@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix dc: . @prefix skos: . vivo:departmentOrSchool "Arts, Faculty of"@en, "Linguistics, Department of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Li, J. Cora R."@en ; dcterms:issued "2009-03-25T23:09:28Z"@en, "1997"@en ; vivo:relatedDegree "Master of Arts - MA"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description """This paper studies the nature of the BEI-construction in Cantonese, with Mandarin as the standard language of comparison. Although the BEI-construction has been much studied in Mandarin, the same in not true for Cantonese. Although this construction has traditionally been termed a "passive", I will show that it can have a different range of semantic interpretations in Cantonese. I argue that BEI is not confined to passive, but is used under certain circumstances to form a causative construction as well. The differences in behaviour between passive-BEI and causative-BEI can be seen in tests with anaphoric binding. I conclude that while the passive structure is mono-clausal, the causative structure must be bi-clausal. The Cantonese BEI-constructions have an obligatory agent-phrase which cannot be dropped. This differs from Mandarin and the challenge is to find an account for this phenomenon, especially if we are to claim that this construction is a passive. The optionality of the agent phrase is characteristic of passives and yet Cantonese deviates from this norm. I argue that passive in Cantonese is a syntactic process and predict that only transitive verbs may participate in this construction. I utilize the universal v-VP structure on transitive verbs, proposed by Chomsky (1995), to guarantee that the external theta role must be retained. I also examine the much debated status of BEI which is used in the BEI-construction. Although this construction can be used to derive both a passives and a causatives, it does not necessarily mean that two separate BEIs must be posited. I conclude that BEI can be treated as a category-neutral element which can interact in both causative and passive structures. To support this proposal I appeal to the functional versus lexical distinction of categories and projections."""@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/6542?expand=metadata"@en ; dcterms:extent "5426714 bytes"@en ; dc:format "application/pdf"@en ; skos:note "BEI A N D T H E P A S S I V E IN C A N T O N E S E by J. Cora R. L i B . A . , The University of British Columbia, 1995 A THESIS S U B M I T T E D IN P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F A R T S in T H E F A C U L T Y O F G R A D U A T E STUDIES (Department of Linguistics) We accept this thesis as conforming to the required standard. T H E U N I V E R S I T Y O F BRITISH C O L U M B I A October 1997 © J . CoraR. L i , 1997 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of ' iL/A^um-l-ic \\ The University of British Columbia Vancouver, Canada Date DE-6 (2/88) Abstract This paper studies the nature of the BEI-construction in Cantonese, with Mandarin as the standard language of comparison. Although the BEI-construction has been much studied in Mandarin, the same in not true for Cantonese. Although this construction has traditionally been termed a \"passive\", I will show that it can have a different range of semantic interpretations in Cantonese. I argue that BEI is not confined to passive, but is used under certain circumstances to form a causative construction as well. The differences in behaviour between passive-BEI and causative-BEI can be seen in tests with anaphoric binding. I conclude that while the passive structure is mono-clausal, the causative structure must be bi-clausal. The Cantonese BEI-constructions have an obligatory agent-phrase which cannot be dropped. This differs from Mandarin and the challenge is to find an account for this phenomenon, especially if we are to claim that this construction is a passive. The optionality of the agent phrase is characteristic of passives and yet Cantonese deviates from this norm. I argue that passive in Cantonese is a syntactic process and predict that only transitive verbs may participate in this construction. I utilize the universal v-VP structure on transitive verbs, proposed by Chomsky (1995), to guarantee that the external theta role must be retained. I also examine the much debated status of BEI which is used in the BEI-construction. Although this construction can be used to derive both a passives and a causatives, it does not necessarily mean that two separate BEIs must be posited. I conclude that BEI can be treated as a category-neutral element which can interact in both causative and passive structures. To support this proposal I appeal to the functional versus lexical distinction of categories and projections. ii Table of Contents PaSe Abstract ii Table of Contents iii List of Tables v List of Figures vi Acknowledgments vii Chapter 1. Introduction 1 1.1 Preliminary Look at the BEI-Construction 1 1.2 Theoretical Assumptions 3 1.3 Some Notes Regarding Chinese: Mandarin versus Cantonese 5 1.3.1 Chinese Word Order 5 1.3.2 Chinese Particles 6 1.3.3 Anaphor Binding: Long-distance versus Local 9 1.4 A n Outline of the Thesis 15 Chapter 2. The BEI-Construction in Mandarin and Cantonese 17 2.1 Obligatory Agents and Colloquial Mandarin 23 Chapter 3. Types of Passives in Cantonese 27 3.1 The Indirect Passive 29 3.2 Indirect Passive as Causatives 34 3.3 Indirect Passives as Possessor-Raising 38 Chapter 4. BEI: Passive versus Causative 41 4.1 What is the Cantonese 'passive'? 41 4.1.1 The Function of BEI 46 4.1.2 BEI-passive as a Preposition 47 4.1.3 BEI-passive as a Verb • 48 4.1.4 BEI-causative as a Verb 51 4.1.5 The Ambiguity of Categories 52 4.2 What is the Cantonese 'causative'? 54 Chapter 5. Previous Analyses of the BEI-construction 57 Chapter 6. The Structure of the BEI-Construction 62 6.1 The Interaction of Time adverbials, Instrumentals and Patient: Review of L i (1990) 62 6.1.1 BEI and the Interaction with Modifiers 67 6.2 The BEI Tree 70 iii 6.2.1 Proposal 1 73 6.2.2 Proposal 2 75 6.2.3 Ritter and Rosen (1993) on 'have' 76 6.2.4 Proposal 3: Functional vs. Lexical Difference 81 6.2.5 Consequences of Proposal 3 88 6.3 A Revision: Proposal 4 90 Chapter 7: Final Comments 92 Chapter 8: A Final Comparison with Mandarin 94 Bibliography 101 A P P E N D I X 1: Exceptional \"no-movement\" Cases 104 iv List of Tables Page Table 1: Summary of BEI-passive and BEI-causative 56 List of Figures Page Diagram 1: Model of Government and Binding 3 Diagram 2: Model of the Minimalist Program 4 Figure 1: Preliminary Trees 72 Figure 2: Proposal 1: Passive-BEI 73 Figure 3: Proposal 1: Causative-BEI 73 Figure 4: Proposal 2: Passive-BEI 75 Figure 5: Proposal 2: Causative-BEI 75 Figure 6: Proposal 3: General BEI-Passive 82 Figure 7: Proposal 3: General BEI-Causative 82 Figure 8: Proposal 3: 0-passive 83 Figure 9: Proposal 3: 0-causative 83 Figure 10: Proposal 3: BEI-passive 87 Figure 11: Proposal 3: BEI-causative 87 Figure 12: Aspect and BEI-causative 89 Figure 13: Aspect and BEI-passive 88 Figure 14: Proposal 4: 0-passive 90 Figure 15: Proposal 4: 0-causative 90 Figure 16: 0-rang I jiao Passive 98 Figure 17: 0-rang I jiao Passive 98 Figure 18: Mandarin BEI 99 v i Acknowledgment I wish to thank all the members of my thesis committee, Dr. Michael Rochemont, Dr. Hamida Demirdache and Dr. Leslie Saxon for all their support, encouragement and invaluable advice. I have benefited immensely from the guidance of Michael Rochemont, who has helped me to overcome monstrous difficulties with the writing of this thesis. He has shaped the foundations for many of the ideas in this paper. His unwavering enthusiasm and willingness to listen to my ramblings have allowed me to expand my writings into something more than I could have imagined. I am indebted to Hamida Demirdache whose vast knowledge and professionalism has never ceased to amaze me. I am grateful for all the hours that she contributed throughout the writing of this thesis, and thank her for enduring my endless onslaught of questions and creased eye-brows. The influence of her guiding hand and constructive advice can be felt throughout this paper. Not only has she taught me how to expand my ideas, but also how to expand as a linguist. No words can express the gratitude I have for Leslie Saxon, who agreed to sit on my committee even with very short notice. Although we have only met once, her generosity, kindness and constant support wi l l not be forgotten. The stretch of water which separates us did not prevent her from providing me with encouragement and renewing my belief that the completion of this thesis was not such an impossible task. Her attention to detail and invaluable comments have greatly improved the contents of this paper. I also wish to extend my gratitude to all the members of the U B C Linguistics department. A big thank-you goes out to Carmen De Silva, who kept the gears of the linguistics office running and always had a smile for me. Special thanks goes to Rose-Marie Dechaine who has reminded me that there is more than one approach to any problem. She has often jump-started my 'little-grey cells' with her encouragement and constant stream of inspirational ideas. I am privileged to have been a student of Douglas Pulleyblank. He has taught me the importance of being clear and concise (else one has to learn to talk really, really fast!). He never hesitated to come to my rescue, even when he was supposed to be \" in hiding.