides an explicit account of which theta-roles are subject to measuring out, an issue which Krifka leaves inexplicit and which Verkuyl treats with the insufficiently structured diacritic [þADD-TO].
23 Thus the ‘location-with-theme’ construction has some of the character of a ‘constructional idiom’, parallel to the to-dative and for-dative constructions (Jackendoff 1990, chapter 9; Goldberg 1992). Compare also to the aspectual particles up and over discussed in Chapter 7, which also have effects on telicity.
218
measuring out, telicity, quantification
Fifth, certain cases of variation of telicity and measuring out are a consequence of pragmatic effects. Such cases include the verbs of creation and consumption and the spray/load class. Thus these verbs do not need to be lexically marked as measuring out verbs. One consequence of such pragmatic construals is that inferences involving telicity are produced by a combination of lexical semantics and nonlexical (or pragmatic) factors. Sixth, the mechanism of projection can be extended to account for atelicity with bare plurals and even for the structure of distributively quantified phrases, either count or mass.
Appendix: A more formal treatment of structure-preserving binding 6.A.1 Strong and weak sp-binding In the main text we notated sp-binding as if it were perfectly symmetrical, as in (A1): (A1)
[1d]a . . . [1d]a
However, as mentioned in section 6.5, such a formulation is inadequate for the general case. The course of a motion event, for instance, progresses from the beginning to the end. But the position of the theme on the path need not be a smooth progression: the theme may stop along the way or even back up. Therefore sp-binding cannot be symmetrical: we must distinguish the smoothly varying constituent (essentially the independent variable) from the potentially jerky one (the dependent variable). So a more adequate notation is that in (A2), where the constituent marked a is the independent variable and that marked f(a) is the dependent variable: (A2)
[1d]a . . . [1d]f(a)
On the other hand, the event structure and the temporal structure in a motion event go smoothly hand in hand; their relation is essentially an isomorphism. Thus we can distinguish strong sp-binding (or isomorphism) from weak sp-binding (the kind that allows backing up). (A3)
Structure-preserving binding (first approximation) Given two spaces X and Y of equivalent dimensionality. A function f is a strong sp-binding (isomorphism) of X into Y iff a. for every xi X there is a yi Y such that yi ¼ f(xi); b. for every yi Y there is an xi X such that yi ¼ f(xi); c. for all xi, xj, f(xi) 6¼ f(xj) , xi 6¼ xj.
measuring out, telicity, quantification (A4)
219
A function f is a weak sp-binding of X into Y iff (conditions a and b above) c. for all xi, xj, f(xi) 6¼ f(xj) ) xi 6¼ xj.
(A4) differs from (A3) only in that different points on X can map into the same point on Y. As a consequence, weak sp-binding allows motion along a path to include stops and backups. It also allows quantified indefinites such as the object in the wide scope interpretation of Three boys lifted a box to refer to anywhere from one to three distinct boxes. In the default interpretation, however, strong sp-binding is assumed: continuous forward motion and a different box for each boy. These definitions do not mention how many dimensions X and Y have. They are therefore adequate not only for the linear cases of measuring out, but also for the twoand three-dimensional cases of covering and filling. They also do not mention whether X and Y are continuous or discrete. They therefore suffice not only for measuring out with motion verbs, but for measuring out by plurals and quantification. For our purposes we must also stipulate that sp-binding preserves boundedness—that is, that a bounded domain X maps into a bounded domain Y, and that the boundaries of X map into corresponding boundaries of Y. Contrapositively, if Y is nonbounded, X must be as well. Under weak sp-binding, it is possible for a nonbounded X to map into a bounded Y, for instance in a (nonbounded) process of moving back and forth between two points. However, strong sp-binding excludes this possibility.
6.A.2 Continuity We next have to assure that, in a motion event, points on the time axis are not associated randomly with points on the path axis. We therefore need to make sure that sp-binding is continuous. This takes two steps: defining the density of the axes, and then defining the continuity of the mapping between them. For linguistic purposes we need a definition of continuity that does not invoke measurement, since we want it to apply also in property spaces (section 6.7.5), for example beauty and redness, over which there is no metric. Therefore we will define density of an axis in terms of betweenness. (A5)
An axis X is linearly ordered iff there is an ordering relation < on X such that for every distinct x1, x2 X, either x1 < x2, or x2 < x1.
(A6)
A linearly ordered axis X is dense iff for every x1, x2 X such that x1 < x2, there is an x3 such that x1 < x3 < x2.
220
measuring out, telicity, quantification
Intuitively, (A6) says that between any two points on a dense axis we can always find another point. To define continuity of a function from X into Y, we build on this idea: inbetween points of Y are mapped from in-between points of X.24 (A7)
Continuity of sp-binding Given two dense axes X and Y, sp-bound by a relation f. The relation f must be continuous, i.e. it must be the case that for any y1 ¼ f(x1) and y2 ¼ f(x2) such that [Case 1]: y1 < y2 and x1 < x2 [Case 2]: y1 < y2 and x1 > x2 there is a y3 and x3, y3 ¼ f(x3) such that [Case 1]: y1 < y3 < y2 and x1 < x3 < x2 [Case 2]: y1 < y3 < y2 and x1 > x3 > x2
6.A.3 Generalization of continuity to covering and filling The above definitions of continuity pertain only to the one-dimensional case. For covering and filling, we need to generalize these notions to two- and three-dimensional spaces. One possible way is this: (A8)
A 2- or 3-dimensional space X is dense iff any two points x1, x2 X determine a dense axis.
(A9)
Continuity of 2- and 3-dimensional sp-binding An sp-binding f from a dense 2- or 3-dimensional space X into a dense 2or 3-dimensional space Y must be continuous, i.e. it must be the case that f is continuous along the continuous axis determined by any two points x1, x2 X.
6.A.4 Definition of bounding on an axis The whole point of developing the notion of measuring out was to explain the relationship between the boundedness of a theme or path and the telicity of an event. It is therefore important to characterize boundedness more precisely. The idea is simple: an ordered axis is bounded if there is a point on it beyond which one can go no further. If there is no such point, the axis is nonbounded.
24 In strong sp-binding, any in-between point of X also maps to an in-between point of Y. But in weak sp-binding, this is not necessarily the case, since within the stipulated interval of X the value of the function can range outside the stipulated interval of Y and then return to it—consider a sinusoidal function, for instance.
measuring out, telicity, quantification (A10)
An interval on an ordered axis X is bounded above iff there is an x0 X such that for any x X, x < x0 or x ¼ x0.
(A11)
An interval is nonbounded above iff for any x X, there is an x’ X such that x < x’.
221
For (non)bounded below replace < with > in these definitions. Notice that (A11) applies either to discrete or dense ordered axes. If a nonbounded axis is discrete, it has to potentially go on forever. But if it is dense, all that is required is that it not reach a definite endpoint; the preposition toward, for example, creates such a nonbounded path of limited extent. This leaves the case of discrete axes with no essential ordering, as found for example in axes induced by quantification. In such cases we must presume an isomorphism between the elements of the axis and the natural numbers, this isomorphism inducing an ordering on the axis—in plain language, we count the elements of the axis. (In Verkuyl 1993, this is the core case, rather than one variation on the theme, as worked out here.) Once this ordering is in place, (A11) can be applied: the set is nonbounded just in case there is no definite last member.
Remarks on Chapters 7–13 The next six chapters concern the notion of constructions—pieces of syntactic structure that are associated with noncanonical meanings. Mainstream generative grammar, of course, takes the position that so-called constructions are an epiphenomenon. When one speaks of ‘the passive construction’, for instance, this is meant as just a convenient way of referring to a particular range of outputs of a maximally general syntactic component. However, in the middle 1980s, Charles Fillmore, Paul Kay, Arnold Zwicky, Adele Goldberg, Laura Michaelis, and others began exploring syntactically and semantically idiosyncratic phenomena that cannot be characterized in these terms. Fillmore, Kay, and O’Connor 1988, for instance, explored the peculiar syntax and semantics of let alone and briefly discussed a considerable number of other constructions such as the more . . . the more. They concluded that the grammar of a language must single these phenomena out as special meaningbearing bits of syntax, and that these supplement the general principles of phrase structure and semantic compositionality. An effort then began, under the banner of Construction Grammar, to reconstruct the whole grammar along these lines. Instead of treating rules of syntax as autonomous general principles of combination, the idea is to view much (and perhaps all) syntactic composition as a reflection of principles of semantic composition. Thus, in addition to syntactically and semantically idiosyncratic constructions such as let alone, the grammar also contains very general constructions such as the ‘transitive VP construction’, which associates the juxtaposition of a verb and its direct object position with the range of possible meaning relations between verbs and objects. I stumbled sort of sideways into this view of grammar in the course of my work on verb argument structure that eventually turned into Semantic Structures. The most striking example I came across was the way construction:1 (1)
1
a. Bill belched his way out of the restaurant. b. Joe joked his way into the meeting. c. Babe Ruth homered his way into the hearts of America.
Since the discovery of this construction in English, cognate constructions with approximately similar semantics have turned up in a number of other Germanic languages (Toivonen 2002a). Some of them use a cognate of way and some use a reflexive instead of his way.
remarks on chapters 7–13
223
Three things are remarkable about these sentences. First, although their VPs have canonical syntactic constituency (V–NP–PP), the complements are not licensed by the verbs, which are normally intransitive. Second, the curious phrase his way apparently occupies direct object position, as it is impossible to either add a direct object (2a) or substitute another direct object for his way (2b). (2) a. Joe told jokes his way into the meeting. b. Babe Ruth hit home runs into the hearts of America. Third, the meaning of these sentences involves the notion of motion, either literal (1a, b) or metaphorical (1c), where the PP expresses the path of motion. The verb is understood as a manner or means modifier of the motion, as shown in the approximate paraphrases in (3). (3) a. Bill came out of the restaurant, belching (all the while). b. Joe went into the meeting, joking (all the while). or Joe got into the meeting by joking. c. Babe Ruth entered the hearts of America by/through homering. Semantic Structures explored three possible approaches to this phenomenon. The first was to use mainstream techniques to derive (1) from an underlying structure that more directly reflects the semantic role of the verb. Essentially, the verb starts out in an adjunct, as in (3), then moves to the main verb position of the sentence. It was not hard to show that such a derivation runs up against considerable difficulty, and I abandoned it without regret. The second approach (later advocated by Levin and Rappaport Hovav 1995) was to treat the construction along the lines of the passive in LFG and HPSG: a rule in the lexicon converts intransitive verbs of the appropriate semantic sort into idioms of the form V one’s way PP, which then undergo lexical insertion and argument satisfaction. This can be made to work, but it has the effect of littering the lexicon with idioms such as belch one’s way and homer one’s way whose existence is entirely predictable. The third approach was to treat V one’s way PP itself as an idiom—a meaningful construction—that combines productively in syntax with intransitive verbs. From the point of view of description, this is a notational variant of the second approach, in that all the same conditions on the verb, the PP, and the interpretation obtain. However, this approach accounts for the fact that the rule is productive, the way syntax is supposed to be, and it is unnecessary to say the lexicon contains all the possible combinations of the idiom with verbs. On the other hand, the price is that one has to abandon the notion that the verb, as head of the verb phrase, is invariably also head in the interpretation—the element that determines the VP’s argument structure. Rather, in the way construction,
224
remarks on chapters 7–13
the construction itself determines the argument structure of the VP, and the verb functions semantically as a means or manner modifier. Semantic Structures, which discussed a number of such VP constructions, left the issue open of which of these two latter approaches is correct. However, at about the same time, Adele Goldberg had written a number of papers on a constructional approach to argument structure (eventually leading to her 1995 book), in which she dealt with some of the same constructions I had been addressing. I was persuaded by her work and other early work on constructions that this was an avenue worth pursuing. Chapters 7, 8, and 9 deal with VP constructions of this sort (another is the sound þ motion construction mentioned in section 1.5); Chapters 10, 11, and 12 deal with constructions that, like the let alone construction, have noncanonical syntax as well as noncanonical meaning. There is an important difference between my position on constructions and some versions of Construction Grammar, particularly those more closely allied with Cognitive Linguistics (e.g. Goldberg 1995, Croft 2001). In these approaches, every syntactic construction is taken to be a sort of Saussurean sign, signifying a meaning. There are no autonomous syntactic principles, free of semantic implications. (However, some versions of Construction Grammar such as Fillmore and Kay 1993 do allow independent syntactic principles.) My own view is that even if meaningful constructions are widespread, the grammar still needs purely syntactic principles. I base my judgment on two sorts of evidence. First, there are syntactic structures that do not support meaning distinctions but are nevertheless present to marshal the form. A clear instance would be a principle of invariable head position within a phrase: it makes no difference for meaning that the English verb is phrase-initial and the Japanese verb is phrase-final. Similarly, English do-support is present only to provide a bearer for tense, not to express a difference in meaning, and expletive it is present only to satisfy the requirement that a tensed clause have a subject. The other sort of evidence against the uniform meaningfulness of constructions is the semantic flexibility of some syntactic structures. One example is the N of NP frame (e.g. leg of a chair) discussed in Chapter 5, which can be used to express a large variety of semantic relations—many of which can also be expressed by NP’s N (a chair’s leg) and/or by a compound (chair leg). About the only general thing one can say about the meaning of N of NP is that the NP complement is semantically related somehow to the head noun. An enumeration of the detailed meanings fails to explain why all these meanings happen to converge on this form, as if it is no more than a massively homonymous word. Another take on the same argument comes from the English transitive verb construction, which likewise is semantically underdetermined until one chooses
remarks on chapters 7–13
225
a verb. One might say that the meaning of the V–NP configuration is that the NP’s meaning is an argument of the V’s meaning. However, this treatment would exclude the way construction, in which the NP is not an argument of the verb. Rather, the way construction is, as it were, parasitic on the syntax of the normal transitive verb construction, using the object for something other than its normal function. But it is impossible to express this parasitism without admitting that the V–NP configuration is set free from any particular semantic function. Chapter 7 develops a similar argument with respect to the English verb-particle construction. Finally, Chapter 13 discusses the semantics of English noun-noun compounds. It turns out that the range of semantic relations between the two nouns of a compound cannot be enumerated: it is best described by a generative system that creates in principle an unlimited number of different semantic relations. Thus it is impossible to say there is a specified collection of nounnoun compound constructions, each with a different meaning. Rather, it is necessary to stipulate the syntax and the semantics independently. The upshot, in my view, is that the grammar should permit pure principles of syntactic phrase structure, interacting with principles of argument structure. In turn, principles of argument structure should be divided into semantic argument structure (what characters are involved in an event and what their roles are) and syntactic argument structure (what phrases may appear with a given head, and what phrases must appear). Normally, semantic and syntactic argument structure are in fairly close alignment, but meaningful constructions provide a striking example of mismatch. (For extensive details, see Jackendoff 1990, 2002; Culicover and Jackendoff 2005.) This partial independence of components is characteristic of the Parallel Architecture. It makes it possible to incorporate the insights of Construction Grammar without committing to a lockstep correspondence between semantic and syntactic combinatoriality.
chapter 7
English Particle Constructions, the Lexicon, and the Autonomy of Syntax (2002)
7.1 The agenda This chapter examines particle constructions in English from the perspective of the following question, discussed at some length in Jackendoff 2002: What parts of an utterance can be constructed online, and what parts must be stored in long-term memory? In order to pursue this question, two important distinctions must be observed. First, it is important to distinguish the notion of lexical item from that of grammatical word. Lexical items are pieces of language stored in long-term memory (DiSciullo and Williams 1987 use the term listeme); grammatical words are grammatical units of a certain size, bigger than affixes and smaller than phrases. Since idioms like kick the bucket, for the most part, and Xi isn’t [This chapter appeared originally in N. Dehe´, R. Jackendoff, A. McIntye, and S. Urban, VerbParticle Explorations (2002), 67–94, which contains many different approaches and much crosslinguistic data on verb-particle constructions. It is reprinted here by permission of the publisher, Mouton de Gruyter.] I am grateful to Ida Toivonen, Silke Urban, Peter Culicover, Joan Maling, and Susan Olsen for discussions crucial to the development of this paper; to Hildy Dvorak, Amy Jackendoff, and Tom Chang for examples of X-out combinations; and to Marcel den Dikken and an anonymous reader for useful comments on an earlier version. This research was supported in part by a fantastic Fellowship from the Wissenschaftskolleg zu Berlin and in part by NIH Grant DC 03660 to Brandeis University.
english particle constructions
227
what proi’s cracked up to be must be stored, they are phrasal lexical items. (Drawing on a considerable concurring literature, Jackendoff 1997b shows that idioms cannot be treated as grammatical words and inserted/merged into syntax 0 under an X node.a) The second important distinction is between productive and semiproductive combinatorial principles in grammar. This is most easily illustrated in morphology. Two productive combinations in English are the present participle (be-ing, manufactur-ing) and the expletive infix (manu-fuckin-facturing, fan-fuckin-tastic). Productive affixes combine with any word that meets their selectional restrictions (semantic, syntactic, and, in cases such as the expletive, phonological), subject to morphological blocking by competing irregulars. Speakers can combine them with words they have never heard before, know exactly what they mean, and in the case of low-frequency examples have no idea whether they have actually heard them before. Thus these combinations need not be listed in the lexicon; they can be built up online by free combination, just like phrases. This requires, however, that we treat regular affixes as lexical items smaller than grammatical words; moreover, grammatical words formed by productive affixation need not be stored in the lexicon (though high-frequency ones may be).b Semiproductive combinations such as English irregular verbs (rang, drew, etc.) and -al nominals (refusal, recital) display different properties. The morphology
a It is sometimes argued that the meaning of an idiom like kick the bucket can be associated with the verb alone, as a special meaning of kick when in the context the bucket. The advantage of this treatment is that it permits meanings to be built up word by word, and, in mainstream generative grammar, that lexical insertion (or Merge) can proceed word by word. However, this proposal also requires special entries for the and bucket, namely with a null reading, when in the context of this special meaning of kick. In other words, this approach actually requires the entire structure of the idiom to be encoded, redundantly, in the lexical entries of all three words. This approach at least looks plausible when applied to a VP idiom, where it makes sense to invest its meaning in its syntactic head. It looks less attractive when dealing with idioms like day in day out, fit as a fiddle, down and dirty, and off the top of my head, where it is unclear how to choose which single word carries the content and which words have null meanings. My conclusion from this evidence (see also section 1.4) is that word-by-word lexical insertion was a technical mistake in the formulation of mainstream generative grammar. The alternative is to build sentences by Unification of lexical items with syntactic structure. This solves the problem of idioms immediately, as it is altogether natural to Unify a phrasal lexical item such as an idiom with a piece of syntactic structure. See Jackendoff 2002, chapter 5; 2008. b This is an advance on the view in Chapter 2, where regular affixes are mentioned in lexical redundancy rules but have no independent existence as lexical entries. The problem with such an approach is that it requires the lexicon to list all members of fully regular paradigms, for instance all English count nouns and their plurals, all English verbs and their present participles. From a purely theoretical point of view this might be unobjectionable. However, in the case of a language with massive regular morphology such as Turkish, the consequence is that the lexicon has to include tens of thousands of possible forms for every verb—a more difficult conclusion to swallow (Hankamer 1989). Moreover, the psycholinguistic evidence suggests that forms derived by regular morphology are not stored in memory unless they are relatively high frequency (Pinker 1999 and references therein).
228
english particle constr uc tions
may impose constraints or selectional restrictions, but within these constraints it is still necessary to list the combinations that exist (ring/rang but not sting/stang; refusal and recital but not infusal or incital). In the case of derivational morphology it is also necessary to know exactly what the derived forms mean: refusal is semantically regular but recital is not. Thus semiproductive combined forms must be listed in the lexicon, although their complexity may be reduced by some form of lexical redundancy rule that accounts for whatever degree of regularity exists [see Chapter 2]. The title of this chapter speaks of particle constructions in the plural because English has a large variety of constructions with different argument structures and semantic structures, all of which share the well-known syntax of verb þ particle. Some combinations of verb þ particle are productive, some are semiproductive, and some are purely idiosyncratic; the patterns interweave in complex fashion. Most of the present chapter will be devoted to documentation of this variety. At the end we will finally be in a position to ask what the syntax of the verbparticle construction is. We will conclude that any internal structure posited for the VP in order to account for one of the constructions is immediately thrown into question by one of the other constructions. Hence the best solution is that the VP is flat. This conclusion is in opposition to many of the papers in the present volume, but it concurs with Culicover’s (2000) arguments on other grounds for a flat VP in English (see also Culicover and Jackendoff 2005). More generally, Culicover (1999, 2000) has proposed the hypothesis of Concrete Minimalism, which supposes that the language learner posits the minimum syntactic structure necessary to relate sound and meaning. (This is restated in Culicover and Jackendoff 2005 as the Simpler Syntax Hypothesis.) This approach contrasts with the Minimalist Program, which attempts to minimize the derivational principles relating sound and meaning, at the expense of proliferating syntactic structure and derivational steps. The evidence presented here, that a relatively simple syntactic structure serves to express a broad range of semantic configurations, can be seen as an argument for the virtues of Concrete Minimalism/Simpler Syntax as well as for the traditional view of the autonomy of syntax.
7.2 Linguistics 101 facts about English particles To review the basic facts: English verb particles form a class that is for the most part homophonous with prepositions. When the verb is intransitive, the particle can serve as the only complement.
english particle constructions
229
(1) George grew up. The house blew up. Fred freaked out. When the verb is transitive, the particle can appear on either side of the object. (2) a. (Left-hand position) Pat put out the garbage. Bill blew up the house. Cathy brought in the cat. The explosion freaked out Fred. b. (Right-hand position) Pat put the garbage out. Bill blew the house up. Cathy brought the cat in. The explosion freaked Fred out. The left-hand position of the particle looks superficially like a preposition whose object is the following NP. However, it is easy to show that the particle and NP do not form a constituent. The standard contrasts are illustrated in (3). (3) a. PP complement: Bill ran [PP up the street]. Bill ran the street up. [no inversion of P and NP] It was up the street that Bill ran. [Cleft] Up which street did Bill run? [Wh-displacement with pied piping] b. Particle þ NP: Bill looked [Prt up] [NP the answer] Bill looked the answer up. [inversion of Prt and NP] It was up the answer that Bill looked. [no cleft] Up which answer did Bill look? [no pied piping] Traditional transformational grammar assumed that one of these positions is the underlying form and the other is derived by movement of the particle around the NP or the NP around the particle. Here we will be less concerned with whether one position is underlying than with the prior question of what the structures are, in both left-hand and right-hand position. But this will have to wait till section 7.8. Right-hand position is mandatory when the direct object is a nonstressed pronoun (4a); left-hand position is mandatory when the object is a sufficiently heavy NP (4b).
230 (4)
english particle constr uc tions a. Lila looked it up. Lila looked up it. b. Lila looked the answer to the question that was on everyone’s mind up. Lila looked up the answer to the question that was on everyone’s mind.
Some particles can bear a specifier; however this is possible only in right-hand position. (By the way, back and together are English particles that unlike most particles do not double as prepositions.) (5)
a. b. c. d. e. f. (6) a. b. c. d. e. f.
I’ll look the answer right up. Bill brought the wagon right back. Fran put the model airplane right together. Clem broke the handle plumb off. [dialectal] Please shut the gas completely off. We turned the situation partway around. I’ll look (right) up the answer. Bill brought (right) back the wagon. Fran put (right) together the model airplane. Clem broke (plumb) off the handle. Please shut (completely) off the gas. We turned (partway) around the situation.
(5e, f) are a bit tricky, since we also have (7a), with the particle and specifier in the opposite order, and (7b), where the particle is in left-hand position, leaving the specifier behind. (7)
a. She shut the gas off completely. He turned the situation around partway. b. She shut off the gas completely. He turned around the situation partway.
Hence completely and partway need not form a constituent with the particle. However, examples like (8a) show that the specifier plus particle in (5e, f) may form a constituent: they can be stranded together by ellipsis. By contrast, a particle followed by a specifier cannot be stranded (8b). (8)
a. She shut the gas completely off, not partly off. He turned the situation partway around, not totally around. b. She shut the gas off completely, not off partly. He turned the situation around partway, not around completely.
Ordinary manner adverbs substituting for completely and partly do not form a constituent with the particle:
english particle constructions
231
(9) He shut the gas slowly off (, not quickly off). He turned the situation laboriously around (, not easily around). The particle must precede any PP complements.1 (10)
a. Jill grew up into a strong woman. Please look out for Harry. Sim ran away to the city. The secretary sent a schedule out to the stockholders. b. Jill grew into a strong woman up. Please look for Harry out. Sim ran to the city away. The secretary sent a schedule to the stockholders out.
The string Prt þ PP in (10a) requires special care, as under some circumstances it forms a constituent. We discuss this problem in n. 4. A particle is sometimes possible in the complement of nominalized verbs. (11)
the rapid looking up of the information (is important) the prompt sending out of reports (is commendable)
The position of the particle in such nominals is always on the left side of the ofNP. Thus, although of-NP is the counterpart of the direct object in such nominals, it behaves as a PP with respect to particle position, just the way it looks. (12)
the rapid looking of the information up the prompt sending of the reports out
Finally, particles can take part in word formation. There are many deverbal nouns of the form [N V þ Prt], for example send-off, lookup, lookout, throwaway, shut-in, put-on, showoff. These seem always to be transparently related to the corresponding verb-particle combination, either as action or result nominals. There are also some deverbal nouns of the form [N Prt þ V], for example input, 1 We can see that the PPs in question are complements rather than adjuncts by applying the standard do-so test. Adjuncts can be stranded by do so (ia), but complements cannot (ib).
(i)
a. John ate his lunch on Thursday, and Fred did so on Friday. b. John put a banana on the table, and Fred did so on the chair.
The examples in (10) show the pattern of complements: (ii)
a. Jill grew up into a strong woman, and Jerry did so into a tall man. b. ??I looked out for Harry, and you did so for Sam. c. Sim ran away to the city, and Aaron did so to the country. d. The secretary sent a schedule out to the stockholders, and the boss did so to the administration.
232
english particle constr uc tions
output, outlook, downdrift (also upbringing). Some of these are semantically transparent, others are not.
7.3 Idiomatic verb-particle combinations English has hundreds of idiomatic verb-particle combinations. Examples are look up (‘search for and find’), blow (NP) up, throw up (‘vomit’), bring NP [e.g. a child] up, chew NP out, freak (NP) out, turn NP [e.g. a situation] around, turn NP on [two senses], turn NP [e.g. a paper, a criminal] in, clue NP in, and so forth. Because of their (at least partly) noncompositional meanings, there is no choice but to list them in the lexicon as complete units. However, recalling the discussion of section 7.1, this does not mean that they have to be listed in the lexicon as grammatical words.c It is often assumed (for example in Chomsky 1965) that the lexical structure of idiomatic verb-particle constructions is as in (12.1a), and the particle extraposes around the object to produce the form in (12.1b). (12.1)
a. [V [V look] [Prt up] ] b. look NP up
This option is chosen in order to preserve the idea that a lexical item is inserted 0 0 under an X category, in this case V . An underlying order like (12.1b), by contrast, would require a discontinuous lexical item. Emonds 1972, however, points out three facts that undermine this analysis. First, anticipating section 7.4, consider the particles in (12.2) that combine with motion verbs. (12.2)
a. Beth carried the food in/up/out/around. b. Beth carried in/up/out/around the food. c. Beth carried the food into the house/up the stairs/out the door/ around town.
The particles in (12.2a) satisfy a PP argument position, normally to the right of the object, as seen from the alternation with full PPs in (12.2c). It would therefore miss a generalization not to ‘base-generate’ the particles in that position as well; it is the position adjacent to the verb (12.2b) that is the marked position.
c For better expository continuity, most of the next four paragraphs has been imported from the original version of Chapter 8. Hence the slightly peculiar numbering.
english particle constructions
233
Second, many verb-particle combinations admit specifiers on the particle 0 (especially right). Since one of the hallmarks of X categories is that they do not admit internal specifiers, it is hard to imagine such cases as lexical verbs. Moreover, in such cases the modified particle must fall after the object, as seen in (6). Again the configuration with the particle adjacent to the verb appears to be the marked case. Emonds’s third argument concerns the basic assumption that it is preferable 0 to insert idiomatic verb-particle combinations as syntactic V s. He points out that there are numerous idioms consisting of a verb plus prepositional phrase, separated by an argument NP, for which no analysis as a grammatical word is feasible. Examples are take NP to task, give NP the slip, bring NPi to proi’s senses, sell NP down the river.2 Other than a rigid dogma that lexical insertion 0 is confined to X categories, there is absolutely no rationale for treating these as 0 lexical V s containing a PP that, coincidentally, must be forced to extrapose to a canonical argument position. But once these cases are admitted as discontinuous lexical items, there is no reason to hold verb-particle idioms to a higher standard. (More extensive discussion of this issue appears in Jackendoff 1997b, chapter 7; 2002; Culicover and Jackendoff 2005; see also Chapter 2.) In short, idiomaticity is evidence for lexical status, but not for grammatical wordhood or even constituency. One curious subclass of idiomatic verb-particle combinations appears to have become semiproductive in the past thirty years. In this class the ‘verb’ need not be a verb or even an independently attested word. They all mean roughly ‘go into an unusual mental state’; the particle is always out. The class is semiproductive because, despite the parallels in meaning, each example (and its exact meaning) must be learned individually, hence lexically listed. (The judgments of relative recency are only my best guesses; and speakers may differ on their judgments of transitivity.) (13)
a. Intransitives [older examples] pass out, black out, conk out, fink out, crap out, chicken out; [more recent] flake out, zonk out, zone out, bliss out, flip out, space out, phase out, crump out, veg out, chill out3 b. Transitives [older] knock NP out, bum NP out, fake NP out; [more recent] gross NP out, weird NP out, creep NP out
2 See section 7.8 for discussion of the possibility that take and to task do form a constituent in underlying structure, in a Larsonian shell (Larson 1988). 3 Veg and chill have still more recently become possible without out, in my children’s dialect but not mine.
234
english particle constr uc tions c. Intransitive or transitive [older] burn (NP) out, poop (NP) out, tucker (NP) out; [more recent] freak (NP) out, wig (NP) out, stress (NP) out, mellow (NP) out
7.4 Directional particles The best-known class of verb-particle combinations aside from idioms was discussed as early as Emonds 1972 and Jackendoff 1973, based in part on still earlier proposals by Klima 1965. Verbs such as toss, take, put, and carry select a directional (Path) PP as an argument, as seen in (14a). Such PPs can be replaced by a particle (14b), which, if it lacks a specifier, can appear before the direct object just like an idiomatic particle (14c). Of course a full PP complement cannot occur to the left of the object (14d), unless the direct object is heavy. (14) a. Beth {tossed/took/put/carried} the food (right) {up the stairs/into the house}. b. Beth {tossed/took/put/carried} the food (right) {up/in/away/back}. c. Beth {tossed/took/put/carried} (right) {up/in/away/back} the food. d. Beth tossed/took/put/carried up the stairs the food. Here the particle satisfies one of the verb’s argument positions. Any verb that selects a directional PP can take any directional particle instead, and the meaning is fully compositional. Hence such combinations are fully productive, and there is no need to list the verb-particle combinations in the lexicon. Rather the particles stand alone as lexical items. Directional particles can appear in the ‘locative inversion’ construction (15a), just like ordinary PPs (15b). (15) a. Up marched the sergeant. Back hopped the frog. Down went the soldiers. Out goes the garbage. b. Up the street marched the sergeant. Into the room hopped the frog. Down the hole went the soldiers. Out the window goes the garbage. They can also appear in a curious exclamative sentence type, PP with NP (Jackendoff 1973), just like ordinary directional PPs.
english particle constructions (16)
235
a. Off with their heads! Down with the king! Out with this garbage! b. Off the shelves with these books! Down the hatch with this wine! Out the window with this garbage!
Of course idiomatic particles lack the appropriate directional meaning, so they cannot appear in ‘locative inversion’ (17a). And they are meaningless without their verb, so they cannot appear in the verbless with-exclamative (17b). (17)
a. Up blew the building. b. Up with your lunch! [in the sense of ‘throw up’, ‘blow up’, etc.]
Klima concludes that the directional particles in (14)–(16) are intransitive prepositions, which explains why so many are homophonous with ordinary prepositions and why they occur in PP positions in (15)–(16). However, the grammar must still specify in addition (a) that PPs headed by a particle precede all other PPs in VP, and (b) that particles without specifiers may precede the direct object. In other words, even if particles are just a kind of preposition, the grammar must take account of their special properties.4 4
Directional particles can be followed by a PP in the locative inversion and with-exclamative constructions, indicating that the particle forms a constituent with the PP. (i)
a.
b.
Back to Hollywood comes that star of stage and screen, Groucho Marx! Up to the soldiers marched the sergeant. Out from the darkness crept a lizard. Off to the police station with these miscreants! Up against the wall with those guys! Down into the hole with the jewels!
In such larger constructions, the P might be head and the particle its specifier, or the particle might be the head and the PP its complement or adjunct. There is evidence that both these structures occur. First consider ordinary preposed locatives with stative verbs. A number of particles occur along with locative expressions in this construction (ii). But they cannot occur alone here, as they can in (15a) and (16a). (ii)
Off/over (in that room), John usually sits for lunch. Down/back (under the ocean), the temperature is pretty constant. Up/out (on Newbury Street), there are lots of great stores. Back (in 1982), we didn’t know a lot about HIV. Off (in the future), we’ll be able to understand these problems.
This suggests that selection is based on the prepositions in (ii), hence that the particles are specifiers. Other cases seem to force a different structure. For instance, the directional prepositions from and to are somewhat uncomfortable in the locative inversion construction, unless preceded by a particle. [continued]
236
english particle constr uc tions
7.5 Aspectual particles A less frequently distinguished use of particles is as aspectual markers. Consider the particle up in the following context. (18) a. b. c. d.
Elena drank the milk (completely) up. Ben glued the chair (right) up. Aaron wiped the counter (right) up. Hilary packed the suitcase (right) up.
This means roughly ‘V NP completely’, i.e. up is not directional as it is in toss the ball up.5 Nor does up satisfy an argument position of the verb: it can be freely omitted. It is often even redundant, as in close up the suitcase. And it does not form an idiomatic combination with the verb: a huge number of verbs can co-occur with it, for instance eat, drink, guzzle, close, finish, clean, wash, boil, fry, bake, fill, cover, pack. [2010: The meaning is fully predictable: the use of up in these circumstances emphasizes the telicity of the event being described.] Thus this ‘aspectual up’ should be listed as an independent lexical item, free to combine with verbs that meet its selectional restrictions. Despite its different semantic and argument structure properties, aspectual up is syntactically indistinguishable from other particles. It can occur in lefthand as well as right-hand position, but only without a specifier: (19) a. b. c. d.
Elena drank (completely) up the milk. Ben glued (right) up the chair. Aaron wiped (right) up the counter. Hilary packed (right) up the suitcase.
An unstressed pronoun can occur only to the left of up (20a); a PP complement must follow up (20b). (20) a. Elena drank it up/drank up it. b. You dissolve the protein up in water/in water up.
(iii)
(Back) from Hollywood comes that star of stage and screen, Groucho! (Up) to the soldiers marched the sergeant. (Down/out) from/to the darkness crept the lizard.
This suggests that the particles rather than the prepositions in (iii) are doing the work of licensing locative inversion; hence on standard assumptions the particle is the head. 5 Note that (18c) is different from Aaron wiped the food up, where up alternates with a directional PP as in Aaron wiped the food off the window. Up in (18c) does not alternate with a PP, and is more or less just a gratuitous insertion. For discussion of this alternation with wipe, see Levin 1993, section 2.3.3.
english particle constructions
237
Two more such aspectual particles are away (21a) and on (21b).6 (21)
a. Bill slept/waltzed/drank/talked/read/sneezed away. b. Bill ran/sang/talked/worked/thought/wrote on.
These mean roughly ‘Bill kept on V-ing’; i.e. away is not directional as in run away, and on is definitely not locational. And again these are not arguments of the verb, since they can be omitted; and they are not idiomatic combinations with the verb, since any verb of a large semantic class can co-occur with them. The meaning is fully compositional: these particles emphasize the atelicity of the event in question. Away and on have the curious syntactic property of forbidding the verb from licensing a direct object. (22)
Dave drank scotch away/on. Dave danced waltzes away/on. Dave read newspapers away/on.
Hence we cannot fully test their membership in the normal set of particles as we could with aspectual up. Moreover, aspectual away does not admit specifiers: Bill slept/waltzed completely away is bad, and Bill slept/waltzed right away is good only on the unrelated reading of right away as ‘immediately’. On the other hand, aspectual on does admit right as a specifier: Dave danced right on. Aspectual away is like particles in that it allows some prepositional complements of the verb to follow it (23a–c), though not all of them (23d). In particular, all ‘conative’ at-PPs such as (23e) seem eminently acceptable. In fact, some verbs such as carve are happy in the ‘conative’ frame only if away is present (23f), a fact not noted in Levin’s (1993) otherwise exhaustive compilation. (23) a. Hilary talked away about her latest project. b. Miriam worked away on her manuscript. c. Sally was screaming away at me. d. Judah jumped away (off the roof).7 e. Billy bashed away at the piano. f. Simmy was carving (away) at the roast.
6
I am indebted to Lombardi 1984 for various details of data on aspectual away; Toivonen 2006 presents far more detailed analysis of aspectual on. 7 Without off the roof, this has, in addition to the desired iterative reading, an inceptive reading along the lines of ‘gather up one’s nerve and V’, as in Jump away! I haven’t looked at this reading of away.
238
english particle constr uc tions
Aspectual on seems to admit a different subclass of PP complements. I see no pattern among the differences. (24) a. b. c. d. e. f.
Hilary talked on about her latest project. ?Miriam worked on on her manuscript. ?Sally was screaming on at me. Judah jumped on off the roof. ?Billy bashed on at the piano. Simmy was carving on at the roast.
Also like particles, aspectual away and on cannot come after a PP complement. This contrasts with right away (‘immediately’), which is not a particle at all.
Hilary talked about her latest project away/on. Miriam worked on her manuscript away/on. b. Hilary talked about her latest project right away. Miriam worked on her manuscript right away.
(25) a.
Two more candidates for aspectual particles are through and over, though their behavior is somewhat more marginal. (26) a. read/scan/skim the book through b. sing/play the aria through c. work/talk/think the problem through (27) a. b. c. d.
cook the food over sing/play the aria over write/type the paper over do the problem over
Like aspectual up, away, and on, the particles through and over in (26)–(27) do not satisfy an argument position of the verb; rather they too contribute an aspectual sense, often paraphrased by some sort of adjunct PP. Read/scan the book through means ‘read/scan the book from beginning to end’; cook the food over means ‘cook the food again’ or ‘re-cook the food’. Aspectual through and over have the opposite restriction from aspectual away: they require the VP to be transitive. (28) a. Bill read/scanned through.8 b. Let’s cook/sing over. These particles are a good deal less comfortable in left-hand position than is up. 8
This is possibly good on another reading, an ellipsis from Bill scanned [through the book].
english particle constructions (29)
a. read/skim through the book b. sing/play through the aria c. talk/work through the problem
(30)
a. b. c. d.
239
cook over the food sing/play over the aria write/type over the paper do over the problem
In (29a, b) we have a strong sense that through NP is a phrase. The tests for phrasehood such as Through which book did she read? confirm this sense. I can find no evidence that through also exists separately as a particle here. (29c) on the other hand yields the opposite result: ??Through which problem did she work? So through seems to be a genuine particle here. Similarly, (30a–c) show no possibility for the particle; but Over which problem did she do? is terrible, so (30d) does seem to contain the particle. In the two cases where left-hand position is acceptable, through and over exclude a specifier and a pronoun object, just like other particles. (31)
a. b. c. d. e. f.
talk the problem completely through talk completely through the problem talk through it [ungrammatical on this reading] do the problem completely over do completely over the problem do over it
Next consider the range of verbs possible with aspectual particles. As observed above, up, away, and on seem altogether productive within the selectional restrictions of each particular particle.9 This means that the possible combinations should not be listed in the lexicon: polka on, epoxy NP up, and many examples above are hardly memorized fixed expressions. They should be able to arise in syntax from a process of free combination parallel to the licensing of free aspectual adjuncts such as from one end to the other, some more, and again. However, examples with through and over seem more limited, and some combinations such as see NP (e.g. the process) through are idiomatic. This 9 Peter Culicover has pointed out (p.c.) that the productivity of aspectual particles seems however to be constrained by a phonological condition: the verb must be (approximately) monosyllabic. For example, we get read the paper through/over and read on but not peruse the paper through/over and peruse on; we get fix my lawnmower up but not repair my lawnmower up. This restriction appears to be related to the better-known restriction on ‘dative-shift’ verbs that permits give Bill the money but not donate the library some books; it is also found in idiomatic verbparticle combinations. However, it does not seem to apply to the constructions to be discussed in the next section: see (32c) for instance.
240
english particle constr uc tions
suggests that combinations with aspectual through and over may be semiproductive, along the lines of the examples with out in (13), hence lexically listed. Moreover, some semantically transparent combinations, for instance finish up, might be listed anyway in the lexicon as relatively high-frequency fixed expressions, sort of like happy birthday. What we find in the aspectual particles, then, is a complex mix of productive and semiproductive combinations, and in some cases further constraints on the structure of the VP. These particles differ from the idiomatic and directional particles in that they function semantically like aspectual modifiers, rather than like arguments of the verb. Nevertheless, their syntax is otherwise indistinguishable from that of the other particles.
7.6 The Time-Away Construction A fourth use of English particles occurs in a construction exemplified in (32), which I’ll call the ‘time-away construction’.d (32) a. Bill slept the afternoon away. b. We’re twistin’ the night away. c. Pat programmed three whole weeks away. These examples contain an intransitive verb, followed by an apparently unlicensed NP expressing a period of time, plus the particle away. As seen from (32), a wide range of verbs is possible in the construction—provided they have an intransitive subcategorization. No postverbal NP may be present other than the time phrase (33), nor may the verb be one that requires a direct object (34). (33) a. Frank drank the night away. Frank drank scotch the night away. b. Ann read the morning away. Ann read the newspaper the morning away. (34) a. Frank devoured the night away. (Frank devoured) b. Ann perused the morning away. (Ann perused) The particle away behaves just like those in other verb-particle constructions, in that it can occur to the left of the NP, especially if the NP is a little longer (35a, b), and it must occur to the right of a definite pronoun (35c).
d The time-away construction is discussed in detail in Chapter 8, so some of the original exposition in this section has been abbreviated.
english particle constructions (35)
241
a. Stan drank away the entire afternoon of his 50th birthday. b. Stan fished away all of Tuesday morning. c. What happened to the morning? I must have slept it away/I must have slept away it.
The particle can also have a specifier, but only in right-hand position: (36)
a. Dan slept the long afternoon entirely away. b. Dan slept entirely away the long afternoon.
Away is the only particle possible in this construction. On general semantic grounds, one might perhaps expect something like sleep the afternoon up, parallel to drink the milk up, but it is impossible. To show that this construction has the syntax of ordinary transitive verbparticle combinations, it is important to establish that even though the time NP is not licensed by the verb, it is nonetheless in direct object position. In particular, it must be distinguished from NP time adjuncts such as those in (37). (37)
Fred hasn’t slept this year. Kate is leaving Monday.
This task is put off to section 8.1. The evidence points to the time expression in the time-away construction being in direct object position; as it were, it usurps this position so that the verb itself cannot license an NP there. Moreover, away is a particle, indistinguishable in syntax from an ordinary verb particle. Turning to the semantics of this construction, it can be paraphrased to a first approximation by a sentence in which the time expression is within a delimiting for-phrase. (44)
Bill slept the afternoon away Bill slept for the (whole) afternoon.
A more subtle semantic aspect of the construction is that the subject is in some sense understood as ‘using’ the time, or even better, ‘heedlessly using the time up’. Some of this flavor appears in the second-approximation paraphrases in (45). The verbs in (45b, c) even duplicate the syntax of the construction: the verbs fritter, while, and the relevant sense of piss require the particle away. These three combinations are of course idiomatic, and therefore must be lexically listed. In fact fritter and while appear only in combination with away, so have a status similar to the cranberry morphs zonk, veg, and crump in (13). (45)
a. Bill spent/wasted the afternoon sleeping. b. Bill frittered/pissed the evening away sleeping. c. I could while away the hours conferring with the flowers.
242
english particle constr uc tions
Further semantic subtleties of the construction are discussed in Chapter 8. In brief, the construction has a complex and rich semantics, approached but not exactly captured by any paraphrase. Again we can ask what is listed in the lexicon. Given that the choices of verb and time NP are completely free within the construction’s selectional restrictions, Chapter 8 proposes that this construction is a lexically listed meaningful construction: a transitive VP with an open verb and object but a specified particle, as in (46a); the semantic structure of this construction is roughly (46b). (46) a. [VP V NP [Prt away] ] b. ‘spend/waste [Time NP] heedlessly V-ing’ What is unusual about this structure-to-meaning pairing is that the verb, although the syntactic head of the VP, is not the semantic head of the meaning. Rather it is embedded as a manner modifier. Moreover, the NP is licensed by the construction rather than by the verb; that is, it is the direct object of the VP but not of the verb. Such a solution is unusual in standard generative grammar, but it is quite natural within a framework such as Construction Grammar (Goldberg 1995; see also Jackendoff 1990, chapter 11, and brief discussion in the remarks preceding the present chapter). The main point for our purposes, though, is that despite the differences in the way the particle combines into the meaning of the VP as a whole, and despite the unusual argument structure licensing conditions, this construction has syntactic properties virtually identical to other verb þ particle constructions in English.
7.7 Idioms consisting of particles plus something other than the Verb 7.7.1 V-d out (13) illustrated a semiproductive family of ‘verb’-particle idioms with out, much of it of recent provenance. Another apparently recent class involving out is totally productive. For example, if I have been knitting or programming for the last six hours straight, I may say (47a). I may conceivably also say (47b), an idiomatic use of the resultative with a so-called ‘fake reflexive’. (47) a. I’m (all) knitted/programmed out. b. I(’ve) knitted/programmed myself out.
english particle constructions
243
But this odd combination is not confined to verbs. If I’ve drunk fourteen cups of coffee in the course of a morning I might utter (48a), and if I’ve watched fourteen Edward G. Robinson movies in a row I might even utter (48b). My daughter, after working at the Olympics, uttered (48c). (48)
a. I’m (all) coffeed out. b. I’m Edward G. Robinsoned out. c. I’m Olympic’d out.
I don’t think that fake reflexive counterparts of (48) are acceptable: I’ve coffeed myself out. Thus the productive extension of this use of out to nouns is confined to the non-fake reflexive form with a past participle. What would seem to be stored in the lexicon for this case is again an idiomatic construction, roughly of syntactic form (49a) and meaning (49b). (49)
a. [AP V/N þ -d [Prt out] ] b. ‘worn out from too much V-ing/too much N’
The model for this construction is rather clear: it is an extension of tired out, worn out, burned out, and so forth. What I find most striking about it is its productivity: one does not need to learn which verbs and nouns can be substituted into it. So the question is: how does the language learner distinguish between this case and the altogether similar class in (13), which remains only semiproductive? The parallel to Pinker’s (1999) word/rule dichotomy ought to be obvious; I leave the resolution for future research.
7.7.2 His head off A final type of idiomatic particle combination, brought to my attention by Jay Keyser, is illustrated in (49.1).e These resemble idioms such as cry one’s eyes out, which are generally taken to be based on the resultative construction. (49.1)
a. Fred talked his head/his ass/his butt off, but to no avail. b. Every night I sit here and sing my heart out; but does anyone listen to me? No! c. The chef was cooking up a storm back in the kitchen.
Unlike cry pro’s eyes out, these permit a variety of intransitive verbs, much like the time-away and way constructions.
e The rest of this section has been supplemented with material imported from the original version of Chapter 8.
244
english particle constr uc tions
(50) a. Susan worked/swam/danced/cooked her head off last night. b. Sam whistled/dreamt/jogged/programmed/yelled his heart out. c. Harry sang/argued/painted/wrote/slept/drew/edited up a storm. Here the construction is strictly fixed in form, so we cannot do any of the usual syntactic tests for particles. In particular, the order of particle and NP cannot be reversed: (51) a. Richard cooked off his head. [OK only with radically different meaning] b. Harold sang out his heart. c. Kelly edited a storm up. Like the aspectual particles away and on and the time-away construction (cf. (22) and (33)), these constructions exclude normal NP objects: (52) a. Harold sang arias his heart out. b. Richard cooked lentils his head off. c. Kelly wrote letters up a storm. And manner adverbs cannot occur before the NP: (53) a. Harold sang happily his heart out. b. Richard cooked busily his butt off. c. Kelly slept peacefully up a storm. Kelly slept up peacefully a storm. Thus the italicized NPs in (49.1), like those in previous constructions, seem to occupy direct object position. Aside from the intransitivity restriction, the choice of verb seems totally open, whereas by contrast, the choice of NP is totally fixed. This suggests that we are dealing with idioms of the structure (54): (54) a. [VP V [NP pro’s heart] [Prt out] ] b. [VP V [NP pro’s head] [Prt off] ] c. [VP V [Prt up] [NP a storm] ] Although the particles out, off, and up are normally associated with telic aspect (see section 7.5), the expressions in (49.1) are atelic. (54.1)
a. Sue worked her butt off for/in an hour. b. The frog sang his heart out for the whole night/in a night. c. Harry cooked up a storm for/in an hour.
So these expressions cannot mean, even idiomatically, what a resultative interpretation would predict: (49.1a), for example, does not mean anything close to
english particle constructions
245
Fred made his head come off by talking, which is telic. Rather, in each case the NP þ Particle combination carries a sort of adverbial force, denoting intense and perhaps passionate activity; each combination also carries its own affective tone, which I will not try to characterize here. In fact, cry pro’s eyes out is also atelic (he cried his eyes out for/in an hour), so it is probably a specialized instance of this construction rather than a resultative, as generally assumed. These cases thus are a mixture of the phenomena we have observed previously. Like aspectual particles, they are not arguments of the verb, and they have a sort of adverbial meaning. But like the time-away construction, they are syntactically complex, being a VP with a specified complement but a free choice of V. In particular, they are not syntactic adjuncts.
7.8 The syntactic autonomy of the particle construction To summarize: we have enumerated the following classes of verb-particle constructions. (55)
a. Verb-particle idioms: V and Prt listed, potentially discontinuously Includes semiproductive subclasses, such as zone out type, where ‘verb’ may be of any category, even a cranberry morph b. Directional particles Prt is an independent lexical item that combines productively with directional verbs, in alternation with directional PPs c. Aspectual particles i. up, away, and on are independent lexical items that combine productively with verbs; meaning parallels aspectual adverbials ii. V over and V through are possibly semiproductive combinations listed individually d. Time-away construction Prt is part of a lexically listed VP that takes a free verb and a free time expression as arguments Includes subclass of idioms with cranberry morphs such as fritter and while e. V/N-d out construction Prt is part of a lexically listed A(?)P that takes a free verb or noun as argument f. his head off family of constructions NP and Prt listed together as a lexical VP that takes a free choice of verb as argument; meaning parallels degree adverbials
246
english particle constr uc tions
In each case except some of the aspectual particles and the his head off family, the particle can occur on either side of an object NP, but must occur before any full PP. An object pronoun is possible only with the particle in right-hand position. When a specifier is possible, it occurs only in right-hand position. Finally, these varieties of particles are in complementary distribution; a single VP cannot have more than one of them. Thus insofar as the syntactic tests permit us to tell, the particles in all of these constructions have identical syntax. This constitutes the classical sort of evidence for the autonomy of syntax: English assigns particular syntactic positions and syntactic properties to particles, no matter how their presence is licensed. Let us consider the consequences for the structure of English VPs that contain particles. (56) lists the relevant questions. (56) a. Do the verb and the particle together form a lexical item? b. Do the verb and the particle form a constituent that excludes the direct object? c. Do Prt and NP form a constituent that excludes the verb? Beginning with (56a): The verb and the particle together form a lexical item only in verb-particle idioms and other semiproductive combinations such as (55c.ii). In the his head off family, the combination of the particle and the NP is lexically listed. Otherwise, the particle is a lexical item on its own, in some cases carrying some extra syntactic structure with it. Turning to (56b): When the particle is in left-hand position, it is sometimes taken to be ‘incorporated’ into the verb in a structure [V V Prt] (e.g. Toivonen 2002b). The main argument for such a structure is that in this position the particle cannot project specifiers and complements; this is what would be expected if it were a quasi-morphological affix. However, such a structure also usually carries the expectation that the verb and particle form a semantic unit. This expectation is confirmed for idiomatic verb-particle combinations, but not for aspectual particles and especially the time-away and up a storm constructions. Thus, although [V V Prt] may be the correct structure on syntactic grounds, it requires one to abandon the semantic implications traditionally associated with it. When the particle is in right-hand position, it obviously does not form a constituent with the verb on the surface. Do they form a constituent at some more underlying level? In the case of idiomatic verb-particle combinations there might be some reason to start with the particle at the left and move it around the NP (or move the NP around it), since the verb and particle together form a lexically listed semantic unit. In no other case is there any reason for ever needing them adjacent. But even in the case of idiomatic combinations there
english particle constructions
247
is little argument for underlying contiguity: as pointed out in section 7.3, movement seems totally unmotivated for the similar case of discontinuous V þ PP idioms. A better solution is to work out a way for lexical insertion to countenance discontinuous idioms (Nunberg, Sag, and Wasow 1994; Jackendoff 1997b, chapter 7; Jackendoff 2002; Culicover and Jackendoff 2005). One might propose to make the verb and right-hand particle a constituent by putting them in an underlying Larsonian shell (Larson 1988), as in (57); the verb would then move to the left of the object by head movement. This proposal would of course generalize to V-PP idioms. (57)
[V’ e [VP NP [V’ V Prt/PP] ] ]
But then consider another kind of VP idiom, for instance take [unfair advantage] [of NP], make [much] [of NP], make [a play][for NP], where the indicated constituency can be tested by the passive: unfair advantage was taken of NP vs unfair advantage of NP was taken. If the verb of such idioms formed an underlying constituent with the PP, as in (57), it would not form an underlying constituent with the NP. Yet the idiom this time consists of V plus NP. In other words, a move that improves the lot of the V. . . PP idioms makes matters worse for V-NP idioms. Another disadvantage of (57) is that the idiomatic particle construction his heart out would start its life discontinuous. So overall this solution confers no advantage. Finally, consider question (56c), the possible constituency of the particle and the NP. If a particle in left-hand position formed a constituent with the object, it would be essentially indistinguishable from a PP, making the facts of section 7.2 difficult to account for. What about a particle in right-hand position? A view championed by den Dikken 1995, and adopted in several papers in Dehe´ et al. 2002, is that a particle in right-hand position is the predicate of a small clause (SC) whose ‘subject’ is the NP, as in (58). (58)
[VP push [SC Fred down] ]
The argument is that (a) SCs are the canonical expression of predication, (b) down is predicated of Fred (i.e. the consequence of (58) is that Fred is down), so therefore (c) they must form a SC. While this reasoning is plausible in the case of (58), it is less so in the following examples, where the relation between the NP and the particle is not predication: (59)
a. Idiom: You’re putting me on. [I’m on.] b. Directional particle: He pushed the truck around/along. [The truck was around/along.] c. Aspectual particle: She fixed the sink up. [The sink is up.]
248
english particle constr uc tions d. Time-away: He knitted three hours away. [Three hours were away.] e. His heart out: They sang their butts off. [Nothing is (literally) predicated of their butts.]
Moreover, most combinations of NP plus particle never occur anywhere but after a verb; they don’t move around. Thus there is no independent evidence for their constituency, of the sort we have with NPs, PPs, and CPs. By contrast, there exists a genuine small clause construction in English that does occur in a variety of positions (Culicover and Jackendoff 2005, 131–5): (60) a. Subject position: [Bill in the hospital] would be a terrible thing. [Everyone drunk at once] would be amazing. b. Object position: I can’t imagine [Bill in the hospital/everyone drunk at once]. (Note anaphora: Can you imagine it?) c. Object of absolutive without: Without [Bill in the hospital/everyone drunk at once], we’ll never escape suspicion. A few particles can be shoehorned into this construction, for instance Fred down, Bill out, everyone away. But, unlike the particles in (59), these are particles that can occur independently with be, and hence function semantically as genuine predicates. Thus in general there is no argument based on predication for NP þ Prt being a small clause. One overarching reason for wanting the particle to form a constituent with either the verb or the NP is the hypothesis—now nearly a dogma in some circles—that syntactic trees are exclusively binary branching (Kayne 1994). This hypothesis would preclude a triply branching VP [VP V NP Prt]. However, there are two reasons to doubt this hypothesis, which I can only sketch here (for details, see Culicover and Jackendoff 2005, chapter 4). First, it has not been demonstrated that requiring binary branching genuinely makes the grammar any simpler. To be sure, it eliminates linear order from c-command, allegedly simplifying the theory of anaphora. But this doesn’t enable us to eliminate linear order altogether from UG or even from the theory of anaphora: linear order is required in any event to state conditions for discourse (intersentential) anaphora. Furthermore, linear order is obviously available in the primary linguistic input, so it is not clear in the long run that eliminating it from UG confers much advantage for the learner. Second, Culicover 2000 and Culicover and Jackendoff 2005 examine all extant arguments for branching structure in VP based on anaphora, ellipsis,
english particle constructions
249
gapping, and the like. They show that all these arguments ultimately lead to contradictory claims of constituency; the phenomena in question are really to be accounted for in terms of conceptual structure, not syntax. Thus there is no need for exclusively binary branching in the VP, and in fact it leads to unnecessary complications. The upshot is that the alleged need for binary branching vanishes, and hence cannot motivate a search for internal structure in VPs containing particles. The arguments of this section have been exceedingly brief, and I have certainly not exhausted all the evidence in the literature concerning constituency in particle constructions.10 My goal here has been primarily to pose a challenge: careful description reveals a wide range of English particle constructions that are syntactically uniform, despite vast differences in semantics, argument structure, and lexical status. Any proposal concerning syntactic structure must apply equally to them all. The null hypothesis is the lowest common denominator: a flat VP.f
10 In particular, various papers in Dehe´ et al. 2002 offer interesting arguments for constituency based on phenomena such as Gapping and extraction. There is no room to address them here. f Returning to one of the issues of the remarks preceding this chapter: There are six different phenomena that make use of the syntactic structure V-NP-Prt. In two of these, the particle is an independent lexical unit, but it plays entirely different semantic roles: directional particles are arguments, and aspectual particles are quasi-adverbial. In verb-particle idioms, the verb and the particle form a lexical unit; in his head off idioms, the particle and the direct object form a lexical unit. Finally, in the time-away construction, the particle commandeers the argument structure of the VP, and the verb functions as a manner modifier. There seems absolutely no semantic unity among these various phenomena, despite their sharing the very same syntax. I take this to be a strong argument for the (partial) autonomy of syntax from semantics. A Construction Grammar approach that insists on matching every syntactic pattern directly to a meaning has to treat the syntactic identity among these six phenomena as fortuitous. It does not allow any independent syntactic principles that restrict the position of the particle with respect to the direct object, object pronouns, adverbs, and PPs. By contrast, in the Parallel Architecture, the syntactic principles behind the verb-particle construction are to some degree independent of its varied semantics. English provides this particular idiosyncratic piece of syntax, and then finds many different ways to fill it.
chapter 8
Twistin’ the Night Away (1997)
This chapter examines the construction illustrated in (1), which I will call the ‘time-away’ construction.1 (1)
a. Bill slept the afternoon away. b. We’re twistin’ the night away.
This construction shares many general properties with the resultative construction, illustrated in (2a), and the way construction, illustrated in (2b). (2)
a. Amy pounded the clay flat. b. Beth whistled her way across America.
In this chapter, however, I will show that the time-away construction has syntactic and semantic peculiarities of its own, so it cannot be reduced to either of these other two. The conclusion will be that it is a distinct member of a family of constructions to which all three belong. Many contemporary theories of [This chapter appeared originally in Language 73 (1997), 534–59, and is reprinted here by permission of the Linguistic Society of America.] 1 Even after twenty-five years, teaching undergraduate Introduction to Linguistics occasionally holds some surprises. In March 1996 I was beginning to teach subcategorization, as usual using sleep as my stock example of an obligatorily intransitive verb. After class, one of the students, Amelia Parkes, came up to ask me, ‘What about He slept the afternoon away?’ As I glibly backed and filled on the topic of time adjuncts, the intrinsic interest of the example rapidly dawned on me. My debt to Ms. Parkes in the present chapter is therefore boundless. I am also grateful to Joan Maling, James Pustejovsky, Peter Culicover, Jay Keyser, Adele Goldberg, and the members of my spring 1996 Syntax class at Brandeis University for discussion of this material, and to Patrick Farrell, Malka Rappaport, and two anonymous referees for important comments on earlier versions. This research was supported in part by National Science Foundation grant IRI 92-13849 to Brandeis University.
twistin’ the night away
251
syntax proceed under the premise that phrase structure is projected exclusively from lexical heads.a If the analysis proposed here is correct, these constructions constitute an interesting challenge to this premise, for in such constructions, the argument structure of the VP is licensed not by the verb, as in the usual situation, but by the construction itself. But before jumping to any theoretical conclusions, let us first investigate the properties of the time-away construction.
8.1 Syntactic properties Example (1) contains an intransitive verb, followed by an apparently unlicensed NP plus the particle away. The issue is what licenses the occurrence of these two components of VP, since they are apparently not subcategorized by the verb. This section will establish the syntactic structure of the VP in this construction; the next section will deal with the semantics. Let us begin with the verb. As was noted in section 7.6 and as will be seen in the examples throughout this chapter, a wide range of verbs is possible in the construction—provided they have an intransitive subcategorization. Reviewing section 7.6, no postverbal NP may be present other than the time phrase (3), nor may the verb be one that requires a direct object (4). (3) a.
Fred drank the night away. Fred drank scotch the night away. b. Ann read the morning away. Ann read the newspaper the morning away.
(4) a. Fred devoured the night away. (Fred devoured) b. Ann perused the morning away. (Ann perused) Sometimes a with-phrase can substitute for the verb’s direct object in this construction, even though such a phrase would be impossible with the verb alone.b (5) a.
Fred drank the night away with a bottle of Jack Daniels. Fred drank with a bottle of Jack Daniels.
a
An accompanying assumption, also challenged by the present analysis, is that semantics is projected uniformly from syntactic structure, so that syntactic heads always correspond to semantic heads. Culicover and Jackendoff 2005 call this principle Interface Uniformity. b The peculiarities of the construction illustrated in (5) and (6) remain unexplained. It has some of the flavor of the with-theme construction, as in butter the bread with rancid butter, in which a semantic theme (or patient) argument for which there is no syntactic position in the VP comes to be realized as a with-phrase (Jackendoff 1990). In the with-theme construction, the theme cannot be licensed by the verb because it is already in the semantics of the verb (butter ¼ ‘put butter on’); in the time-away construction, the theme cannot be licensed by the verb because the construction has taken control of licensing object position. Perhaps.
252
twistin’ th e nigh t away b. Ann read the morning away with a pile of old mysteries. Ann read with a pile of old mysteries.
If the verb subcategorizes a PP, sometimes I find the PP acceptable after away, just where one would expect it (6a)—but this is not always the case (6b). (6)
a. Fran talked the night away with Harry/?to Harry/??about politics. b. Fran walked the afternoon away toward New York.
As mentioned in section 7.6, the particle away behaves like those in other verb-particle constructions, in that it can invert with the NP, especially if the NP is a little longer. (7)
a. Stan drank away the (entire) afternoon of his 50th birthday. b. Stan fished away all of Tuesday morning.
The particle can also be modified, in which case it does not invert: (8)
a. Dan slept the long afternoon entirely away. b. Dan slept entirely away the long afternoon.
Thus it behaves just like ordinary verb particles (look the answer right up/look right up the answer). The NP is a free time expression, permitting a variety of determiners (including quantifiers). (9)
Bill drank the night/three whole weeks/every morning/his entire vacation away.
It must not be confused with NPs used as free time adverbials such as those in (10). (10) a. Fred hasn’t slept this year. b. Kate is leaving Monday. [For one thing, the time-away construction excludes the verb having other NP complements, as noted above in (3). By contrast, the time adjuncts do not: (10.1)
a. Frank drank (scotch) all night. b. Ann read (the newspaper) all morning.]
Unlike the time phrases in (10), those in (7)–(9) are apparently VP complements, since they can undergo passive, given the right contextualization (though I have no account of why such contextualization is necessary). (11) In the course of the summer, many happy evenings were drunk away by the students before they finally realized there was serious work to be done.
twistin’ the night away
253
In fact, even the intransitive sleep can be passivized in this construction: (12)
The evening had been nearly slept away, when I suddenly awoke with a start.
Such passives are of course impossible with the time phrases in (10). (13)
a. This year hasn’t been slept by Kate. b. Monday is being left by Bill.
Similarly, the time phrases in the time-away construction can undergo ‘toughmovement’, whereas time adjuncts like those in (10) cannot. (14)
a. A morning like this is hard for even ME to sleep away. b. A morning like this is hard for even ME to sleep.
Conversely, contrasted time adjuncts can be stranded by VP-ellipsis, but, as is typical of VP complements, the time phrases in the time-away construction cannot. (15)
a. Bill read all of Monday AFTERNOON, and Sally did so most of Tuesday MORNING. b. Bill (blissfully) read away all of Monday AFTERNOON, and Sally did so most of Tuesday MORNING.
The positioning of manner adverbs in the time-away construction is also telling: they can go in the exact places where they occur in verb-particle constructions such as throw away. (16)
a. Sue drank happily the night away. Sue threw quickly the paper away. b. Sue drank the night happily away. Sue threw the paper quickly away. c. Sue drank the night away happily. Sue threw the paper away quickly.
This contrasts again with the position of manner adverbs vis-a`-vis the time adjuncts in (10). (17)
a. Sue drank happily all night. b. Sue drank all night happily.
All this evidence points to the time expression in the time-away construction being in direct object position, as if it usurps this position so that the verb itself cannot license an NP there. Moreover, away is a particle, indistinguishable in
254
twistin’ th e nigh t away
syntax from an ordinary verb particle. We return in sections 8.4 and 8.8 to how both these phrases are licensed.
8.2 Semantic properties For a first approximation, (1) can be paraphrased by a sentence in which the time expression is within a delimiting for-phrase. (18) Bill slept for the (whole) afternoon. Such for-phrases are familiar from the literature on telicity: they require the rest of the sentence to express an atelic (non-intrinsically bounded) situation: (19) a. Paula ate the peanut for the whole afternoon. b. Dave died for the whole afternoon. c. Beth reached the mountaintop for the whole afternoon. If an event is bounded but repeatable, its interpretation is coerced into a repetition reading by the for-phrase (see Chapter 5). (20) a. Saul sneezed for the whole afternoon. (repeated sneezing) b. Jimmy jumped off the roof for the whole afternoon. (repeated jumping) The time-away construction behaves similarly: (21) a. Dave died/awakened the afternoon away. b. Saul sneezed/sprinted the afternoon away. (repeated sneezes, sprints) However, the time-away construction has further constraints. It requires its subject to be acting volitionally: (22) a. The light flashed for two hours. b. The light flashed two hours away. Furthermore, the verb must denote an activity, not a state (even a volitional state): (23)
a. Celia sat for two hours. b. Celia sat two hours away. c. Celia saluted for two hours. (stayed in saluting position or did repeated salutes) d. ?Celia saluted two hours away. (repeated salutes only)
A more subtle aspect of the semantics of the construction is that the subject is in some sense understood as ‘using’ the time, or, even better, ‘using the time up’. Some of this flavor appears in the second-approximation paraphrases in (24). The verbs in (24b, c) even duplicate the syntax of the construction, the verbs
twistin’ the night away
255
fritter, while, and the relevant sense of piss requiring the particle away. In fact, the verb while appears only in this construction. (24)
a. Sam spent/wasted the afternoon sleeping. b. Sam frittered/pissed the evening away gambling. c. I could while away the hours conferring with the flowers.
These verbs require their gerundive complement to express a temporally extended situation. (25)
a. Sam spent/wasted three days dying/awakening. b. Sam whiled/frittered/pissed away three days dying/awakening.
Again, a repeatable bounded event can be construed as iterated. (26)
a. Sam spent/wasted three days sneezing. (multiple sneezes) b. Sam whiled/frittered/pissed away three days diving from the 10-meter platform. (multiple dives)
Alternatively, a bounded (telic) situation is acceptable in the gerundive complement, if of a commensurate time scale. (27)
a. Beth spent four weeks building a cabin. b. I frittered away three days fixing my car. c. This computer program wastes 33 precious milliseconds deleting files.
What makes these paraphrases better than those with for-phrases is that they require the subject to be volitional (directly or indirectly, as in a computer program): (28)
a. The light spent/wasted the afternoon flashing. b. The light whiled/frittered/pissed away the afternoon flashing.
On the other hand, they are still not quite as constrained as the time-away construction, in that they allow volitional states; compare (29) to (23). (29)
a. Amy spent/wasted two hours sitting (in the park). b. Amy whiled/frittered/pissed away two hours (just) sitting.
