'1‘“: "v 'V .x ‘ up”. 2:. av... 3.. A an . “flaw/z .(EU \1, - {awake . , . lg 1» . .. . ‘ . . 6.. . , . .. ‘ . ‘ Mrs; vs 1., \c at: ,. ‘ ,. .3 G. . (it)... 51? {Hutu—w :211vu5hfl.“ ‘ . ‘ . » A . c Alumni! . Iv ”.451. 1.0.? 40“?“ 4‘0. n, .41! F2... , . . , ,r .LUVRJH). 01..“ \31 I . ,. . . {9 4 , 21.. a 55:13,. ,3 . . . .. :1 thumbSEYJfJiJJa. . . : , #49 -. (franhkwfihlwnm‘ J Hr.) . , .4 , . . 5.10;!(4... .‘t’a‘llm‘lul. A . ‘1 . v ‘ Vi‘ixninu....:¢1u t. in 11.555 .1: L9. ‘5 , .r7 ii, II i! l» 3.4 73“ 01;: 014114 . -. .. 10" r. .{lfi L. y. 321$... : a El-.- . . v‘iv-X‘f (7.10...“ Ill»! [(06). A‘ ,(I {up \'0 4’. \ fix: a.- .nlsvynmAIi-IU . . . C! . . .3 Hr- .EC . . . \n... p 1... b In .1. it! . .IIIR%A Ill .. ‘ “Fifi, thhztfnrdhlle 11...! L I.. l! was. .3...§%§x¢§3: q 1 IHESIS 3 MCIIHGANS lllllh llllllllllllllllllllllllllllllllll 01688 0415 This is to certify that the dissertation entitled ON SUBJECT AND SUBJECT-POSTPOSED CONSTRUCTIONS IN CHINESE presented by Jean Yuanpeng Wu has been accepted towards fulfillment of the requirements for Ph o D degree in Lil’flliStiCS flW Major professor Date t, (id? ms MSU is an Affirmative Action/Equal Opportunity Institution 0-12771 LIBRARY MlChlgan State University PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. r DATE DUE MTE DUE DATE DUE , ,- W A R9; it 3109 A _________.__— ________,___.. _ 4 / ________.___—— / _________.____.— _________.___. __ #_ __________._——— ____,____...— ________________ ___,_______.._—— / _ #__ use ammo-mu ON SUBJECT AND SUBJECT-POSTPOSED CONSTRUCTIONS IN CHINESE BY Jean Yuanpeng Wu A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Linguistics 1998 Copyright by JEAN YUANPENG WU 1998 ABSTRACT ON SUBJECT AND SUBJECT-POSTPOSED CONSTRUCTIONS IN CHINESE BY Jean Yuanpeng Wu This dissertation presents a study on the notion of ‘subject' in linguistic theory with an emphasis on its relevance in the syntactic structure of Chinese. ‘Subject’ has been one of the oldest descriptive categories utilized in almost all theories of grammar, and yet it has remained a concept difficult to define. In this dissertation, issues related to the notion of ‘subject’ are explored within the theoretical framework of functionalism. It is proposed that ‘subject’ needs to be treated as a prototype concept so as to accommodate the flexibility and gradation it manifests cross-linguistically. This proto-type approach follows from the theoretical assumption that ‘subject’ is a universal linguistic category, though its functions may vary in different languages. It is further proposed that subject, as a proto—type concept, can be defined in terms of a group of characteristic subject properties (based on Keenan 1976), which are abstracted from a large and diverse corpus of data collected from different languages. These properties are used as the collective criteria for identifying subjects. Some of them are more crucial than others in determining what subject is. For instance, the morphosyntactic coding and behavioral properties are considered more prominent indicators of subjecthood. The assignment of the subject status to an NP (or its equivalent) in a predication is based on the comparison of the different degrees of subjecthood measured according to the ranking of subject properties that NPs exhibit in that sentence. It is recognized however, that some subject properties may not apply in all languages. This study also focuses on a particular syntactic structure in Chinese whose nature crucially involves the notion of ‘subject’. With a clause level configuration of V + NP, it has been frequently characterized as intransitive subjectless in the literature. In this dissertation, a detailed and systematic analysis is given of the syntactic, semantic and pragmatic properties of this construction. Arguments are presented that such a construction is not subjectless, but can be better accounted for as a subject-postposed structure. It is shown that the postposing of the subject is triggered by semantic and pragmatic factors which can be clearly identified. To my beloved family ACKNOWLEDGMENTS I am deeply grateful to Professor Barbara Abbott, the chair of my dissertation committee, for her guidance and encouragement throughout the writing of this dissertation and for her contribution to the development of the ideas contained herein. Thanks also to other members of my committee, Professor Grover Hudson, Professor Yen—Hwei Lin, Professor David Lockwood, and Professor Dennis Preston for their comments and suggestions. I would also like to thank the Department of Linguistics and Languages of Michigan State University for providing the teaching assistantships which made my studies of linguistics possible. I also wish to express thanks to Professor Mary Denman, whose help allowed me to come to the United States for graduate studies; and to Dr. Edward Sullivan, whose tremendous kindness and wisdom has been a source of strength and inspiration to me. Finally, I wish to thank my family, for everything. A very special thanks to my husband, Dr. Yangmin Wang, for his love and faith in me. vi TABLE OF CONTENTS CHAPTER 1 INTRODUCTION 1.1 General statement of purpose ........................... 1 1.2 Hypothesis ............................................. 4 1.3 Theoretical framework .................................. 7 1.4 Organization of the study ............................. 14 CHAPTER 2 THE PROBLEM OF DEFINING SUBJECT 2.0 Introduction ........................................... 17 2.1 On the term of ‘subject’ ............................... 20 2.1.1 Three meanings of ‘subject’ ..................... 20 2.1.2 Subject: its metaphysical, logical and grammatical sense ........................... 21 2.2 Previous definitions of subject and their limits ....... 23 2.2.1 Subject as a pure syntactic notion .............. 23 2.2.2 Subject as ‘topic’ .............................. 24 2.2.3 Subject as ‘point of departure' ................. 28 2.2.4 Subject as ‘agent’ .............................. 31 2.3 Previous definitions of subject in Chinese ............. 34 2.4 Toward a universal definition of subjects: Keenan’s Subject Properties List (SPL) ................ 48 2.4.1 Morphosyntactic properties ...................... 50 2.4.1.1 Coding properties ........................ 50 2.4.1.2 Behavioral and control properties ........ 51 2.4.2 Semantic properties ............................. 56 2.4.2.1 Indispensability ......................... 56 2.4.2.2 Independence existence ................... 56 2.4.2.3 Absolute reference ....................... 57 2.4.2.4 Semantic role: agency .................... 57 2.4.3 Pragmatic properties ............................ 58 2.4.3.1 Autonomous reference ..................... 59 2.4.3.2 Topicality ............................... 59 2.4.3.3 Referentiality ........................... 60 2.4.3.4 Given information ........................ 60 2.4.3.5 Definiteness ............................. 61 2.5 Critique of Keenan’s approach .......................... 63 12.6 Summary ................................................ 71 CHAPTER 3 SUBJECT AS A PROTO-TYPE CONCEPT 13.0 Introduction ........................................... 73 3.21 Definition of basic notions ............................ 75 3.1.1 Predication, predicator, and arguments .......... 75 3.1.2 Arguments and ‘satellites’ ...................... 77 3.1.3 Semantic roles .................................. 79 3.1.4 Predicate frames ................................ 80 vii 3.2 The predication principle .............................. 82 3.3 Subject as a prototype concept ......................... 87 3.3.1 Empirical motivation and theoretical plausibility .................................... 87 3.3.2 Prototypical properties of subject .............. 90 3.3.2.1 Review of Keenan’s SPL ................... 90 3.3.2.2 Subject selection is argument selection ................................ 92 3.3.2.2.1 Verbal valence and grammatical relations .................. 93 3.3.2.2.2 Argumenthood and syntactic properties ................... 98 3. 3.2.3 Proto- -agent properties of subjecthood .. 99 3. 3. 3 Ranking of the subject properties .............. 101 3.3.3.1 Argumenthood ............................ 103 3.3.3.2 Syntactic coding and behavioral properties ................... 107 3.3.3.3 Proto-agent properties .................. 116 3.4 Summary ............................................... 118 CHAPTER 4 ‘SUBJECTLESS CONSTRUCTIONS’ in Chinese 4.0 Introduction .......................................... 120 4.1 Arguments against the 'subjectless' analysis .......... 124 4.2 Arguments against the ‘object' analysis ............... 125 4.3 The subjectless verb hypothesis ....................... 126 4.4 Arguments against the subjectless verb hypothesis ..... 127 4.5 Subject properties of the postverbal NPs .............. 131 4.6 Summary ............................................... 137 CHAPTER 5 SUBJECT-POSTPOSED CONSTRUCTIONS IN CHINESE 5.0 Introduction .......................................... 138 5.1 Syntactic structures .................................. 138 5.2 Semantic properties ................................... 150 5.2.1 Constraints on the verb selection .............. 150 5.2.1.1 Verbs of existence ...................... 151 5.2.1.2 Verbs of appearance ..................... 156 5.2.1.3 Verbs of disappearance .................. 159 5.2.1.4 Restrictions on verbal aspects .......... 167 5.2.2 Constraints on the NP selection ................ 168 5.2.2.1 Restrictions on the postverbal subject NP .............................. 168 5.2.2.2 Restrictions on the preverbal locative NP ............................. 170 55.3 Pragmatic Properties ................................. 171 5.3.1 Information structure: an introduction....172 5.3.2 Pragmatic presupposition vs. pragmatic assertion ....................... 173 5.3.3 The principle of ‘the new’ following ‘the given' ..................... 179 5.41.XVS constructions and their focus structure ........... 186 viii 5.5 XVS constructions and theticity ....................... 193 5.5.1 Categorical vs. thetic sentences ............... 193 5.5.2 Entity-central theticity ....................... 195 5.5.3 Event-central theticity ........................ 196 5.6 Summary ............................................... 199 CHAPTER 6 CONCLUSION 6.0 Introduction .......................................... 202 6.1 Subject as a universal category and proto—type concept ................................ 203 6.2 Subject plays an important role in Chinese syntax ..... 206 6.3 The ‘subjectless constructions’ have postverbal subjects ................................... 207 6.4 Properties of the subject-postposed constructions ..... 207 6.5 Limitations and suggestions for future research ....... 209 REFERENCES ................................................ 2 l 1 ix ON SUBJECT AND SUEJECT-POSTPOSED CONSTRUCTIONS IN CHINESE Chapter 1 Introduction 1.1 General statement of purpose The purpose of this dissertation is to give a detailed and systematic analysis of the syntactic, semantic and pragmatic properties of the three groups of sentences exemplified in (1.1)-(1.3): (1.1)a. Xia yu le. fall rain ASP. 'It is raining.’ b. Xia xue le. fall snow ASP. 'It's snowing.’ c. Gua feng le. blow wind ASP. 'It' windy.‘ d. Chu taiyang le. out sun ASP. ‘The sun came out.’ e. You guo yi hui, yingying xiang qi leisheng. again pass one moment, vague sound ASP. thunder 'A moment later, there came the vague sound of thunder.‘ [Examples (a) through (d) from Zhang & Chen 1981, p83; (e) from Huang & Liang 1991, p.115] (1.2)a. You zhe yang yi ge chuanshuo. exist this kind one CL. legend ‘There is such a legend.’ b. You ge nongchun jiao Zhangjiazhuang. exist Cll village call Zhangjiazhuan ‘There is a village called Zhangjiazhuang.’ ' C1. = Classifier, usually used before NPs to indicate the <:ategories they belong to. c. You yi ge ren zai tiaowu. . 2 ex1st one Cl. person Asp dance ‘There is someone dancing.’ d. Houbian you yi ge xiao huayuan. behind exist one Cl. small garden ‘There is a garden at the back.’ (1.3)a. Lai keren le. come guest ASP. ’Some guest(s) came. = Here comes some guest(s).' b. Houlai you zou 1e xuduo ren. later again leave ASP. many people ‘Later, many more people left.' c. Pao le yi zhi mao. run—away ASP. one CL. cat. 'A cat has run away.’ d. Jinglai 1e ji zhi xiao gou. enter Asp. several C1. small dog ‘Several puppies came in.’ e. Tai shang zuo zhe zhuxituan. stage top sit Asp. presidium ‘On the stage sat the presidium.’ The analysis of these sentences crucially involves the grammatical relations of ‘subject’ and ‘object’ in Chinese. Given the fact that ‘subject’ is still hardly a well- defined notion in Chinese, there is an absence of general agreement over the nature of these sentences among Chinese linguists. For instance, some View all (1.1)-(1.3) as typical ‘subjectless constructions’ (Gao 1957, Yang 1963, Chang & Chen 1981), some View all of them as having subjects (e.g. Li and Thompson 1981), and others analyze some of these sentences as subjectless, and some as having 2 Zai is an aspect marker indicating the on—going status ()f the verb following it. a subject (e.g. Chao 1968). Even when linguists agree that a particular sentence has a subject, they differ as to which constituent is the subject. As a result of lack of agreement on the nature and structure of these sentences, the explanations of the syntactic and semantic structures offered for these sentences in Chinese grammar books are sketchy and superficial. Sentences such as (1)-(3) are simply labeled as 'subjectless constructions', or 'non—subject-predicate sentences' etc. without any unified criteria or adequate explanation. There is as yet no consensus among Chinese linguists as to how to define the notion of ‘subject’ in Chinese. The question of subjecthood in Chinese remains a linguistic problem which has attracted a lot of attention in the field of Chinese linguistics but is still unresolved. Thus, two major research questions in Chinese are still in search of answers: (1) How to define 'subject' and identify subjects in Chinese and (2) How to characterize those sentences that are perceived as subjectless. In order to solve these problems, other fundamental issues need to be resolved: Is 'subject' a universal category? If so, what definitions of this category will hold universally, and how can the category be recognized across languages? Are subjects primitive or predictable? In what way can ‘subject’ be best characterized? What criteria can be used to identify the subject? Can a sentence be without a subject? Are those 'subjectless constructions' in Chinese truly subjectless? If yes, why? If not, what is the subject? What are the functions of 'subject' in the structure of Chinese? In this study, I will explore answers to the above questions by investigating the properties of sentences such as (1)-(3) since their syntactic structures are typical of those which have remained controversial. . In brief, the goals of this dissertation are: (A) to redefine the notion of ‘subject’ in Chinese and to argue for the prominence of subjecthood in Chinese; (B) to re-examine the properties of the so-called ‘subjectless constructions’ in terms of their syntax, semantics and pragmatics, and to argue that these sentences have postverbal subjects; (C) to account for subject—postposing phenomena in Chinese in contrast to the unmarked SVO form, (D) to identify the semantic and pragmatic factors which trigger the particular syntactic structures of the subject—postposed constructions. 1.2 Hypothesis I present the following hypotheses to be examined in this study: (I) Grammatical relations such as 'subject' and 'object' are not independent of semantic and pragmatic factors. Subjects are not primitive, but they are also not independent of non-syntactic factors. (II) Subject can be best viewed as a prototype concept rather than taken as a discrete category. Instead of being defined in terms of a single criterion, which would yield numerous counter—examples, subject can be more plausibly characterized in terms of a cluster of prototypical subject properties. (III) The characteristics of a Chinese subject may be different for different constructions of Chinese. Subject in Chinese can be postverbal as well as preverbal, if the conditions under which it can go postverbal are met. (IV) Subject selection is argument selection: only arguments of the predicator can function as subjects. The relation between a subject and a predicate is fundamentally related to the argument structure or the valence of the predicator. Subject can be properly assigned only after the argument structure of the predicator is determined first. (V) Subject can be functionally construed as the most prominent argument of the sentence in the sense that it is that argument which manifest more core subject properties than other constituents in the sentence. (VI) Sentence patterns frequently construed as 'subjectless constructions' can be classified into three groups or types according to their semantic structure. (3roup one is what I will call 'weather sentences' since they normally have to do with the change of natural phenomena such as the weather; group two consists of existential sentences; group three includes what is often referred to as 'presentative sentences' which mainly involve verbs denoting appearance and disappearance. While the data to be presented in this study are by no means an exhaustive list of all the so—called 'subjectless constructions' in Chinese, they represent the prototypical types commonly given in the grammar books and those which are frequently featured in contemporary linguistic discussion. I claim that all of the sentences in (1)—(3) do involve explicit syntactic subjects, subjects that do not occupy the unmarked preverbal position in Chinese. Rather, their subjects are postposed due to semantic and pragmatic conditioning. Thus, I will characterize them as subject- postposed constructions in Chinese. Furthermore, I argue the subject postposing is motivated by pragmatic factors and principles. I will present evidence to show that Chinese is a pragmatics—driven language in the sense that the organization of basic sentences is very much conditioned by pragmatic considerations. (VII) I will suggest that the theory of information structure and especially the notion of focus structure play an important role in accounting for subject-postposed constructions and syntactic structures in general in Chinese. My claim is that a syntactic structure is often determined by its information structure and the subject- postposed constructions represent certain grammaticalized focus structures in Chinese. 