\" Patricia Shaw and Dr. Kitching have been kind and supportive throughout my years in the department, and are always ready to lend an ear. Thanks is also due to Henry Davis, Laura Downing, Susan Urbanczyk and Martina Wiltschko for attending my talks and providing me with discriminating input. To David Ingram, who I hold in highest regards, thank-you for sparking my interest in linguistics during my first year in university. To my fellow classmates, I don't know where to begin to thank all of you. I am forever indebted to Lisa Chang, who was not only my consultant and my confidante for new ideas, but also my dear friend. I wish to thank Tomio Hirose for his patience, encouragement and helpful suggestions, and Elizabeth Currie for making me feel like a 'big person'. For all those who shared the department hallway with me, Leora Bar-El, Eleanor Blake, Susan Blake, Nicole Horseherder, Darin Howe, Sandra Lai and Takeru Suzuki, it was my privilege to know all of you. Everyone has provided me with inspiration at one time or another and my life as a struggling student was made more pleasant by your acquaintance. Finally, I am deeply grateful to the ones who are close to me, who have shaped me into who I am. M y mom, dad, brother and sister have been the calm inside every storm that crosses my way (even if I created the thunder). I could never do without their love, inspiration and support. Mama, Baba, I love you. A heart-felt thank-you is also dedicated to Jeff, for putting up with my temper fits, running my \"little\" errands, and for your enduring friendship. Thank-you, dear Lord, for everything. Chapter 1: Introduction In the linguistic world many studies have been conducted on the Chinese languages. Surprisingly, for the last twenty years very few of these studies have looked at the Cantonese dialect in particular, but instead focused on Mandarin. Although the Chinese dialects share a uniform writing system, the grammars of the individual dialects are not the same. Though most of the grammatical relations that hold of Mandarin also hold of Cantonese, there are still quite a few syntactic differences between these two non-mutually intelligible dialects. The goal of this paper is to analyze one such aspect: the passive construction in Cantonese. 1.1 Preliminary Look at the BEI-Construction It is generally assumed, from studies conducted in Mandarin, that the passive is signaled by the presence of the word BEI . In both Mandarin and Cantonese, this item carries the approximate meaning translated by the preposition \"by\" when followed by an agent N P . The identity of B E I itself remains unclear. This item ' B E T used in Mandarin to signal the passive also translates as B E I in Cantonese as wel l , with the exception of a difference in tone. Most of the Mandarin BEI-constructions have exact equivalents in Cantonese, except that the words are pronounced differently. One must keep in mind that although the sentences are still written the same way in the orthography for both dialects, the pronunciations of individual words are different. Example (1) illustrates some BEI-constructions in the two dialects. These sentences all carry a passive interpretation. (I)1: (N.B. Mandarin and Cantonese B E I have different tones.) 1 Abbreviations used in this paper ADJ = adjectival marker ADV = adverbial marker PL = plural CL = noun classifier LP = linking particle PRT = sentence particle NEG = negation marker PFV = perfective aspect V-PRT = verbal particle 1 Mandarin2 Cantonese a) Wong pian le Lee c) a Wong gwai jo a Lee Wong cheat PFV Lee Active CL Wong cheat PFV CL Lee 'Wong cheated Lee.' 'Wong cheated Lee.' b) Lee bei Wong pian le d) a Lee bei a Wong gwai jo Lee BEI Wong cheat PFV Passive CL Lee BEI CL Wong cheat PFV 'Lee was cheated by Wong.' 'Lee was cheated by Wong.' The Mandarin passive sentence, signaled by the presence of BEI, can also appear without an Agent NP following BEI. That is, the subject (external theta role) of the active transitive verb can be omitted, as in (2a). However, the same sentence in Cantonese (2b), with omission of the Agent phrase, is ungrammatical. (2) Mandarin Cantonese a) Lee bei pian le b) * a Lee bei gwai jo Lee BEI cheat PFV CL Lee BEI cheat PFV 'Lee was cheated.' 'Lee was cheated.' The ungrammaticality of (2b) leads one to ask why and how the BEI-construction in Cantonese is different from the one in Mandarin, and whether it is the same construction at all. One might easily dismiss this behavior of BEI in Cantonese as a dialectal difference, but how can one explain the obligatoriness of the agent phrase if the BEI in (2b) is the same as the BEI in (2a) and the sentence is a passive? Previous analyses of the BEI-construction (Chu 1973, Cheng 1988, LaPolla 1988, Y.H.Li 1990, E.W.Li 1994, etc.) have all based their studies on the Mandarin BEI-construction which allows omission of the Agent NP. These analyses are not adequate to analyze BEI in Cantonese because the Agent NP cannot be omitted. The rest of this paper is dedicated to probing into the structure of the BEI-construction and also into the status of BEI itself. I will address the following questions: 2 The orthography I have used for the Mandarin and Cantonese examples in this paper are not standard pin-yin conventions. They are broad transcriptions only. 2 1) Is the BEI-construction a passive? Can it be something else? 2) Is the Cantonese BEI-construction the same as the Mandarin BEI-construction 3) If the answer to (2) is 'no', then in what ways are they different? 4) What is the nature and status of BEI in Cantonese? 1.2 Theoretical Assumptions M y analysis will mostly be presented within the framework of Government and Binding, initiated by Noam Chomsky (1981, 1982, 1986). However, I will also include some discussion from the more recent Minimalist framework developed by Chomsky (1993, 1995). To address the controversial debate regarding word order in the Chinese languages I will be adopting Kayne's (1994) proposal of a Linear Correspondance Axiom which forces a fixed word order for all languages. Government and Binding is a version of generative syntax which places much emphasis on the explanation of abstract universal principles of grammar rather than the development of language-particular rules. The main levels of representation for this model of grammar are depicted in Diagram 1: Diagram 1: D-structure S-structure Phonetic Form (PF) Logical Form (LF) A general movement rule called Move a, relates the D-structure (underlying structure) level of representation to the S-structure level (surface structure). PF and L F act as the interface between acoustic/articulatory features and semantic features, respectively, to the formal grammar. To constrain the application of Move a GB-Theory contains several semi-autonomous modules: X-bar Theory, Theta Theory, Bounding Theory, Government Theory, Case Theory, 3 Binding Theory and Control Theory. The reader is referred to Chomsky (1981,1986) or Aoun (1985) for a more comprehensive discussion of the above. The Minimalist Program (Chomsky 1995) is Chomsky's latest development towards a universal theory of grammar. It essentially involves the same basic notions as Government and Binding but approaches the formal grammar in a different way. Diagram 2 depicts the main stages of a derivation within the Minimalist framework: Diagram 2: Numeration I ^ J ^ S p e l l - O u t ) A - P interface C-I interface A main component of Minimalism is the generative procedure. It consists of a single lexicon and a computational system, CHL. A n array of lexical choices called the numeration is mapped by CHL to a pair (n, X) which contain the features of lexical items, n is the PF representation which is interpreted at the A - P (articulatory-perceptual) interface; X is the L F representation interpreted at the C-I (conceptual-intentional) interface; elements interpreted at one interface cannot be interpreted at the other. At some stage X of the derivation, CHL will split the derivation (Spell-Out) into two parts, forming % and X, to be interpreted within their respective components. The mapping and structure building processes are mediated by a complex set of operations called Select, Merge, Move and Erase/Delete. There is also an intricate Checking Theory which can apply to the derivation in both the covert and the overt syntax. The reader is referred to Chomsky (1995) for a more thorough discussion of the applications of the generative procedure in the Minimalist Program. 