All these verbs except while can take other direct objects besides times, referring to the subject’s resources; this is not a possibility in the time-away construction. (30)
a. Bill spent/wasted his energy/hundreds of dollars skiing. b. Bill frittered/pissed/whiled away his energy/hundreds of dollars fishing. (31) Pam fished the afternoon/her energy/hundreds of dollars away.
256
twistin’ th e nigh t away
A further subtlety in the time-away construction is reflected in the verbs of these paraphrases: there is an insinuation that the activity in question was heedless pleasure, or that the subject should have been doing something else, or both. So there is something a little ironic in a sentence like (32), if it is good at all. (32) ?#Ivan worked/toiled/labored three (miserable) hours away. This insinuation can also appear with verbs like spend, waste, fritter, and piss, but it is defeasible, an alternative interpretation being that the subject wished to have spent the time otherwise. While is closer to the sense of (32). (33) a. Ivan spent/wasted three hours working (when he could have been out in the sun). b. Ivan frittered/pissed three hours away toiling on his manuscript (when he could have been drinking). c. #Ivan whiled away three hours toiling on his manuscript. We can also get something close to this sense by adding the adjunct carefree: (34) a. Nancy slept carefree for three hours. b. The light flashed carefree for three hours. c. #Ivan toiled on his manuscript carefree for three hours. A final point on the description of the semantics of the construction. In the context of a journey, say a cross-country flight, the time expression in a forphrase can be replaced by a distance expression D which is read as ‘the amount of time it took to travel/traverse D’: (35) (It was a long boring flight, and so . . . ) Bob slept for 350 miles/for the whole state of Nebraska. The same is true in the time-away construction and in the spend and while away constructions. (36) (It was a long boring flight, and so . . . ) a. Bob slept 350 miles/the whole state of Nebraska away. b. Bob spent 350 miles/the whole state of Nebraska sleeping. c. Bob whiled away 350 miles/the whole state of Nebraska sleeping. However, all these cases other than the time-away construction also permit an activity nominal to be substituted for the time expression: (37) a. Lillian spent/wasted the lecture doodling. [¼ drawing doodles] b. Lillian frittered/pissed/whiled away her appointment with the dean daydreaming.
twistin’ the night away
257
This is not possible with the time-away construction, even though the verbs doodle and daydream are prime candidates for the construction (38c). (38)
a. Lillian doodled the lecture away. b. Lillian daydreamed away her appointment with the dean. c. Lillian doodled/daydreamed the morning away.
In short, the construction has a rather complex and rich semantics, approached but not exactly captured by any paraphrase.
8.3 The aspectual particle awayc As discussed in section 7.5, the particle away occurs by itself with a related aspectual sense. (39)
Bill slept/waltzed/drank/talked/read/sneezed away.
This means roughly ‘Bill kept on V-ing’. Away here is more or less a continuative counterpart of the aspectual particle up, which signals completion: (40)
Elena drank the milk up. Aaron rolled the rug up. Ben glued the chair up.
However, unlike up, and like the time-away construction, aspectual away forbids the verb from licensing a direct object. (41)
Dave drank scotch away. Dave danced waltzes away. Dave read newspapers away.
Like the time-away construction, it allows some but not all prepositional complements of the verb to follow it (compare to (5)). (42)
a. Hilary talked away about her latest project. b. Miriam worked away on her manuscript.
Looking slightly more closely at the semantics of this particle, it seems to carry some of the sense of heedless activity found in the time-away construction, but not everything. For example, (43a) is bad, like (23b). But (43b) is good, unlike (22b); it seems somewhat to personify the light, as though the light didn’t
c Some of the material originally in this section and the next has been incorporated into Chapter 7 for better continuity.
258
twistin’ th e nigh t away
care about us. And (43c) is good, unlike (32); there is no need for the activity in question to be pleasurable. (43) a. Celia sat away. b. The light flashed away (despite all our attempts to turn it off). c. Ivan worked/toiled/labored away. Whatever the similarities, any simple identification of the two uses of away is thwarted by two striking differences. First, sentences with aspectual away are atelic, whereas the time-away construction is telic, i.e. it is understood as occupying a bounded period of time. (However, apparently because the time period is already specified in the time-away construction, the standard tests for both telicity and atelicity fail; (44c, d) provide a possible contrast.) (44) a. Lois and Clark danced away for/in two blissful hours. b. Lois and Clark danced two blissful hours away for/in a month. [OK only under coerced readings, e.g. every day for a month] c. It took a month for Lois and Clark to finally get to dance away. [OK under spatial reading of away] d. It took a month for Lois and Clark to finally get to dance two blissful hours away. Second, away does not admit quantificational modification when used aspectually, but it does in the time-away construction. (45) a. Sally waltzed entirely/partly/half away. b. Sally waltzed the afternoon entirely/partly/half away. Thus the away in the time-away construction looks as though it might be related to aspectual away.d However, aspectual away alone is not sufficient to explain the semantic properties of the construction.
8.4 More on aspectual particles How are the time-away construction and aspectual away to be analyzed? We might, as one reviewer has suggested, simply regard sleep away as a verb with transitive and intransitive variants (sleep (the night) away), and be done with it. However, it is worth seeing what lies behind such a seemingly simple response.
d I am not sure any more that I would endorse any sort of close semantic relations between aspectual away and the time-away construction, given how vaguely the relation is established here.
twistin’ the night away
259
Chapter 7 discussed three well-known sources in English for the verb-particle construction. First is the large collection of verb-particle combinations with idiomatic meanings, for instance look NP up, bring NP [e.g. a child] up, chew NP out, and so forth. Because of their noncompositional meanings, there is no choice but to list them in the lexicon. The second source involves verbs such as throw, take, and carry that select a directional (Path) PP. If the PP happens to consist only of an intransitive preposition, the result behaves syntactically like a verb-particle construction. (46)
a. Beth threw/took/carried the food up/in/up the stairs/into the house. b. Beth threw/took/carried up/in the food.
Here there is no need to list the verb-particle combinations in the lexicon, since the particle satisfies one of the verb’s argument positions, and the meaning is fully compositional. Provision has to be made in syntax only for the alternative positioning of the particle preceding the direct object. This class is remarked as early as Klima 1965 and Emonds 1970. The third source involves the combination of a verb with an aspectual particle. Here is a smattering of such examples to add to those with up and away in the previous section. (47)
run/sing/talk/work/think/write on
(48)
a. read/scan/skim the book through b. sing/play the aria through c. work/talk/think the problem through
(49)
a. b. c. d.
cook the food over sing/play the aria over write/type the paper over do the problem over
The particle does not satisfy an argument position of the verb; rather it contributes an aspectual sense, often paraphrasable by some sort of adjunct PP. Run/sing on, for instance, means roughly ‘run/sing some more’; read/scan the book through means ‘read/scan the book from beginning to end’; cook the food over means ‘cook the food again’ or ‘re-cook the food’. Aspectual up means roughly ‘to completion’ (even when redundant, as in close the suitcase up); aspectual away emphasizes atelicity, as described in the previous section. Thus this meaning of the verb-particle construction is in general fully compositional. This does not preclude idiomatic combinations such as see the process through and work up a proposal built along the same pattern; the examples in (48c) may also be of this type. Moreover, certain regular combinations, for
260
twistin’ th e nigh t away
instance finish up and sleep the night away, might be listed anyway in the lexicon as relatively high-frequency ‘fixed expressions’, sort of like whistle while you work. But, on the other hand, examples like doodle three hours away are hardly fixed expressions. In light of this distribution, how are we to interpret the proposal that sleep away in sleep the morning away ‘is a verb’? Such a proposal invokes an analogy to the idiomatic verb-particle combinations, which are listed in the lexicon. However, we have seen that, in terms of productivity, sleep (the night) away is more like sleep some more and put the towel down—a consequence of productive phrasal combination—than it is like the idiomatic sleep in (‘sleep till late in the morning’). The conclusion, therefore, is that sleep away need not be listed in the lexicon as a unit, and that it is a concatenation of two independent syntactic elements. Such a syntactic and lexical analysis, however, leads to the question of what licenses the NP in the time-away construction. Though it is not licensed as the object of the verb, it behaves as though it is the object of the VP. That is, it is useful to pull apart two notions normally treated as identical: the postverbal NP licensed by the argument structure of the verb, and the postverbal NP that precedes adverbs and participates in passive, ‘tough-movement’, particle movement, and so forth. A simple way to sidestep this issue is to invoke a lexical rule that productively combines a verb with an aspectual particle with a verb to create a new ‘complex verb’ (now not a V0, but something on the order of take to task), which in the case of time-away licenses an object in the usual way. Such an account would be characteristic of the approach of HPSG (Pollard and Sag 1987) and of Levin and Rappaport Hovav 1995. The latter in particular advocate an approach of this sort to the resultative and the way construction.2 By contrast, Jackendoff 1990 and Goldberg 1995 argue for an approach to the latter two constructions in which the construction is an autonomous grammatical unit that supplants the verb in 2
Levin and Rappaport Hovav vacillate in different publications between calling the formation of a complex verb a ‘lexical rule’ or a ‘syntactic rule’. In Levin and Rappaport 1995, the formation of resultatives is called a syntactic rule forming a complex verb, while the use of sound emission verbs in motion contexts (the trolley rumbled around the corner) is called a lexical rule creating a new verb sense. The following, from Levin and Rappaport 1996, clarifies their current thinking on resultatives: . . . we propose that they are derived via a syntactic process of complex V0 formation involving an activity verb and the result phrase. That is, a complex predicate is formed from a verb and either an AP or PP. . . . Although we describe this as a syntactic process of complex predicate formation, more research is necessary to determine whether this process is, in fact, syntactic or whether it is actually a lexical process that involves the composition of two argument structures. What is important for us is that this process does not involve the creation of a new accomplishment lexical entry for a verb . . . [continued]
twistin’ the night away
261
licensing the argument structure of the VP. In such an approach, the object of the VP need not be the object of the verb. In order to evaluate these two contrasting approaches, we need to review the resultative and way construction, and show their distinctness from each other and from the time-away construction.
8.5 Review of the resultative and way constructions As mentioned at the outset, the time-away construction is reminiscent of two constructions discussed in the literature, the resultative and the way construction, shown in (2) (repeated here). (2) a. Amy pounded the clay flat. [resultative] b. Beth whistled her way across America. [way construction] In each of these there is (a) a fixed syntax in the VP, at odds with what the verb licenses, associated with (b) semantic peculiarities, including particular aspectual properties. The resultative (Simpson 1983, Carrier and Randall 1992, Goldberg 1995, Levin and Rappaport Hovav 1991, 1995, Jackendoff 1990, among others; much more detail here in Chapter 9) comes in two syntactic variants. The transitive resultative exemplified in (52) has the fixed VP structure (53); the intransitive resultative (54) has the fixed VP structure (55). (52)
a. b. c. d. e.
Willie wiped the table clean. Cathy cooked the pot dry. I coughed myself awake. (Amy Jackendoff, August 1, 1997) Tara talked us into a stupor. Wally walked the soles off his shoes.
(53)
[VP V NP AP/PP]
(54)
a. b. c. d.
(55)
[VP V AP/PP]
The river froze solid. The toast burned black. The pitcher broke into fragments. The corn grew tall.
I will concentrate on the transitive case here, since it is more closely parallel to the time-away construction. With the exception of (52a), none of the verbs in What is crucial for my argument is that not only are ‘complex predicates’ not listed in the lexicon, but also at no point is there reason to believe that they form V0s in syntax, except on the precise theory-internal grounds that I am questioning here. See section 9.7.4 for more discussion of Levin and Rappoport Hovav on resultatives.
262
twistin’ th e nigh t away
(52) license these direct objects unless the AP/PP is also present: cook the pot, cough myself, talk us, walk the soles. Yet these are undeniably direct objects, since they all undergo passive (except for himself, of course). (56) a. b. c. d.
The table was wiped clean by Willie. The pot was cooked dry by Cathy. We were talked into a stupor by Tara. The soles had been walked off Wally’s shoes by the time he finished the Walk for Hunger.
Thus in present terms, it makes sense to say that the objects in (56) are objects of the VP but not objects of the verb. These direct objects invariably have the role of Patient, as revealed by the do to NP test. (57) a. b. c. d. e.
What Willie did to the table was wipe it clean. What Cathy did to the pot was cook it dry. What I did to myself was cough myself awake. What Tara did to us was talk us into a stupor. ?What Wally did to his soles was walk them off his shoes.
The literature cites many tricky selectional restrictions on what verbs and adjectives can appear in the construction, attesting to its semantic complexity. A first approximation to the meaning is provided by a paraphrase with a causative inchoative main clause, the verb of the sentence appearing in a means adjunct. (58) a. b. c. d. e.
Willie caused the table to become clean by wiping (it). Cathy caused the pot to become dry by cooking (in it). I caused myself to become awake by coughing. Tara caused us to go into a stupor by talking (to us). Wally caused the soles to come off his shoes by walking.
Finally, there are various idioms that borrow the syntax and semantics of the resultative. (59) a. b. c. d.
I cried my eyes out. (6¼ my eyes came out) Bill drank me under the table. (Bill drank me into another bar.) You scared the daylights out of me. (daylights ¼ ?) Jan tickled us pink. (6¼ we became pink from Jan’s tickling)
The way construction (Jackendoff 1990, Goldberg 1995, Marantz 1992, Levin and Rappaport Hovav 1995) has a fixed syntax in which the verb is followed by the NP X’s way and a PP of path.
twistin’ the night away (60)
263
a. b. c. d.
Dora drank her way down the street. James joked his way into the meeting. Babe Ruth homered his way into the hearts of America. They were killing and robbing their way across the Southwest. (Walter Cronkite, January 2, 1997) e. Paulette pushed/elbowed her way through the crowd.
The verb cannot occur with a direct object other than X’s way (61), and must either permit an intransitive frame (60a–d) or else denote an action of applying force so as to move objects out of one’s path ((60e) vs. (62)). (61)
(62)
a. b. c.
Dora drank scotch her way down the street. James told jokes his way into the meeting. Babe Ruth hit home runs his way into the hearts of America.
a. Emma elbowed. b. Emma elbowed Harry. c. ?Peter pulled his way through the crowd.
X’s way is demonstrably in direct object position, unlike all the way, which occurs as a specifier of the path phrase. For example, X’s way can be followed but not preceded by an adverb, whereas the reverse is the case for all the way. (63)
a. b. c. d.
Sue sang happily her way down the street. Sue sang her way happily down the street. Sue ran happily all the way down the street. Sue ran all the way happily down the street.
However, X’s way does not undergo passive, partly because of the bound pronoun. (We return to this issue in section 8.6.2.) (64)
Her way was sung down the street by Sue.
There are interesting selectional restrictions on the verb, in particular that it denote a continuing activity (65a) that is initiated by the subject (65b) and that is usually goal-directed in some sense (65c). (65)
Judy jumped her way over the field/over the ditch. (multiple jumps necessary) b. The butter melted its way off the turkey. c. ??Bill meandered his way over the field. a.
The overall meaning of the construction is close to ‘go/get/come PP (by) V-ing’.
264
twistin’ th e nigh t away
(66) a. Dora went down the street drinking. (¼60a) b. James got into the meeting by joking. (¼60b) c. Babe Ruth went/got/came into the hearts of America by homering. (¼60c) d. Paulette got through the crowd by pushing/elbowing people. (¼60e) e. Sue went happily down the street singing. (¼63b) f. Judy went over the field (by) jumping. (¼65a) And there are particular verbs that appear idiomatically in this construction, wend in particular occurring only with way. (67) Max made/wended/threaded/wormed/worked his way down the narrow alley.
8.6 Comparison of the three constructions Can any of these three constructions be reduced to any of the others? I will argue that the answer is no, and that they all must be regarded as special cases of a larger more abstract generalization.
8.6.1 Resultative versus way construction Jackendoff (1990) and Goldberg (1995) argue that the resultative and way construction are distinct, each with its own peculiar properties. Marantz (1992), however, tries to reduce the way construction to a special case of the ‘fake object’ resultative. He claims that his way refers to the path traveled by the subject, and that it is to be treated as an inalienable possession, parallel to the body parts that alternate with reflexives in (68). (68) a. Bill shouted himself/his vocal chords hoarse. b. Bill cried his eyes red. Goldberg, although sympathetic to the treatment of way as denoting the path traveled, rejects Marantz’s identification: While it is important to recognize the relationship between the way construction and fake object resultatives, it is nonetheless necessary to posit a distinct, albeit related, construction in the grammar to account for the way examples. In particular, the following differences between the two constructions remain; these differences prevent us from claiming that the two are actually the same construction. As observed above [and here as well—RJ], the way construction is available for use with a wide variety of verbs, whereas resultatives in general, and fake object resultives in particular, are highly restrictive. For example, the fake object resultative analogs of [69, shown in 70,] are unacceptable:
twistin’ the night away
265
[(69)]
a. He bludgeoned his way through. b. The players mauled their way up the middle of the field. c. Their customers snorted and injected their way to oblivion.
[(70)]
a. He bludgeoned himself crazy. (with a fake object interpretation wherein he bludgeoned [people in general] until he went crazy) b. He mauled himself silly. (meaning that he mauled [people] until he became silly) c. He snorted and injected himself dead.
Moreover, the resultative construction cannot be used to predict the requisite interpretation of way expressions that a path is created—that the speaker must construe there to be difficulty or obstacles to the motion. (Goldberg 1995, 217)
That is, the semantics and selectional restrictions of the two constructions do not match closely enough to allow us reduce one construction to the other.
8.6.2 Time-away versus way construction The same proves true of the time-away construction vis-a`-vis the other two. Compare it to the way construction. Marantz argues that his way in the way construction ‘measures out’ the path traversed by the subject, along lines discussed by Tenny 1994, for example.e One might say that the time expression in the time-away construction does likewise, for instance that the afternoon measures out Bill’s sleeping in Bill slept the afternoon away. Moreover, one might point out that both require activity extended over a period of time, so that jumping and sneezing, for instance, have to be iterated. So one might suggest that the way-construction describes the subject’s activity while traveling through space, and the time-away construction describes the subject’s activity while traveling ‘through time’. These parallels are all well and good, but they do not provide enough evidence to say the two constructions are the same. For one thing, in the way construction, the direct object his way is lexically fixed and the PP is lexically free, whereas in the time-away construction the direct object is lexically free (within the selectional restrictions) and the particle is lexically fixed. For another, the selectional restrictions on the verb are different, as shown in (71)–(72). (71)
e
a. ?#Ivan toiled three hours away. b. The light flashed the morning away. c. Emma elbowed an afternoon away.
See Chapter 6 for why Tenny’s approach to measuring out is almost entirely off the mark.
266
twistin’ th e nigh t away
(72) a. Ivan toiled his way up the steep slope. b. The light flashed its way into our consciousness. c. Emma elbowed her way into the room. Next consider the status of the direct object. In the time-away construction it behaves like a fully referential phrase: it undergoes passive and other movements such as tough-movement (73a); it can be pronominalized (73b) and questioned (73c). (73) a. Even a whole afternoon isn’t hard for Bill to sleep away. b. Bill slept Monday afternoon away, while Harry drank it away. c. Which mornings is Bill most likely to sleep away? By contrast, his way in the way construction cannot undergo passive, as seen in (64); it cannot undergo tough-movement (74a); it cannot be pronominalized (74b), ellipted (74c), or questioned (74d). Notice that it is distinguished from the referential noun way (‘route’), which does participate in all these processes (75) and also from inalienable body parts, which participate in most of them (76). (This further argues against Marantz’s account of the semantics.) (74) a. His/Bill’s way is hard for Bill/him to push into the room. b. Bill whistled his way into the room, and then he joked it down the hall. c. Bill whistled HIS way into the ROOM, while Harry whistled HIS down the hall. d. Which way/which of his ways did Bill poke into the room? (75) a. b. c. d. e.
Hisi way was blocked by Billj. Bill’s favorite way through the woods is easy to miss. The river didn’t block his way, but the lake blocked it. The river didn’t block HIS way, but it blocked MINE. Which/whose way did Bill block?
(76) a. Hisi/Billi’s arm was broken by Billi/himi. b. Hisi arm wasn’t too hard for Billi to break. c. Bill broke his arm playing football, and then he broke it again falling downstairs. d. Bill broke HIS arm playing football, but Harry broke HIS falling downstairs. e. Which arm did Bill break? This suggests that X’s way in the way construction is not a referential phrase, at least not one of any usual sort.
twistin’ the night away
267
In fact, it is even more constrained than idiomatic uses of body parts, such as his heart in Bill ate his heart out. This latter fails most of the tests above, but pronominalization (77c) and ellipsis (77d) are still possible. (77)
a. His heart was eaten out (by Bill). b. His heart was terrifyingly easy for Bill to eat out. c. Bill ate his heart out over Sally on Wednesday, then he ate it out over Jessica on Thursday. d. Bill ate his heart out over Sally, and Harry ate HIS out over Jessica. e. Whose/which heart did Bill eat out?
Thus, contra Marantz (and Goldberg’s (1995) and Levin and Rappaport Hovav’s (1995) assent to his analysis) X’s way is about as nonreferential as NPs come. We conclude that the way and time-away constructions differ radically in the semantics of their direct objects. Finally, it is not the case that the way construction simply applies to travel through space and the time-away construction to travel ‘through time’. (78a) is a way-construction in which the path is a duration of time; it is not entirely synonymous with (78b). (78)
a. Betty bumbled her way through the whole afternoon. b. Betty bumbled the afternoon away.
The way construction permits the time period to be identified by an event taking place in it; the time-away construction does not. (79)
a. Dave daydreamed his way through the whole movie. b. Dave daydreamed the movie away.
In the proper context, the time-away construction allows the period of time to be identified by a stretch of space traveled during that period. The way construction does not: (80b), if acceptable, is not equivalent to (80a). (80)
During our cross-country flight, a. Bob slept the whole state of Nebraska away. b. Bob slept his way across the whole state of Nebraska. [wrong interpretation]
As far as I can tell, none of these differences between the two constructions can be attributed simply to the difference between ‘travel in space’ and ‘travel in time’. I conclude that neither of the constructions can be reduced to a special case of the other.
268
twistin’ th e nigh t away
8.6.3 Time-away versus resultative construction The direct object in the resultative is of course fully referential and can undergo all the usual syntactic processes. (81) a. The pot was cooked dry by Cathy. b. Tara is real easy to talk into a stupor. c. Bill cried his eyes out on Wednesday, and he cried them out again on Sunday. d. Bill cried his eyes out on Wednesday, and Harry cried HIS out on Sunday. e. What did Helen hammer flat? In this respect it resembles the object of the time-away construction more than does X’s way, reinforcing the view urged in section 8.5 that it is the object of the VP even if not the object of the verb. With this resemblance in mind, one might try to explain the time-away construction in terms of some sort of causative, perhaps like (82). (82) Bill slept the afternoon away. ‘Bill caused the afternoon to go away by sleeping’ But is this realistic? Here are three reasons not to assimilate the two constructions. First, recall that the object in the resultative is a Patient: it is acted upon by the subject and affected by the action, as shown by the do to test in (57). But a time period cannot be affected by what someone does during it. Hence the object of the time-away construction fails the do to test and satisfies only the weaker do with test, as seen in (83). (83) a. What Bill did with/to Monday was sleep it away. b. What Betty did with/to the morning was bumble it away. c. ?What we did with/to the night was twist it away. An example like (84a) might appear to satisfy the happen to test for Patients, but other examples seem less consistent. (84) a. What happened to Monday was I slept it away. b. ??What happened to the morning was Betty bumbled it away. c. What happened to the night is we twisted it away. Consequently it is hard to conceive of the time period as satisfying the same semantic role as the object in a resultative. Another reason not to assimilate the two is the status of away. Recall that we want somehow to relate the time-away construction to the aspectual particle
twistin’ the night away
269
away, which lends some of its properties to the construction. However, the syntax of aspectual away looks much like the intransitive resultative. (85)
a. Beth whistled away. b. Beth grew tall.
But there is no resemblance whatsoever in the semantics of (85a) and (85b). In (85b), tall describes Beth’s state at the end of the event; in (85a), away describes nothing of the sort. (85a) is atelic: we can say Beth whistled away for hours but not Beth whistled away in an hour. By contrast, (85b) is telic: we can say Beth grew tall in a year but not Beth grew tall for a year. Thus, if we wish to capture the rather salient relationship of the time-away construction to aspectual away [but see subfootnote d above], we necessarily lose the (much less salient) parallelism to the resultative. Finally, there is nothing about the resultative that explains the peculiar selectional restrictions on the verb in the time-away construction, for example, the oddness of verbs like labor, work, and toil, and the restriction to volitional subjects (compare the resultative The wind blew the papers away and the timeaway construction The wind blew the night away).f Two more examples to push the point home. The following sentence (pointed out by Joan Maling) is ambiguous between the two constructions. (86)
Bill gambled his life away.
The time-away reading is one in which Bill has spent his whole life gambling. The resultative reading is one in which Bill has bet his life and lost; he ends up perhaps submitting to slavery or killing himself. A resultative that looks a lot like a time-away construction is (87). (87)
Diet those pounds away!
This really does mean ‘make those pounds go away by dieting’; that is, the typical resultative paraphrase works beautifully, unlike in (82). By contrast, the timeaway paraphrases developed in section 8.2, when generalized beyond times, all have to do with using up resources. Under this type of reading, (87) would have to mean something like the implausible ‘spend those pounds by dieting’. Hence (87) is a resultative, not an extension of the time-away construction. Notice also that away does not express continuative aspect either in (87) or the resultative reading of (86); rather Bill’s life and those pounds are gone.
f As will be seen in Chapter 9, many resultatives have their own peculiar selectional restrictions, which have little to do with those of the time-away construction.
270
twistin’ th e nigh t away
From these considerations, I conclude that the resultative and the time-away constructions are distinct. The differences emerge clearly once one examines their semantics in any detail; it is only if one confines oneself to relatively shallow semantics such as listing theta-roles that one is tempted to conflate them.
8.7 Still another family of related constructions/idioms [The original version of this section concerned two further constructions. The first was the his head off construction illustrated in (88), showing that it is an idiom consisting of a fixed direct object plus a particle, connoting intense activity. (88) a. Fred talked his head/his ass/his butt off, but to no avail. b. The chef was cooking up a storm back in the kitchen. c. Every night I sit here and sing my heart out; but does anyone listen to me? No! This material has been incorporated here into section 7.7. The second part of the section concerned a subclass of ‘fake reflexive’ resultatives with a similar interpretation: (92) Dean laughed/danced himself crazy/silly/to death/to oblivion. This part of the discussion has been incorporated into section 9.8.]
8.8 Two accounts We now return to the question of how the NP in the time-away construction is licensed. As remarked at the end of section 8.4, one possible account is to regard sleep away, drink away, dive away, doodle away, and so forth as complex verbs, produced by a lexical rule. These complex verbs license the time NP in object position. Recall, however, that such complex verbs are not syntactic V0s, for like other verb-particle combinations they are discontinuous. Nor are they lexically listed, since they are productive. Such an account is congenial with the approach of Levin and Rappaport Hovav (1995), who speak, for example, of verbs of sound emission undergoing a regular lexical meaning shift to a directed motion sense, as in (95) (though see n. 2). (95) The trolley rumbled across the intersection. In the present case we can speak of activity verbs undergoing a regular lexical meaning shift to a ‘spending time’ sense when combined with the aspectual particle away. (I ignore for now the further refinement that is needed to distinguish the time-away construction from ordinary aspectual away.)
twistin’ the night away
271
Such a lexical rule account appears appropriate within the overall outlook of Government-Binding Theory and the Minimalist Program (Chomsky 1981, 1995), in which phrase structure is taken to be projected from the lexicon. In order to license the NP position in sleep the night away, there must be a verb that projects it; hence we must regard sleep away as the verb in question. Parallel considerations apply to a theory as different as HPSG (Pollard and Sag 1987), in which complementation is determined by the subcategorization feature of the syntactic head. This forces us to treat the night as subcategorized by a verb, presumably again sleep away. Similar treatment would seem to be required in LFG as well (Bresnan 1982a, Bresnan and Moshi 1990): the status of the NP as direct object would have to be encoded in f-structure and licensed by the complex verb sleep away, which would be produced by a morpholexical operation. What these approaches have in common is the assumption, pervasive in the literature, that the direct object of a VP is necessarily licensed by the verb. Under this assumption, the night must be the object of some verb, so the theory forces one to say sleep away is that verb. An alternative is to suppose that licensing of the object by the verb is only the default case, and that the grammars of particular languages may provide other means of licensing it. Such an approach emerges in Jackendoff 1990 and Goldberg 1995, who propose that argument structure can be determined in part by constructional idioms—syntactic configurations whose structure contributes semantic content above and beyond that contained in the constituent lexical items. The existence of a class of constructional idioms is hard to doubt. As argued, for instance, by Fillmore, Kay, and O’Connor 1988 and Michaelis and Lambrecht 1996, there exist some constructions like (96) that are syntactically sui generis and do not follow from independent syntactic principles. (96)
a. One more beer and/or I’m leaving. b. The more you eat, the fatter you get.
(96a) contains the unusual structure NP conj S, which follows none of the standard rules for conjunction; it carries a conditional sort of reading (Culicover 1972; Culicover and Jackendoff 1997, 2005). (96b) contains two paratactic comparative clauses, each beginning with the word the (hardly in its usual determiner use); again there is a conditional or correlative reading (Fillmore, Kay, and O’Connor 1988; Culicover and Jackendoff 1999, 2005; den Dikken 2005; Abeille´, Borsley, and Espinal 2006).g g
See Chapters 10–12 for more constructions with sui generis syntax:
(i) the symbol $ (Chapter 10) (ii) Do you LIKE-him-like-him? (Chapter 11) (iii) year after year (Chapter 12)
272
twistin’ th e nigh t away
Jackendoff and Goldberg both argue that constructional idioms can also assign special meaning to more garden-variety autonomous syntactic structures. Such constructions can be detected because (a) the complement structure is unusual for the verb, (b) the syntactic structure of the VP is unusually restricted, (c) unusual selectional restrictions emerge, and (d) sometimes there are special morphemes such as way that mark the construction. In addition to the way construction and the resultative, for instance, Jackendoff and Goldberg independently conclude that the beneficiary (or for-) dative and even certain cases of the recipient (to-) dative are constructional idioms of English. A third sort of case is discussed by Gleitman et al. 1996, who point out (and cite experimental evidence) that ‘symmetric’ predicates such as be similar are in fact asymmetric when used in the construction (97a) but not in the construction (97b), even though the sentences are nominally synonymous. (97) a. X is similar to Y. b. X and Y are similar. For example, as suggested by Talmy 1978b and demonstrated experimentally by Tversky 1977, speakers find Nepal is similar to China more acceptable than China is similar to Nepal, at least with default pragmatics; but Nepal and China are similar and China and Nepal are similar are equally acceptable. Gleitman et al. suggest (roughly) that the semantic asymmetry comes from casting the sentence syntactically into subject–predicate form, which imposes a figureground organization on the meaning of the sentence that is otherwise absent with the predicate similar. Hence syntactic structure is contributing meaning over and above that derived from the lexical items in the sentence. Goldberg 1996, in a commentary on Jackendoff 1990, goes even further and suggests that all syntactic structures are at bottom meaningful constructions. (One might read Langacker 1987a in a similar spirit; Fillmore and Kay 1993, however, are less committal.) Jackendoff 1996a, 2002 disagrees, arguing that basic phrase structure, structural case marking, and agreement, for instance, are syntactically autonomous, while leaving open how much of what has traditionally been considered autonomous syntax (e.g. the middle and the passive) may yet prove to be constructional idioms. [See a bit more discussion in the remarks preceding Chapter 7.] From a traditional generative view, in which all principles of syntactic configuration are autonomous of semantics, the notion of a constructional idiom makes little sense. On the other hand, once one admits into the lexicon full VPs such as kick the bucket, it is a short step to admitting phrasal idioms with open argument places, for example take NP to task. A next step is to admit idioms which, instead of stipulating the verb and leaving an argument out,
twistin’ the night away
273
stipulate an argument and leave the verb free. Such is the way construction, which can be specified syntactically as (98a) and semantically (informally) as (98b). The idiomatic intensifiers such as his head off likewise can be specified syntactically as (99a) and semantically as (99b). Finally, the fixed X0 elements can be eliminated entirely, leaving an idiomatic skeleton like (100) for the resultative (see further development of this position in Chapter 9).3 (98)
a. [VP V [bound pronoun]’s way PP] b. ‘go PP (by) V-ing’
(99)
a. [VP V [bound pronoun]’s head off] b. ‘V intensely’
(100)
a. [VP V NP AP/PP] b. ‘cause NP to become AP/go PP by V-ing (it)’
In other words, a constructional idiom is treated by the grammar like a special kind of phrasal lexical item, whose meaning is learnable in the same way word meanings are learned. In these constructions, the construction rather than the verb determines the argument structure; the way the verb is integrated into the interpretation of the clause comes from the construction, not from the verb itself; and all the semantic peculiarities come from the meaning associated with the construction. In such a case, the NP in the resultative is the object of the construction, and hence the object of the VP—but not the object of the verb. What the way construction and the resultative share is that the meaning of the verb is treated semantically as a means or manner modifier. This accounts for the paraphrases of the way construction and the resultative as (98b) and (100b) respectively. That is, these two can be considered instances of a more general abstract construction type (101) which we might call the Verb Subordination Archi-Construction, from which they inherit their overall character. (101)
a. [VP V. . . ] b. ‘act (by) V-ing’
The use of sound emission verbs in motion contexts would be another instance of the same construction type: (95) means roughly ‘the trolley went across the
3
If constructions can be marked by special words such as way or by nothing at all, it seems only another short step to positing that they can also be marked by derivational morphology. Such a move might, for example, produce a constructional account of the Bantu applicative as an alternative to the lexical account of Bresnan and Moshi 1990; English re-prefixation and outprefixation seem other likely candidates.
274
twistin’ th e nigh t away
intersection rumbling’ [though see semantic differences discussed in section 9.3.3 and 9.4]. On the other hand, the beneficiary dative construction is not of this type: fix me a sandwich does not mean ‘give me a sandwich by fixing it’ but rather ‘fix a sandwich for my benefit’. Neither is the idiomatic intensifier construction (99), which does not mean ‘take pro’s head off (by) V-ing’. (See Goldberg 1995 and Michaelis and Lambrecht 1996 for discussion of inheritance hierarchies among constructions.h) In this light, the time-away construction emerges as yet another example of the Verb Subordination Archi-Construction, with the syntactic structure (102a) and the approximate meaning (102b). (102)
a. [VP V NP away] b. ‘waste [Time NP] V-ing’
This construction inherits properties not only from the Verb Subordination Archi-Construction, but also from the syntax of the verb-particle construction and [perhaps] from the semantics of the aspectual particle away. On the other hand, the fact that the construction has a number of peculiar semantic properties is a consequence of its being a lexical item of its own. The existence of verbs such as while and fritter that are happy only in this construction—and with which the construction has slightly broader selectional restrictions—is parallel to the verbs like wend and worm that occur only in the way construction. The constructional account of time-away thus claims that the direct object is licensed by the construction, and that the construction’s semantics determines the argument structure of the VP. As in the other instances of the Verb Subordinating Archi-Construction, the argument structure of the verb plays no role in licensing the syntax of the VP. This theory thus makes explicit the intuition that the NP is the object of the VP but not the object of the verb. The constructional account contrasts with the lexical rule approach, which claims that the verb itself undergoes a meaning shift in the construction. The
h The invocation of inheritance hierarchies does not tell us exactly how an inheritance hierarchy helps capture generalizations. The treatment of morphology in Chapter 2 suggests that one should think in terms of how much redundancy in individual VP constructions is accounted for by having them fall under an ArchiConstruction such as (101). Roughly, one could say that the cost of each individual VP construction such as the resultative is the information required to state it, minus the information in it that is predictable by virtue of (101), plus the cost of referring to (101). In turn, the cost of referring to (101) would be calculated in terms of what other interpretations might be assigned. These other interpretations would include, for instance, treatment of the object as a normal verb argument; treatment of the verb and object as together forming a light verb construction (e.g. take a walk to the park); and treatment of the object and PP as together forming a modifier of the verb, as in his head off. Such an approach parallels the measure of cost for morphological redundancy rules developed by Bochner 1993.
twistin’ the night away
275
constructional approach does not have to say, for example, that rumble is polysemous between ‘emit rumbly noise’ and ‘move while emitting rumbly noise’. Rather, rumble always means ‘emit rumbly noise’, and the ‘move’ part comes from the construction. Similarly, in the time-away construction, we do not have to say that drink is polysemous between ‘ingest fluid’ and ‘waste time ingesting fluid’: it always means ‘ingest fluid’, and the ‘waste time’ part comes from the construction. Levin and Rappaport Hovav 1995 discuss a different sort of constructional approach advocated by Dowty 1991 and Hoekstra and Mulder 1990: roughly, constructions can have meanings, and the core meaning of a verb can be underspecified and therefore be compatible with more than one construction. Such an approach may be conceivable for causative or unaccusative-unergative alternations, and it also seems appropriate for the sort of subject–predicate constructional meaning studied by Gleitman et al. However, it is less natural for the use of sound emission verbs in motion contexts—and especially for the use of activity verbs as verbs of wasting time. What would it mean to say that rumble is underspecified for movement and that drink is underspecified for both movement (so it occurs in the way construction) and wasting time (so it occurs in the time-away construction)? Levin and Rappaport Hovav, in offering several arguments against this approach, conclude by saying, ‘the question is whether multiple meanings [of verbs] are handled via principles or rules specific to the lexicon[,] or whether they can be shown to reduce to properties of syntactic configurations’ (208). Their statement, however, presupposes that the issue is multiple meanings of verbs. And though Levin and Rappaport Hovav cite Goldberg in connection with this argument, they do not discuss how her version of construction grammar (as well as mine) differs from the one they criticize, and how it offers another alternative: verb meanings remain the same in these constructions, and do not join with another constituent of VP to form a ‘complex verb’, but rather just play a different role than usual in the meaning of the VP. If the constructional account of time-away is correct, the various instances of the Verb Subordination Archi-Construction—as well as the idiomatic intensifier construction—counterexemplify the standard assumption that verb subcategorization always determines the complement structure of the VP.i Impressionistically, it does not seem a serious problem to incorporate constructions of this sort into
i —thereby creating a mismatch between semantic argument structure and syntactic argument structure. This violates Interface Uniformity and Fregean composition, but it is exactly what one would expect in the Parallel Architecture.
276
twistin’ th e nigh t away
HPSG and LFG; GB/Minimalism might have more difficulty. But I leave this enterprise up to practitioners of the theories in question. As discussed in Jackendoff 1990, the lexical rule and constructional approaches to these phenomena incorporate exactly the same information, and formally the lexical rules will look not unlike the constructions. Hence, considering these constructions alone, either approach can account for the evidence adequately. If there is to be a decision for one account or the other, it must rest on considerations outside the phenomena presented here. Two such considerations have already been hinted at, but let me conclude by pulling them together a bit more explicitly. The first is the independent necessity for sui generis constructions of the sort in (96). In such cases, a piece of syntax that has no independent uses in the grammar is associated with a distinct meaning. There is no obvious account of such phenomena in terms of lexical rules. Hence a notion of constructional meaning is necessary in UG in any event. The second consideration has to do with what is meant by a lexical rule. Two somewhat different notions have not been clearly distinguished in the literature. One type, for instance the relation between English singular and regular plural nouns, is completely productive aside from a finite class of listed irregularities. Hence given a new singular noun, this rule automatically permits one to infer its plural form and precisely what this form means. The other type is what I would like to call a semiproductive regularity. A good example is the formation of English denominal verbs such as saddle and butter. Given such a verb, one can predict approximately what it will mean—but not exactly. Saddling a horse requires putting the saddle in a particular place on the horse; and one cannot saddle a table by putting a saddle on it. Buttering toast cannot be accomplished by placing a stick of butter in the middle of the toast. And so forth.j Moreover it is necessary to learn whether any given noun has a denominal verb or not: bridle and mustard do not (in my lexicon at any rate), and I was surprised and baffled to hear my daughters speak of roofing a frisbee to refer to tossing it onto a roof. (For discussion, see Chapter 2; Jackendoff 1990, chapter 8. Jackendoff 1997 (n. 12 to chapter 5) and Culicover and Jackendoff 2005 (53–6) present arguments against Hale and Keyser’s (1993, 2002) approach to such verbs.)
j The semantics of these denominal verbs is a bit less mysterious in light of Pustejovsky’s (1995) work on the ‘generative lexicon’. Part of the semantic information in the lexical entry of an artifact word is its ‘telic quale’, which specifies the normal or designated use of an object (Millikan’s (1984) notion of ‘proper function’ does approximately the same work). Many denominal verbs denote the action performed on an object in its normal use. For instance, butter is spread on bread, and saddles are placed on horses’ backs in a particular position. However, this account of the semantics still does not explain which cases exist and which do not. See sections 13.4 and 13.6 for other uses of proper function.
twistin’ the night away
277
The upshot is that the ‘outputs’ of semiproductive lexical rules must be listed in long-term memory. But those of productive lexical rules do not: they can be composed ‘online’ in the course of sentence perception and production. That is, in terms of their role in the grammar, the productive rules more closely resemble the productive phrasal rules of the grammar, whose whole point is to avoid listing the indefinitely large number of regular phrasal constituents. What makes the productive lexical rules ‘lexical’ is only that they build constituents of word-size or smaller—not that they add new listings to the lexicon (where I take the lexicon to be a list stored in long-term memory; see section 1.7, Anshen and Aronoff 1988, Pinker and Prince 1991, and Jackendoff 1997b, 2002 for discussion of the productive–semiproductive distinction). Let us now go back to the putative lexical rule for the time-away construction. This is a productive rule, in that any verb that satisfies the construction’s selectional restrictions can participate, and the resultant meaning is fully predictable. Thus the ‘output’ need not be listed in the lexicon; it can be computed online. Moreover, section 8.4 showed that sleep/drink/doodle away are syntactically not verbs. Thus this lexical rule neither adds new items to the lexicon, nor does it build constituents of word size or smaller. Thus the only reason we need to call these items lexical verbs is in order to preserve the assumption that the lexical verb’s argument structure always determines the argument structure of the VP. Other than that, the notion of lexical rule has been stretched beyond recognition. The real issue, then, is this last assumption: in the end the distinction between the lexical rule and the constructional approach rests, as far as I can tell, only on this point. In the light of the need for constructions in any event, I am personally inclined toward the constructional approach, while acknowledging that this may not be a universal preference given the evidence so far. In any event, the time-away construction, aspectual away, and the idiomatic intensifiers can now be clearly added as new members to the family of interesting problems for the theory of argument structure. The fact that in many particulars they differ from two superficially very similar constructions, the way construction and the resultative, highlights the differences between these other two as well. The task for linguistic theory, then, is not to struggle to eliminate the need for such constructions. Rather, it is to discover the range of such constructions permitted by UG such that the child can acquire them.
chapter 9
The English Resultative as a Family of Constructions (2004) Adele E. Goldberg and Ray Jackendoff
9.1 A constructional view of grammar For fifteen years, the English resultative construction has been a focus of research on the syntax–semantics interface. Each of us has made proposals about the resultative (Goldberg 1991, 1995; Jackendoff 1990, 1997b), proposals that share a certain family resemblance. The present chapter is an attempt to consolidate what our approaches have in common and to add some new wrinkles to our common understanding. Our larger purpose is not only to show the virtues of our account of resultatives, but also to justify aspects of what we share in our overall vision of grammar, what we might call the ‘constructional’ view. To the
[This chapter originally appeared in Language 80 (2004), 532–68, and is reprinted here by permission of the Linguistic Society of America.] This chapter originated during RJ’s residency at the University of Illinois as the George A. Miller Visiting Professor in the winter of 2002, when he and AEG had extended opportunities to discuss issues of mutual interest. In particular, we found it important to establish some of the basic points of convergence between our two somewhat disparate approaches to grammar–AEG coming by way of Construction Grammar/Cognitive Grammar with a sprinkling of LFG and HPSG, RJ coming by way of a long path diverging gradually from the Chomskyan tradition. The resultative seemed like an interesting quarry. RJ is grateful for the support of the Miller Professorship, to Marc Hauser, who made an office at Harvard available during RJ’s sabbatical in 2002–3, and to Grant DC 03660 from the National Institutes of Health. We are grateful to two anonymous reviewers and to James McCloskey, Brian Joseph, and Malka Rappaport Hovav for comments on an earlier version of this manuscript.
t he e ngl is h r e s ul t ative
279
extent that our treatment of the resultative can be stated only within the constructional view, it serves as evidence for this view as a whole.a In the interest of being maximally provocative, let’s start by stating some aspects of the overall vision. (1) The Constructional View: a. There is a cline of grammatical phenomena from the totally general to the totally idiosyncratic. b. Everything on this cline is to be stated in a common format, from the most particular, such as individual words, to the most general, such as principles for verb position, with many subregularities in between. That is, there is no principled divide between ‘lexicon’ and ‘rules’. c. At the level of phrasal syntax, pieces of syntax connected to meaning in a conventionalized and partially idiosyncratic way are captured by constructions.1, b These tenets of the constructional view have been developed by each of us in different ways (Goldberg 1992, 1995, 1999, 2005a, 2006; Jackendoff 2002; Culicover and Jackendoff 2005); for other versions see, for example, Ackerman and Nikolaeva 2004, Barðdal 1999, Booij 2002, Croft 2001, Culicover 1999, Fillmore, Kay, and O’Connor 1988, Fried 2005, Kay 2002, Kay and Fillmore 1999, Lakoff 1987, Langacker 1987a, 1991, 1992, Michaelis and Lambrecht 1996, Sag 1997, Zwicky 1994; for a review of various positions that call themselves constructional, see Goldberg 2006, chapter 3. 1
Construction Grammar defines constructions to be any stored pairings of form and function [but see subfootnote b]; according to this definition, words and morphemes are technically constructions as well. In addition, stored (typically highly frequent) regularities between form and meaning are considered constructions even if they are fully compositional. We speak here specifically of partially idiosyncratic phrasal constructions, using the term in a fairly traditional way. a Although the focus of this chapter is on demonstrating the virtues of meaningful constructions as a component of grammar, it also represents the dawning realization (for me, anyway) that semiproductivity appears in syntax as well as in morphology. Goldberg and I were able to show that there are areas in which resultatives are semiproductive. But we were not able to give a detailed topography of the semiproductive parts of the construction, which appears to be extensive and tangled. The theme of semiproductivity in constructions returns, with greater force, in Chapter 12, where the NPN construction (day after day) is explored in detail; and, because of the more limited scope of the construction, it is possible to be more comprehensive. b As suggested in the remarks preceding Chapter 7, and toward the end of Chapter 7, I take a somewhat different view of constructions than Construction Grammar. For me, some constructions are pairings of form and function, i.e. syntactic and semantic structure. These include both words and meaningful constructions such as the way construction. However, other constructions are purely syntactic, for instance the English do-support construction. Moreover, conventionalized coercions such as the ‘Universal Packager’ (I’ll have three coffees—see Chapter 5) can be treated as semantic constructions with no syntactic consequences. Jackendoff 2002 (chapter 6) also suggests that phonotactic principles such as syllable structure rules can be treated as constructions involving only phonological structure.
280
t h e e ngl is h re s ul t ative
Constructional approaches can be distinguished from traditional lexicalist approaches in that the latter approaches emphasize the role of words (particularly lexical heads) in determining phrasal patterns (Bresnan 1982b, Hudson 1990, Pollard and Sag 1987, 1994). The present approach, on the other hand, expands this notion of the lexicon to include phrasal patterns with or without any morphological specifications in the mental lexicon. Some constructions are especially easy to spot because they have unusual syntax and/or bits of specified morphology, as indicated by italics in the examples given in (2). (2)
a. Our friends won’t buy this analysis, let alone the next one we propose. (let alone: Fillmore et al. 1988) b. One more pseudo-generalization and/or I’m giving up. (NP and/or S: Culicover and Jackendoff 1997, 2005) c. The more we study resultatives, the crazier they get. (comparative correlative: Culicover and Jackendoff 1999, 2005, Michaelis 1994) d. Day by day the facts are getting murkier. (NPN: Chapter 12, Williams 1994, Oehrle 1998)
Another type of construction involves garden-variety syntax, but there is some special meaning attached which imposes special restrictions. These are harder to detect, but we each have argued (Jackendoff 1990, Goldberg 1992, 1995) that the examples in (3) involve constructional meaning; (3b) is of course the resultative. (3)
a. I’ll fix you a drink. (Beneficiary ditransitive) b. Fred watered the plants flat. (Resultative)
In between these are constructions where a standard syntactic position is occupied by a special element that marks the construction: (4)
a. Bill belched his way out of the restaurant. (Way construction: Goldberg 1995, Jackendoff 1990) b. We’re twistin’ the night away. (Time-away construction: Chapter 8)
In each of these cases, some special interpretation is associated with the syntactic structure. For instance, the NP and/or S construction (2b) means, informally, ‘If some contextually determined event happens/doesn’t happen that involves NP, then S.’ The time-away construction (4b) means roughly ‘subject spends time frivolously doing V’. And our quarry here, the resultative construction (3b), means roughly ‘subject makes object become AP by V-ing it’. Constructions are like traditional idioms: they are listed in the lexicon with a syntactic structure, a meaning, and (where there is a special morpheme) a partial phonology. Like idioms such as take NP to task, constructions may have
t he e ngl is h r e s ul t ative
281
argument positions. For instance, the apparent object the night in (4b) is actually the object of the construction, not the object of the verb twistin’; and flat in the resultative (3b) is an argument of the construction, not of the verb water. An important innovation in the constructional view is that in VP constructions such as (3) and (4), the VP’s complement structure is not determined by the verb alone, as is assumed in most of mainstream generative grammar as well as in many functionalist traditions (e.g. Bresnan 1982b, Langacker 2003). On our view, argument structure is determined by the composite effects of the verb and the construction.2 One of the crucial issues in the constructional view is to work out how this composite is constructed. The essential point is that the verb does not change its meaning so as to license these extra arguments: for instance belch in (4a) is not ‘converted’ to a motion verb in the lexicon or anywhere else. Its contribution to the meaning of (4a) is the same as its contribution to the meaning of Bill belched loudly; it is, in both cases, a verb expressing bodily function and sound emission. The sense of motion and the sense of repeated belching in (4a) come from the construction and from the way the construction combines semantically with the verb to express a complex event. Similarly, fix in (3a) does not have a beneficiary role; rather this role comes from the ditransitive construction, and the verb remains its own boring self, a verb of creation. One benefit of this approach is considerable reduction in the apparent polysemy of verbs in the lexicon: the meaning of belch is the same in its various uses.c Moreover, Narasimhan 1998 demonstrates that many verbs that are translations of each other in different languages share the same semantic implications and aspectual properties and are used in the same types of discourse contexts. Nonetheless different languages allow the same verbal meanings to appear in distinct constructional contexts. Distinguishing verb meaning from the constructions a verb may appear in allows us to capture these facts. The approach we are suggesting comes at the following obvious cost: (i) we need to admit meaningful constructions as items stored in the lexicon, and (ii) we 2 Certain generative proposals such as Marantz 1997 and Borer 1994 likewise suggest that argument structure is not determined exclusively by the main verb. These accounts, however, differ in critical respects from the present proposal: (i) the syntax proposed is an underlying level of syntax and is claimed to be general across all constructions and all languages, (ii) the meaning involved is also proposed to be general across all constructions and all languages, and (iii) the verbs are claimed to contribute only their encyclopedic semantics without specifying anything at all about their argument structure possibilities. For a critique of this approach, see Goldberg 2006, chapter 4. c Note how this approach parallels the treatment of the plural in section 5.5. There the plural doesn’t change the semantic structure of a word, it just embeds it in a larger structure. Here the constructional meaning doesn’t change the meaning and argument structure of the verb, it just embeds these in further structure which then supplements or supplants the verb in determining the syntactic argument structure of the sentence.
282
t he e ngl is h r e s ul t a tive
need to abandon the rigid view that the verb alone determines the complement structure of its VP. We think the benefit is worth the price. Mainstream generative grammar does not recognize constructions in this sense; phenomena that have more standardly been termed ‘constructions’, such as the passive, are taken to be epiphenomenal outcomes of more general processes in the computational system. Thus a defender of the classical view might well complain, ‘By introducing so-called constructional principles, you’re making an arbitrary addition to the theory of grammar. A true explanation of the resultative would make use of mechanisms we already know.’ This complaint might be justified if the resultative were the only phenomenon that demanded a construction. However, as we see, there are many other such phenomena. Consider the way construction and the time-away construction, illustrated in (4). We simply see no way that general principles of syntactic structure and argument structure can predict that English has such constructions, with the meanings they have. Something has to be stipulated, to the effect that when way or away is present in a VP under the correct conditions, the proper interpretation emerges. Regardless of how you wish to do it, you will have to say something special. The constructional account says it directly, localizing the irregularity in the construction itself rather than elsewhere in the grammar. Von Stechow 1995, within a formal semantics framework, localizes it in a special rule of semantic interpretation that has effects quite similar to our constructions. In particular, because aspects of these constructions are so rare crosslinguistically, and on occasion peculiar to English, we would seriously question an attempt to characterize them in terms of parameter settings in the sense of Principles and Parameters theory. Several other Germanic languages, including Swedish, Norwegian, and Dutch, have a construction that means pretty much the same as the English way construction, but Swedish and Norwegian use a reflexive instead of X’s way, and Dutch uses a reflexive benefactive form instead of X’s way (Seland 2001, Toivonen 2002a, Verhagen 2002). These differences seem to us to be brute facts, stipulations that speakers of each language must learn. When we turn to constructions with abnormal syntax such as those in (2) [and those in Chapters 10–12], the need for English grammar to say something special about both form and interpretation seems obvious. The proposal, then, is to expand the role of the traditional lexicon to include productive or semi-productive phrasal patterns that have previously been assumed to lie within the domain of syntax. The point is that if indeed constructions in our sense are necessary in the theory of grammar, there can be no a priori objection to using them to account for the resultative. The resultative is unusual among the constructions mentioned here only in that it carries no overt marking of its constructional status, such as way or a weird syntactic structure. It is for this reason that practitioners
t he e ngl is h r e s ul t ative
283
of nonconstructional approaches have been able to hold out hope that the resultative can follow from general principles. We know of no attempts to capture the whole range of constructional phenomena in other frameworks; we hereby throw down the gauntlet. We cannot prove that no classical solution is possible, but we show what challenges a classical solution must meet.
9.2 Dimensions of variation in resultatives The resultative construction, unlike some of the other constructions mentioned above, shows a great deal of syntactic and semantic variation. The literature on the whole has treated resultatives as a unified phenomenon (‘the resultative’).d We think this is a mistake. In our own work (Goldberg 1991, 1995, Jackendoff 1990), we have treated resultatives as forming a sort of ‘family’ of constructions (we might call them ‘subconstructions’ of the resultative), sharing important properties but differing in certain specifics, including their degree of productivity.3 By ‘family’, we have in mind the sort of family resemblances recognized to exist in non-linguistic categories (Rosch and Mervis 1975, Wittgenstein 1955). Our strategy here is first to establish a taxonomy of the subconstructions, then to state which of their properties can be explained by the construction as a whole, and finally to pick out some of the differences. We begin by establishing some terminology. An identifying characteristic of a resultative sentence is an AP or PP that occupies the normal position of a verbal argument, for instance the italicized phrases in (5). We call this phrase the resultative phrase or RP. (5) a. Herman hammered the metal flat. b. The critics laughed the play off the stage.
[RP¼AP] [RP¼PP]
Resultatives must be distinguished from depictive or ‘current-state’ phrases which look superficially like resultatives (She handed him the towel wet), but which differ syntactically in that they are clear adjuncts, not argument phrases, and semantically in that they do not designate states that are contingent on the action described by the main verb (i.e. results). A resultative may contain a direct object, in which case the RP follows the object, as in (5); we call such cases transitive resultatives. Or a resultative may
3 Families of constructions are relevantly like morphological constellations (Janda and Joseph 1999, Joseph and Janda 1985); see also Valimaa-Blum 1988 for an extension of the notion of a constellation to the syntactic domain. d
Among the exceptions is Boas 2005, which we discuss in section 9.8.
284
t he e ngl is h r e s ul t ative
lack a direct object, in which case the RP is immediately after the verb, as in (6); we call these intransitive resultatives.4 (6) Intransitive resultatives a. The pond froze solid. b. Bill rolled out of the room.
[RP¼AP] [RP¼PP]
In some transitive resultatives, the direct object is independently selected by the verb; in others it is not. We refer to the former cases as selected transitive resultatives and the latter as unselected transitive resultatives. (7)
(8)
Selected transitive resultatives a. The gardener watered the flowers flat. [cf. The gardener watered the flowers.] b. Bill broke the bathtub into pieces. [cf. Bill broke the bathtub.] Unselected transitive resultatives a. They drank the pub dry. [cf. They drank the pub.] b. The professor talked us into a stupor. [cf. The professor talked us.]
[RP¼AP] [RP¼PP]
[RP¼AP] [RP¼PP]
A special case of unselected transitive resultatives has a reflexive object that cannot alternate with other NPs. This is often called a ‘fake reflexive’ (Simpson 1983). (9)
Fake reflexive resultatives a. We yelled ourselves hoarse. [RP¼AP] Unselected: We yelled ourselves. Does not alternate with other NPs: We yelled Harry hoarse. b. Harry coughed himself into insensibility. [RP¼PP] Unselected: Harry coughed himself. Does not alternate with other NPs: Harry coughed us into insensibility.
When RP¼AP, it normally expresses a property. Some cases of RP¼PP also have this semantics, for example (7b), (8b), and (9b). We refer to these two 4 We set aside here passive and middle resultatives, such as The metal was hammered flat and This metal hammers flat easily. We take it that these expressions are formed by composing the passive and middle constructions with resultative constructions.e e To my knowledge, it is not yet understood how a middle construction composes with a resultative (though see Carrier and Randall 1992). The composition of the passive with the resultative can be accomplished using mechanisms in Culicover and Jackendoff 2005.
t he e ngl is h r e s ul t ative
285
types collectively as property resultatives; when we need to differentiate them we refer to AP property resultatives and PP property resultatives. In other sentences with RP¼PP, such as (5b) and (6b), the RP expresses a spatial path; we call such sentences PP spatial resultatives. There are arguably some adjectives that express spatial paths or configurations, such as free and clear; these may appear in a class of AP spatial resultatives such as He jumped clear of the traffic (see section 9.8). To sum up, there are so far three independent dimensions of variation in resultative sentences, the third of which is further subdivided. (10)
a. RP¼AP vs. RP¼PP b. RP¼property vs. RP¼spatial configuration c. Intransitive vs. transitive i. Within transitive: selected vs. unselected 1. Within unselected: normal vs. fake reflexive
These dimensions are all well known. We must however add one more. One of the NPs in the sentence is normally understood as undergoing a motion or a change of state whose endpoint is expressed by the RP; we call this the host of the RP. Normally the host of a transitive resultative is the object, for instance the metal in (5a) and the play in (5b). And normally the host of an intransitive resultative is the subject, for instance the pond in (6a) and Bill in (6b). That is, the choice of host appears to correlate exactly with transitivity and therefore does not constitute an independent dimension of variation. However, (11) shows a class of transitive examples that has newly emerged in the literature (Rappaport Hovav and Levin 2001, Verspoor 1997, Wechsler 1997) in which the host is the subject. (11)
Transitive resultatives with subject host a. Bill followed the road into the forest. b. We drove Highway 5 from San Diego to Santa Fe. c. The sailors rode the breeze clear of the rocks. d. Fred tracked the leak to its source. e. John danced mazurkas across the room.
A distinct subclass is shown in (12), involving verbs of bodily emission (12a, b) (Goldberg 2005a); we can also add verbs of other sorts of substance emission (12c) and verbs of ingestion (12d, e). In each of these types, the entity in motion is not overtly expressed in the sentence; the host is therefore an implicit argument.5 5
For some of these, one might argue that the host is a deleted cognate object, as in He spit spit onto the floor. But this is not true of all of them: the noun cough, for instance, denotes the act of coughing, not what is coughed up, so She coughed a cough into the sink is bad.
286
t he e ngl is h r e s ul t ative
(12) Intransitive resultatives with implicit (i.e. non-subject) host a. Bill spit/urinated/coughed/sneezed out the window. b. Bill sweated/bled on the floor. c. The toilet leaked through the floor into the kitchen below. d. Bill ate off the floor. e. Bill drank from the hose. Thus the choice of host proves to be an additional independent dimension in the resultative construction.
9.3 The semantics of the resultative In this section we propose structures for the meanings of members of the resultative family, and in this section and the next we show how these meanings predict many of the properties of resultative sentences. As hinted in section 9.1, the basic intuition behind our constructional approach to resultatives is that the meaning of a resultative sentence contains two separable subevents. One of them, the verbal subevent, is determined by the verb of the sentence. The other subevent, the constructional subevent, is determined by the construction. A resultative sentence means more than just the conjunction of the verbal subevent and the constructional subevent. For instance, Willy watered the plants flat does not mean just that Willy both made the plants flat and watered them. Rather, the two subevents are related: Willy made the plants flat by watering them. That is, for the bulk of cases (we discuss one class of exceptions below), the verbal subevent is the means by which the constructional subevent takes place. This paraphrase also shows the distribution of arguments between the two subevents: Willy is the agent of both subevents, the plants is the patient of both subevents, and flat is the resulting property in the constructional subevent. In sections 9.4–5 we show that many properties of the resultative construction follow from the semantics of the two subevents and the connection between them. Using an informal, more or less common-practice semantic notation, we can express the meaning of Willy watered the plants flat as follows:6 (13) Syntax: Willy watered the plants flat: Semantics: WILLY CAUSE [PLANTS BECOME FLAT] MEANS: WILLY WATER PLANTS 6
For ease of exposition, we adopt notation that is neutral between our distinct formalizations.
t he e ngl is h r e s ul t ative
287
If we extract from this example its particular arguments (WILLY, PLANTS, FLAT) and the particular verbal subevent (WILLY WATER PLANTS), we can discern the semantic contribution of the construction itself, shown in (14). The subscripts in (14) indicate the correspondences between the syntactic arguments and the corresponding semantic arguments, correspondences that are discussed more fully in section 9.6.f (14)
Causative property resultative Syntax: NP1 V NP2 AP3 Semantics: X1 CAUSE [Y2 BECOME Z3] MEANS: [verbal subevent ]
With what we have done so far, we can see already that the semantics of the constructional subevent predicts the syntax of the construction. Consider actual verbs that have the meaning ‘X cause Y to become Z’, for instance make and get. These show exactly the same distribution of arguments in syntax as the construction: Willy made/got the plants flat. That is, whatever generalizations are responsible for mapping the semantics of these ‘intrinsically resultative’ verbs to their syntactic argument structure can also be used to map the constructional subevent to its syntactic argument structure.7 (15) The semantic argument structure of the constructional subevent determines the syntactic argument structure of the sentence by general principles of argument linking.8 This is of course no accident—productive constructions probably arise from speakers generalizing over verbs that lexically specify the corresponding forms 7
AEG would amend each of the representations in (14/15) to include INSTANCE as well as the MEANS relations to indicate that verbs that are lexical resultatives such as make, render, become are instances of the construction as well; in fact all overt expression is accomplished via constructions. RJ prefers to treat these verbs as lexically determining their syntax via more traditional linking rules. 8 Our formulation here abstracts away from any particular approach: whatever means one chooses to account for the regular mapping between syntactic and semantic arguments can be applied to the constructional approach as easily as to a completely verb-driven approach. f
In the notation of Conceptual Semantics, this comes out as follows:
(14’)
Causative property resultative Syntax: NP1 V NP2 AP3 Semantics: [CAUSE (X1, [BECOME (Y2, Z3) ] ) ]a; [CAUSE ([verbal subevent ], a) ]
The first line of the semantics is the constructional subevent. The semicolon at the end of the line indicates that the following material is a modifier. The superscript a indicates that this constituent binds the argument notated by a in the second line. The second line, the modifier of the constructional subevent, says that the verbal subevent causes a, which I take to be the meaning of a means modifier.
288
t he e ngl is h r e s ul t ative
and meanings (Casenhiser and Goldberg 2005, Goldberg, Casenhiser, and Sethuraman 2004). Expressions involving verbs that are inherently resultative (let us call these verbal resultatives), often have broader or narrower selectional properties than the general construction. For example, make allows predicate NPs as well as APs: They made him president/angry.9 We can see this is a verbal resultative because the ‘means’ paraphrase, They caused him to become angry by making him, makes no sense. Similarly, drive, as in drive NP crazy, allows only a range of APs and PPs that all refer to demented mental states: nuts, bananas, to distraction, into a tizzy, and so on (Goldberg 1995; Boas 2000). Again, the ‘means’ paraphrase is out: He caused me to become crazy by driving me. This is to be expected if the verbal resultatives are stored as resultatives: as stored patterns, they may acquire their own idiosyncratic properties. In fact, verbs that lexically specify any particular construction generally tend to be somewhat broader or narrower in distribution than productive instances of the pattern. For example, the prototypical intrinsically ditransitive verb, give, has a broader semantic distribution than constructional ditransitives such as throw him the ball.10, g A referee suggests that it would be preferable to derive semantics from syntax, instead of allowing the semantics to predict the syntax as we do here. If the syntactic structure and the semantic structure were isomorphic, it would be impossible to decide between the proposals, but we demonstrate below that there are semantic properties that are independently required, since they do not follow from syntax. That is, the semantic properties are richer, more nuanced— syntax can be mapped from them, but not vice versa. For example, there is more to say about the semantics of the constructional subevent. In particular, there are two dimensions of variation: property vs. path resultatives and noncausative vs. causative resultatives.
9.3.1 Property vs. Path resultatives In property resultatives, the constructional subevent consists in the host coming to have the property expressed by the RP. So, for example, in She watered the 9 Boas (2000) notes that make cannot appear as a resultative with a PP¼RP (They made him to exhaustion). However, it may be possible to treat examples like They made him into a monster as a type of verbal resultative. 10 Many languages, including Japanese and French, only allow verbal resultatives and do not use the construction productively with other verbs (Takami MS; Washio 1997). g This point was also observed in Chapter 8, where the verbs fritter away and piss away, unlike the timeaway construction, were shown to allow resources other than time in object position: She frittered/golfed 3 hours away; She frittered/golfed $100 away.
t he e ngl is h r e s ul t ative
289
plants flat, the plants come to be flat, and in The pond froze solid, the pond comes to be solid. In path resultatives, the constructional subevent normally consists of the host traversing the path expressed by the RP (we discuss a variation with stative verbs in section 9.4.1). For instance, in Bill rolled the ball down the hill, the ball travels down the hill. In the follow-type examples (11), such as Bill followed the leak to its source, the host is subject and travels a path terminating at the source of the leak. In the spit-type examples (12), such as Sue spit out the window, the host is the implicit argument, the spit, and it traverses a path that goes out the window. We leave the spit-type and follow-types aside until section 9.7.
9.3.2 Noncausative vs. causative resultatives When the host is the subject, the constructional subevent is simply a change of state or change of position. When the host is the direct object, the constructional subevent consists in the subject causing the host to do what it does. Example (13) is a causative transitive property resultative, with RP¼AP. The other major subconstructions are shown in (16).h (16)
a. Noncausative property resultative (e.g. The pond froze solid) Syntax: NP1 V AP/PP2 Semantics: X1 BECOME Y2 MEANS: [verbal subevent ]11 b. Noncausative path resultative (‘intransitive motion construction’, e.g. The ball rolled down the hill) Syntax: NP1 V PP2 Semantics: X1 GO Path2 MEANS: [verbal subevent ]
11
In the case of this construction, the constructional subevent is noncausative while the relationship between the verbal subevent and constructional subevent is one of causation, as indicated by MEANS. h
Renotating (16):
(16’) a. Noncausative property resultative Syntax: NP1 V AP/PP2 Semantics: [BECOME (X1, Y2)]a; [CAUSE ( [verbal subevent], a) ] b. Noncausative path resultative (intransitive motion construction) Syntax: NP1 V PP2 Semantics: [GO (X1, PATH2)]a; [CAUSE ( [verbal subevent], a) ] c. Causative path resultative (caused motion construction) Syntax: NP1 V NP2 PP3 Semantics: [CAUSE (X1, [GO (Y2, PATH3)] ) ]a; [CAUSE ( [verbal subevent], a) ]
290
t he e ngl is h r e s ul t ative
c. Causative path resultative (‘caused motion construction’, e.g. Bill rolled the ball down the hill) Syntax: NP1 V NP2 PP3 Semantics: X1 CAUSE [Y2 GO Path3] MEANS: [verbal subevent]
9.3.3 Soundþmotion and disappearance resultatives A different relation between the verbal and constructional subevents appears in (17). This is a class of intransitive path resultatives in English in which the verb expresses emission of a sound (Levin and Rappaport Hovav 1990): (17) a. b. c. d.
The trolley rumbled through the tunnel. The wagon creaked down the road. The bullets whistled past the house. The knee-replacement candidate groaned up the stairs.