1.3 Theoretical framework The theoretical approach that I use to analyze subject and subject postposing in Chinese is that of functionalism, which emphasizes the consideration of the pragmatic purposes served by linguistic forms in the analysis of linguistic phenomena. This is well explained by M. Halliday, a dominant functionalist in the history of linguistics: A functional approach to language means, first of all, investigating how language is used: trying to find out what are the purposes that language serves for us, and how we are able to achieve these purposes through speaking, listening, reading and writing. But it also means more than this. It means seeking to explain the nature of language in functional terms: seeing whether language itself has been shaped by use, and if so, in what ways --how the form of language has been determined by the function it has evolved to serve... (Halliday 1973:7) The fundamental principle underlying functionalist theories is a functional perspective on the language, which is regarded as 'an instrument which human beings use in order to achieve certain goals and purposes’ (Dik 1987: 83). Therefore the structure of linguistic expressions is taken as non-arbitrary but to a large extent influenced by the communicative purposes of the language user. This View is expressed in the non—autonomous assumption widespread among functionalists that ‘language (and grammar) can be neither described nor explained adequately as an autonomous system’ (Givon 1995: xv). The language system is not considered as an autonomous set of rules and principles; rather, it is assumed that the rules and principles composing the language system can only be adequately understood when they are analyzed in terms of the conditions of use. Specifically, functionalists advocate the following theoretical premises (see Givon 1995z9): 0 Language is a social-cultural activity 0 Structure serves cognitive or communicative function 0 Structure is non-arbitrary, motivated, iconic 0 Change and variation are ever—present 0 Meaning is context-dependent and non-atomic 0 Categories are less—than discrete 0 Structure is malleable, not rigid 0 Grammars are emergent 0 Rules of grammar allow some leakage The ultimate goal of functionally-oriented linguistic theories is to determine the relation between the form of a linguistic expression and its linguistic function (Bolkestein et al. 1985, Dik 1991). In the analysis of the subject-postposed constructions in Chinese, one of my goals is to have an understanding of the correlation between the syntactic structures that these constructions display and the functions they fill in verbal interaction. A major point that I argue for is that the syntactic structures under examination are non—arbitrary and cannot be adequately accounted for unless semantic and pragmatic factors are taken into consideration, because these structures are very much conditioned by the language user's judgment of situations in which they are used. Major theoretical assumptions and principles based on which this study is conducted are given as follows: (I) Following Functional Grammar outlined in Dik (1991), I will describe linguistic expressions in terms of a level of abstract underlying predications, built up from predicates and arguments, which together constitute the base of the language. Every predicate is a part of a predicate frame where the semantic and morphosyntactic properties of the predicate are stored. Predicate frames are the most basic components for the construction of predication. In FG outlined in Dik (1991), an underlying predication is formed through the insertion of appropriate terms into the slots of a predicate frame. The constituents of that predication will then be provided with syntactic and pragmatic functions, according to general principles. Underlying predications are mapped onto linguistic expressions through a system of 'expression rules', which govern the form and order of the constituents, depending on the structural properties of the corresponding components of the underlying predication (Dik 1991:249). (II) Linguistic expressions involve three dimensions (Danes 1966, Halliday 1967 & 1985, Dik 1991): (a) the grammatical dimension, which involves the syntactic functions (e.g. 'subject' and 'object'); (b) the semantic dimension, which specifies the roles that the referents of the NPs play within the state of affairs designated by the predication in which the NPs occur (e.g. 'agent', 'goal', 'recipient', 'beneficiary', etc.); (c) the pragmatic dimension, which specifies the informational status of the constituents of a predication within the communicative setting in which they occur (e.g. 'given' and 'new'; ’theme' and 'tail'; 'topic' and 'focus', etc.). In this dissertation, I will explore the relations between these three dimensions of syntax, semantics and pragmatics. I hold the View that syntax is not autonomous of pragmatics and semantics: many syntactic phenomena only make sense against a background of pragmatic and semantic correlates; and many syntactic phenomena in Chinese can be regarded as syntacticization of semantic-pragmatic phenomena. (III) The pragmatic dimension is an essential part that characterizes functionalist theories. It is generally assumed that the form of a linguistic expression may be influenced not only by the semantic and syntactic functions borne by the arguments of various predicates but also by the pragmatic functions associated with constituents. In addition, it is claimed that the structuring of sentences lO has to do with what the speakers assume hearers know and are paying attention to at the time of utterance (Chafe 1976, Prince 1986, Lambrecht 1994, Vallduvi 1995). Vallduvi (1995:123) maintains that 'when communicating a proposition p, a given speaker may encode p in different sentential structures according to his/her beliefs about the hearer's knowledge state with respect to p.‘ Lambrecht (1994) argues that the relationship between speaker assumptions and the formal structure of the sentence is 'governed by rules and conventions of sentence grammar' and views 'information structure' as a grammatical component3. In the analysis of subject postposed constructions in Chinese, I will make the following distinctions (largely based on Lambrecht 1994) which appear to correlate directly with structural properties of the syntactic structures I will investigate in this study. 0 Pragmatic presupposition vs. pragmatic assertion Basically these are concerned with the speaker's assumptions about 'the hearer's state of knowledge and awareness' at the time of utterance. I claim this distinction helps to explain some of the differences 3 The formal definition of 'information structure' proposed by Lambrecht (1994:5) is as follows: 'That component of sentence grammar in which propositions as conceptual representations of states of affairs are paired with lexicogrammatical structures in accordance with the mental states of interlocutors who use and interpret these structures as units of information in given discourse contexts’. 11 between subject—preposed structures and their postposed counterparts. I argue that the variation of subject position in Chinese to a large extent has to do with the nature of the setting in which assertions occur as well as the pragmatic presupposition on the part of the speaker. 0 Topic vs. Subject The notions of 'topic' and 'subject' function in different dimensions (i.e. pragmatic vs. syntactic) and they are not conflated since topics are not necessarily grammatical subjects, and grammatical subjects are not necessarily topics. Thus, the distinction is made between 'non-subject topics' and 'non—topic subjects'. The contrast is also made between 'topicalized sentences' and 'non-topicalized sentences'. 0 Focus structure This refers to 'the conventional association of a focus meaning with a sentence form' (Lambrecht 1994: 222). Three kinds of focus structure are distinguished: Predicate focus, argument focus, and sentence focus. Predicate—focus structure involves a presupposition and an assertion in the unmarked subject-predicate sentence type, in which the predicate is the focus and the subject is in the presupposition. Argument—focus structure is one in which 'the focus identifies the missing argument in a presupposed open proposition’ (p.222). The sentence-focus structure 12 characterizes the event—reporting or presentational sentence type, in which the focus extends across the whole sentence. In this dissertation, I argue that the unmarked focus structure in Chinese is the predicate-focus structure and that subject-postposing occurs only in sentence-focus structures. In other words, the focus structure of a predication determines the subject position in Chinese. 0 Categorical vs. thetic sentences These are two different types of 'judgment’ which involve different structures. This distinction was first proposed by the nineteenth—century philosopher Brentano and his student Marty as a cognitive distinction between two types of human judgment (Kuroda 1992:19), and later further developed by Kuroda (1972 & 1992), Sasse (1987) and Lambrecht (1994). According to Kuroda (1992), the categorical judgment corresponds to the subject—predicate structure (i.e. associating a subject with a predicate) while the thetic form does not. Specifically, a categorical judgment involves both the act of recognition of a subject and the act of affirming or denying what is expressed by the predicate about the subject. In contrast, a thetic judgment does not predicate a property of some entity but simply asserts a fact or state of affairs. In this thesis, I will argue that subject-postposed constructions in Chinese are typical thetic sentences in the sense that they manifest an 'all-new character' by 13 'introducing a new element into the discourse without linking it either to an already established topic or to some presupposed proposition' (Lambrecht 1994: 144). 'Categorical sentences' on the other hand, represent the 'topic-comment' structure. To summarize, I will argue that the grammatical form of subject postposing in Chinese is motivated by the requirement of information structure. I claim that subject postposing only occurs in what Lambrecht (1994) refers to as 'pragmatically structured propositions'. Predications in which subjects are postposed can be analyzed as having grammaticalized 'sentence—focus structures' featuring 'non- topic subjects' and 'thetic' propositions. 1.4 Organization of the study This thesis includes three major parts: (I) A review of previous studies on the notion of ‘subject’ The study of subject-postposing in Chinese crucially involves the notion of ‘subject’: how it features in the grammatical theories in general and how it works in Chinese in particular. Thus, as we can see, while the universality of subject as a grammatical category is widely assumed, the difficulties of determining subject are also well—known. ‘Subject’ is hardly defined in a unified fashion, presumably due to the fact that languages differ considerably in the prominence which they give to 14 subjecthood, as far as surface manifestations of it are concerned (Li and Thompson 1981). In Chapter 2, I will first present the problem of defining 'subject' by examining previous research on the notion of ‘subject’ and then establish the need to redefine the notion of ‘subject’ in Chinese. (II) Redefining the notion of ‘subject’ in Chinese In chapter 3, I will attempt to redefine the notion of ‘subject’ in Chinese. I will first establish the predication principle that ‘all clauses have a subject’ and then argue that subject can be best viewed as a prototype concept. In Chapter 4, I will discuss the properties of the so- called ‘subjectless constructions', and argue that they have subjects, subjects that are postposted. I will therefore claim that these sentences should be accounted for as subject—postposed constructions. (III) A systematic analysis of ‘subject-postposed constructions’ Having shown that the so—called ‘subjectless constructions' are in fact subject—postposed constructions, I will explore in chapter 5 the semantic and pragmatic constraints on the subject—postposed constructions which determine their use. I will attempt to identify the .necessary semantic and pragmatic conditions under which a 15 subject can be postposed in Chinese and then propose some general statements about their use in functional terms. In Chapter 5, I will also discuss the distinction between two types of statement, ‘categorical’ vs. ‘thetic’, and propose that in Chinese all sentences with subject postposed can be uniformly characterized as having the typical ‘thetic’ structure. In Chapter 6, I will summarize major findings in the dissertation and present some suggestions for future research. 16 Chapter 2 The problem of defining subject 2.0 Introduction The practice of dividing a sentence or clause into two parts, subject and predicate, is a long established one. The notion of ‘subject’ has always been considered a basic grammatical relation in the sentential structure of languages. This grammatical relation has always played a central role in descriptive discussions of data in the world’s languages. In recent years, grammatical relations such as subject and object have come to play a prominent role in linguistic theory as well. For instance, Relational Grammar is based on the idea that grammatical relations are primitive elements. In this theory, the phrase structures and derivations of transformational grammar are dispensed with in favor of relational networks (or relational graphs). Various generalizations, including many originally motivating transformations, are captured by allowing constituents to bear distinct grammatical relations in different strata. Grammatical relations are also central to the analysis of clause structure in Lexical Functional Grammar (LFG). LFG mapping theory provides an account of the linking of thematic roles with grammatical relations, based on the idea that the former are organized into a hierarchy and the latter are composed of binary features. 17 In Government and Binding Theory (GB) and Head-Driven Phrase Structure Grammar (HPSG), ‘subject’ is a derivative notion, and it is a key element as well. Within GB, position in phrase structure, which potentially yields different theta-role assignments, different government configurations, and the properties corresponding to the grammatical relations of other theories, is crucial to the statement of both universal principles and language- particular generalizations. Within HPSG, a predicate’s dependents are differentiated by relative positions in the SUBCAT list, a formal device that yields the equivalent of grammatical relations. Thus, as we can see, the grammatical relations apparently serve as a common ground for practitioners of various syntactic theories. As an important grammatical relation, 'subject' is 'frequently used in the description of individual languages and in stating cross-linguistic generalizations' (Comrie 1989:105). Despite its wide use, 'subject' is still a notion that is hardly defined in a unified fashion, presumably due to the fact that languages differ considerably in the prominence which they give to subjecthood, as far as surface manifestations of it are concerned (Li 1981). Research indicates lack of agreement among linguists on subjecthood. Studies have shown that subjects can vary in their properties not only cross- linguistically but also within a specific language (Givon 1996; Anderson 1976; Chung 1976; Craig 1976). For a 18 language like Chinese, where there is not much morphological inflection for case marking or syntactic agreement between subject and the predicate, it is difficult to identify subjects. In the Chinese descriptive tradition, one of the most controversial issues has been which constituents to be assigned the grammatical status of ‘subject’. Many questions concerning the concept of 'subject' are still in search of answers today. For example: Is subject a universal category? If so, what definitions of this category will hold universally, and how can the category be recognized across languages? What role do subjects play in the structure of language? In what way can ‘subject’ be best characterized? What criteria can be used to identify the subject? Can a sentence be without a subject? Are those 'subjectless construction' in Chinese truly subjectless? If yes, why? If not, what is the subject? What are the functions of 'subject' in the structure of Chinese? In this chapter, I will survey issues surrounding the notion of 'subject' and review previous studies on this concept. I will then establish the need to redefine it in Chinese grammar. l9 2.1 On the term of ‘subject’ 2.1.1 Three meanings of the term ‘subject’ The term ‘subject’ has different meanings in different contexts. According to Pope (1995), three meanings of ‘subject’ are current: (A) ‘Subject’ meaning grammatical subject: it controls the verb and is traditionally distinguished from grammatical ‘object’. For example, in the two sentences below, the underlined constituents are the corresponding subjects of each sentence. (2.1)a. She started the car. b. The car was started easily. (B) ‘Subject’ meaning subject matter: What something is about. This can have a generalized sense, as in ‘the subject of the film is...’ or a specialized educational sense, meaning ‘subject of study’ or ‘academic discipline.’ (C) ‘Subject’, as in psychological and modern philosophical usage, meaning ‘subject position: a perceptual location within or orientation towards an event, usually unconsciously and habitually assumed' (Pope 1995:46). In this dissertation, it is primarily the first two meanings that are of relevance. 20 2.1.2 Subject: its metaphysical, logical and grammatical sense The term 'subject' comes from Latin subjectum, which is a translation of Aristotle's to hupokeimenon, meaning 'the material of which things are made', hence 'the subject of an attribute or of a predicate' (Chalker et a1 1994: 378). 