1.3 Some Notes Regarding Chinese: Mandarin versus Cantonese This next section will set out some preliminaries about the Chinese language relevant to the ensuing discussion I will address my assumptions regarding word order, the different particles used in Mandarin and Cantonese, and also some facts relating to the binding of anaphors. 1.3.1 Chinese Word Order The issue of Chinese word order has been a widely debated topic for many years. Huang (1982) presents detailed arguments that Mandarin Chinese is essentially head initial, with the exception of noun phrases which are head final. T o circumvent the controversial debate of head-final versus head-initial I will be adopting Kayne (1993) who proposes a Linear Correspondence Axiom (LCA) which imposes a strict linear order for terminal elements via asymmetric c-command. Chomsky (1995) comments that \"order reflects structural hierarchy universally by means of the L C A . . .any category that cannot be totally ordered by L C A is barred. From Kayne's specific formulation, it follows that there is a universal specifier-head-complement (SVO) order and that specifiers are in fact adjuncts.\" (Chomsky, 1995:335). Kayne's specifications appear compatible with what we observe as the surface order of Mandarin and Cantonese sentences. The canonical order of elements in Mandarin and Cantonese is Subject-Verb-Object, although other orders are possible. Since word order is not the main topic of this paper I now depart from this problem and will be assuming a strict specifier-head-complement structure from this point forward. 5 (3) Mandarin s V o a) wo huwei- tswo nei jien shi I will - do thatCL thing 'I will / am going to do that thing.' Cantonese s V o ngohwuh-joe go geen si I will - do that CL thing T will / am going to do that thing.' b) 1.3.2 Chinese Particles Chinese is not a morphologically rich language. Most of the morphology is conveyed either by word order alone or by an elaborate selection of particles. For example, the Cantonese pronoun keuih (Mandarin ta) can be used for both masculine and feminine subjects or objects. Therefore, it can mean \"he\" (3rd person masculine, singular, nominative case), \"she\" (3 r d person feminine singular, nominative case), \" h i m \" (3 r d person masculine, singular, accusative case) or \"her\" (3 r d person feminine singular, accusative case). Similarly, ngoh (Mandarin wo) can stand for either \"I\" (1 s t person singular, nominative case) or \"me\" (1s t person singular, accusative case). Word order and position in relation to the verb reveal whether these pronouns stand for subject or object because they have no overt case marking. (4) Mandarin Cantonese a) wo ai ta b) ngoh ngoi keuih I love 3p I love 3p In this section I will discuss some of the main particles that are of interest to the ensuing discussion. I have assembled all nominal prefixes (such as a to be discussed below), verbal and aspect particles, question particles and sentence particles together under the one general title of 'Particles'. The fine distinctions among them have no noticeable influence on the 'I love him/her' T love him/her.' c) ta ai wo 3p love me 'He/she loves me.' d ) keuih ngoi ngoh 3p love me 'He/she loves me.' 6 current topic. In particular I w i l l be discussing the perfective aspect particle (PFV) , the linking particle (LP), the noun classifier particle (CL) , and the utterance particle (PRT). The Cantonese particles are not easily translatable into English and they also differ from Mandarin as well . Matthews and Y i p (1994) comments that: Cantonese is especially rich in particles, having some thirty basic forms compared to seven in Mandarin.. .The particles do not correspond straightforwardly to those in Mandarin, although their functions are broadly similar. Moreover, combinations of two or more particles occur readily in Cantonese, whereas in Mandarin only le combines with other particles.. .In casual conversation, many sentences seem incomplete without one or more final particles; in broadcasting, by contrast, they are reduced to a minimum, whether to save time or to preserve a higher degree of formality. .. .Given that it is difficult or impossible to define the function of particles without reference to the speech context, it is arguable more accurate to treat them as associated with individual utterances rather than sentences. Moreover, particles may be used with phrases and fragments which do not form complete sentences. Matthews & Yip (1994: 338-339) The first two particles, the perfective aspect and the linking particle, are quite straightforwardly translatable from one dialect to the other. Their corresponding forms are exemplified in (5) below. I have given the most widely used form for each particle. Others may be used for the same purpose but the difference w i l l not impact the discussion regarding the passive construction. (5) Mandarin Cantonese Perfective Aspect -le Perfective Aspect -jo a) ta lai - le b) keuih lai - jo 3p come-PFV 3p come - P F V 'He came.' 'He came.' Linking Particle de Linking Particle ge c) sihuan wo de ren d) tsounyi ngoh ge yun like me L P people like me L P people 'The people who like me.' 'The people who like me.' 3 For ease of exposition I will only use \"he\" or \"him\" for glosses involving the 3rd person pronoun, although as noted above the feminine form is the same. 7 The perfective aspect is attached to a verb to convey a completed action. It is often accompanied by a time adverbial, such as kumyak 'yesterday', to restrict the time of the event, as Chinese verbs carry no Tense marking. I will return to a discussion of the time adverbials in Section 6.1. The perfective aspect is often used in BEI-constructions, but several other aspect markers, such as the progressive gun, is also possible. The linking particle acts as a relativizer linking two NPs together, much in the same fashion as that in English. The latter two particles, noun classifier and sentence particle occur in Cantonese but not in Mandarin. Since both of these particles are used in Cantonese casual speech it is not surprising that they do not show up in Mandarin, where the spoken speech is often the same as the written speech. In conversational Cantonese, the noun prefix a- (which I have included as part of the class of Noun Classifiers) appears before proper names and kinship terms to denote familiarity. In Mandarin examples, such as (1) from above, the reader should note that this particle is not present: (1) Mandarin Cantonese a) Wong pian le Lee c) _a_ Wong gwai jo _a_ Lee Wong cheat PFV Lee Active C L Wong cheat PFV C L Lee 'Wong cheated Lee.' 'Wong cheated Lee.' Cantonese also has a much wider range of utterance particles, which occur at the end of a sentence or utterance, than Mandarin. They are best conveyed in English by changes in intonation. There is usually no correspondant for these particles in Mandarin except perhaps for ba, which is used informally to accompany an invitation or a polite request; a to accompany questions; and de in exclamations. 8 (6) Mandarin a) chin tswo ba! please sit P R T 'Please sit down!' Cantonese b) tsirj tso la! please sit P R T 'Please sit down!' c) ni bo - shufu a? you not comfortable P R T 'Are you ill?' e) hou de! good P R T 'Great / sure!' g) ni tswotien chu nali ? h) nieh kumyak heui bin jek? you yesterday go where you yesterday go where P R T 'Where did you go yesterday anyways?' 'Where did you go yesterday anyways?' Notice that the particle in the Cantonese example (6h) has no correspondant in Mandarin. However, the particle a can be inserted at the end of (6g) and combined with raising intonation to form a question. 1.3.3 Anaphor Binding: Long-distance versus Local I will be assuming the standard Binding Theory from the Government and Binding framework. Binding Theory (A) An anaphor is A-bound in its governing category. (B) A pronoun is A-free in its governing category. (C) An R-expression is A-free. PRO is ungoverned Chomsky (1981: 188, 191) (5) (i) a is X-bound by P if and only if a and P are coindexed, P c-commands a, and P is in an X-position. (ii) a is X-free if and only if it is not X-bound ...