Here, as in The ball rolled down the hill, the verbal subevent is the subject performing the action expressed by the verb—the trolley is rumbling—and the constructional subevent is the subject moving along the path expressed by the PP—the trolley moves through the tunnel. However, the meaning of the sentence is not that the verbal subevent is the means by which the constructional subevent takes place (e.g. the rolling is the means of moving down the hill), but rather that the motion causes the sound to be emitted: the rumbling is a result of the trolley’s motion. To see this difference more clearly, consider examples like (18), which we find ungrammatical: (18) a. The car honked down the road. b. The dog barked out of the room. c. Bill whistled past the house. These are evidently out because the sound is not a result of the subject’s motion. The car’s honking and the dog’s barking are separate actions from their motion. In particular, compare (17c) and (18c): the whistling noise is a result of the bullet’s motion in (17c), whereas Bill’s whistling in (18c) is a separate volitional act. In (17d) and (18a) we see that what is critical is not whether the entity that emits the sound is animate, but rather that the sound must be causally related to the motion. In (17d) we construe the motion to cause the painful groaning; in (18a) we recognize the honking as a sound not caused by the motion of the car. This is not to say that the subevents in (18) cannot be combined into a single
t he e ngl is h r e s ul t ative
291
clause: the way construction allows the subevents to be simply simultaneous rather than causally related: (19)
a. The car honked its way down the road. b. The dog barked its way out of the room. c. Bill whistled his way past the house.
We conclude that sound-emission resultatives are not licensed by principle (16b), though they share the same syntactic form and the same two subevents. Rather they are a consequence of principle (20), another subconstruction of the resultative. (We recognize that there are speakers who accept (18). For them the semantic relation is COOCCURRENCE rather than RESULT.)i (20)
Sound-emission path resultative Syntax: NP1 V PP2 Semantics: X1 GO Path2 RESULT: [verbal subevent: X1 emit sound]
Note that there is a selectional restriction on the verbal subevent, which constrains the construction to sound-emission verbs. Even the closely related class of light-emission verbs is prohibited. For instance, suppose the trolley in (17a) is emitting sparks from its wheels as it rumbles through the tunnel. We still cannot say The trolley sparked through the tunnel; again the more liberal way construction, The trolley sparked its way through the tunnel is all right.12 Another tiny class of verbal subevents forms resultatives with the pattern (20). (21) gives a couple of examples. Note that an appropriate paraphrase of (21a) is not (22a) but rather (22b). (21)
a. The witch vanished into the forest. b. Bill disappeared down the road.
12 Malka Rappaport Hovav (p.c., June 2003) notes that light-emission verbs are sometimes possible. In fact, the following type of examples are fully acceptable:
(i) The light flashed across the sky.
However, when light flashes, the flashes are not emitted by the light. Rather, flashing is the MEANS of the light’s motion. As stated in (16b), the MEANS relation extends to all verb classes. What appear not to be conventional are instances in which a verb that designates light emission is used as an incidental side-effect or result of the motion. i
Renotating (20):
(20’)
Sound-emission path resultative Syntax: NP1 V PP2 Semantics: [GO (X1, PATH2) ]a; [CAUSE (a, [verbal subevent] ) ]
The difference between this and (16’b) is the reversal of cause and effect in the modifier.
292
t he e ngl is h r e s ul t ative
(22) a. The witch went into the forest by vanishing. b. The witch went into the forest and thereby vanished. In short, sound-emission and disappearance resultatives involve a result relation between the constructional subevent and the verbal subevent instead of a means relation. There exist generalizations about what types of semantic relationships are more or less likely to be allowed between verbal and constructional subevents, within and across languages (Goldberg 1997). But we do not think that the fact that resultatives allow the verbal event to be either means or result is entailed by any general considerations of well-formedness of meaning, given that the way construction readily allows verbs of sound emission in particular to appear without being interpreted as standing in a means or result relation to the motion (as seen in (19)). Therefore, for us this is a crucial point where the resultative begins to fragment into a number of distinct subconstructions with similar syntax, similar argument structure, and similar constructional subevents, and where each subconstruction has its own peculiarities of interpretation and its own selectional restrictions. Moreover, we see no way to semantically unify events of sound emission with events of disappearance. Note for instance that events of disappearance cannot be used in the way construction: The witch vanished her way into the forest. Thus we reluctantly conclude that the selectional restriction on the choice of verbal subevent in (20) is a raw disjunction: there is no explanation to be sought at this level of detail, there is only description. We return to this issue in section 9.8. What makes the (sub)constructions in (14)/(16)/(20) form a ‘family’ is their family resemblance in syntax and semantics. The constructional subevents in the two transitive cases are causative and have arguments in parallel positions; the two property cases involve change of property; the three spatial cases involve motion in space.13 Apart from the sound-emission and disappearance verb classes, the verbal subevent is a means to achieving the constructional subevent. Unlike the way construction and time-away construction, none of them has a special morpheme that marks the construction. The fact that instances of grammatical patterns share family resemblances and not definitional necessary and sufficient features is highly reminiscent of other sorts of
13 Goldberg (1991, 1995) argues that the property resultatives are related to the path resultatives by metaphorical extension insofar as the RP is interpreted as a metaphorical goal. This predicts that AP resultatives are based on PP resultatives and that we should not expect to find languages which have only AP resultatives and not PP resultatives. Jackendoff (1990) treats them as parallel instantiations of thematic structure, differing in semantic field feature.
t he e ngl is h r e s ul t ative
293
conceptual categories (Jackendoff 1983, Lakoff 1987, Murphy 2002, Rosch and Mervis 1975, Taylor 1995, Wittgenstein 1955). We therefore do not see it as a drawback that our analysis requires several different but related types of resultatives, but rather as an indication of a parallel with general cognition.j
9.4 The meaning of resultatives explains their aspectual properties We now turn from the argument structure properties of resultatives to their aspectual and temporal properties.
9.4.1 Telic, atelic, and stative resultatives There seems to be a widespread view in the literature that resultative sentences are invariably telic (e.g. Boas 2000, Rappaport Hovav and Levin 2001, Rothstein 2000, Tenny 1994, 36–7, Van Valin 1990, Wechsler 2001). In fact, the term resultative suggests that the sentence expresses the result of some action. However, this view is false, for three reasons. First, although APs used with predicates of change normally denote the end state of change (Goldberg 1991, 1995, Wechsler 2001), the AP constructions A-er and A-er and ever A-er do not:14 For hours, the mixture got hotter and hotter. When serving as RP, these APs create atelic resultatives: (23) Non-end-bounded state of change denoted by an AP, atelic resultatives: a. For hours, Bill heated the mixture hotter and hotter. [non-repetitive] b. For hours, Bill hammered the metal ever flatter. [non-repetitive] c. For years, Penelope wove the shawl longer and longer. [non-repetitive]
14
It is possible to demonstrate that the A-er and A-er pattern requires a special construction. Note first that usually an adjective cannot conjoin with itself (i); only comparative adjectives can. Second, conjoined comparatives can usually appear as the predicate of either states or events (ii), but A-er and A-er can appear only as the predicate of an event (iii). (i) The building is tall and wide/tall and tall. (ii) This building got/is taller and wider than that one. (iii) This building got/is taller and taller (than that one).
Third, with longer adjectives we get more and more beautiful; ordinary conjoined comparatives come out differently: more beautiful and (more) exciting. Thus at bottom the construction appears to be a reduplication of the comparative morpheme. The meaning involves some sort of quantification over successive stages, and is paralleled by the equally idiosyncratic ever taller, ever more beautiful. j Note that this is a subcase of the argument for semiproductive rules as opposed to mere lexical associations. See section 1.7.
294
t he e ngl is h r e s ul t ative
Second, path resultatives are telic if and only if the RP is end-bounded (in the sense of [Chapter 6]):15 (24) End-bounded spatial PPs, telic resultatives: a. Bill floated into the cave (for hours [on non-repetitive reading]) b. Bill pushed Harry off the sofa (for hours [on non-repetitive reading]) Non-end-bounded spatial PPs, atelic resultatives: c. Bill floated down the river (for hours [non-repetitive]) d. Bill pushed Harry along the trail (for hours [non-repetitive]) Third, there exist stative sentences that are indistinguishable from path resultatives in both syntactic structure and argument structure properties (see Chapter 6 for a formal account). (25) Stative path resultatives a. The road zigzagged down the hill. b. The rope stretched over the pulley. c. The weights stretched the rope over the pulley. (25a) is paralleled by the typical intransitive path resultative Barbara zigzagged down the hill, but it is not a motion sentence. (25b, c) can be construed either as change of state (e.g. the rope gradually stretches)—in which case they fall under standard resultatives—or as continued states of tension in the rope. The latter interpretation is not even an event, much less a telic event. Yet one would like to treat the argument structure properties of both interpretations in terms of a common solution. Thus it appears necessary to admit not only atelic resultatives, but stative resultatives. Let us see how these data follow immediately from the treatment of resultative semantics in (14)/(16). It has often been observed in the literature that the telicity of motion events correlates with (among other things) the end-boundedness of the path of motion (Chapter 6 and references therein). For instance, the PP into the room expresses a path that terminates in the room; and John went into the room expresses a telic event, i.e. one that comes to its conclusion when John is in the room. By contrast, the PP along the river expresses a path whose end is not specific; and John went along the river expresses an atelic event, i.e. one whose termination is not specific. This correlation is brought out explicitly in the formalism of motion and change in Chapter 6: in this formalism, the timecourse of an event of change is homomorphic to the structure of the path of change, in particular preserving end-boundedness. In turn, the telicity of an event amounts precisely to the end-boundedness of its time-course. 15 We abstract away here from factors such as indefinite plural subjects, which can independently induce atelicity (People walked in for hours). See Chapter 5.
t he e ngl is h r e s ul t ative
295
Now consider the account of the semantics of path resultatives in (16). The constructional subevent is the ‘main event’ in the semantic structure. As a consequence, its temporal structure determines the telicity of the sentence. Its final argument, the RP, corresponds to the path of motion. Since the end-boundedness of the RP determines the telicity of the constructional subevent, and the telicity of the constructional subevent determines the telicity of a resultative sentence, we derive directly the correlation observed in (24). Similarly, expressions like hotter and hotter arguably denote an unbounded path of change in the ‘hot direction’; hence when used as RP they too will result in atelic sentences like (23). Next let us consider the statives. In general, a stative or ‘extension’ interpretation of an erstwhile motion verb is possible when the theme is an object idealized as extended in one dimension, i.e. as an ‘elaborated line’. The meaning of the ‘extension’ interpretation is that the theme, rather than moving along the path, occupies the entire extent of the path.16 For example, the road goes along the river asserts not that the road travels but that the road occupies a linear space parallel to and nearby the river. In the formalism of Chapter 6, this is because in an extension interpretation, the theme rather than the time-course is homomorphic to the structure of the path, and there is no inherent time-course. In turn, the absence of an inherent time-course is what makes a sentence stative.17 The stative interpretations in (25) fall under this generalization: the themes are the road and the rope, and these are asserted to be extended over the paths expressed by down the hill and over the pulley respectively. This is exactly what is expected under the semantic analysis in (16): like ordinary verbs of motion, the resultative construction can be coerced into an extension reading, given the right choice of subject. Some other aspects of the extension interpretation are worthy of note. First, the verb in a path resultative often expresses manner of motion. But this cannot be true of zigzag in (25a), since there is no motion. Rather, the verb expresses the detailed shape of the path: at the coarsest idealization, the path just extends down the hill; but at the next level of detail it has a zigzag shape. This analysis can actually be applied just as well to the motion interpretation: zigzagging consists in traversing a zigzag-shaped path. Similarly, in (25b, c) the verb is 16
This interpretation requires some suitable level of abstraction. For example, the telephone poles run alongside the road means not that the road’s edge is covered with telephone poles, but that the poles form a ‘virtual line’ that extends along the road. See Talmy 2000a, Jackendoff 1990, chapter 6; Langacker 1987a. 17 Talmy (2000) treats such sentences as ‘fictive motion’, as though one imagines oneself or one’s gaze traveling the road and thereby traveling along the river. While this interpretation is intuitively seductive, a full semantic analysis still must explain why sentences expressing extension are stative rather than eventive, as far as we know an issue not addressed by Talmy. The stativity is crucial to our analysis here. [2010: It is especially hard to interpret (25c) in terms of fictive motion.]
296
t he e ngl is h r e s ul t ative
interpreted as a manner of motion (or better, manner of adopting a shape), brought about by application of force. But on the extension interpretation it is interpreted as maintenance of shape under application of force. In other words, when the resultative sentence becomes stative, the verbal subevent must be interpreted statively as well.18 In addition, consider the transitive stative resultative in (25c). Causation in this interpretation does not involve change but rather forced maintenance of a state—just like the causation with ordinary verbs in sentences like (26). (26) a. The weights made the rope stretch over the pulley. b. The weights kept the rope stretched over the pulley. The overall generalization is: (27) The aspectual structure of the constructional subevent controls the aspectual structure of the resultative sentence; in turn, the aspectual structure of the constructional subevent is predictable from general principles correlating event structure with change, extension, motion, and paths. Thus our account of resultatives is in this respect also explanatory.
9.4.2 Temporal relation of the two subevents Next let us turn to the temporal relation between the two subevents of the resultative. On the analysis in (14)/(16), the verbal subevent is a means toward the constructional subevent. To do X by means of doing Y, one cannot do X first and then do Y. Thus this correctly predicts that the constructional subevent may not entirely precede the verbal subevent when the latter is interpreted as the means of effecting the former. This prediction leaves open whether the verbal subevent is concurrent with the constructional subevent, overlaps with it, or entirely precedes it. All of these options are possible with means expressions in general, depending on the pragmatics of the situation. (28) provides examples. (28) a. Cause cotemporal with effect: Bill made the ball go up the hill by pushing it. b. Cause overlaps with inception of effect: Bill made the vase fall on the floor by knocking against it. c. Cause completely precedes effect: Bill made himself get sick on Tuesday by eating mushrooms on Monday. 18
The relation between the two subevents must be reinterpreted as well, as ‘means’ is normally a relation between two events rather than two states.
t h e e ngl is h re s ul t a tive
297
The first two are clearly possible with transitive resultatives, as predicted: (29)
a. Cause cotemporal with effect: Bill pushed the ball up the hill. b. Cause overlaps with inception of effect: Bill knocked the vase onto the floor.
It is not clear whether the temporal relation in (28c), in which the means entirely precedes the main event, is an option for resultatives. Goldberg (1995, 194) and Rappaport Hovav and Levin (2001) observe that a temporal delay is not available with transitive resultatives generally. For example, it is not possible to utter (30a) to mean that Sam cut his finger, which caused his captors to release him in order to clean him up. Such a sense is possible in (30b), for instance. Instead, (30a) must mean that the cutting directly and immediately released him from his (implicit) restraints. (30)
a. Sam cut himself free. b. Sam got free (at 6:00) by cutting his finger (at 4:00).
This restriction is in fact expected. Fodor (1970) observes that biclausal causatives permit a means expression in which the means entirely precedes the main event. This is the case in (28c), for instance, as well as examples like (31a). However, Fodor points out that monoclausal causatives such as (31b) do not permit such a temporal relation (for possible reasons, see McCawley 1978, Jackendoff 1990, 150–1, Bohnemeyer et al. 2007). (31)
a. Sue made Bill die on Thursday by poisoning his breakfast on Wednesday. b. Sue killed Bill on Thursday by poisoning his breakfast on Wednesday.
On the analysis in (14)/(16), resultatives are syntactically monoclausal, and the verbal subevent functions semantically like a means expression. Hence, whatever the reason for the temporal constraint on means expressions in monoclausal sentences, this constraint ought to apply to resultatives as well. Rappaport Hovav and Levin (2001) suggest that it is possible to have a time delay with the unselected object cases, citing the following example. (32)
Sam sang enthusiastically during the class play. He woke up hoarse the next day and said, ‘Well, I guess I’ve sung myself hoarse.’ (Rappaport Hovav and Levin 2001, 775)
In the case of unselected objects, the verb by definition does not have a patient argument itself—for example, in (32) there is no entity that the singing actually
298
t he e ngl is h r e s ul t ative
directly acts upon. This would seem to allow for instances in which the action designated by the verb does not cause the result directly or immediately. Still, we feel that (32) is interpreted to mean that Sam only notices his throat becoming hoarse after a time delay. It is still inferable that the injury to his throat had been initiated by the time the singing ended.19 Evidence for the predictive power of our analysis comes from a consideration of sound-emission resultatives. Note for instance that in The door banged open, the banging can be at the end of the opening, say as the door strikes the adjacent wall. If the temporality restrictions observed above were a stipulation on resultatives, rather than falling out of the independently needed facts about the semantic relationship between verbal and constructional subevents, this outcome would not be predicted. However, this possibility follows automatically from (20), our treatment of the semantics of sound-emission resultatives. Here, unlike other resultatives, the verbal subevent (banging) is a result of the constructional subevent (coming open) rather than a means to it. Thus the general constraint on causation requires that the banging cannot precede the opening, precisely the correct outcome. To summarize, we find that the temporal relations between the verbal subevent and the constructional subevent are predicted by three independently necessary factors: (i) the semantic relation between the two subevents, whether it be means or result, (ii) our pragmatic world knowledge of the particular subevents in question, (iii) the strong tendency to interpret means expressions in monoclausal events as cotemporal. In short, (33) The semantics and syntax of resultatives explain the possibilities for temporal relations between the two subevents.
9.5 World knowledge relating to the semantics of the resultative explains additional distributional facts We have already seen that the semantic properties of resultatives explain much of their syntax and their aspectual/temporal properties. The distribution of resultatives is also constrained by general effects of world knowledge as it relates to the semantics of resultatives. For example, our taxonomy in section 9.2 distinguished transitives with fake reflexives from other unselected transi-
19
See Pustejovsky 1995 for further discussion and formalization of various possible temporal relations among subevents.
t he e ngl is h r e s ul t ative
299
tive resultatives: example (34a) has no counterpart with a non-reflexive object (cf. 34b), while the examples in (35a) do (cf. 35b). (34)
a. Bill cried himself to sleep. [fake reflexive] b. Bill cried Sue to sleep.
(35)
a. Bill wiggled himself through the hole. [ordinary reflexive] Bill sang himself to sleep. b. Bill wiggled the puppet through the hole. Bill sang Sue to sleep.
However, the distinction between fake and true reflexives requires no syntactic stipulation, because it arises from our world knowledge of what is likely to cause what. It’s hard to imagine making someone else go to sleep by crying (the putative meaning of (34b)); whereas it is of course possible to make something else wiggle someplace and to make someone else go to sleep by singing (the meanings of 35b). Thus there seems little reason to distinguish fake reflexives as a syntactically special case of resultatives; the semantic requirement that the verbal subevent must be construed to be the means of effecting the event designated by the construction requires such a causal relationship be pragmatically plausible. Another contrast that falls out from the relation between the subevents concerns the following examples: (36)
a. The elevator creaked itself to the ground. b. If the telephone rang, it could ring itself silly. (Rappaport Hovav and Levin 2001, ex. 34b)
The sound-emission resultative (20) applies only to intransitive sentences, as in The elevator creaked to the ground. Hence, despite the fact that the verbs in (36) are sound-emission verbs, it cannot apply here. What possibilities are left? The verb creak cannot be used causatively: Bill creaked the elevator. Therefore, the only applicable resultative pattern for (36a) is the unselected transitive resultative, which would have to mean ‘the elevator made itself go to the ground by creaking’, which is pragmatically implausible. The same pattern applied to (36b) is pragmatically plausible, under the metaphorical interpretation in which the phone is animate and therefore can ‘get silly’: ‘the phone could make itself silly by ringing.’ Alternatively, since ring can also be used causatively (Bill rang the doorbell), the selected transitive pattern is available as well: ‘the phone could make itself silly by ringing itself.’ Both these interpretations seem plausible. An interpretation that does not emerge is ‘the elevator made itself go to the ground, thereby emitting creaks’, ‘the telephone could make itself get
300
t he e ngl is h r e s ul t ative
silly, thereby emitting rings’, a plausible but non-existent extension of the sound-emission resultative to transitive frames.
9.6 How arguments are shared between the two subevents So far we have ignored the effects of the verbal subevent on the resultative, other than the fact that it supplies the overt verb in the syntax. Now we return to an issue set aside in section 9.3: how the semantic arguments of the verb are related to the semantic arguments of the construction.
9.6.1 Full Argument Realization Let us turn more directly to the issue of argument sharing between the verbal and constructional subevents. Our hypothesis can be stated as (37): (37) Full Argument Realization (FAR) All the arguments obligatorily licensed by the verb and all the syntactic arguments licensed by the construction must be simultaneously realized in the syntax, sharing syntactic positions if necessary in order to achieve well-formedness.20 An argument is considered obligatorily licensed by a verb if and only if an expression involving the verb in active, simple past tense without the argument is ill-formed. For example, drink’s second argument is not obligatory because (38) is well-formed; break’s patient argument, on the other hand, is obligatory, because (39) is ill-formed: (38) (39)
She drank. She broke. [in the sense ‘she broke it/something’]
The two of us differ on the nature of the conditions enforcing well-formedness. For AEG, ill-formed structures cannot occur because there is no combination of lexical and phrasal constructions of English that licenses them. For RJ, they cannot occur because the general rules of English syntax and argument linking, interacting with the verb’s lexical entry, do not license them. The 20
This condition is proposed in slightly different form in Goldberg 1995 as the Correspondence Principle. In Goldberg 1995, 2006 it is argued that this principle is a default principle that can be overridden by the specifications of particular constructions; e.g., the passive construction specifically serves to allow a normally obligatory argument to be omitted. Goldberg also suggests that particular constructions determine which arguments must be shared between verb and construction and which can be contributed by the construction alone. Jackendoff 1990 also has a formulation, which however deals only with the sharing of the subject argument.
t he e ngl is h r e s ul t ative
301
difference stems from whether one wishes to assume that all aspects of grammar can be captured by constructions (AEG) or whether one wants to preserve a role for traditional syntax and argument linking that is non-constructional (RJ). Since we wish to emphasize what is shared between our approaches, we leave this issue aside. Let us work through a few examples. Consider first (40). (40)
Willy watered the plants flat.
The verb water is obligatorily transitive, since Willy watered is out.21 Thus by Full Argument Realization (FAR), both arguments of the verb must appear in the syntax in (40)—and they do. The constructional subevent has three arguments: a causer (or agent) mapped into subject position, a patient mapped into object position, and a predicate; FAR demands that all of these be expressed too. So we have a total of five arguments to be mapped into only three syntactic positions. However, FAR permits two of these to be shared, which is in fact the case: Willy and the plants serve as arguments of both subevents and thereby have multiple thematic roles. Suppose instead that we tried to combine the same verbal subevent with a constructional subevent such as ‘Willy made the ground become wet’. Then there would be two distinct patients competing for the same syntactic position, one from the constructional subevent and one from the verbal subevent. This is impossible in English: Willy watered the plants the ground wet. The reason is that there is only one syntactic position in the clause that a patient can map into. And neither of these NPs can be omitted because of FAR. Therefore this combination of events can only be expressed biclausally: Willy made the ground wet by watering the plants. Suppose, however, that the verb is only optionally transitive, for instance drink (the beer), and we again wish to combine it with a constructional subevent that has a different patient, say ‘make the pub dry’. This time FAR permits us to leave the verb’s patient implicit and express only the construction’s patient, so we can get Dave drank the pub dry. Here the only shared argument is Dave, which is agent in both subevents. Similarly, if the verb is intransitive, for instance laugh, there is no patient in the verbal subevent, and the agent is shared with the constructional subevent, as in The critics laughed the play off the stage. 21 Goldberg 2000 discusses a particular construction that allows the patient argument of many causal verbs to be omitted under special discourse conditions. This construction, termed the ‘Deprofiled Object Construction’, can be used to account for special circumstances where water can appear intransitively as in With three new greenhouses to take care of, the gardener had to water all afternoon. Also in the context of two people gardening together, one could say to the other: You weed and I ’ll water.
302
t he e ngl is h r e s ul t ative
Thus FAR accounts for both selected and unselected variants of transitive resultatives. The analysis so far is summed up by (41) (in a notation neutral between the authors’ somewhat different formalizations): (41) Transitive resultativesk a. Selected transitive resultatives Willy watered the plants flat. Constructional subevent: Agent: Willy; Patient: the plants; Predicate: flat Verbal subevent: Agent: Willy; Patient: the plants 2 arguments shared b. Unselected transitive resultatives with optionally transitive verb Dave drank the pub dry. Constructional subevent: Agent: Dave; Patient: the pub; Predicate: dry Verbal subevent: Agent: Dave; Patient: implicit 1 argument shared c. Unselected transitive resultatives with intransitive verb The critics laughed the play off the stage Constructional subevent: Agent: the critics; Patient: the play; Path: off the stage Verbal subevent: Agent: the critics 1 argument shared d. Impossible case Willy watered the plants the ground wet. Constructional subevent: Agent: Willy; Patient: the ground; Predicate: wet Verbal subevent: Agent: Willy; Patient: the plants 1 argument shared; no way to map two independent Patients into the same clause k
(41’)
Let me also do these examples in the notation of Conceptual Semantics. a. Willy watered the plants flat: [CAUSE (WILLYb, [BECOME (PLANTSg, FLAT)] ) ]a; [CAUSE ([WATER (b, g)], a) ]
The first line is the constructional subevent; the second is the verbal subevent. The verbal subevent, following the semicolon, is a modifier of the constructional subevent. The Greek letters bind arguments of the verbal subevent to the constructional subevent: b and g indicate that Willy waters the plants, a indicates that the verbal subevent causes the constructional subevent (which is what I take a means modifier to mean). b.
Dave drank the pub dry. [CAUSE (DAVEb, [BECOME (PUB, DRY)])]a; [CAUSE ([DRINK (b, SOMETHING)], a)] c. The critics laughed the play off the stage. [CAUSE (CRITICSb, [GO (PLAY, OFF (STAGE) )])]a; [CAUSE ([LAUGH (b)], a)] [continued]
t he e ngl is h r e s ul t ative
303
Another sort of impossible case concerns a verb that requires an obligatory PP expressing a path argument. Consider the contrast between walk and pad, both verbs of motion: (42)
a. Bill walked (around the room) for hours. b. Bill padded (around the room) for hours.
Pad obligatorily requires a path phrase, where walk does not. Now notice what happens when these verbs are combined with the transitive property resultative construction.l (43)
a. Bill walked himself into a coma. Constructional subevent: Agent: Bill; Patient: himself; Property: into a coma Verbal subevent: Agent/Theme: Bill; Path: implicit 1 argument shared b. Bill padded himself into a coma. Constructional subevent: Agent: Bill; Patient: himself; Property: into a coma Verbal subevent: Agent/Theme: Bill; Path: implicit 1 argument shared; violates FAR since Path phrase of pad must be overt (42b)
These differ from (41’a) in that the Patient of the constructional subevent is not bound to anything in the verbal subevent. In (41’b) the second argument of the verbal subevent is instead an implicit argument; in (41’c) there is no second argument in the verbal subevent. d.
Willy watered the plants the ground wet. Willy caused the ground to get wet by watering the plants. [CAUSE (WILLYb, [BECOME (GROUND, WET)])]a; [CAUSE ([WATER (b, PLANTS)], a)]
Here the second argument of the verbal subevent is not bound to anything, but there is no place in a resultative to realize it syntactically. However, it is fine in a periphrastic causative. l
In the notation of Conceptual Semantics:
(43’) a. Bill walked himself into a coma. [CAUSE (BILLb, [BECOME (b, IN COMA)])]a; [CAUSE ([WALK (b, PATH)], a)] (Note: the reflexive pronoun is notated as a variable bound to the subject.) b. Bill padded himself into a coma. [CAUSE (BILLb, [BECOME (b, IN COMA)])]a; [CAUSE ([PAD (b, ???)], a)] (Path argument of pad is not realized.) c. Bill padded himself into the room into a coma. [CAUSE (BILLb, [BECOME (b, IN COMA)])]a; [CAUSE ([PAD (b, INTO (ROOM) )], a)] (Path argument of pad is realized but ungrammatically.)
304 c.
t he e ngl is h r e s ul t ative
Bill padded himself into the room into a coma. Constructional subevent: Agent: Bill; Patient: himself; Property: into a coma Verbal subevent: Agent/Theme: Bill; Path: into the room 1 argument shared; no way to map Path PP and property PP into the same clause
(43b) is bad because the required Path argument of pad is not expressed, in violation of FAR; (43c) is bad because two different Path arguments are competing for the same slot.22
9.6.2 The Semantic Coherence Principle An observation that emerges from (41) and (43) is that a shared argument has parallel thematic roles in the two subevents: it is either Agent of both or Patient of both. This idea proves to be a more general constraint on resultatives (and probably other VP constructions); it is stated by Goldberg (1995) as a Semantic Coherence Principle: (44) Principle of Semantic Coherence: roles of the construction (rC) and roles of the verb (rV) may only unify if they are semantically compatible; roles rV and rC are semantically compatible if and only if rV can be construed as an instance of rC. The principle prevents incompatible roles from combining. For example, an agent role cannot combine with a patient role. Note that the principle is stated as a constraint on the combination of roles in a clause, not as a real world constraint on the referent of those roles. It is quite possible for a single referent to be construed as both an agent and a patient, for example, when each role is assigned by a distinct clause He captured the snake (patient), and the snake
22
A referee asks why Bill padded himself into/around the room is ill-formed. Indeed, it should be possible for the construction to add a patient/theme role expressed as direct object. In the case of a reflexive object, the interpretation required would be one involving a ‘split self ’: one’s superego acts on one’s ego (or however one wishes to describe the split-self scenario). As predicted, this is possible with certain intransitive motion verbs such as: (i) She marched herself out of the kitchen.
While it is pragmatically felicitous to force oneself to march somewhere, it is hard to imagine why one’s superego would cause one’s ego to ‘pad’ somewhere. That it is just pragmatically odd and not impossible is supported by the fact that we were able to find the following example using Google, from a story in which a girl forces herself to walk quietly up the stairs with a gun in her hand. (ii) She padded herself up the steps and knocked on the door, heart beating in her chest (www. geocities.com/applejax8831/mailcall.html)
t he e ngl is h r e s ul t ative
305
(agent) bit him, or via a coreferential argument, e.g., He (agent) talked himself (patient) hoarse. At the same time, the principle does not require that the role of the verb come pre-classified as the same type of role as the constructional role it combines with. What is required is that the role of the verb be construable as in instance of the role of the construction. For example, the object of wipe is not necessarily affected or a ‘patient’ (cf. She wiped the table but it was still dirty). However, it is potentially affected by the wiping and thus is construable as a patient. This is what licenses expressions such as Sam wiped the table clean. The principle of Semantic Coherence can be used to account for the contrasts in (45). (45)
a. She yelled hoarse. b. Ted cried to sleep. c. The tiger bled to death.
The construction required in (45) is the noncausal property resultative. The constructional subevent is X BECOME Y; its first argument, which maps to subject position, is a patient (i.e. something happens to X). However, the verbs yell and cry require an agent argument in subject position (i.e. yelling and crying are something that X does). Since an agent role cannot be construed as a type of patient, the combination of yell or cry and the noncausal property resultative is ill-formed. Now note that it is the specific semantics of the verb that is important, not just the general verb class that a verb belongs to. While both bleed and cry are verbs of bodily function, the subject of bleed is a type of patient, since bleeding is something that happens to someone, not something that a person or other being does. Thus Semantic Coherence correctly predicts that bleed can readily appear in the noncausal property resultative construction, as in (45c).m Consider now the bodily functions of coughing, sneezing, and yawning. The single arguments of these verbs can be construed as patients, or, perhaps more typically, as agents, since coughing, sneezing, and yawning can be construed either as things we do or as things that happen to us. As predicted, attested
m
(45’)
In the notation of Conceptual Semantics: a. She yelled hoarse. [BECOME (SHEb, HOARSE)]a; (Person becoming hoarse is Patient) [CAUSE ([YELL (b)], a)] (Person yelling is Agent) c. The tiger bled to death. [BECOME (TIGERb, DEAD)]a; (Individual dying is Patient) [CAUSE ([BLEED (b)], a)] (Individual bleeding is Patient)
306
t he e ngl is h r e s ul t ative
examples (from Google web search engine) exist with these verbs used both intransitively (46a, b, c) and with fake objects (47a, b, c).n (46) a. He coughed awake and we were all overjoyed, especially Sierra. (www.diabloii.net/tdl/stories/destiny2.shtml) b. the kittens yawned awake and played with the other young . . . (www.geocities.com/mistacorjellicle/lifelove.html) c. Zoisite sneezed awake, rubbing his nose and cursing under his breath. (www.geocities.com/Tokyo/Dojo/9951/vaults/love1.html) (47) a. Patamon coughed himself awake on the bank of the lake where he and Gomammon had their play. (www.furnation.com/Furryden/NEW/ Stories/Digi8.doc) b. Ron yawned himself awake. (www.schnoogle.com/authors/jsawyer/SNW06.html) c. She sneezed herself awake as the leaf landed on her nose. (www.transcendzero.co.uk/halloween/hallowitches.htm) The Semantic Coherence Principle predicts just this variability. At the same time, we acknowledge that the examples in (47) are somewhat more natural than those in (46). We believe this results from the fact that coughing, sneezing, and yawning are more easily construed as actions that we do rather than actions that happen to us. We also recognize that examples found on the web must be used with caution, since it is often unclear whether the author is a native speaker. But dozens of instances of the types of examples in (46) were found; we take this as a clear indication that the requisite construals are possible. The noncausative spatial resultative has slightly different properties. Consider the examples Sue ran down the stairs vs. The water ran down the stairs. The subject of run is a theme in both cases. In the former case it is also always an agent. But in the latter case the subject can be interpreted as either doing something (i.e. an agent) or having something happen to it (i.e. a patient).
n
(46’) a. He coughed awake. [BECOME (HEb, AWAKE)]a; (Person becoming awake is Patient) [CAUSE ([COUGH (b)], a)] (Person coughing is Patient) (47’) a. He coughed himself awake. [CAUSE (HEb, [BECOME (bg, AWAKE)])]a; (Person causing waking is Agent) [CAUSE ([COUGH (b)], a)] (Person coughing is Agent)
The distinction between Agent and Patient readings of cough is not notated in the formalism so far. It is necessary to invoke what Jackendoff 1990 calls the Action Tier and Jackendoff 2007a renames as the Macrorole Tier. The macrorole tier involves the roles Actor and Patient/Undergoer, as distinct from the thematic roles Causer (the first argument of CAUSE) and Theme (the individual undergoing motion or change). Motivating this and introducing the notation is beyond the scope of the exposition here.
t he e ngl is h r e s ul t ative
307
In other words, run with an inanimate subject is indifferent as to whether its subject is an agent or a patient. In fact this is the case with many intransitive motion verbs. Given that the intransitive spatial resultative is based on the model of intransitive motion verbs, we believe the same distribution of thematic roles obtains with the GO in the constructional subevent of the intransitive spatial resultative. The consequence is that intransitive spatial resultatives are more liberal than intransitive property resultatives in the range of verbs they admit. So we find both (48a), where the subject of wriggle is an agent, and (48b), where the subject of melt is a patient. (48)
a. The worm wriggled onto the carpet. b. The chocolate melted onto the carpet.23
Both are compatible with Semantic Coherence: (48a) with the agentive reading of GO, and (48b) with the nonagentive reading of GO.o In the transitive spatial resultatives, the situation is again different. The semantics of the constructional subevent is X CAUSE [Y GO Path]. X, which maps into the subject, has to be an agent; Y, which maps into the object, has to be patient. The consequence is that resultatives are more acceptable to the degree that the subject can be understood as instigating the motion of the object in the verbal subevent. This consideration is especially interesting when the object is a reflexive, as in (49a, c, d, e). (49)
a. Bill wiggled himself loose. b. Aliza wiggled her tooth loose. c. The mechanical doll wiggled itself loose. [mechanical doll can instigate wiggling]
23 The causative of this example, the sun melted the chocolate onto the carpet, is semantically parallel to Goldberg’s (1991) He broke the eggs into the bowl. Levin and Rappaport Hovav 1995 (60–1) cite the latter as a potential counterexample to their claim that there can only be one delimiting state per clause, because the eggs both end up broken and end up in the bowl. They take a stab at resolving this problem by saying the broken eggs are whole eggs and the eggs in the bowl are only the eggs’ contents, so that the sentence covertly speaks of two different entities undergoing change. This story cannot be told in the present case: the same chocolate melts as ends up on the carpet. o
(48’)
In Conceptual Semantics notation: a. The worm wriggled onto the carpet. [GO (WORMb, ONTO (CARPET) )]a; (worm is Actor) [CAUSE ([WRIGGLE (b)], a)] (thing wriggling is Actor) b. The chocolate melted onto the carpet. [GO (CHOCOLATEb, ONTO (CARPET) )]a; (chocolate is Undergoer) [CAUSE ([MELT (b)], a) ] (thing melting is Undergoer)
Again, the difference between Actor and Undergoer is encoded on the macrorole tier.
308
t he e ngl is h r e s ul t ative d. The off-center spinning flywheel shook itself loose. [flywheel’s spinning can instigate shaking] e. The ball wiggled itself loose. [ball cannot independently instigate wiggling]
In the case of (49a, b), the subject argument is clearly volitional and agentive. The subject arguments in (49c) and (49d) are also construable as instigating the wiggling—i.e., are construable as agents. Only in (49e) is the ball unable to be construed as instigating its own wiggling action—therefore (49e) is incompatible with the Semantic Coherence principle and is unacceptable. Thus Semantic Coherence, motivated here by the standard examples in (41) and (43), additionally helps explain many otherwise mysterious constraints on resultatives (see also Takami 1998 for a related constraint). A further application of Semantic Coherence is discussed in the following section. To sum up so far, it appears that all the acceptable classes of resultatives observe the conditions of Full Argument Realization and Semantic Coherence. The two principles together account for a great deal of the distributional data. In addition, we have posited that ‘the’ resultative is actually a family of constructions, clearly closely related, but each subject to its own restrictions on verb classes and with its own special syntactic and semantic properties. The specifics need to be accounted for by any theory. Our choice is to formulate the analysis as directly as possible in terms of the mappings between form and meaning.
9.7 Extending the analysis to additional types of resultatives 9.7.1 Follow cases We next consider the follow-type cases introduced in section 9.2. These are distinguished from everything we have looked at so far by the fact that they are transitive expressions, but the subject rather than the object is host of the RP. Semantically they fall into two subtypes (many of these examples are adapted from Rappaport Hovav and Levin 2001). (50) Object determines subject’s path of motion, either by its own motion or by traces it leaves a. Bill followed the thief into the library. (thief’s motion determines Bill’s path) b. Bill tracked the leak to its source. (path of leak’s outflow determines Bill’s path) c. Bill traced the clues to the haunted house. (distribution of clues determines Bill’s path)
t he e ngl is h r e s ul t ative (51)
309
Transitive noncausative spatial resultatives Object is vehicle a. Bill took the train to New York. b. Sara caught a plane to New York. c. Ann rode a limo to the hotel. Object is path of motion d. Bill took/traveled the Taconic Parkway to New York. e. Chris rode the Oregon Trail to Kansas. f. Ray flew the coastal route to Buffalo.
What makes these cases semantically different from previous transitive resultatives is that, while the direct object is an argument of the verb, it is not a patient: one is not acting on the thief, the leak, or the clues in (50), nor on the various vehicles and paths in (51). Thus these verbs cannot combine with the normal causative resultative construction, which marks the object as a patient: that would violate Semantic Coherence. In many cases, it would also mean something bizarre, e.g. ??‘Bill made the leak go to its source by tracking it’, ‘Ray made the coastal route go to Buffalo by flying (it)’. So some different means of mapping these sentences to semantics must be invoked. We can envisage two alternative approaches. One claims that the constructions involve a transitive variant of the noncausative resultative construction, which is normally intransitive; the other involves positing new transitive subconstructions. We sketch the two approaches in turn. The first possibility for the follow-cases is to treat them as a variant of the noncausative motion construction (16b) (e.g. Bill rolled into the room): they entail that the subject argument moves along a path, just as noncausal motion expressions do generally. On this view, the direct object argument can be viewed as an argument of the verb alone; it is not an argument of the constructional subevent. In other words, unlike previous cases, the verbal subevent has two arguments, only one of which is shared with the constructional subevent.p (52)
p
(52’)
Transitive noncausative resultative construction Syntax: NP1 V NP2 PP3 Semantics: X1 GO Path3 MEANS: [VERBAL SUBEVENT: X1 GO [PATH DETERMINED BY Y2 ] ]
In Conceptual Semantics notation: Transitive noncausative resultative construction Syntax: NP1 V NP2 PP3 Semantics: [GO (X1, PATH3)]a; [CAUSE ([VERBAL SUBEVENT: GO (x1, [PATH DETERMINED BY Y2 ] ), a])]
310
t h e e ngl is h re s ul t ative
This construction requires a spatial resultative, not a property resultative. The selectional restriction on the verbal subevent, path determined by y2, generalizes the separate cases in (50)–(51). In each of these cases, the subject traverses a path determined by the object; but they differ in exactly how the object determines this path. This selectional restriction has the effect of ruling out certain other cases that have the same relation between subevents, for instance (53a), with the plausible meaning ‘she went into the room by pushing the door’, as well as examples like (53b), with the implausible meaning ‘we got into the building by noticing the thieves’. (53) a. She pushed the door into the room. (6¼She went into the room by pushing the door) b. We noticed the thieves into the building. According to (52), these are out because the direct objects do not determine the path of the subject. Whether this is the correct restriction or too narrow awaits the discovery of further cases. Finally, (54a) is ruled out because sniff cannot be construed as a verb of motion, and (54b) is ruled out because run is not a transitive verb as required by (52). (54) a. The dog sniffed the trail into the yard. (¼the dog went into the yard by sniffing the trail) b. The dog ran the clues into the yard (¼the dog went into the yard by running the clues) An alternative account of these cases might posit additional constructional subevents such as the following: (55) Syntax: NP1 V NP2 PP3 Semantics: X1 GO-AFTER Y2 Path3 INSTANCE: [VERBAL SUBEVENT]
e.g. follow, track, trace
(56) Syntax: NP1 V NP2 PP3 Semantics: X1 GO-BY WAY OF Y2 Path3 INSTANCE: [VERBAL SUBEVENT] e.g. take or MEANS: [VERBAL SUBEVENT] e.g. ride, sail, drive We observed in section 9.3 that constructions generally need to specify the possible range of semantic relationships between verb and construction (e.g., MEANS or RESULT). In the case of (55) and (56), a different relationship between the verbal subevent and the constructional subevent is indicated by the predicate, INSTANCE. That is, the verbs in (50) and (51) have meanings that closely resemble the semantics of the constructions: e.g., the verb follow
t he e ngl is h r e s ul t a tive
311
designates a specific instance of the more general meaning ‘X goes after Y along a Path’; take as used in (51d) designates an instance of the more general ‘X goes by way of Y along Path.’ Goldberg (1997) proposes that cases in which the verbal subevent is simply an instance or an elaboration of the constructional subevent are the most prototypical case, both for particular constructions and cross-linguistically. In the case of the construction represented in (55), the only acceptable relationship between verb and constructional subevent is the instance relation. Notice MEANS and RESULT are not acceptable: (57)
a. The dog sniffed the trail into the yard. (¼ the dog followed the trail into the yard by sniffing it—MEANS) b. The dog ran the clues into the yard. (¼ the dog followed the clues into the yard by running—MEANS) c. The dog found her into the yard. (¼ the dog went after her into the yard thereby finding her—RESULT)
In the case of the construction represented in (56), it is possible to use a verb that designates the MEANS of going by way of a vehicle or path as in the following examples: (58)
a. Bill rode a train to New York. (Bill went to NY by way of a train by riding the train) b. Bill drove the Taconic Parkway to New York. (Bill went to NY by way of the TP by driving)
The semantic representations in (55) and (56) capture the semantic properties of the follow subclasses and make two additional predictions. For ease of exposition, we refer to the Y2 argument of (55) as a ‘sought-after’ role and the Y2 argument of (56) as a ‘route’ role. First, these representations insure that this class is only productive with verbs that can be construed to involve agent/theme and, if they have a second obligatory role, it must be construable as either a ‘sought-after’ or a ‘route’ type role. As expected in accord with the Semantic Coherence principle, other verbs cannot integrate with these semantic templates, as seen in (53b) (We noticed the thieves into the room). Secondly, the representations predict that only path phrases and not change of state resultatives apply to the follow subclasses. This prediction is borne out: (59)
a. We drove Highway 5 silly. b. We followed the trail silly.
312
t he e ngl is h r e s ul t ative
Moreover, the linking patterns between syntax and semantics specified in (55) and (56) are independently required. That the first argument of GO should be expressed by a subject is familiar from intransitive motion verbs such as run, walk, jog, jump. That the path argument should be expressed by a PP is also expected. In addition, the expression of the ‘sought-after’ and ‘route’ roles as direct object in (55) and (56) is also independently required for other verbs. For example, the direct objects of seek and need are naturally described as bearing a role that can be described as ‘sought-after’;24 the NP arguments of climb and descend arguably involve ‘route’ type arguments. Thus additional constructions may be required, but the semantic structure is required for verbs like follow and the relevant sense of take, and each aspect of the syntax is also independently needed for other verbs.25 In any case, it is clear that something special needs to be said to account for these cases, since they are transitive but noncausal, unlike all the other examples discussed so far. Our suggestion is to posit additional subconstructions that are minimally different from the other, independently required constructions.26
24
The reason seek and need do not appear in the construction in (55) is because they do not designate instances of the meaning ‘X go-after Y Path.’ 25 It is possible to further motivate the constructions in (55) and (56) by demonstrating a chain of semantic relatedness such that they can be seen to be minimal variants of constructions already posited to exist. Notice that the verb chase has a similar meaning to that in (55). Its roles can be captured by the following: (i) chase (agent/theme sought-after/theme path)
Chase is causal in that the agent argument causes the sought-after argument to move. This makes chase itself an unremarkable instance of the transitive path resultative represented in (16c). Its close similarity to the meaning represented in (55) serves to relate the follow-cases to the caused-motion construction. In fact the examples below suggest a chain of semantic relatedness that serves to relate the transitive path resultative we saw earlier to the representations in (55) and (56): (ii)
26
a. She chased the criminals into the building. (cause-go: transitive path resultative (16c) ) b. She followed the criminals into the building. (minimally different from a, but noncausal: instance of (55)) c. She followed the trail of clues into the building. (minimally different from b, also instance of (55)) d. She followed the trail into the building. (minimally different from c, instance of (56) )
The two of us differ on their preference for these two analyses, with AEG preferring the constructions in (55) and (56) and RJ preferring the construction in (52). The difference stems from whether one wishes to allow verbs to license syntactic arguments directly (RJ) or whether one views all argument linking as being done by constructions (AEG). Related to this is the issue of whether one views verbs that are isomorphic with the constructional meaning (e.g., make as it appears in the resultative or give as it appears in the ditransitive) to be licensed by the construction (AEG) or whether one wishes to allow those verbs to link their arguments to syntax directly (RJ).
t he e ngl is h r e s ul t ative
313
9.7.2 Dancing mazurkas The examples in (60) are somewhat different again from what we have seen so far. (60)
Verbþobject together form a predicate a. Martha danced mazurkas across the room. b. The kids played leapfrog across the park. c. John did cartwheels through the crowd.
We note first that these are somewhat less acceptable than (61), where there is no NP object. They are also much better than (62), where the object has some structure beyond a bare NP. (61)
a. Martha waltzed across the room. b. The children leapfrogged across the park. c. John cartwheeled through the crowd.
(62)
a. ??Martha danced mazurkas by Chopin across the room. b. ??The children played games of leapfrog across the park. c. ??John did impressive cartwheels through the crowd.
Our sense is that in (60) the verb plus object together form a sort of complex predicate. Once the object comes to be identified by anything other than its role in the subject’s action—that is, once the object becomes referential, as in (62)— the resultative is unacceptable. The complex predicate analysis of these cases can be used to explain why these examples fail to passivize (Visser 1963–73; Rappaport Hovav and Levin 2001)—the postverbal NP is not a direct object but part of a complex predicate.27 However this distinction may be formalized, at worst it has to be added to (16b), the noncausative path resultative, as another type of possible verbal subevent.28 27 Visser and Rappaport Hovav/Levin attribute the failure of passive to the fact that these expressions have subject hosts. However, the follow-cases discussed above also have subject hosts and they readily passivize.
(i)
a. The leak was tracked to its source. b. The clues were traced to the haunted house.
28 Rappaport Hovav and Levin 2001 have one further example, John swam laps to exhaustion, which is a property resultative. Again it is far worse if the object is more elaborate: John swam laps of the pool to exhaustion. Moreover, we suspect that the judgment here is clouded by an analogy to John swam laps to the point of exhaustion, where the PP is an adjunct: Sue swam laps until she fainted and John did so to the point of exhaustion. Similar considerations may be operative in (60a, b), where substitution of the adjunct all the way across the park improves matters. Moreover, Jackendoff 1990 found examples like (60), but with goal PPs, very dubious: ?Dave danced waltzes into the room. On the whole, then, everything points to the marginality of this resultative type.
314
t h e e ngl is h re s ul t ative
9.7.3 Spit cases Our last case is the spit-type examples introduced in section 9.2. (63) Intransitive resultatives with implicit (i.e. non-subject) host a. Bill spit/urinated/ejaculated/coughed/sneezed out the window. b. Bill sweated/bled on the floor. c. The toilet leaked through the floor into the kitchen below. d. Bill ate off the floor. e. Bill drank from the hose. We note that the examples in (63) do superficially fit the criteria for resultatives— the spit ends up outside the window, the sweat ends up on the floor, and so on. These are interesting because they appear to violate Full Argument Realization. That is, while the verbs do not require the entity in motion to be expressed, the causative resultative construction normally does. So these cases are potential counterexamples to our analysis. However, consider (64). (64) a. Bill entered/left/exited (the room) through the bathroom window. b. Bill crossed (the street) to our side. c. The cream rose to the top. The verbs in (64) express inherent paths: one cannot enter without going into something, exit without going out of something, cross without passing from one side to the other of something, or rise without going in an upward direction. But it is always possible to add a PP that further delineates the path, even when not selected by the verb. Jackendoff (1990, 170–1, 280) formalizes this possibility as what in present terms might be called ‘the PP adjunct path construction’. What is important in the present context is that the construction only serves to modify or amplify an argument of the verbal event, namely the path. Now return to the verbs in (63). They all express some entity or substance in motion, but in these cases the entity is implicit. In addition, like the verbs in (64), the entity or substance in motion has a specified path: in (63a, b, c) out of or away from the subject, in (63d, e) into the subject. The PP simply further delineates the path, just as in (64). Therefore these examples can be produced by the PP-adjunct path construction rather than the resultative. This means that there is no constructional subevent whose theme argument (the entity in motion) must be expressed to satisfy Full Argument Realization. On the other hand, there are other examples of this type in which the verb does not imply any necessary entity in motion: (65) Bill coughed/sneezed on the mirror.
t he e ngl is h r e s ul t ative
315
It is quite possible to cough or sneeze expelling only air. No motion trajectory is normally specified with these verbs. Therefore there is no necessary path implied by the verb that the PP can amplify. So an alternate approach would be to recognize a construction in the grammar that allows a highly inferable theme argument to be unexpressed for certain classes of verbs (namely verbs of bodily emission, ingestion, and contribution). The construction would specify that a particular semantic argument need not be overtly expressed, a sort of specification required anyway for the passive and middle constructions (see Goldberg 2005b).
9.7.4 Rappaport Hovav and Levin’s approach While the present account explicitly recognizes a family of related constructions in order to account for the evident semantic variability, it may seem that other approaches yield greater generalizations. However, few accounts have even attempted to account for the full range of data discussed here. One recent exception that does discuss most if not all of the types of examples explored here is Rappaport Hovav and Levin (2001). But a close look at this account reveals that it classifies resultatives in an extremely heterogeneous fashion. The account is based on a distinction in which certain resultatives are claimed to involve simple event structure and others are based on complex event structure. In their appendix B, RH&L classify the examples in (66) as having simple event structure, where simple event structure is defined such that the verbal and constructional subevents unfold at the same time (2001, 793–4).29 (66)
a. b. c. d.
The pond froze solid. Robin danced out of the room. We pulled the crate out of the water. The wise men followed the star out of Bethlehem.
They classify the examples in (67) as having complex event structure in which two subevents need not be cotemporal: (67)
29
a. b. c. d. e.
The joggers ran the pavement thin. We yelled ourselves hoarse. They drank the pub dry. The cows ate themselves sick. The critics panned the play right out of town.
As they put it, ‘the happening in the world described by . . . [the] resultative is not linguistically construed as two distinct events’ (2001, 782). [For a more rigorous account of this intuition, with much crosslinguistic experimental data, see Bohnemeyer et al. 2007.]
316
t he e ngl is h r e s ul t ative
Note that (66) includes intransitives (66a, b) as well as transitives (66c, d), and the latter have both object hosts (66c) and subject hosts (66d). All of the examples in (67) have object hosts, but so does (66c). (67a–d) are unselected transitive resultatives, but (67e), like (66c), is a selected transitive resultative. Rappaport Hovav and Levin suggest that all of the examples in (66) are noncausative whereas all of the examples in (67) are causal. But it is unclear why (66c) is considered noncausative: it is naturally paraphrased by ‘we caused the crate to move out of the water by pulling’, and it is expressed transitively with agent and patient arguments. Thus it is hard to see any natural division among resultatives that is captured by the notion of temporal dependence in Rappaport Hovav and Levin’s sense. In fact this notion only distinguishes ‘fake reflexive’ property resultatives from intransitive property resultatives, a generalization that accounts for only a proper subset of data from (66) and (67). Other syntactic attributes of resultatives, such as their overt transitivity, and particular semantic properties other than the temporal one on which the account is based, are not addressed. Neither do previous general analyses attempt to account for the partial productivity that is strikingly present in the resultative. We address this issue in the following section.
9.8 On the productivity of resultative constructionsq We have already seen that in order to account for the full range of data, it is necessary to posit a number of related but distinct resultative constructions. As noted above, the family of constructions required is akin to the type of families that are familiar from derivational morphology, as well as from nonlinguistic categorization.r q
This section has been substantially rewritten, in part for expository clarity, in part to incorporate comments on the original version by Boas 2005 and Wechsler 2005. r The family of constructions discussed here can be summed up in an inheritance hierarchy, as follows:
Verb Subordination Archi-Construction way construction time-away
etc. Resultative
Simplex (default intransitive)
Causative (default transitive)
Path Noncausative
MEANS RESULT
subject host trans
Property Causative MEANS
Noncausative MEANS
selected unselected implicit host intrans fake refl
Causative MEANS
selected
unselected fake refl
[continued]
t he e ngl is h r e s ul t ative
317
Also parallel to morphological phenomena is the partially productive nature of the resultative. Two issues must be addressed for each type of resultative: Is the choice of RP productive, and is the choice of verb productive? The answers to these questions prove to be interdependent. What we find is that there are areas of productivity plus many diverse pockets of idiosyncrasy. We go through the various types of resultatives in turn. Spatial resultatives appear to be totally productive, within constraints posed by the meaning of the construction. In English, any motion verb is possible (68a), and any spatial PP that can be construed as a path can be used as RP, whether it be telic or atelic (68a, b) complex or simple (69a, b), goal directed or source directed (70a, b): (68)
a. Pat ran/swam/jumped/hopped/plodded/staggered/flew into the room. b. Pat ran towards the room.
(69)
a. Pat ran down the hall out the door into the alley. b. Pat ran outside.
(70)
a. Pat ran to the river. b. Pat ran from the river.
APs such as free (of NP), clear (of NP), apart, open, and shut are also fairly freely productive within the semantic constraints imposed by the construction. These adjectives can arguably be interpreted as spatial configurations with some force-dynamic overtones: Some explication: . The resultative construction as a whole is a special case of the Verb Subordination Archi-Construction of section 8.8, encompassing also the way construction, the time-away construction, and perhaps others. . Each type of resultative construction inherits properties from two sources: from being either a Path or Property resultative, and from being either simplex or causative. In turn, the relation between simplex and causative forms is the same as that between noncausative and causative forms of verbs such as break. . All four types can express a MEANS relations between the constructional subevent and the verbal subevent, but the noncausative Path resultative also has the possibility of a RESULT relation in the sound emission and disappearance constructions. . The causatives break into selected and unselected resultatives, depending on how the Full Argument Restriction is implemented. .
The unselected resultatives permit the further possibility of a fake reflexive.
The simplex resultatives are typically intransitive, but the special case of subject host with verbs such as follow is transitive. .
. The causative resultatives are typically transitive, but the special case with an implicit Patient with verbs such as bleed is intransitive.
The meaning of the inheritance hierarchy relating these constructions is the same as the meaning in the case of morphological relations, along lines suggested in Chapter 2: the information content of each individual subconstruction is reduced by the information that can be predicted from its ancestors in the hierarchy, plus the information that it is descended from these ancestors and not others.
318
t h e e ngl is h re s ul t ative
(71) a. Willy wiggled/squirmed/pried free/loose (of the ropes). b. Judy jumped/leaped/skated/slid clear of the rocks. c. The bottle broke/spilled/fell/smashed open. Willy ends up in a position in space where the ropes don’t constrain him, and Judy ends up in a position in space where the rocks can’t injure her. Likewise, being open is not only a property but also a spatial configuration, affording free passage between the interior and exterior of the object; being shut precludes such access. We therefore tentatively take these cases to be interpretable as either property resultatives or spatial resultatives. The situation with property resultatives is more complex. First, a number of inherently resultative verbs (e.g. the intransitives become, turn, and get and the transitives make, get, turn, and render) are quite productive in their choice of RP, allowing a broad range of predicates. In particular, make, render, and become allow predicate NPs as well as APs (e.g. make him president, render him a nonentity, become a pariah). Other inherently resultative verbs constrain their choice of RP, often by specifying a semantic restriction on the RP. For example, both pry (transitive) and come (intransitive) can only appear with RPs that express a semantic notion that corresponds roughly with ‘apart’: (72) a. He pried it apart/open/loose/free/flat/straight. b. It came apart/open/loose/free/flat/straight. Similarly, section 9.3 mentioned the verb drive, which in its resultative use selects only predicates denoting negative and extreme mental states: (73) a. drive NP crazy/nuts/bananas/to desperation/to drink/up the wall/ meshuga/frantic b. drive NP happy/sick/silly/clean/calm/thin/sober In fact, most of the RPs that are good with drive occur only with it and with make and go. The overall moral is that what looks like a productive choice of APs in property resultatives often boils down to occurrence with one of these inherently resultative verbs. If the verb is not inherently resultative, the choice of RP is more restricted. For example, in Boas’s (2000, 2003) search of the 100-million-word British National Corpus, some adjectives such as wet, sleepy, brown, and dirty appear as RPs only with inherently resultative verbs. Below is an enumeration of contrasting cases, many drawn from Boas’s appendix A. Note that the list of adjectives that can appear freely as property RPs with verbs other than inherently resultative verbs is quite limited (Goldberg 1995, Green 1972, Takami 1998, Wechsler 2001).
t he e ngl is h r e s ul t ative (74)
fairly productive as RP
RP only with inherently resultative verbs
dry awake black clean flat thin straight smooth sick full dead silly stupid hoarse
wet sleepy brown dirty thick fat crooked rough ill empty famous safe calm tired
319
There are some generalizations about which APs are more productive. Wechsler 2005 shows that adjectives that are more productive as RPs tend to express an extreme point along a scale. Consider the contrast between dry, which is quite productive as an RP, and wet, which is not. Dry is at the endpoint of a scale of wetness: it entails ‘no degree of wetness’, and things can be almost dry. By contrast, wet is not at an endpoint: ‘no degree of dryness’ makes little sense, and things can be termed almost wet only under very special conditions. Similarly for clean vs. dirty and smooth vs. rough. Perhaps even black and brown fit this criterion, in that things can be almost black but it is rather odd to describe things as almost brown. On the other hand, other pairs such as thin vs. fat do not appear to fall under this generalization, and they remain at the moment mysterious.30 Moreover, section 9.4.1 pointed out non-endpoint property resultatives such as heat the mixture hotter and hotter, which also do not conform to this generalization. Beyond these sorts of cases are what appear to be still more idiosyncratic constraints on the choice of property APs and PPs. There exist cases in which an AP is not acceptable but a cognate PP is. (75)
a. He danced himself to fame/famous. (Verspoor 1997, 119) b. We danced ourselves to dust/dusty. (Boas 2000, 257) c. Amy ran herself to exhaustion/weary.
30 Goldberg (1995, 195–6) argues that this generalization holds even for RPs such as sick or hoarse. Although these are normally gradable, when used as RPs, they strongly tend to imply a clearly delimited boundary beyond which the activity cannot continue. That is, He ate himself sick implies that he ate to a point beyond which he could no longer continue eating.
320
t he e ngl is h r e s ul t ative
Here it might be argued (as Wechsler 2005 does) that the adjectives do not express extreme points along a scale (??almost famous/dusty/weary), but that the PPs do, because of the to. This account might be of use as well in explaining an apparently arbitrary restriction on the choice between AP and PP that has been known for a long time (Green 1972; Carrier and Randall 1992): the resultative excludes past participial adjectives as RPs (76a), whereas cognate PPs are acceptable (76b). Again, the past participial forms appear to denote open-ended scales (??almost exhausted/bored/ exhilarated), but the PPs with to treat the same scale as having an endpoint (almost to exhaustion/to the point of boredom/to a state of exhilaration). Moreover, as with the non-endpoint adjectives such as wet and dirty, inherently resultative verbs do allow past participial adjectives (76c). (76) a. He sang himself exhausted/bored/exhilarated. b. He sang himself to exhaustion/to boredom/to (a state of) exhilaration. c. He made himself exhausted/bored/exhilarated. But in other cases, an AP is more acceptable than the cognate PP. (77) a. He danced his feet sore/to soreness. (Verspoor 1997, 119) b. He talked himself hoarse/to hoarseness. c. Denny Butler nearly crushed you flat. (British National Corpus, cited by Boas 2000, 257) d. Denny Butler nearly crushed you to flatness. (Boas 2000, 257) A possible explanation for the illformedness of the PPs in (77) is that they are blocked or pre-empted by the highly conventional verbþAP forms.31 In order to invoke blocking, it is necessary to recognize that the pre-empting forms are stored in the mental lexicon, a recognition that is central to the present approach. This is another virtue of the constructional approach, since conventionalized resultative expressions are assumed to be stored as subregularities along with the more generalized constructions. One particular case of an alternation of AP and cognate PP has received particular attention: dead vs. to death. It turns out that the choice is highly dependent on the main verb (Verspoor 1997; Goldberg 1995; Boas 2000). In Boas’s large corpus search, the verbs stab, bat, put, batter, frighten, crush, scare, burn only occur with the RP to death and never with dead. Shoot on the other hand occurs with dead 408 times compared with to death (11 times). This difference has a semantic basis. Dead is used as an RP only when the end 31
Brian Joseph suggests the pair beat someone senseless and beat someone into submission as possible counterexamples, but we maintain that the two examples are not synonymous. Notice that the following are decidedly odd: ?beat someone to senselessness; beat someone submissive.
t h e e ngl is h re s ul t a tive
321
state is an instantaneous result of the action denoted by the verb. It is possible to shoot someone and cause them to die instantly. If more than one shot is used, only to death is preferable: (78)
a. Riddling him with 16 bullets, Billy Bob shot him to death/??dead. b. Firing a single bullet to the heart, Billy Bob shot him dead/?to death.
Most methods of killing (including stabbing, battering, frightening, crushing, and burning) are not normally understood to cause instantaneous death; thus to death is strongly preferred with the corresponding verbs (a similar analysis is given by Wechsler 2005). Regardless of how these semi-systematic threads are accounted for, there remains a fairly large dose of idiomaticity or conventionalization. For example, ill is a near synonym of sick and yet is not nearly as felicitous as an RP: (79)
a. She ate herself sick b. She ate herself ill.
Asleep describes a clearly delimited state and yet it is far less conventional as an RP than awake or to sleep: (80)
a. ?She cried herself asleep. b. She cried herself to sleep. c. She jerked herself awake.
Here are some other minimal pairs that do not readily admit of an explanation. (81)
a. They stirred the paint smooth/homogeneous. b. His pants wore thin/??threadbare/rough. c. They twirled themselves dizzy/happy.
In addition to these somewhat general patterns, there exist idioms with the syntactic pattern and approximate semantics of the general construction. A good example is drink NP under the table, which patterns like a spatial resultative but has a specialized meaning and a rigid choice of lexical items. There also exist cases such as those in (82), where the choice of verb is quite broad, but the object and the RP are fixed. In all cases, the meaning is approximately ‘do V to excess’. (82a–c) are from Chapter 8; (82d) from Boas 2000. (82)
a. b. c. d.
She talked/sang/cooked/knitted/programmed her head/butt/tush off. She talked/sang/cooked/knitted/programmed her heart out. She talked/sang/cooked/knitted/programmed up a storm. She talked/sang/cooked/knitted/programmed herself blue in the face.
322
t he e ngl is h r e s ul t ative
Here the idiom is the combination of the object and the RP.32 Somewhere in this mix also falls She worked/wore/knitted/?typed her fingers to the bone. A similar case appears in (83).s (83) Dean laughed/danced himself crazy/silly/to death/to oblivion. Expressions like (83) are often taken to be standard resultatives resulting from productive lexical/syntactic processes. But such an account fails to explain three things. First, the construction is not happy with all verbs. For example, it is markedly odd with talk, even though an interpretation is imaginable (Goldberg 1995, 137): ?He talked himself crazy (Goldberg 1995, 137). Second, the construction is nearly always idiomatic: Dean doesn’t end up silly or dead in (83), and perhaps not crazy either. Third, the choice of RPs is hardly productive: it is largely restricted to phrases that idiomatically mean ‘excessively’: (84) Dean laughed/danced himself happy/sleepy/famous/very crazy/nuts/dead/ to a certain death. The expressions with crazy and silly can be used telically or atelically; those with to prefer to be telic. (85) a. Dean laughed/danced himself crazy/silly for/in an hour. b. Dean laughed/danced himself to death/to oblivion in/??for an hour. Thus, like the idiomatic intensifiers in (82), they do not carry typical resultative semantics. This suggests that we are not dealing with a standard property resultative, but rather an idiom of roughly the form is ‘oneselfþ[list of possible RPs]’—not unlike one’s head off. The view that emerges from this little survey is that the grammar contains the property resultative as an overarching generalization; but that particular subclasses, constrained in all sorts of different ways, are learned individually (see also Boas 2000, and for a learning theory in which this is expected, see Tomasello 2000, 2003). This may be an unappealing result to those committed to a maximally general view of grammar, but we think it cannot be avoided. The fact of limited productivity is already widely recognized in other corners of English grammar, for example in derivational morphology (e.g., Chomsky
32
Jackendoff 1997b, 2002 argues that these are constructional idioms of English on a par with the way construction—not really resultatives. Goldberg would prefer to treat them as conventionalized instances of the resultative construction related via a default inheritance hierarchy structure [of the sort in subfootnote r]. s
This paragraph is drawn from the original text of Chapter 8.
t he e ngl is h r e s ul t ative
323
1970, Janda and Joseph 1999). The property resultatives provide a parallel case in syntax. At the same time, an approach that disallows any degree of productivity runs into just as much trouble. For example, we feel Boas (2000, 2005) takes the implications of the semi-idiosyncrasy of the resultative too far when he claims that ‘In order to be able to describe which senses of a verb may occur with a specific semantic and/or syntactic type of result phrase, we must encode this information in the event-frames representing the individual senses of the verb’ (2000, 301): each RP frame a verb appears with is taken to represent a different verb sense. We believe his own corpus data indicate that the construction is in fact partially productive, in that there are many verbs that occur only once in the 100-million-word corpus with a particular RP. This is illustrated in Table 9.1. If a particular verb only appears once in a 100-million word corpus with a particular property RP, we feel it is safe to infer that the use of that verb with that RP is not likely a conventional fact about the verb. The following provide a tiny sample of such examples, and we feel their novel character is apparent. (86)
a. ‘Shadows cast by thin trees yawn and stretch themselves awake’ b. ‘After waiting two years on the backburner for Batmania to boil itself dry. . . ’ c. ‘Plus a recess above the nibs allows the tiles to be stacked virtually flat’ d. ‘That the girls might be unaware of the camera rather than seducing it to death’ e. ‘She held out her dish and Vi spooned it full.’ f. ‘ . . . bloody vultures overhead shitting themselves silly’ g. ‘Shouldering the door shut . . . ’
Table 9.1. Productive uses of RPs: Tally of distinct verbs that appear only once with listed RPs in 100-million-word corpus (based on raw data reported in Boas 2000) Apart Awake Black Clean Dead Dry Flat
13 verbs 11 verbs 5 verbs 5 verbs 3 verbs 9 verbs 16 verbs
Full Hoarse Into pieces Into shape Into the ground Open Shut
4 verbs 4 verbs 4 verbs 41 verbs 21 verbs 10 verbs 13 verbs
Silly Stupid To death To pieces To sleep
9 verbs 7 verbs 35 verbs 11 verbs 9 verbs
324
t h e e ngl is h re s ul t a tive
Table 9.1 includes only common RPs, and the data collected only involve transitive property resultatives. Including more novel RPs and intransitive resultatives would increase the number of productive resultative expressions that much more. We wish to simply acknowledge that resultative constructions are partially, but not fully productive. In particular, property resultatives, by contrast with spatial resultatives, display a great deal of semiproductivity and idiosyncrasy, which would be entered in the inheritance hierarchy of subfootnote r. We do not attempt to provide a full explanation of conservative productivity here (but see Baker 1979, Brooks and Tomasello 1999, Goldberg 1995, Pinker 1989 for discussion and suggestions).
