STUDIES IN LINGUISTICS Edited by Laurence Horn Yale University A ROUTLEDGE SERIES STUDIES IN LINGUISTICS LAURENCE HORN, General Editor ELLIPSIS AND WA-MARKING IN JAPANESE CONVERSATION John Fry WORKING MEMORY IN SENTENCE COMPREHENSION Processing Hindi Center Embeddings Shravan Vasishth INPUT-BASED PHONOLOGICAL ACQUISITION Tania S.Zamuner VIETNAMESE TONE A New Analysis Andrea Hoa Pham ORIGINS OF PREDICATES Evidence from Plains Cree Tomio Hirose CAUSES AND CONSEQUENCES OF WORD STRUCTURE Jennifer Hay THE PHONETICS AND PHONOLOGY OF GUTTURALS A Case Study from Ju\’hoansi Amanda Miller-Ockhuizen TRUE TO FORM Rising and Falling Declaratives as Questions in English Christine Gunlogson LEXICAL, PRAGMATIC, AND POSITIONAL EFFECTS ON PROSODY IN TWO DIALECTS OF CROATIAN AND SERBIAN An Acoustic Study Rajka Smiljanic
LENITION AND CONTRAST The Functional Consequences of Certain Phonetically Conditioned Sound Changes Naomi Gurevich THE INFLECTED INFINITIVE IN ROMANCE LANGUAGES Emily Scida SYNTACTIC FORM AND DISCOURSE FUNCTION IN NATURAL LANGUAGE GENERATION Cassandre Creswell THE SYNTAX-INFORMATION STRUCTURE INTERFACE Evidence from Spanish and English Eugenia Casielles-Suárez THE UPS AND DOWNS OF CHILD LANGUAGE Experimental Studies on Children’s Knowledge of Entailment Relationships and Polarity Phenomena Andrea Gualmini CATEGORY NEUTRALITY A Type-Logical Investigation Neal Whitman MARKEDNESS AND FAITHFULNESS IN VOWEL SYSTEMS Viola Miglio PHONOLOGICAL AUGMENTATION IN PROMINENT POSITIONS Jennifer L.Smith ENRICHED COMPOSITION AND INFERENCE IN THE ARGUMENT STRUCTURE OF CHINESE Ren Zhang
ENRICHED COMPOSITION AND INFERENCE IN THE ARGUMENT STRUCTURE OF CHINESE Ren Zhang
ROUTLEDGE New York & London
Published in 2005 by Routledge Taylor & Francis Group 270 Madison Avenue New York, NY 10016 Published in Great Britain by Routledge Taylor & Francis Group 2 Park Square Milton Park, Abingdon Oxon OX14 4RN © 2005 by Taylor & Francis Group, LLC Routledge is an imprint of Taylor & Francis Group This edition published in the Taylor & Francis e-Library, 2005. “To purchase your own copy of this or any of Taylor & Francis or Routledge's collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” ISBN 0-203-49595-0 Master e-book ISBN
ISBN 0-203-58083-4 (Adobe eReader Format) International Standard Book Number-10:0-415-97100-4 (Print Edition) (Hardcover) International Standard Book Number-13:978-0-415-97100-3 (Print Edition) (Hardcover) No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording or in any information storage or retrieval system, without written permission from the publishers. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging-in-Publication Data Catalog record is available from the Library of Congress
Taylor & Francis Group is the Academic Division of T&F Informa plc. Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the Routledge Web site at http://www.routledge-ny.com
Table of Contents Preface Acknowledgements
CHAPTER INTRODUCTION: VARIATION IN ARGUMENT ONE EXPRESSION
viii x
1
CHAPTER CONSTRUCTIONS, CONCEPTUAL STRUCTURES AND TWO COMPOSITIONALITY
13
CHAPTER THE EAT RESTAURANT CONSTRUCTION: LICENSING THREE UNSELECTED COMPLEMENTS
32
CHAPTER CONCEPTUAL INFERENCE AND PREDICATE FOUR TRANSFER
83
CHAPTER CONCLUDING REMARKS FIVE
148
References
153
Index
163
Preface
This book is a minimally revised version of my doctoral dissertation completed at York University in 2002. The empirical focus of this work is a set of non-canonical argument structure phenomena in Mandarin Chinese in which the external and the direct internal argument bear a non-transparent semantic relation to the eventuality expressed by the verb of a sentence, as with Xiaowang sile fuqin (‘Xiaowang died (his) father’: ‘Xiaowang’s father died’) and Xiaowang jingchang chi zhejia changuan (‘Xiaowang often eats this restaurant’: ‘Xiaowang often has his meal in this restaurant’) Such phenomena pose a challenge to the Projectionist view of the relation between the lexicon and the syntax, according to which the number of syntactic arguments is largely determined by the lexical semantics of the verb. I propose a semantic account of such ‘unselected’ arguments within the framework of Conceptual Semantics (e.g. Jackendoff 1990) and situate my proposals within a general constructionist perspective of linguistic theory (Jackendoff 2000). Discussing a wide range of data, I show that the ‘unselected’ internal and external arguments are licensed in terms of an enriched view of conceptual combination and inferences and that no abstract syntactic mechanisms need to be invoked for such phenomena. Firstly, a schematic conceptual structure of the eat restaurant type of examples is proposed, which can be roughly glossed as ‘one entity undertakes an event by affecting another entity in some way’. Based on such a general conceptual structure as well as the lexical conceptual structures of the elements in the construction, I propose three conditions governing the semantics of the construction: (1) the event is brought into existence by an agentive quale (Pustejovksy 1995) in which some default entity x is affected; (2) the unselected complement expresses an entity m with a telic quale that caters to the need of the event in (1); and (3) there should be no redundancy between x and m. Exceptions to such generalizations are treated under a Preference Rule System of defeasible typicality conditions, based on Lascarides and Copestake (1998). Secondly, three constructions, the Possessum-object construction, the retained-object passive construction and the locative inversion construction, are subsumed under the category of ‘unselected’ subjects and are given a uniform treatment in term of conceptual inferences and Noteworthiness Condition, drawing on Nunberg’s (1995) work on predicate transfer. Both the Possessum-object construction and the locative inversion construction share a conceptual structure in which a Thing or a Place is predicated over by a State as its property, whereas the conceptual structure of a retained-object passive is defined in schematic action-tier terms as one entity z reacting to an event involving its Possessum, which in turn affects z lexically or contextually. The conceptual structures of
these constructions are linked to those of their canonical alternations by two types of inference rules involving set inclusion and inferences from a spatial event with a Place adjunct or a Location argument. Constraints on predicate expressions follow from the conceptual structures of the constructions. In particular, the predicator in Possessumobject and locative inversion constructions must express a State. This is made possible via the semantic contributions of particular aspectual particles such as le and guo. The study shows that an elaborate semantic theory, embedded within a parallel architecture (as manifested in a version of Construction Grammar, for example), offers a more adequate solution to the non-canonical argument structure phenomena than approaches that adopt a syntactocentric outlook. Thus by virtue of the research reported here a valuable alternative is offered to the syntactic approaches in the formal studies of Mandarin Chinese. The starting point of this research is a critical reconsideration of work in Chinese syntax that adopts the syntactocentric perspective, and the result comes close to the claims and the insights associated with construction-based cognitive linguistics, while remaining a certain distance from the latter school of theorizing. In retrospect, there are certain claims in this work that could be modified as well, if I went deeper into Cognitive Linguistic frameworks. Nevertheless, I have decided not to make any substantial change in the book version of the dissertation. I would like to take this opportunity to thank Larry Horn, Diane Massam, Ruth King, Peter Avery and Michael Cummings as well as my editor Mark Henderson at Routledge, without whose patience, suggestion and encouragement this book would never have come out in print. My students at Nanjing University have been very helpful with sundry matters in the course of preparing the book version of this work. March, 2004 Nanjing University Nanjing, China
Acknowledgments
As a linguist of Chinese origin, I have followed the footsteps of many predecessors in searching in the ‘West’ for a theory that could account for the subtleties of the Chinese language. In this long process I have incurred many intellectual and personal debts. Firstly, I was very lucky to have Professor Robert van Valin, Jr. to serve as my external examiner. I have long admired his erudition in many languages and found his own linguistic theory of great potential in exploring the properties of Sinitic languages. I want to thank him for his encouraging comments, perceptive questions and valuable suggestions. Throughout the past five years, my academic life in Toronto has been blessed with two mainstays from two institutions. Ruth King first suggested that I broaden the scope of my research interests. Her advice led me into the fascinating areas of syntactic and semantic studies, which I have since found it hard to give up. I owe it to her for the general orientation of this dissertation. As a good example of a linguist straddling different research areas including syntactic theory, sociolinguistics and women’s studies, Ruth has activated my own interest in the many areas of sociolinguistics, including language and gender. Most importantly, Ruth has never failed to help me see my own strengths and believe in myself in times of academic and career impasses. I wish I could properly convey my gratitude to her in words. If Ruth has been an overall director of my academic growth, Diane Massam at the University of Toronto is directly responsible for my coming of age in theoretical linguistics. Diane’s courses in Intermediate and Advanced Syntax enabled me to reach a thorough understanding of Chomskyan syntactic theories. The training I enjoyed under Diane’s guidance serves as a crucial foundation for me to enter into monostratal linguistic theories and to appreciate larger cognitive science issues. I owe my skills in linguistic analysis and argumentation to Diane, who in the last few years spent countless gratuitous hours going through my every conference paper and offered many valuable suggestions. The final version of the dissertation has again benefited a lot from Diane’s constructive criticisms on my argumentation and presentation. Obviously, this dissertation would not have come into the world without the many thought-provoking works authored by Ray Jackendoff, whose influence should be apparent on practically every page. I would like to thank him for making my trip in 2001 as a visiting scholar to the Linguistics and Cognitive Science Program of Brandeis University a pleasurable and fruitful experience. That trip directly helped clinch the theoretical framework of this research. Subsequent communications with him also clarified some further issues. While Jackendoff may not completely endorse the way I
implement his ideas, I flatter myself to think that some of Jackendovian traits in approaching linguistics, such as non-parochialism, breath of knowledge, clarity and force in style, have rubbed off on me in a modest way. I am grateful to the linguists who responded to my query on the Linguist List on nonselected complements and pointed me to important literature, including Tom Givón, Paul Hopper, Sabine Bergler, Georges Rebuschi, Kiyoko Takahashi, Ressy Ai and Diane Massam. Asya Pereltsvaig (McGill), Jonah Lin (UC Irvine), Tao Hongyin (UCLA) and Gao Qian (Ohio State) sent me their papers related to the topic, which all served as inspirations to my Chapter 3. The general direction of this research took shape during my attendance at the 1999 LSA Institute held in the University of Illinois at UrbanaChampagne (UIUC). I thank Adele Goldberg for offering a research affiliate position and for spending time having discussions with me. At the risk of unwanted omission, I would also like to express my indebtedness to a wider circle of linguists, including Michael Cummings, Bill Greaves, Jim Benson, Barry Miller, Bob Binnick, Chris Butler, Dick Hudson, Heidi Harley, Bill Croft and Urpo Nikanne. My life in Toronto, as my first experience in a different culture, would have been harder without the gracious help and support from many warm-hearted people and friends. Prominent on my mind are the following scholars and friends: Michael Cummings, Bill Greaves, Ray Ellenwood, Jan Pearson, Ruth King, Donna Lillian, Ed and Marg Boldt, Dave Kauffman, Seth Blumenthal and Hendrick Tan. Other friends, old and new, and spread in several countries, have showed their support in various ways. They include Meng Zhiyong, Feng Gang, Wang Yongqun, Zhou Guojun, Xiao Li, Wang Feng, Ouyang Jingsong, Chen Jingsong, Xu Maolei, Wu Yuwen, Tang Duoyi, Li Jun, Zhang Zhe, Niu Kangsheng, Peter Au, Chen Shengli, Li Zheng and Denis Yu. A special thanks also goes to Hu Zhuanglin (Peking University) for trust, support and understanding. Finally, how could all these be possible without the practical and moral support of my parents (Qin Jin and Zhang Weiping) as well as my sisters and brothers-in-law? They could not completely understand why I had to leave home for this faraway land, but what matters is that they let me be myself.
ENRICHED COMPOSITION AND INFERENCE IN THE ARGUMENT STRUCTURE OF CHINESE STUDIES IN LINGUISTICS
CHAPTER ONE Introduction: Variation in Argument Expression
In theory-neutral terms, it is largely agreed that the grammatical form of a sentence expresses a Situation1, or an eventuality in the sense of Bach (1986), with a number of participants or semantic arguments. Very often it appears that a particular conceptual situation can be expressed in varying syntactic forms, as seen in the semantic relation between a passive construction and its active counterpart, as well as many other cases of verb frame alternations (see Levin 1993). Let us call such linguistic variations in the expression of event structure argument structure2 phenomena, which lie at the core of any theory of linguistic structure. The empirical focus of this dissertation is a subset of argument structure phenomena in Mandarin Chinese, as illustrated in the following: (1) Xiaowang gang sile fuqin just dieLE father “Xiaowang has just had his father die (on him).” (2) Xiaowang jingchang chi zhejia canguan often eat thisCL restaurant “Xiaowang often has is meal in this restaurant.” In such examples, the external and the direct internal argument bear a non-transparent semantic relation to the eventuality expressed by the verb. The person who died in (1) is not the external argument Xiaowang, and the internal argument zhejia canguan ‘this restaurant’ in (2) is not what is eaten by Xiaowang, as would be expected of these syntactic positions in typical cases. In both cases it is possible to express the same idea more transparently, as shown in the following: 1. The ontological categories of Conceptual Semantics are presented in capital letters. 2. There is a technical sense of the term argument structure assumed in a variety of frameworks as an abstract lexical representation separate from lexical semantic representations (see e.g. Williams 1984; Grimshaw 1990; Rappaport, Laughren and Levin 1993; Bresnan 2001). Following Jackendoff (1990), I do not assume such a level but still use the terms internal/external arguments for expository purposes.
Enriched composition and inference in the argument structure of Chinese
2
(1) a. Xiaowang de fuqin gang sile DE father just dieLE “Xiaowang’s father has just died.” (2) a. Xiaowang jingchang zai /dao zhejia canguan chi fan often LOC./go thisCL restaurant eat meal “Xiaowang often has his meal in this restaurant/often comes to this restaurant to have his meal.” Argument structure phenomena raise important issues concerning the relations between grammatical form and meaning. Specifically, what are the mapping relations between the structure of the Situation linguistically construed by a sentence and the particularities of the grammatical form of the sentence? Where are such mapping principles located in the architecture of a theory that accounts for the linguistic knowledge of a speaker? Do the alternative grammatical patterns for expressing a Situation indeed share the same semantic structure? How are the arguments in a construction licensed? Under traditional verb-centered views of argument structure (see section 1.1), certain non-canonical cases as exemplified in (1) and (2) are especially relevant, as the relations between the verb and arguments in these cases are by no means straightforward and thus pose a challenge to any theory of argument expression and licensing. The specific goal of this dissertation is thus to propose an account of such non-canonical argument structure phenomena in Mandarin Chinese. I argue that the ‘unselected’ arguments in constructions like (1) and (2) are naturally licensed under a theory of enriched semantic composition and a theory of conceptual inferences, and that there is no need to assign an abstract syntactic structure to such cases to account for certain implicit meanings, nor is it desirable to posit syntactic transformations to take care of what turn out to be conceptual inferential relations. The analyses that I propose aim to account for the full range of data. I point out empirical limitations, as well as certain theoretical clumsiness, associated with a number of syntactic proposals. Thus I intend the study to be an exercise in semantics. Crucially I focus on semantic representations and semantic compositionality as manifested in the apparently simple and formally austere syntactic structures of Mandarin Chinese. The conceptual processes and inferences are explicated on the basis of a system of enriched semantic representations within the framework of Conceptual Semantics as developed by Jackendoff (e.g. 1976, 1983, 1990, 1996a, 2000). Insights from compatible semantic theories will often be referenced and drawn upon as well, especially the Generative Lexicon (Pustejovsky, 1991, 1995 et seq.) as well as works associated with Cognitive Semantics (esp. Langacker 1987, 1990, 1991, etc.). In addition, this research is situated within a more general conception of linguistic structure, one that is akin to but distinct from much recent construction-based work stemming from different frameworks, including, among others, Construction Grammar (e.g. Goldberg 1995; Michaelis 2000; Croft 2001), Cognitive Grammar (Langacker 1987), Head-Driven Phrase Structure Grammar (HPSG) (Sag 1997; Ginsburg and Sag 2001) as well as certain works within the Chomskyan framework (e.g., Hoekstra 1992; Ghomeshi and Massam 1994; Ritter and Rosen 1998, among others; see Levin and Rappaport Hovav 1996 for an overview). The research results presented here underscore two theoretical points. First, syntactic patterns are often underspecified concerning their
Introduction: Variation in argument expression
3
meaning, which is a revised Constructional view advocated by Jackendoff (1996b, 1997, 2000). Second, syntactic machinery can be greatly simplified under a rich system of semantic representations, which is currently being advocated by a number of linguists such as Langacker (1987), Jackendoff (1990), Pustejovsky (1995) and Culicover (1998), as well as many functional linguists. However, in this dissertation, I don’t have particular proposals to make with regard to syntactic representations per se in a theory of Chinese grammar, and I simply assume a simplest version of phrase structure. Given the general Constructionist framework which I adopt (see Ch. 2), I wish to achieve a larger methodological goal in this thesis. I wish to demonstrate that certain important semantic issues can only begin to be raised under a parallel architecture of linguistic theory (cf. Sadock 1991), where semantics is taken as an independent generative component on a par with phonology and syntax. The parallel conception contrasts with the Chomskyan framework in which the syntax is the only generative component, an outlook Jackendoff (1997) terms syntactocentrism. Furthermore, since argument structure phenomena are central in any language, the approach demonstrated by this dissertation could well apply to other phenomena in Chinese grammar. Thus I wish to advocate a methodological point that the structure of Chinese and its typological properties are best approached within a parallel architecture, in contrast to the majority of theoretical work on Chinese that is conducted from a syntactocentric perspective. In this chapter and the next, I will review certain metatheoretic arguments for the parallel conception of linguistic theory and the role of semantic explanations, especially as pertaining to grammatical phenomena in Chinese, in order to contextualize the proposals of this thesis within a range of contrasting options.3 In the remainder of the chapter, I first present the empirical challenges for mapping theories, in particular, issues that arise from English and Mandarin Chinese data and sketch the possible approaches and directions that could be pursued. Then I discuss the general motivations as to why I focus this dissertation on the conceptual-semantic processes in a theory of argument expression. The chapter concludes with an outline of the thesis.
1. Approaches to Lexical-syntactic Flexibility 1.1. Bottom-up vs. Top-down Perspectives A fundamental question that all linguistic theories attempt to answer is what the factors are that determine the basic grammatical structure of a sentence, such as the number of constituents that are allowed in a sentence. There are two contrasting views with respect to this question. Taking a bottom-up perspective, many frameworks, ranging from the classic European dependency grammar (Tesnière 1956) to the Government and Binding 3. Note that it is not my goal in this dissertation to provide thorough arguments for semantics and the parallel architecture, which can be found in many of the works referenced in this study (see esp. Jackendoff 1997, 2000). However, to the extent that the actual analyses proposed in this study are at least empirically superior to syntactic options, this work does constitute a substantial argument against the syntactocentric approaches.
Enriched composition and inference in the argument structure of Chinese
4
framework of Chomsky (1981, 1986), take a verb-centered point of view. Such a position is popular among students of argument structure (‘valency’) in Chinese, as can be seen from two recent collections of papers (Shen and Zheng (ed.)1995; Yuan and Guo (eds.) 1998). Much recent research stemming from this tradition takes the lexical properties, in particular, the lexical semantics of a verb as determining its syntactic behavior, such as the number of syntactic arguments it can occur with. For instance, if we take the meaning of the verb sweep as involving two semantic arguments, a ‘sweeper’ and a surface to be swept, this would account for the following contrast: (3) a. Terry swept the floor. b. *Terry swept the mother the floor. Given the semantic argument structure of the verb sweep, the sentence in (3b) is naturally ruled out. On this approach, the meaning of the sentence is also taken to be composed from verbal semantics and the relevant arguments in a syntactically transparent way. However, an alternative view can be taken with respect to (3). One could argue that the grammatical structure of (3a) instantiates an independent event structure, for example, an idealized cognitive model (Lakoff 1987) of event or a canonical event model in which “one forceful object transfers energy to another through forceful physical contact” (Langacker 1991:286, cf. Talmy 1988; Croft 1998). The event model must be stored in the mind as a conceptual structure template. Under this assumption, (3a) is acceptable because the semantics of its constituents is compatible with the structural elements of the event model embodied in the sentence structure, whereas (3b) is bad because one of its constituents the mother cannot be licensed by the event structure of the sentence that only involves two participants.4 One might ask why (3b) cannot be licensed by a threeparticipant event structure in the way the following is: The problem lies in the fact that verbal semantics of sweep clashes with the threeparticipant event semantics. Thus what distinguishes this approach from the verbcentered view is that supra-lexical meaning and associated syntactic structure are taken as a ready-made resource, in addition to lexical resources. Such an approach focuses on how elements of varying size in a common resource pool (a ‘lexicon’, if you will) get along with each other in forming the grammatical and semantic structure of a sentence. This kind of top-down perspective characterizes the so-called construction-based or sign-based approaches to argument structure and other phenomena (see references above).5 4. Of course there could be numerous accounts of the contrast from different theoretical perspectives, but I only focus on the syntax-semantics interface here and will largely speak in theory-neutral terms at this point 5. Incidentally, the top-down spirit has long been present in Systemic Grammar, a theory much ignored by theoretical linguists (see esp. Davidse 1991). The term ‘top-down’ is borrowed from Goldberg (1995).
Introduction: Variation in argument expression
5
An important feature of constructionism is that semantic structure such as the canonical event model mentioned above is treated as the inherent meaning of particular syntactic patterns. This is clear in the concept of argument structure construction used in Construction Grammar (Goldberg 1995), according to which the structure of a situation is directly associated with basic syntactic structure. Goldberg (1995, 1998a) suggests a scene-encoding hypothesis, claiming that simple sentence types such as Subj V Obj1 Obj2, Subj V Obj Obl, encode basic event types such as “X causes Y to receive Z,” “X causes Y to move Z.” Pushed to an extreme, such an approach implies a biunique connection (4) John baked Mary a cake. between semantic structures and syntactic structures. While the works of Goldberg only posit flat syntactic structures employing grammatical functions, the constructionist work associated with the Chomskyan framework has been positing increasingly abstract syntactic structures6 involving empty functional and verbal heads that serve to encode the intricacies (e.g. aspectual properties and causation) of event structure (Hale and Keyser 1993; Harley 1995; Ritter and Rosen 1998; Borer 1994, 2000; Travis 2000, among others). In other words, while standard work in Construction Grammar does not seem to have taken a stand as to whether all event-structural meanings correspond to syntactic forms, constructional work within the Chomskian framework does seem to be pushing such a view, stressing a tighter fit between semantics and syntax. This view accords well with the syntactocentric outlook mentioned earlier (cf. Lin 1999 for application of eventsyntactic approaches to Chinese data). The work presented in this dissertation will show that the connections between syntactic structures and meaning are more complicated than a biuniqueness view could handle. In fact, the view defended here was stressed at the outset of generative grammar (Chomsky 1957). 1.2. The Challenge of Lexical-Syntactic Flexibility Let us now bring the two perspectives sketched above to bear on the empirical challenge they are both faced with. Recent works on argument expression have underscored the fact that most verbs are capable of occurring in different syntactic environments, a phenomenon captured by the term diathesis alternation (Levin 1993; Jones 1994; Van Hout 1998). Van Hout (1998) has subsumed this sort of argument structure phenomena under the heading of lexical-syntactic flexibility. What follows is a typical set of alternations from English (taken from Rappaport Hovav and Levin 1998): (5) a. Terry swept. b. Terry swept the floor. c. Terry swept the crumbs into the corner. d. Terry swept the leaves off the sidewalk. e. Terry swept the floor clean. f. Terry swept the leaves into a pile. 6. For Hale and Keyser (1993), the event-related syntactic structures are located in the lexicon, which they term l-syntax.
Enriched composition and inference in the argument structure of Chinese
6
As can be seen above, a single verb sweep is allowed to occur in numerous syntactic frames involving different number and type of complements, with accompanying variation in meaning, from activity (5a, b), change of location to a Goal or from a Source(5c, 5d)7, change of state (5e), to creation of an artifact (5f). As Rappaport Hovav and Levin (1998) point out, it is counterintuitive to posit that these examples involve six different verbs sweep. It would be useful to contrast this set of examples with their close equivalents8 in Chinese, as presented in (3) below: (6) a. Zhangsan hai zai sao still Prog. sweep “Zhangsan is still sweeping.” b. Zhangsan saole zhejian wu sweepLE this-CL room “Zhangsan has swept this room.” c. Zhangsan ba mianbao xie sao-jin dong-li BA bread crumb sweep-enter hole-inside “Zhangsan swept the (bread) crumbs into the hole.” d. Zhangsan ba yezi cong jieyuan sao-zou BA leaf from sidewalk sweep-go “Zhangsan swept the leaves off the sidewalk.” e. Zhangsan sou-ganjingle diban sweep-cleanLE floor “Zhangsan swept the floor clean.” f. Zhangsan ba yezi sao-cheng yi dui BA leaf sweep-become one pile “Zhangsan swept the leaves into a pile.” The examples in (6) parallel those in (5) in semantic variation. Thus the Chinese verb for sweep is just as flexible in its syntactic behavior as its English counterpart. This kind of variation is problematic for the bottom-up verb-centered perspective, as pointed out by Rappaport Hovav and Levin (1998), since it means that the verb sao or sweep must be given multiple lexical-semantic representations that are listed in the lexicon, while in the meantime these separate senses are systematically related to each other. The lexical approach represents a sense-enumerative view of the lexicon that has been discredited by recent work such as Pustejovsky (1995), among others, due to its inability to capture the systematic relations between the multiple senses of a verb and the constraints on sense variation. Rappaport Hovav and Levin (1998) propose an improved 7. Note that (5c) and (5d) appear to have the same syntactic structure, and thus may not illustrate lexical-syntactic flexibility as intended by Rappaport Hovav and Levin (1998). However, I leave them here, as their close Chinese equivalents (6c) and (6d) have different syntactic structure. 8. Direct translation, if possible at all, of these English sentences would not be able to maintain their argument structure properties.
Introduction: Variation in argument expression
7
lexical approach by recognizing an inventory of event structure templates in UG and suggesting ways to augment lexical event structure within the constraints of UG (see Fong and Poulin 1998 for an application; Wunderlich 1997, 2000 for a similar lexical approach). But this theory does not differ from constructional approaches in fundamental ways, as Rappaport Hovav and Levin (op. cit.) themselves point out. On the other hand, the lexical-syntactic flexibility is entirely expected in the top-down constructional perspective, according to which the so-called alternations in (5) and (6) are determined by their different syntactic structures associated with differences in event structure, i.e., by a pairing of grammatical forms and meanings. For instance, as noted above, the structural differences exhibited by examples in (5) are correlated with aspectual and other event-structural differences. Most notably, the change of location (5c, 5d) is associated with a PP complement, whereas the change of state is associated with an AP complement (5e). Similar event-structural differences are exhibited in the corresponding Chinese sentences in (6), which are expressed by a different syntax as well, notably, verb-verb compound (6c, 6d, 6f) and verb-adjective compound (6e), as well as the use of the much-studied morpheme ba. The morpheme ba is arguably associated with a telic situation (see Liu 1997, cf. Wong 1999). While much needs to be said regarding the exact correspondences between form and meaning in (5) and (6) as well as how they constrain the behavior of lexical items, I only wish to stress a general point arising from these examples, namely, that the lexical semantics of the verb alone does not determine its syntactic behavior, but interacts with and is constrained by the independent semantics expressed by the relevant constructions (see e.g. Goldberg 1995, 1997, 1998b; Talmy 1985; Matsumoto 1996; see also Ghomeshi and Massam 1994 for a similar view)9. Thus lexical-syntactic flexibility is not cause for concern at all from a constructional perspective, since a particular lexical-semantic representation could in principle be compatible with a number of constructional-semantic representations, each corresponding to a syntactic pattern. Thus the term alternation may be misleading, since different syntactic patterns carry differences in content. Such a view has recently been stressed by Dowty (2000a, 2000b) with respect to the English swarmalternation (see section 4.5.3). In short, the most important feature of the constructional approach is that it argues for functional-semantic motivations of syntactic alternants (see Goldberg 1995 for extended arguments). The semantics-oriented view distinguishes Construction Grammar from purely syntactic approaches, according to which verb frame alternations are purely syntactic phenomena involving syntactic movement that changes grammatical relations (see. e.g. Baker 1988; Belletti and Rizzi 1988). Syntactic movement is sometimes aided by certain abstract operations on argument structure, as in the well-known analyses of passives where the external argument is said to be suppressed or blocked from being projected (cf. Jaeggli 1986; Baker, Johnson & Roberts 1989; cf. Chapter 4 below for discussions of Chinese passives). On such analyses, the alternations are taken to maintain the same set of theta roles postulated in the lexical entry of a verb. The underlying assumption is that the alternations retain the same semantics. However, drawing on data of Chinese 9. Note that meaning expressed by a construction could be, but not necessarily, related in an inherent way to the syntax of the construction. This point will be often stressed in this dissertation.
Enriched composition and inference in the argument structure of Chinese
8
passives, I argue in Chapter 3 that important semantic differences are involved in passive alternations, thus discrediting transformational proposals for Chinese passives that rely on a coarse-grained semantics (cf. Pustejovsky 1995). As Jackendoff (1990) observes, the practice in the literature tends to favor syntactic strategies over other options,10 which can also be seen in the syntactic proposals I discuss in Chapters 3 and 4. In this dissertation, I will not adopt this general strategy, and will attempt to pinpoint in subsequent chapters certain empirical inadequacies in actual applications of the syntactic strategy. As exemplified by (6), the empirical challenge for a theory of argument expression in Chinese is of the same scope as in English. But let us take a look at some more Chinese data of a somewhat different flavor: (7) a. Zhangsan saole neijian wu yige xiaoshi sweepLE thatCL room oneCL hour “Zhangsan swept that room for one hour.” b. neijian wu saole Zhangsan yige xiaoshi thatCL room sweepLE oneCL hour “That room had Zhangsan sweeping it for one hour.” (8) a. ta xiaole. he laughLE “He laughed.” b. ta xiao wo he laugh I “He laughed at me.” (9) a. tade lian heng hong his face very red “His face is very red.” b. ta hongle lian he redLE face “His face reddened.” (10) a. ta ba yizan denlong diao he BA oneCL lantern
zai mengqian
hang Loc. door-front
“He hung a lantern in front of his door.” b. mengqian diaozhe yizan denlong door-front hangZHE oneCL lantern “In front of the door was hung a lantern.” (11) a. Zhangsan zai shui Prog. sleep 10. This kind of tendency seems to be dictated by the general architecture of Chomskyan models in which the syntax is the only generative component (see Jackendoff 1997, 2000 for discussions).
“Zhangsan is sleeping.”
Introduction: Variation in argument expression
9
b. nashi women shui yaodong that-time we sleep cave “At that time we slept in caves.” (12) a. zuotian wo zai mang banjia yesterday I Prog. busy move-home “Yesterday I was busy with moving.” b. ta hanle yihui sangzi he cryLE a-while throat Lit “He exercised his throat by crying out loud for a while.” Meaning. He trained his voice as part of a singing practice. In contrast to (6), the examples presented above do not exhibit any distinctive grammatical markings except the simple Subject V (NP) pattern. However, the semantic structures expressed by these sentences are of varied sorts. For example, in addition to designating an agentive action (7a, 8b), the transitive pattern S-V-NP also expresses causation or extrinsic instigation (7b, see Gu 1997, cf. Jackendoff 1990 for the relation between causation and instigation). In addition, it is possible to allow the Subject of the pattern to be related to the event expressed by the verb in an opaque way. For instance, ‘he’ in (9b) and ‘door-front’ in (10b) do not seem to be direct participants11 of the relevant situation expressed, that is, ‘reddening’ and ‘hanging’: what reddened is strictly the face, not “he’; an action of hanging minimally involves a person who hangs and the thing hung. A similar feature is found with postverbal complements as well, but played out to a much greater extent. For instance, instead of expressing expected roles like Patient, they are found to roughly designate Location (11b), Cause (12a), Purpose (12b: crying out loud is for the purpose of exercising the throat). What makes them more interesting is that the verb with such uses is semantically intransitive in the examples given here, for example, shui ‘sleep’, mang ‘be-busy’, han ‘cry-out’. There are other possible semantic relations between the verb and its complement, as documented in Guo (1999). The discrepancy between the simple syntax and the complex semantic relation means that certain semantic elements are left unexpressed by the surface syntax. For instance, most prominently, the crucial idea of ‘instigation’ in (7b) and the notion of ‘exercising (one’s throat)’ in (12b) are not directly expressed by any syntactic element. This group of examples poses some difficulty for standard constructionist work that stresses formmeaning biuniqueness. For instance, it does not sound like a solution at all to simply claim that these unexpressed semantic elements are invariably cases of construction meaning, or that they ARE expressed by certain invisible syntactic elements (the idea of ‘exercising’ in 12b cannot plausibly be expressed by a particular syntactic construction nor any invisible syntactic element). In the central chapters of this thesis, it will be shown that such phenomena are productive to some extent, provided certain lexical-conceptual conditions are satisfied. 11. I return to a discussion of direct or core participants and circumstantial participants in Chapter 3.
Enriched composition and inference in the argument structure of Chinese
10
2. Motivations for a Semantic Focus As will be seen from my review of the available proposals in the literature, the syntactic strategy, as incarnated in recent developments in mainstream syntactic theories, has almost always been the only favored option in the analyses of argument structure alternations in Chinese. Therefore it might be relevant to make a few general comments as to why I adopt the particular semantic focus I do as opposed to the syntactocentric trend in theoretical linguistics. First, semantics, understood as conceptual structures and the underlying combinatorial system in this dissertation, can be considered to be ontologically prior to syntactic structures, since “the brain organization supporting conceptual structure has had a great deal more time to evolve than that supporting syntax” (Jackendoff 1993:284). Thus conceptual structure is expected to be more complex than syntactic structure. On the grounds of parsimony, a semantic solution is therefore preferred over a syntactic one, if both are available. In the context of a discussion on syntactic categories, Gil (2000) stresses the same point in his Semantics before Syntax principle. Arguments from parsimony are nicely demonstrated in Jackendoff’s (1992) discussion of binding theory in which he contrasts the syntactic and semantic solutions, as summarized below: (13) a. Syntactic Solution Binding condition based on asymmetrical c-command complexity in syntax complexity in semantics b. Conceptual-semantic Solution Binding condition based on cs-superiority simplicity in syntax complexity in semantics Jackendoff (1992) proposes the conceptual-semantic notion of CS-superiority as a replacement for the syntactic c-command as a condition for binding relations to obtain.12 The important point from his careful comparison is that the complexity in semantics cannot be eliminated in either case, but the syntactic solution complicates syntax whereas the conceptual-semantic solution simplifies syntax. It will be shown in the ensuing chapters that this kind of contrast resurfaces time and again in alternative approaches to argument expression in Chinese as well. I demonstrate how my particular proposals account for semantic complexities while preserving a simple syntax. 12. Since this thesis does not focus on binding theory, I leave syntacticians with doubts to consult cognitive-conceptual accounts of binding and anaphora (Kuno 1987; van Hoek 1997; Jackendoff 1992, 1997; Culicover and Jackendoff 1995).
Introduction: Variation in argument expression
11
Second, a focus on semantics helps determine the limits of autonomous syntax.Granting the ontological priority of conceptual structure, it is still an open and difficult issue to draw in practice the proper boundary between syntax and semantics with respect to a particular phenomenon (see Newmeyer 1998 for relevant discussions). Many linguists, notably Jackendoff (1990, 1997, 2000), Culicover (1998, 1999), as well as Goldberg (1995) and many other cognitive-functional grammarians, have converged on a view, more or less explicitly, that syntactic theory has overstepped its bounds in the practical approaches to diverse phenomena, in spite of Minimalist attempts to simplify syntactic machinery. It would be fitting to explore the alternative avenue, as cognitivefunctional linguists of various kinds have long being doing, investigating in detail what an articulated semantic system could offer with respect to various empirical challenges, as part of the attempt to determine the limits of the syntax, as explored in the papers in Culicover and McNally (eds. 1998). In recent works by Culicover (1998, 1999), much of the role played by the conventional syntax has been argued to properly fall within conceptual structure. Thus in this dissertation I will make only minimum assumptions about syntactic theory, relying largely on some well-known grammatical reflexes of conceptual structure. The result of the research reported here should serve to further delimit the boundaries of the syntax. Thirdly, the semantic focus I adopt serves to redress a trend among certain syntacticians, certainly those whose works I discuss in later chapters (see e.g. Feng 2000; Lin 1999; Xu 1999), that basically pushes semantic complexities aside or assumes a coarse-grained view of meaning, as embodied in a set of theta roles, even though all agree that language serves to express and communicate thoughts. My view is that a syntactic theory must be paired with an elaborate semantic theory, which is now made possible by the seminal works of Jackendoff and other cognitive semanticists.13 I show in this dissertation that this sort of semantic engagement is more illuminating of many argument structure phenomena than purely syntactic proposals. This belief is shared by linguists to varying extents who study argument structure, such as Van Valin (1990), Grimshaw (1990), Rappaport and Levin (1998), van Hout (1998), Goldberg (1995, etc), Dowty (2000a), Pustejovsky (1995) and others.
3. Organization and Contents of the Study The rest of this thesis is composed of four chapters. The general framework and theoretical assumptions are laid out in chapter 2, where I present a synopsis of Conceptual Semantics as a generative system, incorporating insights from recent works in lexical, computational and cognitive semantics and embedding this view of semantics within a revised Constructional outlook on the theory of universal grammar. I illustrate different types of construction, with varying amount of syntactic and semantic information. In chapter 3, I focus on what I call the eat restaurant construction in 13. The same view is taken in HPSG, which adopts a version of situation semantics (Barwise and Perry 1983).
Enriched composition and inference in the argument structure of Chinese
12
Chinese and discuss the way the unselected object is semantically licensed by enriched compositional processes of coercion and co-composition. In particular, elaborate lexicalsemantic representations of the verb and its collocated nouns are proposed, incorporating crucial frame-semantic (or qualia) information. The compatibility of the relevant lexical representations predicts the range of the empirical data instantiating the so-called unselected object constructions, while possible overrides to semantic generalizations are taken to fall within the Preference Rule System of the Conceptual Semantics. The semantic proposal is evaluated against three syntactic analyses from somewhat different perspectives (Feng 2000; Lin 1999; Gao 1996). Chapter 4 contrasts with Chapter 3 by concentrating on preverbal argument NP’s, bringing together apparently disparate syntactic constructions including Possessum-object constructions, retained-object passives and locative inversions. It is shown that the subject of these constructions is semantically non-canonical and is licensed by the mechanism of predicate transfer in the sense of Nunberg (1995), reconceived in conceptual-inferential terms. I demonstrate that the verb type in the constructions is conditioned by the conceptual structure of the relevant construction. Finally, the thesis concludes in Chapter 5 with a summary of the findings, and a short discussion of the implications and further issues.
CHAPTER TWO Constructions, Conceptual Structures and Compositionality
In this chapter, I first outline the general framework within which argument structure phenomena in Chinese will be investigated. I briefly clarify the theoretical inclinations, emphases and rationales in this study. This is followed by an overview of the assumptions concerning semantic representation and semantic compositionality. Overall I follow the framework advocated by Jackendoff (1997, especially 2000), but will feel free to incorporate certain technicalities commonly accepted in constraint-based frameworks of grammar.
1. The Tripartite Parallel Architecture As Jackendoff (1997, 2000) has observed, the standard architecture of generative grammar, from Aspects to the Minimalist Program, is syntactocentric in that it conceives of syntax as the only source for the generative capacity of language. Such an outlook has guided most formal research on the grammatical structure of Chinese (see e.g. Huang 1982; Li 1990; He 1995; Sybesma 1999, among many other works). In keeping with the methodological goal mentioned in chapter 1, I will not follow this mainstream tradition in Chinese syntax. My methodological decision is motivated by following considerations. First, Chinese is typologically known to have scant inflectional and derivational morphology, relying largely on word order, certain particles and distinctive constructions (e.g. ‘wh in-situ’ constructions, the double subject construction). Under a universalist syntactic theory characterized by syntactic movement and various functional heads,1 Chinese syntax would appear very abstract and complicated, which is at odds with the typological properties of this language. Second, the unnecessary complication of the syntax makes it harder to integrate syntax with processing, which is a goal argued for by many non-transformational linguists, as instantiated in the strong competence hypothesis 1. Ackerman and Webelhuth (1998) report that 21 different functional heads, often unobservable, have been proposed in various places in the P & P literature.
Enriched composition and inference in the argument structure of Chinese
14
of Bresnan and Kaplan (1982).2 Furthermore, as Jackendoff (2000) argues, the logical directionality of a syntactocentric model is incompatible with the non-directionality of processing, which finds a direct correlate in a parallel architecture. In this study, I assume some form of Concrete Minimalism (Culicover 1998,1999) in the form of the Aspectsstyle phrase structure, leaving open the larger issue of syntactic representation in Chinese grammar. Thirdly, the syntactocentric conception does not lend itself easily to an exploration of semantic issues concerning the semantic structures of linguistic expressions (cf. Talmy 1988, 2000; Langacker 1987, 1991; Pustejovsky 1995; Jackendoff 1983, 1990, among others). In a series of works (Jackendoff 1983, 1990, 1996b, 1997, 2000), Jackendoff has argued in detail for semantics as a generative system, which I give an overview of below. Under the parallel model, we can allow semantics to come into its own, thus able to formally implement the insight of some ‘cultural’ grammarians (Shen 1992; Xu 1997) that stresses the preponderance of semantic structuring in the grammar of Chinese.3 Under the tripartite conception of the grammatical architecture, semantics, syntax and phonology are independent parallel rule systems that are linked through a set of interface rules. This conception has been developed at least since Jackendoff (1983, 1990, see Culicover 1988 for an early discussion) and has been elaborated in detail in the recent works of Jackendoff (1997, 2000). Wiese (2000) develops a close variant on the basis of both Jackendoff (1997) and the Two-level model of semantics (see e.g. Bierwisch 1989). Figure 2.1 summarizes the basic aspects of this framework: 2. Current generative theory maintains a “hard” idealization in the distinction between competence and performance. Chomsky and Lasnik (1993:18) assert that “in general, it is not the case that language is readily usable or ‘designed for us’.”(see Jackendoff 2000 for more discussion) 3. What I have in mind here is a movement in the early 90s within Chinese linguistics, usually branded “Chinese cultural linguistics,” represented by some controversial works such as Shen (1992), whose goal was to describe the ‘unique’ properties of Chinese grammar in a way compatible with what is known about other aspects of Chinese culture (e.g. Chinese painting). I wish to highlight their major insight here, in full knowledge that their ideas are prone to vagueness and imprecision and do not contribute a developed linguistic theory (see Shi 2000 for discussion).
Constructions, conceptual structures and compositionality
15
Figure 2.1. The tripartite parallel architecture (adapted from Jackendoff 2000) The model in Figure 2.1 is defended via a careful critique (see Jackendoff 1997) of various versions of the syntactocentric paradigm. The resulting model bears a number of interesting similarities to (as well as certain differences from) constraint-based approaches such as HPSG (Jackendoff 2000). I situate the analyses proposed in this thesis within this general framework.
2. A Constructionist Deconstruction of the Lexicon An important issue for all linguistic theories is the relation between the lexicon and the extra-lexical principles of the grammar. One important consequence of the framework in Figure 2.1 is that there is no separate lexical interface. Given the commonsense understanding that the lexical item is an association of phonological, semantic and syntactic features stored in long-term memory, the lexicon is naturally taken as part of the interface components (Jackendoff 1990, 1997, 2000). In other words, a lexical item is a three-way mini interface rule that depends on the same set of primitives and combinatorial principles in the syntactic, semantic and phonological components.
Enriched composition and inference in the argument structure of Chinese
16
2.1 Lexicon as Part of the Interface Components of the Grammar Jackendoff entertains two related options (Jackendoff 1997, 2000) in juggling the relations between the lexical and the extra-lexical phenomena, both of which result in a deconstruction of a separate lexical component. This is consistent with the accepted wisdom of recent work in constraint-based theories and computational linguistics (see Briscoe 1996 for a review, cf. Hudson 1986/1995). The first option simply dissolves the lexicon into the grammar, specifically into the interface components. On this view, instead of being inserted (either ‘early’ or ‘late’) into the syntactic derivation, the lexical item, being a three-way linkage itself, serves to license or establish the correspondences of near-terminal symbols of syntactic structure with phonological and conceptual structures, which are all independently generated. In this context, licensing is understood as unification in the sense of Shieber (1986), which we return to below. This option bears a certain resemblance in spirit to similar ideas long favored in Systemic Grammar since the 60s, as embodied in the convenient slogan, “lexis as the most delicate grammar.” (see Butler 1985 for discussion) According to this theory of lexical licensing (as opposed to lexical insertion), the lexicon is part of the grammar in that it shares the same format of interface rules. 2.2. Grammar as the Lexicon With the lexicon taken as part of the interface components, the first option basically maintains the distinction between generative components (i.e. formation rules) and interface components, as shown in Fig. 2.1. The second option, which is the one favored in this thesis, attempts to assimilate all grammar to a repository of correspondence patterns, a lexicon in a more general sense or a ‘constructicon,’ if you will. On this view, both interface and formation rules are taken as pieces (termed construction) stored in long-term memory. Given the various components in the architecture depicted in Fig. 2.1, the stored pieces can be classified with respect to the possible combination of information from these components, as will be seen shortly. This view of the grammar ties in with the recent Constructionist impulse in numerous theories (Langacker 1987; Goldberg 1995; Sag 1997; Kay and Fillmore 1999; Holmes and Hudson 1999; Croft 2001, to cite a few), the consensus that the notion of grammatical construction is central in the organization of the grammar, instead of being epiphenomenonal and derivative as deemed to be the case in some versions of Principles and Parameters theory (Chomsky 1992)4. In the prevalent understanding of the notion (Goldberg 1995, 1999, cf. Pollard and Sag 1994), a construction is a conventional pairing of form and meaning/use (a sign, in HPSG terms). From Jackendovian perspectives, a construction is itself an interface rule, just as a lexical item is. Jackendoff (1990) terms an extra-lexical construction a constructional 4. Chomsky (1992:3) states, “Constructions such as verb phrase, relative clause, passive, etc., are taken to be taxonomic artifacts, collections of phenomena explained through the interaction of the principles of UG, with the parameters fixed.”
Constructions, conceptual structures and compositionality
17
idiom, which is stored in the long-term memory in the same way as lexical items are. Thus the notion of the construction is not defined in terms of the internal complexity of linguistic objects. Following the line of arguments presented in Jackendoff (1996, 1997, 2000), I take the notion of a construction used in sign-based theories (HPSG, Construction Grammar, Cognitive Grammar) to be a “full” sign in the sense that it embodies a structured complex of numerous sorts of information, including phonological, semantic, discourse and contextual information (Pollard and Sag 1994; Engdahl 1998). It should be noted, however, that these different sorts of information do not necessarily have to converge on a single linguistic object simultaneously. Thus we could have “defective” signs to varying extent that lack one or more types of information. The parallel architecture in Figure 2.1 could serve as a resource for working out a typology of “full” and “defective” constructions, in terms of the kinds of information that are being juxtaposed in a construction. What follows focuses on such a typology. First, as the prototype of constructions, we have “full” constructions. These are lexical and phrasal structures (i.e., lexical items and constructional idioms) specified with the triple
. These types of constructions have been the focus of research in much recent work (see, for example, Jackendoff 1990, 1997a, b; Culicover and Jackendoff 1997; Goldberg 1995; 1997c; Michaelis and Lambrecht 1996). It should be noted that “full” constructions do not mean there are no mismatches among the members in the triple, a point emphasized in Construction Grammar. For example, the meaning of this kind of construction is often more than the sum of its parts, as in the English wayconstruction studied by Jackendoff (1990) and Goldberg (1995, 1997c): (1) Bill belched his way out of the restaurant. In (1), nothing in the sentence entails motion, yet, as Jackendoff (1990:214) observes, the unexpressed conceptual function GO is imposed on the conceptual structure of the verb, thus demoting the meaning of the verb to a conceptual modifier (see Levin and Rappaport 1988 for a similar analysis). In other words, the SS and PS in the triple often fail to be in biunique correspondence with the CS. Second, we have “defective” signs that have phonology and semantics, but no syntax, represented as . Examples in Chinese are wei (hello), aiyo (ouch), aiya (expressing surprise). 5 Noteworthy in this category are cases of reduplication that are particularly productive in Chinese, ren-ren (person-person: every person), shi-shi (thingthing: every thing) (cf. Jackendoff 2000; Ghomeshi, Jackendoff, Rosen and Russell (to appear)). On the phrasal level, information structure such as topic and focus often instantiates this sort of defective sign in a number of languages such as English and Chinese in that no special syntax is necessarily invoked in expressing these discourse notions (cf. Zhang 2000b). 5. The statement that they have semantics implies a very broad view of semantics, one that certainly far transcends truth-conditional meaning. But even within the broad scope of Conceptual Semantics, it is not clear if it is possible to encode emotions in Conceptual Structure. Perhaps some items are associated with some separate emotional or other sensory structures, to accommodate the meanings of words like acrid, fragrant, just as spatial representations are associated with expressions for physical objects (Jackendoff and Landau 1991). This falls outside the research issues of this dissertation (see Jackendoff 2000, Allan 2001 for some discussion).