(5) define \"bound\" and \"free\" with \" X \" replaced by \" A \" or \" A \" . Governing category is defined as: P is a governing category for a if and only if p is the minimal category containing Chomsky (1981:185) 9 a, a governor of a, and a SUBJECT accessible to a. .. .the subject of an infinitive, an NP or a small clause... is a SUBJECT. Chomsky (1981: 209-211) In addition to these principles, binding also requires that the antecedent c-command the pronoun or anaphor. Generalized C-command a c-commands p iff a does not dominate P and every y that dominates a dominates P Chomsky (1986:8) Pronouns in both Mandarin and Cantonese are subject to Principle B. (7) Mandarin Cantonese a)* ta, sihwan tai b)* keuihj tsourjyi keuih, 3p like 3p 3p like 3p ' He j likes him,.' ' He j likes him;.' c) ta; sihwan taj d) keuihj tsourjyi keuihj 3p like 3p 3p like 3p 'Hei likes him.j.' 'Hej likes hiirij.' In the above examples the governing domain is the matrix IP. Both (7a) and (7b) violate Principle B because the pronoun is bound within the governing domain of IP. The anaphors of the Chinese languages are a bit more intriguing. Mandarin literature states that the self-anaphor ziji 'self can be locally or long-distance bound to a subject. However, the compound reflexive, such as ta-ziji 'himself, made up of the pronoun ta '3 r d person' and the reflexive ziji 'self , is locally bound only. The following Mandarin examples are taken from Huang and Tang (1991: 263-282)4. They have been re-numbered for ease of exposition. I have placed a # 4 on the left-hand side of some examples because of several inconsistencies to be discussed below. I) a) Zhangsanj renwei [Lisij hai-le zijij] b) Zhangsanj renwei [Lisij hai-le zijij] 'Zhangsanf thought that Lisij hurt himselfj/j' 4 Henceforth H&T 10 II) a) Zhangsanj renwei [Lisij hai-le ta-zij b) Zhangsarij renwei [Lisij hai-le ta-zij ij] 'Zhangsan; thought that Lisij hurt himself^' III) a) Zhangsanf renwei [WOJ hai-le ziji»j] b) Zhangsanj renwei [WOJ hai-le zijij] 'Zhangsan; thought that Ij hurt myself^' IV) a) Nij renwei [Zhangsanj dui ziji.j meiyou xinxin] b) Nij renwei [Zhangsanj dui zijij meiyou xinxin] 'You; think that Zhangsanj has no confidence in self.j/j' V) a) Woj renwei [womenj yinggai dui ziji.j you xinxin] b) Woj renwei [womenj yinggai dui zijij you xinxin] 'Ij think that wej should have confidence in self.j/j' VI) a) Womenj renwei [taj dui ziji.j meiyou xinxin] b) Women; renwei [taj dui zijij meiyou xinxin] 'Wej think that hej has no confidence in self.j/j' ? VII) a) Zhangsanj shuo wo k zhidao [Lisij chang piping ziji.j] b) Zhangsanj shuo wo k zhidao [Lisij change piping z i j i . j c) Zhangsanj shuo wok zhidao [Lisij chang piping zijij] 'Zhangsanj said that Ik knew that Lisij always criticized self*j/j/«k' VIII) **a) Zhangsanj shuo [Wangwuk zhidao [Lisij chang piping zijij] #*b) Zhangsanj shuo [Wangwuk zhidao [Lisij chang piping z i j i j c) Zhangsan, shuo [Wangwuk zhidao [Lisij chang piping zijij] 'Zhangsanj said that Wangwuk knew that Lisij always criticized selfj/j/k' IX) a) Zhangsanj di xin biaoshi [Lisij hai-le ziji.j] b) Zhangsan; di xin biaoshi [Lisij hai-le zijij] 'Zhangsanj's letter indicates that Lisij hurt self.j/j' X) a) Zhangsanj shuo [ruguo Lisij chang piping zijij], ta j iu bu qu. b) Zhangsan; shuo [ruguo Lisij chang piping zijij], ta j iu bu qu. 'Zhangsanj said that if Lisij criticized himself/j, then he won't go. Huang & Tang (1991: 263-282) The above examples demonstrate that the bare reflexive ziji can be both long-distance and locally bound to a subject. However, when the reflexive is long-distance bound all potential antecedent N P s must be identical in person (i.e. 1 s t, 2 n d , 3rd) and number features (i.e. singular, plural). The reflexive in (I) can be bound both locally and long-distance because the potential antecedents are both 3 r d person singular nouns. However, the compound reflexive shown in (II) is strictly locally bound only. Examples (III) to (VI) and (X) show that long-distance binding is not possible i f the available antecedents do not have the same person or 11 number features. In the case of (IX) long-distance binding is not possible because the antecedent is a possessor which is part of a larger D P and as such cannot c-command the reflexive. I verified H & T ' s findings with native Mandarin speakers and found some interesting results. While the speakers presented no counter-evidence to refute H & T ' s claims, they did find (VII) and (VIII) rather hard to process because the anaphor is too deeply embedded within the structure. I have place a '?' before (VII) because speakers found that it was hard to understand and commented that no one would actually say such utterances. However, there was a general consensus that the anaphor could be long-distance bound in these cases. However, none of my consultants agreed with H & T ' s predictions regarding the examples in (VIII), hence the # s symbol. According to the speakers, they preferred the anaphor to be locally bound in these cases and the long-distance binding was impossible to get. However, because the speakers did accept long-distance binding of the self-anaphor in the simpler sentences, and also (with difficulty) in (VII), I suspect that they may not be able to get the appropriate readings in (VIII) because the sentences are too complex to process for speakers who are untrained in linguistics. Therefore, the general conclusion is that H & T ' s claims regarding long-distance binding of the reflexive ziji are valid. A descriptive Cantonese grammar text by Matthews and Y i p (1994) affirms that the behavior of the reflexives in Mandarin also translate over into Cantonese. That is, the bare reflexive, translated as tsigei, can be both long-distance and locally bound; the compound reflexive, translated as keuih-tsigei, is locally bound, However, all the judgment from native speakers that I consulted are contrary to this literature in regards to 12 the bare reflexive. The following examples are translations of the sentences taken from H & T . I have also included the §F symbol beside any examples where speakers' judgments are contrary to Matthews and Yip. i) #*a) Johns yingwei [Paulj hoy-jo tsigeij] b) Johtij yingwei [Paulj hoy-jo tsigeij] ' Johrii thought that Paulj hurt himselfi/j' ii) a) Johnj yingwei [Paulj hoy-jo keuih-tsigei*j] b) Johnj yingwei [Paulj hoy-jo keuih-tsigeij] 'John; thought that Paulj hurt himself*j/j' iii) a) Johxij yingwei [ngohj hoy-jo tsigeij] b) Johnj yingwei [ngohj hoy-jo tsigeij] 'John, thought that Ij hurt myself^ ' iv) a) Neihj yingwei [Johnj duey tsigei*; mouh xeunsum] b) Neihj yingwei [Johnj duey zigeij mouh xeunsum] 'YOUJ think that Johnj has no confidence in self*j/j' v) a) Ngohj yingwei [ngohdeihj yinggoy duey tsigeij youh xeunsum] b) Ngohj yingwei [ngohdeihj dui tsigeij youh xeunsum] Ti think that wej should have confidence in self*i/j' vi) a) Ngohdeihj yingwei [keuihj duey tsigei*j mouh xeunsum] b) Ngohdeihj renwei [keuihj dui tsigeij meiyou xeunsum] 'We; think that hej has no confidence in self*j/j' vii) a) Johnj wa ngohk zhidoe [Paulj sirjyuk peiping tsigei*j] b) Johnj wa ngohk zhidoe [Paulj sinyuk peiping tsigei*k] c) Johnj wa ngohk zhidoe [Paulj sirjyuk peiping tsigeij] 'Johnj said that Ik knew that Paulj always criticized self*j/j/*k' viii) a) Johnj wa Maxk zhidoe [Paulj sinyuk peiping tsigeij] # 4 b) Johnj wa Maxk zhidoe [Paulj sinyuk peiping tsigeij c) Johnj wa Maxk zhidoe [Paulj sinyuk peiping tsigeij] 'Johnj said that Maxk knew that Paulj always criticized self/j/k' ix) a) Johnj ge xeun byeiusi [Paulj hoy-jo tsigei*j] b) Johnj ge xeun byeiusi [Paulj hoy-jo tsigeij] 'Johrij's letter indicates that Paulj hurt self*j/j' x) f*a) Johnj wa [yueguo Paulj sinyuk pieping tsigeij], keuih jow mh-heuy. b) Johnj wa [yueguo Paulj sirjyuk pieping tsigeij], keuih jow mh-heuy. 'Johnj said that if Paulj criticized himselfi/j, then he won't go. 13 The above examples show a consistent pattern. Speakers find the long-distance binding of the bare reflexive completely unacceptable (hence contrary to what M & Y notes. M & Y claim that (i b), where the reflexive is long-distance bound, should be possible. However, this is not verified by speaker's judgments. While we can use the same argument that the potential antecedents in some of the examples do not match in person and number, this still will not explain the ungrammaticality of (i a), (viii a, b) and (x a). The speakers also disagreed with the following Cantonese example taken from Matthews and Y i p (1994:84): A - Mah; wah Mihng-jai mh sik jiugu jihgeij C L grandma say Ming-boy not know take-care self 'Grandma says Ming doesn't know how to take care of her.' Matthews & Yip (1994: 84) According the judgments I received, the reflexive cannot be long-distance bound with ,4-Mah; Mihng-jai is the only available antecedent5. Therefore, the reflexive is bound locally. I did not encounter any processing problems with the Cantonese examples as I had with the Mandarin examples. Reflexives which act as possessors in DPs must also be locally bound: xi) a) Johnj yingwei [Paulj hoy-jo tsigeij go tsey] b) Johnj yingwei [Paulj hoy -jo tsigeij go tsey] John think Paul hurt-?RT self C L son 'John; thought that Paulj hurt se l f^ ' s son.' xii) a) Johnj wa MaXk zhidoe [Paulj sirjyuk peiping tsigeij go tsey] b) Johnj wa Maxk zhidoe [Paulj sirjyuk peiping tsigei*k go tsey] c) Johnj wa Maxk zhidoe [Paulj sirjyuk peiping tsigeij go tsey] John say Max know Paul always critisize self C L son 'Johnj said that Maxk knew that Paulj always criticized self.j/j/*k's son.' However, it was pointed out to me by a consultant that when such a sentence is read out in a formal context, or in narratives, the long-distance binding with A-Mah is possible. This possibility may be attributed to the similarity of Mandarin speech with written texts. However, only local binding is possible in colloquial Cantonese speech. 