9.9 Conclusions What does this corner of English tell us about the nature of the grammar? It tells us that a phrase structure rule V–NP–AP/PP is not enough to characterize the resultative construction. We see no choice but to treat the resultative construction as made up of a family of subconstructions, united by related but not identical syntax and by related but not identical semantics. Moreover, among the subconstructions are cases with only a single example, i.e. classical idioms. Like classical idioms, resultative idioms have the syntax of the (more) productive class and share to some degree in the semantics, yet have to be learned one by one. We differ to some degree in our optimism that a learning theory capable of assimilating these facts is imminent (RJ being less optimistic than AEG). But we are in solid agreement that no learning theory based purely on setting universal parameters (a` la Principles and Parameters Theory) or purely on ranking universal violable constraints (a` la Optimality Theory) is going to come up with anything close to this pattern. We have attempted here to offer a comprehensive, descriptively adequate account of English resultatives. Section 9.3 posited four major subconstructions, repeated here. Note that we have folded the special cases of sound emission and disappearance resultatives into (87c). [See also the formalizations in subfootnotes f, h, and i and the inheritance hierarchy in subfootnote r.] (87) a. Causative property resultative Syntax: NP1 V NP2 AP3 Semantics: X1 CAUSE [Y2 BECOME Z3] MEANS: [verbal subevent ]
t h e e ngl is h re s ul t a tive
325
b. Noncausative property resultative (e.g. The pond froze solid) Syntax: NP1 V AP/PP2 Semantics: X1 BECOME Y2 MEANS: [verbal subevent ] c. Noncausative path resultative (‘intransitive motion construction’, e.g. The ball rolled down the hill; The truck rumbled into the station) Syntax: NP1 V PP2 Semantics: X1 GO Path2 i. MEANS: [verbal subevent ] ii. RESULT: [verbal subevent: X1 emit sound ] iii. RESULT: [verbal subevent: X1 disappear ] d. Causative path resultative (‘caused motion construction’, e.g. Bill rolled the ball down the hill) Syntax: NP1 V NP2 PP3 Semantics: X1 CAUSE [Y2 GO Path3] MEANS: [verbal subevent ] We also offered two possible accounts of the follow subclasses in section 9.7.1, and two of the spit on the floor examples in section 9.7.3; each of these involves adding further specialized members to the family. Any treatment of resultatives has to account for the fact that they have the range of forms and interpretations shown above. For a first approximation, a constructional approach to resultatives says that the syntax–semantics interface of English contains this family of four subconstructions in addition to the more general principles (or constructions!) that link the semantic arguments of a verb to its syntactic arguments (via a thematic hierarchy or whatever). Languages such as Spanish, Japanese, and Hindi have far less general resultative constructions: they allow only atelic path RPs combined with verbs that independently encode motion (Aske 1989, Nakamura 1997, Narasimhan 1998, Takami 1998). It is clear that something about English allows it to license resultatives with the particular range of interpretations that it does. The generalizations in (87) capture what this something is. Overall, it is clear that something must be stipulated in order to equip English with resultative sentences. Perhaps the stipulation is some unusual syntactic structure involving small clauses or the like (Hoekstra 1988); but this syntactic structure must also come with rules of interpretation that result in the interpretations in (87). Or perhaps the stipulation is a rule in the lexicon that amplifies verbs’ argument structures (as in Levin and Rappaport Hovav 1990 and some versions of HPSG and LFG); if so, this rule must also specify the resulting meaning. Or perhaps the stipulation is some extra principle of
326
t h e e ngl is h re s ul t a tive
semantic interpretation or inference (as in von Stechow 1995); if so, this rule again must specify the special meaning. Even if some stipulation is possible in the form of an abstract parameter that connects to other properties of the language (Snyder 2001), the grammar still must predict the allowable forms and meanings of resultative sentences in English. In other words, something like the generalizations in (87) ultimately must be captured. The constructional approach simply states the generalizations directly, thereby avoiding further complexities in (a) the syntax, (b) the lexical entries of verbs, and (c) mappings from abstract parameters to grammatical properties. We have demonstrated that a number of generalizations follow from our hypothesis. In particular, the semantic argument structure of the constructional subevent determines the syntactic argument structure of the sentence by general principles of argument linking. Similarly, the aspectual structure of the constructional subevent controls the aspectual structure of the resultative sentence; in turn, the aspectual structure of the constructional subevent is predictable from general principles correlating event structure with change, extension, motion, and paths. Finally, the semantics and syntax of resultatives explain the possibilities for temporal relations between the two subevents. Using the semantics of the constructional subevent, the semantics of the verbal subevent, and the semantic relationship between the two, we were able to explain many of the distributional properties of resultatives. We were able to prevent many further conceivable overgeneralizations by adopting the Full Argument Realization and Semantic Coherence principles, which appear to apply generally to all VP constructions such as the way construction as well as to the resultative. Finally, it is important to recognize the limits of explanation. For instance, we have seen that two unrelated classes of verbal subevents appear in construction (87cii, iii): sound emission and disappearance. It is unlikely that any general principles of syntax or semantics are going to explain why these particular cases work and many other conceivable ones do not. We have shown that there are pockets of productivity, but there is also a great deal of idiosyncrasy, especially in property resultatives. Many idiosyncratic instances and small subclasses of the construction must be learned and stored individually. So far as we can determine, this pattern can be described only within a constructional approach to grammar. We leave it as a challenge for practitioners of other approaches to develop comparably detailed accounts.
chapter 10
On the Phrase The Phrase ‘The Phrase’ (1984) For Adrian Akmajian, who always appreciated old-time syntax
10.1 Introducing the construction This chapter concerns an NP construction that appears frequently in papers on syntax, but which has not itself been the subject of much syntactic investigation.a The construction consists of a definite article and a noun followed without pause by an expression E which can be of quite varied character; I will refer to it as the the N–E construction. Here is a range of examples, grouped approximately into semantic categories. Examples like (1) are ubiquitous in the discourse of linguistic theory. [This chapter appeared originally in Natural Language and Linguistic Theory 2 (1984), 25–37, in an issue dedicated to the memory of my dear friend Adrian Akmajian, who died while this paper was under review. It is reprinted here by permission of the publisher, Springer Netherland.] I am grateful to Joan Maling, Jane Grimshaw, Dwight Bolinger, Richard Oehrle, and three anonymous readers for discussion of the material presented here. This research was supported in part by National Science Foundation Grant IST-8120403 to Brandeis University, and in part by National Science Foundation Grant BNS 76-22943 to the Center for Advanced Study in the Behavioral Sciences, where the author was a Fellow during the completion of this study. Adrian Akmajian used to speak with affection of the days back in the 1960s when one could do descriptive syntax without heavy theoretical overtones—when every construction was a new frontier. The present study was conceived in that spirit, and it had been my intention before Adrian’s death to dedicate the paper to him as a light-hearted remembrance of those times. Now, regrettably, the dedication must bear more weight. a This chapter originated as a sort of jeu d’esprit, unrelated to anything I or anybody else was doing at the time, but rife in colorful examples, many directly from the texts of the linguistic literature (though unnoticed). From today’s perspective it is an early attempt to characterize an instance of what Peter Culicover has termed a
o n t he p hra se th e p hrase ‘ the phrase ’
328 (1)
(2)
a. b. c. d. e. f. g. h.
a. b. c. (3) a. b. c. d.
the phrase the phrase the word/verb run the prefix un the construction N of NP the sentence Will you marry me the sequence up a the sound ph the syllable pa the letter A the number 14 the note E♭ the song Entzweiflung the play/opera/novel/movie Death in Venice the poem Trees the painting Seated Woman
(4)
a. the element engoopium b. the compound engoopium sulfate c. the material polyacrynilate
(5)
a. the actor John Gielgud b. the artist/poet William Blake c. the pianist/composer Johannes Brahms
(6)
the name Harry
(7)
the color red
(8)
a. the noise [raspberry, imitation of a goat, etc.] b. the pattern da-dum da-dum da-dum
(9)
the symbol $
Most of this chapter will be concerned with determining, though hardly conclusively, what the syntactic structure of this construction might be. First, however, we must clear away a certain amount of semantic and pragmatic debris. (Admittedly, what counts as debris on one occasion may well be of primary interest on another; this is just how I choose to slice the issues up this time.)
syntactic nut—a construction that does not fall under standard syntactic principles but about which speakers nevertheless have robust intuitions. My account tried to cram the construction into a canonical X-bar framework, in particular the three-level X-bar theory of Jackendoff 1977. The artificiality of this solution will be evident. The subfootnotes in this version discuss some alternatives in a more contemporary framework. I also completely missed the significance of the fact that the construction admits all manner of nonlinguistic material. This remained to be pointed out in no uncertain terms by Postal 2004: it shows that the traditional view of lexical insertion is severely compromised. Again, the subfootnotes bring this issue more to the fore.
on th e ph rase the phrase ‘ th e ph rase ’
329
10.2 Semantic and pragmatic constraints It should be obvious that the possibilities for the expression E are semantically constrained by the choice of N: the noun specifies a category to which E is to be understood as belonging. On the other hand, there are no inherent syntactic constraints on E: it need not be a syntactic constituent—as in (1); nor even an expression of English—as in (3a). In fact, if the construction is uttered, E need not be expressible in standard orthography (as I have tried to suggest in (8a)); while if the construction is written, E need not have a pronunciation, as in (9). Hence, like the complements of verbs such as say and go (in the sense ‘make a noise’), E is a constituent whose interior is unconstrained by normal rules of syntax and phonology. We will assume, therefore, that the phrase structure rule responsible for introducing E violates the normal theory of syntactic categories by permitting a totally free expression.b
b Here is the place where I fudged the issue of lexical insertion. Postal 2004 (chapter 6) addresses this construction and points out that it represents a severe challenge to the Minimalist Program’s (Chomsky 1995) proposal to build phrase structure by beginning with a numeration of items chosen from the lexicon (a sort of ‘Scrabble tray’ of words) and joining them into a tree by Merge. In addition to the the N–E construction, Postal offers further cases in which non-English terminal elements appear in English sentences.
(i)
a. b. c. d.
The space alien said ‘klaatu barrada nikto’ to Gort. [Teenspeak:] And then, I’m all like, [gesture of exasperation]. Sklerf does not rhyme with nikto. Jean est mange´ le pain is ungrammatical in French.
These cannot be described using a derivation that starts with a numeration of lexical items, since they include elements that are not part of the lexicon of English. One would not want to counter by saying that klaatu, @, gestures of exasperation, and so on are all lexical items of English, for that swells the lexicon beyond all recognition. In fact, since arbitrary sentences and non-sentences of any other language can be inserted in frames like (i.d), the lexicon would have to be infinite. Nor would one want to simply say that the sentences in (i) are ungrammatical, underivable by the grammar. That would imply, among other things, that linguists, who use metalinguistic utterances like (1a–h) and (i.c, d) all the time, are speaking and writing ungrammatically, making use of some unspecified auxiliary system of communication. It is important also that the content of the non-English elements in (i) is involved in the truth-conditions of the sentences. For example (i.d) becomes false if a is substituted for est. This means that these elements cannot be treated as undigested lumps in the structure of the sentence, inserted under a wild-card category such as Name which somehow evades the numeration. For a constraint-based model such as the Parallel Architecture and HPSG there is a more convenient escape hatch. The fact that sentences must be composed of words of the language, like many other constraints, can be taken to be a violable default. Making it a default allows for particular constructions, such as those in (1) and (i), to license constituents that override this constraint and thereby ‘leak out of the language’. Nevertheless, such constituents are not totally unconstrained: they still have selectional restrictions. For instance, the elements in quotative contexts such as (i.a) have to be linguistic utterances in some (purported) language, the element in (i.b) has to be an expressive linguistic or nonlinguistic gesture, the element in (8) has to be a noise, and so on. And there is nothing to exclude these elements having content that contributes to the meaning of the sentence—it depends on the semantics of the construction in which they are embedded. (One can also imagine extending this argument to an analysis of code-switching.)
330
o n t he p hra se th e p hrase ‘ the phrase ’
Next let us look at possible choices for N in this construction. There are nouns that one might expect to appear but in fact do not, for instance in the taste sour and the profession linguist. (Some readers claim to accept the taste ‘sour’, but I find it dubious; compare to the flavor rum raisin, which is unimpeachable.) Moreover, there are choices of N that are odd unless accompanied by certain modifiers: (10) a. i. ?the doctor Ben Spock ii. the noted doctor Ben Spock b. i. ?the meter 3/4 ii. the difficult meter 13/8 c. i. the friend Harry ii. my friend Harry (10a) and (10b) fall under a somewhat wider generalization, an interesting pragmatic constraint on certain uses of the construction. In these uses (i.e. with these choices of N), the N–E is apparently more felicitous when used to introduce something or someone relatively noteworthy that may be unfamiliar to the hearer. Thus it is perhaps not too significant that Ben Spock is a doctor, but more so that he is a noted doctor. To see that the constraint is not of any simple syntactic sort, note contrasts like these: (11) a. i. ??the pleasant drink wine ii. the pleasant Turkish drink floog b. i. ??the profitable crop tobacco ii. the profitable crop milo c. i. ??the material wood ii. the material polyacrynilate (¼ (4c)) Here it is the unfamiliarity of the noun in E that renders the (ii) examples more acceptable. On the other hand, this pragmatic constraint certainly does not apply in other cases, such as (1)–(2) and (6)–(9). In (1), (6), (8), and (9), the N perhaps serves to mark E as quotational. The special function of the construction in (2) and (7), if there is one, is less obvious. We hesitantly conclude, therefore, that certain nouns can appear unmodified in this construction, for a variety of semantic or pragmatic reasons, while many more can occur in the construction with modification, provided the resulting construction satisfies the pragmatic condition of unfamiliarity.
10.3 E is not an appositive Now let us turn to the main question I would like to pursue here: where E is attached within the NP. Let us assume for now that N is the head of the
on th e ph rase the phrase ‘ th e ph rase ’
331
construction and E is some kind of complement (we will consider the alternative that E is the head in section 10.5). Following the theory of English NP complements of Jackendoff (1977, chapter 4), there are three positions where E could be attached: to N’ along with strictly subcategorized complements; to N’’, along with restrictive modifiers (including restrictive relative clauses); or to N’’’, along with appositives. These alternatives are summarized in (12).c
(12)
N¢¢¢
( = NP)
N¢¢ N¢ N
appositives restrictive modifiers
strictly subcategorized arguments
Since E does not have the internal structure of a syntactic constituent, one’s first impulse is to attach it as loosely as possible to the NP, making it some sort of appositive, as in (13).
(13)
N¢¢¢ Art
N¢¢
E
the
N¢
*****
N noise However, E in (1)–(9) is not preceded by comma intonation, the usual prosodic mark of an appositive. Furthermore, E may be followed by a restrictive modifier, the whole appearing without comma intonation.
c
A terminological point: Throughout the discussion I speak of N’, N’’, and N’’’ complements. Contemporary terminology would speak of the phrases in N’ as complements and the phrases in N’ and N’’’ as adjuncts. The three-level X-bar theory of Jackendoff 1977 never really took hold in the linguistics community at large. The third level was there to accommodate appositives, which can be adjoined to any phrasal category and which are clearly outside other sentential complements and modifiers. To my knowledge, mainstream generative grammar has had little or nothing to say about the phrase structure of appositives in the past thirty years. For that matter, the syntactic framework I currently work in, Simpler Syntax, has no proposals about appositives either. I will return to a possible Simpler Syntax account of the the N–E construction in subfootnote f.
332
o n t he p hra se th e p hrase ‘ the phrase ’
(14) a. the sound ph in English b. the song cycle I Hate Music by Leonard Bernstein This suggests that E must be attached either somewhere in N’ or at the beginning of N’’. On the other hand, one might note that there is another NP construction with a free expression E, in which the typical comma intonation of appositives does occur, and in fact must occur. (15) a. the phrase you mentioned, in the house b. the compound we synthesized, engoopium sulfate c. the noise he made, (16) a. the phrase you mentioned in the house (ungrammatical with intended meaning) b. the compound we synthesized engoopium sulfate c. the noise he made One might therefore suggest that at least some cases of the N–E have E in appositive position, and that the comma intonation is simply suppressed. However, the conditions under which such suppression would take place are rather curious. The construction with appositive, unlike the N–E, may be indefinite: (17) a. a phrase you mentioned, in the house (On the blackboard was written) a phrase, in the house. b. a compound we synthesized, engoopium sulfate (He won the Nobel Prize for the synthesis of) a compound, engoopium sulfate. c. a noise he made, (Right at the end of the triangle solo, there should be) a noise, (18) a. a phrase in the house b. a compound engoopium sulfate c. a noise Thus if there is a rule of comma suppression, it must take place only if N is definite and no relative clause intervenes between N and E. But even such an oddly restricted rule cannot be sustained, for if we look a little more carefully, there is semantic evidence against it. Consider the following examples, to be taken as the inception of a discourse. (19) a. On the nine o’clock news we saw the newscaster, Dan Rather, fall on his face. On the nine o’clock news we saw the comedian, Ernie Kovacs, fall on his face.
on th e ph rase the phrase ‘ th e ph rase ’
333
b. On the nine o’clock news, we saw the newscaster Dan Rather fall on his face. On the nine o’clock news, we saw the comedian Ernie Kovacs fall on his face. c. On the nine o’clock news, we saw a newscaster, Dan Rather, fall on his face. On the nine o’clock news, we saw a comedian, Ernie Kovacs, fall on his face. The difference between (19a) and (19b) appears to have to do with the pragmatic source of definiteness. The function of an appositive is to add information about an already specified individual; it can itself never be the specification by virtue of which the NP in which it occurs is definite. Rather, if the NP is definite, this must be because of other information in the discourse. In (19a), where E is an appositive, the nine o’clock news provides a frame of reference in which a newscaster, but not a comedian, plays a specified role. (Notice that Dan Rather in (19a) is understood as the designated newscaster for this broadcast, not just some newscaster who has happened to appear on the show; for this latter case, (19c) is more appropriate.) Since comedians have no designated role in newscasts, only the indefinite form (19c) is possible with comedian substituted for newscaster. By contrast, in the N–E, E has the function of uniquely specifying some individual of the category named by N. Thus the construction is inherently definite, without any presupposition about discourse context; and both versions of (19b) are acceptable. This inherent definiteness of the N–E can be confirmed by noting that E and superlative are mutually exclusive; evidently the definitizing character of each prevents the other from being used at the same time. (20)
a. the first name Harry (on the list) b. the smelliest compound engoopium sulfate c. the most disgusting noise
Notice that (20c) is acceptable if most disgusting is read ‘extremely disgusting’ instead of as a superlative. This shows that the unacceptability of (20c) is specifically due to the interaction of superlatives and E, not to some syntactic restriction on prenominal APs in the N–E. We therefore conclude that although there is indeed a construction in which E is attached to NP as an appositive, the N–E differs in phonological, syntactic, and semantic properties to an extent sufficient to recommend treating the two constructions as distinct.
o n t he p hra se th e p hrase ‘ the phrase ’
334
10.4 Attachment of E to N’ or N’’ We are thus left with the two possibilities for E illustrated in (21): either somewhere in N’ (with the strictly subcategorized arguments) or at the beginning of N’’ (with restrictive modifiers).
N¢¢¢
(21)
| Art
N¢¢
| the
N¢
|
PP
|
N
E
sound
ph
in English
One apparent problem with these structures is dealt with rather easily. If E is attached as in (21), it should be able to precede a restrictive relative clause, which falls at the end of N’’. But such is not the case: (22) a. ? the phrase in the house that begins the poem b. ?the noise that Harry is always making c. ?the song cycle I Hate Music that Lenny wrote However, the difficulty in (21) is due not to incorrect linear order, but to the inherent definiteness of the N–E observed above. Intuitively, since the reference is already unique, further constraint on the reference by means of restrictive relatives is excluded. This account is enhanced by cases such as (23), where the relative clause has the syntactic form of a restrictive but an interpretation tending toward the appositive: here the relative clauses seem much better than in (22). (23) a. the phrase in the house that, incidentally, begins the poem b. the well-known phrase in the house that begins the poem Thus the infelicity of (22) is a consequence of a semantic, not a syntactic problem. The evidence to decide between the two possibilities in (21) is somewhat more equivocal than the evidence that E is not an appositive. Some, though not all, of the phrases in (1)–(9) have relative clause-like paraphrases, suggesting attachment to N’’, like restrictive relatives.
on th e ph rase the phrase ‘ th e ph rase ’ (24)
a. b. c. d. e.
335
the song entitled Entzweiflung the actor named John Gielgud the name (??pronounced) Harry the color (?that is) red the symbol (?that looks like this:) $
On the other hand, one of the tests for an N’’ complement (Jackendoff 1977, section 4.1) is its ability to be left in the NP when the head is replaced by one, as in the lovely story about Bill and the awful one about Harry. This proves impossible in the construction the N–E. (25)
the (lovely) song Entzweiflung (by Schubert) and the (trite) one Wiegenlied (by Brahms)
This would seem to suggest that E is an N’ complement, which cannot appear after one (e.g. the father of Bill and the one of Harry). However, the characteristic behavior of N’ complements in one-pronominalization is that their interpretation is absorbed into that of one. For instance, in two good quarts of wine and three rotten ones, ones is read as quarts of wine, incorporating the N’ complement. But the N–E excludes this possibility as well: (26)
a.
the song Wer nur die Sehnsucht kennt by Schubert and the one by Tchaikovsky b. ?The sound ph in English and the one in Dutch differ in timing of voicing onset.
(One reader claims to find acceptable the sound ph in English and the corresponding one in Dutch. I find this marginally better than (26b) but not as good as the ph sound in English and the corresponding one in Dutch. Judgment begins to fail me here.) Thus evidence from one-pronominalization, often fairly reliable in distinguishing N’ from N’’ complements, reveals nothing about the N–E except the existence of yet another unexplained restriction. The closest we can get to the intended interpretation of (26a) is (27). (27)
the song Wer nur die Sehnsucht kennt by Schubert and the one of the same title by Tchaikovsky
Of the same title, since it appears after one, must be an N’’ complement. Since it is anaphoric to Wer nur die Sehnsucht kennt, there is weak evidence for N’’ attachment of the latter expression as well, in accordance with the paraphrase judgments in (24a, b).
336
o n t he p hra se th e p hrase ‘ the phrase ’
Suppose then that E is attached to N’’. This predicts that it can be preceded by N’ complements. Such a prediction seems unfortunately to be incorrect: (28) a. the (well-known) variety cabernet sauvignon b. the (well-known) variety of wine cabernet sauvignon (29) a. the (famous) composer Franz Schubert b. ?the (famous) composer of lieder Franz Schubert Notice that the addition of a comma, placing E unequivocally in appositive (N’’’) position, improves (28b) and (29b) considerably: (30) a. the well-known variety of wine, cabernet sauvignon b. the famous composer of lieder, Franz Schubert Thus the N–E appears to exclude N’ complements, arguing that E is itself in N’ position. How is this conflicting evidence to be resolved? A possible factor mitigating the argument for N’ attachment concerns prosody. Oddly enough, (31) seems better than (29b). (31) the famous composer of lieder Johannes Brahms Similarly, compare (32a) and (32b), where the choice of E affects acceptability. (I assume of yours is an N’ complement.) (32) a. ??That friend of yours Harry has been snooping around again. b. That friend of yours Franz Schubert has been snooping around again. The only conceivable difference that could be relevant in distinguishing (29b) from (31) and (32a) from (32b) is the prosodic pattern of E. Pursuing this lead, notice that Franz Schubert is out in (29b) but acceptable in (32b). The difference between these two cases may be tied to the stress patterns of their N’ complements: of yours is unstressed, while of lieder carries the highest stress in the phrase up to that point. Franz carries secondary stress in Franz Schubert; it is acceptable adjacent to unstressed of yours but not adjacent to stressed lieder. On the other hand, primary stressed Harry is acceptable in neither context (?the famous composer of lieder Harry), while Johannes Brahms, which begins with an unstressed syllable, is acceptable in both (that friend of yours Johannes Brahms). What seems to emerge from these observations is the existence of a prescribed prosodic pattern for the N–E which prohibits N’ complements that result in stresses falling too close together. Without working out the details of this pattern, we see that it potentially
on th e ph rase the phrase ‘ th e ph rase ’
337
saves the N’ analysis, at the cost of an area of non-autonomy between phonology and syntax.d
10.5 E is not the head We therefore settle tentatively on the theory that E is attached to N’’. For completeness, however, we should mention a possible analysis radically different from those we have been considering, in all of which N is the head and E a complement. On semantic grounds, it would not be counter-intuitive to claim that E is the head and the N a specifier, for the phrase as a whole refers to the same thing as E alone does. (33) sketches such an analysis.
(33)
NP | NP N¢¢ | the noise N¢ | E | *****
Under this analysis, the N could be seen as simply adding information about the category of E’s referent—in much the same manner as the appositives in (34), whose structure is given in (35). (34)
a. b. c. d. e.
in the house, a PP engoopium, a rare element William Blake, a/the famous poet Harriauschus, an unusual name $, the symbol for dollar
d
Similarly peculiar prosodic conditions show up again in the next chapter in connection with the ‘saladsalad’ construction. A troubling piece of evidence I forgot: If E is an N’’ complement (i.e. an adjunct), it ought to be able to be preceded by other N’’ complements, as these are usually freely ordered. But it cannot be (except with comma intonation, in which case it is the appositive E instead). (i)
a. the song by Schubert Wer nur die Sehnsucht kennt b. the newscaster on MSNBC Dan Rather
On the other hand, these may be out for the same prosodic reasons as (28b) and (29b).
o n t he p hra se th e p hrase ‘ the phrase ’
338
(35)
NP | N¢¢ | N¢ | E | $
NP the symbol for dollar
The analysis could be reinforced by noticing the virtual interchangeability of the two constructions in examples such as (36). (36) The subject of today’s lecture is . . . a. in the house, a PP. b. the PP in the house. c. engoopium, a rare element. d. the rare element engoopium. However, although there is a good deal of semantic overlap between the N–E and the construction in (34)–(35), the parallel is by no means complete. Here is a selection of examples where one construction is acceptable and the other is not. (37) a. b. c. d. e. f. (38) a. b. c. d. e. f.
the important person John the doctor Bill the graduate student Alan Sparks the pleasant drink wine the profitable crop rice the oldest violinist Jascha
John, an important person Bill, a doctor Alan Sparks, a graduate student wine, a pleasant drink rice, a profitable crop Jascha, the oldest violinist
(39) a. the number 4 b. the note E♭ c. that fool John (40) a. 4, a/the number b. E♭, a note c. ?John, that fool
on th e ph rase the phrase ‘ th e ph rase ’ (41)
339
a. We mentioned the disgusting noise . We mentioned (, a disgusting noise). b. We discussed the odd pattern dum-da-dum-dum We discussed dum-da-dum-dum (, an odd pattern). c. Then Capt. Kangaroo uttered the magic words Abracadabra Please and Thank You.
Then Capt. Kangaroo uttered Abracadabra Please and Thank You, the magic words.
These differ for various reasons, some of which have been mentioned already. (37b, c, d, e) and perhaps (37a) violate the pragmatic constraint on the N–E discussed in connection with (10) and (11); this constraint evidently does not apply to E, NP. (37f) violates the constraint against superlatives illustrated in (20); this too does not apply to E, NP. In the next group of examples, (39)–(40), it is the E, NP construction that is unacceptable. I will not attempt to uncover the restrictions responsible for this unacceptability, except to point out that a prosodic account like that invoked for (28)–(29) is unlikely, in view of the fact that (40a) has precisely the same stress pattern as the acceptable (38b). Example (41) shows that the apparent match between the two constructions illustrated in (36) is in fact less than perfect. Unlike the subject of be, the objects of mention, discuss, and utter cannot be free expressions. Thus E is unacceptable in this position, with or without the appositive. This strongly suggests that the N–E and E, NP have different heads, in particular that the former is headed by N, as assumed earlier. As further evidence that this distinction is correct, notice that the construction in which N is unambiguously head and E an appositive is acceptable in these contexts. (42)
a. We mentioned an extremely disgusting noise, . b. We discussed an odd pattern, dum-da-dum-dum. c. Then Capt. Kangaroo uttered some magic words, Abracadabra Please and Thank You.
Hence, in contexts that require a genuine noun and not E as head, the N–E patterns with the N, E and not with E, NP, undermining any hypothesis like (33) for its structure.
10.6 Conclusion We have found three constructions with roughly similar meaning but distinct syntax.
340
(43)
o n t he p hra se th e p hrase ‘ the phrase ’
a. the N–E Art
NP | N¢¢
| N¢ | N b. a/the N, E
NP Art
c. E, NP
E
N¢¢ | N¢ | N
E
NP
|
N¢¢
NP
| N¢ | E It has developed that each has distinct syntactic and semantic properties, and none can be reduced to another at either syntactic or semantic levels. (Furthermore, we have not explored constructions like President Truman or the country of Costa Rica, which have related semantics but do not permit a free expression E.) One reader has remarked that these constructions ‘lie at the edge of linguistic structure—in my judgment, just at the point where linguistic structure slides off into chaos’, and that ‘one would presumably not want to use them to throw light on the core of linguistic theory’.e My conclusion, however, is quite the opposite. The only ‘chaos’ in these constructions lies in the appearance of the free expression E, and the judgments seem to me no more delicate or unreliable than those in contemporary discussions of ‘core’ matters such as control.
e
This is an early version of the core–periphery distinction. My argument here sort of anticipates that in Culicover 1999, although Culicover does it much better: here I am not yet seeing this as a general problem posed by constructions with noncanonical syntax.
on th e ph rase the phrase ‘ th e ph rase ’
341
Moreover, the data presented here clearly are of considerable relevance to the treatment of definiteness, surely a central aspect of grammar. And these constructions present every bit as much of a problem to the language learner as more extensively studied ones. For example, how does the language learner correlate the absence of comma intonation in the N–E with inherent definiteness, and why does the language learner impose the prosodic filter observed in (28)–(29)? In short, all the usual problems arise. My object here, however, has not been to provide solutions to these problems so much as simply to expose them to critical view.f
f Within the framework of Simpler Syntax (Culicover and Jackendoff 2005), the syntactic distinctions among three X-bar levels are not available: the structure of NP is simply (i) (the double line indicates the head of the phrase).
(i)
NP
Det
N
[complements]
[adjuncts]
So some other solution must be found for the N–E. My inclination would be to view it as a double-headed construction along the lines of (ii).
NP
(ii) Det
N
E
The interpretation of this structure appears to be that N and E co-specify each other; in particular, E has to fall under the description offered by N. This approach has to face the question of where the other constituents of NP go in (ii). We have seen that adjuncts all go to the right of E, which is what we would expect if E is a head. The only problem is where complements of N might go. From the evidence presented in section 10.4, it looks like complements of N are possible directly after N—but only when they are prosodically acceptable. To be sure, section 10.5 offered an argument that E is not the head. The most telling evidence in the present context is that verbs like mention and discuss permit the N–E but not E, a/the N as their object. But if syntactic selection is selecting specifically for a noun head, it will find one in structure (ii) and it can ignore the co-head E. That is, this evidence shows that E is not the sole head of the NP. The structure in (ii) is unusual, of course, in having two heads, a major violation of standard X-bar phrase structure. My sense is that in some respects the N–E resembles serial verb constructions (Aikhenvald and Dixon 2007). There too it is often difficult to argue that one verb or the other is head on other than semantic grounds, and different languages play out the possibilities for agreement and complementation in different ways. My reading of the evidence is that analyses are frequently forced into uncomfortable situations by assuming that one verb must always be head and the other subordinate; in many cases a more natural treatment arises by allowing the verbs to be co-heads. This of course is a topic for much further research. There remains the question for Simpler Syntax of how appositives are structured—including of course the appositives in a/the N, E and in E, a/the N. But this is a more general problem that need not be addressed in the present context.
chapter 11
Contrastive Focus Reduplication in English (The Salad-Salad Paper) (2004) Jila Ghomeshi, Ray Jackendoff, Nicole Rosen, and Kevin Russell
11.1 Introduction This chapter presents a phenomenon of colloquial English that has previously received scant attention in the literature, which we call Contrastive Reduplication (CR); Horn 2006 refers to it as Cloning. Examples of this construction are given in (1):1 [This chapter first appeared in Natural Language and Linguistic Theory 22 (2004), 307–57, and is reprinted here by permission of the publisher, Springer Netherlands.] The chapter began as a dinner discussion at the 1999 LSA Linguistics Institute, and we’d like to acknowledge our two other dining companions: Andrea Wilhelm and Qwynten D. Richards. For helpful comments and discussion we thank Peter Culicover, Larry Horn, Marie-Odile Junker, Leslie Saxon, Lisa Travis, Moira Yip, and audiences at the University of Manitoba (Fall 1999 colloquium series), the UCL conference on the Interaction between Syntax and Pragmatics (April 2000), and at the Canadian Linguistic Association (June 2000). Ray Jackendoff is happy to acknowledge the support of NIH Grant 03660 to Brandeis University, and especially of the Wissenschaftskolleg zu Berlin, where he was in residence during much of the time this chapter was being developed. 1 Most of the examples in this chapter come from a corpus we have gathered of examples occurring in natural speech, written texts, or in films and television shows (available at http://www. umanitoba.ca/faculties/arts/linguistics/russell/redup-corpus.html). Examples drawn from the corpus are indicated by the sign Ó. All examples show the reduplicant in small caps even when not in
c o nt ras t iv e f o c us re d uplication
343
(1) a. I’ll make the tuna salad, and you make the salad-salad. Ó b. Like ’em-like ’em? Or, I’d-like-to-get-store-credit-for-that-amount like ’em?2 Ó c. Is he French or French-French? d. No, I didn’t get up-up. Ó e. That’s not Auckland-Auckland, is it? Ó f. My car isn’t mine-mine; it’s my parents’. g. Oh, we’re not living together-living together. As illustrated in these examples, CR targets nouns (1a), verbs (and optionally pronominal material to their right) (1b), adjectives (1c), verb particles (1d), proper names (1e), pronouns (1f), and lexicalized expressions (1g). The semantic effect of this construction is to focus the denotation of the reduplicated element on a more sharply delimited, more specialized, range. For instance, salad-salad in (1a) denotes specifically green salad as opposed to salads in general, and, in the context in which (1e) was used, Auckland-Auckland denotes the city in New Zealand as opposed to other cities that may happen to have this name. For a first approximation, we characterize this effect as denoting the prototypical instance of the reduplicated lexical expression. CR is quite common in North American English. We have recorded it used by speakers in their twenties and their seventies, by speakers of British English, and even by native speakers of other languages (when speaking English). The phenomenon is of course much rarer in written corpora (though we have one example from John Steinbeck, (3e) below); we have however found numerous instances of it in film and television transcripts. CR is of interest for a number of reasons. First, although reduplication has been studied in many languages of the world, it has rarely been cited as a grammatical phenomenon in English. In fact, CR is not the only reduplication process in English: there are at least six others of various degrees of productivity: (2) a. ‘Baby-talk’ reduplication, e.g., choo-choo, wee-wee b. Multiple partial reduplications, e.g., hap-hap-happy (as in song lyrics)
the original text. [Another corpus appears in Hohenhaus 2004, a paper written entirely independently from ours in mutual ignorance.] 2 This example also contains an instance of another construction, the ‘you-can-put-anythingyou-want-before-the-head’ construction. Our corpus of CR, especially in the television scripts, is rife with such examples, which also deserve analysis. See n. 26 for some discussion.
344
contrastive focus reduplication c. Deprecative reduplication, e.g., table-shmablea d. ‘Rhyme combinations’: super-duper, willy-nilly, pell-mell, okey-dokey, hanky-panky. . . e. ‘Ablaut combinations’: flim-flam, zig-zag, sing-song, pitter-patter, riffraff, mish-mash . . . b f. Intensive reduplication:3 You are sick sick sick!
A second reason CR is of interest is its semantics. Crosslinguistically, reduplication phenomena are used to express such factors as plurality, distributivity (each X), perfective aspect, continuous/progressive/habitual aspect (keep Ving, 3
(2f ) shows Intensive reduplication (IR) with adjectives. It also occurs with verbs (ia), nouns (ib), prepositions/adverbs (ic), and pronouns (id, e). (i)
a. b. c. d. e.
Let’s get out there and win win win! All Sandy thinks about is sex sex sex! Prices just keep going up up up. All you think about is you you you. It’s mine mine mine!
At first glance these may appear simply to be examples of the general ability in English to repeat modifiers, as in You are a sick, sick man or You’re really really sick. But IR can apply to categories (e.g. verbs) and in positions (e.g. sentence-finally) that modifier repetition cannot. IR is subject to an interesting constraint: an instance of IR must have at least three items: (ii)
a. You are sick sick sick. b. You are sick sick.
One might think that English IR is like the Telegu and Mokilese reduplication patterns discussed by Moravcsik (1978) that may be triplicated but not duplicated. But it turns out to be more interesting. Sentence-final IR has two possible prosodic patterns. The first (iiia) has balanced stress on each item and the potential for pauses between them. The second (iiib) has an alternating strong-weak-strong stress pattern covering an odd number of items. It is not possible to have any kind of alternating stress pattern with an even number of items (iiic, d). [continued] a [Material excised from the published version for reasons of length:] Cross-linguistically, reduplications which involve a fixed segmental content (such as [m]) in the copy are common, for example, [gi(:)] and [m] in the following examples from Kannada and Persian.
(i)
(ii)
KANNADA (Lidz 2001) ooda ‘run doDDa ‘large’ meele ‘above’ PERSIAN ketaˆb ‘book’ boshqaˆb ‘plate’ shirni ‘sweets’
ooda-giida beeDa doDDa-giDDa meele-giile
‘Don’t run or do related activities’ ‘large and the like’ ‘above and the like’
ketaˆb-metaˆb boshqaˆb-moshqaˆb shirni-mirni
‘books and stuff’ ‘plates and stuff’ ‘sweets and stuff’
b [Material excised from the published version for reasons of length:] The compounds in (2d, e), discussed in Marchand 1969, may at first glance not seem to be reduplications. They are certainly not very productive in modern English. But their forms resemble unquestionable reduplications from other languages, especially given the possibility of pre-specified segments mentioned in subfootnote a. The meanings of many of these forms also fit quite nicely into the crosslinguistic semantic range of reduplication listed below, e.g. the ‘out of control’ semantics of pell-mell and willy-nilly.
c o nt ras t iv e f o c us re d uplication
345
is Ving, Vs habitually), diminutives (little X), augmentatives (big X), intensification (really X), variety and similarity (all different kinds of X, X and such), ‘out of control’, and various other kinds of derivational meaning (e.g. agentive nominal) (this list compiled largely from Moravcsik 1978). All of these uses, of course, are also expressed in other languages by ordinary closed-class morphemes. The prototype/contrastive focus meaning we find for English CR adds a new item to this list. We will see in section 11.2.2 that parallels to English CR exist in other languages, both as reduplicative phenomena and as ordinary closed-class morphemes. What makes CR more than just another curiosity, though, is its difference from other reduplicative phenomena. One of the results of the Prosodic Morphology research program (beginning with McCarthy and Prince 1986) is that in cases of partial reduplication, the reduplicant always forms a well-defined prosodic constituent, such as a foot or a heavy syllable. Each reduplicative morpheme specifies a prosodic category (i.e., it has a prosodically defined ‘template’ in original Prosodic Morphology or it is subject to a prosodic size constraint in Optimality Theory), and as much material as possible will be copied from the base, subject to the prosodic conditions on the reduplicant. From this point of view, there need be nothing qualitatively different about total reduplication: it is simply reduplication where the reduplicant forms a prosodic word rather than a foot or a type of syllable. Unlike the cases of reduplication examined in the Prosodic Morphology tradition, English CR cannot be defined in prosodic terms. This is already evident from (1b, g), which consist of more than a single word. Much of our discussion in this chapter will concern how the ‘scope’ of CR is to be defined. (iii)
a. b. c. d.
You are sick, sick, sick, sick. You are sick-sick-sick (-sick-sick). You are sick-sick-sick-sick. You are sick-sick-sick-sick.
The alternating stress pattern is not available for pre-nominal modifiers, as seen in (iva). However, many speakers can use an alternating stress pattern if the noun occurs in final stressed position, as in (ivb). Overall, then, IR seems to require the construction of binary feet. (iv)
a. ?? You are a sick-sick-sick-sick-sick man. b. You are a sick-sick-sick-sick man.
A variant of (ie) is It is mine, mine, all mine, in which case the final repetition is slightly elaborated. The semantics of CR are much narrower than IR (see section 11.3). In form, the two are disjoint (CR has two items, IR must have at least three), with the result that no example of copying in English will ever be ambiguous between CR and IR. (v)
a. She’s not just goofing off work, she’s SICK-sick. (only CR) b. She’s not just a little odd, she’s SICK-sick-SICK. (only IR)
346
contrastive focus reduplication
The organization of the chapter is as follows. Section 11.2 develops the semantics of CR in more detail, showing a number of alternate meanings it permits, and showing some crosslinguistic parallels. Section 11.3 works through the considerable problems in specifying the ‘scope’ of CR, i.e. the permissible units that can be reduplicated. Section 11.4 works out a solution in the parallel architecture framework of Jackendoff 1997b, 2002, along the way showing how reduplication is to be treated in that framework. On this analysis, CR is a lexical item with syntactic and semantic structure and with reduplicative phonology. What is unusual about its syntactic structure is that it can be adjoined either to a full word, inside an X0, or to some larger phrasal constituent. Section 11.5 sketches an analysis that is more consistent with the assumptions of the Minimalist Program (Chomsky 1995, 2000): the reduplication in CR is treated as a version of head movement where both head and tail of the chain are retained at PF.
11.2 The semantics of CR In this section we first discuss the range of meanings of CR based on our corpus of examples. We conclude that these meanings, taken together, are a kind of contrastive focus. We then present and briefly discuss examples of CR from other languages in order to show that this construction is not found only in English.
11.2.1 Specifying the interpretation Let us explore the semantics of CR in more detail. The use of a word or phrase often leaves open some vagueness, lack of precision, or ambiguity. CR is used as one way to clarify such situations, by specifying a prototypical denotation of the lexical item in contrast to a potentially looser or more specialized reading. This is clearest when CR is applied to simple nouns: (3)
a. I’ll make the tuna salad and you make the salad-salad. Ó b. Look at all the yellow vans on the road. Not vans like ours [i.e. minivans], but van-vans. Ó c. She wasn’t a fancy cow, a Hereford or Black Angus or something, just a cow-cow. Ó d. Should I wear a hat-hat? [as opposed to a yarmulke] Ó e. And Charley is no more like a dog-dog than he is like a cat. Ó f. I had a job-job once. [a ‘real’ 9-to-5 office job, as opposed to an academic job] Ó
c o nt ras t iv e f o c us re d uplication
347
This characterization is precisely the informal one given by Horn (1993). He briefly discusses CR (which he labels, following Dray 1987, the double construction), stating: ‘As a rough approximation, we can say that the reduplicated modifier singles out a member or subset of the extension of the noun [or verb, or adjective, or preposition—JG et al.] that represents a true, real, default, or prototype instance’ (48). As already seen in (1), CR can apply not only to nouns, but to a range of lexical categories. Regardless of the lexical category, however, reduplication signals that the ‘real’ or prototypical meaning of the lexical item is intended: (4) a. Are you leaving-leaving? [i.e., are you ‘really’ leaving (for good), or are you just stepping out for a minute] b. A: Are you nervous? B: Yeah, but, you know, not nervous-nervous. [i.e. not ‘really’ nervous] Ó c. Lily: You have to get up. Rick: I am up. Lily: I mean up-up. Ó The meaning of the construction is nicely illustrated by the following example, from a novel written in English and translated into German, which lacks CR. In the German version the reduplicated sequence rich-rich gets translated as richtig reich:4 (5) a. They are rich, of course; obscenely rich by the world’s standards; but not rich-rich, not New York City rich. Ó [Michael Cunningham, The Hours (New York, 1998), 91] b. . . . aber nicht richtig reich, nicht nach den Maßsta¨ben von New York City. ( . . . but not really rich, not by the standards of NYC.) If the function of CR is to restrict the meaning of an item to its central or prototypical meaning, and if functional items lack the appropriate sort of semantic variation, then CR should not occur with functional items. This prediction is borne out by the ungrammatical examples below: (6)
Are you sick, or are-are you sick? [Auxiliary] I didn’t just read the book, I read the-the book! [Determiner]
4 We thank Roland Pfau for both noting the English example and then looking up the relevant sentence in the German translation. [Hohenhaus 2004 notes one lexicalized example of CR in German, Film-Film, but finds no productive use.]
348
contrastive focus reduplication
This hypothesis is nicely confirmed with the class of prepositions, where some prepositions fall on the more contentful side while others have purely functional uses. As expected, the more contentful prepositions can undergo CR (7a), while the functional ones cannot (7b): (7)
a. A: B: A: b. A: B:
I was sitting across from your husband at dinner. Really? Well, not across-across (but close by). [contentful preposition] Did you go to Montreal? Well, not to-to. [functional preposition]
It may be surprising that CR is possible with proper names, which, as rigid designators, should not be ambiguous. But there are three contexts in which CR occurs with proper names, all involving some context-dependent ambiguity. The first is where a noun is ambiguous between proper and common. (8)
A: B: A: B:
So then who’s coming through the Stargate? Gods. Huh? Not as in God-God. Ra played a god, the sun god. Ó
The second is when the discourse participants know more than one individual with the same name. In this case CR picks out the most salient (important or well-known) referent: (9)
a. So did you go to the movie with Dave-Dave, or with Dave? [i.e. the Dave best known to the speaker and hearer] b. We call him psycho Marcus in order to distinguish him from normal Marcus and Marcus-Marcus. Ó c. Oh, that’s Beacon-Street-Beacon Street! [uttered by a person being given directions, who has just realized that the Beacon Street in West Newton is in fact a continuation of the well-known Beacon Street in Boston] Ó
Third, CR may contrast an individual’s typical behavior against abnormal and uncharacteristic behavior, or one’s inner self against a public persona, or one’s ‘true’ self in cases where a person has been cloned, replaced by a robot, transferred into another body, possessed by supernatural beings, or fallen victim to one of the many other identity-clouding fates that so often befall film and TV characters. (10) A: That doesn’t sound like Murray. B: Remember that he joined that cult the Spiritologists. A: Murray-Murray!? Ó
c o nt r as t iv e f o c us re d uplication
349
Even pronouns may be reduplicated in analogous situations, when the nature of the individual being denoted is in question: (11)
a. It might have been me, but it wasn’t me-me. b. You see me for a couple of hours out of every day, and you think you know me? The me-me? Ó
Horn (2006) categorizes the semantics of this construction into four types: (a) prototype meaning (which we have already discussed), (b) literal meaning, (c) intensified meaning, and (d) ‘value-added’ meaning. An example of literal meaning appears in (12), where reduplication signals that a literal rather than euphemistic interpretation of ‘coming in for coffee’ is intended: (12)
[Dialogue between a married couple, recently separated and now living apart.] A: Maybe you’d like to come in and have some coffee? B: Yeah, I’d like that. A: Just coffee-coffee, no double meanings. Ó
The intensified type of meaning is illustrated in (13) (¼ (4b)). (13)
A: [to B, who is about to give a recital]: Are you nervous? B: Yeah, but, you know, not nervous-nervous. Ó
The ‘value-added’ meaning is illustrated in (14), adapted from Dray (1987) as cited by Horn (1993, 50.13): living together is taken in the sense ‘living together as lovers’. (14)
A: I hear you guys are, um, living together now. B: Well, we’re not living together-living together.
As Dray and Horn note, the choice of reading for CR can be determined by context. Consider (15), a slightly different reply by B to A’s remark in (14). Here the context requires the literal reading for living together, ‘living together as roommates’. (15)
B: Well, we’re only living together-living together.
When used with the more specialized, innuendo-laden meanings, the CR construction is frequently spoken with a distinctive intonation contour, which Horn notes and marks with a ‘raised eyebrow’ diacritic. Another such ‘raised eyebrow’ example is (16). (16)
A: I’m late, Lois. B: Well, if you didn’t spend so much time on your hair. . . A: No, I mean late-late! Ó
350
contrastive focus reduplication
The relationship between CR and potential ambiguity is so strong that it can force listeners to infer the ambiguity of a term even when they were not previously aware of any. Given only the use of CR in the following exchange, A was forced to conclude that there was, after all, more than one type of bowling. (17) A: I’ve been invited to go bowling tonight. B: bowling-bowling? Ó (B was in fact contrasting regular bowling with ‘bingo bowling’ and ‘glow bowling’, barbarisms which we need not go into here.) Finally, there are two examples in our corpus where the meaning of CR seems to be ‘the obvious one’, i.e. merely very high salience with no hint of prototypicality, ambiguity, or contrast. (18) a. A: B: A: b. A: B: A: B: A:
Did you check out the leak in the bathroom? What leak? The leak-leak. [drags her into the bathroom] Ó Just loosening up. Doing a little bit of kong-chi. . . . What the hell does kong-chi mean? Kong-chi! Huh? You know, kong-chi-kong-chi! It’s an ancient oriental artform of loosening up. Ó
Given this variation in the use of CR, it is not clear to us whether CR is itself polysemous or whether it can pick out contextually salient readings in addition to objectively prototypical ones. In this sense, saying that CR signals the ‘prototypical’ reading is too restricted, but we will continue to use it for lack of a better characterization.c
c
Hohenhaus 2004 also notices the prototype and extreme readings. Whitton 2007a, 2007b argues, however, that the range of possible readings of CR is even broader. For example, the CR DRINK-drink has five different interpretations in the following examples (from Whitton’s corpus): (i)
a. You said in an earlier article that if you must have a DRINK-drink, go with the hard liquor. [alcoholic drink] b. ‘Do you want a bottle of wine?’ Mac asks. ‘I think I’ll have a DRINK-drink,’ I say, and when the waiter comes I order a martini. [hard liquor] c. around 3 euros a shot and 8 euros a DRINK-drink [mixed drink] d. [Two people at fast food restaurant sharing a meal deal] A: What do you wanna get? B: I’ll probably just get water so if you want a DRINK-drink get whatever you want. [soda/juice/etc.] e. A: I am on my own with the BBQ! Come on, girls, I need some drink ideas. B: Are you looking for alcohol? Or just a DRINK-drink? [non-alcoholic drink] [continued]
c o nt ras t iv e f o c us re d uplication
351
Our claim that CR serves to restrict the denotation of a lexical item to its prototype bears a strong resemblance to Lasersohn’s (1999) account of linguistic imprecision and his description of pragmatic ‘slack regulators’.5 Lasersohn’s term applies to a word like exactly in the sentence Mary arrived at exactly 3 o’clock, or the word perfectly in a sentence like This ball is perfectly spherical. The idea is that these sentences differ from their counterparts without slack regulators (Mary arrived at 3 o’clock; This ball is spherical) in that they allow for less deviation from the truth—what Lasersohn calls pragmatic slack. Without a slack regulator like exactly, and in a context in which precision is not important, it is possible to say Mary arrived at 3 o’clock even if she arrived at a few minutes before or after 3. Lasersohn calls the set of objects that is associated with the denotation of a proposition (or lexical item) but differs from it in pragmatically ignorable ways, a ‘pragmatic halo’. Slack regulators serve to shrink or tighten the pragmatic halo of the expressions they combine with. What CR and Lasersohn’s slack regulators have in common is that they both have a set-shrinking effect, the effect of narrowing down the range of appropriate referents of a lexical item. How they differ (putting aside the fact that CR cannot apply to propositions) is in the types of sets involved. Lasersohn defines a pragmatic halo in truth-theoretic terms: his slack regulators make fewer truth-conditionally false statements appropriate in a given context. By contrast, CR rules out not denotations that are truth-conditionally false (not false-false), but rather denotations that are less prototypical: many things are salads without being salad-salads. The notion of a set of alternatives against which an expression is evaluated usually comes up in discussions of contrastive focus, and there is certainly some similarity between CR and contrastive focus. In many cases CR is explicitly paired with a non-reduplicated version of the same phrase nearby in the discourse. The copy has a focus accent just like that of a contrastively focused
5
Thanks to an anonymous reviewer for bringing this article to our attention. Lasersohn’s ‘slack regulators’ appear to be the semantic converse of ‘hedges’ such as sort of, discussed by Lakoff 1972. Lakoff also mentions phrases such as strictly speaking, which are precisely slack regulators in Lasersohn’s sense. Whitton also points out that, under an interpretation dealing with prototypical members of a category, one would not expect homonyms, which denote entirely different categories. Yet examples turn up like (ii) (from Whitton’s corpus) and (iii) (from ours). (ii) So . . . do they fly when you try to hit ’em with a bat? I mean a baseball bat not a BAT-bat. (iii) There’s a guy who collects fans, these are not sports fans but FANS-fans. (¼(32a) ) She suggests that CR simply marks an item as contrastive along some scale or another, and the appropriate scale is chosen pragmatically. Horn’s four possibilities, then, represent four prominent scales along which items can be contrasted.
352
contrastive focus reduplication
modifier: not a red book, not a book-book. The semantic difference between the two is illustrated by the following pairs of sentences: (19) a. I didn’t give the book to JOHN. [Contrastive focus] Contrast set: {John, Bill, Dave, Sue . . . } b. I didn’t give the book to John-John. [CR] Contrast set: {John1, John2 . . . } (20) a. It wasn’t a GOAT. [Contrastive focus] Contrast set: {goats, horses, pigs, sheep . . . } b. It wasn’t a goat-goat. [CR] Contrast set: {prototypical goats, non-prototypical or non-literal (euphemistic, figurative) goats} Contrastive focus on a word signals that it is being contrasted with other words of the same type. CR, on the other hand, signals that one meaning of the word is being contrasted with other possible meanings.
11.2.2 CR in other languages CR is not limited to English. Wierzbicka (1991) discusses syntactic reduplication or ‘raddoppiamento’ in Italian—a phenomenon that is very similar, if not identical, to CR in English. Wierzbicka carefully distinguishes Italian syntactic reduplication from two seemingly similar phenomena. She states that a reduplicated expression like adagio adagio ‘slowly slowly’ can be distinguished from repetition (e.g., adagio, adagio) in that the former is a pauseless expression while the latter, as indicated by the use of the comma, is not. She also distinguishes reduplication from what she calls ‘clausal’ repetition (e.g. Come in, come in! All right, all right!) based on the fact that the illocutionary force of this type of repetition involves a sense of urgency. Most of the properties that Wierzbicka attributes to Italian reduplication are precisely those we find for CR in English: both processes apply to more than one lexical category (e.g. adjectives, adverbs, nouns) and both operate on ‘words rather than on morphemes’ (p. 255).6 As for the meaning, Wierzbicka notes that the characterization given in Italian grammars, as well as for a similar phenomenon in Vulgar Latin (Grandgent 6
One possible area for future research would be to examine to what extent Italian and English reduplication overlap. For instance, as we have noted, CR in English can apply to proper names and pronouns while Wierzbicka makes no mention of whether this is possible in Italian. On the other hand, she considers the syntactic reduplication of adjectives and adverbs (but not nouns) in Italian to have an added ‘emotional’ component (p. 266), while we have found no evidence of this in English.
c o nt r as t iv e f o c us re d uplication
353
1908, 32) is ‘intensification’. One problem with such a characterization, however, is that intensifiers like molto ‘very’ are restricted to gradable qualities while reduplication is not. Citing Lepschy and Lepschy (1984), Wierzbicka proposes instead that the communicative import of syntactic reduplication is to insist on the validity of what is said. She writes that ‘[i]n calling someone’s eyes neri neri the speaker insists that these eyes were ‘really black, literally black’ (p. 264) and later gives further examples of nominal reduplication such as: ‘brodo brodo ‘‘broth broth’’ i.e. genuine broth, and lana lana ‘‘wool wool’’, i.e. genuine wool’ (p. 265). This characterization seems compatible with one of the readings discussed in the previous subsection. Spanish and Russian also appear to have CR. Horn (1993) gives the following example from Spanish taken from the movie Women on the Verge of a Nervous Breakdown: (21)
No es una casa-casa. ‘This isn’t a real [sic] house.’ [Horn 1993, 49.10a]
Asya Pereltsvaig (p.c.) has pointed out the following example from Russian, in which the first hyphenated element looks much like CR. She observes that the stress falls on the first element in both zheltyj-zheltyj and limonno-zheltyj since they are being used contrastively. However, compounds in Russian are usually right-dominant, suggesting that, as in English (see section 11.3.1, n. 8), this process is distinct from compounding. (22)
On zheltyj-zheltyj, a ne limonno-zheltyj. he yellow-yellow, and not lemon-yellow ‘It’s yellow-yellow, not lemon-yellow.’
Persian, too, has something akin to CR, as pointed out by Arsalan Kahnemuyipour (p.c.), who has provided the following examples: (23)
a. loxt-e lox ke na-bud naked-EZ naked FOCUS neg-was ‘S/he wasn’t naked-naked.’ b. bi-kaˆr-e bi-kaˆr ke nist, naqqaˆsh-e without-job-EZ without-job FOCUS NEG.is painter-3sg ‘S/he isn’t unemployed-unemployed, s/he’s an artist.’
In contrast to CR in English, Persian permits this type of copying only with adjectives. The meaning differs also, perhaps as a consequence of this fact, and is better characterized as ‘completely X’ rather than ‘really X’. A notable fact
354
contrastive focus reduplication
about these examples is that the Ezafe vowel (EZ) intervenes between the two copies. This vowel appears between nominal heads and their modifiers in Persian but does not appear within compounds (see, for example, Ghomeshi 1997). Mutaka and Hyman 1990 observe that Kinande, unlike other neighboring Bantu languages, has a form of CR, restricted to nouns, e.g. o.ku-gulu ‘leg’, o.ku-gulu-gulu ‘a real leg’. Just as the semantics of other reduplicative phenomena are paralleled by morphological affixes in other languages, the semantics of CR can also be expressed morphologically. To take one example, Poser (1991) discusses a prefix ma- in Japanese that ‘restricts the denotation of the base form to . . . a canonical point that represents the absolute state’ (p. 453). (24) Japanese (Poser 1991, 449–50) mae ‘front’ maNmae fuyu ‘winter’ mafuyu siro ‘white’ massiro kita ‘north’ makita aka ‘red’ makka
‘right in front’ ‘dead of winter’ ‘snow white’ ‘due north’ ‘deep red’
Ma- can also attach to terms that denote a class, and the ma- prefixed form picks out the most common or prototypical member of the class—what Poser calls ‘the cognitive reference point’, following Rosch (1975). (25) Japanese (Poser 1991, 454) kamo ‘wild duck’ azi ‘horse mackerel’ koti
‘flathead’
magamo maazi magoti
‘the mallard duck’ ‘the horse mackerel Trachurus trachurus’7 ‘the common flathead’
We have also found cases in the literature where a similar focus meaning is marked by potentially non-local copying, i.e., the two copies need not be adjacent, as they must be in English CR. We discuss these briefly in section 11.5. 7
It is amusing, though probably nothing more than coincidence, that the Latin species names used by biologists often involve the repetition of the genus name for the prototypical species belonging to the genus—something that looks very much like CR. Consider the following examples: (i) (ii) (iii) (iv)
Vulpes zerda (fennec fox) Vulpes velox (swift fox) Vulpes pallida (African sand fox) Vulpes vulpes (red fox)
(v) (vi) (vii) (viii)
Mephitis macroura (hooded skunk) Mephitis mephitis (striped skunk) Bison bonasus (European bison, or wisent) Bison bison (the real thing)
c o nt r as t iv e f o c us re d uplication
355
11.3 The scope of CR 11.3.1 The problem What makes CR most interesting from a theoretical point of view is the fact that its ‘scope’ (or domain of copying) is not altogether straightforward. The phonologically based reduplications that have been studied in depth all result in the more or less perfect copying of a legitimate prosodic constituent, such as a heavy syllable, a foot, or a prosodic word. However, this is not the case with CR. On the one hand, the copied constituent can be smaller than a phonological word, as when on occasion a word reduplicates without its inflection: (26)
. . . and here are the glove-gloves. Ó
We take up these cases in section 11.3.2. On the other hand, there are two circumstances where CR applies to sequences longer than a word. The first—and the more problematic—is that of the verb plus its clitics to the right-hand side. (27)
a. I don’t like him-like him. b. . . . you mean thought-about-it considered it or just considered-itconsidered it. Ó
We take up these cases in section 11.3.3. The second case where CR extends beyond a word involves idioms. (28)
a. out-of-her-mind-out of her mind b. over-the-hill-over the hill c. sleeping-together-sleeping together
Indeed, trying to copy just a single word of these idiomatic phrases is ungrammatical: (29)
a. out-out of her mind/out of her mind-mind b. over-over the hill/over the hill-hill c. sleeping-sleeping together/sleeping together-together [in intended sense]
By contrast, syntactically parallel but nonidiomatic strings cannot undergo CR: (30)
a. b. c. d.
I didn’t put it over-the-stove-over the stove. I didn’t hide it under-the-sofa-under the sofa. We weren’t sleeping-apart-sleeping apart. We weren’t singing-together-singing together.
356
contrastive focus reduplication
The challenge presented by CR is to find an analysis that encompasses all these possibilities.8
11.3.2 Scope of CR smaller than a word Let us examine more closely the cases where the scope of CR is smaller than a word. Basically, CR can copy less than a word by leaving out inflectional suffixes and copying only the stem. Technically, the copy is still a word, but it is smaller than the word that it ‘copies from’. Our corpus contains many examples of both uncopied and copied inflectional affixes, such as those in (31) and (32) respectively. (Statistically, it is more common for the verb past tense suffix to copy and the noun plural suffix not to copy.) (31) Examples where inflectional morphology does not copy a. . . . and here are the glove-gloves. [real boxing gloves as opposed to smaller practice ones] Ó b. We’re not one of those couple-couples. Ó c. Not vans like ours [i.e., minivans], but van-vans. Ó d. In fact I barely talked to him. Not talk-talked. Ó e. But how can we tell when the growing pains stop and the pain-pains take over? Ó f. [I] didn’t have a lot of friend-friends. Girlfriends. Ó g. Those guy-guys, y’know? Those guys with skills? Ó h. Not kid-kids, men-kids. Ó (32) Examples where inflectional morphology copies a. There’s a guy who collects fans, these are not sports fans but fans-fans. Ó b. You mean cried-cried, or cried because something heavy fell on you? Ó 8 We can see already that CR, despite the similarity of its prosody to that of compounds (compare salad-salad and tuna salad), does not submit to analysis as an unusual form of compound. Aside from its prosody, it bears no resemblance to other forms of compounding in English. In particular, there is no form of compounding that involves phrases, along the lines of (27) and (28). Compounding typically excludes overt inflection on the first element as is found in, for instance, gloves-gloves as opposed to glove( s)-wearer. Furthermore, compounding does not apply to proper names, prepositions, and especially pronouns as in mine-mine and me-me. Furthermore, as observed in section 11.2.2, Russian’s version of CR is left-stressed, while its compounds are right-stressed; and the Persian version of CR uses the Ezafe vowel while Persian compounds do not. So we can discard an analysis in terms of compounding as a nonstarter.d d Hohenhaus 2004 does consider CR a form of compounding, calling it Identical Constituent Compounding. However, his data are derived by scanning online corpora for repetitions of words. He therefore does not notice the cases in which the scope of CR is larger than a word—one of our important arguments that it is not a kind of compounding.
c o nt r as t iv e f o c us re d uplication
357
c. It has will-have-going-to-have-happened happened. But it hasn’t actually happened-happened yet . . . actually. Ó d. I think she’s older-older. [in reference to relative sibling ages] Ó e. No. I’m not leaving-leaving. Ó f. We should approve tonight’s minutes . . . tonight’s-tonight’s minutes. Ó For some cases of deleted inflection, there may seem to be a phonotactic motivation. For example, talk-talked instead of talked-talked avoids a difficult consonant cluster. But failure to copy inflectional suffixes cannot be attributed solely to phonotactics, given other examples such as guy-guys that would not seem to have called for simplification. Furthermore, phonotactically identical sequences inside single morphemes are not simplified (ac-act, pri-prize), so the conditions on CR that allow guy-guys do somehow have to make reference to morphological constituency. Irregular inflectional morphology always has to copy. Compare the attested examples of CR in (33), where the irregularly inflected word has copied whole, with the ungrammatical examples in (34b), where only the stem has copied. (33)
a. When you say she’s getting better, do you mean better-better? Ó b. D’you mean a lot of people-people, or a lot of women-people? Ó
(34)
a. geese-geese taught-taught seen-seen b. goose-geese teach-taught see-seen
Derivational morphology also necessarily copies in CR, as illustrated in (35). (35)
a. relationship-relationship actor-actor married-married b. relation-relationship act-actor marry-married [as adjective]
Examples of derivational morphology in our corpus include those in (36). The only case of non-copying derivation in our corpus is that in (37), which may be related to the suffix -ing’s status on the borderline between inflection and derivation.
358
contrastive focus reduplication
(36) a. I could never be with a guy in a relationship-relationship. Ó b. Well, not surprising-surprising. Ó c. Bowling-bowling? Ó d. He’s a Christian, I mean a Christian-Christian. Ó e. A: Actually, we’re done. B: Done-done. Ó f. There’s cool-geeky and there’s geeky-geeky. I’m only ever going to be geeky-geeky. Ó (37) I like wind-surfing not surf-surfing. Ó It is possible to copy entire compounds, as in (38). (38) a. A: When was the last time you had a boyfriend? B: You mean a boyfriend-boyfriend? Ó b. We have a fireplace-fireplace in the living room. [in discussion of a bricked up fireplace in the speaker’s kitchen] Ó c. Oh, that’s Beacon-street-Beacon-Street. Ó d. . . . green-tea-green-tea . . . [in the context of drinking green tea with rice in it] Ó e. Air-Canada-Air-Canada or Canadian-Air Canada? Ó But it is impossible to copy only part of an established, lexicalized compound, as shown in (39). (39) a. boyfriend-boyfriend fireplace-fireplace pink-slip-pink-slip b. boy-boyfriend fire-fireplace pink-pink-slip c. boy-friend-friend fire-place-place pink-slip-slip The only potential counterexamples to this in our corpus are: (40) a. Totally self-self directed. Ó b. Wednesday is my work-work day. Ó Example (40a) comes from a newspaper ad for tax-deductible retirement investment funds, and copies only part of the lexicalized technical term selfdirected. Nobody we have checked with finds this example acceptable, suggesting that it is one more example of a copywriter trying to use a colloquial construction for its cachet without worrying to get it right. (40b) is more interesting: it is not a reduplication of the lexicalized compound workday, but a newly produced compound meaning ‘my day for work-work’. Summarizing these facts, the definition of the scope of CR must contain some factor of optionality. This optionality permits CR either to apply to a whole word, or else to ignore regular inflectional affixes and apply only to the stem.
c o nt ras t iv e f o c us re d uplication
359
However, neither irregular inflection nor derivational affixes can be ignored by CR. Judging from the single example of (40b), CR appears to be able to go inside a compound just in case the compound is made up on the spot. Combining these facts with the applicability of CR to idioms, seen above in (28), we see that status as a stored lexical unit plays an important role in defining the scope of CR.
11.3.3 CR and object pronouns The other case of optionality in CR is when object pronouns and similar complements are optionally9 copied along with a head (usually a verb). (41)
Object pronouns copied a. Like-’em-like ’em? Or, I’d-like-to-get-store-credit-for-that-amount like ’em? Ó b. . . . you mean thought-about-it considered it or just considered-itconsidered it? Ó c. I mean, I know him, but I don’t know-him-know him. Ó d. Do I like-you-like you? No. You’re a little too neurotic for that. Ó
(42)
PP containing object pronoun copied a. I didn’t sleep-with-her-sleep with her. b. I talked to him that week, but I didn’t talk-to-him-talk to him. Ó c. Did you talk-about-it-talk about it, or did you just mention it?
(43)
Larger combinations copied a. Well, he didn’t give-it-to-me-give it to me (he only lent it to me). b. . . . after we had finally broke-it-off-broke it off, I found out he had bought me an engagement ring. Ó
It is impossible to copy just the preposition but not the following pronoun (44a–c). However, it is possible to copy an object pronoun but not a following PP (44d). (44)
a. b. c. d.
I didn’t sleep-with-sleep with her. I never talked-to-talked to him. He didn’t give-it-to-give it to me. He didn’t give-it-give it to me/Harry. [OK for some but not all speakers]
Copying of object pronouns and complement PPs is not limited to verbs, but is also found with adjectives (afraid-of-him-afraid-of-him) and prepositions 9 There are speakers for whom the copying of some object pronouns is obligatory. See section 11.3.4.
360
contrastive focus reduplication
(across-from-her-across-from-her). For some reason we have been unable to find fully acceptable cases of nouns with their complements, even when pragmatically plausible (e.g. ??news-about-her-news about her). However, if a complement is non-pronominal, it cannot be included in the scope of CR (45a, b). And if the complement is pronominal but focused, CR cannot include it either (45c, d). (45) a. b. c. d.