Enriched composition and inference in the argument structure of Chinese
18
Third, there are items that lack both syntax and meaning. Jackendoff (1997, 2000) cites the nonsense refrains such as fiddle-de-dee and tarum-diddle-i-day, the latter used in folk songs. They could be notated as <Ø, Ø, PS>. 6 Fourth, there are items that have syntax and phonology, but-no semantics, i.e., <Ø, SS, PS>. These are those pure syntactic items that constitute the staple of formal syntactic theories, such as expletive it, do of do-support, nominative and accusative case markings. Alternative frameworks (see e.g. Langacker 1987, 1991) contest the status of such items. It is not clear to what extent such lexical items are found in a morphosyntactically impoverished language like Chinese. However, also belonging with this category are interface constraints in each language that link phonological words, Intonational Phrases (IntP) to syntactic constituents (see Jackendoff 1997, 2000). In other words, at a level of abstraction, such phonology-syntax correspondence rules (e.g. the near-correspondence between an intonation phrase and a syntactic phrase) can be regarded with a construction devoid of semantic content. Fifth, some theories of syntax postulate abstract signs that have syntax and semantics, but no phonology, namely, the triple , such as the empty category PRO (Chomsky 1980) used in GB theories. Certain recent notions of “light verbs” used in syntactic representations of event structure (Hale and Keyser 1993; Harley 1995; Huang1994, 1997; Lin 1999, etc.) also belong in this category. In line with the practice of constraint-based theories such as HPSG (see Webelhuth et al. 1998 for a discussion), I take it as preferable to avoid appealing to such items. This point is going to be crucial in our analysis of unselected object construction in Chapter 3. However, similar “defective” signs are definitely attested on the phrasal level in every language in the form of certain default linking principles (i.e. interface constraints) between semantics and syntax, as instantiated in versions of linking via thematic hierarchies (Baker 1988, Bresnan and Kenerva 1989, Jackendoff 1990, among others) and the correspondences between subject/object and figure/ground. Sixth, pushing the notion of defective lexical items to an extreme, we have syntactic structures with null semantics and phonology, which I notate as < ( ), SS, ( ) >.7 This means that attributes of semantics and phonology are not specified with any values at all, an extreme version of underspecification. But just as with other cases of underspecification, the null values could be specified online and further fleshed out by unification with more specified constructions. Thus our use of the bracket ( ) differs from Ø, in which case no value could be specified. For example, there is no way to unify the item aiyo (ouch) with any other item so that it gains syntactic value.8 This shows that Ø 6. This does not mean they are random noises. They serve to fill up metrical structure in folk songs, as Jackendoff (2000) observes. However, they do not express conceptual structure. Thanks to Diane Massam for raising questions (in p.c.) about this point. 7. Another kind of defective construction (< >, < >, CS) must in principle be stored in the mind as well, since conceptual structures are not modality-specific. I think this is the idea assumed in cognitive semantic theories (see e.g. Langacker 1987, cf. Koenig 1999b), though CS divorced from linguistic structures are not of linguistic interest, as Jackendoff (p.c.) reminds me. 8. Diane Massam (p.c.) calls attention to apparent counterexamples like ouchless bandaid. In such cases, ouch has to first go through a morphdogical change in order to combine with other words.
Constructions, conceptual structures and compositionality
19
implies a greater degree of defectiveness than ( ).9 On this view, traditional phrase structure rules naturally fall under this category. Independently, Van Valin and LaPolla (1997) conceive of the syntactic rules in Role and Reference Grammar in terms of constructional templates as well. Jackendoff (1996, 1997, 2000) argues against the Construction Grammar view that syntactic structures must be paired with meaning (see Goldberg 1995, Langacker 1987, Michaelis 2000), citing evidence from transitive and ditransitive constructions in English and other languages to make the point that not all syntactic configurations are inherently meaningful. We will see that evidence for this point is abundant in Chinese as well. Having sorted out the possible sorts of constructions, we are practically left with a reinterpretation of the model in Figure 2.1 as an inventory of constructions of different sorts, that is, all the rule and interface components are now interpreted as lexicalized grammatical patterns10, which Jackendoff (2000) terms l-rules. Jackendoff proposes a notion of UG as consisting of a collection of skeletal fragments of 1-rules with typed variables built into lexical memory (called “grammatical archetypes” by Ackerman and Webelhuth 1999). This idea resonates with much work in cognitive linguistic frameworks (Langacker 1991; Goldberg 1995; Hudson 1998). UG “shapes” the emerging grammar (i.e., prespecifies the “unmarked” cases) without entirely determining it (i.e., leaving room for “marked” cases); the grammatical fragments in UG are “attractors” that establish points of stability rather than absolute standards for grammatical patterns. This is an interesting alternative to the parameter-based conception of UG, which has been subject to criticisms on both internal and external grounds in recent works (Elman et al. 1996; Tomasello 2000, Deacon 1997, Culicover 1999; Ackerman and Webelhuth 1998). The external grounds hinge on the plausibility of an extremely elaborate and fine-tuned UG, while the internal grounds concern micro-parameters as universal possibilities (cf. Green 2000).
3. Formal Mechanisms Section 2.2 argues for a view of the grammar as consisting of an inventory of “full” and “defective” constructions. Following work in Construction Grammar and HPSG and much of computational linguistics11 (Goldberg 1995; Michaelis and Lambrecht 1996; Sag 1997; Holmes and Hudson 1999; Briscoe nd.; Pustejovsky 1995, among others), I assume that the constructions themselves are organized into inheritance hierarchies, which capture the linguistically relevant generalizations with respect to phonology, morphology, syntax and semantics across lexical and phrasal constructions. On this approach, a more specified linguistic object inherits structure and properties from less specified generic 9. Jackendoff (1996, 1997, 2000) does not appear to make this formal distinction, especially with his notation of this kind of construction as <Ø, SS, Ø>, using the same symbol as for other “defective” items, but it is certainly implicit in his arguments when he says that the meaning of the construction needs to be specialized by other items. 10. Jackendoff (2000) does not explicitly treat interface constraints as a construction, but such a proposal seems a logical result of his ideas. 11. The idea is also implicit in the use of the mechanism ‘system’ in the lesser known framework of systemic-functional grammar (see Hudson 1986/1995 for discussion).
Enriched composition and inference in the argument structure of Chinese
20
constructions higher in the hierarchy, just as DOG inherits properties from MAMMAL, which in turn inherits properties from ANIMAL in a taxonomic semantic network. Thus inheritance structure can be defined in terms of is-a or subtyping relations between elements in a hierarchical inventory of types of linguistic objects. Two aspects of inheritance should be noted in the context of this thesis. First, an item can inherit properties from multiple dimensions of the hierarchy at the same time, hence the term multiple inheritance. Second, I adopt what Goldberg (1995) calls normal mode inheritance or inheritance by default, allowing for exceptions to capture the phenomena of blocking (Briscoe et al. 1995). A case in point is the irregular past tense forms of English verbs that override the expected pattern. Another formal mechanism is the idea of unification (Shieber 1986). Given the constructionist view of the grammar advocated earlier, the only combinatorial operation or the procedural rule of grammar is UNIFY PIECES, and the pieces are the variety of constructions outlined in the last section. As Jackendoff (2000) notes (see also Carnie 2000), this operation shares certain formal properties with the Minimalist idea of feature checking and the operation Merge (Chomsky 1995). Unification is the standard stock-in-trade in constraint-based theories. By this operation, information from two or more sources is combined (unified), without adding anything more. That is, the output is compatible with the inputs (Pollard and Sag 1987; Riehemann 1995). Normally, unification ‘fails’ if the information is incompatible.12 Notions very similar to unification have also been found in Cognitive Grammar (Langacker 1987) and the cognitive semantic theory of Fauconnier and Turner (1998). In the fifth and sixth type of construction described in section 2.2, either SS or CS (and PS) is underspecified to different degrees. The underspecified dimensions, SS and CS, are fleshed out by unification with more specified syntactic and conceptual dimensions of lexical constructions, in a parallel fashion (Jackendoff 1997, cf. Schieber and Schabes 1991 for similar ideas within tree-adjoining grammars). In this dissertation, I specifically focus on unification on the semantic side, while assuming a simple flat syntactic structure. This parallel unification system embodies the claim that SS and CS are in principle separate at least insofar as we are talking about fully specified SS or CS, contra much work in sign-based theories (Langacker 1987, Goldberg 1995, Michaelis 2000, see Jackendoff 1996, 1997, 2000). Therefore, a semantic structure does not need to be inherently associated with a fully specified syntactic structure in order to unify and coerce pieces of CS’s, contra Construction Grammar assumptions (Goldberg 1995, Michaelis 2000). On analogy with productive and semi-productive morphology, I assume that CS/SS pairing is constructed online (cf. Koenig 1994, 1999; Jackendoff 1997, 2000) except for constructional idioms. From this perspective, the notion of ‘construction meaning,’ (Goldberg 1995) the idea reincarnated as ‘structural meaning’ in recent 12. Within computational lexical semantics some variety of default unification has been defined whereby only consistent information is incorporated, ignoring inconsistent information (Lascarides and Copestake 1998).
Constructions, conceptual structures and compositionality
21
versions of Distributed Morphology (Harley and Noyer 1999), has not been defined in a very precise way13. I return to this point in particular analyses.
4. Conceptual Structures, Frames, and Compositionality To summarize the framework so far, a view is developed of the grammar as an inventory of different sorts of constructions, including the formation and interface rules in Fig 2.1, structured into inheritance hierarchies. The constructions can be combined through the formal process of unification. The tripartite nature of the original model in Fig 2.1 is now recast into the three-way interface constructions represented as and the way they unify with other independent three-way constructions of various sorts such as and <( ), SS, ( )>, that is, the three dimensions are simultaneously unified with corresponding dimensions in other constructions. Leaving phonology aside for the purposes of this chapter, let me illustrate how simultaneous unification works with the following verb phrase (see Jackendoff 1997 for more illustrations), omitting certain details: (2) eat apples The available constructions that are implicated in this linguistic expression are listed below: (3) a. (< >, <[vp V NP]>, < >) b. (< >, < >, <[F([Thing] [Thing])]>) c. (<eat>, <[VP V1 [NP]]>, <[EAT1 ([Thing] [Thing EDIBLE]) d. (, <[NP N2]2>, <[Thing APPLE]2>) The first two items in (3) are the most schematic constructions, with (3a) being a pure phrase structure rule and (3b) being a pure CS rule (cf. note 7). First, (3a) and (3b) are joined online, yielding (3e): (3) e (< >, <[VP V NP]>, <[F ([Thing] [Thing])]>) Each dimension in (3e) is then simultaneously unified with the more fully specified lexical items in (3c) and (3d). This has the effect of adding indices and replacing the functional variable F with the conceptual function EAT, producing (3f): (3) f ( <eat apples>, <[VP V1 [NP N2]2>, < [EAT1 ([Thing] [Thing APPLE]2>) The order of unification is not of theoretical importance. (3c) and (3d) could well first unify with each other, and the resulting structure can be said to be licensed by the generic (3a) and (3b). 13. Chomsky (1957:108) explicitly criticizes the notion of structural meaning, stating that “it is questionable that the grammatical devices available in language are used consistently enough so meaning can be assigned to them directly,” cited approvingly by Jackendoff (1990).
Enriched composition and inference in the argument structure of Chinese
22
Within this Constructionist architecture, this dissertation is centrally concerned with unification of semantic structures under the assumption of simple, flat syntactic structures. The view of semantics and the semantic representation language adopted in this thesis derive most directly from the theory of Conceptual Semantics developed by Jackendoff, but I also incorporate certain relevant insights in related theories. 4.1. The Syntax of Conceptual Structures Arguing from internalist perspectives (Chomsky 1986), Jackendoff (1990,1994) points out the parallels between infinity in possible syntactic structures and in the possible thoughts or concepts that can be expressed by syntactic structures. In Jackendoff’s view, this conceptual infinity can only be captured by a separate theory of the grammar of mental concepts, which he calls Conceptual Semantics. The point of having a separate conceptual semantic theory is to emphasize the distinct nature of syntax and semantics and that semantic interpretation cannot be directly read off some syntactic representation such as LF. This point has also been argued for recently in studies in Grammatical Semantics (Mohanan & Wee (eds) 1999). In Conceptual Semantics developed and elaborated over the years by Jackendoff (1976, 1983, 1987a, 1990, 1991, 1996a, 2000; see also Nikanne 1990, 1995; for related work, cf. Pinker 1989; Mohanan and Mohanan 1999;Wierzbicka 1988; Langacker 1987), he proposes a set of universal primitives and principles of combination based on firstorder predicate logic that seeks to account for the expressivity of concepts. On this approach, the basic conceptual constituents belong to a set of ontological categories established on the basis of linguistic and perceptual evidence (Jackendoff 1983). These include Thing, Event, State, Action, Place, Path, Property and Amount. The set covers a broader range than the ontological types standardly assumed in formal semantics. In the latest version of the theory (Jackendoff 2000), the organization of conceptual structure is broken down into three major tiers, the descriptive tier, the referential tier and the topic/focus tier. The referential tier encodes the existential claims about the entities in the conceptual structure of a sentence. In what follows, I briefly describe the descriptive tier, which organizes two dimensions of a conceptualized situation (including state and event). Thus we have two sub-tiers, the thematic and the action tier. 4.1.1. The Thematic Tier The thematic tier is concerned with the internal structure of situations modeled on the structure of location and motion events, often discussed under the rubric of event structure in related instantiations of decompositional theories (Dowty 1979; Hale and Keyser 1993; Rosen 1996; Pustejovsky and Tenny (eds.) 2000) dating back to Generative Semantics. On this tier, ontological categories are broken down into a set of functions whose arguments recursively belong to the same set of categories.14 Thus we could have an infinite class of possible concepts (see Pinker 1989 for a variant). 14. Nikanne (1990, 1995, 1997) develops a dependency representation of conceptual structure within an organization of conceptual functions into three zones, each zone being associated with its distinct set of semantic roles.
Constructions, conceptual structures and compositionality
23
First, we have a family of ‘core functions,’ BE, STAY, GO, EXT, and ORIENT, which organize basic states and events. The particular patterns of organization are illustrated in the following tree structures:
(4)
(5)
(6)
(7) The above tree structures, where functions are attached by double lines and arguments by single lines, represent the formation rules concerning the basic event and state categories. The instantiations of these argument structures are presented in the following set of corresponding examples: (4) a. Shu zai xiangzi-li.15 book Loc. box-inside “The book is inside the box.” (5) a. Ta zongshi dai zai jia-li. she/he always stay Loc. home-inside “She/he always stays at home.” (6) a. tamen zou-dao le gongyuan. they walk-reach LE park “They (have) walked to the park.” 15. Note that the function BE is not expressed in the overt structure of a spatial sentence (as also with ascriptional sentences). This is to varying extents also the case with several other languages (see Jackendoff 1983:249 for discussion). Note that the Chinese equivalent of BE, namely, shi, is often deployed for emphatic purposes or focus marking (see e.g. Shi 1994), and thus may not express the State function BE.
Enriched composition and inference in the argument structure of Chinese
24
(7) a. zhe-tiao lu tong-xiang jichang. (Nontemporal extension: EXT) this-CL road open-toward airport “This road goes toward the airport.” b. na-zuo diaoxiang mian-xiang da-hai. (Orientation: ORIENT) that-CL statue face-toward big-sea “That statue faces toward the sea.” Drawing on the grammatical parallels between the spatial and non-spatial domains (Gruber 1967), Jackendoff (1983, 1990; cf. Iwata 1996) argues that the same sort of semantic patterns underlie both spatial and non-spatial semantic fields such as Possession, Ascription and Scheduling. Thus the particular character of the arguments in these patterns is determined by field features. As with the Path shown in (6), the categories Thing and Place can also be decomposed into function-argument structures (Jackendoff 1983, 1990, 1991). A second family of functions concerns aspectuality (Jackendoff 1991, 1996c), which take events or states as arguments. These are presented in the alternative bracket notation below: (8) [Event INCH ( [State])] (9) [State PERF ( [Event])] As can be seen, the two functions are somewhat like converses of each other, with (8) denoting an event of the state coming about (Inchoative) and (9) the state of an event being complete (Perfective)(cf. Smith 1994, 1997). I discuss in Chapter 4 the relations between these two functions and certain typological complications in exploring the semantics of the Chinese particle le (Zhang 2000a; Koenig 2000). A third family of functions, namely, CAUSE, LET, and HELP, revolve around the notion of causation. Typically they have three arguments, Agent, Patient and Effect. The Patient argument may be optional. I present the typical pattern and its instantiation with the function CAUSE below: (10) [Event CAUSE ( [Thing/Event], [Thing], [Event])] (10) a. zhejian shi jiao wo heng tongku. this-CL thing make me very agonized “This made me feel very bad.” Within this general scheme, it should be noted that the traditional thematic roles are not assigned any independent theoretical status as in versions of GB theories, but can be defined in terms of particular positions in conceptual structure (cf. Levin and Rappaport Hovavl988, 1996; Dowty 1989, 1991; Croft 1998, etc.) The CS-formation patterns in (4-10) are to be understood as rules with typed variables such as Thing, Place and Path. They are to be unified with constituents of appropriate type in order to compose phrasal semantics. But we should note in passing another major mode of composition, namely, restrictive modification, discussed in Jackendoff (1990) (see also Jackendoff 1997, 2000, cf. Coulson and Fauconnier 1999; Sweetser 1999). Following Jackendoff (2000), I use a dotted line in the tree notation to indicate conceptual modifiers, as illustrated in (11) for the CS of red car:
Constructions, conceptual structures and compositionality
25
(11) Adapting the formal logic notation of lambda-extraction, Jackendoff (2000) proposes to model the relative clause in the following way, with the lambda operator binding the variable x: (12) Xiaowang xihuan de neiwei guniang like Mod. that-CL. girl “that girl who Xiaowang likes”
(12’) In this structure, the entity GIRL is further specified by virtue of her participation in the event expressed by the function LIKE. This treatment of modification is similar to that of Napoli (1989), but contrasts with the philosophical literature where modifiers are often taken to be predicates. We will see later that the mechanism of modification can be used to encode richer conceptual elements. 4.1.2. The Action Tier Noting the distinctness of Patient and Actor roles from other thematic roles (cf. Culicover and Wilkins 1984), Jackendoff (1987, 1990) encodes these roles on what he calls the action tier. By contrast to the elaborate localist structure of the thematic tier, the action tier represents a more abstract dimension of event structure that corresponds to the notion of transmission of force, a notion important in causal approaches to event structure (Talmy 1988; Croft 1991, 1993, 1998; see Levin and Rappaport Hovav1996 for discussion). Thus the action tier can be said to formally express the idea of affectedness. Information on the action tier is expressed by the dyadic functions AFF (“affect”) and REACT, attached with different values, “+,” “−,” “0,” and “±vol” (“volitional”). The relevant patterns are laid out below: (13) [AFF± ( [X], [Y])]
Enriched composition and inference in the argument structure of Chinese
26
(14) [AFF° ([X], [Y])] (15) [AFF±vol ( [X], [Y])] These patterns define the role Actor as the first argument of AFF, Beneficiary as the second argument of AFF+ (“positive affecting”), Patient as the second argument of AFF− (“negative affecting”). AFF° denotes the relation of non-opposition between the two arguments, and the second argument is simply a neutral Undergoer. The feature ±vol encodes the volitionality of the Actor. Finally, REACT is something like a converse of AFF, which maps Patient into the subject position in alternations involving Experiencer verbs (see Jackendoff 1990 for discussion). 4.2. Frames, Qualia and Dot Object Conceptual structures as presented in the last section appear to have a simple, digital format. Jackendoff (1990, 1996) stresses that these aspects of conceptual structures are most directly relevant to the syntax-semantics interface. While this position overlaps to some extent with many authors that advocate a linguistically relevant semantics as distinct from general conceptual, encyclopedic or pragmatic information (Pinker 1989; Bierwisch 1989; Grimshaw 1993; Levin and Rappaport Hovav 1996; Mohanan and Mohanan 1999), Jackendoff argues at length that there is no need for such an independent component (see esp. Jackendoff 1983, 2000, cf. Lascarides and Copestake 1998). Indeed recent studies have shown that one can not make any a priori decision as to what aspects of meanings are relevant to linguistic structure. Jackendoff (2000) cites examples like bachelor and point (as in a game) from Lakoff (1987) and Searle (1995), making the point that the conceptual structures of these notions are not separable from larger elaborate social frames. The notion of frames, defined as schematic relational structures consisting of roles and relations as background to certain semantic categories, also figures importantly in the work associated with Fillmore (1985) and associates, who advocate a view of meaning called frame semantics. This view has been utilized in a number of recent works in semantic compositionality, for example, in the blending theory of Fauconnier (see Coulson 1997; Sweetser 1999, cf. Matsumoto 1997). Construction Grammar also adopts a frame-semantic view of meaning (Goldberg 1995, 1997b). Fillmore and Atkins (1992) demonstrate that the English word risk is associated with an elaborate conceptual frame consisting of elements like action, benefit and harm. Each element is highlighted in an independent argument structure associated with the word. Since grammatical issues are at stake in these cases, we are confronted with the question of how to formalize this sort of frame-semantic information in lexicalsemantic representations. Jackendoff (2000) underscores the importance of these lexical-semantic complexities, but does not make any proposal as to how to incorporate the ideas into his Conceptual Semantic formalisms. A formalization of certain facets of frame-semantic ideas appears in the independent theory of the Generative Lexicon (Pustejovsky 1991, 1995), specifically in the component termed qualia structure in this framework (for different perspectives on qualia structures, see Briscoe 1996; Jackendoff 2000; Cruse 2000; Allan 2001). Pustejovsky proposes a partitioning of lexical concepts into four dimensions or qualia, extending work by Moravcsik (1975). Cruse (2000) interprets the four qualia succinctly as different “ways of seeing” or modes of construal, including seeing
Constructions, conceptual structures and compositionality
27
something meronymically, taxonomically, seeing something in terms of its interaction with other things, and seeing something from the point of view of its life-cycle. These correspond to what Pustejovsky terms constitutive, formal, telic and agentive quale roles, respectively. In other words, constitutive and formal qualia concern structural and taxonomic information of an entity. Telic and agentive roles are more interesting in that they involve the entity into certain events. The telic role involves the entity into events that define its purpose, “characteristic function,” characteristic action or occupation (see Jackendoff 2000 for elaborations on this point). For instance, the telic role of book is reading. The agentive role encodes information concerning how an entity comes into existence and what it will develop into. The agentive quale of book is that it is created by writing. The qualia dimensions are also found in verbal concepts, allowing for certain qualia having no values. Turning back to Conceptual Semantics, the qualia information can be taken as elaboration of lexical concepts and is crucial in conceptual combination (Jackendoff 1997, 2000), but it is an open question how these insights can be technically and theoretically incorporated into the framework of Conceptual Semantics. Specific questions include the nature (for instance, the defeasibility) of different sorts of conceptual information (Jackendoff 2000, Lascarides and Copestake 1998) and how they could be incorporated into the structured CS representation. We return to these issues in the analysis of the unselected object construction in Chapter 3. Another sort of complication in lexical semantics is represented by the notion of complex type or dot object introduced by Pustejovsky (1995), the idea that certain entities can be multiply categorized simultaneously. A case in point is newspaper, which can refer to an organization (as in The newspaper has apologized), a physical object (The newspaper fell off the table), and information carried in the paper (He is reading the newspaper). The three senses are clustered together in a single meta-entry. This concatenation is represented as follows: (16) [ORGANIZATION•INFORMATION•PHYSICAL OBJECT] In addition, each of the dotted aspects has independent qualia structures, which are linked to each other. Jackendoff (2000) makes interesting extensions of this innovation to include activities (including speech acts), humans and symbols as dot objects.
4.3. Enriched Composition The elaboration of the syntax of conceptual structure in the last section is motivated in a view of semantic compositionality that transcends the classic Fregean notion which stresses the tight correspondences between syntactic combination and semantic composition, an idea that resurfaces in different guises in various frameworks. In Chomskyan generative grammar, this view of compositionality, termed simple composition in Jackendoff (1997), is behind the postulated principles such as the θCriterion and the Projection Principle (Chomsky 1981) and Baker’s (1988) Uniformity of Theta Assignment Hypothesis (UTAH). However, recent work in semantic and conceptual compositionality (Pustejovsky 1995; Jackendoff 1997; Coulson 1997;
Enriched composition and inference in the argument structure of Chinese
28
Sweetser 1999) shows that simple composition is but a default option within a wider range of possibilities which Jackendoff calls enriched composition. According to Jackendoff (1997), there are two aspects to enriched composition that are absent in the standard view of semantic compositionality. First, certain elements in phrasal semantics may not correspond to anything at all in syntactic and phonological structure, as is the case in (1) discussed in section 2.2. Second, the internal structure of lexical concepts (for example, the qualia roles) is visible to conceptual combination on the phrasal level. This is especially shown in adjective-noun combinations where the adjective modifies the telic quale within the lexical semantics of the noun, as with a good knife (Pustejovsky 1995; Jackendoff 1997). The enriched composition, which overlaps with generalized conversational implicature (Levinson 2000), is a pervasive phenomenon in language. Piñango, Zurif and Jackendoff (1999) and McElree et al (2001) conduct psycholinguistic experiments showing that more processing load is involved in enriched composition as opposed to simple composition. At present a useful taxonomy of particular cases is yet to be established, but Jackendoff (1997) has illustrated various cases in the areas of reference transfer (cf. Jackendoff 1992; Nunberg 1979, 1995; Langacker 1993), aspectual coercion (Jackendoff 1991; Pustejovsky and Bouillon 1995; de Swart 1998; Michaelis 2000), argument structure alternations and adjective-noun modifications, constructional idioms, and even certain discourse phenomena (Jackendoff 2000). We can frame the basic issues within the unification-based model described in sections 2.2.2 and 2.3 by making reference to a now-famous case (Nunberg 1979; Langacker 1993; Jackendoff 1997, 2000; cf. Nunberg 1995). Suppose in a restaurant setting there is a need to express the idea that “the person over in the corner wearing a big hat wants more coffee.” The speaker is at least faced with two options regarding how the subject of the sentence should be expressed. Needless to say, one option is simply to use the wording the person over there wearing a big hat. But a more economical version is the following: (17) The big hat over in the corner wants some more coffee. The two versions differ in the number of lexical items used, but the meaning expressed is basically the same, except that (17) is slightly more general: it could mean a person wearing the hat, holding the hat, having the big hat on the table in front of him, as the case may be. To capture this more general meaning, the meaning of (17) is roughly “the person over in the corner contextually associated with a big hat in some way wants some more coffee.” Put plainly, the question is how it is possible that (18) means (19) in the environment of (17): (18) the big hat (19) [PERSON CONTEXTUALLY ASSOCIATED WITH BIG HAT IN SOME WAY] To better see the semantic relations among the elements in (19), I recast (19) in a more detailed tree notation in (20):
Constructions, conceptual structures and compositionality
29
(20) The function ASSOCIATED in (19) appears as the modification relation in depicted in (20).16 The underspecified function F in (20) corresponds to the vague element IN SOME WAY in (19), which could be elaborated by verbal semantics such as WEAR. To ensure “full interpretation,” the default option (i.e., simple composition) is to unify each of the following elements from (20) with lexical conceptual structures: (21) a. PERSON b. F c. Thing d. the modification relation e. the operator-variable relation An instantiation of this option would be the complex NP with a relative clause, the person who wears a big hat. Suppose we are not provided with enough lexical resources to afford a simple composition, except for the lexical conceptual structure BIG HAT. When BIG HAT unifies with the first argument of WANT, conceptual ill-formedness results. Thus the semantic requirement of WANT forces the structure BIG HAT to “expand,” so to speak, such that it unifies with (a-e) of (21), the output of which in turn could satisfy the first argument variable of WANT. Several sorts of reference transfers have been discussed in the works of Nunberg (1979, 1995), Fauconnier (1985), Langacker (1993), Jackendoff (1992, 1997) from different but comparable perspectives. I will return to some of these analyses and see how similar mechanisms operate in Chinese unselected subject constructions in Chapter 4. This sort of conceptual expansion or accommodation is called coercion, which arises out of the need for conceptual well-formedness and is triggered by particular conceptual or contextual cues. A common type of coercion is argument coercion involving either the subject or the direct object of the sentence, which is usually triggered by the governing conceptual function expressed by the verb (Pustejovsky 1995), as in the big hat case. But 16. Here I depart from Jackendoff (2000) who codes ASSOCIATE as a function in the embedded SITUATION. On this analysis, ASSOCIATE must be able to unify with various verbal concepts such as “hold,” “wear,” or even “bought,” which will leave us no option but stipulations. On the other hand, it seems reasonable to interpret ASSOCIATE as the modification relation itself.
Enriched composition and inference in the argument structure of Chinese
30
event (see Zhang 2000a) and aspectual coercion (see e.g. Jackendoff 1991) may be triggered by conceptual modifiers as well. Coercion is one of the better-studied operations within the Generative Lexicon (Pustejovsky 1995, Pustejovsky and Bouillon 1995, etc.). In Pustejovsky’s theory, there are two other generative operations. One is cocomposition, a case in which the semantics of a complement takes the verb as its argument and shifts its event type. For example, in the phrase bake the cake, the agentive quale for cake, i.e., a baking process, unifies with process expressed by the verb bake, and in that process the formal role of cake becomes the formal role for the entire VP, thus accounting for the creation sense of bake (see Pustejovsky 1995 for details, cf. Levin and Rappaport Hovav 1988). Still another operation, termed selective binding, relates to adjective-noun combinations in which the adjective selectively modifies an event expression in the qualia of the head noun, as with fast typist (see Jackendoff 1997 for several sorts of examples). From the perspective of the present framework in which generic conceptual patterns are part of lexically-stored constructions, which are to be unified with more specific lexical conceptual structures, I take the three operations sketched in the above paragraph to be some of the mechanisms that function in the process of conceptual unification within a general theory of enriched composition. Here we see more clearly that enriched composition entails richer semantic representation discussed in section 2.4.2. As has been pointed out by Jackendoff (2000) and Lascarides and Copestake (1998), semantic composition via qualia information is subject to contextual defeasibility. Thus a theory of enriched composition must interact with a model of contextual knowledge. While this research indubitably overlaps with research in pragmatics, Jackendoff (1992, 1997) points out the grammatical ramifications of enriched composition and argues for an unified rather than a ‘modular’ approach based on a rigid division of semantics and pragmatics. As noted earlier, Chinese “cultural grammarians” tend to make much of the fact that there is often much more semantic content underneath an utterance in Chinese, though they lack the tools to make explicit the intuition. The theory of enriched composition would then provide a tool having the right property. In this thesis, enriched composition will be shown to play a crucial role in argument licensing.
5. Summary The grammatical knowledge of a language is a structured inventory of varieties of generic grammatical patterns or archetypes (l-rules), plus more specific sorts of lexical items, all stored in the lexicon in the same format. These stored items, called constructions, are linked to each other through inheritance hierarchies, which capture linguistic generalizations in an economical fashion. A fully specified construction is a three-way correspondence between conceptual structure, syntactic structure and phonological structure. In addition, there are “defective constructions” that have one or more of the dimensions underspecified or simply having null values. The only combinatorial operation needed in the grammar is unification of compatible information on the three dimensions, in a parallel mode. The semantic side of a construction is treated within the theory of Conceptual Semantics, which is taken to consist of an algebraic
Constructions, conceptual structures and compositionality
31
syntax and possible elaborations based insights from Frame Semantics. How these elaborations should be incorporated into Conceptual Semantics is an open question. I will discuss this issue further in the particular analyses in the rest of the dissertation and will attempt to make specific proposals. In this connection, work in the framework of Generative Lexicon, particularly its innovations such as qualia structure, dot object, coercion and other operations, will serve as a useful basis. I focus on the unification of LCS’s with generic conceptual patterns under the assumption of enriched composition. Throughout the thesis, it will be shown that “simple composition” plus an abstract syntax to deal with syntax-semantics mismatches is not an adequate methodology to deal with the full range of data, aside from its possible ontological and psycholinguistic defects (see Chapter 1 for the priority of semantic solutions).
CHAPTER THREE The Eat Restaurant Construction: Licensing Unselected Complements
A central issue in any theory of grammar concerns the determinants of syntactic structure, in particular, where do elements in syntactic structure come from and what determines the way these elements are expressed in the way they are. The standard assumption in the 1980s was verb-centered and lexical-semantically based: the syntactic expression of arguments is projected from the lexical-semantic properties of verbs and other predicators (for discussions see Wasow 1985; Levin and Rappaport 1996; Arad 1996; Mohanan and Mohanan 1999). The idea is explicit in the Projection Principle (Chomsky 1981), the attempt to derive c-selection from s-selection, the idea of Canonical Structural Realization (Chomsky 1986) and Baker’s (1988) UTAH. On this approach, verbs license arguments by assigning a thematic role to them, in line with the θ-Criterion. However, recent studies in argument expression variation, verbal alternations or lexical-syntactic flexibility from various perspectives pose challenges to this general approach, as discussed in chapter 1. (Ghomeshi and Massam 1994; Rappaport Hovav and Levin 1998; Arad 1996; Goldberg 1995; van Hout 1998; Jackendoff 1990, 1997a, b, 2000, among many others) The solutions to Projectionist problems are of different sorts, ranging from semantic and syntactic strategies to those that invoke multiple lexical entries (see Arad 1996 for discussion). I don’t intend to review various approaches in this chapter (see the literature referred to above for details; see also Chapter 1 above). In keeping with the general goal of elucidating the relation between conceptual structures and syntactic forms, I focus this chapter on the licensing of complements in simple verb phrases in Chinese, which often exhibit lexical-syntactic flexibility. I pursue a semantic direction for the phenomena, drawing on an explicit theory of the syntax of conceptual structures and conceptual combinations, as presented in Chapter 2. As part of the argumentation, comparison will be made to relevant syntactic analyses as well, showing certain empirical insufficiencies of syntactic proposals. The general conclusion is that the independently motivated theory of enriched conceptual structures and composition render abstract syntactic operations unnecessary. Specifically I discuss the variable semantic relations between the verb and its complement in Chinese verb phrases in comparison with English (Section 3.1), focusing on a particular type that has been called unselected object and how it is semantically licensed under an enriched composition approach. In Section 3.2, I discuss the general
The eat restaurant construction
33
semantic properties of a particular type of unselected object construction, various possible approaches and issues involving metaphor and frozen forms. In Section 3.3, I discuss more formal properties of the construction and test the syntactic and semantic status of the postverbal nominal in the construction. Section 3.4 is devoted to the conceptual structure representations of the construction and its major constituents, which are motivated in independent grammatical properties. I present a demonstration of how the conceptual pieces sorted out are combined through enriched composition in section 3.5, and how this analysis fares with a range of cases. It is shown how certain complicated cases are handled under the assumption of “cluster” concepts in a preference rule system (Jackendoff 1983). I show how points of idiomaticity can be located in a principled way in section 3.6. To conclude, I compare my proposal with three syntactic analyses in section 3.7, highlighting the strength of my approach. The general result points to a sort of criteria-governed productivity (see section 5.2.3. cf. Pinker 1989) that controls the learning of such constructions.
1. The Diversity of Verb-Complement Relations It has often been observed that the prototypical relation between a two-argument verb and its complement is that of an Agent (or Actor, to use the terminology introduced in Chapter 2) acting on a Patient, which in turn undergoes a change of state. This has often been adopted as a standard assumption in Cognitive Grammar (see e.g., Horita 1996; Croft 1998; cf. Hopper and Thompson 1980). Langacker (1990:215) observes of a prototypical transitive clause that “it originates with a canonical agent (volitional energy source) and terminates with a canonical patient (energy sink).” The following illustrates: (1) John drank the water. (2) Floyd broke the glass. In other words, a canonical transitive clause is characterized by a fully specified action tier, containing the semantic roles Actor and Patient. However, this is by no means the whole picture, as a variety of other possibilities are available (see Ghomeshi and Massam 1994 for a similar observation). The following examples from Jackendoff (2000) nicely illustrate this: (3) a. John imagined/mentioned his pet elephant. b. The audience applauded the clown. c. Emily likes/fears/despises cats. d. Capitalism frightens/bothers/disgusts Sam. e. Liana left/entered the room. f. A wall surrounds/divides the city. g. This book comprises 15 chapters. h. The meeting lasted seven hours. i. Larry lacks money. j. My patience equals/exceeds my enthusiasm. k. The doctor underwent an operation. l. The outcome means trouble.
Enriched composition and inference in the argument structure of Chinese
34
Every sentence in (3) can be said to have a “defective” action tier in some way, either one of the action tier roles is absent (as in a, b, e, k) or the Actor is not volitional (as in d) or both action tier roles are absent (as in g, h, i, j, l). In addition, the object of the verb in these sentences functions in varying semantic roles, such as Stimulus (as in c), Goal or Source (as in e), and many other roles for which it is difficult to assign a label. Nevertheless, in one way or another, the object complements all intuitively play a certain role in the semantic relation expressed by the verb. In other words, the object is a semantic argument of the verb and thus s-selected by the verb (I return shortly to a refinement of the notion s-selection). Take (3e) for example. Assuming with Jackendoff (1990) that leave/enter are verbs of motion that semantically incorporate a Path function, such as FROM and TO, we can argue that the Goal/Source that is expressed as the syntactic object is s-selected by this incorporated Path function, and hence by the semantics of the verb leave/enter. Likewise, more abstract relations expressed by lack, equal, mean, divide, and comprise all have their relevant argument variables satisfied in the relevant examples. Thus s-selected complement is a property shared by this apparently heterogeneous set of sentences. It would be instructive to contrast the above set of sentences with the set of complement options available in Chinese. One will notice that the complement in Chinese serves as well in a variety of semantic roles, a situation we have seen in (3) above.1 I present the possibilities below, based on A Usage Dictionary of Chinese Verbs (Meng et al. 1994), which documents 14 semantic classes of NP complement, and Xin (1996) (the semantic role labels are as given in Meng et al. 1994 and Xin 1996, and of course are not conclusive): (4) a. Patient wa tu (dig (out) the soil), shaole zheben shu (burn this book); gaibian guanxi (change the relation). b. Result wa dong/zhan hao (dig hole/tunnel: to make a hole/tunnel by digging) c. Purpose wa jinzi/hongshu (dig (to obtain) gold/red potato), bi zu (pressure (to get) the rent), kao yanjiusheng (exam postgraduate: to write the exam to become a graduate student) 1. By assuming that the postverbal NPs in these cases are complements, I am taking the position that they all serve as a sister to a verb in syntactic configuration, thus excluding the possibilities of noun incorporation and postverbal adjuncts, which Diane Massam has raised (p.c.). In section 3.3, I justify this position in some detail regarding a subset of the examples represented by (4). However, the main point of this chapter is how the postverbal element (whatever its syntactic status) composes semantically with verbal semantics. Thus the semantic analysis to be proposed will not be affected by how one resolves the syntactic status of the postverbal element.
The eat restaurant construction
35
d. Agent i. Taishang zuozhe zhuxituan. podium sitZHE presidium “On the podium sat the Presidium.” ii. Duimian paolaile yi tou niu. opposite run-comeLE one CL cow “Right in front came a cow.” e. Equative Wo ti zhong feng I play center-fielder f. Locative shui xiao chuang (sleep small bed), qi wugen (get up the fifth watch of the night: to get up before the dawn), chi shitang (eat restaurant). g. Instrument chou bianzi (lash whip: to lash with a whip), chou yiandou (smoke pipe: to smoke with a pipe), jiujile daliang yaopin (relieve big-amount medicine: relieve (with) a large amount of medicine). h. Manner chang A diao (to sing (in) the A key), cun huoqi (save (in) a savings account), xie fangsongti (write calligraphy (in) Song Style) i. Reason chou jingfei wenti / hunyin dashi (worry (over) funding problem/big issue of marriage) j. Cognate chang ge (sing song), zou lu (walk road: walk), chui qi (blow air), die gentou (tumble a fall), chao jia (quarrel a quarrel). As can be seen in the above set of Chinese verb phrases, the semantic relations between the verb and its complement vary as much as and perhaps more than the set in (3), even though the labels given in the above may still be open to discussion in certain cases, as this chapter will make clear with respect to Instrument and Manner. In the same way as the examples in (3), the canonical Actor-Patient relation is absent in most cases of (4). This diversity shown in (3) and (4) challenges the Projectionist view with the tension between a lexicon burdened with multiple lexical entries and the obvious relations between the entries that need to be captured (see chapter 1 for discussion). A quick and intuitive look at the verbal lexical semantics in (4) shows that several varieties of complements are not s-selected by the verb in any obvious way, by contrast to the uniform s-selection in (3). In other words, these examples violate the Head Constraint of Jackendoff (2000), according to which the syntactic arguments of a clause express the semantic arguments of the verb in the clause (see Jackendoff 1990, 1997b, 2000; Goldberg 1995 for various cases of such violations). The relevant categories include Purpose, Locative, Instrument, Manner, Reason and Cognate. For instance, in (4c), yanjiusheng ‘graduate student’ is apparently not a character in the relation expressed by kao ‘exam,’ but a purpose of this relation, neither is jia ‘quarrel’ in (4j) a semantic argument of chao ‘quarrel’ (see Massam 1990 for a different view concerning the thematic status of cognate object, cf. section 4.2.1).
Enriched composition and inference in the argument structure of Chinese
36
Note however that I have so far been relying on an intuitive notion of s-selection, which needs to be made more precise. Given the need for an elaborate semantic structure incorporating qualia and other frame-semantic information, as discussed in chapter 2, the categories picked out as ‘unselected’ in the last paragraph could all be argued to be selected in some sense. The hierarchical linking theory developed in Conceptual Semantics (Jackendoff 1990), as well as comparable linking theories in other frameworks (Foley and Van Valin 1984; Grimshaw 1990; Bresnan and Kanerva 1989, among others), offers a way to formalize the notion ‘unselected’ argument. Such linking theories suggest an ordered mapping between a thematic hierarchy and a syntactic hierarchy. Given the configurational theory of theta roles (see chapter 2), one can assume an ordering of theta roles and other semantic functions like the following (adapting Jackendoff 1990)2: (5) action tier arguments from left to right→main conceptual clause arguments→ most deeply embedded semantic elements Such a hierarchy of conceptual arguments is then mapped into a hierarchy of syntactic positions, such as (6) (cf. Jackendoff 1990): (6) [s NP…]→[vp V NP]→ [vp V…NP…] Under such a mapping system, the notion ‘unselected’ argument can be (partly) defined as an argument that does not respect the ordered mapping between (5) and (6). A typical case would be that an argument that expresses a lower role in (5) is mapped onto a higher position in (6). To return to the several classes of ‘unselected’ arguments intuitively picked out above, we can now base our judgement on the working definition just given. Postverbal elements that putatively express Purpose, Location, Instrument, Manner and Reason (4c, 4f, 4g, 4h, 4i) are arguably all elements embedded in some conceptual function apart from verbal semantics (see Jackendoff 1990 for an analysis of Instrument in this spirit). For instance, Location is by definition the argument of a Place function. Therefore, a Location expression is unselected if verbal semantics does not incorporate a Place function, which is arguably the case in (4f). The examples in (4f) contrast with an English example such as Kangaroos inhabit Australia, where the postverbal complement is selected because a Place function is incorporated within the semantics of the verb inhabit. In addition, cases involving Instrument, Manner and Reason can all be analyzed as embedded in a modifying conceptual clause, and thus ‘unselected’ by verbal semantics. A residual case not covered by my definition of ‘unselected’ argument concerns the ‘cognate’ object construction, which may be unselected in the sense that the cognate object can be argued to specify the event itself, rather than satisfying an argument variable of the verb (see section 4.2.1, cf. Massam 1990). The immediate question around unselected arguments is how are they licensed, since, by virtue of being unselected, they are not licensed by being assigned a thematic role of the verb. In what follows, I focus on certain uniform 2. Jackendoff (1990) argues further that whether a conceptual argument can be syntactically expressed must be stipulated in the lexical entry, as A-marked, and a set of A-markings on conceptual structure is equivalent to argument structure in other theories (cf. note 1, chapter 1).
The eat restaurant construction
37
compositional properties associated with Locative, Instrument and Manner3 classes of complements listed in (4). The methodology to be illustrated extends to other possible classes of unselected complements. Before we proceed, a short note on productivity is in order, as it will play an important role throughout this chapter in a way paralleling recent discussions of morphological processes (Jackendoff 1997, 2000; Koenig 1999a). First, it should be noted that English does exhibit cases of unselected complement, contra Lin (1999), as with cases such as smile her thanks, snort her disgust, wave one’s goodbye (Levin and Rappaport 1988; Massam 1990). However, the pattern does not appear to be fully productive, witness, ??smile her excitement/happiness, ??sneer her hate. Likewise, certain cases in Chinese are not productive either. For instance, while suo shui (shrink (due to) water) arguably falls into the pattern of V+reason (cf. 4i), we don’t have *suo diwen (shrink (due to) low temperature). Therefore suo shui should not be taken as an instance of the relevant pattern. We will have the chance to return to the issue of productivity below.
2. The Instrumentality in the Eat Restaurant Construction The unselected complement classes mentioned above are best illustrated by the verb chi (eat) combining with a Location or Instrument type of complement, as in chi shitang (eat restaurant), chi kuaizi (eat chopstick), as opposed to its regular complement such as chi neiwan fan (eat that bowl of rice). In (7) and (8) is shown the plausible linguistic context for such phrases. (7) ta meitian gongzuo hen mang, mei shijian zuo fan, zhihao chi shitang. he every-day work very busy NEG time do meal have-to eat canteen “He is busy with work every day and does not have time to prepare his own meal. So he has to have his meals at the restaurant (in his workplace).” (8) ta shang yue shouru bu cuo, zongyu chi le yici guanzi he last month income NEG bad finally eat LE one-time restaurant “He earned a lot last month. Finally he got a chance to eat at a restaurant.” Similar sorts of verb phrases with no sentential context are given below: (9) xi
liang shui wash cool water “to have a shower with cool water”
(10) xie Mama mai de maobi write mother buy Mod.writing-brush “to write with the writing brush that Mother bought” (11) ni chi zhege da wan you eat thisCL big bowl “You eat with this big bowl.” 3. I am using these labels here only for expository purposes. In the course of this chapter I will attempt to decompose them.
Enriched composition and inference in the argument structure of Chinese
38
(12) shui jiali de xiao cuang sleep home-inside Mod.small bed “to sleep in the small bed at home” (13) xie haiban write blackboard “to write on a blackboard” (14) da zhugongshou play centerfielder “to play in a game as a centerfielder” The instances uniformly focus on the means or manner in which a certain action is undertaken. They can be used to address How rather than What type questions. This is illustrated for the verb chi below: (15) How do you solve your meal problem every day? __Wo mei tian chi shitang I every day eat canteen (16) What do you eat everyday? __* Wo mei tian chi shitang In other words, in the answer in (15), shitang contrastively expresses the means by which a meal event is accomplished, as opposed to other means such as doing the cooking oneself. More precisely, shitang serves a role in the means by which the meal event is accomplished, thus instantiating a typical property of the role Instrument (Jackendoff 1990). Jackendoff (1990) has shown that the role Instrument can be decomposed into a modifying action tier configuration (see section 4.1). Thus the postverbal Instrument argument is ‘unselected,’ according to the definition given above, and it is not a role assigned by verbal semantics. The property of having a contrastively expressed Instrument sets this construction apart from similar forms found in certain varieties of languages. Consider a formal analogue in English, such as go home. In examples like this, the postverbal nominal is by no means contrastive. To investigate the way the postverbal element is licensed in this construction, we need to focus on the semantic contribution of the postverbal NP’s, taking as a starting point the intuition described in the last paragraph, namely, that the postverbal element serves a role defined by a modifying action-tier configuration, rather than ‘selected’ by the semantics of the verb. However, most analyses of this sort of construction (e.g. Lin 1999; Feng 2000; Gao 1996) pursue a reverse direction. Implicitly or explicitly, these analyses treat the postverbal NPs as being selected by the main verb, expressing certain semantic roles assigned by the verb4. There are a number of difficulties with this line of analysis. First, the verb-centered approach with respect to such examples leads to a heterogeneous set of roles, for example, location (4f, 12), instrument (4g), manner (4h), which is at odds with 4. Lin (1999) actually uses the term ‘unselected’ object, but she does not explicitly define the notion. Somewhat contradictorily, she also treats the postverbal element as expressing a theta role of the verb.