14 xiii) a) John; wa [yueguo Paulj sirjyuk pieping tsigeij go tsey], keuih jow mh-heuy. b) Johnj wa [yueguo Paulj sirjyuk pieping tsigeij go tsey], keuih jow mh-heuy. John say if Paul always critisize self C L son 3p then not go 'John; said that if Paulj criticized self*j/j's son, then he won't go. The anaphor binding facts of Mandarin and Cantonese are summarized below: (i) The bare reflexive in Mandarin, ziji, can be local or long-distance bound as long as the potential antecedents agree in person and number. (ii) The compound reflexive in Mandarin, ta-ziji, is locally bound only. (iii) The bare reflexive in Cantonese, tsigei, is locally bound only. (iv) The compound reflexive in Cantonese, keuih-tsigei, is locally bound only. In a later section I intend to use the local nature of the bare reflexive in Cantonese in several tests for binding in different structures. These tests will demonstrate that the BEI-passive construction in Cantonese must be in a mono-clausal structure, consisting of one binding domain. The previous section was meant to point out some differences in the translations between Mandarin and Cantonese, since Mandarin will be the standard language of comparison in this thesis. 1.4 A n Outline of the Thesis This thesis will study the nature of the BEI-construction in Cantonese and make comparisons with the same construction in Mandarin. I will show that under different sets of conditions, this construction can derive (i) utterances with a passive interpretation; and (ii) utterances with a causative interpretation as well. This behavior is surprising in that on the one hand, passives are known to suppress a theta role, while on the other hand, causatives are known to introduce an extra theta role into the structure. The BEI-construction in Mandarin does not exhibit the ability to render these dual interpretations. I argue that although two different interpretations are possible, there is no need to postulate the existence of two 15 different BEIs in Cantonese. I maintain that the BEIs in both structures are the same element. The passive and causative structures are generated by a Functional projection containing a null Functional head, and a Lexical projection containing a null Lexical head, respectively. I will also discuss an equivalent alternative strategy whereby BEI itself assumes both tasks because of its specialized feature matrix. I argue that the categorial status of BEI in previous analyses have always been indeterminate because it carries a unique combination of categorial features. Using the binding facts above, I will show that the BEI-passive construction is necessarily mono-clausal while the BEI-causative construction is bi-clausal. This prediction is consistent with my argument that passive is generated by a Functional projection and causative is generated by a Lexical projection. Following Grimshaw (1990), a functional projection above a lexical projection (in this case VP) are extended projections of . the lexical head and do not induce another independent domain of their own. On the other hand then, the formation of a new lexical projection can serve to mark the beginning of a new clause.6 The progression of the thesis is organized as follows. In Chapter 2 I will do a comparison of the BEI-construction in Mandarin and Cantonese and comment on the apparent obligatory nature of the Agent phrase. Chapter 3 examines the different types of passives available in Cantonese, with a special focus on what is termed, the \"Indirect Passive\". The different semantic interpretations available with this type of passive launches us into the discussion of whether the Cantonese BEI-construction would be better analyzed as involving a causative structure instead. 6 I am grateful to Michael Rochemont for pointing this out to me. 16 Chapter 4 investigates the nature of the item B E L I use tests adapted from A . Y . H . L i (1990) to test whether BEI can be classified as a preposition or a verb. This section also explores the properties of the Cantonese passive and causative interpretations that can be rendered by a BEI-construction. Chapter 5 examines two previous analyses of the BEI-construction in Mandarin. I argue that these analyses cannot be adopted into Cantonese because they will not adequately handle the pattern of data associated with the BEI-construction in Cantonese. In Chapter 6 I will establish the syntactic structure of the BEI-construction. I will review A . L i (1990)'s discussion of the placement of time adverbials, instrumentals and patients to determine the location of BEI within a tree structure. I will consider several different possible structures for BEI but will eventually argue that the Functional versus Lexical difference is the best solution. The consequences of such a proposal will also be reviewed. In Chapter 6 I will apply the analysis of the Cantonese BEI-construction to Mandarin and comment on any differences. The final chapter will summarize the points made in this paper. Chapter 2: The BEI-construction in Mandarin and Cantonese As previously mentioned, the BEI-construction in Mandarin has been accepted by many as the passive construction (Chu 1973, Xiaobo 1991, Po-Ching & Rimmington 1997, etc.) The BEI-construction is essentially a sentence that contains the particle BEI followed by an optional N P . The basic form of this 'passive' construction is [Object + BEI + Subject + verb]. The N P in initial position corresponds to the underlying object of an active sentence, and usually bears the thematic role of Patient. The NP following BEI corresponds to the 17 subject of an active sentence, and bears the thematic role of Agent of the verb. BEI in these passive sentence is often translated as \"by\" in English. M (8) a) Lisi pian - le Zhang Active Lisi cheat-PFV Zhang 'Lisi cheated Zhang.' b) Zhang BEI Lisi pian-le Passive Zhang BEI Lisi cheat-PFV ' Zhang was cheated by Lis i . ' Although there are Mandarin sentences lacking BEI which are best translated into English using the passive construction, they do not carry the same semantic flavour or characteristics as BEI-passive constructions. While BEI-constructions can be dynamic passives, these BEI-less constructions only carry the meaning of statives. Verbs which participate in the BEI-construction usually cannot be used in these statives. The only exception I know of, and which has been noted in by Chu (1973:445) is the verb chi 'eat' (Cantonese sik). A n additional constraint is that BEI-passives are normally associated with an adversative reading, while there is no such requirement on the statives. M (9) a) pingguo bei chi -le apple BEI eat P F V 'The apple was eaten.' (by someone) b) pingguo chi -le apple eat PFV 'The apple was eaten.' (describing the state of the apple) c) * shu bei chuban -le book BEI publish P F V 'The book was published.' (by someone) d) shu chuban -le book publish P F V 'The book was published.' (describing the state of the book) 7 I will be adopting the convention of listing M for Mandarin examples, C for Cantonese examples and E for English examples from this point forward. 