'Subject' has been used in at least three different senses, not only grammatically but also metaphysically and logically. As a metaphysical concept, 'subject' is 'a unit of existence-—whatever is assumed to exist as an individual entity' (Kuroda 1976: 1): A subject is that in which various properties are contained, those properties which can be affirmed of it as an entity. The entity as a subject may be identified by the properties that belong to it, and these properties may be considered as making up the notion of the entity as a subject. ‘Subject’, in this sense, overlaps the concept of ‘substance’. (Kuroda 1976: 1) According to Kuroda (1976), the logical concept of 'subject' is derived from the metaphysical concept of 'subject'. If we assume that the universe consists of subjects, (i.e. entities with their respective properties), Kuroda argues, then 'the content of a single judgment, must take the form of affirming or denying the attribution of some property to some subject' (p.2). Kuroda goes on to explain: ‘That to which a judgment affirms or denies the attribution of a property is called the subject of the 21 judgment, and that which the judgment affirms or denies of the subject is called its predicate’ (ibid.). In this dissertation, the focus of study is 'subject' as a grammatical concept. As such, it has been discussed extensively in the linguistic literature. Major research in recent linguistic history on the notion of ‘subject’ includes Keenan (1976); Li and Thompson (1976); Schachter (1976); Foley and van Valin (1977 & 1984); Cole et al.(1980); Tomlin (1983 & 1986); Comrie (1989); Arnold et al. (1989); Chafe (1994); Palmer (1994); Naylor (1995). A significant consensus among many of these researchers is that there is no universal definition of ‘subject‘, and it is almost impossible to discuss the notion of 'subject' outside a particular grammatical theory. As a result, the notion of 'subject' is defined differently within different theoretical frameworks, and different theories employ different analytical devices to account for grammatical relations such as ‘subject’ and ‘object’ etc. For example, GB theory defines ‘subject’ structurally in terms of dominance relations in phrase structure trees, Relational Grammar and Lexical Grammar treat it as a primitive notion, and Functional Grammar characterizes it in terms of a cluster of properties that subjects share cross- linguistically. In this dissertation, the study of the grammatical category of ‘subject’ is oriented towards a functionalist theoretical framework, and ‘subject’ is characterized as a prototype concept, definable in terms of 22 its prototypical subject properties attested cross— linguistically. 2.2 Previous definitions of ‘subject’ and their limits In the following sections, I will review previous definitions of 'subject' proposed within the framework of functional theories and discuss their advantages and limits. 2.2.1 'Subject' as a pure syntactic notion 'Subject' is frequently regarded as a pure syntactic notion. As such, it refers to the primary syntactic relation borne by a NP with respect to the verb. It is generally identifiable through syntactic coding features such as case marking and agreement. Viewed as strictly a syntactic category, subject has no semantic or pragmatic attributes. Notions such as 'semantic role', 'position', 'theme', 'topic', or 'old information', which are often used in the literature to characterize subject, represent neither features of subjects nor their identifying characteristics. Instead, what is emphasized is the morphological marking and syntactic behavior of subjects. The syntactic function of 'subject' seems to be essential in the sentence structure of languages such as English. So much so, that a dummy subject must sometimes be introduced: 23 (2.2)a. It’s raining. b. It’s me. While this syntactic definition of subject seems to work well with Indo-European languages such as English, where a grammatical relation may be immediately recognized on the basis of the coding features in ordinary clauses, it runs into a major problem in Chinese which lacks morphological inflections as basis for analysis. Morphosyntactic features such as cross—referencing (frequently called 'agreement'), case-marking, etc. do not apply to Chinese and thus cannot be used as a sole criterion for identifying subject. 2.2.2 Subject as 'topic' Subject is often defined as 'what is being talked about'. Subject and predicate are often used respectively to refer to 'that which is spoken of' and 'that which is said of the subject'. In this sense, subject is defined in terms of the concept of 'topic'. There are several problems with confusing 'subject' with 'topic'. First of all, the division of a sentence into grammatical subject and predicate does not always correspond to the distinction between 'topic' and 'comment', that is, what the sentence is about and what is said about it. This point is well illustrated in Jespersen (1924: 146), where given the sentence 'Jo n promised.Mary a gold ring', one could say there are four things of which something is said, and which 24 might therefore be said to be subjects, namely (a) John, (b) a promise, (c) Mary and (d) a ring. If we were to claim that all of the four choices are equally valid, we would be led to say that the subject of the sentence varies according to the speaker. But then a sentence would not have a fixed subject, it would have multiple potential subjects, relative to different individuals, given that different individuals tend to have different intuitions as to what a sentence is about. The following sentences further illustrate the problem with defining subject as what the sentence is about. They involve the same linguistic string, but are spoken in different ways, with the capital letters indicating where the audible stress peak of the sentence falls: (2.3)a. JOHN likes Mary. b. John LIKES Mary. c. John likes MARY. It goes without saying that the discourse contexts differ in which these sentences are uttered with these different intonation contours. For example, if the speaker assumes that the hearer knows someone likes Mary but does not know who likes Mary, (2.3a) might be used. If the speaker assumes that the hearer knows that John has something to do with Mary, but does not know in what respect, (2.3b) might be used. If the speaker assumes that the hearer knows that John likes someone, but does not know ‘who he likes, (2.3c) might be used. Now if these different 25 situations mean that each of the sentences in (2.3) is about different things, then we could be forced to say that the subject of each of these sentences is different. However, syntactically speaking, these sentences are obviously identical, given that they involve the same syntactic structure with the same lexical items in the same word order. If the notion of ‘subject’ is supposed to have clear syntactic correlates, such as manifested in (2.3), the definition of ‘subject’ as ‘what the sentence is about’ certainly fails in this regard. Therefore, it can be concluded, just as Jespersen (1924: 146) did years ago, that 'this popular definition, according to which subject is identified with subject—matter or topic, is really unsatisfactory'. The current literature also attests a widespread objection against conflating 'subject' and 'topic'. This objection is clearly reflected in Napoli’s statement quoted below, which echoes a general consensus in the linguistic community with regard to the relation of ‘subject’ and ‘topic’. Although in subject-prominent languages like English, the topic is usually the subject, we may not in principle subsume topic under subject, and the definition of subject must not simply appropriate the notion of topic without further ado. Topic is a discourse function realized at various levels of structure. Although in languages like English topic function is typically coded in clause syntax as subject, topic and subject are nonetheless separate entities. (Napoli 1995: 161) 26 Conflating ‘subject’ and ‘topic’ also presents another kind of problem for language internal data where there are clear cases of subject being non-topical, as in (2.2), repeated below, and topics being non—subject, as in (2.4). Compare: (2.2)a. It’s raining. b. It’s me. (2.4)a. The play, John saw it yesterday. b. As for the play, John saw it yesterday. [Example (2.4) from Chafe 1976: 49] In (2.2), one would agree that the subject of the sentences is ‘it’, given its syntactic agreement feature with the verb. However, ‘it’ is only what is called a ‘dummy subject’ with no referential or semantic content, and therefore cannot be considered topic of the sentences. In contrast, sentences may involve topics that are not subjects. ‘The play’ in (2.4a), marked as the topic of the sentence by a comma which separates it from rest of the sentence or a pause in speech, is generally considered topical, but not the subject of the sentence. Nor is ‘the play’ in (2.4b) a subject, though overtly marked as the topic of the sentence by the topic expression ‘as for..’. Both sentences in (2.4) may be characterized as contrastive sentences with topicalization of the object. It is generally considered that the subject of the sentences is ‘John’, given its semantic role of agent, and given the 27 fact that in basic active sentences, the subject is usually the agent, if there is one. 2.2.3 Subject as 'point of departure' In this sense, ‘subject’ primarily refers to the sentence-initial NP in the sentence. The theoretical foundation for this point of View is largely cognitive and appears to parallel the term 'psychological subject' introduced by grammarians of the nineteenth century. For example, von der Gabelenz (1891: 351) defines the psychological subject as 'the idea which appears first in the consciousness of the speaker...what makes him think and what he wants the hearer to think of'. He considers that this notion is determined by word order and the psychological subject needs to be the first element in the sentence (ibid.). The psychologist G. F. Stout, who was quoted by Jespersen in The Philosophy of Grammar (p.246), shares the same View and considers the subject to be the 'product of previous thinking which forms the immediate basis and starting-point of further development.‘ Chafe (1994:83) also argues for the conception of subject as 'point-of—departure': Clauses do not express a random collection of independent events or states, floating in the air like so many disconnected bubbles. Rather, each has a point of departure, a referent from which it moves on to provide its own new contribution. It is this starting point referent that appears grammatically as the clause's subject. 28 This view of subject is closely related to the discourse notions of 'given' vs. 'new' information. Chafe (1994) maintains that the referents of subjects tend to be either 'given' or 'accessible', but not new information. 'It makes sense', he says, 'that one would employ as a starting point a referent that is already active in the discourse' (p.85). Studies have shown that one of the striking properties of subjects manifested in discourse is the fact that a high proportion of them do express 'given information' (Prince 1992, Chafe 1994). The idea of subject as the expression of starting- points undoubtedly gives us an insightful perspective on the nature and function of subject in the sentence. However, there are several problems with designating subject as the sentence—initial NP. First, compare the following sentences: (2.5)a. He likes beans, not peas. b. Beans he likes, not peas. In (2.5a), we would be led to claim that he is the subject, but in (2.5b), beans would be the subject. However, these sentences appear to express the same eventuality -— the same person likes and dislikes the same kind of things. It is true that the two sentences differ to a certain extent: (2.5b) involves a more marked word order with its object topicalized. However, one could hardly characterize the difference between the two as one of the difference between subjects, given the identical 29 syntactic agreement features between the pronoun he and the verb likes in both sentences. Generally speaking, if a language clearly has overt morphosyntactic agreement features, these features are normally considered clear indications of grammatical relations such as ‘subject’ and ‘object’. Now consider the following pairs: (2.6) a. Mary left. b. Yesterday Mary left. c. Mary left yesterday. Again, the definition of the ‘subject’ as the sentence-initial NP would lead to the conclusion that the subjects of the three sentences in (2.6) are different: The subject of (2.6a) would be Mary while that in (2.6b) would be yesterday. However, one could not help notice that yesterday as a temporal NP, is syntactically and semantically peripheral in (2.6b) compared to the NP Mary. It serves here to modify the event denoted by the rest of the sentence, that is, Mary left. Claiming that the subject is yesterday would mean that the subject of the sentence has changed as a result of this modification. But then consider (2.6c), the subject remains unchanged (i.e. still is.Mary) with the same type of modification. The apparent contradiction between (2.6b) and (2.6c) suggests that defining the subject as the sentence initial NP is less than satisfactory. 3O There is yet another problem. Designating the sentence—initial NP as the grammatical subject would also mean that a sentence is subjectless if no NP precedes the verb. Given the fact that 'almost all major syntactic categories in Chinese can appear either before or after the verb' (J. Lu 1956:589), a large number of basic sentences in Chinese, would have to be construed as subjectless when, on the contrary, they could be analyzed more plausibly as containing subjects either in terms of their syntax, semantics or pragmatics. In these cases, the definition of subject which solely depends its position in a sentence, irrespective of its semantic and syntactic properties would be too mechanical and restricted to be reasonable. Cross—linguistically, there are also problems. What about those verb-initial languages such as Tagalog and Arabic? They would be all denied subjects if ‘subject’ is defined as the sentence—initial NP. 2.2.4 Subject as 'agent' In traditional grammar, the subject is often defined as the 'doer' or 'agent' of the verbal action (Chalker and Weiner 1994:379). In other words, the subject of the sentence is the NP whose referent is understood to do some kind of an action. There seems to be some psychological justification for this perspective. Studies indicate that speakers tend to select as subjects participants they empathize with most, and agents are more likely to be 31 selected as subjects because prototypical agents are volitional and most likely human (Dowty 1991, Croft 1991). There is also the factor of 'animacy' that is often associated with the semantic role of 'agent'. It is suggested that ‘high animacy correlates high topicality’ and chances are agents are more animate than any other semantic roles (Croft 1991:154). The limitation of this definition is obvious when one considers the following sentences: (2.7)a. Mary is lovely. b. John is frightened by the ghost story. In (2.7a), one would agree that the subject of the sentence is Mary, but Mary performs no action. Being lovely is a kind of quality or state rather than an action. (2.7b) is a passive sentence, and the subject is John. Yet, John performs no action either. In fact, all passive sentences involve subjects with typically the semantic role of ‘patient' rather than ‘agent’. While it is true that subject is frequently associated with the semantic role of 'agent', subjects are not restricted to agents only. Consider the following examples from Fillmore (1968: 33). (2.8)a. John opened the door. b. The key opened the door. c. The door opened. One would agree that in (2.8b) and (2.8c) the key and the door are subjects respectively, but they do not carry theesemantic role of 'agent'. Instead, the key carries the 32 semantic role of 'instrument', and the door that of 'patient'. These examples indicate that the connection between semantic role and choice of subject is not at all obvious, and that agents are not equivalent to syntactic subjects. It is misguided to equate a grammatical relation with a single semantic function. This definition would have the undesirable consequence of rendering all ergative languages subjectless, where subject is formally identical with 'patient' in the active construction (see Palmer 1994: 12). Equating the grammatical category of ‘subject’ with the semantic role of ‘agent’ would certainly not work for Chinese, where basic sentences often do not involve NPs with the semantic role of 'agent'. For instance, one would agree that the initial NPs in (2.9) below are subjects of the sentences, however, neither is an agent. (2.9)a. Fan chi le. meal eat Asp. ‘The meal is finished.’ b. Xin fa le. letter send Asp. 'The letter has been sent.’ Thus, the definition of ‘subject’ as ‘agent’ would again lead to the undesirable abundance of subjectless sentences in Chinese when in fact many of them manifest a perfect subject-predicate structure. In the discussion of the different definitions of ‘subject’ so far, I have shown that each of these 33 definitions is based on emphasis on a particular function that a subject displays in a language. The major problem with these definitions, I argue, is that the notion of 'subject' is taken as a one-dimensional (or one—condition) concept, either as a pure syntactic notion, or equated with a pragmatic notion such as 'topic', or a semantic notion such as 'agent'. I have presented counter examples which indicate that none of these adequately describes linguistic reality. In order to adequately account for the diverse cross—linguistic data, the notion of ‘subject’ needs to be redefined. 