The eat restaurant construction
39
the uniform meaning of such constructions described above. Second, taking such roles as ‘selected’ by the verb cannot explain certain internal relations between the suggested roles. Take shui jiali de xiao cuang ‘sleep in the small bed at home’ in (12). It may be plausible to construe “small bed” as the location for sleeping, but “small bed” could equally plausibly be taken as the instrument that enables the sleeping event, as well as the manner in which the sleeping action is undertaken (see Goldberg 1997c for the relation between Instrument and Manner).5 The ‘unselected’ analysis, to be made precise later, avoids such a problem, since the postverbal element is taken to assume a uniform role in a modifying conceptual clause independent from verbal semantics. Third, on the verbcentered role analysis, there are cases where it is hard to see how the postverbal element fulfills a role in the semantics of the verb, as also noted by Xin (1996). The following illustrates: (17) Wode didi mei zhou pao-lai chi wo my little-brother every week run-come eat I “My brother comes here to eat on my expenses every week” In this case, there is a sense that “I” is negatively affected due to the event of eating, thus an argument of the function AFF¯. As Diane Massam (p.c.) suggests, wo ‘I’ could be assigned a role label, say, ‘Maleficiary’ (thanks to Larry Horn (p.c.) for suggesting the term). However, the lexical semantics of chi does not plausibly specify such a human maleficiary as an argument; what is affected in an eating event is only the thing that is consumed. Again, the semantic role of wo is easier to grasp if we assume that it is assigned by AFF embedded in a modifying conceptual clause, i.e., not assigned by the verb. Finally, where the postverbal element apparently expresses a location, the construction cannot be used to answer an adverbial Where question: (18) ta zai nali chi fan? he Loc. where eat meal “where is he having his meal?” __* ta zai chi guanzi. he Loc. eat restaurant (19) ta zai nali shuijao6? he Loc. where sleep “where is he sleeping?” __* ta zai shui safa he Loc. sleep sofa Therefore the semantic role Location is not an accurate characterization of the semantic role of the postverbal element. If it were, it would make wrong predictions concerning problematic cases such as *chi chufang (eat (in) kitchen). Pushing the argument further, I 5. Note that the problem here is not about the role labels, but that certain roles are better decomposed and separated from verbal semantics, on finer-grained conceptual analysis, as Jackendoff (1990) has done for the putative role Instrument. 6. However, there is another way to ask the question, with the question word put after the verb:
Enriched composition and inference in the argument structure of Chinese
40
suggest that the postverbal element in the type of patterns we are concerned with does not directly express a semantic role in the relation encoded by the verb. Otherwise we would not be able to account for other problem cases of a similar sort, such as *chi wan (eat (with) bowl), *shui chuang (sleep (in) bed). The analysis to be proposed in this chapter is intended to cover such cases (see section 5.2.3) In conclusion, the original intuition, that the postverbal element is part of a means/instrument expression independent of the semantics of the verb, seems to be on the right track. Furthermore, I subsume Manner with Means/Instrument, following Goldberg (1997c). Given this basic semantic characterization of the contruction, let us discuss the possible analyses one could adopt One option is to suggest that the postverbal element is simply juxtaposed with the verb without special case marking of its semantic roles. This could be taken as a trace of protolanguage, understood to be an evolutionary stage on the pathway to full language (Bickerton 1990), characterized by meaningful words and little or no syntax. Such properties can i. Ni shui nali? you sleep where “where do you (want to) sleep?” In this case, shui da chuang (sleep big bed) can serve as an answer, and da cuang would then be a real Location. This of course is not surprising, since the V+complement combination expresses various relations, as we have seen. Thus shui da cuang is potentially ambiguous.
be found in modern languages as well,7 as well as showing up where full language is unavailable. However, this suggestion is based on the assumption that the postverbal element takes up a semantic role of the verb. This assumption is convenient but not really viable, as I have shown above. It also leaves unexplained the problem cases mentioned earlier, such as the contrast between xie maobi (write (with) writing-brush) and *xie bi (write (with) writing-tool). These problem cases might drive us to the conclusion that there is something idiomatic about this sort of example, a second option we could adopt.8 For instance, we could take shui da chuang (sleep big bed) as a fixed form, thus ruling out the variant shui chuang (sleep bed). However, all the problem cases seem to fall into a pattern in which the postverbal element expresses some supposedly redundant information, as in the “sleep bed” case. This suggests that there must be something more regular than an idiom analysis could allow. In addition, unlike typical idioms, the general meaning of the construction is closely related to (though more than) its constituent meanings. Thirdly, the pattern V+NP (expressing means) is replicated in many examples: instances of the 7. In personal communications, both Ray Jackendoff and Talmy Givón made suggestions to this effect, on the basis of a representative subset of the data presented in this chapter. In particular, Givón takes this as a case of under-grammaticalization common in pidgin and in the speech of “pre-grammatical” children. 8. Discussions with Ray Jackendoff have pushed me to consider this possibility, for which I am grateful.
The eat restaurant construction
41
pattern at least number on the order of dozens from my random search of a verb dictionary (Meng et al. 1994, see also Gao 1996 for the same point and more examples). Some more examples are given below: (20) a. bangzhu dier qian help a-bit money “help with some money” b. bao xiao bao wrap small bag “wrap into small bag” c. cai gudian/yangge diao stamp drumbeats/Yangge tune “dance to the drumbeats/the Yangge tune” d. chang qinyi “sing Qingyi (a particular sort of female role in Chinese opera)” e. chao kapian/da gaozi copy card/big writing-paper “write (on/with) the card/big writing-paper” f. cui hua fei expedite chemical fertilizer “expedite the growth with chemical fertilizers” It should be noted that it is not as easy to collect relevant examples for our purposes as to collect those showing, say, particular case-marking properties, since the examples must fit the semantic pattern we are looking at, which is itself subject to linguistic and contextual coercion. This however does not detract from the productivity of the construction under discussion. Xin (1996) cites a nonce example, which is only acceptable in context, (21) haizimen, nimen zhenme zai kezuo shang huluan xiezi? child-PL you-PL why Loc. desk on careless write-character “Kids, why are you guys scribbling on the desk?” yihou bu zhun zai xie kezhuo, zhidao ma? hereafter not permit again write desk, know Q “From now on don’t write on the desk again, got it?” I take this as showing that the pattern is psychologically productive. Thus there is more to the idiom proposal: we need to be more precise about what is stored, what is constructed online (Jackendoff 1997, 2000). A third option, which is the one to be illustrated in this chapter, is to take an alternative approach, drawing on the intuition suggested earlier. The essence of the idea is that the postverbal element is not directly related to the semantics of the verb, but to a certain semantic function embedded in an elaborate semantic structure which is paired with the syntactic pattern (cf. Goldberg 1995; Jackendoff 1990, 1997b, 2000). This approach is essentially a Constructional view.
Enriched composition and inference in the argument structure of Chinese
42
Before going further, let us stop and consider certain related patterns. I believe that native speakers have a strong sense of certain connections between the examples cited so far and those of the following: (22) a. chi lixi eat interest “to benefit from the interest” b. chi laoben eat old-capital “to live off one’s past gains” c. kao shan chi shan, kao shui chi shui rely mountain eat mountain, rely water eat water “those living on a mountain live off the mountain, those living near the water live off the water” (making use of local resources) However, the expressions in (22) do have a strong idiomatic flavor and often involve metaphor, as can be seen from the translations. How to characterize the relations between this set of instances and the more transparent cases in (7–14)? I take this as another challenge for any proposal that sets out to deal with such constructions.
3. More on Formal Properties In the last section, I discussed the general semantic property of the construction and stated the general approach that will be demonstrated in this chapter. While I am mainly concerned with the semantic structures of the examples, it may be enlightening to discuss briefly the status of the postverbal element with respect to both its syntax and semantics. I briefly discuss in this section the syntacticosemantic properties of the postverbal element in examples like (7–14) and its argumenthood. First, while it may appear that the postverbal element is often a bare noun, as with chi shitang in (7) (see also 13, 14), noun phrases are certainly allowed as well, as seen in (10–12), which instantiate relative clauses and determiners. Second, the postverbal element can be separated from the verb by a frequency expression, such as yici ‘once’ in (8). Third, in terms of referential properties, the postverbal noun phrase can be either generic, as in (7) and (8), or definite and specific (10–12). Fourth, it is easy to see that the postverbal noun phrase in the relevant examples is not optional, cf.: (23) a. ta kanle liangge xiaoshi he readLE twoCL hour “He read for two hours.” b. ta kanle he readLE “He read (it).” c. ta chang chi guangzi he often eat restaurant “He often eats at a restaurant.” d. *ta chang chi
The eat restaurant construction
43
he often eat These properties point to the independent noun phrase status of the postverbal element, ruling out the possibilities of noun incorporation as well as postverbal adjunct, Furthermore, many of the verbs in such constructions naturally take an internal argument, such as chi ‘eat,’ xie ‘write.’ But the internal argument is disallowed in such constructions: (24) a. *chi mian shitang eat noodle restaurant b. *chi shitang mian eat restaurant noodle It seems then that the postverbal noun phrase occupies the internal argument position, which I assume is a sister position to the verb, that is, a complement position. This is corroborated by further tests below. A number of tests are standardly used to probe argument/adjunct status, a topic much discussed in recent work in constraint-based theories, especially HPSG (see Huang 1982 for GB ideas; cf. Pollard and Sag 1987; Bender and Flickinger 1999; Bresnan 2001). However, it’s not always easy to tease apart the semantic and the syntactic dimensions, And it’s not clear if the tests apply to all languages. Some of the problems are discussed in Verspoor (1997). Based on Verspoor’s discussion, the following tests will suffice for our purposes. Syntactic Test 1: A standard syntactic test for argument structure is the do so test (Lakoff and Ross 1976).The idea is that do so can substitute for a V’ constituent. In the case of a string such as Vxy, if Vx can be replaced by do so in isolation from y, then y is acting as a V’ adjunct. If only the whole string can be replaced by do so, then both x and y must be complements. A simple example follows, which shows that lunch must be a complement to the verb: (25) *Sue cooked lunch, and Fred did so dinner. Employing such a test, Verspoor (1997) shows that give and tell are to-dative verbs that take a PP complement, whereas send and kick are to-dative verbs that take a PP adjunct (see also Jackendoff 1990). The following contrasts are from Verspoor (1997): (26) a. Adam gave a book to Debbie and Brian also did so. b. *Adam gave a book to Debbie and Brian did so to Susan. (27) a. Sam sent a letter to Bill and Mark also did so. b. Sam sent a letter to Bill and Mark did (so) to Susan. As an equivalent to the English do so test, we look at the substitutional possibilities of zheyang (zhu) (“do so”) in Chinese. With respect to our examples, if the verb alone can be replaced by zheyang zuo, it shows the postverbal nominal is an adjunct. If only the full string {verb NP} can be replaced, the NP must be construed as a complement. (28) a. Zhangsan jingchang chi guanzi, Lisi ye zheyang (zuo) often eat restaurant too so do “Zhangsan often has his meals at restaurants, and Lisi does so too.”
Enriched composition and inference in the argument structure of Chinese
44
b. *Zhangsan jinchang chi guanzi, Lisi que zhihao zheyang ziji but have-to so self “Zhangsan often has his meals at restaurants, but Lisi has to have his meal (on) himself.” As (28) shows, Chinese “do so” can only replace the full phrase chi guanzi ‘eat restaurant,’ not the verb chi ‘eat’ alone. Thus the postverbal nominal occupies a syntactic complement position. This conclusion confirms the optionality test illustrated in (23) above. Syntactic Test 2: The relative order of complements and adjuncts has been used as a test for the English data. In English, adjuncts tend to be ordered after complements, as Pollard and Sag (1987) point out. The following contrast illustrates: (29) a. Adam gave a book to Debbie in the library. b. *Adam gave a book in the library to Debbie. (ok if the PP is adjunct to NP) In Chinese, however, things are more complicated (see Ernst 1999 for a recent discussion): there are different types of adjuncts distributed both before and after the verb, and adjuncts do not appear to be ordered after complements the way they are in English. According to Ernst (1999), the only types of postverbal adjuncts in Chinese are duration and frequency expressions, as well as manner and result expressions marked by the morpheme de, though earlier studies (e.g. Huang 1982; Li 1990) suggested that Chinese adjuncts are preverbal. Note that the postverbal noun phrase in examples like chi neijia guanzi ‘eat (in) that restaurant” may express some kind of manner or means, as argued in section 3.2, but it cannot be marked by de. Furthermore, duration and frequency expressions are known to occur either before or after the postverbal complement, as shown below: (30) a. ta kanle naben shu liang tian/liangci he readLE thatCL book two day/twice “He read that book for two days/twice.” b. ta kanle liangci/liang tian naben shu he readLE twice/two day that book The same pattern of variation is attested in chi guanzi type of examples: (31) a. ta zhongyu shuile da chuang liang tian he finally sleepLE big bed two day “He finally slept in a/the big bed for two days.” b. ta zhongyu shuile liang tian da chuang he finally sleepLE two day big bed “He finally slept for two days in a/the big bed.” c. ta shangge yue cai chile guanzi liangci he lastCL month just eatLE restaurant twice “He only ate in a restaurant twice last month.” d. ta shangge yue cai chile liangci guanzi he lastCL month just chiLE twice restaurant “He only ate twice in a restaurant last month.”
The eat restaurant construction
45
The pattern of variation concerning the order of frequency and duration expressions confirms again the complement status of the postverbal noun phrase in our relevant examples. More relevant to our purposes, let us invoke two semantic tests with respect to the status of the postverbal element. One test involves constancy of semantic contribution (Pollard and Sag 1987), the idea being that an adjunct contributes to the meaning in a uniform fashion. As discussed in section 3.2, the postverbal nominal is best construed as part of a means/instrument or manner expression: apparent cases of location should arguably be reconstrued as expressing means as well. Goldberg (1997c), in discussing the English way-construction, invokes several cases of lexical polysemy to motivate an inheritance link between means and manner, taking manner as an extension of means. In light of Goldberg’s claims, the semantic contribution of the postverbal nominal in our examples appears rather constant across different verbal heads. Thus it follows that the postverbal element is a semantic adjunct. Another semantic test appeals to entailment: complements, rather than adjuncts, are entailed by the sentences containing the verb. This is in accord with the widely accepted idea that complements fill an argument role in the semantics of the selecting head, while adjuncts never do. Consider for example the sentence below: (32) Zhangsan chile (fan) eatLE (meal) In this sentence, it is entailed that there is something that Zhangsan ate, but it does not entail that Zhangsan’s eating something is causally related in any way to another participant such as a restaurant This again points to the semantic adjuncthood of the postverbal element. The same notion of adjunct is adopted in Jackendoff (1990). To conclude, the postverbal element is a complement syntactically, but an adjunct, or part of an adjunct, to be more precise, in a semantics sense. Therefore we are confronted with a deep mismatch between syntax and semantics in the sense that the hierarchical mapping between (5) and (6) above is not respected, with part of a semantic adjunct (an argument deeply embedded in a modifying conceptual clause) mapped into a syntactic complement position. Without appealing to ad hoc devices, it is hard to see how syntactic approaches invoking transformations could smooth out such mismatches (cf. Huang 1994, 1997; Lin 1999). On the other hand, the parallel architecture adopted in this dissertation seems a more natural way to handle such cases of mismatches, as will be demonstrated in the remainder of the chapter.
4. Sorting Out the Conceptual Pieces To recapitulate the discussion so far, I have argued that the construction instantiates a simple V+complement NP pattern, which expresses the means/instrument through which a particular action is undertaken. The postverbal nominal should be interpreted as part of a means expression, which subsumes apparent cases of location and manner. Now the question is how the postverbal element is semantically licensed under the unification approach. Note that semantic adjuncts by themselves do not pose the problem of licensing in the way syntactic arguments do: they simply compose with matrix conceptual
Enriched composition and inference in the argument structure of Chinese
46
structure through restrictive modification (Jackendoff 1990, 2000; see also Chapter 2) or by some other means favored in formal semantics. With respect to the examples under discussion, the licensing issue arises because of two crucial facts: (1) the mismatch between the syntactic and the semantic argument structure discussed in the last section; (2) the incomplete expression of means/manner information by a noun phrase: the surface syntax of chi guanzi does not serve to flag such information. Formulating the issue in somewhat different terms, we are then concerned with how the postverbal complement NP in our data set semantically composes with the verbal head. To answer this question, we must determine the various conceptual constituents in the construction. In what follows, I largely focus on the verb chi (eat) and its unselected object such as shitang (canteen) and da wan (big bowl). It will be shown later how this analysis extends to other verbs. In addition, while unification is assumed to proceed in a parallel fashion on the dimensions of syntax and semantics (Jackendoff 1997), the emphasis is given to a detailed demonstration of semantic operations. 4.1. The CS Representation of the Unselected Object Construction I have suggested in the above that there is a certain mismatch between syntactic form and meaning associated with the examples under discussion, given the hierarchical mapping assumed between (5) and (6). Another factor contributing to the mismatch is, as also mentioned, that the Means/Manner information is signaled neither in the morphosyntax nor in the semantic constituents of the construction (cf. the English markings such as with, by means of). Such a mismatch in fact suggests the existence of a construction, in the sense of Construction Grammar (Goldberg 1995; see Chapter 2). However, it does not seem plausible that a V+NP pattern is conventionally (i.e., very tightly) associated with the kind of semantic properties discussed in section 2, and this is not a theoretical necessity, given the notion of defective construction discussed in Chapter 2.9 This consideration leads to an analysis in which the eat restaurant type of examples are taken to instantiate two separate constructions, as will be proposed later in this section. In such cases of mismatch, we could, following Jackendoff (1990, 2000), appeal to conceptual paraphrases, where conceptual and syntactic structures are better matched, to reveal the conceptual structures left unexpressed in the syntactic structure. Let us first return to the Instrument/Means status associated with the postverbal element. There are two general properties of Instrument noted by Jackendoff (1990). First, the Actor (defined on the action tier, see Ch.2) acts on the Instrument. This seems true in our cases as well: the diner must somehow act on or affect the shitang ‘canteen’ or da wan ‘big bowl’10 in order to eat shitang/da wan. Second, the Instrument acts on the 9. See below for certain complicating cases relating to the (non)conventionality between the meaning and the form in our cases. In section 5.1, I offer an important qualification to the defective construction analysis adopted here. 10. The diner acts on shitang ‘canteen’ in the sense that he/she must in one way or another deal with the staff in a canteen, to get a meal. Thus “canteen” is both a physical object and an institution, a dot object (Pustejovsky 1995; see Ch. 2). The diner acts on da wan ‘big bowl’ in the sense that he/she must manipulate it as a physical object, in an eating event. The event of affecting is defined in terms of transmission of force (Talmy 1988).
The eat restaurant construction
47
Patient, as with the relation between key and door in Phil opened the door with a key. However, in our corpus, there is no Patient for the Instrument to act on. For instance, shitang does not act on one’s meal, nor would maobi (writing-brush) act on one’s calligraphy. The fact that shitang provides a meal and maobi creates calligraphy, as Diane Massam suggests (p.c.), does not amount to an affecting event defined in terms of force dynamics (see Talmy 1988; Jackendoff 1990; cf. Chapter 2). Thus the Instrument that figures in our construction differs somewhat from the sort discussed by Jackendoff. This is borne out by the tests from Jackendoff (1990), according to which the Patient should be able to occur in the frame What happened/What Y did to—was: (33) a. *What the canteen/writing-brush did to the meal/calligraphy was provide/create it. b. *What happened to the meal/calligraphy was the canteen/writing-brush provided/created it. As will be shown later, the verbs in our corpus are actually not semantically transitive11. So there is no Patient for the Instrument to act on, thus differentiating the Instrument from the more usual sort (cf. Ono 1992). This kind of Instrument that does not act on a Patient appears in English as well, as shown in the following examples from Ono (1992): (34) She was running with new shoes. (35) Bill ate the meat with a fork The Instruments new shoes and fork do not directly act on anything, even though in (35) there is the Patient meat. In the terminology of Marantz (1984), what we have here are “facilitating” Instruments, as opposed to “intermediary” ones discussed in Jackendoff (1990) (see also Ono 1992 for a full discussion of the distinction). The two kinds of Instruments can be differentiated by structural configurations of conceptual structures. There is another property generally shared by both kinds of Instrument. For instance, with chi shitang ‘eat canteen,’ it is not exactly clear, given the wording of the sentence, how a diner uses shitang for the meal event: she could go buy and eat on the premises on a particular occasion or she could have a more regular meal plan and go fetch the meal and eat in her own home. Note that where she chooses to eat is not the issue and hence shitang cannot be taken as a Location (contra Gao 1996, see section 3.7).12 The more important point here is that how the Instrument serves its role for an event or how the Actor acts on the Instrument is left implicit. Likewise, in (35), how Bill actually acts on or manipulates the fork is left unclear. We have to invoke conceptual inferences (or pragmatics) to determine the actual way the Instrument functions, as Jackendoff (1990) points out. 11. Verbs like chi ‘eat’ and xie ‘write’ can be either semantically transitive or intransitive, depending on whether a ‘cognate’ event or an event participant in their LCS is realized as their syntactic complement (see section 4.2.1). As will be shown later, the ‘cognate’ event could be simply activated, without being expressed, in which case the verb is still intransitive. 12. Shitang ‘canteen’ conceptually can be of any function, such as location of acquiring food. But this would not be relevant to the semantics of chi ‘eat.’
Enriched composition and inference in the argument structure of Chinese
48
We could thus paraphrase (7) in a deliberately underspecified way (omitting irrelevant details): (36) He has to have his meal by involving the shitang ‘canteen’ into a relevant event. For a manner-related role such as in da zhugongshou (see (14)) ‘play the centerfielder,’ a paraphrase along the same line also sounds plausible: (37) …go about the game by involving the centerfielder as a particular role in a relevant event (such as taking on that role). Implicit in the underlined portions of the paraphrases is the idea that there is a somewhat abstract superordinate event whose point is to engage in the event expressed by the verb. In other words, the sentence is actually not about the event expressed by the verb itself, but more directly about this more abstract event. The following fact supports this hypothesis. For instance, such constructions cannot be used to answer a question about a particular event (cf. (15) and (16) in section 2): (38) Did you have your meal? —*dui, wo chile shitang yes I eatLE canteen In (38), the question concerns the existence of a specific meal eyent whereas the answer is about the way one engages in the meal event, hence the problem of coherence, The same line of reasoning lies behind the problem of the following: (39) Did you have a sleep just now? —*dui, wo shuile xiao cuang yes I sleepLE small bed Given the hypothesized hierarchical relation of an engaging event embedding a more concrete event, we also expect to detect repercussions on the use of adverbs in such constructions. This is indeed the case, as shown in the following contrast: (40) a. *ta jinchang hen kuai chi canguan he often very quick eat restaurant “He ate quickly (had a quick meal) in a restaurant.” b. ta jinchang guyi chi canguan he often deliberately eat restaurant “He often deliberately engages in a meal event by going to a restaurant.” In (40), the existence of the abstract higher event in conceptual structure demands a proper modifier such as guyi ‘deliberately’ instead of hen kuai ‘very quickly’ which modifies the embedded event. Thus I have given two arguments for the existence of the higher event. An interesting question that arises is why English sentences with “facilitating” Instruments (see 34, 35) do not pass the two tests and so don’t have this higher event, consider: (41) Did Bill eat the meat?
The eat restaurant construction
49
—Yes, Bill ate the meat with a fork. (42) Bill deliberately/often ate the meat with a fork. I suggest that this has to do with differences in the event semantics expressed by the verb in these English and Chinese sentences (see section 4.2.1), but I leave the issue outside of the scope of this chapter. From the above discussion, we can deduce from the conceptual paraphrases the following points concerning the conceptual structure of the eat canteen construction: (43) i. Two events are involved in the construction, a means event (such as the one shitang is involved in) and the higher event of undertaking a certain action (such as eating a meal). The means event is subordinated to the higher event UNDERTAKE, which has an action as its argument ii. The postverbal nominal directly participates in the subordinated event and gets its semantic role there. iii. The subordinated event is underspecified. The action tier of the syntax of conceptual structures introduced in Chapter 2 provides a way to encode the underspecified subordinated event mentioned above. Following Jackendoff (1990), I suggest that the subordinated event is specified only in terms of action tier information, but lacking in thematic information. The postverbal NP takes the role of Patient in that tier. Pulling the above threads together, we get a preliminary representation of the CS of the construction, still taking (7) as an example (cf. Jackendoff 1990:253 on the LCS of the verb eat): (44) [UNDERTAKE ([HE], [CAUSE ([HE], [GO ([MEAL], [TO [IN [MOUTH-OF [HE]]]])])]) [BY [AFF ([HE], [CANTEEN])]] In (44), I have postulated UNDERTAKE as a provisional superordinate conceptual function, which could in principle be decomposed into more primitive functions. Since the examples in our corpus share the same semantic pattern, as suggested in section 3.2, we can generalize (44) as (45), replacing the constants in (44) with typed variables and adopting Jackendoff s (2000) notation. In (45), F1 represents the undecomposed variable for verbs such as chi (eat) with two argument variables, the second one being something that is eaten. This element is not linked to any syntactic counterpart (see 46), though it must be present in conceptual structure. The Means phrase is a conceptual modifier of the main event, with the ‘unselected’ complement filling the variable Thing4 within it. (45) captures the semantic pattern shared by the instances13: 13. Note that (45) is intended to be the sentential semantic pattern expressed by all the sentences in our data set. Thus it is not a lexical representation. This is where the present approach differs from (lexical) Projectionist approaches.
Enriched composition and inference in the argument structure of Chinese
50
(45) Roughly, (45) can be glossed as “one entity undertakes an event (causative or not) by means of affecting another entity,” with the event of affecting being defined in terms of force dynamics (see note 9 and Chapter 2). In our instances, (45) is expressed by the following syntactic structure: (46) [s NP2 V1 NP4]5 The numeral subscripts in (45) and (46) indicate the syntax-semantics correspondences. From these indexes, it can be seen easily that there are elements in (45) that are not expressed by (46) directly, most importantly, the relation of the modifying event to the main event as well as the function UNDERTAKE. As mentioned at the beginning of this section, a potential option would be to say that (45) and (46) comprise a construction in the sense of Goldberg 1995 (cf. Langacker 1987), that is, they are inherently paired together, with (45) being the construction meaning of (46). However, as shown in section 1, the V+complement pattern in (46) is capable of expressing various other conceptual relations in both English and Chinese. On the other hand, the means modifier in (45) is not necessarily expressed as a complement of the verb: it could correspond to another VP in a serial verb construction, as in (47), which basically retains the conceptual content in (45): (47) ta
zai yong mabi xie zi he Prog. use writing-brush write character “He is writing with a writing-brush.”
This shows that (45) and (46) are in principle independent of each other and may be paired together online. In fact, in the syntactic analyses to be reviewed later in this chapter, something like (47) is taken to be the D-structure of such constructions. I share with such proposals the idea about the conceptual equivalence of the two patterns. This means that the particular meaning expressed by the V+NP pattern in our cases is not the so-called construction/structural meaning. However, there are certain complicating cases. Not all cases can be paraphrased by a serial verb construction. For example, one does not say *yong shitang chi fan ‘use the canteen to have a meal,’ ??yong xiaochuang shuijiao ‘use a small bed to have a sleep,’ *yong zhugong shou da qiu ‘use the centerfielder to have a ballgame,’ etc. It appears then that the pairing between (45) and (46) is conventional in some cases. This raises a problem for determining the nature of such pairings, in view of their general online character and conventionality in particular cases. I return to a discussion of this issue in
The eat restaurant construction
51
section 5.1, but for now stick to the idea of defective construction in analyzing such cases. The first argument in (45), Thing2, is arguably an Actor on the action tier Within Jackendoff s (1990) linking theory, it must be linked to the Subject in the syntax (cf. also Croft 1998). Thus it is only this argument in (45) that has a fixed syntactic correspondent. Therefore (45) corresponds minimally to a SS with only its subject matched with the Actor in (45), and the syntactic expression of the modifying event being left open. Thus (45) pairs with its minimal syntax (45a): (45) a. [s NP2 […]] I take (45) and (45a) as comprising a “defective” construction in the sense of Chapter 2. By the same token, I argue that (46) by itself corresponds to an underspecified conceptual structure, thus yielding the following sort of “defective construction”: (46) a. [s NP V NP] CS:[ ] In short, I am analyzing the eat restaurant type of constructions as instantiating two construction types, namely, (45, 45a) and (46). According to the conceptual structure (45), verbs as well as the complement NPs in the examples above are both the semantic arguments of that structure. I will show that the unexpressed elements of (45) are derived from enriched composition. To do this, I must first propose an analysis of lexical conceptual elements in the construction. 4.2 Lexical Semantics Since the examples in our corpus are of the simple V+complement NP structure, I consider the lexical conceptual structures (LCS’s) of the two elements in turn. 4.2.1. The LCS of chi ‘eat’ Since I focus on the examples instantiating the verb chi, I discuss this verb’s LCS in this subsection, while making references to properties shared by the verbs in the data. A. Implicit element vs. semantic argument structure
To begin with, in the examples where chi appears (i.e., 7, 8, 11), there is always implicit in conceptual structure a non-random element MEAL (expressible as fan in Chinese). This element can be made overt in a topical sentence, such as (47a) and (48a): (47) a. shuodao chi fan ah, ta zong xihuan chi guanzi speak-Part. eat meal Part. he always like eat restaurant “Speaking of meals, he always likes to go eat in a restaurant.”
Enriched composition and inference in the argument structure of Chinese
52
b. *shuodao chi ji/mian/zhou/lingshi etc., ta zong xihuan chi guanzi “Speaking of eating chicken/noodles/conge/snacks, etc., he always likes to go eat in a restaurant.”14 (48) a. shuodao chi fan ah, ta zong xihuan chi da wan big bowl “Speaking of meals, he always likes to eat with a big bowl.” b. *shuodao chi ji/mian/zhou/lingshi etc., ta zong chi da wan. “Speaking of eating chicken/noodles/conge/snacks, etc, he always eats with a big bowl.”15 The contrast between the (a) and (b) sentences in the above shows the non-arbitrariness of the element MEAL implicit in chi guanzi/da wan. Examples featuring other verbs in the earlier examples also imply a non-random element, as illustrated in the following contrasts: (49) a. ta xi zao zongsi xi liang shui he wash bath always wash cool water “He always has his shower with cool water.” b. ?? ta xi yi zongsi xi liang shui clothes “He always washes his clothes with cool water.” (50) a. xie zi ah, ta zhengzai xie maobi write character Part. he right-now write writing-brush “As for writing characters, he is writing with a writing-brush.” b. * zhepian wenzhang ah, ta zhengzai xie maobi “As for this article, he is writing with a writing-brush.” (i) shuodao chi mian, ta zong xihuan chi Beifang MianGuan speak-Part. eat noodle he always like eat North Noodle-Restaurant “Speaking of eating noodles, he always likes to eat in Northern Noodle-Restaurant.” The contrasts above show that for examples with verbs like xi ‘wash’ and xie ‘write,’ a fixed element such as zao ‘bath’ and zi ‘character’ is implicit in such types of examples. The non-randomness of the implicit element is even clearer with the verbs shui ‘sleep’ and dal6 ‘play,’ as only yiao (sleep) and qiu (ball; ball-game) are possible as implied elements. 14. As Diane Massam suggests (p.c.), if guanzi ‘restaurant’ is specialized by modification into a particular type of restaurant, the sentence does appear to be acceptable: Interestingly, the postverbal element Beifang MianGuan in this case still appears to refer to Means rather than Location, and the modification is added in concert with the content of the implied element, mian. 15. The sentence is acceptable if, instead of being interpreted as “eat with a big bowl,” (Means reading) chi da wan were given a metonymic reading, meaning “eat a big bowl of chicken/noodles/conge/snacks, etc.” 16 There is another homophonic da which means ‘hit.’
The eat restaurant construction
53
Note that the implicit element argued for above does not belong within the semantic argument structure of any of these verbs in that it does not elaborate any prominent role in the verbal semantic structure. Take the verb chi, whose semantic argument structure consists of an eater, and a thing eaten. The implicit element, MEAL, does not belong in this argument structure, even though the thing eaten may constitute a meal, which is an event. In addition, as for syntactic expression, the internal semantic argument and the implicit element are mutually exclusive, as the following demonstrate: (51) a. * ta chile nei wan mian fan he eatLE that bowl noodle meal “He ate that bowl of noodles meal.” b. *ta chile fan nei wan mian he eatLE meal that bowl noodle “He ate meal that bowl of noodles.” This shows that the implicit element takes the same structural position as the s-selected semantic argument in syntactic structure. Note that a parallel situation holds with unselected complements, which are mutually exclusive with s-selected arguments as well. This has been noted earlier with (24), which is repeated below: (24) a. *chi mian shitang eat noodle restaurant b. *chi shitang mian eat restaurant noodle Thus the verb chi is at least found in the following argument structure alternations: (52) a. ta chile nei wan mian. he eatLE that bowl noodle “He ate that bowl of noodles.” b. ta chile fan. Meal “He has had his meal.” c. ta chi guanzi. restaurant “He eats at a restaurant.” Now we are faced with the classic issue, that of capturing the alternations without postulating multiple lexical entries, since the verbs used in (52) are obviously related. I take this as a boundary condition on the way we set up the LCS of the verb chi. The very same sort of alternations is possible with other semantically transitive verbs in the data, such as xi ‘wash,’ xie ‘write,’ and da ‘play.’ For unergative verbs, for example, shui ‘sleep,’ the situation is simpler, with only the following mutually exclusive alternations: (53) a. ta shuile yi jiao he sleepLE one sleep “He slept a sleep.” b. ta zhihao shui huochezhan
Enriched composition and inference in the argument structure of Chinese
54
he have-to sleep railway-station “He had to sleep at the railway-station.” In this case, there is no s-selected argument that competes for the postverbal complement slot. B. Implicit element and “cognate” event
Having discussed the presence of a non-random implicit element and its structural complementarity with the semantic argument structure of the verb, let us turn to the semantic properties of the implicit element in our examples, still focusing on the verb chi. First, the implicit element for sentences with the verb chi in (7, 8, 11), i.e., MEAL, is ontologically an event:17 it adds further property to the event of eating, specifying that it is a specific kind of eating, eating as in a meal event, that is at issue. In other words, the implicit element intuitively predicates some i. ta ba
zhe wan fan chile he BA this bowl rice eatLE “He ate this bowl of rice.” ii. qin fan qian xi shou please meal before wash hand “Please wash your hands before a meal.” property of the event expressed by the verb. 18 Therefore such examples cannot be used to answer questions focused on the category Thing, as the following shows: (54) ni
ba fan zhenme le? you BA meal how LE “What did you do to the rice?” —*wo ba fan chile yi-ci shitang I BA rice eatLE once canteen “I ate the rice at a restaurant once.”
17. The Chinese expression of the conceptual element MEAL, fan, is polysemous, since it can also mean a Thing: 18. Another possibility is that the implicit event MEAL does not predicate over an eating event, but is created by the eating event, under a lexical subordination analysis by Massam (1990). In other words, it is an argument serving a Patient role. However, it is hard to apply such an analysis to the data here (arguably to some of the English cognate object construction instances as well), since fan in the event reading of MEAL is conceptually dependent on the verbal event of ‘eating’ (has ‘eating’ as part of its semantic structure) whereas ‘eating’ itself does not necessarily involve the notion of a meal. On such considerations (Langacker 1987), ‘eating’ is more like an argument of MEAL. This point is given further support below.
The eat restaurant construction
55
In the answer in (54), chile yi-ci shitang, with MEAL being implied, describes a specific meal event and does not allow any additional argument (its syntactic argument places are already saturated), that is, the implied element contributes to the specification of the event chi. Turning to other verbs, the implied elements are evidently events as well, such as jao (sleep) for shui (to sleep), zao (bath) for xi (wash). What about the implied zhi (characters) for xie (write) then, as in xie maobi (write writing-brush)? Arguably it is still predicative in such constructions, as shown in the following, which parallels (54): (55) Did you write those characters? —*dui, wo xiele zhezhi maobi yes I writeLE thisCL writing-brush “Yes, I wrote those characters with this writing brush.” The problem in the answer above remains the same: the implicit element zi ‘character’ is used as a nominal predicate specifying the nature of a writing event, whereas the question is about the argument “those characters.” The argument above concerning the eventive or predicative status (see Massam 1990 for a different view, cf. note 17) of the implicit element drives us to a somewhat novel conception of the so-called “cognate object” (CO) phenomena, exemplified by examples such as sleep a peaceful sleep. The claim then is that these implicit elements can actually appear as “cognate” objects with these verbs, as in chi fan (eat meal), xi zao (wash bath), xie zi (write characters), shui jiao (sleep sleep), da qiu (play ballgames), even though curiously none of these cases are recognized in Zhou’s (1997) discussion of several classes of CO’s in Chinese, which draws on the early work of Chao (1968)19. Drawing on data from several unrelated languages, Pereltsvaig (1998) argues at length for a differentiation of two types of CO’s, argument and adverbial CO’s, in part reconciling the disputes in the earlier literature on this issue. These unrecognized CO’s in Chinese exhibit all the relevant properties for adverbial CO’s discussed by Pereltsvaig.20 For one thing, they cannot be definite (as with (b) sentences below) or occur with strong determiners (as with (c) sentences), consider: (56) a. ta zai chi fan he Prog. eat meal “He is having his meal.” b. *ta zai chi neidun fan21 thatCL “He is having that meal.” 19. Cases cited by Zhou (via Chao 1968) as “cognate objects” all appear to measure or quantify the event expressed by the verb, as in ti yi-jiao (kick a foot). These would not be regarded as “cognate objects” on the view adopted here (for a similar view, see Teng 1975). 20. A full study of “cognate” object in Chinese, or “cognate event”, as it will be called later, will not be attempted in this book. 21. Since the verb chi is optionally transitive, the sentence might be acceptable on the transitive reading, but this does not affect our general point.
Enriched composition and inference in the argument structure of Chinese
56
c. *ta chile wo meidun fan I everyCL “He had his every meal on my expenses.” (57) a. ta chile liang xiaoshi de fan he eatLE two hour DE meal “He was having his meal for two hours.” b. *ta chile liang xiaoshi de neidun fan he eatLE two hour DE thatCL meal c. *ta chile liang xiaoshi de meidun fan he eatLE twohour DE everyCL meal The pattern in (56) and (57) can easily be tested on CO’s of other verbs, such as shui (sleep), xie (write), etc. This property, as Pereltsvaig points out, shows that the CO’s above are predicative in nature. Secondly, adverbial CO’s cannot be pronominalized (cf. Massam 1990), as shown in (58): (58) * Zhangsan chile ta (=fan) eatLE it (=meal) In addition, adverbial CO’s cannot delimit the event expressed by the main predicate. The following illustrates:
22
(59) * Laowang yao Zhangsan wu fenzhong lei chi fan ask five minute within eat meal “Laowang asked Zhangsan to have his meal within five minutes.” This shows that the CO fan ‘meal’ does not make the event expressed by chi ‘eat’ a bounded event, as shown by the incompatibility with the frame adverbial ‘within five minutes.’ Thirdly, as should be expected, the CO’s in our cases cannot be passivized. Consider: (60) *fan bei ta chile meal BEI he eatLE “The meal was eaten by him.” (61) *zi bei ta xiele character BEI he writeLE “The characters have been written by him.” The sentences above can be accepted if the subject were given a Thing reading, and as such it would be definite, This of course would change the meaning of the active counterparts. Fourthly, an adverbial CO cannot be coordinated with a direct object, as in: 22. Tenny (1994, cf. Massam 1990) cites cases where a “cognate” object delimits the event, as with: Bill sighed a weary sigh. From the present perspective, this should be taken as argument CO and therefore not relevant. As will be pointed out later, “cognate” object is something of a misnomer for argument CO’s if we assume a uniform semantic characterization of the phenomenon.
The eat restaurant construction
57
(62) *ta chilefan he neige pinguo he eatLE meal and thatCL apple “He had his meal and that apple.” Again, fan here is given an Event reading, not a Thing, and thus cannot be coordinated with another Thing token. Finally, adverbial CO’s are not allowed in cleft sentences: (63) *ta chide shi fan he eatDE be meal “What he ate was a meal.” cf.(63) a. ta chide shi neige pingguo he eatDE be thatCL apple “What he ate was that apple.” (64) * ta shuide shi jiao he sleepDE be sleep “What he slept was a sleep.” Since fan is polysemous between a Thing and an Event (note 16), the sentence in (64) makes the point much more clearly, where jiao is unambiguously interpreted as an Event and disallowed by the cleft construction. In the above, I have argued at some length for the “cognate” object status of the implicit element in the eat restaurant type of constructions if it were overtly expressed. This leads us to the question as to the CS of expressions like chi fan ‘eat meal’ in Chinese. In the literature, “cognate object” phenomena are usually characterized from the morpho-syntactic perspective (see e.g. Pham 1996; Matsumoto 1996; Pereltsvaig 1998, among the considerable literature), as indicated by the term “cognate,” with the semantics left implicit or somewhat vague, in particular, how the cognate object contributes to the meaning of the construction as a whole.23 I believe a unified characterization of the phenomena lies in an explicit characterization of the semantic structures involved, while allowing for varying surface manifestations. However, there are only a few welldeveloped semantic proposals that deal with “cognate” object. In Systemic Grammar, “cognate” object has been taken as expressing a particular semantic role termed Range, which specifies “the extent of the scope or reference” of an eventuality (Halliday 1967); in a sense, the “cognate” object defines the co-ordinates of an event (Halliday 1985). As Halliday (1967:59) remarks, “cognateness is best thought of as ‘extension inherent in the process’ leading to a mutual expectancy of collocation between the noun and the verb involved.”24 In a much similar vein, in Cognitive Grammar, the “cognate” object is taken to profile the 23. Massam (1990) may be one exception, with her analysis of CO as thematic object However, she may have dealt with only one type of CO’s, given Pereltsvaig’s (1998) classification of CO’s. 24. It might be noted that the scope of Range as conceived by Halliday seems too broad, encompassing the noun phrase in examples like climb the mountain, which on finer-grained conceptual analysis should be taken as an argument of an implicit Path function (see Jackendoff 1983, 1990).
Enriched composition and inference in the argument structure of Chinese
58
“abstract region” inherent in verbal semantics (Langacker 1991; Horita 1996). From a philosophical tradition associated with Davidson (1966), Mittowich (1997) takes the “cognate” object as a realization of the event argument. For present purposes, these proposals can all be interpreted as saying basically the same thing: a certain specificational relation between two elements that together express one single event. Thus from a conceptual-semantic perspective, we generalize the relevant phenomena as follows: one event which is conceptually “defective” in some respects (an extreme case is light verb constructions, e.g. have a bath) finds its full specification in the combination with a fully specified event of compatible nature. I take this combinatorial relation to be one of conceptual modification in the sense of Chapter 2 (though it could be taken as a predicate-argument relation in other traditions), and the fully specified compatible event to be a “cognate” event. Defining the phenomena in this way, I am in effect avoiding the usual morphosyntactic bias in studies on “cognate” object. The point is that formal “cognateness” is not a necessary requirement even in English, let alone languages like Chinese. Remarks in this spirit are made independently by Halliday (1985) and Horita (1996), citing instances such as sleep a fitful slumber, play tennis, which happen to resemble the Chinese type of CO’s being discussed here. For a quick review of where we have been going, I have argued that in the unselected object constructions, there is always an element that is implicit. This implicit element is then argued to fall within the class of adverbial CO’s discussed by Pereltsvaig (1998). In addition, I have adopted a semantic view of CO’s as a “cognate” event. Before proposing a CS for relevant verbs that allow such elements, we must first look more closely at the conceptual categoriality of the implicit “cognate” event in unselected object constructions. This may help us see in what sense the event expressed by the verb may be “defective” and is fully specified by a cognate event. Take chi shitang (eat canteen) again. Granted that a non-random element such as MEAL is present in the CS of this construction, we can observe that this implicit element presupposes a default event frame that relates to MEAL by a causal chain, which might include, among other things, buying and/or cooking events. It is this causing event sequence, whatever its actual content such as buying/cooking, that brings about the event of having a meal. This is coded as the agentive quale of the event (Pustejovsky 1995; Busa 1999). In other words, the conceptual category of MEAL makes necessary reference to an agentive quale. Such categories are subsumed within the functional type of Pustejovsky’s (2001) tripartite concept lattice, as opposed to the natural type that does not refer to telic or agentive roles. The same observation holds for implicit “cognate” events of other verbs, such as CHARACTER for xie ‘write,’ BATH for xi ‘wash,’ BALLGAME for da ‘play,’ which arguably all involve events that bring them into existence; they are artifactual. What about SLEEP (expressible as jiao in Chinese) for shui ‘sleep,’ as implicit in shui xiao chuang ‘sleep small bed’? Pustejovsky extends the same natural/functional division to Qualities, with natural attributes being exemplified by HEAVY, BIG, etc. Note that a natural event cannot be modified by a functional attribute, e.g., *He touched the ceiling very well.25 For examples like shui jiao ‘sleep a sleep,’ it may be observed that evaluative modifiers may be used only when jiao is implicitly or explicitly present, as suggested by the following examples: 25. The sentence is bad when very well is understood to evaluatively modify the event of touching.
The eat restaurant construction
59
(65) a. haohaode shui (yi jiao) ba good-goodDE sleep (one sleep) Part. “Have a good sleep.” b. ta shuile yige hao jiao he sleepLE oneCL good sleep “He had a good sleep.” c. wo huiguo tou lai, faxian ta jinran (*haohao) shuizhaole I turnGUO head Part. find he unexpectedly (very-well) sleep-fall “I turned over and found him sleeping very well.” The sleeping event in the context of (c) above is presumably the most natural of its kind, not facilitated by any external force, unlike (65a) and (65b). It is infelicitous to have the evaluative modifier in such cases, in contrast to (65a) and (65b). Thus I claim jiao, even though synonymous with shui ‘sleep,’ contributes a sense of functionality to a natural sleeping event. A question that arises at this point is whether the ontological functional nature is contributed by the “cognate” event alone, or verbal semantics has its share as well. More concretely, does chi (eat) itself already designate a functional event? Put differently, in a combination of the verb with its s-selected argument, such as chile neige pingguo (ate that apple), can this be interpreted as a functional event? This possibility is ruled out, since this event cannot be modified by functional attributes either, witness *heng rongyi de chile neige pingguo (ate that apple very easily)26. Note that in this example, there is no implicit element such as MEAL. Thus I argue that the “cognate” event in our cases serves to contribute a functional character to the main event which it specifies. Pustejovksy (2001) attributes the functional nature of events to the functional nature of at least one argument. However, “that apple” in the above example is of a natural type; thus the whole event is of a natural type. Thus, apart from Pustejovksy’s observation, the ontological character of “cognate” events may be another source of event functionality. C. Representing the LCS of the verb chi ‘eat’
From the above discussion of the lexical-semantic properties of the verb chi and its like, we should have gotten a sense of the non-trivial nature of lexical semantics. The important point is to design the lexical-semantic representation in such a way as to capture significant generalizations, particularly the variable uses of verbs in our case. Thus for verbs like chi, the discussion above has called our attention to the following points in designing its LCS representation: 26. This may be subject to coercion triggered by certain contexts, where eating that particular apple is something that has to be forced. In the infelicitous reading, ‘very easily’ is taken to modify ‘eat’ alone. The infelicity may be better highlighted when embedded: * John faxian Mary zhengzai heng rongyidechi neige pingguo find right-now very easyDE eat thatCL apple “John found Mary very easily eating that apple.”
Enriched composition and inference in the argument structure of Chinese
60
(66) a. The presence of the implicit “cognate” event and its conceptual relation with the main event; b. The complementarity between the “cognate” event and the second (or internal) semantic argument in syntactic expression; c. The functional nature of the “cognate” event. To incorporate these points into the representation, the LCS would necessarily appear to be rich in structure. Drawing the various threads together from previous sections, I suggest in (67) (see overleaf) the LCS for the verb chi (eat). Let us see how this representation takes care of the factors listed in (66). First, the “cognate” event (MEAL2) is related to the main event by the relation of conceptual modification, as indicated by the dotted line. Second, numeral indexes indicate elements that are to be realized syntactically, the Jackendovian counterpart to the notion of syntactic argument structure (see note 1, Chapter 1). Thus the indexes show that both the
(73) second argument of EAT and its “cognate” event MEAL could be linked to the same syntactic argument position, but the braces (à la Jackendoff 1990) mark them as mutually exclusive. Since MEAL2 is the head, its mother should aiso be marked with braces.27 This captures the fact that if the verb is used as a transitive verb, the “cognate” event and its agentive modifier would not be relevant at all (see (52a, b; cf. 68 below). Third, the functional nature of the “cognate” event is indicated by a further modification between MEAL2 and Eventii via lamda extraction (cf. Chapter 2). The agentive quale of the meal 27. This can be attributed to certain head-driven semantic principles implemented in HPSG (Pollard and Sag 1994). I assume that marking an element in the conceptual modifier with the braces is tantamount to saying that they are not relevant under some conditions: whereas we have implicit arguments, the idea of implicit semantic modifiers does not make sense. The point is that when the “cognate” event MEAL is expressed as a syntactic argument, Thing2 is left as an implicit argument. But when Thing2 is expressed as an argument, MEAL and the whole Event are simply out of the picture.
