Prominence and Locality in Grammar: The Syntax and Semantics of Wh-Questions and Reflexives 0367220911, 9780367220914

This book challenges the current consensus on the analysis of wh-questions and reflexives from the perspective of the sy

802 54 7MB

English Pages 208 Year 2019

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Prominence and Locality in Grammar: The Syntax and Semantics of Wh-Questions and Reflexives
 0367220911, 9780367220914

Citation preview

 i

Prominence and Locality in Grammar

This book challenges the current consensus on the analysis of wh-​questions and reflexives from the perspective of the syntax–​ semantics interface. An integrated approach incorporating analyses of the interaction between different levels of linguistic knowledge is proposed. It argues that the derivation and interpretation of wh-​ questions and reflexives are not purely syntactic in nature but are regulated by principles operating at the syntax–​semantics interface. Two general principles underlying our knowledge of language and cognition are proposed in this work. One is the Principle of Locality, and the other is the Principle of Prominence. It shows that although wh-​ quantification and reflexivization belong to two different domains of study in gene­ rative grammar, their derivation and interpretation are basically constrained by the complex interaction between prominence and locality in grammar. The first part of the book discusses how wh-​questions are formed and interpreted in Chinese and English and shows that the formation and interpretation of wh-​questions are constrained by the interaction between prominence and locality. It is shown that in wh-​interpretation prominence is used to define the set generators so as to licence other wh-​words in the pair-​list reading in multiple wh-​questions. It also discusses wh-​island effects in English and Chinese, and unlike previous claims made in the literature (cf. Huang 1982a, 1982b), it argues that the so-​called wh-​island effects in English are also observed in Chinese. The second part of the book investigates the role that prominence and locality play in reflexive binding. It is shown that in reflexive binding, the binding domain of the reflexive is defined by prominence. It proposes a unified account for both the noncontrastive compound reflexive and the bare reflexive in Chinese and shows that they are constrained by the same reflexive binding condition proposed in this work, though they employ different definitions of the most prominent NPs to determine their binding domains. Prominence and Locality in Grammar: The Syntax and Semantics of Wh-​Questions and Reflexives is an important theoretical contribution to the syntax–​semantics interface studies and can serve as a valuable text for graduate students and scholars in the field of Chinese, linguistics, and cognitive science. Jianhua Hu is Professor of Linguistics at the Institute of Linguistics, Chinese Academy of Social Sciences, China.

ii

Routledge Studies in Chinese Linguistics Series editor: Hongming Zhang

Mandarin Chinese Words and Parts of Speech A Corpus-​based Study Chu-​Ren Huang, Shu-​Kai Hsieh, Keh-​Jiann Chen A Study of Sino-​Korean Phonology Its Origin, Adaptation and Layers Youyong Qian Partition and Quantity Numerical Classifiers, Measurement and Partitive Constructions in Mandarin Chinese Jing Jin Mandarin Loanwords Tae Eun Kim Intensification and Modal Necessity in Mandarin Chinese Jiun-​Shiung  Wu The Architecture of Periphery in Chinese Cartography and Minimalism Victor Pan Focus Manifestation in Mandarin Chinese and Cantonese A Comparative Perspective Peppina Po-​lun Lee Prominence and Locality in Grammar The Syntax and Semantics of Wh-​Questions and Reflexives Jianhua Hu For more information about this series, please visit:  www.routledge.com/​ languages/​series/​RSICL

 iii

Prominence and Locality in Grammar The Syntax and Semantics of Wh-​Questions and Reflexives Jianhua Hu

iv

First published 2019 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN and by Routledge 52 Vanderbilt Avenue, New York, NY 10017 Routledge is an imprint of the Taylor & Francis Group, an informa business © 2019 Jianhua Hu The right of Jianhua Hu to be identified as author of this work has been asserted by him in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. British Library Cataloguing-​in-​Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging-​in-​Publication Data Names: Hu, Jianhua (Linguist), author. Title: Prominence and locality in grammar : the syntax and semantics of wh-questions and reflexives / Jianhua Hu. Description: New York, NY : Routledge, [2019] | Series: Routledge studies in Chinese linguistics | Includes bibliographical references and index. Identifiers: LCCN 2019004641 | ISBN 9780367220914 (hardback : alk. paper) | ISBN 9780429273209 (ebook) Subjects: LCSH: Grammar, Comparative and general–Interrogative. | Interrogative (Grammar) | Question (Logic) | Grammar, Comparative and general–Syntax. | Semantics, Comparative. | Chinese language–Interrogative. Classification: LCC P299.I57 .H823 2019 | DDC 415–dc23 LC record available at https://lccn.loc.gov/2019004641 ISBN: 978-​0-​367-​22091-​4  (hbk) ISBN: 978-​0-​429-​27320-​9  (ebk) Typeset in Times New Roman by Newgen Publishing UK

 v

To the memory of my mother AN Yulan

vi

 vii

Contents

Preface  Acknowledgements  Abbreviations  1

Introduction  Notes  5

ix xii xiv 1

PART I

Wh-​questions 

7

2

9

The syntax and semantics of wh-​questions  Introduction  9 2.1  To move or not to move?  11 2.1.1  LF wh-​movement  11 2.1.2  D-​linking  17 2.1.3  Pied-​piping  18 2.1.4  Unselective binding and choice function  22 2.1.5  Referentiality and nonreferentiality  25

2.2  Prominence and locality: Theorizing the interpretation of wh-​questions  29 2.2.1  Focus and wh-​interpretation via choice functions  30 2.2.2  Clausal typing and wh-​feature strength: Two parameters in classifying wh-​questions  34 2.2.3  The C Typing Condition and wh-​interpretation  43 2.2.4  A-​not-​A and multiple wh-​questions  45 2.2.5  Some problems solved  62

2.3  The Principle of Economy in wh-​interpretation  64 2.3.1  The division of labour between clausal typing and wh-​interpretation  64 2.3.2  Economy and wh-​interpretation  66 2.3.3  The structure of Chinese wh-​questions  74 2.3.4  Multiple wh-​questions revisited  75

viii

viii Contents 2.3.5  A note on weishenme and A-​not-​A, and PCTC further revised  79

2.4  Further discussion  84 2.4.1  Wh-​island effects in Japanese  84 2.4.2  Additional wh-effects  87

2.5  Economy in wh-​interpretation revisited  90 2.6  Summary  92 Notes  93 PART II

Reflexives 

95

3

97

Prominence and locality in reflexive binding  Introduction  97 3.1  The predicate-​based binding theory  98 3.1.1  R&R’s predicate-​based theory of reflexives  98 3.1.2  Lidz’s condition R  102 3.1.3  Limitations of the predicate-​based theory of reflexives  103 3.1.4  The missing type  111 3.1.5  Further discussion: Complex predicates and reflexive-​marking  115

3.2  Characterizing reflexives in Chinese: Deriving referential dependency and reflexivity from primitive features  117 3.3  The Chinese bare reflexive  126 3.3.1  Syntactic approaches  126 3.3.2  Nonsyntactic approaches  127 3.3.3  The binding properties of ziji  128 3.3.4  Pan’s (2001) self-​ascription theory  133 3.3.5  Huang and Liu’s (2001) account  136

3.4  Prominence and locality in Chinese reflexive binding  141 3.4.1  The binding of the compound reflexive  141 3.4.2  The binding of the bare reflexive: Two search engines  144

3.5  Further discussion  160 3.5.1  The derivational cycle  160 3.5.2  The noncontrastive and contrastive compound reflexives  163 3.5.3  Multiple occurrences of ziji  165

3.6  Summary  168 Notes  169 4

Conclusion 

171

Bibliography  Author index  Language index  Subject index 

177 185 187 188

 ix

Preface

This book challenges the current consensus on the analysis of wh-​questions and reflexives from the perspective of the syntax–​semantics interface. An integrated approach incorporating analyses of the interaction between different levels of linguistic knowledge is proposed. It argues that the derivation and interpretation of wh-​questions and reflexives are not purely syntactic in nature but are regulated by principles operating at the syntax–​semantics interface. Two general principles underlying our knowledge of language and cognition are proposed in this work. One is the Principle of Locality, and the other is the Principle of Prominence. Previous studies on wh-​questions and reflexives have attached much importance to the role that locality may play in grammar. However, less attention has been paid to the role that prominence plays in the grammatical system of natural language. The purpose of this work is to show that prominence also plays a crucial role in grammar, and as a matter of fact, the role that it plays is even more important than that of locality in the grammar of Chinese. The Principle of Locality is basically a syntactic condition if locality is viewed in terms of the Closeness Condition or the Minimal Link Condition. The Principle of Prominence, however, involves factors pertaining to semantics, pragmatics, and cognition, though it may include a syntactic de­finition of c-​command. Although these two principles are proposed for the two separate domains of study (i.e., syntax and semantics), factors related to prominence and locality are often interrelated or interacted in the actual derivation and interpretation of linguistic expressions. For instance, in the current study of reflexives in Part II, prominence plays a major role in the de­finition of the local domain of reflexive binding, whereas locality is viewed as a kind of prominence condition. The work carried out in this book is descriptive in a sense, given that its focus is largely placed on a fine-​grained description of the complex interaction between prominence and locality regulating the derivation and the interpretation of the linguistic structures and expressions. However, the descriptive work of this study is by no means a continuation of the tradition of Descriptive Linguistics. Rather, it adheres to the basic scientific assumptions of the Universal Grammar (UG) and the spirit of Generative Linguistics. It

x

x Preface aims to give an in-​depth description of the mechanism and conditions underlying our linguistic knowledge and cognition by using a set of theoretical and descriptive tools developed at the syntax–​semantics interface in the spirit of New Descriptivism (Hu 2006, 2009, 2018). This book contains two parts. The first part studies wh-​questions, and the second one investigates reflexives. The goal of this study on these two parts is to show that, although wh-​questions and reflexives belong to two different domains of study, their derivation and interpretation are both constrained by the complex interaction between prominence and locality in grammar. It is argued that locality is used to guarantee the priority of the prominent wh-​ word in wh-​movement so as to derive the pair-​list reading in multiple wh-​ questions, whereas prominence defines the binding domain of reflexives in natural language. In the first part of the book, the discussion is centred upon the question of how wh-​interrogative structures are derived and how wh-​elements are interpreted in Chinese and English. It shows that the derivation and interpretation of wh-​ questions are constrained by some economy considerations, specifically, the Pure Clausal Typing Condition (PCTC) and the Principle of Economy (PE), both of which incorporate the abovementioned two factors:  prominence and locality. PCTC is a pure locality condition established on the notion of closeness, according to which a clause prefers to be typed by the closest wh-​phrase. PE involves the interaction of prominence and locality, according to which a wh-​phrase closer to the left periphery of the clause should generally move into [Spec, CP] first in multiple wh-​questions, not only because the relevant movement is preferred by the locality condition but also because the relevant wh-​phrase is prominent in the Prominence Hierarchy, due to the fact that only a prominent wh-​phrase can license wh-​ phrases following it to produce the pair-​list reading. This book also discusses wh-​island effects in English and Chinese, and unlike previous claims made in the literature, it argues that the so-​called wh-​ island effects in English are also observed in Chinese. This book employs PCTC and the Principle of Economy (PE) to explain why wh-​island effects are observed in natural language. It also discusses the A-​not-​A question and the asymmetry between weishenme “why” and other wh-​expressions in Chinese and argues that when weishenme or the A-​not-​A element occurs in a strong island, they cannot take wide scope since they cannot be interpreted with the relevant C via either the Agree operation or the choice function. The second part of the book proposes a unified account for both the noncontrastive compound reflexive and the bare reflexive in Chinese and shows that they are constrained by the same reflexive binding condition proposed in this work, though they employ different definitions of the most prominent NPs to determine their binding domains. On the basis of an assumption made in Huang and Tang (1991), this book derives the long-distance (LD) binding properties of ziji from its lack of phi-​features and referential features and shows that it is constrained by the same binding condition that applies to the

 xi

Preface xi compound reflexive in Chinese. This book uses the feature ​search engines to define the binding domains of reflexives. It claims that the binding domains for both compound and bare reflexives are determined by the candidate set related to the most prominent NP chosen by the relevant search engines. Under the analysis assumed in this work, the compound reflexive, due to its lack of only the referential feature, depends on the R-​engine (referential feature search engine) to determine its binding domain, whereas the bare reflexive ziji employs the R-​engine and the P-​engine (phi-​feature search engine) to determine its binding domain, because, different from the compound reflexive like ta-​ziji, ziji lacks both the referential feature and the phi-​feature. Hence, the binding domain of ziji is determined by the union of the two candidate sets related to the two most prominent NPs chosen by the two search engines. When the binding domain of the reflexive is defined, the blocking effect will be derived if it is bound outside of that domain. This work shows that prominence and locality are the two most important factors in grammar, which can account for not only the interpretation of wh-​ questions but also that of reflexives.

xii

Acknowledgements

My research on prominence and locality in grammar has been greatly influenced by Haihua Pan, Liejiong Xu, and Thomas Hun-​tak Lee. I  owe enormous debt to Haihua Pan, my PhD thesis supervisor, for introducing me to the topic of prominence and for making me delve into the complex interaction between prominence and locality in grammar through numerous discussions over the years. His insightful observations, critical comments, and penetrating criticism have helped me toward a better understanding of the issues discussed in this work. I have benefited greatly from discussions on the role that prominence may play in grammar with Liejiong Xu, who inspired many very specific aspects of the analyses assumed in this book, and his influence can be found in every part of the work. I would also like to take this opportunity to acknowledge the debt I owe to Thomas Hun-​tak Lee, who has always been an inspiration. I am deeply grateful to him for making me understand the nature of linguistic theorizing from the perspective of child language acquisition. His linguistic knowledge and insights, as well as his philosophical thinking, had an enormous influence on me in my study of the relationship between prominence and locality. I would like to acknowledge the debt I owe to the series editor Hongming Zhang for his encouragement, understanding, and patience, without which the completion of this book might have become impossible. I am grateful to the editor at Routledge, Claire Margerison, for her generous help and professional advice. Special thanks should go to Candice Roma for her excellent copyediting work. I  would also like to take this opportunity to extend my gratitude to the two anonymous reviewers of the book for their helpful suggestions and constructive comments. This book has also benefited a lot from my students at the Graduate School of Chinese Academy of Social Sciences (CASS), Na Sui, Xiuli Guo, Jialei Zhu, Lulu Peng, Mengmeng Yang, Lei Wang, Xiaolan Liu, Chengzhang Ge, Mengqi Yang, and Liling Wu, for their questions and feedback when I used the draft of this work as a text for a course on Linguistic Interfaces at the Department of Linguistics of the Graduate School, CASS. Special thanks should go to Lei Wang for combing through the manuscript in its entirety and helping me with editing and to Xiaolan Liu, Chengzhang Ge, and

 xiii

Acknowledgements xiii Mengqi Yang for preparing the index and proofreading. I would like to thank Mengmeng Yang for her observations on a number of errors in the manuscript and her corrections of the proofs. I was indebted to my colleagues at the Institute of Linguistics, Chinese Academy of Social Sciences, who have supported my academic life. The preparation of the present work was supported in part by the Key Project “Prominence and Locality in the Distribution and Selection of Arguments” of the National Social Sciences Foundation of China (Grant No. 14AYY016), the CASS (Chinese Academy of Social Sciences) Five-​Year Research Plan Key Project “Psycholinguistic Studies on Language Acquisition and Development,” as well as the CASS-​HAS (Hungarian Academy of Sciences) Scientific Cooperation Project “Argument Structure, Event Structure, and their Relationship.” I would also like to take this opportunity to thank the members of the Linguistics Institute of China (LINC), Thomas Hun-​ tak Lee, Yafei Li, Danqing Liu, Haihua Pan, Hongming Zhang, and Hongbo Zhou, for their suggestions, encouragement, and support. This book grew out of my PhD thesis submitted to the City University of Hong Kong in 2002. The analyses assumed in the book have been left essentially in the form they were presented in the thesis, though some typos were corrected, some lines were added, and some sentences were rephrased for the purpose of stating more clearly the views of the book. I  am especially grateful to the Department of Chinese, Translation and Linguistics, the School of Graduate Studies, and the FHS (Faculty of Humanities and Social Sciences) Research Expense Grant for their support. I consider myself fortunate to work with Jiujiu Xu, Danqing Liu, Dongfan Hua, Peppina PoLun Lee, Fang Hu, Qin’an Hu, Yuan Shen, Qing Wu, Chenglong Huang, Fuminobu Nishida, Ling Wang, Kai Sum Ho, Wanmin Zhang, and Haiying Wang at the City University of Hong Kong. Jiujiu Xu and Peppina Lee were always ready to give me a lot of help. Dongfan Hua introduced me to wh-​ quantification. Many ideas of mine on wh-​questions arose from our talks over beer in Dan Ryan’s located in the Festival Walk. I owe special gratitude to Fuminobu Nishida, who offered me a lot of help with his native speaker’s judgments on Japanese data. I owe a special debt to my teacher and MA supervisor at Shandong University, Yanfu Li, who introduced me to formal linguistics and changed the way I view linguistic problems in general. Without him, it is unlikely that I would have become a linguist. Lastly, I want to express my deepest love for my parents, my brother, my wife, and my son. I am very much grateful to them for their encouragement and unfailing support. This book is dedicated to the memory of my mother, who left this world in 2014. My mother’s influence on me could be felt in many aspects. She taught me all the essentials, including the attitude toward knowledge and education, the way of thinking and reasoning, and the philosophy of life. Her love and encouragement not only gave me confidence to finish this work, but also helped me understand the meaning of life.

xiv

Abbreviations

A accusative case AGR ​agreement AH animacy hierarchy ASP aspect marker AUX auxiliary BA ba marker (preverbal object marker) BEI bei marker (passive marker in Chinese) C complementizer CFC complete functional complex CH closeness hierarchy CL classifier CP complementizer phrase CTC C Typing Condition D determiner DE modifier marker or postverbal complement marker DP determiner phrase ECP empty category principle EH empathy hierarchy F focus FM focus marker G genitive case GB government and binding GC governing category GFH grammatical function hierarchy INFL inflection IP inflection phrase LD long-​distance LF logical form MLC Minimal Link Condition N nominative case OBJ object Op operator OT Optimal Theory

newgenprepdf

 xv

Abbreviations xv P past tense PCA prominence computing algorithm PCTC Pure Clausal Typing Condition PE Principle of Economy PH person hierarchy PMC Principle of Minimal Compliance PR prominence ranking Q-​p question particle QP quantifier phrase R Referential independence RBC reflexive binding condition SFP sentence-​final particle SH structural hierarchy SHI … DE shi … de construction SPH syntactic prominence hierarchy SUBJ subject Top topic marker UG Universal Grammar

xvi

 1

1  Introduction

According to Chomsky (1981), the relationship between a reflexive and its antecedent is characterized by a locality condition and a syntactic prominence condition. Locality is defined in terms of governing category (GC), and syntactic prominence is defined by the notion of c-​command (cf. Baker 1995; Pan 1998). In his study of Chinese compound reflexives, Pan (1998) finds that Chinese has a different definition of prominence for its reflexives and that locality is, in fact, a relative rather than absolute condition in Chinese. He shows quite convincingly that Chinese employs the animacy hierarchy rather than c-​command in defining its prominence condition and that locality should be regulated by a closeness condition in Chinese. Xu (1999) emphasizes the importance of the interaction between prominence and locality in the interpretation of reflexives. He claims that prominence means dependability, locality means availability, and that different languages may have different definitions of prominence and locality.1 Xu and Lee (1990) investigate the role that prominence plays in the determination of scope. They find that scope-​ taking is regulated by thematic hierarchy. Bresnan (2001) also discusses the interaction between prominence and locality and claims that their interaction may derive different possibilities in pronominal binding. Inspired by these ideas on prominence and locality, this work explores the roles that prominence and locality play in grammar. The basic idea is that prominence and locality will interact to determine the derivation and interpretation of language. Assume that there are two general principles, as shown below, at work in the grammatical system: (1)  The Principle of Prominence The more prominent one is preferred to the less prominent one in syntactic derivation and interpretation. (2) The Principle of Locality The closer one is preferred to the less close one in syntactic derivation and interpretation.

2

2 Introduction The above two principles are very simple, since they are just two general principles. The actual cases involved, however, would be quite complicated. Suppose that what is prominent is less close and what is closer is less prominent. All this will result in linguistic variation. If viewed cross-​linguistically, things will be made even more complicated, since different languages may have different definitions of prominence, though they may have the same definition of locality if locality is viewed in terms of the Closeness Condition (cf. Pan 1998) or the Minimal Link Condition (Chomsky 1995).2 That is to say, what is prominent in one language may not be prominent in another language. If this is true, it will be of great significance if it can be proven that part of the differences among languages can be reduced to the parameterization of the notion of prominence. The goal of this work is to explore how the interaction between prominence and locality can account for some basic constraints that underlie linguistic interpretation. Specifically, it investigates how the interpretation of wh-​questions and reflexives is determined by the interaction between prominence and locality. It shows that in multiple wh-​questions, locality is used to guarantee the priority of the prominent wh-​word in wh-​movement so as to derive the pair-​list reading, whereas in reflexive binding, the binding domain for reflexives is defined by the notion of prominence. This book contains two parts. The first part studies the syntax and semantics of wh-​questions. It shows that, although Chinese and English employ different strategies to type a wh-​ question, the interpretation of wh-​expressions is basically constrained by the same set of conditions. It is observed that in multiple wh-​questions, a more prominent wh-​word prefers to precede a less prominent one since in multiple wh-​questions, wh-​words prefer to be interpreted in the pair-​list reading, which can be derived only by the prominent wh-​word that functions as a set generator. This work makes a distinction between the clausal typing condition and the wh-​interpretation condition and argues that wh-​questions must be properly typed according to the following condition: (3)  The Pure Clausal Typing Condition (PCTC) a. For wh-​raising languages, a clause is typed as a wh-​question iff there is a wh-​word that moves overtly into [Spec, CP] via cyclic movement without crossing any strong island. b. For Q-​particle languages, a clause is typed as a wh-​question iff there is a wh-​word interpreted with the closest C[+Q] via either the Agree operation or the choice function application. Besides being typed, a wh-​question also needs to be properly interpreted. This book shows that wh-​interpretation is constrained by the Principle of Economy, which stipulates that wh-​interpretation in syntax is preferred to wh-​ interpretation in semantics (cf. Reuland 2001 for similar ideas in accounting

 3

Introduction 3 for anaphoric binding). It points out that the Principle of Economy is, in fact, used by the grammatical system to help place the more prominent wh-​ word before the less prominent one in multiple wh-​questions so as to properly derive the pair-​list reading. The reason why in (4) the wh-​object cannot move across the wh-​subject in syntactic derivation is because the wh-​object is less prominent than the wh-​subject and thus cannot derive the pair-​list reading for the wh-​subject. (4)  *What did who buy? (5) ?Which book did how many people buy? (Comorovski 1996: 85) However, if the relevant wh-​object is an inherently D-​linked which-​phrase, it can move across the wh-​subject, as shown in (5), because a D-​linked which-​ phrase is prominent in discourse and can thus derive the pair-​list reading for the wh-​subject that it c-​commands. This book also discusses the A-​ not-​ A question and the asymmetry between weishenme “why” and other wh-​expressions in Chinese. Following the analysis proposed in Huang (1991) and Shi (1994), this book assumes that the A-​not-​A element is formed by the incorporation of the Q operator into the Inflection (often shortened to INFL) and argues that when the A-​not-​A element occurs in a strong island, the sentence is ungrammatical because the Pure Clausal Typing Condition (PCTC) would be violated if the A-​not-​A element is associated with the matrix Q operator. This book claims that when a wh-​element occurs in a strong island, it cannot move out of it in either overt syntax or LF and argues that weishenme cannot take matrix scope in strong islands because it cannot range over an individuated set and thus fails to be interpreted via the choice function application, given that wh-​expressions occurring in an island cannot be interpreted via the Agree operation introduced in this book. As a result, weishenme has no interpretation when it occurs in a strong island. This book also discusses wh-​island effects in English and Chinese, and unlike previous claims made in the literature, it argues that the so-​called English wh-​island effects are also observed in Chinese. This book employs PCTC and the Principle of Economy (PE) to explain why wh-​island effects are observed in natural language. With the help of the four criteria for distinguishing echo questions from original questions, this book argues that the reason why one may have the impression that the wh-​island effects are not observed in Chinese is because echo questions are often wrongly taken for original questions, given that the distinction between the echo question and the original question is not reflected in word order in Chinese (cf. Xu 1990; Liu 1986). The second part of the book shows how the interaction between prominence and locality can account for the binding properties of reflexives in Chinese. On the basis of an assumption held in Huang and Tang (1991), this

4

4 Introduction book derives the long-distance (LD) binding properties of ziji from its lack of phi-​features and referential features and shows that ziji is constrained by the same binding condition that also applies to the compound reflexive in Chinese, which is given below: (6)  Reflexive Binding Condition (RBC) a. A reflexive can be bound to an accessible prominent NP in its binding domain. b. The binding domain of the reflexive is the minimal complete functional complex (CFC) that contains all the members of the candidate set and the reflexive. c. A binds B iff A is co-​indexed with B, and A and B are compatible in phi-​features. d. A is accessible to B iff the assignment of the index of A to B would not violate *[γ …δ …], where γ and δ bear the same index. This book argues that the crucial concept accounting for the binding of reflexives in Chinese is prominence, and the binding domain for reflexives is actually determined by the notion of prominence. It uses the feature s​ earch engines (i.e., the R-​engine (referential feature search engine) and the P-​engine (phi-​feature search engine)) to search for the candidates and the most prominent NP (noun phrase) to define the binding domain of reflexives. Whenever the most prominent NP is found, the search engine will stop its searching and produce a candidate set. It is the candidate set that defines the binding domain of the reflexive. When the binding domain of the reflexive is defined, the reflexive cannot be bound to an antecedent outside of it. Following Huang and Tang (1991), this book claims that the bare reflexive ziji is different from ta-​ziji in that it lacks not only the referential feature but also the phi-​feature and thus must depend on two feature s​ earch engines for defining the candidate set and the most prominent NP(s). Since there are two feature search engines for ziji,  the binding domain of ziji is defined by the union of the sets determined by the two most prominent NPs selected by the two search engines. The blocking effect for ziji is derived if ziji is bound outside of the binding domain defined above. This work claims that the binding domain defined by the candidate set related to the most prominent NP is similar to the governing category. The governing category is also defined by the notion of prominence (i.e., the subject/​SUBJECT). To summarize, it can be seen from the above discussion that prominence and locality, which are often interrelated and interact with each other, are the two most important factors in the grammar of natural language. Their interaction can account for not only the interpretation of wh-​questions but also that of reflexives in Chinese and English.

 5

Introduction 5

Notes 1 Xu (personal communication) also points out that it is theoretically and empirically possible that what is constrained by a strict locality condition in one language may be free from it in another language, while what is free from the locality condition in one language may be strictly constrained by it in another language. 2 If locality is viewed in terms of the local domain, things will be different as the definition of the local domain may depend on the definition of prominence.

6

 7

Part I

Wh-​questions

8

 9

2  The syntax and semantics of wh-​questions

Introduction Although English and Chinese use different strategies to type a wh-​question (Cheng 1991, 1997), and thus differ from each other with respect to syntactic wh-​movement, both languages involve wh-​expressions that remain in-​situ in overt syntax. In English, wh-​in-​situ only occurs in multiple wh-​questions, as illustrated in (1), whereas in Chinese, wh-​in-​situ is the norm, as shown in (2). (1)  Who t bought what? (2) Ni  mai-​le 

shenme?

you buy-​ASP what ‘What did you buy?’ There are also languages such as Bulgarian, Polish, and Hungarian that require multiple instances of overt wh-​movement to the left periphery of CP (complementizer phrase) in multiple wh-​questions. In the following Bulgarian multiple wh-​questions, all the wh-​expressions move overtly to the sentence-​ initial position, even though all the wh-​ words originate from the same embedded clause, as shown in (3b). (3c) shows that a wh-​word in Bulgarian may not remain in-​situ. Of course, if the relevant wh-​word in (3c) is replaced by a non-​wh-​word, it can stay in-​situ. (3) Bulgarian a.  Koj kakvo na kogo  e   dal?   who what to whom has  given ‘Who gave what to whom?’ (Rudin 1988: 461) b. Koj  kŭde misliš  [če  e  otišŭl _​ _​]?   who where think-​2s that has gone   ‘Who do you think (that) went where?’

10

10 Wh-questions c. *Koj  misliš  [če  e 

 otišŭl  _​  kŭde]?

  who think-​2s that has  gone       where (Rudin 1988: 450) Despite the above differences in overt syntax among languages with respect to syntactic wh-​movement, Huang (1998 [1982b]) proposes that all wh-​expressions in natural language undergo movement to some clause-​initial position to form an operator–​variable relation, though languages may differ in terms of where this movement applies, in overt syntax or at LF. On this view, at some point the English wh-​question (1) and the Chinese wh-​question (2) will be assigned a representation, as demonstrated in (4) and (5) in their syntactic derivation: (4)  [whatj whoi [ti bought tj]] (5) [shenmei [ni mai-​le ti]] Another approach to wh-​interpretation is a nonmovement one, which proposes that all wh-​ expressions, whether moved or not, be bound by an abstract Q(uestion)-​ morpheme. The nonmovement approach is first proposed by Baker (1970), according to whom a Q-​morpheme, as an operator, can unselectively bind all the free variables in its c-​command domain. If this approach is adopted, the LF representations for (1) and (2) would be (6) and (7): (6) [Q [Whoj tj bought whati]]? (7) [Qi [ni mai-​le shenmei]]? In the last 30 years or so, the unselective binding approach has been further developed in Nishigauchi (1986), Pesetsky (1987), Cheng (1991), Aoun and Li (1993), Cole and Hermon (1994), Shi (1994), and Tsai (1994b). Although both the movement and the nonmovement approaches have their respective theoretical and empirical advantages in accounting for wh-​phenomena, they also have their respective limitations when examined carefully against linguistic data in natural language. In the following discussion, I  first investigate these two approaches and present evidence and arguments to show where they fail and then give my own account of wh-​phenomena. Specifically, I will argue that the interpretation of wh-​expression is basically regulated by two factors: locality and prominence. Previous studies pay little attention to prominence except for the notion of c-​command, though enough attention has been paid to locality. I will show that it is the interaction of prominence and locality that determines the interpretation of wh-​elements in Chinese as well as English. The discussion in this part will be organized as follows. In section 2.2, I will review different approaches to wh-​questions and argue that these approaches have their respective limitations in accounting for wh-​phenomena. In section

 11

The syntax and semantics of wh-questions 11 2.3, I  will set up two parameters to classify wh-​questions across languages and develop a clausal typing condition to account for wh-​ phenomena. I  will also discuss the island effects on wh-​interpretation and the multiple wh-​phenomena in both Chinese and English. I  will show that Chinese and English obey the same constraint on the interpretation of wh’s-​in-​situ, though Chinese and English employ different strategies to type wh-​clauses and argue that wh-​island effects are also observed in Chinese in multiple wh-​questions. In section 2.4, I will redefine the wh-​clausal typing condition, distinguishing it from the wh-​interpretation condition and show that wh-​interpretation is constrained by economy considerations. In section 2.5, I will show that the so-​ called additional wh-​effects can be adequately accounted for under the analysis assumed in this part of the book.

2.1  To move or not to move? To move or not to move? That is the question. If one does not assume LF movement for wh-​expressions, one has to explain why some wh-​elements that stay in-​situ exhibit locality effects. If one assumes that all wh-​expressions move either in syntax or at LF, one has problems related to movement. The LF movement approach tries to establish a correspondence between wh-​interpretation and some locality conditions. The ideal picture for this approach is that if a derivation is out, it must result from a violation of some locality conditions at some point of the derivation, in either overt syntax or LF. The problem for this approach is that the picture often cannot be idealized as expected. There are always some sentences that should be out, but are actually in, and some sentences that should be in, but are actually out. Different locality conditions have been proposed to account for wh-​quantification, but these locality conditions often cannot make adequate predictions since they are both too strong and too weak. 2.1.1  LF wh-​movement 2.1.1.1  Superiority effects The Superiority effect is a restriction on the distribution of wh-​expressions involved in multiple wh-​questions, as illustrated below: (8)  a. Who t bought what? b. *What did who buy t? (Pesetsky 2000: 15) (9) a. Whom did John persuade t to visit whom? b. *Whom did John persuade whom to visit t? (Comorovski 1996: 91)

12

12 Wh-questions This effect can be captured by Kuno and Robinson’s (1972) constraint given in (10): (10)  A wh-​word cannot be preposed crossing over another wh-​word. Chomsky (1973) proposes that (10) be replaced by a more general condition, termed Superiority Condition: (11)  Superiority Condition No rule can involve X, Y in the structure … X … [α … Z … -​WYV …] where the rule applies ambiguously to Z and Y and Z is superior to [m-​commands] Y. What Kuno and Robinson (1972) and Chomsky (1973) try to account for is the fact, as exhibited in (8) and (9), that the derivation is made worse by the movement of the lower wh-​expression when more than one wh-​expression is involved. In Minimalist Program, Chomsky (1995: 311) reduces the above arbitrary and structure specific restrictions on overt movement to a more general locality condition on movement, which can be viewed as an instance of the economy strategy of preferring shorter links. (12) Minimal Link Condition (MLC) K attracts α only if there is no β, β is closer to K than α, such that K attracts β. One may wonder if the Superiority Condition can apply to LF structures. At first sight, it seems that the covert LF movement may also exhibit Superiority effects. As observed in Huang (1998[1982b]), wh-​adjuncts are not allowed to occur in the following multiple wh-​questions. (13) a. *What did you buy t how? b. *Who t arrived why? If we assume that wh-​in-​situ must undergo movement at LF, then the unacceptability of (13) can be straightforwardly accounted for by the same independently motivated condition that rules out (8b) and (9b). However, the Superiority Condition fails to account for the acceptability of (8a) and (9a), though capturing (13). This indicates that the Superiority Condition can only apply to the overt syntax. 2.1.1.2  Subjacency In GB (government and binding) syntax, overt wh-​movement is constrained by Subjacency and the Empty Category Principle (henceforth ECP) (Chomsky

 13

The syntax and semantics of wh-questions 13 1973, 1981). Subjacency states that no rule may move an element from Y to X in the configuration in (14) below: (14)  … X … [α … [β … Y …] …] … X … where α and β are bounding nodes. The bounding nodes defined in (14) may be parameterized across languages, and in English, they are two cyclic nodes, NP and S (=IP). As discussed in Huang (1998 [1982b]), the Subjacency condition covers a number of constraints formerly proposed by Ross (1967) such as the Complex NP Constraint (CNPC) and the Sentential Subject Constraint. Hence, the Subjacency condition can adequately account for the ungrammaticality of the sentences below (Huang 1998 [1982b])): (15) *[CP Whoi [IP do you like [NP the books [CP that [IP describe ti]]]]]? (16) *[CP Whoi [IP did [NP [CP that [IP she married ti]]] surprise you]]? In (15) the wh-​expression moves to the matrix C node by crossing an island formed by NP, CP, and IP. In (16) the movement of the wh-​expression crosses an island formed by NP, CP, and IP. In both cases, Subjacency is violated. Although Subjacency can correctly rule out the illicit overt movement of wh-​expressions in (15–​16), it fails to rule in the grammatical covert wh-​ movement at LF. In the following sentences, all the wh’s-​in-​situ are assumed to move to the matrix C at LF. (17) [Who likes [NP books that [IP criticize whom]]]? ‘Which person x and which person y such that x likes books that criticize y.’ (18) [Ni  xihuan [NP[IP piping   you like  



shei]  de 

shu]]?

  criticize  who  DE  book

‘Which person x such that you like books that criticize x.’ (19) [[IP Shei kan  zhe ben shu]   zui   

heshi]?

 who read this CL book  most appropriate

‘What is x such that it is most appropriate for x to read this book.’ In the above English and Chinese examples, all the wh’s-​in-​situ are allowed to move and take matrix scope at LF by crossing two or more bounding nodes on the LF movement approach. Although the assumed LF movement of wh-​expressions in (17–​19) violates Subjacency, these sentences are not ungrammatical.

14

14 Wh-questions 2.1.1.3  ECP Although the covert movement of arguments does not obey Superiority or Subjacency, it seems that the covert movement of adjuncts is constrained by these locality conditions, as demonstrated below: (20)  *Who e arrived why? (21) *[[Ta weishenme xie]  de shu]    zui    youqu   he why      write  DE book  most  interesting Lit. ‘Books that he wrote why are most interesting?’ (Huang 1998 [1982b]: 373) According to Huang (1998 [1982b]), the difference between overt and covert movement with respect to Superiority and Subjacency can be reduced to ECP, which is defined in (22) below. (22) The Empty Category Principle (ECP) A nonpronominal empty category must be properly governed. Proper government is defined below: (23) Proper Government A properly governs B iff A (i) lexically governs B, or (ii) antecedent-​governs B. In (22), “nonpronominal empty category” refers to NP-​traces and wh-​ traces. Since traces created through Quantifier Raising at LF (i.e., raising of a quantified NP or covert movement of a wh-​expression) are also A'-​bound nonpronominal empty categories and are thus on a par with wh-​traces at S-​ Structure, they are also assumed to be subject to ECP. According to Huang, there is an asymmetry between overt syntactic movement and covert LF movement of wh-​expressions with respect to Subjacency and ECP. In overt syntax, both Subjacency and ECP apply, but at LF, only ECP applies. This account captures correctly the adjunct cases in (20–​21) and the Superiority effects with subjects in (8b–​9b). In GB syntax, objects are assumed to be lexically governed by the verb, whereas subjects and adjuncts are not since (i)  INFL, the potential governor for the subjects, is a functional, but not lexical, head incapable of lexically governing the relevant subjects, and (ii) adjuncts are not sisters of verbs and are thus not under the government of verbs. Now if wh-​in-​situ must move at LF, the ungrammaticality of (20–​21) and (8b–​9b) can be accounted for since the traces left by the wh-​in-​situ in these sentences are neither lexically governed nor antecedent-​governed. For

 15

The syntax and semantics of wh-questions 15 instance, in (8b) the wh-​in-​situ who adjoins to what in [Spec, CP] at LF, as shown in (24). From that position, it cannot antecedent-​govern its trace since the Spec of CP is already assigned the index of what. Since the trace is neither antecedent-​governed nor lexically governed, given the assumption that INFL is not a lexical governor, the sentence is ruled out as ungrammatical by ECP. (24)  [CP [Whoi [whatj]]j [C’ did [IP ti buy tj]]]? The same is true for (20–​21). In (20), the trace left by the adjunct why is not lexically governed, and it is not antecedent-​governed, either, since it is adjoined to who at LF, which has already indexed the Spec of CP. In (21), the trace left by the adjunct weishenme “why” is not antecedent-​governed after weishenme moves to [Spec, CP] at LF, since the antecedent-​government is blocked by a sentential subject island. Huang’s (1998 [1982b]) account can explain not only the ungrammatical cases but also the grammatical cases like (8a–​9a), and (17–​19). In (8a), (9a), (17), and (18), the traces left by the LF moved wh-​expression are lexically governed since they function as objects. In (19), the subject trace is also lexically governed since Huang assumes that the INFL in Chinese is lexical, rather than functional, thus significantly differing from the INFL in English. Although Huang’s analysis successfully accounts for some adjunct and Superiority cases, it still leaves other ECP-​related cases unexplained, as demonstrated below: (25) Who reads the books that who writes? (Reinhart 1998: 33) (26) *Whatj did you persuade whom to read tj? (27) a. Which book did which person buy?1 b.*What did who buy? (28) a. *What did who give t to Mary? b. What did who give to whom? (Pesetsky 2000: 17) (29) Zenme shao de  

dan  zui    haochi?

How  cook DE  egg most  delicious Lit. ‘Eggs that are cooked HOW are most delicious?’ (Xu 1990: 370) (30) Shei weishenme mei lai? Who why 



  not come

Lit. ‘Who did not come why?’

16

16 Wh-questions In (25), the wh-​in-​situ would adjoin to the sentence-​initial who at LF and thus fail to antecedent-​govern its trace. Since INFL in English is functional, the wh-​trace is not lexically governed. Hence, the ECP account would predict (25) to be ungrammatical. However, (25) is grammatical. As a matter of fact, the ECP account is not only too strong in ruling out some grammatical sentences but it is also too weak in ruling in some ungrammatical sentences. Specifically, the ECP account would rule in the unacceptable (26) since the traces left by the syntactically moved wh-​expression and the LF moved wh-​ expression are both properly governed as they are objects of a verb, thus satisfying ECP. The ECP account will also have problems in accounting for the contrast between (27a) and (27b) in grammaticality. Although it can correctly rule out (27b), it will incorrectly exclude the grammatical (27a). (28a) and (28b) are also problematic for the ECP account. Although the movement of the wh-​subject who at LF in both cases violates ECP, (28a) is out, but (28b) is in. (29) shows that the Chinese wh-​adjunct zenme “how” can occur within an island. (30) is regarded as an ungrammatical sentence under the ECP account since the trace of weishenme “why” is not antecedent-​governed at LF. However, (30) is not ungrammatical according to our judgment. Shi (1994: 310) points out that the question in (30) asks about the pairing of a set of individuals and a set of reasons. Hence, a statement like (31) can be taken as an appropriate answer. (31)  Zhangsan yinwei   tai mang  mei  lai;   Lisi yinwei shengbing mei  lai. Zhangsan because too busy

not

come  Lisi because be-​sick  not come

‘Zhangsan did not come because he was too busy; Lisi did not come because he was sick.’ (Shi 1994: 311)

2.1.1.4  Crossover phenomena Strong Crossover phenomena are often taken as evidence in support of LF wh-​movement in the literature. (32) a. *Whoi did hei say that Mary helped ti? b. *When did hei say that Mary helped whoi? In (32a) the wh-​expression crosses over the binding pronoun in overt syntax and thus results in a Binding Condition C violation since the trace left by the moved phrase is assumed to be an R-​expression. In (32b), it is assumed that the wh-​in-​situ moves at LF and thus results in a similar violation. Although this kind of analysis can account for the data above, it fails to account for (33). Simpson (2000) notes that in (33) below the wh-​expression does not cross the co-​indexed NP subject either in overt syntax or at LF, but the co-​reference

 17

The syntax and semantics of wh-questions 17 between them is equally impossible. This fact shows that wh-​expressions, just like r-​expressions, are also constrained by the Binding Condition C. (33)  *Janei wanted to know whoi Mary had seen ti. (Simpson 2000: 42) Simpson (2000) further notes that the Strong Crossover account also fails to explain the following sentences: (34) a. *[Whose gossip about which womani]k did Janei fervently deny tk? b. *[Whosei book]k did Johni borrow tk? (Simpson 2000: 42) In (34) the wh-​traces bear different indices from the elements that have been crossed by the moved wh-​ expressions. In order to account for the ungrammaticality of the above sentences, the bracketed wh-​expressions have to be assumed to be reconstructed or copied to their original positions (cf. Chomsky 1993). As a result, the co-​indexation between the c-​commanding subjects and the elements within the reconstructed or copied wh-​expression will be ruled out by Condition C. However, if Binding Condition C can rule out (34), it can also rule out (32b). Hence, it is unnecessary to assume that the wh-​in-​situ in (32b) moves at LF. 2.1.2  D-​linking Pesetsky (1987) argues that it is necessary to make a distinction between two types of wh’s-​in-​situ in terms related to discourse. One type undergoes LF movement, and the other does not. Consider the following examples: (35) a.?*Whatj did you persuade whom to read tj? b. Which bookj did you persuade which man to read tj? According to Pesetsky, the LF movement of the wh-​in-​situ in (35a) would violate Superiority and is thus marked as unacceptable. In contrast, the wh-​ in-​situ in (35b) remains in situ and thus violates no grammatical principles. How is it possible for the wh-​in-​situ in (35b) to involve no LF movement if it needs to be assigned scope? Pesetsky argues that which-​phrases are D-​ linked wh-​expressions, and D-​linked wh-​expressions are not quantifiers and thus have no quantificational force. Following Baker’s approach, he suggests that which-​phrases be unselectively bound by an operator Q. As a result, no Superiority effects are expected because wh-​in-​situ does not move in sentences like (35b). Here, Pesetsky uses two strategies to resolve wh-​quantification: wh-​ movement and unselective binding. For those wh’s-​in-​situ that must obey

18

18 Wh-questions locality constraints, they must move at LF. For those wh’s-​in-​situ that are not constrained by locality conditions, they need not move. Although this account is plausible, it is just a restatement of the facts. The question why D-​linked wh-​expressions need not move is unanswered. Another problem for Pesetsky’s account is that it fails to explain why the following sentence is not fully acceptable, as noted in Comorovki (1996: 85): (36)  ?What did which student read t? If a D-​linked wh-​expression can remain in-​situ, (36) should be fully grammatical since which student does not move at LF and thus has no chance to violate ECP, according to Pesetsky’s analysis. However, (36) is not fully acceptable. Note that it is of no use to say that a which-​phrase should be given the priority to move in overt syntax since the following sentence is still less acceptable even though the which-​phrase is overtly moved. (37) ?Which book did how many people buy? One might argue that the less acceptability results from the LF movement of the non-​D-​linked subject wh-​expression. However, this account will have problems in explaining (i) why (37) is not fully marked as ungrammatical if it violates ECP, and (ii) why (25) is fully grammatical since the LF movement of the non-​D-​linked embedded subject who will also violate ECP if non-​D-​ linked wh-​in-​situ must move at LF, as assumed by Pesetsky. 2.1.3  Pied-​piping The previous discussion shows that wh-​in-​situ fails to exhibit the full range of island effects that characterize overt wh-​movement. In order to account for the scope property of the wh’s-​in-​situ that are not constrained by islands, Huang (1998 [1982b]) has to assume that Subjacency does not apply at LF. Although this is not implausible, there is no independent motivation for assuming that LF is not constrained by Subjacency, as argued in Shi (1994). To circumvent this stipulation, Nishigauchi (1986, 1990) suggests that in cases where a wh-​ expression occurs in an island, the entire island containing the wh-​expression be pied-​piped to the operator position (cf. Choe 1987; Pesetsky 1987). As a result, the wh-​expression is still contained within the island, and Subjacency is still observed at LF. Nishigauchi (1990: 48) gives the following evidence to show why it is the entire island, rather than the wh-​expression contained in it, that is affected by LF movement. (38) Kimi-​wa [NP[IP dare-​ga  kai-​ta] hon] –​o  yomi-​masi-​ta ka? you -​Top 

  who –​N write-​P book-​A read 

‘You read books that who wrote?’

  -​P -​Q

 19

The syntax and semantics of wh-questions 19 Nishigauchi observes that apart from a full-​fledged answer that repeats the entire sentence, an appropriate answer for a question like (38) is (39), instead of (40), though (40) is also a possible answer if treated as a truncated form derived from (39): (39)  Austen-​ga  kai-​ta    

hon   desu.

  -​N  write-​P book  be

‘(It’s) the book that Austen wrote.’ (40) Austen-​desu. ‘(It’s) Austen.’ Hence, what (38) asks for is not the identity of an author, but the identity of the book making crucial use of the identity of the person who wrote it. Along this line, then, the LF representation for (38) should be (41), instead of (42): (41)  [You read y] [NP[[S x wrote] whox] books]y (42)  You read [[x wrote] books][Comp whox ka] In (41), the entire complex NP occupies the operator position and thus matches the appropriate answer (39). Attractive as it may appear, however, Nishigauchi’s pied-​piping analysis runs into a number of problems. It predicts that wh’s-​in-​situ contained in a complex NP cannot take scope over the head of the complex NP, since this analysis attempts to preserve island conditions at LF by assuming that the island pied-​pipes with the wh-​in-​situ to some operator position, and the wh-​ in-​situ never leaves the island. But, as pointed out by Fiengo et al. (1988), certain standard cases of “inversely linked quantification” of the type discussed in May (1977) would require a wh-​expression contained in a syntactic island to be interpreted out of the island, and this is at variance with Nishigauchi’s pied-​piping analysis. Consider example (44) below (Fiengo et al. 1988: 86): (43) Meige ren   dou mai-​le    [yiben [shei xie  

de] shu]?

every man  all buy-​ASP    one  who write  DE book ‘Everyone bought a book that who wrote?’ (44) Daduoshude ren    dou mai-​le    [[shei xie  de]  meiben  shu]? Most  



man  all buy-​ASP   who write DE  every  

book

‘Most people bought every book that who wrote?’ According to Fiengo et  al. (1988), (43) contains three quantifiers, whose scope order, under a natural reading, may be shei “who”> meige ren

20

20 Wh-questions “everyone”> yiben shu “a book.” Although the wh-​expression stays in-​situ, it is separated from the existential quantifier in scope order by the universal. Thus (43) has the interpretation “Who is the person x such that everybody bought one book or another that x wrote?” This scope interpretation is identical to that of the following English sentence in which the wh-​expression overtly moves to the operator position: (45)  Who did everybody see a picture of t? However, if we assume the pied-​piping analysis for (43), this scope interpretation would not be available, for, if the entire complex NP is pied-​piped to the matrix Comp, (i) shei “who” would be unable to take scope over the existential quantifier yiben shu “a book,” and (ii) the wh-​expression and existential quantifiers would not be separated in scope by the universal quantifier meige ren “everyone.” Another problem for the pied-​piping analysis, according to Fiengo et al., is posed by the semantics of the operators like shei xie de meiben shu “every book that who wrote” in (44). The pied-​piping analysis implies that the wh-​operator in the form of a complex NP is existential in nature, adopting the question semantics of Karttunen (1977), but it is clear that this complex NP is also a universal quantifier, as it takes a universal determiner meiben “every.” If it is the whole complex NP that is quantified, then the pied-​piping analysis would predict that the wh-​expression shei “who” and the universal quantifier meiben shu “every book” cannot take on different quantificational forces, which is inconsistent with the quantificational facts in (44). A third type of problem with Nishigauchi’s analysis, as pointed out in Fiengo et al., is that sometimes the wh-​word contained in an island prefers to have an elliptical answer that does not repeat the island. In the following sentence, contrary to those cases involving complex NPs, the most natural elliptical answer to a question whose wh-​expression occurs in the sentential subject is the one which only provides the value of the wh-​expression, without repeating the whole sentential subject (Fiengo et al. 1988: 85): (46) [shei  kan   zheben  shu]   zui    heshi? who  read  this 

  book  most appropriate

‘That who read this book is most appropriate?’ a.  *Zhangsan kan zheben shu.     ‘That Zhangsan read this book.’ b.   Zhangsan As shown in (46), the most appropriate answer for the above Chinese question is not the elliptical answer that repeats the entire sentential subject but the answer that spells out only the value of the wh-​expression. If repeating

 21

The syntax and semantics of wh-questions 21 the entire island in an elliptical answer is taken to be symptomatic of pied-​ piping, then the impossibility of doing this in (46b) will suggest that no pied-​ piping of the entire clause should take place. But if this is really the case, the pied-​piping analysis cannot be maintained since this would indicate that island conditions are violated at LF, contra the basic assumption taken in the pied-​piping analysis that island conditions are not violated at LF. Apart from the problems pointed out in Feingo et al. (1988), Nishigauchi’s pied-​piping analysis also has problems in accounting for the contrast in grammaticality among the following sentences: (47)  [Ni xihuan [NP[IP piping   you like  



shei] de   shu]]?

  criticize  who DE book

‘Which person x such that you like books that criticize x.’ (48) *[[Ta weishenme xie]  

de

shu] 

zui 

youqu

  he why      write  DE book most interesting Lit. ‘Books that he wrote why are most interesting?’ (49) [Zenme shao de   dan] zui    haochi? how  cook DE 

egg  most  delicious

Lit. ‘Eggs that are cooked HOW are most delicious?’ To account for the contrast between (47) and (48), Nishigauchi suggests that a wh-​expression must be identical in syntactic category with the dominating node in order for the [+WH] feature to be percolated to the latter. In (47) the wh-​expression shei “who” is [+N] so that its [+WH] feature can climb up to the complex NP as a result of feature percolation. In (48), the wh-​expression weishenme “why” is [-​N], so its [+WH] feature cannot be percolated to the complex NP. As a result, the relevant complex NP will have no chance to be identified as [+WH], and, accordingly, will not be pied-​piped to the operator position to take scope. If weishenme alone moves to the operator position, Subjacency will be violated. Hence, the ungrammaticality of (48). Although this analysis can account for the ungrammaticality of (48), it fails to do so for the grammaticality of (49). In (49) zenme “how” is also [-​N], but the sentence is grammatical as a wh-​question. In order to circumvent the problems found in Nishigauchi’s analysis while maintaining the idea that Subjacency Condition holds at LF, Fiengo et  al. take Nishigauchi’s analysis one step further to allow for both pied-​piped movement and extraction of wh-​expressions out of the pied-​piped constituent. Adopting the stipulation from Chomsky (1986b) that an A'-​binder will lose its barrierhood, they propose that the wh-​in-​situ first be pied-​piped together with the entire island to the IP-​adjoined position, and the wh-​expression then be raised out of the island to the operator position since the relevant island loses its barrierhood when IP-​adjoined.

22

22 Wh-questions Under Fiengo et al.’s analysis, wh-​expressions contained in a complex NP are now able to take scope over the head of that complex NP, as they end up in an operator position outside the complex NP. It seems that their analysis can solve some of the problems associated with Nishigauchi’s pied-​piping analysis. However, Fiengo et al.’s analysis has its own share of problems, as pointed out by Shi (1994: 304). Shi argues that their analysis has problems in accounting for the following Chinese sentence which involves a wh-​expression in a complex NP that is inside another complex NP. (50) Ni renshi [NP1[ti zhuadao [NP2[tj sha shei] de [xiongshouj]] de neige jingchai] you know   

catch     

  kill who DE murderer  

DE that policeman

ne? Q ‘You know the policeman that caught the murderer that killed whom?’