2.3 Previous definitions of subject in Chinese There was a heated linguistic debate during the mid- 50's in the Chinese linguistic community on grammatical relations of ‘subject’ vs. ‘object’ in Chinese, but no conclusion was made nor consensus reached. The debate still goes on today. Major traditions in the analysis of subject in Chinese can be roughly classified into five groups: (A) ‘Subject’ as topic: Chao (1948 & 1968); Hong (1956); Householder & Chen (1967); Alleton (1973); Huang (1973); Zhu (1981); (B) ‘Subject’ as the sentence—initial NP in the sentence: Zhang J. & Y. Chen (1981); Huang & Liao (1991); (C) ‘Subject’ as the agent or actor: S. Lu (1942); T. Tang (1989); Gao (1994); 34 (D) ‘Subject’ as as either the topic or as the actor: S. Lu (1979); L. Li (1985); (E) ‘Subject' as the NP which has a ‘doing’ or ‘being’ relationship with the verb: Li and Thompson (1976 & 1981). Among these five groups, the most dominating and influential views are the first two. These two views are in fact essentially the same: subject is defined as ‘topic’, i.e. what the sentence is about. The sentence- initial position is designated as the topic position, and subject is designated as the sentence—initial NP of the sentence. The major proponent of this analysis of subject was Yuan-Ren Chao, one of the most renowned Chinese grammarians. He claimed that 'subject', as a grammatical term, cannot be defined syntactically within the framework of Chinese grammar. In Mandarin Primer (1948) and especially in his A Grammar of Spoken Chinese (1968), Chao said: (a) The phonetically loose connection between subject and predicate is paralleled by a semantic looseness. In a Chinese sentence, the subject is literally the subject matter and the predicate is just something said about the subject matter. (Mandarin Primer, p.35) (b) The grammatical meaning of subject and predicate in a Chinese sentence is topic and comment, rather than actor and action. (A Grammar of Spoken Chinese, p.69) 35 (c) Note that we are using the terms of ‘topic and comment’ as semantic terms and not as grammatical terms as used by many writers in discussing Chinese grammar. (A Grammar of Spoken Chinese, p.69) It is clear from these statements that Chao did not treat ‘topic’ as a grammatical or discourse entity distinct from the ‘subject’, but rather as a way to talk about the meaning that subject conveys in Chinese. In Chao’s view, the functions of ‘subject’ and ‘topic’ are simply the same, and the subject of the sentence is the subject matter that is being talked about. Put differently, most Chinese sentences consist of a subject (the first noun phrase) and a predicate (the rest of the sentence), but the meaning or function of ‘subject’ and ‘predicate’ is ‘topic’ and ‘comment’ respectively. However, this definition presents a problem for the analysis of sentences such as (2.10) below, given that one could hardly agree on what such a sentence is about, and yet one is reluctant to say that these sentences do not have a subject. (2.10)a. Shui chi le dangao? who eat Asp. cake ‘Who ate the cake?’ b. Shui dou meiyou lai. who all not come ‘Noboby came.’ Zhu (1981: 95-96) also characterizes the subject in Chinese as the sentence-initial element. He claims that 36 the subject in Chinese is preverbal and functions as the topic of the sentence. As such, it can be optionally marked off from the predicate by a pause and/or particles. This subject-as-topic approach is featured in contemporary Chinese grammar books and is the major approach currently being adopted by the Chinese linguistic community in China: ‘subject’ is defined as the ‘tOpic’, and the first NP of a sentence is designated as subject, because the sentence- initial position is considered as the topic position. (2.11)a. Women zuotian kai le yi ge hui. we yesterday hold ASP one CL. meeting <—Subject—> < ------------------ Predicate -------------- > ‘We held a meeting yesterday.’ b. Zuotian women kai le yi ge hui. yesterday we hold ASP one CL. meeting <-8ubject-> < ------------- Predicate ------------------- > <-8ubject-> < --------- Predicate ————— > ‘Yesterday we held a meeting.’ (2.12) Zhe ge ren xinyan’r hao. this CL. person heart good <—--8ubject ----------- > < ————— Predicate ----- > <—Subject-> <—Predicate—> ‘This person is kind—hearted.’ [Examples (2.11)-(2.12) are from Fang et al., 1995:235] In this analysis, the predicate of a clause may itself be a subject-predicate predication: a sentence may contain two subjects, a ‘major subject’ and a ‘minor subject’, the former being the subject of the whole clause, the second the subject of only the predicate segment of the clause. If a sentence begins with a non—nominal constituent, it is considered subjectless. 37 The problems with this type of analysis, I argue, are two fold: on the one hand, it unduly complicates the structure of Chinese sentences, and on the other hand, it oversimplifies the structure of Chinese sentences. For instance, e.g. (2.11a) and (2.11b) have the same constituents, the only difference being that in (2.11a), the pronoun women 'we' is sentence-initial, while in (2.11b) the adverbial of time zuotian 'yesterday' begins the sentence. According to this analysis, the subject is different for each sentence because their initial elements are not the same, and (2.11b) would have to be considered as having a double-subject when in fact, it could be treated as a very simple sentence with a single subject, just like in (2.11a). Let me give another example. (2.13)a. Ta mingtian lai. s/he tomorrow come <-Subject-> <----Predicate ——————— > ‘S/he will come tomorrow.’ b. Mingtian ta lai. tomorrow s/he come <-8ubject—> < ------ Predicate ----- > <-Subject—> <-Predicate-> ‘Tomorrow s/he will come.’ Under this analysis, (2.13b), which is a simple sentence, would have to be considered a complex double— subject construction. This analysis would yield the result that a majority of Chinese sentences have a double-subject structure, when in fact many of them can be accounted for as simply having just one subject. 38 The Chao/Zhu analysis oversimplifies the syntactic structures of Chinese in the sense that all subjects are sentence-initial, then there is really no variation in terms of the order of the constituents to account for. Analysis of syntactic structures in Chinese is rendered uninteresting and unmotivated, since there is are really no variations to account for. The motivating pragmatic factors underlying the word order variations are largely ignored in the Chao/Zhu analysis. This analysis also ignores the syntactic nature of the first two NPs of the main clauses: one of them in fact shares a closer relationship with the verb. For instance, in (2.11b), women 'we' is selectionally more closely related to the verb kai, 'held', than its subject counterpart zuotian 'yesterday'. In (2.13b), ta 'he' is more closely related to the verb lai 'come' than its subject counterpart mingtian. Dubbing 'subject' as 'topic' also has the undesirable consequence of having the syntactic properties of subject in Chinese ignored and ‘subject' treated as a marginal notion in Chinese syntax. I claim that 'subject' as a syntactic notion is as important in Chinese as in English, because, it is the notion of 'subject' (not the notion of 'topic') that plays a central role in syntactic processes such as serial verb construction, reflexivization, imperativization, etc. These issues will be further discussed in the next chapter. 39 Li and Thompson’s (1976 and 1981) distinction of the notion of ‘subject’ from the notion of ‘topic’ marks a significant departure from the traditionally dominating (despite being problematic) definition of ‘subject’ via the notion of ‘topic’ or sentence position. For the first time, ‘subject’ and ‘topic’ are considered as distinct notions in Chinese, and each is characterized with different properties. The following briefly summarizes how Li and Thompson account for each notion. In essence, they distinguish ‘subject’ from ‘topic’ via a semantically based criterion which basically says a subject has to be semantically or selectionally involved with the predicate while a topic does not have to be thus restricted. The subject of a sentence in Mandarini is the noun phrase that has a ‘doing’ or ‘being’ relationship with the verb in that sentence. The precise nature of this relationship depends on the semantic makeup of the verb. (Li and Thompson 1981: 87) While their definition of ‘subject' may sound a bit too broad, their definition of ‘topic’ is more vague and intuitive: basically they say that semantically ‘topic’ is what the sentence is about, and that functionally it ‘sets a spatial, temporal, or individual framework within which the main predication holds’ (Li and Thompson 1981: 85), a ‘ Madarin is a major dialect of Chinese. In this dissertation, the term Chinese is used to refer to this major variety of Chinese, as is the general practice in the literature. 40 functional characterization which was initially proposed in Chafe (1976:50). However, they are specific in laying out the following characteristics that they maintain that a tOpic has. First, a topic has to be definite or generic in the sense that it ‘always refers either to something that the hearer already knows about...or to a class of entities’ (Li and Thompson 1981: 85). In comparison, a subject does not have to be definite. Secondly, a topic always occurs in the sentence initial position (unless it is preceded by a connector that links it to the preceding sentence) whereas a subject does not need to be so. Therefore, in their account, time phrases and locative phrases which occur at the beginning of a sentence are all considered topics (ibid., p. 94). Lastly, a topic can be optionally followed by a pause or pause particle? which separates it from the rest of the sentence. On the basis of these distinctions between ‘subject’ and ‘topic’, Li and Thompson classify simple declarative sentences in Chinese into the four basic types given below (examples (2.14) through (2.16)are from Li and Thompson 1981:87-93). (A) Sentences with both subject and topic (2.14) Nei zhi gou wg yijing kan guole. that. Cl. dog I already see ___ Asp. ‘That dog I have already seen.’ 2 From the functional perspective, pause particles can be considered topic markers. They are particles such as a, ,ya, ne, or ba. 41 Li and Thompson identify nei zhi gou ‘that dog’ as the topic and we ‘I ‘ as the subject of the sentence. (B) Sentences in which subject and topic are identical (2.15) W9 xihuan chi pinguo. I like eat apple ‘I like to eat apples.’ (C) Sentences with no subject (2.16) Nei ben shu chuban le. that Cl. book publish Asp. ‘That book, (someone) has published it.’ Li and Thompson claim that nei ben shu ‘that book’ in (2.16) is the topic but not the subject, because it is not in a doing relationship with the verb chuban. In other words, the book does not publish itself; someone or some institution publishes it. (Li and Thompson 1981:89) Similarly, Li and Thompson consider sentences in (2.17) are sentences involving topics only with no subjects. (2.17)a. Fangzi zao hao le. house build finish Asp. ‘The house, (someone) has finished building it.’ b. Yifu tang wan le. cloth iron finish Asp. ‘The clothing, (someone) has finished ironing it.’ c. Fan zhu jiao le yi-dian. rice cook burnt Asp. a bit ‘The rice, (we) burnt it a little bit.’ 42 (D) Sentences with no topic Li and Thompson distinguish two types of sentences with no topics. The first type is one where the topic (i.e. the phrase representing the topic) is omitted because it is understood (and recoverable if necessary) from the linguistic context in which the sentence occurs. (2.18) and (2.19) are the examples provided by Li and Thompson (1981: 90) for this type. (2.18) A: Ni kan guo Lisi meiyou? you see Asp. Lisi not ‘Have you seen Lisi?’ B: Mei kan guo. not see Asp. ‘(I) haven’t.’ (2.19) A: Juzi huai le ma? orange spoil Asp. Q‘ ‘Are the oranges spoiled?’ B: Huai le. spoiled Asp. ‘(They) are spoiled.’ Li and Thompson maintain that topics are omitted in both of B’s responses because they are understood from the context since they have been previously introduced in A’s remarks. The second type of sentences with no topics, according to Li and Thompson, is one in which ‘no noun phrase is definite, or in which the definite or generic noun phrase 3 Q is a question marker. When it is attached to the end of a statement, it make that sentence into a yes-no question. 43 is not what the sentence is about’ (Li and Thompson 1981: 91-92). The sentences in (2.20) are Li and Thompson’s examples of this type (ibid., p.91): (2.20)a. Jin lai le yi ge ren. enter come Asp. one CL. person ‘A person came in.’ b. You ren zai da dianhua gei Zhangsan. exist person Asp. hit telephone to Zhangsan ‘Someone is making a phone call to Zhangsan.’ c. Xia yu le. descend rain Asp. ‘It is raining.’ However, Li and Thompson do not consider these sentences subjectless. They claim that ‘in such sentences, the subject is usually an indefinite noun phrase, which cannot occur in sentence—initial position and cannot be a topic’ (ibid.). The reason they give for identifying the postverbal NPs as subjects is that they have a ‘doing’ relationship with the corresponding verbs in the sentences. Whereas Li and Thompson’s analysis has the advantage of recognizing two distinct grammatical categories of ‘subject’ and ‘topic’, their definition of ‘subject’ is not without problems. First of all, ‘an NP that has a doing or being relationship with the verb’ is hardly a definition of ‘subject’, given that more than one NP (an object, for instance) could share such a relationship with the verb. It is unclear as to how such a ‘doing’ or ‘being’ relationship should be interpreted or understood. For instance, one may wonder why those sentences in (2.16) and 44 (2.17), repeated below, are considered subjectless by Li and Thompson, given their own definition. One could claim that the sentence initial NPs in (2.16) and (2.17) are subjects since they are selectionally related to their predicates, and therefore may be characterized as having a ‘doing’ or ‘being’ relationship with the verb. Yet, they are denied subjecthood in Li and Thompson’s analysis. (2.16) Nei ben shu chuban le. that Cl. book publish Asp. ‘That book, (someone) has published it.’ (2.17)a. Fangzi zao hao le. house build finish Asp. ‘The house, (someone) has finished building it.’ b. Yifu tang wan le. cloth iron finish Asp. ‘The clothing. (someone) has finished ironing it.’ c. Fan zhu jiao 1e yi—dian. rice cook burn Asp. a bit ‘The rice, (we) burnt it a little bit.’ From Li and Thompson’s classification of the four sentence types given above (not by their definition of ‘subject’ ), it appears that what they mean by a ‘doing relationship with the verb’ is simply that the subject is the NP with the semantic role of ‘agent’. That seems to be the only reason why sentences in (2.16) and (2.17) are all considered by Li and Thompson as subjectless, because, the doer of the action (or the NP carrying the agent role) does not appear in these sentences. As a matter of fact, the way these sentences are translated by Li and Thompson seems to suggest so. 45 According to Li and Thompson, sentences in (2.16) and (2.17) are topic and comment constructions without subjects. They do not consider these sentences passive sentences in Chinese, claiming that passive sentences in Chinese are limited to those involving the use of the bei phrase“, roughly the equivalent of the English by phrase in passive constructions. My objection to Li and Thompson’s analysis of (2.16) and (2.17) has to do with their restricting the passive sentences in Chinese to those with the bei phrase. Sentences (2.16) and (2.17) should be characterized as passive sentences both syntactically and semantically. The bei phrase is optional rather than obligatory. For instance, a bei phrase can be inserted in these sentences without changing their general structure or meaning, except ‘ Here is how Li and Thompson define passive sentences in Chinese: ‘The term passive in Mandarin is generally applied to sentences containing the coverb bei with the following linear arrangement: NP1 bei NP2 verb For example: Ta bei jiejie ma le. s/he BEI elder-sister scold Asp. ‘S/He was scolded by (his/her) sister.’ This type of construction has the object noun phrase, that is, the thing or person affected by the action of the verb, in sentence-initial position. This direct object noun phrase is followed by the passive coverb bei, which introduces the agent of the action. We will call this the bei noun phrase. (Li and Thompson 1981:492) 46 to provide the specific informtion on the agent of the action, as in the sentences below: (2.16)’ Nei ben shu bei xuexiao chuban le. that Cl. book BEI school publish Asp. ‘That book has been published by the school.’ (2.17)’a. Fangzi bei San. Shu zao hao le. house BEI third uncle build finish Asp. ‘The house was built by Third Uncle.’ b. Yifu bei wo tang wan le. cloth BEI I/me iron finish Asp. ‘The clothes were ironed by me.’ c. Fan bei ta zhu jiao le yi—dian. rice BEI him/her cook burn Asp. a bit ‘The rice was burnt a little bit by him/her.’ Since these sentences can be considered passive sentences, the sentence initial NPs with the semantic role of theme or patient may be construed as subjects of the sentence. Though whether a sentence should be analyzed as passive or not is a separate issue to the grammatical notion of ‘subject’, the problems involved with the analysis of (2.16) and (2.17) suggest that Li and Thompson’s definition of ‘subject’ as the NP sharing a doing or being relationship with the verb appears to be too vague or ambiguous to be well-defined. To summarize, though Li and Thompson’s definition helps to distinguish ‘subject’ and ‘topic’, it still lacks what is needed for the proper identification of subject in Chinese. I have so far argued that the major definitions of ‘subject’ in Chinese are still problematic. Subject 47 defined as 'topic' or designated as the sentence—initial NP has the undesirable consequence of oversimplifying and overcomplicating the syntactic structure of Chinese at the same time. In order to better characterize the notion of 'subject', it is necessary to have a clear understanding of the properties of ‘subject’. To do so, I now turn to a 'milestone' article by Keenan (1976). 2.4 Toward a universal definition of subjects: Keenan’s 'Subject Properties List' (SPL) In an attempt to provide a definition of the notion of 'subject' which would be universally valid in the sense that it will enable us to identify the subjects of arbitrary sentences from arbitrary languages, Keenan (1976) proposes a complex if 'cumbersome' (his own word) definition in the form of a Subject Properties List (SPL). The universality of the subject properties on this list is justified by the fact that they are abstracted from a large and diverse corpus of data collected from different languages. Keenan points out that ‘such a definition is needed in universal grammar in order for the many universal generalizations which use this notion to be well defined’ (1976:305). If different criteria are used to identify subjects in different languages, he argues, then ‘subject’ is simply not a universal category and as a consequence all those universal generalizations stated in terms of that 48 notion cannot be understood as generalizations at all (ibid.). Keenan distinguishes between basic subjects and non- basic subjects, the former being the subjects of the ‘semantically basic sentences’, which is what he calls ‘a privileged subset of sentences in any language’ (ibid., 306). The subjects of non-basic sentences are defined as those NPs, which ‘present a clear preponderance of the properties characteristic of b—subjects’ (ibid., 307). He also maintains that subjecthood of an NP in a sentence is a matter of degree, and that subjects in some languages will be more subject—like than those of other languages in the sense that they will in general, 'present a fuller complement of the properties which universally characterize b-subjects' (ibid.). This subject property list that Keenan proposed is composed of four sections: (a) autonomy properties; (b) case-marking properties; (c) semantic role and (d) immediate dominance. Each of these in turn incorporates sub-lists of properties, altogether 30 of them. If one NP in the sentence has a clear preponderance of the subject properties, then it will be called the subject of that sentence. In other words, a subject in any language can be understood as the combination of a subset of Keenan's Subject properties. The properties of subjects specified on Keenan’s SPL can be classified into three types: morphosyntactic 49 properties, semantic properties, and pragmatic properties. In the following sections, I will discuss these subject properties and illustrate how they help pick out the subject in Chinese sentences. 