The eat restaurant construction
61
event is implemented as the first argument of a CAUSE function, i.e., an event, which in turn is underspecified in that it is only characterized by action tier information. The second argument of the action tier function AFF¯ is an argument constant m, which should be interpreted as any entity within a default semantic frame, such as the typical things, e.g., a bowl, chopsticks or forks, that could be involved in order to engage with a meal event. It should be noted that, for the sake of simplicity, I do not decompose the structure of EAT (cf. 44) in (67). The full structure in (67) can then be glossed approximately as “an eating event between two entities, which is a meal event, which is in turn caused by the eater negatively affecting some understood entity m.” Given the representation in (67), we can account for the ambiguity of the following sentence: (68) ta
zai chi fan he Prog. eat meal/rice
i. “He is eating the rice.” ii. “He is having a meal.” As observed in note 16, the concept expressed by fan is polysemous between Thing and Event. On the Thing reading, the LCS of fan unifies with the second argument of EAT, i.e., Thing2 in (67). On the event reading, the LCS of fan unifies with the “cognate” event MEAL2, and Thing2 would be implicit in the conceptual structure. D. “Cognate “event and event quantification
The important point about (67) is that what is usually understood as (adverbial) “cognate” object is explicitly represented as a modifying “cognate” event, which is further modified by its agentive quale.28 The full significance of this representation can only be appreciated when we turn to the way it combines with nominal semantics. But before proceeding, we briefly turn to an independent phenomenon involving event quantification in Chinese and show how the notion of “cognate” event simplifies the analysis of apparent syntax-semantics mismatches. The relevant examples are given below: (69) ta chile liang xiaoshi (de) fan (cf. 57) he eat two hour DE meal “He was having/had his meal for two hours.” (70) ta kanle san tian (de) shu he readLE three day DE book “He read (books) for three days.” In each sentence above, a phrase of temporal duration occurs in a position that modifies the object noun. That such combinations form a constituent is supported by their variable distribution in different sentence patterns. Huang (1994) cites examples like the following, attributed to a personal communication with Zhu Dexi: 28. Though this does not mean all “cognate” events have an agentive quale.
Enriched composition and inference in the argument structure of Chinese
62
(71) ta lian yi tian shu dou mei kan he even one day book DOU not read “He did not even for one day read a book.” This shows that the duration+noun combination can occur in a different position in lian..dou.. constructions (see Zhang 2000 on this construction).29 In the view of Huang (1994, 1997), such examples instantiate syntax-semantics mismatches because duration phrases are event quantifiers and cannot be used to quantify objects like books or meals. Huang therefore proposes an underlying structure of gerundive nominalization, along the following lines: (72) he [vp DO [IP [+N] three days readv bookN] The operation of verb raising is then applied which raises the verb read to the empty light verb DO and assigns Case to the gerundive IP. In (72), the duration phrase is then taken to legitimately quantify the event of reading books. This all seems like the standard approach to such issues within the Principles & Parameters (P & P) theory. On this approach, apparent surface mismatches are accounted for by abstract syntactic mechanisms involving movement and null categories. However, the underlying assumption of this approach, that shu ‘book’ and fan ‘meal’ denote primitive objects or Things, is questionable, in the context of such examples, as argued earlier. Such an assumption leads to an analysis where duration phrases are treated as misplaced modifiers, on a par with examples like a sad cigarette, a grouchy cup of coffee.30 A related assumption is that the conceptual content of shu is of a simple or natural type, and so semantic composition is transparent (cf. Chapter 2). However, from the perspective being elaborated in this section, both these assumptions cannot be maintained. I suggest that the apparent verbal complements in (69–70) actually express “cognate” events. Note that neither of them can be definite, even when they appear in the lian…dou…construction:
(i) a. ta pipingle liang nian Zhangsan he criticizeLE two year Zhangsan “He criticized Zhangsan for two years.” b. *ta lian yi tian/yi ci Zhangsan dou mei piping. even one day/one time DOU not “He did not for one day/once criticize Zhangsan.” 29. Note that not all examples presented by Huang (1994) have such lian…dou… alternations. Cf. This shows that the postverbal duration phrase simply modifies the main verb, not forming a constituent with Zhangsan, contra Huang’s claim. This is supported by the fact that the modifier marker de cannot be added in (a). There are two other examples of this nature in Huang (1994) that cast doubt on his arguments. 30. Such examples were drawn to my attention by Diane Massam (p.c.).
The eat restaurant construction
63
(73) * ta kanle san tian de neiben shu he readLE three day DE that book “He read that book for three days.” (74) *ta chile liang xiaoshi de neidun fan he eatLE two hour DE thatCL meal “He had that meal for two hours.” (75) *ta lian san tian de neiben shu dou mei kan he even three day DE that book DOU not read “He did not even for one day read that book.” Furthermore, it sounds odd to add typical Thing modifiers in such examples: (76) a. * ta kanle san tian de youyisi de shu he readLE three day DE interesting DE book “He read interesting books for three days.” b. * ta chile liang xiaoshi meiwei de fan he eat two hour delicious DE meal “He had the delicious meal for two hours.” This supports the predicative, rather than the argumental, status of the verbal complements in the above.31 I therefore suggest that shu and fan actually express the “cognate” event that further specifies the main event. As discussed earlier, fan can directly express an event. As for shu ‘book,’ its content is of the functional type; its LCS must refer to the telic dimension, i.e., the reading event. It is this property that makes it possible to further specify the event of reading. Thus, given the eventive status of both fan and shu in such constructions, it is quite straightforward that they can be quantified by temporal duration phrases. Therefore, no abstract lexical items like DO or verb raising need to be invoked. On the present view, when duration+N phrases appear in lian…dou… constructions, the noun is also of the type Event (see Zhang 2000b for the variability of phrases after lian) as shown by the following paraphrase of (71), (77) ta lian kan yi tian shu dou mei kan he even read one day book DOU not read “He did not even for one day read a book.” Thus given the potential for enriched composition, there is no need for ad hoc syntactic mechanisms (see discussions in Chapter 2). There is one remaining issue that poses an apparent difficulty to my alternative analysis. I have been arguing that temporal duration phrases directly modify cognate object in examples like (69) and (70). The question is, how, on my proposal, can one make sure that the temporal duration phrase scopes over the event expressed by the verb,32 since the theory should rule out examples like (78): 31. The reader might try out the tests/properties of adverbial CO’s discussed in section B. 32. I am grateful to Diane Massam for drawing my attention to this issue.
Enriched composition and inference in the argument structure of Chinese
64
(78) *ta lianxu yi xiaoshi chile liang xiaoshi de fan he continuously one hour eat two hour DE meal “He had his two hour meal for one hour.” (contradictory on intended reading) Admittedly, the syntactic analysis takes care of the problem by invoking a deep-structure representation, where the duration phrase modifies the verbal event directly (see 72). It appears that the LCS proposed for the verb chi in (67) does not offer a ready solution for the problem of scope, since the cognate event is represented as a conceptual modifier to EAT and thus it is not obvious that a duration phrase modifying the cognate event MEAL in (67) will also modify EAT. However, this difficulty may be an artifact of analyzing the cognate event as a modifier. It is likely then that this cognate event modifier is better represented as a predicate taking the verbal event as its argument, a view not unfamiliar from the literature on CO (e.g. Moltmann 1990; see Massam 1990; Pereltsvaig 1998 for more references). In other words, the cognate event could well be taken as the head of the conceptual structure in (67). Such a scenario naturally solves the scope problem. However, I leave the LCS in (67) as it is, since nothing in what follows rests on such an issue. In short, our independently motivated semantic structures render otiose in this case the need for abstract syntactic mechanisms at the syntax-semantics interface and uphold the assumption of “concrete” syntax in Chapter 2 (cf. Culicover 1998). 4.2.2. Nominal Semantics: the Case of guanzi ‘restaurant’ I have devoted much space to the lexical-semantic properties of the verb chi (eat) and verbs with similar properties in the previous section. Now I am in a position to discuss the nominal element in the unselected object construction, focusing on the LCS representation of guanzi ‘restaurant’ while pinpointing the properties shared with other unselected complements. Basically I apply the same richer view of lexical semantics that I assumed for verbal semantics in the above. In the case of guanzi ‘restaurant’ or shitang ‘canteen,’ the important thing to note is that it is again of the functional type in the sense of Pustejovsky (2001); it makes necessary reference to its telic role, namely an event that facilitates accomplishing a meal event. But exactly how it facilitates the meal event is left implicit, as already noted in section 4.1. I formalize this underspecified content in terms of action tier information alone, a déjàvu from the discussion of verbal LCS in the last section. Thus the complete LCS for guanzi is presented in (79):
(79)
The eat restaurant construction
65
In (79), the telic quale of RESTAURANT is coded as a CAUSE relation between two events. This can be glossed roughly as “restaurant is a Thing that is affected by some Thing. This affecting event causes the event of engaging with a meal.” The important point about this representation is of course the presence of the telic information, which is again implemented as conceptual modification through lamda extraction, as in the cases of verbal semantics discussed earlier. We expect that other nominal complements in our data be characterized by similar telic information. This is indeed the case. For example, the “proper function” (Jackendoff 2000) (i.e., the telic role) of maobi (writing-brush) and heiban (blackboard) (see 10, 13) is to produce (Chinese) characters or general writings, respectively; that of xiao chuang (small bed) (see 12) is to sleep in. The concept of zhugongshou (centerfielder) (see 14) is more abstract. But in a somewhat general interpretation of the telic role (Jackendoff 2000), the proper function of a centerfielder lies in its role that helps make up a ballgame. The same point can be made for examples in (20), though we turn to some of the more complicated cases in Section 5.2.
5. Unifying the Pieces via Enriched Composition In the forgoing sections, I have proposed conceptual representations for the unselected object construction as well as the verbal and nominal constituents in such constructions on the basis of the grammatical properties of these elements. In the course of doing so, the focus has been on the specific construction chi shitang ‘eat restaurant’ while making extrapolations to instances of the same sort. Let me summarize the results as an inventory of constructions in the following, making explicit the simple assumptions concerning the syntax of the elements (As always, I omit the phonological dimension and numeral indexes are used throughout sequentially): (80) a. CS (for the unselected object construction, from 45)
b. SS [s NP2 […]]1 (81) a. CS (for the most general syntactic pattern, from 46) [] b. SS [s NP [vp V NP]] (82) a. CS (for chi (eat), from 67)
Enriched composition and inference in the argument structure of Chinese
66
b. SS33 [s NP7 [V6NP8]]5 (83) a. CS (for guanzi (restaurant), from 79)
b. SS [NP (DET) N10]9 5.1. Conceptual Unification and Coercion Given the list of constructions above, some of which are “defective” in the sense discussed in Chapter 2, I am now in a position to demonstrate how parallel unification proceeds on both syntactic and conceptual dimensions. But again I mainly focus on the conceptual dimension in the following. Since (80a) is the generalized meaning of the unselected object constructions in our data, it is the target CS I am attempting to derive or license. To achieve this goal, each of its variables must in principle be licensed by LCS’s. In (80a), the functional variable F3 is unified with EAT6 in (82a),34 the two argument variables of F3 being unified with the two arguments of EAT6. Note that their indexes are unified in this process as well, since each element is a linked constituent. 33. Intransitive verbs like shui (sleep) can be assumed to have a transitive use, taking its “cognate” object (cf. Hale and Keyser 1993). 34. It seems easy to understand that EAT6 cannot be unified with any other function in (80a), since the configuration of (80a) already imposes conceptual constraints on how unification proceeds.
The eat restaurant construction
67
Now, for the CS in (83a), there are two options as to how it integrates into (80a). One is for it to unify with the second argument variable of AFF¯, Thing4, conveying the meaning that guanzi (restaurant) is the Thing (an institution) that is affected in the force-dynamic sense, causing the meal event. The other option is for it to unify with the second argument variable of F3. But this would cause conceptual ill-formedness, since, as we have seen, F3 must be unified with EAT6 and a restaurant is not an edible Thing, in the default world we work with for now. A related issue is that there are other conceptual elements in (80a) that need to be licensed, including the functions CAUSE, AFF¯, the lamda operator needed for the modifying relation and the Event variable. As Jackendoff has observed (1991, 1997, 2000; see also Chapter 2 above), conceptual ill-formedness triggers enriched composition, which in turn supplies the needed conceptual elements noted above. Following Pustejovsky (1995, 2000b), I take it that these elements can be supplied from within the LCS of guanzi ‘restaurant’ in (83a) through processes that have been called metonymic reconstruction. More concretely, we can extract by coercion the event UNDERTAKE A MEAL by inserting a lamda operator that binds the resultant event variable. This in effect turns (83a) into (83a’) (shown overleaf), shifting the type of (83a) from Thing to Event and de-extracting the original argument variable. This then is a proper structure to be integrated into (80a). Evidently, the undecomposed event UNDERTAKE A MEAL unifies with and thus licenses both the superordinate function UNDERTAKE in (80a) and the “cognate” event MEAL8 in (82a) through a process termed co-composition (cf. Chapter 2). In addition, the modifying event in (83a’) can now unify with the modifying event in (80a) perfectly, with the two variables x and y unified together. The result of conceptual unification, involving the CS’s in (80a), (82a) and (83a) and (83a’), is presented in (84). The inner structure of the event that modifies the cognate MEAL event is simplified, only highlighting the argument constant m.
(83a’)
Enriched composition and inference in the argument structure of Chinese
68
(84) Note that a whole verbal argument structure (82a) is incorporated into (80a) as an argument, with the effect that none of its own arguments can be linked to the syntax except through argument binding as indicated by co-indexing for Thing2 Therefore, Eventi or more precisely MEAL8, which could have been expressed as a “cognate” object, will not be expressed in the syntax, leaving the syntactic position to RESTAURANT10. But the “cognate” event is constrained to be contextually accessible, as I demonstrated in section 4.2.1. Before leaving this section, a short note on syntactic unification is in order, if only to complete the picture. This is assumed to be happening in parallel with conceptual unification (Jackendoff 1997, see also Jackendoff 1999). Syntactic unification saturates syntactic variables. When (82a) is integrated into (80a), the syntactic dimensions of the two structures are unified as well, yielding the following:
(85) The underspecified SS in (80b) is thus specialized by the syntactic structure (82b), since the two do not clash. Then the syntactic dimension of guanzi in (83b) satisfies the syntactic variable NP8 in (85). The resultant structure is licensed by the independent construction (81), which has only its syntax specified. I have shown that cases of unselected complements can be accounted for entirely within the syntax of conceptual structure, thus allowing for a simple syntactic dimension, leaving us nothing more than a simple set of phrase structure rules (see section 7 for discussion of more abstract syntactic proposals). While I have glossed over many details as to how syntax works in parallel with conceptual combinations, I trust that a solid tradition of unification and construction-based syntax is already in place (see e.g. Pollard and Sag 1994; Sag 1997; Ginzburg and Sag 2001; Croft 2001). The parallel unification demonstrated here rests on the assumption of defective construction (see (80), (81)). In other words, I am not taking the meaning expressed by such constructions to be conventionally associated with the syntax of the relevant cases, as mentioned in section 4.1. However, I also suggested certain cases that are problematic
The eat restaurant construction
69
for this position. Another difficulty raised by the defective construction analysis concerns the issue of mapping: why should a complicated conceptual structure (84) map onto a simple syntactic structure (85)? A simple construction grammar answer is that (84) and (85) are just stored together as a construction, as mentioned in section 4.1. The parallel unification approach adopted here suggests that they are paired online drawing on a variety of constructions in (80–83). However, pairings like (84) and (85) could become entrenched as a fixed relationship through frequent use and hence is redundantly stored as linguistic information (i.e., a full construction), in line with a usage-based model of grammatical representation (Bybee 1985; Langacker 1987; see also Jackendoff 2000; Croft 2001). The entrenchment could be enhanced by further steps of idiomaticization (see section 6). Thus whether a particular meaning/form pairing is conventional or not is no trivial issue, and it has to be resolved through psychological processes of categorization (see Croft 2001 for discussion of this point). 5.2. Further Illustrations and the Defeasibility of Conceptual Elements Before going on further to demonstrate how the analysis proposed above extends to other cases, it is useful to take a closer look at the major points embodied in the proposal for unselected object constructions. There are three interrelated components in my analysis, which are informally summarized as semantic conditions in (86): (86) i. The event expressed by the verb should be specified by an artifactual “cognate” event (i.e., one containing an agentive quale) or the verb inherently expresses such a content. ii. The unselected object must contain telic information that brings about the “cognate” event, e.g., in (83a’), UNDERTAKE A MEAL is extracted from the qualia of guanzi, and MEAL happens to be the “cognate” event of the verb chi. iii. The unselected object that satisfies the variable Thing4 in (80a) must not be redundant with the lexically specified constant m in (82a). Condition I in (86) has the effect of producing a functional event, in the sense of Pustejovsky (2001): in other words, the unselected object construction cannot express an ontologically natural event. This can also be viewed from the perspective of the superordinate function UNDERTAKE, which only selects a functional event (one does not “undertake a coughing/laughing/giving event,” unless one intentionally feigns to do so!). Likewise, condition II above specifies that the unselected object must also denote a functional entity that is compatible with the functional event in condition I.35 Thus this condition serves to further specify the functional event. The third condition above captures the contrastive effect mentioned in Section 2. Similar non-redundancy conditions are suggested by Jackendoff (1990) in his discussion of adjuncts. This also ties in with Pustejovsky’s (2000a) work on lexical shadowing, where a certain element in verbal semantics blocks its expression as an argument, as exemplified in *Mary buttered 35. The coordination of the first two conditions appears as intuitive comments in the introduction to Meng et al’s (1999) dictionary, which points out the compatibility between Instrument object and Patient object. Obviously they lacked the appropriate theoretical tools to formalize the insight.
Enriched composition and inference in the argument structure of Chinese
70
her bread with butter. (Though the sentence is fine in the right contrastive context, as Larry Horn (p.c.) points out.) 5.2.1. Various Cases Let us take a quick look at how the conditions in (86) predict the acceptability of examples earlier in the data (see 9–14, and 20) without going into technical representations again. Those that are immediately captured by the proposal include xie maobi/heiban ‘write writing-brush/blackboard,’ chi da wan ‘eat big bowl,’ shui xiao chuang ‘sleep small bed,’ da zhugongshou ‘play centerfielder,’ chang Qingyi ‘sing the role of Qingyi,’ etc. For instance, xie ‘write’ is a functional event because the writing action itself is specified by its implicit “cognate” as discussed earlier, namely, zi ‘characters,’ which itself makes references to a process of producing them, i.e., its agentive quale. In addition, maobi/heiban ‘writing-brush/blackboard’ is designed to help with the event of producing characters/writings. And they are in a sense non-redundant with the default constant argument m in the semantics of xie ‘write,’ which could be either the generic bi (writing-tool/pen) or zi (paper). Thus xie maobi/heiban contrasts with the problematic *xie bi/zi (write writing-tool/paper), meaning “use writing tool/paper to write characters.”36 This brings us to some of the more complicated cases in the earlier data. First, there are cases where the verb does not seem to take any “cognate” event, as in (20a), (20b), and (20e), repeated below: (87) a. bangzhu dian qian help little money “help with some money” b. bao xiao bao wrap small bag “wrap (something) into a small bag” cf. * bao bao (wrap bag: wrap into a bag) e. chao kapian copy card “copy (something) onto cards” The verbs in the above do not seem to be specified by a “cognate” event, but are inherently functional: the event of helping makes implicit reference to what constitutes a helping event (i.e., it has its own agentive quale); a copying event: of course presupposes a certain preparatory event as well, one that involves Instrument; a wrapping event involves Manner, a different sort of preparatory condition for the event. These cases without “cognate” event are already accommodated in Condition I of (86). 36. Of course, if bi/zi is made definite by determiners, it can be said to be new with respect to the default element a. Thus the following is acceptable:
The eat restaurant construction
71
A different class of complications involves Condition II in (86), i.e., the LCS of the unselected object must contain the right telic information that brings about the functional event expressed by the verb together with its “cognate” object, if any. In the data cited earlier, there are a number of apparently exceptional cases to this condition, which I repeat below: (88) a. wode didi mei zhou pao-lai chi wo (cf. 17) my little-brother every week run-come eat I “My little brother comes here to eat on my expenses every week.” b. xi liang shui (cf. 9) wash cool water “to have a shower with cool water” c. xie kezhuo (cf. 21) write desk “to use the desk to write on” i. xie zhegan/zhang bi/zi write thisCL pen/paper “write with this piece of paper/this pen” The problem presented by these examples is that the unselected object in each case does not in any obvious way express the right telic information. In other words, the senses of “I,” “cool water” and “desk” intuitively do not properly contain a telic dimension that helps in the events of meal, shower and producing characters.37 Let us first consider (88a). Presumably, the conceptual content of the personal pronoun “I” must in part involve the notion of a person. Note similar examples such as chi fumu ‘eat parents.’ Without going into philosophical discussions of what a person is, I simply assume with Jackendoff (2000) that the concept of a person is the most important dot object in the conceptual system, conceptualizable as a physical object and a mind simultaneously. Each aspect has its own telic property. I suggest that it is the mind aspect that is relevant in this example, with its telic predicates including thinking and acting in particular social roles. Given such telic information, a person can certainly be “affected” in a way that helps with a meal event, similar to an Instrument (see 3.4.1). This affecting event is a more specific subtype of the telic quale of the mind aspect of the person concept. Thus the unselected object must satisfy the more general constraints imposed by the telic of the mind, including its reference to social roles. Consider, *chi zhou-guo-lai zhe wei ren (eat the one who is walking over), in which the relevant person does not bear any social relation with the subject of chi. Another issue is how the more general telic of the mind, i.e., acting in social roles, gets specialized as one related to a meal event in the meaning of the sentence, being acted upon for the purpose of a meal event. This is achieved through enriched composition with verbal semantics: the “cognate” event (MEAL) serves to co-specify the general mind telic, a natural outcome under the assumption of co-composition. Note that in this case the unselected object still satisfies 37. The telic role of desk might be to help with a particular sitting arrangement involving oneself and stationery to be used in order to write, rather than directly helping with writing.
Enriched composition and inference in the argument structure of Chinese
72
the condition of being a functional entity, granted the complex type expressed by the concept of a person. A somewhat different case is represented in (88b), where the unselected object liang shui ‘cool water’ does not express any conventional telic information at all: it does not seem to make sense to say that cool water has any normative function, though of course it could be manually put to different uses. One condition then in the conceptual category of “cool water” is arguably that it does not have a proper use. But this condition is not satisfied in (88b), where liang shui appears as if it were designed for a showering event. On the other hand, the event expressed by xi ‘wash’ and the implicit “cognate” event zhao ‘bath’ is functional: taking a bath typically entails some preparatory event, though this might itself be subject to exceptions in certain possible worlds. The functional nature of “taking a bath” entails further that certain functional entities must be involved.38 In other words, a shower must involve things designed for showering purposes. Taking the concepts of both “taking a bath” and “cool water” into account, we can see that the functionality conditions of the two concepts are in conflict: the verbal concept involves functional entities whereas “cool water” is natural. But this interaction results in the winning out of the verbal functionality: “cool water” accordingly appears as if it were a functional entity. One may suggest that the quale of “cool water” is underspecified, subject to coercion by its collocated verbal concept 5.2.2. Preference Rule System The upshot of the above analysis is that certain elements of lexical semantic representation may be assumed to have default values and thus may be defeasible under certain circumstances.39 These are called typicality conditions by Jackendoff (1983), which, together with necessary and central (i.e. graded) conditions, define the statuses of different elements in word meaning. Furthermore, the typical conditions are conceived as preference rules that interact according to principles of reinforcement and conflict.40 Thus a scale of stereotypicality can be defined in terms of how many conditions are satisfied. Given a system of preference conditions, two issues at this point need to be addressed: (89) i. The statuses of relevant conceptual elements, especially, necessary and typicality conditions. ii. The strength of the typicality conditions in interaction. 38. Pustejovsky (2001) defines a functional event as one in which at least one argument is a functional type, making reference to either agentive or telic qualia. 39. This point has been much stressed in recent work in lexical semantics (Verspoor 1997; Lascarides and Copestake 1998, and references therein). 40. The resemblance of the Preference Rule System (Jackendoff 1983, 1987) to Optimality Theory cannot be missed here (see Jackendoff 1997 for some discussion). Thus Conceptual Semantics is arguably a version of OT Semantics, a direction that is being pursued in recent formal semantics (e.g. Hendriks and de Hoop 2001), though the role of the Preference Rule System in Conceptual Semantics awaits full development (Jackendoff l996b).
The eat restaurant construction
73
Since both issues are yet to be fully explored, I can only offer some suggestions here. Returning to xi liang shui ‘wash cool water’(88b), what it has suggested to us is that the telic and agentive information is subject to exceptions. This point is also exemplified in other cases of conceptual coercion, as noted by Jackendoff (2000) and given a full analysis by Lascarides and Copestake (1998) in the context of a discussion of Pustejovsky’s Generative Lexicon. The following example illustrates: (90) Mary enjoyed the book. On Pustejovsky’s (1995) theory, and put in present terms, an event of reading is extracted from within the LCS of book and satisfies the second argument variable of enjoy, which is lexically restricted to be an event. However, the reading event would be cancelled if Mary refers to a goat, and the sentence would instead be interpreted as “Mary ate the book.”41 Similar scenarios are not unimaginable where agentive information is defeasible. Thus I assume both telic and agentive information are typicality conditions. Since I have proposed to represent these kinds of content by means of conceptual modification, the ontological category of a conceptual head and its argument structure must then fall into necessary conditions. For example, the fact that book in (90) is a Thing is indefeasible, even though its conventional use may be subject to exceptions. Furthermore, as Lascarides and Copestake (1998) have shown, lexical typicality conditions also interact with conditions of discourse coherence, which they formalize in a theory of Discourse in Commonsense Entailment (DICE)(Lascarides and Asher 1993). In present terms, this discourse theory can be recast as a set of typicality conditions as well.42 As we know, the grammatically relevant part of conceptual structures is usually represented as head-dependent structures. It seems reasonable to assume that the typicality conditions associated with a conceptual head must be ranked higher than those associated with its dependents and modifiers. Furthermore, we can also adopt a principle of Discourse Win, along the lines of Lascarides and Copestake (1998), whereby typicality conditions associated with discourse-pragmatic information override those associated with LCS’s. This short discussion leads us to a preliminary model of interactions of preference conditions, as summarized below: (91) i. Necessary conditions: ontological categories; (semantic) argument structure Typicality conditions: telic & agentive qualia; discourse coherence ii. Ranking a. Head>Dependent/Modifiers b. Discourse>LCS 41. The telic role of book can be maintained if the goat were able to read, as in a fairy tale context. 42. The DICE model is supposed to be a theory of pragmatics as well, which covers real world knowledge such as “goats don’t read”. Since I don’t assume a theoretical division of linguistic and encyclopedic information, this theory overlaps with conceptual semantics as well.
Enriched composition and inference in the argument structure of Chinese
74
This model addresses the issues raised in (89) above. It should be noted that the two ranking principles in the above coincide with triggering conditions for conceptual coercion (see e.g. Pustejovsky 1995; Jackendoff 1991; 1997). Thus the preference rule model offers a new perspective to coercion in enriched composition, which I intend to explore in further research. Given the model in (91), let us turn to the last example in (88), xie kezhuo ‘write desk,’ where the “desk” is construed as something similar to a blackboard. In other words, the conventional telic role of “desk” suggested in note 37 is overridden by the “cognate” event of “write,” which, as noted earlier, contains a default argument representing typical writing tools such as paper or blackboard. As “desk” is part of the conceptual modifier on the analysis argued for in this chapter, we can take this override as an instance of the first ranking principle in (91ii). In view of the context of the example, the principle of Discourse Win seems to be in effect as well. For another case of Discourse Win, let us look at the eat restaurant case again: (92) laowang jingchang chi canguan Laowang often eat restaurant “Laowang often goes to restaurant for his meals.” As I have been arguing throughout, canguan is taken to express a means event that facilitates the meal event. But paralleling (90) above, this reading is defeasible as well if Laowang were some sort of strange mythical monster that cannot be expected to go to a restaurant to have a regular meal, but could devour the restaurant as a whole. Thus the telic condition of canguan is overridden by the conditions associated with laowang. The important point from this discussion is that the theory that underlies the proposed analysis has the potential to accommodate exceptions, though at present a full development of this point is yet to come (see note 40). 5.2.3. Predicting Problematic Cases The discussion of the preference rule system above suggests that semantic generalizations are subject to exceptions. It follows that conceptual well-formedness is partially relative to the interactions of typicality conditions, as the analysis of xie kezhuo ‘write desk’ has shown. With this caveat in mind, the three semantic conditions summarized in (86) still account for the following problematic cases. These conditions rule out following expressions for conceptual ill-formedness: (93) a. ?? he da beizi (violating condition i) drink big cup “drink water with big cup” (bad on intended reading, ok if interpreted metonymically as drinking big cup of water/tea) b. *? chi (na jian) fang (violating condition ii) eat thatCL room “have one’s meal by entering that room” c. * meng shafa (cf. shui shafa: sleep sofa) (violating condition i) dream sofa “dream by (sleeping in a sofa)”
The eat restaurant construction
75
d. * ta xihuan xie bi (violating condition iii) he like write writing-instrument “He likes to write with BI.” e. ta xihuan shui cuang (violating condition iii) he like sleep bed “He likes to sleep in a bed.” Examples like (93d) and (93e) above are included here for completeness, as they are already discussed earlier with respect to the non-redundancy condition in (86iii) (see also note 36). In the above examples, (93a) and (93c) violate the first condition of (86). The implied “cognate” object shui ‘water’ in (93a) is ontologically natural and thus cannot contribute a functional flavor to the event of drinking. In (93c), the event of dreaming is supposed to be of a natural nature, in contrast to the event expressed by shuijiao ‘sleep a sleep.’ Thus there is no need for any means event. (93b) is similar to our xie kezhuo ‘write desk’ example in that “that room” typically does not seem to express a telic condition that facilitates a meal event, but again this might be plausible where “that room” is contextually understood as a food-serving facility. The problematic cases in (93) raise issues concerning the productivity of the unselected object construction in this chapter, and thus pose a classic problem of the learnability (Baker 1979) of such constructions, given that the child will not be expected to have recourse to the negative evidence. This kind of problem is typically associated with restrictions regarding argument structure alternations (Pinker 1989). As will be seen from section 3.7, the alternative analyses in the literature do not possess the right elements that could serve as the basis of a solution. On the other hand, my conceptualsemantic representations seem to constrain the productivity of the construction well, as illustrated in this section. I take the constraints that derive from the conceptual-semantic representations, summarized in (86) above, as well as those to be discussed in Chapter 4, as belonging with the same type of semantic constraints often discussed with various alternations (see Pinker 1989 and references therein). Thus I am essentially taking Pinker’s approach of criteria-governed productivity to the learnability paradox.
6. The Idiomaticity of chi ‘eat’ While I was searching for the relevant data of unselected object construction, a striking fact that has emerged is that the construction seems to be very productive with the verb chi ‘eat’ This may be due to the fact that a large variety of functional entities contain or can be construed as having telic information that is compatible with a meal event, including chopsticks, bowls, knife and forks, socially related persons, banquet, mountain, and various other sources (cf. Tao 2000 for various interesting examples). It might not be surprising that certain conceptual frames involving MEAL have undergone conceptual transformation. I presented some of the relevant examples in section 3.2, which are repeated below: (94) a. chi lixi eat interest “to benefit from the interest”
Enriched composition and inference in the argument structure of Chinese
76
b. chi laoben eat old-capital “to live off one’s past gains” c. kao shan chi shan, kao shui chi shui rely mountain eat mountain, rely water eat water “those living on a mountain live off the mountain, those living near the water live off the water” (making use of local resources) In these examples, the verb chi could still in principle be analyzed as taking MEAL as its “cognate” event. However, the verb in these examples is more systematically associated with a conceptual structure such as GET BENEFIT/ADVANTAGE. One cannot take this as another sense of the verb chi, since this sense only appears in restricted contexts such as (94), e.g. ta chi le ‘He has eaten’ does not mean “he has got some benefit.” So we can take this as a case of lexical metaphor,43 which in turn contributes a sense of idiomaticity to these instances. But note that these examples follow the regular syntactic and conceptual combinatorial principles discussed in this chapter: GET BENEFIT is evidently a functional event that can be aided by certain telic information deriving from the concepts of “interest,” “past gains” or “local resources.” Basically the same lexical-metaphorical process is found in (94c) as well, except that it goes further in its conventionalization in two ways: (a) the unselected complement is conceptually restricted to be MOUNTAIN and WATER, both of which denotes a kind, not an individual, cf. *chi zhezuo shan ‘eat this mountain’; (b) the phrase must be embedded in a larger parallel (paratactic) construction, where shan ‘mountain’ appears in the first conjunct, and shui ‘water’ in the second. The important point about this and other examples above is that while they follow the regular principles of combination, they are at best semi-productive (as with (94a) and (94b)) or unproductive (as with (94c)). This fact is often ignored in analyses of the chi shitang ‘eat restaurant’ type of constructions. On the present approach, we can precisely locate the points of unproductiveness.
7. Alternative Analyses To my knowledge, there are no systematic analyses of such constructions in the literature. By “systematic” I mean an adequate collection of data AND a formally explicit analysis. Tao (2000) attempts to meet this first criterion, making statistical use of corpora with a focus on the verb chi alone. But his functionalist proposal is not close to being explicit Certain pieces of insight are also scattered in a few recent descriptive studies on this topic (Wang 2000; Xin 1996), which are however impressionistic to varying extents. Formal analyses of unselected object phenomena are few. All come from a syntactic orientation and seem to stand in need of a better appreciation of the properties of the data, in addition to having certain other problems briefly mentioned in previous sections. In the following, I briefly discuss three formal analyses and try to determine where we think alike and where we differ, 43. This treatment differs from the Thematic Relations Hypothesis as well, which deals with the systematic parallels between spatial and other conceptual fields (Jackendoff 1983).
The eat restaurant construction
77
Feng (2000) proposes to derive structures like xie maobi ‘write writing-brush’ from (95) (cf. Lefebvre 1991) through verb incorporation (Baker 1988): (95) [v’ Ø [VP maobi [v’ xie e]]] In this structure, the null verb Ø expresses something like USE, and the object of the verb xie (write) is an empty pronoun referring to characters ‘zi.’ Xie moves to the null verb, deriving the surface pattern. Feng argues that this movement is prosodically motivated such that maobi, which expresses new information, appears sentence finally, gaining the stress assigned by the verb. Lin (1999) presents a paradigm of unselected objects for the verb chi ‘eat,’ including chi da wan ‘big-bowl’/guanzi ‘restaurant’/xiawu ‘afternoon’(“dine in the afternoon”)/toutong ‘headache’(“eat for [curing] headache”). She proposes the following underlying structure:
(96) In the structure above, the surface object expressing Instrument, etc., is selected by a corresponding light verb such as YONG. The main verb chi incorporates into YONG, and the chi-YONG ‘eat-use’ complex incorporates into the light verb DO, leaving the Instrument behind as a surface object. In a different vein, Gao (1996) presents an HPSG perspective on the unselected object construction. For the verb chi for instance, it expresses a relation whose second argument corresponds to a disjunction of three roles forming a hierarchy: {EATEN>INSTRUMENT>LOCATION}. Accordingly, the second syntactic argument is given a role interpretation in line with the sequence of the semantic roles in the hierarchy. This hierarchy is employed to account for the following examples where the three roles are all present: (97) a. Lisi xihuan yong kuaizi zai fanguan chi fan Lisi like use chopsticks at restaurant eat meal “Lisi likes to eat his meal with chopsticks in restaurants.” b. *Lisi xihuan yong kuaizi ba fan chi fanguan Lisi like use chopsticks BA meal eat restaurant
Enriched composition and inference in the argument structure of Chinese
78
c. *Lisi xihuan zai fanguan ba fan chi kuaizi Lisi like at restaurant BA meal eat chopsticks d. *Lisi xihuan yong kuaizi chi fanguan Lisi like use chopsticks eat restaurant e. *Lisi xihuan ba fan chi fanguan Lisi like BA meal eat restaurant f. ?Lisi xihuan zai fanguan chi kuaizi Lisi like at reataurant eat chopsticks “Lisi likes to eat with chopsticks in restaurants.” g. *Lisi xihuan ba fan chi kuaizi Lisi like BA meal eat chopsticks According to Gao, the examples above illustrates the priority of Theme (EATEN), and when it is absent, Instrument is given priority over Location. Since it is impossible to given a complete assessment of the above three analyses here, I only want to show how they relate to the analysis proposed in this chapter and the theoretical assumptions discussed in Chapter 2. In the first place, Lin (1999) and Feng (2000) form a natural group, both being couched within the recent P & P model and employing contentful but phonologically null light verbs (cf. Harley 1995), a notion of light verb that contrasts with the semantically null notion of light verbs (Larson 1988; Chomsky 1995). Given the assumptions discussed in Chapter 2, we don’t have access to such “defective” items in the present theory. Admittedly, the appeal to light verbs such as USE in these analyses is equivalent to my use of telic information within LCS’s. This shows that Lin (1999), Feng (2000) and I share a certain semantic intuition. However, I differ from them in expressing that intuition in lexical semantics itself Furthermore, the telic information appealed to in my analysis is not just “light verbs” at a different level, as might sometimes be suggested, since such information often cannot find a light verb counterpart such as YONG ‘use.’ For instance, chi shitang ‘eat canteen’ does not literally mean “use canteen to eat.” In addition, the light verbs such as ZAI ‘at,’ WEI ‘for’ postulated by Lin (1999) cannot be motivated in telic information and would not be needed on the present analysis.44 Finally, lexical telic information is independently motivated whereas null light verbs are theoryinternal. Neither Lin nor Feng are not explicit about the alternative verb frames associated with, for example, xie ‘write,’ which can take either a s-selected argument (e.g. ‘that article’), unselected argument (e.g. ‘that pen’) or a ‘cognate’ argument 45 (e.g. ‘characters’). In the approach advocated in this chapter, these differences are made in CS’s, in which sselected arguments and ‘cognate’ arguments are mutually exclusive in syntactic expression, and ‘cognate’ arguments are suppressed in conceptual unification. In 44.I have argued earlier that restaurant in eat restaurant type constructions does not express a Location role (see 18, 19, in section 3.2), thus Lin’s light verb ZAI ‘at’ is simply out of place. 45. From theory-internal perspectives, they could assume the verb is intransitive, which is derived from an underlying transitive with a ‘cognate’ object, following Hale and Keyser (1993). But Lin does not make this clear, and Feng assumes an awkward empty pronoun in order to attain the surface results.
The eat restaurant construction
79
addition, on these syntacticized analyses as presented in (95) and (96)46, the problematic cases mentioned in (93) would surely be generated, such as *shu chuang ‘sleep bed,’ *xie bi ‘write writing-instrument.’ It should also be noted that some of the examples presented in Lin (1999) are of a dubious nature. For instance, chi xiawu ‘eat afternoon’ and chi toutong ‘eat head-ache,’ both cited by Lin to support her arguments, seem at least very forced, and the latter is downright ‘gibberish’ according to my intuition. But if they are unacceptable, the analysis proposed in this chapter can account for them straightforwardly.47 Turning back to the HPSG analysis, we would expect Gao (1996) to fare better, since the framework adopted in his study shares the spirit that is being advocated in this thesis (cf. Chapter 2). However, empirically, it is not clear how Gao’s thematic hierarchy could account for the data in (93) that prove difficult for Feng and Lin as well. The problematic data in (97) that Gao cites as supporting his thematic hierarchy can in fact be accounted for on independent principles, for instance, that the ba morpheme appears only with bounded events (Yang 1995; Liu 1997), and that a clause should have only one focus projection (cf. van Valin and LaPolla 1997) (cf. 97d). Furthermore, Gao stipulates inherent semantic features such as [+location] and [+instrument] for Chinese nouns in order to interpret postverbal NPs as different semantic roles. This cannot be correct For instance, Gao claims that shitang has a feature [+location], but the feature is nowhere to be found in examples like they demolished the shitang ‘canteen.’ The deeper problem underlying this seems to be associated with the rigid division between CONTENT and CONTEXT domains within HPSG, whereas under the assumptions of enriched composition within Conceptual Semantics both domains contribute to semantic compositionality. Finally, all three approaches reviewed here do not seem to provide any basis on which to explain the contrastive effects in the construction, though only Feng (2000) takes note of the fact. As suggested earlier, this effect follows from the proposed elaborate semantic representations. Thus it appears that, to varying extents, the power of complex semantic categories (i.e., decomposition into primitives) is yet to be appreciated in these approaches. For instance, recent P & P work casts much of decomposed verbal semantics in terms of syntactic categories, while failing to note the equally complex structure of nominal semantics (Pustejovsky 1995; Busa 1999). The elaborate semantic analyses proposed in this chapter with respect to verbal and nominal domains seem to indicate that such analyses are best conducted on a separate level in a parallel architecture. 46. For more critical discussions of approaches that in a way revive the Generative Semantics tradition (stemming from Hale and Keyser 1993), see Jackendoff (1997), Kiparsky (1997), Newmeyer (1999). 47. For instance, ‘afternoon’ does not plausibly have a telic role that facilitates a meal event.
Enriched composition and inference in the argument structure of Chinese
80
8. Summary The essential elements of my analysis have been presented as three semantic conditions associated with the functional (in the sense of Pustejovsky 2001) character of the event and its second argument, as well as their compatibility. These elements and their interrelations lend themselves easily to a unification-based approach applied to semantic compositionality. In this chapter, I have demonstrated this general approach through a close examination of a particular kind of unselected object constructions, which exemplifies one among a range of semantic relations between verb and its complement. On my approach, the unselected object is semantically licensed by enriched compositional processes of coercion and co-composition. This entails elaborate lexical-semantic representations of verbs and nouns incorporating crucial telic and agentive information. In particular, the LCS of the verb contains a “cognate” event that is modified by agentive information, which is in turn compatible with the telic of the nominal semantics of the complement. It is shown how these representations are independently motivated in grammatical properties, and how they render otiose certain abstract syntactic analyses (e.g. Huang 1994) such as those relating to event quantification. The analysis underscores specific lexical-semantic constraints, which restrict the productivity of the construction. I have applied the proposal to a variety of more complicated cases, thus demonstrating that it is of the nature of semantic generalizations that they are subject to overrides and how this fact can be captured within a Preference Rule System consisting of defeasible typicality conditions. The proposed analysis provides a principled way to study the relations between the unselected object construction and certain closely related frozen expressions where bits of CSs are conventionalized. Finally I set the present analysis against a comparative discussion of three syntactic proposals (Feng 2000; Lin 1999; Gao 1996) that either attempt to syntacticize verbal semantics along the lines of Hale and Keyser (1993) or rely on a hierarchy of thematic roles. I have highlighted their empirical insufficiencies and argued that the power of complex semantic categories is yet to be fully appreciated by these approaches. The approach demonstrated in this chapter has certain obvious implications for current work that proposes to syntacticize event-semantic information in terms of syntactic categories (Tenny 1994; Borer 1994; Rosen 1996; Ritter and Rosen 1998; van Hout 1998; Travis 2000, among many others). A central line of thinking in these approaches is that basic syntactic structures such as Direct Object or alternative syntactic notions encodes aspectual structures of eventualities, but the unselected object construction tells us that Direct Object may also be deployed to encode more peripheral elements of the event. This whole body of work remains to be evaluated in terms of empirical coverage as well as representational parsimony, given a principle of semantic priority assumed in Chapter 1. In addition, as discussed in Chapter 2, the approach I advocate in this dissertation is a modified Constructional stance based on the work of Jackendoff (1997, 2000). The analysis in this chapter supports one of the central ideas of our Constructionalist theory, namely, semantic and syntactic structures are in principle separate, allowing for a variety of “defective” constructions. As shown in my analysis of
The eat restaurant construction
81
unselected object constructions, instead of being fully specified in the first place, syntactic structures can be fleshed out as a result of parallel unification on the syntax and semantics fronts, contra much standard work in Construction Grammar (Goldberg 1995; Michaelis 2000; see also Langacker 1987). The approach based on enriched composition can be applied to other kinds of semantic relations, for example, the complement expressing the purpose of the event (e.g. ji gonggong qiche “struggle to get onto the public bus”), but I leave this for future research.
CHAPTER FOUR Conceptual Inference and Predicate Transfer
As discussed in previous chapters, it is commonly assumed that syntactic arguments express the semantic arguments of the verb in a sentence. In the last chapter we have seen the range of variable relations between the verb and its direct object in Chinese. Certain of these relations pose a challenge to this assumption, as demonstrated by the case of unselected object discussed there, where the noun phrase in the object argument position does not express a direct semantic argument of the verb, but refers to an argument in a modifying conceptual clause. The proposed analysis, based on enriched composition and an elaborate syntax of conceptual structures, shows that underlying such cases are much richer conceptual structures than meet the eye. In this chapter, I focus on the relations between the syntactic subject in Chinese and semantic predication, showing that cases of semantic ‘unselectiveness’ are attested in this context as well, frequently in Chinese and in other languages. But the grammatical properties of the particular cases of ‘unselected subject’ in Chinese to be discussed in this chapter demand a treatment that is essentially different from the approach adopted in Chapter 3 for ‘unselected object’ I propose an analysis of ‘unselected subject’ in Chinese within the theory of Conceptual Semantics, taking as a point of departure the basic insights of Nunberg (1995). I discuss three apparently unrelated constructions, showing that the subject of these constructions is licensed through the same mechanism of conceptual inference rules. The related alternations of the construction are also accounted for within the same Inference Component. Again, in keeping with the methodological goal of this dissertation, I show that the proposal fares better with respect to the range of data than do relevant syntactic analyses presented in the literature. I begin in section 1 with a discussion of the canonical semantic relation between the subject and the predicate and present a range of cases where non-canonical relations are exemplified. In section 2, I discuss Nunberg’s (1995) approach to predicate transfer and reinterpret his insight in conceptual-semantic terms. I develop an approach to predicate transfer that highlights the role of the Inference Component. Following this, three case studies are presented in subsequent sections, covering the Possessum-object construction (section 3), the retained-object passive (section 4) and the locative inversion construction (section 5). I focus on the CS of the construction, the restrictions imposed on the distribution of verb types, the inferential connections to related alternations and the encoding of the external possessive relation in conceptual structure. The chapter is closed
Enriched composition and inference in the argument structure of Chinese
84
with a conclusion in section 6, where I summarize the major findings in a tabular form and try to unify them from the general Constructionist perspective developed in chapter 2. Throughout the chapter, alternative syntactic proposals and their inadequacies are discussed as well.
1. Unselected Subject in Chinese In the conceptual situation (including event and state) expressed by a simple clause, the semantic participants do not involve themselves in the situation on an equal footing. The asymmetries among the participants are often formalized in terms of prominence hierarchies of semantic roles (Foley and Van Valin 1984; Bresnan and Kanerva 1989; Jackendoff 1990, among others). In Cognitive Grammar, the notion of prominence is captured by the idea of an asymmetric action chain plus the concept of profiling with respect to portions of the chain (see e.g. Talmy 1988; Croft 1990, 1998; Langacker 1990). On any of these prominence metrics, the most prominent participant (or semantic role), which in Role and Reference Grammar is generalized as Actor (as opposed to Undergoer) and in Cognitive Grammar generalized to a greater extreme as the Figure (as opposed to the Ground) (see Culicover 1999 for a similar suggestion; cf. also Chapter 2 above), is expressed by the syntactic subject. The simple assumption underlying these hierarchical mapping proposals (see also (5) and (6) in Chapter 3) is that the syntactic subject must be semantically licensed by taking up the most prominent semantic role available in a given conceptual situation expressed by the verb. In other words, the subject argument is semantically selected by verbal semantics,l just like the object argument. The following examples from Chinese (Xin 1996) should give a feel of standard cases: (1) a. ta jintian xiawu xiehaole yipian guanggao he today afternoon write-good-LE one-CL advertisement “He finished writing an ad this afternoon.” b. yifu mo, zhao jiu fenghaole clothes part. early part. sew-good-LE “The clothes were made a long time ago.” c. nage kuangzi zhineng zhuang baicai that basket only hold cabbage “That basket can only be used to store cabbage.” d. youyong bu rongyi swim not easy “To swim is not easy.” 1. The case of the expletive it subject may pose a difficulty for this claim, but the relevant phenomena can be approached from a constructional perspective according to which the CS of a clause is fused with the null CS of it in a correspondence rule (see Jackendoff 1990 for suggestions). However, the notion of subject and related ideas involve complex theoretical and typological issues (for recent discussions, see Van Valin and LaPolla 1997; Croft 2001) beyond the scope of this dissertation. I return below to my working assumption of syntactic subject in Chinese as used in this chapter.