18 (9a) would be used in a situation where someone was adversely affected because the apple had been eaten. It is unnatural to use (9b) in the same situation because it only describes the physical state of the apple, and not the effect of its having been consumed. Also, notice that a verb such as chuban 'publish' cannot be used in a BEI-construction. Without digressing into an elaborate discussion about BEI-less passives I will be assuming that this class of statives are different from the BEI-constructions. For the rest of the paper, I will only concentrate on those passives which involve the morpheme BEI. The reader is referred to Cheng (1986) for further discussion of these BEI-less constructions. In Mandarin, the agent N P can sometimes be left out. Because the N P , if present, must immediately follow BEI, previous analyses have suggested that BEI serves to introduce the agent. (Peyraube 1989; L i 1994; Matthews & Yip 1994, among others) Example (10) below illustrates that nothing can intervene between BEI and the overt agent N P . The time adverbial zuotian 'yesterday' can be inserted before the BEI-phrase (10a, c) or after the BEI-phrase (lOd) as long as it does not intervene between BEI and the N P following it (10b). If there is no overt agent N P , some speakers find it acceptable to have the time adverbial immediately after B E I 8 (lOe). M (10) a) zuotian Lisi BEI Zhang sha si - le yesterday Lisi BEI Zhang kill die-FFV 'Yesterday, Lisi was killed by Zhang.' b) * Lisi BEI zuotian Zhang sha si - le Lisi BEI yesterday Zhang kill die-VFW c.f. c) Lisi zuotian BEI Zhang sha si - le Lisi yesterday BEI Zhang kill die-FFV d) Lisi BEI Zhang zuotian sha si - le Lisi BEI Zhang yesterday kill die-VFW 8 A. Li (1990) supports this view. 19 e) ? Lisi BEI zuotian sha si - le Lisi BEI yesterday kill dze-PFV Cantonese also has a corresponding passive9 construction which is signaled by the presence of the same character, BEI. Although little work has been done on the BEI-construction in Cantonese, it is generally assumed that it behaves much like its Mandarin counterpart. Thus, if we translate Example 10 into Cantonese we would expect to get the same results in grammaticality. C (11) a) kumyak LeisaBEI a Tseurj saksei - jo yesterday Leisa BEI C L Tseurj kill die - P F V 'Yesterday, Leisa was killed by Tseurj.' b) * Leisa BEI kumyak a Tseurj sak sei - jo Leisa BEI yesterday C L Tseurj kill die - P F V c.f. c) Leisa kumyak BEI a Tseurj sak sei - jo Leisa yesterdayBEl C L Tseurj kill die - P F V d) Leisa BEI a Tseurj kumyak saksei - jo Leisa BEI C L Tseurj yesterday kill die-PFV e) * Leisa BEI kumyak sak sei - jo Leisa BEI yesterday kill die - P F V As expected, (1 lb) in Cantonese, like it's correspondant (10b) in Mandarin, is ungrammatical. However, notice that unlike Mandarin (lOe), (1 le) is completely unacceptable in Cantonese. The ungrammatically is not simply due to the intervention of the time adverbial between BEI and the verb, but to a more general constraint. Compare (12) and (13) below: I use 'passive' to refer to sentences where surface subject is construed as the underlying object of the verb and is being affected in some way by the verb i.e. has the role of patient 20 (12) Mandarin (13) Cantonese a) jincha da shang - le tufei a) ginchak da sheung - jo go chak police hit injured-PFV thief Active police hit injured - P F V C L thief 'The police injured the thief 'The police injured the t h i e f b) tufei bei jincha da shang-le Passive b) go chak bei ginchak da sheung - jo thief BEI police hit injured-PFV (overt Agent) C L thief BEI police hit injured-PFV 'The thief was injured by the police.' 'The thief was injured by the police.' c) tufei bei da shang - le Passive c) * go chak bei da sheung-jo thief BE\\ hit injured- PFV (no overt Agent) C L thief BEI hit injured - PFV 'The thief was injured.' 'The thief was injured.' As previously noted, the agent N P after BEI can be omitted in Mandarin. However, the same does not apply in Cantonese. (13c) shows that omission of the agent N P will render the sentence ungrammatical. The same explanation is offered for (lOe) above. However, consider the following sentences: Mandarin Cantonese (14) a) tufei bei da shang - le Passive b) * go chak bei da sheung-jo thief BEI hit injured- P F V (no overt Agent) C L thief BEI hit injure - PFV 'The thief was injured.' 'The thief was injured.' c) go chak beih da sheung -jo C L thief BEIH hit injure -PFV 'The thief was injured.' Example (14) above shows that there is a different phonological form of BEI in Cantonese which does not seem to obligatorily require an agent phrase. This BEI is written as beih (BEI with a low-level tone). (14c) shows that his special form of BEI must be used if the agent N P is omitted, else the sentence would be ungrammatical (14b). This form of BEI is reserved for formal speech only, such as literary contexts and news reports. In casual, everyday speech only bei is used, followed by the agent N P 1 0 . 1 0 In this section I will not discuss cases such as (14a) or (14c) at great length because they conform to the generalization offered by Keenan in regards to passives and omission of the agent phrase. I will return to a fuller discussion of these agentless forms at a later time. 21 C (15) a) howtseun beih pai heui toywan formal speech school-principalBEIH send to Taiwan 'The principal was being sent to Taiwan.' b) howtseun bei howgouk pai heui toywan casual speech school-principal BEI school-administration send to Taiwan 'The principal was being sent to Taiwan by the adminstration.' Because of its restricted use, Matthews and Yip (1994) suggest that its appearance is likely influenced by Mandarin. However, although this phonologically different BEI (beih) does not appear to require a following agent N P , it is usually not used in conjunction with an overt agent phrase. C (16) a) ?? go chak beih ginchak da sheung - jo C L thief BEI police hit injure - P F V b) go chak bei ginchak da sheung-jo C L thief BEI police hit injure - P F V While there is no question that (14c) is the accepted form, (16a) is also understandable but might cause the listener to assume that the speaker has uttered a wrong tone on BEI. That is, the use of beih with a low-level tone in conjunction with an Agent-phrase sounds unnatural. This behaviour suggests that perhaps the difference in tone acts as a pragmatic signal that there is a non-overt agent present, or tells the listener that the speaker will omit the agent in that particular utterance. I leave this open for now and will concentrate on constructions with the non-specialized form of BEI, bei, in the rest of the paper. In English the passive may be used whenever an agent cannot be specified because it is unknown. In Cantonese, if the agent is unknown or generic, an indefinite N P , yahn 'person' for human agents, or yeh 'thing' for non-human agents, is inserted after BEI. C (17a) go chak bei yahn da sheung-jo C (17b) go chak bei yeh da sheung-jo C L thief BEI person hit injure-PW C L thief BEI thing hit injure-PFY 'The thief was injured by someone.' 'The thief was injured by something.' 2 2 Assuming that the BEI-construction is a passive, the ungrammaticality of (13c) is surprising because it is a general property of passives that the agent phrase can be omitted. (17a) and (17b) are even more surprising in that the insertion of an agent will save the structure. According to a typological classification of passives offered by Keenan (1985), one characteristic of a true passive construction is the optionality of the external theta role. This seems to be the pattern across languages of the world. That is, as a general rule, if a language has passives, it would have what Keenan terms 'basic passives'. These are the type where there is no agent phrase and the main (non-passive form) verb is transitive. G-2.1 If a language has passives with agent phrases then it has them without agent phrases (Keenan 1985:249) Mandarin Chinese presents no deviation from this generalization, but Cantonese Chinese appears to have an obligatory agent phrase associated with its BEI-construction which cannot be omitted. The question is how do we account for this deviance? Three possibilities come to mind: 1) Keenan's generalizations are incorrect; 2) the Cantonese BEI-construction is not a passive; or 3) the BEI-construction is a passive but something else is forcing the retention/appearance of the agent. Another issue which must be considered is the nature of the BEI-phrase itself. That is, where it is generated, what properties does it have and what effect does it have on the predicate. Any previous analyses which only deal with Mandarin data cannot adequately account for the Cantonese BEI-constructions because the Agent cannot be omitted. 2.1 Obligatory Agents and Colloquial Mandarin Probing further into Mandarin, we find that the characteristic of retaining an agent may not be isolated to Cantonese alone. Referring to colloquial Mandarin, Po-Ching and Rimmington (1997) make the following comment: 23 \"In colloquial speech, rang or jiao may be used instead of bei... it is possible for the construction to be used without an agent. In these cases, bei.. .but not rang or jiao, is placed before the verb.