2.4.1 Morphosyntactic properties The morphosyntactic properties of subjects include properties related to overt coding, syntactic behavior and sentential configuration. 2.4.1.1 Coding properties The coding features that are attributed to subject properties in Keenan’s SPL are primarily of three kinds: case—marking, cross—referencing, and word order. In a great many languages, the coding features of ordinary main clauses clearly mark a subject grammatical relation. For instance, in Ancient Greek, subjects of ordinary main clauses occupy no definite position, but are for the most part clearly marked by nominative case, which always and only appears on the subjects of finite clauses. Furthermore, verbs agree with their subjects in person and number via an extensive system. (Andrews 1985: 104). Subject can also be coded in terms of its unmarked sentence position with respect to the verb. Since all languages have a linear order of constituents, position is one possible device for all of them. In English, for example, subjects of ordinary main clauses are primarily 50 marked by the coding feature of preverbal position. They are also indicated to a limited extent by case marking and subject-verb agreement. Chinese on the other hand, is known for its lack of inflectional marking to indicate grammatical relations, thus case marking and agreement do not apply to Chinese. However, it is widely recognized that Chinese is fundamentally a SVO language where a subject has the tendency to occur preverbally in unmarked cases (Tsao 1977, Light 1979, Sun and Givon 1985, Wei 1989). This phenomenon conforms to what SPL specifies: basic subjects are normally the initial occurring NP in basic sentences (Keenan 1976: 319). Though subjects are not always the leftmost occurring NPs, this generalization holds in many basic sentences. For example, it would have the effect of picking out the subjects in (2.18), which are underlined accordingly. (2.18)a. Mama bu xihuan zhe ge ren. mother not like this Cl. person ‘Mother does not like this person.’ b. Wg hua 1e qian xiangshou. I spend Asp. money enjoy ‘I spent money and had a good time.’ [Example (2.12b) from Li and Thompson 1976:478] 2.4.1.2 Behavioral and control properties Behavior and control properties refer to general behavior characteristics that subjects share with other NPs in grammatical processes such as coreferencing (e.g. reflexivization, pronominalization), deletion (e.g. equiaNP 51 deletion, imperative-deletion) and movement (e.g. relativization, clefting, passivization, raising, etc.). All these processes appear to involve subject as a controller in English. Keenan’s SPL states that, cross— linguistically, basic subjects tend to have the following behavioral and control properties: a). They are always among the possible controllers of stipulated co-reference; b). They in general can control reflexive pronouns; c). They are among the possible controllers of coreferential deletions and pronominalizations; d). They are the easiest NPs to stipulate the coreference of across clause boundaries. (Keenan 1976:315—16) Now Consider the following sentences in Chinese: (2.18)b. W9 hua le qian xiangshou. I spend Asp. money enjoy ‘I spent money and had a good time.’ (2.19)a. Zhangsan ti le qiu. Ta ye da le ren. Zhangsan kick Asp. ball he also hit Asp. person ‘Zhangsan kicked the ball, he also hit someone.’ b. Zhangsan ti le qiu ye da le ren. Zhangsan kick Asp. ball also hit Asp. person ‘Zhangsan kicked the ball and hit someone.’ (2.20) Zhangsan xihuan ta-ziji. Zhangsan like himself 'Zhangsan likes himself.‘ (2.21)a. (Ni) Qu. you go ’Go.’ b. (Ni) Bie qu. you not go ‘Don't go.’ Let me assume that the underlined preverbal NPs in (2.18) - (2.21) above are the subjects of the sentences, 52 given that Chinese is a SVO language, and that subjects tend to occur preverbally. I will now illustrate that these subjects do have the behavior and control properties that Keenan’s SPL specifies. In other words, these properties help predict the subjecthood of an NP in a sentence. (2.18b) shows that the subject controls coreferential deletion in the serial verb construction, which normally involves at least two verbs. Here, in (2.18b), the two verbs are hua 'spend' and xiangshou 'enjoy'. The sentence literally means ‘I spend money and (I) enjoy’. The subject, WC ‘1’, is both the controller and the target of deletion, as indicated by a blank where the coreferential deletion controlled by the subject occurs. (2.18b)Wg hua le qian ___ xiangshou. I spend Asp. money enjoy ‘I spent money and had a good time.’ Now look at the sentences in (2.19). (2.19a) shows that the subject stipulates the coreference across clause boundaries and controls coreferential pronominalization: the subject pronoun ta ‘he’ of the second clause is used to refer back to the subject of the prior sentence, Zhangsan, a proper name. (2.19) also show that subject controls conjunction reduction. In fact, the ungrammaticality of (2.21) seems to suggest that identical NP deletion in a coordinate structure is limited to the subject. Compare (2.19), which is repeated here, to (2.22) below: 53 (2.19)a. Zhangsan ti le qiu, ta ye da le ren. Zhangsan kick Asp. ball he also hit Asp. person ‘Zhangsan kicked the ball, he also hit someone.’ b. Zhangsan ti le (Ihl ye da le ren. Zhangsan kick Asp.ball also hit Asp. person ‘Zhangsan kicked the ball and hit someone.’ (2.22)a. Wo mai 1e Ii. ni chi le 11. I buy Asp. pear you eat Asp. pear ‘I bought the pear; you ate the pear.’ b. *Wo mai le ni chi le, 1i. I buy Asp. you eat Asp. pear ‘I bought and you ate, the pear.’ In (2.19b), the two identical subjects ta ‘he’ are reduced to one, and the sentence is OK. However, if identical NPs that are non—subjects are reduced in the same way, such as the identical objects 1i 'pear' in (2.22b), the sentence becomes unacceptable. Now, let us take a look at (2.20), repeated below: (2.20) Zhangsan. xihuan ta—ziji. Zhangsan like himself 'Zhangsan likes himself.‘ Keenan’s SPL specifies that ‘b—subjects in general can control reflexive pronouns’ (Keenan 1976:313). In other words, the subject is what can serve as the antecedent of a reflexive. This is apparently true in (2.20), where Zhangsan, is the antecedent that controls the reflexive pronoun ta-ziji, ‘himself’. In terms of word order, subjects tend to occur before their co-referential reflexives. According to Keenan, this preferred word order follows from the autonomous-reference 54 property that is attested in subjects of many different languages. Cross-linguistically, subjects tend to be autonomous in reference, and 'the referent of the subject must be determinable independently of that of a following NP' (Keenan 1976: 313). Thus in English, the sentence 'Mary admires herself’ is OK but 'Herself admires Mary' is not where.Mary is coreferential to herself. SPL (Keenan 1976: 321) states that ’subject normally expresses the addressee phrase of the imperatives.‘ This is apparently true in Chinese. (2.21), repeated below, shows that subjects of imperative sentences, whether overtly expressed or omitted, are NPs denoting the addressee(s). If the subject does not refer to the addressee(s), then the sentence cannot be interpreted as imperative. (2.21)a.(Ni) Qu. you go ’Go.’ b. (Ni) Bie qu. you not go ‘Don’t go.’ So far I have illustrated that Chinese subjects share major coding and behavior properties that are attested in Keenan’s universal subject list. Thus I conclude that ‘subject’ is a notion needed in Chinese grammatical theory in order to describe certain syntactic processes. 55 2.4.2 Semantic properties Semantic properties of subject specified on the SPL (Keenan 1976:312) include the following: ‘indispensability’, ‘independent existence', ‘absolute reference’ and ‘semantic role'. 2.4.2.1 Indispensability: Basic subjects cannot be eliminated from a sentence without rendering the sentence incomplete. Thus (2.23b) is ungrammatical without a subject. (2.23)a. John hunts lions (for a living). b. * hunts lions (for a living). In Chinese, though subjects tend to be suppressed whenever understood from the context, such as in imperative sentences, basic sentence subjects cannot be omitted unless recoverable. The equivalent of (2.23) in Chinese would have almost the same effect as in English, as illustrated in (2.24). If the subject is omitted, this sentence is incomplete unless understood as an imperative sentence. (2.24) a. Yuehan 1m; shizi (wei sheng). John hunt lion (for living) ‘John hunts lions (for a living).’ b. Bu shizi (wei sheng). hunt lion (for living) ‘Hunt lions (for a living).’ 2.4.2.2 Independent existence: The SPL specifies that ‘the entity that a b-subject refers to (if any) exists independently of the action or 56 property expressed by the predicate’ (Keenan 1976: 312). For non-subjects, it says, this seems less true. For example, in (2.25) below, the existence of the object ‘a new song’ is dependent on the act of writing, whereas the existence of the subject, ’the singer’, is not. (2.25) The singer wrote a new song. In a sense, this property helps to explain why semantic features such as animacy and agency are closely associated with subjects more than any other grammatical functions. 2.4.2.3 Absolute reference: The SPL states that ‘in the overwhelming majority of cases, if a b-sentence is true then we understand that there is an entity (concrete or abstract) which is referred to, or has the property expressed by, the b—subject’ (Keenan 1976: 317). Thus if (2.26) is true, it requires that there exist someone named JOhn, but not the report. (2.26) John owes his professor a report. 2.4.2.4 Semantic role: agency Linguists have been concerned with the nature of the relation between an NP and the element to which it looks for its semantic role. A common position in the literature regarding the semantic role of an argument in a sentence is that ‘semantic roles have no a priori or independent 57 existence but rather are entailed by the lexical semantics of individual predicates’ (O'Brady 1987:56). A verb such as ‘employ’, for instance, has a meaning that entails the existence of an ‘employer’ and an ‘employee’, while ‘dance’ entails a ‘dancer’ and so on. In other words, the semantic role of an argument in a sentence is determined by the semantic make-up of the predicate. Along the same line, the SPL attests that the semantic role of the referent of a basic subject is predictable from the form of the main verb. Though a subject may potentially carry any specific semantic role according to the semantic—make up of its predicate, more often than not, subject expresses the agent of the action, if there is one (Keenan 1976: 321). Indeed, there is a cross—linguistic tendency to mark the agent as the subject of a sentence. 2.4.3 Pragmatic properties Pragmatic properties have to do with the ways speakers categorize their information based on the assumptions of their shared knowledge with the listeners. They are related to pragmatic notions such as ‘definiteness’, ‘topicality’, ‘referentiality’, ‘given vs. new information’, etc. A pragmatic property can be identified on the basis of its relationship to the discourse context. 58 2.4.3.1 Autonomous reference: SPL specifies that 'the reference of a b(asic) subject must be determinable by the addressee at the moment of utterance' (Keenan 1976: 313). This means that the reference of the subject NP must not be dependent upon the reference of an NP that follows the subject. This property is largely manifested in the reflexive binding in syntax, where the reflexive tends to be coded after its co—indexed antecedent, not before it. 2.4.3.2 Topicality: Topicality is an area where the two notions of ‘subject’ and ‘topic’ often overlap. SPL states that ‘b— subjects are normally the topic of the b—sentence, i.e. they identify what the speaker is talking about’ (Keenan 1976: 318). I have shown in the previous chapter that subject is frequently defined via the notion of topic. Studies on the relation between ‘subject’ and ‘topic’ are numerous. Topics being an elusive notion, researchers seem to agree more upon what properties they do not have rather than what they do have. From my research, the current consensus seems to be that subjects are obligatory arguments which are syntactically and semantically integrated into the predicate—argument structure of the clause, while topics are normally represented by NPs which do not necessarily have selectional restrictions with the 59 verbs in the proposition. Often, the relation between the topics and the rest of the predication is taken to be ‘a matter of pragmatic construal’ (Lambrecht 1994: 118). Chinese has been described as a topic prominent language, where the overlap between topic and subject is common. When subject and topic conflate, the subject becomes highly topical. What could be construed as topic becomes a constituent of the predication proper, rather than being only loosely associated with the predication. 2.4.3.3 Referentiality: An NP is referential if the speaker intends for it to refer to a particular entity which exists within a particular universe of discourse, with continuous identity over time (Givon 1978: 293, and LaPolla 1990: 19). SPL indicates that subjects tend to be highly referential. Since personal pronouns, proper nouns and demonstratives are highly referential by nature, they can always occur as subjects. 2.4.3.4 Given information: The SPL also states that subjects tend to be definite and express 'given information'. That is, the entity referred to by the subject is normally known to both the speaker and the hearer. Clark (1973) points out that when speaker—hearers engage in talk, they abide by a 'Given/New contract': the speaker is responsible for marking 60 syntactically as 'given' that information that he assumes to be known by the listener, and marks as 'new' what he assumes not to be known by the listener. Givenness as a subject property is apparently related to the topicality that subject tends to exhibit. Topicality is established on the basis of shared common knowledge and given information. Givenness is also related to the unmarked preverbal position of subject in a sentence. Studies on discourse analysis and information structure (e.g. Chafe 1994, Lambrecht 1994) suggest that new information overwhelmingly follows given information in discourse packaging. 2.4.3.5 Definiteness: Definiteness represents another way that NPs can be categorized in communication. In the communicative situation, if a speaker assumes that the listener knows and can identify the referent she or he has in mind, she or he would mark this item as definite. For instance, English speakers can use the definite article or a demonstrative pronoun to mark an NP as given and definite when they presume the referent to be known to and uniquely identifiable by the addressee. In Chinese, definiteness is marked by demonstrative pronouns or by word order. Definite NPs tend to occur preverbally and indefinite NPs postverbally (see Li and Thompson 1976). 