Conceptual inference and predicate transfer
85
In these examples, the most prominent argument available always relates to the situation expressed by the sentence by taking up one of the roles such as Agent, Patient, Instrument, etc., according to their place in a prominence hierarchy. In this sense, the subjects in these sentences are all s-selected by the verb. However, apart from the canonical cases that abide by the hierarchical mapping principle, there are several classes of examples that raise interesting questions as to the semantic relations between the subject and the conceptual situation expressed by the verb. The following three groups of examples (from Huang, C-R. et al. 1998; Huang 1987; Chappell 1999; Xu 1997) illustrate: (2) a. fuqin baile jigeng toufa father white severalCL hair Lit “(My) father had several hairs gone gray.” b. Saiweng de erzi duanle tui Mod. son breakLE leg “Saiweng’s son broke his leg.” c. wo qi-hongle erduo I angry-redLE ear “My ear went red with anger.” d. lao xiansheng zoushile yipi ma old gentleman wander-awayLE oneCL horse “The elderly gentleman lost a horse.” e. Wangmian qishui sile fuqin seven-year dieLE father “Wangmian’s father died when Wangmian was seven.” f. najia gongchang tale yidu qiang thatCL factory collapseLE oneCL wall “That factory had a wall collapsed.” g. najia gongsi chenguo yitiao cuan thatCL company sinkGUO oneCL ship “That company had a ship sunk.” h. zhejian guazi diaole liangge kouzi thisCL blouse loseLE twoCL button “This blouse has two buttons lost.” i. wo xin han 1sg. heart cold “I felt discouraged.” j. neige nuhaizi gezi xiande bijiao da that girl stature appear comparatively big “That girl, she appeared to be quite big in build.” k. ta pengyou duo, renyuan hao he friend many popularity good “He had lots of friends, and was very popular.” (3) a. zhuguan de deguoren…rang laoxiang yong buqiang da-diaole in-charge Mod. German Pass. folk use rifle hit-awayLE yitiao tui
Enriched composition and inference in the argument structure of Chinese
86
one leg “The German in charge…had his leg broken by the folks with a rifle.” b. lisi bei da-duanle yitiao gebo Pass. hit-brokenLE one arm “Lisi’s arm was broken.” c. pinguo bei xuele pi apple Pass. peelLE peel (n.) “The apple has been peeled.” d. ta zongyu bei mianqule zuihou yige zhiwu he finally Pass. removeLE last one post “He was finally removed from the last post.” (4) a. Zhangsan de tou-shang zhuangle yige bao Mod. head-top bumpLE one swelling “Zhangsan’s head got a swelling from bumping.” b. Chuang-shang tang-zhe yige bingren bed-top lie-Dur one patient “In the bed lies a patient” c. shu-dixia zhan-zhe liangge xiaohai tree-bottom stand-Dur two child “Under the tree are standing two children.” d. fangjian-li fang-le xuduo xingli room-inside put-Perf many luggage “Inside the room are placed many pieces of luggage.” e. qiang-shang gua-zhe yiding maozi wall-top hang-Dur one hat “On the wall is hung a hat.” f. Shulinli paochulai liangge shibin woods-inside run-out-come two soldier “Out of the woods ran two soldiers.” The examples above are cases where there is a non-canonical relationship between the syntactic subject and the situation expressed by the verb in the sentence. But let me first make precise my assumption of syntactic subject as applying to Chinese, before going into a detailed description of the non-canonical relationship exhibited in the examples above. Traditionally, various criteria have been used as tests for subjecthood, such as case, agreement, position, binding, “null NP” in conjunction reduction coordination, etc. Croft (2001) presents a detailed critique of these tests from typological perspectives and argues that syntactic roles are construction-specific. However, for the purposes of this chapter, I only wish to note certain formal properties shared by the initial phrases in the examples (2) to (4) above. First, they are all obviously preverbal. Second, these phrases are subject to deletion in conjunction reduction coordination, as illustrated below: (5) a. Wangmiani qishui sile fuqin, binqie Øi jingchang shengbin seven-yr dieLE father and often fall-ill “Wangmian had his father die when he was seven and (Wangmian) fell ill very often.”
Conceptual inference and predicate transfer
87
b. neike shui huar haokan, erqie Øi bu gui that tree flower good-look and not expensive “That tree has beautiful flowers, and is not expensive.” c. Lisii bei da-duanle yitiao gebo, dan Øi hai lai ganhuo Pass. hit-brokenLE one arm but still come work “Lisi was wounded in the arm, but still came to work.” d. Zhangsan de tou-shang, zhuangle yige bao, Mod. head-top bumpLE one swelling erqie Øi meiyou toufa and not-have hair “Zhangsan’s head got a swelling from bumping, and was hairless.” The examples above represent the different kinds of constructions in (2–4) and the initial phrases are all coindexed with a null phrase in a coordinated clause. Thirdly, these phrases are not optional, as shown below: (6) a. *(Wangmian) qishui sile fuqin seven-year dieLE father b. *(neike shu) huar haokan that tree flower look-good c. *(Lisi) ben da-shangle yitiao gebo Pass. hit-woundedLE one arm d. *(Zhangsan de tou-shang) zhuangle yige bao Mod. head-top bumpLE one swelling This property further supports the argument status of the initial phrase. I assume that the properties mentioned above determine the subject status of the sentence-initial phrases in these examples, without going into the theoretical complications of the syntactic concept subject itself (Van Valin and LaPolla 1997; Croft 2001). However, not all researchers in Chinese linguistics may agree with my decision. For instance, some may take huar ‘flower’ in (6b) as the subject. However, the conjunction reduction coordination test does not seem to yield acceptable results. Consider the following: (7) a. ?? neike shu kuari haokan, erqie Øi heng xiang that tree flower good-look and very fragrant “That tree has beautiful flowers, and (its flowers) are very fragrant.” b. * ta pengyoui duo, erqie Øi heng youhao he friend many and very friendly “He had lots of friends, and (his friends) are very friendly.” The same test also shows that the postverbal NP in (4) is not the subject (contra Tsao 1990). For instance: (8) * chuang-shang tang-zhe yige bingreni, erqie Øi heng pijuan bed-top lie-Dur one patient and very tired “In the bed lies a patient and (he) is very tired.” In short, I take the initial phrase in these sentences to be the syntactic subject, and in the rest of this chapter I will show that such a syntactic structure is entirely compatible with a
Enriched composition and inference in the argument structure of Chinese
88
finer-grained semantic view of such examples, by which I mean a view of the meaning of the sentences that in some sense reaches beyond verbal semantics itself. This brings me back to the issue mentioned earlier of the non-canonical relationship between the syntactic subject and verbal semantics. Let me describe in greater detail this non-canonical property with respect to the examples (2)-(4). First, the sentences in (2) exemplify what is commonly called External Possessor Constructions (EPC) (see Payne and Barshi 1999) or possessor ascension, where the relation of possession, broadly construed as embracing partitive relations, is instantiated in discontinuous syntactic positions. In (2a) through (2h), this possessive relation involves two arguments in subject and object positions respectively. For instance, in (2a), fuqin ‘father’ is the subject, but also stands in an inalienable possessive relation to the object noun phrase jigeng toufa ‘a strand of hair.’ In this sentence, the conceptual situation described by the predicator is ‘something went gray,’ an inchoative state of having the property of being gray. In precise terms, the entity that satisfies the only argument variable in this situation is jigeng toufa ‘a strand of hair,’ and ‘the father,’ while functioning as the syntactic subject of the sentence, does not strictly participate in the situation of ‘something went gray.’ Likewise, for (2b), the argument that saturates the single variable in the situation of ‘something being broken’ is ‘the leg,’ and ‘Saiweng’s son’ does not strictly speaking participate in the relevant situation (in other words, Saiweng’s son is not broken, but the son’s leg is), even though the two are again involved in a possessive relation. Examples (2i-2k) in this group make up a subtype, in which the Possessum appears in a sort of secondary subject position and the Possessor in the primary subject position, as with ‘that girl’ and ‘stature’ in (2j). Sometimes called the double subject construction (see Chappell 1999; Langacker 1993, 2001), this pattern displays the same argument structure property as in other examples, i.e., the primary subject does not directly relate to the situation described by the predictor, though its Possessum does. It should also be noted that the possessive relation construed in the sentences may be either inalienable, as is often the case, or alienable, as in (2d, 2g, 2k). A similar situation concerning semantic argument structure appears in (3), which consists of an interesting group of passive sentences. They are interesting in that they deviate from the typical syntactic pattern of passives found cross-linguistically. In other words, instead of moving the direct object of the verb to the subject position to form the passive, as is commonly the case, the direct object of the verb in the passive sentence (marked by bei or a bei phrase preceding the verb) stays in its canonical position, while the subject of the sentence is construed as the Possessor of the entity expressed by the direct object. For instance, in (3b), the subject is Lisi, which relates to the direct object ‘one arm’ in terms of inalienable possession. The situation ‘hit-broken’ is marked as passive, Given the definition of the EPC construction in the last paragraph, the examples in (3) represent another class of EPC constructions, though apparently they exhibit a very different syntax under the standard understanding of the syntax of passives (see 4.3 and Huang (to appear) for the peculiarities of Chinese passives). Let’s call them passives with retained object, following Xu (1997). Again in these cases, the subject does not directly participate in the passivized situation expressed: what is ‘hit-broken’ is strictly speaking Lisi’s arm, not the Possessor Lisi alone. Note again that the alienability of the possessive relation is not at issue here, since in (3d) ‘the post’ and its possessor ‘he’ are not in an intrinsic possessive relation, a property instantiated in other examples in (3) as well.
Conceptual inference and predicate transfer
89
Still another class of examples is presented in (4), in which the subject position is occupied by a locative NP (J.Zhang 1987; Huang 1987; Li 1990; Gao 1996), and an apparently inverted subject occupies the structural object position (cf. Bresnan and Kanerva 1989, 1992). While constructions of this type have been known as cases of locative inversion, they are also sometimes called presentative or existential sentences, for they appear to be concerned with the existence or (dis)appearance of the entity expressed by the postverbal NP. In (4b), for example, the sentence seems to be focusirig on the existence of a patient in the bed. On the other hand, each sentence expresses a state or a state resulting from a causing action (let’s call it a resultative state), such as the state of the patient lying in the bed in (4b) and the state of the luggage being in the room after being put there, in (4d). An interesting question of semantics arises at this point: does the construction assert the existence or disappearance of an entity or only assert the relevant state in a location or both? Given our assumption about the basic syntax of the construction, another problem that confronts us is that the locative subject argument is not licensed in any obvious way by either an existential situation or a resultative state. Nothing in these examples directly expresses an existential predicate, thus making the licensing of the subject argument a mystery with this possible reading. In addition, a (resultative) state by itself cannot license the locative subject either, since a state is itself a saturated conceptual situation with no open variables. In a word, the locative inversion construction presents the same difficulty for licensing the subject argument, though in this case possessive relations in the common sense are not involved.2 To sum up, I have presented several classes of examples where the semantic relatedness of the syntactic subject and the relevant conceptual situation expressed by the predicator is not immediately straightforward. We will come back in greater detail to the particular properties of these constructions in the subsequent sections. But now we are faced with deciding among a variety of options in the analysis: Is a syntactic movement approach capable of capturing the non-canonical relation discussed above? Can we apply the same technology of enriched composition as we did in the last chapter? What could Construction Grammar offer with respect to our issues? I will contrast the different alternatives in the coming discussion, but again the particular analysis I adopt is dictated by the general approach outlined in Chapter 2 where conceptual structures are assigned a crucial role. In addition, the proposal is driven by considerations of empirical adequacy and analytical economy in an attempt to make generalizations over apparently disparate phenomena.
2. Predicate Transfer in Conceptual Semantics As the analyses to be proposed in sections 4.3–4.5 are directly inspired by Nunberg (1995), I discuss in this section the predicate transfer analysis proposed by Nunberg and develop a conceptual semantics approach incorporating Nunberg’s insight. 2. Note that there is always a conceptual relation between possession and location: they can be taken to have a parallel conceptual structure, something like [X be at Y], under the Thematic Relations Hypothesis (Jackendoff 1983). Hence my qualification above.
Enriched composition and inference in the argument structure of Chinese
90
2.1. Nunberg (1995) The non-canonical relationship between the subject and the predicator discussed in the last section points to a kind of tension in their respective meanings. When semantic tensions (for example, semantic type mismatches) arise between arguments and the predicator in a sentence, one has to make principled judgements as to the source of the tension, whether it comes down to certain semantic processes relating to a particular argument or to the predicate. For instance, with respect to the semantic tension found in the following sentence: (9) The man with the cigar is parked out back. a commonly employed strategy to tackle this type of mismatch is to say that the man with the cigar is coerced by the predicate ‘be parked’ into a reading something like ‘the car of the man with the cigar,’ with the semantics of the NP subordinated to a semantic head ‘the car’ which is not syntactically expressed.3 Such an analysis is distinctly of a Jackendovian style, which aligns (9) to well-known cases like the following, discussed by many researchers (Nunberg 1979; Fauconnier 1985; Jackendoff 1997, 2000, among others; see also Chapter 2): (10) The ham sandwich over in the corner wants more coffee. For (10), Jackendoff offers an enriched composition analysis whereby the ham sandwich over in the corner is coerced to be an argument of a function ‘contextually associated with,’ which takes another unrealized argument ‘person.’ The subject of the sentence really designates ‘the person in the corner contextually associated with a ham sandwich.’ (see also Chapter 2) Thus both (9) and (10) could, in principle, be given a referential transfer analysis in the way sketched here. However, rejecting the early reference transfer analysis proposed in Nunberg (1979), Nunberg (1995) shifts the attention onto the semantics of the predicate and argues that in (9) the transfer occurs on the predicate, rather than on the subject argument. If we take BE PARKED OUT BACK as a property, the property must be attributed to a very specific entity, i.e., CAR, which can be expressed by the syntactic subject in a sentence. However, the entity expressed by the subject in (9) is ‘the man with the cigar,’ not ‘the car.’ According to Nunberg, there is no meaning transfer in the nominal semantics of the man with the cigar, but the predicate be parked out back contributes, under conditions to be specified shortly, a property of ‘the man with the cigar,’ the property the man possesses in virtue of the location of his car. In present terms, the predicate is extended to cover the ‘unselected’ subject argument. Thus, the predicate expression be parked out back in (9) is given the following two Property readings: (11) a. Original Property: the property of the man’s car being parked out back. b. Derived Property: the property of the man having a car being parked out back. 3. I agree with the arguments of Jackendoff (1997) and Langacker (1984) that abstract syntactic proposals involving a logically consistent deep structure are not adequate in these cases. For some of the arguments, see section 2.2.
Conceptual inference and predicate transfer
91
The italicized parts in (11) indicate the exact meanings being related. From this analysis, something interesting emerges about the semantics of the predicate phrase4 be parked out back. In (1 la), it contains an event-related variable of the type Patient, which is satisfied by an expression denoting a car, whereas in (11b) this Patient variable is already satisfied, though it is not expressed in the syntax of (9). However, the closest equivalent to (9) in Chinese would have to have this variable overtly satisfied, in a double subject construction (cf. 2i-k), as shown in (12): (12) a. *wo ting zai waimian I park Loc. outside “My car is parked outside.” b. wo che ting zai waimian I car park Loc. outside “I have a car that is parked outside.” This point is not made explicit by Nunberg (1995), but will be important in conceptual structure representation (see 4.2.2). Nunberg (1995) is anticipated by an analysis in the same spirit proposed by Langacker (1984), who highlights the widely attested discrepancy between the strictly s-selected argument (active zone, in Langacker’s terminology) and the actual argument that appears in a sentence. 5 In ways similar to Nunberg (1995), Langacker posits verbal polysemy based on semantic extension to account for the alternate patterns of verb/argument combination. Focusing on a case where the ‘unselected’ subject argument is a pronoun, such as I am parked out back and its equivalent in Italian, Nunberg motivates his predicate transfer analysis in terms of agreement in number (if the sentence involves two cars, the subject is still singular) and gender (i.e. the adjective for ‘parked’ in Italian does not need to agree in gender with the Italian word for ‘car’). Pollard and Sag (1994) also support the analysis with facts from person agreement, pointing to examples such as the following: (13) * I is parked on 12th Street. On the predicate transfer analysis, the subject of (13) remains the first person and the sentence is correctly predicted to be ungrammatical. One might object to Nunberg, Pollard and Sag’s arguments from agreement facts, claiming that reference transfer does not need to affect agreement.6 So even if I in (13) refers to a car, i.e., third person in meaning, its syntactic feature as the first person must remain the same and still needs the first person agreement on the auxiliary. This counterargument represents a conventional view of agreement as syntactic information, on a par with other syntactic 4. I stress the relevance of phrasal semantics because it appears the verb park itself cannot enjoy the same kind of predicate transfer without proper context: ??I am parked (‘My car is parked’). 5. A suggestion similar in spirit to both Nunberg and Langacker is made by Paul Bloom in a personal communication to Jackendoff, as reported in Jackendoff (1997:223). Bloom points out the relevance to examples like (5) of the conceptual and proprioceptive sense of having a vehicle or tool act as an extension of one’s body. However, Jackendoff does not pursue this possibility. 6. I thank Diane Massam for raising this issue with me.
Enriched composition and inference in the argument structure of Chinese
92
categories such as part of speech and case. However, there is strong evidence against this conventional wisdom. Pollard and Sag (1994) cite a variety of cases, including singular plurals and collectives, among others, where the syntactic number specification of a subject NP is overridden by some kind of semantic construal. Most relevant to the issue here, they argue that “agreement is in many cases guided by this transferred reference, rather than any inherent properties of the phrase itself.” (Pollard and Sag 1994:69) This is shown in the following contrast they give: (14) a. The hash browns at table nine are/*is getting cold. b. The hash browns at table nine is/*are getting angry. The subject of the sentence in the above is inherently plural, as shown in (14a), but this is overridden by the nonaggregate reading of the subject in (14b). Following this logic, we will have to conclude that there is no reference transfer associated with the subject I in (13). On the other hand, Croft (2001) points out the difficulties with the traditional view of agreement as expressing syntactic relations. The alternative theory Pollard and Sag (1994) propose is to put agreement information in the internal structure of referential indices within the Content attribute of an expression. In other words, agreement serves to keep track of discourse entities by indexing the referent through encoding contextually relevant properties of the referent Essentials of such a conception have been adopted by Barlow (1988) and Croft (2001) as well (see Pollard and Sag 1994 for details and references to similar proposals). In the case of (13) again, the auxiliary is specifies that its subject referent have a third-singular index, which is incompatible with the first-singular index associated with I. On this view, the reference transfer on I does not happen, or the third-singular index would be available. Another argument mentioned by Pollard and Sag (1994) as well as Nunberg is that contrasts in coordination also favor the predicate transfer analysis: (15) a. I am parked out back and have been waiting for 15 minutes. b. *I am parked out back and may not start. In (15a), the predicates can be conjoined to describe the speaker, i.e., the car owner ‘I,’ but it is not always possible to conjoin another predicate that describes the car, as shown in (15b).7 In short, such arguments point to strong evidence supporting semantic transfers on predicates. (i) I am parked out back and I might not start up again. Given the existence of reference transfer discussed earlier, one may ask if predicate and reference transfers relate to each other in any interesting way. This does seem to be the case, according to Nunberg (1995). Certain well-known cases that apparently involve reference transfer can be reanalyzed from a predicate transfer perspective, assuming that there is property predication within nominal semantics, an assumption not unfamiliar in formal semantics. For example, Take (16): (16) The ham sandwich is at table 7. 7. Diane Massam points out (p.c.) that the following may still be acceptable:
Conceptual inference and predicate transfer
93
The relevant predicate transfer within nominal semantics can be laid out roughly as follows: (17) Predication expression: ham sandwich Original Property: the property of some entity being a ham sandwich Derived Property: the property of a person having ordered a ham sandwich Thus the property of ‘being a ham sandwich’ is extended to cover the property of some person ‘having ordered a ham sandwich,’ with the semantic transfer happening within the semantics of the noun ‘ham sandwich.’8 The parallel with cases of predicate transfer is evident (see Nunberg 1995 for more detailed argument). According to Nunberg, predicate transfer is subject to conditions of correspondence and noteworthiness, which I informally present as follows, adapting Nunberg’s formulation: (18) a. Correspondence Condition The original property and the derived property associated with a predicate expression correspond. b. Noteworthiness Condition The derived property is noteworthy in either of two ways, (i) The derived property classifies its bearer in a useful way relative to the immediate conversational interests. (ii) The derived property has an abiding interest or consequence for its bearer, beyond the immediate conversational purposes. This is not a case of predicate coordination, and both predicates in the two conjuncts may be said to have transferred in meaning. In (9) above, the original property of some car being parked out back corresponds to the derived property of the man with the cigar having a car9 parked out back (see 11). This correspondence is mediated by the correspondence between cars and their owners, which are the bearers of the two related properties. In addition, (9) meets the Noteworthiness Condition in that the derived property of ‘having a car being parked out back’ usefully classifies its bearer, presumably a garage customer, so that the attendant’s responsibilities could be discharged, which is the immediate conversational purpose of the utterance. As an example of the effect of (18b ii), Nunberg notes the following contrast: (19) a. I am in the Whitney Museum. b. ??I am in the second crate on the right. 8. An analysis similar in spirit to Nunberg’s view for this type of example is proposed by Langacker (1993) within Cognitive Grammar in terms of the notion reference point. But Langacker does not seem to have unified his analyses of active zone and reference point phenomena, in the way Nunberg unifies transfers on predicate and referring positions. 9. The bearer of the original property (i.e., ‘car’ in this case) could be given either an existential or a universal reading, as Nunberg points out.
Enriched composition and inference in the argument structure of Chinese
94
Given the context of a painter talking of one of her paintings, (19a) expresses a significant property of ‘abiding interest’ the speaker acquires in virtue of her painting being on display in the museum. However, the property the speaker acquires in (19b), namely, ‘my painting being put in the second crate on the right,’ is usually not taken as a notable property. 10 Nunberg takes the Noteworthiness Condition as a special case of relevance, presumably understood in the standard Gricean sense. Thus it appears that predicate transfer is conditioned by pragmatic principles. 2.2. Jackendoff Meets Nunberg As discussed in the last section, Nunberg (1995) argues for semantic transfers on predicates whereas Jackendoff (1992, 1997) as well as Nunberg (1979) focus on the semantics of arguments, though they both pursue a purely semantic direction in tackling semantic mismatches in the constituents of a sentence. The present chapter supports Nunberg’s (1995) proposal and attempts to incorporate his insight into a conceptualsemantic framework. More specifically, in this section, I discuss the nature of Nunberg’s proposal from the standpoint of conceptual semantics and thereby develop a Jackendovian approach to predicate transfer. Once Nunberg’s arguments for predicate transfer are accepted, the most important idea in Nunberg’s proposal comes from the conditions of correspondence and noteworthiness in (18). The Correspondence Condition basically relates two predicates, for example, the property of a car and the property of a car owner, via the mediation of the two property bearers. The two properties correspond because the bearers of the properties are related in a possessive relation. Before going further, one might stop and ask if a syntactic derivational analysis could capture the idea of correspondence and thus simplify the grammar. In the case of (9), for instance, we might want to claim that it derives from an underlying structure such as the options in (20) below: (20) a. The man with the cigar [‘s car] is parked out back. b. [NP e [The man with the cigar]] is parked out back. (20a) involves deletion of a specified noun and the possessive morpheme, which is not a solution attractive to syntacticians (cf. Jackendoff 1997 for similar cases). For one thing, we don’t want the syntactic analysis to change the semantic predication structure (or thematic relations) of the sentence, but predicate transfer entails just such changes, thus precluding syntactic solutions. Of course agreement facts mentioned earlier does not favor such a solution either. Langacker (1984) cites more radical cases to make the same point, one of which is presented below: (21) [The color sensation associated with the outer surface of] the ball is yellow. 10. Nunberg acknowledges the possibility of fiddling with the context in a way that might change the conversational interest and hence the noteworthiness of the property, thus making utterances like this acceptable (see Nunberg 1995, Note 7). This possibility can be captured by the model of preference rules interaction sketched in Chapter 3, where discourse coherence is a high-ranking condition (Lascarides and Copestake 1998).
Conceptual inference and predicate transfer
95
For obvious reasons no one would take (21), which arguably also involves predicate transfer or active zone (see Langacker 1984 for a compatible analysis in a different framework), as the underlying structure for the sentence The ball is yellow, Turning now to (20b), which involves a null syntactic element, it must still involve a special rule of semantic composition to interpret the null element, as Jackendoff (1997) argues. In other words, we still have to rely on the notion of predicate transfer to interpret the null element as ‘a car.’ Furthermore, it is still unclear how the dual predication readings (relating to the two property bearers) are captured by the syntax in (20b). If we reject possible syntactic solutions as we did above and stick to the general assumptions of conceptual semantics, the Correspondence Condition comes down to a relation between the conceptual structures of two related properties. The question that arises is how do we capture this pragmatically conditioned relation between conceptual structures within the framework of Conceptual Semantics? This brings us to a less explored component within Conceptual Semantics, the Rules of Inference, which map conceptual structures back into conceptual structures. These include rules of pragmatics, logical and invited inference (Jackendoff 1976, 1983, 1990). Jackendoff argues that inference rules are stated over full CS configurations, not just a list of thematic roles. I therefore propose to capture the correspondence relation between the two properties in terms of conceptual inference, thus defining this relation over conceptual structures, rather than syntactic structures. In the case of (9), the predicate transfer involves not just ‘being parked out back,’ but a fuller conceptual structure ‘having a car being parked out back,’11 (see 11 above) which is taken as a property of the subject the man with the cigar through an inference rule. We can make explicit the inferential relation between the two conceptual structures in the following fashion, omitting definiteness information: (22) [State [CAR [Modifier AT MAN WITH CIGAR]] BE AT [Property PARKED OUT BACK]] → [State [MAN WITH CIGAR]α BE AT [Property SUCH ([State CAR [AT α] BE AT PARKED OUT B ACK])]] The CS preceding the arrow in (22) expresses the property of the man with the cigar’s car. The possessive modifier of ‘the car’ is coded as a locative element12 restrictively modifying CAR. From this it is inferred that the argument in the restrictive modifier, i.e., MAN with CIGAR, is necessarily attributed with a property of having his car parked out back, as made clear in the CS following the arrow in (22). In this derived CS, the external possessive relation between the man and his car follows automatically from the conceptual binding relation between MAN WITH CIGAR and the modifier of CAR, as indicated by the symbol a. Note that the function SUCH (Iwata 1999) in the derived CS turns a saturated state into a property that entails a new variable, the bearer of the property serving the Theme role. I take this function as something added at no cost in this inferential process, i.e., as part of the inference rule, therefore not as something that is coerced in enriched composition. Still another important property of this inferential 11. Otherwise a conceptual anomaly would result, since ‘being parked somewhere’ is not a usual property of a person, metaphorical readings aside. 12. This follows from the Thematic Relations Hypothesis according to which spatial and some other semantic fields are structured in a parallel fashion (see Jackendoff 1983), though the details do not need to concern us here.
Enriched composition and inference in the argument structure of Chinese
96
relation is that the bearer of the original property, i.e., the car, appears again in the derived CS, though it is not expressed in the syntax of (9). This is important because it makes a precise claim about the underlying CS of (9) and also captures the fact that the Chinese equivalent of (9) must have this argument expressed in the syntax, as noted earlier (see 12 above). Thus it is important to represent this argument in the CS, though whether and how it is expressed in the syntax is open to cross-linguistic variation.13 Given the role played by the corresponding bearers in the inference, we might want to ask why the two bearers of the properties, the car and the car owner, play such a role in the inference. I believe the reason lies in the fact that the possessive relation between the two property bearers permits a subset inclusion reading of the relevant properties. In other words, the property of the car being parked somewhere is properly included in the set of properties associated with the car owner. This is captured in an abstract inference rule proposed in Jackendoff (1976): (23) Inference Involving Set Inclusion [BE(x, y)]→ [BE (x, w)] Condition:w y The rule says in effect that every subset property of X is a property of X. Sentences such as If Bill is in Kenya, Bill is in Africa exemplify the rule above, where ‘being in Africa’ is included in the property of ‘being in Kenya.’ Likewise, the property of ‘being mortal’ is included in the property of ‘being human.’ This shows that the rule in (23) generalizes to the Identificational field, i.e., the field of property ascription. The rule helps us get a better grip on the property inclusion in the car parking example of (9), the car’s property as a subset property being predicated of the car owner, exactly in line with the rule in (23). Turning again to the relation in (22), the configuration of the CS preceding the arrow satisfies the set inclusion condition in (23) by virtue of the possessive relation between the car and its owner: the car’s property is a sub-property of the car owner. Therefore the inference follows in accordance with (23). I take the rule in (23) as a general principle of inference, and (22) as being a particular instantiation of it. But it will be useful to generalize (22) itself as a concrete inference rule defined over conceptual structures: (24) [State BE ([X [AT Y]], [AT [Property Z]])]w < --- > [State BE ([Y], [AT [property SUCH ([State W])]] ]. The rule above involves turning the chunk W into a new property predicated of Y. Since W is a shorthand for the whole CS preceding the arrow, it is possible to make the reverse inference from the derived property to the original one. This means that the property of Y may have an elaborate substructure itself with its own argument structure. Thus the rule in (24) is bidirectional, unlike the general rule (23). We will have a chance to come back to the particular rule in (24) in the coming discussion. 13. In the terminology used by Langacker (1984), this Possessum argument is the active zone of the event, which, according to him, might be optionally expressed through optional periphrasis. This observation does not cover the range of facts: clearly there are cases where the active zone is obligatorily expressed, as in Chinese.
Conceptual inference and predicate transfer
97
So far I have been focusing on the Correspondence Condition on predicate transfer, one of the two conditions proposed in Nunberg (1995). What of the other condition, the one involving Noteworthiness? As pointed out by Nunberg, the condition is a special case of relevance. I propose to retain its strong pragmatic flavor by leaving it as a pragmatic principle within the component of the Rules of Inference. Its role is thus to regulate the syntactic expression of conceptual structures: a CS derived in a principled way from an inference rule may not get to be expressed due to the effect of Noteworthiness (see 19b). Thus, a driver may be said to have acquired a property of having her car hit, but this property would not be relevant or noteworthy if her car was hit after she died, witness ?? Susan was hit in the fender by a truck two days after she died (cf. Jackendoff l992). To summarize this section, two semantic approaches to semantic tensions in the composition of sentence meaning are compared, one focusing on the meaning of arguments, one on the meaning of predicates, while ruling out the possible syntactic options. Nunberg (1995) presents a strong case for predicate transfer with respect to a number of cases involving the relation between subject and predicate. Predicate transfer is governed by the important conditions of Correspondence and Noteworthiness. From a conceptual-semantic perspective, the correspondence condition boils down to a principle of conceptual inference involving subset inclusion, which conditions the mapping between conceptual structures. The noteworthiness condition is considered a principle of pragmatics within the component of Inference Rules, which regulates the syntactic expression of conceptual structure. Thus, instead of positing syntactic derivationai relations, I underscored the need for systematic relations between conceptual structures and principles for their expression.
3. The Possessum-Object Construction Having developed a conceptual-semantic approach to predicate transfer in the last section, I am now in a position to examine in more detail the types of constructions involving non-canonical subject-predicate relations, as described in section 2. 3.1. The Issues Let us call the first type of construction the Possessum-object construction. This is exemplified by the group of examples in (2a-h) above, which instantiates EPC constructions, as mentioned in Section 4.1. I repeat these instances below: (25) a. fuqin baile jigeng toufa father whiteLE severalCL hair “(My) father had a strand of hair gone grey.” b. Saiweng de erzi duanle tui Mod. son breakLE leg “Saiweng’s son broke his leg.” c. wo qi-hongle erduo I angry-redLE ear “My ear went red with anger.” d. lao xiansheng zoushile yipi ma
Enriched composition and inference in the argument structure of Chinese
98
old gentleman wander-awayLE oneCL horse “The elderly gentleman lost a horse,” e. Wangmian qishui sile fuqin seven-year dieLE father “Wangmian’s father died when Wangmian was seven.” f. najia gongchang tale yidu qiang thatCL factory collapseLE oneCL wall “That factory had a wall collapsed.” g. najia gongsi chenguo yitiao cuan thatCL company sinkGUO oneCL ship “That company had a ship sunk.” h. zhejian guazi diaole liangge kouzi thisCL blouse loseLE twoCL button “This blouse has two buttons lost.” The examples above can be said to have the general meaning that a certain entity is affected by some event that affects some part of the entity. As discussed in Section 2, each of the examples above construes a possessive relation distributed over the grammatical relations of subject and object. Thus one of the issues we try to tackle in this section is where this possessive relation is coded in grammatical knowledge. In addition, each instance is systematically related to a pattern with a closely related meaning, as shown in the following (cf. 25): (25’) a’ fuqin de jigeng toufa baile father Mod. severalCL hair whiteLE “A strand of hair of my father has gone gray.” b’ Saiweng de erzi de tui duanle Mod. son Mod. leg breakLE “The leg of Saiweng’s son broke.” c’ wo de erduo qi-hongle I Mod. ear anger-redLE “My ear went red with anger.” d’ lao xiangsheng de yipi ma zoushile old gentleman Mod. oneCL horse wander-awayLE “A horse of the elderly gentleman went missing.” e’ Wangmian de fuqin sile14 Mod. father dieLE “Wangmian’s father died.” f” najia gongchang de yidu qiang tale thatCL factory Mod. oneCL wall collapseLE “A wall in that factory collapsed.”
14. Note that the idea ‘being seven years of age’ in ‘Wangmian being seven years of age (when his father died)’ cannot be expressed in this pattern. I come back to the significance of this observation later.
Conceptual inference and predicate transfer
99
g’ najia gongsi de yitiao cuan chenle15 thatCL company Mod. oneCL ship sink “A ship of that company sank.” h’ zhejian guazi de liangge kouzi diaole thisCL blouse Mod. twoCL button loseLE “Two buttons of this blouse went missing.” The sentences in (25’) contain the same verb as in the their counterparts in (25), but the possessive relation is made explicit by the modifier relation in the NP subject position in (25’). Granted the evident relatedness of the sentences in (25) to their counterparts in (25’), we need to decide what principles of grammar account for such systematic relations. Still another issue concerns the nature of the verbs that occur in such constructions. The verbs attested in (25) are intransitive, but the following instances with transitive or intransitive verbs are not acceptable (Xu 1999): (26) a. *lisi zheguai didi16 scold brother Intended meaning, “Somebody scolded Lisi’s brother.” b. *Zhangsan shuile fuqin sleep father Intended meaning, “Zhangsan’s father has gone to bed.” c. *Zhangsan shanghaile fuqin hurtLE father Intended meaning, “Somebody hurt Zhangsan’s father.” d. *Lisi kule didi cryLE brother Intended meaning, “Lisi’s brother cried.” Thus an important research issue is what conditions the verb that can occur in such constructions. In the discussion that follows, I focus on the three issues just described, namely, the grammatical encoding of possessive relations, the relations between the alternations in (25) and (25’), and conditions that predict the verb type required in the construction. The discussion of these issues will throw some light on the double-subject construction in (2i-k) as well. 3.2. The Conceptual Structures of the Construetion Following the strategy employed in Chapter 3, as a first step, we need to make precise the conceptual structures expressed by the examples in question, an important aspect of which concerns the aspectual structures expressed. Most of the sentences in (25) contain 15. The aspectual particle guo in the original is replaced by le here to avoid extra pragmatic implication. 16. In (26a) and (26c), the sentences are acceptable if the subject is construed as the agent of the action.
Enriched composition and inference in the argument structure of Chinese
100
the so-called aspectual particle le, with one containing the particle guo. Note that predicators suffixed with le in (25) alternate between an Event reading and a State reading (more on this below), regardless of the aspectuality of the entire sentence. For example, the adverbial expression ‘seven years of age’ in (25e) invites a temporally specified Event reading of change whose final state is that the father is dead. This is a typical inchoative relation (Jackendoff 1990). The Event reading can be made even clearer if an adverb like turan ‘suddenly’ is inserted preverbally, as in (27a): (27) a. Wangmian qishui turan sile fuqin seven-year suddenly dieLE father “Wangmian’s father died suddenly when Wangmian was seven years old.” This contrasts with the following where the adverb is put at the sentence-initial position: (27) b. *turan Wangmian qishui sile fuqin “Suddenly, Wang had his father die when he was seven years old.” Thus the scope of the adverb ‘suddenly’ is limited to the verb in (27a), and cannot apply to the whole sentence, as (27b) shows. This suggests that while the verb-le can be interpreted as an Event, the whole sentence may however express a State, an issue I come back to below. Note that the sentence expresses a change of state associated with the father, but due to the possessive relation between father and son, the relevant change associated with the father affects the son Wangmian and thereby assigns the resultant state of the change to the son as a property. Thus it appears that a situation like change of state can be transitive in a sense, given the appropriate connection between participants. The other reading possible with predicate expressions suffixed with le in (25) is a stative reading, which can be made apparent if a duration phrase is appended, as shown in (27c): (27) c. Wangmian sile fuqin duo nianle dieLE father many yearLE “Wangmian‘s father has been dead for many years.” Now, when the verb-le is given a stative reading, the whole sentence expresses a stative situation as well. The two types of adverbials cannot appear in the same sentence, supporting Jackendoff’s point that the Event and State readings associated with inchoatives are independent, as shown in (27d): (27) d. * Wangmian qishui turan sile fuqin duo nianle seven-year suddenly dieLE father many yearLE The same test can be applied to other instances with le in (25), showing that they all allow the two readings of the same sort for their predicate expressions. Thus, the sentence (25e) as a whole expresses State, but it may contain an element that is either State or Event. This gives two readings, which can be glossed as either “Wangmian is in the state of being characterized by the state of his father being dead” or “Wangmian is in the state of being characterized by a change of state which terminates in his father being dead.” These are the two alternative readings elaborated in more precise terms, but both readings are stative, with differences in internal complexity. We can
Conceptual inference and predicate transfer
101
correspondingly represent these readings as in (28) (see overleaf), ignoring for the moment the semantics of the particles. The State and the inchoative Event readings associated with the predicate-le expression in (25e) are represented in (28a) and (28b) respectively, with the latter being distinguished by the function SUCH taking an Eventincho as its argument, instead of a State, as in (28a). Thus the possible argument structure of the function SUCH (cf. section 2.2) needs to be expanded: it can map either a State or an inchoative change-of-state Event into a Property. This extension is motivated in the inferential relation between an inchoative Event and a State: a change of State entails a State (‘x die’ entails ‘x be dead.’).
(28) a.
b. Another property of the representations above concerns the well-known external possession relation in the sentence, for example, that between WANGMIAN and FATHER, which is captured by the conceptual binding between WANGMIAN higher in
Enriched composition and inference in the argument structure of Chinese
102
the conceptual tree and the modifier of FATHER down in the tree,17 as indicated by α. The conceptual binding relations in the above representations are important because they enable the inferences that are needed to license such conceptual structure configurations, according to (24). From either (28a) or (28b), a conceptual structure like [WANGMIAN’S FATHER BE DEAD] can be inferred. Thus Nunberg’s Correspondence Condition is implemented in concrete conceptual-semantic terms. Turning now to the example containing the particle guo, i.e., (25g), repeated below in (29): (29) najia gongsi chenguo yitiao cuan thatCL company sinkGUO oneCL ship “That company had a ship sunk.” This sentence, with the experiential particle guo presenting a perfective viewpoint (Smith 1994, 1997), expresses the State of the company being characterized by the property of the ship being sunk. As evidence, the sentence can appear with durative adverbials such as ‘two years,’ but not ‘suddenly,’ ‘quickly.’ I come back later to the issue of how perfective elements fit in with the general stative meaning expressed by the sentence. Thus the conceptual structure expressed by (29) is of the same kind as the State reading in (28a). Without going into a case-by-case demonstration, I suggest that the conceptual structures of other examples in (25) can all be analyzed along the same lines as either (28a) or (28b). For instance, for (25a), it permits the reading that ‘the father is in the state of being characterized by a change of state which terminates in a strand of his hair being gray.’ Likewise, for (25b), it expresses the State of the son being characterized by the State of his leg being broken or the property the son acquires due to the change of state with respect to his leg. Thus it will be useful to generalize (28) into a schema with variables in the following fashion: (30) where the dashed underline follows Jackendoff’s (1990) notation for the alternative inchoative reading. I take (23) as the general meaning shared by Possessum-object constructions. Adopting a Construction Grammar (CG) approach, C. -R. Huang et al. (1999) formulate the construction meaning of the Possessum-object construction, suggesting that the subject NP is the generalized possessor of the event. This formulation seems somewhat obscure. It is not clear to me in what sense Wangmian “possesses” the inchoative event that his father died. They explain that the event directly makes contact with or affects the entity expressed by the subject NP. The essence of this idea is captured by the predicational relation between the Property and the first argument variable y in (30), as well as the Noteworthiness Condition in the present framework. The predicational relation seems evident, since all the examples in (25) can be conjoined with clauses describing the subject NP, not the 17. Jackendoff (1992) defines binding over conceptual structures in terms of the notion CScommand, analogous to the syntactic C-command.
Conceptual inference and predicate transfer
103
postverbal NP, as noted earlier (see 5a above). Consider again the following: (31) a. Wangmian qishui sile fuqin, heng buxin seven-year dieLE father very unfortunate “Wangmian had his father die when he was seven, he was very unfortunate.” b. *Wangmian qishui sile fuqin, yisheng qinlao whole-life hard-working “Wangmain had his father die when he was seven, (his father) worked hard his whole life.” Huang et al (1999) suggest further that as part of the construction meaning the subject NP expresses the location of the new event, thus aligning the pattern to the existential (also called locative inversion) construction. They leave unclear the relations between the two parts of their construction meaning: in what sense can the location of the event be affected by the event, as they suggest? I argue in section 4.4 that locative inversion involves a different sort of conceptual inference. In short, the semantic theory underlying their CG proposal seems nothing more than intuitive. Another point where I differ from Huang et al is that I don’t take the entire CS of (30) to be the meaning inherent in the relevant syntactic construction, which itself is not clearly specified in Huang et al’s account, thus rendering the claims of construction meaning dubious. As in Chapters 2 and 3, I take the conceptual structures expressed in the sentence to be independent in principle from (though linked to) the syntactic structure, an essentially Jackendovian standpoint (1996, 1997, 2000). 3.3. Conceptual Structure, Inference and Verb Type Given the approach developed in section 2.2 and the conceptual structure postulated in section 3.2, we are now in a position to come back to the issues discussed in section 3.1, namely, the systematic alternations in (25) and (25’), the external possessive relation and the distribution of verb types in the construction. 3.3.1. Semantics vs. Syntax of the Alternation and External Possession The generalized schema in (30) is essentially the same as the part of the inference rule following the double-headed arrow in (24), except for the optional inchoative event. Thus by the rule (24), it is possible to infer from (30) the State embedded as the argument of SUCH. Put simply, if Wangmian is characterized by the property of his father being dead, we can infer that Wangmian’s father is dead. Furthermore, appending the function INCH as in (30) does not affect this inference: if Wangmian is characterized by a change of state which terminates in his father being dead, we can still infer that Wangmian’s father is dead. This conceptual inference relationship captures the systematic relations between the Possessum-object construction in (25) and the counterparts in (25’). I take the addition of the function SUCH, which is not expressed in the syntax, as part of the inference rule: thus we need no additional mechanisms such as coercion for its presence in conceptual structure. On this approach, the alternations in (25) and (25’) are based on different though related semantic structures, a result that is compatible with much recent
Enriched composition and inference in the argument structure of Chinese
104
work on verbal alternations (see, e.g., Pinker 1989; van Hout 1996; Dowty 2001, among others). The view that the relevant relations are syntactic in nature (see e.g., Xu 1999) is therefore rejected, and I claim that the present view is motivated in independently established principles (i.e., conceptual inferences) and therefore avoids certain ad hoc assumptions associated with syntactic proposals, One such analysis using the syntactic strategy is Xu (1999), who suggests within the GB framework that the patterns in (25) and (25’) are derived from the same underlying structure through Possessor raising. Take (25e) for example. According to Xu, its underlying structure would be something like the unaccusative structure in (32): (32) [s e [VP die [NP Wangmian de father]]] In this structure, either the Possessor NP Wangmian or its containing NP moves to the Subject position to get Nominative Case, thus yielding the alternations in (25e) and (25e’). There are at least two immediate problems with this approach (see also 4.2.2). The most fundamental problem is the hidden assumption that the two relevant alternations share the same semantic structure, for which I have argued to the contrary. The gist of the discussion in the last paragraph is that ‘Wangmian’s father died’ is an inchoative event or state embedded within the meaning of the Possessum-object construction (see 30) and thus can be inferred from the latter. On the present view, the empty Subject position is actually a θ position. When the Possessor NP Wangmian moves to it, the Possessor will be predicated by the extended property of ‘the father being dead.’ Thus the movement would violate the standard θ-Criterion (Chomsky 1980) on the present analysis, with a Possessor moving to a Property bearer position18. Secondly, Xu does not have a natural way to deal with the possessive morpheme de other than ad hoc stipulations to get the correct result19. The problem would not arise if we assume that the alternations simply have different underlying structures, an assumption that is compatible with my conceptual inference analysis. Chappell (1999) also argues for the independence of the two alternations, mainly noting their distinct semantic and discourse properties. Her most important argument concurs with mine: the two alternations basically show distinct semantic predication relations. It is clear that the focus of the syntactic approach as proposed by Xu (1999) is on the apparent possessive relation and encodes the relation explicitly in the underlying structure. It does not attempt a finer-grained semantic analysis, thus overlooking important predication structures. The present approach takes care of both aspects of semantics, highlighting the distinctness of predication structures and at the same time encoding the possessive relation between elements in different CS positions as a conceptual binding relation, as noted in 4.3.2. The idea of conceptual binding readily allows for the following cases where the relation is spelled out with pronominal anaphors in the possessive form: 18. This syntactic movement analysis can be salvaged, as Diane Massam suggests (p.c.), by allowing for some idea of thematic overlay. This derives part of the semantic effects from my approach, and indeed it would be a near notational variant of my semantic approach, though the other problems mentioned below remain. 19. Indeed Xu (1999) only leaves the issue to a footnote, suggesting that the morpheme de is either absent in the deep structure or deleted with the raising of the possessor.
Conceptual inference and predicate transfer
105
(33) a. [Wangmian]i diule [ ]i shubao loseLE school-bag “Wangmian lost his schoolbag.” b. [Wangmian]i diule [tade]i shubao loseLE his schoolbag “Wangmian lost his schoolbag.” (34) a. [Zhangsan]i zoushile [ ]i yipi ma walk-awayLE oneCL horse “Zhangsan lost a horse.” b. [Zhangsan]i zoushile [tade]i yipi ma walk-awayLE his oneCL horse “Zhangsan lost one of his horses.” In (33b), the binding relation between the subject and the Possessor of ‘schoolbag’ is made explicit by the pronoun tade. This poses a difficulty for the Possessor raising analysis, but is entirely expected on the conceptual binding analysis of the possessive relation. 3.3.2. Conditions on Verb Types On the present approach, the semantic type of the verb that occurs in the construction is restricted both in terms of the larger conceptual structure to be expressed and the pragmatic condition of Noteworthiness discussed earlier (see 18b). Given the schematic conceptual structure (30) taken as the general meaning of the Possessum-object construction, the limiting conditions for the semantics of verbs or predicate expressions should be as follows: (35) a. The verb or the predicate expression must designate a State or an inchoative Event to unify with the argument variable of the function SUCH. b. The State expressed contains a Property element. c. The first argument of the Property-based State is conceptually modified (via conceptual binding) by the highest Thing argument. Thus, the schematic conceptual structure in (30) above makes strong predictions as to the type of predicate expressions that are allowed in such constructions. But what is more, the sernantics of the predicate expression is further restricted by the Noteworthiness Condition (see 18b) in two ways, i.e., the predicate expression must characterize the highest Thing argument in a useful way, for or beyond the immediate conversational purposes. It should be noted that the condition in (35a) should be taken to suggest that the State component in the semantics of the verb must be foregrounded, rather than simply in the background frame semantics of the verb (cf. Chap. 2). The idea is captured by the term profile in Cognitive Grammar (Langacker 1987) or the more narrowly defined notion of event headedness within the framework of Generative Lexicon (Pustejovsky 1995).20 20. Nakajima (2001) formulates restrictions on verb types in locative inversion constructions in terms of event headedness and qualia structure, which we will discuss briefly in section 5.2.