In (50), the LF movement of the wh-​ in-​ situ will inevitably violate Subjacency no matter what pied-​piping strategy is taken. If the inner complex NP (NP2) is pied-​piped, the raising will violate Subjacency by crossing the outer complex NP (NP1). Of course, one can assume that NP1 is adjoined to the matrix IP first, and NP2 is then adjoined to the IP of the inner relative clause of NP1, and after adjoining to the IP of the inner relative clause, the wh-​expression finally moves to the Spec of the matrix CP. However, as pointed out in Shi (1994), the final process of raising the wh-​expression to the matrix operator position in (50) would cross two CP bounding nodes which are not L-​marked and thus barriers. Hence, the grammaticality of (50) would be left unaccounted for if one assumes that Subjacency is inviolable at LF. In addition to this problem unsolved, Fiengo et al.’s analysis also fails to account for the grammaticality of (49) since the LF movement of zenme “how” would violate ECP. The above discussion shows that neither the pied-​piping analysis proposed by Nishigauchi (1986) nor the one further revised in Fiengo et al. (1988) can adequately explain why sometimes locality conditions are violated, but the relevant sentences are still grammatical. 2.1.4  Unselective binding and choice function The alternative to the LF movement analysis of wh’s-​in-​situ is an operator binding approach, which makes use of the notion of unselective binding developed in Kamp (1981) and Heim (1982). On this approach, all wh-​expressions, whether moved or not, should be treated as variables and unselectively bound by an operator in its c-​command domain. In the last 30  years, the unselective binding approach has been further developed in Nishigauchi (1986), Pesetsky (1987), Cheng (1991), Aoun and Li (1993), Cole and Hermon (1994), Shi (1994), and Tsai (1994b).

 23

The syntax and semantics of wh-questions 23 An advantage of treating wh-​expressions as variables is that it captures the fact that in many languages wh-​expressions do not have fixed quantificational force and can be assigned different interpretations by different operators (cf. Haspelmath 1997). For instance, a wh-​expression in Chinese is often ambiguous between a question and statement reading, as pointed out by Xu (1990: 357): (51)  Zheli que-​le 

shenme

here miss-​ASP what a. What is missing here? b. There is something missing here. In (51) the wh-​phrase shenme can be interpreted either as an interrogative phrase, meaning “what,” or as an existential indefinite, meaning “something.” As discussed in Huang (1998 [1982b]), Lee (1986), and Shi (1994), Chinese wh-​ expressions like shenme “what,” shei “who,” nali “where,” zenme “how,” and so on can also be interpreted as universal quantifiers when they are licensed by dou “all.” Since wh-​expressions can achieve different quantificational forces when licensed by different operators, it is not unreasonable to assume that they are interpreted as interrogative phrases only when they are licensed by a Q operator. Although the unselective binding approach can adequately account for the fact that wh-​expressions may function either as interrogative NPs or indefinite NPs in natural language, it has its own problems. First, it fails to explain why the scope interpretation of some wh-​expressions like why and how in English and weishenme “why” in Chinese is highly constrained, as demonstrated in (13) and (21). The unselective binding approach assumes that a Q operator can license any number of wh-​expressions as long as they occur in its c-​ command domain (Baker 1970). Since the relation between the operator and the relevant variables is not constrained by locality conditions, it is expected that ECP or Subjacency effects will not arise when a wh-​phrase occurs in an island. But (21) shows that locality effects do arise when weishenme “why” occurs in an island. Another problem with the unselective binding approach, as pointed out in Reinhart (1997, 1998), is that it fails to assign an adequate interpretation to sentences involving a conditional. Consider the following examples adopted from Reinhart (1998: 36–​37): (52) Who will be offended if we invite which philosopher? Wrong: (53) a. for which , if we invite y and y is a philosopher, then x will be offended b. {P│(∃)(P=^((we invite y & philosopher(y)) → (x will be offended)) & true(P))} c. Lucie will be offended if we invite Donald Duck.

24

24 Wh-questions Right: (54)  a. for which , y is a philosopher and if we invite y, x will be offended b. {P│(∃) ((y is a philosopher) & P=^((we invite y) → (x will be offended)) & true(P))} c. Lucie will be offended if we invite Donald Duck. In (52), a wh-​in-​situ (i.e., which philosopher) occurs in an if-​clause. If we apply the unselective binding analysis to represent (52), we get the LF representation (55): (55) Q[whoi will be offended if we invite which philosopherj] (55) is equivalent to (53a), but not to (54a). The semantic representation for (55)/​(53a) is (53b), where wh-​expressions are treated as existential NPs, and the question denotes the set of propositions which are true answers to it. The problem with (55)/​(53a) is that the restriction is left in-​situ. If ​(53a) is the question expression for (52), anything that is not a philosopher could be a value for y in (53a), given the representation (53b). Hence, (53c) could be one of the possible answers to (52), which is an undesirable result unexpected by the unselective binding approach. (54a) yields the correct set of answers for (52), since in (54a), the restriction is pulled out of the antecedent clause of an implication. To circumvent the interpretative problems with the unselective binding approach, Reinhart proposes that wh’s-​ in-​ situ be interpreted via choice functions, and it is the function variable, rather than the individual variable, which is long-​distance bound by the existential Q operator. The choice function can assign wide scope to wh-​in-​situ by allowing existential quantification over choice function (notated as “CH(f)”), and at the same time restricts the values that could possibly be assigned to the wh-​variable to the domain defined by the N-​restriction. According to Reinhart, the semantic representation for (52) is (56), where the choice function f bound by the question operator selects a value from the philosopher set. (56) {P│(∃)(CH(f) & P=^((we invite f(philosopher)) → (x will be offended)) & true(P))} What (56) states is that a function f exists such that for some person x, if we invite the philosopher selected by f, x will be offended. In (56), f(philosopher) is in argument position, and it denotes the value of the function f. I think that Reinhart’s choice function analysis is the right approach to solve the problems with wh’s-​in-​situ, but one issue that this analysis has to address is how to account for the assumed argument-​adjunct asymmetry in wh-​ interpretation. Reinhart (1998) claims that adverbial adjuncts cannot

 25

The syntax and semantics of wh-questions 25 be interpreted via choice functions and thus cannot be interpreted in-​situ. However, this account still cannot explain why (29) and (30) are grammatical, though a wh-​adverbial occurs in an island in these sentences. 2.1.5  Referentiality and nonreferentiality Xu (1990) observes that Chinese wh-​adjunct zenme “how” can take matrix scope while occurring in an island, as shown in (29), repeated below as (57): (57)  Zenme shao de  

dan zui

  haochi?

How  cook DE egg most   delicious ‘Eggs that are cooked HOW are most delicious?’ But it is also observed that Chinese wh-​expression weishenme “why” is highly restricted in scope interpretation, as shown in (21), repeated below as (58): (58) *[[Ta weishenme xie]   de shu] 

zui    youqu

  he why      write  DE book most interesting ‘Books that he wrote why are most interesting?’ Based on these facts, Lin (1992) claims that there is an asymmetry between zenmeyang “how,” a variant of zenme,2 and weishenme “why” in regards to LF movement. In order to account for the asymmetry between zenmeyang and weishenme in scope interpretation, Hua (2000) stipulates that zenmeyang is a variable, but weishenme is an operator. Although this account captures the fact that weishenme cannot be interpreted out of an island, it is just a restatement of the fact since the question why zenmeyang is a variable and weishenme is an operator is left unexplained. Tsai (1994a) argues that the relevant asymmetry at LF is not between zenmeyang and weishenme, in particular, but between referentiality and nonreferentiality among wh-​expressions in general. He suggests that wh-​arguments and referential wh-​adjuncts (when, where, instrumental how, and purpose why), as one group, be differentiated from nonreferential wh-​adjuncts (manner how and reason why) as another group in LF scope interpretation since only the former, but not the latter, are allowed to take scope out of islands at LF. To show that such a distinction may be necessary, Tsai observes that the questions below differ significantly in their acceptability: (59) a.  [Ta zenmeyang chuli   he how   

zhe-​jian shi]    bijiao qiadang?

  handle  this-​CL matter  more appropriate

‘what is the means x such that it is more appropriate [for him to handle this matter by x]?’

26

26 Wh-questions b. *[Zhe-​jian shi,   ta chuli-​  de  zenmeyang] bijiao qiadang?   more  appropriate

This-​CL matter  he handle DE  how 

‘what is the manner x such that it is more appropriate [for him to handle this matter in x]?’ (60)  a.  [Women wei(-​le)shenme nianshu] cai   we 

  for    what   study

you   yiyi?

just  have

meaning

‘which purpose x such that it is just meaningful [for us to study for x]’ b. *[Women weishenme nianshu] cai   you     we    why      study  

just have

yiyi? meaning

‘what is the reason x such that it is just meaningful [for us to study for x]’ (59a) is acceptable when the wh-​expression is associated with an instrumental reading, while (59b) is unacceptable when it seeks for an answer with a manner reading. (60a) is grammatical when the wh-​expression concerns the purpose, whereas (60b) is ungrammatical when it concerns the reason. In order to account for the contrast in acceptability among the above sentences, Tsai (1994a) labels the instrumental how and the purpose why as referential, and the manner how and the reason why as nonreferential. For Tsai, the referential vs. nonreferential distinction plays a crucial role in accounting for the differences of these wh-​expressions in taking scope. He assumes that sentential subjects, relative clauses, appositive clauses,3 and [+N] complement clauses subcategorized for by verbs such as yihan “regret,” jide “remember,” and tongyi “agree” constitute nominal islands that block the extraction of nonreferential wh-​expressions at LF.4 Although it is insightful and very close to the solution of the problems by proposing a referentiality/​nominality vs. nonreferentiality distinction in accounting for wh-​quantification in Chinese, Tsai’s (1994a) account is still problematic in a number of aspects. First, the notion of referentiality is not well-​defined, as noted by Tsai (1994a:  130) himself. If there is no reliable criterion to make a distinction between referentiality/​nominality and nonreferentiality, then it is stipulative to say that the instrumental how and the purpose why are referential, and the manner how and reason why are nonreferential. Since no clear distinction can be made between referentiality and nonreferentiality, this account is just a restatement of the fact that these two groups of wh-​expressions are different, and the reason why they are different in scope interpretation is still left unaccounted for. In fact, the notion of referentiality has been widely explored in the literature. Rizzi (1990) associates referentiality with referential theta roles. Cinque (1990) further refines Rizzi’s (1990) analysis by defining the notion of referentiality as “the ability to refer to specific members of a set in the mind of the

 27

The syntax and semantics of wh-questions 27 speaker or preestablished in discourse” (Cinque 1990: 16). Based on Rizzi’s and Cinque’s theory of referentiality, Chung (1994: 38) uses the notion of familiarity and descriptive content to define referentiality. Kiss (1993) also employs the notion of referentiality (under the term SPECIFICITY) to account for the wh-​phenomena. According to Kiss (1993: 87), a wh-​phrase is specific if it ranges over a set which is familiar to the participants of the discourse. Although various definitions of referentiality are available, Tsai does not specify clearly which definition of referentiality is employed in his analysis. Since he does not show how to tie the instrumental how and the purpose why to the notion of referentiality, we fail to see why the instrumental how and the purpose why are referential, but the manner how and reason why are nonreferential. Second, Tsai’s analysis fails to explain why in the following sentence, the purpose why (i.e., wei(-​le)shenme “for what”) still cannot be interpreted out of the assumed A-​not-​A island even when it is referential, according to Tsai: (61)  *Ni xiangzhidao [ta wei(-​le)shenme lai-​bu-​lai]? you wonder 





he for-​what 

come-​not-​come

‘What is the purpose x such that you want to know that he will come or not come for x?’ According to Tsai’s analysis, the purpose why, as a referential wh-​adjunct, should be unrestrained when occurring in an island, but in the above sentence, it is obviously blocked for taking matrix scope. Third, as pointed out in Hua (2000), zenmeyang as a manner how can also be interpreted out of the island, as shown in the following sentence, where (62b) indicates that zenmeyang can have the manner reading: (62)  [Li xiaojie  zemeyang tiaowu]   zui    haokan? Li miss 

  how 



dance    most   beautiful

‘what means/​manner x such that Miss Li looks the most beautiful to dance by/​in x.’ a.  Yong  jiaojian tiao   (zui    haokan).     by    toe    dance  most

beautiful

    ‘By dancing on toes.’ b.   Man-​man   de    tiao      slow-​slow

DE

(zui    haokan)

dance  most

beautiful

    ‘Slowly.’ Fourth, Tsai fails to explain why in the following sentence the A-​not-​A operator, a nonreferential wh-​operator, can be interpreted out of a nominal island:

28

28 Wh-questions (63)  Ni dasuan 

[xia-​ge   yue    qu-​bu-​qu meiguo]?

you plan   next-​CL month  go-​not-​go America Lit. ‘Do you plan to go or not to go to America next month?’ (64) Ni  jihua  you plan 

[xia-​ge  yue 

qu-​bu-​qu  meiguo]?

next-​CL month go-​not-​go  America

Lit. ‘Do you plan to go or not to go to America next month?’ According to Tsai (1994a: 150–​151), the following sentences are ungrammatical because the complement clauses selected by dasuan “plan” and jihua “plan” are [+N] and thus constitute nominal islands that will block the extraction of nonreferential wh-​elements. (65) *Lisi

dasuan [weishenme mai diannao]?

Lisi  plan 

why   

  buy computer

‘What is the reason x such that Lisi plans [to buy a computer for x]?’ (66) *Lisi jihua Lisi plan  

[weishenme mai diannao]? why   

  buy computer

‘What is the reason x such that Lisi plans [to buy a computer for x]?’ However, this analysis obviously fails to account for the grammaticality of (63) and (64). It seems that the only way out is to treat the A-​not-​A operator as a referential element, but this will leave the ungrammaticality of sentences like (67) unaccounted for: (67) *[Ta qu-​bu-​qu meiguo] bijiao hao? he  go-​not-​go America more good Lit. ‘Is it better whether he goes to America or not?’ In (67), the A-​not-​A operator cannot take scope out of the subject island. Fifth, according to Shi (1994: 306), zenme “how” can also be used as a verb in Chinese, and in this case, we have good reasons to believe that zenme is not referential/​nominal in any syntactic or semantic sense. But in the following sentence, the verb zenme can take matrix scope out of an island: (68) a.  [Ta ba ni

zenme le]

  he BA you how

rang ni  zheme shangxin?5

ASP let 

you so    sad

  Lit. ‘He treated you HOW so as to make you so sad?’ b. [[Ta ba Lisi zenme le]   he BA Lisi how

de shuofa]  bijiao kexin?

ASP DE claim 

more reliable

  Lit. ‘The claim that he treated Lisi HOW is more reliable?’

 29

The syntax and semantics of wh-questions 29 In (68), Tsai (1994a) would predict that the matrix scope interpretation of the nonreferential wh-​verb should be blocked by the nominal subject and Complex NP islands. But the nonreferential wh-​verb does take matrix scope in (68), without resulting in any ungrammaticality of the sentence.

2.2  Prominence and locality: Theorizing the interpretation of wh-​questions Hua (2000: 43) claims that wh-​expressions in natural language may be ordered along a scale as in (69), with those towards the left more likely to function as variables, and those towards the right to function as operators. Hence, according to him, the difference between Chinese and English with respect to wh-​interpretation lies in the fact that Chinese allows more wh-​expressions to function as variables, down to the manner adverb zenmeyang “how,” whereas English allows more wh-​expressions on the scale to function as operators, up to the manner adverb how. (69)  which/​who/​what ← when/​where ← how ← why Under Hua’s analysis, a binary distinction is made between wh-​operators and wh-​variables on the two extremes of the scale. This distinction implies that the wh-​expressions on the two extremes of the scale given in (69) are of two absolutely different types, each of which is homogeneous in terms of scope interpretation. Obviously, this account fails to cover the fact that wh-​ expressions on the two extremes can both function as operators and variables. For instance, in Bulgarian all wh-​ expressions are operators. Nishigauchi (1990:  93) notices that the acceptability of the following Japanese sentence would be improved when another wh-​expression of the category NP appears within a complex NP island, in addition to the wh-​adjunct why: (70) ?(?)[NP [Dare-​ga naze kai-​ta] hon]-​ga omosiroi-​desu-​ka?       Who –​N why wrote book-​N  interesting-​be-​Q Lit. ‘Books that who wrote why are interesting?’ If why is an absolute operator, the improved acceptability of the above sentence is unexpected under Hua’s analysis, since this sentence should be ruled out by ECP no matter what additional wh-​elements are added. Besides, Hua (2000) fails to explain why it becomes easier to take scope out of an island for the wh-​expressions closer to the left extreme of the scale, while it becomes more and more difficult to do so for those closer to the right extreme of the scale. Specifically, he fails to explain why some wh-​expressions tend to be operators while others can be variables. To simply make a distinction between wh-​operators and wh-​variables is just a restatement but not an explanation of the relevant facts that wh-​expressions may behave differently with respect to scope interpretation.

30

30 Wh-questions 2.2.1  Focus and wh-​interpretation via choice functions The previously discussed data seem to suggest that there is an asymmetry between the reason weishenme “why” and other wh-​expressions in Chinese. In the following, we will explain why the reason weishenme “why” cannot be interpreted out of islands. Xu (1990: 371) observes that there are two stress patterns for weishenme. When weishenme is pronounced with stress on the first syllable wei, it cannot be interpreted out of islands. When stress falls on shenme instead of wei, it can be interpreted out of islands. To distinguish these two stress patterns, I represent the former as weishenme, and the latter as weishenme. It is obvious that weishenme corresponds to Tsai’s reason why, and weishenme corresponds to his purpose why, which can be substituted for by wei(-​le)shenme (cf. Huang 1998 [1982b]).6 When we associate these two stress patterns with their respective focus properties, we can see the reason why weishenme cannot be interpreted out of islands. It is well-​known that inherently D-​linked which-​ phrases can violate locality constraints for interpretation in English, as shown in (27a), repeated in (71) below: (71)  Which book did which person buy? The question to ask is why D-​linked wh-​phrases can violate the locality constraint. I think that the answer lies in the prominence of these wh-​phrases since an inherently D-​linked wh-​phrase ranges over some set that constitutes the common ground of communication, and it is in this sense that it is D-​linked. What is important here is that an inherently D-​linked wh-​phrase is licensed by a restricted set that belongs to the shared knowledge of the speaker and the hearer. As a first approximation, I assume that if a wh-​element can be licensed by an accessible alternative set when it is in focus, it can be interpreted out of islands. Here lies an important difference between weishenme and weishenme. When wei “for” is focused, no accessible alternative set can be generated. Whereas, when shenme “what” is focused, at least an accessible alternative set of lexical items can be generated. When shenme is focused in weishenme, it is licensed by an accessible set of lexical items whose members may include zhe “this” and na “that,” and thus weishenme can be answered by wei-​zhe “for this” or wei-​na “for that.” In fact, we can further claim that wei in weishenme can only receive phonological stress, but not semantic focus, since it cannot generate an alternative set when it is focused by the contrastive focus marker shi, as shown in (72a). (72) a. *[Women shi weiFsheme  nianshu] cai   you  yiyi?     we    FM why 

  study   just  have meaning

‘what is the reason x such that it is just meaningful [for us to study for x]’

 31

The syntax and semantics of wh-questions 31 b. *Women  shi  weiFsheme nianshu?     we  

  FM  why   

study

Lit. ‘Why is it that we study?’ c. [Women shi wei(-​le)shemeF nianshu] cai   you  yiyi?     we   FM for    what   study  

just  have meaning

‘which purpose x such that it is just meaningful [for us to study for x]’ Notice that weishenme cannot be licensed by an alternative set when it is focused by shi even though it does not occur in an island, as shown in (72b). I assume that the contrastive focus marker shi, different from the copula shi “BE,” presupposes the existence of a set. This means that when an element is focused by shi, it always implies that there is an alternative set available, of which the focused element is a member. Wei cannot be focused by shi simply because wei is a function word that does not have any lexical content to receive semantic focus. Different from wei in weishenme, shenme in weishenme can be licensed by an alternative set when it is focused by shi, as shown in (72c) above. Huang (1982a) notices that when the focus marker shi co-​occurs with a wh-​word in the same clause, shi cannot focus an element other than the wh-​ word, as shown in (73a) and (73b) below. This indicates that in wh-​questions a wh-​word bears an inherent focus feature and thus must be the only element that is overtly marked by a focus marker. In this aspect, weishenme “why” is, again, different from other wh-​words. In a wh-​question sentence that is typed by the wh-​word weishenme, the focus marker shi can focus elements other than weishenme, as shown in (73c). As a matter of fact, it is the wh-​word weishenme, not elements other than it, that cannot be focused by shi, as shown in (72b) above and (73d) below. This fact demonstrates that weishenme does not bear an inherent focus feature, since if it does, it will not allow elements other than it to be focused by shi in its domain. Since weishenme cannot be focused by the focus marker shi, it can be concluded that it, different from other wh-​words, not only bears no inherent focus feature, but also rejects it. (73)  a. *Shi ZhangsanF da-​le      shei?   FM Zhangsan beat-​ASP who   Lit. ‘Who is it Zhangsan that beat?’ b.   Shi sheiF

da-​le    ta?

  FM who  beat-​ASP him   Lit. ‘Who is it that beat him?’ c.  Weishenme shi ZhangsanF bixu cizhi?   why      FM Zhangsan  must resign   Lit. ‘Why is it Zhangsan who must resign?’

32

32 Wh-questions d. *Shi weiFshenme Zhangsan bixu cizhi?   FM why  



  Zhangsan must resign

  Lit. ‘Why is it that Zhangsan must resign?’ The focus marker test can show clearly that zenme(yang) patterns with weishenme rather than weishenme since zenme(yang), when focused by shi, can be licensed by an alternative set of lexical items like zheme(yang) “like this” or name(yang) “like that,” which may provide a value for the focused element: (74)  Shi Zenme(yang)F shao de  dan cai zui   haochi? FM how        cook DE egg just most  delicious ‘Eggs that are cooked HOW are most delicious?’ Zenme(yang) can be focused by focus marker shi since zenme is a lexical rather than a function element, and thus focusing zenme can also generate an accessible set of lexical items over which the focused zenme ranges. Hence, zenme(yang), at least, can be answered by zheme(yang) or name(yang). Notice that in this respect, A-​not-​A morpheme patterns with weishenme since it also cannot be focused by shi: (75) *Ni shi   qu-​bu-​quF meiguo? you FM  go-​not-​go America ‘Do you go or not go to America?’ Pan (personal communication) suggests that the A-​not-​A element is a compound. As a first approximation, I assume that the A-​not-​A element is a compound formed by the incorporation of the Q morpheme (cf. Shi 1994). If this is really the case, the reason why it cannot be focused by shi is obvious since it is not possible to construct a set of entities out of part of a compound. For instance, one cannot construct a set of books from the compound “book-​reading.” Now, we can see why weishenme and the A-​not-​A element cannot be interpreted out of islands. The reason is very simple: they cannot receive contrastive/​semantic  focus. Then, the question to ask is why a focused element can be interpreted out of islands. This is because under our analysis a wh-​in-​situ must be interpreted via the choice function application, and only when a wh-​word can be licensed by an alternative set can it be interpreted via the choice function application. Given that a semantically focused element always implies the existence of an alternative set, it thus becomes possible for it to be interpreted out of islands via the choice function. The difference between weishenme “why” and zenme(yang) “how” in Chinese and why and how in English is that for the

 33

The syntax and semantics of wh-questions 33 former, there is an accessible set of lexical items to license it, but for the latter, such a set is not available. It is at this point that Tsai’s (1994a) idea of referentiality vs. nonreferentiality distinction is insightful, though I would prefer to say that the real distinction is between lexical/​notional elements and function elements. The Chinese weishenme “why” and zenme(yang) “how” are made of more than one element and thus can be disintegrated into two separate parts, of which one part is a lexical element, and the other part is only a grammatical morpheme. Since the lexical element can receive semantic focus and be licensed by an accessible alternative set, it is interpretable in an island when the choice function operation applies. The English why and how, however, are different. Because they are function elements, they cannot be licensed by an alternative set when focused. Hence, they are uninterpretable when occurring in an island since the choice function cannot apply to them. Notice that we do not mean that why, how in English, and weishenme in Chinese cannot range over some sets that might be possibly imagined when we say they cannot be licensed by an alternative set. What we mean is that they neither range over a set of elements that are of discrete properties or can be individuated nor range over a set of elements that can be presupposed by the speakers and hearers. I argue that only the set that is tied to the shared knowledge of the speakers and hearers can be presupposed and thus constitute the common ground of communication. And only a presupposed set can have the possibility of being, at least conceptually, individuated. The key point here is that why, how in English, and weishenme in Chinese are the very wh-​words that simply reject presupposition. Hence, even if there were a set associated with these wh-​ words, it is not in the common ground of communication. If this analysis is on the right track, then it is safe to say that the sets (if there is one) associated with these wh-​words for interpretation are different from the sets that the wh-​words other than why, how, and weishenme range over. The sets that why, how, and weishenme may range over are open sets. They are not presupposed and are thus not supposed to be in the common ground of the speakers and hearers. Other wh-​words like wh-​arguments denoting entities or wh-​adjuncts denoting time or location range over some sets that belong to the shared knowledge of the speakers and the hearers. Hence, they are presupposed and are thus in the common ground of communication. Based upon the above discussion, we hypothesize that wh-​ elements can be interpreted out of islands if it satisfies the following Wh Focus Condition: (76)  The Wh Focus Condition a. A wh-​word can bear focus iff it can range over a presupposed set of entities. b. A wh-​element can be interpreted out of an island iff it can bear focus.

34

34 Wh-questions Note that under the present analysis, wh-​expressions that are not inherently D-​linked can also be interpreted via the choice function if they range over some presupposed set, and this is actually what Reinhart (1998) has proposed. According to Reinhart’s analysis, pronominal wh-​phrases can be treated as determiners. For instance, the wh-​word who can be treated as a determiner, with the noun position empty. The empty N can be viewed either as denoting the set of entities in the model or as containing the selectional-​ restriction of the determiner (such as animacy). 2.2.2  Clausal typing and wh-​feature strength: Two parameters in classifying wh-​questions Cheng (1991, 1997) assumes that the overt syntactic wh-​movement is triggered neither by the strength of the wh-​feature nor by the need for scope-​marking. Instead, she proposes that wh-​movement be driven by the need for a clause to have a specified type. Her hypothesis for clausal typing is given below: (77)  Clausal Typing Hypothesis (Cheng 1997: 22) Every clause needs to be typed. In the case of typing a wh-​question, either a wh-​particle in C° is used or else fronting of a wh-​word to the Spec of C° is used, thereby typing a clause through C° by Spec-​head agreement. Following Cheng, I assume that different languages may use different strategies to type a question. In English, a wh-​word must move to the sentence-​ initial position to type the relevant sentence. In Chinese, a question is typed by a Q(uestion) particle. Hence, wh-​movement is unmotivated. As a first approximation, I assume that languages can be classified into three types with respect to wh-​quantification:  wh-​operator languages, wh-​variable languages, and wh-​ operator/​variable languages. Languages such as Bulgarian and Chinese are on the two extremes, with English standing in between. In wh-​operator languages, all wh-​expressions must move overtly to the operator position, whereas in wh-​ variable languages, all wh-​elements should stay in-​situ. In wh-​operator/​variable languages such as English, at least one wh-​element needs to move to the operator position to type the sentence, while the rest of the wh-​elements can stay in-​situ and be interpreted in-​situ as variables, as suggested by Reinhart (1997, 1998). I assume that all these three types of languages need to do two things to make a question:  (i) to choose an appropriate wh-​clausal typer, and (ii) to associate the wh-​word with C for interpretation. Since both English and Chinese have wh’s-​in-​situ, it is not unreasonable to assume that, although English uses a different strategy to type a wh-​question, there is no fundamental difference between wh-​elements in English and in Chinese in terms of wh-​quantification force. That is, wh-​elements in these two languages do not have inherent quantificational force. The only reason why English raises one wh-​element to the operator position is that the Q particle strategy used in Chinese is not available in this language. In this respect, Bulgarian-​type

 35

The syntax and semantics of wh-questions 35 languages are fundamentally different from both English and Chinese since wh-​elements in these languages are inherently quantificational. The question to ask is why these three types of languages exhibit different quantificational force. I think that the answer lies in the different morphological properties of wh-​expressions in these languages. Cheng (1991, 1997) observes that in multiple wh-​fronting languages, such as Bulgarian, Polish, and Hungarian, wh-​words can be used to form indefinite NPs by either prefixing or suffixing certain particles to them, as shown by the following Bulgarian data cited from Cheng (1997: 65): (78)  kój     who      njákoj    someone kudé   where    njakude   somewhere koga    when 

     njakoga    sometimes

kakvó  what sort of  njakakvo  some sort of Based on the above data, Cheng claims that the indefinite reading of the wh-​ words comes from the presence of an affix, which contributes existential force. Take the Bulgarian njákoj “someone” as an example. Under Cheng’s analysis, this word, made up of njá and koj, forms a DP (determiner phrase) structure as given in (79a), with koj as the core, which has no quantificational force, and njá as the determiner that contributes the existential quantificational force. As for the bare wh-​words, they may have a DP structure like the one given in (79b). (79)

a.

DP

D’

D

NP

njá

koj

b.

DP

D’

D

NP

Φwh

wh-word

36

36 Wh-questions Cheng proposes that when the bare wh-​word is interpreted as an interrogative, the D position is occupied by a null determiner which can be considered as the null counterpart of which in English and thus has interrogative force: [D Φ[+wh]]. According to Cheng, the whole DP has to move to [Spec, CP] since the null determiner [D Φ[+wh]] must be licensed by a C° marked with [+WH]. Tsai (1994b, 1999a, 1999b) claims that in English wh-​expressions form an operator–​variable structure at the lexical level. He observes that English wh-​words and pronominals are more or less built on the same materials, as shown below: (80)  a. wh-​words        b. pronominals    wh-​o    wh-​en   

th-​ey    th-​en

   wh-​om   wh-​ere  

th-​em    th-​ere

   wh-​at         



th-​at

The only difference between the above two groups of words is that the prefix for pronominals is th-​, while that for wh-​words is wh-​. Tsai notices that there is a crucial difference between these two morphemes: th-​, as a reduced form of English definite article the, blocks binding from the suffix –​ever (*theyever) since it can license the indefinite morpheme it attaches to. Whereas, wh-​ does not block binding from ever (e.g., whoever, whatever, whenever, etc.), which assigns universal force to the indefinites, since it cannot act as a determinant of quantificational force. On the basis of this observation, Tsai (1999b: 46) proposes the following morphological structure for English interrogative wh’s: (81)





Wh-

Opx [Q]

ind. (x)

I adopt Tsai’s basic assumption without further argument that wh-​does not have quantificational force in English, but I do not think that an operator–​ variable structure is formed for English wh-​words at the lexical level since there is no convincing evidence to show that the operator–​variable structure is formed at that level. Tsai suggests that only wh-​expressions, such as what, that can be “quantified” by -​ever or some-​, form an operator–​variable structure at the lexical level. However, as pointed in Hua (2000), if this morphological criterion were adopted, the English wh-​adverb how would also form an operator–​ variable structure at the lexical level since we can derive such expressions as however and somehow from how. But this would leave the ungrammaticality of the following sentence, repeated from (13a), unaccounted for:

 37

The syntax and semantics of wh-questions 37 (82)

*What did you buy how?

In (82), the clause is typed by what, and since how can be quantified by –​ever or some-​, and thus form an operator–​variable structure at the lexical level in English, it can have an interpretation. Since both what and how in (82) are licensed, the sentence should be grammatical, but it is not. Note that how need not move to the operator position either for interpretation or for clausal typing in (82) since it is already interpreted by the relevant operator at the lexical level, and the clause is already typed by what, and it is also of no use to assume that the operator-​bound wh-​words in English have a strong [WH+S] feature and thus must move since wh-​in-​situ is tested in this language, and English, different from Bulgarian, does not need front all wh-​words in multiple questions. I think that the difference among Bulgarian, English, and Chinese wh-​ words lies in the content of D. Following Cheng, I assume that the wh-​words in Bulgarian contain a strong and thus fixed/​specified [+WHS] feature in D, whereas those in English contain a weak wh-​feature [+WHW], instead of a strong [+WHS] feature in D, and those in Chinese contains a weak and undetermined/​unspecified wh-​feature characterized as [±WH] ([+, -​WH]) in D. Adopting both Cheng’s and Tsai’s idea of wh-​word structure, I propose the following structure (83) for Bulgarian wh-​words, structure (84a) for English wh-​words, and structure (84b) for Chinese wh-​words: (83) 

DP

D’

(84)



NP

s [+WH ]

wh-word

a.

DP

D’



wh-word [+WHw]

NP

38

38 Wh-questions b.

DP

D’



NP

wh-word [±WH]

It can be seen from above that the crucial difference between the wh-​ words in Bulgarian-​type languages as shown in (83) and those in English-​ type languages as shown in (84a) is that the former have a null D containing fixed/​specified strong [+WHS] feature, whereas the latter have a D filled by a wh-​word with a weak wh-​feature. As for Chinese wh-​words, the wh-​feature in D, specified as [±WH], is not only weak but also undetermined or unspecified. The reason why all wh-​elements in Bulgarian-​type languages must move is simply because the wh-​words in these languages are specified as [+WHS] in wh-​questions and thus must move to the operator position to check off their strong wh-​feature. Since the Bulgarian-​type is not the focus of the present study, I will not discuss it any further here. Languages like Chinese and English are fundamentally different from the Bulgarian-​type since wh-​words in these languages can be interpreted in-​situ. In English, wh-​words that do not function as the wh-​clausal typer can be interpreted in-​situ by a Q operator since they do not have strong wh-​features and thus do not need to move into [Spec, CP] to check off their features. In Chinese, no wh-​word needs to be raised overtly into [Spec, CP] since a clause can be typed as a wh-​clause by a Q particle, and wh-​words can be interpreted in-​situ by operators. When a wh-​word in Chinese is interpreted in-​situ by an operator, the relevant operator can be either a Q operator or a non-​Q operator. It is in this sense that we say that the features of the wh-​words in Chinese are represented as [+, -​WH] (or [±WH]) and thus undetermined/​unspecified. Notice that, although the wh-​ word in English also does not have a strong wh-​feature, wh-​fronting occurs in English wh-​questions. The reason why there must be one wh-​word being fronted in English wh-​questions is not because the relevant wh-​word has a strong wh-​feature but because the clause needs to be overtly typed. If English does not use any strategy to label a sentence as a wh-​question, then the relevant sentence may be ambiguous between a wh-​question reading and an echo question reading. In many cases, Chinese fails to disambiguate these two

 39

The syntax and semantics of wh-questions 39 readings out of context. For instance, the following sentence is ambiguous between a wh-​question reading and an echo reading: (85)  Zhangsan mai-​le 

  shenme?

Zhangsan buy-​ASP  what a. ‘Zhangsan bought what?’ b. ‘What did Zhangsan buy?’ The above fact is consistent with the characteristics of Chinese, which also shows ambiguity in many other structures. While Chinese often resorts to context for disambiguation of sentence meanings, English tends to disambiguate them by structural means whenever possible. Since wh-​fronting can disambiguate an echo question from an original question in English, it will be consistently used whenever an original question is to be formed. Of course, there is another possible way to explain why wh-​fronting is obligatory in English wh-​questions. We can assume that in English wh-​questions C always bears a strong [+WHS] feature, and this strong wh-​feature has a morphological requirement, which can be satisfied only by a wh-​element which enters into local agreement with C. One thing that should be pointed out is that the Chinese sentence (85) can also be structurally disambiguated. If a Q particle is added to this sentence, we can only get a wh-​question reading, as shown below: (86) Zhangsan mai-​le 

shenme ne?

Zhangsan buy-​ASP what  

Q

‘What did Zhangsan buy?’ Although the Q particle strategy is not obligatorily used, it is always available in Chinese. But this Q particle strategy is not available in English. In order to type a wh-​question, it has to use another strategy, which is wh-​ fronting. In the process of clausal typing, the fronted wh-​word in English will be assigned an interpretation by C° marked with strong [+WHS] under agreement. Since one wh-​word will suffice to do the job of labelling the sentence or satisfying the morphological requirement of C, all other wh-​words need not move in English multiple wh-​questions. For those wh-​words that stay in-​situ, I  assume that they are interpreted via the operator binding. It can be seen that the main difference between English and Chinese with respect to wh-​question formation lies in the Q typing. In English, the Q typing is realized by wh-​raising, whereas in Chinese, it is done by the Q particle. Another issue that should be taken into consideration is whether a Q particle is equal to a Q operator. As a first approximation, I  assume that it is,

40

40 Wh-questions though later on I  will reformulate this account. Note that sentence (51) is ambiguous between a wh-​question reading and an indefinite reading, but when a Q particle is inserted, it can only get a wh-​question reading, as shown below: (87)  Zheli que-​le    shenme  ne? here miss-​ASP  what    Q What is missing here? The following sentence discussed in Hua (2000: 176) constitutes another piece of evidence that supports the operator property of the Q particle. (88) *Lisi mai-​le  

zhe-​ben shu   ne?

Lisi buy-​ASP this-​CL book Q (88) is ungrammatical because the Q particle as an operator binds nothing in the sentence, thus violating the ban on vacuous quantification (May 1977; Kratzer 1995; Partee 1988). The use of the Q particle has two functions: (i) assigning a Q feature to C; and (ii) binding the wh-​elements in its c-​command domain. (Notice that the Q particle ne can also bind an empty wh-​word. For detailed discussion on the use of ne, see Shao 1996.) The first function is clausal typing, and the second function is to disambiguate the identity of wh-​ words and thus assign them an interpretation. Note that the wh-​feature on wh-​words is not assigned by the Q operator, but triggered out by the Q operator. I do not assume that the Q operator ne in Chinese is a wh-​particle as does Cheng (1991, 1997), since it does not have a wh-​feature (see Miyagawa (2001) for a discussion on Japanese Q particles, which, he also assumes, do not have any wh-​feature). If Q operators bear a wh-​feature, the grammaticality of the following sentences are unexplained: (89) a. Lisi lai-​bu-​lai      ne?   Lisi come-​not-​come Q   ‘Does Lisi come or not come?’ b. Lisi

lai   haishi bu   lai 

  Lisi  come or 

ne?

  not come  Q

  ‘Does Lisi come or not come?’ The above sentences are not wh-​questions, though each of them is compatible with the Q particle ne. (89a) is an A-​not-​A question and (89b) a disjunctive question. Note that neither the A-​not-​A question nor the disjunctive question is compatible with a wh-​word, as shown in (90), but two wh-​words are compatible with each other, be it a wh-​argument or a wh-​adjunct, as shown in (91):

 41

The syntax and semantics of wh-questions 41 (90)  a. *shei  lai-​bu-​lai?   who come-​not-​come   Lit. ‘Does who will come or not?’ b. *shei  lai 

  haishi  bu  

lai?

  who   come  or    not  come   Lit. ‘Does who will come or not?’ (91) a. Shei mai-​le  shenme?   who buy-​ASP what   ‘Who bought what?’ b. Shei weishenme mei lai?   who why      not come   Lit. ‘Who did not come why?’ But, of course, there is a division of labour between the Q particles though they all bear a Q feature. For instance, the Q particle ma in Chinese, a yes-​no question marker, is always in complementary distribution with ne. Thus, it cannot replace ne in (92) below: (92) a. *Lisi  lai-​bu-​lai      ma?   Lisi come-​not-​come

Q

  Lit. ‘Will Lisi come or not?’ b. *Lisi lai  haishi bu   lai    Lisi come  or 

  ma?

not  come  Q

  ‘Does Lisi come or not come?’ This also indicates that the A-​not-​A question and the disjunctive question are not yes-​no questions since these two kinds of questions are incompatible with the yes-​no question marker ma. Given that the questions typed by the Q particle ne are not of a homogeneous class, which may include wh-​questions, disjunctive questions, and A-​not-​A questions, we conclude that the Q particle ne does not have a wh-​feature. To make a distinction between these two Q particles, we may specify ma as a Q particle containing the feature [+Q, +yes-​ no], and ne as one containing the feature [+Q, -​yes-​no]. If we assume that wh-​words in Chinese are associated with undetermined/​ unspecified [±WH] features, two possibilities may arise: it is +wh, or it is –​wh. Note that +wh is associated with the Q feature since the former entails the latter. Hence, if it is bound by a Q operator, it becomes +wh. Now that the wh-​feature on a wh-​word is triggered, the wh-​word, as an activated goal, must identify itself with a probe for interpretation. C is such a probe which is also

42

42 Wh-questions looking for a matching goal. They end up in matching their wh-​features via the operator-​binding. In English both the Q feature and the wh-​feature are on C. When a wh-​word moves to [Spec, CP], the wh-​feature on C is checked off by the wh-​word and the sentence is typed as a wh-​question. As for the other wh-​words in English multiple questions, they do not need to move since they are not inherently marked with a strong wh-​feature, and typing is no longer necessary. But they need to get an interpretation. Hence, the operator binding via the choice function will apply. We see that two processes are involved in the formation of wh-​questions: (i) Q typing; and (ii) wh-​feature matching. Q typing employs two strategies: (i) the Q particle typing (see also Pan 2019 for a discussion on the prosodic licensing of wh’s-in-situ); and (ii) wh-​raising typing. Wh-​feature matching can be done either by moving a wh-​word into the Spec of CP or by the operator–​variable binding. If a wh-​word is inherently specified as [+WHS] in the lexicon, it has a strong wh-​feature. If it is specified as [+WHW] or [±WH] in the lexicon, its wh-​feature is weak. Q typing is a process of interpreting (labelling) the clause type, while wh-​feature matching is a process of interpreting both C and wh-​ words. Now, we can construct two parameters from these processes to give a more precise prediction of the typological variation among wh-​questions in different languages. One is the typing parameter, as shown below: (93)  The Typing Parameter a. wh-​raising. b. Q particle. The other is the [WH] strength parameter, as shown below: (94) The Wh Parameter a. [+WHS]. b. [+WHW]/​[±WH]. The interaction of these two parameters predicts that there are four types of wh languages: (i) the wh-​raising and [+WHS] type; (ii) the Q particle and [±WH]/​[+WHW] type; (iii) the wh-​raising and [+WHW]/​[±WH] type; and (iv) the Q particle and [+WHS] type. If we follow Chomsky (1993) in assuming that operators have a universally strong wh-​feature, we can say that (94a) specifies the feature for operators, and (94b) for variables. Then, we can reformulate (94) as (95): (95) The Wh Parameter [± Op(erator)] (93) amounts to saying that if a language does not use the Q particle strategy to type wh-​questions, it has to use the wh-​raising strategy to do so. Thus, we can reformulate (93) as (96):

 43

The syntax and semantics of wh-questions 43 (96)  The Typing Parameter [± Q-​p(article)] Bulgarian belongs to the [+Op, -​Q-​p] type in which clauses needs to be typed by wh-​raising, and all wh-​words need to be identified with C because they are operators and thus bear a strong wh-​feature. Chinese belongs to the [-​Op, +Q-​p] type in which clauses are typed by a Q particle, and wh-​words stay in-​situ with their wh-​features being fixed (triggered) by an operator. English belongs to the [-​Op, -​Q-​p] type in which clauses need to be typed by wh-​raising, and wh-​words that do not participate in clausal typing stay in-​situ and are interpreted by a wh operator. I leave open the question whether the [+Op, +Q-​p] type exists in natural language, though theoretically its existence is possible.7 The establishment of these two parameters, at least, shows that simply making a distinction between operators vs. variables in classifying wh-​ expressions may not adequately characterize the properties of wh-​expressions in natural language. What matters is the interaction between the strength of the wh-​feature on wh-​words and the typing requirement of the language. 2.2.3  The C Typing Condition and wh-​interpretation The present analysis suggests that a wh-​element be matched with C via wh-​ agreement for wh-​ interpretation. If a wh-​ word is inherently specified as [+WHS] in the lexicon, it is an interrogative NP with a strong wh-​feature and must be interpreted in the left periphery of CP. If it is specified as [+WHW] in the lexicon, it is an interrogative NP with a weak wh-​feature and may be interpreted either in [Spec, CP] or via operator binding. If it is specified as [±WH] in the lexicon, it may be interpreted as either an interrogative NP (for a wh-​question reading) or an indefinite NP (for an indefinite reading) in syntax, depending on the nature of the operator that binds it. In Bulgarian, the wh-​ feature on a wh-​word is inherently specified as [+WHS] in the lexicon. That is why its wh-​feature is strong. In both English and Chinese, the wh-​feature of a wh-​word is not strong. We assume that the wh-​feature on a wh-​word in Chinese should be interpreted with C through either wh-​agreement or operator binding in syntax. We can use the “probe-​goal” system of Chomsky (2000, 2001) to account for the agreement process between C and wh-​words. We say that it is the operation Agree that establishes an agreement relation between the probe (i.e., the C containing the [+WH] feature) and the goal (i.e., the wh-​expression). I assume that Agree can be done either between a head and its Spec by Spec-​head agreement or between C and a wh-​word in a way similar to binding. If their features match, the wh-​clause gets an interpretation. If their features do not match, the wh-​clause does not have an interpretation, and the sentence will be ruled out as uninterpretable. In the Agree operation, the probe either attracts the closest goal to its Spec or enters into a binding relation with the closest goal.

44

44 Wh-questions Although Chinese uses a Q particle to type a wh-​question, I assume that the wh-​feature on C can be assigned only when C enters into agreement with a wh-​word since, as shown in the previous discussion, a Q particle does not bear a wh-​feature. I assume that in English it is the closest wh-​element that is attracted to the operator position to agree with C, while in Chinese it is the wh-​element closest to C that establishes a matching relation with C. I will soon show that locality alone cannot fully account for this probe-​goal operation between C and the wh-​word. Another factor we should take into consideration is prominence. It seems that C prefers to attract the closest and the most prominent wh-​element. If this is really the case, a condition on the probe-​goal operation can be formulated below: (97)  A probe C should agree with the most prominent goal, a wh-​word, that is closest to it via either Spec-​head agreement or operator–​ variable binding. Prominence can be understood in terms of the following Prominence Hierarchy: (98) The Prominence Hierarchy a. Subject > Nonsubject b. Argument > Nonargument c. Lexical Element > Function Element Now, consider the following sentences: (99) *[C°wh [weishenmewh shei  mei  lai]]?    

  why     

who not

come

Lit. ‘Who did not come why?’ (100)  *[C°wh [Zhangsan weishenmewh xihuan shei]]?       Zhangsan  why   

  like   who

Lit. ‘Why does Zhangsan like who?’ (101) *[CP Whatwh C°wh [did who buy t]]? The ungrammaticality of (99–​101) results from a violation of (97). In fact, (97) can be regarded as a clausal typing requirement. It is reasonable to think that if C needs to choose an element from the sentence as its clausal-​type label, it tends to choose the closest one since it conforms to the general Principle of Economy, and it also tends to choose the most prominent one since it is more representative. Hence, we say that (101) is unacceptable because English needs the closest and the most prominent

 45

The syntax and semantics of wh-questions 45 wh-​word to type the clause. But what about the above Chinese sentences? Note that the clauses are already typed by Q particles under our analysis. I  think that the ungrammaticality of the above Chinese sentences also results from the violation of the clausal typing requirement. Note that since the Chinese Q particle does not contain a wh-​feature, it can only specify C as a Q feature containing head. If a clause is to be interpreted as a wh-​ clause, its C should be interpreted with a wh-​word. I  assume that the Q particle is head-​adjoined to C, and the binding between Q and a wh-​word will get the wh-​feature of the wh-​word transmitted to C. As a result, C is typed as a wh-​feature bearing C. In English, a clause is typed as a direct wh-​ question by (i) raising a wh-​word into the Spec of CP, and (ii) moving AUX (auxiliary) or inserting do into C.  Whereas in Chinese, what are involved in direct wh-​question typing are these steps: first, a Q particle types C as Q by merge and, then, a wh-​word types C as a wh-​feature containing head via operator-​binding and feature transmission. Since, no matter whether C has a strong wh-​feature or not, it always needs to be associated with a wh-​ word for wh-​interpretation in natural language, I tentatively propose (102) instead of (97) as a C Typing Condition: (102)  The C Typing Condition (CTC) a. For wh-​raising languages, a clause is typed as a wh-​clause if the most prominent wh-​word moves overtly into [Spec, CP] via cyclic movement. b. For Q-​particle languages, a clause is typed as a wh-​clause if the closest Q operator inserted in [Spec, CP] binds the most prominent wh-​word that is the closest to it. Prominence is defined in (98), and closeness is defined below, on the basis of Pan (1998: 793): (103) The Closeness Condition α is closer to X than β is iff the path from α to the minimal maximal projection dominating X is a proper subset of the path from β to the minimal maximal projection dominating X. 2.2.4  A-​not-​A and multiple wh-​questions It is claimed that in Chinese the LF movement of wh-​arguments does not exhibit wh-​island effects (Huang 1982a, 1998[1982b], 1991; Tsai 1999b). Thus, neither an A-​not-​A element nor a wh-​phrase can block a wh-​argument for scope interpretation at LF. It is also claimed in the literature (cf. Nishigauchi 1990; Watanabe 1992) that Japanese exhibits wh-​island effects at LF, though it lacks Complex NP effects. This difference between Chinese and Japanese

46

46 Wh-questions seems very interesting. Since Chinese and Japanese share similar properties in many other aspects in terms of wh-​interpretation, one may ask why they should behave differently with respect to wh-​island constraints. Although this difference between Chinese and Japanese will leave enough room for linguists to exercise their syntactic imagination, one thing is often ignored. That is the truth of the facts. Due to the lack of knowledge of a foreign language, the picture of wh-​island effects presented in the literature is often taken for granted. However, upon closer examination, one may find that the assumed facts are not always true. Liu (1986) and Xu (1990) claim that linguists arrive at wrong conclusions because they take echo question readings for original question readings. Xu (1990) is the first to try to show what factors cause linguists to take the echo question for the original question. In the following, I  will show that (i) wh-​island effects are also observed in Chinese and the claimed nonexistence is due to the fact that echo question readings and original question readings are not correctly distinguished (cf. Liu 1986; Xu 1990), and (ii) many WH facts in Chinese can be accounted for by the interaction of prominence and locality conditions subsumed under the C Typing Condition (henceforth CTC). 2.2.4.1  Feature clash in C Typing Now, consider why the following Chinese sentences are ungrammatical, though the wh-​word closest to C is also the most prominent one: (104)  *shei xi-​bu-​xihuan ta? Who like-​not-​like him Lit. ‘Does who like him?’ (105) *Lisi xi-​bu-​xihuan shenme? Lisi like-​not-​like what Lit. ‘Does Lisi like what?’ According to Xu (1990), (104) and (105) are ungrammatical because they are semantically uninterpretable. Following the propositional approach to the semantics of questions (Hamblin 1973; Karttunen 1977), Xu (1990) claims that the semantic representation of a question should make reference to the set of possible answers, represented by a disjunction of statements, and the failure of (104) and (105) results from the fact that the set of possible answers to it does not exist. According to Xu, in asking (104), the speaker would be saying (106): (106) Tell me which of the following is true: who like him or who does not like him.

 47

The syntax and semantics of wh-questions 47 But neither of the disjuncts in (106) has a truth value. This is why (104) is unacceptable since it is uninterpretable.8 Under my analysis, C in (104) would be typed by two interrogative elements specified for two conflicting features. When bound by a Q operator, shei has the feature [+WH], but the A-​not-​A morpheme is specified as [-​WH]. Under this situation, the feature of C would be left undetermined. Hence, the ungrammaticality of (104). But it seems at first sight that our feature clash analysis fails to explain the grammaticality of (107): (107)  Ni  xiang-​zhidao shei xi-​bu-​xihuan ta? you wonder      who like-​not-​like him (108) Wo xiang-​zhidao Lisi xi-​bu-​xihuan ta. I   wonder      Lisi like-​not-​like him ‘I wonder whether Lisi likes him or not.” (109) Wo xiang-​zhidao shei

xihuan ta.