\" (Po-Ching & Rimmington, 1997:124) In the cases where rang or jiao is used, they are associated with a meaning that is close to the get-passives in English. When they are used in other constructions rang has a literal meaning of 'let' and jiao has the meaning of 'call' . The correspondant of rang in Cantonese is yoerj 'let', while jiao is gui 'call' . Compare the following examples. These are the same sentences of (12) and (13) except that BEI has been substituted with rdng/yoerj and jiao Igui. Observe the grammaticality differences. M(18) Colloquial Mandarin a) jincha da shang - le tufei police hit injured-PFV thief 'The police injured the thief C(19) Cantonese a) ginchak da sheung - jo go chak Active police hit injured - P F V C L thief 'The police injured the t h i e f b) tufei jiao jincha da shang-le Passive b) *go chak giu ginchak da sheung - jo thief got police hit injured-PFV (overt Agent) C L thief call police hit injured-PFV 'The thief was injured by the police.' 'The thief was injured by the police.' c) tufei rang jincha da shang-le thief got police hit injured-VFV 'The thief was injured by the police.' d) *tufei jiao / rang da shang - le thief got I got hit injured- P F V 'The thief was injured.' c) *go chak yoerj ginchak da sheung - jo C L thief let police hit injured-VW 'The thief was injured by the police.' d) *go chak gui/yoerj da sheung-jo C L thief call / let hit injured - P F V 'The thief was injured.' From (19b, c), notice that Cantonese does not appear to correspond with Mandarin in using gui andyeorj for a passive interpretation. They only mean either 'let' or 'call' and do not have the semantics of get-passives like the Colloquial Mandarin examples. However, the ungrammaticality of (18d) shows that Colloquial Mandarin behaves like Cantonese in having 24 an obligatory agent phrase. Like (17a) above, (18d) can be saved by the insertion of a generic agent: C o l l o q . M (20) tufei jiao / rang ren da shang - le thief J I A O / R A N G person hit injure-PFV 'The thief was injured by someone.' Perhaps Cantonese BEI has subsumed the passive semantics conveyed separately by Colloquial Mandarin jidolrdng and bei. Cantonese BEI participates in a structure which gives rise to a passive meaning, like the Mandarin BEI-constructions, but it does not allow for the omission of the agent phrase. A further argument for this proposal comes from observations made by Chappell (1996) regarding the adversity constraint on jiao and rang passives: ...The bei passive of translatese broadened in its application to include the expression of passive events both neutral and fortunate in nature and as a result lost its adversity constraint entirely. N O U N P H R A S E BEI ( N O U N PHRASE) V E R B P H R A S E (undergoer) (agent) (23) Ta bei ( ) xuanwel zhuxf 3sg BEI ( ) elect: as chariperson 'She was elected as the chairperson.' ... A further piece of evidence showing that agentless passives are not a feature of the spoken language in Chinese is provided by the two passives formed by rang and jiao. As these colloquial passives are exclusively agentful, no sentences analogous to the bei of (23) exist. (27) *Ta rang (pao) xuanwei zhiixi. 3sg R A N G (JIAO) elect: as chairperson .. .even though the requirement of an overt agent is fulfilled, the adversative constraint is violated. Chappell (1996 : 1033-1034) The Chinese passive is usually considered an Adversity passive because these constructions are only used in contexts where something bad has happened the Patient or Experiencer. However, many scholars suggest that due to the translation of Western texts into Chinese, and vice versa, the BEI-passive construction has lost its strict adversity 25 constraint because the English passive has no such a requirement. Most passives in English are best translated using a BEI-construction (see discussion in Section 2.0). This has led researchers to suggest that the contact with the prestige literature (English) has caused the Chinese to accept and generate non-adversity passives with the BEI-construction. To prove this hypothesis, Cowen and Reed (1988) conducted an experiment where they ^compared the acceptability rate for non-adversity passives of Taiwanese residents, who have not had much exposure to Western literature texts in translation, and educated Mainland Chinese residents who have had more exposure to such texts. Their results indicated that the Mainland residents had a higher rate of acceptability for non-adversative BEI-constructions than the Taiwanese residents did. They concluded that language change could be influenced by contact with another language. They also predicted that once the change was accepted widely (i.e. no further need for an adversity effect on passives) the modified language rule would displace the original one and non-adversity passives might be accepted as the norm in future language development. Cantonese BEI-construction also seems to display an adversity requirement, in that if the verb has no adversative effect, it usually cannot participate in a passive construction. However, just like Mandarin it has also recently developed a tolerance for non-adversative meanings (see (21a, b)). C (21) a) keuih bei yun soorjwey joojik (c.f. Chappell's #23) 3p BEI people elect-as chairperson 'He was elected as the chairperson by the people.' b) ngoh bei lousi tsaan ngoh lsk I BEI teacher praise me smart T was praised by the teacher that I was smart.' 26 c)* Paul bei Mary ngoy-do mhduklieu Paul BEI Mary love-to-the-extent-that extremely 'Paul was loved by Mary to the point of desperation.' c.f. M d) Paul bei Mary ai de budeliao Paul BEI Mary love-to-the-extent-that extremely 'Paul was loved by Mary to the point of desperation.' However, a comparison of (21c) to (2Id) shows that Cantonese still retains more of an adversative requirement than Mandarin because the verb 'to love' cannot be passivized due to its lack of an inherent adversative meaning. Although the modifier budeliao (mhduklieu) 'desperation' adds an adversative connotation to the sentence, the sentence is still ungrammatical in Cantonese. This might be due to the notion that the verb itself, 'to love', cannot be adversative. However, the same sentence translated into Mandarin, (2Id) is fine. Thus it would seem that Cantonese BEI has retained a mixture of properties from bei (non-adversative passives possible), rang and jiao (non-adversative passives not possible) in Mandarin. Chapter 3: Types of Passives in Cantonese In addition to the regular passive constructions, Cantonese also has other kinds of passives. According to the discussion found in Matthews and Yip (1994)11, Cantonese has at least 4 different types of passives. They include the Regular passive, Indirect passives, Resultative passives and Adjectival passives. Each is exemplified below for the Cantonese dialect: C (22) a) gachebei a Paul touh-jo. Regular C L car BEI C L Paul steal P F V Passive 'The car was stolen by Paul.' 11 henceforth M&P (1994) 27 b) a John bei a Paultouh-jo ga che. Indirect C L John BEI C L Paul steal P F V C L car Passive 'John had his car stolen on him by Paul.' (Literally: John was affected by Paul's stealing his car.) c) gachebei a Paul tek laahn-jo. Resultative C L car BEI C L Paul kick broken P F V Passive 'The car was broken by Paul, (by his kicking of it).' d) bei Paul tek laahn-jo ge che mo yuhn wuhyew. Adjectival BEI Paul kick broken P F V L P car none people will want Passive 'No one will want the car that was broken (by kicking) by Paul.' (22c) and (22d) above illustrate the Resultative and Adjectival passives respectively. Notice, that these types of constructions are not easily translated into English. The Resultative passive describes a state resulting from an action and is formed by the addition of either a resultative verbal particle or an adjective of result/completion after the verb of a regular passive. M & Y suggest that the particle or adjective forms a verbal compound with the verb and is obligatory in these constructions . This is comparable to the constructions in English such as \"to word carefully\", where the modifier of the verb cannot be omitted. E (23) a) John worded the letter carefully b) *John worded the letter. The adjectival passive consists of a BEI-phrase with a passive verb that is used to modify the head noun. In (22d) the head noun of the phrase is che 'car'. These types of passives have the flavour of a relative clause. What is of main interest in the present discussion are the passives given in (22a) and (22b), the Regular and Indirect passives. Notice that in all of the passive constructions BEI must always be followed by its agent N P . An alternative to their suggestion is that these passives are exactly like Regular passives, except that they are formed with serial verbs instead of a simplex (single) verb. Notice in (22c) that the perfective marker actually attaches to the so-called \"adjective\" laahn 'broken'. This suggests that perhaps it is verbal in nature, so could plausibly be the second member of a serial verb construction. 