Enriched composition and inference in the argument structure of Chinese
106
The simplest way to meet the conditions a ove is to have a predicate expression whose conceptual structure directly expresses a Property, thus satisfying the variable z in (30), provided that the Property bearer conceptually relates to the highest argument as stipulated in (35c). The double-subject constructions in (2i-k) are examples employing this strategy. Take (2j) for example, repeated below as (36): (36) neige nuhaizi gezi xiande bijiao da that girl stature appear comparatively big “That girl, she appeared to be quite big in build.” The predicate da above, being an adjective, directly expresses a Property, whose bearer gezi ‘stature’ is necessarily modified by the first subject nege nuhaizi ‘that girl’ through conceptual binding. (36) can be conjoined with more clauses describing aspects of ‘that girl,’ such as xinge que heng wenhe ‘but (her) personality is very good,’ and the conceptual binding is sustained in the conjunct.21 In cases like these, if the predicate expresses a Property, Conditions (35a) and (35b) are satisfied immediately. However, most of the examples being discussed in this chapter present more complicated cases of unification. One sign of the complexity relates to the use of aspectual particles in the construction. For instance, the particle le is obligatory in (25e), whose conceptual structure is discussed in detail in 4.3.2. The Possessum-object construction such as (25e) is found to alternate with the double-subject construction as well, as in (37b) below. The relevant patterns are as follows: (37) a. Wangmian sile fuqin dieLE father a’ *Wangmian si fuqin. b. Wangmian fuqin sile b’ *Wangmian fuqin si The aspectual particle le is shown to be required in the two patterns in (37). The question is how the verbal semantics coupled with the semantic contribution made by le fits into the schematic conceptual structure in (30), This leads us to the issue concerning the semantics of the particle le, a notoriously tricky problem in Chinese linguistics (for recent discussions of le, see Sybesma 1997; Wang 2000; Klein, Li and Hendriks 2000). 3.3.2.1. The Semantics of the Particle le Without going into a lengthy discussion of the huge literature concerning the Chinese particle le at this point (see the references above for reviews), I would like to propose a semantic analysis of le based on Zhang (2000a). The essential idea of Zhang (2000a) is that the particle le is a lexical construction in the CG sense. Its morphosyntactic side is characterized by its being attached to either verbs, resultative verbal compounds (e.g., dacuile ‘hit-broken’) or adjectives.22 21.I leave aside the issue that the first subject is most likely a Topic element as well. 22. The morphosyntax of le is even more complicated than this: it can also be attached to particular noun phrases, as in (i) below:
Conceptual inference and predicate transfer
107
On its semantic dimension, the particle le designates (or profiles) the transition point between dynamic segments and a resultative stative segment within the frame of a causalchain-based model of event structure (Croft 1991, 1993, 1998, and see Levin and Rappaport 1996 for discussion). I continue to advocate this conception of the semantics of the particle le because it appears to capture in a predictive way the various uses of the particle, depending on how the semantics of the particle composes or unifies with predicate expressions. For instance, if the particle combines only with a dynamic event, a perfective reading will be induced, as the following illustrates (from Wang 2000): (38) ta zai wu fenzhong nei xiele sanfeng xin he in five minute within writeLE threeCL letter “He finished writing three letters in five minutes.” The frame adverbial (cf. Dowty 1979) ‘in five minutes’ in (38) indicates that the event of letter writing has come to its natural terminal point, i.e., the event is delimited. If the particle combines with a stative event (as in 39) or a complex event containing a stative result (as in 40), we get a change of state reading, as shown in the following: (39) zhe haizi zongyu congminle this child finally smartLE “This child finally became smart.” (40) ta you dacuile yige wan he again hit-brokenLE oneCL bowl “He broke another bowl.” The change of state reading is called a new situation by Wang (2000), who also cites examples such as the following: (41) a. ta youle yun she haveLE pregnancy “She is pregnant now.” b. ta youle yi da bi qian she haveLE one big CL money “She has now got a large sum of money.” i. dou daxueshenle, yingan dongshile. already university-studentLE should understand-thingLE “(You) are already an university student. (You) should be more understanding.” In addition, I leave aside the issue of sentence-final le, though it appears that a unified account might be plausible, as has already been suggested in the literature (e.g. Shi 1990).
It appears from the above that the verb phrase headed by you ‘have’ denotes a stative situation, which is then turned into a change of state when the state is incorporated into the meaning of le. The different ways of unification of verbs and the particle illustrated above show that the transition point in a complex event is capable of functioning in two converse ways: it could either map an event into a state (as with perfective reading) or map a state into an event (as with change of state reading). This corresponds to the two converse aspectual
Enriched composition and inference in the argument structure of Chinese
108
functions postulated in Conceptual Semantics, PERF, which maps an Event argument into a State, and INCH, which maps a State argument into an Event of this state being brought about (Jackendoff 2000; see also Chapter 2 above). The examples above thus indicate that the semantics of the particle le can be broken down into the two converse aspects. Since the two conceptual functions are arguably associated with the same transition point in a geometrically represented event structure (Croft 1998), I propose a coarser-grained Transition function (TRAN) that subsumes the two converse functions and represent its argument structure as follows:
(42) In other words, the function TRAN either maps a State into an Event, as the function INCH does, or maps an Event into a State, as the function PERF does. The meaning of the aspectual particle le will henceforth be assumed to be captured by the above underspecified conceptual structure. 3.3.2.2 Conceptual Unification: V-le With the semantics of the particle le in place, let us return to the issue of how verbal semantics coupled with the contributions of le unifies with the schematic conceptual structure in (30) and thereby satisfies the conditions on verbal semantics in (35). Returning to the by-now-familiar case of (25e), repeated in (37a), we are faced with three issues: (1) the semantics of the verb si ‘die’; (2) how the verb semantics unifies with the semantics of the le construction; (3) how the semantics of the complex V-le unifies with the larger schematic structure in (30). First, the verb si ‘die’ is ambiguous between a stative and a change of state reading, as illustrated below: (43) a. ta bu xiang mashang jiu si he not want immediately then die “He does not want to die immediately” b. ta si(le) hengduo nianle he die(LE) very-many yearLE “He has been dead for very many years.” c. *ta turan si(le) hengduo nianle he suddenly die(LE) many yearLE The adverb ‘immediately’ in (43a) brings out the change of state Event reading, and the duration phrase in (43b), where an optional le is attached to the verb, shows the stative reading of the verb si. An analogous but clearer case of a stative reading combining with le is given below: (44) zhe fangjian zhi ganjinle liangtian this room only cleanLE two-day “This room became clean just for two days.”
Conceptual inference and predicate transfer
109
The predicate expressions in both (43b) and (44) designate a state that lasts for a specified period of time, as indicated by the durative modifier ‘very many years’ or ‘two days,’ though in both cases the particle le contributes an inchoative twist to the state. The independence of the two readings is shown in (43c), where the two relevant phrases cannot co-occur. The two senses of the verb si are represented as in (45):
(45) (45a) shows the lexical conceptual structure of si as a change of state verb, with the function INCH being incorporated in the lexical semantics, whereas the stative reading shown in (45b) does not contain such a function. Now, if (45a) unifies with the Event argument variable in (42), the result is a State, as shown in (46a) (overleaf). Put differently, the particle le in this case functions in effect as a PERF function. This structure accounts for (43a) above, where the adverb ‘immediately’ modifies the Event headed by INCH in (46a). It is interesting to note that, in this case, the State that results from the function TRAN is lexically entailed by the LCS of (45a), i.e., being recursively the same as [State BE (x, [AT DEAD])], though this is not always the case in Chinese, Thai and a number of other languages (Depraetere 1995; Croft 1998; Koenig and Musansuwan 2000)23. Thus it appears that the addition of the function TRANS serves to profile or pick out the State within (45a).
(46) a. The unification of (45b) with the State argument variable in (42) yields an inchoative Event, as shown in the following:
(46) b.
23. The central problem here for studies of aspectuality of events is formulated by Koenig and Muansuwan (2000) as ending an event without ever finishing it.
Enriched composition and inference in the argument structure of Chinese
110
In this case, the particle le serves in effect as the function INCH. The State above can be further modified by durative adverbials, as in (43b). The final step is to integrate either (46a) or (46b) into the schematic structure in (30) above, both unifying with the single argument variable of the function SUCH. This is allowed since neither (46a) nor (46b) contains any information that clashes with any element in (30). So (46a) directly unifies with the State argument of (30), yielding (47a): (47) a. [State y BE AT [Property SUCH ([State TRAN([Event INCH([State x of y BE AT [Property DEAD]])])])]] This can be taken as the more detailed version of the reading in (28a), where the combination of the verb (itself inchoative) plus the particle le (which is PERF accordingly) yields a stative event which in turn functions as the argument of SUCH, mapping into a Property. Alternatively, (46b) satisfies the variable Eventincho in (30) through unification, yielding a structure like the following: (47b) [State y BE AT [Property SUCH ([inchoEvent TRAN ([State x of y BE AT [Property DEAD]])])]] Thus we license the reading in (28b) from integrating the LCS’s of the verb and the particle. What all this means for the distribution of verb types in the construction is that either the stative and the change of state sense of the verb chi ‘eat’ fits in with the schematic structure in (30), though, as we have seen, the LCS of the particle le plays an important part in adjusting the verbal conceptual structure. Cases in (25) that involve unifying an unambiguous inchoative Event into (30) include (25a) and (25c), and others are ambiguous between the inchoative and the change of state readings. Thus the predicate expressions in these examples are licensed in the same way. 3.3.2.3 Conceptual Unification: V-guo/zhe Turning now to cases with particles other than le. Since the particle guo can be treated as a PERF function as noted earlier, we can treat the distribution of verbs in sentences like (29) in a similar fashion. For a different sort of case, Huang et al (1999) cite a class of Possessum-object examples that contain verbs attached with still another particle zhe, as illustrated in the following: (48) a. ta meizhe liandan she beautifulZHE face “She’s got a face that’s beautiful.” b. ta hongzhe lian she redZHE face “She blushed.” An issue that arises at this point is whether the particle zhe functions to adjust the verbal conceptual structure, in the way le does, for unification with the general schema in (30). This does not seem likely, since the semantics of the predicates in (48) are all statives containing a Property element, already satisfying the requirements on predicate types.
Conceptual inference and predicate transfer
111
Thus we can use the double-subject construction, which does not need zhe24, to express the relevant conceptual structure, as in (48c): i. ta
liandan meizhe ne she face prettyZHE Part. “She has a pretty face.”
This again shows that the particle does not play a role in licensing the particular predicate expression in such constructions.
(48) c. ta liandan hen mei25 she face very pretty “She as a very pretty face.” Then, what is the function of zhe in this construction? Following Smith (1994, 1997) and Pan (1996), I take the particle zhe, called stative imperfective by Smith, as marking the predicate as stage-level (Carlson 1977). Thus it contributes an independent perspective to the stative situation expressed by the predicate irrespective of the general conceptual schema in (30). This contribution adds to and does not contradict the schema (30). 3.3.2.4. Further Predictions We are now in a position to see how the theory I propose rules out examples of the construction where inappropriate types of predicate expressions are used. A representative set of examples was given earlier as (26), which is repeated below: (49) a. *Lisi zheguai didi scold brother Intended meaning, “Somebody scolded Lisi’s brother. b. *Zhangsan shuile fuqin sleep father Intended meaning, “Zhangsan’s father has gone to bed.” c. *Zhangsan shanghaile fuqin hurtLE father Intended meaning, “Somebody hurt Zhangsan’s father.” d. *Lisi kule didi cryLE brother Intended meaning, “Lisi’s brother cried.”
24. In certain dialects, the particle zhe is optionally found in double-subject constructions as well, as in: 25. For phonological reasons which I leave aside, the predicate would sound better if it were bisyllabic. The adverb hen in the example serves this purpose.
Enriched composition and inference in the argument structure of Chinese
112
The verb zheguai ‘scold’ in (49a) is immediately ruled out because, being an activity verb, it does not contain a State or an inchoative Event and therefore cannot be integrated into the CS schema in (30). Of course it is also ruled out for an independent reason, since, under the intended meaning, the Agent of the action ‘sold,’ a necessary element in argument structure indexed in CS as the external/first argument, is not realized in the syntax. This same reason accounts for (49c) as well (‘hurt’ needs to have its Agent expressed), though it appears that the complex shanghaile does meet the requirement for a State component. In terms of verb types, shuile ‘sleep’ and ‘kule’ both can be construed as stative. However, under normal circumstances, the properties ‘father being asleep’ and ‘brother crying’ are not ‘noteworthy’ to another individual (even though related by kinship) in any useful way. I claim therefore that they are ruled out by the pragmatic condition of Noteworthiness (see section 4.2.1). The notion of Noteworthiness nicely accounts for the following contrast as well: (50) a. Kongzide houyi sile ConfuciusDE descendent dieLE “The descendent of Confucius died.” b. *Kongzi sile houyi “Confucius had his descendant die (on him).” (50b) is an instance of Possessum-object construction. Teng (1975) and Chappell (1999) discuss the contrast in the above in terms of the presupposition associated with an Experiencer role, which the subject of the sentence assumes. In present terms, even though the semantics of the predicate expression sile meets the condition of having a State component, the sentence is ruled out due to the fact that ‘the descendent being dead’ is evidently inconsequential to an ancient (Confucius), a failure to satisfy Noteworthiness. Nunberg (1995) makes the same point with examples adapted from Jackendoff (1992): (51) ??Ringo was hit in the fender by a truck two days after he died. Assuming Ringo refers to the car owner, the deceased Ringo does not acquire any property due to anything that happens to his car. In other words, the potential property does not matter in this case. It remains to be seen the extent of the instances that are covered by this pragmatic condition. The following are some additional examples where the situations reported are not ‘noteworthy’ enough to the subject argument under usual circumstances: (52) a. ?wo puopuo sile xi hua I grandma dieLE some flower “Some flowers of my grandma’s died.” b. ?Wangmian sile pengyou dieLE friend “Wangmian’s friend died.” c. ??ta binle taitai he be-illLE wife
Conceptual inference and predicate transfer
113
“His wife is ill.” Chappell (1999) suggests that the possessive relation involved in such constructions must be of an inalienable nature. According to Chappell, the problem in (52a) and (52b) above is due to the failure to satisfy this requirement. From present perspectives, inalienability of the possessive relation enhances the noteworthiness of the situation reported by the predicate expression. However, not all problematic cases can be related to inalienability, consider (52c) above. The conditions on predicate types laid out in (35) cast the net relatively wide, subsuming the class of unaccusatives, which Chappell (1999) suggests as the only verb class allowed in the construction. Chappell’s proposal follows naturally from the present account, if we assume with Pustejovsky and Busa (1995), along with Pustejovsky (1995), that lexical unaccusatives contain a binary right-headed event structure, which means that they designate/profile a State, thus satisfying the condition (35a). Chappell (1999) takes the following contrast as showing that unergative verbs do not meet the condition (adapted from Chappell 1999): (53) a. *tamen paole haizi they runLE child “Their child ran away (Lit. they had their child run away).” b. tamen de haizi paole Mod. “Their child ran away.” However, while verb pao ‘run’ does not profile a State component in its LCS and therefore cannot be unified into the schematic conceptual structure in (30), the particle le can function to adjust the LCS of the verb, as argued above. In this case, the combination paole does express a State containing the property of ‘being away’ and should be allowed under the present theory. Thus the data present a problem to my approach. I believe the problem brings up a separate issue, which has to do with the directionality of predicate transfer, discussed to some extent by Nunberg (1995), Note that the problematic example (53a) is similar to (54b) below: (54) a. zhege xiaohai sile fuqin thisCL little-boy dieLE father “This little child had his father die.” b. ??zhewei fuqin sile xiaohai “This father had his child die.” In both (53a) and (54b), an expression denoting parents appears as the subject, whereas one denoting child occurs in the object position. This arrangement is reversed in the acceptable (54a). In other words, the death event associated with the father can be extended to affect the child, but not conversely. Thus there is an asymmetry between father and child. This kind of asymmetry is independently instantiated in the following contrast: (55) a There comes a child/Mary from across the street. b. There comes a father from across the street. c. There comes Mary’s father from across the street.
Enriched composition and inference in the argument structure of Chinese
114
It appears that the expression child or a proper name like Mary enjoys some sort of referential independence, which the expression father lacks and must be modified by a possessor (55c). Such expressions fall into the category of relational nouns (Langacker 1987; Taylor 2002). Therefore, to return to (53a) and (54b), since the expression denoting ‘child’ is referentially independent, it will not be guaranteed that the ‘child’ is conceptually related to the subject by a possessive relation through CS binding. This configuration would be ruled out by the CS I postulated for the Possessum-object construction (30). In addition, the Noteworthiness condition precludes such a conceptual structure as well, since the father cannot be affected by the death of a child unrelated to him. A contrast of the same nature appears in the following: (56) a. tamen jia paole laomazi they household runLE maid “The maid ran away from their home.” b. ??tamen jia paole haizi they household run child “The child ran away from their home.” Chappell (1999) takes (56a) as an exception to her unaccusativity condition on the verb type, but sweeps away the problem by taking it as an instance of locative inversion construction. In my view, there is nothing about (56a) that is structurally distinct from other examples in this section (see section 5 for locative inversion). Arguably, laomazi ‘maid’ is referentially dependent (*duimian guolai yige laomazi ‘right in front comes a maid’) and thus allows for conceptual binding in (56a). Thus the issue of directionality in predicate transfer comes down to the lexical semantics of argument expressions within the predicate, a point made by Nunberg (1995) as well. One way to formalize the intuition regarding referential dependency noted above within the lexical semantics theory sketched in Chapter 2 is to suggest that the relational nominal within the predicate must be construable as a stage nominal (Busa 1999) or belonging to the functional type (Pustejovsky 2001; see Chapter 3), i.e., it makes reference to an agentive quale, in the way the concept of pedestrian refers to some walking-on-the-street event. In the present cases, the agentive quale is conveyed by the possessive relation noted in these examples. For instance, ‘child’ is what makes possible the content of ‘father,’ but not vice versa, and the same goes for the relation between a ‘househoid’ and a ‘maid.’ However, as noted in the last chapter, the functionality of an entity is subject to contextual coercion, especially with respect to the alienable cases in the present set of data (25), such as ‘horse’ in (25d) and ‘ship’ in (25g). My claim would be that these entities are usually of the natural type, but can be construed as functional given certain contextual and encyclopedic knowledge. At this point I leave a case-by-case study to a future project. 3.4. Summary I have been discussing the three issues surrounding the Possessum-object construction from the perspective of conceptual structure and semantic unification. I have argued that the systematic relations between (25) and (25’) are better encoded as conceptual inference relations rather than syntactic derivational relations. The external possessive
Conceptual inference and predicate transfer
115
relation involved in such constructions is coded via conceptual binding relation. Finally, based on the conceptual structure of the construction, I proposed a set of conditions restricting the semantic types of the predicates, the crucial component of which is a State containing a property. The bare essentials of such a proposal can be detected in the less rigorous works of Huang et al (1999)26 and Chappell (1999). I further proposed a lexicalsemantic solution to the directionality issue of predicate transfer, suggesting that the nominal argument within the predicate expression must have some functional need, which is satisfied by the possessive relation with the subject.
4. Retained-Object Passives The general approach advocated above for the Possessum-object construction in Chinese can be naturally extended to a class of passive constructions. The relevant examples from section 4.1 are reproduced below: (57) a. zhuguan de deguoren…rang laoxiang yong buqiang da-diaole in-charge Mod. German Pass. folk use rifle hit-awayLE yitiao tui one leg “The German in charge…had his leg broken by the folks with a rifle.” b. Lisi bei da-duanle yitiao gebo Pass. hit-brokenLE one arm “Lisi has had his arm broken” c. pinguo bei xuele pi apple Pass. peelLE peel (n.) “The apple has been peeled.” d. ta zongyu bei mianqule zuihou yige zhiwu he finally Pass. removeLE lastone post “He was finally removed from the last post.” Two important properties of this construction, mentioned in section 1, parallel those of the Possessum-object construction. First, these examples again exhibit external possessive patterns: the subject of each sentence above must be construed as a Possessor (in a broad sense) to the postverbal NP, as with Lisi and ‘one arm’ in (57b). Huang terms the construction the possessive passive, which is commonly found in East Asian languages such as Japanese, Korean and Taiwanese (Huang, to appear). Similar to the case with the Possessum-object construction, the Subject of these examples is not directly related to the passivized situation designated by the predicate of the sentence, since, for (57b) for example, what is affected by hitting and thus got broken is strictly ‘one arm,’ not Lisi. This has been noted by descriptive grammarians as early as Wang (1954), who observes that in this construction the object after the verb is the direct patient and the subject is the indirect patient Secondly, for these examples, where the Agent is 26. Huang et al (1999) note that the second NP in the construction must be in the State described by the intransitive verb.
Enriched composition and inference in the argument structure of Chinese
116
contextually available,27 an active alternation can be provided, where the possessive relation is expressed by an integrated NP. Compare (57a) with its active counterpart (58): (58) laoxiang yong buqiang da-diaole zhuguan de deguoren de folk use rifle hit-awayLE in-charge DE German DE yitiao tui one leg “The folks broke the leg of the German in charge with a rifle.” In this alternation, the direct Object designates the relevant possessive relation, expressed by the complex NP ‘the leg of the German in charge.’ Any theory of retained-object passives must at least account for the above two properties, among others. While it should be expected that the mainstream syntactic tradition might want to derive the above properties from syntactic representations and derivations (see e.g. Ting 1998; Huang (to appear); Xu 1999 for syntactic accounts, with some technical differences), yet given the parallels drawn above to the Possessum-object construction, the approach illustrated in the last section provides a ready and I believe simple semantic alternative to understanding the properties under discussion. In the remainder of this section, I develop a semantic perspective on passives in general and show how the two properties above are captured under this general conception. The general perspective sheds light on certain additional properties of Chinese passives as well. 4.1. Toward a Semantic Perspective on Passives The passive construction has been one of the central empirical issues since the inception of generative grammar (Chomsky 1957) and has been pivotal in the development of alternative theories such as Relational Grammar and LFG (see Bresnan 2001 for a recent textbook summary). Most accounts of passives capture the apparent grammatical function changing involved in terms of certain formal lexico-syntactic apparatus largely in abstraction from meaning, such as suppression of the external argument (Williams 1981; Grimshaw 1990, among many others), similar operations on the theta structure (an argument hierarchy abstracted from the Semantic Form) (Wunderlich 1997, 2000), plus case-marking properties of the participle invoked in GB theories (see e.g. Jaeggli 1986) that trigger movement for Case purposes. The data from the Chinese passive construction raises interesting questions for the standard argument structure accounts described above. A GB-type of account (see e.g. Li 1990; Xu 1999) has often been applied to standard cases of Chinese passives such as (59): (59) Zhangsan bei Lisi dale BEI hitLE “Zhangsan was hit by Lisi.” 27. The examples (57b, c, d) happen to exemplify what is called the short passive, where no Agent appears after bei. See Huang (to appear and references therein) for discussion.
Conceptual inference and predicate transfer
117
In such sentences, the morpheme bei is taken to ‘absorb’ the Accusative Case of the object and dethematize the subject position of the verb, in a way not unlike the function usually attributed to the English participial morphology. In other words, bei changes the argument structure of the verb as well as its putative case-assigning properties. However, in such analyses, the morpheme bei is also taken to be a P forming a PP with the following Agent NP. Thus it functions in the same way as the by-phrase in English passives. This leaves a lot of questions unanswered, among which is why the Chinese preposition bei (with a function similar to the English by) functions like English -en at the same time in manipulating the argument structure and the syntactic properties of the verb. In addition, unlike the English by-phrase, the morpheme bei is obligatory in Chinese passives, even though the Agent NP after it may be absent (see note 27). Thus it seems a correct analysis of Chinese passives hinges crucially on a proper understanding of the nature of the morpheme bei. Further problems with the traditional GB-type of approach for Chinese passives are discussed in detail by Huang (to appear), who argues convincingly that the NP following bei forms a constituent with the following VP, rather than with bei itself, as indicated in the following: (60) a. [IP Zhangsani [v’ bei [IP Lisi dale ei]]] Pass. hitLE Thus Huang basically supports the traditionai complementation analysis of Chinese passives which goes back to Hashimoto (1969), treating bei as a verb that takes a phrasal complement. In addition, he incorporates the null operator movement analysis due to Feng (1995) to deal with the obligatorily null object in (60). In other words, the embedded null object is adjoined to the embedded IP and gets bound by the matrix subject Zhangsan under predication. Thus in a way Huang combines the strengths of both complementation and movement proposals. For present purposes, it is important to note that on Huang’s and other complementation approaches to Chinese passives, the argument structure of the verb ‘hit’ does not change in the passive, the obligatorily null object notwithstanding28: the Agent remains the external argument of the verb in the complement IP. Instead, a verb-like element bei is employed which takes a complement phrase. The usual argument suppression approaches mentioned above do not seem to work well with respect to Chinese data. In view of the diversity of passive forms reviewed in a variety of languages, Huang (to appear) attempts to propose a universal syntactic notion of the passive, which is bound to be very abstract, due to syntactic differences crosslinguistically. For instance, Huang’s universal notion of the passive invokes various invisible 28. As Huang (to appear) and others note, there are cases where the empty object is filled with a resumptive pronoun, as in the following:
i. Zhangsan bei Lisi da-le ta yixia BEI hitLE him once “Zhangsan was hit once by Lisi.” Huang takes this as supporting the null operator movement analysis.
Enriched composition and inference in the argument structure of Chinese
118
elements such as traces and null operators, which would be unavailable to the present framework (see Chapter 2). Given the same kind of diversity of the passive which Huang tackled, I believe it important and perhaps more feasible to arrive at a functional (and hence universal) definition of passives. An initial attempt is found in Koenig (1999), who draws a parallel between the ‘low topic persistence’ of the Agent in passives and ‘discourse inertness’ of what he calls the ‘ultra-indefinite’ on construction in French, which satisfies an argument position without being accessible to discourse reference. The following illustrates: (61) *Oni a tué la présidente. Ili était Berry. (Koenig’s 18) INDEF have.PR kill the president he be from “Someonei killed the president. Hei comes from Berry.” Or “The president was killed. He comes from Berry.” The parallel with the on construction is brought about above with the short passive English translation. While (61) does not share any morphosyntactic property with the English passive, it does seem to share certain underlying functions expressed by passives. Thus Koenig (1999:249) defines a passive construction as “any rule, schema, or category whose effect is to satisfy a predicate’s external-argument valence requirement without introducing a corresponding discourse referent.” This amounts to saying that there is no universal morphosyntactic passive construction, but there is a universal passive function, a position which has been argued for in detail by Croft (2001) from his Radical Construction Grammar perspective. In this section, I show how attempts at a functional notion of the passive are facilitated by the representational devices of Conceptual Semantics. 4.1.1. The Conceptual Structure of Passives in English It should be noted that argument structure accounts must rely on some implicit assumptions about lexical semantics. For instance, in Grimshaw’s (1990) theory, the external argument is defined in terms of prominence on both the thematic and the aspectual dimensions of the predicate. I maintain that the suppression of the external argument would then be ultimately motivated by certain changes in the underlying lexical semantic operations. Such a view is advocated by Pustejovsky and Busa (1995) in their semantic account of unaccusativity. Building on ideas from Chierchia (1989), Pustejovsky and Busa propose that a causative event underlies unaccusative forms, with the resulting event (the formal quale) being the head of the event structure, and the causing event (the agentive quale) in the background, which can be accessed only through adjunction, as in the boat sank from the explosion. Pustejovsky and Busa (1995) (see also Pustejovsky 1995) hinted at a possibly similar treatment for passive participles, but did not offer an explicit analysis. This line of ideas with respect to passives is elaborated in greater detail by van Hout (1998) in a discussion of the semantic motivations of verbal alternations in Dutch. On her view, the active-passive alternation reflects different focus patterns on the same event, with the passive focusing on the passive character and the ‘theme’ participant of the event while defocusing the active character and the agent participant (see van Hout 1998:149). However, van Hout stops short of a concrete semantic representation for passives.
Conceptual inference and predicate transfer
119
Basically working from a cognitive linguistic perspective, Croft (1998) proposes a semantic representation of passives (cf. Langacker 1991) based on his theory of event structure (Croft 1991, 1993, 1994, 1998, see Levin 1996 for discussion). On this view, a complex event structure is a causal chain consisting of three subevents, CAUSE, PROCESS and STATE, with the CAUSE segment being defined in terms of force dynamics in the sense of Talmy (1988). Thus the CAUSE subevent in Croft’s theory corresponds in part to the action tier in Conceptual Semantics (cf. Chapter 2 above).29 Adopting the frame-semantic notion of profile/base from Fillmore (1985), as do Goldberg (1995) and Langacker (1987), which roughly corresponds to distinctions like presupposition/assertion (see discussion in Ch.2), Croft proposes a representation of the passive in which the PROCESS and STATE subevents are profiled, leaving the CAUSE segment (together with its ‘Agent’ role) in the background or base. This profiling configuration determines how semantic roles are linked to grammatical relations (see Croft 1998 for details). As noted previously, the cognitive-semantic concept profile corresponds to the Pustejovskian notion of event headedness (Pustejovsky 1988, 1995). It seems plausible that the idea of verbal profile can be applied to give more substance to van Hout’s notion of event focus. Thus it appears that from independent angles these authors converge on a particular semantic methodology in treating passive as an argument structure alternation, with Croft coming up with the most specific implementation. While the idea of profile/base or event headedness or event focus, which plays a big role in argument linking in the approaches discussed above, does not seem to have a technical counterpart in the framework of Conceptual Semantics, it does correspond partly to a distinction in conceptual structure representations between conceptual arguments that are indexed to be linked to syntactic arguments and those that are not, namely, the distinction between implicit and explicit arguments. For instance, for a transitive verb such as sink, the CS of its passive form such as (62a) could be represented as (62b), with the argument variables filled in (adapted from Jackendoff 1990:179): (62) a. The ship was sunk by the captain. b. In (62), where the action tier is shown, the external argument is given the subscript i, and the internal argument indexed as j. In the passive, the external argument of the verb is suppressed, which is indicated by the brackets around the index i. Thus the internal argument j would either be reindexed i in a way compatible with lexicalist theories and 29. The notion of causation is more finely decomposed in conceptual semantics and distributed among both the thematic and the action tier, thus tearing apart the ‘doer of action’ and the ‘extrinsic instigator of action,’ each being defined by independent functions such as AFF and CS+(=CAUSE) on separate tiers (see Jackendoff 1990 for details).
Enriched composition and inference in the argument structure of Chinese
120
linked to the grammatical subject30 or moved to the subject position through the casemarking properties of the verb, as in GB approaches. Given the idea of profile or event headedness described in the last paragraph, we could interpret this operation on the linking indexes over CS as a sort of defocusing or deprofiling the functions AFF¯ and CS+ as well as their arguments.31 I show in boid print the profiled portion of the conceptual structure of the passive in the following:
(62) c. From the standpoint of cognitive construal of events, there are two elements of universal significance that follow from this conception of the semantics of the passive. First, the notion of affectedness is a necessary element of the passive in that the grammatical subject of the construction must be related to the Patient on the action tier through a conceptual binding relation (cf. note 29), which in turn entails the presence of an Actor in conceptual structure, though both are part of the base/frame of the passive semantics. Second, the second argument of CS+, which is underlying an Effect, is foregrounded in the passive, due to the defocusing of CS and AFF. Thus the Theme, assigned by GO within the Effect component, is expressed as the Subject in the passive. The defocused argument(the Actor) can be optionally referenced only through adjunction or oblique case marking (Pustejovsky 1995; Croft 1998), though it is known to bind the subject in a rationale clause, as with the famous example cited in Jackendoff (1990), the ship was sunk to collect the insurance. Due to the connection of the by-phrase adjunct to the Actor (Agent) role in the argument/conceptual structure, it is dubbed A-adjunct by Grimshaw (1990). 4.1.2. Do Chinese Passives Mean Something Different? An important question at this point is whether the CS proposed for English passives such as (62b) applies to Chinese passives as well. In view of the distinct properties of the long passive in Chinese discussed earlier and the syntactic structure proposed in (63a) repeated below:
(63) a. [IP Zhangsani [v’ bei [IP Lisi dale ‘hitLE’ ei]]] Pass.
hitLE
30. Note that a different set of indices is presented with Greek superscripts, which are meant for argument binding. On Jackendoff s theory, the action tier roles takes precedence over thematic roles in argument linking, concurring in essence with Croft (1998) (cf. Nikanne 1995). So the action tier roles are binders indicated by superscripts, binding the roles on the thematic tier. 31. Due to defocusing, the index j might need to shift to its bindee on the thematic tier, which could be taken as part of the changes involved in the LCS of the passive participle.
Conceptual inference and predicate transfer
121
a CS representation of the sort in (62b) does not fit Chinese data. The crucial fact concerns the constancy of the verbal argument structure, which basically remains unchanged in this analysis and shows up in the IP complement, as observed earlier. Thus there should be no CS defocusing in this case, contrary to the CS representation in (62c). Therefore the Chinese passive might correspond to a CS configuration that is somewhat different from (62b) and (62c). Certain crucial grammatical and lexical properties of the Chinese passive provide the best clues to the conceptual structure it expresses. First, there is strong evidence that the NP following bei forms a constituent with the following VP, rather than with bei. Huang (to appear) presents three arguments for the analysis as presented in (63a). First, the beiNP sequence does not stand the standard constituency test: it cannot move across a time phrase or to a sentence-initial position. The following (from Huang, to appear) illustrate: (64) a. Zhangsan zuotian bei Lisi dale yesterday BEI hitLE (cf. John was hit by Bill yesterday.) b. *Zhangsan bei Lisi zuotian dale BEI yesterday hitLE (cf. John was hit yesterday by Bill.) c. *bei Lisi Zhangsan zuotian dale BEI yesterday hitLE (cf. It was by Bill that John was hit yesterday.) Secondly, the Agent NP after bei forms a phrasal constituent with the following VP, as shown by the following coordination test: (65) ta bei Lisi male liang-sheng, Wangwu tile san-xia he BEI scoldLE twice kickLE three-times “He was scolded twice by Lisi and kicked three times by Wangwu.” The same fact is noted by Li (1994) as well in a descriptive study of Chinese passives, with attested examples such as the following: (66) Qinfeng yige yue mei xia di, bu bei feng cui ri shai, oneCL month not go-down field not BEI wind blow sun bask shenzi geng fengyi qilai body more plump Part. “Qinfeng hasn’t been to the field for a month, and so has not been exposed to the wind or the sun. Her body is becoming more plump.” What is interesting about this example is that the coordinated subject-predicate structure feng cui ri shai following bei fits the typical four-syllable template for Chinese idiomatic phrases (‘four-character idiom,’ in plain terms), which most often assume an extended meaning. This supports the independent status of the sequence following bei. Thirdly, Huang (to appear) cites examples like the following to show that the NP after bei can antecede a subsequent reflexive pronoun ziji, thus showing the subject status of the NP in the embedded clause: (67) nei-feng xin
bei Lisi
dai-hui
ziji
de
jia
qu le
Enriched composition and inference in the argument structure of Chinese
122
thatCL letter BEI Lisii bring-back selfi DE home go LE “That letter was brought back to his (=Lisi’s) home by Lisi.” Apart from the arguments above for the constituency status of the sequence following bei, there is some independent evidence that points to certain verbal (rather than prepositional) characteristics of the morpheme bei. Li (1994) surveys the disputes in the literature of Chinese linguistics on the grammatical status of bei, including some evidence for its verbal character. To begin with, as a historical note, Li (1994) traces the semantics of the morpheme to the original meaning “put on (clothes),” which then extends to senses such as “cover,” “undergo,” “experience.” Such verbal uses are still attested in Modern Chinese, as the following shows (cited in Li 1994)32: (68) Liyin buyou de li-qilai, bei shufu yanguang de yinglin cannot-help DE stand-upBEI uncle gaze DE direction you yiyuweifa de zuoxia and without-a-word DE sit-down “Liyin could not but stand up, followed (lit. ‘experiencing’) the gaze of his uncle, and then sat down without a word.” In the example above, bei is followed by a complex NP describing an event, which makes bei susceptible to the verbal reading of “experience.” A similar use of bei is found in the following example from Li (1994): (69) Yuandi miao beile dao temple BEI-LE theft “The Temple of Yuandi was burglarized (lit. ‘underwent a burglary’).” In this case, bei is affixed with the particle le, which is usually taken to be indicative of verbal character.33 If we are on the right track in hypothesizing that bei has some verbal character and even denotes some sort of event like ‘undergo’ or ‘experience,’ the subject of the Chinese passive would be assigned some independent thematic role by the event structure denoted by bei, a role such as Experiencer. Thus it would not be surprising to find subject-oriented adverbs in the Chinese passive, as shown in the following example from Huang (to appear): (70) Zhangsan guyi bei Lisis dale intentionally BEI hitLE “Zhangsan intentionally got hit by Lisi.” 32. Note incidentally that Li (1994) still treats bei as a preposition, regarding the evidence presented in the following as not overwhelming. In view of the large literature on ‘light’ verbs, the data presented by Li can be better accommodated. 33. A process termed ionization after Chao (1968) might be at work here, according to which beidao in (69) is a VV compound and is reanalyzed as a V’ phrase due to the historical development of disyllabifiction of verbs. cf.
(i) a. bei -bu BEI-arrest “be arrested”
Conceptual inference and predicate transfer
123
b. beile bu Thus some historical restriction is placed on this process, but it does not affect the point being made here. See Huang (to appear) for discussion of historical issues.
The presence of the subject-oriented adverb in the above indicates that the subject must take a more dynamic role such as Experiencer (though the Patient role can be bound to it, see 62c above), as Huang points out. However, in spite of these pieces of evidence that point to the verbal character of bei, this morpheme does not seem to function like a full verb. As pointed out by Li (1994), unlike full verbs, bei cannot take a nominal object, cannot be affixed with aspectual particles like zhe or guo and cannot be reduplicated (*bei-bei, cf. kan-kan ‘read’). The grammatical properties discussed above would seem to fall out naturally if we simply assume that a typical Chinese passive (which I take to be the long passive; I ignore the short passive for sake of argument) such as (59) expresses an event of undergoing or experiencing some particular event that affects the experiencer in a relevant way. Granting the particular sense of ‘experiencing’ attributed to bei, its IP complement could be taken to correspond to the event being experienced (cf. English getpassive). Thus we arrive at a tight alignment between syntax and semantics, a welcome result in theories of interfaces. Note that an event of ‘experiencing’ can be construed as some sort of reacting, which involves energy delivered to an event exerting an impact. in other words, there is at work some force-dynamic relationship (Talmy 1988) between two opposed force entities. The case at hand also presupposes a converse force interaction in which one character (antagonist or Actor) directly impacts (acts on) another (agonist or Patient). Jackendoff (1990) suggests the function REACT to capture the agonist’s reaction toward an antagonist. Typical instantiations of this function are found with English verbs of reaction such as give in, withstand, resist, which differ with respect to whether the antagonist gets his/her way (see Jackendoff 1990:137). I use REACT+ for cases of antagonist success, as with give in, and REACT− for antagonist failure, as with withstand. Resist is a case of undetermined outcome. Now, if the meaning of passives is construed as expressing the function REACT, an interesting generalization emerges: the function REACT in the passive should always take a plus value: as a defining feature of the passive, the agonist always gives in to the antagonist. In view of the considerations above, the Chinese passive in (60a) receives the following CS representation (simplified somewhat) in (60c): (60) a. Zhangsan bei Lisi dale BEI hitLE “Zhangsan was hit by Lisi.” b. [IP Zhangsani [v’ bei [IP Lisi dale ‘hitLE’ ei]]] c. The conceptual structure analysis above makes explicit the tight alignment between syntax and semantics mentioned earlier, with the Patient (being a reactor) linked to the subject position and an affecting Event linked to the complement IP in (60b). On this
Enriched composition and inference in the argument structure of Chinese
124
analysis, the morpheme bei is a ‘light’ verb in that it designates nothing more than the function REACT+, a function on the action tier in Jackendoff s system. Thus it differs from verbs of reaction in not having lexically encoded thematic content, a property underlying its less-than-full verb status stressed by Li (1994). Let us compare (60c) with the CS for English passives (see (62b) above). As shown in (60c), the first argument of REACT+ must be related to the Patient within the Event argument through a conceptual binding relation. Thus the affectedness property of passives noted earlier is replicated, though in a fashion somewhat different from the CS for English passives (cf. (62b)). However, as the argument structure of the verb remains unaffected within the second argument of REACT+, there is no defocusing of the CAUSE function, contrary to the other property of the CS for English passives (see (62c)). Through formal comparison, my analysis shows that the affectedness property of the subject of the passive is attested cross- linguistically, with variations as to how this property is realized. 4.1.3. Directness of Passives In (60c), the Goal of the hitting event on the thematic tier must also be bound to the first argument of REACT+. This raises questions as to whether this is a necessary conceptual condition on Chinese passives in general. Phrased more generally, the issue is whether the first argument of REACT+ must bind any thematic tier element in the embedded Event, as is the case in (60c). The following examples show the range of the cases that are possible: (71) a. Zhangsan bei Lisi pipingle BEI criticize LE “Zhangsan was criticized by Lisi.” b. Zhangsan bei Lisi daduanle yitiao tui BEI hit-breakLE one leg “Zhangsan had a leg broken by Lisi.” c. *Zhangsan bei Lisi taopaole fleeLE “Zhangsan had Lisi fleeing away (on him).” d. Zhangsan zuile yizhen, haishi bei Lisi taopaole chaseLE a-while still BEI fleeLE “Zhangsan chased (Lisi) for a while, (but) still had Lisi fleeing away (on him).” e. *Zhangsan bei Lisi baozhu banzi BEI Lisi hold birch “Zhangsan had the birch held by Lisi.” f. Zhangsan hai yao dashi, zhao bei Lisi baozhu banzi still want hit time early BEI hold birch “Zhangsan wanted to beat (him) again, but had the birch held by Lisi.” In terms of conceptual structure, three possibilities emerge from the above examples and contrasts. First, the subject of the passive conceptually binds a thematic role within the predicate, as is the case in (71a). Second, the subject binds a possessive modifier within the predicate, as illustrated by (71b). Note that (71b) contrasts with (71e), another case of
Conceptual inference and predicate transfer
125
possessive modifier being bound to the subject, which is however ungrammatical. Nevertheless, with enough discourse context as in (71f), the sentence is made acceptable. Thirdly, the subject is not conceptually related to any element within the predicate, which sounds bad as shown in (71c), but is again made acceptable under the discourse context as made explicit in (71d). These cases instantiate the major types of passives recognized by typologists, particularly regarding East Asian languages (Washio, 1990; Huang, to appear). For instances, cases like (71b, 71e, 71f) have been called possessive passives found in some East Asian languages, a kind of indirect passive, as opposed to the direct sort in (71a). (71d) exemplifies what are called adversative passives (see also Wierzbica 1988 on Japanese adversative passives), another kind of indirect passives. The intervention of the discourse context in indirect passives shows that the relevant binding relation on the thematic tier is not a necessary condition in a conceptual definition of passives. These facts are better captured by a looser semantic definition of the role Patient. As Jackendoff (1990:294, note 1) points out, Patient role can be assigned either by the semantics of the verb or by discourse context. It is the availability of the discourse Patient that broadens the scope of passives. In summary, I schematize the necessary elements of the CS of Chinese passives as follows: (72) [REACT+ ([ ]α, [Event AFF− ([ ], [α])])] What this says essentially is that an entity experiences a particular event in which the said entity is somehow negatively affected. And the action tier shown in the embedded event could derive from the discourse context as well as the lexical semantics of the embedded event. Again, only the first property of the English passive noted in section 4.1.1, namely, affectedness associated with the external argument, is maintained in the Chinese passive. 4.2. Passive Alternation as CS-Inference The schema in (72) crucially relies on force dynamic relations that define the action tier. A force-dynamic relation involves tension between two opposed force entities. This kind of relationship is captured by the function AFF, as well as the function REACT. Jackendoff (1990) points out that REACT is a sort of mirror image of AFF, which means that from a particular action tier configuration featuring AFF can be inferred another configuration headed by REACT. Thus we may have an inference rule like the following: (73) [Event AFF (x, y)] <---> [Event REACT (y, x)) In the rule above, the thematic information is not specified, except being implied by the node Event. According to this schema, x engages in a causative event that affects y. And from y’s point of view, y reacts to x. Furthermore, I assume an additional inference rule according to the set inclusion principle (see 23): (74) [Event REACT (y, x)] → [Event REACT (y, [Event (x, y)] ) ] This means that the event in which x is engaged is part of the property of being x. If y reacts to x, y reacts to the event in which x (and y) engages. The rule (74) captures the following nominal alternations (Jackendoff 1990:137):
Enriched composition and inference in the argument structure of Chinese
126
(75) a. Sam gave in to/withstood/resisted Harry. b. Sam gave in to/withstood/resisted Harry’s pressure on him. From (73) and (74) follows naturally the rule (76): (76) [Event AFF (x, y)] < --- > [Event REACT (y, [Event (x, y)]) ] I suggest this is how the relations between Chinese active and passive sentences are coded, that is, within the CS-inference component, in contrast to operations that work on lexical conceptual structures (similar to the lexical rules of LFG) for English passives (see 62b), though neither of the approaches resorts to an underlying syntactic structure as with derivational approaches. Note that the event that y reacts to in this case is exactly the affecting event preceding the arrow in (76). For instance, for a simple long passive such as (59), its CS derives from an affecting event in which Lisi hit Zhangsan and negatively affects Zhangsan, which is exactly the event embedded within the conceptual structure in (60c), obeying the rule in (76). What about passive alternations involving external possessions such as (57a)? The alternation itself is based on the same inference rule as (76), but the external possessive relation entails a further rule of inference again based on the set inclusion principle. The rule involves the second argument y in an affecting event, which can be schematized as follows (cf 24): (77) [Event AFF (x, [y [at z]]) <---> [Event AFF (x, z)] Recall that possessive relations are coded as a spatial property in conceptual semantics, as shown in (77). What (77) says is that if x affects y, x affects the possessive modifier of y, another case of predicate transfer in the sense of Nunberg (1995). Thus this rule is subject to Nunberg’s condition of noteworthiness discussed earlier. From (76) and (77) we can derive the CS such as the following: (78) [Event REACT (z, [Event (x, [y [at z]])])] In this structure, z within the modifer will be bound in CS by the higher argument It is this struture that underlies the possessive passives in (57). Take the simpler one in (57b): (57b) Lisi bei da-duanle yitiao gebo Pass. hit-brokenLE one arm Lit. “Lisi was hit-broken an arm.” “Lisi had his arm broken.” The present analysis entails several CS conditions that follow from (77) for such possessive passives to be felicitous. Informally, one of Lisi’s arms must be negatively affected by some force entity. Furthermore, ‘Lisi’ is a conceptual spatial modifier to the ‘arm’ being affected. Thirdly, to return to Nunberg (1995) once again, ‘Lisi’ must be relevantly affected by the same event in a negative way, in line with Nunberg’s Noteworthiness Condition. This third condition ties in with the notion of discourse Patient noted earlier. When ‘Lisi’ is relevantly affected due to its Possessum being affected by some event, ‘Lisi’ becomes a discourse Patient, since the role is not assigned by the lexical semantics of the verb. While these conditions are perfectly satisfied in
Conceptual inference and predicate transfer
127
(57b), the noteworthiness condition is apparently (i.e., in normal context) flouted in the following possessive passive (Huang, to appear:54): (79) ?*Zhangsani bei wo wenle eibaba le BEI I askLE father Part. “Zhangsan had his father asked (a question) by me.” In this example, Zhangsan is construed as a spatial modifier of ‘father,’ but Zhangsan is not obviously being affected due to his father being affected by a questioning event. In other words, Zhangsan is not easily construed as a discourse Patient, and hence the problem in its acceptability.34 The approach being advocated here contrasts with both Huang (to appear) and Xu (1999). For theory-internal (and syntactocentric) reasons, Huang (1991; to appear) proposes that indirect passives (including adversatives) are characterized by a complex V’ or VP predicate that takes a null ‘outer’ object that, being coindexed with the subject, controls the null possessor in possessive passives and serves in the role of an Indirect Affectee. The outer object moves to IP by null operator movement, and gets an index from the subject under predication. Thus, for (57b), it would be analyzed along the following lines:
(80) Huang’s analysis is consistent with the framework he adopts and he basically captures the same idea of affectedness associated with the external argument (cf. 72), though he syntacticizes the idea with an abstract syntactic representation. Apart from the usual methodological criticisms of syntactocentrism (see Chapters 1 and 2 for some discussion and references), such an analysis overlooks the inferential relation schematized in (77). In addition, the notion of Patient (or Affectee, in Huang’s terminology) cannot be rigidly defined over a syntactic configuration, since discourse context plays a crucial role, as pointed out earlier. The ‘outer object’ analysis does not account for the problem in (71e). The contrasts in (71c, d) and (71e, f) clearly point to the need for discourse context in these cases.35 For instance, as shown in the contrast between (71c) and (71d), it is clearly due to the fact that Zhangsan had been chasing Lisi that Lisi’s fleeing away makes Zhangsan an Affectee or Patient. It should be noted however that Huang and I agree on some crucial points 34. This analysis is compatible with Huang’s (to appear) ‘outer object’ analysis (see below) whose essence is to capture Patienthood in terms of a covert grammatical relation. As will be pointed out shortly, such syntactocentric analyses run into empirical problems due to the possibility of discourse Patient. 35. Huang (to appear, 56) gives an example similar to (57c):
(i) *Zhangsan bei Lisi pao hui jia qu le BEI run back home go LE “Zhangsan had Lisi run away home (on him).” This is intended to show that Mandarin does not normally allow adversative passives, but he ignores the fact that the sentence can be acceptable under proper discourse contexts.