I   wonder      who  like 

him

‘I wonder who likes him.’ (Huang 1982a: 390) Huang (1982a) claims that different verbs subcategorize for different clauses as their complements. Verbs like wen “ask” are characterized as [+WH] verbs, which require indirect questions as their complements. Verbs like xiangxin “believe” are characterized as [-​ WH] verbs, which cannot take indirect questions as their complements. Verbs like zhidao “know” are characterized as [+, -​WH], which can take either declarative sentences or indirect questions as their complements. Notice that, under Huang’s analysis, [-​WH] means [-​WH, -​Q], but under the present analysis, when the A-​not-​A element is interpreted as [-​WH], it only means that it lacks the relevant wh-​ feature but does not mean that it lacks interrogative force. Hence, a complete feature characterization of the A-​not-​A element should be something like [-​WH, +Q]. Since, according to Huang, in (107) xiang-​zhidao “wonder” is a verb subcategorizing for an interrogative complement, either shei or the A-​not-​A element in (107) must take the embedded scope to satisfy the subcategorization requirement of the matrix verb.9 Although the A-​not-​A element is a phonetic realization of the INFL with Q under Huang’s (1991) analysis, he finds that (107) can only be answered by (108), but cannot by (109). The sentence given in (108), as the only possible answer to (107), shows that only shei can take matrix scope. This means that the A-​not-​A element must take the embedded scope. Huang derives this result from his ECP account. He treats the A-​not-​A element as an adjunct operator and thus predicts that its LF movement to the matrix operator position will violate ECP since the relevant

48

48 Wh-questions movement will cross the wh-​island formed by shei, letting the trace left by the A-​not-​A element ungoverned. Since shei, as an argument operator, can always be properly governed, it can cross the island formed by the A-​not-​A element to take matrix scope without violating ECP. Huang (1991) has shown convincingly that the A-​not-​A question has an interrogative INFL constituent, and the phonetic realization of the INFL with [+Q] may take different forms in different Chinese dialects. Shi (1994) follows Huang’s basic assumption but rejects his idea that the A-​not-​A element is an operator. He claims that the A-​not-​A element is a morphological realization of the Q morpheme, which has the function of a verb and not that of an operator and thus cannot license wh-​words. According to Shi, (104) is ungrammatical because the wh-​word is not licensed by a Q operator since the relevant operator has been incorporated into the verb. (107) is grammatical because there is a matrix operator to license the wh-​word though the embedded operator has been incorporated. Under Shi’s analysis, (107) cannot have a reading in which the A-​not-​A element takes matrix scope because it is impossible for the embedded clause to generate two operators. I agree with Shi (1994) that the A-​not-​A element is formed by the incorporation of the Q operator and that the A-​not-​A element cannot bind wh-​words. Since we assume that the Q feature on C is assigned by the Q operator, we can further assume that the Q feature in INFL is also assigned by the Q operator. Assume that the Q operator can be assigned to either the C or the INFL of a sentence. In both ways, the sentence is typed. If the Q operator is assigned to C, it needs to bind a wh-​word. If it is assigned to INFL, it is incorporated into INFL to trigger the formation of the A-​not-​A question, as suggested by Huang (1991) and Shi (1994). If we assume that a question can only be typed by one Q operator in the Q particle languages, the reason why sentences like (104) are ungrammatical is obvious since the wh-​word shei “who” is not bound by a Q operator, given that the Q operator has already been incorporated into INFL. An interesting result of this analysis is that it predicts that (107) should also be ungrammatical since the wh-​word shei “who” in this sentence should also be unbound and thus uninterpreted. We will argue that this prediction is correct. But, first, consider the question why the A-​not-​A cannot take wide scope in (107). Law (2001) finds that the A-​not-​A element cannot take matrix scope in the following sentence and thus argues that it does not move at LF: (110)  *Wangwu zhidao [ni juede [Lisi hui-​bu-​hui shengqi]]? Wangwu know  you think Lisi will-​not-​will angry Lit. ‘Wangwu knows that you think whether Lisi will be angry?’ If the A-​not-​A element can move at LF, nothing blocks it in (110), and thus it should have no problem in taking matrix scope by making cyclic head-​ movement via C.  The impossibility for it to do so in (110) indicates that it cannot move at LF. McCawley (1994) argues that the following example

 49

The syntax and semantics of wh-questions 49 chosen by Huang (1991) to show that the A-​not-​A element can take wide scope is not convincing since the supposed higher clause ni juede “you think” can be treated as a parenthetical expression. (111)  Ni  juede [ta   hui-​bu-​hui  shengqi]? you feel  

he  will-​not-​will get-​angry

‘Do you think he will be angry?’ McCawley claims that in the following sentence, the A-​not-​A element cannot take matrix scope because the higher clause does not admit any interpretation as parentheticals: (112) *Lisi yiwei [Edison you-​mei-​you  faming dianhua]? Lisi think   Edison have-​not-​have invent telephone ‘Does Lisi think Edison invented the telephone?’ If Law and McCawley are right, then it is not the case that the specific movement of the A-​not-​A element will be blocked by the wh-​island constraint but that the movement of an A-​not-​A element out of its own clause is generally banned. If the A-​not-​A element cannot move out of its own clause at LF in (107), what about the wh-​word shei “who”? Can shei move to the matrix operator position at LF in (107)? If it moves, it means that it seeks to be interpreted as a Q operator, assuming that any wh-​word can be interpreted as a Q operator if it moves to the operator position. Although this option is possible, it seems that the wh-​word shei “who” in (107) may not move since its movement must be carried out cyclically. Given that the embedded C is already typed as [-​WH] under agreement with the feature contained in the A-​not-​A element, LF movement of shei with the feature [+WH] will be blocked when it moves cyclically through the embedded C because of feature clash. Notice that, if it does not move, it is a variable and should thus be bound by an operator, but it is not bound by an operator in (107), given that the Q operator has been incorporated into INFL. I  further assume that positing another operator in the embedded clause to bind shei should generally be banned by the grammatical system. Given that one operator can unselectively bind all the variables in its c-​command domain, the grammatical system is not motivated to generate more operators of the same nature than needed. Notice that, even if it is allowed to insert another Q operator into the embedded CP domain in (107), the sentence is still ruled out because of feature clash, given the fact that the two operators that bind two different wh-​elements have different features. Inserting an operator into the matrix Spec of CP will also not work since the closest operator for the wh-​word is the LF raised operator in the embedded CP domain. Only when the matrix operator and the most prominent wh-​word are the closest to each other can the matrix clause be typed as a wh-​question via operator binding.

50

50 Wh-questions If the present analysis is correct, then shei “who” in (107) must be left uninterpreted since it is not bound by an operator. It can be seen that there is a welcome result if we adopt Shi’s (1994) assumption that the Q operator has been incorporated into the A-​not-​A form in the A-​not-​A question, and at the same time insist that a question can be typed by inserting only one Q operator. If this line of analysis is on the right track, we can explain more satisfactorily why the A-​not-​A element cannot occur in an island in A-​not-​A questions such as (67). Assume that LF movement should also be constrained by Subjacency. In (67), the A-​not-​A element cannot take wide scope since it is contained in an island. It also cannot be interpreted via the choice function since there is no Q operator to bind it, given that there is only one Q operator per sentence and the Q operator has already been incorporated into the INFL of the subject clause. As there is no way to help the A-​not-​A element take scope out of the subject clause, it cannot have the matrix scope interpretation. If the wh-​word shei “who” is not bound in (107), then the next question to ask is why (107) is regarded as an acceptable sentence under Huang’s analysis. According to Xu (1990), the seemingly acceptable status of (107) results from the possibility that it can be easily interpreted as an echo question. Note that (104) can also be heard as an echo question, and in this case, it is acceptable. (Note that, if an example discussed in this part of the book is starred as unacceptable, I mean that it is ungrammatical under the intended original question reading.) I agree with Xu that (107) can only be interpreted as an echo question. Consider the following sentence in which zenmeyang “how” fails to be interpreted as an echo wh-​word: (113)  *Ta wen ni  Lisi zenmeyang za-​mei-​za   

kai  na-​dao men?

he ask you Lisi how      break-​not-​break open that-​CL door Lit. ‘He asked you by what means Lisi had broken or not that door?’ Although a possible answer like (114) is available for (113), (113) is not an acceptable wh-​question. (114) Ta wen wo Lisi yong  tie-​chui    he ask I  Lisi with

za-​mei-​za      kai  na-​dao men.

iron-​hammer break-​not-​break open that-​CL door

‘He asked me whether Lisi had broken that door with an iron hammer.’ If Tsai (1994a) is right in claiming that zenmeyang “how” is referential when having the instrumental or means reading and thus can be extracted out of an island at LF, the ungrammaticality of (113) is unexpected. Note that under Tsai’s (1994a) analysis, a PP like yong jinpodingren “by a one-​on-​one approach” in (116) has an instrumental reading. If we replace this PP with zenmeyang “how,” the sentence should be equally acceptable since zenmeyang should also have an instrumental reading and thus can be freely extracted out

 51

The syntax and semantics of wh-questions 51 of an island. However, (115) shows that zenmeyang cannot take matrix scope though it is referential, according to Tsai. (115)  *Ta wen ni  tamen hui-​bu-​hui  zenmeyang da   zhe-​chang lanqiu? He ask you  they will-​not-​will  how      play 

this-​CL 

basketball

Lit. ‘He asked me by what means they would or would not play this basketball match?’ (116) Ta wen he  ask 

wo tamen  hui-​bu-​hui  yong jinpodingren I  they  

will-​not-​will  with one-​person-​on-​one person

da  zhe-​chang  lanqiu. play this-​CL    basketball ‘He asked me whether they would play this basketball match by a one-​on-​one approach.’ Under our analysis, (113) and (115) are unacceptable because it is difficult for zenmeyang “how” to be interpreted as an echo wh-​word in these sentences. (107) is acceptable because it is easy for shei “who” to have an echo question interpretation. Evidence can be given to show that the wh-​word in (107) must be an echo wh-​word. First, (107) cannot be answered by a single word that directly provides the value for the wh-​word. In providing an answer for (107), one has to repeat the whole sentence, as in (108). If (107) is really an original question, one would ask why it does not accept a short answer that directly provides a value for the relevant wh-​word. It should be noted that an echo question, as a simple sentence, can also be answered by directly providing the value for the relevant wh-​word, but the crucial point is that an original question can always be answered by simply providing a value to the relevant wh-​word, besides receiving a full-​fledged answer that repeats the entire sentence, whereas an echo question, though it can receive a short answer in some cases, generally rejects short answers that only provide a value to the wh-​word in many other cases, especially when it is embedded in a complement clause. Hence, if a certain question can receive both a full-​fledged answer and a short one, we certainly would fail to make a distinction between an echo question and an original question with this “answer-​length” criterion. But if a question simply rejects short answers and can only receive a full-​fledged answer that repeats the entire sentence, we have good reasons to believe that it is an echo question. Second, the wh-​word in (107) cannot be focused by a focus marker that will make it range over a set of individuals in the extralinguistic world. An echo question is usually used to express either the speaker’s surprise at information just made available to him or the speaker’s request for repeating what is just said by the hearer. According to Dayal (1996: 124), an echo question seeks to identify the previous utterance rather than establish the facts in the actual world. In Janda’s (1985) view, an echo question

52

52 Wh-questions word quantifies over linguistic expressions rather than individuals in the extralinguistic world. Hence, echo questions are just metalinguistic devices for requesting information about an unperceived or disbelieved string.10 If this is true, then an echo question must be built on previously uttered sentences.11 Since an echo wh-​expression is used to identify the previous utterance rather than establish the facts in the actual world, I assume that the echo reading of a wh-​word is incompatible with the focus reading of a normal wh-​word in the original questions that presupposes a set of extralinguistic individuals. Hence, an echo wh-​word cannot maintain its echo reading if focused by a contrastive focus marker that always requires a set of individuals. There is one such contrastive focus marker in Chinese. It is jiujing-​shi “exactly/​actually (BE),” which roughly means: “what is it exactly/​ actually?” Or, “what is it on earth?” When this focus marker is associated with an individual, it always implies that this individual is a member of a set.12 In the following interrogative sentences, either the wh-​subject or the wh-​object can be focused by jiujing-​shi: (117)  Mingtian de  hui    jiujing-​shi   

sheiF   qu canjia?

tomorrow DE meeting exactly-​BE  who

go attend

‘Who is actually going to the meeting tomorrow?’ (118) Nimen  jiujing-​shi yao

shenmeF?

you    exactly-​BE want what ‘What exactly do you want?’ But the wh-​word in (107) obviously refuses to be focused by jiujing-​shi, as shown below: (119) *Ni xiang-​zhidao jiujing-​shi sheiF xi-​bu-​xihuan ta? you want-​know  exactly-​BE who like-​not-​like him Lit. ‘You wonder who exactly it is that likes him or not?’ (119) is unacceptable because the echo question reading of the wh-​word is cancelled by the focus marker and thus has to be interpreted as a normal wh-​ word in an original question. However, the relevant wh-​word cannot have an original wh-​question reading since it is not allowed to occur within the domain marked by the A-​not-​A element because of the wh-​feature incompatibility. Third, when the Q particle ne is attached to the end of the sentence in (107), it becomes unacceptable, as shown below: (120) *Ni xiang-​zhidao shei xi-​bu-​xihuan ta ne? you wonder 



who like-​not-​like him Q

Lit. ‘You wonder who likes him or not?’

 53

The syntax and semantics of wh-questions 53 The unacceptability of (120) is expected because, as I have argued in the previous discussion, the Q particle ne can be used to disambiguate an echo question reading from an original question reading, given that ne is used to type an original question. But in (120) neither the wh-​word shei “who” nor the A-​not-​A element can have an original question reading. Since an echo question word is incompatible with neither jiujing-​shi nor the Q particle ne, I assume that the most reliable way to make a distinction between an original question word and an echo question word is to use both jiujing-​shi and ne to test the relevant question word, as shown below: (121)  *Ni xiang-​zhidao jiujing-​shi sheiF xi-​bu-​xihuan ta ne? you wonder      exactly-​BE who like-​not-​like him Q Lit. ‘You wonder who exactly it is that likes him or not?’ 2.2.4.2  Scope-​taking in multiple wh-​questions In the previous discussion, I have claimed that the use of the overt Q particle ne in Chinese can disambiguate an echo question from an original question. In fact, we can simply say that ne is an original or direct question marker since it can also disambiguate an indirect question from a direct (original) question. Notice that English also uses a similar strategy to distinguish a direct question from an indirect question. In English, direct questions are formed either by moving AUX to C or by inserting do under C, whereas indirect questions reject AUX-​raising and do-​support. It should be pointed out that in Chinese the covert Q operator is ambiguous between a direct question reading and an indirect one, but the overt Q operator is not ambiguous. Consider the following sentence: (122) Ta wen ni  [shei lai-​le]. he ask you who come-​ASP ‘He asked you who had come.’ Since the embedded clause in (122) can only be interpreted as an indirect question, the use of ne will be excluded, as shown below: (123) *Ta wen  ni  [shei lai-​le  He ask

ne].

you who come-​ASP Q

‘He asked you who had come.’ If ne occurs in the embedded clause, it will assign the embedded C a direct Q feature and will thus violate the subcategorization requirement of the matrix verb, which selects an indirect question as its embedded clause. Now, consider the following sentence in which ne is put in the matrix clause-​final position.

54

54 Wh-questions (124)  *Ta wen ni  [shei lai-​le] 

ne?

he ask you who come-​ASP  Q Lit. ‘Who did he ask you had come?’ When ne is placed at the matrix clause final position, it will assign a direct question reading to the matrix C. In that case, the embedded C will have no chance to get the relevant Q feature so as to be interpreted as an indirect question, since (i)  the Q operator is not adjoined to the embedded C, and (ii) the wh-​word shei in the embedded clause does not enter into agreement with the embedded C. Note that the Q operator ne has to bind the wh-​word shei “who” in the embedded clause in order to avoid violating the ban on vacuous quantification. If shei is bound by ne in the matrix C, it will enter into agreement with the matrix C and be interpreted with the matrix C. As a result, the embedded C has no way to be assigned a Q feature, thus violating the subcategorization requirement of the matrix verb. Note that the English translation of (124) is also ungrammatical because of the same reason. In neither case, the subcategorization requirement of the matrix verb can be satisfied. Notice that, if the matrix verb does not subcategorize for a question clause, then the relevant sentence will be acceptable, as shown below: (125)  Ta renwei shei lai-​le      ne? he think  who come-​ASP Q ‘Who does he think has come?’ Now, consider the following ungrammatical English examples: (126)  a. *Whoi do you wonder [CP ti [IP ti bought what]]? b. *Whati do you wonder [CP ti [who bought ti]]? The reason why (126a) is ungrammatical is obvious. The matrix verb wonder subcategorizes for a wh-​clause, but this requirement is not satisfied since the embedded clause is not headed by a wh-​word. It seems that the trace left by the wh-​word cannot be used to label a wh-​clause. This means that a clause can be identified as a wh-​clause only when it is overtly labelled by a wh-​word in English. One may ask why the embedded C is not assigned a wh-​feature, given the fact that there is a wh-​trace left in the Spec of CP. We can assume that this is because when the wh-​word moves through the Spec of the embedded CP, it does not enter into agreement with the embedded C. If it enters into agreement with it, its wh-​feature will be identified with the embedded C. As a result, it will be anchored there in agreement. In that case, the sentence, as shown in (127), will be grammatical since the subcategorization requirement of the matrix verb is satisfied. (127) You wonder who bought what.

 55

The syntax and semantics of wh-questions 55 But in (126a) what the wh-​word who agrees with obviously is not the embedded C, but the matrix C, and thus the subcategorization requirement of the matrix verb is not satisfied. Under the classic analysis assumed in generative grammar, (126b) constitutes a typical violation of wh-​island constraint, but under my analysis, its ungrammaticality simply results from a violation of CTC given in (102). Given that the matrix C needs to be typed as Q, it is the closest and the most prominent wh-​word who instead of the less close and less prominent what that should be moved to the matrix operator position. Now, consider another possibility: if both the matrix clause and the embedded clause need to be typed as Q in sentences like (126), what about moving the most prominent wh-​word to the matrix C and the less prominent one to the embedded C? (128)  *Whoi do you wonder [CP whatj [IP ti bought tj]]? The above sentence shows that this possibility does not exist. In fact, the ungrammaticality of (128) also can be accounted for by CTC: the embedded C is not typed by the most prominent wh-​word that is closest to it. Notice that the locality condition requires that the Agree operation between the probe and the goal should be done locally. Given that there is a local probe (i.e., the embedded C) for the goal who, it is unreasonable for who not to match with the local probe but with the matrix probe. Now, let’s consider one more possibility: moving who to type the embedded C and then moving what to type the matrix C by passing through the embedded CP domain, thus deriving the representation in (129), which is slightly different from the one given in (126b). (129) ?*Whati do you wonder [CP ti [CP whoj [tj bought ti]]]? However, the above derivation is also impossible since it violates CTC, too. Given that the matrix C needs to be typed as Q, the typing wh-​word that should be chosen is who instead of what since it is the most prominent NP that is the closest to the matrix C. Although who is impossible to move since it already agrees with the embedded C, moving what to the matrix C will still be blocked by CTC. This amounts to saying that even when a qualified element has lost its ability to move to the matrix operator position, a less qualified element is still disallowed to replace it because of CTC. Of course, this is unfair for what, but this is the truth. The above analysis shows that the wh-​island constraint can be reduced to CTC, which is more general and less ad hoc. I will show that when CTC is adopted, a clearer picture of Chinese multiple WH phenomena can be obtained. Since Huang (1998 [1982b]), it is generally assumed in the literature that Chinese lacks wh-​island effects. In the following, I will argue that wh-​island effects are also observed in Chinese and show that the existence of wh-​island effects in Chinese can be predicted by CTC.

56

56 Wh-questions Consider the following sentence: (130)  Ni  xiang-​zhidao [shei mai-​le    shenme]? you want-​know  

who buy-​ASP what

‘You wonder who bought what?’ According to Huang (1998 [1982b]), (130) is ambiguous between two direct question readings, as shown in (131). On one reading, as indicated in (131a), the wh-​subject is interpreted with the matrix C and the wh-​object is interpreted with the embedded C. On the other reading, as indicated in (131b), the wh-​ object is interpreted with the matrix C, whereas the wh-​subject is interpreted with the embedded C. (131) a. Who is the person x such that you wonder what x bought? b. What is the thing x such that you wonder who bought x? The LF representations of these two possible readings are given in (132a) and (132b): (132) a. [sheii [ni xiang-​zhidao [shenmej [ti 

mai-​le tj]]]]?

  who  you want-​know  what      buy-​ASP   Lit. ‘Who do you wonder bought what?’ b. [shenmej [ni  xiang-​zhidao [sheii [ti 

mai-​le tj]]]]?

  what    you want-​know   who    buy-​ASP   Lit. ‘What do you wonder who bought?’ Under Huang’s LF movement analysis, one wh-​word has to move into the embedded [Spec, CP] to satisfy the matrix verb’s subcategorization requirement in (132). After that, the other wh-​word can move to the matrix [Spec, CP]. In Xu’s (1990: 365) words, this amounts to saying that long-​distance wh-​ movement is possible in (132) only when there is a wh-​island to violate. How is this possible? Under Huang’s analysis, this is possible because LF movement is constrained by ECP only, and the wh-​island constraint, which is subsumed under the Subjacency condition, applies only in overt syntax. The difference between the Chinese sentences in (132) and those English sentences in (128–​ 129) is that in Chinese wh-​words move at LF, whereas in English they move in overt syntax, which is constrained by Subjacency. According to Huang’s judgment, the following sentences are unambiguous: (133) Ni   xiang-​zhidao shei weishenme mai-​le    shu? you want-​know

who why     

buy-​ASP book

‘Who is the person x such that you wonder why x bought books?’

 57

The syntax and semantics of wh-questions 57 (134)  Ni   xiang-​zhidao shei zenme mai-​le    shu? you want-​know

who how  buy-​ASP book

‘Who is the person x such that you wonder how x bought books?’ The reason why the wh-​arguments in (133–​134) as well as those in (132) can take matrix scope follows from ECP since the traces they left after LF movement can always be head-​governed. And ECP can also explain why wh-​ adjuncts like weishenme “why” and zenme “how” cannot take matrix scope in (133–​134) since the LF movement will leave their traces ungoverned, thus violating ECP. Although this account is attractive, Liu (1986) and Xu (1990) point out correctly that it is incorrect due to the fact that Huang fails to make a distinction between echo questions and original questions. According to Xu, neither shei “who” nor shenme “what” in (130) can have the original (direct) question reading, though both of them can trigger the echo question reading. If Xu is correct, then there is no difference between the Chinese multiple question in (130) and the English one in (127): in both languages, the multiple wh-​words contained in an embedded Q clause subcategorized for by the matrix verb can only have indirect question readings by taking embedded scope. Although Xu presents a picture that is completely different from the widely assumed one, it is, however, a true picture. In the following, I will show that Xu’s account is not only expected by CTC but also supported by independent evidence which demonstrates that neither of the two wh-​words in (130) can have a direct question reading. The CTC accounts for (130) the same way as accounting for its English counterpart. If we associate the wh-​subject shei “who” with the matrix C, the matrix verb’s subcategorization requirement for the embedded clause is not satisfied. If shei is associated with the embedded C, it is anchored there in agreement and thus cannot agree with the matrix C. The possibility of associating the wh-​object shenme “what” with either the matrix C or the embedded C will be blocked by CTC because of the existence of a more prominent wh-​ phrase (i.e., shei). Hence, CTC can exclude the direct question reading of the two wh-​words in (130) without assuming the existence of an LF-​formed wh-​ island. At this point of discussion, something interesting emerges. It seems that different theories predict different pictures of the same set of linguistic facts. Under Huang’s analysis, the wide scope reading of wh-​words in (130) is expected, but under my analysis, it is excluded. Now, what is crucial is the true picture of these linguistic facts. I have shown that there are three ways to disambiguate echo questions from original ones in discussing the co-​occurrence of the A-​not-​A element and the wh-​words in embedded questions. These three criteria will work equally well here in disambiguating echo questions from original questions in embedded multiple wh-​questions.

58

58 Wh-questions First, (130) cannot be answered by simply providing the value for the wh-​ word as in answering an original question. In answering (130), repeated as (135) below, one has to repeat the previously uttered question (i.e., the whole sentence), besides providing a value for the wh-​word, as shown in (136). (135)  Ni  xiang-​zhidao [shei mai-​le   shenme]? you want-​know    who buy-​ASP what ‘You wonder who bought what?’ (136) a. Wo xiang-​zhidao Zhangsan mai-​le   shenme.   I  want-​know   Zhangsan buy-​ASP what  

‘I wonder what Zhangsan bought.’

b. Wo xiang-​zhidao shei mai-​le  shu.   I  want-​know  who buy-​ASP book   ‘I wonder who bought books.’ In answering an original wh-​question, besides repeating the entire sentence as an answer, one can directly provide a value for the relevant wh-​word, no matter how complex the sentence is, as shown in (137) and (138): (137) Ni renwei Lisi zhidao shei hui lai? you think Lisi know who will come ‘Who do you think Lisi know will come?’ (138) a. Zhangsan. b. Zhangsan hui lai.   Zhangsan will come. In answering (137), one can either say (138a) or (138b). In neither case, one needs to repeat the whole sentence, though one can. But in answering (135), one must repeat the whole sentence. For instance, (136a) as one of the possible answers can be replaced neither by Zhangsan nor by Zhangsan mai-​le shenme “Zhangsan bought what.” Second, according to the native speakers’ judgment, (135) is incompatible with the Q particle ne, though it is compatible with the Q particle ma. (139) a. *Ni xiang-​zhidao [shei mai-​le   shenme] ne?   you want-​know   who buy-​ASP what    Q   ‘You wonder who bought what?’ b. *Ni xiang-​zhidao [shei mai-​le   shenme ne]?   you want-​know   who buy-​ASP what  Q  

‘You wonder who bought what?’

 59

The syntax and semantics of wh-questions 59 c. Ni xiang-​zhidao [shei mai-​le 

  shenme] ma?

  you want-​know  who buy-​ASP what    Q   ‘Do you wonder who bought what?’ In (139a) ne takes matrix scope, but fails to bind a wh-​variable. This is expected by CTC since neither of the wh-​words in (135) can have a matrix scope reading, though they both can have an embedded scope reading as an indirect matching question. If shei is associated with the matrix C, the subcategorization requirement for the embedded C is not satisfied. Note that, as discussed before, the wh-​object is not qualified to type the embedded C, given that there exists a more prominent wh-​word shei. The reason why (135) can be answered by the sentences in (136) is because either of the wh-​ words can have an echo question reading. Since an echo wh-​word is incompatible with an original question marker ne, the unacceptability of (139a) is expected. Note that ne also cannot take embedded scope in (139b) since the matrix verb does not select a direct question as its complement clause. (139c) is acceptable because neither of the wh-​words in the embedded clause can take matrix scope, and thus the whole sentence can be interpreted as a statement before ma is added. When ma is added, it is turned into a yes-​no question. Third, if it is true that the wh-​word in (135) is interpreted as an echo wh-​ word when answered by the sentences in (136), then the relevant echo wh-​ word cannot be focused by a focus marker like jiujing-​shi “exactly BE,” which forces the element in focus to be a member of an alternative set. (140)  a. *Ni xiang-​zhidao [jiujing-​shi sheiF mai-​le    shenme]?   you want-​know  exactly-​BE who buy-​ASP what   Lit. ‘You wonder who exactly it is that bought what?’ b. *Ni xiang-​zhidao [shei jiujing-​shi     you want-​know 

mai-​le    shenmeF]?

who exactly-​BE  buy-​ASP  what

  Lit. ‘You wonder EXACTLY what who bought?’ As pointed out by Xu (1990), xiang-​zhidao “want-​know” is not a Chinese compound word though it is extensively cited in the GB literature. We may replace xiang-​zhidao with a real word in Chinese like wen “ask,” which subcategorizes for an interrogative clause. But the relevant sentences are still unacceptable, as shown below: (141) a. *Ta wen ni [jiujing-​shi sheiF mai-​le   

shenme]?

he ask you exactly-​BE who buy-​ASP what

  Lit. ‘He asked you who exactly it is that bought what?’

60

60 Wh-questions b. *Ta wen ni [shei jiujing-​shi   

mai-​le    shenmeF]?

he ask you who exactly-​BE buy-​ASP what

  Lit. ‘He asked you EXACTLY what who bought?’ According to the native speakers’ judgment, (141) cannot be answered by the sentences in (142) when the relevant wh-​word is focused by jiujing-​shi: (142)  a. Ta wen wo Lisi mai-​le    he ask I  Lisi buy-​ASP

  shenme. what

  ‘He asked me what Lisi bought.’ b. Ta wen wo shei mai-​le    shu.   he ask I  who buy-​ASP books   ‘He asked me who bought books.’ The native speakers’ judgment on (140) is the same as that on (141): (140) cannot be answered by the sentences in (136) when the relevant wh-​word is focused by jiujing-​shi. This result is expected since (140) and (141) cannot be interpreted as original questions, though they have echo question readings as well as embedded indirect question readings. When the relevant wh-​words in (140) and (141) are focused by jiujing-​shi, the possibility to interpret them as echo wh-​words is excluded, and they have to be interpreted as wh-​words in original questions. That is why they cannot be answered by sentences which are used to answer echo questions. It can be seen from the above discussion that our analysis predicts that there is no difference between Chinese and English with respect to the wh-​ island constraints since both languages must obey CTC in making a wh-​ question. Xu (personal communication) points out that it is unreasonable to assume that an ungrammatical sentence such as (104) will become grammatical when it is used as an embedded clause in (107). The present analysis predicts that both (104) and (107) are uninterpretable as original questions because they violate CTC. It seems to be true that a wh-​word cannot move across another wh-​word at LF in original questions in Chinese. What are extensively discussed in the literature are only sentences like (107) and (135). Sentences like the following are not considered: (143) *Zhangsan wen ni   [shei yiwei [Lisi mai-​le    shenme]] ne? Zhangsan ask you    who think Lisi buy-​ASP what    Q Lit. ‘What did Zhangsan ask you who thought Lisi bought t?’ (144) *Ni xiang-​zhidao [ta wen-​mei-​wen Lisi [shei xi-​bu-​xihuan zhe-​ben shu]] you want-​know  he ask-​not-​ask Lisi who like-​not-​like this-​CL book

 61

The syntax and semantics of wh-questions 61 ne? Q Lit. ‘You wonder whether he asked Lisi who likes or not likes this book?’ (145)  *Ni xiang-​zhidao [shei wen ta [Lisi xi-​bu-​xihuan shenme]] ne? you want-​know  who ask he Lisi  like-​not-​like what    Q Lit. ‘You wonder who asked him whether Lisi likes what?’ (146) *Ni xiang-​zhidao [ta weishenme wen Lisi [shei xi-​bu-​xihuan zhe-​ben shu]] you want-​know  he why      ask Lisi who like-​not-​like this-​CL book ne? Q Lit. ‘You wonder why he asked Lisi who likes or not likes this book?’ If wh-​movement can freely cross wh-​islands at LF, the ungrammaticality of the above sentences is unexpected. Note that in the above sentences, none of the boldfaced wh-​words can take matrix scope, though Huang (1998 [1982b]) predicts that they can. The wh-​object shenme “what” in (143), as a head-​governed argument, cannot take matrix scope by being associated with the Q particle (i.e., the matrix Q operator). Note that under Huang’s analysis, nothing can ban the LF movement of shenme to the matrix operator position in (143) since (i) wh-​islands can be crossed by wh-​arguments at LF in Chinese, (ii) the matrix operator position is not filled in (143) and thus moving shenme into it would not violate the Doubly Filled COMP Filter, and (iii) the subcategorization requirement of the matrix verb wen “ask” is satisfied by shei in (143), and shenme has to move because the clause in which it occurs is subcategorized as -​Q by the verb of the intermediate clause. In (144), (145), and (146), none of the boldfaced wh-​words can take matrix scope though nothing in the ECP account can block them. In each of these sentences, the matrix operator position is empty and thus can be filled by one of the wh-​words at LF without violating the Doubly Filled COMP Filter. Since all the boldfaced wh-​words are head-​governed arguments, their LF movement is expected to be free and thus can cross other wh-​words, including the A-​not-​A element, under the ECP analysis. However, the facts are that none of the boldfaced wh-​words can take matrix scope. Under the present analysis, shenme in (143) cannot take matrix scope because of CTC, and shei also cannot take matrix scope because it has to be interpreted with the C of the intermediate clause to satisfy the matrix verb’s subcategoriztion requirement. Hence, the only possibility to make (143) interpretable is to delete the Q particle ne and assign the two wh-​words an indirect matching question reading. Under our previous analysis, the ungrammaticality

62

62 Wh-questions of (144), (145), and (146) is accounted for by a feature clash in the most deeply embedded C. Now, we can also account for it in another way. In (144) the A-​ not-​A element in the intermediate clause cannot take matrix scope because it has to be interpreted with the intermediate C to satisfy the subcategorization requirement, and shei cannot take matrix scope because it is blocked by the A-​not-​A element in its own clause as well as the intermediate clause. Since the A-​not-​A element is formed by the incorporation of the Q operator, there is no Q operator that can bind shei in (144). Even if we posit a Q operator in the matrix CP, the matrix CP still cannot be typed as a wh-​question since the operator and the wh-​word are not the closest to each other and thus fail to satisfy CTC. Notice that the A-​not-​A element cannot bind shei even if it is LF raised since the former is incompatible with the latter in wh-​feature. The same analysis can be applied to (145) and (146). These two sentences are unacceptable simply because the boldfaced wh-​word fails to be bound by a Q operator. 2.2.5  Some problems solved The previous discussion shows that the English sentence (147) below, repeated from (15), is not grammatical, but its Chinese counterpart (148), repeated from (18), is. The reason why they contrast in grammaticality is obvious: the English wh-​phrase who has to move to the operator position to type the sentence. If it does not move, the sentence is not typed and thus will have no interpretation. If it is moved, island conditions would be violated. (148) is grammatical because the wh-​word does not need to move out of the island to type the sentence since the sentence is typed by a Q particle which binds the wh-​word in Chinese. Since the wh-​word in Chinese can be interpreted in-​situ via the choice function, it has an interpretation. Hence, nothing rules out (148). (147)  *[CP Whoi [IP do you like [NP the books [CP that [IP describe ti]]]]]? (148) [Ni xihuan [NP[IP piping  shei] you like 



de  shu]]?

  criticize who  DE book

‘Which person x such that you like books that criticize x.’ It should be pointed out that in English those wh-​words that do not participate in the typing of clauses also need to be interpreted via the choice function. (149), repeated from (17), and (150), repeated from (25), are grammatical because the wh-​in-​situ can be interpreted via the choice function. (149) Who likes books that criticize whom? (150) Who reads the books that who writes? In (149) the wh-​word occurs as an object, and in (150) it occurs as a subject. Since subjects are not assumed to be lexically governed, the grammaticality

 63

The syntax and semantics of wh-questions 63 of (150) cannot be accounted for by LF movement and ECP. But it can be accounted for under our analysis. The ungrammaticality of the following sentence, repeated from (13a), is also expected under our analysis since the wh-​in-​ situ how cannot be interpreted via the choice function application. (151)  *What did you buy how? Now, consider why the following two sentences contrast in acceptability: (152) *What did who buy? (153) Which book did which person buy? The unacceptability of (152) is expected by CTC since the wh-​word that agrees with C is neither the most prominent nor the closest one. But, how to account for the acceptability of (153)? Since in (153) both wh-​phrases are inherently D-​linked, we can assume that they are both prominent in discourse, and in fact, for those D-​linked wh-​phrases, it is the discourse that will determine which one is more prominent. Assume that in (153) the relevant discourse determines that which book is more prominent, and then it can cross another which-​NP that is nearer to C.  What happens here is that prominence override locality. But, it seems that there are other cases that prominence cannot override locality. For instance, in (147) even if who is replaced by which-​NP, the island effects still hold. This indicates that prominence can override locality only in some cases in overt movement in English, but not in all cases. Specifically, Superiority Condition and Minimal Link Condition (MLC) can be violated, but island conditions seem to be inviolable in English. Previously discussed examples like (36) and (37), repeated here as (154) and (155), seem to pose a problem for our account of D-​linked wh-​phrases. (154) ?What did which student read t? (155) ?Which book did how many people buy t? If a D-​linked wh-​phrase is more prominent, why can it be crossed by a less prominent wh-​element in (154), given that it is also a subject? If a D-​ linked wh-​phrase is more prominent and thus can violate the locality constraint, why is (155) not a fully acceptable sentence? It seems that a plausible characterization of (154) and (155) is that a D-​linked wh-​phrase does not need to move, but can move, across another wh-​phrase. How is this possible? I assume that since a D-​linked wh-​phrase can be interpreted in discourse, it does not need to move to [Spec, CP] for interpretation, and since a D-​linked wh-​phrase is prominent in discourse, it can move across another wh-​word. In case of (153) where two wh-​phrases are both D-​linked, either of them can move because either can be treated as prominent. Either can also stay in-​situ

64

64 Wh-questions because either can be interpreted in discourse as a discourse-​related element. Although this account seems plausible, it fails to explain why (154) and (155) are not fully acceptable. In section 3.2, I will try to provide a better account for (154) and (155). Now, let us see if the following sentences, repeated from (28), can be accounted for in the same way: (156)  a. *What did who give t to Mary? b. What did who give to whom? It seems that the contrast in acceptability between the above two sentences cannot be accounted for the same way since none of the wh-​elements in these sentences are D-​linked. In next section I will discuss why the two sentences in (156) contrast with each other in acceptability.

2.3  The Principle of Economy in wh-​interpretation 2.3.1  The division of labour between clausal typing and wh-​interpretation First, consider the question: Why is CTC needed in forming wh-​questions? This amounts to asking what motivates CTC. It seems that CTC is motivated for both syntactic and semantic reasons. It is motivated for syntactic reasons because a wh-​ question needs to be typed syntactically. It is motivated for semantic reasons because the prominence of wh-​phrases needs to be considered in forming a wh-​question. Then, the next question to ask is why we should consider the prominence of wh-​words in forming a wh-​question. Before answering this question, let us consider the following sentences: (157) a. *Ni weishenme tou-​le      shenme?   you why      steal-​ASP

what

  Lit. ‘Why did you steal what?’ b. Ni   wei(-​le)shenme tou-​le    shenme?   you  why     

steal-​ASP what

  Lit. ‘For what reason did you steal what?’ The unacceptability of (157a) is predicted by CTC, but the acceptability of (157b) is not. Since CTC is related to multiple wh-​questions, we need to consider the relation among wh-​elements in multiple wh-​questions. First, consider why CTC should not be violated in most cases. I think that the answer can be found in the contrast between (158) and (159): (158) a. Who bought what? b. *What did who buy?

 65

The syntax and semantics of wh-questions 65 (159)  a. Which person bought which book? b. Which book did which person buy? Hornstein (1995: 132) assumes that pair-​list readings require set generators, and a noninherently linked wh-​phrase can be a set generator only when it occupies an A'-​position. This assumption can account for the unacceptability of (158b) since what is less prominent than who and thus fails to function as a set generator, as it is not a D-​linked wh-​word. (159b) is acceptable since the wh-​subject in (159b) is inherently D-​linked and can thus function as a set generator even when it is in-​situ. Following Hornstein (1995: 126), I assume that multiple wh-​questions obligatorily receive pair-​list readings. If this is true, then the interpretation of multiple wh-​phrases must be built on the dependency between the relevant wh-​phrases. It is at this point that the notion of set generator becomes important. I assume that if two wh-​phrases are involved, one wh-​phrase must function as a qualified set generator so as to generate an accessible set that the other wh-​phrase can depend on for interpretation. Take (158a) for an example. First, a set over which who ranges must be generated, and then the wh-​phrase what can be paired with the elements of this set for interpretation. It is reasonable to think that the preferred set generator should be the most prominent wh-​phrase (i.e., the subject), if prominence is understood in terms of grammatical functions, because it is easy for a subject to generate an accessible set since subjects are always associated with discourse saliency. When discourse saliency is taken into consideration, it is natural to assume that a subject is more prominent than an object, and a wh-​argument is more prominent than a wh-​ adjunct. (157a) is uninterpretable because weishenme “why” cannot generate an accessible set for the other wh-​phrase to depend on for interpretation when wei is focused. (157b) is acceptable because wei(le)-​shenme can generate a set of reasons or purposes when shenme is focused. Notice that if shenme in (157b) is not focused, the sentence will be out since in that case, weishenme fails to generate an accessible set. Another way to account for the acceptability of (157b) is basically theory-​internal. If we treat the focused shenme in weishenme as an argument, we can explain why (157b) is acceptable since there is no ranking between shenme in weishenme and the wh-​object in (157b) according to the prominence hierarchy given in (98). In (158b), what, as an object, is less prominent than the subject who and is thus not a qualified set generator. Since what fails to generate a set for who to depend on for interpretation, who is not licensed. (159b), contrasting with (158b), is acceptable because a D-​linked wh-​phrase, be it in [Spec, CP] or in-​ situ, can always function as a set generator. Now, consider why the following English sentence (160a) is grammatical, though the wh-​word why in [Spec, CP] is not a qualified set generator. (160) a. Whyi did Bill buy what ti? b. *I wonder whyi you bought what ti

66

66 Wh-questions In fact, (160a) is unacceptable as an original question. Hornstein (1995) argues that the acceptability of (160a) comes from the ease with which it can be heard as a sort of echo question. Hornstein (1995: 148) shows that if a sentence like (160a) is embedded under I wonder, the relevant sentence, as shown in (160b), becomes unacceptable because it does not tend to be interpreted as an echo question. Notice that, if in (160b) the matrix subject is not the first person pronoun I, the wh-​object what can still be interpreted as an echo wh-​word, and if it is interpreted as an echo wh-​word, the relevant sentence becomes acceptable. Under the present analysis, the unacceptability of (160) as an original question is expected since why, as an adjunct, is less prominent than the wh-​object what and thus cannot function as a set generator. The above discussion shows that prominence is needed only because multiple wh-​questions depend on a prominent set generator for interpretation. It seems that prominence is only relevant to wh-​interpretation, but not to clausal typing. If this is true, our CTC has incorporated two different things: clausal typing and wh-​interpretation. If the wh-​clausal typing condition is treated as a pure typing condition, it may not consider how wh-​elements are interpreted. In that case, the typing condition may not need to consider prominence, though it has to consider locality. To make a distinction between clausal typing and wh-​interpretation, I redefine CTC as below, which is a Pure Clausal Typing Condition. (161)  The Pure Clausal Typing Condition (PCTC) a.  For wh-​raising languages, a clause is typed as a wh-​clause iff there is a wh-​word that moves overtly into [Spec, CP] via cyclic movement without crossing any strong island. b.  For Q-​particle languages, a clause is typed as a wh-​clause if the closest Q operator inserted in [Spec, CP] is interpreted with a wh-​word that is the closest to it. (161a) implies that in wh-​raising languages, a clause would fail to be typed as a wh-​question if the wh-​word placed in [Spec, CP] is extracted from a strong island. Strong islands are defined below (Cinque 1990): (162) Strong islands are Complex NPs and subject/​adjunct clauses. 2.3.2  Economy and wh-​interpretation As for wh-​interpretation, I  think that it is necessary to make a distinction between syntactic and semantic interpretations. (163) Wh-​Interpretation in Syntax A wh-​word can be directly interpreted in syntax iff it matches with C via the Agree operation.

 67

The syntax and semantics of wh-questions 67 I think that the Agree operation should be generalized to include the operator-​variable binding since binding means co-​indexation that results in agreement. Hence, I define the Agree operation as below: (164)  The Generalized Agree Operation A probe can enter into the Agree operation with a goal via either Spec-​head agreement or operator–​variable binding iff the following two conditions are met: a. The probe and the goal are the closest to each other before the application of the Agree operation. b.  The goal is not contained in an island. The above definition requires that Agree obey strict locality constraints. This is reasonable since Agree is a syntactic operation. However, if we adopt the above strict definition of the Agree operation, a natural question that one may ask is why only the operator–​variable binding involved in syntactic interpretation is sensitive to island constraints. If there is no evidence to show that the operator–​variable binding should be constrained somewhere but free elsewhere, it is stipulative to say that it should be constrained only in syntactic interpretation. A possible solution to this problem is to assume that the operator–​variable binding is free from island constraints in general, and only movement, be it overt or covert, should be sensitive to islands. Following this line of reasoning, we can assume that the Agree relation between the probe and goal defined by the operator–​variable binding in (164) is not realized by binding but by the Spec-​head agreement at LF. Let us assume that there is an LF movement on a par with overt syntactic movement and that this LF movement is as highly constrained as is overt movement. Since there is no evidence to show that the LF movement is not constrained by island conditions, I assume that the following condition on wh-​movement is true: (165) The Condition on Wh-​Movement A wh-​element can move to the Spec of CP in either syntax or LF iff it moves for clausal typing or wh-​feature checking, and the movement does not violate the strong island constraints. Now, we can redefine the Agree operation as below: (166) The Agree Operation (Revised) A probe can enter into the Agree operation with a goal via Spec-​head agreement in either syntax or LF iff the following two conditions are met: a. The probe and the goal are the closest to each other before the application of the Agree operation. b.  The goal is not contained in an island.

68

68 Wh-questions In our previous discussion, we treat the choice function application as a means to interpret wh’s-​in-​situ. If we treat the Agree operation as a way to interpret wh-​expressions in syntax, we can treat the choice function application and the pair-​list reading as a condition that regulates wh-​interpretation in semantics. (167)  Wh-​Interpretation in Semantics A wh-​word is interpreted in semantics if it is interpreted via the choice function application or in the pair-​list reading. The following is the condition on the pair-​list reading: (168)  The Condition on the Pair-​List Reading B can be interpreted with A in the pair-​list reading if (i) A can function as a set generator, and (ii) B can be paired with the members of the set generated by A. The following is the condition on the set generators: (169) The Condition on Set Generators A is a set generator for B if (i) A precedes B and is more prominent than B, or (ii) A is an inherently D-​linked wh-​phrase that c-​commands the trace of B. The definition of prominence, given in (98), is reformulated as follows: (170) The Prominence Hierarchy a. Subject > Nonsubject b. Argument > Nonargument c. Lexical Element > Function Element d. D-​Linked > Non-​D-​Linked Given that if a wh-​word is contained in an island, it cannot move to [Spec, CP] even at LF according to the revised Agree operation given in (166), it becomes clear that a clause with the wh-​word contained in an island can be typed as a wh-​clause in Q-​particle languages only by the semantic operation (i.e., the choice function application). However, not all wh-​words can be interpreted with the Q operator through the choice function application as the application of the choice function is constrained by the following condition. (171) The Condition on the Choice Function Application A wh-​expression can be interpreted via the choice function if it can range over a set of elements that are of discrete properties or can be, at least conceptually, individuated.

 69

The syntax and semantics of wh-questions 69 According to our analysis, a set can be considered to be composed of elements that are of discrete properties or can be, at least conceptually, individuated if it can be presupposed as in the common ground of communication. As shown by the following examples, when a wh-​word occurs in an island defined by the Complex NP, the head of the Complex NP must be a bare noun. If the head of the Complex NP is a definite NP, the sentence would be ungrammatical with a wh-​word in the structure. (172)  a. [[Shei xie]

de  shu] zui   youqu

    who write DE book most interesting    ‘Books that who wrote are most interesting?’ b. *[[Shei

xie] de  zhe ben shu] zui  youqu

  who  write DE this CL book most interesting  

‘This book that who wrote are most interesting?’

The ungrammaticality of (172b) follows from the fact that the definite head noun of the Complex NP cannot be interpreted as a set of books that can be used to match the set of authors ranged over by the wh-​word shei contained in the island. What (172b) asks for is the answer related to the matching relation between a set of authors and a set of books. If the matching relationship does not hold, the sentence would become unacceptable as it is unanswerable. The ungrammaticality of the following example discussed previously may also come from the fact that no matching relation could be possibly identified between the wh-​word and the head noun of the Complex NP. (173) *[[Ta  weishenme xie]   he

de shu] zui    youqu

why      write  DE book most

interesting

‘Books that he wrote why are most interesting?’ In (173), weishenme “why” does not range over a set of elements that are of discrete properties or can be individuated, though the head noun shu “book” does range over a set of books. As no matching relation could be established between weishenme and shu, the sentence gets no interpretation. Notice that even if it is argued that weishenme can range over some set, it is an open set in nature and is thus not presupposed as in the common ground of communication. Hence, no possible matching relation could be established between weishenme and the head noun shu that are associated with a set consisting of discrete members presupposed as in the common ground of communication. Since all wh-​expressions need to be interpreted in some way, we can formulate a wh-​interpretation condition as below:

70

70 Wh-questions (174)  The Wh-​Interpretation Condition A wh-​expression must be properly interpreted, and it is properly interpreted if it is interpreted in syntax or semantics. Now, it can be seen that there is a division of labour between wh-​clausal typing and wh-​interpretation. Wh-​clausal typing can be done by either wh-​ raising or the Q operator insertion. Wh-​interpretation in syntax is done by the Agree operation, whereas wh-​interpretation in semantics is realized via the choice function application or the pair-​list reading. In a single wh-​question, the wh-​expression can be interpreted via either the Agree operation or the choice function application. If it is interpreted via the Agree operation, it is directly interpreted in syntax. If it is interpreted via the choice function application, it is interpreted in semantics. In a multiple wh-​question, one of the wh-​expressions can be interpreted in syntax if the Agree operation is available, and the wh-​expressions that are not interpreted via the Agree operation prefer to be interpreted via the pair-​list reading. The above discussion shows that wh-​expressions can be interpreted in either syntax or semantics. Following the spirit of minimalism, I claim that wh-​interpretation in syntax is always preferred to the one in semantics since the former is the most economical way to interpret a wh-​expression (cf. Reuland 2001 for similar ideas in his discussion on reflexive interpretation). The syntactic operation is the most economical one because it does not need the application of any extra nonsyntactic mechanism. If this line of analysis is on the right track, then the interpretation of wh-​expressions must be constrained by the following economy consideration: (175) The Principle of Economy (PE) Choose the most economical operation whenever possible unless it is intended to cancel the interpretation associated with it. a. Syntactic Interpretation > Semantic Interpretation b. Default Interpretation > Nondefault Interpretation   (where A > B means that A is more economical than B) Now, consider why the two sentences in (156), repeated as (176) below, contrast in acceptability: (176) a. *What did who give t to Mary? b. ?What did who give to whom? In (176a), the movement of what violates PE because the movement of the wh-​object what cancels the opportunity for the wh-​subject who to be interpreted in syntax by the Agree operation. Although what is moved to [Spec, CP] in (176a), it is not interpreted in syntax by the Agree operation, according to (166).

 71

The syntax and semantics of wh-questions 71 Note that what is worse is that the fronting of what also cancels the possibility for what to be interpreted in the pair-​list reading since what is less prominent than who and is thus not a qualified set generator. Assuming that in multiple wh-​questions the pair-​list reading is the default interpretation of wh-​words, the movement of what in (176a) results in a second violation of PE since what is not interpreted in the pair-​list reading. Since both who and what fail to be interpreted economically, the ungrammaticality of (176a) is expected. In the above representation, (176b) is marked as “?” since this sentence is not fully acceptable according to the native speakers’ judgment. This is expected since PE is violated when what moves to the operator position, as the fronting of what deprives the wh-​subject who of the possibility of being interpreted in syntax by the Agree operation. (176b) is better than (176a) since, although who fails to be interpreted in syntax, it can still be interpreted in semantics as it can generate a set for whom to be paired with. When two wh-​phrases co-​occur in a sentence and if the prominent one c-​commands the less prominent one, they will try to derive the pair-​list reading. Since who is more prominent than whom and can thus function as a set generator (notice that this shows that a set generator does not necessarily occupy an A'-​position though it must be prominent), both who and whom can get interpreted. What is interesting is that, although the fronted what cannot license other wh-​words that follow it in the pair-​list reading, what itself is licensed in the pair-​list reading. Since the wh-​subject can function as a set generator for the indirect wh-​object whom, and a matching relation can be established between who and whom, the fronted what can be reconstructed to its original position at LF to be interpreted with who and whom in the pair-​list reading. The drastically improved acceptability of (176b) shows that what matters to the multiple wh-​question is whether the wh-​words involved can be interpreted in the pair-​list reading. Suppose that the pair-​list reading (as the default interpretation) is the most economical way to interpret wh-​words involved in the multiple wh-​question. Then, the reason why (176b) is made acceptable is explained. Note that the present analysis predicts that (159b) should not be as good as (159a) since the derivation in (159a), repeated below as (177a), is the optimal one according to PE. In fact, this prediction is consistent with some native speakers’ judgment. In (159b), repeated below as (177b), PE may not be violated, though the fronting of the wh-​object cancels the chance for the wh-​subject to be interpreted in syntax via the Agree operation. However, an inherently D-​linked wh-​phrase may not need to be interpreted via the Agree operation. (177b) is not as good as (177a) since, everything being equal, a closer goal is always preferred to a less close one by the probe. Notice that in (177b) the pair-​list reading is not destroyed by the movement of the wh-​object, given that the inherently D-​linked wh-​subject c-​commands the trace of the moved wh-​object and can thus function as a set generator. (177)  a. Which person bought which book? b. (?)Which book did which person buy?