I can’t say I like-like Mary. I can’t say I like-mary-like Mary. I like him, but I like-like her. I like him, but I like-her-like her.
In all the acceptable cases, the post-head material has been in some sense cliticized to the head. But this does not correlate precisely with cliticization in the phonological sense. Object pronouns have both a full, stressed form (which, however, need not bear a pitch-accent) and a phonologically reduced, unstressed form which cliticizes onto the verb. But a verb þ pronoun sequence can undergo CR regardless of whether the object pronoun is in its full or reduced form: (46) a. Does he like-them-like them? b. Does he like-’em-like ’em? A few other verb þ X sequences may undergo CR; and again this does not depend on the X being phonologically reduced and cliticized onto the verb. Indeed, many speakers find the unreduced forms preferable. Compare the copying of the full infinitival [tu:] with the slight oddness of copying a cliticized wanna. (47) a. Do you want [tu:] or want-[tu:]-want [tu:]? b. ?Do you wanna or wanna-wanna? The question arises of whether this optionality in CR can be characterized phonologically, say by optionally allowing CR to incorporate clitics on the right of the element to be copied. We think not, for several reasons. First, it is impossible to copy just any unstressed syllable that happens to be phonologically cliticized onto the verb, e.g. the indefinite article in the following: (48) a. I wouldn’t date-date a linguist. b. I wouldn’t date-a-date-a linguist One possibility in accounting for this might be to refer to the clitic group, a constituent proposed by Hayes (1989) and Nespor and Vogel (1986) as a level
c o nt ras t iv e f o c us re d uplication
361
of the Prosodic Hierarchy intermediate between the word and the phrase. Hayes defines a clitic group as a prosodic word plus the clitics to its right or left. When a clitic falls between two prosodic words, it belongs to the clitic group of the one to which it is more closely related syntactically.10 So in (48), date and a do not form a copyable clitic group, since a must belong to the same clitic group as linguist. However, Hayes’s definition of clitic group includes clitics on the left as well as the right, and these never reduplicate.11 (49)
I wouldn’t date [cg a-linguist]-[a linguist]
Even restricting optional incorporation to right-hand clitics, it is not clear that unreduced pronouns such as in (46a) should count as clitics, not to mention disyllabic prepositions such as about in (42c). And even were we to develop a sufficiently generous interpretation of ‘clitic’ to accommodate these cases, there would be no obvious generalization to the optionality observed with inflectional morphology, where the reduplicant in, e.g., guy-guys is less than the prosodic word that it reduplicates. Nor would there be a natural generalization to idioms such as living-together-living-together and over-the-hill-over the hill, which contain multiple prosodic words and are therefore larger than a clitic group. A telling example involves the discontinuous idiom take NP to task, which has an open direct object slot. It is possible to reduplicate the idiom if the direct object is an unstressed pronoun (50a). But if it is a full NP or a stressed pronoun, CR is impossible (50b); and of course it is impossible to reduplicate just part of the idiom instead (50c, d). (50)
a. b. c. d.
take-her-to-task-take her to task take-sandy/her-to-task-take Sandy/her to task take-take Sandy to task take Sandy to task-task
This shows that the distinction between unstressed pronouns and other NPs holds of positions other than just the right periphery of the reduplicated item. 10
Recently, the more usual practice in phonology has been to use ‘recursive prosodic words’ to handle cases that were previously analyzed using clitic groups. For arguments for the clitic group as a unique level of the prosodic hierarchy, see Vogel (1990); for arguments against, see Selkirk (1996) and Peperkamp (1997). Essentially the same strengths and weaknesses would carry over if the possible analysis under discussion were to be recast in terms of recursive prosodic words. 11 The exception to this is when the determiner is part of a lexicalized proper name, as in The Hague or The Pas. These can undergo CR: The casino isn’t in The-Pas-The Pas, but in Opaskwayak. (The reserve of the Opaskwayak Cree Nation includes land that is in the northern Manitoba town of The Pas, but not legally part of it.)
362
contrastive focus reduplication
This leads us to question whether the scope of CR has to do with prosody at all. We therefore turn to a (morpho)syntactic characterization; we return to some secondary prosodic considerations in section 11.3.5.
11.3.4 The generalization Another dimension of the distinction between unstressed pronouns and other NPs is that unstressed pronouns are noncontrastive closed-class items. These characteristics are shared by unstressed prepositions, of whatever length. More importantly, they are also shared by inflectional affixes. Thus a characterization in these terms brings out a parallelism between optional object pronouns and optional inflectional affixes in CR. In order to generalize across these two, however, we have to step back and look at the scope of CR in (morpho)syntactic terms. First, an invariable condition of CR is that its scope must include a full lexical item, to whose meaning the semantic effect of CR is applied. This allows CR to apply to stems to which regular inflectional affixes are attached, and it allows CR to apply to idioms, which are lexically listed. But it excludes reduplication of part of an idiom, since the words within an idiom do not have their normal meaning. It also (perhaps more controversially) excludes words composed by derivational morphology, since the meaning of the stem is (typically) not transparent and is not available for contrast within the meaning of the whole word. Next let us turn to the size of the base. First consider examples without optional material. The smallest unit that CR applies to is an X0 inside a word such as glove-[N glove]-s. The largest unit that CR applies to is in idioms such as over-the-hill-[over the hill]. Characterizing this unit depends on one’s theory of phrase structure. In older X-bar theory it was characterizable as X1, the constituent of XP containing the head and its complements, and not containing the specifiers. In more contemporary approaches to phrase structure it is XP, the complement of a functional category. We will use the neutral notation XPmin to stand for this unit. Strikingly, CR never applies to idioms larger than XPmin. For instance, one might imagine applying CR to the sentential idiom My goose is cooked, to mean something like (51a); but it is ungrammatical (51b), and as usual, no part of the idiom can be reduplicated either (51c). (51) a. I’m not just sort of in trouble, my goose is really cooked. b. My-goose-is-cooked-my goose is cooked. c. My goose is cooked-cooked.
c o nt ras t iv e f o c us re d uplication
363
So one important part of the optionality in CR comes from choice in the syntactic size of the base: CR applies to either X0 or XPmin.12 That the same operator should apply to both X0 and XP is not without precedent. An example is English –ing, which can be applied productively to a verb to form a noun (52a), or to a VP to form an NP (52b) (Jackendoff 1977, chapter 9; we leave aside here the process by which ing is attached to the verb rather than to the outside of the VP). (52)
a. [DP/NP his compulsive [N [V drink]-ing] of beer] b. [DP/NP his [NP [VP compulsively drinking beer] ] ]
There is even another example of a reduplication that can apply to bases of variable size: Kannada echo-reduplication (ER), as described by Lidz (2001). In ER, an element X is repeated with the first CV syllable replaced by gi- or gi:-, expressing the meaning ‘X or related things’. Like CR, the size of the copy in Kannada ER can range from part of a word to an entire phrase. All three of the following sentences (Lidz 2001, 378–9) mean ‘Don’t say that I closed the door or did related activities.’ (In (53), the base has been enclosed in brackets and the reduplicant is in boldface.) (53)
a. baagil-annu [much]- gich-id-e anta heeLa-beeDa door-acc closeredup-past-1sg that say-proh b. baagil-annu [much-id-e]gichide anta heeLa-beeDa door-acc close-past-1sg- redup that say-proh c. nannu [baagil-annu much-id-e] giigilannu muchide anta I-nom door-acc close-past-1sg redup that heeLa-beeDa say-proh
Again, in (53a), this looks like morphology; in (53c), like syntax. In the inbetween case (53b), there is no way to tell.13 12
We have a few examples in the corpus where something to the left of the head is within the scope of CR. Most of them, for example green-tea-green tea and all-done-all done, are arguably compounds and may not offer a problem. If it should prove necessary, an alternative would be to enlarge the scope of CR to include left-hand adjuncts, though still excluding specifiers. 13 The constraints on the base of Kannada ER differ from those for CR in two respects. First, as seen in (53c), ER allows more than one non-functional morpheme to be reduplicated; compare ?Well, I didn’t open-the-door-open the door. Second, it permits parts of idioms as well as whole idioms to be reduplicated. The acceptable example (ii) would be paralleled in CR by the ungrammatical Hari kicked the bucket-bucket. (i)
Hari kannu much-id-a Hari eye close-past-3sg:m ‘Hari died’ (lit. ‘Hari closed his eyes’)
(ii)
Hari [kannu]-ginnu much-id-a ‘Hari died and did related things’
(iii)
Hari [kannu muchida] ginnu muchida ‘Hari died and did related things’
364
contrastive focus reduplication
We now return to the issue of the ‘extra’ reduplicated material: the inflectional affixes, the unstressed pronouns, and the PPs with unstressed pronouns. As mentioned at the outset of this subsection, all of it seems to belong to the class of ‘grammatical’ or ‘functional’ morphemes. To be sure, the boundaries of this class are somewhat uncertain. However, we find that this hazy boundary is reflected in CR: as we shade from clearly ‘grammatical’ morphemes to more ‘contentful’ alternatives, our judgments of CR get worse (or more of us find the examples dubious). (54) illustrates with pronouns: see it is fine, see one may be worse, see some is definitely worse. (54) a.
An alien, huh? Did you actually see-it-see it, or just sort of guess it was there? b. ?Aliens, huh? Did you actually see-one-see one, or just sort of guess one was there? c. ? Aliens, huh? Did you actually see-some-see some, or just sort of guess they were there?
The examples in (55)–(56) illustrate the same thing with more- versus lesscontentful prepositions. The prepositions in (55) are more or less ‘grammatical’, that is, they are more or less default prepositions for the verb they occur with. Those in (56), however, add information that would not be predicted by the verb. We find them worse in CR; note especially the minimal pairs (55a–56a) and (55b–56b). (55) a. Did you sit-on-it-sit on it, like, squash it? b. Did you look-at-it-look at it, or just sort of glance at it? c. Did you talk-about-it-talk about it, or just mention it? (56) a. ?Did you sit-near-it-sit near it, or just sort of hang around it? b. ?Did you look-in-it-look in it, or just stick your hand in? c. ?Did you run-around-it-run around it, or just sort of jog? d. Does Superman’s X-ray vision really work, like, can he seethrough-her-see through her? Our conclusion is that in addition to a single contentful lexical item, the scope of CR may include only noncontrastive ‘functional/grammatical morphemes’ (however the latter is to be defined). Putting together all the conditions on CR so far, we get the following definition: (57) a. The scope of CR is either X0 or XPmin. b. The scope of CR must include a full lexical item, to whose meaning the semantic effect of CR is applied.
c o nt ras t iv e f o c us re d uplication
365
c. In addition to a single contentful lexical item, the scope of CR may include only noncontrastive ‘functional/grammatical morphemes’. We summarize the results of this section by showing how the scope of CR is defined in a variety of cases we have discussed. In each case, the scope of CR is marked by underlining, the contentful lexical item is marked in bold, and the leftover functional/grammatical morphemes are marked in italics. (58)
a. [ [N glove]s] ¼) glove-gloves b. [V [talk]ed] ¼) talked-talked c. [PPmin off the wall] ¼) off-the-wall-off the wall d. [VPmin talk about it] ¼) talk-about-it-talk about it e. [APmin proud of her] ¼) proud-of-her-proud of her f. [VP [VPmin give it] to him] ¼) give-it-give it to him14 g. [VPmin take her to task] ¼) take-her-to-task-take her to task (cf. (50a) ) h. [N [N work] [day]] ¼) work-workday (‘a day where I do real work’, cf. (40b) )
Condition (57c) is violated if contentful material is substituted for the italicized material in (58), e.g. talk-about-math-talk about math, proud-of-daveproud of Dave. Condition (57b) is violated if the scope is less than a contentful lexical item, e.g. one element of a lexicalized compound (boy-boyfriend), one word of an idiom (kicked-kicked the bucket), or the stem of a word containing derivational affixes (relation-relationship). Condition (57a) is violated in a-linguist-a linguist (cf. (47) ): although the determiner is a grammatical morpheme, it is outside of NPmin, so it cannot be within the scope of CR. Similarly, as mentioned earlier, sentential idioms such as my goose is cooked are impossible because they involve more than VPmin.15, e
14
The bracketed constituent over which CR has scope in this example is based on an analysis under which double complements in VP have a recursive left-embedded structure, a precedented but not uncontroversial assumption. We will not argue the point here. Recall also that speakers’ judgments differ on this example. 15 In section 11.6 we briefly discuss the similarity between CR and modification by words like really (for verbs and adjectives) and real (for nouns). If we view CR as being of the same class as these modifiers, some of the restrictions on its scope fall into line. For instance, the restriction e This configuration is harder to specify in the framework of Simpler Syntax, where there is less articulated syntactic structure. I think the scope can instead be stated in terms of linear order, by substituting (i) for (57a):
(i)
The scope of CR is X0, plus, optionally, sister constituents to its right. It looks as though conditions (57b, c) take care of everything else.
366
contrastive focus reduplication
11.3.5 Prosodic constraints on CR Although on our story the scope of CR is defined primarily in morphosyntactic terms, there do seem to be some prosodically based preferences which have a gradient effect on the acceptability of CR. These vary among speakers. First, there is a group of speakers for whom the copying of inflectional affixes is completely optional while the copying of object pronouns is strongly preferred. Such speakers judge (59a) and (59b) to be equally acceptable, but prefer (60a) to (60b): (59) a. guys-guys b. guy-guys (60) a. like ’em-like ’em b. ?like-like ’em While this may seem puzzling if taken as a fact about inflectional morphemes vs. object pronouns, it is less so if considered as a fact about prosody. Indeed, when an inflectional affix constitutes a separate syllable, these same speakers prefer to copy it. Hence, in (61)–(62), they make a distinction in the (a) examples, where the affix is syllabic, but not in the (b) examples, where it is not (‘>’ signifies ‘more acceptable than’; ‘¼’ signifies ‘equally acceptable’). (61) a. peaches-peaches > ?peach-peaches b. apples-apples ¼ apple-apples (62) a. voted-voted > ?vote-voted b. played-played ¼ play-played ruling out a-linguist-a linguist may be the same one that dictates the order a real linguist rather than real a linguist. And the restriction that rules out My goose is cooked-my goose is cooked is the same one that rules out Really my goose is cooked (with the relevant reading of really). More specifically, within a framework in which the functional categories Det and I are heads of phrasal categories of which NP and VP respectively are complements, CR falls in with the class of modifiers that can only adjoin to NP and VP—the lexical rather than functional categories. Assuming that this restriction can be stated for the whole class of modifiers, the constructionspecific conditions on CR are then the following: (i) CR can adjoin below the X0 level, provided its sister is a contentful lexical item. (ii) In addition to a single contentful lexical item, the scope of CR may include only noncontrastive functional/grammatical categories. (iii) The adjunction structure of CR cannot be dominated by any node licensed by a lexical entry (or a node that is coindexed with a constituent in the semantics).
The first two conditions account for the facts regarding inflectional morphemes and object pronouns. The third condition rules out the application of CR to parts of compounds or idioms. That is, unlike really, CR cannot apply within a sentential idiom (My goose is really cooked vs. My goose is cooked-cooked). Since this alternative view of CR is more difficult to work out with the parallel architecture model, we will pursue the version given in the main text.
c o nt ras t iv e f o c us re d uplication
367
Moreover, if the affix always constitutes a syllable, such speakers strongly prefer the version of CR in which it is copied along with the stem. (63) illustrates with the -ing forms of verbs and -est forms of adjectives. (63)
a. I was talking to him that week, but I wasn’t talking-talking [> talktalking].16 b. I’ve been reading Finnegan’s Wake. My friends told me it’s the hardest book in the English language. Not the hardest-hardest [> hardhardest].
This suggests that the relevant constraint is to prefer an application of CR which results in parallel prosodic structure in the reduplicant and the prosodic constituent containing the base. Two further phonological preferences we have found among speakers pertain to length and prosodic contour. Many speakers report degraded judgments correlating with increased length, as shown in (64a). A few speakers (RJ among them) prefer initial main stress over late main stress preceded by secondary stress (i.e. trochaic rhythm is preferred over iambic). The distinction in (64a) is a matter of prosodic contour as well as length, but (64b, c) are minimal pairs differing in prosodic contour alone.17 (64)
a. Be´acon Stre`et-Beacon Street > Co`mmonwealth a´venue-Commonwealth Avenue b. ba´sset ho`rn-basset horn > e`nglish ho´rn-english horn c. cla´rine`t-clarinet > cla`rine´t-clarinet
These prosodic constraints on CR are summarized informally as (65). (65)
a. The reduplicant preferably contains the same number of syllables as the prosodic constituent containing the base of CR. [some speakers] b. The scope of CR should not be too long. [many speakers] c. The scope of CR preferably has trochaic prosody. [a few speakers]
It is worth mentioning two factors that may contribute to this variation among speakers. First, CR most often occurs in either questions or sentences
16
Not everyone dislikes examples involving the omission of -ing, suggesting that the constraint on syllabic parallelism may be stronger or weaker for some. For those of us who find an example like talk-talking acceptable, we do not accept it in a context where the suffix -ing is derivational rather than inflectional. For example, in reference to someone who is a professional dancer one could say When I said his dancing drove me nuts I didn’t mean his dancing-dancing/dancedancing. 17 Both pronunciations of clarinet in (64c) are available in RJ’s speech, the initial-stressed variant being ‘clarinetplayerspeak’.
368
contrastive focus reduplication
containing negation, i.e. contexts in which main verbs will not be inflected. The rarity of fully inflected candidates for CR may explain why there is variation with respect to whether inflectional material gets copied. A second factor that muddies the water has to do with the fact that CR marks contrastive focus. As such, there is a preference for reduplicated expressions to occur alone, sentence-finally.18 To illustrate, here are some examples from our corpus in which material that is presupposed (shown here struck out) has been elided: (66) a. Felix: Tim! I’d be careful. That’s instant glue you’re using. Tim: It’s not like instant-instant [glue]. Ó b. That’s ok, I’m familiar with these young ladies. Well, I’m not familiarfamiliar [with these young ladies], I know them. Ó c. In fact I barely talked to him. Not talk-talked [to him]. Ó Because contrastively focused words tend to occur finally, it is difficult to determine the scope of CR beyond the ‘word’ (X0)—not only for the linguist but presumably for speakers/learners also. Hence, it is possible that speakers will vary as to the kinds of constraints they adopt for the construction. These constraints can be phonological (e.g. length restrictions), syntactic (e.g. whether the scope of CR is restricted to X0 or beyond), semantic/pragmatic (e.g. whether non-focus-bearing elements can copy), or some combination of these, UG being silent on the matter.
11.4 An analysis of CR in the parallel architecture framework We seek an analysis of CR in which it is possible to incorporate its meaning, its syntactic conditions, its reduplicative phonology, and, for those speakers who have them, its prosodic conditions. The syntactic conditions must make it possible for CR to apply to phrases as large as XPmin and as small as stems minus their inflectional affixes; thus we seek a framework where morphosyntax and phrasal syntax are not distinguished by being in separate impenetrable modules. One such framework is the parallel architecture of Jackendoff (1997b, 2002); we will work out an analysis of CR as a test of this framework on an unusual phenomenon. In order to present the analysis, we first review
18 The association between sentence-final position and new information/rheme/focus has long been noted in the literature. Recent analyses offering a syntactic account for this correlation include Vallduví (1995) for Catalan, Zubizarreta (1998) for Spanish, and Alboiu (2000) for Romanian.
c o nt ras t iv e f o c us re d uplication
369
some basics of the parallel architecture and show how it might be applied to reduplication in general; we then are in a position to state CR.
11.4.1 Basics of the parallel architecturef The fundamental postulate of the parallel architecture is that phonological, syntactic, and semantic/conceptual structures are each the product of a set of combinatorial primitives and principles of combination. The role of syntax is thus not as prime generative engine but rather as an intermediate structure that aids in more precise mapping between phonology and meaning. In order for the three independent structures to form a unified representation for linguistic expressions, they are linked by interface rules. A sentence is well-formed when its phonological, syntactic, and semantic/conceptual structures are all well-formed and a well-formed correspondence among them has been established by the interfaces. (67) illustrates the structure of the phrase the cats; the subscripting indicates the connections established by the interfaces between the parts of the three structures. Notice that the lowest nodes in the syntactic tree are syntactic features, not the customary notation the cat-s. This is because, in the parallel architecture, syntax contains not full lexical items, but rather only their syntactic features. The phonological and semantic features are in their respective trees. The semantics uses the Conceptual Structure notation of Jackendoff (1983, 1997b, 2002; see also Chapter 1 above); readers invested in other frameworks should feel free to substitute their own notations.
(67) Phonology
Syntax
Conceptual structure
NP1
[PLUR4 ([CAT]3); DEF2]1
Wd1 Cl2
ð
Wd
Det2
Wd3
Af4
kæt
s
N
[N, 3 sg]3 [Af, plur]4
In the parallel architecture, as in every theory, a lexical item is an association of phonological, syntactic, and semantic features (or structures). However, it is
f The original text of section 11.4.1 has been abridged here, given that the basics of the parallel architecture appear in Chapter 1.
370
contrastive focus reduplication
not inserted into a syntactic tree. Rather, a lexical item tells the grammar that its three sets of features can be linked in the three independent linguistic components. That is, a lexical item is a small-scale interface rule that plays an active role in licensing the construction of sentences. Unlike in mainstream generative grammar, idioms may be XPmin or larger, and productive affixes may be treated as lexical items syntactically smaller than X0 (Jackendoff 2002, chapter 6). (68) shows the three lexical items from which (67) is built. The only connection in (67) that does not come from the lexical items is the subscript 1 on the entire constituent; this comes from principles of phrasal correspondence.
(68)
Phonology a. Cli | ð
Syntax Deti
Conceptual structure DEFi
CATj
b.
Wdj | kæt
[N, 3sg]j
c.
Wd
N0
Wdk
Afm
[PLURm([Xk])]
N0k [Af, plur]m
z A crucial piece for our purposes here is the regular plural suffix (68c). The parts in italics are contextual features: they show how the affix attaches to a stem in phonology and syntax, how it applies as an operator to its stem in semantics, and how the stem corresponds in the three components. Similar contextual features can also be used to specify the connection between syntactic subcategorization of verbs and semantic argument structure (see section 1.5). More generally, the formal devices for encoding morphological composition are altogether parallel to those for phrasal composition. At the same time, this does not preclude there being differences between morphosyntax and phrasal syntax. Such differences will be reflected in the grammar (and quite possibly UG) by different sorts of compositional principles for units smaller than X0 (morphosyntax) than for units of X0 and larger (phrasal syntax). The point is that it is not necessary to say that (all) morphology ‘takes place in the lexicon’ (or in some other component altogether
c o nt ras t iv e f o c us re d uplication
371
distinct from the syntax), as is the practice in HPSG, for example. Rather, regular morphology is as much a matter of free (potentially online) composition as are phrases. (For the important distinction between regular and ‘semiregular’ morphology, see Jackendoff 2002, chapter 6; see also section 1.7 and Chapter 2.) The matching among phonology, syntax, and semantics in (67) and (68) is straightforward. For a slightly less canonical case, (69) shows the idiom take NP to task.
(69)
Phonology Wdi
Wd Clj
Wdk
teyk tuw
tæsk
Vi
Syntax
Conceptual structure
VPn
[CRITICIZE (Xe, Ym)] PP
NPm Pj
NP Nk
First look at the syntax–semantics correspondence. The verb, preposition, and noun are not coindexed with anything in the conceptual structure; rather, only the entire VP is coindexed with a Conceptual structure, namely ‘x criticize y’. This is precisely what it means for something to be an idiom: the syntactic parts do not contribute to the meaning of the whole. Next look at the two arguments of CRITICIZE, which bear subscripts. The first argument has the subscript e, which by convention picks out the external argument; the second argument is subscripted to the direct object in the idiom’s syntax, which is to be filled in by an argument. Finally, look at the syntax–phonology correspondence. The verb, preposition, and noun are coindexed with phonological content, as expected. However, the variable for the direct object is not coindexed with phonological content. Such a coindexation is unnecessary, because when an actual direct object fills in the variable, it will carry its own index to phonological structure. In turn, since the direct object falls between the verb and the PP in syntax, its phonology will be have to fall between that of the verb and that of the PP, because of the general interface principle that linear order corresponds in syntax and phonology.g
g In Simpler Syntax, the index for the external argument does not have to be stipulated, as it follows from more general principles. The same is probably true for the index for the direct object.
372
contrastive focus reduplication
This brief exposition puts into place most of the pieces we need to approach CR in the parallel architecture. The one remaining piece is a treatment of reduplication.
11.4.2 A treatment of reduplication in the parallel architecture A standard treatment of reduplicative morphology (Marantz 1982, McCarthy and Prince 1986) sees it as the realization of an abstract affix whose phonological content is an empty (or partially empty) frame; its segmental content is copied from the base in accordance with various phonological constraints.19 The emphasis in phonological theory has been on the mechanics of the copying: the nature of the empty frame and the phonological constraints that result in the reduplicated form. What is often neglected is the role of the reduplicative construction in the grammar as a whole. But it is clear that reduplicative morphology is (often) associated with a meaning to be expressed; that is, a reduplicant is (often) the phonological realization of a morpheme with a meaning. As mentioned in section 11.2.2, reduplication can express a wide range of possible meanings crosslinguistically (and in some languages such as Tagalog there are many different reduplicative morphemes). Some of these meanings, for instance plural and past, correspond to traditional inflectional morphemes; others, such as forming verbs from nouns, fit in with derivational morphology. But the meanings are typical for affixal morphemes: the literature does not cite reduplicative phenomena that express things like ‘picture of X’ or ‘sit on X’ that are typically expressed by open-class items. From the point of view of morphosyntax and semantics, then, it makes sense to view reduplicative constructions as ordinary morphological affixes. The only difference is that instead of ordinary phonological content, they have empty phonological frames or some ‘metaphonological’ content COPY X, where X specifies the base to be reduplicated. In this larger context, the discussion of reduplication in phonological theory concerns what UG needs in order to realize this ‘metaphonological’ content properly across the languages of the world. English CR might therefore be seen as such a morpheme. We are encouraged in this approach by the fact that, like other reduplicative morphemes, CR is paralleled semantically by an ordinary morpheme, in this case the Japanese maprefix discussed in section 11.2.2. The fact that CR is realized phonologically as 19 This view of reduplication is not universally accepted. For example, Travis (1999, 2000) has been pursuing a syntactic approach to reduplication. Inkelas and Zoll (2000) argue that all cases of reduplication involve the copying of a bundle of morphosyntactic features, not the copying of phonological material—even those cases which are the classic examples of the need for phonological copying. However, they do not elaborate on how they think the morphosyntactic copying is accomplished.
c o nt r as t iv e f o c us re d uplication
373
reduplication will be taken to be a consequence of its having the ‘metaphonological’ content that, in the context of an appropriate base, triggers automatic reduplication processes in phonology. There turns out to be a straightforward formal way to express COPY X in the parallel architecture. A reduplicative morpheme, like any other affix, has to specify the base to which it attaches. The base will have an index that connects its phonological content to its syntactic content. In order to create reduplication, we can attach this very same index to the phonology of the affix. (70) is a hypothetical reduplicative plural prefix. It is virtually the same as the English plural (68c), except for its phonological content.
(70)
Phonology
Syntax
Conceptual structure
Wd
N0
[PLURj (Xk)]
Afj
Wdk
Wdj,k
N0k
The important difference here is that the prefix has no specified phonological content. The phonological content is still coindexed j with the syntactic prefix, so we know the reduplicant is a realization of the prefix. But it is also coindexed k with the base; hence the base’s phonological content shows up in the prefix as well. If (70) were substituted for the English plural (68c) in the expression the cats, the result would be (71), the cat-cat.
(71)
Phonology
Syntax NP1
Wd1 Cl2
Det2
Wd Wd3,4 Wd4
ð
Conceptual structure
kæt
[Af, plur]3
[PLUR4 ([CAT3]); DEF2]1 N [N, 3sg]4
kæt
11.4.3 Formulating CR We now have most of the tools we need to formulate CR as a free reduplicative morpheme in the parallel architecture. There are still a few problems to be solved, though, some notational and some substantive. One thing that is different from the cases of reduplication so far is that the base is of variable size, in both syntax and phonology. The syntactic structure of CR will have to stipulate
374
contrastive focus reduplication
that either X0 or XPmin is possible (condition (57a) of section 11.3.4). In turn, the size of the phonological structure will follow automatically from whatever the syntactic scope happens to be. So we need not specify the phonological constituency at all; we will notate the reduplicant, the base, and their combination in phonology with the neutral label P, which may encompass anything from a Word to a Phonological Phrase. (It is not yet clear whether the reduplicant and the base together always form a larger phonological constituent and, if so, whether the size of that constituent is consistently related to the size of its daughters. In what follows, we will not try to diagram higher levels of phonological constituency.) A further terminological issue concerns what syntactic category to assign to CR. When it applies inside a word, as in glove-gloves, it ought to be an affix; when it applies to an XPmin, as in over-the-hill-over the hill, it has the aura of a modifier of some sort. We’ll provisionally call its category ‘CRsyn’, this term standing for the set of CR’s purely syntactic features.20 Given these terminological choices, the phonological and syntactic parts of CR can be stated as (72). For those speakers who have prosodic conditions or preferences on CR, these can be added; we will not work out the details. (Recall that italics denote contextual variables.)
(72)
Phonology
Syntax
P
X0/XPmin
Pj,k
Pk
CRsynj
X0/XPmink
Turning to the semantics of CR, recall that it has two components. First, CR delimits the denotation of its base, restricting it to the most prototypical, most extreme, or most contextually salient case or range of cases. Second, CR contrasts this sense with less prototypical, less extreme, or less contextually salient cases. There are no standard notations for either of these components; we will adopt the usual strategy of simply notating the relevant notions in capitals, pending further research on how they are decomposed into more basic feature systems (Jackendoff 2002, section 11.2). The crucial thing in CR, though, is that contrast is a modifier of prototypical/extreme/salient (henceforth p/e/s) [alternatively, see subfootnote c in section 11.2.1]. In turn,
20
One possibility for such features is suggested by only and even, which occur before VP, NP, AP, PP. Like CR, they are related to focus. Interestingly, also like CR, there is no standard account of their syntactic category. Thus CR might be syntactically whatever only and even are, except that it adjoins to X0 and XPmin instead of to XP.
c o nt r as t iv e f o c us re d uplication
375
p/e/s is a modifier of the base—which is why it is so often paraphrasable by a modifier such as real(ly). Some examples from section 11.2 provide a useful illustration; we append here the conceptual structure of the focused element (recall that the material after a semicolon is a modifier). (73)
a. It wasn’t a GOAT. [Contrastive focus] Contrast set: {goats, horses, pigs, sheep . . . } [GOAT; [CONTRAST] ] b. It wasn’t a goat-goat. [CR] Contrast set: {prototypical goats, non-prototypical or non-literal goats} [GOAT; [P/E/S; [CONTRAST] ] ]
Given this representation, we can formulate a version of CR containing semantics as follows:
(74) Phonology
X0/XPmin
P Pj,k
Syntax
Pk
CRsynj
Conceptual Structure [Zk; [P/E/S; [CONTRAST]]j]
X0/XPmink
If the meaning of the base does not offer an appropriate contrast, for instance if it is a grammatical morpheme or a piece of an idiom that lacks independent meaning, P/E/S will be a semantically anomalous modifier, exactly parallel to green ideas or contrastive kick the bucket with idiomatic meaning. Thus this formulation automatically accounts for condition (57b) of section 11.3.4: the scope of CR must contain a full lexical item. (74) contains nothing that specifies the special stress pattern of CR, because this is taken care of automatically. Recall that there is a component of interface rules that correlates information structure (topic/focus) directly with stress and prosody. These rules are responsible for the contrastive stress and intonation in (73a) and in sentences like It’s not a real dog, where the modifier is contrastive. Thus these rules can apply also to the contrast feature marked by CR. The contrasting constituent in semantics is subscripted j; therefore the stress will go on the constituent subscripted j in phonology, i.e. the reduplicant. We thereby see the virtue of the double subscripting in reduplication: the segmental content comes from the base, and the stress comes from the meaning of the reduplicative affix. One important piece is missing: we still have not accounted for condition (57c), which requires no more than one ‘contentful’ lexical item. (74) correctly applies to single words, and to combinations such as guys-guys (with an appended plural) and see-him-see him (with an appended pronoun). It also applies correctly to XPmin idioms such as over-the-hill-over the hill, and to idioms with interpolated pronoun objects such as take-her-to-task-take her
376
contrastive focus reduplication
to task. However, as it stands, it also applies, incorrectly, to combinations of multiple contentful lexical items, such as see-Dave-see-Dave. The brute-force way to fix this would be to add the following stipulation to (74): (75) Addendum to (74): The base material, subscripted k, must contain exactly one contrastable (or non-functional) lexical item. This stipulation simply rephrases condition (57c). It might conceivably be shown to follow somehow from the semantics of p/e/s þ contrast; perhaps it will turn out that an appropriate well-formed contrast cannot be constructed if more than one lexical item is competing for it. Should such a semantic/pragmatic solution not prove feasible, (75) will remain stipulative. It will then be necessary to add to the theory of UG the possibility that lexical items can countenance such ‘lexicality’ conditions within their contextual features. To make this sort of move plausible, other instances of such conditions will have to be sought. We leave the ultimate resolution of the stipulation in (75) as a matter for future research.21 Here are structures that show what CR looks like adjoined to various bases. (76a) is glove-gloves, where CR is adjoined inside the word; (76b) is criedcried, where CR is adjoined to the word including its affix. (76c) is feet-feet, the reduplication of an irregular plural. (76)
Phonology
Syntax
Wd
N
a.
Conceptual structure [PLUR3[GLOVE2;
Wd
Af3
Wd2,4 Wd2 gl v
N CRsyn
4
Wd1,4
N2
V CRsyn4
Wd1 Wd2 Af3
krayd kray c.
[P/E/S; [CONTRAST]]4]]
gl v z Wd
b.
[Af, plur]3
V1
[[PAST3([CRY]2)]1; [P/E/S; [CONTRAST]]4]
V2 [Af, past]3
d
Wd
N
Wd1,4
Wd1
fiyt
fiyt
CRsyn4
[[PLUR3([FOOT]2)]1; N1
N2
[P/E/S; [CONTRAST]]4]
[Af, plur]3
21 Note that Kannada Echo Reduplication (section 11.3.4) does not share this condition, in that ‘door-acc close’ can be reduplicated (53c).
c o nt ras t iv e f o c us re d uplication
377
Notice that in the parallel architecture, feet is sort of a phonological counterpart of an idiom: it has the syntactic structure of a normal plural, but its constituents in syntax are not coindexed with anything in phonology. Rather the inflected noun is coindexed with the phonology as a whole. Now, since CR must apply to a base that appears in both syntax and phonology, it cannot apply to the inner N in (76c) to produce foot-feet. Thus the formulation in (74) automatically explains how CR applies to irregularly inflected forms. (76) shows X0-size cases of CR. (77) shows an XPmin-size case: takeher-to-task-take her to task. It builds on the structure of the idiom given above in (69).
PhonPhr
(77)
PhonPhr1,4 Wd
Wd
VP7 PhonPhr1 Wd
Wd
Wd Cl Cl Wd Wd2 Cl3 Cl5 Wd6 | | | | | | | | teyk hr tuw tæsk teyk hr tuw tæsk
CRsyn
4
VP1
[[CRITICIZE(Xe, HER3)]7; [P/E/S; [CONTRAST]]4]
V2 [NP, pro 3sg fem]3
PP P5
NP | N6
Her in (77) is a noncontrasting functional item. If a phrase with lexical content is substituted for it, for instance take Susan to task, condition (75) is violated and therefore CR cannot apply. Similar considerations apply to rule out all nonfunctional complement phrases. We should also note why CR cannot apply to the inside of the idiom, say take her to task-task. This is because CR’s sister in syntax must be coindexed with a constituent in semantics. As observed in section 11.4.1, what makes an idiom idiomatic is precisely that its internal syntactic constituents are not coindexed to semantic constituents. Hence CR cannot apply in a well-formed fashion. Our conclusion is that an approach to CR in the parallel architecture is altogether feasible. Its parallelisms with other forms of affixation and with other reduplicative phenomena emerge quite naturally. In addition, the ways in which CR diverges from other known phenomena are clear and formally expressible in a natural fashion: in particular, its ability to adjoin to either X0 or XPmin is easily statable, because the framework does not require a strict division of modules between morphology and syntax. The contrastive stress with CR appears as an automatic consequence of its semantics. The fact that irregular inflected forms must reduplicate in toto is a consequence of the way such forms
378
contrastive focus reduplication
are lexically encoded. The only serious difficulty we see is the stipulation in (75), which as far as we can see is shared by accounts in any other framework (we leave it to the reader to check this out in his or her favorite).
11.5 A Minimalist Program approach to CR For those of different theoretical tastes, we now sketch a preliminary account of CR based on the Minimalist Program. The basic idea is that the reduplication in CR results from the spelling out of both the head and the tail of a chain created by head movement. Perhaps the prime virtue of a Minimalist analysis is that almost all the tools required are already available. Head movement—movement of an X0 to a c-commanding Y0 to check features—is a core part of the Minimalist Program. Focus features in general, and the feature [þcontrastive focus] in particular, have been employed to trigger movement (though not necessarily head movement) and/or to require checking (see, for example, Brody 1995, Rizzi 1997, Erteschik-Shir 1997, Kiss 1998, Zubizarreta 1998). Finally, the idea that traces are copies of a moved element (Chomsky 1995) has been adopted by many, if not most, Minimalist syntacticians. Our point in this section, then, is that a copy theory of movement buys us a movement theory of copying. The analysis goes as follows. The CR morpheme consists of the features [P/E/S, þcontrast], but no phonology. Akin to a modifier of category A (adjective or adverb), it heads a CR phrase that can take any lexical phrase (NP, VP, AP, etc.) as its complement (see n. 16). Assuming that the features of the CR morpheme are strong, they must be associated with a syntactic head that is lexically filled. This triggers head-movement of the adjacent X0.
(78)
CRP XP CR0 | | [____][+P/E/S, contrast] X0
Prior to the (re)introduction of the copy theory of movement, it was always the head of a chain created by overt syntactic movement that was taken to be pronounced. The idea that lower copies may be relevant for interpretation at LF (e.g., for ‘reconstruction’) opened up the possibility that the interface levels (PF and LF) may be choosy about which copy they ‘privilege’. Thus, Richards (1997), for example, has recast overt and covert movement (a distinction
c o nt ras t iv e f o c us re d uplication
379
attributable to strength of features) in terms of which copy of a chain gets pronounced: (79)
a. PF must receive unambiguous instructions about which part of a chain to pronounce. b. A strong feature instructs PF to pronounce the copy in a chain with which it is in a feature-checking relation. [Richards 1997, 122.3–4]
More recently, Bobaljik (2002) has argued that all four possible ‘privileging’ options (head vs. tail; PF vs. LF) are attested. So far the idea that more than one copy may be pronounced has not been explored, though it has not been explicitly ruled out either. Richards (1997) assumes ‘for the sake of simplicity’ (n. 1, p. 122) that only a single element in a chain will be pronounced, but acknowledges that it could be otherwise. Indeed, a copy theory of movement should be more credible if cases where more than one copy is pronounced are found. The question, in the case of CR, is why this should be so. Let us consider the possibilities. Given head movement of an X0 to the head of the CR phrase there are three options: (80a) that the tail alone will be pronounced; (80b) that the head alone will be pronounced; or (80c) that both the head and the tail will be pronounced. (80)
a. [ [X0][ þ P/E/S contrast] [X0] ] b. [ [X0][ þ P/E/S contrast] [X0] ] c. [ [X0][ þ P/E/S contrast] [X0] ]
The first option is ruled out if we assume, following Richards, that strong features serve as instructions to pronounce the item they check or are checked by. The second option can be ruled out for functional reasons, specifically, (80b) is indistinguishable from a situation in which an unreduplicated X0 is contrastively focused. (That is, It wasn’t red-red would sound exactly the same as It wasn’t red, which we have already seen has a different meaning). Having ruled out the possibility of pronouncing only the head or tail of the chain resulting from movement to check the [ þ contrast] feature, we are left with the option of pronouncing both. In other words, relatively uncontroversial Minimalist assumptions provide a derivational explanation for the copying that characterizes CR.22 Apart from the fact that this analysis makes use of independently motivated features of the Minimalist Program, it also accounts for some of the features of 22
In previous versions of this work we have entertained the idea that the CR ‘slot’ does not project to a phrase and is simply adjoined to the head being modified. This draws on the proposal made in Travis (1988) that adverbs and prenominal adjectives may be base-generated
380
contrastive focus reduplication
CR in a natural way. It also invites parallels with other, non-local, copying phenomena, which may be amenable to a similar sort of analysis. One such example is offered by Koopman (1984): in Vata, a Kru language spoken in the Ivory Coast, a focused verb appears at the front of the sentence, with a copy remaining in the base position.23 (81) a. pA$ n2 ka’ me´ pA’ a$ throw you fut-a it throw Q ‘Are you going to throw it?’ [Koopman 1984, 155 (4a)] 1 mU2 b. zA$lK$ n’ zA$lK$ za redden I redden sauce ‘I really reddened the sauce. [1984, 157 (11g)] Another example is found in ASL, as discussed by Petronio and Lillo-Martin (1997). The authors discuss the ‘double construction’—a construction used for focus or emphasis in ASL—in which elements such as modals, quantifiers and verbs can be doubled sentence-finally, as shown in (82).24 (82) a.
cond
KNOW PROBLEM SITUATION, CANNOT J-U-R-Y CANNOT ‘ . . . If [you] are aware of the problem, the situation, then [you] cannot be on the jury’ [P & L-M 1997, 30.25] b. MY HIGH SCHOOL FIVE DEAF KID FIVE ‘My high school had [only] five deaf kids.’ [30.29] c. t HE HATE LIGHTS-FLASHING-ON HATE ‘He hates the lights flashing on and off.’ [30.31]
in head-adjunction structures. In fact, in her recent work on reduplication, Travis (2001) explicitly puts forth this view of CR (though she does not adopt a head-movement analysis). However, as an anonymous reviewer points out, if the modifier slot is head-adjoined, then head movement will result in a situation in which the antecedent and its trace mutually c-command each other. The reviewer goes on to make the intriguing suggestion that this may be why both copies must be spelled out. While this looks like a promising alternative to pursue, we leave it aside for now. 23 Koopman shows that many of the properties that hold of wh-constructions in Vata also hold of this construction; hence she argues that both involve movement to CP. As to why ‘Focus-Vmovement’ (her name for the construction), unlike wh-movement, must leave a copy, Koopman argues that the copy is a ‘resumptive verb’ that appears to satisfy the ECP. 24 Petronio and Lillo-Martin use standard conventions for transcribing ASL: An uppercase English word is used as a gloss to represent an ASL sign. Capital letters with dashes between them indicate a fingerspelled word. . A line above the signs represents the cooccurrence of a nonmanual marker, and the symbol at the end of the line indicates the types of marker: cond (conditional) ¼ brow raise, side tilt on the antecedent clause; hn (assertion) ¼ head nods; t (topic) ¼ brow raise, upward head tilt. . .
c o nt ras t iv e f o c us re d uplication
381
Petronio and Lillo-Martin (1997), based on Petronio (1993), propose that the final double is base-generated in the head of a [þF] (for focus) CP. It is unclear whether a plausible movement analysis would ultimately be possible instead.25 Nevertheless, the phenomenon resembles CR and Vata ‘FocusV-movement’ in that in all three cases there is a connection between copying and focus. For a third case, Rosen (2003) proposes that the surface position of demonstratives in Michif can be derived through different pronunciation instructions at PF. Demonstratives in Michif may be pre- or post-nominal, as in (83a, b). But in Cree (the source language for Michif demonstratives, Michif being a FrenchCree mixed language), the demonstratives are sometimes doubled (83c). Rosen suggests that all three possibilities are derived from a single syntactic structure and that in (83c) both copies are pronounced. (83)
a. awa la fij ‘that girl’ dem det girl b. la fij awa ‘that girl’ det girl dem c. awa nisıˆmis awa dem my younger sister dem ‘this little sister here of mine’ [Rosen 2003, 64]
It is a virtue of the Minimalist analysis, therefore, that it can link CR with other syntactic phenomena (e.g. cases of non-local copying connected with focus) rather than other phonological phenomena (e.g. partial reduplication)—a virtue given that CR cannot be purely phonologically defined. The Minimalist analysis of CR is relatively straightforward, relying on existing assumptions within the theory. Indeed, many of these current assumptions appear to be ideal for a phenomenon such as CR. On the other hand, this analysis cannot account for some of the core properties of CR. As we have seen in section 11.3, CR optionally copies inflectional morphology, something that is hard to explain with a head-movement account. Specifically, it is unclear how inflectional morphemes such as plural marking or tense marking can show up on the tail of a chain and not on the head if movement to check both inflection and CR is upwards. It is possible however, that 25 Matsuoka (1997) proposes an analysis of a similar construction of ASL, the so-called ‘verb sandwich’, where the verb is repeated (possibly non-locally) later in the clause, the second time with aspect marking. Unlike English CR and the doublings discussed by Petronio and LilloMartin, verb sandwiches seem not to be motivated by focus considerations. But the mechanics of Matsuoka’s analysis is nearly identical to ours: the verb undergoes head-movement to Asp0, and the copy left behind may optionally fail to be deleted.
382
contrastive focus reduplication
Minimalism augmented with a theory of post-syntactic morphology, such as Distributed Morphology (Halle and Marantz 1993), may be able to solve this puzzle. Another problem for a head-movement account is that it requires all strings that undergo CR to be X0s. For VPs and APs with object pronouns and PPs, this would require a well worked-out theory of clitic adjunction or the like, a task far beyond the scope of our analysis here. The fact that head movement moves X0s is also problematic for idioms (cf. Jackendoff 1997b, 2002): while it may just be possible to insert over the hill as a single X0, it stretches credulity to do the same for a syntactically transparent idiom like take X to task, which, as we have seen, has an open position and still can (in our judgments) undergo CR (I didn’t take-him-to-task-take him to task). In other words, although the Minimalist account deals nicely with the central cases where a single word is reduplicated, it does not so easily extend to the cases smaller and larger than a word. We leave an amplification of this approach for future research.
11.6 Final remarks We would like to reflect briefly on why the parallel architecture analysis may offer the virtues it does. One reason is that it allows us to integrate the phonological, syntactic, and semantic constraints of CR in a natural fashion. A second concerns the scope of CR. Because CR applies both within words and to phrasal-size constituents, it creates difficulties for any theory that makes a strict difference between morphology and phrasal syntax, say by putting morphology in the lexicon but leaving syntax as an independent combinatorial component. Notice that when CR applies inside a word, it looks like a morphological process, paralleling for instance the semantically similar Japanese affix ma- cited in section 11.2.2. On the other hand, when it applies to an XPmin, it looks more like a syntactic phenomenon: it can often be paraphrased by real(ly) and is often contrasted with other modifiers. This is shown in (84), where the reduplicants can be paraphrased by the expressions in square brackets. In cases where the contrast is made explicitly, we have underlined the counterpart to the reduplicant. (84) a. It is part of the highway-highway [real/concrete highway], not the information highway. Ó b. Yeah, but, you know, not nervous-nervous [really/very nervous]. Ó c. like-’em-like ’em [really/genuinely like ’em]? Or, I’d-like-to-get-storecredit-for-that-amount like ’em? Ó
c o nt ras t iv e f o c us re d uplication
383
d. So when you say you’ve considered it . . . you mean thought about it considered it or just considered-it-considered it [superficially considered it]. Ó CR especially looks like a syntactic modifier in (84c, d), in that the contrasting slot is filled by an entire phrase such as I’d-like-to-get-store-credit-for-thatamount in (84c) or thought-about-it in (84d).26 The overall point is that it is not appropriate to ask whether CR is morphological or syntactic: there is no fact of the matter. Rather, when applying to its smallest scope, X0 inside of a word, it has the feel of other things that attach there, i.e. morphological affixes; when applying to its largest scope, XPmin idioms, it has the feel of other things that attach there, i.e. syntactic modifiers. At the in-between scope of a single word, the stereotypical application of CR, the choice makes no difference. Notice that there is nothing in the semantics of CR that demands it be either morphological or syntactic. We commonly find doublets of morphological affixes and syntactic constructions with the same 26
However, before drawing any strong conclusions on the basis of these examples, one would have to think more carefully about the status of the ‘you-can-put-anything-you-want-before-thehead’ construction, already mentioned in n. 2. We have found brief mentions of this construction in a number of papers but no serious analysis. The shared insight among those who discuss this construction is that it involves a phrase that acts like a word. For instance, Carnie (2000) focuses on cases where a ‘sentence’ functions as a verb, even taking verbal affixes such as the past tense marker ed, as in: (i) He I-don’t-cared his way out of the room. (ii) She I’m-from-New-Yorked her way into the men’s room.
[Carnie 2000, 91.66]
Déchaine and Wiltschko (2002) address the issue of why examples such as the following are possible if there is a general ban against using 1st/2nd person pronouns in compounds: (iii)
a me-first attitude; a holier-than-thou attitude; an I-don’t-give-a-flying-fuck attitude
[Déchaine and Wiltschko (2002), n. 16] They suggest that such examples should be considered ‘phrasal compounds’ (cf. Di Sciullo and Williams 1987) whereby an XP, even one including functional categories, can serve as the input to compounding. Haiman (1991) notes the orthographic convention, adopted by many authors, of using hyphens to suggest that phrasal modifiers have the status of words. (iv) But now those Democrats can find easy cover in the weak-kneed it’s-just-not-politicallyfeasible argument. (David Corn, The Nation, 1989) (v) A: Show business is a dog-eat-dog world. B: No, it’s worse. It’s a dog-doesn’t-return-the-other-dog’s-phone-calls world (Woody Allen, Crimes and Misdemeanors) [Haiman 1991, 54]
The idea that phrases are acting like ‘words’ in the above examples seems to be different from the way in which phrasal idioms are word-like. The above modifiers are not all well-established idioms, and most instances of them in speech and writing have a novel, one-time-only, flavor. The sense in which they are word-like seems to be in terms of their syntax. In short, it seems to be possible for a zero-level modifier position to host a phrase (provided the right quotation pragmatics are present).
384
contrastive focus reduplication
meaning, for example rewrite and write over, or phrasal vs. morphological causative. Because the parallel architecture regards all free combination, whether below the word level or above, as accomplished online by the same process of unification, the issue of dividing CR between syntax and morphology need not arise. The issues that arise within this framework lie for the most part in finding the appropriate formal categories to state the generalizations. These issues include the precise characterization of the category XPmin that determines the largest possible scope of CR, the syntactic features of CR, and the precise characterization of the phonological category that is sufficiently general to encompass the various cases of CR from large to small. We have been very clear about hedging in the course of addressing many of these questions. The major substantive issue concerns how the ‘lexicality constraint’ (75) is stated so as to permit ‘grammatical’ morphemes to fall within in the scope of CR. We can for the moment find no non-stipulative way to deal with this problem, so we leave it as a puzzle for the field.h
h Whitton 2007a, 2007b (see subfootnote c) suggests that the lexicality constraint might follow from the fact that when there are two or more contentful lexical items within the scope of CR, it is not clear which of them should furnish the scale for the interpretation of CR. Her suggestion is consistent with the difference between English CR and the Kannada reduplication construction, which violates the lexicality constraint: the Kannada construction means ‘and things like that’, which does not involve such a range of possible pragmatically chosen scales. We do not know whether CR constructions in other languages extend beyond the word level.
chapter 12
Construction after Construction and its Theoretical Challenges (2008)
12.1 Basic facts1 The English NPN construction, exemplified by construction after construction, turns out to illustrate several themes that have been developing in recent theories of syntax and the syntax–semantics interface, and it presents numerous challenges to most extant approaches. The construction is a prime example of what Culicover (1999) calls a ‘syntactic nut’—an entrenched noncanonical structure. But it is hardly unusual in speech or writing (readers are challenged to count its occurrences in the text of the present chapter). It has been mentioned here and there in the literature (e.g. Beck and Stechow 2006, Huddleston and Pullum 2002, Ko¨nig and Moyse-Faurie 2007, Matsuyama 2004, Oehrle 1998, Pi 1995, Postma 1995, Williams 1994) but, as far as I know, the full range of data has not been explored.
1 [This chapter originally appeared in Language 84 (2008), 8–28, and is reprinted here by permission of the Linguistic Society of America.] The chapter was inspired by the brief discussion of this construction in Williams 1994. It was originally intended as a brief addendum to Goldberg and Jackendoff ’s (2004) study of the family of resultative constructions (here Chapter 9), but it rapidly outgrew that context. I am grateful to Adele Goldberg, Peter Culicover, Barbara Citko, Geoffrey Pullum, Paul Postal, Marcel den Dikken, Tetsuya Matsuyama, Ekkehard König, audiences at SUNY Stony Brook and University of Toronto, and the reviewers for Language for help in shaping it into its final form.
construction after construction
386
Constructions like this appear in a variety of languages, including Dutch (Postma 1995), Japanese (Matsuyama 2004), German and Mandarin (Ko¨nig and Moyse-Faurie 2007), Polish, Russian, Italian, and French, to cite just a random sampling.2 The present chapter focuses on the facts of English, which are detailed and idiosyncratic enough to suggest that none of these other languages is likely to be exactly parallel. The construction includes a number of one-off idioms, such as those in (1). (1)
cheek by jowl, hand over fist, head over heels, hand in glove, tongue in cheek, hand to mouth, tit for tat
Unlike these cases, most instances of NPN require the two nouns to be identical.3 The idiomatic cases in (2) observe this requirement. (2)
hand over hand, hand in hand, arm in arm, (tear NP) limb from limb
In the examples in (1) and (2), the choice of noun is relatively limited (e.g. tongue in tooth, finger over finger, leg from leg). However, the NPN construction is productive with a handful of prepositions; that is, the choice of noun is quite free. These prepositions are by, for, to, after, and upon (with the variant on). Examples appear in (3).4
(3)
a. b. c. d. e.
day by day, paragraph by paragraph, country by country dollar for dollar, student for student, point for point face to face, bumper to bumper term paper after term paper, picture after picture book upon book, argument upon argument
The construction is highly constrained. The nouns cannot be mass nouns (4a) (with certain exceptions to be seen later); they are not allowed to have determiners (4b) or plurals (4c) (again with exceptions). They are also not allowed to have postnominal modifiers (4d), except for one important case shown in (4e), which is discussed in section 12.4.5. However, some of the prepositions permit prenominal adjectives on either the second noun or both (4f). 2
Thanks to Barbara Citko and Vera Gribanova respectively for pointing out relevant Polish and Russian data. 3 Coinages with non-identical nouns are possible if they have clear intended analogues: I recall a tongue-in-cheek science fiction story in which a human and an alien, romantically involved, walk off at the end hand in pseudopod, the phrase evidently intended as a joke. The e.e. cummings poem ‘anyone lived in a pretty how town’ contains coinages such as when by now, tree by leaf, bird by snow, and if by yes; but of course, as evidenced by the title, the poem consists largely of ungrammatical sentences. 4 A further possibility suggested by a referee is X against Y, as in hawk against dove. However, this does not observe the constraints on NPN pointed out in (4), e.g. the clever French against the indomitable British is fine. I suspect therefore that this is a different construction.
construction after construction
387
(4) a. No mass nouns: water after water, dust for dust b. No determiners: the man for the man, a day after a day, some inch by some inch c. No plurals: men for men, books after books, weeks by weeks d. No postnominal complements or modifiers: father of a soldier for father of a soldier, day of rain to day of rain, inch of steel pipe by inch of steel pipe e. —except with after and upon: day after day of rain f. Prenominal adjectives: day after miserable day, tall boy by tall boy The positions in a clause in which the construction can occur are predominantly adjunct positions: presententially (5a), after the subject (5b), and in VP after the complements (5c). It can also appear within an NP, in the position normally occupied by a prenominal adjective (5d). N after N and N (up)on N furthermore can appear in positions characteristic of NPs (5e–h).5 (5) a. Page for page, this is the best-looking book I’ve ever bought. [presententially] b. John and Bill, arm in arm, strolled through the park. [after subject] c. We went through the garden inch by inch. [post-complement] d. Your day-to-day progress is astounding. [prenominally in NP] e. Student after/upon/by student flunked. [subject] f. We filled crack after/by crack. [object] g. We looked for dog after/ by dog. [object of PP] h. Student after/by student’s parents objected. [possessive NP]
12.2 The productive subconstructions of NPN Each of the productive prepositions has its own associated meanings in this construction, which we now go through one by one. Speakers vary in their judgments of some of these cases. For example, one referee disputes my judgment that century to century is unusual, finding it ordinary, while another referee finds the same example unacceptable. What is important to the present argument is not so much the exact facts of the NPN construction as its overall
5
Huddleston and Pullum 2002 (633 n. 15) notice this difference between N after N and the other cases of NPN. They claim that only temporal nouns are acceptable in N after N in adjunct positions: they cite day after day but quarrel after quarrel. The examples throughout this article show that this is not the case, and I find the following example perfectly acceptable: Quarrel after quarrel, those two somehow manage to remain friends.
388
construction after construction
texture—its interplay of regularity and idiomaticity, which is maintained despite varying details in speaker after speaker of English.
12.2.1 N by N The main sense of N by N is of some sort of succession: (6)
a. We examined the neighborhood, house by house. b. Inch by inch, row by row, got to make this garden grow. c. We went through the reptile house snake by snake. d. We examined the moon crater by crater.
[Children’s song]
N by N has a pronominal form, one by one. It is a bit unclear whether one is actually a pronoun or a numeral here, as we also have two by two, as in the animals came in two by two. On the other hand, I find larger numerals rather marginal, though there may well be interpersonal variation. N by N also has the special case side by side, which denotes juxtaposition of two entities, and which therefore fits into a semantic paradigm with face to face and back to back. We discuss this paradigm in section 12.2.3.1. Another special case is little by little, which has the characteristic succession sense but conjoins two quantifiers. Note that instead of the expected parallel more by more, we have more and more. A semantically unrelated use of N by N refers to measures in orthogonal dimensions, as in (7). Notice that the constituents connected by by need not be identical, and that not only nouns but dimensional adjectives can be connected. These differences suggest that this construction is perhaps only a distant cousin of the NPN family. (7)
a. a 200 by 400-pixel matrix b. 3 feet by four feet c. 3 feet long by 10 inches wide by ½ inch thick
12.2.2 N for N The next productive case is N for N. This has the sense of matching the members of two sets being compared or exchanged (8). (8)
a. Snake for snake, this is the best reptile house in the world. b. This article, line for line, has more mistakes than the previous version.
[comparison] [comparison]
construction after construction c. Gandia and Qumar exchanged prisoners man for man. d. The candidates matched each other insult for insult. e. a syllable-for-syllable replacement that improved comprehensibility
389 [exchange] [exchange] [exchange]
In this subconstruction, the two nouns must be identical, even in a situation where the comparison would pragmatically allow non-identity. Suppose the Wellesley debating team is all women and the Harvard debating team is all men. (9a) is pragmatically plausible, but the construction forces us to reduce all participants to a common category, as in (9b). (9) a. Woman for man, Wellesley’s got the better debating team. b. Member for member/Student for student, Wellesley’s got the better debating team. A special case of this subconstruction is word for word, which expresses an exact matching in repetition or translation of a text. (10)
a. I bet you can’t learn/recite Syntactic Structures word for word. b. a word-for-word translation
This seems to be idiomatic, at least in my speech. I use syllable by syllable and paragraph by paragraph in the corresponding cases.6 Another special case is found with numerals, as in a 3-for-2 swap of prisoners.
12.2.3 N to N 12.2.3.1 Close contact or juxtaposition N to N has a number of uses. One, mentioned in section 12.2.1, expresses close contact or juxtaposition of similar parts of similar objects, particularly body parts.
6 Again there are evidently individual differences. To my surprise, one referee finds (10a) unacceptable. Also, a Google search yields a few examples like these:
(i) (ii) (iii) (iv) (v)
I can recite their liturgy syllable for syllable to this day. . . . take an Ella Fitzgerald solo and learn it syllable for syllable They know the Qur’an word by word, syllable for syllable I do wonder exactly how this work would read if read backwards, paragraph for paragraph. All you can do is read out word for word, sentence for sentence and paragraph for paragraph someone else’s ideas.
In (iii) and (v) the usage is perhaps motivated by a desire to extend word for word (although, perversely, (iii) uses word by word ). My own strong preference would be to use by in these examples, in contrast with (8e), in which for is perfect.
construction after construction
390
(11) hand to hand, face to face, cheek to cheek, eye to eye, toe to toe, shoulder to shoulder, base to base, bumper to bumper This case can occur after be (12a), unlike any of the other cases of NPN (12b). It cannot, however, occur after seem (12c). And unlike other cases of N to N, it can occur as the object of a preposition (12d). (12) a. b. c. d.
Bill and Harry are face to face/shoulder to shoulder. Bill and Harry are boy by boy/student after student. ?Bill and Harry seem face to face/shoulder to shoulder. Go from bumper to bumper to face to face! [billboard on Boston’s Southeast Expressway, summer 2002]
Many instances of this subconstruction are a bit idiomatic. For instance, eye to eye is used primarily in the idiom see eye to eye, hand to hand in the context of combat, and cheek to cheek in the context of dancing. I find near neighbors of the nouns in (11) unusual, as in (13).7 (13) ?foot to foot (cf. toe to toe); ?finger to finger (cf. hand to hand); ?arm to arm (cf. shoulder to shoulder); ?front to front (cf. back to back); ?lip to lip (cf. cheek to cheek) I find that many pragmatically plausible cases that are not body parts are only semi-acceptable—they come with a sense of stretching the paradigm (14a, b). On the other hand, such cases are certainly not unattested: (14c) comes from the Boston Globe (16 April 2007). And there are also cases in which the nouns are not identical, such as (14d, e) (the latter suggested by a referee). (14) a. b. c. d. e. 7
?The cars sat in the parking lot, trunk to trunk. ?The houses are facing each other, porch to porch. . . . two cocktail glasses placed stem to stem back to front teapots lined up spout to handle
For what it is worth, a Google search for hand to hand, eye to eye, shoulder to shoulder, and back to back yielded citations in the hundreds of thousands, whereas front to front and lip to lip had about 5,000 each, and wrist to wrist had less than 1,000. In other words, the ratio of citations for idiomatic cases to those for ‘productive’ cases is two to three orders of magnitude. This situation parallels the family of denominal verbs such as butter (the bread), shelve (the books), pocket (the money), and so on. Speakers must learn which denominal verbs are conventionalized (e.g. butter is but mustard and mayonnaise are not). At the same time, speakers can ‘stretch’ to new cases they have not heard before, with appropriate contextual support. For instance, in the context of recipes, a Google search indeed found instances of the phrases mayonnaise the bread and mayo and mustard the bread. But their numbers were minuscule compared to butter the bread (one each versus about a thousand). This situation exactly parallels hand to hand vs. foot to foot. More systematic corpus work would be of interest here in assessing the status of such semiproductive cases.
construction after construction
391
Overall, then, this juxtaposition reading of N to N is not completely productive; it is more like a ‘stretchable’ collection of idioms. (A particular sort of juxtaposition is also expressed by idiomatic cases with in such as hand in hand, arm in arm, hand in glove, and tongue in cheek.) Side to side also has an idiosyncratic meaning that fits in a semantic paradigm with the family of idioms in (15): (15)
back and forth, up and down, to and fro, round and round
12.2.3.2 Succession Another reading of N to N involves iterated transitions or successions. This reading can optionally be prefaced with from in adjunct position (16a), though not prenominally (16b), and it is productive with nouns that denote time periods: (16)
a. (from) minute to minute, (from) day to day, (from) millisecond to millisecond, (from) year to year, (from) century to century b. our (from) day-to-day progress on the building
This reading is paraphrasable by from one N to the next, as in from one week to the next. A related use is the idiom case to case, denoting a succession of cases. This can also be used with nouns that denote what something is a case of, as in (17). (17)
a. b. c. d.
We have to make up our minds on a sentence-to-sentence basis. Adult coloration is highly variable (from) snake to snake. (From) situation to situation, conditions change. item-to-item differences/similarities/variation.
Again, there are paraphrases along the lines of From one situation to the next, conditions change. Further subcases of succession N to N appear in (18). (18)
a. search (from) house to house (¼ ‘successively from one house to the next’) b. travel (from) country to country (¼ ‘successively from one country to the next’) c. sell books (from) door to door (¼ ‘going successively from one door to the next’)
This reading of N to N is close in meaning to succession N by N. But an example-by-example comparison reveals semantic differences. For instance, I judge the examples in (19a–c) better with by than with to; those in (19d, e)
construction after construction
392
are better with to than with by, and in (19f–h) they seem equally good. The examples in (19i–k) show a very delicate semantic context-dependence that I have been unable to characterize. In general, N by N seems more productive than N to N. (19) a. b. c. d.
We went through the reptile house snake by/??to snake. We examined the moon crater by/to crater. We took out the garbage bag by/to bag. We traveled country to/??by country. [but: travel country by country through Asia] e. Adult coloration is highly variable snake to/by snake. f. We searched house to/by house throughout the neighborhood. g. His condition keeps improving day to/by day. h. We have to make up our minds on a sentence-to/by-sentence basis. i. We looked house by/to house at the unusual architecture in New Orleans. j. We looked house by/to house for traces of the criminals. k. We looked article by/to article for examples of the NPN construction.
12.2.3.3 Single transition Another case of N to N involves transitions in time or space. The two nouns are usually not identical. (20) a boy-to-man transition, a rags-to-riches story, the Boston-to-New York commute, cradle-to-grave health care, the syntax-to-semantics mapping, a one-to-many mapping, the Christmas-to-New Year’s break Normally, these cases of NPN are only possible prenominally. Most examples require from in adverbial positions (21a–e), though a few idiomatically allow from to be omitted (21f–h) or even require it to be omitted (21i). (21) a. b. c. d. e. f. g. h. i.
He changed (from) boy to man in less than six months. His life altered (from) rags to riches. The break lasted (from) Christmas to New Year’s. (From) Boston to New York, it took us over eight hours. (From) syntax to semantics, not all information is preserved. We lived (from) hand to mouth. We traveled (from) coast to coast. This plan protects you (from) cradle to grave. This function maps syntax to semantics (from) one to one. [but This function maps syntax to semantics (from) one to many.]
construction after construction
393
12.2.3.4 Comparison/juxtaposition with differing nouns A last subcase of N to N is shown in (22). It occurs only in prenominal position, and the to can be omitted. (22)
a boy-(to-)girl comparison, a student-(to-)professor matchup, the sulfur(to-)kryptonite ratio, finger-(to-)hand proportions, tongue-(to-)palate contact
Here the nouns that NPN modifies (or is an argument of) are all predicates that involve comparison or juxtaposition. Where counterparts with verbs exist, the complements do not have NPN structure. (23)
a. b. c. d.
We compared boy to girl. We compared (the) boys to (the) girls. We matched (the) student(s) up to/with (the) professor(s) The tongue contacted the palate. ?He contacted tongue to palate. the ratio of sulfur to kryptonite [no verbal counterpart]
Since this case is licensed by the semantics of the noun that it precedes, this subconstruction has the flavor of a compound, parallel to, say, student association (¼ ‘association of students’). Notice how different this case is from the other one that expresses comparison, N for N (section 12.2.2, especially (8a, b) ). There the construction is perfectly acceptable in adverbial positions as well as prenominally, and the two nouns must be identical, as shown in (9). This illustrates vividly the idiosyncrasy of the various subconstructions.
12.2.4 N after N and N Upon N The last productive cases of the general NPN construction are N after N and N upon N. N after N always denotes a succession of Ns, either temporal or spatial. (24)
a. Day after day, alone on a hill, the man with the foolish grin is keeping perfectly still. b. Telephone pole after telephone pole stretched along the road toward the horizon. c. That new series of books on Imperialist Grammar is totally riveting, volume after volume.
[temporal] [spatial]
Essentially the same sense is expressed by a slightly different construction, one N after another, as seen in (25). (25)
a. One day after another, the man with the foolish grin is keeping perfectly still.
394
construction after construction b. One telephone pole after another stretched along the road toward the horizon. c. That new series of books on Imperialist Grammar is totally riveting, one volume after another.
N upon N and its variant N on N likewise express spatial succession, but vertically and upward, as in (26). This sense is also preserved metaphorically, as seen in (26c). (26) a. Layer (up)on layer of mud lay on the seabed. b. We threw brick (up)on brick onto the pile. c. She scornfully piled up argument (up)on argument against my position. N after N and N upon N differ syntactically from the other three cases, in that, as seen earlier, they can occur in NP as well as adjunct positions. In NP positions they offer further syntactic opportunities that are examined in section 12.4.5. Pi 1995 points out a further use of N upon N that occurs with plural numerals and plural measure phrases, with the meaning of unexpected large quantity (27a, b). The numerals may be different. This case parallels a similar construction with the conjunction and instead of the preposition (27c, d). (27) a. Hundreds upon hundreds/thousands of demonstrators converged on the Capitol. b. buckets upon buckets of paint c. hundreds and hundreds/thousands of demonstrators d. buckets and buckets of paint This use of the construction seems to have the semantic value of an intensifier, not unlike the reduplication of prenominal adjectives for intensification, as in a long, long, long parade.