61 SPL specifies that subjects of basic sentences tend to be definite (Keenan 1976: 319). In some languages, Malagasy, Tagalog and Philippine languages, subjects of basic sentences must be definite (ibid.). In Chinese, definiteness is a prominent feature of subjects. In the 50’s, there were debates in the Chinese linguistic community on whether indefinite NPs can serve as subjects in Chinese. The general agreement reached nowadays is that indefinite NPs can serve as subjects, but the majority of subjects in Chinese tend to be definite. So far I have discussed the major subject properties on Keenan’s SPL, which are summarized as follows: Syntactic properties: 0 unmarked position: preverbal O behavior: subject controls reflexive binding, imperativization, co-referential NP deletion, conjunction reduction Semantic properties: 0 indispensability 0 independent existence 0 absolute reference 0 selectional relation with the predicate 0 semantic role Pragmatic properties: 0 autonomous reference 0 referentiality 0 topicality 0 givenness 0 definiteness 62 In my current research, I found that, except for the morphological marking properties, almost all other properties on the SPL are attested in Chinese. I will further discuss some of these properties in the rest of the dissertation. Given that fact that crosslinguistic subject properties are largely attested in Chinese, I conclude that 'subject' is an important notion in Chinese, rather than a marginal one. 2.5 Critique of Keenan’s approach Keenan's approach can be characterized as a prototype clustering approach to the grammatical relation of ‘subject’, given the fact that it treats 'subject’ as a 'cluster concept' (his own word) and a prototype, with many features to determine subjecthood, even when none of the features is by itself necessary or sufficient. It differs significantly from previous definitions of 'subject' in the following aspects. (A) For the first time, 'subject' is no longer treated as a single-dimension or single-condition concept. Rather, it is viewed as a multi-factor notion, with attributes that are syntactic, semantic and pragmatic in nature. Results of typological studies of subjects apparently support this perspective. For instance, Farrlund (1988) concludes from his typological investigation of subjects cross-linguistically that subjects typically code the semantic information of agent and/or the pragmatic 63 information of topic. From my own research, I have reached the conclusion that if 'subject' is treated as a single- condition concept, the notion of 'subject' in universal grammar would collapse. If we isolate the semantic and pragmatic attributes from the concept of 'subject', what remains would be the morphosyntactic attributes of subject. It is precisely in terms of these properties that Philippine languages such as Tagalog (see Schnachter 1976) would depart significantly enough from the Romance language model to render the notion of 'subject' as an essential component of clause syntax unviable in such languagess. .As a consequence, the notion of 'subject' in universal grammar would break down. This is exactly where the problem lies for the discrete single-criterion approaches that I have discussed in their characterization of 'subject'. Just as Givon (1995:228) points out, these approaches would lead to ‘considerable mischief’ in typological comparisons. NPs in languages which seem to conform to the single ad hoc criterion are designated as subjects, while NPs in languages where this single criterion does not play a significant role would be treated as non—subjects, despite sSchachter (1976), in his examination of Tagalog, concludes that ‘there is no single syntactic category in Philippine languages that corresponds to the category identified as the subject in other languages’ (p. 513). Rather, the syntactic properties of the subject in Tagalog are divided between the topic and actor in a clause, with a few subject-like properties reserved for the intersection of the topic and the actor (ibid.) 64 their strong candidacy for subjecthood if other criteria are taken into account. An example of this type of typological error can be seen in Li and Thompson’s (1976) typological study of 'subject'. Here, Chinese is treated as a non—subject- prominent language. A close look at that study will show that the criterion based on which this classification is made is largely morphological (i.e. case—marking and verbal agreement). I argue that this criterion of morphology by itself as an indicator of subjecthood may be questionable. Instances of mismatches between grammatical relations and morphology have been described in typological studies. For example, in the South Caucasian language Georgian, agreement markers which indicate features of indirect object will instead indicate features of subject in a construction called ‘inversion’. Here, markers which otherwise indicate features of the subject instead indicate features of the direct object (see Kathman 1995). In comparison, Keenan's proposal of the SPL as a means to define the universality of subjecthood has a clear theoretical advantage. Keenan’s new approach has the benefit of avoiding potential problems associated with those single-criterion analyses. His is a unified treatment of the universal notion of 'subject', incorporating separate but parallel morphological, syntactic, semantic and pragmatic features into the subject 65 prototype. This way, the notion of 'subject' as a universal concept would be applicable and cross—linguistic data would be comparable. (B) The second aspect in terms of which Keenan's approach differs from other definitions is that 'subject' is treated as a relative rather than an absolute notion. This is in View of the fact that not all properties in the SPL will be associated with any one NP in a clause in any given language. In addition, empirical accounts of grammatical relations have shown that formal subject properties such as behavior and control properties are not equally distributed cross—linguistically. Not all properties in the SPL will be associated with any one NP in a clause in any given language, nor are all properties attested in all languages. As a result, it is virtually impossible to isolate any combination of the subject properties which is both necessary and sufficient for an NP in any sentence in any language to be the subject of that sentence. Thus Keenan concludes that 'subject' cannot be treated like a prime number, and subjecthood of an NP in a sentence is a 'matter of degree'. Subjects in some languages will be more subject-like than those of other languages in the sense that they will in general, 'present a fuller complement of the properties which universally characterize b(asic) subjects' (Keenan 1976: 307). (C) A third aspect in terms of which Keenan’s approach differs from previous definitions is his distinction of two 66 types of subjects: basic and non—basic subjects. Keenan (1976: 307) states that basic subjects (coded by him as b- subjects) are the subjects of the 'semantically basic sentences', which is ‘a privileged subset of sentences in any language.‘ The subjects of non—basic sentences are those NPs which ‘present a clear preponderance of the properties characteristic of b(asic)-subjects'. This distinction is based on the observation that the subjects of basic sentences tend to display more subject properties than the subjects of non-basic sentences. Put another way, there is a gradation of subjecthood even within the same language: subjects of simple sentences tend to be more subject—like (i.e. more prototypical) than subjects of complex sentences. The procedure that Keenan uses for defining language- specific subjects using the SPL are as follows: (A) Defining basic sentencesb; (B) Defining basic subjects: Basic subjects of a given language are defined to be any NPs of basic sentences of that language that manifest a majority of the properties of the universal SPL. 6Keenan's definition of 'basic sentence' is given below: For any L, a. a syntactic structure x is semantically more basic than a syntactic structure y if, and only if, the meaning of y depends on that of x. That is, to understand the meaning of y it is necessary to understand the meaning of x. b. a sentence in L is a basic sentence (in L) if, and only if, no (other) complete sentence in L is more basic than it (Keenan 1976:307). 67 (C) Establishing a language specific SPL: Once the basic subjects are determined for a particular language, one can establish a SPL which is associated with the basic subjects of that language. This list may include properties that are idiosyncratic to the basic subjects in that language. (D) Defining non—basic subjects The non-basic subjects are defined in terms of the language specific SPL. They include those NPs which ‘present a clear preponderance of the properties characteristic of b(asic)subjects’ (Keenan 1976: 307). To summarize, what the SPL seems to have achieved is to provide us with a description of different characteristics of subjects as manifested in different languages. What Keenan has given us is an explicit clustering framework based on which we could examine each language individually and come up with a language-specific definition of subject, which would, in some degree, share the universal subject properties. In the rest of the dissertation, I will assume a cluster—prototype approach to the problem of subject, though the mechanism of operation of subject assignment is different. Keenan's article, written over 20 years ago, has had a profound impact on the study of ‘subject’ as a universal notion. From my research, it is by far the closest approximation to a universal definition of 'subject'. Though there are problems with Keenan's 68 definition in the form of a Subject Property List (SPL) to be discussed in the next section, his theory provides us with some fundamental assumptions and principles with regard to the notion of subject that are still being widely adopted today in the functionalist camp (e.g. Givon 1996). Major objections to the validity of Keenan’s SPL as the universal definition of 'subject' are raised in Johnson (1977). Johnson's major objections are the following: (A) Objection against the definition of 'basic sentences' as a formal component of the actual definition Johnson points out Keenan's concept of basic sentences 'does not correlate in any straightforward way with syntactic simplicity' and that this definition of basic sentence by Keenan 'has the unfortunate consequence that any paraphrase of a given sentence is more basic than S' (p. 676). (B) Objection against ‘clear preponderance’ (CP) used as a general criterion in the statement of Keenan’s definition. According to Johnson, there are at least two interpretations: On the absolute reading, this phrase would mean that for an NP to be a b-SUBJ it must have at least a majority of the properties on the SPL, i.e. at least 16 properties. On the 'relative' reading, the phrase would mean that for an NP to be a b-SUBJ it must have more of the SUBJ properties than any other NP in its clause. (Johnson 1977:677) Johnson's major objection with both readings of CP is Keenan's 'lack of necessary and sufficient criteria' for 69 the identification of subject. He argues that the relative interpretation is inconsistent with Keenan’s assumption that the same defining criteria should be used in every language in order for subject to be well defined as a universal category. If an NP merely needs more properties than any other NP in its clause to be, e.g., a b-SUBJ, then the definition would allow the case in which SUBJs in a language J are defined by a subset S of the SPL and SUBJs of some distinct language M are defined by another subset of the SPL, where S and T are disjoint. That is, totally different criteria could be used in the definition of SUBJ in J and M. (Johnson 1977: 768) However, in my view, Johnson’s concerns might be unwarranted, since the disjoint cases that he specifies may be cases where a number of the defining subject properties simply do not apply, rather than cases where criteria for subjecthood conflict. For instance, in Chinese, the morphosyntactic coding properties (e.g. agreement features) do not apply. As to the cases where criteria or properties conflict with each other (i.e. one NP has some properties and a different one has others), the problem can be resolved by distinguishing the relative weight of subject properties, that is, some properties may be more vital (or play a more important role) in defining subjecthood. This issue will be dealt with in the next chapter, where I propose that subject can be treated as a prototype notion with properties that can be ranked. 70 My View regarding Keenan's SPL is that it is of undoubtedly tremendous importance in the theory of linguistics because of the insights for the understanding of the properties that subjects share in many different languages. It has also provided a convenient point of departure for further research on the notion of ‘subject’ and other grammatical relations. The fact that not all of the subject properties apply in every language does not nullify subject as a universal grammatical category defined in terms of prototypical subject properties. While different languages may display different subject properties, within individual languages, subject properties can be used to distinguish the subject from other grammatical categories in the sense that the subject is more accessible to these properties than any other grammatical categories. The possibility is ruled out that in a language, the object displays a certain subject property, but the subject does not. 2.6 Summary Given the research that I have done, I have come to the conclusion that many previous researchers have expressed that there is no universal definition of subject (Bavin 1980; Platt 1971; Keenan 1976; Van Valin 1977; Foley & van valin 1977 & 1984; Perlmutter 1982; Comrie 1989; LaPolla 1990) in the sense that it is virtually impossible to set up necessary and sufficient criteria for defining 71 universal subjects. Based on previous research and given the diverse cross—linguistic data, I have come to the conclusion that there are only a limited number of options to deal with this dilemma. One could claim that ‘subject’ is simply not a universal category. However, I believe there is a better choice than that. One could still maintain the universality of subjecthood, as is so commonly assumed in the traditional and current linguistic theories, by treating subject as a prototype concept with a variety of properties that may be manifested in different degrees in different languages. Subjects of individual languages can be defined individually within the range of these subject properties. In the next chapter, I will further explore issues surrounding subject as a prototype concept and discuss how subjects in Chinese can be properly identified via prototypical subject properties. 72 Chapter 3 Subject as a proto-type concept 3.0 Introduction In the previous chapter, I examined the problem of defining 'subject‘ by reviewing various definitions that have been proposed on this notion. My research of previous studies indicates that there is as yet no universal definition of ‘subject’ due to the fact that subjects vary in their properties within and across languages. Furthermore, it is impossible to give a strict universal definition of ‘subject’ in terms of necessary and sufficient conditions. As a consequence, we are faced with two major research questions: Is subject a universal grammatical category (given the fact that no universal definition is applicable)? If the answer is yes, how do we go about defining it in such a way that it is applicable cross— linguistically? The focus of this chapter is to deal with the two questions above. In section 3.1, I cover the basic notions that will be used in the rest of this dissertation, terms such as 'predication', 'predicator', 'arguments', and 'semantic roles'. A clear understanding of these fundamental concepts and how they relate to each other is essential for further discussion of the syntactic and semantic relations of the elements in a clause. 73 In section 3.2, I argue in favor of characterizing subject as a universal grammatical category by presenting a fundamental theoretical assumption that has been regarded as 'an exemplary principle of Universal Grammar' (Droste & Joseph 1991: 29), the predication principle that 'all clauses must have a subject'. After establishing the status of subject as a universal grammatical category, I will then attempt to present a solution to the second question. I propose that 'subject' can be optimally characterized as a proto-type concept, even though it can be hardly defined in terms of necessary and sufficient conditions. Section 3.3 explores and identifies the proto-typical subject properties. In section 3.4, I attempt to establish a subject selection hierarchy based on the relative weight of these prototypical subject properties . Such weighting is necessary because the subject properties are not equally prominent in their contribution to subjecthood. Some are more significant in the sense that they help distinguish ‘subject’ from other grammatical categories. In other words, some properties are more critical than others. Having established the subject selection hierarchy, I will further illustrate in the rest of the dissertation how it helps to predict the appropriate subject assignment in problematic cases in Chinese. 74 3.1 Definition of basic notions In this section, I will introduce a few fundamental notions frequently used in analysis of sentences. These are 'predication', 'predicator', 'arguments', and 'semantic roles'. As pointed out earlier, it is necessary to be clear about how they are used in order to discuss the nature of sentential constituents. The definitions given below are largely based on Functional Grammar (FG) as outlined in (Dik: 1991). 3.1.1 Predication, predicator, and arguments Traditionally, a predication is divided into two parts, subject and predicate, and the subject is notionally ‘what is being talked about’ and the predicate ‘what is said about it'. Thus in (3.1), the subject is the boy and the predicate is hit the ball. (3.1) The boy hit the ball. An alternative View, which is common in current linguistic theories, holds that a predication consists of a predicator and one or more arguments or terms. The predicatorl is the center of the predication. Often lBasic predicators are given in the lexicon, and derived predicates are formed by means of ‘predicate formation rules’ which specify the productive processes of the predicate formation. Examples of derived predicates are: modified predicates such as (be) very happy; causative predicates such as make..happy, let go; predicates derived from terms, such as (be) a doctor. 75 realized by a verb, the predicator presupposes a number of participants——one, two or three, depending on the verb and its semantic type concerned. The participants are called arguments or terms. Predicators are expressions that designate properties or relations; arguments or terms are expressions which can be used to refer to entities. In (3.1), the predicator is hit, and there are two arguments: the boy and the ball. Notionally, the predicator expresses the relationship (here the act of hitting) between the arguments (here the boy and the ball). The two arguments the boy and the ball refer to two entities, which are related to each other through the predicator hit. The structure of this sentence would be: argument—:predicator—-argument. In this dissertation, I adopt this second view of predication in an attempt to recharacterize the notion of 'subject'. What I want to argue for is that subject is functionally the most prominent argument of the . predication. What this means is that a subject has to be an argument of the predicator, unless the predicate verb is one which does not take arguments, verbs such as rain and seem, as in ‘It is raining’ and ‘It seems that Mary has left’. The subject NP has a special status in the sentence in terms of its syntactic behavior and semantic function. By semantic function, I mean the different roles played by participants (referred to by the arguments) in verbally described states of affairs such as action, processes and 76 states. Before I elaborate on this point, I will first discuss how arguments differ from non-arguments and specify how the distinction of ‘arguments’ vs. ‘non arguments’ is related to the characterization of ‘subject’. 