72

72 Wh-questions Following this line of analysis, we can also account for (154) and (155), repeated as (178) and (179) below: (178)  ?What did which student read t? (179) ?Which book did how many people buy t? The difference between (178) and (176a) is that in (178) the movement of what may not destroy the pair-​list reading of the two wh-​expressions since the wh-​subject is an inherently D-​linked which-​phrase that c-​commands the trace of what and can thus function as a set generator. The less acceptable status of (178) may result from the fact that the grammatical system cannot determine whether PE is violated. Note that PE may be violated in (178) because the movement of what deprives the which-​phrase of its chance to be interpreted in syntax via the Agree operation. PE may not be violated in (178) since the wh-​subject is an inherently D-​linked which-​phrase and thus may not depend on the Agree operation for interpretation, assuming that an inherently D-​linked which-​phrase may depend on discourse rather than syntax for its default interpretation. Now, consider (179). As in (176), although the wh-​ subject fails to be interpreted in syntax by the Agree operation due to the fronting of the D-​linked wh-​phrase in (179), the sentence is not fully ruled out since the pair-​list reading is not destroyed, given that the D-​linked wh-​phrase may be more prominent than the non-​D-​linked wh-​subject according to (170), and it also precedes the wh-​subject. One may ask why (179) is not starred as completely ungrammatical if it violates PE. If we regard PE as a mechanism to derive the preferred interpretation, we can see why (179) is not starred as ungrammatical. In multiple wh-​questions, the preferred interpretation for wh-​ elements is the pair-​list reading. If PE is just treated as a condition to help derive the pair-​list reading, then the reason why (178) and (179) are not ruled out is clear because the pair-​list reading is still maintained. According to Bolinger (1978), sentences like (158b) can also be made more acceptable in a certain context if the relevant wh-​words are forced to be D-​ linked through contextualization. We think that there are basically two possible ways to license multiple wh-​words arranged in a marked order in a single clause: lexical device (D-​linked which-​phrase) or contextualization. But, notice that even when the acceptability of the sentences like (158b) is improved by forcing the wh-​words to be D-​linked through contextualization, the wh-​words cannot be interpreted in the pair-​list reading since the fronted wh-​object is not a qualified set generator. This is also expected by PE since the costly derivation in (158b) may be acceptable only when it is intended to cancel the interpretation associated with the most economical one. It seems that the most economical way to derive wh-​questions is to keep the order of wh-​words that is established by the default or canonical word order since only from such an order can the pair-​list readings be obtained by the wh-​words that are not inherently D-​linked. But this account seems to fail to account for the following sentence:

 73

The syntax and semantics of wh-questions 73 (180)  who did everyone like? The above sentence can have a pair-​list answer, in addition to an individual answer and a functional answer, as shown below: (181) a. John and Mary. (individual answer) b. His mother. /​Their mothers. (functional answer) c. Bill likes John, and Tom likes Mary. (pair-​list answer) The question to ask is why the pair-​list reading can be obtained in (180) even when the wh-​object is fronted. I claim that (180) can still obtain its pair-​ list reading not only because everyone is an inherent set generator that c-​ commands the trace of who but also because moving the wh-​object who in this sentence is the only licit way to form a wh-​question. In this sense, movement is the last resort. I assume that if a licit wh-​question can be formed without moving the wh-​word, then moving the wh-​word will result in the loss of pair-​ list readings. This is exactly what happens in Chinese. Consider the following Chinese sentences: (182) a. Meigeren dou xihuan shei  ne?   everyone all  like   who   Q   ‘Who does everyone like?’ b. shei, meigeren dou xihuan ne?   who everyone all   like   Q In (182a), the pair-​list reading obtains, but in (182b), it is lost. All this suggests that an unmarked derivation is to keep an unmarked meaning. Whenever a canonical or unmarked derivation is available in syntax, the use of a marked one is to cancel the semantic implication associated with the unmarked one. In this sense, PE can be seen as a means to guarantee the unmarked derivation of wh-​questions in syntax. Although PE is not an absolute grammatical constraint, any derivation that violates it has a price to pay because PE specifies the most economical way to interpret wh-​questions. In the above discussion, we have shown that PE is basically a mechanism to help derive the pair-​list reading in multiple wh-​questions. In fact, what lies behind PE is the prominence of the relevant wh-​words. PE is employed to guarantee that the less prominent wh-​word can depend on the most prominent one for interpretation since only in this way can the pair-​list reading be properly derived, given the fact that the production of the pair-​ list reading depends on a set generator, and only a prominent wh-​word can function as the set generator. Consider the following sentence, repeated from (100):

74

74 Wh-questions (183)  *Zhangsan weishenme xihuan shei? Zhangsan  why 

  like 

who

Lit. ‘Why does Zhangsan like who?’ In the above sentence, the wh-​adjunct is interpreted in syntax via the Agree operation, but the wh-​object fails to be interpreted via the pair-​list reading since the wh-​word preceding it is not prominent and thus cannot function as a set generator. Since the wh-​object fails to be interpreted in the pair-​list reading, PE is violated in (183), and hence, the sentence is out. Notice that we have argued that in multiple wh-​questions, all the wh-​words involved prefer to be interpreted in the pair-​list reading. In (183), both the wh-​object and the wh-​adjunct fail to be interpreted in the pair-​list reading, given that the wh-​adjunct precedes the wh-​object but fails to function as a set generator. The facts discussed above indicate that in multiple wh-​questions, the most important thing is to derive the pair-​list reading. If the violation of PE results in the loss of the pair-​list reading, the sentence will be excluded, whereas, if the violation of PE does not result in the loss of the pair-​list reading, as shown in (178) and (179), the sentence may not be ruled out. Notice that the present analysis does not explain whether the wh-​object shei “who” in (183) can be interpreted via the choice function application, though it predicts that (183) should be ruled out because the wh-​object in it fails to be interpreted in the pair-​list reading. I  think that the wh-​object in (183) cannot be interpreted via the choice function application. If it can, the reason why (183) is unacceptable will not be adequately explained. In section 2.3.5, I will explain why the wh-​object in (183) cannot be interpreted via the choice function application. 2.3.3  The structure of Chinese wh-​questions In Chinese wh-​questions, the Q particle is located at the sentence-​final position, but in our previous discussion, we seem to assume that C is located at the left periphery of the sentence. How is this possible, given our proposal that the Q particle is an operator in C? Obviously, the structure of wh-​questions in Chinese is not stated clearly in our previous discussion. Following Ning’s (1993) proposal for Chinese relative clauses, I propose the following structure for Chinese wh-​questions: (184) 

CP

Spec

Opx

C’

IP



[... whx ...]

Qx

 75

The syntax and semantics of wh-questions 75 In (184) the Q particle is under C°, which is head-​final and located at the right periphery of the sentence. In this case, my proposal that the Q particle is an operator must be inaccurate. Let us say that the Q particle is a realization of the C head, and a Q operator is independently introduced into the Spec of CP to support and agree with C°. All this done, the sentence becomes a Q sentence, but not a wh-​question. In order to form a wh-​question, C needs to be interpreted with a wh-​word. This is done by associating the Q operator Op with the closest wh-​word, according to PCTC. When Op is associated with a wh-​word, it will share the wh-​feature with it. Since Op is in an agreement relation with the Q particle, it will transmit its wh-​feature to it. As a result, a wh-​question is formed. 2.3.4  Multiple wh-​questions revisited With the distinction between clausal typing and wh-​interpretation, we can account for the multiple wh-​phenomena in a more satisfactory way. Consider (126), repeated below as (185): (185)  a. *Whoi do you wonder [CP ti [IP ti bought what]]? b. *Whati do you wonder [CP ti [who bought ti]]? (185a) can only be derived via cyclic movement. If not, the cyclicity requirement given in PCTC will be violated. As a result, the sentence would fail to be typed. However, if the wh-​word moves cyclically, it cannot move to the matrix CP domain when it moves into the embedded [Spec, CP] since it would be anchored there in agreement, given that the embedded C has a wh-​ feature as required by the matrix verb. Note that the derivation in (185a) also violates PE since the probe and the goal are not closest to each other before movement. As a result, the wh-​word in (185a) is not interpreted in syntax by the Agree operation. Now, consider (185b), which contrasts with (129), repeated below as (186), in derivational history. (186) ?*Whati do you wonder [CP ti [CP whoj [tj bought ti]]]? (185b) is an illicit derivation since (i)  what is not allowed to make further movement when it agrees with the embedded C, and (ii) the fronting of what is excluded by PE. In (186), who agrees with the embedded C to satisfy the subcategorization requirement of the matrix verb, and what moves to the matrix CP domain via embedded CP adjunction. (186) constitutes a standard wh-​island violation, but is not that worse than (185a), a phenomenon usually accounted for under proper government. Under the present analysis, (186) is not that worse because it does not violate PCTC since the subcategorization requirement for the embedded C is satisfied by who. The derivation is not well-​formed because the movement of what violates PE. Note that, according to our condition on wh-​interpretation in syntax, what is not interpreted in syntax, though it is located

76

76 Wh-questions in the matrix CP domain. It is obvious that the most economical way for what to be interpreted is that it pairs with who in the pair-​list reading. However, its movement destroys that preferred interpretation since what is less prominent than who and thus fails to function as a set generator though it precedes who. That is why it is not a good sentence. Now, consider (128), repeated here as (187): (187)  *Whoi do you wonder [CP whatj [IP ti bought tj]]? Although what is moved into the embedded CP domain in (187), it is not interpreted in syntax by the Agree operation, according to our definition. Since the preferred interpretation for who is cancelled by the movement of what, PE is violated in (187). Besides, the movement of what also cancels its own opportunity to be interpreted in the pair-​list reading, which constitutes another violation of PE. The movement of who to the matrix [Spec, CP] in (187) also violates PE since the probe and the goal are not closest to each other before the movement and who thus fails to be interpreted in syntax by the Agree operation. Since PE has been violated several times, there is no way for (187) to be grammatical. Now, consider the following Chinese multiple wh-​question: (188) Ta wen ni  [shei mai-​le    shenme]. he ask you  who buy-​ASP  what ‘He asked you who bought what.’ The present analysis expects that the multiple wh-​words in (188) can only have the embedded indirect question reading. Since the embedded C has a Q feature as required by the matrix verb, the wh-​subject cannot take matrix scope by crossing the embedded C, according to PCTC and PE. If the wh-​subject, as a goal, is associated with the matrix probe (Q operator) for interpretation, it is not interpreted in syntax via the Agree operation since they are not the closest to each other. The closest probe for the wh-​subject is the embedded Q operator. Since the wh-​object is not contained in a strong island, it can move at LF, but such a movement would cancel its opportunity to be interpreted in the pair-​list reading and thus violates PE, given that the most economical way for the wh-​ object to be interpreted is to pair with the embedded wh-​subject in the pair-​list reading. Notice that even if the wh-​object is associated with the matrix Q operator for interpretation, it is not interpreted in syntax via the Agree operation since they are not the closest to each other. Under the present analysis, if we insert a Q operator into the Spec of the matrix CP in (188), that means that we want to type the sentence as a wh-​question. But associating the matrix Q operator with the wh-​object cannot make the sentence a wh-​question since PCTC is not satisfied, given that the wh-​object is not the closest to the matrix Q operator. It should be noted that the above examples in English and Chinese do not indicate that wh-​words in embedded multiple wh-​questions cannot take wide

 77

The syntax and semantics of wh-questions 77 scope in general. In fact, they only indicate that a Q feature containing C of the embedded clause blocks the wide scope interpretation of the wh-​word. If the relevant C does not contain a Q feature, the wh-​word can take wide scope, as shown below: (189)  a. Who do you think bought what? b. Ni renwei shei mai-​le   shenme?   you think who buy-​ASP what ‘Who do you think has bought what?’ In the above sentences, the wh-​words in the embedded clauses can have a wide scope reading. Now, consider the following sentence: (190) Who wonders who bought what? (190) can be answered by (191a) and (191b) but not (191c) and (191d) since the latter two answers are not grammatical sentences themselves.13 (191) a.  John does. b. John wonders who bought the book, and Bill wonders who bought the pen, and … c.  *John wonders Bill bought what, and Mary wonders Ted bought what, and … d.  *John wonders Bill bought the book. Neither (191c) nor (191d) is grammatical because the embedded C is not typed by a wh-​word thus failing to satisfy the subcategorization requirement of the matrix verb. The problem is how to account for the acceptability of (191b). It seems that the embedded wh-​object can be interpreted out of the embedded CP domain which is typed as [+WH]. Following Kuno and Robinson (1972), I argue that the embedded wh-​object may not be interpreted as a direct question word in (190), though it can have the pair-​list reading with the matrix wh-​word. Notice that the embedded wh-​object in (190) can be left without being provided a value, as indicated by (191a). In (192), both of the wh-​words must be obligatorily provided a value in the answer, as shown in (193), since the two wh-​words are direct question words. (192) Who bought what? (193) a. John bought a book, and Bill bought a pen, and … b. *John does.

78

78 Wh-questions It can thus be assumed that only indirect questions can be left unanswered. Since I  assume that the embedded wh-​object in (190) is a wh-​word in an indirect question, I  must explain why it can have an answer, given that an indirect question does not need an answer. I think that the distinction between a direct question and an indirect question does not lie in the possibility of having an answer, but in the obligatoriness of having an answer. It is obligatory for a direct question to seek an answer, whereas it is unnecessary for an indirect question to have an answer. This is their fundamental difference. For instance, the following Chinese sentence (194) can be optionally answered by (195), though it is an indirect question: (194)  Wo xiang-​zhidao  shei hui lai. I  want-​know 

who will come

‘I want to know who will come.’ (195) Lisi hui lai. Lisi will come. ‘Lisi will come.’ The English translation of (194) can also be answered by the English translation of (195). Suppose that one speaks to an addressee with a sentence like “I want to know who will come.” It is quite natural for the addressee to respond with a sentence like “Lisi will come,” though the former sentence is not a wh-​question. The following examples discussed in Kuno and Robinson (1972: 481) further show that answerability may not be taken as evidence to support the claim that an answerable wh-​word should be interpreted as a direct question word. (196) Who remembers where we bought these books? (197) a.  John does. b. John remembers where we bought the physics book, and Martha and Ted remember where we bought The Wizard of Oz. (196) can be answered by (197a), but what is interesting is that it can also be answered by (197b). Of course, no one wants to treat these books as a question word. If (197b) as a possible answer to (196) cannot prove that the NP these books is a direct question word, (191b) as a possible answer to (190) also cannot prove that the embedded wh-​object is a direct question word. Notice that, although I argue that the embedded wh-​object in (190) may not be interpreted as a direct wh-​question word, I do not claim that it cannot be interpreted with the matrix wh-​subject in the pair-​list reading. In fact, nothing under the present analysis can prevent it from having the pair-​list reading with the matrix wh-​subject. Under the present analysis, wh-​island effects are

 79

The syntax and semantics of wh-questions 79 predicted by PCTC. But PCTC won’t be violated when the embedded wh-​ object is interpreted with the matrix wh-​subject since typing is done by the matrix wh-​subject and the embedded wh-​object plays no role in typing. 2.3.5  A note on weishenme and A-​not-​A, and PCTC further revised Since we make a distinction between the Agree operation and the choice function application, we can explain why there is a difference in grammaticality between the following sentences: (198)  a. Ta weishenme xie  shu?   he why      write book  

‘Why did he write books?’

b. *[Ta weishenme xie  de shu]   zui 

youqu?

    he why      write DE book most interesting   Lit. ‘Books that he wrote WHY are most interesting?’ In (198a) the wh-​word weishenme is interpreted in syntax via the Agree operation since it can move to [Spec, CP] at LF. In (198b) the wh-​word weishenme cannot move to the matrix [Spec, CP] since it is contained in an island. Hence, it cannot be interpreted in syntax by the Agree operation. Since the Agree operation is not available, it is quite natural for us to proceed to the next step and see whether the choice function can be applied. According to our condition on the choice function application given in (171), choice function can only apply to wh-​expressions that can range over a set of elements that can be individuated. If this is true, then the reason why weishenme cannot be interpreted via the choice function is clear. It cannot range over a set of elements that can be individuated. Consider the following sentences: (199) a. Ta zenme ye/​dou  bu ken qu.   he how  FM/​all

not will go

  ‘He won’t go any way.’ b. *Ta weishenme ye/​dou  bu  ken  qu.     he  why     FM/​all    not will  

go

‘He won’t go for any reason.’

In the above sentences, zenme “how” can be focused by an emphatic marker ye or quantified by dou “all,” but weishenme “why” cannot.14 Xu (1990: 371) notes that zenme as well as other wh-​words in Chinese can be interpreted as either indefinite pronouns or adverbs, and only weishenme shows some peculiarity in this aspect. I think that the reason why weishenme cannot be interpreted as an indefinite adverb results from the fact that it refuses individuation. Because it cannot range over a set of elements that can be individuated, it

80

80 Wh-questions cannot be used as an indefinite adverb and thus refuses to be focused by ye or quantified by dou, which requires the individuation of the relevant wh-​phrase. Now, let us consider another question related to weishenme in (199b). If weishenme cannot be properly interpreted when contained in a strong island, are sentences like (198b) still wh-​questions? This amounts to asking whether weishenme in sentences like (198b) can type the sentence as a wh-​question. According to the definition of PCTC given in (161), (198b) should be a wh-​ question. But we intuitively feel that it is not a qualified wh-​question. Note that the typing condition defined for wh-​raising languages in (161a) is a strict condition, according to which wh-​words extracted from strong islands cannot type a clause as a wh-​question. Hence, one of the possible reasons why the following sentence is ungrammatical is that it fails to be typed as a wh-​question. (200)  *Whoi do you like [the book that ti wrote]? According to (161b), a clause would be typed as a wh-​question if the Q operator is interpreted with a wh-​word that is the closest to it. However, (161b) does not make it clear what “interpret” means. Let us assume that a wh-​word is interpreted with a Q operator if it is interpreted with it via either the Agree operation or the choice function. If this is true, sentences like (198b) can be excluded by reformulating PCTC as below: (201) The Pure Clausal Typing Condition (PCTC) a.  For wh-​raising languages, a clause is typed as a wh-​question iff there is a wh-​word that moves overtly into [Spec, CP] via cyclic movement without crossing any strong island. b.  For Q-​particle languages, a clause is typed as a wh-​question iff there is a wh-​word interpreted with the closest C[+Q] via either the Agree operation or the choice function application. Following the analysis assumed in section 2.3.3, let us assume that in Chinese interrogatives C has a Q feature, and the Q particle is an overt instantiation of C.  All this done, the clause becomes a Q clause, but not a wh-​question. In order to form a wh-​question, C needs to be interpreted with a wh-​word. If the relevant wh-​word is not contained in a strong island, it can directly move to [Spec, CP] at LF to agree with C and be interpreted with C. If the relevant wh-​word is contained in a strong island, it cannot move to [Spec, CP] because of the island constraint. Since the operator insertion strategy is available in Chinese, a Q operator can be inserted into [Spec, CP] to bind the relevant wh-​word. Since binding means feature transmission and the Q operator agrees with C, the clause is typed as a wh-​question. In (198b), the wh-​word weishenme “why” is contained in a strong island, it cannot move to [Spec, CP] at LF. Then, it has to be interpreted with C via the choice function.

 81

The syntax and semantics of wh-questions 81 However, since weishenme cannot range over a set of elements that can be individuated, it cannot be interpreted via the choice function. Since there is no way for C to be interpreted with a wh-​word, the clause fails to be typed as a wh-​question. As a result, (198b) violates PCTC. Assuming that a sentence that violates PCTC is uninterpretable, we can see that the ungrammaticality of (198b) must result from its uninterpretability. A welcome result of this analysis is that we can make a distinction between the untyped sentences and the illicit derivations that violate PE. Consider the following sentences ((202b) is repeated from (70)): (202)  a. *[NP [Kare-​ga naze kai-​ta] hon]-​ga omosiroi-​desu-​ka?         he -​N why wrote book-​N 

interesting-​be-​Q

  Lit. ‘Books that he wrote why are interesting?’   (Nishigauchi 1990: 93) b.  ?(?)[NP [Dare-​ga naze kai-​ta] hon]-​ga omosiroi-​desu-​ka?         Who –​N why wrote book-​N  

interesting-​be-​Q

  Lit. ‘Books that who wrote why are interesting?’ According to Nishigauchi (1990), (202a) is completely ungrammatical, but (202b) is somewhat better than (202a), if not perfectly grammatical, when another wh-​phrase is added. This difference in grammaticality between the above two sentences is anticipated by the present analysis. Under the present analysis, (202a) is in fact uninterpretable since the sentence fails to be typed as a wh-​question, given that naze “why” cannot be interpreted with C when it occurs in an island. The acceptability of (202b) is improved since PCTC is satisfied and the sentence is typed as a wh-​question by the wh-​word dare “who,” which can be interpreted with C via the choice function. It is not fully grammatical because naze fails to be interpreted. The difference in acceptability between (202a) and (202b) shows that it is necessary to make a distinction between the uninterpretability of the sentence and the uninterpretability of the wh-​word. (202a) is worse because, besides having an uninterpretable wh-​word in it, the sentence itself is uninterpretable since it fails to be typed as a wh-​question. (202b) is better because, although there is an uninterpretable wh-​word in it, the sentence is interpretable since it is typed as a wh-​question. In our previous analysis, we assume that the A-​not-​A element is formed by the incorporation of the Q operator into the INFL, following Shi (1994). On the basis of this assumption, let us use our newly revised PCTC to give a more satisfactory explanation to the ungrammaticality of the following sentence, repeated from (67). (203) *[Ta qu-​bu-​qu  meiguo] bijiao hao? He go-​not-​go America more good Lit. ‘Is it better whether he goes to America or not?’

82

82 Wh-questions Following Huang (1991), let us assume that the A-​not-​A element is the Chinese counterpart of whether in English and thus a counterpart of the wh-​ element with a feature specified as [-​WH].15 If the A-​not-​A element is a wh-​ element, it cannot move out of the subject island at LF according to (165) since LF movement is constrained by island conditions under the present analysis, though it can move to the C of the subject clause. In (203), the Q operator is directly adjoined to the INFL of the subject clause to trigger the A-​not-​A formation. (If the Q operator is adjoined to the matrix INFL, it will be the matrix INFL that forms an A-​not-​A.) According to (201b), only when a wh-​word is interpreted with the closest C can the relevant sentence be typed as a wh-​question. In this case, it is the subject clause rather than the matrix clause that is typed as an A-​not-​A question since the A-​not-​A element containing the operator can move to [Spec, CP] of the subject clause. Since the A-​not-​A element cannot move out of the subject island, the only way for it to take matrix scope is to insert a Q operator in the matrix [Spec, CP]. Even if we assume that we can insert another Q operator into the matrix [Spec, CP] without considering the economy requirement of the grammatical system, the matrix clause still fails to be typed as a question since the matrix operator binds nothing. First, the matrix operator cannot bind the A-​not-​A element because it already agrees with the Q operator that is incorporated into it. This means that the choice function cannot apply to the A-​not-​A element since it is already interpreted via the Agree operation in the subject clause. Second, the matrix clause cannot be typed as a wh-​question because PCTC cannot be satisfied, due to the fact that the newly inserted matrix operator is not the closest operator to the A-​not-​A element, as compared with the Q operator introduced into the subject clause and incorporated into the INFL of the subject clause. According to Law (2001), (203) is ungrammatical since the subject clause is interpreted as a direct question and the matrix predicate is incompatible with the argument, which is a direct question. The present analysis is consistent with Law’s account. According to the present analysis, (203) is ungrammatical because, besides violating the ban on vacuous binding, it also violates PCTC and thus fails to be typed as an A-​not-​A question, though the sentence needs to be typed and interpreted as a question. Consider the following sentence: (204)  [Wo qu meiguo haishi bu qu meiguo] bijiao hao?  I  go America or    not go America more good ‘Is it better that I go to America or that I do not go to America?’ (Huang 1991: 313) According to Huang (1991), the above sentence is a true disjunctive question. Following Huang (1991), I assume that true disjunction questions are not formed by the incorporation of the Q operator into the INFL. Since the Q operator can be inserted in the matrix [Spec, CP], and the questioned

 83

The syntax and semantics of wh-questions 83 elements contained in the subject clause can be licensed and interpreted via the choice function, (204) satisfies PCTC and is thus typed and interpreted as a question. Pan (personal communication) asks if the sentences like (205a) would cause any problem for the newly defined PCTC. In the following sentence, the A-​not-​A element can satisfy PCTC, and the wh-​word shenme “what” can be interpreted via the choice function. It seems that our further refined analysis would wrongly predict that it is a well-​formed wh-​question. (205)  a. *Wo xiang-​zhidao Lisi xi-​bu-​xihuan   shenme?  

I   want-​know   Lisi like-​not-​like  what

  Lit. ‘I want to know whether Lisi likes or not like WHAT?’ b. [CP Who [IP t bought what]]? c. *Zhangsan weishenme    xihuan      shei?   Zhangsan   why      like     who   Lit. ‘Why does Zhangsan like who?’ In fact, the ungrammaticality of (205a) is just predicted by the present analysis. First, since the embedded Q operator is incorporated into the embedded INFL, there is no Q operator to bind the wh-​word shenme. Inserting another Q operator into the embedded CP domain is not allowed by the grammatical system. Even if it is allowed, it will result in the feature conflict, given that the A-​not-​A element and the wh-​word have different wh-​features. Second, the matrix clause cannot be typed as a wh-​question by inserting another Q operator into the matrix CP domain. According to PCTC, a clause is typed as a wh-​question if there is a wh-​word interpreted with the closest Q operator. Although in (205a) the wh-​word can be interpreted via the choice function if an operator is inserted into the matrix [Spec, CP], it cannot type the matrix clause as a wh-​question because the matrix operator is not the closest Q operator to the wh-​word. The Q operator closest to the wh-​word is the embedded Q operator, which is incorporated into the INFL, and raised to the embedded Spec of CP at LF. Although this operator is available, it cannot bind the wh-​ word because of the feature incompatibility. Reinhart (1998) suggests that wh’s-​in-​situ be bound by an existential Q operator and interpreted as function variables via choice functions. Under her analysis, a Q operator must be independently introduced into the sentence to bind the function variables. I think that in multiple wh-​questions, if the pair-​ list reading is intended, the operator that binds the function variables does not need to be independently introduced if there is a wh-​word that moves into [Spec, CP] either in syntax or at LF. I assume that a fronted wh-​word can function as the Q operator that binds not only its own trace but also other wh-​words in ​situ. Notice that the LF raised wh-​operator in fact does not exist in English under the present analysis since (i) in English a wh-​question must

84

84 Wh-questions be typed by an overtly moved wh-​word according to PCTC, and (ii) when the clause is typed, no wh-​word is allowed to move into [Spec, CP] according to the Condition on Wh-​Movement given in (165), since the relevant wh-​ movement is unmotivated. It is not motivated for clausal typing because the clause is already typed. It is not motivated for wh-​feature checking because the wh-​feature on the wh-​word is not strong in English. Although the LF raising of a wh-​operator is not available in English, it is available in Chinese. In Chinese, a wh-​word can raise to [Spec, CP] at LF if it is not contained in an island, and the relevant wh-​movement is motivated for wh-​clausal typing, given that a wh-​word must be interpreted with the closest C[+Q] in wh-​clausal typing according to PCTC. I think that the insertion of an operator is the last resort. An operator can be independently inserted into the clause only if no wh-​word in the clause is allowed to function as an operator or when a wh-​in-​ situ is intended to take wide scope over the moved wh-​element. If this line of analysis is on the right track, then in (205b) the wh-​word who, as an operator, binds not only its own trace but also the wh-​object interpreted as a function variable. In (205c), repeated from (183), the wh-​word weishenme functions as an operator after it moves into [Spec, CP] at LF. However, since weishenme cannot range over a set of entities, it cannot interpret the wh-​object in ​situ in the pair-​list reading, though it can interpret its own trace.

2.4  Further discussion 2.4.1  Wh-​island effects in Japanese It is assumed that Japanese differs from Chinese with respect to wh-​island effects (cf. Wantanabe 2001). I have shown in the previous discussion that this assumption is based upon a false picture of Chinese multiple wh-​questions, and Chinese also exhibits wh-​island effects. If there is no difference between Chinese and Japanese with respect to multiple wh-​interpretation, what makes linguists think that they should be different? This seems to be related to the fact that no linguist is a native speaker of both Chinese and Japanese. What is interesting is that if we go through the relevant literature on Japanese wh-​ questions, we will soon find that it is not the case that all linguists assume that Japanese is a wh-​island effect sensitive language, and in fact, it is still under debate whether Japanese exhibits genuine wh-​island effects (cf. Lasnik and Saito 1984, 1992; Nishigauchi 1990; Pesetsky 1987; Watanabe 1992). A natural question to ask is: why do linguists differ in their judgments on the same set of data? A possible answer is that if one can get an echo question reading for the relevant wh-​word in multiple wh-​questions in Japanese, one tends to think that there is no wh-​island effects, whereas if one fails to get an echo question reading for the relevant wh-​word, one tends to conclude that wh-​island effects exist. Consider the following Japanese sentence (206a) cited from Nishigauchi (1990: 33):

 85

The syntax and semantics of wh-questions 85 (206)  a. Tanaka-​kun-​wa [dare-​ga nani-​o tabe-​ta-​ka] oboe-​te-​i-​masu-​ka?             Top who-​N what-​A eat-​P  Q remember is    Q  

‘Does Tanaka know who ate what?’

b. For which x, x a person, does Tanaka know what x ate? According to Nishigauchi, for some speakers, (206a) can be interpreted as (206b). However, he also notes that even for those speakers that can interpret (206a) as (206b), there are some restrictions, one of which is that the wh-​ word that purports to take matrix scope must be pronounced with a marked intonation (i.e., with a heavy stress on it). This is interesting since I find that the Chinese wh-​words in (107) and (130) also need a heavy stress support when it is intended to be interpreted as a wide scope-​taking wh-​word in an echo question. Hence, I  think that when the relevant wh-​word can obviate wh-​island effects in Japanese multiple wh-​questions, it is in fact interpreted as an echo question as in Chinese. Now, we can have a fourth criterion to distinguish an original question from an echo question. A wh-​word in an original question does not need any extra stress support, but an echo wh-​word needs it. Another assumed difference between Chinese and Japanese with respect to wh-​quantification lies in the interaction between wh-​in-​situ and QP (quantifier phrase), as shown below: (207) a. ?* Daremo-​ga nani-​o katta  no?  (Japanese)     everyone  what  bought Q b. Nani-​o daremo-​ga 

t  katta Q?

c. Meigeren dou mai-​le    shenme?  (Chinese)   everyone all  buy-​ASP what  

‘What did everyone buy?’ (Watanabe 2001: 215)

Hoji (1986:  88) observes that if a wh-​phrase is c-​commanded by a QP, the sentence becomes ungrammatical in Japanese, as indicated by (207a), but when the relevant wh-​phrase is scrambled over the QP, the sentence becomes grammatical, as shown in (207b). Watanabe (2001) notices that the Chinese counterpart (207c) to (207a) is not only grammatical but also allows a pair-​list reading like (207), which is rendered in English: (208) John bought beer, Mary a bottle of wine, … Watanabe finds that in this respect, Japanese is different from Chinese since not only is (207a) unacceptable but also the acceptable (207b) lacks a pair-​list reading, as observed by Hoji (1986). According to Watanabe, even

86

86 Wh-questions those who find (207a) acceptable do not get the pair-​list reading. Based on the above facts, Watanabe assumes that Chinese differs from Japanese in whether wh-​movement is involved. Japanese is a wh-​movement language, and the Japanese sentence (207a) is unacceptable because wh-​movement is blocked by a c-​ commanding QP daremo-​ga. Chinese is a language that involves unselective binding but not wh-​movement, and thus (207c) is grammatical since unselective binding is not blocked by an intervening QP. Here, two things should be considered. First, why is (207a) unacceptable but (207c) is acceptable? Second, why is the pair-​list reading available in (207c) not available in (207b)? I think that the answer to the first question should not be so complicated as to investigate whether some abstract wh-​movement is involved in Japanese or Chinese. If we look at the relevant QPs in (207), we can find that the comparison is made between two different QPs in these two languages. The Japanese QP daremo is made of the wh-​word dare “who” plus the domain-​widening particle mo. The Chinese QP meigeren “everyone” is made of the distributive quantifier mei “every,” the classifier ge, and the noun ren “man.” Hence, meigeren can be counted as a Chinese counterpart to the English QP everyone, but not to the Japanese QP daremo. Notice that Chinese has a counterpart to daremo, which is shei dou “who all.” The Chinese dou “all” is basically equivalent to the Japanese mo in meaning and function, and when a wh-​word is quantified by dou, the relevant wh-​word becomes a universal QP, as shown below: (209) Shei dou  xihuan zhe-​ben shu. Who all   like    this-​CL book ‘Everyone likes this book.’ The interesting thing is that if we replace the object in (209) with a wh-​ word and make a Chinese sentence similar to (207a), we get a sentence that is as unacceptable as (207a) if shei “who” is treated as a QP, as shown below: (210)  ?*Shei dou mai-​le    shenme? Who all  buy-​ASP what ‘What did everyone buy?’ Another interesting thing is that if we topicalize the wh-​object, the sentence becomes acceptable, just as the Japanese sentence (207b) is. (211) Shenme, shei dou mai-​le? What  who all  buy-​ASP What will happen if we replace the QP in (207a) with another one that is not derived from a wh-​word? There is such a word in Japanese, which is minna “everyone.” If we replace the QP in (207a) with minna, the sentence

 87

The syntax and semantics of wh-questions 87 is not only acceptable but also allowed for a pair-​list reading, as shown below:16 (212)  Minna-​ga nani-​o katta no? Everyone what  buy  Q ‘What did everyone buy?’ Now, we can see that the first question is not a real question since two different things are being compared. In fact, the first question should be asked in this way: why are (207a) and (210) unacceptable? I think that the answer can be derived from PCTC. If we apply PCTC to (207a) and (210), the Q operator will automatically be associated with the closest wh-​elements. However, the closest wh-​words in these two sentences are also chosen as bindees by another operator which is a universal quantifier, mo in Japanese and dou in Chinese. Since a wh-​word bound by two different quantifiers will result in a feature conflict, (207a) and (210) are thus ruled out. (207b) and (211) are grammatical because in each sentence, the fronted wh-​word is the one that is the closest to the Q operator, and thus the Q operator and the universal quantifier can bind two different wh-​words, thus avoiding a struggle between them. Now, let us turn to the second question: Why is the pair-​list reading available in (207c) not available in (207b)? Notice that the pair-​list reading is also unavailable in the Chinese sentence (211). This is because, in both (207b) and (211), the optional fronting of a wh-​word will cancel the semantic implication that obtains when the wh-​word stays at its original site, as we have discussed previously. Notice that both (207c) and (212) can have a pair-​list reading. But when the wh-​objects are fronted, the pair-​list reading is lost, as shown below: (213) a. Shenme meigeren dou mai-​le?   what  

everyone all buy-​ASP

b. Nani-​o minna-​ga katta no?   what  everyone buy  Q If we account for this phenomenon in minimalist terms, we can say that movement is never cost free. The so-​called optional movement is not optional since it can derive new meanings by cancelling old ones.17 2.4.2  Additional wh-effects Grewendorf (2001) proposes a quite interesting account for multiple wh-​ questions. He argues that in multiple wh-​fronting languages like Bulgarian, a wh-​word can be the landing site for another wh-​word in overt wh-​movement, and the wh-​cluster formed by this movement then moves to the Spec of CP to form a wh-​question. He further claims that in Japanese there is a covert

88

88 Wh-questions process of wh-​cluster formation, which is thus an LF-​type analogue of overt multiple wh-​fronting languages like Bulgarian. The idea to adjoin one wh-​ word to another one and then to move the newly formed wh-​cluster away is not totally new. At least, it can be traced back to Saito (1994). What is new is that all wh-​words can be adjoined to one another to form one cluster in either syntax or LF. Although this account is attractive, it is neither conceptually nor empirically adequate. Consider the following sentence, repeated from (70): (214)  ?(?)[NP [Dare-​ga naze kai-​ta] hon]-​ga omosiroi-​desu-​ka?       Who –​N why wrote  book-​N interesting-​be-​Q Lit. ‘Books that who wrote why are interesting?’ Under Grewendorf’s analysis, the improvement of acceptability of the above sentence is due to the fact that the wh-​adjunct can adjoin to the wh-​ argument, and then the newly formed wh-​cluster can covertly move out of the island. Since his analysis is based on Minimalism (Chomsky 2000, 2001), one would ask how it is possible for a wh-​element to be moved out of islands without violating the spirit of minimalism. Notice that since government is abandoned in minimalism and ECP plays no role, Grewendorf has to explain why movement should be sensitive to the distinction between wh-​arguments and wh-​adjuncts, besides accounting for what motivates an element to move out of islands. There are also empirical problems with Grewendorf’s analysis. First, his account fails to explain why in the following Bulgarian examples discussed in his paper (cited from Richards 1997: 242), (215b) is more acceptable than (215a): (215)  a. *Koja knigai otreče senatorăt  [mălvata   če  iska    da zabrani ti]?   which book denied the-​senator the-​rumor that (he)-​wanted to ban   ‘Which book did which senator deny the rumor that he wanted to ban?’ b. ?Koj senator   koja  knigai otreče [mălvata  če  iska    da zabrani ti]?   which senator which book denied the-​rumor that (he)-​wanted to ban   ‘Which book did which senator deny the rumor that he wanted to ban?’   ‘Which senator denied the rumor that he wanted to ban which book?’ In (215a) the sentence is typed by a wh-​expression that is extracted from a complex NP, and in (215b), the sentence is typed as a wh-​question by the matrix wh-​subject, and a wh-​phrase is extracted from the complex NP. If Bulgarian is a wh-​cluster formation language, then the two wh-​words in (215b) must form a wh-​cluster. However, it is unknown at what stage the wh-​cluster is formed since the two wh-​words are separated by a Complex NP island. If no wh-​cluster is formed in (215b), then one may ask why and when the wh-​ cluster formation process can be optional. Besides, if no wh-​cluster is formed

 89

The syntax and semantics of wh-questions 89 in (215b), the question why (215b) is better than (215a) is left unanswered in Grewendorf’s analysis. Consider the following Japanese sentence in (216). According to my informants, there is no difference between (216) and (214) in acceptability. Notice that in (216), the wh-​subject dare “who” does not occur in the island, and thus there is no chance for the wh-​adjunct in the island to adjoin to it first and then move together with it out of the island. When we take (216) into consideration, we can see clearly that the improved acceptability of (216) has nothing to do with the assumption that a wh-​adjunct can move out of an island if it forms a wh-​cluster with a wh-​argument within the island. (216)  ?(?)Dare –​ga [[John-​ga naze kai-​ta] hon]-​o katta  ka?   Who –​N  

John  why wrote-​P book-​N bought Q

Lit. ‘Who bought the book that John wrote why?’ Second, if Japanese is a covert wh-​cluster formation language, one may also ask why the two wh-​words in (206a) do not form a cluster and then move to the matrix operator position to be interpreted as a direct matching question. Notice that if the two wh-​words form a wh-​cluster in (206a), there should be no wh-​island effects since there is only one wh-​word which is a wh-​cluster. I think that the so-​called additional WH effects have nothing to do with the assumed wh-​cluster formation process. The acceptability of (214) is improved with a wh-​subject added simply because the wh-​subject can be interpreted with the matrix C via the choice function application, and hence, the relevant sentence is typed as a wh-​question, thus satisfying PCTC. It is not fully acceptable because the wh-​adjunct naze “why” is uninterpretable in a strong island. The same analysis applies to (216). The acceptability of (215b) is improved because the sentence is successfully typed as a wh-​question and PCTC is satisfied. It is not fully acceptable because the movement of the wh-​object violates Subjacency. (215a) is worse than (215b) and thus fully ungrammatical because it violates PCTC, besides violating Subjacency. Notice that the present analysis established on the interaction between PCTC and other grammatical constraints not only captures Richards’ (1997, 1998) observation expressed in his Principle of Minimal Compliance (PMC) but also fares better than his analysis. The following is the definition of PMC (Richards 1998: 601): (217) Principle of Minimal Compliance For any dependency D that obeys constraint C, any elements that are relevant for determining whether D obeys C can be ignored for the rest of the derivation for purposes of determining whether any other dependency D' obeys C.

90

90 Wh-questions According to PMC, (215a) is unacceptable because Subjacency is violated, and (215b) is better than (215a) in acceptability because Subjacency is obeyed by adding a wh-​subject. The basic idea underlying Richards’ PMC is that a certain constraint need be obeyed only once in a clause. Since Subjacency is obeyed once in (215b), it can be ignored by the movement of the wh-​object. Although PMC can account for the contrast between (215a) and (215b), it fails to account for the contrast between (215b) and the following Bulgarian sentence, as pointed out by Grewendorf (2001: 92): (218)  *Koji kakj ti iska   da kaže molitva [predi da intervjuitame Marija tj] who how  wants to say prayer  before to we-​interview Maria According to Grewendorf (2001), the above example, which is provided to him by Marina Stojanova, is ungrammatical, though PMC is satisfied. If Subjacency need be obeyed only once in a clause, the movement of the wh-​ adjunct from the adjunct clause should be allowed, given that Subjacency is obeyed by the wh-​subject in (218). It is obvious that PMC cannot account for the contrast between (215b) and (218) in acceptability. Under the present analysis, (218) is unacceptable because when the wh-​adjunct moves to the matrix CP domain from the adjunct island, it not only violates Subjacency but also fails to be interpreted. It cannot be interpreted with the matrix C in syntax via the Agree operation, due to the fact that the matrix C and the wh-​adjunct contained in the adjunct clause are not the closest to each other. It also cannot be interpreted in semantics because it is a wh-​adjunct and can be interpreted neither via the choice function application nor in the pair-​list reading. Notice that under the present analysis, only the first fronted wh-​ word licensed by the Agree operation is interpreted in syntax in multiple wh-​fronting languages, and wh-​expressions other than the one licensed by the Agree operation can only be interpreted in semantics, though they need to move to the CP domain to check off their strong wh-​features.18

2.5  Economy in wh-​interpretation revisited This work shows that a wh-​ question must satisfy the following two requirements: (219) A wh-​question must be properly typed. (220) A wh-​expression must be properly interpreted. Neither (219) nor (220) needs to be stipulated in the grammar since (219) can be derived from the assumption that all clauses have complementizers (Bresnan 1970), and (220) can be derived from the Principle of Full Interpretation (Chomsky 1986a). On the basis of Reuland’s (2001) study of reflexives, I assume that when multiple wh-​questions are considered, the abovementioned

 91

The syntax and semantics of wh-questions 91 two requirements can be satisfied in the following four different ways, which can be ranked according to the Principle of Economy: (221) a. Discourse interpretation Semantic interpretation              wh1 + wh2 + …whn = 1 Syntactic derivation and interpretation  Typing + Interpreting = 1 (The number of cross-​modular operations is 2.) b. Discourse interpretation Semantic interpretation           

  wh1 + wh2 + …whn = 1, …, n

Syntactic derivation and interpretation  Typing + Interpreting = 2 (The number of cross-​modular operations is 3.) c. Discourse interpretation              wh1 + wh2 + …whn = 1 Semantic interpretation           

  wh1 + wh2 + …whn = 2, …, n

Syntactic derivation and interpretation  Typing + Interpreting = 1 (The number of cross-​modular operations is 4.) d. Discourse interpretation              wh1 + wh2 + …whn = 1 Semantic interpretation           

  wh1 + wh2 + …whn = 2, …, n

Syntactic derivation and interpretation  Typing + Interpreting = 2 (The number of cross-​modular operations is 5.) (221a) can account for the following sentence: (222)  Who bought what? In (222) the typing of the clause and the interpretation of who can be done by one operation in syntax. Hence, the number of the syntactic operation is 1.  Since who and what can be interpreted in the pair-​list reading by one semantic operation, the number of cross-​modular operations for (222) is 2. (221b) and (221d) can account for the following contrast in acceptability: (223) a. *What did who give t to Mary? b. ?What did who give to whom?

92

92 Wh-questions In (223a) typing does not mean the interpretation of what. Hence, the number is 2, since typing and the interpretation of what have to be done by two different processes in syntax. As pointed out in our previous discussion, in (223a) the movement of what cancels the opportunity for what and who to be interpreted in the pair-​list reading, and since what and who have to be interpreted by two independent operations, the number of semantic operations is thus 2. Due to the fact that who and what cannot be interpreted as members of one pair by one operation in semantics and thus have to be interpreted as members of one pair by accessing the discourse, the total number of cross-​ modular operations is 5. Now, consider why the grammaticality of (223b) is greatly improved. Since in (223b) typing also does not mean the interpretation of what, the number of operations involved in syntax is 2.  (223b) is better than (223a) because all the wh-​elements involved in (223b) can be interpreted in the pair-​list reading by one semantic operation. Hence, the number of the semantic operation involved is 1, and the total number of cross-​modular operations involved is 3. Notice that the present analysis can also correctly predict that (223b) should not be as good as (222) since the former involves more operations than the latter. (221c) can predict the unacceptability of the following sentence: (224)  *Zhangsan  weishenme xihuan shei? Zhangsan  

why      like    who

Lit. ‘Why does Zhangsan like who?’ In (224) the typing of the clause and the interpretation of weishenme can be done by one operation in syntax. Hence, the number of the syntactic operation is 1.  Since weishenme and shei cannot be interpreted as members of one set in the pair-​list reading by one semantic operation, the number of operations involved in semantics is 2, and the total number of cross-​modular operations is 4, given the fact that these two wh-​elements have to be interpreted as members of one set by accessing the discourse. The above analysis shows that economy can be evaluated in terms of the number of operations involved. The basic idea is that the relevant derivation will become worse if the computational complexity of the syntax increases.

2.6  Summary This part of the book discusses how wh-​questions are formed and how wh-​ elements are interpreted in Chinese and English and shows that the formation and interpretation of wh-​questions are constrained by economy considerations, specifically, PCTC and PE, which incorporate two factors: prominence and locality. It is shown that the derivation and interpretation of wh-​questions are affected not only by locality conditions but also by prominence considerations. It is the interaction between prominence and locality that determines how wh-​questions can be derived and interpreted in the most economical way.

 93

The syntax and semantics of wh-questions 93

Notes 1 Examples like (27a) were first noted by Kayne (1984). 2 Although zenmeyang, like zenme, can have two distinctive readings, means, and manner, it does not have the causal reading that zenme has. 3 According to Tsai (1994a), both relatives and appositives are located in the projection of N. 4 Tsai argues that verbs may be distinguished by their selection of complement clauses with different [N]‌features in their COMPs. Tsai suggests that verbs such as yihan “regret,” jide “remember,” tongyi “agree,” are distinct from verbs such as renwei “think,” cai “guess,” and shuo “say” since the former select a [+N] complement clause, whereas the latter can only select a [-​N] complement clause. The distinction roughly corresponds to Lin’s distinction between verbs of opinion and verbs of conjecture. 5 Note that if weishenme “why” occurs in the bracketed clause in (68a), as shown below, the sentence would be ungrammatical, and this shows that the bracketed part and the rest of the sentence should not be treated as two sentences. (i)  *[Ta  weishenme bu lai]   rang ni

zheme shangxin?

  He  why      not come let  you so

  sad

Lit. ‘WHY he did not come so as to make you so sad?’ 6 Note that weishenme can also be used to express reason. 7 If Miyagawa (2001) is right in claiming that in Japanese the wh-​feature is on T and wh-​phrases must move to the Spec of TP to enter into agreement with T, then Japanese might marginally belong to the fourth type in which clauses are typed by a Q particle, and wh-​words are licensed by T, though not by C. 8 Pan (personal communication) points out that the two disjuncts in (106) can have truth values and both of them can be true, though the two who’s cannot have the same value. That is, the first who can have a value for the first disjunct, and the second who for the second disjunct. But the values for the two who’s have to be different, though the value for who in (104) has to be the same, which may be the reason why (104) is not acceptable. 9 Although xiang-​zhidao “want-​know” is extensively cited in the WH literature, it is not a compound word in Chinese, as pointed out by Xu (1990). Xu (personal communication) further points out that the fact that xiang-​zhidao must be followed by a wh-​clause as its complement clause shows that it is the whole predicate rather than one single verb that subcategorizes for a complement clause. 10 Janda (1985) maintains that echo questions are derived by lexical substitution, and an echo wh-​word may substitute not just for a Determiner or an NP but also for a verb, for a VP, for lexical subconstituents, for a syllable, or for a nonconstituent string of syllables. The following is one of Janda’s examples: (i) Speaker A: She believes in ad-​/​sub-​jacency. Speaker B: She believes in what … -​jacency? 11 Pan (personal communication) asks how the echo question is typed. A plausible answer may be that echo questions are universally typed by an intonation operator, which binds a wh-​element that bears contrastive stress. Since standard English employs different strategies for typing original questions and echo questions, a

94

94 Wh-questions wh-​question is never ambiguous between the wh-​reading and the echo reading in syntactic form. Chinese wh-​questions are ambiguous between these two readings in surface form since wh-​words are not raised in this language. However, this ambiguity can be resolved at the phonological level since the wh-​element in an echo question is bound by an intonation operator and thus bears contrastive stress. 12 Note that shi “BE” is different from jiujing-​shi. The former is compatible with an echo wh-​word since it can be used as a copula and thus may be ambiguous between a copula and a focus marker uses, but the latter is incompatible with an echo wh-​word because it is not ambiguous in its identity and cannot be used as a copula. 13 Pan (personal communication) points out that the reason why (191c) and (191d) are not acceptable is because the embedded who is merged with the embedded C and selected by the matrix verb, and thus one cannot supply an answer to it. The present analysis is consistent with Pan’s view. (191c) and (191d) are ungrammatical because the subcategorization of the matrix verb is not satisfied. Notice that if one does supply a value to who, one has to move the wh-​object what to satisfy the subcategorization requirement of the matrix verb. However, if what is fronted, PE will be violated since it is not the closest wh-​word to the embedded C. Moving what, besides cancelling the possibility for who to be interpreted in syntax, will also cancel its own possibility to be interpreted in semantics via the pair-​list reading. 14 When ye means also, (199b) is grammatical, but on this reading, ye is not used to focus the wh-​word. 15 When an element is specified as [-​WH], it means that the relevant element belongs to the WH family but cannot be interpreted as a wh-​question word. Note that a wh-​word interpreted as an indefinite NP is also specified as [-​WH]. Since both the indefinite wh-​word and the A-​not-​A element are specified with the same feature, they can co-​occur within the same sentence, as shown below: (i)  Ni  xiang-​bu-​xiang chi 

dian  shenme?

you want-​not-​want  eat  some  what ‘Would you like to eat something?’ 16 The difference between Chinese and Japanese is that Japanese tends to use the QP derived from the wh-​word to denote the notion of everyone, but such a preference is not found in Chinese. 17 Pan (personal communication) points out that the loss of the pair-​list reading in (213) may be related to the linear order and the contrast between the movement of the wh-​element and the nonmovement of the wh-​element. If not moving the wh-​element can produce the pair-​list reading, then moving it will lose the relevant reading. In English, there is no such contrast in sentences like those in (213), as the wh-​element has to move. 18 Since the present study does not focus on multiple wh-​fronting languages, I will not discuss whether multiple [Spec, CP]s or multiple [Spec, IP]s, and the notion of “tucked in,” as proposed by Richards (1997), should be employed to account for the distribution of the multiple fronted wh-​words. For the relevant discussion, see Rudin (1988), Richards (1997), and Grewendorf (2001).

 95

Part II

Reflexives

96

 97

3  Prominence and locality in reflexive binding

Introduction In this part of the book, I propose a unified account for both the noncontrastive compound reflexive and the bare reflexive in Chinese (as well as the reflexive in English) and show that they are constrained by the same reflexive binding condition, which stipulates that a reflexive should be bound to an accessible prominent NP in its binding domain defined by the candidate set related to the most prominent NP. Since different languages may have different definitions of the most prominent NP, then the parameterization of the prominence of NPs will determine the parameterization of the binding domain of reflexives across languages and thus account for the different properties of reflexives either in different languages or within the same language. This part of the book contains six sections in addition to the Introduction. In the section 3.1, I investigate Reinhart and Reuland’s (1993), Reuland and Reinhart’s (1995) (henceforth R&R) and Lidz’s (2001a, 2001b) predicate-​ based binding theory and their characterization of anaphoric expressions and show that neither of their theories can adequately characterize the binding properties of anaphoric expressions in Chinese. In section 3.2, I give my own characterization of Chinese anaphoric expressions. In section 3.3, I discuss the peculiar binding properties of the Chinese bare reflexive ziji and review two accounts of ziji. In section 3.4, I  show that the binding of the noncontrastive compound reflexive and the bare reflexive is governed by the same binding condition established on the prominence of NPs. Based upon Huang and Tang’s (1991) assumption that ziji has neither referential features nor phi-​features, I  develop an approach of two feature search engines to account for the binding properties of the bare reflexive ziji and show that the blocking effect exhibited by ziji results from the union of the two most prominent NPs selected by the two search engines. In section 3.5, I give a further discussion on the derivational cycle, the noncontrastive and contrastive compound reflexives as well as the multiple occurrences of ziji. Section 3.6 is the summary of this part of the book.