Enriched composition and inference in the argument structure of Chinese
128
regarding Chinese passives, notably, the verbal character of the morpheme bei and the associated complementation structure.36 Xu (1999) proposes a possessor-raising sort of syntactic analysis for possessive passives similar to his proposal for Possessum-object construction (see 4.3.3.1). For instance, for (81a) below: (81) a. Lisi bei toule yige qianbao BEI stealLE one purse “Lisi had a purse stolen.” Xu (1999) proposes a D-structure as follows: (81) b. [e [VP bei toule [NP Lisi de [N’ yige qianbao] ]]] On this analysis, the possessive relation is coded as an integrated NP in the underlying structure. Then either the whole NP Lisi de yige qianbao moves to the empty subject position for Case, yielding a normal passive, or only Lisi moves to the subject position, deriving a retained object passive. Such an analysis is subject to many criticisms empirically as well as theoretically. For one thing, some of the problems mentioned in section 3.3.1 rear their heads again, such as the ad hoc treatment of the morpheme de. In addition, on the present analysis, Chinese passives have their distinct structures of semantic predication (the subject can get its own role such as Experiencer), which is not captured by such movement approaches, as criticized as well by Huang (to appear). Furthermore, it is not clear raising the possessive Spec of the NP is a welcome move, since it violates the well-known Left Branch Condition (Ross 1969), as Huang (to appear) points out. The present approach (and to some extent Huang’s account as well, though paying the price of having a more abstract syntax) avoids such problems by assuming a more active semantics and inferential component, thus shifting away the burden from the syntax. 4.3. Summary Two basic issues discussed with respect to the Possessum-object construction in section 4.3 are studied again in the context of the retained-object passives, alias possessive passives, namely, the external possessive relation and the active-passive alternation. The results are essentially in the same spirit as in the last section: the alternation is coded as a conceptual inference relation between two converse action tier functions AFF and REACT. The possibility of the external possessive relation in such passives is motivated by a further inference rule involving an extension in the Nunbergian sense of an affecting predicate: if entity x negatively affects y, x relevantly affects z that modifies y spatially. And z can therefore react to the same event that affects y, which is the basic meaning of Chinese passives. The analysis in this section underscores differences as well as similiarities between English and Chinese passives. The role Patient is given prominence in the passives of both languages, but in 36. I don’t intend to give a full review of Huang (to appear) here, which contains a lot of interesting data and important insights.
Conceptual inference and predicate transfer
129
somewhat different ways. Whereas in English this is achieved through defocusing the function AFF and the Agent, leaving the Patient bound to the Theme on the thematic tier, in Chinese, the top argument (i.e., the first argument of REACT37) must bind the Patient role within the embedded event, and the availability of the Patient role is determined either by verbal semantics or by discourse context, which significantly leaves the door open for the so-called indirect and adversative passives found in East Asian languages. It should be emphasized that these results are made possible by an explicit CS represenation of the semantics of the passive in Chinese as well as English, building on the insights of Pustejovsky (1995), van Hout (1998), and Croft (1998) as well as Jackendoff (1990).
5. Locative Inversion I have so far examined two classes of examples in which the grammatical subject does not bear a straightfoward semantic relation to the semantic structure of the verbal predicate. Reinterpreting Nunberg’s (1995) insight on predicate transfer in a ConceptualSemantic framework, I have proposed that the subject of such sentences is licensed by rules of conceptual inference, and the (external) possessive relation exhibited in both types of examples makes it possible for the conditions of inference to be met. This section brings in another class of sentences which illustrate a type of predicate transfer and hence of conceptual inference which is not obviously mediated by possessive relations. The construction under discussion has standardly been termed locative inversion, and to my knowledge it has never been discussed in conjunction with the other two types of constructions that are of concern to this chapter.38 Locative inversion is a much discussed topic in the generative literature, presumably because it involves mismatches of argument role, grammatical function and categorial structure and thus may throw some light on the overall architecture of UG, as Bresnan (1994) observes. Important recent studies on the topic include Bresnan and Kanerva (1989, 1992), Bresnan (1994), Levin and Rappaport Hovav (1995), and Culicover & Levine (2001), among others. Citing examples such as the following from English (examples from Bresnan 1994): (82) In the corner was a lamp. (83) Among the guests was sitting my friend Rose. (84) Back to the village came the tax collector. 37. This comparison raises an interesting contrast: the subject of the Chinese passive is an action tier role, whereas that of the English passive is a thematic tier role which binds the Patient (see note 31). Thus English passives are in a sense more ‘passive’ than Chinese passives, accounting for the presence of subject-oriented adverbs in Chinese passives, as noted by Huang (to appear). 38. Huang et al (1999) make reference to Chen (1995) who proposes to relate the meaning of the Possessum-object construction to locative inversion via metaphorical extension of construction meaning (cf. Goldberg 1995).
Enriched composition and inference in the argument structure of Chinese
130
Bresnan (1994) and Levin and Rappaport Hovav (1995) note the formal properties of the construction such as the following, (i) postverbal NP subject; (ii) preverbal locative PP; (iii) intransitiveness of the verb in the construction, a property best known for the construction due to its connection to unaccusativity. The first two properties are clear from contrasts with the canonical counterparts (Bresnan 1994): (82’) A lamp was in the corner. (83’) My friend Rose was sitting among the guests. (84’) The tax collector came back to the village. The third property is illustrated by the obvious ungrammaticality of the following: (85) a. The mother scolded the kid in the room. b. *In the room scolded the kid the mother. c. *In the room scolded the mother the kid. Chinese can be said to have the same construction to the extent that there are constructions that instantiate the pattern locative phrase+V+NP, where the NP is a Theme whose location or existence is of concern. I repeat the relevant examples from (4) below: (86) a. Zhangsan de tou-shang zhuangle yige bao Mod. head-top bumpLE one swelling “Zhangsan’s head got a swelling from bumping.” b. chuang-shang tang-zhe yige bingren bed-top lie-Dur one patient “In the bed lies a patient” c. shu-dixia zhan-zhe liangge xiaohai tree-bottom stand-Dur two child “Under the tree are standing two children.” d. fangjian-li fang-le xuduo xingli room-inside putLE many luggage “Inside the room are placed many pieces of luggage.” e. qiang-shang gua-zhe yiding maozi wall-top hang-Dur one hat “On the wall is hung a hat.” f. shulinli paochulai liangge shibin woods-inside run-out-come two soldier “Out of the woods ran two soldiers.” Given the set of putative locative inversions in Chinese as presented above, it is not clear the set exhibits the same basic properties noted by Bresnan (1994) and by Levin and Rappaport (1995). First, they usually don’t have a so-called canonical counterpart and indeed the preverbal locative phrase has been argued to be a base-generated subject (Gu 1992, cf. Li 1990). Consider (86b) and (86b’), for example: (86) b.’ *yige binren tangzhe chuangshang one patient lieZHE bed-top
Conceptual inference and predicate transfer
131
Second, the initial locative phrase is a NP (Huang 1987, Li 1990, Gao 1996), rather than a PP, thus similar to Chicheŵa (Bresnan 1994) in this respect. I have argued in section 1 that it occurs in the subject position. Third, transitive verbs are instantiated in such examples, as in (86d,e) (see also Pan 1996). The literature on the topic has most often focused on the syntactic structure of the construction, in particular the syntactic status of the preverbal locative phrase and the postverbal NP (see Bresnan and Kanerva 1989, 1992; Bresnan 1994; Levin and Rappaport Hovav 1995; Culicover and Levine 2001 and references therein). The present section will not be concerned with choosing among alternative syntactic analyses of such constructions, but will draw on some well-known grammatical properties of the construction and see how they align with a proper semantic analysis, which may turn out to simplify and vitiate certain syntactic proposals. Bresnan (1994) and Levin and Rappaport (1995) make some specific proposals concerning the semantics and the discourse function of the construction without however invoking an articulate semantic theory.39 In this section, I discuss the semantics of the locative inversion construction in Chinese (5.1). This provides a perspective as to the way the subject is licensed and the distributional restrictions on the verb in such constructions (5.2). The approach is compared with the proposal of Dowty (2000a, b) on the English swarm-with construction in (5.3), such as the garden swarms with bees. Section 5.4 summarizes and discusses the implications. 5.1. Predicating the Location In the above, I noted that unlike locative inversion in English, the construction in Chinese often does not have a formal ‘canonical’ equivalent (cf. (86b) and (86b’), (82)–(84) and 82’–84’). However, Chinese locative inversion patterns clearly correspond to a set of ‘canonical’ instances that roughly paraphrase them: (87) a. yige bao zai Zhangsan de tou-shang chuxianle one swelling Loc. Mod, head-top appearLE40 “A swelling appeared on Zhangsan’s head.” b. yige bingren tang zai chuangshang one patient lie Loc. bed-top “A patient is lying on the bed.” c. liangge xiaohai zhan zai shu-dixia two child stand Loc tree-bottom “Two children are standing under the tree.” d. Zhangsan fangle xuduo xingli zai fangjian-li putLE many luggage Loc. room-inside “Zhangsan put many pieces of luggage in the room.” 39. It should be noted that Bresnan’s (1994) proposal concerning the argument structure restriction of the construction comes closest to the basis of the present proposal. 40. It is difficult in this paraphrase to maintain the verb zhuang ‘bump,’ due to the possible resultative reading associated with (86a).
Enriched composition and inference in the argument structure of Chinese
132
e. Zhangsan zai qiangshang guale yiding maozi Loc. wall-top hangLE “Zhangsan hung a hat on the wall.” f. liangge shibin cong shulinli paochulai two soldier from woods-inside run-out-come “Two soldiers ran out from within the woods.” In the above sentences, the Theme appears as the subject of the sentence, rather than taking the object position, as in (87). Given the list of paraphrases above, especially (87b) and (87c),41 it is tempting to claim that sentences in (86) mean roughly same as their paraphrases in (87). This means that the conceptual structure from which is derived verbal argument structure (Jackendoff 1990) must remain the same for the equivalents in the two sets of sentences, for instance, between (86b) and (87b), (86c) and (87c). This view in fact underlies one of the common assumptions in the literature concerning the semantics of the construction. For instance, it has been proposed, in the influential works of Bresnan and Kanerva (1989) and Bresnan (1994) that the argument structure of the verb in locative inversion must be of the form . The implicit assumption of this proposal is that the verbs used in, for example, (86b, c) and (87b, c) maintain the same argument structure and thus the same conceptual structure, each taking a Theme and a Location argument. 5.1.1. Aspectual Restriction on Locative Inversion There are three reasons that we might want to drop the assumption concerning the constancy of verbal meaning in the two patterns of (86) and (87), if we look more widely than just semantic roles. First, the case of the transitive verb in (87e) suggests that its argument structure must be changed in the equivalent (86e) where it is suffixed with the marker zhe.42 Pan (1996) proposes that the suffix zhe induce the morphological operation of deleting the Agent argument. This of course does not affect Bresnan’s argument structure condition on locative inversion, but it does show that the meaning of a transitive verb changes when it occurs in locative inversion as opposed to the canonical pattern. Given the corresponding paradigms in (86) and (87), we might wonder if the meanings of intransitive verbs change as well in locative inversion. Second, (86) and (87) are characterized by grammatical function reversal, i.e., what occupies the subject position in (86) (Theme) becomes the object in (87). It would be a welcome result to find motivations for such formal changes, especially if we assume something like maximal functionality of grammatical form as a psychological principle organizing grammatical information, now being entertained in some version of Construction Grammar as a principle of no synonymy (see Goldberg 1995). Indeed the literature has motivated the formal change in terms of the discourse notion of presentational focus or ideas along this line (Bresnan 1994; Rochemont 1986; see Levin 41. Other examples involve transitive verbs and unergative verbs, whose argument structure may change across (86) and (87). Thus apparent differences exist as to what constituents of meaning get expressed. 42. The argument structure does not change when the aspectual marker is le in locative inversion, but the Agent may be syntactically deleted (see 86d), as Pan (1996) argues.
Conceptual inference and predicate transfer
133
and Rappaport 1995 for discussion of related proposals). Levin and Rappaport (1995) in particular attempt to reduce Bresnan’s argument structure restriction on locative inversion verbs to the discourse-oriented notion of ‘informational lightness’ of the verb, drawing ideas from Birner (1992). However, it is very likely that the differences between (86) and (87) are motivated not only in discourse functions, but also in their respective semantic structure. In an articulate framework of semantics, conceptual and focus structures are different tiers of meaning associated with a single syntactic structure (Van Valin and LaPolla 1997; Jackendoff 2000). In this spirit, Bresnan formulates conditions on both the argument structure (which is based in CS) and the focus structure of the construction: the two sorts of conditions are in principle independent of each other. Thus the reductive discourse account of Levin and Rappaport (1995) is in this sense problematic theoretically: it is still necessary to formulate precise conditions on the conceptual structure of the construction.43 Thirdly and most importantly, we do find evidence pointing to conceptual differences between (86) and (87), apart from discourse-pragmatic differences. This is shown in the following contrasts: (88) a. yige bingren xunshu tang zai chuangshang one patient quick lie Loc. bed-top “A patient laid on the bed quickly.” b. *chuangshang xushu tangzhe yige bingren Lit. *“In the bed lies a patient quickly.” c. zhe jitian, chuangshang tangzhe yige bingren this several-day “In the bed lies a patient for the last several days.” (89) a. Zhangsan hen kuai fangle xuduo xingli zai fangjianli very quick putLE many luggage loc. room-inside “Zhangsan quickly put many pieces of luggage in the room.” b. *?Fangjianli hen kuai fangle xuduo xingli44 Lit. ??“In the room are put many pieces of luggage quickly.” c. zhe jitian fangjianli fangle xuduo xingli this several-day “In the room are put many pieces of luggage for the last several days.” (90) a. Zhangsan hen kuai zai qiangshang guale yiding maozi very quick Loc.wall-top hangLE one hat 43. See Nakajima (2001) for a related criticism of Levin and Rappaport’s (1995) discourse account of locative inversion, and Du (1999) for counterexamples to the ‘informational lightness’ story of the verb. 44. The sentence is worse if the verb is suffixed with zhe, rather than le:
(i) *fangjianli hen kuai fangzhe xuduo dongxi room-inside very quick putZHE many thing Lit. “In the room are put many things quickly.” This is presumably because the Agent can be expressed in locative inversion when the verb occurs with le:
Enriched composition and inference in the argument structure of Chinese
134
(ii) fangjianli Zhangsan hen kuai fangle xuduo dongxi Lit. “In the room Zhangsan put many things quickly.” “Zhangsan hung a hat on the wall quickly.” b. *qiangshang hen kuai guazhe yiding maozi Lit.?? “On the wall is hung a hat quickly.” c. zhe jitian qiangshang guazhe yiding maozi this several-day “On the wall is hung a hat for the last several days.” The (b) sentences above show that manner or rate adverbials cannot be present in locative inversion, in contrast to the cases in (a), and examples in (c) show that locative inversion allows duration adverbials. This generalization applies to locative inversion in English as well, as the status of the translations above show. This leads us to the following conclusion: (91) Aspectual Restriction on Locative Inversion: locative inversion designates a State. Note that verbs in the canonical pattern in (87) are not so restricted: often they denote events, as the tests above shows, but the verb zhan ‘stand’ in (87c) denotes a State as well. 5.1.2. Conceptual Structure and Inference A natural question at this point is why locative inversion must express a State, instead of an Event. In my view, this aspectual restriction may find a plausible explanation if we hypothesize that there is a conceptual predication relation between the Location subject and the VP in locative inversion constructions. Du (1999) points out two distinct functions of the construction: a presentational focus function and one that comments on or describes a given location. This second function has not to my knowledge been noted by other authors,45 but is consistent with the present location predication view. Independent support for the predication view comes from examples such as the following: (92) shuxia zhan-manle ren tree-under stand-fullLE person Lit. “Under the tree stood-full people.” In this example, the verb zhan ‘stand’ forms a compound with the adjective man ‘full,’ which, as Du (1999) points out, could only be taken to predicate of the Location ‘under the tree’: the place is full of people. This means that the preverbal locative phrase does not merely express a Location argument as dictated by the semantics of the verb zhan, more importantly it fills in an argument variable of the secondary predicate man. 45. Aissen (1975, cited in Bresnan 1994) takes the function of the locative inversion pattern as ‘to set a scene in which the locative phrase serves as a backdrop,’ i.e., it asserts the existence of a scene. Such comments can be understood in the present light.
Conceptual inference and predicate transfer
135
Furthermore, since the locative phrase is a NP, it seems to have a semantics close to the English place NPs (cf. Bresnan 1994 for a discussion of this point concerning English and Chicheŵa). It would thus not be surprising to find it being predicated and assigned an attribute. The following example, drawn from conversation, serves to make this point: (93) A. women dao shuxia qu ba we to tree-under go Part. “Let’s go to the place under the tree.” B. bu, shuxia zhan-manle ren Lit. “No, under the tree stood-full people.” “No, there are too many people under the tree.” In the above, shuxia means something like ‘the place under the tree,’ which is assigned a property in (B). Further support comes from passivized locative inversions such as the following (from Pan 1996): (94) zhuozishang bei John man-le shu table-top BEI put full-LE book “John put books all over the table.” Given the view of the conceptual structure of Chinese passives discussed in the last section, the phrase that begins with bei must semantically predicate over its locative subject Thus the sentence expresses something like ‘the table-top underwent John’s putting books all over it’ The fact that the preverbal locative phrase can serve as the subject of a passive sentence in Chinese shows that it is a direct participant of the event expressed in a sentence, rather than an adjunct. This is further supported by the following fact. Sentence (94) can be related to a ba-variant, the Chinese disposal construction (Liu 1997; Wong 1999; Bender 2000): (94’) John ba zhuozishang fang man-le shu BA table-top put fullLE book “John covered the table with many books.” The morpheme ba is usually taken to introduce the object of the sentence. So it marks a direct argument Thus the location predication view provides a way to license the locative phrase as expressing a direct argument when it appears preverbally. In addition to the above empirical support for the predication relation between the preverbal locative phrase and the VP in locative inversion, there is theoretical bonus to be gained if we adopt the location predication view. First, the view would provide a tighter semantic connection to current syntactic analyses of locative inversion. In a series of works, Bresnan and Kanerva (Bresnan and Kanerva 1989, 1992; Bresnan 1994) have argued convincingly for the subject status of the preverbal locative phrase in English and Chicheŵa. Within the GB framework, Gu (1992) argues for the D-structure subject status of the locative phrase in Chinese. But this kind of syntactic analysis would entail a serious mismatch between syntax and semantics, as stressed by Bresnan (1994), in having a non-prominent Location argument mapped onto the subject. The Location predication view avoids this difficulty by suggesting a different predication relation not based on the verb itself, which means that the subject no longer expresses a Location role (w.r.t. the
Enriched composition and inference in the argument structure of Chinese
136
verbal semantics), but some role46 bearing a somewhat abstract property which naturally maps onto the subject. Second, coming back to the aspectual restriction in (91), we are now in a position to see its rationale. Locative inversion must express a stative situation because that is the only way to predicate over a Place argument, that is, by assigning the Place a property: a Place itself cannot be a direct participant of any event. I schematize the location predication view that is being argued for so far as the following conceptual structure for the locative inversion construction: (95) [State BE ([Place], [Property SUCH ([State])] )] On this view, the locative phrase that denotes a Place no longer serves a Location role, but is being predicated by a Property, which consists in a State. However, even though we have rid the locative phrase of the Location role, there is an obvious connection between the preverbal locative phrase and the Location role expected by the verbal semantics in locative inversion, which Bresnan captures with her argument structure restriction, as mentioned earlier. The question is how do we capture this connection in the present approach? In view of the aspectual differences between locative inversion and canonical patterns noted earlier, I claim that the Location is a semantic role associated with the canonical pattern, which changes into a different role in the locative inversion pattern. Following the strategy adopted in this chapter, I suggest that the two patterns be related by conceptual inference (not by syntactic transformation). In other words, there is an inference rule postulated as follows: (96) [Event f({[Thingl ]}, [Thing2 Theme], [Place ([State f([Thing2]) ])])]
])] <---> [State BE ([Place], [Property SUCH
In the above rule, the semantics of locative inversion can be roughly glossed as ‘the place has the property of being in a certain state.’ It is an output inferred from the semantics of a canonical pattern47, indicated to the left of the bidirectional arrow, where the braces indicate an optional argument, thus accommodating the occurrences of transitive verbs in locative inversion (e.g. 86d, 86e). The Place argument in the canonical pattern serves the role of Location for the event, a role not assumed by the same Place argument in the locative inversion pattern to the right of the arrow because the Place argument is in a different position in CS. The same Thing2 argument serves the Theme role on both sides of the rule. Again, a déjà vu in this rule is that the predicate semantics of the locative inversion construction is taken to be an extension of the predicate semantics in the canonical pattern along the lines of Nunberg (1995), which has been reinterpreted as conceptual inferences in this chapter. On the right of the arrow, the predicate is not the original function in the input ƒ, but is the constituent labeled Property containing the function ƒ. 46. For the sake of convenience, I continue to label the locative phrase as having a Location role, as in our phrase ‘location predication view,’ though this is not exactly correct on the present view. 47. I assume the function SUCH is a free rider accounted for by the inferential process, and need not be expressed syntactically.
Conceptual inference and predicate transfer
137
On this conception, Bresnan’s argument structure restriction on the locative inversion verb (i.e., ) is subsumed by the CS configuration in the left input to the rule in (96). As will be seen, I am proposing a wider set of restrictions (to be discussed in the next section), covering both the inferential relations associated with the locative inversion construction and the conditions on the construction itself. This is necessary to accommodate locative inversion data that appear to feature different verb types, rather than just unaccusatives (cf. Levin and Rappaport 1995; Nakajima 2001). As noted at the beginning of this chapter, locative inversion constructions have often been characterized as having a presentative function (see e.g. Bresnan 1994) or being concerned with the existence of the entity expressed by the postverbal NP (see Levin and Rappaport 1995 for discussion). On the present view, the existential meaning is not asserted by the locative inversion pattern itself, but rather is another possible inference from the left of the rule (96). In other words, the input in (96) (call it an event in a spatial field, see e.g. Jackendoff 1983) will necessarily entail that the relevant Theme (i.e. Thing2) will end up existing in the place involved: (97) [Event f({[Thing1
]}, [Thing2 Theme], [Place ])]--->[State BE ([Thing2 ], [Place ])]
By virtue of the inferential relation expressed in (96), the existential meaning as the output of (97) gets attached to the locative inversion construction. In other words, both the existential meaning in the output of (97) and the locative pattern on the right of the arrow in (96) are associated with the same input CS. The existential meaning may be made more prominent when the output of (96) interacts with something like a ‘focus tier’ (Jackendoff 2000), resulting in the Theme being focused, as proposed by Bresnan (1994). Given that the existential meaning itself associated with locative inversion is part of a conceptual inference process, there would be no need to posit any abstract syntactic element expressing this meaning and serving to license the locative subject. Lin (1999) however does exactly this, proposing an abstract light verb EXIST, which selects a VP, the head of which moves to the light verb in a syntactic operation. The proposed light verb EXIST is formally close to the function BE in the locative pattern in (95), but its content is too specific. Similar approaches like this have been reviewed throughout this book. They tend to demonstrate how a syntactic approach solves problems of meaning without adequate regard to how semantics works. The rule proposed in (96) shows that the locative subject is licensed in a subtle way involving conceptual inference, rather than a postulated syntactic element. Another point about the inference rule in (96) is worth mentioning. Since the upshot of the inference rule in (96) is that the predicate in locative inversion is an extension of the predicate in the canonical pattern, a case of predicate transfer in the sense of Nunberg (1995), the output of the transfer must also observe the Noteworthiness Condition introduced by Nunberg (see section 4.2). In other words, the predicate in the locative inversion construction must serve to characterize the locative subject in a contextually useful way. This provides a principled explanation of the contrast noted by Aissen (1975, cited in Bresnan 1994): (98) a. *On the wall never hung a picture of U.S. Grant. b. On the wall hangs not a picture of U.S. Grant but one of Jefferson Davis.
Enriched composition and inference in the argument structure of Chinese
138
The generalization from this is that locative inversion in English does not allow sentential negation though phrasal negation is fine. Larry Horn (p.c.) points out that in these cases phrasal negation with sentential scope is fine too, witness: On the wall hangs nothing at all. It seems to me that this case is not much different from (98b): the sentence still predicates an impacting action or an associated state, rather than an inaction, over the locative subject. Note that the equivalent example in Chinese is ambiguous between the two kinds of negation: (98) c. qiangshang meiyou guazhe Zhangsan de zaopian wall-on not hang Mod. photo “On the wall hangs not (or: *does not hang) Zhangsan’s photo.” Again on the sentential negation reading, (98c) is bad as well. However, the conceptual equivalent to (98) can be expressed in a canonical pattern in both languages, along the lines of someone did not hang a picture of X on the wall. In these circumstances, the negative event (or inaction) associated with the Theme does not have any impact on the locative subject; it is not noteworthy at all. Hence such sentences are ruled out by the pragmatic condition. Such an account provides some theoretical backup to Aissen’s (1975) somewhat intuitive comment that the assertion of the non-existence of a scene contradicts the scene-setting function of the construction (cf. note 45). 5.2. Verbs in Locative Inversion Restriction on verb types in locative inversion is one of the central topics in the literature, especially the question whether or not the construction is limited to unaccusative verbs (see for example discussions by Levin and Rappaport 1995, Bresnan 1994, Nakajima 2001). The data from Chinese suggest that locative inversion constructions cannot be restricted by a formally-defined verb type, since a transitive activity verb appears to occur freely in the construction (e.g. 86d, 86e, see also Lin 1999), in addition to intransitive verbs. From the present perspective, the locative inversion construction should be conditioned by the conceptual event structure expressed by predicate expressions (i.e., verb types alone are not sufficient). The basic restriction provided by the rule in (95) on event structure is that the predicate expression of locative inversion patterns must designate a State, as discussed above. As also discussed in the last section, this restriction is motivated by the nature of the Place argument being the external argument that is to be predicated over. This condition on conceptual event structure is very close in spirit to, but I believe better motivated than, the proposal of Nakajima (2001). Working from the perspective of Generative Lexicon (Pustejovsky 1995), Nakajima (2001:46) restricts the verb type in locative inversion in terms of what he calls a “state subevent condition”: (99) Verbs can occur in locative inversion constructions only if their lexical representations (or those of the verb phrases containing them) involve the headed final subevent structure designating a state, the quale for which means that something is AT some place. The condition above relaxes the usual verb-type oriented restrictions on the construction by making reference to the verb phrase, and invoking the notion of headed subevent.
Conceptual inference and predicate transfer
139
Nakajima shows how this condition accounts for unaccusative as well as unergative verbs in the construction through co-composition in Generative Lexicon. This means that elements other than verbs are also crucial in satisfying the restriction. In addition, the condition proposed by Nakajima shares with my proposal the emphasis on the State requirement, but it remains something of a riddle on his account as to why the locative inversion construction must express a State subevent whereas a close conceptual paraphrase can express an Event, as discussed earlier. Let us now see how the requirement for State predicts the distribution of verb types in Chinese locative inversion constructions. In other words, the conceptual structure of verbs must not contradict the conceptual structure of the locative inversion construction repeated below: (100) [State BE ([Place], [Property SUCH ([State f ([Thing2 ]) ])])] In particular, the verbal conceptual structure must unify with the function variable ƒ in (100). First of all, the distribution of verbal particles in the construction is accounted for. As can be seen in the examples in this section, locative inversion in Chinese almost always needs a particle attached to the verb,48 typically le, guo, zhe, all being related to aspect or event structure in some way. For instance, the experiential particle guo presents a situation as stative (Smith 1994, 1997), though it also has other temporal meanings. Thus an accomplishment verb attached with it can appear in locative inversion: (101) zhetiao huli faxianguo guai shou thisCL lake-inside discoverGUO strange animal “In this lake was (once) discovered a monster.” The sentence would be bad without the particle guo, showing that the verb itself cannot unify with f in (100). The particle in this case serves to profile the state component of an accomplishment event. The other two particles have related uses. As proposed in (42) earlier, the so-called perfective particle le could map an Event into a State, thus making a combination like fang-le ‘be placed’ in (86d) able to unify with ƒ and express a State. So (86d) has a meaning something like ‘the inside of the room has the property of being placed many pieces of luggage.’ The particle zhe is compatible only with stative Stage-level predicates in the sense of Carlson (1977), as noted earlier (see section 4.3). For a complex event comprising an event and a resultative state, zhe serves as an operator that changes the event profile or headedness (Pustejovsky 1988, 1995; see discussion in chapter 2 above) to the State component,49 as is the case with guazhe ‘be hung’ in (86e), which as a whole 48. An exception is found in (86f) which features a resultative compound verb paochulai ‘run-outcome’ without any particle, but in this case the relevant State element is explicitly expressed by the resulting state element of the compound and is lexically profiled. 49. As noted earlier, Pan (1996) takes the verbal particle zhe as inducing the morphological operation that deletes the Agent of the verb in locative inversion, observing that the zhe locative inversion does not allow the presence of the Agent, but the le kind does, as in:
(i) zhuozishang John fang-le/*-zhe yi ben shu table-top put LE/ZHE oneCL book ‘On the table John put a book.’
Enriched composition and inference in the argument structure of Chinese
140
However, this is not as conclusive as Pan claims, cf.
unifies with (100) without a problem. There are cases where the locative inverting verb can be affixed with either le or zhe: (ii) luzishang John zhuzhe yi guo tang stove-top cookZHE one pot soup ‘On the stove John was cooking a pot of soup.’ The analysis of zhe as related to event headedness here avoids this problem. In (i), the State result of ‘putting’ is ‘book being on the table,’ which does not involve the Agent of the ‘putting’ action, whereas the focused State element of a cooking event in (ii) is ‘John being in the State of cooking,’ still retaining the original Agent. (102) a. chuang-shang tang-zhe/-le yige bingren bed-top lieZHE/LE one patient “In the bed lies a patient.” b. fangjian-li fang-le/zhe xueduo xingli room-inside putLE/ZHE many luggage “In the room are placed many pieces of luggage.” However, the verb zhuang ‘bump’ in (86a) can only be affixed with le, rather than zhe. The question then is what decides the choice of the two particles. I suggest that zhe is used only when the LCS of the verb entails a resultative (stage-level) State. For instance, the verbs tang ‘lie’ and fang ‘put’ entail something be at some place as a result of the action, but the verb zhuang ‘bump’ does not have such an entailment. Thus zhe and le in (102) each operates on a different element in verbal semantics, a stative one and an eventive one respectively, yielding a State either way. Secondly, the prevalence of aspectually-related particles in Chinese explains the more liberal distribution of verb types in Chinese locative inversion than for example in English. In addition to unaccusative verbs50, as is often required in English (Bresnan and Kanerva 1989, Bresnan 1994, cf. Levin and Rappaport 1995), transitive and unergative verbs (e.g. (86f)) are allowed in Chinese locative inversion as well, as shown in the examples of this section (see also (104) below), provided that they compose semantically with other elements such as aspectual particles and resultative predicates to yield a State. Thus the proper generalization should be on the level of conceptual structure. Thirdly, note that the inference rule in (96), as repeated below, is bi-directional: (96) [Event f({[Thing1 ]}, [Thing2 Theme ], [Place ])] <---> [State BE ([Place], [Property SUCH ([State f([Thing2 ]) ])])] 50. Given the view of the passive developed in the last section as well as that in Huang (to appear), there is no equivalent to the passivized verb itself as a category in Chinese, which is used in English locative inversion. Concomitantly, there is nothing like a by-phrase restriction (Bresnan 1994) found in locative inversion in English (see note 42).
Conceptual inference and predicate transfer
141
This means that the lexical conceptual structure of the verb that unifies with the function variable ƒ in the output of the rule must at the same time satisfy the conceptual structure requirements imposed from the input of the rule. In other words, the LCS of the verb must define a Theme and a Location (expressed by the Place element), as stressed by Bresnan (1994). Given a locative inversion sentence, we should be able to infer a conceptual structure conforming to the input of the rule in (96). However, as Lin (1999) points out, many Chinese locative inverting verbs do not contain a Location role in conceptual structure, in contrast to the argument structure condition proposed by Bresnan (1994). The contrast with English is underscored by the following pair (Lin 1999): (103) a. *In the oven baked an apple pie. b. kaoxiangli kaozhe pinguo bin oven-inside bakeZHE apple pie What is interesting about the pair is that the verb ‘bake’ does not have an internal Location argument, which accounts for the problem in (103a), as predicted by the rule in (96) and by Bresnan’s argument structure condition. However, the rule as such does not predict the grammaticality of (103b). In fact, Lin (1999:4) goes so far as claiming that “all kinds of active verbs (without an internal Location argument) can take locative subjects,” with the following examples (the glosses are mine) as support: (104) a. chufangli
dunzhe
yiguo niu-rou
kitchen-inside stewZHE one-pot beef Lit. “In the kitchen is stewing a pot of beef.” b. shan-shang gaile henduo fangzi mountain-top buildLE many house Lit. “In the mountain built many houses.” c. zhezuo huli faxianguo baozang thisCL lake-inside discoverGUO treasure Lit. “In this lake have discovered treasure.” All of the transitive verbs of different aspectual types in the above do not subcategorize for a Location argument, and therefore are not predicted by the rule in (96). The Place element in the above serves as a locative adjunct to the eventuality expressed, rather than playing a participant role in the event semantics, This is evident from their canonical counterparts: (105) a. ta zai chufangli dun yiguo niurou he Loc. kitchen-inside stew one-pot beef “He is stewing a pot of beef in the kitchen.” b. tamen zai shan-shang gai-le hengduo fangzi they Loc. mountain-top buildLE many house “They built many houses in the mountain.” c. tamen zai zhezuo huli faxianguo baozang they Loc. thisCL lake discoverGUO treasure “They discovered treature in this lake.
Enriched composition and inference in the argument structure of Chinese
142
The Place elements in the above are all italicized, and they are optional in the structure and semantics of the sentences. Thus it seems necessary to relax the conditions imposed from the input of the rule in (96), in particular the requirement for a Location argument. Note that the canonical examples in (105) are all characterized by a common entailment: the internal argument of the verb is located in the space designated by the adjunct. For (105a), for example, the sentence entails that ‘a pot of beef is in the kitchen.’ What is important is that this inference conforms in its CS to the output of the rule postulated in (97), which is repeated below: (97) [Event f({[Thing1 ]}, [Thing2 Theme], [Place])]--->[State BE ([Thing2 ], [Place ])] that is, ‘something is in some place.’ In other words, the sentences with a Place adjunct and those with a Location argument share a common sort of entailment. This suggests the possibility to supplement the rule in (96) with the following informally stated condition: (106) If a sentence entails that the internal argument is located in the space designated by the adjunct, it entails the semantics of locative inversion constructions as stated in (95). (95) [State BE ([Place ], [Property SUCH ([State ])] )] Now, the canonical examples in (105) all satisfy the condition in (106). So it is possible to infer from them the conceptual structures of locative inversion, which are expressed in (104). Under the present conception, the differences between locative argument and locative adjunct, as invoked in Bresnan (1994), do not play a role in the conditioning of locative inversion. What is important is whether the relevant inference stated in (106) obtains. This approach easily accounts for examples such as (107) below that do not feature a argument structure, as discussed by Bresnan (1994): (107) Through the window on the second story was shooting a sniper. The logic at work is roughly as follows. If somebody is shooting on the second story, it entails that he is on the second story. Thus, by (106), locative inversion semantics will obtain.51 The present inference-oriented analysis has the same effect as, but perhaps works more explicitly than, Bresnan’s (1994) overlaying approach, according to which a canonical argument structure is contextually ‘coerced’ into a presentational argument structure: an Agent changes into a Theme, Finally, let us see how certain problematic examples are ruled out under the present theory: 51. Conceptually, ‘through the window’ expresses a Path constituent, which differs from a Place element. I don’t go into the question of whether the preverbal Path phrase functions the same as a locative phrase.
Conceptual inference and predicate transfer
143
(108) a. *On the freeway drive many cars. b. *lushang kaizhe xuduo che road-top driveZHE many car c. tianshang feizhe xuduo niao sky-top flyZHE many bird “In the sky are flying many birds.” d. *jiaoshili xiezhe yipian wenzhang classroom-inside writeZHE oneCL article Intended: “In the classroom is being written an article.” Lin (1999) claims that the equivalent of (108a), which is (108b), is acceptable in Chinese, but it is unacceptable according to my informant’s judgement as well as my own. Both (108a) and (108b) are ruled out for the same reason: their canonical counterparts, i.e., ‘people drive the cars,’ do not strictly entail the stative situation that the cars are ON the road. Put differently, the adjunct ‘on the road’ does not predicate over ‘the cars,’ but specifies the location of the event Thus by (106) the locative inversion semantics cannot obtain. They contrast with (108c), which entails that ‘many birds are in the sky,’ at the same time they are flying. What about the problem in (93d)? The sentence meets the stative requirement for locative inversion, the inferential requirement that the product of the writing event is strictly located in the space expressed by the phrase ‘in the classroom.’ Nevertheless it fails on grounds of noteworthiness in Nunberg’s (1995) sense. The state of an article being written somewhere more relevantly characterizes things like blackboard, desks, notebooks, etc, which are all located in the classroom, than the classroom itself. To sum up, the verb types that are allowed in locative inversion constructions are constrained both by the need for the State element in locative inversion semantics and by the proper CS configurations that can yield by inference rules the conceptual structures of locative inversion constructions. 5.3. The Swarm-Alternation and Predicate Transfer Before closing this section, it is worth noting certain parallels between the Chinese locative inversion construction and the English intransitive swarm-alternation with with, exetnplified by the (b) sentences in the following (from Dowty 2000a): (109) a. Bees swarm in the garden. b. The garden swarms with bees. (110) a. Snails are crawling in the garden. b. The garden is crawling with snails. (111) a. Fish abound in the pond. b. The pond abounds with fish. (112) a. Music resounded in the hall. b. The hall resounded with music. (113) a. Garlic reeked on his breath.
Enriched composition and inference in the argument structure of Chinese
144
b. His breath reeked with garlic. Such alternations have been noted as early as in Jespersen (1933), discussed more recently by Salkoff (1983), and most recently by Dowty (2000a, 2000b) (see Levin 1993 for other references to the topic), which I briefly discuss here. As noted by Dowty (2000a), the verb in the (b) sentences expressess an atelic activity or stative situation. What is of interest to us is that the subject of the swarm-alternation does not bear a straightforward relation to the situation expressed by the verb, and, in addition, the subject is in a sense locative, being the Reference Object in the Location argument of the (a) sentences. In fact, the Chinese counterpart to the swarm-alternation is exactly of the Chinese locative inversion form, for instance: (114) a. huayuanli pazhe xuduo woniu garden-inside crawlZHE many snail “The garden is crawling with many sails.” b. *huayuan pazhe xuduo woniu garden crawlZHEmany snail The contrast above shows that the subject of a swarm-alternation must be explicitly a locative phrase in Chinese (i.e., with the locative suffix –li), though it is not so in English.52 Note also that the event type of the sentence (not just the verb used) is stative in each case, even where the activity verb is used. Consider the contrast in the following: (115) a. The garden is crawling with snails b. *The garden is crawling slowly with snails. c. Snails are crawling slowly in the garden. The fact that (115b) does not allow the manner adverbial shows that the sentence expresses a stative situation, obeying the aspectual restriction on locative inversion in (91). Given the obvious conceptual relatednesss between the variants in (109–113), the approach developed in this section carries over to this set of examples naturally. The central claim would then be that there is a predication relation between the subject and the predicate, such as the garden and swarm with bees in (109b). And the predicate in (109b) can be taken as a Nunbergian extension from that in (109a).53 Such an analysis is essentially the upshot of Dowty’s analysis as well, as can be seen from the following central claim of Dowty’s: 52. It is an open issue whether the subject of the English swarm-alternation expresses a Location/Space. If Location/Space must consist of a locative function with its Thing argument, according to Jackendoff (1990, see also Bresnan 1994 for discussion of this point concerning Chicheŵa), the subject of the English examples may simply be a Thing (see esp. 98b). However, this issue does not necessarily affect the general point being made by Dowty. 53. Dowty (2000a) draws parallels between the L-subject form and ‘container metonymy’ sentences:
Conceptual inference and predicate transfer
145
(i) a. The water is boiling in the teakettle. b. The teakettle is boiling. The LS-form (i.e., the location-subject form) ascribes an abstract property (expressed by the predicate) to a Location (denoted by the subject NP): the property a place or space has when it is ‘characterized’ by an activity taking place within it (Dowty 2000a: 122) This characterization is very much in the same spirit as the CS-representation for the locative inversion construction proposed in (95). Dowty (2000a, b) in addition elaborates on the nature of the Property being attributed to the Location subject as some kind of DYNAMIC TEXTURE, building on a suggestion attributed to Ray Jackendoff in a personal communication. The dynamic texture is, roughly, some pattern of movement on a surface, just as shininess, flatness, etc. are the static textures distributed over most or all sub-portions of a surface, recognizable by vision or other senses. Thus it appears that the approach to locative inversion developed in this section and Dowty-Jackendoff s texture analyses of the swarm-with construction nicely resonate with each other. Nonetheless, there are some subtle but important differences between them concerning the source of the Property being attributed to a locative subject in the two kinds of constructions. The representation proposed for the locative inversion construction shows that the pattern is consistently associated with a spatial event with its specific existential inference (see (97)). The needed Property derives from the resultative State associated with the event. By contrast, for the swarm-with construction, the Property comes from the event itself, including its extent, intensity, frequency and perceptual salience, which can categorize a Location in a relevant way, as Dowty (2000a) claims. 5.4. Summary In this section, I have once again demonstrated the explanatory power of predicate transfer as conceptual inferences by examining locative inversion constructions in Chinese. The goals focus on the conceptual structure representation of the construction and the restrictions on the distribution of locative inverting verbs. Semantically, the locative inversion construction is argued to differ from its canonical paraphrases in expressing a State. This aspectual requirement is accounted for by proposing a conceptual predication relation between the locative subject and the VP in the construction. Empirical support is provided for the proposal, which is argued to make possible a tighter syntax/semantics connection. Based on these arguments, the CS representation Dowty argues that the case in (b) is a case of predicate transfer, in the Nunbergian sense. He suggests that the LS-sentences may be treated in the same way. of the locative inversion construction is proposed, which is linked to its canonical paraphrase through a conceptual inference rule (96). In other words, the predicate in locative inversion is taken to be extended from that in its canonical counterpart. Thus the
Enriched composition and inference in the argument structure of Chinese
146
semantics of locative inversion is subject to the Noteworthiness Condition on predicate transfer (Nunberg 1995). On this conception, the so-called existential meaning is argued to follow from an additional inference from the CS of the canonical paraphrase, and there is no need to posit an empty syntactic head EXIST. As for the distribution of locative inverting verbs, they are subject to the State requirement imposed by the CS of the locative construction, thus accounting for the aspectually-related particles zhe, le and guo that all serve to present a State situation, depending on the LCS of the verb. Furthermore, the verbs must also either allow a Place argument in conceptual structure or entail that the internal/only argument is located in the Place expressed by an adjunct. Finally, the independent proposals by Dowty (2000a, b) strengthen the present approach, suggesting that I may have hit on a generalization covering diverse phenomena.
6. Conclusion In this chapter, I have been focusing on the semantic relations between the grammatical subject in Chinese and the semantics of predicate expressions, including the crucial contributions made by the so-called aspectual particles le, zhe, and guo. In particular, I have explored in some detail the ‘unselected subject’ as shown in the three types of constructions, the Possessum-object construction, the retained-object passive and the locative inversion construction. The general conclusion from the study in this chapter (including the related studies by Dowty 2000a, b) is that the ‘unselected subject’ in all these constructions is licensed by the mechanism of predicate transfer in the sense of Nunberg (1995), which has been reinterpreted in this chapter in conceptual-semantic terms as falling into the Inference Rules component. The machinery needed is independently motivated within the framework of conceptual semantics, and thus there is no need to posit ad hoc syntactic devices such as empty heads and movement. Furthermore, as all three constructions are argued to exhibit systematic correspondences between two related predicates, such as ‘die (father)’ and ‘father dead (son),’ the way the subject is licensed is necessarily different from the enriched composition mechanisms employed in chapter 3. Particular issues discussed with respect to the three constructions include the conceptual structure representation of the construction, certain inference rules involved that license the construction and its alternant, the way external possessive relations are represented, and the conditions imposed on the distribution of the verb and other predicate expressions. To see any possible interconnections between them, I summarize the major findings (with gross simplifications of the Noteworthiness Condition) in Table 1.
Conceptual inference and predicate transfer
Construct -on Syntax Possessum Object
Conceptual Structure Inference Rules
NPposs V NPum
147
CS Condition on verb type State or inchoative Event, with Property element Not discussed
Set inclusion inference [State y BE AT [Property··x of y BE AT [Property]]] Retained-obj. NP bei [IP [REACT+( [ ]α, [Event AFF<->REACT Set Passive NP V) AFF¯ ([ ], [α ])])] inclusion Inference from a spatial State Locative [NP N-loc.] [State Place be at event, with a Place adjunct Inversion V-particle [Property State]] or Location argument NP
Table 1. Constructions and Predicate Transfer Taking a ‘surface’ view of the syntax, as is customary in unification-based theories, the three constructions can be said to share a Subject-Predicate structure, but their conceptual structures seem less uniform. The Possessum-object construction and the locative inversion construction can be grouped together for their similar conceptual structures: an entity being characterized by a state, but they involve different inference rules. On the other hand, the conceptual structure of the retained-object construction stands out from the rest with its distinct action tier configuration, but it involves the set inclusion inference as well, a feature of the Possessum-object construction. The variations in the conceptual structures of the three constructions with similar syntactic structures raise doubt as to any uniform treatment, as in some constructionist accounts (C.-R. Huang et al. 1999; Zhang 1995), but support again the relative independence of SS and CS, as argued by Jackendoff (1996b, 1997a, 2000, see chapter 2). However, certain abstract element can be extracted from the three conceptual structures. For instance, they all contain a most prominent argument (Figure), which participate in the situation (state or event). One could pair this abstract feature with syntactic Subject-Predicate structures, yielding a most schematic construction. In other words, each construction can be said to inherit (see chapter 2 above) syntactic and CS features from the maximally schematic construction, with more idiosyncratic features contributed by particular lexical items. Therefore, it does not seem appropriate to consider the three conceptual structure representations above as the construction meaning of the three constructions,54 since a subject-predicate structure is not inherently associated with any of them. Finally, the semantic strategy adopted in this chapter crucially involves the use of inference rules and the principle of no synonymy (Goldberg 1995), which motivates the different CS-representations for the three constructions. This greatly reduces the apparent syntax-semantics mismatches as stressed by Bresnan (1994). 54. Something along this line is found in C.-R.Huang et al. (1999), which to my mind greatly proliferates the number of constructions stored in long-term memory, an undesirable result.
CHAPTER FIVE Concluding Remarks
1. Achievements This dissertation started out with laying out the empirical focus of the research, namely, the argument structure phenomena in Chinese involving non-transparent relations between the predicator (often a verb) of a sentence and its direct internal and external arguments. Issues raised by such phenomena are discussed, particularly with respect to the traditional verb-centered, Projectionist view of the relation between the Lexicon and the Syntax. The specific goal of this study was therefore to account for what I called noncanonical argument structure phenomena from an alternative perspective, a perspective that shares the spirit of construction- and unification-based grammatical theories. The idea of fundamental importance under this perspective is that there is no principled boundary between the lexicon and extra-lexical grammatical information, in contrast to the special role accorded to the lexicon under Projectionist approaches. A particular constructionist approach implementing this perspective is presented. Under this approach, which grows directly from work by Jackendoff (1997, 2000), a grammar is taken to consist of an inventory of constructions of varying schematicity, with semantic and syntactic information specified to varying extents. The semantic dimension of a construction is approached in terms of Conceptual Semantics (Jackendoff 1983, 1990, 2000), which is also the theoretical focus of this dissertation. The study pushes the non-canonical argument structure phenomena to the semantic side, focusing on how the ‘unselected’ external and internal arguments are licensed in terms of an enriched view of conceptual combination and conceptual inferences while leaving the syntax of the constructions involved as simple as possible. This view presupposes an enriched conceptual-semantic representation that covers crucial frame-semantic or qualia information. A feature shared by the analyses proposed is that, in each case, the conceptual structure of a relevant construction, whether it be the eat restaurant construction, the possessum-object construction, the retained-object passive, or the locative inversion construction, is motivated independently by the grammatical properties of the construction involved, rather than derived from a simple semantic composition of the constituents in a construction. A schematic conceptual structure representation of the construction is then proposed, which is taken to constrain how conceptual structures of particular constituents unify with the general meaning of the construction.