3.1.2 Arguments vs. ’satellites’ In Functional Grammar (as described in Dik 1991), predications are construed by combining predicators and terms. Predicators are expressions that designate properties or relations, and terms are expressions which can be used to refer to entities. The terms in a predication are distinguished according to whether they are obligatory or optional for the predication. Obligatory terms are called ‘arguments’. These arguments can be understood in two ways. In the underlying logical-level configuration, a verb’s predicate frame specifies the argument positions of the verb inherently required by its subcategorization. Arguments can be understood as referring to those argument positions. In the clause—level structure, they are the nominals filling in those argument positions. From a functional point of view, what they specify are obligatory participants in the state of affairs characterized in the predication. The predicator and its arguments form what is referred as ‘nuclear predication’, a minimal predication without the modification of adjuncts. These arguments are considered essential in the predication because they are absolutely necessary for defining the 77 state of affairs designated by the predication, even though they may be omitted if clearly understood given the appropriate contextual or situational clues. In contrast, optional terms, called 'satellites', are non-arguments which are viewed as more peripheral in the predication in the sense that they do not specify the obligatory argument positions of the predicator but only provide additional details to the information given in the nuclear predication (formed by the predicator and its arguments). What satellites specify is usually information such as time, location, manner, cause, condition etc. of the states of affairs described in the predications. Example: (3.2) The baker bought a new car yesterday because his old one had broken down. [Example from Dik 1991:252] In this sentence, there is a nuclear predication formed by the predicator bought applied to its two arguments, the baker and a new car. This nuclear predication is modified by two satellites, yesterday and because his old one had broken down, which function as adjuncts of Time and Reason to the nuclear predication of The baker bought a new car. This example also shows that while arguments normally refer to entities, satellites do not. The functional differences between them parallel the distinction between ‘complements’ and ‘adjuncts’ in generative grammar. 78 The distinction between ‘arguments’ and ‘satellites’ appears to be significantly relevant to the characterization of subjecthood. I think there is an important generalization to be made here. Generally speaking, only arguments can function as subjectsz. This is because arguments have a more prominent function in the predication, given that they usually specify sentence participants whereas 'satellites' only add optional details about these participants. 3.1.3 Semantic roles Arguments also differ in their semantic relationships to the predicator and are clearly distinguished from one another through grammatical marking. For example, in the sentence ‘the boy hit the ball’, the distinction between the two arguments is shown by the word order. Switching the positions would alter the semantic relationship of the arguments to the predicate and yield a quite different predication. In Functional Grammar, each term in a predication is characterized by some semantic function. This semantic function specifies the role of the entity to which the term refers within the state of affairs designated by the predication. Theoretically speaking, it is possible to identify a large number of roles played by the terms of a predication. 2 Two exceptions to this requirement are to be specified in Section 3.3.2.2. 79 Among them the common ones are Agent, Patient, Theme, Experiencer, Recipient, Beneficiary, Instrument, Source, Goal, Location and Timei, etc. All these semantic roles relate the arguments of a verb to the meaning of that verb. In the sentence 'the boy hit the ball', for example, the verb hit has two arguments that carry the semantic roles of Agent and Patient respectively. The semantic role of Agent (here the boy) identifies a participant who does or causes something, while the Patient role (here the ball) identifies the participant undergoing the effect of the action. The point to be made here about the semantic roles in relation to subjecthood is that there is no one-to-one correspondence between an underlying semantic role and the subject. Any semantic role could potentially be mapped onto subjecthood given the appropriate syntactic structure in which the NP bearing that role is involved. 3.1.4 Predicate frames In Functional Grammar, a predication which consists of only the predicator and its arguments is called a 'nuclear predication'. By ‘nuclear predication’ is meant the application of a predicator to an appropriate number of arguments of that predicator without satellites. The fundamental structure of the nuclear predication is determined by the combinatory possibilities of the 3 Following the conventions in the FG, all semantic roles henceforth are capitalized. 8O predicator, which are specified in its ‘predicate-frame’. The information that a predicate frame specifies about a predicator includes: (a) its lexical form; (b) its syntactic category; (c) its number of arguments; (d) its selectional restrictions with its arguments; and (e) the semantic roles of the arguments. For example, the predicate-frame for a predicate like English eat would look like (3.3): (3.3) eatv(x1: animate(xl))Ag (x2)pa This predicate frame spells out that eat is a two— place verb, taking as arguments two terms, one of which is animate with the semantic role of Agent, and the other in the function of Patient. If the details are left out, the predicate—frame for eat can be simplified as (3.4), which says that the predicate eat takes two arguments. (3.4) eat (x1)(x2) Verbs may have alternative predicate-frames given the fact that many verbs are both intransitive and transitive (often with a difference in meaning), as in (3.5) and (3.6), where open and run serve as intransitive and transitive verbs respectively. (3.5) a. The door opened. b. He opened the door. 81 (3.6) a. He ran the test. b. He ran in the 100—meter dash. Due to their role of coding the basic semantic and morphosyntactic properties of the predicator, ‘predicate frames’ have been considered the most basic components for the construction of predications. 3.2 The predication principle The prototype approach that I adopt here to the categorization of subject follows from the theoretical assumption that subject is a universally applicable notion. However, the universality of subjecthood is not so obvious and deserves elaboration. Subject as a universal category can be maintained via a universal property of subject (as expounded in Keenan 1976) that subjects are indispensable in basic sentences: each basic sentence involving a predication requires a subject, expressed or unexpressed. This assumption is coded as ‘the predication principle’ and regarded as 'an exemplary principle of Universal Grammar' (Droste & Joseph 1991: 29). This principle has long been reflected in the traditional view that the sentence consists of two parts, subject and predicate, the subject being notionally 'what is being talked about' and the predicate 'what is said about it'. It is also one that is implicit in much of modern theoretical linguistics. The adoption of this 82 subject—predicate analysis of the sentence is clearly shown in Chomsky’s Syntactic Structures (1957:26), where the rule of S —> NP + VP states that the sentence consists of a noun phrase and a verb phrase, which corresponds closely to the traditional subject and predicate. In his Knowledge of Language (1986), Chomsky maintains that 'a predicate (in particular, a VP) must have a subject' (p.93). This principle is arrived at on the basis of generative analysis of many syntactic constructions, most of which involve interclausal relations: control of nonfinite complements, deletion of coreferential arguments in conjoined clauses, the relation between a relative clause and its head noun in the main clause, reflexivization across clause boundaries, etc. To give an example, let us look at the arguments for the predication principle in the generative analysis of infinitive clauses. The earliest treatment of null subjects in infinitival clauses was advanced on the basis of examples like (3.7). (3.7) John wanted [_ to shave himself]. Even though infinitive clauses generally have no lexicalized subject, a syntactic subject is assumed to be invariably present. It is argued that in infinitivals, the presence of the subject can be deduced from (though it is not required by) the need for reflexives to find an antecedent. 83 Given, however, the ungrammaticality of sentence (3.8) on the one hand, and the grammaticality of (3.9) on the other, it is concluded that the implicit subject is syntactically present in the form of a phonologically empty pronoun. (3.8) *Himself was shaved. (3.9) How to shave oneself is the question. (3.8) shows that reflexives cannot occur independently. To be well formed, they have to occur in the environment of a syntactically realized antecedent. Since (3.9) is grammatical, one has to conclude there is a syntactic antecedent for the reflexive oneself, the only candidate being the phonologically null subject. The relevant antecedent is the empty correlate of the lexicalized pronoun one. If one paraphrases the infinitival clauses of (3.8) and (3.9) in terms of finite sentences, one would invariably use lexicalized pronominal subjects. To give another example, in (3.10), the presence of a semantic pronominal subject (I, we, you or one) is implied, though it has no phonological content. (3.10) What __ to do? In Government and binding theory (GB), the conclusion is that null subjects of infinitival clauses are present in 84 syntax and have to be based—generated. To visualize them, the term ‘PRO’(standing for pronoun) is used. The distribution of PRO is accounted for in terms of government theory and case theory, and its interpretation by control theory. However, it is recognized that subjects of finite clauses are not universally lexicalized. In Italian, for example, it is possible to drop a pronominal subject in a finite clause. In GB theory, this is called the pro-drop or null subject phenomenon. Since the finite verb shows agreement morphology, however, it is maintained that an empty subject (labeled ‘small pro’) is present, which would keep the Predication Principle intact. The minimal difference between pro—drop languages and non pro-drop languages is considered to reside in the richness and visibility of the agreement features on the finite verb. On the whole, though the predication principle is universal and bears on the finite clauses as well as infinitivals, there is a lexicalization parameter associated with it. Two choices are allowed: either a language obligatorily lexicalizes the subject of a finite clause or it does so optionally. In English, for instance, it is obligatorily lexicalized in finite declarative sentences and is predictable in infinite clauses. In Italian, however, it does so optionally. In Chinese, subjects tend to be suppressed when understood from context. Unlike Italian, Chinese does not 85 have a rich morphological coding system to rely on for the identification of the subject. However, it should be emphasized that the non-lexicalized subject in Chinese is always recoverable from the context. This is illustrated in (3.11), which describes a Chinese custom: each year on the fifteenth of January (i.e. the date for the traditional family reunion festival), Chinese people would serve a special kind of food called yuanxiao -- delicate stuffed round-shaped rice balls made with sticky rice powder, which symbolizes good fortune and family reunion. The unexpressed subject of the sentence is understood as ‘people there in general’. (3.11) Zhengyue shiwu chi yuanxiao January fifteen eat yuanxiao 'On January 15th, (people) eat_yuanxiao.' It should be pointed out that in Chinese, subjects must be lexicalized whenever they are unpredictable or unidentifiable from the context. So far I have illustrated that subjects as understood in the predication principle are inherently arguments of the predicate, identifiable primarily in terms of their syntactic behavior or semantic relation to the predicate. In the following, I will explore other properties of ‘subject’ and argue that ‘subject' can be most optimally characterized as a proto-type concept. 86 3.3 Subject as a prototype concept 3.3.1 Empirical motivation and theoretical plausibility I argue that ‘subject’ needs to be treated as a prototype notion in universal grammar in order for the many universal generalizations which use this notion to be well defined. If only single—condition definitions were used, certain languages would be excluded from having subjects, then ‘subject’ is not applicable cross-linguistically. As a consequence, ‘subject’ cannot be construed as a universal category, and all those universal generalizations stated in terms of this notion cannot be understood as generalizations at all. A proto—type approach to ‘subjecthood’, I argue, is both empirically motivated and theoretically plausible. This proto-type approach is based on two major conclusions drawn from findings of empirical studies on subjecthood. One is that subjecthood is indeed ‘a matter of degree’, and the other is that subject is a ‘cluster concept’. Both conclusions are made originally in Keenan (1976), and they are still widely accepted as valid and important principles in functionally—oriented linguistic theories. For instance, Givon (1995: 247—248) points out that ‘empirical account of grammatical relations has no recourse to this taxonomic luxury’ of ‘single-trait definitions’ that produce ‘unambiguous discrete classes’, and argues for the necessity of recognizing the ‘gradations and indeterminacy of grammatical relations’. 87 The necessity of such a proto-type approach is sustained by its theoretical plausibility. According to Shore (1996:242), a category can either be seen 'in Platonic terms as something that is discrete or absolute,‘ or it can be seen 'in terms of a prototype with some instances considered typical representatives of the category while others are less representative.‘ The benefit of such a prototype approach to categorization, according to Givon (1995: 12-13), is that ‘it can accommodate distinct phenomena that are nevertheless in partial overlap’. Givon maintains that ‘there are profound functional reasons’ why natural categories must retain a margin of flexibility as well as considerable rigidity. The main reason for the need to retain a margin of flexibility, he explains, is that Context-dependent processing cannot proceed without some flexibility and graduality in construing and adjusting interpretations to the relevant context. Givon (1995: 13) I consider subjecthood essentially a context—dependent phenomenon, in the sense that subjects do not exist independent of the constructions in which they occur. Different constructions (i.e. syntactic structures) may yield different subjects. For instance, in an active declarative sentence, the subject is usually the agent, but in a passive sentence, the subject is usually the patient. This notion of a prototype has been long implicit in functional grammar. Firth (1968: 46—47) points to the need 88 to recognize indeterminacy both in the metalanguage of linguistics and in the language being described. Halliday (1961:254) points out that Likeness, at whatever degree of abstraction, is of course a Cline, ranging from ‘having everything in common’ to ‘having nothing in common.’ In recent years, a proto—type approach to categorization has been favored in functionalist-oriented linguistic theories. Givon (1995:12) points out that Most functionalists are currently working, whether explicitly or implicitly, within a distinct approach to categorization, Roschean prototypes. (boldface in original) According to the Roschean prototype approach to human categorization elaborated in Rosch (1973a, 1973b, & 1975), membership in a natural category does not need to be determined by a single feature, but rather by a set of characteristic features. The members of a category that display the greatest number of these features are considered most typical members of a category and thus construed as the category’s prototype. The majority of members display a great number of these features and are regarded as closely resembling the prototype. A minority of the members may display fewer of the characteristic features, and thus are less like the prototype. Subject being essentially a syntactic notion, I would view its syntactic properties as the core of the subject prototype. In comparison, its semantic and pragmatic 89 attributes can be treated as more marginal. The advantage of the prototype approach, I argue, is that it would allow for not only the solidity of syntactic features at the core of subjecthood, but also the flux of semantic and pragmatic features at its margins. A convenient point of departure for characterizing subject as a proto-type concept is Keenan's (1976) Subject Property List (SPL), which I argue, has incorporated all the properties that a subject can potentially possess, given the fact that it was arrived at on the basis of empirical accounts of a large corpus of data from a variety of languages. Therefore, they can be construed as jointly present in what I would call the prototypical subject. However, in line with prototype theories, I do not claim that all properties in the SPL will be associated with any NP in a clause in any given languages. In the following section, I will first review the bulk of the proto-typical subject properties on Keenan's (1976) SPL, and then suggest a few changes that I find necessary. 3.3.2 Prototypical properties of subject 3.3.2.1 Review of Keenan’s SPL The following are the major subject properties that I have selected from Keenan's List. Syptaptic properties: 0 unmarked position: preverbal 0 behavior: subject controls reflexive binding, 90 imperativization, co—referential NP deletion, conjunction reduction Semantic properties: 0 indispensability 0 independent existence 0 absolute reference 0 selectional restrictions 0 semantic role: agency Pragmatic properties: 0 autonomous reference 0 referentiality 0 topicality 0 givenness 0 definiteness One of the major changes on Keenan’s SPL that I would like to make is to add ‘argumenthood’ as a subject property to take into account of cases such as (3.12). (3.12) be admired. Though the predicate be admired does not impose any selectional restrictions on its subject, since any NP can go into the underlined subject position, the NP nevertheless is an argument of the predicate. ‘Selectional restriction’ as stated on Keenan’s SPL, appears to be too restrictive by itself to take care of cases such as (3.12). In what follows, I will argue that argumenthood is not only a prominent subject property but also a necessary condition on subjecthood. 