98

98 Reflexives

3.1  The predicate-​based binding theory R&R (1993, 1995) propose to replace Chomsky’s argument-​based binding theory with their predicate-​based nonstructural binding theory that requires a reflexive-​marked predicate be reflexive (Condition A) and a reflexive predicate be reflexive-​marked (Condition B). They establish a typology of anaphoric expressions with the property Reflexivizing function and the property R (referential independence). Their analysis predicts that the SELF anaphor—​a complex reflexive—​cannot be long-​distance (LD) bound when occupying an argument position, since it can reflexive-​mark a predicate, whereas the SE anaphor—​a bare reflexive—​which does not reflexive-​mark a predicate can be LD bound. Lidz (2001a, 2001b) tries to extend R&R’s theory by distinguishing Pure-​reflexives from Near-​reflexives. He claims that the former are anaphors that occur with lexically reflexive-​ marked predicates and require a complete identity between the reflexive and its antecedent, whereas the latter are anaphors that syntactically reflexive-​mark the predicate and do not impose the abovementioned identity requirement. In the following discussion, I will show that neither R&R’s (1993, 1995) nor Lidz’s (2001a, 2001b) theories can be extended to Chinese to correctly characterize the properties of its reflexives, since their theories are both too strong and too weak and will thus make wrong predictions regarding the binding and referential possibilities of reflexives in Chinese. Besides failing to differentiate the bare reflexives in Chinese such as ziji “self,” benren “self,” and benshen “self,” R&R’s typology of reflexives, if extended to Chinese, would wrongly exclude the existence of reflexives like ziji “self ” and ta-​ziji “he/she-self ” in Chinese since they can not only reflexivize their predicate but also find long-​ distance antecedents when occurring in an argument position. I argue that, although it is necessary to make a distinction between Near-​reflexives and Pure-​reflexives, Lidz’s (2001a, 2001b) Condition R is not a universal principle since, when applied to Chinese reflexives, besides wrongly grouping simplex reflexive ziji with complex reflexive ta-​ziji, it also fails to make a distinction between reflexives such as ta-​ziji that do not require identity and reflexives like ziji, benren, and benshen that require identity. This section also discusses a possible way to reanalyse the Chinese data so as to save R&R’s predicate-​ based theory of reflexivization but only finds that this is impossible, and it concludes that reflexives in natural language are not of a homogeneous class and thus cannot be adequately characterized simply by the two properties given in R&R (1993) and the Condition R proposed by Lidz (2001a, 2001b). 3.1.1  R&R’s predicate-​based theory of reflexives 3.1.1.1  R&R’s typology of reflexives Following Faltz’s (1977[1985]) and Pica’s (1987) observation that complex reflexives are universally locally bound, whereas simplex reflexives are

 99

Prominence and locality 99 universally long-​distance bound, R&R (1993) classify the former as SELF anaphors and the latter SE anaphors. They establish a typology of anaphoric expressions with the property Reflexivizing function and the property R (referential independence), as illustrated below: (1)                          SELF 

  SE   

  PRONOUN

Reflexivizing function      +   

    -​        -​

R(eferential independence)  -​   

    -​        +

The property R (referential independence) applies to both R-​expressions and pronouns, whereas the property Reflexivizing function only applies to SELF anaphors. As seen in (1), SE anaphors not only form a group with SELF anaphors with respect to the R property, as both are referentially deficient (-​R) expressions, but also form a group with pronouns with respect to the Reflexivizing function since none of them can reflexivize a predicate. Besides, SE anaphors also pattern with pronouns in having an empty head, as illustrated in (2), whereas SELF anaphors have the head N position occupied by “self,” as shown in (3). (2) a. [NP Pron [N’ …e…]] b. [NP SE [N’…e…]] (3) [NP Pron/​SE [N’ self]] According to R&R, both SELF and SE anaphors allow (discourse) uses known as logophoric. They argue that the terms local anaphor and long-​ distance anaphor are highly misleading because both kinds of anaphors can be used logophorically when they do not occupy argument positions, especially the object position, and can thus occur at all kinds of distances. They think that there are two domains for the occurrence of an anaphor:  one is the domain of reflexivity, and the other the domain where SE anaphors are allowed to be bound. The first is the domain where a SELF anaphor obligatorily reflexivizes a predicate and where both pronouns and SE anaphors are excluded. The second is the domain where the binding of the SE anaphors must obey the Tensed-​S Constraint (R&R 1993). 3.1.1.2  R&R’s predicate-​based binding theory R&R (1993) formulate the following conditions for the theory of reflexive binding. (4) Conditions A: A reflexive-​marked syntactic predicate is reflexive. B: A reflexive semantic predicate is reflexive-​marked.

100

100 Reflexives (5)  Definitions a. The syntactic predicate of a (head) P is P, all its syntactic arguments, and an external argument of P (subject).   The syntactic arguments of P are the projections assigned θ-​role or Case by P.   The semantic predicate formed of P is P and all its arguments at the relevant semantic level. b.   A predicate is reflexive iff two of its arguments are co-​indexed. c.  A predicate (formed of P) is reflexive-​marked iff either P is lexically reflexive or one of P’s arguments is a SELF anaphor. Condition A  draws a dividing line between the bound use and the logophoric use of SELF anaphors. It requires that SELF anaphors reflexive-​ marking a syntactic predicate be interpreted reflexively and also implies that SELF anaphors can be used logophorically if they do not occupy an argument position of a syntactic predicate and thus do not reflexive-​mark the relevant predicate. This analysis can adequately account for the contrast between the sentences in (6) (R&R 1993: 670): (6) a. Max boasted that the queen invited Lucie and himself for a drink. b. *Max boasted that the queen invited himself for a drink. (6a) is grammatical because the SELF anaphor in (6a) does not occupy an argument position by itself and thus does not reflexive-​mark the relevant predicate. In this case, Condition A does not apply, and the SELF anaphor can be used logophorically. (6b) is ungrammatical because Condition A  is violated. The SELF anaphor in (6b) occupies an argument position and thus reflexive-​marks the predicate. However, it is not reflexive since it is not co-​ indexed with its co-​argument, the queen. Condition B requires that a reflexive predicate be reflexive-​marked. (7) is ungrammatical because the reflexive predicate is not reflexive-​marked. (7) *Maxi criticized him i. However, if no reflexive predicate is formed, reflexive-​ marking is not required. Consider the following sentence: (8) a. Max i likes jokes about him i. b. Max (λx (like (x, jokes about x))) In (8a) the pronoun is not an argument of the predicate like, but an argument of the nominal predicate joke, as illustrated by the semantic representation given in (8b). In this case, the co-​indexation between Max and the

 101

Prominence and locality 101 pronoun him does not yield a reflexive predicate. Hence, Condition B is vacuously met in (8a). Since it is a condition on reflexivization rather than on the distribution of pronouns, Condition B predicts that SE anaphors, like pronouns, cannot occur as an argument of a predicate that is not lexically reflexive. (9) *Max i haat zich i. Max  hates SE In the Dutch example (9), a reflexive predicate is formed, as the two arguments of the predicate are co-​indexed. Since zich is not assumed to be a reflexive-​marker and the predicate is also not lexically reflexive-​marked, (9) is ruled out as ungrammatical by Condition B.  Now, consider the following sentence: (10)  Willem i shaamt zich i /​*hem i. Willem shames SE

/​*him.

Although it can correctly rule in the co-​indexation between Willem and the SE anaphor zich in (10) since the reflexive predicate formed by the co-​ indexation of the subject and zich is intrinsically reflexive-​marked by the inherent lexical property of the verb, Condition B cannot properly rule out the illicit co-​indexation between Willem and the pronoun hem in (10) since the reflexive predicate formed by the co-​indexation of the subject and hem is also intrinsically reflexive-​marked. Note that the relevant co-​indexation cannot be ruled out by Condition A, either, as the lexically reflexive-​marked predicate is indeed made reflexive by the relevant co-​indexation. In order to rule out such an unwarranted co-​indexation, R&R (1993) propose a Condition on A-​ chains, which can regulate the distribution of SE anaphors and pronominals. The definition of an A-​chain and the condition on its well-​formedness are given below: (11) Definition (R&R 1993: 693) An A-​chain is any sequence of co-​indexation that is headed by an A-​position and satisfies antecedent government. (12) General Condition on A-​chains (R&R 1993: 696) A maximal A-​chain (α1,…, αn) contains exactly one link—α1—that is both +R and Case-​marked. Now, the illicit co-​indexation between Willem and hem in (10) will be ruled out by the Chain Condition. According to the feature analysis given in (1), SE anaphors are different from pronouns in terms of R(eferential independence). SE anaphors have the feature [-​R] and pronouns, the feature [+R]. The

102

102 Reflexives co-​indexation between Willem and hem in (10) is illicit because the chain formed from such a co-​indexation will contain two [+R] elements, violating the Chain Condition, whereas that between Willem and zich is licit, since the tail of the chain zich is [-​R], not violating the Chain Condition defined in (12). 3.1.2  Lidz’s condition R Lidz (2001a, 2001b) argues that there are limitations in R&R’s (1993) predicate-​based theory of reflexives since their theory would wrongly predict that reflexive-​marked predicates are semantically uniform. He claims that predicates which are reflexive-​marked by SELF anaphors should be distinguished from those which are lexically reflexive-​ marked. According to Lidz (2001a, 2001b), there is a distinction between Near-​reflexives and Pure-​reflexives, and the former are anaphors that syntactically reflexive-​mark the predicate, while the latter are anaphors that require a lexically reflexive-​ marked predicate. These two types of anaphors differ from each other in their semantic interpretations:  Pure-​reflexives require a complete identity between the reflexives and their antecedents, whereas Near-​reflexives do not impose such a requirement. The difference between Pure-​reflexives and Near-​ reflexives is reflected in their respective semantic representations, given in (13): only the former is translated as a bound variable, whereas the latter is translated as a function related to the antecedent (Lidz 2001b). (13)  a. λx [P(x, x)] 

  (Semantic/​Pure-​reflexive)

b. λx [P(x, f(x))]   (Near-​reflexive) In (13a) the two arguments of the predicate P are directly related to each other and thus have the same identity, whereas in (13b) the two arguments are related to each other via a Near-​reflexive function f and thus need not be identical, though in most cases the two are extensionally equivalent. Lidz (2001a, 2001b) proposes the following Condition R to account for the fact that lexically or morphologically reflexive-​marked predicates disallow the Near-​reflexive interpretation. (14) Condition R λx [P(x, x)]  ↔   (θ1=θ2) semantics        theta-​grid The left side of the formula is the semantic representation of reflexivity, and the right side is the theta-​grid requirement of a lexically/​morphologically reflexive predicate. Condition R states that if a predicate is semantically reflexive, then it must be lexically/​morphologically reflexive, and if a predicate is lexically or morphologically reflexive, then it must be semantically reflexive. Like R&R’s theory, Condition R can also rule out the illicit co-​indexation as shown in (15). This is because this sentence is semantically reflexive, as zich is

 103

Prominence and locality 103 assumed to introduce the Pure-​reflexive function, but not lexically reflexive. Notice that, if a predicate is lexically reflexive, it is inherently reflexive-​marked. For instance, the verb in (16) is inherently reflexive-​marked, but the one in (15) is not. If a predicate is morphologically reflexive, it is reflexive-​marked by a morpheme. Lidz (1995, 2001b) extends the R&R’s definition of reflexive-​ marked predicates by including morphological reflexivity, which can be found in languages like Kannada, which has a morpheme reflexive-​marking a predicate. Lidz points out that in Kannada the anaphor tannu cannot be bound by a co-​argument if the verbal reflexive morpheme –​kol (-​koND in past tense) is absent, as illustrated by the contrast in (17). (15)  a. *Max i haat zich i. b. *Maxi hates him i. (16) a. Max scheert zich.   Max shaves self   ‘Max shaves himself.’ b. Max scheert zichzelf.   Max shaves self-​self   ‘Max shaves himself.’ (17) a. *Hari tann-​annu hoDe-​d-​a   Hari self-​A

  hit-​P-​3SM

  ‘Hari hit himself.’ b. Hari tann-​annu hoDe-​du-​koND-​a   Hari self-​A 

hit-​PP-​REFL.PST-​3SM

  ‘Hari hit himself.’ In addition, Condition R can predict that the reflexive in (16a) and (17b) can only get a Pure-​reflexive interpretation, while that in (16b) can get a Near-​ reflexive interpretation. The reflexives in (16a) and (17b) cannot get a Near-​ reflexive reading because the predicates are lexically/​morphologically reflexive and must thus be semantically reflexive, according to Condition R. (16b) can have a Near-​reflexive reading because zichzelf syntactically reflexive-​marks the predicate in question and thus introduces the Near-​reflexive function.1 3.1.3  Limitations of the predicate-​based theory of reflexives 3.1.3.1  Problems with R&R’s characterization and their typology of reflexives When we apply R&R’s (1993) predicate-​based theory of reflexives to the analysis of reflexives in Chinese, the first problem we encounter is that we do

104

104 Reflexives not know how to classify the simplex reflexive ziji. Since there is a contrast between the simplex reflexive ziji and the complex reflexive ta-​ziji, it seems that ziji should be analysed as a SE anaphor. This analysis can account for the fact that ziji can be long-​distance bound. However, if ziji is a SE anaphor, then it cannot reflexivize the predicate and thus cannot be bound by its co-​ argument when the relevant predicate is not lexically reflexive, though ziji can obviously be locally bound by its co-​argument, as illustrated below: (18)  Johni xihuan zijii. John like   self ‘John likes himself.’ (18) indicates that ziji should not be a SE anaphor but can be a SELF anaphor since it can reflexivize a predicate. However, if ziji is a SELF anaphor, it should not be long-​distance bound when occurring in an argument position. But this prediction is not borne out. (19) Johni renwei Billj xihuan zijii/​j. John think  Bill like    self ‘Johni thinks that Bill j likes himi/​himselfj.’ In (19) ziji can refer to the matrix subject across the local one. If ziji is a SELF anaphor, the long-​distance binding in (19) should be ruled out, according to R&R, since ziji, occupying an argument position, reflexive-​ marks the local predicate and should thus be co-​indexed with its co-​argument. What is worse is that ziji in (19) can be both locally and long-​distance bound. In order to account for these facts, R&R would have to assume that there are two kinds of ziji: one is a SELF anaphor and the other is a SE anaphor. However, nothing in their theory can help us distinguish these two kinds of ziji in Chinese. In fact, R&R’s theory is too weak because it cannot rule out the illicit binding in the following sentences: (20) a. [Johni de baba]j hai-​le 



ziji*i/​j.

  John DE father hurt-​ASP self  

‘John’s father hurt himself.’

b. Johni zhidao woj bu xihuan ziji*i/​j.   John know  I  not like    self  

‘John knows that I do not like myself.’

If we apply R&R’s analysis to (20), the illicit binding of ziji would be ruled in as grammatical since ziji can be a SE anaphor and can thus be bound by a non-​co-​argument antecedent. The blocking effect of ziji as exemplified

 105

Prominence and locality 105 in (20b) also presents a problem for R&R, as they would fail to predict the impossibility of long-​distance binding in (20b), as their theory, if extended to Chinese, would predict that the LD binding should be possible, with the assumption that ziji can be a SE anaphor. Notice that, although we can appeal to ziji’s property of being a SELF anaphor to account for (20), sentences like (19) will be problematic for such an appeal. Hence, it is not easy for R&R to have a consistent criterion to differentiate SELF ziji from SE ziji. These problems not only arise for the simplex reflexive ziji but also for the complex reflexives like ta-​ziji in Chinese. According to R&R’s (1993) typology of reflexives, complex reflexives like ta-​ziji should be classified as SELF anaphors. Although R&R’s theory can correctly predict the local binding of the complex reflexive ta-​ziji in (21), it fails to do so in (22) because their theory is not only too weak, as discussed above, but also too strong. Their Condition A would wrongly rule out the well-​formed sentences in (22). (21)  Johni renwei Bill j xihuan ta-​ziji*i/​j. John think  Bill like  

himself

‘Johni thinks that Bill j likes himself*i/​j.’ (22) a. Johni de jiaoao hai-​le    ta-​zijii.   John DE pride hurt-​ASP   himself  

‘John’s pride hurt him.’

b. Johni shuo [zhejianshi]j hai-​le    ta-​zijii/​*j.   John say  this-​matter hurt-​ASP himself  

‘John said that this matter hurt him.’

R&R’s theory would predict that, when occupying an argument posi­ tion, complex reflexives, being a SELF anaphor, cannot be LD bound, which contradicts with the fact that complex reflexives like ta-​ziji in argument positions can be LD bound, as shown in (22) and noted in Pan (1998). Although the complex reflexive ta-​ziji can reflexive-​ mark the predicates in (22a) and (22b), respectively, both sentences should be ungrammatical, according to R&R, since the reflexive-​marked predicates in (22a) and (22b) are not reflexive due to the fact that the co-​arguments of verb hai “hurt” are not co-​indexed, thus violating R&R’s Condition A. Notice that one cannot account for the LD binding possibility in (22) by claiming that the occurrences of ta-​ziji there are logophoric, as they are not contrastive and occupy an argument position. Besides, sentences like (22b) also present a problem for R&R’s A-​chain Condition. One property of the A-​chain Condition is that the A-​chain domain of a given NP is a subset of the binding domain of this NP, and there should be no barrier between any two of the links in the A-​chain. Hence, the A-​chain for the reflexive in (23) should be confined to the embedded clause:

106

106 Reflexives (23)  Johni thinks [CP that hei likes himselfi]. In (23) even though the embedded subject he bears the same index as the matrix subject John, it does not form an A-​chain with it since they are in different binding domains. The same analysis can be applied to the following Chinese sentence: (24)  Johni renwei [CP tai xihuan ta-​zijii]. John think 

  he like  himself

‘John thinks that he likes himself.’ In (24), the embedded subject ta “he” also does not form an A-​chain with the matrix subject John. If it formed an A-​chain with John, the A-​chain condition would be violated since the relevant A-​chain would contain two [+R] elements. These facts indicate that in Chinese, CP is a barrier for A-​chain formation. However, if it is really the case that in Chinese an A-​chain can only be formed within the CP domain, the following sentence, repeated from (22b), should be ruled out as ungrammatical, according to R&R’s A-​chain Condition. (25)  John shuo [CP zhejianshi John say 

hai-​le    ta-​ziji].

  this-​matter  hurt-​ASP himself

‘John said that this matter hurt him.’ In (25) ta-​ziji does not form an A-​chain with the matrix subject because they are separated by a CP barrier. However, if the A-​chain domain of ta-​ziji is restricted to the embedded CP, it will not have a [+R] element in its A-​chain and should thus be ruled out as ungrammatical by the Chain Condition. Obviously, (25) is grammatical. If we extend the A-​chain domain of ta-​ziji to the whole sentence, the A-​chain Condition will be satisfied in (25), but (24) will be wrongly ruled out since its A-​chain contains two [+R] elements. Note that ta-​ziji in (25) occupies an argument position and is not in focus, and thus it is not a logophor, according to R&R’s theory. In fact, (25) will be wrongly ruled out twice under R&R’s analysis, as it is also ruled out by Condition A: although ta-​ziji, a SELF anaphor, reflexive-​marks the predicate, the predicate in the embedded clause is not reflexive, as the co-​arguments of the predicate in question are not co-​indexed. Although one may change the barrier definition for Chinese ta-​ziji by stipulating that a barrier is a CP with a human subject to account for sentences like (25), this change will cause problems when we consider the cases involving ziji (e.g. (19)). In (19) the embedded clause will be considered as a barrier for chain formation, and this would predict that ziji cannot form an A-​chain with the matrix subject and thus cannot have the matrix subject as its antecedent, a wrong prediction.

 107

Prominence and locality 107 Finally, R&R’s two-​way classification of reflexives fails to differentiate the three bare reflexives in Chinese: ziji, benren “self,” and benshen “self,” as discussed in Pan (1997). If we apply R&R’s typology of reflexives to these reflexives in Chinese, they should all be classified as SE anaphors, as they do not have systematic variation in terms of number, gender, or Case. However, as discussed in Pan (1995, 1997), these bare reflexives exhibit different properties. For instance, the bare reflexive benren, different from ziji, cannot refer to any c-​ commanding subject when occupying an object position, as illustrated below: (26)  Johni renwei Bill j kanbuqi    benren*i/​*j. John  think Bill contemplate self ‘John thinks that Bill contemplates me.’ Although benren in (26) can refer to neither the local subject nor the matrix subject and can only refer to the speaker, it can, nevertheless, be used anaphorically, as illustrated below (Pan 1997: 187–​188): (27) a. Tai shi wo de tongshi, jiating chushen   buxiang, benreni chengfen   He be I

DE colleague family background unknown self 

class-​status

  xuesheng.   student   ‘He is my colleague. His family background is not clear. His own class status is student.’ b. Nashuiren i chu

benreni mianshui’e    wai hai keyi xiangshou

  Tax-​payer except self    exempt-​amount out also can  enjoy   beifuyangren de mianshui’e.   foster-​child

DE exempt-​amount

  ‘Besides the exempt amount of their own, tax payers can also enjoy the exempt amount for the foster child.’ In (27a) benren has an antecedent across a clause boundary, and in (27b) benren occurs in an adjunct phrase and has the local subject as its antecedent. Note that, although benren cannot be replaced by ziji in (27a), it can in (27b). If we follow R&R’s analysis and simply group benren and ziji into the same type (i.e., the SE anaphor type), their different properties will be left unaccounted for. Reuland (2001) recasts some of the ideas in R&R (1993) in the minimalist terms. Since he still maintains that only complex reflexives can license reflexivization, the question why (18) is grammatical is still left unanswered. According to Reuland (2001), the bare reflexive ziji and subject in (18) should form a CHAIN which functions as the only argument of the predicate. Hence,

108

108 Reflexives (18) should be ruled out since there is a conflict between the arity of xihuan “like,” which is a two-​place verb and the fact that the CHAIN makes it a one-​place predicate. However, (18) is grammatical, which casts serious doubt on Reuland’s (2001) analysis. Note that, although it is possible to analyse the Chinese bare reflexive ziji as a complex reflexive with an empty pronoun pro-​ziji, similar to ta-​ziji, in Reuland (2001), this analysis will not recognize the differences between ziji and ta-​ziji and thus leave them unaccounted for. Another problem with Reuland (2001) is that English reflexives like himself have to be treated as simplex reflexives incapable of bearing an independent theta role when they are associated with inherently reflexive-​marked verbs as in “John behave himself,” but as complex reflexives capable of bearing an independent theta role when associated with verbs that are not inherently reflexive-​ marked as in “John likes himself.” Since simplex reflexives have different structures from complex reflexives under Reuland’s analysis, it is unknown how English reflexives should be structurally represented when they are associated with these two kinds of verbs. 3.1.3.2  Problems with Lidz’s Condition R The extension to R&R’s theory by Lidz (2001a, 2001b), who makes a distinction between Pure-​reflexives and Near-​reflexives, will not work for Chinese, either. Lidz (2001a, 2001b) distinguishes syntactically reflexive-​ marked predicates from lexically and morphologically reflexive-​marked predicates, as pointed out in section 3.1.2. He argues that anaphors licensed by a syntactically reflexive-​marked predicate are Near-​reflexives, and those licensed by a lexically/​morphologically reflexive-​marked predicate are Pure-​reflexives. Specifically, any anaphor that can be locally bound in the absence of lexical reflexivity introduces the Near-​reflexive function. According to Lidz (2001a, 2001b), the simplex reflexive ziji in Chinese is a Near-​reflexive since it can be locally bound by its co-​argument in the absence of lexical reflexivity. Therefore, in the following sentence, ziji can either refer to John or his statue: (28)  John ba   ziji qiangbi le John BA

self shoot

ASP

‘John shot himself (=statue or John).’ According to Lidz, (28) can be uttered in a situation in which John goes into a wax museum where he finds a statue depicting himself, and he gets enraged at seeing this statue. Under this situation, (28) can be uttered to mean either that John shot himself or the statue, according to Lidz. This is interesting, but not true. I have consulted many native speakers, and none of them thinks that (28) can mean that John shot the statue. If one wants to have the relevant reading, ta-​ziji has to be used instead of ziji.

 109

Prominence and locality 109 There are cases where ziji can indeed refer to the statue, though this is usually limited to situations in which the relevant sentences are spoken humorously. Suppose that Queen Elizabeth II and John are in front of the wax figure of Queen Elizabeth II. The Queen finds that there is something wrong with the nose of the wax figure, and she raises her fingers to touch the nose, but when she touches the nose, the nose suddenly falls off. Under this context, John may say the sentence below with humour, and in this case, ziji can refer to the statue.2 (29)  Nüwangi, nii  zenme ba zijii de bizi   Queen 

nong diao le?

you how  BA self De nose  make off ASP

Lit. ‘Queen, why did you make your nose come off ?’ Although ziji in (29) can get the Near-​reflexive reading, it should be pointed out that ziji is different from ta-​ziji in obtaining the Near-​reflexive reading. It is easy for ta-​ziji to get the nonidentity (Pan 1997) or Near-​reflexive reading, but difficult for ziji to get the same kind of reading. In fact, only when used humorously can ziji get the Near-​reflexive reading. According to the native speakers’ judgement, if used in some humorous situation, even the Chinese reflexive forms such as pronoun-​benren (pronoun-​self) and ziji-​benshen (self-​ body) that clearly point to the real self in lexical meaning can have a Near-​ reflexive reading. Suppose that Queen Elizabeth II and John are in front of the wax figure of Queen Elizabeth II. The Queen dislikes the wax figure depicting herself so that she kicks the wax figure several times. As a result, one of the legs of the wax figure gets broken. In this situation, John may say the sentence below to the Queen with humour, with benren and zishen—​an abbreviated form of ziji-​benshen—​referring to the statue: (30)  Nüwangi, nii   nong duan  de     ke shi ni-​benreni/​zisheni  de Queen   you  make broken DE  just be you-​self  self    DE yu      tui  a. jade-​like leg  SFP Lit. ‘Queen! What you have broken is just your own jade-​like leg.’ However, there are other situations in which ta-​benren and zishen cannot get the Near-​reflexive reading even when these reflexives are used humorously. For instance, in (31) neither ta-​benren nor zishen can refer to the duplicated self, even when they are used humorously. (31) Johni kanbuqi 



ta-​benreni/​zisheni.

John contemplate  he-​self    self ‘John contemplates himself.’

110

110 Reflexives Besides, (31) poses another problem for Lidz’s Condition R since it predicts that if a predicate is semantically reflexive, it must be lexically reflexive. In (31), although the reflexives are semantically reflexive, as they only allow the identity reading, namely the Pure-​reflexive reading, they are not lexically reflexive. I  think that the difference between (30) and (31) in reflexive interpretation lies in the semantics of the verb in question. For certain verbs expressing the meaning of physical evaporation and psychological process, it is difficult for the reflexives selected by them to get the Near-​reflexive reading. Hence, I think that there is a fundamental distinction between ta-​ziji and ziji in referring to the duplicated self. Ta-​ziji can directly refer to the duplicated self because it does not require identity whereas ziji can refer to the duplicated self only under some special situation because it requires identity (Pan 1997). I argue that the identity-​requiring reflexives like ziji can get the Near-​reflexive reading in sentences like (29) only when there is a metaphorical process that can project the real person and the wax figure into a possible world where the wax figure is personified so that both the real person and the personified wax figure share the same identity. Now, we can see that some special conditions must be satisfied in order for ziji and other identity-​requiring reflexives to obtain the Near-​reflexive reading. Lidz fails to account for these differences between ta-​ziji and identity-​requiring reflexives like ziji in their referential possibilities and constraints. Furthermore, Lidz’s criterion for identifying ziji as Near-​reflexive is also inappropriate. According to Lidz (2001b), only a sloppy interpretation is allowed in a Pure-​ reflexive sentence under Comparative Deletion in Dutch and Kannada, whereas both the strict and sloppy interpretations are allowed in Near-​reflexive sentences. Lidz claims that since ziji allows both a strict and sloppy interpretation in (32) under Comparative Deletion, it should be a Near-​reflexive. (32)  Zhangsan  bi  Lisi wei ziji  bianhu de hao. Zhangsan

than Lisi for self defend DE well

‘Zhangsan defended himself better than Lisi defended himself.’ ‘Zhangsan defended himself better than Lisi defended him.’ There are two points that should be made clear here. First, (32) is not an acceptable comparative sentence in Mandarin Chinese. An acceptable comparative sentence expressing the meaning of (32) in Mandarin Chinese should be something like (33), but (33) does not allow the sloppy reading, as indicated below. (33) Zhangsan wei ziji  bianhu bi    Lisi bianhu de  

hao.

Zhangsan for self defend  than Lisi defend DE  well *‘Zhangsan defended himself better than Lisi defended himself.’ ‘Zhangsan defended himself better than Lisi defended him.’

 111

Prominence and locality 111 Note that (33) is different from (32) in that ziji in (32) is c-​commanded by both Zhangsan and Lisi in the comparative construction, whereas ziji in (33) is c-​commanded only by Zhangsan, which may be the reason that in (32) ziji can refer to both NPs, whereas in (33) it can only refer to Zhangsan. Second, although sentences like (33) are still compatible with the Near-​reflexive interpretation of ziji according to Lidz’s criterion, I think that the fact that a sentence allows both the strict and sloppy interpretations cannot prove that a reflexive is a Near-​reflexive. If we make a similar awkward sentence like (32) in which ziji is replaced by ta-​benren, ta-​benren will also allow the strict and sloppy interpretations, but note that ta-​benren can never be used as a Near-​ reflexive in such a neutral context. (34)  ?Zhangsan bi  Lisi

wei  ta-​benren  bianhu de   hao.

Zhangsan than Lisi  for

he-​self    defend DE  well

‘Zhangsan defended himself better than Lisi defended himself.’ (sloppy reading) ‘Zhangsan defended himself better than Lisi defended him.’ (strict reading) The above fact shows that, even if the strict/​sloppy interpretation test under Comparative Deletion works in Dutch and Kannada, it does not work in Chinese since it would wrongly classify a Pure-​reflexive as a Near-​reflexive. The above discussion shows clearly that Lidz’s extension to the predicate-​ based theory of reflexives is also both too strong and too weak when applied to Chinese reflexives. It is too strong since it would wrongly rule out the licit sentences like (31) in which a semantically reflexive predicate is not lexically/​morphologically reflexive, thus violating Lidz’s Condition R.  It is also too weak since it would wrongly rule in the impossible and thus illicit Near-​ reflexive reading of reflexives in (28), (31), and (34). 3.1.4  The missing type Anagnostopoulou and Everaert (henceforth A&E) (1999) note that in R&R’s typology of reflexives a fourth type, characterized as [+SELF, +R], is predicted but not attested in natural language. According to A&E (1999), there is a theory-​internal reason why the missing type is not attested, which follows from the interaction between R&R’s Conditions A and B and their Chain Condition. Specifically, if a predicate is reflexive-​marked by a SELF anaphor, the co-​arguments of the predicate must be co-​indexed, according to R&R. When the co-​arguments of the predicate are co-​indexed, the A-​chain formed from this co-​indexation must obey the A-​chain Condition, which excludes the occurrence of an element with the feature [+R] in the tail of the chain. This is why anaphors with the feature [+SELF, +R] are missing in R&R’s typology. Nevertheless, A&E argue that anaphors with the feature

112

112 Reflexives [+SELF, +R] can still occur in natural language if they do not enter into chain formation. According to A&E, the Greek o eaftos tu is such an anaphor which consists of the definite determiner o “the,” the head noun eaftos “self,” and the possessive pronoun tu “his.” Consider the following sentence taken from A&E (1999: 105): (35)  [O Petros]j  

agapai [ton eaftoi tuj]i.

The Petros(N) loves  the self(A) his(G) ‘Petros loves himself.’ In (35) the two co-​indexed elements are the subject of the predicate and the possessor inside the object DP, a complex anaphor. Since the two co-​indexed elements occur in two different governing categories, they do not form an A-​ chain. Note that, although in (35) the SELF anaphor that reflexive-​marks the predicate is not co-​indexed with its co-​argument, the sentence is not ungrammatical. A&E argue that (35) is grammatical because an abstract incorporation occurs in it, as illustrated below: (36) [O Petros]j eaftoi-​agapai [ton ti tuj]j. In (36) the noun and the verb form a complex predicate, and as a result of this noun–​verb incorporation, the possessor is “promoted” to the status of an argument of the complex predicate, and thus the possessor and the subject become co-​arguments, satisfying R&R’s Condition A. Notice that A&E seem to suggest that the head noun eafto, by incorporating to the verb, reflexive-​ marks the predicate, and the co-​indexation of the promoted possessive pronoun and the subject makes the predicate reflexive. Hence, no violation of Condition A  or Condition B.  The crucial point is that the reflexive marker need not share the same index with the subject or object. At the first sight, it seems that the incorporation analysis solves the problem that a [+SELF, +R] anaphor might pose for R&R’s theory and can adequately explain why the missing type can exist in natural language. However, after closer examination, I find that A&E’s incorporation analysis is, in fact, implausible since it has its own problems that cannot be resolved. The first problem is that they fail to explain why the reflexive-​marker can be counter-​indexed with the subject of the predicate. According to the standard analysis, as assumed in Cole, Hermon, and Sung (1990) and Cole and Wang (1996), if a reflexive is adjoined to the predicate by head movement, it will be co-​indexed with the subject via Spec-​head agreement. However, in (36) the reflexive is counter-​indexed with the subject. How is this possible? Note that Fox (1993) has already pointed out that if a reflexive marker does not need to participate in reflexive binding, nothing can block the derivation of illicit sentences like “Youi showed myself to yourselfi” or “Jani showed myself to SEi (Dutch) .” In these sentences, the predicates are reflexive since their two

 113

Prominence and locality 113 arguments are co-​indexed, and the predicates are also reflexive-​marked by a third argument, namely myself. This may be the reason why R&R (1993: 662) relativize their Conditions A and B to an index in order to exclude such illicit structures in their system. They stipulate that an i-​reflexive marked syntactic predicate must be i-​reflexive, and an i-​reflexive semantic predicate must be i-​ reflexive marked. Hence, what is assumed in A&E is just what is banned by the analysis of R&R, since it will produce wrong results. Another problem with A&E’s analysis of (36) is that it runs against the attested case of incorporation in Chinese that the possessive pronoun must be counter-​indexed with its co-​argument (i.e., the subject of the sentence) when incorporation occurs and the possessive pronoun is promoted to the argument position. (37)  Johni zai  sheng  tade*i/​j  qi.    

at   grow   his  

  anger

‘John is being angry at him.’ In (37) qi “anger” is incorporated into the verb sheng “grow” to form a verb–​noun compound shengqi “be angry at,” and the possessive pronoun is promoted to the argument position and functions as the object of the compound sheng-​qi. In this case, the possessive pronoun must be counter-​indexed with the subject. If incorporation does not occur and the possessive pronoun is not promoted to the argument position, the possessive pronoun can be co-​ indexed with the subject of the sentence because they are not co-​arguments, as demonstrated below: (38) Johni zai  kan   tadei/​j  shu.  



at   read  his    book

‘John is reading his book.’ Hence, I  think that it is impossible for the possessive pronoun to be co-​ indexed with the subject in (36) when it is promoted to the argument position since such a co-​indexation will violate the standard Binding Condition B, as we have a pronoun bound in its local domain. Note that there is no evidence to show that the possessive pronoun in (36) can change its status and become an anaphor when it is promoted to the argument position, and I  think that such a status changing process should generally be prohibited. Since the possessive pronoun cannot be co-​indexed with its co-​argument if it is promoted to the argument position, I think that such a promotion process as indicated in (36) is impossible. Besides, there is a potential conflict in A&E’s treatment of eaftos. On the one hand, they say that the noun–​verb incorporation is possible in (36) because eaftos is a semantically defective noun, and it is the semantic defectiveness of eaftos that triggers the abstract incorporation. On the other hand, they classify

114

114 Reflexives it as referential. The question is: can semantic defectiveness coexist with referentiality? A&E do not define clearly what semantic defectiveness means, but it seems that they take particles and idiomatic nouns/​adjectives as semantically defective elements, as indicated in their note 12. However, if eaftos is grouped with these semantically defective idiomatic nouns/​adjectives, they are not referential in the normal sense. Hence, it should not be classified as referential. Notice that this is a general problem in R&R’s definition of the R properties. According to R&R (1993), an element is defined as an R element if it has a full specification of phi-​features and structural Case. Phi-​features are grouped into classes, which include person, number, and gender. An element will be defined as a [-​R] element if it lacks consistent contrast in phi-​features and structural Case. Under this analysis, the referential dependence of anaphors does not necessarily mean that it cannot refer independently, but means that it lacks contrast in phi-​features and Case. If we follow R&R’s definition, we would find it difficult to classify elements which are fully specified for phi-​ features and Case but are semantically defective. The Greek eaftos is such an element. A&E adopt R&R’s definition of R and conclude that eaftos in Greek is an R element since it is marked for gender and fully inflected for number and case. But, if eaftos is defined as an R element, how can we account for its semantic defectiveness? Obviously this way of defining referential independence does not work for Chinese reflexives. Simplex reflexives like ziji, benren, benshen, and zishen do not exhibit any contrast in phi-​features or Case, and complex reflexives like ta-​ziji exhibit a contrast only in person and number, but not in gender3 or Case. If we use R&R’s criteria to classify these reflexives, two possibilities will arise: (i) all the simplex reflexives are classified as [-​R] elements, but the complex reflexives like ta-​ziji are classified as [+R] since they exhibit contrast in person and number; and (ii) all of them are classified as [-​R] elements since, although the complex reflexives show contrast in person and number, they do not show such a contrast in gender or Case, and hence, they do not exhibit a full contrast in phi-​features. But, note if we apply R&R’s criteria to the pronouns in Chinese, they should also be classified as [-​R] elements since, although pronouns in Chinese exhibit a contrast in person and number, they do not exhibit a contrast in gender or Case, thus showing no full contrast in phi-​features. Clearly, there is a problem in R&R’s phi-​feature based definition of R, as Chinese third person singular pronoun ta can apparently be used independently. I think that if an element is an R element, it should be able to refer independently. That is, it can be used deictically and need not be dependent on an antecedent for its interpretation within the sentence domain. In this case, our definition of R is consistent with Chomsky’s (1981) definition of pronouns and names in the sense that they need not be bound within their local domain. If we define referential independence as the ability to refer independently, the Chinese reflexive pronoun-​benren “pronoun-​self ” can be characterized as a [+SELF, +R] element. For instance, the compound form ta-​benren “he-​self ” given in (31), repeated as (39), can refer independently, besides being bound to

 115

Prominence and locality 115 the co-​argument of its own predicate. It can refer to some NP other than the local subject, for example, some person under discussion or mentioned in the previous discourse, besides being anaphoric to the local subject. (39)  Johni kanbuqi    ta-​benren?i/​j. John contemplate

himself

‘John contemplates himself.’ Sentences like (39) indicate that ta-​benren is really a [+SELF, +R] anaphor, which cannot be explained away by the incorporation analysis. First, if the bare reflexive benren “self ” is incorporated into the predicate, it will reflexive-​ mark the predicate. Otherwise, such an incorporation process is unmotivated. However, benren by itself is not qualified as a reflexive-​marker since it must obviate its co-​argument in reference, cf. (26). Second, benren, different from the Greek anaphor eaftos, is not semantically defective and is thus not allowed to be incorporated into the verb. 3.1.5  Further discussion: Complex predicates and reflexive-​marking In this section we will see if it is possible to solve the problems presented by Chinese reflexives while still maintaining R&R’s (1993) basic assumption that binding is a process of reflexive-​marking of the predicate. It seems that the above problem can be possibly solved if we assume that different clauses in a Chinese sentence can optionally form a complex predicate. We may assume that complex predicate formation is constrained by the finiteness of clauses. Different finite clauses cannot be absorbed into one predicate whereas different nonfinite clauses can. We may further assume that complex predicate formation is sensitive only to predicates which exhibit overt morphological marking of finiteness. Given that Chinese does not have overt morphological marking of finiteness or nonfiniteness (cf. Hu, Pan, and Xu 2001), it is conceptually possible that complex predicate formation in Chinese is not constrained by the above finiteness condition. Hence, Chinese exercises more freedom in forming a complex predicate than those languages which exhibit overt morphological marking of finiteness. If this is true, then in (40) we may treat renwei Bill xihuan “thinks that Bill likes” as a complex predicate which takes two arguments:  the matrix subject and the reflexive. Under this kind of analysis, ziji is treated as a SELF anaphor instead of a SE anaphor. Since complex predicate formation is optional in Chinese, both local and nonlocal binding of ziji in (40) can be accounted for. (40) Johni renwei Billj  xihuan  zijii/​j. John think  Bill   like 

  self

‘Johni thinks that Bill j likes himi/​himselfj.’

116

116 Reflexives Although the complex predicate account can explain why ziji can be both locally and nonlocally bound when occupying an argument position, it cannot explain the binding properties of the following sentences: (41)  Johni zhidao woj bu xihuan  ziji*i/​j. John know  I  not like     self ‘John knows that I do not like ​myself.’ (42) Woi  zhidao  Johnj  bu  xihuan  zijii/​j. I 

  know   John   not  like    self

‘I know that John does not like himself/​me.’ (43) a. Johni de jiaoao hai-​le 



(ta-​)zijii.

  John DE pride hurt-​ASP  himself   ‘John’s pride hurt him.’ b. Johni shuo [zhejianshi]j hai-​le 

   (ta-​)zijii/​*j.

  John say   this-​matter hurt-​ASP  

himself

‘John said that this matter hurt him.’

(44) Johni renwei Bill j  xihuan ta-​ziji*i/​j. John think  Bill   like    himself ‘Johni thinks that Bill j likes himself*i/​j.’ In (41) ziji cannot be bound by the matrix subject, and it is unclear why the embedded clause and the matrix predicate cannot form a complex predicate when the subject of the embedded clause is a pronoun of different phi-​features. We may assume that complex predicates must be formed of clauses with subjects of the same phi-​features. However, even if this is possible, it cannot explain why the reflexive can be bound by the matrix subject in (42), though the subject of the embedded clause has different phi-​features from the matrix subject. Another problem with the hypothetical analysis is that it still cannot explain why the sentences in (43) are grammatical, though bluntly violating Binding Condition A, given that both ta-​ziji and ziji are SELF anaphors and must obey Binding Condition A. (44) further shows that the complex predicate analysis might not be valid since in (44) the reflexive cannot be bound by the matrix subject, though the complex predicate analysis assumes that it can. From the discussion of the above sections we can see that the predicate-​ based binding theory, as proposed by R&R and Lidz, and their relevant typology of reflexives cannot adequately characterize the properties of reflexives in Chinese, when extended to Chinese. It seems that reflexives in natural language are not of a homogeneous class and thus cannot be adequately characterized simply by the two properties given in R&R (1993) (i.e., the

 117

Prominence and locality 117 property Reflexivizing function and the property R(referential independence)). Although the reflexives in natural language may have one of these two properties, they also have other properties that cannot be covered by these two properties. I think that the predicate-​based binding theory need be further improved if it is intended to be a general theory of reflexives in Universal Grammar, and no typology of reflexives can present a complete picture of reflexives in Universal Grammar if Chinese reflexives are not adequately covered and accounted for.

3.2  Characterizing reflexives in Chinese: Deriving referential dependency and reflexivity from primitive features Chomsky (1981) uses two binary features, [±anaphor] and [±pronominal], to characterize the properties of nominal expressions. R&R (1993) use [±Reflexivizing function] and [±R(eferntial independence)] to characterize the properties of anaphoric expressions. Although their feature characterization of anaphoric expressions is quite useful in establishing a preliminary typology of anaphoric expressions across languages, it fails to capture the binding properties of anaphoric expressions in Chinese. Simply making a distinction among nominal expressions with features like [±anaphor, ±pronominal], besides failing to predict the long-​distance (LD) binding property of the Chinese bare reflexive ziji, has nothing to say about the blocking effect exhibited by ziji. I have shown that R&R’s (1993) feature characterization of anaphoric expressions also fails in this aspect. Besides failing to explain why and when ziji would exhibit the blocking effect, their characterization also cannot predict the varieties of anaphoric expressions in Chinese. Moreover, their feature characterization of anaphoric expressions is not primitive enough and thus fails to answer the question what properties of nominal expressions make themselves anaphors or pronominals and why some reflexives are reflexivizers, but others are not. In the following discussion, I  will give my own feature characterization of Chinese anaphoric expressions and show that some of the binding properties of Chinese anaphoric expressions can be derived from it. If we view nominal expressions in general, we can make a first distinction between them with the binary feature [±dependency]. It is well-​known that reflexives are referentially dependent, but r-​expressions and pronouns are not. Assuming that it is true that nominal expressions in natural language can be classified by the feature [±dependency], the next question to ask is why some expressions are referentially dependent but some are not. It seems that there must be some syntactic primitives that are responsible for [±dependency]. In accounting for the difference in referential behaviour between the bare reflexive and the compound reflexive in Chinese, Huang and Tang (1991) point out that the bare reflexive is more “anaphoric” than the compound reflexive in the sense that it has neither references nor phi-​features. It is a “double anaphor” since it needs to obtain not only phi-​features but also

118

118 Reflexives references from its antecedent. This view is very insightful since it does reveal the crucial difference between the bare reflexive and the compound reflexive. Obviously, it is less ad hoc to tie the referential behaviour of the reflexives to their primitive properties. In later discussion, I  will further develop this feature-​driven account of reflexives and show that the blocking effect of the Chinese bare reflexive can in effect be derived from it. If we characterize the anaphoric expressions in Chinese with these more primitive features, [±phi-​features] (henceforth [±phi]) and [±reference] (henceforth [±ref]), we can obtain the following result: (45)  The Characterization of Chinese Anaphoric Expressions             ziji    pronoun+ziji  pronoun phi(-​feature)   -​        +          + ref(erence)    -​         -​          + The above feature characterization shows that pronouns have phi-​features and referential features. According to Huang and Tang (1991: 274), all NPs, except quantifiers, have both inherent phi-​features and inherent referential features. A pronoun has inherent phi-​features and may have independent reference or obtain reference from its antecedent. If lacking either phi-​features or reference can result in the dependency property of the nominal elements, then it is reasonable to analyse the bare reflexive ziji as a “double anaphor.” Since the Chinese compound reflexive lacks only a referential feature, it can be assumed that whenever it obtains this feature from a proper antecedent (I will explain later what a proper antecedent is), it will not continue its antecedent searching process since further searching is unmotivated. If this is true, then the relative local nature of the compound reflexive binding may result from the fact that it lacks only one feature. As for the bare reflexive, it is possible that there are two searching processes involved: one for phi-​features and the other for the referential features. I will soon show that this is the case. Another question related to the above feature characterization is whether reflexivity can be derived from it. At first sight, it seems that reflexivity could be derived from the lack of reference. But in fact this is not true. Consider the empty pronominals in Chinese. According to Chomsky (1981: 20, 60, 322), PRO has phi-​features such as person, number, and gender but lacks inherent reference. Chomsky (1981: 65) claims that the overt pronoun is free in reference, but PRO is not since its reference is determined by its co-​indexed antecedent, and PRO has to be interpreted under control. Let us simply follow Chomsky’s assumption without further argument that PRO has phi-​features but lacks reference. Huang (1991) suggests that both PRO and pro fall under the same category Pro. Following Huang, I simply label both PRO and pro as Pro. Although I do not make a distinction between PRO and pro, I still maintain the distinction between the obligatorily controlled Pro and the optionally controlled Pro. I claim that only the former has no reference. As for the

 119

Prominence and locality 119 latter, it can be either optionally controlled if it has an antecedent within the sentence or used deictically when referring to a nonlinguistic antecedent, as claimed by M.-​D. Cole (2000). When Pro is used deictically, it has reference. If we use the above feature matrix to characterize the controlled Pro and the uncontrolled Pro, the former should be characterized as [+phi, -​ref] ([phi] means phi-​feature, and [ref] means referential feature), and the latter should be characterized as [+phi, +ref]. Now, consider the following sentences: (46)  a. Lisii dasuan [Proi mingtian  lai].   Lisi plan   



tomorrow come

  ‘Lisi plans to come tomorrow.’ b. Lisii shuo Proi/​j mingtian   lai.   Lisi say      tomorrow  come   ‘Lisi said that he will come tomorrow.’ c. Lisii pa    laoshij   hui piping Proi/​*j/​k.   Lisi afraid teacher will criticize   ‘Lisi is afraid that the teacher will criticize him.’ In (46a) Pro has no reference before entering a dependency relation with an antecedent since it is an obligatorily controlled Pro. The Pro’s in (46b) and (46c) are optionally controlled. In (46b), Pro can depend on either the matrix subject or a sentence external antecedent for these referential features. (46c) shows that, although Pro can be identified by either the matrix subject or a sentence external antecedent, it cannot depend on its co-​argument for reference. This fact demonstrates that in this aspect, Pro exhibits different binding properties from reflexives, though Pro is sometimes termed as the “zero anaphor” in the literature. I claim that the distinction between reflexives and Pro, be it obligatorily controlled or optionally controlled, does not lie in whether they have the relevant dependency features (i.e., the referential feature or the phi-​feature) but in whether they have the SELF feature. I will show that a distinction must be made between dependency and reflexivity as this distinction is necessary for an adequate characterization of anaphoric expressions. I  argue that reflexivity requires the presence of the SELF feature, though dependency does not necessarily needs it. I think that SELF can be best understood when it is set in contrast with OTHER. Hence, [+SELF] means [-​OTHER]. But it should be pointed out that [-​OTHER] does not necessarily mean [+SELF], as will be shown soon. In Safir (1996), OTHER is a semantic primitive, but SELF is not. Safir makes a distinction between SELF and SAME and points out that in Germanic languages, the SELF morpheme is used to form reflexives, while in Romance languages, it is the SAME morpheme, for instance, même “same” in lui-​même “himself ” in French, that is used. He further claims that, although both of them participate in identity

120

120 Reflexives relations, they differ in the sense that the former can be used to denote an individual, whereas the latter cannot. Since SAME is irrelevant to the definition of Chinese reflexives, we will not discuss it here. Under Safir’s (1996) analysis, SELF is an instance of MET applied to the body part self. The following is Safir’s (1996: 548) definition of MET: (47)  MET is a two-​place metonymic identity relation parasitic on the argument of BODYPART, whereby the part argument stands for or is identified with an unsaturated whole argument. BODYPART is defined as below (Safir 1996: 549): (48) BODYPART is a two-​place relation between a part and the whole inalienably associated with it. The argument corresponding to the whole may be unsaturated. The body part in question inherently saturated the part argument. Under Safir’s analysis, the anaphoric SELF is a metonymic anaphor, which represents a part of an individual as being identical to the whole of that individual, with the result that the atom SELF is used as a two-​place identity relation. Safir treats the second argument of SELF as self, which corresponds to the “part” argument. Hence, according to Safir, the sentence John loves himself would mean that John loves the part of him that is his “self,” as shown below: (49) a. John loves himself. b. Johni loves [xi SELF] c. where [xi SELF] means xi –​MET-​selfi. According to Safir, the interpretation of (49b) is that John loves the x represented by self and MET requires that self stands for whatever is the antecedent of x, namely John. As a result, the reflexive reading for love mediated by MET is derived. Haiman (1985, 1995) also argues that “self ” is an instantiation of the body, and he claims that the use of reflexives suggests a (subject) mind–​(object) body dualism. If we take into consideration Chinese contrastive reflexives benren “this person” and benshen “this body,” we can see why their analyses are reasonable. If we adopt Safir’s (1996) analysis, then in the representation for benren, the second argument of SELF is realized as “person,” while in the representation for benshen, the second argument of SELF is realized as “body.” These facts show that Safir’s characterization of reflexives really can capture their semantic properties. However, this characterization, though plausible and insightful, fails to explain why benren and benshen cannot be used as a reflexivizer, but the bare reflexive ziji can. The only difference between ziji and benren and benshen is that in the representation for ziji, the second argument (i.e., the “part” argument) is realized as

 121

Prominence and locality 121 “self ” rather than the person or the body. This fact shows that in reflexivization what matters is not MET or BODYPART, but “self ” in Safir’s terms. Let us represent Safir’s “self ” as SELF. Note that SELF is different from the Self defined by Sells (1987). SELF can be understood literally as the SELF morpheme in reflexives. (Note that Safir (1996) also occasionally uses SELF to mean the SELF morpheme.) The crucial point is that SELF is used in contrast with OTHER. If we include SELF and OTHER in anaphoric classification, we can characterize Chinese anaphoric expressions as below: (50)  The Characterization of Chinese Anaphoric Expressions                 phi(-​features)   ref(erence)   SELF  OTHER uncontrolled Pro   

  +            +        -​        +

controlled Pro     

  +            -​     

ziji                   -​   



pronoun+ziji          +    benren 

     

benshen 



ziji-​benshen 

         



    -​     

            -​   

ta-​benren 





    +          -​     



  -​        +

   -​         + 

      -​

    -​         + 

     -​

  +        -​   

    +        -​ 

   -​



   -​

    +        -​        +        + 

   -​  

  -​

In Chomsky (1981), PRO (i.e., the controlled Pro) is characterized as [+pronominal, +anaphor]. Under the present analysis, its [+anaphor] feature is derived from its lack of reference, and its [+pronominal] feature is derived from the fact that it has a [+OTHER] feature but lacks the [+SELF] feature. Ziji derives its dependency from its lack of both phi-​features and referential features, and its reflexivity from the fact that it has a SELF feature but lacks the OTHER feature. Compound reflexives like ta-​ziji differ from ziji only in that it has phi-​features. If dependency results from the lack of phi-​features or referential features, it can be reasoned that lacking one of these features will result in the local dependency whereas lacking both of these features may result in LD dependency. This is exactly what is suggested in Huang and Tang (1991). Although lacking phi-​features or referential features may result in dependency, a nominal expression cannot be used reflexively if it does not have the SELF feature. It is the SELF feature that results in the co-​reference of the co-​arguments of a predicate. If this is true, then a plausible assumption is that if a nominal expression has the SELF feature but lacks only one of the dependency features (i.e., either the phi-​feature or the referential feature), it does not tend to be LD bound if it can obtain its reference locally. This assumption not only conforms to the Principle of Economy but is also supported by the facts. It is well-​known that the Chinese compound reflexive like ta-​ziji prefers to be locally bound whenever possible. It can be LD bound

122

122 Reflexives only if it cannot find an ideal antecedent locally (see Pan 1995, 1997, 1998 for detailed discussion). Since we will discuss ta-​ziji later, we will not go into the details of it here. Another Chinese reflexive that obeys a relatively restrictive locality condition is ziji-​benshen. Yu (2000) notes that in the following sentence ziji-​benshen can be bound to the embedded subject but cannot be LD bound to the matrix subject across the embedded subject: (51)  Zhangsani shuo [Lisij  zhi guanxin ziji-​benshen*i/​j]. Zhangsan say   Lisi

only care    self-​self

‘Zhangsan said that Lisi only cared about himself.’ (Yu 2000: 51) However, if the embedded subject is not a possible antecedent, LD binding of ziji-​benshen to the matrix subject across the local one is possible, as shown below: (52) Ta zhidao [zhe yinggai guai

ziji-​benshen].

he know  this should blame self-​self ‘He knew that this should be blamed on him himself.’ (Yu 2000: 18) As a first approximation, the present analysis can assume that the relative local nature of ziji-​benshen is derived from the fact that it lacks only one kind of feature: the phi-​feature. Whenever this feature is obtained from a prominent antecedent, its antecedent searching process will stop. In this aspect, ziji is different from ta-​ziji or ziji-​benshen. Since ziji lacks both the phi-​feature and the referential feature, it needs two feature-​searching processes: one for the phi-​feature, and the other for the referential feature. Later on, I will show that the LD binding property of ziji can be derived from its lack of these two features. Benren and benshen are two bare contrastive reflexives under Pan’s (1995, 1997) analysis. Benren means this person, and benshen means this body. Although these two contrastive reflexives do not have the SELF feature, they do not point to the OTHER. This shows that [-​SELF] does not necessarily mean [+OTHER]. Take benren for example. Pan (1995, 1997) notes that, although benren usually refers to the speaker of the utterance or the author of the article, it can be used anaphorically and have antecedents across clauses or sentences, as shown below: (53) a. Ta shi wo de tongshi,  

jiating chushen 

  buxiang, benren

  he be  I   DE colleague family background unknown self   chengfen  xuesheng.   class-​status student

 123

Prominence and locality 123    

‘He is my colleague. His family background is not clear. His own class status is student.’ (Pan 1997: 187)

b. Nashuiren i chu  benreni mianshui’e    wai hai keyi xiangshou   Tax-​payer except self    exempt-​amount out also can  enjoy   beifuyangren de mianshui’e.   foster-​child DE exempt-​amount  

‘Besides the exempt amount of their own, tax payers can also enjoy the exempt amount for the foster child.’