12.3 The place of NPN in the lexicon and grammar With the data so far, let us ask how the lexicon and grammar encode our knowledge of the NPN construction. The difficulty comes from the construction’s complex interweaving of regularity and irregularity. Of course, the idiomatic cases have to be listed in the lexicon. But this alone does not account for the fact that they all have the idiosyncratic syntactic pattern NPN—if simply listed, they might as well be as syntactically different as the idioms in (28). (28) by and large, all of a sudden, day in day out, time and (time) again, now and again, how dare NP VP, far be it from NP to VP, what/who/ where/ . . . the devil
construction after construction
395
The productive cases such as N after N present more serious problems. Under standard derivational frameworks such as GB or the Minimalist Program, these cases require two things: first, a general way of generating phrases of the form NPN, with all the constraints on their internal structure; and second, a special sense of the preposition after that means multiple succession and that can be inserted only in a structure where it is surrounded by identical nouns. I leave it for adherents of such frameworks to work out an account (bearing in mind all the data).8 A perspective that I find more amenable to the facts of the NPN construction is that of Construction Grammar (Fillmore, Kay, and O’Connor 1988; Goldberg 1995, 2006; Goldberg and Jackendoff 2004 [here Chapter 9]; relevant aspects of this approach also appear in Jackendoff 2002, Culicover and Jackendoff 2005; [see also the remarks preceding Chapter 7]). The basic insight of Construction Grammar is that languages can contain numerous offbeat pieces of syntax with idiosyncratic interpretations. Some samples are in (29). (29)
a. Off with his head! Into the house with you! b. One more beer and I’m leaving. Another incident like this and you’re finished. c. How about some lunch? [Hair stylist:] How about a little off the back? d. John drunk? Him in an accident? (I don’t believe it!) e. Everyone in the car! Seatbelts fastened! f. The Red Sox 10, the Yankees 3. g. three dollars a gallon; a My Lai a month [headline in The Nation]
Construction Grammar lists these utterance frames as syntactic constructions that carry with them a piece of interpretation. They all have open variables that can be filled productively: for instance the frame in (29a) stipulates a directional PP, filled productively, followed by with, followed by an NP. The utterance types in (29a–c) stipulate a piece of phonology (with, and, and how about respectively); but those in (29d–f) have no characteristic phonology—only a syntactic frame and a constructional meaning. Three features of the Constructional approach recommend it for cases like (29). First, it allows pieces of syntactic structure to be listed in the lexicon with associated meanings, just as individual words are; these are the meaningful constructions of 8 Anticipating one possible response, it is certainly possible to consider NPN part of the ‘periphery’ rather than a ‘core’ phenomenon. But such a move does not free the theorist of responsibility for an account of NPN. Moreover, the fact that variants of the construction are so widespread in the world’s languages suggests that there is something cognitively natural about it: in its broad outlines, it may be no more an accidental historical artifact than the passive. [See also comments on core vs. periphery at the end of Chapter 10.]
396
construction after construction
the language. Second, Construction Grammar makes no principled distinction between words and rules: a lexical entry is more word-like to the extent that it is fully specified, and more rule-like to the extent that it contains variables that have to be filled by other items in the sentence. The extreme case is a phrase structure rule like VP ! V–NP (or, in constraint-based format, [VP V NP]), which is made up entirely of variables and carries little or no independent meaning. The third relevant feature of the constructional approach, shared with many other constraint-based approaches, is that lexical entries are arranged in an inheritance hierarchy, so that commonalities or redundancies among words and constructions can be captured by entries at a higher level in the hierarchy. That is, Construction Grammar gives up the assumption, inherited from structuralist linguistics, that the grammar captures all regularities, and the lexicon is simply a nonredundant list of exceptions, totally distinct from the grammar. One further feature must be added to this approach in order to account for the NPN construction: the distinction between productive and semiproductive rules. A productive rule has a variable that can be filled freely by anything that meets its conditions, and so the rule can be applied to novel items, for instance amphibian for amphibian, or for that matter to brand-new nouns, for instance wug after wug. By contrast, with a semiproductive rule, one has to learn the acceptable cases such as hand to hand individually, and one recognizes novel cases, such as foot by foot, as relatively unusual. Although this distinction is rather uncommon in syntax, it is quite familiar in morphology, for instance in the distinction between the principles for regular and irregular past tenses in English (see Chapters 1 and 2). Given this distinction, we can then say that the NPN construction as a whole is semiproductive—meaning that one has to learn which cases are acceptable. Among its listed subcases are the idioms in (1) and (2). But also among its listed subcases are the productive rules N by N, N for N, N after N, and N upon N, plus the semiproductive rule N to N. This last case in turn subdivides into the semiproductive juxtaposition sense, the productive succession sense, the productive transition sense, and the productive comparison sense.a Figure 12.1 is a rough representation of this hierarchy. Some of the outlier cases, such as the orthogonal dimensions use of N by N, have been omitted; these complicate the picture further. Productive subconstructions are in bold, semiproductive in italics; approximate meanings are in parentheses. For now I a The semiproductive part of this hierarchy should be understood in the context of Chapter 2 and its subfootnotes. All entries in the hierarchy are fully specified, but the information content of an entry is reduced to the extent that it is predictable from entries higher in the hierarchy. Chapter 2 also adds the cost of referring to the higher elements in the hierarchy, based on their degree of regularity. However, in the case of this construction, unlike morphological inheritance hierarchies, it is not at all clear how to compute the degree of regularity of NPN as a whole or that of its particular semiproductive subcases.
construction after construction
397
NPN
N upon N N to N N after N hand over hand N by N N for N hand over fist (succession) (matching/exch) (succession) hand in glove (succession) Nums upon Nums (juxtaposition) (succession) (transition) (comparison) tongue in cheek (large quantity) N1 to N2 N1 to N2 hand to hand cheek to cheek face to face etc.
Fig. 12.1 Inheritance hierarchy for NPN
leave aside the important question of how the productive/semiproductive distinction is to be formulated less stipulatively. (See section 1.7 and Jackendoff 2002, chapter 6 for some discussion.) Viewed at a distance, many of the meanings associated with the NPN construction are vaguely iconic, in that they mostly deal with pairing or succession of some sort. This observation at first glance might be taken as grist for the mill of Cognitive Grammar in the sense of Langacker 1987a or Goldberg’s (1995) formulation of Construction Grammar, in which every syntactic configuration is taken to bear meaning. On the other hand, the variety of precise meanings associated with the construction—succession, spatial juxtaposition, transition, comparison, exchange, ratio, and dimension—is not predictable.9 A different perspective arises by considering not just all the meanings associated with this construction, but also all the structures or constructions that the language has for expressing the same meaning. Many of the meanings associated with NPN have alternative syntactic realizations that are not directly related to NPN. For instance, N after N has the semantically indistinguishable but syntactically distinct variant one N after another (see section 12.4.6); the quantificational reading of N(s) upon N(s) also shows up as N(s) and N(s); some instances of succession N to N can be paraphrased by from one N to the next; and the succession reading of little by little is paralleled by more and more. Moreover, all the succession readings are fairly close to many Ns in succession. This leads to the position that the syntax of English happens to have this offbeat frame NPN—a bit of noncanonical autonomous syntax—which perhaps comes with a vague iconic meaning of pairing. But the varieties of precise meaning associated with this frame, either idiomatically or in certain subconstructions, are not entirely predictable; nor is the fact that some of the subconstructions are productive and others are not. 9
One could of course respond by saying the NPN construction is inherently meaningful but multiply ambiguous. But such a position has little empirical content. If another meaning were to crop up, it would just be added to the inherent readings of the construction. In other words, no empirical discovery could falsify the position. [See the remarks preceding Chapter 7 and discussion of a similar but more severe issue with compounds in section 13.1.]
398
construction after construction
Learning the construction as a whole, then, requires learning (a) the syntactic frame and (b) the particular instances of this frame and how they match to meanings. This is not so different from learning a collection of words that fall together into a semiproductive paradigm of derivational morphology. Of course, the meanings associated with NPN are not entirely arbitrary. Looking across these cases, there appear to be at least three independent meaning components that mix and match in the various subconstructions: pairing, multiplicity, and time. A rough analysis of the cases then goes like this: (30) Juxtaposition ¼ pairing (juxtaposition N to N) Transition ¼ pairing þ time (first one place then the other) (transition N to N) Matching ¼ pairing þ multiplicity (many pairs matched up) (N for N) Succession ¼ multiplicity þ time (many instances in succession) (N by/ to/after N) This analysis is far too coarse and omits many of the details pointed out in section 12.2. In particular, many subconstructions include a sense of intensification (day after day being more intensified than every day, for instance). However, it suggests an inheritance hierarchy of meanings that cross-classifies with the inheritance hierarchy of forms in Figure 12.1, yielding a multiple inheritance hierarchy along lines familiar from HPSG (e.g. Sag, Wasow, and Bender 2003) and Construction Grammar (e.g. Goldberg 1995).b One might ask whether the taxonomy in Figure 12.1 should be considered as ‘in the grammar’ or ‘in the lexicon’. Its being ‘in the grammar’ would suggest that the construction is predominantly productive, perhaps overlaid with a few exceptions. Yet this is hardly the situation: the pockets of productivity are systematically interspersed with and embedded in idiomatic and semiproductive cases. The alternative, that the construction is ‘in the lexicon’, might suggest that unusual cases such as snake for snake are stored in the lexicon. This is unlikely, given that the construction generalizes to new nouns. Another option is that they are ‘built’ in the lexicon ‘prior to’ lexical insertion. However, such an account makes a distinction where there is none: free combination of phrases is normally taken to be in the province of phrasal syntax, not the lexicon. If ‘lexical rules’ include principles of phrasal combination such as NPN, it is hard to see why the composition of transitive verb phrases does not belong ‘in the
b This cross-classification of the semantic features with morphosyntactic form is reminiscent of the separation of morphological and semantic regularities in Chapter 2, where denominal verbs display a choice of affixes (-tion, -al, -ment, etc.) cross-cutting a choice of semantic relations to their roots (process, result, agent, etc.).
construction after construction
399
lexicon’ as well. [Recall the similar discussion in Chapter 9 with respect to semiproductive regions of the resultative construction.] A better way of framing the issue comes from reconsidering the nature of the lexicon. It is not a repository of words: it is a repository of whatever pieces of linguistic structure have to be stored in long-term memory. Some of these pieces are relatively idiosyncratic, and some include variables which must be filled productively or semiproductively in the course of building a sentence in working memory (Jackendoff 2002, chapter 6). From this perspective as well as that of Construction Grammar (Langacker 1987a, Goldberg 1995), there is no distinction between ‘in the grammar’ and ‘in the lexicon’; there are only distinctions of productivity in the lexicon.
12.4 Syntactic puzzles posed by NPN 12.4.1 What syntactic category is NPN? Let us look in more detail at the syntax of the construction. What category is it? Reviewing material from earlier (5a–d): NPN occurs in adjunct positions in the clause. It is presentential in (31a), after the subject in (31b), after the complements in VP (31c), and prenominal in (31d). (31)
a. Page for page, this is the best-looking book I’ve ever bought. [presentential] b. John and Bill, arm in arm, strolled through the park. [after subject] c. We went through the garden inch by inch. [VP adjunct] d. Your day-to-day progress is astounding. [prenominal]
N after N and N upon N can also appear in positions characteristic of NPs, as seen in (32). The other prepositions are not good here, even when they might mean something appropriate, as by does in (32) (repeated from (5e–h)). (32)
a. b. c. d.
Student after/upon/by student flunked. We filled crack after/by crack. We looked for dog after/by dog. Student after/by student’s parents objected.
[subject] [object] [object of PP] [possessive NP]
It might be reasonable to call the cases in (31) PPs. PPs are typical sentential adjuncts, and some PPs—mostly idiomatic ones—can appear prenominally, as shown in (33).c c Is there any relation between the restriction to idiomatic PPs here and the restriction to idioms found in chapter 11 for Contrastive Focus Reduplication (take-him-to-task-take him to task but take-him-toBrooklyn-take him to Brooklyn)?
400
construction after construction
(33) a. an in-your-face insult (but an in-your-eyes light) b. these out-of-the-way locations (but these out-of-the-sink dishes) c. an after-the-fact conclusion (but an after-the-concert depression) It has never been clear how the grammar licenses the idiomatic PPs but not the productive ones in (33). But however it is accomplished, presumably nothing stands in the way of also admitting prenominal PPs like (31d). In addition, at least some NPNs can occur postnominally, the position more typical for PPs, e.g. [A search case by case for the predicted paradigms] turned up no evidence. An alternative view might be that NPN is an AP when dominated by NP, and an AdvP when dominated by S or VP. In addition, NPN sometimes semantically parallels secondary AP predicates, suggesting it can be AP even when dominated by VP. For instance, the semantic role of NPN in (31b) is not so far from that of the AP in John and Bill, naked as jaybirds, strolled through the park. On the other hand, NPN displays no relationship at all to the structure of canonical APs/AdvPs, so this solution is perhaps even less comfortable than treating NPN as a PP. A referee suggests the further alternative that NPN is an NP marked with the grammatical function oblique, rather like the time expression in He sat there ten minutes. However, the latter case alternates with the PP for ten minutes, whereas there is no comparable paraphrase for adjunct NPN. In general, the adjunct positions where NPN appears are categorially unselective: they care more about semantics than syntactic category. So yet another alternative is that adjunct NPN has no syntactic category, and its position is determined on purely semantic grounds.d At this point, deeper theoretical considerations come into play than I can deal with here. For convenience I categorize adjunct NPNs as PPs, while recognizing that there is no standard category into which they fit comfortably.
d Jackendoff 2002, chapter 8, argues that the principles licensing sentence adverbials in English are categorically unselective, in that it doesn’t matter what category they are—only their semantics is relevant. The drift of the argument there is that some principles of grammar belong to an evolutionarily earlier layer of the language faculty: they are relatively ‘protolinguistic’. NPN, being anomalous in terms of standard X-bar theory, could well belong to this layer as well. Chapter 13 argues that the principles of English noun-noun compounds are also protolinguistic.
construction after construction
401
12.4.2 What is the head of NPN, and what is the rest? The internal structure of these phrases is also a puzzle. What follows the preposition is not a normal prepositional object, since, as noted, it cannot have a determiner.10 (34)
a day by a day, one face to one face, this book upon that book
The normal specifiers of PPs and APs are degree phrases and quantifiers (35a). They also include measure phrases (35b), which are arguably NPs but which, unlike the N in NPN, require specifiers.11 A bare noun acting as a PP or AP specifier is unprecedented. (35)
a. high on the hill, much further down the road, three days after the accident, very/quite beautiful, much more beautiful b. three inches long
So even if the NPN construction is a PP or AP, it is a very unusual one. The cases of N to N that allow from, as in (20)–(21) (e.g. boy-to-man transition), look like they may be at bottom compound PPs of the form from N to N, with the from deleted, as in the two alternative possibilities (36a, b). Williams 1994 suggests this solution, but it is implausible for any of the other NPN cases (36c). (36)
a. [PP from NP to NP] b. [PP [PP from NP] [PP to NP] ] c. [PP from NP by/for/after NP]
Now consider cases like (32) with after and upon, in which NPN behaves like an NP. If indeed it is an NP, what is the head? Huddleston and Pullum 2002 take the first noun to be the head and the rest a complement, as in (37). Pi 1995, Postma 1995, and Beck and von Stechow 2006 propose similar structures, though derived from something far more complex.12
10
Brian Joseph (p.c.) observes that there is a class of preposition-noun expressions in which the noun lacks a determiner, for instance in/at school, in/to bed, at/in/to college, and in British English in hospital and so on. This is quite a different construction from NPN. For instance, it occurs in normal locational PP positions. 11 Recall, however, that measure phrases do occur in the orthogonal dimensions case of N by N, as in three feet by two inches. 12 Beck and von Stechow 2006, within a formal semantics approach, propose that dog after dog is derived by deletion from one dog after the other dog, though they admit that the means of accomplishing this derivation is far from obvious. Pi 1995 derives N upon N from an unusual underlying structure in which the preposition is marked ‘iterative’, and the two nouns both originate as its complements, projected from the same position but differentiated in the third dimension. Pi’s focus is generalizing quantificational N upon
402
construction after construction
(37) [NP N [PP P NP]] This structure does account for the fact that the construction behaves like an NP. But it is quite a peculiar NP: notice again that the noun must lack a determiner, as seen in (4b) and (34).13 Furthermore, the only place that the special characteristics of the construction could be specified would be in the entry for after. This entry would have to say that after can occur as a nominal modifier that denotes iterated succession, just in case its complement and the noun it modifies are identical bare nouns (as opposed to ordinary after in the day after the accident, which denotes a single succession). This is outrageously distant from the restrictions normally found with prepositional modifiers. I would suggest that if there is any head at all in N after N, it is the preposition, and the construction as whole is simply a violation of the X-bar schema—it is an NP headed by P. As with other heads, the preposition is what determines the character of the phrase’s interpretation, and the preposition determines what nouns are possible and whether they have to match. Such a situation, in which an NP is not headed by a noun, has a precedent in the familiar English gerund construction, for instance John’s leaving the room so soon, which is arguably an NP headed by a verb (see Jackendoff 1977, Pullum 1991, Subramanian 1992).
12.4.3 Prenominal adjectives Some cases of the NPN construction—the ones with by, after, and upon— permit an adjective before the second noun, and, if that is present, they also permit an identical adjective before the first noun.
N with things like boxes and boxes of cookies; but all the differences between boxes upon boxes and the other NPN constructions are basically matters of stipulation. Note also that juxtaposition, transition, and comparison N to N are not fundamentally quantificational, so such an approach leaves them out. Postma 1995 derives NPN from an underlying (i), in which PROk moves to the specifier of DP. This is supposed to be analogous to an alleged derivation of each dog from an altogether parallel structure with a null preposition. Thus parallelism of two individually implausible structures is taken to account for the quantificational force of N after N, again ignoring the nonquantificational cases of NPN altogether. (i)
[DP [D’ D0 [PP [NP PROi dog] after [NP PROk dog] ] ] ] 13
Under the DP hypothesis, one might be able to say that the construction has the structure [DP NP–P–NP] rather than the expected [DP–P–DP]. Still, the unusual structure of NP not dominated by DP needs to be explained.
construction after construction (38)
a. b. c. d.
403
day by/after miserable day miserable day by/after miserable day awful day by/after miserable day awful day by/after day
Such adjectives are not possible with the other prepositions, not even with the succession sense of to (39a), where it would semantically parallel (38). (39)
a. The situation is getting worse week to (miserable/long) week. b. They walked along hand in (strong/greasy) hand. c. They walked along arm in (muscled) arm. d. They stood side by (firm/resolute) side. e. They stood face to (ugly) face. f. We lived hand to (hungry) mouth. g. Woman for (brilliant/wonderful) woman, you’ve got the best team.
Crucially, consider the semantics when there is only one adjective, as in (38a): day by miserable day is not speaking about a succession of days followed by miserable days, but rather a succession of miserable days. So the adjective is modifying the whole construction, even though it appears to form a constituent with only the second noun. We return to this fact in section 12.6.
12.4.4 Triplication These same cases, but especially N after N, can also be triplicated, as in day after day after day—or iterated even further, as in page after page after page after page. Triplication carries the semantic force of an intensifier. The conditions on prenominal adjectives parallel (38) and show that the adjective appears either on the last noun or on all of them, as seen in (40a–d).14 It is also possible for the second and third nouns to carry different adjectives if they grow in intensity (40e). (Curiously, the sense of (40e) is that the weeks get successively worse, while the literal meaning of after suggests that the worst week comes first.) (40)
14
a. week after week after miserable week b. miserable week after miserable week after miserable week c. week after miserable week after miserable week
This is reminiscent of the conditions on multiple conjunction: X and Y and Z and W or X, Y, Z, and W, but not X, Y and Z and W. However, it is hard to see how to draw a connection between the two phenomena.
construction after construction
404
d. miserable week after week after week e. week after miserable week after thoroughly rotten week The other prepositions resist triplication, e.g. page for page for page, a babyto-child-to-adult transition, etc.
12.4.5 Complements and postnominal modifiers to NPN in NP position A further puzzle is that when NPN is in an NP position, it can be followed by an appropriate NP complement (41a–d) or modifier (41e) (this is noticed in Huddleston and Pullum 2002, 634 n. 15).15 (41) a. b. c. d.
We endured day after day of rain. We looked at picture after picture by Renoir of his wife. We demolished argument after argument for the war. Page after page of tattered, fraying paper with the notations of a bean counter spill from the Archives. [Stefan Fatsis, Word Freak, 94] e. Country after country that I had never heard of before condemned the US invasion.
However, this is possible only when N after N is used in an NP position; elsewhere it is ungrammatical (42), paralleling other cases of NPN. (I have taken care to make the examples pragmatically plausible.) (42) a. We endured the cold, day after/by day (of rain). b. Picture after/by picture (by Renoir of his wife), we see the growth of the artist’s style. c. We compared the documents, page after/for page (of every volume). d. Country after/by country (that I had never heard of), the UN representatives condemned the US invasion. e. We traveled country to country (of Asia). [n.b.: We traveled [country to country] in Asia] Matsuyama 2004 notices that these complements and modifiers to N after N can also appear in extraposed position, given proper prosodic conditions: (43) a. Picture after picture fell off the wall of the late ambassador from Trafalmadore. b. Country after country denounced us that I’d never heard of. 15 Further idiomatic cases should also be mentioned, all in juxtaposition senses: face to face with a lion, side by side with my brothers, arm in arm with Harriet, bumper to bumper with a truck.
construction after construction
405
As with the prenominal adjectives, the complements and modifiers in (41) pertain to the entire NP: it is not days of unspecified weather after days of rain that we endured, it is all days of rain. This would suggest that the complement is attached outside both nouns, perhaps with a structure like (44). (44)
[NP [NP N P N] PP]
But there is evidence for a different structure: a short complement can be reduplicated fairly comfortably after the first noun (45a, b). Longer complements, however, become uncomfortable (45c, d).16 As with the prenominal adjectives, the first noun can have a complement only if the second noun does (45e–f). (45)
a. b. c. d. e. f.
gallon of paint after gallon of paint picture of Bill after picture of Bill gallon (??of green paint) after gallon of green paint picture (?of his sister-in-law) after picture of his sister-in-law gallon of paint after gallon picture of Bill after picture
The variant with three or more nouns likewise follows the pattern of prenominal adjectives: compare (46) to (40). (46)
a. b. c. d.
gallon of paint after gallon of paint after gallon of paint gallon after gallon after gallon of paint gallon after gallon of paint after gallon of paint gallon of paint after gallon of paint after gallon
These parallels suggest that if prenominal adjectives form a constituent with the nouns, then postnominal complements do as well. The conclusion is that there is inevitably a mismatch between syntactic and semantic behavior (Matsuyama 2004 arrives at a similar conclusion). These complements present the additional puzzle of how the NPN construction can license ordinary NP complements and modifiers despite its peculiar internal syntax—in particular despite its being headed by a preposition.
12.4.6 Comparison with one N after another and one N at a time Before going on with further properties of N after N, it is useful to digress a moment to discuss two closely related constructions, one N after another and one N at a time. The former, as noted in section 12.2.5, appears to mean the 16
The possibility of reduplication does not extend to postnominal adjuncts, no matter how short: meeting ( yesterday) after meeting yesterday; student ( in Bonn) after student in Bonn; woman ( with Phil) after woman with Phil.
construction after construction
406
same as N after N; the latter is very close to succession N by N. They both allow complements to the noun, even in adjunct position. Compare (47) to (42), in which adjunct NPNs are unacceptable. (47) a.
We endured the cold, one day of rain after another/one day of rain at a time. b. One picture by Renoir after another, we see the artist’s attention to detail. c. We compared the documents, one page of the volume after another/ at a time. d. ?One country that I had never heard of after another/at a time, the UN representatives condemned the US invasion.
Alternatively, the complements and modifiers can be placed at the end of the construction, with the flavor of Heavy Shift, as in (48). (48) a. one day after another of heavy rain/one day at a time of distressing news b. one picture after another by Renoir c. one page after another of the volume d. one country after another that I had never heard of One N after another and one N at a time both allow a full NP before the preposition, but the former prohibits a head noun or any modification on another: (49) a. one very happy child from Siberia after another/at a time b. one happy kid after another kid c. one day after another miserable one [cf. day after miserable day] After another and at a time can sometimes be extraposed as if they were complements or adjuncts of one N (50a); this is clearly impossible for N after N (50b). (50) a. One unhappy kid came in after another/at a time. b. Child came in sick after child. These constructions have the pronominal forms one after another and one at a time (note also two/three at a time, paralleling two by two); the N after N construction has no pronominal form one after one that might parallel one by one. These cases are altogether consistent with an analysis in which the noun is head, and after another and at a time are PP adjuncts, i.e. a structure like (37). They point up the extraordinary character of NPN, where none of the standard options for analysis work consistently (or even at all).
construction after construction
407
12.5 N after N as a quantifier Oehrle 1998, Pi 1995, Postma 1995, and Matsuyama 2004 observe that N after N has quantificational force and can bind indefinite NPs and pronouns from the same positions as ordinary quantified NPs can. For instance, in (51a), student after student binds a complaint, parallel to Every studenti came in with a complainti: there is one complaint per student, and they might all be different. Similarly, in (51b), his refers to each individual student, just as in Every studenti talked about hisi attitudes. (51)
a. b. c. d.
Student after student came in with a complaint. [Student after student]i talked about hisi attitudes. Teachers of student after student of ours came in with a complaint. We could see that there was a serious error on page after page of the document. e. From student after student, we’ve received a serious complaint.
Notice that one N after another can be substituted for NPN in any of these. (51) shows cases in which N after N functions as an NP. In contrast, when it is an adjunct, it cannot quantify over a singular indefinite (52) but can quantify over a bare plural (53). (51)–(53) are matched as closely as possible, configuration for configuration. The only syntactic difference is that in (51d, e), N after N is the object of a preposition (hence functioning as an NP), while in (52)–(53), N after N is a stand-alone adjunct. (52)
a. Student after student, we’ve received a serious complaint. [only good if it’s the same complaint each time] b. We could see that there was a serious error, page after page/one page after another. [where it’s a different error on each page]
(53)
a. Student after student, we’ve received serious complaints. b. We could see that there were serious errors, page after page.
Adjunct N by N, even though it means about the same as N after N, cannot be quantificational at all: compare (54) with (51e), (52a), and (53a). (54)
a. Student by student, we’ve received a serious complaint. [if OK, only one complaint] b. Student by student, we’ve received serious complaints.
One N after N can even take scope over another, as in (55). (55)
Student after student turned in lousy paper after lousy paper.
408
construction after construction
Oehrle reads (55) as having a branching quantifier: there is a succession of student–paper pairs. I can also read it as one student after another turning in a succession of lousy papers, that is, with the subject taking scope over the object. It is hard to specify exactly what kind of quantifier this is. Pi 1995, Oehrle 1998, and Matsuyama 2004 note that N after N in subject position takes singular agreement, like every/each N (56a). Richard Larson has pointed out (p.c.) that like every/each N, N after N cannot be the antecedent of each other (56b). (57) a. Page after page is/are sprinkled with errors.17 b. Dog after dog bit the others/each other. However, Matsuyama also observes that, with respect to aspectuality, N after N behaves more like a bare plural (57b) than like every N (57c). It also does not pattern exactly with many Ns in succession, which is often a decent paraphrase (57d). (58) a. For hours/In an hour, student after student walked into my office. b. For hours/In an hour, students walked into my office. c. In an hour/For hours, every student walked into my office. [ in reading where each student walked into my office once] d. For hours/In an hour, many students in succession walked into my office. So although N after N is syntactically singular, it is evidently semantically some sort of plural (as might be expected). The basic problem posed by these data is how to get a quantificational interpretation out of a syntactic structure NPN, with nary a quantifier in sight. In particular, in a framework such as GB or the Minimalist Program, quantificational interpretations are derived by raising quantified phrases to the position of their scope at LF. But here there is nothing overt to trigger raising. One could of course designate N after N as something that raises, but such syntactic raising is patently not based on its syntax—only on its semantics. One could alternatively propose a special phonologically null quantifier that triggers raising. However, consider how such a null quantifier would have to be licensed: it must somehow be associated with the use of after and (up)on, specifically in the context of NPN. In other words, the preposition must specify
17 I am told (Moira Yip, p.c.) that in British dialects where words like team can take plural agreement, N after N does too: Team after team are/is drinking beer.
construction after construction
409
not only the superficial syntax of the construction, but also the presence of this null quantifier—a kind of syntactic stipulation otherwise unknown (at least to me). Even if we admit such a derivation from something more obviously quantificational, such a proposal fails to capture the generalization that the overt syntax of quantificational NPNs has the exact same internal form as nonquantificational cases such as N by N. In short, such a proposal would be just a stratagem to preserve the hypothesis of syntactic quantifier raising. An alternative conclusion is that the syntax of NPN simply does not match its semantics. Such a conclusion is nothing new. Culicover and Jackendoff 2005 argue that a strong match between syntax and semantic structure is only a default alignment, and there exist numerous mismatches. In particular, Construction Grammar is founded on arguments that syntactic structure too can carry meaning, independent of the words in it.18 Section 12.3 argued that the NPN construction presents another example of a meaningful construction; the present section has shown that part of the meaning of N after N used as an NP is that it is quantificational—whatever its peculiar syntax. In turn, however, this means that there cannot be a syntactic level of LF in which all scope of quantification is overt. Rather, scope of quantification is encoded in semantic structure only, even if it may depend in part on the syntactic position of the quantificational phrase. This is the position urged by Culicover and Jackendoff 2005; NPN presents especially striking evidence. Let us return to the NP complements and modifiers in (41) and (43). In light of our conclusion about quantification in N after N, we might conclude as well that the licensing of complements and modifiers of N after N is determined by semantics, where N after N has roughly the same semantic structure as many Ns in succession. That is, in semantics there are not two Ns that license complements and modifiers; there is only one. Furthermore, the complements and modifiers are realized in syntax exactly as they should be. It is only the weird syntactic realization of the semantic quantifier–head configuration as NPN that leads to the strange configuration in (41) and (43); at a semantic level everything is as it is supposed to be. As with quantification, this conclusion leads to broader questions about how licensing of NP complements and modifiers is accomplished, and it suggests that semantics plays a larger role than is usually assumed.
18
One need not agree with some strains of Construction Grammar that all structure is meaningful. But certainly some structure is meaningful, as seen in the constructions enumerated in (29) above. [See again the remarks preceding Chapter 7.]
410
construction after construction
12.6 A lexical entry for the NPN construction A final question is how the rule producing the NPN construction is stated. I propose that this is a reduplicative construction, along the lines of the ‘contrastive focus reduplication’ construction discussed in Ghomeshi et al. 2004 [here Chapter 11] (see also Horn 1993) and illustrated in (58). (58) a. I’ll make the tuna salad, and you make the salad-salad. b. Would you like some wine? Or would you like a drink-drink? Most work on reduplication has focused on how the phonological component accomplishes reduplication. However, Chapter 11 asks a different question: how the lexicon encodes a morpheme that, like other morphemes, carries a meaning and syntactic structure, but which has reduplicative phonology. Among the possible analyses explored in Chapter 11, the most promising is a constructional analysis based on the parallel architecture of Jackendoff 2002, in which the phonological realization of the focus operator is a reduplication of the word being focused. A similar approach appears to work for those versions of NPN that require reduplication. The principle can be stated as a constructional idiom that maps between meaning and syntax roughly like (59): (59) Meaning: MANY Xi’s IN SUCCESSION [or however it is encoded] Syntax: [NP Ni Pj Ni] Phonology: Wdi afterj Wdi The entity being counted maps into both nouns, as notated by the shared subscript i. The preposition in syntax maps into after in phonology but does not map directly into anything in semantics. This is what makes the construction idiomatic or noncompositional. In effect, then, this account parallels reduplicative expressions of quantification in other languages, such as Hebrew boker boker, ‘every morning’, except that it has this idiosyncratic preposition in the middle. In order to incorporate adjective modifiers, (59) can be amplified to (60). There are two cases, depending on whether the adjective is reduplicated along with the noun.19 (60) Meaning: MANY (MODk) Xi’s IN SUCCESSION a. Syntax: [NP Ni Pj (Ak) Ni] Phonology: Wdi afterj (Wdk) Wdi
19
I leave open the issue of whether the A–N sequences are constituents.
construction after construction
411
b. Syntax: [NP (Ak) Ni Pj (Ak) Ni] Phonology: (Wdk) Wdi afterj (Wdk) Wdi In (60a), the modifier is mapped to a prenominal adjective preceding the second noun; in (60b) it is copied before the first noun as well. This treatment accounts for the fact that a single adjective before the second noun modifies the entire construction: in the semantics, there is only one counterpart of the two nouns, and this is what is modified. A more extreme variant of the formalization might be stated as (61) (omitting the adjective for convenience). (61)
Meaning: MANY Xi’s IN SUCCESSION Syntax: [NP Pj Ni] Phonology: Wi afterj Wi
Here the first noun is even absent in the syntax—it appears only in the phonology. For the case of English, there appears to be no reason to have the first noun in syntax: it doesn’t move around or take its own modifiers. However, the Polish NPN construction does provide some evidence for the presence of two separate nouns in syntax, in that the second noun receives the instrumental case governed by the preposition za, but the first noun receives the case governed by the larger environment (Barbara Citko, p.c.). (62)
a. Prof. Kowalski oblewa studenta za studentem. Prof. K flunks student-ACC behind student-INSTR (‘student after student’) b. Miasto za miastem glosowalo przeciw tej ustawie town-NOM behind town-INSTR voted against this measure (‘town after town’)
This account of NPN as reduplication is stipulative, but it does the trick, and it lays the groundwork for a more detailed investigation of the syntax and semantics of the NPN construction in the context of Construction Grammar and the parallel architecture.
12.7 Inconclusion The discussion here has hardly ‘solved’ the NPN construction. It has however exposed some of the construction’s unexpected complexity and has reflected on the consequences for theoretical alternatives. The syntax of NPN and the way it maps to semantics are sufficiently distant from canonical structures that it
412
construction after construction
appears hopeless to find a classical derivation: we have to accept it as a ‘syntactic nut’ that is simply a special case in the grammar of English. The distribution of NPN vividly illustrates the continuity between idiosyncrasy, semiproductivity, and full productivity argued for by Construction Grammar and related approaches. It thus provides strong arguments for a theory in which all knowledge of language is encoded in terms of stored pieces of structure organized into an inheritance hierarchy, without a strict separation between lexicon and grammar. Finally, the ways in which NPN deviates from canonical structures lead to the conclusion that at least some licensing of complements and modifiers is a function of semantics rather than syntax, and still more striking, that scope of quantification cannot be canonically represented at a syntactic level of LF. The parallel architecture framework of Jackendoff 2002 and Culicover and Jackendoff 2005 accommodates these conclusions nicely. I leave it as a challenge for practitioners of competing frameworks to account for them.
chapter 13
The Ecology of English Noun-Noun Compounds (2009)
13.1 Compounds: On the cusp between grammar and lexicon Compounding forms a subsystem of grammar somewhat distinct from ordinary syntax and morphology. The present chapter explores the properties of this subsystem and its place in linguistic theory. On grounds of manageability I have confined the analysis to English noun-noun compounds, which appear to be representative of compounding more generally in English and other languages.1
This chapter is an expanded and revised version of Jackendoff 2009, incorporating material from a much longer manuscript of 1998 whose completion was interrupted by the more pressing need to write Foundations of Language. The study is based on examination of a corpus of approximately 2,500 compounds. About half of these were collected by students in my Semantics course at Brandeis University in spring 1997. I thank Erica Goldman, Joy Budewig, and Kristen Lauer for their help in organizing this material. The corpus also includes many examples from the linguistics literature, in particular all of Lees’s (1960) examples, as well as about 400 examples from random issues of the Boston Globe and The Nation. I am grateful to James Pustejovsky for much useful discussion; to him and Joan Maling, Federica Busa, Hildy Dvorak, and Dan Dvorak for further examples; and to Nigel Love, Herbert Brekle, Christian Bassac, Geert Booij, Peter Culicover, Susan Olsen, Biljana Mišić Ilić, Andrew Carstairs-McCarthy, and Shelly Lieber for comments on earlier versions of the chapter. 1 The corpus happens to include a number of compounds that quite possibly are V-N compounds rather than N-N, for example watchdog, flashlight, and repairman. In most cases the first word is both a noun and a verb form, so it is difficult to judge. In any event, there seemed no interesting reason to omit them from the corpus once they were there.
414
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
The surface pattern of English noun-noun compounds (teabag, oil drum, sunflower, bedframe, copy machine, engagement ring) is easy to characterize: two concatenated nouns, henceforth N1 and N2, that together form a noun. (1)
[N N1N2]
Typically the stress falls on N1, as in the examples above. However, there are classical well-known exceptions such as apple pı´e, forming a minimal pair with a´pple cake (another such pair is a´lto horn vs. alto reco´rder). I will not be too concerned with the reasons for the occasional shift of stress, though there are cases such as aluminum fo´il, contrasting with tı´nfoil, where it turns out to make some difference (see n. 20). Compounding is recursive, often productively. The list in (2) includes some conventionalized combinations, but also several that are no doubt novel to many speakers, yet perfectly interpretable. (2)
[ice cream] cone [line item] veto [accident prevention] program neighborhood [liquor store] [campaign finance] indictment [screen door] [key hole] [winter weather] [skin troubles] [ [ [health management] cost] containment] services (or is it [ [health management] cost][containment services]?) [ [diesel electric] engine] [dump truck] [ [ [Volume Feeding] Management] [Success Formula] ] Award (from Gleitman and Gleitman 1970)
As pointed out by Levi 1978 and ten Hacken 1994, the nouns can also be elaborated in quasi-syntactic fashion: (3)
a. Conjunction: [health and welfare] fund [primary and caucus] states [cock and bull] story baseball [gloves and helmets] [media, retail, and property] empire [relaxation and [focus concentration] ] techniques [ [stress relief] and [aroma therapy] ] product
the ecology of english n o u n - n o u n c o m p o u n d s
415
[Fresh Pond Parkway] [sewer separation and surface enhancement] project b. Adjective-Noun [foreign exchange] flow [sexual harassment] laws [First Amendment] values [smoked [pork shoulder] ] [boiled dinner] c. Paired arguments [love-hate] relationship [Port-cornstarch] mixture [town-gown] tensions d. Numeral-Noun [two-car] garage [2300-word] corpus [five-inch] hole [three-part] harmony (note car garage, inch hole, part harmony) [two-axle] [ [ [diesel electric] engine] [dump truck] ] The frequency of novelty demonstrates that the compounds of English cannot just be listed; rather compounding must include a productive rule system. Even with the quasi-syntactic complications of (3), the morphosyntax of noun-noun compounds is rather simple. My interest here lies in a different question: how the meaning of a compound is built from that of N1 and N2. It is well known (e.g. Selkirk 1982) that N2 usually is the semantic ‘head’ of the compound, in the sense that the compound usually denotes a particular subtype of the type denoted by N2, as in the partial paraphrases in (4). For now, let us call this the Head Principle; it will be stated more formally in section 13.5. (4) a. sunflower ¼ ‘a kind of flower’ b. engagement ring ¼ ‘a kind of ring’ c. health and welfare fund ¼ ‘a kind of fund’ Despite the semantic asymmetry, there is no syntactic reason to elaborate the morphosyntactic structure (1) into an asymmetric structure in which N2 is somehow syntactically distinguished as the head. In particular, since N2 is at the end, the plural ending will go on it with nothing further said. Two sorts of exceptions to the Head Principle are well known. One sort are exocentric compounds (called bahuvrihi by Panini, as conveyed into modern tradition through Whitney to Bloomfield). These have a tacit semantic head (5),
416
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
and sometimes use N2 metaphorically, as it were (6). The other sort (called by Panini dvandva) equates the two components (7).2 a. blockhead ¼ ‘someone with a head like a block’ 6¼ ‘a head like a block’ b. duck call ¼ ‘something that makes a noise like the call of a duck’ 6¼ ‘the call of a duck’ (6) a. seahorse 6¼ ‘a kind of horse’ ¼ ‘something that resembles a horse that lives in the sea’ b. pigtail 6¼ ‘a kind of tail’ ¼ ‘a (human) hair arrangement that resembles the tail of a pig’ (7) a. tractor-trailer ¼ ‘something that consists of both a tractor and a trailer’ b. fisherman ¼ ‘someone who is both a fisher and a man’ c. pantyhose ¼ ‘something that is both a panty and hose’
(5)
Many writers on compounds explicitly exclude these three types from their analysis. Sections 13.5.3 and 13.7 will show how they fall under a suitably generalized form of the Head Principle. Even disregarding these cases, the Head Principle is far from sufficient to determine the meaning of a compound. Consider the variety of meaning relations between N1 and N2 in (8)–(9). (8)
a. b. c. d. e. f. g.
chocolate cake ¼ ‘a cake made with chocolate in it’ birthday cake ¼ ‘a cake to be eaten as part of celebrating a birthday’ coffee cake ¼ ‘a cake to be eaten along with coffee and the like’ marble cake ¼ ‘a cake that resembles marble’ layer cake ¼ ‘a cake formed in multiple layers’ cupcake ¼ ‘a little cake made in a cup’ urinal cake ¼ ‘a (nonedible) cake to be placed in a urinal’
(9)
a. b. c. d. e. f. g. h. i.
railroad car ¼ ‘a car that runs on a railroad’ cattle car ¼ ‘a (railroad) car to carry cattle’ dome car ¼ ‘a (railroad) car that has a dome’ refrigerator car ¼ ‘a (railroad) car that serves as a refrigerator’ dining car ¼ ‘a (railroad) car in which one dines’ observation car ¼ ‘a (railroad) car from which one observes’ handcar ¼ ‘a (railroad) car powered by hand’ police car ¼ ‘a car (automobile) used by the police’ grease car ¼ ‘a car (automobile) equipped to be fueled by grease’
2 There are also some families of left-headed compounds in English, such as attorney general, mother-in-law, blowup, and pickpocket. But they are not noun-noun compounds.
the ecology of english n o u n - n o u n c o m p o u n d s
417
This range of semantic relations is not confined to conventionalized compounds: Wisniewski and Gentner 1991 show that a similar wide range is found in people’s judgments of novel compounds. Moreover (as observed in Chapter 5), the compound construction overlaps considerably with N of NP. Each pair in (10) shows a paraphrase between a compound and an N of NP, but with a different meaning relation between the two nouns in each pair. (10)
a. ant heap ¼ heap of ants b. heart beat ¼ beat of the heart c. donut hole ¼ hole of a donut d. power supply ¼ supply of power e. shoemaker ¼ maker of shoes f. love song ¼ song of love
It therefore seems more reasonable to say that the meaning of a noun-noun compound is (for a first approximation) compatible with any relation between N1 and N2, provided it conforms to the Head Principle. Sections 13.6–7 will in fact show that the class of possible meaning relations between the two nouns is the product of a generative system, so it is impossible to enumerate them. This conclusion, if correct, decisively refutes the assumption that the meaning of a phrase can be constructed solely from the meaning of its constituents plus the way they are syntactically combined (‘Fregean composition’). Thus no theory that derives meaning just from words plus syntactic structure—be it early transformationalist approaches such as Lees 1960 and Levi 1978, the Minimalist Program, or standard formal semantics—can adequately account for the semantics of compounds. This conclusion also requires us to abandon the strong version of Construction Grammar (Langacker 1987a, Goldberg 1995), in which each piece of syntactic structure can be assigned a constructional meaning, and the meaning of a phrase is constructed from the meanings of its constituents plus the meaning of the construction. In some cases such as the resultative (Chapter 9) and the NPN construction (Chapter 12), the possible meanings can be enumerated, and one can justifiably say the syntactic structure is ambiguous. But in the case of compounds, where the possible meanings cannot be listed, this move cannot be sustained. One possible way out is to claim that compounds, like idioms, are stored units. This is certainly true for some compounds: speakers store thousands of lexicalized compounds with semi-idiosyncratic meanings. One knows not just that peanut butter is a buttery substance made from peanuts, but exactly what peanut butter is and what it tastes like. But compounds can’t all be stored in the
418
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
lexicon: as we have already observed (and as Levi 1978 points out in response to the strictly lexical account of Jackendoff 1975 (¼ Chapter 2) ), new compounds can be built on the fly. For instance, among the several hundred compounds from the Boston Globe in my corpus, at least several dozen were novel to me, including cotinin [sic] level, politician tycoon, aid inflow, locality pay, and spring seepage. Downing 1977 stresses the frequent coining of compounds in particular discourse situations, citing for instance bike girl being used for a girl who left her bike in the vestibule, and apple juice seat for a seat at which apple juice was set on the table. Some years ago, my daughter left something for our cats, accompanied by the note ‘Maggie and Peanut’s heat wave present’. Gleitman and Gleitman 1970, Brekle 1986, and Sadock 1998 make similar observations. There is also evidence from acquisition: Clark, Gelman, and Lane 1985 observe that children begin understanding novel compounds and coining their own between about 2½ and 3½ (see Berman 2009 for crosslinguistic documentation). In short, compounding falls under the classical arguments that the grammar cannot just be a list but rather must include a productive rule system. On the other hand, the difficulty with a productive rule system is that (at least as we have traditionally understood productive syntactic and phonological rules) it suggests relatively exceptionless regularity. How is it possible to reconcile the productivity of compounds with their rampant idiosyncrasy? Levi suggests isolating the lexically listed compounds from the productively generated compounds. But this raises the question: Why should the listed compounds look at all like the productive ones? For instance, if listed compounds are exceptional, then why aren’t some of them left-headed? Looking more closely, we can see that one cannot draw a principled line between listed and generated compounds. For example, soccer ball, like peanut butter, is listed in my lexicon: I can connect it to particular physical objects. On the other hand, although I may not have ever heard bocce ball before, I can guess on the fly that it is a ball for playing bocce. But it certainly is listed in the lexicons of bocce players, so speakers may differ in whether they list or ‘generate’ particular compounds. In short, in a theory with a strict division between lexicon and grammar, compounds are problematic.3 Some perspective comes from considering VPs. VPs are certainly generated by a productive rule system. At the same time there are hundreds of idioms consisting of whole or partial VPs, and these too must be stored in the lexicon.
3
Ten Hacken 1994 takes his task to be to define a sharp line separating compounding from both lexicon and syntax. Such an approach, whatever its a priori appeal, is demanding joints in nature where there aren’t any.
the ecology of english n o u n - n o u n c o m p o u n d s
419
And the same problems of redundancy arise here: how can the lexicon specify the idiom give NP the once-over without somehow mentioning that it contains the good old irregular verb give?4 Of course, the syntactic variety of VPs, especially freely generated ones, is far greater than that of compounds. But the problem of reconciling productivity with lexical listing is qualitatively exactly the same. So compounds do not present a unique problem in this respect. The Parallel Architecture’s treatment of the lexicon, illustrated for VPs in Chapters 7–9 and for the NPN construction in Chapter 12, generalizes nicely to compounding: it requires no principled line between freely generated compounds and morphosyntactically complex listed items with the same structure. In this approach (shared by HPSG and some versions of Construction Grammar), lexical items and rules of grammatical combination are couched in identical formalisms, so that there is an easy formal transition between one and the other. Moreover, this approach takes for granted the redundancy between general principles (e.g. the free composition of VPs and compounds) and their more specialized elaborations (e.g. VP idioms and conventionalized compounds), and their relationships are stated in terms of inheritance hierarchies. Thus lexically listed complex items that conform to more general structural principles (e.g. kick the bucket and snow man) can be treated simply as specialized instances of general schemas.5 As in the approach of Chapter 2, a conventionalized compound is fully listed. Its ‘cost’ is measured in terms of its own information content, minus the degree of its redundancy with more general principles, plus the cost of referring to these more general principles. In the case of compounds, the cost of referring to general principles includes the information that one possible relation rather than another is the right one. For example, the ‘eaten with’ relation in coffee cake is predictable by general rule (see sections 13.6–7), but there should still be a cost to the choice between ‘cake eaten with coffee’ and the equally plausible ‘cake made with coffee’, parallel to chocolate cake, both of which are products of general rules. Of course, the lexicon being what it is, we also expect some
4
Note that it also contains the totally exceptional once-over, which looks suspiciously like a compound noun. 5 Ryder 1994, working within a Cognitive Grammar framework, recognizes the existence of patterns in compounding ranging from the very general to the very particular, which govern both the listed compounds and the production and comprehension of novel compounds. However, she seems to think that the regularities of compounds essentially result from the pull of statistical patterns in what people hear. I acknowledge such influences, particularly at very specialized levels (e.g. if Ryder is correct about the data, why should there be so many compounds beginning with sea and so few with ocean?). But this chapter shows that there is more going on in compounds than statistical patterns.
420
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
proportion of lexicalized compounds to exhibit inexplicable irregularity due to historical drift and capricious coinages. These will be less closely related to general schemas, and will therefore ‘cost’ more to list. The task for the language learner, then, is to learn the lexicalized compounds and to acquire the rules for interpreting novel compounds. These two processes are obviously interdependent. At the outset, all compounds are novel, and children somehow list some of them in their lexicon. The child’s evidence for the general principles, insofar as they are not innate, must come from generalizing over the learned compounds (see Jackendoff 2002, section 6.9 for more on this view of acquisition). One still might wonder whether speakers really have general principles for compounding. Gleitman and Gleitman 1970 and Ryder 1994 asked subjects to give putative definitions of constructed novel compounds. In both experiments, subjects were far from reliable in giving answers that conformed to linguists’ intuitions, for example giving definitions like (11). (11) a. Ryder (137) willow forest: ‘a willow that grows in the forest’ giraffe land: ‘a giraffe on land’ bottle flower: ‘a bottle in which one puts a flower’ b. Gleitman and Gleitman (156–7, 167) bird-house glass: ‘a birdhouse made of glass’ house-bird wash: ‘a bird that washed the house’ house-bird boot: ‘the boot, the shoe, the house-boot of the bird’ These definitions are reminiscent of the performance of agrammatic aphasics on reversible passives and object relatives: the subjects seem to be driven by a need to create maximal semantic plausibility, grammatical principles be damned. Such results lead us to ask whether there is any grammatical competence at all involved in interpreting novel compounds. I believe there is, but it is not of the sort we are led to expect from standard syntax. Standard syntax pretty well determines the meaning of a novel phrase from the meanings of the words and the principles—including constructional meanings—for mapping syntactic structure into combinatorial relations among constituent meanings. In the case of a novel compound, though, the general principles yield only a range of possible meanings. The language user must home in on the intended meaning of a novel compound by making use of (a) the semantic details of the constituent words and (b) the discourse and extralinguistic context. In the experimental situations presented by the Gleitmans and Ryder, the compounds are presented in isolation with no contextual support. And the semantic details of the constituent words press for a meaning relation at
the ecology of english n o u n - n o u n c o m p o u n d s
421
odds with that demanded by the grammar, in particular right-headedness. Evidently the grammatical principles are sufficiently unstable (for some speakers) that they can be overwhelmed by semantic plausibility in such situations. My conclusion is that, although compounding is indeed productive, the productivity is rather fragile by the usual standards of productive grammar. When I teach compounding in Introductory Linguistics, students enjoy building up a compound like (12a) piece by piece, in such a way that everyone understands it at the end. But if I were to present it as a whole to a naive class, few would get it. By contrast, the syntactic paraphrase (12b), while unwieldy, is nevertheless easier to comprehend. (12)
a. an inflectional morphology instruction manual software programming course b. a course in programming the software that accompanies manuals that teach inflectional morphology
Similarly, example (13a), from the New York Times (3 June 2007), is initially hard to understand, because one’s tendency is to parse it pairwise, like (13b). There is no grammatical support for the proper parsing (13c). (13)
a. child camel jockey slavery b. [ [child camel] [jockey slavery] ] c. [ [child [camel jockey] ] slavery], i.e. ‘slavery of [children serving as camel jockeys]’
The upshot is that the fragility of productive compounding also has to be part of the account.
13.2 Compounds as an evolutionary throwback Bickerton 1990 and Givo´n 1979 propose that the language capacity evolved in two stages, ‘protolanguage’ and modern language; they conceive of the former as having had a vocabulary and pragmatics, but no syntax or morphology as such. This stage for them is not just hypothetical: in the standard manner of evolution, it was not thrown away when modern language evolved. Rather, modern language was built on top of it, as a refinement. Moreover, in situations where the complexity of modern language is disrupted or impaired, elements of protolanguage still emerge. The sorts of situations Bickerton and Givo´n have in mind include pidgin languages, the two-word stage of language learning, agrammatic aphasia, the
422
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
language acquired by late first language learners such as Genie (Curtiss 1977), and what apes instructed in sign language can learn. In each of these cases, vocabulary is acquired and words are concatenated into larger utterances. However, inflectional morphology is at best highly defective, functional categories are rarely used, and syntax does not go much beyond simple basic clauses. Subordination is largely absent, replaced by parataxis (jamming two independent clauses together); and many of the connections between words and between clauses are left up to the hearer’s understanding of context. These situations of course differ from each other in many respects, due to their different genesis (in particular, the apes appear to string words together rather randomly, if we are to believe Terrace 1979). What is interesting is that when less than full language is in evidence, the parts that appear (or remain) are remarkably similar. Bickerton and Givo´n argue that these are all instances of the resurfacing of protolanguage, which is in some sense more resilient in the brain than the refinements of modern language. Three further cases can be added to this collection. One is the degree of language competence achieved by the right hemisphere (Baynes and Gazzaniga 2005). Another is the ‘home sign’ created by deaf children of non-signing parents (Goldin-Meadow 2003). The third comes from a massive longitudinal study of immigrant second language learners by Klein and Perdue (1997). All of them turn out to pass through a stage of second-language competence that Klein and Perdue call the Basic Variety (BV)—and some fail to progress beyond this point despite years of exposure. Klein and Perdue say (333) that although ‘the BV is a highly efficient system of communication . . . it lacks some of the structural characteristics which we typically find in fully fledged languages’. Speakers of BV acquire vocabulary, but there is hardly any morphology or closed-class vocabulary, and no syntactic subordination—only parataxis. Word order in BV is determined in terms of rudimentary functional principles, primarily Agent First and Focus Last6— principles that also apply broadly in pidgins and function as strong defaults in modern language as well. These principles do not make use of the syntactic categories and fully articulated phrase structure of standard syntax: they can be regarded as mappings directly between semantic roles and linear order of words in phonology. In short, BV too has the telltale symptoms of protolanguage. Jackendoff 2002 (chapter 8) suggests that protolanguage is a cognitive ‘scaffolding’ on which modern language is built, both in evolution and in 6 As pointed out by some of the commentators on Klein and Perdue’s paper in the same issue of Second Language Research, some of the word order effects may be artifacts of the range of target languages studied by Klein and Perdue. I leave this question open, as it is not too important to the point at hand.
the ecology of english n o u n - n o u n c o m p o u n d s
423
development. Under this view, various ‘modern’ aspects of language can be added or lost piecemeal in different situations, revealing different amounts or aspects of the ‘scaffolding’. The surprise is how robust this protolinguistic scaffolding is, emerging over and over again under different conditions. In terms of the Parallel Architecture, protolanguage is a subsystem of modern language: it consists of just phonological and semantic components, plus a direct interface between phonology and semantics. By contrast, mainstream generative grammar has to consider protolanguage a completely unrelated system, since it claims that phonological and semantic combinatoriality—and a correspondence between them—cannot exist without the generative capacity of syntax. This view of modern language as ‘laid over’ a protolinguistic substrate leads to the intriguing possibility that the coverage is incomplete: that relics of earlier stages of the language capacity remain as pockets within modern language. These relics would have more rudimentary grammatical structure, and such grammatical structure as there is would not do much to shape semantic interpretation. Rather, we would expect semantic interpretation to be highly dependent on the pragmatics of the words being combined and on the contextual specifics of use. Fanselow 1985 and Jackendoff 2002 (chapter 8) suggest that compounding is a strong candidate for such a relic of the simpler system. Fanselow draws parallels between compounding, the two-word stage of language acquisition, Broca’s aphasia, deep dyslexia, right-hemisphere competence, the competence of language-trained apes, and the grammatical properties of newspaper headlines,7 seeing in this ensemble of phenomena manifestations of a ‘rudimentary language capacity’. In confirmation, Klein and Perdue observe that although BV lacks inflectional and derivational morphology, one kind of morphology does appear (in the target languages that permit it): compounding. The point is solidified by the early appearance of creative compounding in child language, well before other morphology. This view of compounding explains the properties that make it not look like the rest of morphology (differences stressed by Sadock 1998): compounding is actually not a grammatical phenomenon, but a protogrammatical one. Even the right-headedness of (English) compounds, their most grammatical feature, really relies only on a language-specific correlation of linear order with semantic headedness, not on X-bar head-argument structure. In this respect it resembles the Agent-First and Focus-Last principles of BV.8 7
The relation of newspaper headlines to child language is also explored by De Lange, Vasic, and Avrutin 2009. 8 Notice, by the way, that the standard Head Parameter in syntax is also a correlation of linear order with headedness. So perhaps it is no surprise that compounds often generalize with X-bar structures in their headedness, as Lieber 1992 suggests.
424
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
Pushing this point further, the syntax of English compounding is to some extent even blind to syntactic category. Nominal compounds need not be constructed from two nouns (doghouse): there are also compounds like (14), most of which have characteristic left-hand stress of compounds.9 (14) a. [longAbowN]N, [watchV dogN]N b. [underP currentN]N, [overP coatN]N, [underP pantsN]N, [underP classN], [inP fluxN]N, [inP seamN]N, [downP sideN]N, [outP houseN]N, [inP-jokeN]N c. [blowV upP]N, [castV offP]N, [turnV overP]N (two senses), [popV overP]N, [pullV overP]N, [pushV overP]N, [driveV inP]N d. [overP killV]N, [inP flowV]N, [outP lookV]N e. [speakV easyA]N f. [hearV sayV]N, [lookV seeV]N g. [once?overP]N h. [skinN deepA]A, [wormN eatenV]A Moreover, there are pairs of semantically indistinguishable N-N and A/P-N compounds, such as atom bomb vs. atomic bomb, sea life vs. marine life, and topcoat vs. overcoat, suggesting that syntactic category hardly makes a difference. And for some compounds such as guard dog, there is no fact of the matter as to whether they are V-N (‘a dog that guards’) or N-N (‘a dog that serves as a guard’). In short, compounding is only barely syntactic. It has right-headedness and the recursive properties illustrated in (2)–(3), but that’s about all. Under this hypothesis, compounding takes on an unexpected status in grammatical theory. It is not some odd peripheral aspect of morphology; it is a system that reveals some of the evolutionary roots of modern language, as it were a coelacanth of grammar. Such a conclusion should not be taken as too radical. After all, the semantic relations that link discourse together are not marked syntactically either. Rather, hearers use the meanings of the sentences plus understanding of the context—including social context—to create the semantic linkages from sentence to sentence, whether spoken by the same or different
9
Lieber 2004 argues that V-P compounds such as blowup are simple zero-nominalizations of verb-particle combinations. I find this plausible for many cases, but not for exocentric V-P compounds such as pullover (a sweater), pushover (a person), and turnover (a pastry). Lieber also suggests that P-V compounds such as overkill are actually prefixed nominalizations. Here I am less convinced, as there are many semantically idiosyncratic examples, such as undertow, outlook, downdrift, outcast. (See also Huddleston and Pullum 2002, 1654–5.) In any event, the main point about categorial unselectivity stands, even if less robustly. I am grateful to Lisa Pierce (p.c.) for bringing [V-V]N compounds to my attention. Some instances, such as make-believe and stirfry, are probably zero nominalizations of compound verbs, but those listed in (14f ) are not.
the ecology of english n o u n - n o u n c o m p o u n d s
425
speakers. Compounding is just the same sort of phenomenon writ small. We return to this point in section 13.8. (Jackendoff 2002 suggests some other possible phenomena in the corners of modern grammar where earlier evolutionary stages might be showing through.)
13.3 Preliminaries to semantic analysis of English N-N compounds The framework of Conceptual Semantics is deeply concerned with details of word meaning and how these interact with composition of phrase meanings. At the same time, it incorporates a great deal of what is usually called pragmatics: aspects of meaning that are not encoded in word meanings or in relations conveyed directly by syntactic structure. This outlook impacts on the approach to compounding: the goal is an account of compound meaning that is as rich as the account of word meaning. The basic intuition, as in other approaches, is that the meaning of a compound is a function of the meanings of its constituents. Thus the problem is: given two nouns N1 and N2 meaning X1 and Y2 respectively, what is the function F(X1, Y2) that yields the meaning of the compound [N1 N2]? Of course, it is important to recognize the limits of compositionality in compounds. For novel compounds (such as backgammon ball), compositionality should be all that is available. But lexicalized compounds usually also incorporate idiosyncratic information. For instance, nothing principled predicts the difference in shape between stereotypical soup bowls and fish bowls, or that a boxcar is a kind of railroad car but a kiddy car is a kind of toy. And general principles cannot account for cranberry morphemes (underlined parts of (15a) or what I like to call ‘strawberry’ morphemes (real words within compounds that play no role in the compound’s meaning, e.g. underlined parts of (15b) ). (15)
a. cranberry, basset horn, bogeyman, pratfall, fascia board, ferris wheel, linchpin, iceberg, fig newton, nightmare b. strawberry, cottage cheese, polka dot, bobby pin, dogwood, horseradish, monkey wrench, gangway, Charley horse, tea cozy, hillbilly, sheepfold, sidekick, cotton gin, airplane, water moccasin, lawsuit
So sometimes lexicalized meaning has to ignore the semantics of one noun or the other, just as it ignores bucket in kick the bucket. Still, on the whole there is significant compositionality.10 10
If the measure of lexical ‘cost’ is as proposed in Chapter 2, the fact that the strawberry morphemes are phonologically independent words ought to count as a redundancy, and subtract from the cost of these compounds in the lexicon, even though their meanings play no role. This might account for the common misspelling of linchpin as lynchpin, where the cranberry morpheme has been converted into a strawberry morpheme.
426
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
There are also morphemes that (in modern English) appear only as bound morphemes in compounds, but, unlike cranberry morphemes, appear in several compounds each, so we can see what they mean. (16) a. cookware, dinnerware, stoneware, warehouse (also A-N compounds hardware, software, wetware) b. alleyway, doorway, pathway, raceway, breezeway, gangway (also A-N compound highway and V-N compounds driveway and runway; related to but not identical to free morpheme way as in which way did they go) c. playwright, millwright d. swingset, TV set (bound -set ¼ ‘apparatus’, not ‘set’) e. housekeeper, beekeeper, bookkeeper, doorkeeper, gatekeeper, shopkeeper, zookeeper (keep house/bees/the books but not keep a zoo/a door/a gate These families of compounds parallel the rootless pairs and triples of Chapter 2 such as retribution/retributive and aggression/aggressor/aggressive: the cost in the lexicon should reflect the common parts. Unlike cranberry morphemes, they lend themselves to creativity, as in the relatively recent coinages software, wetware, and malware.11 A common technique for analyzing compounds has been to establish a phrasal paraphrase and to assign the meaning of the paraphrase to the compound. However, it has often been noticed that it is sometimes impossible to establish a single best paraphrase for a compound. Lees 1960 (123), for instance, discusses the example pontoon bridge: . . . it is not even obvious which interpretation is the most commonly used, but the following might occur to us: [17] bridge supported by pontoons bridge floating on pontoons bridge made of pontoons pontoons in the form of a bridge
11
(like steamboat) (like seaplane) (like blockhouse) (like cell block)
Another minor peculiarity is compounds in which a word has evidently been truncated. The clearest example comes from the names of sports equipment that incorporate the name of the game, e.g. soccer ball, lacrosse ball, pingpong ball, baseball bat, football helmet, pinochle deck. By analogy, the ball used in baseball ought to be a baseball ball, not a baseball, and similarly for football, volleyball, and racquetball. A random collection of less systematic cases, with the possible truncated part in brackets: roller[skate] blades, jailbreak[out], cocktail [party] dress, string/woodwind [instrument] ensemble/quartet/player, bank[note] roll, tap/toe [dance] shoe, folk [song/music] singer. Jespersen 1942 suggests also news[paper]boy and waste[paper]basket.
the ecology of english n o u n - n o u n c o m p o u n d s
427
Gleitman and Gleitman (1970, 95) make similar remarks: ‘We suspect that the person who says lion-house would consider it rather odd if someone asked him: ‘‘Did you mean a house for a lion, a house suitable for lions, or a house lions can live in?’’ Obviously the speaker meant any of these indifferently.’ Levi 1978 also discusses this issue at length. This problem presents three different cases. First, the paraphrases may be pure semantic variants, as in Gleitman and Gleitman’s ‘house for a lion’ and ‘house suitable for a lion’, or felafel ball ¼ ‘ball made of/created from felafel’. Here, a proper semantic analysis should abstract F(X1,Y2) away from the particular way it happens to be expressed in the paraphrase. Slightly more complex multiple paraphrases arise in cases like (18). (18)
ticket window ¼ ‘a window at which tickets are bought/at which tickets are sold’ toll booth ¼ ‘a booth at which tolls are paid/at which tolls are collected’ movie theater ¼ ‘a theater where movies are seen/at which movies are shown’ grammar school ¼ ‘a school where grammar is taught/at which grammar is learned or studied’
Here the multiplicity arises from different ways of orienting the same connection between N1 and N2. Following ideas of Langacker 1987a and Fillmore and Atkins 1992, the two paraphrases really express the same conceptualized ‘event schema’, changing only the perspective. Buy places the initiative with the recipient of goods, sell with the original owner; in a real transaction of course both must play an active role. Similarly with the payment and collection of tolls, the seeing and showing of movies, and the teaching and the learning of grammar. The solution, then, is that F(X1,Y2) likely consists of the simple event schema, bereft of perspective or focus. The paraphrases, however, create alternative perspectives, because the overt use of a verb forces us to choose a particular frame of thematic roles. The following examples, like pontoon bridge, present a third situation. (19)
box car ¼ ‘car that carries boxes/that resembles a box/that serves as a box’ refrigerator car ¼ ‘car that has a refrigerator as significant part/that serves as a refrigerator’ elevator shaft ¼ ‘shaft that an elevator travels in/that is part of an elevator’ file folder ¼ ‘folder in which one places a file/that forms part of a file’
428
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
Unlike (18), these cases are not lexical or perspectival variants of the same relation. Resembling a box and serving as a box are quite distinct relations.12 Lees, Levi, and the Gleitmans suggest that there is no fact of the matter about which paraphrase is correct. Someone learning these words is typically given no evidence (e.g. ‘This is called a box car because it looks like a box’): the relation is normally taken to be self-evident and without need of explanation (‘That one is called a boxcar’). Boxcar is not ambiguous: it picks out the same objects no matter which reading is assigned to it. It is not like football the game vs. football the ball. Nor is boxcar vague: it does not leave open a continuous range of possibilities on a scale, the way, say, cold does. This being the case, what can the learner do? I propose that there is another way that boxcar can have multiple meanings besides being ambiguous or vague: it can have all the meanings in (19) simultaneously, in cooperation rather than competition. We might call such a word promiscuous (by contrast with ambiguous). If such a solution seems like giving up, one must remember that a word meaning is an entity in a brain, not in a logical system. It is altogether in the style of the brain to arrive at multiple solutions to the same result, and for a result so obtained to be more stable in perception and memory. A standard example of this is the system for depth perception, which relies on partially redundant evidence from sensory cues (e.g. lens accommodation), perceptual cues—both monocular (e.g. occlusion) and binocular (e.g. stereopsis)—and cognitive cues (e.g. knowing what size familiar objects should be). These all converge on a single perceptual feature: the absolute and relative distances of objects in the visual field. The claim, then, is that pontoon bridge and boxcar are promiscuous rather than ambiguous. A learner attempts all possible strategies for combining N1 and N2 (presumably in parallel), and since there are multiple satisfactory strategies that do not conflict, all such semantic combinations are stored in memory as part of the meaning of the compound. (Of course, if asked to define the compound, speakers will likely give only one of the combinations and be satisfied with that.) A linguist seeking to analyze these compounds faces the same problem as the learner. The insistence on a single best solution is only a prejudice, which, I admit, is well grounded in scientific and commonsense practice. But in dealing 12
Some of these examples play on an ambiguity in one of the constituent nouns. In elevator shaft, is the elevator the box that one rides in, or is it the whole piece of machinery including motor, cables, pulleys, controls, and the shaft? File folder plays on an ambiguity in file: in the first reading, the file is an individual piece in a collection of information; in the second, the file is the physical instantiation of the entire collection. Another possible example is lipstick: is it something shaped like a stick that one puts on one’s lips, or is it something that one sticks to one’s lips? Here the relevant readings of stick are totally unrelated.
the ecology of english n o u n - n o u n c o m p o u n d s
429
with brain processes it is, I believe, sometimes counterproductive and should be judiciously abandoned when inappropriate.