91 3.3.2.2 Subject selection is argument selection The major point that I want to argue for here is that ‘argumenthood’ is a core property of subjecthood, and that subject selection is first of all argument selection. This claim is based on the generally indisputable assumption in modern grammatical theories that subject is primarily a syntactic notion. As such, a subject must be syntactically involved with the predicate. That the subject must be an argument of the clause means that it has to be an NP syntactically related to the verb in the sense that it is a nominal or its equivalent that fills one of the argument positions as specified in the its predicate-frame. It should be noted, however, that there appear to be a few counter—examples to the necessity of argumenthood for subjecthood. One has to do with data in Tagalog, where it is suggested (via personal communication by Dr. David Lockwood) that certain overtly marked subject NPs are not necessarily arguments of their predicates. The other has to do with English sentences such as ‘It is raining’ and ‘It seems that Mary has left’, where it functions as the subject of both sentences, and yet is not an argument of either verbs. In light of these observations, I think two exceptions can be made to the requirement of argumenthood for subjecthood. One is the case where the putative subject bears overt morphosyntactic marking, and the other is the case where the predicate verbs involved are those which do not take arguments, but the syntax involved 92 requires a surface subject. Except for these special cases, an NP has to be an argument to be a subject. To a certain extent, the subject of a sentence is determined by its predicate verb in the sense that its valence (to be explained below) plays an important role in the subject assignment. In the next section, I will present the relations between verbal valence and grammatical relations such as subject and object, based on researches in the theory of verbal valence and functional grammar (e.g. Allerton 1982, Dixon 1989, Speas 1990, Klaiman 1991, Palmer 1994). 3.3.2.2.1 Verbal valence and grammatical relations Researchers have long noted that there is a correspondence between the clause level (e.g. surface level) syntactic relation between a verb or verbal element with nominals, and the underlying logical level relation of a predicate with its arguments (see Allerton 1982). In recent accounts of grammatical relations, it has been suggested that grammatical relations, such as 'subject-of', 'object—of', 'indirect-object—of', are not monostratal, but rather, are mapped onto clause-level verb nominal structure from their underlying (i.e. logical level) arguments through linking (see Dixon 1989, Speas 1990, Klaiman 1991, Palmer 1994). The logical-level arguments are of two kinds: core arguments, which are those nominals essential to form a 93 predication, and non-core arguments, which are optional to form a predication. Functionally speaking, core arguments correspond to the logically required participating entities in the situation denoted by the verb, and as such, they can be identified with grammatical statuses of subject, object, and indirect object (see Matthew 1981, Speas 1990, Klaiman 1991, Palmer 1994). Noncore arguments correspond to peripheral accessories specifying the circumstances such as time, place and manner in which a process or act takes place. They can be identified with the grammatical statuses of adjuncts. This distinction of core arguments vs. peripheral arguments parallels the distinction of arguments and satellites (i.e. non-arguments) in Functional Grammar mentioned in a previous section of this chapter and I will continue to use the latter two terms in the rest of this dissertation. The particular number of arguments with which a verb combines to form a nuclear predication is called its ‘valence’. The valence of a verb is generally assumed to be an inherent property of that particular verb in the sense that verbs idiosyncratically specify the number and type of arguments they take. For example, different classes of verbs are described as having different valences, or taking different sets of valents. A one—place (i.e. univalent) predicate (e.g. vanish) is a predicate in an essential relation with exactly one argument and corresponds to an intransitive verb. A two-place (i.e. 94 bivalent )predicate (e.g. love) involves two arguments, corresponding to a transitive verb; and a three—place predicate (e.g. give) involves three arguments, corresponding to a bitransitive verb. The relation of verbal valence and the grammatical relation is characterized as follows: univalent verbs or verbal elements take a subject element, bivalent verbs take a subject and an object, and trivalent verbs take a subject, a direct object and indirect object (see Tesniere 1959, Matthew 1981, Klaiman 1991, and Palmer 1994). To summarize, a verb takes up to three arguments to form a nuclear predication, which can be mapped onto the clause level as subject, direct object and indirect object through linking. The following is a summary of how the linking works in basic (e.g. active) clauses. For a univalent verb (i.e. an intransitive verb), its sole argument is mapped onto the clause-level as the subject, whatever semantic role it assumes (Dixon 1979 & 1989). Evidence in support of this analysis can be seen in the fact that in English and many other languages, the single argument of an intransitive sentence has the same grammatical marking as the Agent of an active transitive sentence, as illustrated by the following pairs in (3.13): (3.13) He laughs They laugh He loves them They love him 95 The three features of word order (i.e. preverbal), morphology (he, they; rather than him, them) and agreement with the verb (with —s for he, without —s for they), clearly establishes the identity of the single argument of the intransitive sentences with the agentive subject of an active transitive clause. This identity is also reflected in syntax. There are certain specific syntactic possibilities which are restricted to subject, a status that the single argument of the intransitive clauses share with the agent of an active transitive clause. For instance, subjects, but not objects, can be omitted in the second clause of a coordinate construction, if they are coreferential with the subject of the clause. This is illustrated in (3.14), where the omitted subject is shown in brackets: (3.14) The thief escaped and (the thief) chased a taxi. In this coordinate construction, the first clause has an intransitive verb (i.e. escape) with a single argument (i.e. the thief). The second clause has a transitive verb (i.e. chase) taking two arguments (i.e. the thief and a taxi respectively). The subject of the second clause can be omitted because it is identical with the subject of first clause, which happens to be the single argument of an intransitive verb. However, one cannot omit the object in a similar way: 96 (3.15) *The thief escaped and the taxi chased (the thief). What has been illustrated so far is that the single argument of an intransitive verb (whatever its semantic role) behaves like the Agent argument of a transitive verb in a basic structural configuration, in that they share a subject identity: the single argument of an intransitive verb always maps onto the clause—level as the subject, just like the Agent argument of a transitive verb always becomes the subject in a basic active (i.e. non-passive) sentence. For most bivalent verbs, their two arguments, which usually can be identified as Agent and Patient semantically, will be mapped unto the clause level as subject and object respectively in basic structural configuration (i.e. the unmarked active sentence structure). The semantically linked arguments can in principle be remapped from one grammatical relation to another by syntactic rules. In other words, the basic grammatical relations assigned to particular NPs may be altered through what is called ‘role—remapping rules’ (Klaiman 1991: 14). For instance, the object of an active sentence can be remapped as the subject of a passive sentence. For satellites (i.e. those adverbials specifying Location, Instrument, Beneficiary, etc.). the grammatical relations to which they are mapped in basic structures are termed ‘oblique relations’(see Klaiman 1991:14), and their 97 respective corresponding basic—level grammatical relations are called ‘oblique—locative’, ‘oblique-instrument’, ‘obligue—benefactive’ (ibid.). What I have presented so far is that researches indicate there is a well recognized correlation between the grammatical relation of ‘subject' and the valence or the argument structure of the verb. A natural generalization that can be drawn from the correlation is that subject has to be an argument of the verb, and it is likely to be selectionally restricted to that verb (except for cases involving linking verbs such as ‘to be’). Therefore, I argue that subject selection is argument selection in the sense that the subject is mapped onto the clause level from the verb’s underlying argument structure and is to some extent predictable from the valence of the verb involved in the sentence. 3.3.2.2.2 Argumenthood and syntactic properties Another argument for the criterion that a subject has to be an argument of the verb (unless otherwise exempted as specified in 3.3.2.2) is the evidence that all those NPs that manifest syntactic subject properties such as control, raising, reflexivizations, etc. are all arguments of their corresponding verbs. This is true because all syntactic and behavior properties of the subject entail that the subject must be selectionally related to the predicate. Selectional restrictions entail argumenthood. In section 98 3.3.3, I will discuss the syntactic coding and behavioral properties of the subject. The examples in that section all illustrate that argumenthood and the syntactic subject properties go together. In this section, I have presented some of the theoretical assumptions and arguments for my claim that a subject has to be an argument of the verb in the sentence (though two exceptions may apply as specified in 3.3.2.2). One of the advantages of argumenthood as a necessary subject property is that it eliminates adjuncts from subject candidacy, therefore significantly narrowing down the range of subject assignment. Another advantage is that it helps to distinguish ‘subject’ from ‘topic’, a closely related but distinct notion that is primarily discourse— oriented (rather than syntactic-oriented) in nature. A major distinction between ‘subject’ and ‘topic’ is that the latter does not have to be an argument of the predicate. This is because ‘topic’ is primary a discourse notion, whose selection is said to be ‘independent of the verb’ (see Li and Thompson 1976:463). What determines the topic selection is pragmatic factors such as discourse context and the speaker presupposition, a point that will be elaborated on in Chapter 5. 3.3.2.3 Proto-agent properties of subjecthood In this section, I propose another change to Keenan’s Emibject Property List, namely, replacing ‘the semantic role 99 of agency’ with ‘proto—agent properties’. Keenan’s idea is that subjects ‘normally express the agent of the action, if there is one’ (Keenan 1976:321). As Keenan himself notes, this property cannot be used to identify subjects of sentences in which there is no agent (ibid.). However, I argue that this property of 'agency' should be extended to accommodate cases where arguments of the predicate are not agents, but clearly display agent—like properties. For instance, the arguments for intransitive verbs like ‘break’, ‘die’, ‘grow’, and ‘smile’, do not control or initiate the activity like typical agents do, however, they closely resemble arguments for verbs like ‘walk’, ‘sit’, ‘speak’, which control or initiate the activity. More examples are given in (3.16). (3.16) Qppp is being polite to Mary. Se accidentally fell.. Water filled the boat. [Examples from Dowty 1991:572] To accommodate cases such as (3.16), I propose to restate ‘agency’ as ‘proto-agent properties’, a term originally suggested in Dowty (1991), which incorporates the following characteristics specified in Dowty (1991:572). Each of these properties is ‘hypothesized to be semantically independent’ (ibid.), although more than one such property may apply to each argument of certain transitive verbs. lOO 0 volitional involvement in the event or state e.g. gohn is ignoring.Mary. 0 sentience (and/or perception) e.g. John knows/believes/is disappointed at the statement; John sees/fears Mary. 0 causing an event or change of state in another participant e.g. His loneliness causes his unhappiness. 0 movement (relative to the position of another participant) e.g. Se accidentally fell; water filled the boat. 0 exists independently of the event named by the verb e.g. JOhn needs a car. To summarize, evidence given above indicates that it is plausible to generalize the feature of ‘agency’ on Keenan’s SPL into ‘proto—agent properties’ since ‘agency’ is not a discrete notion but rather a cluster concept itself. 3.3.3 Ranking of the subject properties What I have presented so far is a list of prototypical properties of subjects largely adopted from Keenan (1976), but not without refinements. These properties are construed as jointly present in the prototypical subject, and.the more of these properties an NP displays in a sentence in comparison to other NPs, if there are any, the rmore subject—like it becomes. Recall that the criterion for subject selection proposed in Keenan (1976) is the Inanifestation of a clear preponderance of his SPL. 101 However, there are some problems with this approach. One of them is its operational difficulty, since it is not very convenient to check NPs against a list of 30—odd properties. A more serious problem arises when two NPs display equal numbers of properties (though not identical) in the same clause. A solution to this dilemma is to weigh the prototypical subject properties and decide what properties count more than others in determining what the subject is. This is not only theoretically plausible in the sense that it is in line with the prototype theory, but also practically necessary since these subject properties do not hold equal status, given the fact that some properties are usually subject—exclusive (e.g. syntactic agreement features) while others (e.g. definiteness) may be shared by non-subject grammatical categories as well. Evidently, some properties are more significant than others in marking subjecthood. It seems that a detailed ranking of all 30 odd properties on Keenan’s SPL is beyond the scope of this dissertation. However, a principled relative ranking can be worked out by which some subject properties can be shown to count more than others in determining what the subject is. In the rest of this chapter, I present arguments and evidence that the following general ranking (from top to bottom) of subject properties applies to English and Chinese: 102 (A) Argumenthood, (B) Syntactic coding and behavioral properties, (C) Proto-agent properties. 3.3.3.1 Argumenthood In an earlier section, I argued that ‘argumenthood’ is a core property of subjecthood, and that subject selection is first of all argument selection. What this means is that subjects are typically NPs, or other units that are treated as equivalent of NPs. Moreover, they are obligatorily involved with the predicates of the sentences in which they occur. While argumenthood does not directly indicate what subjects are, it certainly indicates what they are not. This conclusion is drawn based on the theoretical assumption that arguments and non-arguments (i.e. satellites) are not functionally equivalent in a predication. Recall that arguments are terms that are obligatory and satellites are terms that are optional. Arguments are more essential in that they refer to entities and indicate sentence participants. In contrast, satellites function as adjuncts or modifiers of these participants, usually specifying details such as Time, Place, Manner, Degree, and Reason, etc. When both arguments and non—arguments are involved in a sentence, it is one of the arguments that will be selected as the subject. This is illustrated in (3.17). 103 (3.17) Jintian tamen meiyou lai. today they did not come ‘Today they did not come.’ There are two NPs in (3.17), tamen ‘they’ and jintian ‘today’. The former is an argument, indicating the participant of the predicator lai, ‘come’. The latter is a ‘satellite’ (i.e. non—argument), specifying the information of Time. Given that tamen ‘they’ is the only argument and therefore is more essential than the non—argument jintian 'today', it is selected as the subject of the sentence. This analysis, I argue, is far more plausible than the one that is current in the Chinese linguistic community, which would pick as subject jintian ‘today’ for (3.17). According this analysis (e g. Zhu 1981: 95—96), subjects are invariantly sentence-initial. Therefore, the structures of (3.17) would be characterized as below. (3.17) Jintian tamen meiyou lai. today they did not come <—Subject—> < ————— Predicate ------------ > <-Subject-> <--Predicate--> ‘Today they did not come.’ In this analysis, the predicate of a clause may itself be a subject-predicate predication. A sentence may contain two subjects, a ‘major subject’ and a ‘minor subject’, the former being the subject of the whole clause, as jintian ‘today’ in (3.17), the second the subject of only the Igredicate segment of the clause, as tamen ‘they’ in the same sentence . 104 The problem with this type of analysis, as I mentioned in Chapter 2, is that it unduly complicates the structure of Chinese sentences on the one hand, and oversimplifies it on the other. For instance, (3.17), which is a simple sentence, would have to be considered a complex double- subject construction. There is nothing wrong with a complex double-subject construction per se, the problem here is that this analysis would render a majority of Chinese sentences as having a double—subject structure, when in fact many of them can be more simply accounted for as having just a single subject. Given the stipulation that all subjects are sentence— initial, this model also has the consequence of oversimplifying the syntactic structures of Chinese, in the sense that variation in terms of the order of the constituents motivated by pragmatic factors is ignored. Analysis of syntactic structures in Chinese is therefore rendered uninteresting and unmotivated, since there are no ‘variations to account for. In contrast, the property of argumenthood would help the proper identification of :subjects. It is simple and syntactically and functionally Inotivated. What I have illustrated using the example of (3.17) is tillat the property of argumenthood helps to narrow down the ESIlbject choices in a sentence with the effect of excluding Ei