  (Pan 1997: 188) Pan (1995, 1997) further notes that benren cannot refer to any c-​ commanding NP, as exemplified below: (54) Johni gaosu Markj Billk bu xihuan benren*i/​*j/​*k de zhaopian. John tell     Mark Bill not like  

self 

    DE picture

‘John tells Mark that Bill does not like my picture.’ The anaphoric property of benren is expected under the present analysis since it can be derived from the [-​OTHER] feature, but its anti-​c-​commanding property is unexpected. Note that lacking the SELF feature does not mean that it cannot refer to a c-​commanding NP. Pronouns also lack the SELF feature, but they can refer to c-​commanding NPs if the c-​commanding NPs are not their co-​arguments, as shown below. (55) Zhangsani shuo Lisi xihuan tai. Zhangsan say   Lisi like 

him

‘Zhangsan said that Lisi liked him.’ The anti-​c-​commanding property of benren must be derived from some other features. But none of the features given in (50) can derive this property. Lasnik (1989) (see also Thráinsson 1991) has argued quite convincingly that the binary feature [±r-​expression] (henceforth [±R]) is needed in the characterization of the so-​called pronominal epithets (the bastard, etc.) since, besides having some pronominal properties, it also has certain things in common with names (John, etc.). Assume that the feature [±R] is really needed, as suggested by Lasnik (1989) and Thráinsson (1991). Then, the next question to ask is whether it is redundant to posit both a [ref] feature and an [R]‌feature in the present system and whether these two features can be reduced to one. In fact, these two features are independently needed. The [ref] feature means “independent reference,” whereas the [R] feature means that the relevant element containing it is an R-​expression. Thráinsson (1991)

124

124 Reflexives argues that both the independent reference feature and the [R] feature are needed in the classification of anaphoric expressions since, although an NP having the [R] feature may also have the [ref] feature, an NP having the [ref] feature may not have the [R] feature. For instance, a pronoun has the [ref] feature, but does not have the [R] feature. In Thráinsson’s typology of NPs, NPs are first classified by the [±ref] feature, and then further classified by the [±R] feature and other features. I think that NPs should be first classified by the feature [±R], and then further classified by other features. In Chomsky’s (1981) typology of NPs, r-​ expressions are characterized as [-​ anaphor, -​ pronominal]. If Chomsky’s (1981) typology is adopted, then quantifiers and wh-​expressions should also be classified as r-​expressions with the feature [-​ anaphor, -​pronominal]. However, it is well-​known that quantifiers do not have reference. And according to the standard analysis (e.g., Chomsky 1981), wh-​ expressions are also nonreferring expressions since they do not pick up a specific individual or entity in a given domain of discourse. That is why they are often treated as quantifiers. If quantifiers and wh-​expressions do not have reference, they should not be classified as r-​expressions. Since they are also not anaphors or pronominals, they are not included in Chomsky’s typology of NPs. This shows that Chomsky’s typology may not cover all possible NPs in natural language. If we use the feature [±R] to make a first distinction between NPs, then names have the [+R] feature, and all other nominal expressions, including anaphors, pronominal, quantifiers, and wh-​words, have the [-​R] feature. Assuming that NPs can be further classified by [±SELF] and [±OTHER] and that [±ref] is relevant only to the [-​R] NPs, but not to the [+R] NPs, then we can see the necessity of using both [±ref] and [±R] in our system. If this analysis is on the right track, we can reformulate our feature characterization of Chinese anaphoric expressions as below. In the following feature characterization, the [±phi] feature and the [±ref] feature apply only to the [-​R] NPs, but not to the [+R] NPs. (56)  The Characterization of Chinese Anaphoric Expressions A.                 SELF      OTHER      R uncontrolled Pro

-​

+

-​

controlled Pro

-​

+

-​

ziji

+

-​

-​

pronoun+ziji

+

-​

-​

benren

-​

-​

+

benshen

-​

-​

+

ta-​benren

-​

-​

-​

ziji-​benshen

+

-​

-​

 125

Prominence and locality 125 B.                   phi(-​features)       ref(erential) uncontrolled Pro

+

+

controlled Pro

+

-​

ziji

-​

-​

pronoun+ziji

+

-​

ta-​benren

+

+

ziji-​benshen

-​

+

The above feature characterization can explain why benren cannot be bound to a c-​commanding antecedent. This is because it has the feature [+R]. It is well attested that r-​expressions always reject c-​commanding antecedents. Notice that, as mentioned in our previous discussion, benren also has a [-​ OTHER] feature. That is why it can be used anaphorically. Pan (1995, 1997) notes that the compound contrastive reflexive ta-​benren can refer to a c-​commanding antecedent, as shown below: (57) Johni shuo Billj kanbuqi    ta-​benreni/​?j/​k. John say   Bill contemplate he-​self ‘John says that Bill contemplates him/​himself.’ According to Pan (1997), in (57) the preferred antecedent for ta-​benren is the matrix subject, and the less preferred one is the embedded subject. Note that in (57) ta-​benren can also refer to some sentence external antecedent. Under our analysis, ta-​benren does not prefer to be bound to its local subject because it is not a strong reflexivizer. It is not a strong reflexivizer due to its lack of a SELF feature, though it has a [-​OTHER] feature. I think that only those nominal elements that have the [+SELF] feature can function as strong reflexivizers. In (56), benshen “this body” has the same feature characterization as benren “this person.” However, there is an important difference between benshen and benren. Pan (1997: 202) notes that benshen differs from benren in that it cannot be used as an agent, as shown below. (58) *John/​ta benshen da-​le  Bill yixia’r. John/​he self   

hit-​ASP Bill once

‘John/​He himself hit Bill once.’ Although this property of benshen is not predicted by the above feature characterization, it can be worked out independently from its lexical meaning. Since the lexical meaning of benshen is this body, and what it refers to is the

126

126 Reflexives body rather than the person, it does not seem to have a strong animacy feature. Assuming that agentivity requires the relevant element to be ranked high in the animacy hierarchy or have a strong animacy feature, the reason why benshen cannot be used as an agent is explained. Note that the above feature characterization of anaphoric expressions can also be used to characterize the SELF-​anaphor and the SE-​anaphor defined by R&R (1993). Both the SELF-​anaphor like himself in English and the SELF-​ anaphor like zichzelf in Dutch have the features [+SELF, -​OTHER, -​R]. Their difference lies in that the former is specified as [+phi, -​ref], whereas the latter, just like ziji-​benshen in Chinese, is specified as [-​phi, +ref]. The reason why zichzelf has a [+ref] feature is because the reference of zelf in zichzelf is defined by zich and thus points to the self of zich. Since himself and zichzelf both have the [+SELF] feature but only lack one of the dependency features (i.e., they lack either phi-​feature or the referential feature), they are predicted to be reflexivizers. The features that the SE-​anaphor has must be [-​SELF, -​OTHER, -​R] and [-​phi, -​ref]. The SE-​anaphor cannot be a reflexivizer because it does not have the [+SELF] feature.

3.3  The Chinese bare reflexive The Chinese bare reflexive ziji has attracted the attention of many theoretical linguists not only because it can access an NP outside its Governing Category (GC), and thus demonstrates some peculiar properties that are not captured by Chomsky’s (1981) Binding Condition A (BCA), but also because it exhibits the blocking effect, and the blocking effect is asymmetrical, as shown in Pan (1995, 1997, 2001). To account for the peculiar properties of Chinese reflexives, different linguists have proposed different analyses which can be divided into two general lines of enquiry:  one, syntactic, and the other, nonsyntactic. 3.3.1  Syntactic approaches The syntactic approach attempts to show that LD binding of Chinese reflexives is basically a syntactic phenomenon and can be dealt with under BCA with some revision. One analysis proposes to parameterize the notion of GC for different languages in the spirit of Huang (1983), Yang (1983), and Manzini and Wexler (1987). Another analysis, represented by Wang and Stillings (1984) (see also Mohanan 1982), suggests that ziji represents a new type of NP, called “anaphoric pronoun,” the interpretation of which is regulated by a new binding principle. The third approach, based largely on Pica (1987), is to assume that the long-​distance binding of ziji is a result of covert cyclic movement at LF. Although in the literature there is disagreement about the details of the LF movement of ziji, these analyses all agree that the apparently unbounded dependencies between the antecedent and the reflexive are covertly bounded in fact and are thus local in nature. In Huang and Tang

 127

Prominence and locality 127 (1991), the LF movement is a A'-​movement, realized through IP-​adjunction, similar to the process of quantifier raising. In Battistella (1989), Cole et al. (1990), Cole and Sung (1994) as well as Cole and Wang (1996), it is a head movement, with ziji moving to the I (or AGR) of its own clause and optionally moving (via C) to the next clause up. The fourth analysis, proposed by Progovaç (1992, 1993), is a nonmovement variant of the LF movement analysis, called relativized subject (RS) approach (adopted and revised in Tang 1994), and it assigns X° (namely AGR) as the SUBJECT for X° reflexives (e.g., ziji) but assigns XP as the SUBJECT for XP reflexives (e.g., ta-​ziji “he-​ self ”). Tang and Gu (1998) also use the notion of subject to account for the binding properties of ziji. Following Bower’s (1993) Predication Theory, they claim that the binding of ziji is regulated by the notion of the primary subject and the secondary subject in Bower’s terms. In the spirit of Baker (1995), Xue, Pollard, and Sag (1994) and Pollard and Xue (1998) argue for the separation of syntactic binding from discourse prominence and propose an HPSG (Head-driven Phrase Structure Grammar) account of ziji that posits a fourth NP category called Z-​pronoun of which Chinese ziji is one, and it is subject to a new binding condition, in the same spirit as Wang and Stillings (1984). Cole, Hermon, and Lee (2001) also argue for the separation of syntactic binding from discourse prominence, while Huang and Liu (2001) suggest that the dividing line for the local ziji and the LD ziji be the traditional notion of Governing Category (GC) and that the blocking effect exhibited by ziji be accounted for by a theory of perspectivity based on Kuno (1972). 3.3.2  Nonsyntactic approaches The nonsyntactic approach claims that LD binding of Mandarin reflexives should be accounted for by nonsyntactic factors. Chou (1992) argues that LD reflexives are not subject to syntactic binding but to some animacy and thematic conditions. Xu (1993, 1994) shows that the blocking effect is not consistently observed in ziji binding. He argues that the antecedent of ziji is constrained by subjects or the thematic hierarchy. He also recognizes the interaction between prominence and locality in the interpretation of Mandarin ziji (Xu 1999). Others follow a discourse-​pragmatic approach to account for the binding properties of Mandarin reflexives and claim that the antecedent of Mandarin reflexives is determined by discourse factors such as logophoricity (Y. Huang 1994; Maling 1984; Reinhart and Reuland 1991, 1993; Yu 1991; Zribi-​Hertz 1989), perspectivity (Iida 1992; Kuno 1987; N.-​C. Li 1991; Sells 1987; Zubin et  al. 1990), emphasis, or intensive pronouns (Baker 1995). Following this line of research, Yu (1991) and Y.  Huang (1994) claim that Chinese ziji is a logophor and must thus be constrained by logophoricity, and Li (1991) argues that the interpretation of unbound and LD bound ziji should be determined by perspectivity, while Chen (1992) uses TOPICALITY and Sells’ PIVOT notion to account for LD binding of ziji.

128

128 Reflexives Following the spirit of Baker (1995), Pan (1995, 1997, 2001) proposes a three-​way partition of Chinese reflexives: locally bound reflexives should be distinguished from nonlocally bound ones which are further divided into de se anaphor ziji and contrastive reflexives like benren and benshen. He claims that discourse, specifically self-​ascription and discourse prominence, plays an essential role in the interpretation of Chinese reflexives. Pan claims that LD bound ziji is a de se anaphor and should be constrained by self-​ascription. Pan (1995, 1997, 2001) argues that the blocking effect is asymmetrical, and the crucial factor in inducing the blocking effect is not the local subject or the unlike person feature conflict, as suggested in the literature (Huang and Tang 1991; Xue, Pollard, and Sag 1994). Rather it is the asymmetry between first/​second and third person noun phrases that plays a crucial role in the blocking effect, and syntactic functions other than subject can also induce the blocking effect. Pan (1998) further shows that noncontrastive compound reflexives like ta-​ziji do not need a c-​commanding or subcommanding antecedent, can have LD antecedents, and also exhibit the blocking effect, which casts serious doubt on Pica’s (1987) widely accepted generalization that only morphologically simplex reflexives not compound reflexives can be LD bound. Since I  have already reviewed both the syntactic approaches and nonsyntactic approaches elsewhere (cf. Hu 1998b; Hu and Pan 2002) and a detailed and insightful review of the previous studies on Chinese reflexives can be easily accessed from Pan (1995, 1997, 2001), it is deemed unnecessary to give another review of the relevant literature if new comments and criticism cannot be offered. In what follows, after a brief discussion on the binding properties of ziji, I will only review two studies on ziji done by Pan (2001), Huang and Liu (2001), and then propose my own analysis. Before starting the discussion on ziji, it is necessary to point out that, although in the last 30 years or so an increasing volume of research has been conducted on the bare reflexive ziji, a unified account of both the compound reflexive and the bare reflexive in Chinese is still lacking. Besides, how to give a more satisfactory account of the blocking effect exhibited by ziji is still a great challenge to theoretical linguistics. A satisfactory account should explain not only why ziji exhibits the blocking effect but also why sometimes it is strong and sometimes it is weak. Most of the analyses treat the blocking effect as a hard inviolable constraint, but in fact it is not since it is not a grammatical constraint according to my analysis. In the following discussion, I will show that the blocking effect, in fact, can be reduced since the blocking effect is derived from the prominence of NPs, and it is reasonable that the blocking effect would be weakened if the prominence of the relevant NP is reduced or if another NP more prominent than it comes into the picture. 3.3.3  The binding properties of ziji The following examples show that the bare reflexive ziji (i) can be LD bound, (ii) can be bound only to a subject, (iii) can be bound to a subcommanding

 129

Prominence and locality 129 NP, and (iv) cannot be LD bound to an antecedent across an intervening NP with different person features (C.-​T. J. Huang 1982b; Y.-​H Huang 1984; Tang 1989; Wang and Stillings 1984). (59)  a. Zhangsani renwei Lisij xihuan zijii/​j.   Zhangsan think  Lisi like 

  self

  ‘Zhangsani thinks that Lisij likes himi/himselfj.’ b. Zhangsani mei-​you gaosu Lisij zijii/​*j de fenshu.   Zhangsan not-​have tell   Lisi self  DE grade   ‘Zhangsan did not tell Lisi his own grade.’ c. Lisii zhidao nij xihuan ziji*i/​j.   Lisi know you like 

self

  ‘Lisi knows that you like yourself.’ d. [Zhangsani de jiao’ao]j hai-​le    zijii/​*j.   Zhangsan DE pride 

hurt-​ASP self

  ‘Zhangsan’s pride hurt him.’ (59a) shows that ziji can be bound either locally or nonlocally. (59b) illustrates the subject-​orientation property of ziji since it can be bound only to a subject. (59c) shows that LD binding of ziji to the matrix subject is blocked by an intervening NP with different person feature from the matrix subject. (59d) shows that ziji can be bound to a subcommanding subject, with the definition of subcommanding given below (Tang 1989): (60) β subcommands α iff β is contained in an NP that c-​commands α or that subcommands α, and any argument containing β is in subject position. Although a local subcommander can bind ziji, a nonlocal one cannot bind it, as observed by Huang and Tang (1991). (61) Zhangsani de   xin  biaoshi [Lisij hai-​le    ziji*i/​j]. Zhangsan DE  letter indicate Lisi hurt-​ASP self ‘Zhangsan’s letter indicates that Lisi hurt himself.’ Huang and Tang (1991) also notice that the possible blockers of LD ziji include not only c-​commanding subjects but also subcommanding NPs, as shown below: (62) a. Zhangsani shuo woj de   jiao’ao hai-​ le    ziji*i/​j.   Zhangsan say  I  DE pride   hurt-​ASP self   ‘Zhangsan said that my pride hurt myself.’

130

130 Reflexives b. Zhangsani shuo [[nij zheyang zuo] dui ziji*i/​j buli].   Zhangsan say  

you thus    do  to self  not-​beneficial

  ‘Zhangsan said that your doing this will do yourself no good.’ Li (1993) notes that a subcommander with different person feature can induce blocking effect even though it is not a possible antecedent for ziji when it functions as the possessor of a [+Human] NP. (63) Lisii yiwei woj de  xueshengk  bu   xihuan ziji*i/​*j/​k. Lisi think  I  DE student 

  not like    self

‘Lisi thinks that my student does not like himself.’ Furthermore, Xue et  al. (1994) find that direct objects and obliques can also induce blocking effect and that blockers may not be potential binders of ziji, contra Huang and Tang’s (1991) claim, as shown below: (64) a. Zhangsani gaosu woj  Lisik  hen   ziji*i/​*j/​k.   Zhangsan tell  

me   Lisi   hate  self

  ‘Zhangsan told me that Lisi hates himself.’ b. Zhangsani dui woj  shuo Lisik chang  piping     Zhangsan to  me 

say  Lisi often  

ziji*i/​*j/​k.

criticize  self

  ‘Zhangsan told me that Lisi often criticized himself.’ c. Zhangsani cong nij  nar tingshuo  Lisik  chang piping  ziji*i/​*j/​k.   Zhangsan from you there hear-​say

Lisi   often criticize self

  ‘Zhangsan heard from you that Lisi often criticized himself.’   (Xue et al. 1994: 437) (65) a. Zhangsani zhidao woj de   xin  biaoming  Lisik hai-​le    ziji*i/​*j/​k.   Zhangsan know  I  DE  letter show-​clear Lisi harm-​ASP self   ‘Zhangsan knows that my letter makes it clear that Lisi harmed himself.’ b. Zhangsani tingshuo nij  

de wenzhang  jielu  Lisik hen  ziji*i/​*j/​k

  Zhangsan hear-​say you  de    taitai. DE  wife

DE article 

reveal Lisi hate self

  ‘Zhangsan heard that your article revealed that Lisi hated his wife.’   (Xue et al. 1994: 437–​438) According to Xu (1994), the nominals that follow the passive morpheme bei or the preverbal object marker ba (hereafter bei and ba nominals) can be

 131

Prominence and locality 131 antecedents of ziji and do not block the binding of ziji to the subject of the sentence with unlike person features. Cole and Wang (1996) note that these bei and ba nominals do not induce the blocking effect even in LD binding, as shown below: (66)  a. Zhangsani yiwei Lisij  hui ba  

nik ling hui  zijii/​j/​k 

de  

jia.

  Zhangsan think Lisi   will BA  you lead back self    DE  home   ‘Zhangsan thought Lisi would take you back to his/​your home.’ b. Zhangsani yiwei  Lisij hui bei nik   ling hui  zijii/​j/​k

de   jia.

  Zhangsan think   Lisi will BEI you  lead back self    DE  home   ‘Zhangsan thought Lisi would be taken by you back to his/​your home.’   (Cole and Wang 1996: 360–​361) Cole et al. (2001) further note that although in (66) it is possible for the reflexive ziji to co-​refer with the matrix subject Zhangsan, this co-​reference can be more easily obtained in (67). Based upon this fact, they conclude that a mild blocking effect occurs in (66) due to the occurrence of a second person pronoun that follows ba/​bei. (67) a. Zhangsani yiwei Lisij hui ba Xiaomingk  dai   hui  

zijii/​j/​k

  Zhangsan think Lisi will BA Xiaoming   take  back  self   de    jia. DE   home   ‘Zhangsan thought Lisi would take Xiaoming back to his home.’ b. Zhangsani yiwei Lisij hui bei  Xiaomingk dai hui

zijii/​j/​k de  

jia.

  Zhangsan think Lisi will BEI Xiaoming take back self  DE home   ‘Zhangsan thought Lisi would be taken by Xiaoming back to his home.’   (Cole et al. 2001: 32) Cole et al. make a distinction between grammatical blocking and logophoric blocking. In (66) the mild blocking effect is derived from the logophoric blocking, whereas in the following sentences the grammatical blocking is derived from LF head-​ movement and feature percolation. According to their observation, the grammatical blocking is stronger than the logophoric blocking. (68) a. Zhangsani yiwei woj  hui ba   nik dai hui  ziji*i/​j/​k de    Zhangsan think  I  

will BA  you take back self 

jia.

DE home

  ‘Zhangsan thought I would take you back to *his/​my/​your home.’

132

132 Reflexives b. Zhangsani yiwei

woj hui bei  nik dai hui

ziji*i/​j/​k  de 

  Zhangsan think  I  will BEI you take back  self 

jia.

  DE  home

  ‘Zhangsan thought I would be taken by you back to *his/​my/​your home.’   (Cole et al. 2001: 31) It should be pointed out that, although the blocking effect is sometime strong and sometimes weak, this does not mean that the stronger one is a grammatical blocking, whereas the weaker one is a logophoric one. According to Cole et  al. (2001), if the blocker occupies the subject position, it will produce the strong blocking effect. Under their analysis, in sentences given in (68), ziji will first adjoin to the local AGR at LF and percolate its features to it, and since ziji can subsequently adjoin to the AGR of the next clause up, a person feature conflict will occur if the NP occupying the Spec of AGR has a different person feature. Hence, the blocking effect. This kind of feature percolation mechanism is widely assumed in various LF approaches to the study of the bare reflexive ziji. The basic idea is to derive the blocking effect from a failure in grammatical agreement. It is a well-​established fact that grammatical agreement cannot be violated, as shown below: (69) *He am a student. The above sentence is excluded simply because the Spec of the AGR and the AGR itself have different person features. If the blocking effect can really be derived from the agreement failure, as suggested by Cole et al. (2001), the following sentences should also be ruled out by the grammar. (70) a. Ni  pa  

ta   chaoguo  ziji ma?

  you fear  he  surpass  

self Q

  ‘Are you afraid that he may surpass you?’ b. Woi bu   xihuan Lisij guan   I 

  zijii/​j  de  shi.

not  like    Lisi interfere  self

DE matter

  ‘I don’t like Lisi interfering in my/​his (own) business.’   (Pan 2001: 283) However, the above sentences are perfectly acceptable, though the embedded subject and matrix one have different person features. The above fact demonstrates that the blocking effect exhibited by ziji cannot be derived from the agreement failure and thus should not be treated as a grammatical blocking. It is unreasonable to assume that a derivation violating a grammatical constraint can remain grammatical if the constraint is really a grammatical one.

 133

Prominence and locality 133 3.3.4  Pan’s (2001) self-​ascription theory Pan (1995, 1997, 2001) makes a distinction between local and LD bound reflexives and claims that only LD bound noncontrastive ziji exhibits the blocking effect. He finds that the blocking effect of ziji is not symmetrical, and the crucial factor in inducing the blocking effect is not the local subject or the unlike person feature conflict, as suggested in the literature (Huang and Tang 1991; Xue, Pollard, and Sag 1994), but the asymmetry between first/​ second and third person NPs. He claims that first/​second person pronouns can block third person NPs from LD binding ziji, though third person NPs do not necessarily block first/​second person pronouns from LD binding ziji, as shown below. (71)  a. Woi bu xihuan  Lisij guan 

  zijii/​j  de  shi.

  I   not like     Lisi interfere  self

DE matter

  ‘I don’t like Lisi interfering in my/​his (own) business.’ b. Nii   xihuan Lisij guan    you  like  

  zijii/​j  de   shi 

  ma?

Lisi interfere  self   DE  matter  Q

  ‘Do you like Lisi interfering in your/​his (own) business?’ c. Lisii bu   xihuan wo/​nij guan 



ziji*i/​j de   shi.

  Lisi not  like    I/​you interfere  self  DE  matter   ‘Lisi does not like me/​you interfering in my/​your (own) business.’ d. Lisii bu xihuan Zhangsanj guan    Lisi not like  

  zijii/​j  de 

shi.

Zhangsan interfere  self   DE  matter

  ‘Lisi does not like Zhangsan interfering in his (own) business.’   (Pan 2001: 283) (71a) and (71b) show that the co-​reference between the first/​second person pronoun and ziji is not blocked by the intervening third person NP Lisi. (71c) shows that when the first/​second person pronoun intervenes between the matrix subject and ziji, their co-​reference is blocked. Pan (1995, 1997, 2001) also notes that other grammatical functions filled by first/​second person pronouns, not just subject or NPs contained in the subject, can induce the blocking effect, as shown in (64). Based upon these facts, he develops a theory of self-​ascription to account for the blocking effect exhibited in ziji binding. Pan (1995, 1997, 2001) claims that LD ziji points to the carrier of belief and is thus constrained by self-​ascription in interpretation. He accounts for the blocking effect by appealing to the fact that only first/​second person pronouns are obligatory self-​ascribers and can thus block LD binding of ziji if they intervene between the potential third person NP and ziji, while third person

134

134 Reflexives NPs are optional self-​ascribers and thus do not necessarily block ziji from being LD bound by first/​second person pronouns. Pan (1995, 1997) makes a distinction between the locality ziji and the self-​ascription ziji. He claims that only the self-​ascription ziji is sensitive to the blocking effect. Pan (2001: 297) proposes the following binding condition for the self-​ascription ziji: (72)  The Condition for Self-​Ascription Ziji Ziji can be bound to the carrier of belief, the most prominent self-​ ascriber, in a linguistic domain γ iff there is no blocker in the believed proposition contained in γ. The Prominence Condition is defined below (Pan 2001: 298): (73) The Prominence Condition α is the most prominent self-​ascriber in γ iff there is no β in γ such that β appears higher in one of the following hierarchies than α. a.  SUBJ > OBJ or OBLIQUE b.  Dominating NPs > Dominated NPs Under Pan’s analysis, blockers are separated from antecedents, and hence, a blocker is not necessarily among the possible antecedents. It seems that the Prominence Condition is only responsible for the identification of possible antecedents, but is irrelevant to that of blockers. That is why Pan (2001: 298) defines blockers separately. (74) α is a blocker for β if α is a self-​ascriber such that (a) α precedes ziji; and (b) neither α nor the NP controlled by it is an argument of an irreflexive predicate containing ziji. Although Pan’s analysis fares better than previous accounts, it still has a number of problems left unaccounted for. First, the addition of the notion of the irreflexive predicate to the definition of blockers weakens the explanatory power of the self-​ascription theory since it is unclear why self-​ascription should be sensitive to irreflexive predicates if first/​second person pronouns are really obligatory self-​ascribers, as claimed in Pan (1995, 1997, 2001). The reason why irreflexive predicates should be considered in the definition of blockers is mainly because there exist sentences like the following which are not constrained by self-​ascription. (75) a. Ta pa wo  chaoguo  ziji.   he fear I  

surpass  

self

  ‘He is afraid that I may surpass him.’

 135

Prominence and locality 135 b. Zongtong qing wo zuo zai ziji de pangbian.   president ask  I  sit  at self DE side   ‘The president asked me to sit beside him.’   (Xu 1993: 136) In (75) the first/​second person pronoun does not block the binding of ziji to a third person NP. Pan (2001) explains these examples away by claiming that self-​ascription does not occur in the irreflexive predicate. If this is true, it means that first/​second person pronouns may not be obligatory self-​ascribers. It is true that there are some pragmatic or world-​knowledge factors that exclude the possibility for ziji to be co-​referential with the closest NP in (75). But we can modify the above sentences slightly to make it possible for ziji to be bound to either the closest NP or the matrix subject, as shown below: (76)  a. Tai pa   nij  chaoguo  zijii/​j  de   erzi.   he fear  you surpass  

self   DE  son

  ‘He is afraid that you may surpass your/​his son.’ b. Zongtongi qing woj zuo zai  zijii/​j de  taitai  pangbian.   president  ask  I   sit  at  self  DE wife  side   ‘The president asked me to sit beside my/​his wife.’ In (76) the interpretation of the reflexive ziji is not constrained by the irreflexive predicate since it is not an argument of it. Hence, it becomes possible for ziji to refer to the local antecedent. Since binding to the matrix subject is not the only possible choice, the self-​ascription theory would predict that ziji cannot be bound to the matrix subject, contra the above fact. Second, the self-​ascription theory also fails to explain why first/​second person pronouns are not obligatory self-​ascribers when they occur after ba/​bei as shown in (66). Third, there exists a potential conflict between Pan’s (2001) account of self-​ ascription and the claim that first/​second person pronouns are obligatory self-​ ascribers. According to Pan (2001:  295), a self-​ascriber ascribes a property to him/​herself. In the following sentence (77), ziji cannot refer to the matrix subject since there is an obligatory self-​ascriber, the first person pronoun wo “I,” in the believed proposition. If the first person pronoun is an obligatory self-​ascriber, then one may ask why it cannot function as the antecedent. A ready answer for this question is that blockers may not be antecedents. Even if we accept this claim, we may still ask why the first person pronoun in (77) is a self-​ascriber and what property it self-​ascribes. It seems that there is no believed proposition for it to self-​ascribe in (77). If the first person pronoun in (77) does not have a property to ascribe, to label it as a self-​ascriber in (77) may not be well-​motivated since it is inconsistent with Pan’s claim that a self-​ ascriber should ascribe a property to him/​herself. This problem is also pointed

136

136 Reflexives out in Huang and Liu (2001). They argue that to call the embedded subject in (71c) a self-​ascriber is not appropriate for what the term self-​ascription means because it is simply the subject of some event-​denoting predicate without ascribing any property. (77)  Lisii yiwei woj de xueshengk  bu   xihuan ziji*i/​*j/​k. Lisi think  I 

DE student  

not  like  

self

‘Lisi thinks that my student does not like himself.’ 3.3.5  Huang and Liu’s (2001) account Huang and Liu (2001) also make a distinction between the local reflexive and the LD reflexive and claim that the dividing line between them is the traditional notion of the Governing Category (GC), as given below (Huang and Liu 2001: 141): (78)

The Governing Category (cf. Huang 1983; Chomsky 1981) α is the Governing Category for β iff α is the minimal category containing β, a governor of β, and a SUBJECT accessible to β.

Huang and Liu (2001) claim that the GC can make a correct distinction between embedded subject and object reflexives: (79) a. Zhangsan yiwei ziji de  

erzi  zui  congming.

  Zhangsan think self DE  son   most clever   ‘Zhangsan thought that his son was the cleverest.’ b. Zhangsan yiwei Lisi zui xihuan ziji de  

erzi.

  Zhangsan think Lisi most like  self DE

son

  ‘Zhangsan thought that Lisi liked his son most.’ (80) a. Zhangsan shuo ziji kanjian-​le  Lisi.   Zhangsan say  self see-​ASP   Lisi   ‘Zhangsan said that he saw Lisi.’ b. Zhangsan shuo Lisi kanjian-​le ziji.   Zhangsan say  Lisi see-​ASP self   ‘Zhangsan said that Lisi saw him.’   (Huang and Liu 2001: 168) According to Huang and Liu, in (79a) and (80a), ziji is ambiguous between a logophor and a local reflexive when being bound by the matrix subject Zhangsan. That is to say, it can refer to the matrix subject under either a de se scenario or a non-​de se scenario. Whereas in (79b) and (80b), ziji can only be

 137

Prominence and locality 137 bound by the embedded subject Lisi, but not by the matrix subject Zhangsan under the non-​de se scenario. Under their analysis, this difference is derived from the fact that the GC for ziji in (79a) and (80a) is the main clause and that in (79b) and (80b) is the embedded clause. They claim that the distribution of the blocking effect in the following sentences further shows that the dividing line between the local reflexive and the LD reflexive is the GC. (81)  a. Zhangsani gaosu wo zijii de     Zhangsan tell  

erzi  zui  congming.

me self DE  son   most clever

  ‘Zhangsan told me that his son was the cleverest.’ b.??Zhangsani gaosu wo Lisi zui  xihuan zijii de erzi.   Zhangsan  tell 

me Lisi most like 

  self DE son

  ‘Zhangsan told me that Lisi liked his son most.’ (82) a. Zhangsani dui wo shuo zijii piping-​le    Lisi.   Zhangsan to  me say  self criticize-​ASP Lisi   ‘Zhangsan said to me that he criticized Lisi.’ b.??Zhangsani dui wo shuo Lisi piping-​le    zijii.   Zhangsan  to  me say Lisi criticize-​ASP self   ‘Zhangsan said to me that Lisi criticized him.’   (Huang and Liu 2001: 169) In (81b) and (82b) the binding of ziji to the matrix subject is blocked by a first person pronoun since ziji is a LD reflexive, whereas in (81a) and (82a) the binding of ziji to the matrix subject is not blocked since ziji is a local reflexive which is bound within its GC. Huang and Liu further claim that in the following sentence ziji is also a local reflexive when it takes a subcommanding NP as its antecedent. (83) [Zhangsani de  jiao’ao]j hai-​le    zijii/​*j. Zhangsan DE pride   hurt-​ASP self ‘Zhangsan’s pride hurt him.’ Since under their analysis only logophors exhibit the blocking effect, they correctly predict that in the following sentence the binding of ziji to the subcommanding antecedent is not blocked. (84) Zhangsani de   biaoqing  gaosu  woj [zijii/​*j shi  wugude]. Zhangsan DE expression tell 

  me  self  be

innocent

‘Zhangsan’s [facial] expression tells me that he is innocent.’ (Huang and Liu 2001: 170)

138

138 Reflexives Although it may be both theoretically and empirically necessary to make a distinction between local ziji and LD ziji, Huang and Liu’s criterion to distinguish them fails to capture an important distinction between the local reflexive and the LD reflexive.4 It is well-​known in the literature that the local reflexives are in complementary distribution with pronouns, whereas the LD reflexives are in free variation with them. It is obvious that in (79a), (80a), (81a), (82a), (83), and (84), ziji can be replaced by a pronoun that maintains the reference for ziji. If ziji in these sentences is really a local reflexive, one may ask why it can be in free variation with a pronoun. Note that in the following sentence, ziji cannot be replaced by a pronoun, which indicates that ziji is a real local reflexive in this sentence: (85)  Johni xihuan zijii/​ta*i. Johni likes  himselfi/​him*i. I think that if it is really necessary to make a distinction between local ziji and LD ziji, the distinction should be kept only to those syntactic environments where ziji is in complementary distribution with the pronoun. Huang and Liu (2001) employ a theory of perspectivity to account for the blocking effect exhibited in the binding of ziji and claim that the blocking effect can be explained by taking literally Kuno’s (1972) direct discourse representation hypothesis. According to Huang and Liu’s analysis, the LD binding of ziji is blocked in (86a) because there is a conflict between the perspective of the external speaker and that of the internal speaker. (86) a. Zhangsan juede

wo zai  piping  ziji.

  Zhangsan think   I  at  

criticize self

  ‘Zhangsan thinks that I am criticizing self.’ b. Zhangsan juede, “wo zai   piping  wo.”   Zhangsan think   I  at  

criticize me

  Zhangsan thinks, “I am criticizing me.”   (Huang and Liu 2001: 161) Following Kuno (1972), Huang and Liu suggest that the LD ziji of (86a) be derived from an underlying first person pronoun wo “I/​me,” of (86b). (86b) is considered as the underlying representation of (86a) under their analysis. In (86b), there are two occurrences of wo “I.” The first one refers to the (external) speaker of the entire sentence, and the second one refers to Zhangsan, the (internal) “speaker” of the direct discourse complement. Because there is a conflict between the perspectives of two different speakers, (86a) is not acceptable if ziji is bound to the matrix subject. As pointed out in Pan (2001:  301), the problem with the perspectivity account is that it would wrongly predict that the LD binding of ziji is blocked in sentences like the following:

 139

Prominence and locality 139 (87)  John shuo Bill ba  ziji de   shu

songgei-​le 

wo.

John say  Bill BA self DE book send-give-​ASP  I ‘John said that Bill gave his own books to me.’ (Pan 2001: 300) Under the analysis of Huang and Liu (2001), (87) can be represented as (88): (88) John shuo, “Bill ba wo  de   shu song-​gei-​le  John say  

  wo.”

Bill BA I   DE book send-​give-​ASP  I

John said, “Bill gave my own books to me.” In (88), the first wo “I” refers to John, the (internal) “speaker” of the direct discourse complement, and the second one refers to the (external) speaker of the entire sentence. Because there is a conflict between the perspectives of two different speakers, the analysis assumed in Huang and Liu would predict that (88) is not acceptable if ziji is bound to the matrix subject, contra the fact. Note that sentences like those in (66), (75), (76), and (77), repeated below as (89), (90), (91), and (92) are also problematic for the perspectivity-​based account. (89) a. Zhangsani yiwei Lisi hui ba  

ni  ling hui  

zijii  de  jia.

  Zhangsan think Lisi will BA  you lead back  self  DE home   ‘Zhangsan thought Lisi would take you back to his home.’ b. Zhangsani yiwei Lisi hui bei   ni  ling hui 

zijii de  

jia.

  Zhangsan think Lisi will BEI you lead back  self DE home   ‘Zhangsan thought Lisi would be taken by you back to his home.’ (90) a. Ta pa wo  chaoguo ziji.   he fear I   surpass  self   ‘He is afraid that I may surpass him.’ b. Zongtong  qing wo zuo  zai ziji   de  pangbian.   president   ask  I   sit   at  self  DE side   ‘The president asked me to sit beside him.’ (91) a. Tai pa  nij  chaoguo zijii/​j   de erzi.   he fear you surpass      self  DE son   ‘He is afraid that you may surpass your/​his son.’ b. Zongtongi qing woj zuo zai  zijii/​j de  taitai pangbian.   president  ask  I   sit  at 

self DE wife side

  ‘The president asked me to sit beside my/​his wife.’

140

140 Reflexives (92)  Lisii yiwei woj de  xueshengk  bu  Lisi think  I  DE student 

xihuan ziji*i/​*j/​k.

  not  like

self

‘Lisi thinks that my student does not like himself.’ Take (89a) as an example. When ziji refers to the matrix subject, (89a) can be represented as (93): (93) Zhangsan yiwei, “Lisi hui ba  ni  ling  hui 

wo de  jia.”

Zhangsan think  Lisi will BA you lead back  I

DE home

Zhangsan thinks, “Lisi would take you back to my home.” According to the analysis assumed in Huang and Liu (2001), in (93) ni refers to the addressee with respect to the external speaker, and wo refers to the internal speaker Zhangsan. Since there is a conflict between the internal Source (to whom wo is “anchored”) and the external Source (to whom ni is anchored), (89a) is predicted to be unacceptable under the intended reading according to the analysis assumed in Huang and Liu (2001), again contra the fact. Although Huang and Liu can dispense with (90) by claiming that it involves the irreflexive predicate, they cannot do so with (91). Another problem with their analysis is that it would wrongly predict that in (92) ziji can be bound to the subcommanding NP, as shown below. (94) Lisi yiwei, “wo de 

xuesheng bu xihuan wo.”

Lisi think  I  DE  student  not like   me Lisi thinks, “My student does not like me.” In (94), the direct discourse representation for (92), there are two occurrences of wo “I.” Since both the first and the second wo would refer to the external speaker under the intended reading that the first wo and the second wo co-​refer, there is no perspective conflict in (94). As a result, the perspectivity-​based account would wrongly predict that ziji can refer to the subcommanding NP in (92). Of course, Huang and Liu could argue that the subcommander wo cannot antecede the object wo in (94) since the subcommander wo is an NP dominated by another NP. But the question that is left unanswered is why perspectivity should be sensitive to the difference between the dominating NP and the dominated NP. Another problem with the analysis assumed in Huang and Liu (2001) is that it would wrongly predict that the logophoric use of the compound reflexives in Chinese should also be blocked when the perspective conflict occurs. Consider the following sentence: (95) Lisii bu xihuan wo  guan 

  ta-​zijii   de   shi.

Lisi not like    I   interfere he-​self   DE matter ‘Lisi does not like me interfering in his (own) business.’

 141

Prominence and locality 141 When used contrastively, the compound reflexive ta-​ziji can be LD bound. Note that since the compound reflexive has a clearly specified person feature, it does not exhibit the blocking effect in LD binding. If we use the Direct Discourse Representation approach to represent (95), the result is (96): (96)  Lisii bu xihuan, “wo guan    woi  de    shi.” Lisi does not like, “I interfere in my (own) business.” Since the first wo refers to the external speaker, and the second wo refers to the internal speaker, Huang and Liu’s analysis would expect that the LD binding of logophoric ta-​ziji to the matrix subject should be blocked, but in fact, it is not. Besides these problems, it also seems strange to represent sentences like (95) as (96) no matter whether the reflexives involved is a compound one or a bare one since verbs like xihuan “like” cannot be used to introduce a direct discourse.

3.4  Prominence and locality in Chinese reflexive binding 3.4.1  The binding of the compound reflexive Pan (1995, 1997, 1998) notes that the Chinese compound reflexive (e.g., ta-​ ziji “he/she-self ”) can be bound not only to an LD antecedent but also to a non-​c-​commanding/​subcommanding antecedent, as shown by the following sentences cited from Pan (1998): (97) a. John i shuo naben shu    fang  zai  ta-​ziji i de   jiali.   John say  that  book   put   at   he-​self DE  home   ‘John said that that book was put at his home.’ b. Johni de   jiao’ao hai-​le    ta-​zijii.   John DE  pride  hurt-​ASP he-​self   ‘John’s pride hurt him.’ c. Wo wei John zhaodao-​le ta-​ziji de zhaopian.   I  for  John find-​ASP

he-​self DE photo

  ‘I found John’s photo for him.’ d. Johni shuo [Billj de 

xiaocongming]k hai-​le    ta-​ziji i/​j/​*k.

  John say  Bill DE 

little-​trick      hurt-​ASP 

  ‘John said that Bill’s little trick hurt him.’ e. *Johni zhidao Bill xihuan ta-​zijii.   John know Bill like     he-​self   ‘John knows that Bill likes himself.’

he-​self

142

142 Reflexives In (97a), ta-​ziji is LD bound across a local subject. In (97b), it is bound to a subcommanding NP. In (97c), it is bound to a non-​c-​commanding antecedent. In (97d), it can be bound to either the subcommanding NP or the matrix subject. In (97e), it cannot be bound to the matrix subject. Based upon the above facts, Pan (1998) claims that an NP can be the antecedent of an anaphor if there is no closer blocker in its linguistic domain. Under his analysis, the blocker covers two types of NPs with equal or higher prominence, namely, the subject and the nonsubject that dominates the candidate in question. Based on Pan (1998), Pan and Hu (2001) claim that the Chinese compound reflexive cannot be bound to a remote antecedent by crossing a closer one which is not less prominent than the remote one. They give the following condition to constrain the binding possibilities of the Chinese compound reflexive. (98)  An anaphor cannot be bound to α across an intervening NP β which is not less prominent than α. According to the above condition, the compound reflexive in (97e) cannot be bound to the matrix subject simply because the local subject is not less prominent than the matrix subject. In Pan and Hu (2001), prominence is defined on the basis of the animacy hierarchy and the grammatical function hierarchy, as shown below: (99) The Animacy Hierarchy (AH) (Chou 1992) [+Human] > [+Animate (-​Human)] > [-​Animate] (100)  The Grammatical Function Hierarchy (GFH) Subject > Nonsubject Although (98) can correctly constrain the binding of the Chinese compound reflexive, it would wrongly predict that in the following sentence the bare reflexive ziji cannot be bound to the matrix subject. (101) Zhangsani renwei Lisij xihuan  zijii/​j. Zhangsan think  Lisi like  

  self

‘Zhangsani thinks that Lisij likes himselfj/​himi.’ Pan (1995, 1997, 1998) as well as Pan and Hu (2001) account for this difference in binding between the compound reflexive and the bare reflexive by assuming that the compound reflexive and the bare reflexive should be constrained by different binding conditions. Although this is possible, it will be more welcome if these two kinds of reflexives can be constrained by the same condition. In previous discussion, I claim that if ta-​ziji finds a proper

 143

Prominence and locality 143 antecedent, it is not motivated to find another one. Assume that this proper antecedent for the compound reflexive as well as the bare reflexive is the most prominent NP, though it is possible that ziji and ta-​ziji require different definitions of the most prominent antecedents. If reflexives do not tend to find another antecedent whenever it finds the most prominent one, then the blocking effect exhibited by reflexives can be derived from the prominence of NPs. If this is true, then we can recast Pan and Hu’s (2001) generalization as a general reflexive binding condition: (102)  Reflexive Binding Condition (RBC) A reflexive prefers to be bound to a more prominent NP, but cannot be bound across the most prominent one. Now, the remaining task is to define the most prominent NP. I  think that the variation in reflexive binding just lies in the definition of the most prominent NP. A plausible assumption is that different languages may have different definitions of prominence (cf. Pan 1998), and different reflexives are constrained by different prominent NPs. According to the feature characterization given in (56), the Chinese compound reflexive needs to obtain a referential feature from its antecedent. Assume that it can pick up the referential feature from any antecedent available under the condition that the Reflexive Binding Condition (RBC) is not violated. Under this view, the reflexive in question can be bound to any binder available as long as it does not violate RBC. Since we have two prominence hierarchies for the compound reflexive, it can be easily seen that the most prominent binder for it is an NP with the feature [+Human, +Subject]. Suppose that its feature searching process is somewhat like a feature replacing process. Whenever it finds a binder, it is bound to that binder, but whenever it finds a more prominent binder, it will replace the less prominent one. But, whenever it picks up the referential feature from the most prominent binder, its feature searching process will stop since the most prominent binder does not allow its feature to be replaced by the feature of a less prominent binder. It can be seen that this feature searching process can derive not only different binding possibilities but also the constraint on these possibilities. For instance, in (97d), the compound reflexive can get its referential feature from the subcommanding NP since RBC is not violated when the reflexive is bound to it. If the feature searching process for the compound reflexive optionally stops, it will end up being co-​indexed with the subcommander. But, if it continues its feature searching, it will find the matrix subject as its binder and replace the feature obtained from the less prominent binder with the feature of the more prominent one. In (97e), when the compound reflexive is bound to the embedded subject, it is anchored there and cannot continue its feature searching process because the embedded subject is the most prominent one that does not allow the reflexive to replace its feature with that of another binder that is not more prominent than it.