13.4 Aspects of compound meaning that come from semantics of nominals The semantics of compounding involves a number of distinct components. This section sketches three components implicated in the combinatorial semantics of nouns in general; the next two sections add components that are specifically involved in compounding.
13.4.1 Profiling The first general component might be called profiling (roughly following the usage of Langacker 1987a; Brekle 1975 calls it topicalization): picking out a character in an event and designating this character as the one being referred to. For instance, the action of driving involves an agent directing the motion of a vehicle; the nominal driver picks out the agent. A standard way to notate this is through lambda-abstraction, which binds an argument within an expression to a variable outside (20b). For my purposes a slightly different notation for profiling proves useful. In (20c), the head of the expression is PERSON and the expression after the semicolon is a modifier. What makes the expression a well-formed modifier is that it contains a variable a which is bound by the superscript on PERSON. Profiling an argument of a function, then, consists in binding it to something outside the function; this is the semantic counterpart of a relative clause in syntax. (20)
a. DRIVE (A, B) ¼ ‘A drives B’ b. lx[DRIVE (x, INDEF)] ¼ ‘individual who drives something’ c. [PERSONa; [DRIVE (a, INDEF)] ] ¼ ‘a person a such that a drives something’
Any argument can be profiled; for instance the distinction between employer and employee is shown in (21a, b). The nominal orientation profiles the Path argument of the function ORIENT, yielding the direction in which the theme is oriented (21c). (21)
a. employer: [PERSONa; [EMPLOY (a, INDEF)] ] b. employee: [PERSONa; [EMPLOY (INDEF, a)] ] c. orientation: [PATHa; [ORIENT (X, a)] ]
More generally, the distinction between ‘process’ and ‘result’ nominals is that the latter has profiled the theme argument, as in (22).
430
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
(22) a. the composition of the song (process nominal): [Event COMPOSE (X, SONG)] b. John’s new composition (result nominal): [MUSICa; NEW; COMPOSE (JOHN, a)] As noted in previous chapters, the Parallel Architecture notates the relation between syntactic and semantic constituency in terms of coindexing. Thus (20a) and (20c) can be notated more precisely as (23a) and (23b) respectively, and the productive use of the -er suffix can be encoded as the schemas (23c, d), where F is a variable function of some unspecified number of variables.13 ‘¼’ now stands for the interface relation between syntax and semantics. (23) a. A1 drives2 B3 ¼ [DRIVE2 (A1, B3)] b. drive1-er2 ¼ [PERSON2a; [DRIVE1 (a, INDEF)] ] c. V1-er2 ¼ [PERSON2a; [F1 (a, . . . )] ] (agentive -er) d. V1-er2 ¼ [OBJECT2a; [F1 (INDEF, . . . ); WITH a)] ] (instrumental -er) The semantic structure of (23b) also appears in morphologically different nominals such as (24a), as well as in compounds such as (24b). The differences among them show up in the coindexation between the morphosyntax and the semantic structure. (24) a. violin1-ist2 ¼ [PERSON2a; [PLAY (a, VIOLIN1)] ] b. violin1 play2-er3 ¼ [PERSON3a; [PLAY2 (a, VIOLIN1)] ]
13.4.2 Action modality Busa 1997 develops an analysis of agentive nominals—nouns that denote characters individuated by their actions. She points out, for example, that although violinist denotes someone who plays the violin, it is actually ambiguous between an occupation (25a), a habitual activity (25b), or an ability (25c). It can even be used in a case where playing the violin is a specific activity on a specific occasion (i.e. a stage-level predicate). For instance, (25d) might be used if all the players in the orchestra are rank beginners, or if the players have switched instruments as a joke (a situation I actually experienced once). All this is unchanged, of course, if we substitute the compound violin player. 13 This formalization is not too different from many others in the literature, for example Rappaport Hovav and Levin 1992, Lieber 2004, and Booij 2007. I omit the details for specifying that X is an agent; see Jackendoff 1990. I have oversimplified the rest of the semantics of -er and -ee nominals, as not especially relevant to the present discussion.
the ecology of english n o u n - n o u n c o m p o u n d s (25)
431
a. She’s a violinist in the Pittsburgh Symphony but hasn’t played since they went on strike. b. She’s an occasional violinist. c. She’s a good violinist, but hasn’t played since she sold her violin ten years ago. d. None of the violinists can play the violin!
I’ll call these variant interpretations the action modalities under which a nominal can be understood. (25) might suggest that the choice of action modality is just a matter of pragmatics. But there are action nominals whose action modality is an essential part of the lexical meaning (and this is Busa’s main concern). For instance, pedestrian is a stage-level predicate: someone on foot on a particular occasion. I don’t remain a pedestrian when I’m driving my car. Similarly, passengers are individuated by their trips: when one counts passengers carried by American Airlines, I count as a different passenger on each trip. By contrast, someone who only happens to discuss economics on a particular occasion is unlikely to be called an economist (except perhaps sarcastically); being an economist is an occupation. A customer may be either current (stage-level) or habitual; for the occupation, the term is buyer. A convict is someone who has been convicted of a crime—once. And the difference between a whore and a slut is whether the action in question is an occupation or a habit. Among compounds, some (e.g. milkman, garbage man, mailman) lexically denote occupations; others (fisherman, bartender, violin player) are more open in their action modality. Novel coinages in particular may often be understood as stage-level, with specific function and specific action. For instance, Downing’s (1977) bike girl, ‘girl who left her bike in the hallway on this particular occasion’, is of this sort, parallel to pedestrian. In the context of a recipe, starch bowl, ‘bowl currently containing starch’, has this same action modality, which I’ll call ‘current’. An important action modality is Ruth Millikan’s (1984) notion of proper function.14 Roughly, ‘[h]aving a proper function is a matter of having been ‘‘designed to’’ or of being ‘‘supposed to’’ (impersonal) perform a certain function’ (Millikan 1984, 17). Crucially, an object need not actually ever perform its proper function. Millikan’s striking example is a sperm: only one of millions ever performs its proper function of fertilizing an egg.
14
Millikan develops the notion of proper function in the context of a theory of language that I find difficult to endorse in general. Nevertheless this notion, extracted from her overall approach, is of great utility in a mentalistic analysis of concepts.
432
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
Three major classes of things can have proper functions. The first class is artifacts: concrete objects constructed by people15 who have some function in mind for them, or who benefit from their functioning. The second class is parts. For parts of artifacts, such as the back of a chair, the proper function is clear: the part serves part of the proper function of the artifact. But parts of organisms also have proper functions: the heart is to pump blood, the leaves of a plant are to perform photosynthesis, and so on. A third class of objects with proper functions is objects that are ‘destined’ to become something: the proper function of a seed is to become a plant, of an egg to become an animal, and of a fiance´e to become a wife—whether or not these situations actually come to pass. Action modality will be formalized as an operator on a profiled action. So, for instance, the occupation reading of violinist can be notated as (26a), and the ‘current’ reading of starch bowl as (26b). For a noun that denotes an artifact such as book, the proper function is part of its lexical entry, as in (26c). (26) a. violin1ist2 ¼ [PERSON2a; [OCC (PLAY (a, VIOLIN1) )] ] b. starch1 bowl2 ¼ [BOWL2a; [CURRENT (CONTAIN (a, STARCH1) )] ] c. book1 ¼ [BOOKa; [PF (READ (PERSON, a) )] ]1 It is an interesting empirical question what the full range of action modalities is.16
13.4.3 Cocomposition An important way in which natural language semantic composition goes beyond simple Fregean compositionality is cocomposition, first explored extensively by Pustejovsky 1995. The best-known example involves the complement of verbs such as enjoy, which semantically must be an activity. We enjoyed singing and We enjoyed the work undergo ordinary composition, as their 15 Or other intention-having beings, since it makes sense to include beaver dams among artifacts. 16 It is important to distinguish proper function from what might be called generic or habitual function. In order to do so, we have to distinguish two senses of ‘generic’. Because compounds, like all common nouns, denote a type, any characteristic action ascribed to them will necessarily be generic over the members of the type. But this does not mean that the action ascribed to a single individual is itself generic. For instance, sperms in general have a certain proper function; but an individual sperm fulfills its proper function at most once. By contrast, an individual table may fulfill its proper function many times. Thus a sperm’s proper function is a token action, but a table’s is a generic action. Nor should proper function be conflated with habitual action or characteristic activity. Building dams is a characteristic activity of beavers, but it is not the proper function of beavers. It is, however, a generic activity: a beaver keeps doing it over and over again. But another characteristic activity of beavers, also not its proper function, is dying, and a beaver only does this once. These examples suggest that action modality may be further differentiated into features, such that the distinction of proper function vs. characteristic action crosscuts the distinction of generic vs. one-time action.
the ecology of english n o u n - n o u n c o m p o u n d s
433
complements are activities. However, the complements in we enjoyed the book and we enjoyed the beer do not denote activities. Nevertheless, their interpretations do incorporate an activity, most probably ‘reading the book’ and ‘drinking the beer’ (other possibilities depend on context). Crucially, the default activity is dependent on the choice of noun. Where does this extra piece of meaning come from? The obvious source is the internal structure of the noun’s meaning, in particular from the noun’s proper function. Let me formalize just enough of this to make it useful in the analysis of compounds. (27a) is what would result from composing enjoy and book in simple Fregean fashion; it is ill formed because a book is not a kind of activity. (27b) is a slightly more complex but well-formed expression; the unspecified function F serves as a sort of ‘adapter plug’ ‘coerced’ into the interpretation, so that all selectional restrictions can be met (Jackendoff 1997b, chapter 3). The first argument of F, the actor, is bound to BILL by the as; this is the semantic expression of control (Culicover and Jackendoff 2005, chapter 12), so that it is Bill who is performing the action F. (27) a. Bill1 enjoyed2 the book3 ¼ [ENJOY2 (BILL1, [Activity BOOK3])] b. Bill1 enjoyed2 the book3 ¼ [ENJOY2 (BILL1a, [Activity F (a, BOOK3)])] ‘Bill enjoyed doing something (F-ing) with the book’ The content of the coerced function F is filled out by incorporating material from the proper function of book. (28a) makes this proper function explicit, following the analysis of book in (26c). It is now possible to fill out F by unifying into it a copy of the function from book’s proper function. This will be notated as in (28b): the copy is in italics, and the co-superscripts g show what it is a copy of. (28) a. [ENJOY2 (BILL1a, [Activity F (a, [BOOKb; PF (READ (PERSON, b) )]3)])] b. [ENJOY2 (BILL1a, [ActivityREADg (a, [BOOKb; PF (READg (PERSON, b) )]3)])] One can imagine other formal ways of working this out; Pustejovsky’s own formalization is quite different. But the general idea is clear. First, when pieces of meaning that are expressed cannot link up semantically, it is sometimes possible to add unspoken functions in order to create well-formed semantic connections (coercion). Second, it is possible to fill out these functions by copying them from inside the meanings of nouns (cocomposition).17 17
Lascarides and Copestake 1998 work out a version of this type of cocomposition within a somewhat more standard HPSG/formal semantics treatment. In particular, they propose general principles for filling out control relations like that in (27)–(28) and the many more complex cases below. [continued]
434
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
13.5 Semantic structure of (relatively) simple compounds 13.5.1 The compounding schemata We now return to compounds proper. Determining the conceptual structure of a compound N1N2 involves two factors: establishing the semantic relation between N1 and N2, and designating a head—in English, N2. There are two routes for connecting N1 to N2. First, N1 can be an argument of N2, as in violin player. These are so-called synthetic compounds; the general schema appears in (29a), where X and Y are the meanings of N1 and N2 respectively. Second, N1 and N2 can both be arguments of another function F, as in (29b). (29) a. [Y2 ( . . . , X1, . . . )] b. [F ( . . . , X1, . . . , Y2, . . . )] The fact that N2 must be head of a compound can be formalized as (30). Note that this assigns no role to N1. (30) Head Principle [N1 N2] ¼ [Y2 ( . . . ); ( . . . )] There are now two ways that the meaning of N1 can be combined with N2. First, if N1 is an argument of N2, (29a) can be unified directly with (30) to yield (31a). Alternatively, if N1 and N2 are arguments of a function F, then Y2 has to be profiled in (29b), so that it can serve as head of the compound. In this case F and its arguments and modifiers serve as a modifier of Y. This yields (31b) as the general schema for this type of compound. (31) N-N compound schemata (or constructions) a. Argument schema: [N1 N2] ¼ [Y2 ( . . . , X1, . . . )] ‘a N2 by/of/ . . . N1’ b. Modifier schema: [N1 N2] ¼ [Y2a; [F ( . . . , X1, . . . , a, . . . )] ] ‘an N2 such that F is true of N1 and N2’ What is the range of possibilities for F? Many accounts in the literature, for instance Downing 1977, Selkirk 1982, Ryder 1994, and Lieber 2004, have despaired at finding a systematic account of the possibilities. Jespersen (1942,
A fine point on the binding of READ in (28): This differs from the usual binding relation, such as the as and bs in (28a, b). The latter indicate identity of reference: the same person is both enjoying and reading. In the notation, the binding variable is superscripted to the referential constituent, and the individual being bound is simply indicated by a bound variable. The binding of READ, however, is identity of sense (parallel to ones-anaphora with NPs): the act of Bill’s reading this book cannot be the same token action as the generic act of reading a book to fulfill its proper function, and it is subject to different semantic constraints. This is notated with co-superscripts on the function itself.
the ecology of english n o u n - n o u n c o m p o u n d s
435
137–8) says: ‘Compounds . . . say nothing of the way in which the relation is to be understood. That must be inferred from the context or otherwise. . . . The analysis of the possible sense-relations can never be exhaustive.’ Other accounts such as Lees 1960 and Levi 1978, noting that F is not entirely arbitrary, have attempted to enumerate a set of functions that accounts for all compounds, either in semantic or syntactic terms (e.g. for Lees, in terms of a set of deletable verbs in the underlying forms of compounds). In the present account there is a generative system that creates an unlimited set of possibilities for F. This generative system includes: .
.
. .
A family of basic functions or relations, many of which can be profiled on either variable The set of action modalities, which are applied to the function F to provide further possibilities Cocomposition of aspects of noun meaning with the function F A set of structural principles that can be combined to build structurally more complex realizations of F
Section 13.4.2 dealt with action modalities. This section deals with the basic functions; section 13.6 with cocomposition; section 13.7 with the structural principles.
13.5.2 Reversibility of basic functions First I must discuss the reversibility of the basic functions, which has not to my knowledge received much notice in the literature, but which proves very important in appreciating the protogrammatical character of compounding. Consider the pair beef stew and stew beef. In semantic structure, they both must encode that the stew is made of beef, that is, the same semantic relation obtains between the nouns in both compounds. The difference lies in profiling and action modality: beef stew is telling us the makeup of this kind of stew, and stew beef is telling us the proper function of this kind of beef. Another example is lunch bag, ‘bag meant to carry a lunch in’, and bag lunch, ‘lunch meant to be carried in a bag’. (32) illustrates. Both meanings are instances of the Modifier schema (31b), and both have the same modifier, just profiled differently. (32)
a. beef1 stew2 ¼ [STEW2a; [MADE-FROM (a, BEEF1)] ] b. stew1 beef2 ¼ [BEEF2a; [PF (MADE-FROM (STEW1, a) )] ]
A slightly more complicated case is the pair helicopter attack and attack helicopter. In semantic structure, they both must encode that the helicopter is attacking. In the former case, attack takes an argument. Hence the meaning of the compound, (33a), is an instance of the Argument schema (31a).
436
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
However, consider what happens in attack helicopter, where helicopter serves as N2. Like beef, helicopter does not take an argument, so the Argument schema cannot be invoked. Rather, attack must be part of a modifier. Within the Modifier schema (31b), ATTACK does not fit in as an argument of F. Rather, it is cocomposed with F, and then HELICOPTER can be bound to its argument, resulting in (33b). ( (33b) also says the modifier is a proper function.) (33) a. helicopter1 attack2 ¼ [ATTACK2 (HELICOPTER1, INDEF)] ‘an attack on something by helicopter(s)’ b. attack1 helicopter2 ¼ [HELICOPTER2a; [PF (ATTACK1 (a, INDEF) )] ] ‘a helicopter whose proper function is to attack things’
13.5.3 Fourteen basic functions Here is a list of the (most prominent) basic functions that can fill out F in English noun-noun compounds. Please bear with me through this possibly tedious enumeration. First, in the loosest possible relation, CLASSIFY (X, Y), the meaning of N1 plays only a classificatory role. (34) [Y2a; [CLASSIFY (X1, (a)] ], ‘N1 classifies N2’: beta cell, X-ray, Leyden jar, Molotov cocktail The second case is the argument schema Y(X) (35). (35) [Y2(X1)], ‘N2 of/by N1’: wardrobe color, food surplus, sea level, union member, wavelength, hairstyle, helicopter attack, tooth decay, speed limit, birth order, birth rate, collar size, particle shape, crystal structure, bandwidth, used-car prices, onion smell, attention span, German grammar, ship model Notice that the noun expressing a function (N2 in (35) ) has an inherent semantic argument. Because the other noun satisfies this argument position, the argument is no longer available for phrasal expression, e.g. the wardrobe color of her clothes, the food surplus of potatoes. On the other hand, if the argument noun (N1) itself takes an argument, the compound can inherit this argument, as illustrated in (36).18
18
In interpreting a novel compound, say John’s elbow strength, one must therefore compute a composite argument structure of the type in (36) online. For this to be possible, the grammar must provide more than one system for checking and satisfying theta-roles (or saturating arguments). One is the usual system for composing phrasal meanings, and one is the system of composing compound meanings—which is partly pragmatic.
the ecology of english n o u n - n o u n c o m p o u n d s (36)
437
a. the wavelength of the light ¼ the length of [a wave of the light] b. Bill’s hairstyle ¼ the style of [Bill’s hair]
Within the class with this relation we also find the vast subclass in which N2 is morphologically composed of V þ er, for instance bus driver and screwdriver, or is a zero derivative of a verb, like supply. As seen above, the semantics of V þ er nominals (and some zero nominals) is precisely ‘someone/something that V’s’; that is, the noun is essentially nothing but a profiled agent or instrumental argument of the verb. The exact morphology is irrelevant: for instance, screwdriver and paperclip are parallel in all relevant respects. (37)
a. [N1 [N V3-er]2] ¼ [Z2a; [Y3 (a, X1)] ], ‘someone who V3’s N1’: woodcarver, junk dealer, hairdresser, dogcatcher, gravedigger, bus driver, cheerleader, store manager, gym teacher, bartender, bookseller b. [N1 [N V3]2] ¼ [Z2a; [Y3 (a, X1)] ], ‘someone who V3’s N1’: life guard, cowherd, fighter pilot, talk-show host c. [N1 [N V3-er]2] ¼ [Z2a; [Y3 (INDEF, X1, WITH a)] ], ‘something that someone V3’s N1 with’: hairdryer, windbreaker, aircraft carrier, snowblower, flycatcher, carpet cleaner, hedge clipper, metal detector, screwdriver, fire extinguisher, bird feeder, coffee grinder d. [N1 [N V3]2] ¼ [Z2a; [Y3 (INDEF, X1, WITH a)] ], ‘something that someone V3’s N1 with’: power supply, doorstop, wine press, hair dye, noise filter, bookmark, stomach pump, wing support, mouth/eyewash, nail/tooth brush, paperclip, bear/booby/fly/ion/mouse trap, nail/shoe polish, neck brace, ear plug, chicken feed, nail file, toothpick, dog whistle, hearing aid
The argument relation is sometimes reversible (X(Y)), with the extra structure shown in (38a, b). Usually N1 is a zero- or -ing derivative of a verb, so considerations parallel to those in (37) apply. (38)
a. [ [N V3(-ing)]1 N2] ¼ [Y2a; [X3 (a, . . . )] ], ‘N2 that V3’s (things)’ attack helicopter, curling iron, guard dog b. [ [N V3(-ing)]1 N2] ¼ [Y2a; [X3 (INDEF, a)] ], ‘N2 that people V3’ chewing gum, drinking water, cooking apple
The third basic relation is BE (Y, X), ‘Y is (also) an X’, which yields dvandva compounds (39a). As pointed out by Olsen 2001, there are a couple of variants of this function. One denotes objects that are some sort of mixture, lying on the boundary between the two categories (39b); another variant denotes an object composed of both N1 and N2 (39c).
438
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
(39) a. [Y2a; [BE (a, X1)] ], ‘N2 that is an N1’: boy king, politician-tycoon, maiden aunt, woman doctor, child prodigy, poet-painter, junk mail, waste paper, compound noun, torah scroll, boy toy, washerwoman, fisherman, driver ant, killer shark/bee b. witch doctor, pantyhose, prose poem, sweater vest, apeman, man-god c. tractor-trailer, Alsace-Lorraine The fourth basic function is SIMILAR (X, Y). (40) [Y2a; [SAME/SIMILAR (a, X1)] ], ‘an N2 similar to N1’: piggy bank, string bean, sunflower, kidney bean, I beam, hairpin bend, marble cake, rock candy, pie chart, fiddler crab, animal cracker, tree diagram, bulldog, baby doll, kettledrum, catfish, starfish, zebrafish, sunflower, dragonfly, star fruit, elbow macaroni, Batman, tissue paper, T-shirt, garter snake, hairspring, sandstone, angleworm, ringworm, tapeworm This function is not reversible, because it is symmetric. Similarity can only be rendered asymmetric due to profiling, most commonly by making one argument subject (Gleitman et al. 1996). For instance, Chicago and Akron are similar is symmetric, but Chicago is similar to Akron and Akron is similar to Chicago are not entirely synonymous due to the profiling. It is thus impossible to construct a compound fish zebra, ‘fish such that a zebra is similar to it’. Such a compound would have conflicted profiling: zebra is profiled because it is subject, but fish is profiled because it is relativized. The fifth basic function involves a relation among kinds, KIND (X,Y). It is reversible. (41) a. [Y2a; [KIND (a, X1)] ], ‘an N2 of kind N1’: puppy dog, ferryboat, limestone, pine tree, gemstone, girl child, alleyway, pathway b. [Y2a; [KIND (X1, a)] ], ‘an N2 that is a kind of N1’: seal pup, bear cub (there are other possible analyses as well, perhaps promiscuously) The sixth function involves location, BE (X, AT/IN/ON Y). It is reversible (42a, b). A special case is temporal location, ‘while’ or ‘during’ (42c). (42) a. [Y2a; [BE (a, AT/IN/ON X1)] ], ‘N2 that is located at/in/on N1’: sunspot, window seat, lake dwelling, nose hair, donut hole, basement apartment, tree house, background music, focal-plane shutter, blood sugar, groundwater, earwax, brain tumor, back/ear/head/neck/toothache, leg cramp
the ecology of english n o u n - n o u n c o m p o u n d s
439
b. [Y2a; [BE (X1, AT/IN/ON a)] ], ‘N2 with N1 at/in/on it’: raincloud, garlic bread, inkpad, stairwell, icewater, water bed, beanbag, sandbag, teabag, sandbox, wax paper, bubble bath, ice pack, theater district, sandpaper19 c. [Y2a; [BEtemp (a, AT X1)] ], ‘N2 that takes place at time N1’: spring rain, morning swim, 3 a.m. blues This class of compounds also includes many where location is involved in the proper function of the object— (43)
a. [Y2a; PF([BE (a, AT/IN/ON X1)])], ‘N2 whose proper function is to be at/in/on N1’: door mat, gravestone, street light, kitchen sink, hair ribbon, bathroom scale, urinal cake, shoelace, pocket knife, bag lunch, pocket watch, desk calendar, lawn chair, tablecloth, farmhouse, wallpaper, ceiling fan, coffee-table book, seat cushion b. [Y2a; PF([BE (X1, AT/IN/ON a)])], ‘N2 whose proper function is to have N1 at/in/on it’: steam room, boiler room, oyster/snail/tortoise shell, hot-air balloon, lamp post, doghouse, birdcage, henhouse, beehive, insane asylum, cow shed, pigpen
some where the location is characteristic rather than a proper function— (44)
a. [Y2a; CHAR ([BE (a, AT/IN/ON X1)])], ‘N2 characteristically at/in/ on N1’: seashell, house plant, housefly, seabird, water buffalo, bedbug, caveman, fieldmouse, earthworm, fruitfly, farm boy b. [Y2a; CHAR ([BE (X1, AT/IN/ON a)])], ‘N2 with N1 characteristically at/in/on it’: bear country, duck pond, Indian territory
and some where the thing being located is information. (45)
[Y2a; PF ([BE (X1, AT/IN/ON a)])], ‘N2 whose proper function is to have N1 in/on it’: address book, notebook, notepad, graph paper, order blank, index card
Several further elaborations of this case will appear in section 13.6.
19 These cases verge closely on ‘X with Y as a part’, below. It is not clear to me whether they are distinct.
440
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
The seventh function is COMP (X,Y), ‘X is composed of Y’. This is one of the basic ‘parts and boundaries’ functions discussed in Chapter 5. It is reversible.20 (46) a. [Y2a; [COMP (a, X1)] ], ‘N2 composed of N1’: felafel ball, rubber band, rag doll, tinfoil, brass instrument, jellybean, inkblot, corkboard, card catalog, wood chip, crop circle, cloud cover, sugar cube, steel drum, dungheap, bearskin rug, ice sculpture, soap scum, bloodstain, case tier, scar tissue b. [Y2a; [COMP (X1, a)] ], ‘N2 that N1 is composed of’: wallboard, bathwater, brick cheese, sheet metal, book matches, plate glass, pack ice, sheet ice, head lettuce, loaf sugar, lump sugar, brick tea, leaf tobacco, foam rubber The eighth function is ‘X is made out of Y’, MADE (X, FROM Y). This differs from COMP (X, Y) in that in this case the object or substance Y is no longer in evidence. For instance, one can still find the onions in an onion roll (COMP), but one can no longer find the olives in olive oil (MADE FROM). The distinction is however slippery. This function too is reversible.
20
The compounds in (46a) have characteristic compound stress on N1. Another class with the same meaning relation has stress on N2. (i)
N2 composed of N1, N2 stressed cardboard box, tin can, fur coat, liquid detergent, aluminum foil, leather jacket, glass jar, ceramic mug, wool sweater, oak table, paper towel, brick wall
As observed by Jespersen 1942, these have a bit more syntactic freedom than standard compounds: (ii)
a.
The box is cardboard. It’s a cardboard box. ?The ball is snow. It’s a snowball. b. a part/mostly cardboard box a part/mostly snowball c. a wood(en) and cardboard box an ice and snowball d. a cardboard box and a wood(en) one a snowball and an ice one
These differences suggest that the examples in (i) are not compound nouns but rather nouns preceded by a syntactic modifier. This is a bit curious, in that tin cán and aluminum fóil fall under the ‘false compounds’ in (i), but tínfoil falls under the ‘true compounds’ in (46a). That’s just how the lexicon is, I guess. In any event, just because a ‘false compound’ is phrasal, this does not mean it has to be constructed online. It is as easy to store in the lexicon as a ‘true compound’. After all, we do store lots of phrasal collocations in memory, all the way from salt and pepper to When in the course of human events . . . So the type in (i), although it may be structurally different from compounds, need not necessarily be radically different in its status in the lexicon.
the ecology of english n o u n - n o u n c o m p o u n d s (47)
441
a. [Y2a; [MADE (a, FROM X1)] ], ‘N2 made from N1’: apple juice, olive oil, grain alcohol, cane sugar, cornstarch, tomato paste, bean curd, goose grease, petroleum jelly, coal-tar product, maple syrup, elderberry wine b. [Y2a; [MAKE (X1, FROM a)] ], ‘N2 that N1 is made from’: sugar beet, rubber tree
Another of the functions of Chapter 5 is PART (X,Y), ‘X is a part of Y.’ Again it is reversible, with two variant paraphrases (48b, c), depending on whether the part is count or mass.21 (48)
a. [Y2a; [PART (a, X1)] ], ‘N2 that is part of N1’: backbone, whalebone, cigarette butt, suit coat, oar handle, apple core, doorknob, fingertip, computer screen, bicycle seat, pigskin, bedspring, ticket stub, tree trunk, bra strap, razor blade, shoelace, stovetop, mold cavity, stew beef, cake flour, lunch meat b. [Y2a; [PART (X1, a)] ], ‘N2 that has N1 (count) as a part’: snare drum, lungfish, string instrument, wheelchair, rattlesnake, fur seal c. [Y2a; [PART (X1, a)] ], ‘N2 that is composed in part of N1’: gingerbread, cinnamon bun, cheesecake, noodle soup, dill pickle, jelly roll
The tenth function is (CAUSE X,Y), ‘X causes Y.’ (49)
[Y2a; [CAUSE (X1, a)] ], N2 that is caused by N1’: sunburn, diaper rash, knife wound, surface drag
A closely related function is MAKE (X,Y), ‘X makes Y.’ It is reversible. (50)
a. [Y2a; [MAKE (X1, a)] ], ‘N2 made by N1’: moonbeam, anthill, foot/fingerprint, horse shit, gopher hole, snake poison, suntan, bullet hole, knife wound, beeswax b. [Y2a; [MAKE (a, X1)] ], ‘N2 that makes N1’: honeybee, lightbulb, musk deer, textile mill, lighthouse, silkworm, songbird, candy factory, sweat gland, polio virus
It is sometimes hard to distinguish MAKE from CAUSE. Perhaps MAKE (X,Y) decomposes as CAUSE (X, (COME INTO EXISTENCE (Y)). 21 The difference between COMP and PART can be illustrated by the ambiguity of clarinet quartet. On the COMP reading it means ‘quartet of four clarinets’; on the PART reading it means ‘quartet of which a clarinet is a distinctive member’, e.g. a clarinet and three strings. I note that PART is formalized here differently than in Chapter 5. In particular, the inverse cases in (48b, c) would be formalized in Chapter 5 using CONT (‘containing’), the inverse function of PART. Further work is needed to reconcile the two treatments.
442
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
The twelfth function might be paraphrased as ‘X serves as Y.’ This can be reduced to a more basic analysis: ‘function of X is as a Y’ or even ‘function of X is to do what Y does’. (51) [Y2a; [BE (PF (a), PF(X1) )] ], ‘N2 whose (proper) function is to function as an N1’: handlebar, feature film, extension cord, farmland, buffer state, guard dog, guidebook, fighter plane, ferryboat, retainer ring, booster shot, retainer fee, i.d. card, endpoint The thirteenth function is HAVE (X,Y), ‘X has Y’, in many senses of ‘have’. It too is reversible. (52) a. [Y2a; [HAVE (a, X1)] ], ‘N2 that has (an) N1’ AIDS baby, career girl, glamour girl b. [Y2a; [HAVE (X1, a)] ], ‘N2 that N1 has’: writer’s cramp, shepherd’s dog, gangster money The fourteenth basic function is the only one that does not seem especially ‘basic’: PROTECT (X, Y, FROM Z), ‘X protects Y from Z.’ It creates two families of compounds, depending which two of its three arguments are realized in the compound. (53) a. [Y2a; [PROTECT (a, X1, FROM Z)] ], ‘N2 protects N1 from something’: chastity belt, lifeboat, safety pin, safety lock b. [Y2a; [PROTECT (a, Z, FROM X1)] ], ‘N2 protects something from N1’: mothball, flea collar, cough drop, mosquito net, sun hat, speed bump, mud flap, gas mask, lightning rod, snow shed, bug spray, rain boots, dust jacket, firescreen, headache pill, windshield, surge protector, bird sanctuary, game preserve This list of functions is not far off others that have been proposed in the literature. With the exception of PROTECT, they seem rather plausible as functions that are readily available pragmatically, i.e. to a protogrammatical capacity for building meanings. An important question is how many of these functions are available for compounding crosslinguistically, and what other functions might appear in other languages’ compounds (this question is addressed explicitly and implicitly in many of the articles in Lieber and Sˇtekauer 2009).
13.6 Using material from the meanings of N1 and N2 If this were all there were to filling out the interpretation of compounds, the number of possible relations in compounds would be fourteen, or allowing for
the ecology of english n o u n - n o u n c o m p o u n d s
443
all the variants and reversibility, somewhere in the twenties or thirties—clearly not enough. To create a larger range of relations, two other devices come into play: material from the internal semantic structure of the two nouns, and coercion of more structure to create more distant relations between N1 and N2. We take these up in turn. Consider the range of locative relations illustrated in (42)–(45). The basic relation appears to invoke only locations paraphrasable by at, in, or on—perhaps the least marked spatial relations. However, other spatial relations do in fact occur within the meanings of compounds. In most of these cases, we discover that the spatial relation in question is involved in the proper function of one or the other of the nouns, usually N2. For instance, a water fountain is a fountain that water flows out of; but the proper function of a fountain is for liquid to flow out of it. Now notice: since the proper function of N2 is a modifier, it can be used to fill out the content of F in the modifier schema (31b). (54a) shows the internal structure of fountain. Its proper function, ‘liquid flows out of’, cocomposes with F to produce the semantic structure (54b) for water fountain.22 (54)
a. fountain1 ¼ [FOUNTAIN1a; [PF (FLOW (LIQUID, OUT-OF a) )] ] b. water1 fountain2 ¼ [FOUNTAIN2a; [PF (FLOW (WATER1, OUT-OF a) )] ]
Similar cases are coal mine (‘dug out of’), gas pipe (‘flows through’), Charles River bridge (‘crosses over’), and Downing’s (1977) toe-web (‘extends between’). In all these cases, N2 has a proper function, and N1 is an argument of the proper function. This approach accounts for large families of compounds such as those in (55). (55a–d) are further sources of locative relations; (55e) involves a different sort of proper function. (55)
a. N2 is a container: coffee1 cup2 ¼ [CUP2a; [PF (HOLD (COFFEE1, IN a) )] ] also photo album, carbarn, soapdish, fishtank b. N2 is a vehicle: cattle1 car2 ¼ [CAR2a; [PF (CARRY (CATTLE1, IN a) )] ] also baby carriage, garbage/icecream/oil truck c. N2 is an article of clothing: pinky1 ring2 ¼ [RING2a; [PF (WEAR (INDEFb, a, ON [PINKY(b)]1) )] ]
22 Brekle 1986 calls this kind of cocomposition a stereotype compound. Bassac 2006 analyzes this process—correctly, in my opinion—in terms of Pustejovsky’s (1995) qualia structure internal to noun meanings, most prominently the telic quale, which specifies proper function of an object.
444
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s also face mask, necktie, ankle bracelet, skullcap, earring, backpack, fanny pack, wristwatch d. N2 is itself a location: liquor1 store2 ¼ [STORE2a; [PF (BUY/SELL(INDEF, LIQUOR1; IN a) )] ] also fruit market, movie theater (SHOW/SEE), law school (LEARN/TEACH) e. N2 is an incipient stage of something else: rose1bud2 ¼ [BUD2a; [PF (BECOME (a, ROSE1) )] ] also chick embryo, grass seed, dinosaur egg
This structure also accounts for cases in which N2 is agent or instrument of an action but is not derived from a verb. (The agentive and instrumental cases in (37) thus also fall redundantly under this schema.) (56) a. N2 is an agent or causer: silk1 merchant2 ¼ [MERCHANT2a; [OCC (SELL (a, SILK1) )] ] also eye doctor, pork butcher, sanitation engineer, locksmith, brick mason, car thief, Beatles fan, computer virus, rocket fuel b. N2 is an artifact: glue1 gun2 ¼ [GUN2a; [PF (SHOOT (INDEF, GLUE1, FROM a) )]] also fishnet, ant bait, bread/butter/steak knife, peppermill, handsoap, snow shovel, grapefruit spoon Another such case arises when N2 denotes an information-bearing item such as a song, in which case N1 can describe the topic of the information (what the information is about). (57) love1 song2 ¼ [SONG2a; [BE (INFORMATION (a), ABOUT LOVE1)] ] also Passion play, research paper, success story, fairy tale, horror film, newsletter, lexicalization problem, grammar lesson, cookie recipe In all the cases so far, F has been filled out from material in the lexical entry of N2. It is also possible to use material from N1. Consider cannonball, ‘ball whose proper function is to be shot from a cannon’. The notion of shooting comes not from the meaning of ball but from the proper function of cannon. This time, since SHOOT is in a modifier of N1, it is not automatically in a position where it can satisfy F, as seen in (58). (58) cannon1ball2 ¼ [BALL2a; PF (F(a, [CANNON1b; PF (SHOOT (INDEF, BALL, FROM b) )]) )]
the ecology of english n o u n - n o u n c o m p o u n d s
445
However, we can fill out F by means of cocomposition, which copies material from the proper function of a complement into an unspecified function, just as in enjoy the book (28). This yields (59).23 (59)
cannon1ball2 ¼ [BALL2a; PF (SHOOTg (INDEF, a, FROM [CANNON1b; PF (SHOOT (INDEF, BALL, FROM b) )]) )] ‘a ball whose proper function is for people to shoot it from a cannon (whose proper function is for people to shoot balls from it)’
Further cases of cocomposition from N1 will appear in the next section.
13.7 Generative schemata for F So far we have seen cases in which either N1 is an argument of N2 (helicopter attack, hair dryer, etc.) or vice versa (attack helicopter), cases in which N1 and N2 are co-arguments of a basic function (limestone, sunspot, etc.), cases in which N1 is an argument of a modifier within the lexical meaning of N2 (coffee cup, silk merchant, etc.), and cases in which the proper function of N1 serves as a modifier of N2 (cannonball). Suppose none of these possible relations makes sense. Then further options have to come into play, in which the relation is more complex. A first case involves combining two of the basic functions. A swordfish is a fish with a part that is like a sword, so it involves both PART and the SIMILAR functions. (60)
sword1fish2 ¼ [FISH2a; [PART ([Zb; SIMILAR (b, SWORD1)], a)] ] also alphabet soup
A second case involves the locative relation again. A street singer is not someone who sings streets (like a ballad singer), but someone who sings in the street. Here the basic locative relation is involved, but it is not composed with F itself, as in street sign. Rather, it is added as a modifier to F, which in this case is the occupation or characteristic activity of singing: (61)
street1 sing3-er2 ¼ [PERSON2a; [OCC/CHAR ([SING3b(a); [BE (b, IN STREET1)] ])] ] also skywriter, water skier (ON instead of IN), grasshopper (INSECT instead of PERSON)
23 Exactly how all the material SHOOT (INDEF, . . . , FROM ( . . . ) ) is copied over is an important formal detail that must await further investigation.
446
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
(61) falls under schema (62), which is a specialized version of the modifier schema. This has a second function G to be filled out, which is a modifier of the modifier F. In the case of (61), F is filled out from the meaning of N2 and G is filled out by the basic locative function.24 (62) N1 N2 ¼ [Y2a; [Fb ( . . . a . . . ); [G ( . . . X1, . . . b, . . . )] ] ] But the locative function is not the only basic function that can appear as G in (62). Another basic function is found in (63), steamboat. Here F is filled in from the lexical semantics of boat, ‘something whose proper function is to move in water’, and G is the basic function CAUSE, encoding the means by which the boat moves. (63) steam1boat2 ¼ [BOAT2a; [PF(MOVEb (a) ); [CAUSE (STEAM1, b)] ] ] ‘a boat that moves by steam causing it to do so’ Next consider barbershop. Like store and market in (55d), a shop is a place whose proper function is for goods and services to be bought and sold in it. However, a barber shop does not sell barbers (like a cheese shop), it sells what barbers do, namely cutting hair. In other words, the thing being sold is found in the proper function of N1. Hence it is necessary again to invoke cocomposition, and the resultant function linking N1 and N2 uses material from the meanings of both nouns. The general schema under which barbershop falls is (64). This time the second function, G, is an argument of the modifying function F, and instead of being filled in by a basic function, it is filled in by cocomposition from the meaning of N1. (64) [N1 N2] ¼ [Y2a; [F ( . . . , G( . . . X1 . . . ), . . . , a, . . . )] ] The meaning of barbershop then comes out as (65). F is realized as BUY, from the proper function of shop; G is realized as OCC(CUT ( . . . HAIR) ), from the action modality of barber. The fact that G arises by cocomposition is notated in italics. (65) barber1shop2 ¼ [SHOP2a; [PF (BUY (INDEFd, [CUT g ([BARBERb; OCC (CUTg (b, HAIR(INDEF) ) )]1, HAIR(d); IN a) )] ] ‘a shop in which someone buys the action of a barber cutting his hair (which is what barbers do for a living)’ also toll booth, ‘a booth at which one pays tolls (which is what one does with tolls)’. 24 In (61), the binding of b is to the event of singing, not to the singer. Similarly in (62), b is bound to the event F.
the ecology of english n o u n - n o u n c o m p o u n d s
447
For a still more complex case, consider piano bench, ‘bench on which one sits while playing the piano’. Sitting comes from the proper function of bench, and playing comes from the proper function of piano. These two functions are connected by the basic function of temporal location, ‘while’. Thus there are three independent components involved in linking N1 and N2, two of which come from what might be called ‘encyclopedic knowledge’ of pianos and benches. The general schema is (66), in which F is the part contributed by N2, G is the part contributed by cocomposition from N1, and H is a basic function that connects them. The structure of piano bench is shown in (67), where F is PF (SIT), G is PF(PLAY), and H is BETemp ( . . . AT). (66) [N1 N2] ¼ [Y2a; [Fb ( . . . , a, . . . ); [H (b, [G( . . . X1 . . . )])] ] (67) piano1 bench2 ¼ [BENCH2a; [PF (SITb (PERSON g, ON a); [BEtemp (b, AT [PLAYe (g, [PIANOz; [PF (PLAYe (PERSON, z) )] ]1)])])] ] ‘a bench on which one sits, such sitting being while one plays a piano (which is what one does with a piano)’ also bass stool, bike helmet, lobster bib, dessert wine, coffee cake A final case is exocentric compounds such as blockhead, which violate the Head Principle. These result from combining the modifier schema for compounds with a general coercion schema for metaphor: (68) says that one can refer to an object by using the name of something that resembles it. (68) Metaphor coercion N1 ¼ [Za; SIMILAR (a, X1)], ‘something that is similar to X’ Notice first that this schema is invoked in swordfish (60) to characterize the part of the fish that is like a sword. But now consider what happens if this schema is the outermost material in linking X and Y together. In this case, the semantic head of the compound corresponds to Z, i.e. it is unexpressed, in violation of the Head Principle. However, N2 is still ‘headish’ in that it is the thing that Z is being compared to. Aside from that, the meanings of N1 and N2 are incorporated in the same way as in ordinary compounds, interweaving basic functions and cocomposition. Here are five cases, each slightly different in how the constituent nouns are incorporated. Since these are exocentric compounds, the head has to be lexically stipulated. (69)
a. pig1tail2 ¼ [HAIRa; [SIMILAR (a, [TAIL2b; PART (b, PIG1)])] ] ‘hair that is similar to the tail of a pig’ b. canvas1back2 ¼ [DUCKa; [SIMILAR ([BACK2(a)], CANVAS1)] ] ‘duck whose back resembles canvas’
448
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s c. bird1brain2 ¼ [PERSONa; [SIMILAR ([BRAIN2b (a)], [BRAINb (BIRD1)])] ] ‘person whose brain is similar to that of a bird’ d. sea1 horse2 ¼ [ANIMATEa; [SIMILAR (a, HORSE2)]; [CHAR (BE (a, IN SEA1) )] ] ‘animate entity, similar to a horse, that is characteristically in the sea’ e. coat1tail2 ¼ [Za; [SIMILAR (a, TAIL2)]; [PART (a, COAT1)] ] ‘something that is similar to a tail and that is part of a coat’
13.8 Closing remarks By the time we get to the semantic structures in (59)–(69), a large proportion of the meaning is connective tissue: unexpressed basic functions, bound variables, and cocomposed functions. Nevertheless, the overall result should be clear. The semantic relation between N1 and N2 arises by coercing extra functions into the structure, either in argument or in modifier positions, and by filling these functions out either with basic functions or with internal semantic structure from N1 and N2. The generative system for compound meanings can be summed up as follows: (30) Head Principle [N1 N2] ¼ [Y2 ( . . . ); ( . . . )] (31) N-N compound schemata (or constructions) a. Argument schema: [N1 N2] ¼ [Y2 ( . . . , X1, . . . )] ‘a N2 by/of/ . . . N1’ b. Modifier schema: [N1 N2] ¼ [Y2a; [F ( . . . , X1, . . . , a, . . . )] ] ‘an N2 such that F is true of N1 and N2’ (68) Metaphor coercion N1 ¼ [Za; SIMILAR (a, X1)], ‘something that is similar to X’ (70) Auxiliary principles a. Cocomposition of F in (31b) with i. a basic function ii. a modifier of Y2 (e.g. Y2’s proper function) iii. X1 (if X1 takes arguments, as in attack helicoper) iv. a profiled modifier of X1 (e.g. X1’s proper function) b. Expansion of an argument or modifier of F as G( . . . ) i. Cocomposition of G with a basic function ii. Cocomposition of G with a profiled modifier of X1 c. Recursion on step b: Expansion of an argument or modifier of G This is fairly straightforward when there is only one coercing function F, as seen in sections 13.5–6, but complexity increases quickly with multiple
the ecology of english n o u n - n o u n c o m p o u n d s
449
coercions, as in section 13.7. The resulting number of options for the semantic relation between N1 and N2 also increases quickly, which seems consistent with the literature’s limited success at enumerating them. In the present approach, the repertoire of possible relations is created by a generative system which, aside from the rudimentary linking of N1 and N2 into the structure, is entirely within the semantics. So in a sense Jespersen is correct in saying, ‘The analysis of the possible sense-relations can never be exhaustive’—and yet it is systematic. One might object that piano bench does not feel as complex as (67). I offer three replies to such an objection. First, for decades we have been accustomed to vaunting the covert complexity of language beneath its intuitive transparency. The covert complexity for Conceptual Semantics should not be any less than for other formal theories of syntax and semantics, particularly given its aspiration to semantic explicitness. In fact, by virtue of the Parallel Architecture, it has been possible to keep all complexity out of the syntactic component—it is only the meaning that is complex. Second, for those who find any sort of formalism objectionable, the challenge is to find a notation that (a) is more perspicuous, (b) still retains all the necessary semantic distinctions, and (c) does so with a constrained set of basic relations and schemata. A third, more substantive reply is that another important system in language has properties similar to compounding: discourse. The semantic connections among sentences in discourse are by definition not expressed in syntax, which is confined to the structure of individual sentences. As is well known, the semantic connections among sentences are highly dependent on world knowledge: (71)
a. Max fell. Bill pushed him. (S2 causes/explains S1) b. Max fell. Bill helped him up. (S2 follows S1) (72) a. The car collided with the building. The headlights broke. (S2 is result of S1) b. The car collided with the building. The clutch broke. (S2 causes/ explains S1) Asher and Lascarides 2003 (from which (72) is taken) work out a formal logic of discourse connection that depends on about ten basic relations, which can link up sentences or groups of sentences in the discourse—and not always sequentially, but rather in a plethora of possible ways. (See also Clark 1996 for the hierarchical complexity of connections among sentences in conversation.) Some of the relations posited by Asher and Lascarides parallel basic relations in compounding. For instance, cause/result appears
450
t h e ec o l o g y o f e n g l i s h n o u n - n o u n c o m p o u n d s
in both systems, and their Elaboration function, where S2 specifies details of S1, can be construed as parallel to PART in the compound system: the event denoted by S2 is a part of the event denoted by S1. Asher and Lascarides call their system of rules a ‘glue logic’: it fills in plausible pieces of meaning left inexplicit by lexical and syntactic content, thereby gluing the sentences’ meanings together. The system proposed here for compounding has parallel properties, but it glues together nouns inside a compound instead of sentences. A serious consideration of discourse, then, makes it clear that only part of language understanding can be controlled by the orderly composition of lexical meanings, guided by syntactic structure. Discourse requires much wilder ‘seatof-the-pants’ composition, guided only by semantics and pragmatics. The relevance for the present enterprise, of course, is that it provides a point of comparison for analyses like those in section 13.7, where a great deal of unexpressed ‘glue’ is necessary to connect the nouns in a compound. It also further justifies the view of compounding as protolinguistic, that is, as a subsystem that is less grammatically structured than the standard phenomena investigated by syntactic theory. The questions raised by this account are the same ones that have persisted in the literature, but they can perhaps now be couched more precisely. Here are five; other researchers will surely have more. .
.
. .
.
What is the full set of basic functions, and how uniform are they crosslinguistically? To what extent are these functions generally available for pragmatics (including discourse) and nonlinguistic conceptualization? To what extent are these functions special to language—or to English? How extensive can coerced and cocomposed functions be in compounds, and to what extent does the answer differ between lexicalized and novel compounds? What other basic morphosyntactic patterns must be added to the simple N1N2 structure in (30)–(31) in order to account for the quasi-syntactic elaborations in (3) (e.g. health and welfare fund), and what is their status vis-a`-vis morphology, syntax, and protolanguage?
The first and third of these questions may be answered in part by the typological descriptions in Lieber and Sˇtekauer 2009; the fifth may be addressable in the framework of Construction Morphology laid out in Booij 2009. Other traditional questions have been answered by the present account, for example why the possible relations between N1 and N2 are so varied yet not altogether wild, precisely how the meanings of the two nouns contribute to the meaning of the
the ecology of english n o u n - n o u n c o m p o u n d s
451
compound, and how a balance can be struck between regularity and idiosyncrasy in the evaluation of the ‘cost’ of compounds. Above all, because the Parallel Architecture liberates the generative capacity of semantics from that of syntax, it has been possible to give a semantically based account of compounds that is sufficiently formal to see what is going on, while keeping the syntax as absolutely simple as it looks.
References Abeille´, A., R. D. Borsley, and M.-T. Espinal. 2006. The syntax of comparative correlatives in French and Spanish. In S. Mu¨ller (ed.), Proceedings of the HPSG06 Conference. Stanford, Calif.: SCLI Publications, 6–26. Ackerman, Farrell, and Irina Nikolaeva. 2004. Comparative Grammar and Grammatical Theory: A Construction-Based Study of Morphosyntax: Stanford Monographs in Linguistics. Stanford, Calif.: Center for the Study of Language and Information. Aikhenvald, Alexandra Y., and R. M. W. Dixon (eds.). 2007. Serial Verb Constructions: A Cross-Linguistic Typology. Oxford: Oxford University Press. Alboiu, Gabriela. 2000. The features of movement in Romanian. PhD dissertation, University of Manitoba. Alderete, John, Jill Beckman, Laura Benua, Amalia Gnanadesikan, John McCarthy, and Suzanne Urbanczyk. 1999. Reduplication with fixed segmentism. Linguistic Inquiry 30, 327–64. Anderson, Stephen. 1971. On the role of deep structure in semantic interpretation. Foundations of Language 7, 387–96. Anshen, Frank, and Mark Aronoff. 1988. Producing morphologically complex words. Linguistics 26, 641–55. Asaka, T. 2002. A lexical licensing analysis of the adjectival noun construction. English Linguistics 19, 113–41. Asher, Nicholas, and Alex Lascarides. 2003. Logics of Conversation. Cambridge: Cambridge University Press. Aske, Jon. 1989. Motion predicates in English and Spanish: A closer look. Berkeley Linguistic Society 15, 1–14. Baayen, Harald, Willem Levelt, and Alette Haveman. 1993. Research reported in Annual Report of the Max-Planck Institut fu¨r Psycholinguistik, 6. Nijmegen: MaxPlanck Institut. Bach, Emmon. 1986. The algebra of events. Linguistics and Philosophy 9, 5–16. Baker, C. L. 1979. Syntactic theory and the projection problem. Linguistic Inquiry 10, 533–81. Barðdal, Jo´hanna. 1999. Case in Icelandic: A construction grammar approach. Tijdschrift voor Skandinavistiek 20, 65–100. Barwise, J., and J. Perry. 1983. Situations and Attitudes. Cambridge, Mass.: Bradford/ MIT Press. Bassac, Christian. 2006. A compositional treatment for English compounds. Research in Language 4, 133–53.
references
453
Baynes, Kathleen, and Michael S. Gazzaniga. 2005. Lateralization of language: Toward a biologically based model of language. Linguistic Review 22, 303–26. Beck, Sigrid, and Arnim von Stechow. 2006. Dog after dog revisited. Universita¨t Tu¨bingen, MS. Berko, Jean. 1958. The child’s learning of English morphology. Word 14, 150–77. Berman, Ruth. 2009. Children’s acquisition of compound constructions. In Lieber and Sˇtekauer 2009, 298–322. Bickel, Balthasar. 1994. Mapping operations in spatial deixis and the typology of reference frames. Working paper no. 31, Cognitive Anthropology Research Group, Max-Planck Institut fu¨r Psycholinguistik, Nijmegen. —— 1997. Spatial operations in deixis, cognition, and culture: Where to orient oneself in Belhare. In Jan Nuyts and Eric Pederson (eds.), Language and Conceptualization. Cambridge: Cambridge University Press, 46–83. Bickerton, Derek. 1990. Language and Species. Chicago: University of Chicago Press. Biederman, Irving. 1987. Recognition-by-components: A theory of human image understanding. Psychological Review 94, 115–47. Bierwisch, Manfred. 1967. Some semantic universals of German adjectivals. Foundations of Language 3, 1–36. —— 1986. On the nature of semantic form in natural language. In F. Klix and H. Hagendorf (eds.), Human Memory and Cognitive Capabilities: Mechanisms and Performances. Amsterdam: Elsevier/North-Holland, 765–84. —— and Ewald Lang (eds.). 1989. Dimensional Adjectives. Berlin: Springer-Verlag. Bloom, Paul. 1994. Possible names: The role of syntax–semantics mappings in the acquisition of nominals. Lingua 92, 297–329. —— Mary A. Peterson, Lynn Nadel, and Merrill F. Garrett (eds.). 1996. Language and Space. Cambridge, Mass.: MIT Press. Boas, Hans. 2000. Resultatives at the crossroads between the lexicon and syntax: Where are they formed? Proceedings of the 1999 Western Conference on Linguistics 11, 38–52. —— 2003. Resultative Constructions in English and German. Stanford, Calif.: CSLI Publications. —— 2005. Determining the productivity of resultatives: A reply to Goldberg and Jackendoff. Language 81, 448–64. Bobalijk, Jonathan David. 2002. A-chains at the PF-interface: Copies and ‘covert’ movement. Natural Language and Linguistic Theory 20, 197–267. Bochner, Harry. 1993. Simplicity in Generative Morphology. Berlin and New York: Mouton de Gruyter. Bohnemeyer, Ju¨rgen, Nicholas J. Enfield, James Essegbey, Iraide Ibarretxe-Antun˜ano, Sotaro Kita, Friederike Lu¨pke, and Felix K. Maeka. 2007. Principles of event segmentation in language: The case of motion events. Language 83, 495–532. Bolinger, Dwight. 1965. The atomization of meaning. Language 41, 555–73. Booij, Geert. 2002. Constructional idioms, morphology and the Dutch lexicon. Journal of Germanic Linguistics 14, 301–29.
454
references
Booij, Geert. 2007. The Grammar of Words. An Introduction to Linguistic Morphology, 2nd edition. Oxford: Oxford University Press. —— 2009. Compounding and construction morphology. In Lieber and Sˇtekauer 2009, 201–16. Borer, Hagit. 1994. The projection of arguments. In Elena Benedicto and Jeffrey Runner (eds.), Functional Projections (University of Massachusetts Occasional Papers in Linguistics 17). Amherst, Mass.: GLSA, University of Massachusetts, Amherst, 19–47. Boroditsky, Lera. 2001. Does language shape thought? English and Mandarin speakers’ conceptions of time. Cognitive Psychology 43, 1–22. Botha, Rudolf P. 1968. The Function of the Lexicon in Transformational Generative Grammar. The Hague: Mouton. Bowerman, M. 1996. Learning how to structure space for language: A crosslinguistic perspective. In Bloom et al. 1996, 385–436. Brekle, Herbert E. 1975. Generative Satzsemantik im System der Englischen Nominalkomposition. Munich: Wilhem Fink Verlag. —— 1986. The production and interpretation of ad hoc nominal compounds in German: A realistic approach. Acta Linguistica Academiae Scientiarum Hungaricae 36, 39–52. Bresnan, Joan W. 1982a. Polyadicity. In Bresnan 1982b, 149–72. —— (ed.). 1982b. The Mental Representation of Grammatical Relations. Cambridge, Mass.: MIT Press. —— 2001. Lexical Functional Syntax. Oxford: Blackwell. —— and Lioba Moshi. 1990. Object asymmetries in comparative Bantu syntax. Linguistic Inquiry 21, 147–85. Brody, Michael. 1995. Focus and checking theory. In I. Kenesei (ed.), Approaches to Hungarian V: Levels and Structures. Szeged: JATE, 29–44. Brooks, Patricia, and Michael Tomasello. 1999. How children constrain their argument structure constructions. Language 75, 720–38. Burzio, Luigi. 2002. Missing players: Phonology and the past-tense debate. Lingua 112, 157–99. Busa, Federica. 1997. Compositionality and the semantics of nominals. PhD dissertation, Dept. of Computer Science, Brandeis University. Carnie, Andrew. 2000. On the definition of X0 and XP. Syntax 3, 59–106. Carrier, Jill, and Janet Randall. 1992. The argument structure and syntactic structure of resultatives. Linguistic Inquiry 23, 173–234. Casenhiser, Devin, and Adele E. Goldberg. 2005. Fast mapping of a phrasal form and meaning. Developmental Science 8, 500–8. Cheney, Dorothy, and Robert Seyfarth. 2007. Baboon Metaphysics: The Evolution of a Social Mind. Chicago: University of Chicago Press. Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. —— 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. —— 1970. Remarks on nominalization. In Roderick A. Jacobs and Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar. Waltham, Mass.: Ginn and Company, 184–221. Also in Chomsky 1972a. —— 1972a. Studies on Semantics in Generative Grammar. The Hague: Mouton.
references
455
Chomsky, Noam. 1972b. Some empirical issues in the theory of transformational grammar. In Stanley Peters (ed.), Goals of Linguistic Theory. Englewood Cliffs, NJ: Prentice-Hall, 63–130. Also in Chomsky 1972a. —— 1981. Lectures on Government and Binding. Dordrecht: Foris. —— 1986. Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. —— 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. —— 2000. Minimalist enquiries: The framework. In R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, Mass.: MIT Press, 89–155. —— and Morris Halle. 1968. The Sound Pattern of English. New York: Harper & Row. Clark, Eve V., Susan A. Gelman, and Nancy M. Lane. 1985. Compound nouns and category structure in young children. Child Development 56, 84–94. Clark, Herbert H. 1973. Space, time, semantics, and the child. In T. E. Moore (ed.), Cognitive Development and the Acquisition of Language. New York: Academic, 27–64. —— 1996. Using Language. Cambridge: Cambridge University Press. —— and W. G. Chase. 1972. On the process of comparing sentences against pictures. Cognitive Psychology 3, 472–517. Collins, Chris, Simanique Moody, and Paul M. Postal. 2008. An AAE camouflage construction. Language 84, 29–68. Coventry, Kenny R., and Simon C. Garrod. 2004. Saying, Seeing, and Acting: The Psychological Semantics of Spatial Prepositions. New York: Psychology Press. Crick, Francis. 1994. The Astonishing Hypothesis: The Scientific Search for the Soul. New York: Charles Scribner’s Sons. Croft, William. 2001. Radical Construction Grammar. Oxford: Oxford University Press. Culicover, Peter W. 1972. OM-Sentences: On the derivation of sentences with systematically unspecifiable interpretations. Foundations of Language 8, 199–236. —— 1999. Syntactic Nuts: Hard Cases, Syntactic Theory, and Language Acquisition. Oxford: Oxford University Press. —— 2000. Concrete minimalism, branching structure, and linear order. In Proceedings of GliP (Generative Linguistics in Poland). Warsaw. —— and Ray Jackendoff. 1997. Semantic subordination despite syntactic coordination. Linguistic Inquiry 28, 195–217. —— —— 1999. The view from the periphery: the English comparative correlative. Linguistic Inquiry 30, 543–71. —— —— 2005. Simpler Syntax. Oxford: Oxford University Press. Curtiss, Susan. 1977. Genie: A Linguistic Study of a Modern-Day ‘Wild Child’. New York: Academic Press. Cutting, J. 1981. Six tenets for event perception. Cognition 10, 71–8. De´chaine, Rose-Marie, and Martina Wiltschko. 2002. Decomposing pronouns. Linguistic Inquiry 33, 409–42.
456
references
Declerck, Renaat. 1979. Aspect and the bounded/unbounded (telic/atelic) distinction. Linguistics 17, 761–94. Dehe´, Nicole, Ray Jackendoff, Andrew McIntyre, and Silke Urban (eds.). 2002. VerbParticle Explorations. Berlin: Mouton de Gruyter. De Lange, Joke, Nada Vasic, and Sergey Avrutin. 2009. Reading between the (head) lines: A processing account of article omissions in newspaper headlines and child speech. Lingua 119, 1523–40. den Dikken, Marcel. 1995. Particles: On the Syntax of Verb-Particle, Triadic, and Causative Constructions. New York: Oxford University Press. —— 2005. Comparative correlatives comparatively. Linguistic Inquiry 36, 497–532. DiSciullo, Anna Maria, and Edwin Williams. 1987. On the Definition of Word. Cambridge, Mass.: MIT Press. Downing, Pamela. 1977. On the creation and use of English compound nouns. Language 53, 810–42. Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. —— 1991. Thematic proto-roles and argument selection. Language 67, 547–619. Dray, Nancy. 1987. Doubles and modifiers in English. Unpublished MA thesis, University of Chicago. Emmorey, K. 1996. The confluence of space and language in signed languages. In Bloom et al. 1996, 171–210. Emonds, Joseph E. 1970. Root and structure-preserving transformations. PhD dissertation, MIT. —— 1972. Evidence that indirect object movement is a structure-preserving rule. Foundations of Language 8, 546–61. Erteschik-Shir, Nomi. 1997. The Dynamics of Focus Structure. Cambridge: Cambridge University Press. Fanselow, Gisbert. 1985. Die Stellung der Wortbildung im System kognitiver Module. Linguistische Berichte 96, 91–126. Farah, Martha, K. Hammond, D. Levine, and R. Calvanio. 1988. Visual and spatial mental imagery: Dissociable systems of representation. Cognitive Psychology 20, 439–62. Fauconnier, Gilles. 1985. Mental Spaces: Aspects of Meaning Construction in Natural Language. Cambridge, Mass.: Bradford/MIT Press. Fiengo, Robert. 1974. Semantic conditions on surface structure. PhD dissertation, Department of Linguistics and Philosophy, MIT. Fillmore, Charles. 1968. The case for case. In Emmon Bach and Robert Harms (eds.), Universals in Linguistic Theory. New York: Holt, Rinehart & Winston, 1–88. —— 1971. Santa Cruz Lectures on Deixis. Bloomington, Ind.: Indiana University Linguistics Club. Republished as Lectures on Deixis, Stanford, Calif.: CSLI Publications, 1997. —— and Beryl T. Atkins. 1992. Toward a frame-based lexicon: The semantics of RISK and its neighbors. In Adrienne Lehrer and Eva Feder Kittay (eds.), Frames, Fields, and Constrasts. Hillsdale, NJ: Erlbaum, 75–102.
references
457
Fillmore, Charles and Paul Kay. 1993. Construction grammar coursebook. MS, University of California, Berkeley. —— —— and Mary Catherine O’Connor. 1988. Regularity and idiomaticity in grammatical constructions. Language 64, 501–38. Fodor, Janet D., Jerry A. Fodor, and Merrill Garrett. 1975. The psychological unreality of semantic representations. Linguistic Inquiry 6, 515–32. Fodor, Jerry A. 1970. Three reasons for not deriving ‘kill’ from ‘cause to die’. Linguistic Inquiry 1, 429–38. —— 1975. The Language of Thought. Cambridge, Mass.: Harvard University Press. —— 1980. Methodological solipsism considered as a research strategy in cognitive psychology. Behavioral and Brain Sciences 3, 63–73. —— 1981. The present status of the innateness controversy. In Jerry A. Fodor, Representations. Cambridge, Mass.: Bradford/MIT Press, 257–316. —— 1983. Modularity of Mind. Cambridge, Mass.: Bradford/MIT Press. —— 1987. Psychosemantics. Cambridge, Mass.: Bradford/MIT Press. —— Merrill F. Garrett, Edward Walker, and C. Parkes. 1980. Against definitions. Cognition 8, 263–367. Fraser, J. Bruce. 1965. An examination of the verb-particle construction in English. PhD dissertation, MIT. —— 1970. Idioms within a transformational grammar. Foundations of Language 6, 22–42. —— 1974. The Verb-Particle Construction in English. New York: Academic Press. Fried, Mirjam. 2005. A frame-based approach to case alternations: The swarm-class verbs in Czech. Cognitive Linguistics 16, 475–512. Ghomeshi, Jila. 1997. Non-projecting nouns and the Ezafe construction in Persian. Natural Language and Linguistic Theory 15, 729–88. —— Ray Jackendoff, Nicole Rosen, and Kevin Russell. 2004. Contrastive focus reduplication in English (the salad-salad paper). Natural Language and Linguistic Theory 22, 307–57. Givo´n, Talmy. 1979. On Understanding Grammar. New York: Academic Press. Gleitman, Lila R., and Henry Gleitman. 1970. Phrase and Paraphrase. New York: Norton. —— —— Carol Miller, and Ruth Ostrin. 1996. Similar, and similar concepts. Cognition 58, 321–76. Goldberg, Adele E. 1991. A semantic account of resultatives. Linguistic Analysis 21, 66–96. —— 1992. The inherent semantics of argument structure: The case of the English ditransitive construction. Cognitive Linguistics 3, 37–74. —— 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. —— 1996. Jackendoff and construction-based grammar. Cognitive Linguistics 7, 3–20. —— 1997. Relationships between verb and construction. In Marjolijn Verspoor and Eve Sweetser (eds.), Lexicon and Grammar. Philadelphia: John Benjamins, 383–98.
458
references
Goldberg, Adele E. 1999. The emergence of argument structure semantics. In Brian MacWhinney (ed.), The Emergence of Language. Hillsdale, NJ: Lawrence Erlbaum Publications. —— 2000. Patient arguments of causative verbs can be omitted: The role of information structure in argument distribution. Language Sciences, 34, 503–24. —— 2005a. Argument realization: The role of constructions, lexical semantics and ¨ stman (eds.), Construction Gramdiscourse factors. In Mirjam Fried and Jan-Ola O mar(s): Cognitive and Cross-language Dimension. Philadelphia: John Benjamins. —— 2005b. Constructions, lexical semantics and the Correspondence Principle: Accounting for generalizations and subregularities in the realization of arguments. In Nomi Erteschik-Shir and Tova Rapoport (eds.), The Syntax of Aspect. Oxford: Oxford University Press. —— 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press. —— Devin Casenhiser, and Nitya Sethuraman. 2004. Learning argument structure generalizations. Cognitive Linguistics 15, 289–316. —— and Ray Jackendoff. 2004. The English resultative as a family of constructions. Language 80, 532–68. —— —— 2005. The end result(ative). Language 81, 474–7. Goldin-Meadow, Susan. 2003. The Resilience of Language. New York: Psychology Press. Grandgent, Charles H. 1908. An Introduction to Vulgar Latin. Boston: Heath. Green, Georgia. 1972. Some observations on the syntax and semantics of instrumental verbs. Chicago Linguistic Society 8, 83–97. Grimshaw, Jane. 1990. Argument Structure. Cambridge, Mass.: MIT Press. Gruber, Jeffrey S. 1965. Studies in lexical relations. PhD dissertation, MIT, Cambridge, Mass. Reprinted in J. Gruber, Lexical Structures in Syntax and Semantics. Amsterdam: North-Holland, 1976. —— 1967. Functions of the Lexicon in Formal Descriptive Grammar. Santa Monica, Calif.: Systems Development Corporation. Reprinted in J. Gruber, Lexical Structures in Syntax and Semantics. Amsterdam: North-Holland, 1976. Habel, Cristof. 1994. Paper presented at Nijmegen space conference. Haiman, John. 1991. Motivation, repetition and emancipation: The bureaucratization of language. In H. C. Wolfart (ed.), Linguistic Studies Presented to John L. Finlay, Algonquian and Iroquoian Linguistics, Memoir 8, University of Manitoba, 45–69. Hale, Kenneth, and Samuel Jay Keyser. 1993. On argument structure and the lexical expression of syntactic relations. In K. Hale and S. J. Keyser (eds.), The View from Building 20. Cambridge, Mass.: MIT Press, 53–109. —— —— 2002. Prolegomenon to a Theory of Argument Structure. Cambridge, Mass.: MIT Press. Halle, Morris. 1959. The Sound Pattern of Russian. The Hague: Mouton. —— 1973. Prolegomena to a theory of word formation. Linguistic Inquiry 4, 3–16. —— and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In K. Hale and S. J. Keyser (eds.), The View from Building 20. Cambridge, Mass.: MIT Press, 111–76.
references
459
Hankamer, Jorge. 1989. Morphological parsing and the lexicon. In W. Marslen-Wilson (ed.), Lexical Representation and Process. Cambridge, Mass.: MIT Press, 392–408. Harris, Randy. 1993. The Linguistics Wars. New York: Oxford University Press. Hauser, Marc, Noam Chomsky, and W. Tecumseh Fitch. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298, 1569–79. Hayes, Bruce. 1989. The prosodic hierarchy in meter. In Paul Kiparsky and Gilbert Youmans (eds.), Phonetics and Phonology, i: Rhythm and Meter. New York: Academic Press, 201–60. Herskovits, Annette. 1985. Semantics and pragmatics of locative expressions. Cognitive Science 9, 341–78. —— 1986. Language and Spatial Cognition. Cambridge: Cambridge University Press. Hess, R. H., C. L. Baker, and J. Zihl. 1989. The ‘motion-blind’ patient: Low-level spatial and temporal filters. Journal of Neuroscience 9, 1628–40. Higginbotham, J. 1985. On semantics. Linguistic Inquiry 16, 547–94. Hinrichs, E. 1985. A compositional semantics for aktionsarten and NP reference in English. PhD dissertation, Department of Linguistics, Ohio State University. Hoekstra, Teun. 1988. Small clause results. Lingua 74, 101–39. —— and R. Mulder. 1990. Unergatives as copular verbs: Locational and existential predication. Linguistic Review 7, 1–79. Hohenhaus, Peter. 2004. Identical constituent compounding: A corpus-based study. Folia Linguistica 38, 297–331. Horn, Lawrence. 1993. Economy and redundancy in a dualistic model of natural language. In S. Shore and M. Vilkuna (eds.), SKY 1993, Yearbook of the Linguistic Association of Finland, 31–72. —— 2006. Speaker and hearer in neo-Gricean pragmatics. Wai Guo Yu/Journal of Foreign Languages 164, 2–26. Huddleston, Rodney, and Geoffrey K. Pullum. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Hudson, Richard. 1990. English Word Grammar. Oxford: Blackwell. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press. —— 1973. The base rules for prepositional phrases. In S. Anderson and P. Kiparsky (eds.), A Festschrift for Morris Halle. New York: Holt, Rinehart & Winston, 345–56. —— 1975. Morphological and semantic regularities in the lexicon. Language 51, 639–71. —— 1976. Toward an explanatory semantic representation. Linguistic Inquiry 7, 89–150. —— 1977. X-Bar Syntax: A Study of Phrase Structure. Cambridge, Mass.: MIT Press. —— 1978. Grammar as evidence for conceptual structure. In M. Halle, J. Bresnan, and G. Miller (eds.), Linguistic Theory and Psychological Reality. Cambridge, Mass.: MIT Press, 201–28. —— 1983. Semantics and Cognition. Cambridge, Mass.: MIT Press. —— 1987a. The status of thematic relations in linguistic theory. Linguistic Inquiry 18, 369–411.