Concluding remarks
149
The conceptual structure of the construction instantiated by the eat restaurant kind of examples can be glossed schematically as ‘one entity undertakes an event by affecting another entity in some way.’ Three conditions governing the constituents in such constructions follow from this general conceptual structure. (1) The event expressed is functional (Pustejovsky 2001): it makes reference to an agentive quale, i.e., another event in which some entity x is affected by default, thereby bringing the functional event into existence. The functionality of the event is specified either in lexical semantics or through combination with a‘cognate’ event. (2) The unselected complement must express an entity m attached with a specific telic quale, i.e., m is affected for the purpose of the functional event in (1). (3) There should be no redundancy between x and m. Elaborate lexical-conceptual structures of relevant verbs and nouns are proposed, which are shown to abide by these conditions and compose with each other through coercion and cocomposition. Furthermore, a preliminary Preference Rule System of defeasible typicality conditions is developed on the basis of work by Lascarides and Copestake (1998). The system takes care of exceptions to the conceptual-semantic generalizations made. Both the Possessum-object construction and the locative inversion construction share a conceptual structure in which a Thing or a Place is predicated over by a State as its property, whereas the conceptual structure of a retained-object passive is defined in action-tier terms like this: ‘an entity z REACTS to an event involving x and z’s Possessum y, and z is thereby AFFECTED by that event either lexically or contextually.’ In other words, z must be construable as a Patient, by lexical semantics or by discourse context. It is shown that passives in Chinese and English. differ significantly with respect to their conceptual structure, in alignment with their distinct syntactic properties. The conceptual structures proposed for these three constructions are linked to those of their canonical alternations by inference rules. Two types of inferences are invoked, one involving set inclusion, and one involving a spatial event with a Place adjunct or a Location argument. The ‘unselected’ external argument is licensed by the relevant inference rules. The approach draws on Nunberg’s (1995) theory of predicate transfer, which is reconceived in conceptual-semantic terms as conceptual inference rules and a Noteworthiness Condition in the Inference Component. Thus the three apparently disparate constructions are given a unified treatment, which appears to be supported by the independent work of Dowty (2000a, b) on the English swarm-alternation. The conceptual structures of the possessum-object construction and the locative inversion construction, which are summarized in the last paragraph, impose constraints on the type of predicate expressions that are allowed to occur in such environments. In particular, the predicator in a possessum-object construction must express a State with a Property element. It is shown that the aspectual particles le, guo, and zhe are deployed to satisfy this State requirement. A similar requirement for a State element is suggested for locative inversion as well, which again accounts for the distribution of aspectual particles in this construction. In addition, Bresnan’s (1994) argument structure condition on locative inverting verbs—namely, they have an argument structure of the form —is relaxed and replaced by a general condition on the conceptual structure of the canonical counterpart, i.e., that the canonical counterpart entails that the internal argument is located in the space designated by either a Place adjunct or a Location argument
Enriched composition and inference in the argument structure of Chinese
150
2. Implications and Further Work The study has focused on four types of empirical phenomena that relate to different areas of meaning. Since the general approach of this work is construction-based, a natural question that arises is whether or not we are dealing with four separate constructions. Indeed, suggestions in this spirit have been found in the Construction Grammar literature (see e.g. C.-R.Huang et al 1999). Following Jackendoff (1996, 1997, 2000), I have adopted the notion of a defective construction according to which a syntactic structure may not be inherently associated with a concrete meaning (except perhaps a maximally generalized conceptual schema, in some cases). The ‘unselected’ object construction is a case in point, as a complement position is in this case found to be associated with a conceptual modifier, rather than expressing any thematic role of the predicator, as would normally be expected. Conversely, a particular conceptual schema may also find its expression in varying syntactic structures, a point Koenig (1999) makes as well. In other words, the syntactic and the semantic dimensions of a phrasal construction could in principle be separate, and their pairing together could be a function of a productive online process that incorporates lexical information. This of course does not preclude the cases where syntax and semantics are more tightly locked together, including patterns of systematic mapping, as studied in much construction-based work. This somewhat abstract view of the notion of construction offers a general organizational perspective on the results of this study. Instead of proposing four constructions corresponding to the four empirical areas discussed in this dissertation, one could suggest two schematic constructions with underspecfied semantic information: (1) <( ), (V-Complement), ( )> (2) <( ), (Subject1-VP), (Figure1…)> We may also assume the availability of certain independent conceptual schemas such as those involving predicating a Property over a Thing, those involving force dynamics (AFF and REACT) and those involving conceptual modification, which have all been independently motivated in conceptual and cognitive semantics (Jackendoff 1983, 1990; Talmy 2000). Thus specific meanings derive from unifying particular lexical constructions (with their LCS’s) with (1) and (2), as well as other relevant conceptual schemas. However, as noted in Chapter 3, such online construction of syntax-semantics pairings may become entrenched and thereby stored as fully specified grammatical constructions, a kind of redundancy that is allowed in usage-based models of grammatical representation (Bybee 1985; Langacker 1987). This dissertation, while embedded in a constructonist framework, has largely focused on processes of conceptual unification and inference, thus leaving to future work a more explicit and detailed treatment of the syntax-semantics pairing and relations between different types of constructions. The theory of Conceptual Semantics developed by Jackendoff was found to be useful in modeling certain exceptions to semantic generalizations, an issue studied in much work in computational lexical semantics. Based on related work, I developed a preference rule analysis of the exceptional cases. The preference rule system is another interesting area that is worth much further effort, since it is one of the less developed areas in conceptual semantics (cf. Jackendoff 1996). Another area with a similar status is the
Concluding remarks
151
Inference Component, which to my knowledge has not been studied in detail within conceptual semantics since Jackendoff (1976). The present study has pointed to its importance in regulating syntactic expression of conceptual structures. A third area that may require more attention is the representational system used in conceptual semantics, especially given the need to incorporate certain frame-semantic information into the representation. The rich LCS’s proposed in this study also raise questions for processing. For instance, given the rich LCS of chi ‘eat’ proposed in Chapter 3, with its Thing (‘thing eaten’) argument and ‘cognate’ event, which compete with each other for syntactic expression, how does a listener decide which part of the LCS of the verb is relevant in a particular instance, such that eat restaurant is not understood as ‘consuming restaurant’? It would seem that the meaning of its complement plays a crucial role in this case of lexical processing, and this is nicely correlated with the rich LCS’s proposed. Finally, let us return to the methodological goal of this dissertation set out in Chapter 1, i.e., to demonstrate the value of an approach to the structure of Chinese that accords semantics an independent status, in contrast to the majority of formal studies on Chinese that deal with certain semantic issues from an abstract syntactic perspective. While I am fully aware of certain important insights from the mainstream literature on Chinese syntax, the work reported in this study shows that much of the available literature on the topics dealt with here suffers from a lack of full appreciation of the semantic issues involved. For instance, the existential meaning associated with locative inversion constructions is coded by Lin (1999) as a null light verb EXIST, without argumentation as to whether or not the construction itself directly asserts such a meaning. The proposal in Chapter 4 is that such a meaning arises from an inference associated with the conceptual structure of the canonical counterpart of a locative inversion construction. Thus semantics may not be as flat and straightforward as certain syntacticians seem to think. Furthermore, the syntactic proposals reviewed cannot deal with a full range of data, whereas a parallel architecture of linguistic theory (Construction Grammar being one version) allows one to directly tackle semantic issues and offers an empirical edge with respect to the range of data. As mentioned in Chapter 3, the approach I adopted there does not seem to sit well with much work that syntacticizes verbal event semantics, sometimes called event syntax (e.g., van Hout 1998; Ritter and Rosen 1998; Travis 2000). A full comparison of this approach with mine awaits further study.
References
Ackerman, Farrell and Gert Webelhuth. 1998. A Theory of Predicates. Stanford: CSLI. Aissen, Judith, 1975. Presentational-there insertion: A cyclic root transformation. Chicago Linguistic Society 11:1–14. Allan, Keith. 2001. Natural Language Semantics. Oxford: Blackwell. Alsina, Alex, Joan Bresnan and Peter Sells. (eds.) 1997. Complex Predicates. Stanford: CSLI. Arad, Maya ,1996. A minimalist view of the syntax-lexical semantics interface. UCL Working Papers in Linguistics, 8:215–42. Bach, Emmon, 1986. The algebra of events. Linguistics and Philosophy 9:5–16. Baker, C.Lee. 1979. Syntactic theory and the projection problem. Linguistic Inquiry 10: 533–581. Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Baker, Mark, Kyle Johnson and lan Roberts. 1989. Passive arguments raised. Linguistic Inquiry 20:219–251. Barlow, Michael. 1988. A Situated Theory of Agreement. Ph.D. dissertation, Stanford University (Published by Garland Press, New York). Barwise, Jon and John Perry. 1983. Situations and Attitudes. Bradford Books. Cambridge, MA: MIT Press. Belletti, Adriana and Luigi Rizzi. 1988. Psych-verbs and θ-theory. Natural Language and Linguistic Theory 6:291–352. Bender, Emily. 2000. The syntax of Mandarin ba: reconsidering the verbal analysis. Journal of East Asian Linguistics 9:105–145. Bender, Emily and Dan Flickinger. 1999. Diachronic evidence for extended argument structure. In G.Bouma, E.Hinrichs, G.Kruijff and R.Oehrle (eds.). Constraints and Resources in Natural Language Syntax and Semantics, 3–19. Stanford: CSLI. Bickerton, Derek. 1991. Language and species. Chicago: University of Chicago Press. Bierwisch, Manfred. 1989. Focusing on dimensional adjectives: introductory remarks. In Manfred Bierwisch & Ewald Lang (eds.). Dimensional adjectives: Grammatical structure and conceptual interpretation 1–11. Berlin: Springer. Birner, Betty. 1992. The Discourse Function of Inversion in English. Ph.D.dissertation. Northwestern University Blutner, Reinhard. 2002. Lexical semantics and pragmatics. In Fritz Hamm and Thomas Ede Zimmermann (eds.). Linguistische Berichte, Sonderheft 10, 27–58. Borer, Hagit. 1994. The projection of arguments. In Functional Projections, University Of Massachusetts Occasional Papers 17:19–47. Elena Benedicto and Jeff Runner (eds.). ——2000. The grammar machine. To appear in A.Alexiadou and M.Everaert (eds.). The Unaccusativity Puzzle. New York: Oxford University Press. Bresnan, Joan. 1994. Locative inversion and the architecture of Universal Grammar. Language 70:72–131. ——2001. Lexical Functional Syntax. Oxford: Blackwell.
References
154
——and Ronald Kaplan. 1982. Introduction: grammar as mental representations of language. In J.Bresnan (ed.) The Mental Representation of Grammatical Relation, xvii–lii. Cambridge, MA.: MIT Press. ——and Jonni M.Kanerva. 1989. Locative inversion in Chicheŵa: A Case Study of Factorization in Grammar. Linguistic Inquiry 20, 1–50. ——and Jonni M.Kanerva 1992. The thematic hierarchy and locative inversion in UG: a reply to Schachter’s comments. In Syntax and Semantics 26:111–125. Briscoe, Edward. 1996. Computational lexical semantics in review. Ms. University of Cambridge. ——, Ann Copestake, and Valeria de Paiva. 1994. Inheritance, Defaults, and the Lexicon. Cambridge: Cambridge University Press. Busa, Federica 1999. The semantics of event-based nominals. In Saint-Dizier, Patrick (ed.). Predicative Forms in Natural Language and in Lexical Knowledge Bases. Kluwer, 349–374. Butler, Christopher. 1985. Systemic Linguistics: Theory and Applications. London: Batsford. Butt, Miriam and Wilhelm Geuder, eds. 1998. The Projection of Arguments. Stanford: CSLI. Bybee, Joan L. 1985 Morphology: An Inquiry into the Relation between Meaning and Form. Amsterdam: John Benjamins. Carlson, Greg. 1977. A unified analysis of the English bare plurals. Linguistics and Philosophy 1:413–457. Carnie, Andrew. 2000. Syntax. Oxford: Blackwell. Chappell, Hilary.1999. The double unaccusative construction in Sinitic languages. In Payne, Doris L. and Immanuel Barshi (eds.). 1999. Amsterdam: John Benjamins. Chen, Chaoran. 1995. Existential construction in Chinese: a Construction Grammar analysis. Masters Thesis, National Tsinghua University, Taiwan. Chierchia, Gennario. 1989. A semantics for unaccusatives and its syntactic consequences. Ms. Cornell University. Chomsky, Noam. 1957. Syntactic Structures. Cambridge, MA: MIT Press. ——1981. Lectures on Government and Binding. Dordrecht: Foris. ——1986. Knowledge of Langwge: Its Nature, Origin, and Use. New York: Praeger. ——1992. A minimalist program for linguistic theory. In K.Hale and S.J.Keyser (eds.). The view from Building 20, 1–52. Cambridge, MA.: MIT Press. ——1995. The Minimalist Program. Cambridge, Ma.: MIT Press. ——and Howard Lasnik. 1993. Principles and parameters theory. In J.Jacobs A., von Stechow and W.Sternfeld (eds.). Syntax: An International Handbook of Contemporary Research, pages 506– 569. Berlin: Walter de Gruyter. Coulson, Seana. 1997. Semantic Leaps: the Role of Frame-Shifting and Conceptual Blending in Meaning Construction. Ph.D.dissertation. University of California, San Diego. ——and Gilles Fauconnier. 1999. Fake guns and stone lions: conceptual blending and privative adjectives. In B.Fox, D.Jurafsky and L.Michaelis (eds.). Cognition and Function in Language, 143–158. Stanford: CSLI. Culicover, Peter W. 1988. Autonomy, predication and thematic relations. Syntax and Semantics, 21:37–60. ——1998. The minimalist impulse. In P.W.Culicover and L.McNally (eds.), 47–77. ——1999. Syntactic Nuts: Hard Cases, Syntactic Theory, and Language Acquisition. New York: Oxford University Press. ——and L.McNally (eds.). 1998. The Limits of Syntax. New York: Academic Press. ——and Wendy Wilkins. 1984. Locality in Linguistic Theory, New York: Academic Press. ——and Ray Jackendoff. 1995. Something else for the binding theory. Linguistic Inquiry 19:521– 581. ——and Ray Jackendoff. 1997. Semantic subordination despite syntactic coordination. Linguistic Inquiry 28:195–217. ——and Robert D.Levin. 2001. Stylistic inversion in English: a reconsideration. Natural Language & Linguistic Theory 19:283–310.
References
155
Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago: University of Chicago Press. ——1993. Case marking and the semantics of mental verbs. In J.Pustejovsky (eds.). 1993. ——1998. Event structure in argument linking. In Butt and Geuder (eds.), pp. 1–43. ——2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. Cruse, D.Alan. 2000. Aspects of the micro-structure of word meanings. In Ravin, Yael and Claudia Leacock (eds.), pp. 30–51. Davidse, Kristin. 1991. Categories of Experiential Grammar. Ph.D.dissertation, Katholieke Universiteit Leuven. Deacon, Terence. 1997. The Symbolic Species. London: Penguins. Depraetere, Ilse. 1995. On the necessity of distinguishing between (un)boundedness and (a)telicity. Linguistics and Philosophy 18:1–19. Dowty, David. 1979. Word Meaning and Montague Grammar. Reidel: Dordrecht. —— 1989. On the semantic content of the notion ‘thematic roles’. In G.Chierchia, B. Partee and R.Turner (eds.). Properties, Types and Meaning II, 69–129. Dordrecht: Kluwer. ——1991. Thematic proto-roles and argument selection. Language 67:547–619. ——2000a. ‘The garden swarms with bees’ and the fallacy of ‘argument alternations’. In Ravin, Yael and Claudia Leacock (eds.), pp. 111–128. ——2000b.The semantic asymmetry of ‘argument alternations’ (and why it matters). Presented at the conference Making Sense: From Lexeme to Discourse, Groningen. Du, Juliet Wai-hong. 1999. Locative inversion and temporal aspect in Chinese. Talk presented at Chicago Linguistic Society 35th Regional Meeting. Engdahl, Elisabet. 1998. Integrating pragmatics into the grammar. In L.Mereu (ed.). Boundaries of Morphology and Syntax. Amsterdam: John Benjamins. Ernst, Thomas. 1999. Universal grammar and the order of adjuncts in Mandarin Chinese. Presented at 10th North American Conference on Chinese Linguistics, Stanford University. Fauconnier, Gilles. 1985. Mental spaces: Aspects of Meaning Construction in Natural Language..Cambridge, MA: MIT Press. ——and Mark Turner. 1998. Conceptual integration networks. Cognitive Science 22.2, 133–187. Feng, Shengli. 1995. The passive construction in Chinese. Studies in Chinese Linguistics 1. 1–28. ——2000. xie maobi yu yunlu chufang de dongci binru (Xie maobi and prosodically triggered verb incorporation). Language Teaching and linguistic Studies, 1, 25–31. Fillmore, Charles. 1985. Frames and the semantics of understanding. Quaderni di semantica 6:222– 54. ——and B.T.Atkins. 1992. Towards a frame-based lexicon: the semantics of RISK and its neighbours. In A.Lehrer and E.F.Kittay (eds.).Frames, Fields and Contrasts. Hillsdale, NJ: Lawrence Erlbaum. ——and Paul Kay. 1999. Grammatical constructions and linguistic generalizations: the what’s x doing y? construction. Language 75:1–33. Foley, William and Robert Van Valin. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Fong, Vivienne and Christine Poulin. 1998. Locating linguistic variation in semantic templates. In Jean-Pierre Koenig (ed). 1998, 29–39. Gao, Qian. 1996. Against lexical rules: two case studies in Chinese. Ms. Ohio State University. Ghomeshi, Jila and Diane Massam. 1994. Lexical/syntactic relations without projection. Linguistic Analysis 24:175–217. ——, Ray Jackendoff, Nicole Rosen and Kevin Russell. To appear. Contrastive focus reduplication in English. Natural Language and Linguistic Theory. Gil, David. 2000. Syntactic categories, cross-linguistic variation and universal grammar In Vogel, Petra M. & Bernard Comrie (eds.). 2000. Approaches to the Typology of Word Classes, 173– 216. Berlin, Mouton de Gruyter.
References
156
Ginzburg, Jonathan and Ivan A.Sag. 2001. Interrogative Investigations: The Form, Meaning, and Use of English Interrogatives. CSLI Publications. Goldberg, Adele E. 1995 Constructions: A Construction Grammar Approach to Argument Structure. Chicago, University of Chicago Press. ——1997a. The relationships between verbs and constructions. In Verspoor, Marjolijn et al. (eds.) 1997. ——1997b. Construction Grammar. In K.Brown and J.Miller (eds.). Concise Encyclopedia of Syntactic Theories. London: Elsevier. ——1997c. Making one’s way through the data. In A.Alsina, J.Bresnan and Peter Sells (eds). ——1998a. Patterns of experience in patterns of language. In Tomasello, Michael, (ed.) 1998. ——1998b. Semantic principles of predication. In Koenig, Jean-Pierre (ed.) 1998. Green, Georgia. 2000. Modeling grammar growth: Universal Grammar without innate principles or parameters. Ms.: University of Illinois at Urbana-Champaign. Grimshaw, Jane 1990. Argument Structure. Cambridge: MIT Press. ——1993. Semantic structure and semantic content. Ms. Rutgers University. Gruber, Jeffery. 1965. Studies in Lexical Relations. Ph.D. dissertation, MIT. Gu, Yang. 1992. On the locative existential construction in Chinese. In D. Bates (ed.). The 10th West Coast Conference on Formal Linguistics. Stanford: The Stanford Linguistic Association, 183–195. ——1997. On defining causativity and the significance of discourse information. Text 17(4): 435– 455. Guo, Jimao. 1999. Shitan ‘fei shanghai’ deng bu jiwu dongci dai binyu xianxiang. (Observations on the object of intransitive verbs as with fly shanghai) Zhongguo Yuwen (Chinese Language and Writing) 5:337–346. Hale, K. and J.Keyser.1993. On argument structure and lexical expression of syntactic relations. In Hale and Keyser (eds.). A View from Building 20, A Festschrift for Sylvain Bromberger. Cambridge, MA: MIT Press. Halliday, M.A.K. 1967. Notes on transitivity and Theme in English. Journal of Linguistics 3.2:199–244. ——1985. Introduction to Functional Grammar. London: Edward Arnold. Harley, Heidi. 1995. Subjects, Events and Licensing. Ph.D.dissertation, MIT. ——and Rolf Noyer. 1999. State-of-the-Article: Distributed Morphology. Glot 4.4: 3–9. Hashimoto, Mantaro. 1969. Observations on the passive construction. Chi-Lin 5:59–71. He, Yuanjian. 1996. An Introduction to Government-Binding Theory in Chinese Syntax. Lewiston: E.Mellen Press. Hendriks, Petra and Helen de Hoop. 2001. Optimality theoretic semantics. Linguistics and Philosophy 24 (1): 1–32. Hoekstra, Teun. 1992. Aspect and theta theory. In I.M.Roca (ed.). Thematic Structure: Its role in grammar, 145–174. Berlin: Foris. Holmes, Jasper and Richard Hudson.1999. Constructions in Word Grammar. Ms. University College London. Hopper, Paul and Sandra Thompson. 1980. Transitivity in grammar and discourse. Language 56:251–299. Horita, Yuko. 1996. English cognate object constructions and their transitivity. English Linguistics 13:221–247. Huang, Chu-Ren, Li-ping Chang, Kathleen Ahrens and Chao-ran Chen. 1999. The interaction of lexical and constructional meanings. Chinese Languages and Linguistics, 5:413–438, Academia Sinica. Huang, C.-T.James. 1982. Logical Relations in Chinese and the Theory of Grammar. Ph.D.dissertation, MIT.
References
157
——1987. Existential sentences in Chinese and (in)definiteness. In Reuland, Eric J. and Alice G.B.ter Meulen (eds.). The Representation of (In)Definiteness, 226–253. Cambridge, MA: MIT Press. —— 1991. Complex predicates in control. In U.Lahiri, S.Iatridou, R.Larson and J. Higginbotham (eds.). Control and Grammar, 109–147. Dordrecht: Kluwer. ——1994. Verb movement and some syntax-semantics mismatches in Chinese. Chinese Languages and Linguistics 2:587–613. ——1997. On lexical structure and syntactic projection. In F.Tsao and H.S.Wang (eds.). Chinese languages and linguistics 2, 145–187. Taipei: Institute of History and Philology, Academia Sinica. ——(to appear). Chinese passives in comparative perspective. To appear in Tsing Hua Journal of Chinese Studies. Hudson, Richard. 1986/1995. Identifying the linguistic foundations for lexical research and dictionary design. In D.Walker, A.Zampolli and N.Calzolari (eds.). Automating the Lexicon: Research and practice in a Multilingual Environment. Oxford: Oxford University Press. pp. 21– 52. ——1998. Language as a cognitive network. Ms. University College London. Iwata, Seizi. 1996. Motion and extent: two sides of the same coin. Studia Linguistica 50.3, 256– 285. ——1999. On the status of an implicit arguments in middles. Journal of Linguistics. 35:527–553. Jackendoff, Ray. 1976. Toward an explanatory semantic representation. Linguistic Inquiry 5, 481– 506. —— 1983, Semantics and Cognition. Cambridge, MA: MIT Press. ——1987a. The status of thematic relations in linguistic theory. Linguistic Inquiry 18, 369–411. ——1987b. Consciousness and the Computational Mind. Cambridge, MA: MIT Press. ——1990. Semantic Structure. MIT Press, Cambridge, MA. ——1991. Parts and boundaries. Cognition 41:9–45. ——1992. Mme. Tussaud meets the binding theory. Natural Language & Linguistic Theory l0, 1– 31. ——1993. The role of conceptual structure in argument selection: A reply to Edmonds. Natural Language & Linguistic Theory 11:279–312. ——1994. Patterns in the Mind. New York: BasicBooks. ——1996a. Semantics and cognition. In Shalom Lappin (ed.). The Handbook of Semantic Theory, 539–559. Oxford: Blackwell. ——1996b. Conceptual semantics and cognitive semantics. Cognitive Linguistics, 7, 93–129. ——1996c. The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language & Linguistic Theory 14:305–354. ——1997a. The Architecture of Language Faculty. Cambridge, MA: MIT Press. ——1997b. Twistin’ the night away. Language 73:534–59. ——1999. Parallel constraint-based generative theories of language. Trends in Cognitive Sciences 3.10, 393–400. ——2000. Foundations of Language. Draft, Brandeis University. ——and Barbara 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.: Lawrence Erlbaum. Jaeggli, Osvaldo. 1986. Passive. Linguistic Inquiry 17, 587–622. Jespersen, Otto. 1933. The system of grammar. In Linguistica: Selected Papers in English, French and German, 304–345. Copenhagen: Levin and Munksgaard. Jones, Douglas A. ed. 1994. Verb Classes and Alternations in Bangala, German, English and Korean. Artificial Intelligence Laboratory, MIT. Kiparsky, Paul. 1997. Remarks on denominal verbs. In A, Alsina, J.Bresnan and P.Sells (eds.), 473–499.
References
158
Klein, Wolfgang, Ping Li and Henriette Hendriks. 2000. Aspect and assertion in Mandarin Chinese. Natural Language & Linguistic Theory 18:723–770. Koenig, Jean-Pierre. 1999a, Lexical Relations. Stanford: CSLI Publications. ——1999b. On a tué le president! The nature of passives and ultra-indefinites. In B. Fox, D.Jurafsky and L.Michaelis (eds.). Cognition and Function in Language, 235–251. Stanford: CSLI. ——and Daniel Jurafsky. 1994. Type underspecification and the on-line type construction in the lexicon. Proceedings of the 13th West Coast Conference on Formal Linguistics, pp. 270–285. Stanford: Stanford Linguistics Association. ——and Nuttanart Muansuwan. 2000. How to end without ever finishing: Thai semi-perfectivity. Ms. State University of New York at Buffalo. ——(ed.). 1998 Discourse and Cognition. Stanford: CSLI. Kuno, Susumo. 1987. Functional Syntax. Chicago: University of Chicago Press. Lakoff, George. 1987. Women, Fire and Dangerous Things: What Categories Reveal about the Mind. Chicago: University of Chicago Press. ——and John R.Ross. 1976. Why you can’t do so into the sink. In J.McCawley (ed.). Notes from the linguistic underground. Vol. 7 of Syntax and Semantics 101–111. New York: Academic Press. Langacker, Ronald W. 1984. Active zones. In C.Brugman & M.Macaulay (eds.). Proceedings of the Annual Meeting of the Berkeley Linguistics Society 10:172–188, reprinted in Langacker (1990). ——1987. Foundations of Cognitive Grammar. Vol. l. Stanford: Stanford University Press. ——1990. Concept, Image, and Symbol: The Cognitive Basis of Grammar. Berlin: Mouton de Gruyter. ——1991. Foundations of Cognitive Grammar. Vol. 2. Stanford: Stanford University Press. ——2001. Topic, subject, and possessor. In Hanne Gram Simonsen and Rolf Theil Endresen (eds.) A Cognitive Approach to the Verb: Morphological and Constructional Perspectives, 11–48. Berlin: Mouton de Gruyter, Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19:335–332. Lascarides, Alex and Ann Copestake. 1998. Pragmatics and word meaning. Journal of Linguistics 34.2, 387–414. ——and Nicholas Asher. 1993. Temporal interpretation, discourse relations and commonsense entailment Linguistics and Philosophy, 16.5, 437–493. Lefebvre, Claire. 1991. Take serial constructions in Fon. In C.Lefebvre (eds.). Serial verbs, 37–79. Amsterdam: Benjamins. Levin, Beth. 1993. English Verb Classes and Alternations. Chicago: University of Chicago Press. ——and Malka Rappaport Hovav. 1988. Lexical subordination. In Papers from the 24th Chicago Linguistics Society, 275–289. Chicago: University of Chicago. ——1996. From lexical semantics to argument realization. Ms. Northwestern University and Bar Ilan University. ——1995 Unaccusativity at the Syntax-Lexical Semantics Interface. Cambridge, Mass: MIT Press. Levinson, Stephen. 2000. Presumptive Meanings: the Theory of Generalized Conversational Implicature. Cambridge, MA: MIT Press. Li, Yen-hui Audrey. 1990. Order and Constituency in Mandarin Chinese. Dordrecht: Kluwer. Li, Shan 1994 Xiandan Huyu Bei-ziju Yanjiu (A Study of the bei-Construction in Modern Chinese). Beijing: Peking University Press. Lin, T.-H.Jonah. 1999. Unselectiveness of subject and object in Chinese and a typology of phrase structure. Paper presented at GLOW in Asia. Nagoya: Nanzan University. Liu, Feng-Hsi. 1997. An aspectual analysis of ba. Journal of East Asian Linguistics 6: 51–99. Ma, Qinzhu. 1992. Verb and Verbal Structure in Chinese. Beijing: Beijing Languages Institute Press. Marantz, Alex. 1984. On the Nature of Grammatical Relations. Cambridge, Ma.: MIT Press.
References
159
Massam, Diane. 1990. Cognate objects as thematic objects. Canadian Journal of Linguistics 35(2):161–190. Matsumoto, Masumi. 1996. The syntax and semantics of the cognate object construction. English Linguistics 13:199–220. Matsumoto, Yo. 1996. Complex Predicates in Japanese: A Syntactic and Semantic Study of the Notion ‘Word’: Stanford: CSLI. Matsumoto, Yoshiko. 1997. Noun Modifying Constructions in Japanese: A Frame-Semantic Approach. Amsterdam: John Benjamins. McElree, Brian, Matthew J.Traxler, Martin J.Pickering, Rachel E.Seely, Ray Jackendoff. 2001. Reading time evidence for enriched composition. Cognition 78, B17–B25. Meng, Zong, Huaide Zhen, Qinhai Meng and Wenlan Cai (eds.). 1987 Dongci Yongfa Cidian (A Dictionary of Verb Usage). Shanghai: Shanghai Lexicographic Publishers. Michaelis, Laura A. and Knud Lambrecht. 1996. Toward a construction-based theory of language functions: the case of nominal extraposition. Language 72:215–247. ——2000. Aspectual meaning as construction meaning. Proceedings of the BFG 2000 Conference Workshops, (eds.) Miriam Butt and Tracy Holloway King, CSLI. Mittwoch, Anita. 1997. Cognate objects as reflections of Davidsonian event arguments. In S.Rothstein (ed.). Events in Grammar. Dordrecht: Kluwer. Mohanan, Tara and Lionel Wee (eds.). 1999. Grammatical Semantics: Evidence for Structure in Meaning. Stanford: CSLI. ——and K.P.Mohanan. 1999. On representations in grammatical semantics. In T. Mohanan and L.Wee (eds) 23–76. Moltmann, Frederika. 1990. Nominal and clausal event predicates. In Papers from the 25th Regional Meeting of the Chicago Linguistics Society. Chicago: Chicago Linguistics Society. Moravcsik, Julius. 1975. Aitia as generative factor in Aristotle’s philosophy. Dialogue 14:622–636. Nakajima, Heizo. 2001. Verbs in locative constructions and the generative lexicon. The Linguistic Review 18:43–67. Napoli, Donna Jo. 1989. Predication Theory: A Case Study For Indexing Theory. Cambridge: Cambridge University Press. Newmeyer, Federick J. 1998. Language Form and Language Function. Cambridge, MA: MIT Press. Nikanne, Urpo. 1990. Zones and Tiers: A Study of Thematic Structure. Helsinki: Finnish Literature Society. ——1995. Action tier formation and argument linking. Studia Linguistica, 49:1–31. ——1997. On locative case adjuncts in Finnish. Nordic Journal of Linguistics 20: 155–78. Nunberg, Geoffrey. 1979. The non-uniqueness of semantic solutions: polysemy. Linguistics and Philosophy. 3.1:143–184. ——1995. Transfers of meaning. Journal of Semantics 12:109–132. Ono, Naoyuki. 1992. Instruments: a case study of the interface between syntax and lexical semantics. English Linguistics 9, 196–222. Pan, Haihua. 1996. Imperfective aspect zhe, agent deletion, and locative inversion in Mandarin Chinese. Natural Language & Linguistic Theory 14:409–432. Payne, Doris L. and Immanuel Barshi (eds.). 1999. External Possession. Amsterdam: John Benjamins. Pereltsvaig, Asya. 1998. A cross-linguistic study of cognate objects and predication of events. Ms. McGill University. Pesetsky, David. 1995. Zero Syntax-Experiences and Cascades. Cambridge, MA: MIT Press. Pham, Hoa. 1996. Cognate objects in Vietnamese transitive verbs. Ms. University of Toronto. Piñango, M.M., Edgar Zurif and Ray Jackendoff. 1999. Real-time processing implications of enriched composition at the syntax-semantics interface. Journal of Psycholinguistic Research 28:395–414. Pinker, Steven. 1989. Learnability and Cognition. Cambridge, MA: MIT Press.
References
160
Pollard, Carl and Ivan Sag. 1987. Information-based Syntax and Semantics. Stanford: CSLI Publications ——1994. Head Driven Phrase Structure Grammar. Chicago & Stanford: University of Chicago Press and CSLI Publications. Pustejovsky, James. 1991a. The Generative Lexicon. Computational Linguistics 17(4): 409–441. ——1991b. The syntax of event structure. Cognition 41:47–81. ——(ed). 1993. Semantics and the Lexicon. Dordrecht: Kluwer. ——1995a. The Generative Lexicon. Cambridge, MA: MIT Press. ——1995b. Linguistic constraints on type coercion. In Patrick Saint-Dizier and Evelyne Viegas (eds.) Computational Lexical Semantics. CUP. ——2000a. Lexical shadowing and argument closure. In Ravin, Yael and Claudia Leacock (eds.), pp. 68–90. ——2000b. Syntagmatic processes. To appear in D.Cruse (ed.). Handbook of Lexicology and Lexicography. Berlin: Mouton de Gruyter. ——2001. Type construction and the logic of concepts. In P.Bouillon and F.Busa (eds.). The Syntax of Word Meaning. Cambridge: Cambridge University Press. ——and Federica Busa. 1995. Unaccusativity and event composition. In Bertinetto, Pier Marco, V.Bianchi, J.Higginbotham, M.Squartini (eds.). Temporal Reference, Aspect and Actionality. Vol 1: Semantic and SyntacticPerspectives. Torino: Rosenberg & Sellier, pp. 159–177. ——and Pierrette Bouillon. 1995. Aspectual coercion and logical polysemy. Ms. Brandeis University and University of Geneva. Ramchand, Gillian C. 1998. Deconstructing the lexicon. In Butt and Geuder (eds.). Rappaport Hovav, M. and Beth Levin. 1998. Building verb meanings. In Butt and Geuder (eds.), pp. 97–134. Rappaport, Malka, Mary Laughren and Beth Levin. 1993. Levels of lexical representation. In J.Pustejovsky (ed.) 1993. Ravin, Yael and Claudia Leacock (eds.). 2000. Polysemy: Theoretical and Computational Approaches. Oxford University Press. Riehemann, Susanne. 1995. The HPSG formalism. Ms. Ritter, Elisabeth and Sara Rosen.1998. Delimiting events in syntax. In Butt and Geuder (eds.), pp. 135–164. Rochemont, Michael. 1986. Focus in Generative Grammar. Amsterdam: John Benjamins. Rosen, Sara Thomas. 1999. The syntactic representation of linguistic events. Glot 4.2, 3–10. Ross, John Robert. 1969. Constraints on Variables in Syntax. Ph.D. dissertation, MIT. Sadock, Jerrold. 1991. Autolexical Syntax. Chicago: University of Chicago Press. Sag, Ivan. 1997. English relative clause constructions. Journal of Linguistics 33, 431–483. Salkoff, Maurice.1983. Bees are swarming in the garden. Language 59.2:288–346. Searle, John. R. 1995. The Construction of Social Reality. New York: Free Press. Shen, Yang and Dingou Zheng 1995 Xiandai Hanyu Peijia Yufa Yanjiu (Studies in the Valency Grammar of Modern Chinese). Beijing: Peking University Press. Shen, Xiaolong. 1992. Yuwen de Chanshi (Interpretations of Language). Shenyang: Liaoning Education Press. Shi, Dingxu. 1994. The nature of Chinese emphatic sentences. Journal of East Asian Linguistics 3:81–100. ——2000. The flexibility of Chinese and the theory of syntax. Contemporary Linguistics 2.1,18– 26. Shi, Ziqiang. 1990. Decomposition of perfectivity and inchoativity and the meaning of the particle le in Mandarin Chinese. Journal of Chinese Linguistics 18, 95–123. Shieber, Stuart. 1986. An Introduction to Unification-Based Approaches to Grammar. Stanford: CSLI. ——and Yves Schabes. 1991. Generation and synchronous tree adjoining grammars. Journal of Computational Intelligence 7:220–228.
References
161
Smith, Carlota S. 1994. Aspectual viewpoint and situation type in Mandarin Chinese. Journal of East Asian Linguistics 3:107–146. ——1997. The Parameter of Aspect (2nd ed.). Dordrecht: Kluwer. De Swart, Henriëtte. 1998. Aspectual shift and coercion. Natural Language & Linguistic Theory 16 (2): 347–385. Sweetser, Eve. 1999. Compositionality and blending: semantic composition in a cognitively realistic framework. In Theo Janssen & Gisela Redeker eds. Cognitive Linguistics: Foundations, Scope, and Methodology. Mouton De Gruyter), Sybesma, Rint. 1997. Why Chinese verb-le is a resultative predicate. Journal of East Asian Linguistics 6:215–261. ——1999. The Mandarin VP. Dordrecht: Kluwer. Talmy, Leonard. 1985. Lexicalization patterns: semantic structure in lexical forms. In T. Shopen et al. eds. Language Typology and Syntactic Description, vol. 3, 57–149. New York: Cambridge University Press. ——1988. Force dynamics in language and cognition. Cogitive Science 12:49–100. ——2000. Toward a Cognitive Semantics. Cambridge, MA.: MIT Press. Tao, Hongyin. 2000. cong shi kan dongci lengyuan jiegoude dongtai tezhen (The dynamic property of argument structure: a view from the verb chi). Yuyan Yanjiu 3, 21–38.\ Taylor, John R. 2002. Cognitive Grammar. New York: Oxford University Press. Teng, Shou-hsin. 1975. A Semantic Study of Transitivity Relations in Chinese. University of California Publications in Linguistics 80 Tenny, Carol. 1994. Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. ——and James Pustejovsky. 2000. Events as Grammatical Objects: The Converging Perspectives of Lexical Semantics and Syntax. Stanford: CSLI. Tesnière, Lucien. 1959. Elements de Syntaxe Structurale. Paris: Klincksieck. Ting, Jen. 1998. Deriving the bei-construction in Mandarin Chinese. Journal of East Asian Linguistics 7:319–354. Tomasello, Michael. 2000. Do young children have adult syntactic competence? Cognition 74:209– 253. ——(ed.) 1998 The New Psychology of Language: Cognitive and Functional Approaches to Language Structure. Mahwah: Lawrence Erlbaum. Travis, Lisa. 2000. Event structure in syntax. In C.Tenny and J.Pustejovsky. (eds.) 2000, 145–186. Tsao, Fong-fu. 1990. Sentence and Clause-Structure in Chinese: A Functional Perspective. Taipei: Student Books Co. Van Hoek, Karen. 1997. Conceptual Structure and Anaphora. Chicago: University of Chicago Press. Van Hout, Angeliek. 1998. Event Semantics of Verb Frame Alternations. New York: Garland. Van Valin, Robert. 1990. Semantic parameters of split intransitivity. Language 66, 221–260. ——and Randy J.LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Vendler, Zeno. 1968. Linguistics in Philosophy. Ithaca: Cornell University Press. Verspoor, C.M. 1997. Contextually-Dependent Lexical Semantics. Ph.D.dissertation, Center for Cognitive Science, University of Edinburgh. Verspoor, Marjolijn, Kee Dong Lee and Eve Sweetser (eds.). 1997. Lexical and Syntactical Constructions and the Construction of Meaning, Amsterdam: John Benjamins. Wang, Chuan-chih. 2000. Delimitation: Evidence from Mandarin. Ph. D.dissertation, University of Kansas. Available at http://www.usc.edu/dept/LAS/ealc/chinluig/%20synformal.htm. Wang, Li. 1954. Zhongguo Xiandai Yufa (A Modern Chinese Grammar). Beijing: China Books Co. Wang, Zhanhua. 2000. ‘chi shitang’ de renzhi kaochao (A cognitive examination of chi shitang). Language Teaching and linguistic Studies, 2, 58–64. Washio, Ryuichi. 1989–1990. The Japanese passive. The Linguistic Review 6:227–263.
References
162
Wasow, Tom. 1985. Postcript. In P.Sells. Lectures on Contemporary Syntactic Theories, 193–205. Stanford: CSLI. Webelhuth, Gert, Jean-Pierre Koenig and Andreas Kathol. 1998. Introduction. In Webelhuth, Gert, Jean-Pierre Koenig and Andreas Kathol (eds). Lexical and Constructional Aspects of Linguistic Explanation. Standord: CSLI. Wierzbicka, Anna. 1988. The Semantics of Grammar. Amsterdam: John Benjamins. Wiese, Heike. 2000. Towards an integrated model of semantic and conceptual representations: a discussion of mental models. Ms. Humbolt University Berlin. Williams, Edwin. 1981. Argument structure and morphology. The Linguistic Review 1: 81–114. ——1984. Grammatical relations. Linguistic Inquiry 15:639–674. Wong, Geraint. 1999. Representing the semantics of BA in Mandarin. In T.Mohanan and L. Wee (eds.). 159–210. Wunderlich, Dieter. 1994. Models of lexical decomposition. In Edda Weigand and Franz Hundsnurscher (eds.). Lexical Structures and Language Use. Tubingen: Niemeyer, 169–183. ——1997. Cause and the structure of verbs. Linguistic Inquiry 28:27–68. ——2000. Predicate composition and argument extension as general options: a study in the interface of semantic and conceptual structure. In B.Stiebel and D.Wunderlich (eds.). Lexicon in Focus, 247–270. Berlin: Akademie Verlag. Xin, Fuyi 1996. Hanyu Yufa Xue (A Theory of Chinese Grammar), Changchun, .Northeastern Normal University Press. Xu, Jie. 1999. Some theoretical issues of the two types of Chinese ‘retained-object’ constructions. Contemporary Linguistics 1.1, 16–29. Xu, Tongqiang. 1997. Yuyan lun: yuyixing yuyan de jiegou yuanli he yanjiu fangfa (On language: the structural principles and research methodologies of semanticized languages). Changchun: Northeastern Normal University Press. Yang, Suying. 1995. Ba and bei constructions in Chinese. Journal of the Chinese Language Teachers Association 30.3:1–36. Yuan, Yuling and Rui Guo (ed.) Xiandai Hanyu Peijia Yufa Yanjiu (Studies in the Valency Grammar of Modern Chinese) II, Beijing: Peking University Press. Zhang, Jing. 1987. Hanyu Yufa Wenti (lssues in Chinese Grammar). Beijing: Chinese Social Science Press. Zhang, Ren. 2000a. Event integration, the semantics of the particle le and ‘implicit’ resultatives in Chinese. In A.K Melby and A.R.Lommel (eds.). LACUS FORUM XXVI: The Mental Lexicon, 489–500. The Linguistic Association of Canada and the US. Zhang, Ren. 2000b. Think constructionally: the case of lian x dou y reopened. Presented at International Symposium on Topic and Focus in Chinese, Hong Kong Polytechnic University. Zhou, Hong. 1997. Cognate objects in Chinese. Ms. University of Toronto.
Index
agentive quale 33, 36, 74, 136, 137, 141, 178 argument structure 3–8ff, 17, 28f, 32, 34, 44, 62ff, 72, 82, 88, 90, 105, 116, 120, 129, 133, 139, 140ff, 157–9, 156, 162f, 168, 170, 177, 179 Asher, N. 88 bei 102, 105, 138ff, 144ff, 151–3, 161 175 Bresnan, J. 3, 18, 23, 43, 52, 156, 160 168, 172, 179 Chappell, H. 101, 105, 125, 134, 135f 137 Chomsky, N. 11, 5f, 8, 18, 21f, 25, 27, 34, 39, 94, 124, 139 coercion 15, 34, 36f, 50, 71, 80f, 87f, 89, 96, 124, 137, 178 “cognate” object 66ff, 74, 79, 82, 85, 90, 94 Cognitive Grammar 5, 7, 21, 25, 40, 69, 100, 111, 126 Conceptual Semantics 3, 5, 15, 22, 27, 33, 37, 99, 107, 113, 129, 141f, 177, 180 Construction Grammar 7f, 11, 21, 23ff, 32, 56, 97, 107, 122, 141, 158, 179, 181 co-compositon 15, 36, 81, 86, 96, 166, 178 Copestake, A. 25, 32f, 37, 87f, 112, 178 Correspondence Condition 111, 113, 116,122 Croft, W. 5, 7, 20, 30f, 40, 61, 82f, 100, 101, 103f, 110, 128f, 130, 141–3, 154, Culicover, P. 5, 14, 18, 21, 24, 31, 77, 100, 155f defeasibility 33, 37, 83 “defective” signs 21ff DICE 88 Discourse Win 88–89 dot object 32, 33, 37, 56, 86 Dowty, D. 10, 15, 28, 30, 124, 128, 157, 171ff, 179 eat restaurant construction 15, 39–97, 180, 183 enriched composition 34, 37, 40, 62, 76, 78, 80, 86, 89, 95ff Ernst, T. 53 External Possessor Construction 105 Fauconnier, G. 25, 30, 32, 36, 108 Feng, S. 47, 92, 94ff, 140 Fillmore, C 20, 32, 142
Index
164
force-dynamic relationship 147 frame 3, 5f, 8, 11, 15, 17ff, 22, 24, 26, 32ff, 36–7, 43, 57, 68, 70, 73, 91, 94, 95, 112–3, 123, 128, 140, 142, 143, 152, 154, 158, 162, 174, 178, 180 functional type 70, 76ff, 136 Gao, Q. 15, 47, 49, 57, 93ff, 106, 156 Generative Lexicon 5, 32, 36–7, 87, 126, 165–6 Goldberg, A. 7, 20–1, 23–6, 32, 142, 154, 158, 176 Grimshaw, J. 3, 15, 32, 43, 139, 141, 144 Hale, K. and Keyser, S.J. 8, 23, 28, 79, 94,96 HPSG 5, 15, 19, 21, 23–24, 52, 73, 93 Huang, C.-T. J. 17, 23, 52–3, 55, 74–5, 96, 101, 106, 138ff, 149, 151ff, 154, 156, 168 Huang, C.-R. 101, 122, 123, 132, 137, 175, 176, 179 inference rule 99, 114ff, 124, 150, 154, 163–4, 168, 171, 174ff, 178–179 inheritance 24ff, 37, 54 Instrument 42–48, 54–58, 83, 85–86, 90, 93–95, 101 ionization 146 Jackendoff, R. 3, 5, 6, 11, 13–5, 17–27, 29–37, 39–41, 43–6, 44, 47, 49, 50, 52, 55–61, 69, 73, 78, 80, 82–4, 86–8, 91, 94, 97, 100, 101, 106–109, 112–6, 114, 119, 120, 122, 129, 134, 142–4, 147, 149, 150, 154, 157, 159, 163–4, 172, 173, 176, 177, 179–180 Lakoff, G. 7, 52 Langacker, R. 5, 18, 20, 22–5, 27, 34, 40, 61, 66, 69, 83, 97, 100, 105, 107, 36, 109, 111, 113, 115, 126, 136, 142, 180 Lascarides, A. 25, 32–33, 37, 87–88, 112, 180 Levin, B. and Rappaport, M. 3, 5, 8ff, 15, 22, 30ff, 36, 39, 45, 128, 142, 155–6, 158, 163, 165, 168, 171 lexicon 7, 9, 19, 20, 43, 177 Li, A. 17, 53, 106, 127, 139, 145–146, 148, 156 Lin, J. 8, 15, 23, 45, 47, 55, 92, 94ff, 164–5, 168, 170, 183 locative inversion construction 100, 106, 126, 136, 156, 160, 162ff, 171 173ff, 178–9, 181 l-rules 24, 37 Massam, D. 5, 10, 22–3, 39, 41, 42, 43–5, 47, 56, 63, 66–9, 75–77, 109, 110, 125 Noteworthiness Condition 136, 151, 164, 179 Nunberg, G. 15, 34, 36, 99–100, 107ff, 116, 174, 122, 134ff, 151, 154, 163–4, 171, 179 particle le 29, 119, 127–132 passives 11, 15, 105–106, 137ff, 144–5, 147–154 Pereltsvaig, A. 67, 69–70, 77 Pinker, S. 27, 28, 32, 40, 90, 124 Possessum-object construction 15, 100, 116–139, 153, 154, 174–5, 178–179 Pollard, C. 21, 25, 52–54, 73, 82, 109–110 predicate transfer 15, 173, 108, 179 Preference Rule System 15, 40, 87–89, 96
Index
165
Projectionist 39, 43, 60, 177 Pustejovsky, J. 5, 10–1, 15, 18, 24, 28, 32, 33–4, 36, 56, 69, 70, 77, 80, 83–4, 86–8, 95–6, 126, 135–6, 141ff, 154, 165, 167, 178 qualia 15, 32ff, 36, 37, 43, 83, 86, 88, 126, 178 relational noun 136 retained-object passives 15, 137–154 Sag, I. 5, 20–21, 24–25, 52–54, 73, 82, 109–110 Shi, D. 18,29,128 swarm-alternation 10, 171–173, 181 Talmy, L. 7, 10, 18, 31, 57, 49, 56, 59, 100, 142, 147 Tripartite Parallel Architecture 17, 19 unselected complements 18, 39, 45, 64, 77, 82 usage-based model 83, 178 Van Hout, A. 8, 15, 39, 96, 124, 142, 154, 181 Van Valin, R. 15, 23, 43, 95, 100, 101, 104, 159 Webelhuth, G. 18, 23, 24 Wiese, H. 18 Xu, J. 15, 18, 101, 106, 118, 124, 125, 139, 152, 153