144

144 Reflexives 3.4.2  The binding of the bare reflexive: Two search engines It is shown in the above section that the prominence condition for the Chinese compound reflexive can be defined by the Animacy Hierarchy (AH) and the Grammatical Function Hierarchy (GFH). Now, let us see how the prominence condition for the bare reflexive can be defined. In defining the prominence conditions for the bare reflexive, two extra things must be considered in addition to AH and GFH. One is person feature, and the other is the notion of empathy (Kuno 1987). Person feature should be considered because the bare reflexive ziji lacks phi-​features. The notion of empathy is included because it affects the binding of ziji. In order to avoid being involved in the potential controversies around the definition of empathy, let us restrict the notion of empathy to the use of certain verbs. Consider the following sentences: (103)  a. ?Lisii  shuo laoshi  hui   piping Proi.   Lisi   say  teacher will  criticize   ‘Lisi said that the teacher would criticize him.’ b. Lisii pa  laoshi hui  piping Proi.   Lisi fear teacher will

criticize

  ‘Lisi is afraid that the teacher would criticize him.’ Although the empty object in (103a) does not tend to refer to the matrix subject, the one in (103b) can be easily associated with the matrix subject for interpretation. This difference results from the fact that the matrix verb in (103b) is a verb of empathy, but the one in (103a) is not. Verbs like pa “fear,” gan “dare,” danxin “worry,” baoyuan “complain,” and huaiyi “suspect” are verbs of inherent empathy in the sense that the use of these verbs shows the speaker’s empathy towards the subject (i.e., the referent) of these verbs. These verbs express some subjective feelings or judgment of the speaker towards the subject NP that these verbs are predicated of. Since it is often the case that the complement clauses selected by these verbs describe some event that may have an effect on the subject NP that these verbs are predicated of, it is quite natural for the anaphor involved in the complement clause to refer to the subject NP of these verbs. For instance, in (103b), if what the subject NP Lisi is afraid of has nothing to do with him, he may have no reason to feel afraid and the speaker may have no reason to believe that Lisi should be afraid. In the relevant case, the use of pa “fear” in that sentence would be unreasonable. Assuming that the subject of the verb that induces the speaker’s empathy is more prominent than the one that does not, we can explain why the subject of the empathy verbs can be easily chosen as the antecedent of the anaphor. Now, we can formulate the following prominence hierarchy involved in empathy. (104) The Empathy Hierarchy (EH) The Subject of the Inherent Empathy Verb > Other Subjects

 145

Prominence and locality 145 Since ziji lacks phi-​ features, it can be conceived that its phi-​ features searching process should be constrained by the following person-​ feature hierarchy. (105)  The Person Hierarchy (PH) First/​Second Person Pronoun> Third Person NP As for the reference of ziji, I think that, besides AH, it is also determined by the following Closeness Hierarchy (CH) and the Syntactic Prominence Hierarchy (SPH) which incorporates notions such as grammatical functions and c-​command. I propose that it is the combination of AH, CH, and SPH that determines the binding of ziji. (106) The Syntactic Prominence Hierarchy (SPH) a. The Grammatical Function Hierarchy (GFH)   Subject/​Subject-​Possessor > Object (Indirect Object > Direct Object)   > Adjunct b. The Structural Hierarchy (SH)   C-​Commander > Subcommander > Non-​C-/​Subcommander (107) The Closeness Hierarchy (CH) [+Closer] > [-​Closer] Notice that closeness also plays a role in determining the prominence of NPs, as shown in (107). That is, an NP closer to the reflexive is more prominent than an NP that is less close to it unless the former appears lower than the latter in other higher hierarchies. The notion of closeness is defined below: (108) The Closeness Condition (Pan 1998: 793) α is closer to X, the reflexive, than β is iff the path from X to the minimal maximal projection dominating α is a proper subset of the path from X to the minimal maximal projection dominating β. Since we have different hierarchies in our system, one important thing is how to apply these hierarchies to compute the prominence of NPs and determine the binding of reflexives. I think that the application of these hierarchies will be regulated by something like the ordering of rules in early generative grammar or the ranking of the relevant constraints in the Optimality Theory (OT). The basic idea is that some of the hierarchies are ranked higher than others in the computation of the prominence of NPs. I propose that the relevant hierarchies can be ranked below: (109) The Prominence Ranking (PR) EH > AH > GFH > CH > SH

146

146 Reflexives I further propose that the computing of the prominence of the relevant NPs is regulated by the following algorithm. (110)  The Prominence Computing Algorithm (PCA) If A appears higher than B in a higher hierarchy in PR, A is more prominent than B. Following Huang and Tang (1991), I assume that there are two processes involved for ziji to acquire two kinds of features, though, different from Huang and Tang (1991), I do not assume that the phi-​feature is acquired at S-​structure while the referential feature is obtained at LF. I assume that both the phi-​feature and the referential feature of ziji are acquired in overt syntax. Suppose that there are two search engines at work: one for phi-​features, and the other for referential feature. Let us call the former P-​engine (phi-​feature search engine) and the latter R-​engine (referential feature search engine). The function of these two engines is to find a candidate set and the most prominent NP according to their respective prominence conditions so as to help ziji to delimit its binding domain. The P-​engine will choose the most prominent NP in PH, whereas the R-​engine will choose the most prominent NP in PR, according to PCA. After the two candidate sets and the most prominent NP(s) being chosen, the binding domain of ziji is in fact delimited, and since ziji should be bound within the binding domain defined by the union of the sets related to the two most prominent NPs chosen by the two engines, the blocking effect is also derived if the reflexive in question is bound outside of its binding domain. If this line of analysis is on the right track, RBC can be redefined as below on the basis of Chomsky (1981: 212) and Pan (1998, 2001): (111) Reflexive Binding Condition (RBC) a. A reflexive can be bound to an accessible prominent NP in its binding domain. b. The binding domain of the reflexive is the minimal complete functional complex (CFC) that contains all the members of the candidate set and the reflexive. c. A binds B iff A is co-​indexed with B, and A and B are compatible in phi-​features. d. A is accessible to B iff the assignment of the index of A to B would not violate *[γ …δ …], where γ and δ bear the same index. Notice that in the above definition, neither c-​command nor subcommand is required since it is already incorporated into the definition of the prominent NPs. Now, the most important task left is how to produce the candidate set for the bare reflexive ziji. Suppose that the candidate set for ziji

 147

Prominence and locality 147 results from the union of the two candidate sets selected by the R-​engine and the P-​engine. Suppose that the P-​engine and the R-​engine work according to the same principle that governs the antecedent searching process for the compound reflexive. Hence, whenever the P-​engine finds the most prominent NP (i.e., the first/​second person pronoun) in PH, it will stop its phi-​ feature searching process. The R-​engine works similarly. Whenever it finds the most prominent NP in PR according to PCA, its referential feature searching process will stop. The union of the two candidate sets produced by the P-​engine and the R-​engine delimits the domain of ziji binding. If we use the search engine to search for the candidates, the first thing that we should do is specify its searching domain. I think that the relevant searching domain should be confined to the sentence. Suppose that all the NPs in the sentence that precede the reflexive form a sequence according to their relative closeness to the reflexive: N = , and the NPs that follow the reflexive but are contained in the derivational cycle of the reflexive also form a sequence: N' = . And among the NPs that precede the reflexive, the NP that is the closest to the reflexive is αi, where i = 1. Among the NPs that follow the reflexive but are contained in the derivational cycle of the reflexive, the NP that is the closest to the reflexive is α'-​i, where -​i = -​1. For all the NPs that linearly precede α1, αi = αi+1. For all the NPs that linearly follow α'-​1 in the derivational cycle of the reflexive, α'-​i  =  α'-​i+-​1. After the searching domain is defined, we can formulate the following condition to regulate the operation of the search engines: (112)  The Condition on the Operation of the Search Engine (COSE) Search for candidates in N and N’, where N = , and N’ = . Suppose that there is an N, and N = the most prominent NP. Assuming that the operation of the search engine is constrained by the Closeness Condition, the search engine will: (i)  check if αi = 0. If αi = 0, go to (iii). If αi ≠ 0, pick up αi, and put it into the candidate set, and then go to (ii-​a). (ii)    a. check if αi = N. If αi = N, go to (iii). If αi ≠ N, repeat (i).     b. check if α'-​i = N. If α'-​i = N, go to (iv). If α'-​i ≠ N, repeat (iii). (iii)  check if α'-​i = 0. If α'-​i = 0, go to (iv). If α'-​i ≠ 0, pick up α'-​i and then go to (ii-​b). (iv)   stop searching and produce the candidate set. In (112), αi = 0 or α'-​i = 0 means that there is no NP available for the search engine to pick up. Of course, it is possible that the candidate set may be completely empty. In that case, the reflexive would have no binding domain and may thus be free in reference in the relevant sentence. As for the prominent NPs, they can be defined below:

148

148 Reflexives (113)  The Definition of the Prominent NPs A.  In Chinese, an NP is the most prominent NP for the P-​engine iff it appears the highest in PH, and an NPA is the most prominent NP for the R-​engine iff there is no B such that B appears higher than A in a higher hierarchy in PR according to PCA. In English, the most prominent NP for the R-​engine is the subject/​SUBJECT (i.e., [NP, IP], [NP, NP], or AGR). B. (i) The assignment of antecedents to the reflexive is constrained by the following PR'. PR' = AH > GFH > SH In Chinese, an NP that appears the highest in PR' in its own derivational cycle is a prominent NP that can antecede the reflexive iff there is no closer NP that appears higher than it in PR' in the binding domain. And, an NP that appears the highest in PR' in its own derivational cycle but lower than a closer NP in PR' in the binding domain of the reflexive is also a prominent NP that can antecede the reflexive, iff this NP can bear the speaker’s empathy. (ii) In English, A is a prominent NP for the reflexive if A c-​ commands the reflexive. I use PR' rather than PR to regulate the assignment of antecedents to the reflexive because the latter is used to choose only the most prominent NP for the R-​engine and thus cannot correctly predict which NP is the prominent NP that can antecede the reflexive. Notice that in the above definition, the notion of prominence has two uses. The most prominent NP is defined for the search engines. The prominent NP is used to assign the antecedent to the reflexive. In defining the possible antecedents, the notion of the derivational cycle is involved. A derivational cycle is a domain upon which rules operate (e.g., the clause (IP)). Later on, I will show why the notion of the derivational cycle is needed in defining the possible antecedents for the reflexive. Now, consider the following sentences: (114) Lisii yiwei woj  de xueshengk bu xihuan ziji*i/​*j/​k. Lisi think  I   DE student  not like  self ‘Lisi thinks that my student does not like himself.’ (115) a. Zhangsani shuo Lisij gaosu-​guo wok Wangwun hen ziji*i/​?j/​*k/​n.   Zhangsan say  Lisi tell-​ASP

me Wangwu hate self

  ‘Zhangsan said that Lisi told me that Wangwu hates himself/​him.’ b. Zhangsani yiwei Lisij dui wok shuo-​guo Wangwun   Zhangsan think Lisi  to me say-​ASP   Wangwu     chang piping ziji*i/​?j/​*k/​n   often criticize self

 149

Prominence and locality 149   ‘Zhangsan thinks Lisi told me that Wangwu often criticized himself/​him.’ c. Zhangsani yiwei Lisij cong nik   nar

tingshuo  Wangwun

  Zhangsan think Lisi  from you there hear-​say  Wangwu   chang piping    ziji*i/​?j/​*k/​n.   often criticize

self

  ‘Zhangsan thinks that Lisi heard from you that Wangwu often criticized himself/​him.’ (114) is repeated from (63), and the sentences given in (115) are constructed on the basis of (64). What is interesting about (114) is that although the first/​second person pronoun in it is not a possible antecedent and thus may not be treated as the most prominent binder, it produces the blocking effect. How is this possible? If the first/​second person pronoun is not the most prominent binder in these sentences, one expects that the reflexive can be bound across it, but in fact, it cannot. According to Xue et  al. (1994), ziji cannot refer to the intermediate subject in (115), but according to Pan (2001) and my own judgment, it can, though a mild blocking effect occurs when ziji refers to the intermediate subject. Notice that, although ziji can refer to the intermediate subject, it cannot refer to the matrix subject in (115). Based upon Huang and Tang (1991), I assume that there are two feature search engines that operate simultaneously in overt syntax to find possible antecedents for ziji. The LD binding property of ziji is derived from the fact that the phi-​feature searching process and the referential feature searching process are governed by two different prominence conditions. Under the present analysis, the P-​engine’s feature searching process is regulated by PH only, and PR is thus irrelevant to the P-​engine’s searching. Since the P-​engine’s searching process is constrained by the prominence condition, it stops its searching whenever it finds the most prominent NP defined by PH. When it stops its searching, the P-​engine will produce a set of candidates. Consider (114). The P-​engine will first find the embedded subject xuesheng “student,” and since it is a third person NP and is thus not the most prominent one in PH, the P-​engine will continue its searching. The next one found by the P-​engine is the first person pronoun dominated by the embedded subject. Since the first person pronoun is the most prominent NP according to PH, the P-​engine stops its searching process here. Now, the candidate set produced by the P-​engine includes the first person pronoun wo “I” and the embedded subject xuesheng. The R-​engine will also produce a set in its searching. It stops its searching when it finds the embedded subject xuesheng since it is the most prominent NP in PR according to PCA. Xuesheng is more prominent than wo in SH and CH, though they are equally prominent in GFH, AH, and EH. Hence, xuesheng is more prominent than wo. If we compare the prominence

150

150 Reflexives of xuesheng with that of the matrix subject, we find that the former is more prominent than the latter in CH, though they are equally prominent in SH, GFH, AH, and EH. Since no other NP is more prominent than xuesheng in (114), xuesheng is chosen as the most prominent NP by the R-​engine. When xuesheng is picked up by the R-​engine, the searching process is stopped. The union of the candidate sets produced by the P-​engine and the R-​engine is {wo, xuesheng}. Hence, the binding domain for ziji is the embedded clause since the embedded clause is the minimal CFC that contains the reflexive and the candidate set defined for the reflexive. In this domain, xuesheng is the only possible antecedent for ziji since the first person pronoun wo “I,” as a subcommander NP dominated by the commander NP xuesheng, is less prominent than xuesheng in SH in the derivational cycle (i.e., the clause that contains both wo and xuesheng). Hence, wo cannot function as a possible antecedent for ziji. Since the matrix subject is not in the binding domain, ziji cannot be bound to it, given that ziji can only choose its possible antecedents in its binding domain. Now, consider (115). In (115a), the candidate set produced by the P-​ engine is {wo, Wangwu}. The matrix subject Zhangsan and the intermediate subject Lisi are not included since the P-​engine stops its searching when it finds the most prominent NP wo “I” defined by PH. The candidate set produced by the R-​engine has only one member which is {Wangwu} since Wangwu is the most prominent NP in (115a) in PR according to PCA. Wangwu is the most prominent NP because there is no other NP that is more prominent than it, given that it is more prominent than the matrix object wo in CH and GFH, and more prominent than the matrix subject Zhangsan and the intermediate subject Lisi in CH. The union of these two sets is {wo, Wangwu}. Since the minimal CFC that contains the candidate set and the reflexive is the intermediate clause followed by the embedded clause, the combination of the intermediate clause and the embedded clause constitutes the binding domain of the reflexive. Notice that the binding domain cannot be the embedded clause since the embedded clause does not contain wo, which is a member of the candidate set. Since Wangwu is the most prominent NP and there is no closer NP that appears higher than it in PR' in the binding domain, ziji can be bound to it. In (115a), wo cannot antecede ziji since wo, as an object, is less prominent than Lisi in PR' in its own derivational cycle (i.e., the intermediate clause). Lisi, as a prominent NP in its own derivational cycle, can antecede ziji since it is contained in the binding domain. In (115b) and (115c), the story is basically the same. In (115b), the P-​engine stops when it finds the first person pronoun wo with a candidate set {Lisi, wo, Wangwu}. Lisi is included in the candidate set because the operation of the search engines is constrained by the Closeness Condition proposed by Pan (1998), repeated below:

 151

Prominence and locality 151 (116)  The Closeness Condition (Pan 1998: 793) α is closer to X, the reflexive, than β is iff the path from X to the minimal maximal projection dominating α is a proper subset of the path from X to the minimal maximal projection dominating β. According to the above condition, there is no subset relation between the path from the reflexive to the intermediate subject and the path from the reflexive to the adjunct dui wo “to me” in (115b), as shown in (117). (117)

IP1 NP1

VP1

V

IP2

Zhangsan yiwei NP2 VP2

Lisi

PP

VP3

P NP3 V

IP4

dui wo shuo-guo

VP4

NP4

Wangwu chang piping ziji

As shown in (117), the path from ziji to the PP node dominating NP3 is {VP4, IP4, VP3, VP2, PP}, and the path from ziji to the IP2 node dominating NP2 is {VP4, IP4, VP3, VP2, IP2}. As there is no path containment relation between the path from ziji to the adjunct and the one from ziji to the intermediate subject Lisi, the grammatical system cannot determine which one the P-​engine finds first. As a result, both are chosen as the members of the candidate set. Since the adjunct in (115b) is a first person pronoun, the P-​engine stops its searching after picking up the adjunct and the intermediate subject. The candidate set produced by the R-​ engine includes

152

152 Reflexives only one member which is Wangwu. The union of these two sets is {Lisi, wo, Wangwu}. Since the minimal CFC containing the union of these two sets and the reflexive is the intermediate clause followed by the embedded clause, the binding domain of the reflexive is the combination of the intermediate clause and the embedded clause. Since wo, as an adjunct, is a non-​ c-​commanding NP and is thus less prominent than the intermediate subject Lisi in PR' in its own derivational cycle, it cannot antecede ziji. Wangwu can be the antecedent of ziji since it is the most prominent NP in its own derivational cycle and there is no closer NP that appears higher than it in PR' in the binding domain. Lisi can antecede ziji since it is contained in the binding domain and is also prominent in its derivational cycle. In (115c), the candidate set produced by the P-​engine is {Lisi, ni, Wangwu}, and the one produced by the R-​engine is {Wangwu}. The R-​engine picks up both Lisi and ni because it cannot determine which one is closer to the reflexive. The union of these two sets results in a set {Lisi, ni, Wangwu}. Since Wangwu, as an embedded subject, is the most prominent NP in PR' in its own derivational cycle (the embedded clause) and there is no closer NP that is more prominent than it in PR' in the binding domain, it can antecede ziji. The second person pronoun ni “you” cannot antecede ziji since it is an adjunct and thus less prominent than the intermediate subject Lisi in PR' in its own derivational cycle. Lisi can be the antecedent since it is contained in the binding domain and prominent in its derivational cycle. Now, consider the following sentence, repeated from (71a): (118)  Woi bu   xihuan  Lisij  guan  I 

not  like 





zijii/​j  de shi.

Lisi  interfere  self

DE matter

‘I don’t like Lisi interfering in my/​his (own) business.’ In (118), the candidate set produced by the P-​engine is {wo, Lisi}, and the candidate set produced by the R-​engine is {Lisi} since Lisi is the most prominent NP for the R-​engine. The union of these two sets is {wo, Lisi}. Since the minimal CFC that contains the candidate set and the reflexive is the whole sentence, the whole sentence is the binding domain. Since both the matrix subject wo and the embedded subject Lisi are prominent in their own derivational cycles in PR', both of them can be the antecedents of ziji. Notice that the closer NP Lisi does not block the binding of the reflexive to the matrix subject since Lisi does not appear higher than the matrix subject in PR'. Our system of two engines will also predict that if (118) is embedded as a complement clause, the union of the candidate sets produced by the P-​engine and the R-​engine is the same, as shown in the following sentence, where ziji can be bound to either the local subject or the intermediate subject, but cannot be bound across the intermediate subject since the intermediate clause defines the upper boundary of the binding domain.

 153

Prominence and locality 153 (119)  Zhangsani zhidao woj bu xihuan Lisik guan   Zhangsan know  I  not like  

  ziji*i/​j/​k de shi.

Lisi interfere  self  DE matter

‘Zhangsan knows that I don’t like Lisi interfering in my/​his (own) business.’ The present analysis can also account for the so-​called local binding of ziji discussed in Huang and Liu (2001). Consider the following sentences repeated from (81a) and (82a): (120) a. Zhangsani gaosu wo zijii de erzi  zui  congming.   Zhangsan tell  

me self DE son   most clever

  ‘Zhangsan told me that his son was the cleverest.’ b. Zhangsani dui wo shuo zijii piping-​le    Lisi.   Zhangsan to me say  self criticize-​ASP Lisi   ‘Zhangsan said to me that he criticized Lisi.’ In (120a), the candidate set produced by the P-​engine includes only one member (i.e., {wo}), and the one produced by the R-​engine includes two members (i.e., {Zhangsan, wo}). The union of these two sets produces the candidate set {Zhangsan, wo}, and the binding domain is the whole sentence that contains the candidate set. Since Zhangsan and wo are contained in the same derivational cycle, and the former, as a subject, is more prominent than the latter, which is an object, in a higher hierarchy (i.e., GFH) in PR' and there is no closer NP that is more prominent than it in the binding domain, it is chosen as the antecedent of ziji. In (120b), the candidate set produced by the P-​engine is {Zhangsan, wo}, and the one produced by the R-​engine is also {Zhangsan, wo}. The union of these two sets results in the candidate set: {Zhangsan, wo}, and the binding domain is the whole sentence that contains the candidate set. Since Zhangsan is more prominent than wo in PR' and there is no closer NP that appears higher than it in the binding domain, it is chosen as the antecedent of ziji. Note that our analysis can correctly capture the fact that in (120a) and (120b) the first person pronoun wo does not block the binding of ziji to the matrix subject without assuming that ziji in these sentences is a local reflexive constrained by a different binding condition. Obviously, the present analysis fares better since it gives a unified account for ziji. Note that as I have already pointed out, it is dubious to treat ziji in (120a) and (120b) as a local reflexive since it is in free variation with the pronoun. Now, consider the following sentences repeated from (66): (121) a. Zhangsani yiwei Lisij hui ba   nik ling   hui  ziji?i/​j/​k de  

jia.

  Zhangsan think Lisi will BA  you lead  back self   DE  home   ‘Zhangsan thought Lisi would take you back to his/​your home.’

154

154 Reflexives b. Zhangsani yiwei Lisij hui bei   nik   ling hui  ziji?i/​j/​k de  

jia.

  Zhangsan think Lisi will BEI  you  lead back self   DE  home   ‘Zhangsan thought Lisi would be taken by you back to his/​your home.’ In (121a), the candidate set produced by the P-​engine is {Lisi, ni}. Lisi is included in the candidate set because the operation of the search engines is constrained by the Closeness Condition, according to which there is no subset relation between the path from the reflexive to the embedded subject and the path from the reflexive to the ba/​bei nominal in (121a) because the ba/​ bei nominals are adjuncts in these sentences, as shown in (122). (122)

IP1

NP1 I’

I

Zhangsan

VP1

V

IP2

yiwei

NP2

I’

Lisi

I

VP2

hui

PP

VP3

P

NP3

ba

ni ling hui ziji de jia

As shown in (122), the path from ziji to the PP node dominating NP3 is {VP3, VP2, PP}, and the path from ziji to the IP2 node dominating NP2 is {VP3, VP2, IP2}. As there is no path containment relation between the path from ziji to the ba nominal and the one from ziji to the embedded subject Lisi, the grammatical system cannot determine which one the P-​engine finds first. As a result, both are chosen as the members of the candidate set. Since the ba nominal is a second person pronoun, the P-​engine stops

 155

Prominence and locality 155 its searching after choosing the ba nominal and the embedded subject. The candidate set produced by the R-​engine includes {Lisi, ni}. The R-​engine stops when picking up the embedded subject Lisi since Lisi is the most prominent NP for the R-​engine, as it is more prominent than the matrix subject in CH and more prominent than the ba nominal ni in SH and GFH. The union of the sets produced by the P-​engine and the R-​engine is {Lisi, ni}. Since the embedded subject Lisi is more prominent than the ba nominal, which is in an adjunct position, in PR' and there is no closer NP that is more prominent than it in the binding domain, Lisi is chosen as the antecedent. However, the ba nominal obviously can also be the antecedent. How is this possible? This is possible because the candidate set contains not only the embedded subject and the ba nominal but also a Pro, which is co-​indexed with the ba nominal. Consider the following representation for part of the structure of (121a): (123)  [IP Lisii hui [VP ba nij [VP ling tj [Proj hui  

zijii/​j de jia]]]]

  Lisi will   BA you  lead      back self DE home The structural representation given in (123) shows that the ba nominal is underlyingly the object of the verb ling “lead,” and the subject of the verb hui “return” is a Pro co-​indexed with the ba nominal. Chomsky (1981:  20, 60, 322)  assumes that PRO, which is the controlled Pro under the present analysis, has phi-​features such as person, number, and gender, but lacks independent or inherent reference. Following Chomsky (1981), let us assume that the controlled Pro has phi-​features, though it has no reference. Assume that its reference can only be obtained from an antecedent under control. If Pro has phi-​features, then it can be included in the candidate set produced by the P-​engine. Since Pro in (121a), as shown in (123), is co-​indexed with the ba nominal, the ba nominal can be associated with Pro via co-​indexation for interpretation though the ba nominal itself is not a possible antecedent for ziji. Note that the present analysis predicts that there is a blocking effect exhibited in (121a) when ziji is bound to the matrix subject. Now, consider (121b). In (121b), the candidates selected by the P-​engine are Lisi and ni, as there is no subset relation between the path from ziji to Lisi and the one from ziji to the bei nominal. Under the present analysis, bei is treated as a preposition, and hence, the bei nominal is treated as an adjunct, as shown in (124) below. Note that under the analysis assumed by Huang (1999) and Tang (2001), the nominal after bei is treated as a subject. One obvious problem with this analysis is that it fails to account for the blocking effect. If the bei nominal is really a subject, ziji cannot be bound across it in (121b). It is a well-​tested fact that when functioning as the subject, the first/​second person pronoun will block the binding of ziji. However, in (121b) there is no blocking effect at all when ziji is bound to Lisi across the second person pronoun that follows bei. This fact clearly demonstrates that the nominal that follows bei is not a subject.

156

156 Reflexives (124)  [IP Lisii hui [VP bei  nij [VP ling ti [Proi hui  zijii/​j  de   jia]]]]  

Lisi will  

BEI you  

lead      back self   DE home

The candidate set produced by the R-​engine contains {Lisi, ni}. The union of the two sets gives us the set {Lisi, ni}. Since Lisi is more prominent than ni in PR' in its own derivational cycle and there is no closer NP that is more prominent than it in the binding domain, it is chosen as the antecedent. But, it should be noted that ni can also function as the antecedent, though it is less prominent than Lisi. I will soon explain why ni can also antecede ziji. What is interesting about the sentences in (121) is that native speakers’ judgments vary. Some native speakers feel that there is a blocking effect exhibited in (121), whereas some feel that there is not. Cole et al. (2001) also note that the blocking effect is not strong in sentences like (121). The question to ask is why some native speakers feel that the blocking effect does not exist in (121). One possible reason for this is that the ba/​bei nominals in these sentences are members of a chain. Consider (123). In (123), the chain formed by the ba nominal contains three members before the binding of ziji: {nii, ti, Proi}. Suppose that when the P-​engine finds a chain in its searching, it does not select the whole chain but only one member of the chain. In fact, it is redundant to put the whole chain into the candidate set, given the fact that members of a chain share their phi-​features via co-​indexation. If it is true that selecting one member is enough, the P-​engine, of course, will select the member that is the closest to ziji. In (121a), the member in the chain {nii, ti, Proi} that is the closest to ziji is Pro. Assume that, although Pro has phi-​ features, it does not have a full specification of phi-​features. Specifically, it does not have clearly specified person features before being co-​indexed with an antecedent. If this is true, this means that the person feature on Pro is not strong. Under this situation, since Pro, instead of the second person pronoun, is selected from the chain, and since Pro does not have a strong person feature, and is thus not regarded as the most prominent NP, the P-​engine can continue its searching and thus produce a candidate set including {Zhangsan, Lisi, Pro}. Since all the members of this candidate set are subjects and thus prominent in their derivational cycles, they all can function as the antecedent of ziji. This result reflects the intuition shared by those native speakers who do not feel that there is a blocking effect in (121a). The same analysis can be applied to the bei nominal in (121b). Part of the structure of (121b) can be represented below: (125) [IP Lisii hui [VP bei  nij [VP Proj ling ti [Proi hui  zijii/​j de  jia]]]]   Lisi will  

BEI you   

  lead      back self  DE home

I have argued that the bei nominal is not subject, but adjunct. Here, I further argue that the NP that follows bei is not moved from the VP-​internal subject position. Rather it is a base-​ generated adjunct. The VP-​ internal

 157

Prominence and locality 157 position is occupied by a base-​generated Pro bearing the agent role. It should be pointed out that in contemporary generative grammar, the by-​phrase in English passive sentences is also not treated as a movement-​derived phrase. Although English and Chinese share this similarity in passives, there is an important difference between English and Chinese with respect to the formation of passives. In English, passivization is realized in morphology and the agent role is absorbed in morphology and thus cannot be realized as an argument in syntax, though it may still have a semantic effect on sentence interpretation as an implicit argument. In Chinese, passive sentences are not derived from a morphological process, and hence, the agent role is not absorbed in morphology and can thus be realized as an argument. Hence, one important difference between English and Chinese with respect to passivization is that the agent role is not realized in syntax in English but may have syntactic realization in Chinese. Notice that the above structure given in (125) is very plausible for Chinese passive sentences. If passives in Chinese do not have the above structure, it is hard to explain why the blocking effect in the bei sentence is not strong. As shown in (115), the blocking effect for ziji is stronger when the first/​second person pronouns are in object positions or other adjunct positions. If the present analysis is not assumed, there is no reason why the blocking effect formed by the bei nominal should be weak, given that it is also an agent. Since both ba nominals and bei nominals exhibit the weak blocking effect when they are first/​second person pronouns, they must have properties in common. If we assume that control structures are involved in both ba and bei sentences and that the P-​engine can choose the controlled Pro as the candidate, we can explain why the blocking effect found in (115) is stronger though the blockers are not possible antecedents in these sentences. This is because in these sentences, the first/​second person pronouns do not form a chain with controlled Pro’s, and hence, they themselves will be picked up by the P-​engine. In the following sentence, the blocking effect is also weak because the first/​second pronoun forms a chain with a Pro. (126)  Zhangsani yiwei Lisij hui quan 

  nik [Prok tou   ziji?i/​j/​k  de

Zhangsan think Lisi will persuade you   

cast

self 

piao].

  DE  vote

‘Zhangsan thinks that Lisi may persuade you to vote for him/​yourself.’ In the above sentence, the blocking effect is also not strong when ziji refers to the matrix subject. If the sentences in (121) are treated on a par with the above sentence, a unified account can be achieved. Notice that if the present analysis is adopted, the reason why the bei nominal can be the antecedent of the reflexive is explained. This is because it controls a Pro that can antecede the reflexive. If (121b) does have a structure as (125), the reason why some native speakers feel that there is no blocking effect in it is explained. In (121b), as represented in (125), the candidate set produced by the P-​engine is {Zhangsan, Proj, Proi}.

158

158 Reflexives The candidate set produced by the R-​engine is {Lisi, ni}. Note that Pro will not be selected by the R-​engine since Pro has no reference. The union of these two sets produces the candidate set {Zhangsan, Lisi, nij, Proj, Proi}. Since the minimal CFC containing the candidate set and the reflexive is the whole sentence, the binding domain of the reflexive is the whole sentence. Since Zhangsan, Proj, and Proi are subjects and thus prominent in PR', and since Proj and Proi can obtain reference from their controllers, ziji can be bound to either Zhangsan, Lisi, or ni in (121b). If it is co-​referential with Zhangsan, it is directly bound by it. If it is co-​referential wtih Lisi or ni, it is indirectly co-​ indexed with them via the controlled Pro. Native speakers’ judgments on sentences in (121) vary because there are two strategies available. For those speakers who adopt the Pro strategy, they do not feel the existence of the blocking effect. For those who do not adopt the Pro strategy, they do feel that the blocking effect exists in these sentences. Note that since the blocking effect is derived from the prominence of NPs, it can be assumed that it is possible for the strength of the blocking effect to be weakened in some cases. This is exactly what happens in the following sentences: (127)  a. Ta pa wo chaoguo

ziji.

  he fear I surpass  

self

  ‘He is afraid that I may surpass him.’ b. Zongtong qing wo zuo zai ziji de pangbian.   president ask  I   sit  at self DE side   ‘The president asked me to sit beside him.’   (Xu 1993: 136) (128) a. Tai pa  nij 

chaoguo zijii/​j  de erzi.

  he fear you  surpass   self 

DE son

  ‘He is afraid that you may surpass your/​his son.’ b. Zongtongi qing woj zuo zai zijii/​j de taitai pangbian.   president  ask  I  sit  at self DE wife side   ‘The president asked me to sit beside my/​his wife.’ Sentences like (127b) and (128b) involve control structures. In these sentences, the verb qing “ask” is a control verb, and its object controls the empty subject of the embedded clause. For instance, (127b) can be represented below: (129) Zongtongi qing woj [Proj zuo zai zijii/​j  de   president  ask  I   

pangbian].

  sit  at self   DE side

 159

Prominence and locality 159 In (129), the candidate set produced by the P-​engine is {zongtong, Pro}, and the one produced by the R-​engine is {zongtong, wo}. The union of these two sets is {zongtong, wo, Pro}. Since wo is co-​indexed with Pro, and ziji can refer to either the matrix subject zongtong “president” or Pro, both zongtong and wo can be the antecedent of ziji. Note that in (127b), ziji does not prefer to be bound to Pro which is controlled by wo because their co-​reference does not conform to our world knowledge. In (128b), ziji can be bound to the matrix object controlled Pro because their co-​reference is not in conflict with our knowledge of the world. In (127a) and (128a), ziji is bound across the first/​person pronoun. This means that the prominence defined by the first/​second person pronoun is not absolute. The prominence of the embedded first/​second person pronoun subject in these sentences is first reduced by the embedded predicate which is termed as irreflexive predicate in Pan (2001) and then surpassed by the prominence of the subject of an inherent empathy verb. Note that if the matrix verb in these sentences is replaced by a nonempathy verb, the situation may be different. (130)  a. Tai shuo woj chaoguo-​le  ziji?i/​?j.   he say  I  

surpass-​ASP self

  ‘He said that I had surpassed him/myself.’ b. Tai shuo  nij    chaoguo-​le 

ziji?i/​j  de erzi.

  he say  you  surpass-​ASP self 

DE son

  ‘He said that you had surpassed your/​his son.’ In (130), it becomes difficult for ziji to be bound to the matrix subject across the first/​second person pronoun since, although the prominence of the embedded subject is reduced by the semantics of the embedded predicate, the matrix subject is not more prominent than the embedded subject. Under the present analysis, in (127a) and (128a), the subject of the matrix predicate is more prominent than the embedded subject in terms of the Empathy Hierarchy (EH), the highest hierarchy in PR. Hence, ziji can be bound to the matrix subject in these sentences. For instance, in (128a), the candidate set chosen by the P-​engine is {ni}, and the candidate set chosen by the R-​engine is {ta, ni}. The matrix subject ta “he” is included by the R-​engine because it appears higher in PR. The union of these two sets results in the set {ta, ni}. Since both the matrix subject and the embedded subject appear higher in PR' in their derivational cycles, both of them can antecede ziji. Now, consider the following examples, repeated from (62a) and (97d): (131) a. Zhangsani shuo woj de   jiao’ao hai-​le    ziji*i/​j.   Zhangsan say   I  DE  pride  hurt-​ASP self   ‘Zhangsan said that my pride hurt myself.’   (Huang and Tang 1991: 269)

160

160 Reflexives b. Johni shuo [Billj de  xiaocongming]k hai-​le    John say   Bill DE 

  ta-​ziji i/​j/​*k.

little-​trick      hurt-​ASP 

he-​self

  ‘John said that Bill’s little trick hurt him.’  

(Pan 1998: 785)

In (131a), the candidate set produced by the P-​engine is {wo}, and the candidate set produced by the R-​engine is {wo, jiao’ao}. The union of these two sets produces the candidate set {wo, jiao’ao}, and the binding domain is the embedded clause. The matrix subject is not chosen by the R-​engine because the subcommander wo “I” is the most prominent NP. Wo is more prominent than the embedded subject jiao’ao “pride” in AH. Although jiao’ao appears higher than wo in SH and CH, SH and CH are ranked lower than AH in PR. Hence, wo is more prominent than jiao’ao since wo appears higher than jiao’ao in a higher hierarchy. Notice that wo is also more prominent than the matrix subject Zhangsan in (131a) since it appears higher than Zhangsan in CH. Although Zhangsan appears higher than wo in SH, SH is ranked lower than CH. Since there is no other NP that is more prominent than wo in (131a), wo is chosen as the most prominent NP by the R-​engine. It is obvious that wo is the only NP that can antecede ziji in (131a) since it is the most prominent NP in its derivational cycle and there is no closer NP that is more prominent than it in the binding domain. The matrix subject cannot antecede the reflexive since it is not included in the binding domain. Now, consider (131b). Pan (1998) claims that in (131b) both the subcommander and the matrix subject can antecede the reflexive. If there is no difference between the embedded subcommander and the matrix subject in anteceding the reflexive, Pan’s claim will present a problem for the present analysis. However, since the present approach aims to work out a preferred interpretation for reflexives, and in fact, there exists a difference between the embedded subcommander and the matrix subject in anteceding the reflexive in (131b), his claim would not present a real problem for the present analysis. According to the native speakers’ judgment, it is difficult for the compound reflexive to refer to the matrix subject across the subcommander in (131b), and there is no substantial difference between ziji in (131a) and ta-​ziji in (131b) in their binding possibilities. This fact suggests that there is something in common between (131a) and (131b). In (131b), the most prominent NP for the R-​engine is also the subcommander contained in the embedded subject.

3.5  Further discussion 3.5.1  The derivational cycle According to the definition of the prominent NPs given in (113), the reflexive in Chinese can be bound to an accessible prominent NP that appears the

 161

Prominence and locality 161 highest in PR' in a derivational cycle if there is no closer NP that appears higher than it in PR' in the binding domain of the reflexive, and the reflexive can also be bound to an NP that is less prominent than a closer NP in PR' in the binding domain of the reflexive but appears the highest in a given derivational cycle in PR' if and only if this NP bears the speaker’s empathy. The notion of the derivational cycle can be defined below: (132)  A derivational cycle is a domain upon which rules operate (e.g., the clause). One may ask why the notion of the derivational cycle is needed. The idea of including the notion of the derivational cycle in defining the possible antecedents for the reflexive is inspired by Pan’s (2001:  297) Condition for Self-​Ascription Ziji. The basic idea is that the prominent antecedent for the reflexive can also result from the comparison of the prominence of the relevant NPs according to the relevant prominence hierarchies within the derivational cycle (i.e., the clause). Consider the following sentences: (133) a. Lisii gaosu Markj Suek pian-​le  

zijii/​*j/​k.

  Lisi tell   Mark Sue cheat-​ASP self   ‘Lisi told Mark that Sue cheated herself/​him.’ b. Wuqing de   shishii gaosu Markj Suek  pian-​le    cruel  

  ziji*i/​j/​k.

DE  fact   tell  Mark Sue   cheat-​ASP  self

  ‘The cruel fact told Mark that Sue cheated herself/​him.’   (Pan 2001: 301) c. Zhangsani de   xin  biaoshi [Lisij  hai-​le    ziji*i/​j].   Zhangsan DE  letter indicate Lisi   hurt-​ASP self   ‘Zhangsan’s letter indicates that Lisi hurt himself.’   (Huang and Tang: 267) d. Zhangsani de xin  anshi Lisij  hai-​le    Zhangsan DE letter hint  Lisi

  zijii/​j.

harm-​ASP self

  ‘Zhangsan’s letter hinted that Lisi harmed him/​himself.’   (Pan 2001: 282) In (133a), ziji can be bound to either the embedded subject or the matrix subject, as these two subjects are the most prominent NPs in PR' in their derivational cycles (i.e., the embedded clause and the matrix clause). In (133a) the reflexive cannot be bound to the matrix object, because the matrix object is less prominent than the matrix subject in its derivational cycle in PR'. Notice that, although the reflexive cannot be bound to the matrix object in

162

162 Reflexives (133a), it can be bound to it in (133b), according to Pan (2001). The contrast between these two sentences is just predicted by the present analysis. In (133b), ziji can be bound to the matrix object because the matrix object is prominent in its derivational cycle (i.e., the matrix clause), given that the matrix subject, as inanimate NP, is less prominent than it in PR'. Notice that, although in (133b), the matrix object Mark is less prominent than the closer NP Sue in PR' in the binding domain, it is an NP that the speaker can empathize with, because it can be interpreted as an indirect experiencer, as suggested by Xu (1994). Hence, according to the definition of the prominent NP given in (113B), it can antecede the reflexive. In (133c), it is difficult for the reflexive to refer to the matrix subcommander, though the latter is prominent in PR' in its derivational cycle. The matrix subcommander cannot antecede the reflexive because there is a closer NP (i.e., the embedded subject Lisi) that appears higher than it in PR' in the binding domain. Notice that, according to (113B), an NP prominent in its own derivational cycle but less prominent than a closer NP in PR' in the binding domain can antecede the reflexive if and only if it bears the speaker’s empathy. But in (133c), the matrix subcommander does not bear the speaker’s empathy since the verb biaoshi “indicate, show” can only indicate the speaker’s objective judgment, but not subjective empathy. In (133d), the reflexive can refer to the matrix subcommander. The matrix subcommander Zhangsan in (133d) is prominent in its own derivational cycle. Although it is less prominent than the closer NP Lisi in PR' in the binding domain of the reflexive, it can antecede the reflexive since it can receive the speaker’s empathy. The use of the verb anshi “hint” indicates the speaker’s empathy with the subcommander. Notice that, although the subcommander is not the subject of the verb anshi in syntax, it is in fact associated with the agent role of the verb in semantic interpretation. The subject xin “letter” can be interpreted as an instrument in semantics since the verb anshi requires an animate agent. Hence, the meaning of the sentence is Zhangsan hinted in (with) his letter that Lisi harmed him. It should be pointed out that different languages may have different definitions of prominent NPs in assigning the possible antecedents to the reflexive. For instance, the prominent NP that can antecede the reflexive in English is the c-​commanding NP. In (134a), both the subject and the object can antecede the reflexive as they are both prominent in terms of c-​command: (134)  a. Johni talked to Billj about himselfi/​j.   (Kuno 1987: 154) b. Lisi gaosu le    Lisi tell  

  ta  ziji de  fenshu.

ASP  him self DE grade

  ‘Lisi told him his own grade.’

 163

Prominence and locality 163 However, in the Chinese sentence (134b), the preferred antecedent for the reflexive is the subject since it appears higher than the object in PR'. Notice that the notion of prominence can account for not only the blocking effect but also the subject-​orientation exhibited in Chinese reflexive binding. Reflexives tend to choose subjects as their antecedents because they are more prominent. If elements other than subjects are also prominent, as shown in (133b) and (133d), they can also be chosen as antecedents. 3.5.2  The noncontrastive and contrastive compound reflexives As mentioned in section 3.4.1, according to Pan (1998), the compound reflexive in Chinese can be bound to an adjunct antecedent in the following sentences. (135)  a. Wo wei  Johni  zhaodao-​le ta-​zijii de zuoye.   I   for   John   find-​ASP  he-​self DE homework   ‘I found John’s homework for him.’ b. Lisii

zai Zhangsanj jiali

tingshuo ta-​zijii/​j

de wenzhang fabiao-​le.

  Lisi   at   Zhangsan home hear    he-​self   DE article 

publish-​ASP

  ‘At Zhangsan’s home Lisi heard that his article was published.’ c. Lisii cong Zhangsanj nar   tingshuo zanyang ta-​zijii/​j de   Lisi from Zhangsan 

there hear 

wenzhang

praise   he-​self DE article

  fabiao-​le.   publish-​ASP   ‘Lisi heard from Zhangsan that the article that praised him was published.’  

(Pan 1998: 781–​782)

However, according to my and other native speakers’ judgment, the compound reflexive cannot refer to the adjunct antecedents in the above sentences if used noncontrastively. Note that according to Xue et al. (1994), the compound reflexive in the following sentence cannot refer to the object. (136) Zhangsani gaosu Lisij ta-​zijii/​*j Zhangsan tell  

de shenshi.

Lisi he-​self  DE life-​story

‘Zhangsan told Lisi the story of his life.’ (Xue et al. 1994: 435) If it cannot refer to the object, one wonders how it is possible for it to refer to the adjunct in (135), given that an adjunct is obviously less prominent than an object since adjuncts do not c-​command the reflexive. Notice that in (135a), the adjunct is the only choice for the reflexive since the subject is incompatible

164

164 Reflexives with the reflexive in person feature. If the matrix subject is replaced by a third person NP, the reflexive does not tend to refer to the adjunct, as shown below: (137)  Lisii wei

Johnj zhaodao-​le ta-​zijii/​??j de zuoye.

Lisi for   John find-​ASP

he-​self DE homework

‘Lisi found his homework for John.’ Here, the question to ask is why the grammatical judgments vary. I think that the answer lies in the fact that the compound reflexive in Chinese can be used as an intensive pronoun. Note that, when the compound reflexive like ta-​ziji is used as an intensive pronoun, it can be used deictically and refer to a contextual antecedent. (138) Ni   you

qu  wen go

ta-​ziji   ba.

ask   he-​self SFP

Lit. ‘You may go to ask him HIMSELF.’ Since the compound reflexive like ta-​ziji can be used as an intensive pronoun, it is quite possible that the native speakers who can get the co-​referential reading between the reflexive and the adjunct in (135) and (137) may take the contrastive reading of the reflexive as the noncontrastive reflexive reading. Pan (1998) discusses the distinction between the contrastive compound reflexive and the noncontrastive one in Chinese and claims that the contrastive ta-​ziji is stressed in the ziji part, whereas the noncontrastive ta-​ziji is pronounced neutrally. Obviously, this phonological criterion is not reliable since the intensive pronoun in (138) can also be pronounced neutrally, and the noncontrastive ta-​ziji can also be pronounced with stress on ziji. Since he does not offer any reliable syntactic means to make a distinction between the noncontrastive compound reflexive and the contrastive one in Chinese, Pan (1998) fails to show that the compound reflexive is not used as an intensive pronoun when it refers to the adjunct antecedents in (135). In fact, there is a way to make a syntactic distinction between the noncontrastive compound reflexive and the contrastive one in Chinese. When I ask a native speaker of Beijing dialect if it is possible for the compound reflexive to refer to the adjunct antecedents in sentences given in (135), she tells me that if the adjunct is intended as the antecedent, she prefers to replace the compound reflexive with a form like NP+ziji. For instance, she prefers to say the following if the adjunct is treated as the antecedent: (139) Zhangsan wei Lisii  zhaodao-​le Lisi-​zijii de  zuoye. Zhangsan for Lisi   find-​ASP

Lisi-​self DE homework

Lit. ‘Zhangsan found Lisi’s own homework for Lisi.’

 165

Prominence and locality 165 Notice that in the following sentence, the noncontrastive ta-​ziji cannot be replaced by an NP+ziji form, even though ta-​ziji is LD bound. (140)  Lisii shuo na-​ben shu

fang zai *Lisi-​zijii

(=ta-​ziji) de  

Lisi say that-​CL book put  at  Lisi-​self  he-​self

jiali.

DE  home

Lit. ‘Lisi said that that book was put at Lisi’s own home.’ As shown in (139) and (140), when ta-​ziji refers to the adjunct antecedent, it is used as an intensive pronoun and can thus be replaced by NP+ziji, and when it refers to a prominent antecedent within its binding domain, it is used as a noncontrastive reflexive and cannot be replaced by NP+ziji. Notice that RBC will not be responsible for the binding of the contrastive reflexive. Since I tie the binding possibilities of the reflexive to the prominence of the NP in RBC, it can be seen that RBC can only work out a preferred interpretation for the reflexive since prominence is a relative rather than an absolute concept. Although the compound reflexive does not prefer to be bound to the adjunct antecedent, it can refer to it if it is made prominent. If the adjunct antecedent is made prominent, even ziji can refer to it. There are a variety of ways to make an NP prominent. Contextualization is one of the possible ways. Consider the following sentence:  

ziji de  

zhaopian  yizhi   

Lisi look-​for

self DE

photo 

(141) Lisi zhao 

  mei  zhaodao,

  all-​the-​time not  find

zuihou haishi Zhangsan wei  tai  zhaodao le   zijii de   zhaopian. finally still

Zhangsan for

he  find    ASP self DE photo

‘Lisi tried to find his photo for a long time, but did not find it. In the end, it is still Zhangsan who found his photo for him.’ 3.5.3  Multiple occurrences of ziji According to Pan (1995, 1997, 2001), the two occurrences of ziji in (142) can have their possible readings in (143a) and (143b), but not in (143c) or (143d). (142) John renwei Bill zhidao Mark ba   ziji de  shu  jie  gei-​le John think  Bill know Mark BA self DE book loan ​to-​ASP ziji   de  pengyou. self DE friend ‘John thinks Bill knows that Mark has loaned his book to his friend.’ (Pan 1997: 167)

166

166 Reflexives (143)  a. Johni renwei Billj zhidao Markk ba   zijii de   shu   jie-​gei-​le   John think Bill know  Mark BA self DE book loan-​to-​ASP   zijii de   pengyou.   self DE  friend   ‘John thinks Bill knows that Mark has loaned his book to his friend.’ b. Johni renwei Billj zhidao Markk ba   zijij de  

shu  jie-​gei-​le

  John think  Bill know  Mark BA  self DE  book loan-​to-​ASP   zijij de  pengyou.   self   DE  friend   ‘John thinks Bill knows that Mark has loaned his book to his friend.’ c. *Johni renwei Billj zhidao Markk ba   zijii de   shu   jie-​gei-​le   John think Bill know  Mark BA  self DE  book loan-​to-​ASP   zijij   de 

pengyou.

  self DE friend   ‘John thinks Bill knows that Mark has loaned his book to his friend.’ d. *Johni renwei Billj zhidao Markk ba zijij de   shu

jie-​gei-​le

  John think   Bill know Mark BA self DE book loan-​to-​ASP   zijii de  pengyou.   self DE friend   ‘John thinks Bill knows that Mark has loaned his book to his friend.’   (Pan 1997: 167) Pan (1995, 1997, 2001) further notes that, although the mixed readings are not possible in (143c) and (143d), they are possible in the following sentences: (144) a. Johni renwei Billj zhidao Markk ba zijii de shu   jie-​gei-​le   John think  Bill  know Mark BA self DE book loan-​to-​ASP   zijik de   pengyou.   self DE  friend   ‘John thinks Bill knows that Mark has loaned his book to his friend.’ b. Johni renwei Billj zhidao Markk ba zijij de  shu  jie-​gei-​le   John think  Bill know  Mark BA self DE book loan-​to-​ASP   zijik   de   pengyou.   self   DE  friend   ‘John thinks Bill knows that Mark has loaned his book to his friend.’

 167

Prominence and locality 167 c. Johni renwei Billj zhidao Markk ba   zijik de   shu 

jie-​gei-​le

  John think  Bill know  Mark BA  self DE book loan-​to-​ASP   Zijii de   pengyou.   self DE  friend   ‘John thinks Bill knows that Mark has loaned his book to his friend.’ d. Johni renwei Billj zhidao Markk ba zijik de   shu

jie-​gei-​le

  John think  Bill know  Mark BA self DE book loan-​to-​ASP   zijij de   pengyou.   self DE  friend   ‘John thinks Bill knows that Mark has loaned his book to his friend.’   (Pan 1997: 168) Pan (1995, 1997, 2001) points out that in (143c) and (143d), the mixed readings are not allowed because the reflexives involved are LD bound, and that in (144) the mixed readings can be derived because one of the reflexives involved is locally bound. Based on this fact, Pan (1995, 1997, 2001) argues that locally bound ziji and LD bound ziji are constrained by different conditions. I think that the above evidence cannot be used to support the distinction between the locally bound ziji and the LD bound ziji in these sentences. First, in all the sentences given in (144), the so-​called locally bound ziji can be replaced by a pronoun ta “he.” If it is really a local reflexive, the reason why it can be replaced by a pronoun is left unanswered.5 Second, although it is difficult to get the mixed readings in (143c) and (143d), native speakers find that the mixed readings in (144) are also not easy to obtain. The preferred interpretation for the multiple occurrences of ziji is that they refer to the same antecedent. This fact shows that the difficulties in acquiring the mixed readings in the relevant sentences may not result from the fact that one of the two occurrences of ziji is locally bound, whereas the other is LD bound in these sentences, but from some processing constraints that do not prefer the mixed readings since the mixed readings are more difficult for the interpreter to process. Consider (145) and (146), where the two occurrences of LD ziji are intended to have mixed readings with the subject of the matrix clause and the subject of the intermediate clause. According to the native speakers’ judgment, there is no difference between (144) and (145–​146) in obtaining the mixed readings. (145)  a. ?Lao nainaii zhidao Wang xiaojiej danxin  na-​xie   old granny know Wang Miss   worry    haizi cong zijij de  

jia-​li 



dai-​zou.

  child from self DE home-​in  take-​go

renk  hui ba  zijii de

that-​CL people will BA self DE

168

168 Reflexives   ‘The old granny knew that Miss Wang was worrying that those people would take her child away from her home.’ b. ? Lao nainaii zhidao Wang xiaojiej danxin   na-​xie   old granny know Wang Miss   worry 

renk   hui  ba  zijij de

that-​CL en people will BA self

  haizi cong zijii   de  jia-​li    dai-​zou.   child from self

DE home-​in take-​go

  ‘The old granny knew that Miss Wang was worrying that those people would take her child away from her home.’ (146) a .?Zuotian Zhangshengi he-​zui     

  le 

jiu,  

Yingyingj Xiaojie zhidao tai

yesterday zhangsheng drink-​drunk ASP spirits Yingying Miss

  genben jiu

know he

bu  hui jide      shi sheik ba zijii song-​dao zijij de

  at-​all  just not will remember SHI who BA self send-​to  self DE   jia-​li 



qu de.

  home-​in  go DE  

‘Zhangsheng was drunk yesterday. Ms. Yingying knew that he could not remember at all who took him back to her home.’

b. ? Zuotian zhangshengi  he-​zui      le    jiu,     yesterday Zhangsan   zhidao tai know   he

Yingyingj Xiaojie

drink-​drunk ASP spirits Yingying Miss

  genben jiu bu   hui   jiede    shi sheik  ba  zijij   at-​all  just not will   jia-​li 

remember SHI who

song-​dao zijii

de

BA self  send-​to self  DE

  qu de.

  home-​in go DE   ‘Zhangsheng was drunk yesterday. Ms. Yingying knew that he could not remember at all who took her back to his home.’

3.6  Summary In this part I have shown that it is necessary to make a distinction between not only dependency and reflexivity but also SELF and OTHER. I  have demonstrated that my feature characterization of anaphoric expressions fares better than R&R’s in capturing the referential properties of Chinese anaphoric expressions. On the basis of an assumption made in Huang and Tang (1991), I derive the LD binding properties of ziji from its lack of phi-​ features and referential features and show that it is constrained by the same binding condition that applies to the compound reflexive in Chinese. What is crucial for the present analysis is Prominence. It is Prominence that defines the binding domain of reflexives, which is consistent with the conception that

 169

Prominence and locality 169 Prominence plays a crucial role in Chinese reflexive binding, as pointed out in Pan (1998) and Xu (1999). According to Pan (1998), Chinese compound reflexives like ta-​ziji also exhibit the blocking effect. Under the present analysis, ta-​ziji is blocked by the most prominent NP selected by the R-​engine since it only lacks the referential feature. Ziji is different from ta-​ziji in that it lacks not only the referential feature but also the phi-​features and must thus depend on two feature search engines to define the most prominent NP for its binding domain. When the binding domain of ziji is defined by the union of the sets determined by the two prominent NPs selected by the two search engines, the blocking effect is derived if ziji is bound outside of its binding domain. Notice that under the present analysis, the binding domain is determined by the size of the candidate set. Since in this work the binding domain of the reflexive is defined as the minimal CFC that contains the candidate set and the reflexive, the actual size of the binding domain of the reflexive may be larger than that of the candidate set, as shown in (115a), though in many cases they have the same size.

Notes 1 According to Lidz (2001b), Near-​reflexivity is primarily a property of anaphors and derivatively a property of predicates. 2 This example was constructed by me in May 2000 and was used by Dr. Haihua Pan in reviewing a paper submitted to a journal. 3 In spoken Chinese, pronouns do not show any contrast in gender. 4 Xu (personal communication) points out that since Huang’s (1983) definition of the GC does not consider animacy, it would wrongly predict that there is a difference between ziji in (79a), (80a), (81a), and (82a), and the one in the following sentence: (i)  Lisi shuo na-​feng xin 

   hai   le   ziji.

Lisi say this-​CL letter  harm-​ASP self ‘Lisi said that that letter harmed him.’ 5 Pan (personal communication) points out that in the following sentence, the second occurrence of ziji cannot be replaced by a pronoun when it refers to the embedded subject: (i) John renwei Bill zhidao Mark wei ziji de John think  Bill know Mark for self DE

  pengyou da-​le 

ziji 

yi-​xia’r.

friend beat-​ASP self  once

‘John thinks Bill knows that Mark has beaten him for his friend.’ I agree with Pan that, when the second occurrence of ziji refers to the embedded subject in (i), it is a local reflexive. But the point is that it is also possible for the two occurrences of ziji to have mixed readings with the matrix subject and the intermediate subject in (i). For instance, in (i) the first occurrence of ziji can refer to the intermediate subject, with the second occurrence of ziji referring to the matrix subject. Notice that in the following sentences, it is even easier for the first

170

170 Reflexives occurrence of ziji to refer to the intermediate subject, and the second occurrence of ziji to refer to the matrix subject. (ii)  a. John pa  Bill hui rang Mark ti  ziji de   pengyou lai   John fear  Bill will let

da  ziji.

Mark for self DE friend  come beat self

  ‘John is afraid that Bill will ask Mark to come and beat him for his friends.’ b. John pa  Bill hui yaoqiu Mark ti  

ziji  lai 

da  ziji.

  John fear Bill will request Mark for self come beat self   ‘John is afraid that Bill will ask Mark to come and beat him for him.’ The fact that the two occurrences of ziji can obtain mixed readings with the two nonlocal subjects in (ii) shows that it may not be the distinction between the locally bound ziji and the LD bound ziji that determines whether the multiple occurrence of ziji can have the mixed readings with the relevant subjects.

 171

4  Conclusion

In the first part of this book, I have shown that wh-​questions must be properly typed, and the wh-​clausal typing is constrained by the following condition: (1)  The Pure Clausal Typing Condition (PCTC) a.  For wh-​raising languages, a clause is typed as a wh-​question iff there is a wh-​word that moves overtly into [Spec, CP] via cyclic movement without crossing any strong island. b. For Q-​particle languages, a clause is typed as a wh-​question iff there is a wh-​word interpreted with the closest C[+Q] via either the Agree operation or the choice function application. Besides, all wh-​ expressions must have an interpretation, and the wh-​ interpretation condition is given below: (2) The Wh-​Interpretation Condition A wh-​expression must be properly interpreted, and it is properly interpreted if it is interpreted in syntax or semantics. Wh-​interpretations in syntax or semantics are defined below: (3) Wh-​Interpretation in Syntax A wh-​word can be directly interpreted in syntax iff it matches with C via the Agree operation. (4) Wh-​Interpretation in Semantics A wh-​word is interpreted in semantics if it is interpreted via the choice function application or in the pair-​list reading.