460
references
Jackendoff, Ray. 1987b. On Beyond Zebra: The relation of linguistic and visual information. Cognition 26, 89–114. —— 1987c. Consciousness and the Computational Mind. Cambridge, Mass.: Bradford/ MIT Press. —— 1990. Semantic Structures. Cambridge, Mass.: MIT Press. —— 1991a. Parts and boundaries. Cognition 41, 9–45. —— 1991b. The problem of reality. Nouˆs 25, 411–33. Reprinted in Jackendoff 1992. —— 1992. Languages of the Mind. Cambridge, Mass.: MIT Press. —— 1994. Lexical insertion in a postminimalist theory of grammar. Unpublished MS, integrated into Jackendoff 1997b. —— 1996a. Conceptual semantics and cognitive semantics. Cognitive Linguistics 7, 93–129. —— 1996b. The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language and Linguistic Theory 14, 305–54. —— 1996c. The architecture of the linguistic–spatial interface. In Bloom et al. 1996, 1–30. —— 1997a. Twistin’ the night away. Language 73, 534–59. —— 1997b. The Architecture of the Language Faculty. Cambridge, Mass.: MIT Press. —— 2002. Foundations of Language. Oxford: Oxford University Press. —— 2007a. Language, Consciousness, Culture. Cambridge, Mass.: MIT Press. —— 2007b. Linguistics in cognitive science: The state of the art. Linguistic Review 24, 347–401. —— 2008. Alternative minimalist visions of language. In R. Edwards, P. Midtlyng, K. Stensrud, and C. Sprague (eds.), Chicago Linguistic Society 41: The Panels. Chicago: Chicago Linguistic Society, 189–226. Also to appear in Robert D. Borsley and Kersti Bo¨rjars, Nontransformational Syntax: A Guide to Current Models. Oxford: Blackwell. —— 2009. Compounding in the Parallel Architecture and Conceptual Semantics. In R. Lieber and P. Sˇtekauer (eds.), Handbook of Compounding. Oxford: Oxford University Press, 105–28. —— and B. Landau. 1991. Spatial language and spatial cognition. In D. J. Napoli and J. Kegl (eds.), Bridges between Psychology and Linguistics: A Swarthmore Festschrift for Lila Gleitman. Hillsdale, NJ: Erlbaum, 144–69. Reprinted in Jackendoff 1992. Expanded version: Landau and Jackendoff 1993. Janda, Richard, and Brian Joseph. 1999. The modern Greek negator mi(n)(-) as a morphological constellation. In G. Babiniotis (ed.), Greek linguistics ’97: Proceedings of the 3rd International Conference on Greek Linguistics. Athens: Elinika Gramata Press, 341–51. January, David, and Edward Kako. 2007. Re-evaluating evidence for linguistic relativity: Reply to Boroditsky (2001). Cognition 104, 417–26. Jeannerod, Marc. 1994. The representing brain: Neural correlates of motor intention and imagery. Behavioral and Brain Sciences 17, 187–201. Jespersen, Otto. 1942. A Modern English Grammar on Historical Principles, vol. vi. London: Allen & Unwin.
references
461
Johnson-Laird, P. 1996. Space to think. In Bloom et al. 1996, 437–63. Jolicoeur, P., and Stephen Kosslyn. 1981. Coordinate systems in the long-term memory representation of three dimensional shapes. Cognitive Psychology 15, 301–45. Joseph, Brian D., and Richard D. Janda. 1985. One rule or many? Sanskrit reduplication as fragmented affixation. Paper presented at Proceedings of the Second Eastern States Conference on Linguistics, Columbus. Katz, Jerrold J. 1973. Compositionality, idiomaticity, and lexical substitution. In S. Anderson and P. Kiparsky (eds.), A Festschrift for Morris Halle. New York: Holt, Rinehart & Winston, 357–76. —— and Jerry A. Fodor. 1963. The structure of a semantic theory. Language 39, 170–210. Kay, Paul. 2002. English subjectless tagged sentences. Language 78, 453–81. —— and Charles J. Fillmore. 1999. Grammatical constructions and linguistic generalizations: The what’s X doing Y? construction. Language 75, 1–34. Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Kiss, Katalin. 1998. Identificational focus versus information focus. Language 74, 245–73. Klein, Wolfgang, and Clive Perdue. 1997. The basic variety (or: Couldn’t natural languages be much simpler?). Second Language Research 13, 301–47. Klima, Edward S. 1965. Studies in diachronic syntax. Unpublished PhD dissertation, Harvard University. Koch, Christof. 2004. The Quest for Consciousness. Englewood, Colo.: Roberts. Ko¨hler, Wolfgang. 1927. The Mentality of Apes. London: Routledge & Kegan Paul. Ko¨nig, Ekkehard, and Claire Moyse-Faurie. 2007. Spatial reciprocity: Between grammar and lexis. MS, Freie Universita¨t Berlin. Koopman, Hilda. 1984. The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar. Dordrecht: Foris. Kosslyn, Stephen. 1980. Image and Mind. Cambridge, Mass.: Harvard University Press. Krifka, Manfred. 1989. Nominalreferenz und Zeitkonstitution. Munich: Wilhelm Fink Verlag. —— 1992. Thematic relations as links between nominal reference and temporal constitution. In I. Sag and A. Szabolcsi (eds.), Lexical Matters. Stanford, Calif.: CSLI Publications, 29–54. Lakoff, George. 1970. Irregularity in Syntax. New York: Holt, Rinehart & Winston. —— 1971. On generative semantics. In D. Steinberg and L. Jakobovits (eds.), Semantics: An Interdisciplinary Reader. Cambridge: Cambridge University Press, 232–96. —— 1972. Hedges: A study in meaning criteria and the logic of fuzzy concepts. In Papers from the Eighth Regional Meeting, Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 183–228. —— 1987. Women, Fire, and Dangerous Things. Chicago: University of Chicago Press. —— and Mark Johnson. 1980. Metaphors We Live By. Chicago: University of Chicago Press.
462
references
Landau, Barbara. 1996. Multiple geometric representations of objects in languages and language learners. In Bloom et al. 1996, 317–64. —— and Ray Jackendoff. 1993. ‘What’ and ‘where’ in spatial language and spatial cognition. Behavioral and Brain Sciences 16, 217–38. Langacker, Ronald. 1983. Foundations of Cognitive Grammar. Bloomington, Ind.: Indiana University Linguistics Club. —— 1987a. Foundations of Cognitive Grammar, vol. i. Stanford, Calif.: Stanford University Press. —— 1987b. Nouns and verbs. Language 63, 53–94. —— 1991. Foundations of Cognitive Grammar, vol ii. Stanford, Calif.: Stanford University Press. —— 1992. Reference point constructions. Cognitive Linguistics 4, 1–38. —— 2003. Construction grammars: Cognitive, radical and less so. Paper presented at the 8th International Cognitive Linguistics Conference, University of La Rioja, Logron˜o, Spain. Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19, 335–91. Lascarides, Alex, and Ann Copestake. 1998. Pragmatics and word meaning. Journal of Linguistics 34, 387–414. Lasersohn, Peter. 1999. Pragmatic halos. Language 75, 522–51. Lasher, M. 1981. The cognitive representation of an event involving human motion. Cognitive Psychology 13, 391–406. Lees, Robert. B. 1960. The Grammar of English Nominalizations. The Hague: Mouton. Lehrer, Adrienne, and Eva Feder Kittay (eds.). 2002. Frames, Fields, and Constrasts. Hillsdale, NJ: Erlbaum. Lepschy, Anna, and Giulio Lepschy. 1984. La lingua italiana. Milan: Bompiani. Lerdahl, Fred, and Ray Jackendoff. 1983. A Generative Theory of Tonal Music. Cambridge, Mass.: MIT Press. Levelt, Willem J. M. 1984. Some perceptual limitations in talking about space. In A. van Doorn, W. van de Grind, and J. Koenderink (eds.), Limits in Perception. Utrecht: Coronet Books. —— 1996. Perspective taking and ellipsis in spatial descriptions. In Bloom et al. 1996, 77–108. Levi, Judith. 1978. The Semantics and Syntax of Complex Nominals. New York: Academic Press. Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investigation. Chicago: University of Chicago Press. —— and Tova Rapoport. 1988. Lexical subordination. In Papers from the Twentyfourth Regional Meeting of the Chicago Linguistics Society. Chicago: Department of Linguistics, University of Chicago, 275–89. —— and Malka Rappaport Hovav. 1990. Wiping the slate clean: a lexical semantic exploration. Cognition 41, 123–55.
references
463
Levin, Beth, and Malka Rappaport Hovav. 1995. Unaccusativity: At the Syntax–Lexical Semantics Interface. Cambridge, Mass.: MIT Press. —— —— 1996. Two types of derived accomplishments. Proceedings of the First LFG Conference. Grenoble. —— —— 2005. Argument Realization. Cambridge: Cambridge University Press. Levinson, Stephen C. 1996. Frames of reference and Molyneux’s question: Crosslinguistic evidence. In Bloom et al. 1996, 109–70. —— 2003. Space in Language and Cognition. Cambridge: Cambridge University Press. Li, Peggy, Linda Abarbanell, and Anna Papafragou. 2005. Spatial reasoning skills in Tenejapan Mayans. Poster presented at Cognitive Science Society. http://www.psych. unito.it/csc/cogsci05/frame/poster/1/f537–li.pdf. —— and Lila Gleitman. 2002. Turning the tables: Language and spatial reasoning. Cognition 83, 265–94. Lidz, Jeffrey. 2001. Echo reduplication in Kannada and the theory of word-formation. Linguistic Review 18, 375–94. Lieber, Rochelle. 1992. Deconstructing Morphology. Chicago: University of Chicago Press. —— 2004. Morphology and Lexical Semantics. Cambridge: Cambridge University Press. —— and Pavol Sˇtekauer (eds.). 2009. The Oxford Handbook of Compounding. Oxford: Oxford University Press. Link, G. 1983. The logical analysis of plurals and mass terms: A lattice-theoretic approach. In R. Bauerle, C. Schwarze, and A. von Stechow (eds.), Meaning, Use, and Interpretation of Language. Berlin: Walter de Gruyter, 302–23. Lombardi, Linda. 1984. Away with the passive. MS, Brandeis University. McCarthy, John. 1982. Prosodic structure and expletive infixation. Language 58, 574–90. —— and Alan Prince. 1986. Prosodic morphology. MS, University of Massachusetts, Amherst, and Brandeis University. McCawley, James D. 1968. Lexical insertion in a transformational grammar without deep structure. Papers from the 4th Regional Meeting, Chicago Linguistic Society, 71–80. —— 1978. Conversational implicature and the lexicon. In P. Cole (ed.), Syntax and Semantics 9: Pragmatics. New York: Academic Press, 245–59. Macnamara, John. 1978. How can we talk about what we see? Unpublished mimeo. Department of Psychology, McGill University, Montreal. Mandler, Jean M. 1992. How to build a baby: II. Conceptual primitives. Psychological Review 99, 587–604. —— 1996. Preverbal representation and language. In Bloom et al. 1996, 365–84. Marantz, Alec. 1982. Re reduplication. Linguistic Inquiry 13, 483–545. —— 1992. The way-construction and the semantics of direct arguments in English: A reply to Jackendoff. In T. Stowell and E. Wehrli (eds.), Syntax and Semantics 26: Syntax and the Lexicon. New York: Academic Press, 179–88.
464
references
Marantz, Alec. 1997. No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. Philadelphia: University of Pennsylvania Working Papers in Linguistics 4/2, 201–25. —— 2005. Generative linguistics within the cognitive neuroscience of language. Linguistic Review 22, 429–46. Marchand, H. 1969. The Categories and Types of Present-Day English Word-Formation: A Synchronic-Diachronic Approach, 2nd edn. Munich: Beck. Marr, David. 1982. Vision. San Francisco: Freeman. —— and H. K. Nishihara. 1978. Representation and recognition of the spatial organization of three-dimensional shapes. Proceedings of the Royal Society of London B200, 269–94. —— and L. Vaina. 1982. Representation and recognition of the movements of shapes. Proceedings of the Royal Society of London B214, 501–24. Matsuoka, Kazumi. 1997. Verb raising in American Sign Language. Lingua 103, 127–49. Matsuyama, Tetsuya. 2004. The N after N construction: A constructional idiom. English Linguistics 21, 55–84. Michaelis, Laura. 1994. A case of constructional polysemy in Latin. Studies in Language 18, 23–48. —— and Knud Lambrecht. 1996. Toward a construction-based theory of language function: The case of nominal extraposition. Language 72, 215–47. Michotte, A. 1954. La perception de la causalite´, 2nd edition. Louvain: Publications Universitaires de Louvain. Miller, G., and P. Johnson-Laird 1976. Language and Perception. Cambridge, Mass.: Harvard University Press. Millikan, Ruth. 1984. Language, Thought, and Other Biological Categories. Cambridge, Mass.: MIT Press. Moravcsik, Edith. 1978. Reduplicative constructions. In Joseph H. Greenberg (ed.), Universals of Human Language, iii: Word Structure. Stanford, Calif.: Stanford University Press. Moscoso del Prado Martı´n, Fermin, Aleksandar Kostic, and R. Harald Baayen. 2004. Putting the bits together: An information theoretical perspective on morphological processing. Cognition 94, 1–18. Mourelatos, A. 1978. Events, processes, and states. Linguistics and Philosophy 1, 199–220. Murphy, Gregory L. 2002. The Big Book of Concepts. Cambridge, Mass.: MIT Press. Mutaka, Ngessimo, and Larry M. Hyman. 1990. Syllables and morpheme integrity in Kinande reduplication. Phonology 7, 73–119. Nakamura, Tsuguro. 1997. Actions in motion: How languages express manners of motion. In Masatomo Ukaji, Toshio Nakao, Masaru Kajita, and Shuji Chiba (eds.), Studies in English Linguistics: A Festschrift for Akira Ota on the Occasion of his Eightieth Birthday. Tokyo: Taishukan Publishing Company, 723–38. Narasimhan, Bhuvaneswari. 1993. Spatial frames of reference in the use of length, width, and height. Unpublished paper, Boston University.
references
465
—— 1998. The encoding of complex events in Hindi and English. PhD dissertation, Boston University. Neisser, Ulric. 1967. Cognitive Psychology. Englewood Cliffs, NJ: Prentice-Hall. Nespor, Marina, and Irene Vogel. 1986. Prosodic Phonology. Dordrecht: Foris. Nooteboom, Sieb, F. Weerman, and F. Wijnen (eds.). 2002. Storage and Computation in the Language Faculty. Dordrecht: Kluwer, 23–58. Nunberg, Geoffrey. 1979. The nonuniqueness of semantic solutions: Polysemy. Linguistics and Philosophy 3, 143–84. —— Ivan Sag, and Thomas Wasow. 1994. Idioms. Language 70, 491–538. Oehrle, Richard. 1998. Noun after noun. Paper delivered at Annual Meeting of the Linguistic Society of America. O’Keefe, John, and Lynn Nadel. 1978. The Hippocampus as a Cognitive Map. Oxford: Oxford University Press. Olsen, Susan. 2001. Copulative compounds: A closer look at the interface between morphology and syntax. Yearbook of Morphology 2000. Dordrecht: Kluwer, 279–320. Olson, David, and Ellen Bialystok. 1983. Spatial Cognition. Hillsdale, NJ: Erlbaum. Paivio, A. 1971. Imagery and Verbal Processes. New York: Holt, Rinehart, & Winston. Reprinted by Erlbaum, 1979. Partee, Barbara. 1993. Semantic structures and semantic properties. In E. Reuland and W. Abraham (eds.), Knowledge and Language, ii: Lexical and Conceptual Structure. Dordrecht: Kluwer, 7–30. Pelletier, F. J. (ed.). 1979. Mass Terms: Some Philosophical Problems. Dordrecht: Reidel. Peperkamp, Sharon. 1997. On the prosodic representation of clitics. In Ursula Kleinhenz (ed.), Interfaces in Phonology, Studia Grammatica 41. Berlin: Academie Verlag, 102–27. Perlmutter, David. 1970. The two verbs ‘begin’. In Roderick A. Jacobs and Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar. Waltham, Mass.: Ginn and Company, 107–19. Peterson, P. 1985. Causation, agency, and natural actions. In Chicago Linguistic Society, 21st Regional Meeting and Parasession on Causative and Agentivity, Department of Linguistics, University of Chicago. Petronio, Karen. 1993. Clause structure in American Sign Language. PhD dissertation, University of Washington. —— and Diane Lillo-Martin. 1997. WH-movement and the position of Spec-CP: Evidence from American Sign Language. Language 73, 18–57. Phillips, Colin, and Ellen Lau. 2004. Foundational issues (Review article on Jackendoff 2002). Journal of Linguistics 40, 1–21. Pi, Chia-Yi Tony. 1995. The structure of English iteratives. In Pa¨ivi Koskinen (ed.), Proceedings of the 1995 Annual Conference of the Canadian Linguistic Association. Toronto Working Papers in Linguistics, 434–45. Pinker, Steven. 1984. Visual cognition: An introduction. Cognition 18, 1–63.
466
references
Pinker, Steven. 1989. Learnability and Cognition: The Acquisition of Argument Structure. Cambridge, Mass.: Bradford/MIT Press. —— 1999. Words and Rules. New York: Basic Books. —— 2007. The Stuff of Thought. New York: Viking. —— and Ray Jackendoff. 2005. The faculty of language: What’s special about it? Cognition 95, 201–36. —— and Alan Prince. 1988. On language and connectionism: Analysis of a parallel distributed processing model of language acquisition. Cognition 26, 195–267. —— —— 1991. Regular and irregular morphology and the psychological status of rules of grammar. In Laurel A. Sutton, Christopher Johnson, and Ruth Shields (eds.), Proceedings of Berkeley Linguistics Society 17. Berkeley: Berkeley Linguistics Society, 230–51. Platzack, Christer. 1979. The Semantic Interpretation of Aspect and Aktionsarten. Dordrecht: Foris. Pollard, Carl, and Ivan Sag. 1987. Information-Based Syntax and Semantics, vol. i. Stanford, Calif.: Center for the Study of Language and Information. —— —— 1994. Head-Driven Phrase Structure Grammar. Stanford, Calif.: CSLI. Poser, William J. 1991. Ma. In Carol Georgopoulos and Roberta Ishihara (eds.), Interdisciplinary Approaches to Language. Dordrecht: Kluwer Academic Publishers, 449–58. Postal, Paul M. 1972. The best theory. In P. S. Peters (ed.), The Goals of Linguistic Theory. Englewood Cliffs, NJ: Prentice-Hall. —— 2004. Skeptical Linguistic Essays. Oxford: Oxford University Press. Postma, Gertjan. 1995. Zero-semantics: The syntactic encoding of quantificational meaning. In M. den Dikken and K. Hengeveld (eds.), Linguistics in the Netherlands 1995. Philadelphia: Benjamins, 175–90. Pullum, Geoffrey K. 1991. English nominal gerund phrases as noun phrases with verb phrase heads. Linguistics 29, 763–99. Pustejovsky, James. 1991a. The syntax of Event Structure. Cognition 41, 47–81. —— 1991b. The generative lexicon. Computational Linguistics 17, 409–41. —— 1995. The Generative Lexicon. Cambridge, Mass.: MIT Press. Putnam, Hilary. 1975. The meaning of ‘meaning’. In K. Gunderson (ed.), Language, Mind, and Knowledge. Minneapolis: University of Minnesota Press, 131–93. —— 1988. Representation and Reality. Cambridge, Mass.: MIT Press. Pylyshyn, Zenon. 1973. What the mind’s eye tells the mind’s brain: A critique of mental imagery. Psychological Bulletin 80, 1–24. —— 1980. Computation and Cognition. Cambridge, Mass.: Bradford/MIT Press. Raphael, Bertram. 1968. SIR: A computer program for semantic information retrieval. In Marvin Minsky (ed.), Semantic Information Processing. Cambridge, Mass.: MIT Press, 33–145. Rappaport, Malka, and Beth Levin. 1985. The locative alternation: A case study in lexical analysis. MS, Center for Cognitive Science, MIT.
references
467
Rappaport, Malka, and Beth Levin. 1988. What to do with theta-roles. In W. Wilkins (ed.), Syntax and Semantics, 21: Thematic Relations. New York: Academic Press, 7–36. Rappaport Hovav, Malka, and Beth Levin. 1992. -er nominals: Implications for a theory of argument structure. In T. Stowell and E. Wehrli (eds.), Syntax and Semantics 26: Syntax and the Lexicon. Academic Press, New York, 127–53. —— —— 2001. An event structure account of English resultatives. Language 77, 766–97. Richards, Norvin. 1997. What moves where and in which language. PhD dissertation, MIT. Rizzi, Luigi. 1997. The fine structure of the left periphery. In L. Haegeman (ed.), Elements of Grammar: Handbook of Generative Syntax. Dordrecht: Kluwer Academic Publishers, 281–339. Rosch, Eleanor. 1975. Cognitive reference points. Cognitive Psychology 7, 532–47. —— 1978. Principles of categorization. In E. Rosch and B. Lloyd (eds.), Cognition and Categorization. Hillsdale, NJ: Erlbaum, 27–48. —— and Carolyn Mervis. 1975. Family resemblances: Studies in the internal structure of categories. Cognitive Psychology 7, 573–605. Rosen, Nicole. 2003. Demonstrative position in Michif. The Canadian Journal of Linguistics 48, 39–69. Ross, John Robert. 1967. Constraints on variables in syntax. PhD dissertation, MIT. Published as Infinite Syntax! Norwood, NJ: Ablex, 1986. Rothstein, Susan. 2000. Secondary predication and aspectual structure. In E. Lang, C. Fabricius-Hansen, and C. Maienborn (eds.), ZAS Papers in Linguistics 17: Papers from the Oslo Conference on Adjuncts. Rumelhart, David E., and James L. McClelland. 1986. On learning the past tense of English verbs. In J. McClelland, D. Rumelhart, and the PDP Research Group, Parallel Distributed Processing, vol. ii. Cambridge, Mass.: MIT Press, 216–71. Russell, Kevin. 1999. MOT: Sketch of an OT approach to morphology. MS, University of Manitoba. Ryder, Mary Ellen. 1994. Ordered Chaos: The Interpretation of English Noun-Noun Compounds. Berkeley and Los Angeles: University of California Press. Sadock, Jerrold. 1991. Autolexical Syntax. Chicago: University of Chicago Press. —— 1998. On the autonomy of compounding morphology. In Steven G. Lapointe, Diane K. Brentari, and Patrick M. Farrell (eds.), Morphology and its Relation to Phonology and Syntax. Stanford, Calif.: CSLI Publications, 161–87. Sag, Ivan A. 1997. English relative clause constructions. Journal of Linguistics 33, 431–84. —— Tom Wasow, and Emily Bender. 2003. Syntactic Theory: A Formal Introduction, 2nd edition. Stanford, Calif.: CSLI Publications. Schank, Roger. 1973. Identification of conceptualizations underlying natural language. In R. Schank and K. Colby (eds.), Computer Models of Thought and Language. San Francisco, Calif.: Freeman, 187–248. Schein, Barry. 1994. Plurals and Events. Cambridge, Mass.: MIT Press. Schiffer, Stephen. 1987. Remnants of Meaning. Cambridge, Mass.: MIT Press.
468
references
Seland, Grete. 2001. The Norwegian reflexive caused motion construction: A Construction Grammar approach. PhD dissertation, University of Oslo. Selkirk, Elisabeth. 1982. The Syntax of Words. Cambridge, Mass.: MIT Press. —— 1996. The prosodic structure of function words. In Jill N. Beckman, Laura Walsh Dickey, and Susanne Urbanczyk (eds.), Papers in Optimality Theory. University of Massachusetts Occasional Papers 18, 439–69. Shepard, Roger, and Lynn Cooper. 1982. Mental Images and their Transformations. Cambridge, Mass.: MIT/Bradford. Simpson, Jane. 1983. Resultatives. In Lori Levin, Malka Rappaport, and Annie Zaenen (eds.), Papers in Lexical-Functional Grammar. Bloomington, Ind.: Indiana University Linguistics Club, 143–57. Slobin, Dan. 1996. From ‘Thought and language’ to ‘Thinking for speaking’. In John J. Gumperz and Stephen C. Levinson (eds.), Rethinking Linguistic Relativity. Cambridge: Cambridge University Press, 70–96. Smith, Edward, and Douglas Medin. 1981. Categories and Concepts. Cambridge, Mass.: Harvard University Press. Snyder, William. 2001. On the nature of syntactic variation: Evidence from complex predicates and complex word-formation. Language 77, 324–42. Stanley, Richard. 1967. Redundancy rules in phonology. Language 43, 393–436. Steriade, Donca. 2001. Lexical conservatism and the notion base of affixation. MS, UCLA. http://www.linguistics.ucla.edu/people/steriade/papers/LexicalConservatism.pdf. Stowell, Tim. 1981. Origins of phrase structure. PhD dissertation, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Subramanian, Uma M. 1992. On internal and external syntax mismatches. PhD dissertation, Ohio State University. Takami, Ken-ichi. 1998. Unaccusativity and the resultative constructions in English and Japanese. MS, Tokyo Metropolitan University. Talmy, Leonard. 1978a. The relation of grammar to cognition: A synopsis. In D. Waltz (ed.), Theoretical Issues in Natural Language Processing 2. New York: Association for Computing Machinery. Revised version in Talmy 2000. —— 1978b. Figure and ground in complex sentences. In Joseph Greenberg, Charles Ferguson, and M. Moravcsik (eds.), Universals of Human Language, vol. iv. Stanford, Calif.: Stanford University Press. Revised version in Talmy 2000. —— 1980. Lexicalization patterns: Semantic structure in lexical forms. In T. Shopen et al. (eds.), Language Typology and Syntactic Descriptions. New York: Cambridge University Press. Revised version in Talmy 2000. —— 1983. How language structures space. In H. Pick and L. Acredolo (eds.), Spatial Orientation: Theory, Research, and Application. New York: Plenum. Revised version in Talmy 2000. —— 1985. Force dynamics in language and thought. In Papers from the Twenty-First Regional Meeting of the Chicago Linguistic Society. Chicago: University of Chicago Linguistics Department. Also in Cognitive Science 12, 49–100 (1988). Revised version in Talmy 2000.
references
469
—— 1996. Fictive motion in language and ‘ception.’ In Bloom et al. 1996, 211–76. —— 2000. Toward a Cognitive Semantics, i: Concept Structuring Systems; ii: Typology and Process in Concept Structuring. Cambridge, Mass.: MIT Press. Taylor, John. 1995. Linguistic Categorization. Oxford: Oxford University Press. —— 1996. On running and jogging. Cognitive Linguistics 7, 21–34. Teller, Paul. 1969. Some discussion and extension of Manfred Bierwisch’s work on German adjectivals. Foundations of Language 5, 185–217. ten Hacken, Pius. 1994. Defining Morphology: A Principled Approach to Determining the Boundaries of Compounding, Derivation, and Inflection. Hildesheim: Olms. Tenny, Carol. 1987. Grammaticalizing aspect and affectedness. PhD dissertation, Department of Linguistics and Philosophy, MIT. —— 1992. The Aspectual Interface Hypothesis. In I. Sag and A. Szabolcsi (eds.), Lexical Matters. Stanford, Calif.: CSLI Publications, 1–28. —— 1994. Aspectual Roles and the Syntax–Semantics Interface. Dordrecht: Kluwer. Terrace, Herbert. 1979. Nim. New York: Knopf. Toivonen, Ida. 2002a. The directed motion construction in Swedish. Journal of Linguistics 38, 313–45. —— 2002b. Swedish particles and syntactic projection. In Dehe´ et al. 2002, 191–209. —— 2006. On continuative on. Studia Linguistica 60, 181–219. Tomasello, Michael. 2000. Do young children have adult syntactic competence? Cognition 74, 209–53. —— 2003. Constructing a Language: A Usage-Based Theory of Language Acquisition. Cambridge, Mass.: Harvard University Press. Travis, Lisa. 1988. The syntax of adverbs. McGill Working Papers in Linguistics: Proceedings of the IVth Workshop on Comparative Germanic Syntax, 280–310. —— 1999. A syntactician’s view of reduplication. Paper presented at AFLA VI, 16–18 April 1999, University of Toronto. —— 2000. The syntax of reduplication. Paper presented at NELS 31, 6–8 October 2000, Georgetown University. —— 2001. The syntax of reduplication. In Minjoo Kim and Uri Strauss (eds.), Proceedings of the North Eastern Linguistics Society Annual Meeting (NELS 31). Amherst, Mass.: GLSA, 455–69. Tversky, Amos. 1977. Features of similarity. Psychological Review 84, 327–50. Tversky, Barbara. 1996. Spatial perspective in descriptions. In Bloom et al. 1996, 463–92. Ungerleider, Leslie, and Mortimer Mishkin. 1982. Two cortical visual systems. In D. Ingle, M. Goodale, and R. Mansfield (eds.), Analysis of Visual Behavior. Cambridge, Mass.: MIT Press. Vaina, L. 1983. From shapes and movements to objects and actions. Synthese 54, 3–36. Valimaa-Blum, Riitta. 1988. Finnish word order as a set of syntactic constructions. Proceedings of the Fifth Eastern States Conference on Linguistics, Ohio State University.
470
references
Vallduvı´, E. 1995. Structural properties of information packaging in Catalan. In K. E. Kiss (ed.), Discourse Configurational Languages. Oxford: Oxford University Press, 122–52. van der Zee, Emile, and Jon Slack. 2003. Representing Direction in Language and Space. Oxford: Oxford University Press. Van Valin, Robert D. 1990. Semantic parameters of split intransitivity. Language 66, 221–60. —— and Randy LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Vendler, Zeno. 1957. Verbs and times. Philosophical Review 56, 143–60. Verhagen, Arie. 2002. From parts to wholes and back again. Cognitive Linguistics 13, 403–39. Verkuyl, Henk. 1972. On the Compositional Nature of the Aspects. Dordrecht: Reidel. —— 1989. Aspectual classes and aspectual composition. Linguistics and Philosophy 12, 39–94. —— 1993. A Theory of Aspectuality. Cambridge: Cambridge University Press. Verspoor, Cornelia Maria. 1997. Contextually-dependent lexical semantics. PhD dissertation, University of Edinburgh. Visser, Fredericus Theodorus. 1963–73. An Historical Syntax of the English Language. Leiden: E. J. Brill. Vogel, Irene. 1990. The clitic group in prosodic phonology. In Joao Mascaro and Marina Nespor (eds.), Grammar in Progress. Dordrecht: Foris, 447–54. von Stechow, Arnim. 1995. On the proper treatment of tense. Semantics and Linguistic Theory 5, 362–86. Washio, Ryuichi. 1997. Resultatives, compositionality and language variation. Journal of East Asian Linguistics 6, 1–49. Wechsler, Stephen. 1997. Resultative predicates and control. In R. Blight and M. Moosally (eds.), Texas Linguistic Forum 38: The Syntax and Semantics of Predication. Austin, Tex.: University of Texas Department of Linguistics, 307–21. —— 2001. An analysis of English resultatives under the event–argument homomorphism model of telicity. Proceedings of the 3rd Workshop on Text Structure, University of Texas, Austin. Revised version in Nomi Erteschik-Shir and Tova Rapoport (eds.), The Syntax of Aspect. Oxford: Oxford University Press. —— 2005. Weighing in on scales: A reply to Goldberg and Jackendoff. Language 81, 465–73. Wertheim, A. 1994. Motion perception during self-motion: The direct versus inferential controversy revisited. Behavioral and Brain Sciences 17, 293–311. Whitton, Laura. 2007a. What can be contrasted in Contrastive Reduplication? Paper presented at Annual Meeting of Linguistic Society of America. —— 2007b. The semantics of Contrastive Focus Reduplication in English: Does the construction mark prototype-prototype? MS, Stanford University. Wierzbicka, Anna. 1991. Cross-Cultural Pragmatics: The Semantics of Human Interaction. Berlin: Mouton de Gruyter.
references
471
—— 2007. Theory and empirical findings: A response to Jackendoff. Intercultural Pragmatics 4, 399–409. Williams, Edwin. 1984. Grammatical relations. Linguistic Inquiry 15, 639–74. —— 1994. Remarks on lexical knowledge. In Lila Gleitman and Barbara Landau (eds.), The Acquisition of the Lexicon. Cambridge, Mass.: MIT Press, 7–34. Winston, M., R. Chaffin, and D. Herrmann. 1987. A taxonomy of part–whole relations. Cognitive Science 11, 417–44. Wittgenstein, Ludwig. 1955. Philosophical Investigations. Oxford: Basil Blackwell. Wisniewski, Edward J., and Dedre Gentner. 1991. On the combinatorial semantics of noun pairs: Minor and major adjustments to meaning. In G. B. Simpson (ed.), Understanding Word and Sentence. New York: Elsevier Science Publishers, 241–84. Zubizarreta, Maria Luisa. 1998. Prosody, Focus and Word Order. Cambridge, Mass.: MIT Press. Zwicky, Arnold. 1994. Dealing out meaning: Fundamentals of syntactic constructions. Berkeley Linguistics Society 20, 611–25.
This page intentionally left blank
Index Abeille´, Annette 271 Abstract Syntax 36, 43n.1 Ackerman, Farrell 279 acquisition 324, 341, 420, 423; of productive vs. semiproductive rules 32–33, 243; of redundancy rules 81; of related forms 43 action modality 430–432 Action tier 306n.n actions 116; in 3D model 95, 98–99, 100 adjustment rules 78 A-er and A-er construction 293n.14 affectedness 180–182 affixes as interface rules/lexical items 19, 31, 44n.a, 61n.j, 146, 227 African-American Vernacular 23n.15 Afrikaans compounds 76–77, 80 agentive nominals 430–431 Aikhenvald, Alexandra 341n.f aktionsarten 142 Alboiu, Gabriela 368n.18 (all) X-d out construction 33, 242–243, 245 American Sign Language double construction 380–381 A-movements 27 anaphora 248 Anshen, Frank 277 ape sign language 421, 423 aphasia 420 appositives 330–335, 339, 341n.f argument structure 10–11, 15, 234; distinctness of syntactic and semantic 17n.12; in constructions 281, 434–437; linking 44, 287, 300–312;
syntactic 232, 251–252, 259, 262–263, 268, 271–274, 277; with aspectual particles 236–238 Aronoff, Mark 277 Asaka, T. 23 Asher, Nicholas 449–450 Aske, Jon 325 aspectual functions 165–168 Aspectual Interface Hypothesis 180, 183n.9 aspectual particles (up, away, on) 236–240, 257–259, 268–269 Atkins, Beryl 427 Atsugewi 24 auditory localization 117 Autolexical Syntax 3 Avrutin, Sergey 423n.7 axes, nonspatial 132–133; of events 196–197, 202, 208–213, 219–221; of irregular shapes 128–130; of objects 93–94, 102–104, 106, 123–133, 155–159, 209 axial vocabulary 123–133 Baayen, Harald 32, 38, 76n.s Bach, Emmon 143, 153, 158 back-formation 55 Baker, C. L., 324 Bantu applicative 273n.3 Barðdal, Jo´hanna 279 Barwise, Jon 89 Basic Variety 422–423 Bassac, Christian 443n.22 Baynes, Kathleen 422 Beck, Sigrid 385, 401n.12 Belhare 131n.12, 132n.13
474
index
Bender, Emily 398 beneficiary ditransitive 280 Berman, Ruth 418 Bialystok, Ellen 104, 124 Bickel, Balthasar 131n.12, 132n.13 Bickerton, Derek 420, 422 Biederman, Irving 91n.c, 94, 116 Bierwisch, Manfred 9, 124, 128, 130 binary branching 25–26, 248–249 binding 12, 24, 26, 27 Bloom, Paul 87, 120n.8, 123 Bloomfield, Leonard 37, 415 Boas, Hans 283n.d, 288, 293, 316n.q, 319–323 Bobalijk, Jonathan 379 Bochner, Harry 38, 57n.g bodily emission and function verbs 285, 305–306, 314–315 Bohnemeyer, Ju¨rgen 297, 315n.29 Bolinger, Dwight 9 Booij, Geert 23, 430n.13, 450 Borer, Hagit 281n.2 Boroditsky, Lera 132n.13 Borsley, Robert 271 Botha, Rudolf 76 boundaries 159–163 boundedness 137, 142–155, 176, 205, 293–295; correlation with telicity, 186–187, 190, 192–203; of events 217; vs. definiteness 151 Bowerman, Melissa 24, 33 bracketing mismatches 5 Brekle, Herbert 418, 429, 443n.22 Bresnan, Joan 3, 271, 273n.3, 280, 281 Brody, Michael 378 Brooks, Patricia 324 Burzio, Luigi 54n.f Busa, Federica 430–431 Carnap, Rudolf 20 Carnie, Andrew 383n.26 Carrier, Jill 261, 284n.e, 320 case 21, 411; abstract 27
case grammar 67–68 Casenhiser, Devin 288 categorization in 3D model 95, 97–98 causative 33, 42n.1; alternation 67–72; functions 11; nominalizations 71; in resultatives 289–290 c-command 248 Chaffin, R. 147 Cheney, Dorothy 7n.6 Chinese 24, 142n.2 Chomsky, Noam 3, 4n.2, 7, 8, 10, 20n.13, 29, 31, 35–37, 40–42, 48, 49n. d, 68, 71, 78, 83n.16, 110, 232, 271, 322, 346, 378 Citko, Barbara 386n.2, 411 Clark, Eve 418 Clark, Herbert 102, 104, 449 cliche´s, 18 clitics 359–362 cocomposition 432–433, 443–447 code-switching 329n.b coercion 23, 27, 136, 141, 150–152, 166–169, 176n.2, 177n.3; 433, 443, 447–448; as a construction 279n.b cognate object 285n.5 Cognitive Grammar 2, 3, 8, 11n.8, 19, 224 Collins, Chris 23n.15 competition between productive and semiproductive forms 33 composition of objects (COMP) 148–151, 440 compositionality 15–16, 20–24, 417, 425, 432–433, 450; enriched 24; Fregean 24, 135, 136, 151n.f, 153n.g, 275n.1 compounds 22, 32, 63–67, 80, 81, 154, 172, 225, 358, 401, 413–451; dvandva 416, 437–438; exocentric 66–67, 154, 415–416, 447–448; left-headed 416n.2; Russian and Persian 356n.8 conative construction 237
in d ex conceptual categories 52 n.e Conceptual Semantics 1, 6–14; many-to-many mapping in 172 Conceptual Structure (CS) 7, 9–16, 26, 85; criteria vis-a`-vis Spatial Structure 115–116 connectionism 31, 85, 173 Construction Grammar 18, 19, 222, 224–225, 242, 249n.f, 275, 279, 395–399, 409, 417, 419 Construction Morphology 450 constructional meaning 276 constructions 18, 172, 222–225, 271, 279–283 consumption verbs 176–179, 181–182, 201–204 continuity of motion 186 contrast 374–375 Contrastive (Focus) Reduplication 342–384, 399n.c, 410 control 12, 26, 27, 433 Cooper, Lynn 85 Copestake, Ann 433n.17 copy theory of movement, 378–379 core-periphery distinction 340–341, 395n.8 cost of referring to redundancy rule 46–47, 53, 58, 65, 69, 75–80 Coventry, Kenny 87 covering and filling verbs 197–198, 204, 216, 220 cranberry morphemes 65, 425–426 creation verbs 176–179, 201–203, 204 creativity in lexicon 81–82 Crick, Francis 186n.12 criterion of economy (CS/SR) 121, 130 criterion of grammatical effect (CS/ SR) 122, 131 criterion of interfacing (CS/SR) 122, 132–133 criterion of nonspatial extension (CS/ SR) 122–123 Croft, William 224, 279
475
Culicover, Peter 2, 4n.2, 12, 19, 20, 24, 25–28, 44, 67n.9, 225, 228, 233, 239n.9, 247, 248, 251n.a, 271, 276, 279, 280, 284n.e, 340n.e, 341n.f, 385, 395, 409, 433 Curtiss, Susan 421 Cutting, James 95 dative alternation 27, 272 Davidson, Donald 89, 189n.14, 206 De´chaine, Rose-Marie 383n.26 Declerck, Renaat 135, 140n.a, 142, 167, 182, 208 deep dyslexia 423 definiteness 332–333, 341 Dehe´, Nicole 247, 249n.10 deictic anaphora 107n.m De Lange, Joke 423n.7 den Dikken, Marcel 247, 271 denominal verbs (butter) 29, 33, 276, 390n.7, 398n.b depictive phrases 283 Deprofiled Object construction 301n.21 derived forms, unpredictability of 29 deverbal nouns (contribution) 29, 35–36, 42, 55–57, 426, 429–430; from verb-particle combinations 231–232 dimensional adjectives 124 dimensionality 136–137; of boundaries 159–163; of events 158, 165–169, 174, 188–191; of objects 155–163, 187–88; see also axes directed axes 157–159, 162–163 disappearance verbs 291–292 DiSciullo, Anna Maria 18, 226, 383n.26 discontinuous lexical items 74, 247, 253 discourse 449–450 distinguishers 85–86, 99 Distributed Morphology 78 n.u, 382 ditransitive construction 280–281, 288 Dixon, R. M. W. 341n.f do-support 224, 279n.b Downing, Pamela 418, 431, 434, 443
476
index
Dowty, David 123, 142, 167, 176, 181–184, 275 Dray, Nancy 347, 349 dual-coding hypothesis 115 Dutch 186, 194, 282 element of aggregate (ELT) 147–148 ellipsis 26, 230, 266, 267 Emmorey, Karen 113 Emonds, Joseph 74, 232, 234 empty nodes 26 epsilon dimensionality 160–162, 168 Erteschik-Shir, Nomi 378 Espinal, M.-T. 271 evaluation measure (full-entry theory) 41–42, 45–48, 50, 83 event structure 123, 142, 315; see also axes of events, dimensionality of events, subevents of construction, telicity event variable 206 event-process distinction 145 eventuality 158 evolution 86, 207n.20; see also protolanguage exception features 35, 37, 50, 54, 59, 63, 66, 78n.14, 83, 84 exclamative (P with NP!) 234–235 exemplar theory 32 explanatory adequacy 41, 110 expletive infixation 28, 227 extension of linear objects 11 extension, verbs of 197, 294–296 extracting functions 148, 151–152, 153, 155, 166, 172 fake reflexive (in resultative) 242, 272, 284, 298–299, 316, 322 family resemblance among constructions 283, 292, 324, 387 Fanselow, Gisbert 423 Farah, Martha 118 Fauconnier, Gilles 23, 205 features, syntactic 36
fictive motion 295n.17 Fiengo, Robert 143 figure-ground distinction 102 Fillmore, Charles 67–68, 70, 124, 222, 224, 271, 272, 279–280, 395, 427 Finnish 24, 140 Fitch, W. Tecumseh 7n.6 flow verbs 208–211 Fodor, Jerry 7n.5, 8, 68, 86, 89, 99, 173, 297 follow verbs 308–311 force 119–120 for-dative construction 272 formal semantics 2, 433n.17 frames of reference 24, 124–133 Fraser, Bruce 59n.6, 72, 74 free phrase order 27 Frege, Gottlob 20, 37 French 24, 105; resultatives 288n.10 Fried, Mirjam 279 Full Argument Realization in argument structure constructions 300–304, 314, 326 Full Entry Theory 37 functional definitions 100 Gapping 26 Garrod, Simon 87 Gazzaniga, Michael 422 Gelman, Susan 418 generalized cone 155–156 generalized transformations 36 Generative Semantics 2, 20, 27, 36, 42, 68, 110 Genie 421 Gentner, Dedre 417 geons 116 Ghomeshi, Jila 354, 410 Gibson, J. J. 89 Givo´n, Tom 420, 422 Gleason, Jean Berko 28 Gleitman, Henry 418, 420, 427, 428
in d ex Gleitman, Lila 104n.l, 272, 275, 418, 420, 427, 428, 438 glue logic 450 Goldberg, Adele 217n.23, 222, 224, 242, 260, 261, 264–265, 267, 271–272, 275, 278, 279, 283, 285, 288, 292, 293, 297, 300n.20, 301n.21, 315, 319, 320, 324, 395, 397, 399, 417 Goldin-Meadow, Susan 422 Government-Binding Theory 20, 27 Grammatical Constraint (on syntaxsemantics interface) 20 grammatical function tier 3, 27 Green, Georgia 319, 320 Gribanova, Vera 386n.2 Grimshaw, Jane 17, 45n.a, 182n.9 group nouns 144, 146, 149 Gruber, Jeffrey 11, 102, 122, 132, 140, 143, 183, 195–197 Guggu Yimithirr 24, 104n.l Habel, Christopher 117n.4 Haiman, John 383n.26 Hale, Kenneth 67, 276 Halle, Morris 31, 37, 38, 40, 41, 48–49, 59n.5, 62, 77–78, 82n.15, 83n.16, 382 Hankamer, Jorge 38 haptic representation 117 Harris, Randy 35 Harris, Zellig 36 Hausa, 126 Hauser, Marc 7n.6 Haveman, Alette 32 Hayes, Bruce 360–361 Head-Driven Phrase Structure Grammar 19, 27, 31, 67n.m, 260, 271, 276, 280, 325, 329n.b, 371 Head Parameter 423n.8 Head Principle 415–417, 434, 447 headedness of 3D model 94 headlines 423 Hebrew 410 hedges 351n.5
477
Herrmann, D. 147 Herskovits, Annette 87, 102 Higginbotham, James 16 Hindi 106 n.k; resultatives 325 Hinrichs, E. 123, 140, 176, 186n.13 Hoekstra, Teun 275, 325 Hohenhaus, Peter 343n.1, 347n.4, 350n.c, 356n.d home sign 422 Horn, Laurence 342, 347, 349, 353, 410 host of resultative phrase 285–286, 289, 308, 316; implicit 314–315 Huddleston, Rodney 385, 387n.5, 404, 424n.9 Hungarian 24, 140 Hyman, Larry 354 hypothetical lexical items 51, 53 see also potential lexical items idioms 17–18, 22n, 29, 38, 72–75, 80, 223, 226–227, 272, 280, 321–322, 324, 361, 365, 382, 386, 391, 394, 399n.c, 417, 419; based on resultative 262; based on way construction 264, 274; based on timeaway construction 254–255, 274; syntactically irregular 72; undergoing transformations 73; verbparticle 232–234, 239, 242–244, 246–247, 259–260 image schemas 115, 118 imagery, verbal 7n.6; visual 85 imagistic representations 110 imperfective paradox 137, 153 Impoverished Entry Theory 37; vs. fullentry theory 44–55 inalienable possession 96 inchoative 201; alternation 71; decomposition of 165 including functions 148, 151, 154–155, 172 incremental theme 176, 178, 180, 183–184, 201
478
index
inference 4, 8 infinitival VPs 26 information content of lexicon 37, 39, 42, 45, 83, 419, 425n.10; definition 47 information measure 46, 57–58, 60, 64, 69, 72, 76–80; formal statement 58 information structure tier 13 ingestion verbs 285 inherently resultative verbs 287–288, 318–319 inheritance hierarchy 31–32, 54n.f, 99, 316n.1, 324, 396–398 Inkelas, Sharon 372n.19 Intensive Reduplication 344n.3 interface linking 14–15 interface rules (correspondence rules) 90–92, 369 interface uniformity 20, 153n.g, 251n.a, 275n.i interfaces 2–3; acoustics to phonology 5–6; as ‘dirty’ 21–24, 114–115; as many-to-many mapping 172; conceptual structure to 3D model/spatial representation 86, 96–99, 101–107, 119–134; morphophonology to morphosyntax 57n.g; phonology to semantics 4n.1; phonology to syntax 371–373; semantics to perception and action 9; semantics to vision 4; syntax to semantics 2, 9, 20–24, 138; within phonology 2; within syntax 3 internal structure feature (+ i) 144–155 Interpretive Semantics 27 intonation 4n.1 Italian raddoppiamento 352–353 iteration (repetition) 23–24, 135, 139–140, 147, 166–169, 171, 254, 398, 402 Janda, Richard 283n.2, 323 January, David 132n.13
Japanese 24, 105, 288n.10, 325, 354 Jeannerod, Marc 118 Jespersen, Otto 426n.11, 434, 449, 440n.20 Johnson, Mark 122 Johnson-Laird, Philip 8, 102, 115, 124, 185 Joseph, Brian 283n.3, 320, 323, 401n.10 Kahnemuyipour, Arsalan 353 Kako, Edward 132n.13 Kannada echo reduplication 363, 376n.21 Katz, Jerrold 73 n.12, 99 Kay, Paul 222, 224, 271, 272, 279, 395 Kayne, Richard 248 Keyser, S. Jay 67, 243, 276 Kinande 354 Kiss, Katalin 378 Kittay, Eva 8 Klein, Wolfgang 422 Klima, Edward 234–235 Koch, Christof 117n.c Ko¨hler, Wolfgang 7n.6 Ko¨nig, Ekkehard 385, 386 Koopman, Hilda 380 Kosslyn, Stephen 85, 90, 94n.1, 116n.2 Kostic, Aleksandar 38, 76n.s Krifka, Manfred 176, 184–185, 187, 190, 191, 194, 196, 203, 205, 217 Kuhn, Thomas 99 Lakoff, George 8, 11n.8. 20, 35, 36, 42n.1, 45n.b, 50, 52n.e, 54, 67, 68, 71, 77n.13, 83, 84, 118n.5, 122, 132, 279, 293, 351n.5 Lambrecht, Knud 271, 279 Landau, Barbara 86, 103, 118, 124 Lane, Nancy 418 Lang, Ewald 124, 128, 130 Langacker, Ronald 3, 102, 118 n.5, 119, 122, 132, 140, 185, 272, 279, 281, 295n.16, 397, 399, 417, 427, 429
in d ex Language of Thought 7n.5, 89 language-vision connection 85–134; see also interfaces, conceptual structure to 3D model LaPolla, Randy 3 Larson, Richard 233n.2, 247, 408 Lascarides, Alex 433n.17, 449–450 Lasersohn, Peter 351 Lasher, Margot 104 Lau, Ellen 4n.2 Lees, Robert 35–36, 43n.1, 63, 417, 426, 428, 435 Lehrer, Adrienne 8 Lepschy, Anna 353 Lerdahl, Fred 6, 85 let alone construction 222, 280 levels of representation 89–92, 113–114; parsimony 110 Levelt, W. J. M. 32, 127–128 Levi, Judith 64n.1, 414, 417, 418, 428, 435 Levin, Beth 17, 45n.a, 141, 195n.16, 223, 236n.5, 237, 260, 261, 267, 270, 275, 285, 290, 293, 297, 299, 307n.23, 308, 313, 315–316, 325, 430n.13 Levinson, Stephen 24, 87, 104n.1 Lewis, David 150 lexical access 39, 45n.b lexical entries for aspectual verbs 166; for axial vocabulary 128–133; for constructions 260; for idioms 272; for spatial prepositions 163–165; including spatial representation 99–101, 120–121; verb-particle constructions 245–246 Lexical-Functional Grammar 3, 27, 67n. m, 271, 276, 280, 325 lexical insertion 227n.a, 233, 328n.a, 329n.b, 369–370; of idioms 73 lexical items 6; definition of 226, 277 lexical relatedness 43, 46 lexical rule (HPSG/LFG) 37, 223, 260–261, 270, 274–277
479
Lexicalist Hypothesis 40, 42, 43, 71 lexicalization patterns 105 lexicon 1–2, 14–20, 28–34, 38, 329n.b, 396, 398–399, 419; as interface 138; constructions in 242–244; vs. grammar, 19–20, 31, 37–38, 243 Li, Peggy 104n.1 Lidz, Jeffrey 344n.a, 363 Lieber, Rochelle 423n.8, 424n.9, 430n.13, 434, 442, 450 light emission verbs 291 Lightner, Ted 35 Lillo-Martin, Diane 380–381 linear order 21, 25–27, 248 linguistic semantics 9 Linguistics Wars 35 Link, G. 153 load verbs 213–217 locative inversion 234–235 Logical Form 27, 207, 408, 412 Lombardi, Linda 237n.6 long-distance dependencies 27 long-term memory 17, 95, 226, 277 Macnamara, John 85, 87, 88, 89 macrorole tier 306n.n mainstream generative grammar 3, 4n.2, 19, 20, 22, 73n.p, 78n.u, 135, 271, 276; vs. constructions 282 Maling, Joan 269 Mandarin, 132n.13 Mandler, Jean 115, 118n.5 manner adverbs 230–231, 253 Marantz, Alec 37, 78n.u, 264–267, 281n.2, 372, 382 Marchand, H. 344n.b markedness 41, 48, 83n.16 Marr, David 85, 90, 91–93, 95, 96, 98, 103n.h, 107, 114, 116, 117, 120, 130, 155–156, 187–188, 209 mass-count distinction 122–123, 137, 152–153, 177; parallel to aspect 143–145, 153
480
index
Matsuoka, Kazumi 381n.25 Matsuyama, Tetsuya 385, 386, 404, 405, 407–408 McCarthy, John 28, 345, 372 McCawley, James 36, 42n.1, 68, 297 McClelland, James 31 meaning postulates 173 measure phrases 131 measuring out 175–185, 190–192, 203, 207–209, 265 Medin, Douglas 31 mental rotation 116n.2 mentalism in semantics 7 Merge 73n.p, 329n.b Mervis, Carolyn 120, 283, 293 metaphor 11n.8, 447 Michaelis, Laura 222, 271, 279, 280 Michif demonstratives 381 Michotte, A. 120 middle construction 181, 284n.4, 284n.e Miller, George 8, 102, 124, 185 Millikan, Ruth 100n.f, 276n.j, 431 Minimalist Program 4n.2, 20, 25n.16, 27, 228, 271, 276, 329n.b, 346, 378–382, 408 Mishkin, Mortimer 118 model-theoretic semantics 7, 8, 108n.m, 176n.1; see also formal semantics modifier-head relations 12 modularity 86 Moody, Simanique 23n.15 Moravcsik, Edith 344n.3, 345 morpheme structure rules 48 morphological relation without semantic relation 58, 60–62, 65 morphology 17, 19, 28–29, 35–38, 40–84; derivational 322, 357–358, 362, 372; inflectional 77–78, 356–358, 366, 372, 421 Moscoso del Prado Martı´n, Fermin 38, 76n.s, 80n.w Moshi, Lioba 271, 273n.3 motion 104–105, 119, 186, 223
motion verbs 178–182, 192–195, 204, 218, incorporating path vs. manner 24; manner of 104–105 Mourelatos, A. 142 Moyse-Faurie, Claire 385, 386 Mulder, R. 275 Murphy, Gregory 31, 293 music cognition 6, 85 Mutaka, Ngessimo 354 Nadel, Lynn 118 Nakamura, Tsuguro 325 Narasimhan, Bhuvana 106n.k, 128–130, 281, 325 natural kinds 98 necessary and sufficient conditions 8 Neisser, Ulric 97 Nespor, Marina 360 nonlinguistic thought 7 nonspatial semantic fields, parallels to space 122 Nooteboom, Sieb 32 Norwegian 282 NP and S construction 280 NPN construction 29, 32, 385–412, 419 numeration (in MP) 329n.b Nunberg, Geoffrey 23, 73n.o, 141, 247 O’Connor, Mary Catherine 222, 271, 279, 395 O’Keefe, John 118 object constancy 116–117 object shape 86 Oehrle, Richard 280, 385, 407–408 Olsen, Susan 437 Olson, David 104, 124 one-anaphora 335 ontological categories 108n.m, 143–145, 158, 172, 189 Optimality Theory 324, 345 orientation 11 orientation verbs 181, 203
in d ex Pacin, Phyllis 66n.8 Paivio, A. 115 Panini 415–416 Papafragou, Anna 104n.l parataxis 421, 422 Partee, Barbara 9 partitive construction 153–155 part-whole relation 96, 119, 136, 153–155, 441 passive 27, 313n.27 past tense 33, 77–78; as lexical entry 19; debate 31, 38; English irregular 28–29 Path 10–11, 15, 119, 158, 163–165, 189; in 3D model 104–107; without motion 186n.13 Peperkamp, Sharon 361n.10 Perdue, Clive 422 Pereltsvaig, Asya 353 performance verbs 178, 181–182, 200–201, 204 Perlmutter, David 67n.9 Perry, John 89 Persian 353–354 perspective (on scene) 117 Peterson, Philip 98 Petronio, Karen 380–381 PF deletion 27 Pfau, Roland 347n.4 Phillips, Colin 4n.2 phonological features 14 phonotactics, 48, 279n.b, 357 phrase structure rules, as lexical items 19, 38; as redundancy rules 74–75, 81; for VP 28 physics 14, 172 Pi, Tony 385, 394, 401n.12, 407–408 pidgins 421–422 Pinker, Steven 7n.6, 8, 9, 31, 38, 91, 213, 216, 243, 277, 324 Place 10–11, 102–103, 107, 119, 158, 189 Platzack, Christer 135, 142
481
plural 24, 28, 136, 140, 143–147, 166, 370, 408; as axis of event 204–207, 211–213 Polish NPN construction 411 Pollard, Carl 45n.b, 52n.e, 260, 271, 280 Poser, William 354 possession verbs 198–199 possible worlds 7 Postal, Paul 4n.2, 23n.15, 110, 328n.a, 329n.b Postma, Gertjan 385, 386, 402n.12, 407 potential lexical items 48–50, 64 pragmatic halo 351 pragmatics (vs. semantics) 8–9, 24, 201, 204, 214, 420, 431 predication 247–248 prefix-stem (Latinate) verbs (transmit) 59–62, 80 prepositions, spatial 102–104, 124 primal sketch 91–92, 114 primitives, justification of 13–14; semantic/conceptual 90, 136, 172–173, 191–192, 203; visual 92 Prince, Alan 31, 38, 277, 345, 372 Principles and Parameters 324, 326 PRO 26 processing 91 profiling 429–430, 435, 437–438 progressive aspect 152, 165 projection rules 27 promiscuous meaning 428 proper function 100n.f, 276n.j, 431–432, 439, 442, 444 proper names 348 proprioception 117 Prosodic Hierarchy 361 Prosodic Morphology 345 prosodic words 361 prosody 28, 336, 337n.d, 341, 344n.3, 345, 356n.8; in contrastive reduplication 366–368 protolanguage 400n.d, 420–425
482
index
prototypicality 347–352 Pullum, Geoffrey 385, 387n.5, 402, 404, 424n.9 Pustejovsky, James 8, 100n.f, 123, 142, 168, 185, 195n.16, 276n.j, 298n.19, 432, 443n.22 Putnam, Hilary 86, 120, 173 Pylyshyn, Zenon 109–111 qualia structure 100n.f, 276n.j, 443n.22 quantification 206–213, 219, 407–409 Quantifier Raising 207 quantifier scope 115, 211–213 raising 27 Randall, Janet 261, 284n.e, 320 Raphael, Bertram 31 Rapoport, Tova 141, 195n.16 Rappaport Hovav, Malka 17, 45n.a, 223, 260, 261, 267, 270, 275, 285, 290, 293, 297, 299, 307n.23, 308, 313, 315–316, 325, 430n.13 readjustment rules 48 recursion 36; in 3d model 93–94 redundancy 419 redundancy rules 30, 37–38, 43–55, 82–83; as lexical entries 57n.g; separation into morphological and semantic 55–59, 63, 69 reduplication 5n.3, 19, 343–345, 372–373, 410 reference 7–8, 92n.d; to actions, Places, Paths, amounts 108n.m reference transfer 23–24, 141 referential tier 13 referentiality 266–267 reflexive pronouns 12 Relational Grammar 27 relative clauses 12 repetition 140–141 see also iteration resultative 29, 31, 32, 200, 244, 250, 260n.2, 261–265, 268–270, 273; 278–326
resultative phrase (RP), definition 283 retinotopic representation 114, 117–118 Richards, Norvin 378–379 right hemisphere language 422, 423 Rizzi, Luigi 378 Role and Reference Grammar 3 Romance 24 rootless derived forms 49–55, 61, 65, 80 Rosch, Eleanor 86, 120, 283, 293, 354 Rosen, Nicole 381 Ross, John Robert 36 Rothstein, Susan 293 route directions 127–128 rules of construal see coercion Rumelhart, David 31 Russell, Bertrand 37 Russian 353 Ryder, Mary Ellen 419n.5, 420, 434 Sadock, Jerrold 3, 418, 423 Sag, Ivan 45n.b, 52n.e, 73n.o, 247, 260, 271, 279, 280, 398 Schank, Roger 185 Schein, Barry 206 schematization of objects 157, 159–163 Schiffer, Stephen 173 Seland, Grete 282 Selkirk, Elizabeth 361n.10, 415, 434 Semantic Coherence Principle in argument structure constructions 304–309, 311, 326 semantic memory 31 semantic networks 8 semantics, combinatoriality 20 semiproductive vs. productive rules 31–32, 79 semiproductivity 28–34, 38, 227–228, 233, 240, 276–277, 279n.a, 293n.j, 316–324, 390–391, 396–397, 418 serial verb constructions 341n.f Sethuraman, Nitya 288 Seyfarth, Robert 7n.6
in d ex Shepard, Roger 85 sign, Saussurean 224 signed languages 113n.1 Simpler Syntax 2, 4, 25–28 Simpson, Jane 261, 284 Slack, Jon 87 slack regulators 351 small clause 247–248 smear verbs 213 Smith, Edward 31 Snyder, William 326 social predicates 11 social status 133 sound emission verbs 290–292, 298–300, 326 soundþmotion construction 18, 22–23, 27, 31, 270, 273, 275, 290–292 Spanish 105, 353; resultatives 325 spatial concepts extended to nonspatial fields 11 spatial coordinate systems 103–104 spatial layout 117 spatial relationships in 3D model 102–107 spatial representation 114; criteria for, 116–119 specifier of particle 233, 246; of PP 230, 235n.4 spray-load verbs 70, 179, 213–217 stage-level predicates 430 Standard Theory 36 Stanley, Richard 48 State vs. Event 158 Steinbeck, John 343 Sˇtekauer, Pavol 442, 450 stereotypes in categorization 86, 95, 100; in lexical entries 120 Steriade, Donca 54n.f storage vs. computation of productive forms 32 Stowell, Tim 16 ‘strawberry’ morphemes 65, 425 stress 344n.3
483
structure-preserving (sp-) binding 191–192, 196–197, 198–207, 216, 217; formalization 218–221 subcategorization see argument structure, syntactic subcategorization, in idioms 74 subevents of construction 301–304, 310–311; of resultative 286–293, 295 Subramanian, Uma 402 swarm alternation 70 Swedish 282 symmetric predicates 272, 275 synecdoche 154 syntactic nuts 19, 328n.a, 385 Tagalog 372 Takami, Ken-ichi 288n.10, 319, 325 Talmy, Leonard 23, 24, 86, 102, 105, 106, 119, 122, 132, 135, 136, 139, 140, 142, 143, 147, 155n.i, 177, 272, 295n.16–17 Tarski, Alfred 89 taxonomic relations 115 taxonomies of concepts 31; of categories 99 Taylor, John 21n.14, 293 telicity 137, 174–221, 236, 244, 254–255, 259, 269, 408; in resultatives 293–296, 317, 322; see also boundedness Teller, Paul 124 temporal relation in resultative 296–298, 316 temporal structure 136 ten Hacken, Pius 414, 418n.3 Tenny, Carol 175–176, 180–183, 185, 202, 211n.21, 215–217, 265, 293 tense as syntax-semantics mismatch 22 Terrace, Herbert 421 the N-E construction 327–341 thematic roles 15; protoroles 183–184 theory of mind 11 theta-grid 16
484
index
theta-marking 17 Thompson, Sandra 35 3D model 85–86, 91–96, 110, 114, 120 tiers, conceptual structure 13; phonological 2; semantic 3; syntactic 3 time 123, 132, 186–191, 267 time expressions 252–253 time-away construction 240–242, 250–277, 280, 282, 288n.10 Toivonen, Ida 222n.1, 246, 282 Tomasello, Michael 33, 322, 324 tough-movement 253, 266 trace 27 transformations 35–37, 42–43, 67n.m, 68, 75, 81, 229; in compounds 63, 66 Travis, Lisa 372n.19, 379n.22 triplication 344n.3, 403 truth-conditional semantics 89, 173 Turkish 38 Tversky, Barbara 115, 117, 127–128, 272 2½D sketch 91–92, 94n.1, 114, 118 type-token distinction 12–13, 97, 115, 120 Tzeltal, 104n.l Ungerleider, Leslie 118 unification 73n.p, 227n.a, 384 Universal Grammar 138 Universal Grinder 151–153, 164–167 Universal Packager 150; 279n.b Vaina, Lucia 95, 98, 100 Valimaa-Blum, Riitta 283n.3 Vallduvı´, E. 368n.18 van der Zee, Emile 87 Van Valin, Robert 3, 293 Vasic, Nada 423n.7 Vata reduplication 380 Vendler classes 166–168 Vendler, Zeno 137, 142, 166–168, 176
Verb Subordination ArchiConstruction 273–275, 316n.r verb-particle construction 29, 74, 226–249, 252–253, 259–260, 270; idioms 273 Verhagen, Arie 282 Verkuyl, Henk 123, 135, 142, 168, 176, 179, 181, 184–187, 191, 204–206, 208, 212n.22, 217, 221 Verspoor, Cornelia 285, 320 vision 4, 85–111 Visser, Fredericus 313 visual imagery 118; differences from 3D model 109 ‘visual meaning’ 92 Vogel, Irene 360, 361n.10 von Stechow, Arnim 282, 326, 385, 401n.12 VP shell 233n.2, 247 Washio, Ryuichi 288n.10 Wasow, Thomas 73n.o, 247, 398 way construction 222–225, 250, 261–267, 273, 280, 282, 326; parallels in Swedish, Norwegian, Dutch 282; vs. all the way 263 Wechsler, Stephen 285, 293, 316n.q, 319, 320–321 Weerman, F. 32 Wertheim, A. 131n.10 ‘what’ and ‘where’ systems 118 Whitton, Laura 350n.c, 384n.h Wierzbicka, Anna 21n.14, 352–353 Wijnen, F. 32 Williams, Edwin 16, 18, 226, 280, 383n.26, 385, 401 Wiltschko, Martina 383n.26 Winston, M. 147 Wisniewski, Edward 417 with-theme construction 251n.b Wittgenstein, Ludwig 283, 293 word grammar vs. phrasal grammar 82 word, definition of 17–18
in d ex words as interface rules 17 wugs test 28 X-bar theory 25, 36, 328n.a, 331, 362 Yip, Moira 408 Yngve, Victor 150n.6
485
Yoneyama, Mitsuaki 105 you-can-put-anything-you-want-beforethe-head construction 383n.26 zero rule (dimensionality) 157 Zoll, Cheryl 372n.19 Zubizarreta, Maria Luisa 368n.18, 378 Zwicky, Arnold 222, 279