172

172 Conclusion The following is the definition of the Agree operation: (5)  The Agree Operation A probe can enter into the Agree operation with a goal via Spec-​head agreement in either syntax or LF iff the following two conditions are met: a. The probe and the goal are the closest to each other before the application of the Agree operation. b.  The goal is not contained in an island. The conditions on the choice function application and the pair-​list reading are given below: (6) The Condition on the Choice Function Application A wh-​expression can be interpreted via the choice function if it can range over a set of elements that are of discrete properties or can be, at least conceptually, individuated. (7) The Condition on the Pair-​List Reading B can be interpreted with A in the pair-​list reading if (i) A can function as a set generator, and (ii) B can be paired with the members of the set generated by A. The following is the condition on set generators: (8) The Condition on Set Generators A is a set generator for B if (i) A precedes B and is more prominent than B, or (ii) A is an inherently D-​linked wh-​phrase that c-​commands the trace of B. Prominence is defined below: (9) The Prominence Hierarchy a. Subject > Nonsubject b. Argument > Nonargument c. Lexical Element > Function Element d. D-​Linked > Non-​D-​Linked It is also claimed that the interpretation of a wh-​element out of an island is regulated by the Wh Focus Condition proposed in this book. It is shown that a wh-​element can be interpreted out of an island if it can bear focus. This is

 173

Conclusion 173 so because only a wh-​element that can range over a presupposed set of entities can be focused. The Wh Focus Condition is given below. (10)  The Wh Focus Condition a. A wh-​word can bear focus iff it can range over a presupposed set of entities. b. A wh-​element can be interpreted out of an island iff it can bear focus. Since wh-​expressions can be interpreted by different applications in syntax and semantics, there should be a principle that regulates the preference of their application. This is the Principle of Economy, which is given below: (11) The Principle of Economy (PE) Choose the most economical operation whenever possible unless it is intended to cancel the interpretation associated with it. a. Syntactic Interpretation > Semantic Interpretation b. Default Interpretation > Nondefault Interpretation (where A > B means that A is more economical than B) Now, we can see that the derivation and interpretation of wh-​questions are constrained by PCTC, the Wh-​Interpretation Condition, and PE. PCTC is a pure locality condition, whereas the Wh-​Interpretation Condition and PE show the interaction between prominence and locality. In the second part of this book, I use the following features to characterize the properties of anaphoric expressions in Chinese. (12) The Characterization of Chinese Anaphoric Expressions A.

SELF

OTHER

R

uncontrolled Pro

-​

+

-​

controlled Pro

-​

+

-​

ziji

+

-​

-​

pronoun+ziji

+

-​

-​

benren

-​

-​

+

benshen

-​

-​

+

ta-​benren

-​

-​

-​

ziji-​benshen

+

-​

-​

174

174 Conclusion B.                     phi(-​features)          ref(erential) uncontrolled Pro

+

+

controlled Pro

+

-​

ziji

-​

-​

pronoun+ziji

+

-​

ta-​benren

+

+

ziji-​benshen

-​

+

I claim that the binding of the compound reflexive and the bare one is constrained by the same condition, as given below: (13)  Reflexive Binding Condition (RBC) a. A reflexive can be bound to an accessible prominent NP in its binding domain. b. The binding domain of the reflexive is the minimal complete functional complex (CFC) that contains all the members of the candidate set and the reflexive. c. A binds B iff A is co-​indexed with B, and A and B are compatible in phi-​features. d. A is accessible to B iff the assignment of the index of A to B would not violate *[γ …δ …], where γ and δ bear the same index. The difference between the compound reflexive and the bare one with respect to binding lies in the fact that they have different definitions of the most prominent NP, and it is the most prominent NP that determines the size of their candidate sets, which, in turn, determines the size of their binding domains. Since the compound reflexive lacks only the referential feature, the most prominent NP for it is determined by following hierarchies: (14) The Empathy Hierarchy (EH) The Subject of the Inherent Empathy Verb > Other Subjects (15) The Animacy Hierarchy (AH) (Chou 1992) [+Human] > [+Animate, -​Human] > [-​Animate] (16) The Syntactic Prominence Hierarchy (SPH) a. The Grammatical Function Hierarchy (GFH)   Subject/​Subject-Possessor > Object (Indirect Object > Direct Object)   > Adjunct

 175

Conclusion 175 b. The Structural Hierarchy (SH)   C-​Commander > Subcommander > Non-​C-/​Subcommander (17)  The Closeness Hierarchy (CH) [+Closer] > [-​Closer] Since we have several hierarchies, one important thing is how to apply these hierarchies to compute the prominence of NPs and determine the binding of reflexives. I think that the application of these hierarchies will be regulated by something like the ordering of rules in early generative grammar or the ranking of the relevant constraints in the Optimality Theory. The basic idea is that some of the hierarchies should be ranked higher than others in the computation of the prominence of NPs. I propose that the relevant hierarchies can be ranked below: (18) The Prominence Ranking (PR) EH > AH > GFH > CH > SH I further propose that the computing of the prominence of the relevant NPs in PR is regulated by the following algorithm. (19) The Prominence Computing Algorithm (PCA) If A appears higher than B in a higher hierarchy in PR, A is more prominent than B. I assume that the binding domain is determined by the most prominent NP chosen by the feature search engines. Since the compound reflexive lacks only the reference, its binding domain is defined by the most prominent NP chosen by the R-​engine (referential feature search engine) in PR according to PCA. The bare reflexive, different from the compound reflexive, lacks both the referential feature and the phi-​feature. Hence, besides PR, it needs another prominence hierarchy, as given in (20), to determine the prominent NP for it. (20​) The Person Hierarchy (PH) First/​Second Person Pronoun> Third Person NP Under the present analysis, the most prominent NP for ziji is determined not only by the R-​engine but also by the P-​engine (phi-​feature search engine). And it is the union of the sets related to the two prominent NPs chosen by the two engines that determines the binding domain for ziji. When the binding domain for ziji is defined, it cannot be bound outside of it. Notice that under the present analysis, the blocking effect is derived from the prominence condition, which is not a pure syntactic constraint. Hence, what the present analysis derives is only a preferred reading.

176

176 Conclusion In this work, I  have shown that prominence and locality are the two important factors in the grammar of natural language, which are often interrelated and interact with each other to determine the relevant linguistic derivation and interpretation. Although linguistic derivation and interpretation are generally constrained by prominence and locality, the relevant prominence condition and locality condition may act on the derivation and interpretation of different linguistic structures differently. It can be seen from this work that the derivation and interpretation of wh-​questions are constrained more by locality, whereas the interpretation of reflexives in Chinese is constrained more by prominence. In wh-​interpretation, prominence is used to define set generators so as to license other wh-​words in the pair-​list reading in multiple wh-​questions. In reflexive binding, prominence is used to define the binding domain of the reflexive. Although both wh-​ interpretation and reflexive binding employ the notions of prominence and locality, these two notions are defined differently in these two areas of study. One may ask if there is some universal definition of prominence and locality. It seems that if locality is viewed in terms of closeness, it can be defined universally. If it is viewed in terms of the local domain, it may be difficult for it to have a universal definition. Compared with locality, prominence is more flexible since it is difficult to achieve a unified definition of prominence even for different linguistic structures in the same language. If we probe into the nature of prominence and locality, we can see why they exhibit different properties in their definition. Locality in terms of closeness is basically a syntactic constraint, whereas prominence may not be considered as a syntactic constraint since it incorporates not only syntactic conditions like c-​command but also semantic conditions like animacy hierarchy. If all languages were constrained by the same set of syntactic conditions and if the parametric variation among languages could be derived only from syntactic conditions, the task of finding linguistic universals would become easy. However, the reality is not that ideal since what is constrained by syntactic conditions in one language may be constrained by the combination of syntactic and semantic conditions in another language. The definition of prominence in reflexive binding is such an example.

 177

Bibliography

Anagnostopoulou, E. and Everaert, M. (1999). Toward a more complete typology of anaphoric expressions. Linguistic Inquiry, 30(1), pp. 97–​119. Aoun, J., Hornstein, N., Lightfoot, D., and Weinberg, A. (1987). Two types of locality. Linguistic Inquiry, 18(4), pp. 537–​577. Aoun, J. and Li, Y-​H. A. (1993). Wh-​elements in-​situ:  Syntax or LF? Linguistic Inquiry, 24(2), pp. 199–​238. Baker, C. L. (1970). Notes on the description of English questions:  The role of an abstract question morpheme. Foundations of Language, 6(2), pp. 197–​219. Baker, C. L. (1995). Contrast, discourse prominence, and intensification, with special reference to locally free reflexives in British English. Language, 71(1), pp. 63–​101. Barss, A. (2000). Minimalism and asymmetric wh-​interpretation. In:  R. Martin, D. Michaels, and J. Uriagereka, eds., Step by Step:  Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA: MIT Press, pp. 31–​52. Battistella, E. (1989). Chinese reflexivization:  A movement to INFL approach. Linguistics, 27(6), pp. 987–​1012. Battistella, E. and Xu, Y. (1990). Remarks on the reflexive in Chinese. Linguistics, 28(2), pp. 205–​240. Berman, S. R. (1994). On the Semantics of Wh-​Clauses. New York & London: Garland Publishing. Boeckx, D. (2012). Syntactic Islands. Cambridge: Cambridge University Press. Bolinger, D. (1978). Asking more than one thing at a time. In: H. Hiz, ed., Questions. Dordrecht: D. Reidel, pp.107–​150. Bowers, J. (1993). The syntax of predication. Linguistic Inquiry, 24(4), pp. 591–​656. Bresnan, J. (1970). On complementizers:  Toward a syntactic theory of complement types. Foundations of Language, 6(3), pp. 297–​321. Bresnan, J. (2001). Lexical-​Functional Syntax. Oxford: Blackwell. Burzio, L. (1998). Anaphora and soft constraints. In: P. Barbosa, D. Fox, P. Hagstrom, M. McGinnis, and D. Pestetsky, eds., Is the Best Good Enough? Cambridge, MA: MIT Press, pp. 81–​105. Chen, P. (1992). The reflexive ziji in Chinese:  Functional vs. formalist approaches. In:  T. H.-​ T. Lee, ed., Research on Chinese Linguistics in Hong Kong. Hong Kong: Linguistics Society of Hong Kong, pp. 1–​36. Cheng, L. L.-​ S. (1991). On the typology of wh-​ questions. Ph.D. Dissertation. Massachusetts Institute of Technology.

178

178 Bibliography Cheng, L. L.-​S. (1997). On the Typology of Wh-​Questions. New York & London: Garland Publishing. Chierchia, G. (1995). Dynamics of Meaning: Anaphor, Presupposition, and the Theory of Grammar. Chicago: University of Chicago Press. Choe, J. W. (1987). LF movement and pied-​ piping. Linguistic Inquiry, 18(2), pp. 348–​353. Chomsky, N. (1973). Conditions on transformations. In: S. Anderson and P. Kiparsky, eds., A Festschrift for Morris Halle. New York: Holt, Rinehart and Winston, pp. 232–​286. Chomsky, N. (1976). Reflections on Language. New York: Pantheon. Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1986a). Knowledge of Language: Its Nature, Origin and Use. Westport, CT: Praeger. Chomsky, N. (1986b). Barriers. Cambridge, MA: MIT Press. Chomsky, N. (1991). Some notes on economy of derivation and representation. In: R. Freidin, ed., Principles and Parameters in Comparative Grammar. Cambridge, MA: MIT Press, pp. 417–​454. Chomsky, N. (1993). A minimalist program for linguistic theory. In: K. Hale and S. J. Keyser, eds., The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge, MA: MIT Press, pp. 1–​52. Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, N. (2000). Minimalist inquiries: The framework. In: R. Martin, D. Michaels, and J. Uriagereka, eds., Step by Step:  Essay on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA: MIT Press, pp. 89–​156. Chomsky, N. (2001). Derivation by phase. In: M. Kenstowicz, ed., Ken Hale: A Life in Language. Cambridge, MA: MIT Press, pp. 1–​52. Chou, X-​ L. (1992). An alternative approach to Chinese reflexives. MA report. University of Texas at Austin. Chung, S. (1994). Wh-​agreement and “referentiality” in Chamorro. Linguistic Inquiry, 25(1), pp. 1–​44. Cinque, G. (1990). Types of A'-​Dependencies. Cambridge, MA: MIT Press. Clements, G. N. (1975). The logophoric pronoun in Ewe: Its role in discourse. Journal of West African Languages, 10(2), pp. 141–​177. Cole, M. D. (2000). The syntax, morphology and semantics of null subject. Ph.D. Dissertation. University of Manchester. Cole, P. and Hermon, G. (1994). Is there LF wh-​movement? Linguistic Inquiry, 25(2), pp. 239–​262. Cole, P., Hermon, G., and Lee, C. L. (2001). Grammatical and discourse conditions on long distance reflexives in two Chinese dialects. In:  P. Cole, G. Hermon, and J. Huang, eds., Long Distance Reflexives. Syntax and Semantics Series. Waltham, MA: Academic Press, pp. 1–​46. Cole, P., Hermon, G., and Sung, L-​M. (1990). Principles and parameters of long-​ distance reflexives. Linguistic Inquiry, 21(1), pp. 1–​22. Cole, P. and Sung, L.-​M. (1994). Head movement and long distance reflexives. Linguistic Inquiry, 25(3), pp. 355–​406. Cole, P. and Wang, C. (1996). Antecedents and blockers of long distance reflexives: The case of Chinese ziji. Linguistic Inquiry, 27(3), pp. 357–​390. Comorovski, I. (1996). Interrogative Phrases and the Syntax-​ Semantics Interface. Dordrecht: Kluwer Academic Publishers.

 179

Bibliography 179 Dayal, V. (1996). Locality in Wh Quantification:  Questions and Relative Clauses in Hindi. Dordrecht: Kluwer Academic Publishers. Faltz, L. M. (1977). Reflexivization: A study in universal syntax. Ph.D. Dissertation. University of California, Berkeley. Faltz, L. M. (1985). Reflexivization: A Study in Universal Syntax. New York: Garland Publishing. Fiengo, R., Huang, J.  C.-​T., Lasnik, H., and Reinhart , T. (1988). The syntax of wh-​in-​situ. Proceedings of the 7th West Coast Conference on Formal Linguistics, pp.  81–​98. Fox, D. (1993). Chain and binding:  A modification of Reinhart and Reuland’s “Reflexivity”. Manuscript. Massachusetts Institute of Technology. Grewendorf, G. (2001). Multiple wh-​fronting. Linguistic Inquiry, 32(1), pp. 87–​122. Haiman, J. (1985). Natural Syntax. Cambridge: Cambridge University Press. Haiman, J. (1995). Grammatical signs of the divided self: A study of language and culture. In: W. Abraham, T. Givon, and S. Thompson, eds., Discourse Grammar and Typology: Papers in Honour of John W.M. Verhaar. Amsterdam: John Benjamins Publishing Company, pp. 213–​234. Hamblin, C. L. (1973). Questions in Montague English. Foundations of Language, 10(1), pp. 41–​53. Haspelmath, M. (1997). Indefinite Pronouns. Oxford: Oxford University Press. Heim, I. (1982). The semantics of definite and indefinite noun phrases. Ph.D. Dissertation. University of Massachusetts. Hoji, H. (1986). Scope interpretation in Japanese and its theoretical implications. Proceedings of the 5th West Coast Conference on Formal Linguistics, pp. 87–​101. Hornstein, N. (1995). Logical Form: From GB to Minimalism. Oxford: Blackwell. Hu, J.-​H. (1998a). Binding, predication and feature checking: A study of reflexivization within the framework of the minimalist program. Journal of Foreign Languages, 5, pp.  58–​64. Hu, J.-​H. (1998b). The syntactic study of long-​distance reflexivization in Chinese. Contemporary Linguistics, 3, pp. 33–​40. Hu, J.-​H. (2006). Focus and quantification. Paper presented at The International Symposium on Formal and Functional Studies in Chinese. PLA Foreign Languages University. Hu, J.-​ H. (2009). Focus and quantification. In:  G. Cheng and D.-Q. Liu, eds., Formal  and  Functional Studies in Chinese. Beijing:  The Commercial Press, pp.  83–​91. Hu, J.-​ H. (2018). What is new descriptivism? Contemporary Linguistics, 20(4), pp. 475–​477. Hu, J.-​H. and Pan, H.-​H. (2002). NP prominence and the Chinese reflexive ziji. Contemporary Linguistics, 4(1), pp. 46–​60. Hu, J.-​H., Pan, H.-​H., and Xu, L.-​J (2001). Is there a finite vs. nonfinite distinction in Chinese? Linguistics, 39(6), pp. 1117–​1148. Hua, D.-​ F. (2000). On wh-​ quantification. Ph.D. Dissertation. City University of Hong Kong. Huang, J. C.-​T. (1982a). Move wh in a language without wh movement. Linguistic Review, 1(4), pp. 369–​416. Huang, J. C.-​T. (1982b). Logical relations in Chinese and the theory of grammar. Ph.D. Dissertation. Massachusetts Institute of Technology.

180

180 Bibliography Huang, J. C.-​ T. (1983). A note on the binding theory. Linguistic Inquiry, 14(3), pp. 554–​561. Huang, J. C.-​T. (1991). Modularity and Chinese A-​not-​A questions. In: C. Georgopoulos and R. Ishihara, eds., Interdisciplinary Approaches to Language: Essays in Honor of S.-​Y. Kuroda. Dordrecht: Kluwer Academic Publishers, pp. 305–​332. Huang, J. C.  -​T. (1998) [1982b]. Logical Relations in Chinese and the Theory of Grammar. New York: Garland Publishing. Huang, J. C.-​T. (1999). Chinese passives in comparative perspective. Tsing Hua Journal of Chinese Studies, 29(4), pp. 423–​509. Huang, J. C.-​T. and Liu, L. C.-​S. (2001). Logophoricity, attitudes and ziji at the interface. In: P. Cole, G. Hermon, and J. Huang, eds., Long Distance Reflexives. Syntax and Semantics Series. Waltham, MA: Academic Press, pp. 141–​196. Huang, J. C.-​ T. and Tang, J.  C.-​ C. (1991). The local nature of long distance reflexive in Chinese. In: J. Koster and E. Reuland, eds., Long Distance Anaphora. Cambridge: Cambridge University Press, pp. 263–​282. Huang, Y. (1994). The Syntax and Pragmatics of Anaphora: A Study with Special Reference to Chinese. Cambridge: Cambridge University Press. Huang, Y-​H. (1984). Reflexives in Chinese. Studies in English Literature and Linguistics, 10, pp. 161–​195. Iida, M. (1992). Context and binding in Japanese. Ph.D.  Dissertation. Stanford University, Stanford, CA. Janda, R. (1985). Echo-​questions are evidence for what? Papers from the 21st Regional Meeting of the Chicago Linguistic Society. Kamp, H. (1981). A theory of truth and discourse representation. In: J. Groenendijk, T. Janssen, and M. Stokhof, eds., Formal Methods in the Study of Language. Amsterdam: Mathematical Center, pp. 277–​322. Karttunen, L. (1977). Syntax and semantics of questions. Linguistics and Philosophy, 1(1), pp. 3–​44. Kayne, R. S. (1984). Connectedness and Binary Branching. Foris: Dordrecht. Kiss, K. (1993). Wh-​ movement and specificity. Natural Language and Linguistic Theory, 11(1), pp. 85–​120. Kratzer, Agenlika. (1995). Stage-​level and individual-​level predicates. In: G. Carlson and J. Pelletier, eds., The Generic Book. Chicago:  University of Chicago Press, pp. 125–​175. Kuno, S. (1972). Pronominalization, reflexivizaiton, and direct discourse. Linguistic Inquiry, 3(2), pp. 161–​195. Kuno, S. (1987). Functional Syntax:  Anaphora, Discourse, and Empathy. Chicago: University of Chicago Press. Kuno, S. and Robinson, J. J. (1972). Multiple wh questions. Linguistic Inquiry, 3(4), pp. 463–​487. Lasnik, H. (1989). On the necessity of binding conditions. In: H. Lasnik, ed., Essays on Anaphora. Dordrecht: Kluwer Academic Publishers, pp. 149–​167. Lasnik, H. and Saito, M. (1984). On the nature of proper government. Linguistic Inquiry, 15(2), pp. 235–​289. Lasnik, H. and Saito, M. (1992). Move-​α: Conditions on Its Application and Output. Cambridge, MA: MIT Press. Law, P. (2001). Focus and disjunctive questions in Chinese. Manuscript. Freie University.

 181

Bibliography 181 Lee, T. H.-​T. (1986). Studies on quantification in Chinese. Ph.D. Dissertation.  University of California at Los Angeles. Li, N.-​C. (1991). Perspective-​taking in Mandarin discourse. Ph.D. Dissertation. State University of New York at Buffalo. Li, Y.-​F. (1993). What makes long distance reflexives possible? Journal of East Asian Linguistics, 2(2), pp. 135–​166. Lidz, J. (1995). Morphological reflexive marking: Evidence from Kannada. Linguistic Inquiry, 26(4), pp. 705–​710. Lidz, J. (2001a). Anti-​antilocality. In: P. Cole, G. Hermon, and J. Huang, eds., Long Distance Reflexives. Syntax and Semantics Series. Waltham, MA: Academic Press, pp. 227–​254. Lidz, J. (2001b). Condition R. Linguistic Inquiry, 32(1), pp. 123–​140. Lin, J-​W. (1992). The syntax of zenmeyang ‘how’ and weisheme ‘why’ in Mandarin Chinese. Journal of East Asian Linguistics, 1(3), pp. 293–​331. Liu, C-​S. (1999). Anaphora in Mandarin Chinese and binding at the interface. Ph.D. Dissertation. University of California, Irvine. Liu, F.-​H. (1986). On topic-​traces in Chinese. Proceedings of the 5th West Coast Conference on Formal Linguistics, pp. 142–​153. Maling, J. (1984). Non-​clause-​bounded reflexives in modern Icelandic. Linguistics and Philosophy, 7(5), pp. 211–​241. Manzini, M. R. and Wexler, K. (1987). Parameters, binding theory, and learnability. Linguistic Inquiry, 18(3), pp. 413–​444. May, R. (1977). The grammar of quantification. Ph.D. Dissertation. Massachusetts Institute of Technology. May, R. (1985). Logical Form: Its Structure and Derivation. Cambridge, MA: MIT Press. McCawley, J. (1994). Remarks on the syntax of Mandarin yes-​no questions. Journal of East Asian Linguistics, 3(2), pp. 179–​194. Miyagawa, S. (2001). The EPP, scrambling, and wh-​in-​situ. In: M. Kenstowicz, ed., Ken Hale: A Life in Language. Cambridge, MA: MIT Press, pp. 293–​338. Mohanan, K. P. (1982). Grammatical relations and anaphora in Malayalam. In: A.  Marantz and T. Stowell, eds., MIT Working Papers in Linguistics, Papers in Syntax. Cambridge, MA: MIT Press, pp. 163–​190. Ning, C.-​Y. (1993). The overt syntax of relativization and topicalization in Chinese. Ph.D. Dissertation. University of California at Irvine. Nishigauchi, T. (1986). Quantification in syntax. Ph.D. Dissertation. University of Massachusetts, Amherst. Nishigauchi, T. (1990). Quantification in the Theory of Grammar. Dordrecht: Kluwer Academic Publishers. Pan, H.-​H. (1995). Locality, self-​ascription, discourse prominence, and Mandarin reflexives. Ph.D. Dissertation. The University of Texas at Austin. Pan, H.-​H. (1997). Constraints on Reflexivization in Mandarin Chinese. New  York: Garland Publishing. Pan, H.-​H. (1998). Closeness, prominence and binding theory. Natural Language and Linguistic Theory, 16(4), pp. 771–​815. Pan, H.-​H. (2001). Why the blocking effect? In: P. Cole, G. Hermon, and J. Huang, eds., Long Distance Reflexives. Syntax and Semantics Series. Waltham, MA: Academic Press, pp. 279–​316.

182

182 Bibliography Pan, H.-​H. and Hu, J.-​H. (2001). An optimality-​theoretic account of Mandarin complex reflexive ‘ta-​ziji’ (s/​he-​self). Proceedings of the LFG01 Conference, pp. 380–​393. Pan, V. J. (2019). System repairing strategy at interface:  Wh-​in-​situ in Mandarin Chinese. In:  J. Hu and H. Pan, eds., Interfaces in Grammar. Amsterdam:  John Benjamins Publishing Company, pp. 133–​166. Partee, B. (1988). Many quantifiers. In: Proceedings of the Eastern States Conference on Linguistics. Columbus: Ohio State University, pp. 383–​402. Pesetsky, D. (1982). Paths and categories. Ph.D. Dissertation. Massachusetts Institute of Technology. Pesetsky, D. (1987). Wh-​in-​situ: Movement and unselective binding. In: E. J. Reuland and A. G. B. ter Meulen, eds., The Representation of (In)definiteness. Cambridge, MA: MIT Press, pp. 98–​129. Pesetsky, D. (2000). Phrasal Movement and Its Kin. Cambridge, MA: MIT Press. Pica, P. (1987). On the nature of reflexivization cycle. Proceedings of the North Eastern Linguistics Society, 17(2), pp. 483–​499. Pollard, C. and Xue, P. (1998). Chinese reflexive ziji: Syntactic reflexives vs. nonsyntactic reflexives. Journal of East Asian Linguistics, 7(4), pp. 287–​318. Pollard, C. and Xue, P. (2001). Syntactic and nonsyntactic constraints on long-​distance reflexives. In: P. Cole, G. Hermon, and J. Huang, eds., Long Distance Reflexives. Syntax and Semantics Series. Waltham, MA: Academic Press, pp. 317–​342. Pollock, J. Y. (1989). Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry, 20(3), pp. 365–​424. Progovaç, L. (1992). Relativized SUBJECT:  Long-​ distance reflexives without movement. Linguistic Inquiry, 23(4), pp. 671–​680. Progovaç, L. (1993). Long-​ distance reflexives:  Movement-​ to-​ Infl vs. relativized SUBJECT. Linguistic Inquiry, 24(4), pp. 755–​772. Reinhart, T. (1997). Quantifier scope: How labor is divided between QR and choice functions. Linguistics and Philosophy, 20(4), pp. 335–​397. Reinhart, T. (1998). Wh-​in-​situ in the framework of the minimalist program. Natural Language Semantics, 6(1), pp. 29–​56. Reinhart, T. and Reuland, E. (1991). Anaphors and logophors: An argument structure perspective. In:  J. Koster and E. Reuland, eds., Long Distance Anaphora. Cambridge: Cambridge University Press, pp. 283–​321. Reinhart, T. and Reuland, E. (1993). Reflexivity. Linguistic Inquiry, 24(4), pp. 657–​720. Reuland, E. (2001). Primitives of binding. Linguistic Inquiry, 32(3), pp. 439–​492. Reuland, E. and Reinhart, T. (1995). Pronouns, anaphors and case. In:  H. Haider, S. Olsen, and S. Vikner, eds., Studies in Comparative Germanic Syntax. Dordrecht: Kluwer Academic Publishers, pp. 241–​268. Richards, N. W. (1997). What moves where when in which language? Ph.D. Dissertation. Massachusetts Institute of Technology. Richards, N. W. (1998). The principle of minimal compliance. Linguistic Inquiry, 29(4), pp. 599–​629. Rizzi, L. (1982). Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. (1990). Relativized Minimality. Cambridge, MA: MIT Press. Ross, J. R. (1967). Constraints on variables in syntax. Ph.D. Dissertation.  Massachusetts Institute of Technology. Rudin, C. (1988). On multiple questions and multiple wh fronting. Natural Language and Linguistic Theory, 6(4), pp. 445–​501.

 183

Bibliography 183 Safir, K. (1996). Semantic atoms of anaphora. Natural Language and Linguistic Theory, 14(3), pp. 545–​589. Saito, M. (1994). Additional-​wh effects and the adjunction site theory. Journal of East Asian Linguistics, 3(3), pp. 195–​240. Sells, P. (1987). Aspects of logophoricity. Linguistic Inquiry, 18(3), pp. 445–​479. Shao, J.-​M. (1996). Xiandai hanyu yiwenju yanjiu [Studies on questions in modern Chinese]. Shanghai: East China Normal University Press. Shi, D.-​ X. (1994). The nature of Chinese wh-​ questions. Natural Language and Linguistic Theory, 12(2), pp. 301–​333. Simpson, A. (2000). Wh-​Movement and the Theory of Feature-​Checking. Philadelphia/​ Amsterdam: John Benjamins Publishing Company. Szabolcsi, A. and Zwarts, F. (1993). Weak islands and an algebraic semantics for scope taking. Natural Language Semantics, 1(3), pp. 235–​284. Tang, C.-​C. J. (1989). Chinese reflexives. Natural Language and Linguistic Theory, 7(1), pp. 93–​121. Tang, C.-​C. J. (1994). A note on relativized SUBJECT for reflexives in Chinese. In: B. Lust, G. Herman, and J. Kornfilt (eds.), Syntactic Theory and First Language Acquisition: Cross-Linguistic Perspectives. Vol. 2, Binding, Dependencies, and Learnability. Hillsdale, NJ: Lawrence Erlbaum, pp. 71–82. Tang, G. W-​L. and Gu, Y. (1998). Subject orientation in the Chinese reflexive ‘ziji’. In:  Y. Gu, ed., Studies in Chinese Linguistics. Hong Kong:  Linguistic Society of Hong Kong, pp. 195–​216. Tang, S.-​ W. (2001). A complementation approach to Chinese passives and its consequences. Linguistics, 39(2), pp. 257–​295. Thráinsson, H. (1991). Long-​distance reflexives and the typology of NPs. In: J. Koster and E. Reuland, eds., Long Distance Anaphora. Cambridge: Cambridge University Press, pp. 49–​76. Tsai, W.-​T. D. (1994a). On nominal islands and LF extraction in Chinese. Natural Language and Linguistic Theory, 12(1), pp. 121–​175. Tsai, W.-​T. D. (1994b). On economizing the theory of A-​bar dependencies. Ph.D. Dissertation. Massachusetts Institute of Technology. Tsai, W.-​T. D. (1999a). On Economizing the Theory of A-​bar Dependencies. New York & London: Garland Publishing. Tsai, W.-​T. D. (1999b). On lexical courtesy. Journal of East Asian Linguistics, 8(1), pp.  39–​73. Tsai, W.-​T. D. (1999c). The hows of why and the whys of how. Paper presented at the 8th Annual Conference of International Association of Chinese Linguistics. Wang, J.-​L., and Stillings, J.  T. (1984). Chinese reflexives. In:  X. Y. Li et  al., eds., Proceedings of the First Harbin Conference on Generative Grammar. Heilongjiang University, pp. 100–​109. Watanabe, A. (1992). Subjacency and s-​structure movement of wh-​in-​situ. Journal of East Asian Linguistics, 1(3), pp. 255–​291. Watanabe, A. (2001). Wh-​ in-​ situ languages. In:  M. Baltin and C. Collins, eds., The Handbook of Contemporary Syntactic Theory. Oxford: Blackwell, pp. 203–​225. Xu, L.-​J. (1990). Remarks on LF movement in Chinese questions. Linguistics, 28(2), pp. 355–​383. Xu, L.-​J. (1993). The long-​distance binding of ziji. Journal of Chinese Linguistics, 21(1), pp. 123–​142.

184

184 Bibliography Xu, L.-​J. (1994). The antecedent of ziji. Journal of Chinese Linguistics, 22(1), pp. 115–​137. Xu, L.-​J. (1995). Definiteness effects on Chinese word order. Cahiers de Linguistique—​ Asie Orientale, 24(1), pp. 29–​48. Xu, L.-​J. (1999). The antecedents of reflexive pronouns. In:  L.-J. Xu et  al., eds., Gongxing yu gexing [Language Universals and Particulars]. Beijing:  Beijing Language and Culture University Press, pp. 39–​158. Xu, L.-​J. and Lee, T.  H.-​T. (1990). Scope ambiguity and disambiguity in Chinese. Papers from The 25th Chicago Linguistic Society Meeting:  Part One, General Session, pp. 451–​466. Xue, P., Pollard, C., and Sag, I. (1994). A new perspective on Chinese ziji. Proceedings of the 13th West Coast Conference on Formal Linguistics, pp. 432–​447. Yang, D. W. (1983). The extended binding theory of anaphors. Language Research, 19(2), pp. 169–​92. Yu, W. X.-​F. (1991). Logophoricity in Chinese. Paper presented at the 3rd North-​ American Conference on Chinese Linguistics, Cornell University, Ithaca, NY. Yu, W. X.-​F. (2000). Chinese Reflexives. Leuven: Peeters. Zhang, Z.-​S. (1997). Focus, presupposition and the formation of A-​not-​A questions in Chinese. Journal of Chinese Linguistics, 25(2), pp. 227–​257. Zribi-​Hertz, A. (1989). Anaphor binding and narrative point of view: English reflexive pronouns in sentence and discourse. Language, 65(4), pp. 695–​727. Zubin, D. A., Chun, S., and Li, N.-​C. (1990). Misbehaving reflexives in Korean and Mandarin. In: K. Hall and J-​P. Koenig, eds., Proceedings of the 16th Annual Meeting of the Berkeley Linguistics Society. Vol. 16. General Session and Parasession on the Legacy of Grice. Berkeley, CA: Berkeley Linguistics Society, pp. 338–​352.

 185

Author index

Anagnostopoulou, Elena 111–114 Aoun, Joseph 10, 22 Baker, C. L. 1, 10, 17, 23, 127–28 Battistella, Edwin 127 Bolinger, Dwight 72 Bresnan, Joan 1, 90 Chen, Ping 127 Cheng, Lisa L.-S. 9–10, 22, 34–7, 40 Choe, J. W. 18 Chomsky, Noam 1–2, 12, 17, 21, 42–3, 88, 90, 98, 114, 117–118, 121, 124, 126, 136, 146, 155 Chou, Xiao-Li 127, 142, 174 Chung, Sandra 27 Cinque, Guglielmo 26–7, 66 Cole, Melvyn Douglas 119 Cole, Peter 10, 22, 112, 127, 131–32, 156 Comorovski, Ileana 3, 11 Dayal, Veneeta 51 Everaert, Martin 111–114 Faltz, Leonard M. 98 Fiengo, Robert 19–22 Fox, Daniel 112 Grewendorf, Günther 87–90, 94n18 Haiman, John 120 Hamblin, C. L. 46 Haspelmath, Martin 23 Heim, Irene 22 Hermon, Gabriella 10, 20, 112, 127 Hoji, H. 85 Hornstein, Nobert 65–6 Hu, Jian Hua 115, 128, 142–43

Hua, Dongfan 25, 27, 29, 36, 40 Huang, C.-T. James 3–4, 10, 12–15, 18, 23, 30–1, 45, 47–50, 55–7, 61, 82, 97, 117–118, 121, 126–30, 133, 136–41, 146, 149, 153, 155, 159, 161, 169 Huang, Yan 127 Iida, M. 127 Janda, Richard 51, 93n10 Kamp, Hans 22 Karttunen, Lauri 20, 46 Kayne, R. 93n1 Kiss, Katalin 27 Kratzer, Agenlika 40 Kuno, Susumu 12, 77–8, 127, 138, 144, 162 Lasnik, Howard 84, 123 Law, Paul 48–9, 82 Lee, Cher Leng 127 Lee, Hun-tak Thomas 1, 23 Li, N.-C./ Li Naicong 127 Li, Yafei 130 Li, Yen-hui Audrey 10, 22 Lidz, Jeffery 97–8, 102–03, 108, 110–111, 116, 169n1 Lin, Jo-Wang 25, 93 Liu, Chen-Sheng 127–28, 136–41, 153 Liu, Feng-hsi 3, 46, 57 Maling, Joan 127 Manzini, M. R. 126 May, Robert 19, 40 McCawley, James 48–9 Miyagawa, Shigeru 40, 93 Mohanan, K. P. 126

186

186  Author index Ning, Chunyan 74 Nishigauchi, Taisuke 10, 18–22, 29, 45, 81, 84–5 Pan, Haihua 1–2, 32, 45, 83, 93n8, 93n11, 94n13, 94n17, 105, 107, 109–110, 115, 122–23, 125–26, 128, 132–35, 138–39, 141–43, 145–46, 149–51, 159–67, 169, 169n2, 169n5 Pan, J.-N. Victor 42 Partee, Barbara 40 Pesetsky, David 10–11, 15, 17–18, 22, 84 Pica, P. 98, 126, 128 Pollard, Carl 127–28, 133 Progovaç, Ljiljana 127

Shi, Dingxu 3, 10, 16, 18, 22–3, 28, 32, 48, 50 Simpson, Andrew 16–17 Stillings, J. T. 126–27, 129 Stojanova, Marina 90 Sung, Li-May 112, 127 Tang, C.-C. Jane 3–4, 97, 117–118, 121, 126–30, 133, 146, 149, 159, 161, 169 Tang, Gladys Wai-lan 127 Tang, Sze-wing 155 Thráinsson, Höskuldur 123–24 Tsai, Wei-tien Dylan 10, 22, 25–30, 33, 36–7, 45, 50–1, 93n3, 93n4

Reinhart, Tanya 15, 23–4, 34, 83, 97–108, 111–117, 126–27, 169 Reuland, Eric 2, 70, 90, 97–108, 111–117, 126–27, 169 Richards, Norvin W. 88–90, 94n18 Rizzi, Luigi 26–7 Robinson, Jane J. 12, 77–8 Ross, J. R. 13 Rudin, Catherine 9–10, 94n18

Wang, Chengchi 112, 127, 131–32 Wang Jialing 126–27, 129 Watanabe, Akira 45, 84–6 Wexler, Ken 126

Safir, Ken 119–121 Sag, Ivan 127–28, 133 Saito, Mamoru 84, 88 Sells, Peter 121, 127 Shao, Jingmin 40

Yang, D. W. 126 Yu, X. F. Williams 122, 127

Xu, Liejiong 1, 3, 5n1, 15, 23, 25, 30, 46, 50, 56–7, 59, 60, 79, 93n9, 115, 127, 130, 135, 158, 162, 169 Xue, Ping 127–28, 130, 133, 149, 163

Zribi-Hertz, Anne 127 Zubin, David A. 127

 187

Language index

Bulgarian 9, 29, 34–5, 37–8, 43, 87–8, 90 Chinese 1–4, 9–11, 13, 15–16, 20, 22–3, 25–6, 28–30, 32–5, 37–41, 43–6, 48, 52–3, 55–7, 59–62, 73–4, 76, 78–80, 82, 84–7, 92, 93n9, 94n11, 94n16, 97–8, 103–111, 113–118, 120–22, 124, 126–28, 140–44, 148, 157, 160, 162–64, 169, 173, 176 Dutch 101, 110–112, 126 English 3–4, 9–11, 13, 15–16, 20, 23, 29–30, 32–9, 42–5, 53–4, 56–7, 60, 62–3, 65, 76, 78, 82–6, 92,

93n11, 94n17, 97, 108, 126, 148, 157, 162 French 119 Greek 112, 114–115 Hungarian 9, 35 Japanese 29, 40, 45–6, 84–7, 89, 94n7, 94n16 Kannada 103, 110–111 Polish 9, 35

188

Subject index

A-not-A 3, 27–8, 32, 40–1, 45, 47–50, 52–3, 57, 61–2, 79, 81–3, 94; element 3, 32, 45, 47–50, 52–3, 57, 61–2, 81–3, 94n15; operator 27–8, 47–50 Agree 2–3, 26, 43–4, 55, 57, 63, 66–8, 70–2, 74–6, 79–80, 82, 90, 93n4, 171–72; operation 2–3, 43–4, 55, 66–8, 70–2, 74–6, 79–80, 82, 90, 171–72; Generalized Agree Operation 67 Agreement (AGR) 34, 39, 43–4, 49, 54, 57, 67, 75, 93n7, 112, 132, 172; Spec-head 34, 39, 43–4, 112 anaphoric 3, 97–9, 115, 117–26, 169, 173 animacy 1, 34, 126–27, 142, 144, 169n4, 174, 176; Animacy Hierarchy (AH) 126, 142, 144–45, 148, 150, 160, 174–76 answer: full-fledged 19, 51; functional 73 barrier 22, 105–6; barrierhood 21 benren 98, 107, 109, 111, 114–115, 120–25, 128, 173–74 benshen 98, 107, 109, 114, 120–22, 124–26, 128, 173–74 binding 2–4, 10, 12, 16–17, 22–4, 36, 39–40, 42–5, 48–50, 54, 59, 62, 67, 80, 82–4, 86–7, 93n11, 97–9, 104–06, 111–113, 115–119, 122, 126–31, 133–35, 137–38, 141–50, 152–53, 155–56, 158, 160–63, 165, 169, 174–76; argument-based 98; Condition A 98, 105–06, 111–113, 116, 126; Condition B 98, 111–113; Condition C 16–17; domain 2, 4, 97, 105–06, 146–58, 150, 152–53, 155–56, 158, 160–62, 165, 169, 174–76; predicate-based 97–9, 102–03; reflexive 2–4, 97–9, 112, 115–119, 126–28, 141–50, 153–63, 169, 174–76 see also reflexive

blocking: effect 4, 97, 104, 117–118, 126–28, 130–34, 137–38, 141, 143, 146, 149, 155–58, 163, 169, 175; grammatical 131–32 c-command 1, 3, 10, 17, 22–3, 40, 49, 68, 71–2, 85–6, 111, 123, 125, 128–29, 141–42, 145–46, 148, 152, 162–63, 172, 175–76; domain 10, 22–3, 40, 49 choice function 2–3, 22, 24–5, 30, 32–4, 42, 50, 62–3, 68, 70, 74, 79–83, 89–90, 171–72; application 3, 32, 63, 68, 70, 74, 79–80, 89–90, 171–72; Condition on the Choice Function Application 68, 172 closeness 1–2, 45, 145, 147, 150–51, 154, 175–76; Condition 1–2, 45, 145, 147, 150–51, 154, 176; Hierarchy (CH) 145, 150, 155, 160, 175–76 co-argument 100, 103–06, 108, 111–113, 115, 119, 121, 123 co-indexation 17, 67, 100–02, 111–113, 155–56 co-reference 16, 121, 131, 133, 159 common ground 30, 33, 69 Complete Functional Complex (CFC) 4, 146, 150, 152, 158, 169, 174 Complex NP 13, 19–22, 29, 45, 66, 69, 88 compound word 59, 93n9 computational complexity 92 Condition R 98, 102–03, 108, 110–111 counter-indexed 112–113 cross-modular operation 91–2 crossover 16–17 CTC (The C Typing Condition) 45–6, 55, 57, 59–64, 66 cycle 97, 147–48, 150, 152–53, 156, 159–62

 189

Subject index 189 D-linking /D-linked 3, 17–18, 30, 34, 63–5, 68, 71–2, 172 de se 128, 136–37 dependability 1 dependency 29, 117–119, 121, 126, 168; see also LD Direct Discourse Representation 138, 140–41 discourse saliency 65 distributive quantifier 86 domain-widening particle 86 Doubly Filled COMP 61 echo question 3, 38–9, 46, 50–3, 57, 59–60, 66, 84–5, 93n10–11, 94n12 empathy 144, 148, 159, 161–62, 174; Hierarchy (EH) 144, 145, 148, 150, 159, 174 Empty Category Principle (ECP) 12, 14–16, 18, 22–3, 29, 47–8, 56–7, 61, 63, 88 external speaker 138–41 familiarity 27 feature: checking 67, 84; clash 46–47, 49, 62; conflict 83, 87, 128, 132–33; incompatibility 52; percolation 21, 131–32; focus 31; phi- 4, 97, 114, 116–119, 121–22, 126, 144–47, 149, 155–56, 169, 174–75; primitive 117–118; Q feature[±Q] 2, 40–43, 45, 47–48, 53–54, 76–77, 80, 84, 171; referential 4, 97, 118–119, 121–22, 126, 143, 146–47, 149, 169, 174–75; wh-feature/[±Wh] 34, 37–38, 40–45, 47, 54, 62, 75, 83–84, 93 finiteness 115 focus: contrastive 30–32, 52; marker 30–32, 51–52, 59, 94; reading 52; semantic 30–33 function element 32–33, 44, 68, 172 GB 12, 14, 59 General Condition on A-chains 101 Governing Category (GC) 1, 4, 126, 127, 136 government 14, 88; antecedent 15, 101; proper 14, 75 Grammatical Function Hierarchy (GFH) 142, 144–45, 174 head: head-movement 48, 131; head-governed 57, 61; head-final 75

incorporation 3, 32, 48, 62, 81–82, 112–113, 115 individual answer 73 individuation 79–80 instrumental reading 26, 50 internal speaker 138, 140–41 interpretation: default 70–72, 173; discourse 91; semantic 66, 70, 91, 102, 162, 173; syntactic 67, 70, 173; wh-interpretation 2, 10–11, 24, 29–30, 43, 45–6, 63–4, 66, 68–70, 75, 84, 90, 171, 173, 176 island: adjunct 90; strong 2–3, 66–7, 76, 80, 89, 171; syntactic 19; wh-island 3, 11, 45–6, 48–49, 55–7, 60–1, 75, 78, 84–5, 89; wh-island constraint 6, 49, 55–6, 60 LD: binding 4, 105, 117, 122, 126–27, 131, 133, 137–38, 141, 149, 169; dependency 121; reflexive 127, 136–38; see also reflexive LF 3; movement 11–14, 17–18, 22, 25, 45, 47, 49–50, 56–7, 61, 63, 67, 82, 126–27; representation 10, 19, 24, 56; wh-movement 11, 16 linear order 94 local anaphor 99 local antecedent 135 locality 1–5, 10, 18, 29, 30, 44, 63, 66, 67, 92, 97, 127, 134, 141, 173, 176; condition 5, 11–12, 14, 18, 22–3, 46, 55, 92, 122, 173; effect 11, 23 logophor 106, 136–47 logophoric: blocking 131 logophoricity 127 long-distance: anaphor 99; antecedent 98; binding 104, 126; see also LD binding; wh-movement 56 manner reading 26–7 marked intonation 85 marked order 72 matching relation 44, 69, 71 m-command 12 means reading 50 Minimal Link Condition (MLC) 2, 12, 63 movement: covert 14; LF 45, 47, 49–50, 56–7, 61, 63, 67, 82, 126, 127; optional 87; overt 12–14, 63, 67 multiple wh-question 2–3, 9–12, 39, 45, 53, 57, 64–6, 70–4, 76, 83–5, 87, 90, 176

190

190  Subject index nominal island 26–8 nominality 26 operator: binding 22, 39, 42–3, 45, 49; operator-variable relation 10; operator-variable structure 36–7; Q operator 3, 23, 28, 38–41, 45, 47–50, 54, 61–2, 66, 68, 70, 75–6, 80–3, 87; existential Q operator 24, 83; Optimality Theory (OT) 145, 175 pair-list reading 2–3, 65, 68, 70–4, 76–8, 84–7, 90–2, 94, 171–2; condition on 68, 172 particle typing 42 see also typing PCTC (Pure Clause Typing Condition) 2–3, 66, 75–6, 79–84, 87, 89, 92, 171 see also typing Person Hierarchy (PH) 145, 175 perspectivity 127, 138–40 pied-piping 19–22 predicate: complex 112, 115–116; irreflexive unfound; predicate-based theory 97–9, 102–03, 117; reflexive unfound; semantic 99–100, 113; syntactic 99–100, 113 presupposition 33 Principle of Economy (PE) 44, 64, 70, 91, 121, 173 Principle of Full Interpretation 90 Principle of Locality 1 Principle of Minimal Compliance (PMC) 89 Principle of Prominence 1 probe 41, 43–4, 55, 67, 71, 75–6; local 55; matrix 55, 76; probe-goal 43, 44 prominence 92, 97, 127–28, 142–46, 148–50, 158–59, 161, 163, 165, 173, 175–76; Hierarchy (PH) 44, 68, 172; Computing Algorithm (PCA) 175; Ranking (PR) 145, 175; Syntactic Prominence Hierarchy (GFH) 142, 144–45, 148, 150, 153, 155, 174–75 Q-particle 38–45, 52–3, 58, 61–2, 74–5, 80, 93; language 2, 45, 48, 66, 68, 80, 171 quantificational force 17, 20, 23, 34–6 quantifier phrase (QP) 85–6, 94 Quantifier Raising 14, 127 question: A-not-A 3, 40–1, 48, 50, 82; direct 53–4, 56–7, 59, 77–8, 82; disjunctive 40–1, 82; indirect 47, 53–4,

57, 60, 76, 78; multiple wh-question 2–3, 9–12, 39, 45, 53, 57, 64–6, 70–4, 76, 83–5, 87, 90, 176; original 3, 39, 46, 50–3, 57–60, 66, 85, 93; wh- 2–4, 9–12, 21, 31, 34, 38–45, 49–50, 52, 57–8, 60, 62, 64–6, 70–6, 78, 80–5, 87–90, 92, 94, 171, 173, 176; yes-no 41, 59 referential: dependence 114; features 4, 97, 118–119, 121–2, 126, 143, 146–47, 149, 169, 174–75; wh-adjunct 25, 27 referentiality 25–7, 33, 114 reflexive: bare 4, 97–8, 107–08, 115, 117–118, 120, 126, 128, 132, 142–44, 146, 175; complex 98, 104–05, 107–08, 114; compound 1, 4, 97, 114, 117–118, 121, 125, 128, 140–44, 147, 160, 163–65, 169, 174–75; i-reflexive 113; local 136–38, 153, 167, 170; nearreflexive 98, 102–03, 108–111, 169; noncontrastive compound 97, 128, 164; pure-reflexive 98, 102–03, 108, 110–111; binding 2, 7, 97, 99, 112, 118, 141, 143, 146, 163, 169, 174, 176 see also binding; Binding Condition (RBC) 4, 143, 146, 174; interpretation 70, 102–03, 110–111; reflexive-marked 100–03, 105, 108, 111, 113; reflexivemarking 100, 103, 115; simplex 98, 104–05, 108, 114, 128 reflexivity 99, 102–03, 108, 117–119, 121, 168–69 reflexivization 98, 101, 107, 121 reflexivizing function 98–99, 102–03, 108, 117 R-expression 16–7, 99, 117, 123–5 set: alternative 30–33, 59; candidate 4, 97, 146–47, 149–60, 169, 174; individuated 3, 33; open 33, 88; presupposed 33–34, 173 set generator 2, 65–68, 71–74, 76, 176; Condition on the 68, 172 scope 1, 3, 17–19, 20–22, 24–25, 28–29, 47–49, 50, 57, 61, 77; interpretation 20, 25–26, 29, 45, 50, 77; scopemarking 34; scope-taking 1, 53, 85 SE anaphor 98–99, 101, 104–05, 107, 115, 126 search engine 4, 97, 144, 146–48, 150, 154, 169; Condition on the Operation of the (COSE) 147; feature 4, 97, 146,

 191

Subject index 191 149, 169, 175; phi-feature (P-engine) 4, 146–57, 159–60, 175; referential feature (R-engine) 4, 146–48, 150–53, 155–56, 158–60, 169 self-ascriber 109, 110 self-ascription 128, 133–36; Condition for Self-Ascription Ziji 134, 161 self: duplicated 109, 110; SELF anaphor 98–100, 102–06, 111–112, 115–116, 126 sentence-final particle 74, 109, 164 sentential subject 20, 26; Constraint 13; island 15 shared knowledge 30, 33 specificity 27 stress 30, 85, 164; contrastive 93–94n11 Structural Hierarchy (SH) 145, 175 subcategorization 47, 53–7, 59, 61–2, 75, 77, 94 subcommand 128–30, 137, 140–43, 145–46, 150, 160, 162, 175 Subjacency 12–14, 18, 21–33, 50, 56, 89, 90 Superiority Condition 12, 63 Superiority Effect 11–12, 14, 17 ta-ziji 4, 98, 104–06, 108–10, 114, 116, 121–22, 127–28, 140–43, 160, 163–65, 169 Tensed-S Constrain 99 topic 18, 85–6 TOPICALITY 127 typing 2–3, 11, 34, 37, 39, 40, 42–6, 55, 62, 64, 66–7, 70, 75, 79, 80, 84, 91–3, 171; see also particle, PCTC universal quantifier 20, 23, 87 vacuous binding 82 vacuous quantification 40, 54

variable 49; free 10; function 24, 83–4; individual 24; wh-variable 24, 29, 34, 59 weishenme 3, 14–6, 21, 23, 25–6, 28, 30–1, 33, 41, 44, 56–7, 61, 65, 69, 74, 79, 80–1, 83, 92–3 wh-adjunct 25, 27, 29, 33, 40, 65, 74, 88–90 wh-agreement 43 wh-cluster 87–9 wh-effect 3, 11, 70, 87, 89 wh-element 3, 10–1, 28–30, 33–5, 38–9, 40, 43–4, 49, 63–7, 84, 87–8, 92–4, 172–3 wh-feature matching 42 Wh-Feature Strength 34 Wh Focus Condition 33, 172–73 wh-fronting 35, 38–9, 87–8, 90, 94 wh-in-situ 9, 12–22, 24–5, 32, 37–9, 42–3, 47, 49, 53, 62–3, 65, 68, 83–5 wh-interpretation 2, 10–1, 24, 29–30, 43, 45–6, 64, 66, 68–70, 75, 84, 90, 171, 173, 176; Condition 70, 171, 173 wh-island effects 3, 11, 45–6, 55, 78, 84–5, 89 wh-movement 2, 9, 10–12, 16, 18, 34, 61, 84, 86–7; Condition on 67 wh-operator 20, 27, 29, 34 wh-quantification 11, 17, 26, 34, 85 wh-raising 39, 42, 45, 70; typing 42–3; language 2, 45, 66, 80, 171 zenme 15–6, 21–3, 25, 28, 32–3, 57, 79, 93 zenmeyang 25–7, 29, 32–3, 50–1, 93 ziji 4, 97–8, 104–111, 114–118, 120–22, 124–62, 164–70, 173–75 ziji-benshen 109, 121–22, 124–26, 173–74 zishen 109, 114