28 Indirect passives present yet another puzzle; there appears to be three arguments in these types of constructions: Experiencer, Agent, and Patient. experiencer agent patient (22b) a John bei a Paultouh-jo ga che. Indirect C L John BEI C L Paul steal P F V C L car Passive 'John had his car stolen on him by Paul.' Passive operations do not typically introduce another argument; rather, they are usually analyzed as suppressing an argument (external theta role). Theta grid augmentation is characteristic of causative-type constructions, however. Would Indirect passives be better analyzed as causatives instead? We will consider this possibility in the next section. 3.1 Indirect Passives The Indirect passives is unlike a Regular passive because it does not have corresponding active form. That is, while the grammatical subject of the Regular passive form is actually the logical object, this is not the case with Indirect passives. losical object logical subject C (24) a) a John bo che BEI a Paul touh-jo Regular Passive C L John C L car BEI C L Paul steal P F V 'John's car was stolen by Paul.' logical subject losical object b) a Paul touh-jo a John bo che Active of (24a) C L Paul steal P F V C L John C L car 'Paul stole John's car.' affectee(?) logical subject losical object c) a John BEI a Paul touh -jo bo che Indirect Passive C L John BEI C L Paul steal P F V C L car 'John was affected (adversely) by Paul stealing his car on him.' logical subject logical object d) a Paul touh-jo a John bo che Not the Active of (24c) C L Paul steal P F V C L John C L car 'Paul stole John's car.' 29 A passive sentence describes essentially the same event as an active sentence. For example, \"John hit Mary\" denotes the same event as \"Mary was hit by John\", although the syntactic structure is different. Like the English example, (24a) and (24b) both describe the same event, but (24c) and (24d) do not. The difference lies in the fact that (24c) describes an event where John was adversely affected by something; (24d) describes an event involving John's car. Another peculiarity of Indirect passives, which may help explain why there is no direct active correspondant form, is that the N P in sentence initial position is always construed as the possessor of the N P in object position. For example, in (22b), for a passive reading, John is construed as the owner of ga che 'the car.' This possessor relation can be reinforced by placing a pronoun, keuih, which acts as an overt possessor before ga che. This pronoun can only be bound by John and not by Paul. However, after much consideration and speaker consultation, it appears that although (25) is acceptable, it is rather awkward. (22b) is the preferred utterance. This point will be elaborated further in section 3.3. C (25) Passive Interpretation a John; bei a Paulj touh-jo keuih(/*j ga che c.f. (22b) C L John BEI C L Paul steal-PTV 3p C L car 'John's own car was stolen on him by Paul.' This possessor-relation also holds true in Mandarin as well. This characteristic was noted by Ting (1993), although his analysis is based on a gapped/non-gapped hypothesis for the direct object: ...indirect passives are allowed in Chinese with a coreferential possessor, overt or empty... (9) a. Gapped direct: Zhangsani bei wo da - le [e]j yixia. BEI I hit-ASP once 'Zhangsan was hit once by me.' 30 b. Non-gapped direct: Zhangsani bei wo da - le ta; yixia. BEI I hit-ASP he once 'Zhangsan was hit once by me.' c. Indirect (with post-verbal retained object): Zhangsan; bei tufei sha - le (taj) baba BEI bandit kill-ASP he father 'Zhangsan was affected by the bandit's killing his father.' One thing crucial about (9) is that the matrix subject is in an obligatory coreferential relation with a post-verbal element... .. .if the post-verbal pronominal is not coreferential with the matrix subject, it is ill-formed, as shown in (39): (39) *Zhangsanj bei wo da - le taj. BEI I hit-ASP he 'Zhangsan; was affected by my hitting himj.' Ting (1993:240-241, 247) Now consider the following examples: C (26) a) myu go yun bo che dou BEI a Paul touh-gwo every person's car all BEI C L Paul steal-VFV 'Every person's car had been stolen by Paul before.' b) myu-go yun dou BEI a Paul touh-gwo bo che every person all BEI C L Paul steal-PFV (possessive) car 'Every person was affected by Paul's having stolen their cars.' c) myu-go yiuij dou BEI a Paulj touh-gwo tsigei boj che every person all BEI C L Paul steal-PYV self 's car 'Every person was affected by Paul's having stolen their own cars.' Regular Passive Indirect Passive Indirect Passive A s discussed in Section 1.3.3, tsigei is a locally bound, subject-oriented reflexive. Example (26c) demonstrates three properties of the Indirect Passive: (i) The quantified subject must be in an Argument-position (i.e. [spec, IP]) in order to bind the anaphor in object position (ii) The quantified subject must be in the governing domain of the anaphor to be an eligible binder for the anaphor (iii) Paul cannot be an antecedent for the self-anaphor In fact, (i) can be generalized to the Regular passive as well. The N P in sentence initial position of a passive is the subject of the clause. It is in an Argument-position and can bind pronouns and anaphors. (26c) crucially shows that the subject must be in an A-position. 31 Quantifiers can be dislocated into A'-positions, but in such a position they cannot bind any anaphor . E (27) a)* Every womani, I hate her,. b) Every womanj says Bill hates herj. C c)* myu go neuyunj, ngoh tsun keuihj every C L woman lp hate 3p 'Every woman, I hate her.' d) myu go neuyunj dou wa a Bi l l tsurj keuihj every C L woman all say C L Bill hate 3p 'Every woman says Bill hates her.' However, because the quantified expression in (26c) is able to bind the possessive self-anaphor in object position, it must be in an A-position. (26c) further shows that the subject and the BEI-phrase must be in the same governing domain since the local subject-oriented self-anaphor tsigei can be bound by the subject, (ii) is a reflex of Principle A of the Generalized Binding Theory: Principle A : An anaphor must be bound in its local domain. (Aoun, 1985) However, Paul in (26c) is also a subject (subject of touh 'steal' in [spec, vP]) and is a closer potential binder for the anaphor than myu-go yun 'every person' is. However, it is not possible for Paul to serve as the binder of tsigei in (26c) for the sentence to have a passive interpretation. If Paul were to be construed as the antecedent for the anaphor, the meaning of the sentence would change. It would no longer have the semantics of an Indirect passive but rather means something like 'Every person has let Paul steal his own car before.' With Paul as the antecedent of tsigei, (26c) would carry the flavor of a permissive causative instead (also see (32b) below). I am grateful to Hamida Demirdache for this suggestion. 32 Due to this seemingly unusual restriction, I argue that the possessor relation is crucial for a passive interpretation. If this relation is blocked or disrupted then no passive meaning is possible. Ting's example (39) given above shows that the possessor relation also holds in Mandarin as well. Compare (22b) with (28) below. In (28a) the subject John no longer binds the possessor pronoun in object position, so that John is no longer construed as the owner of the car. In (28b) an overt nominal possessor is substituted. Again the possessor relation between the subject and the object does not hold. C (28) a) a Johnj BEI a Paulj touh - jo keuihk ga che C L John let C L Paul steal-PFV 3p C L car 'Johnj let Paulj steal hisk car.' b) a Johnj BEI a Paulj touh-jo Marykgache C L John let C L Paul steal-VFN Mary C L car 'Johnj let Paulj steal Maryk's car.' (27a) and (28b) also have the ring of a permissive causative rather than a passive sentence. Thus, we observe from (25) and (28) that the possessor-relation is crucial in deriving a 'passive' meaning from the sentence. When there is no possessor-relation between the N P in subject position and the N P in object position, the sentence cannot be interpreted as a passive. In fact, BEI can no longer be translated as meaning \"by\" in these cases. Instead, it conveys the meaning of \"let\". This line of argumentation also explains why Paul cannot be the binder for the anaphor in object position, although it could qualify as one. The possessor-relation between the subject myu-goyun 'every person' would be disrupted if Paul was the antecedent of tsigei. At this point, the disruption of the possessor-relation serves as one explanation why Paul cannot bind tsigei, however, we shall see later how the structure can actually rule out this possibility as well. We now come to the question of whether these Indirect passives are better analyzed as causatives instead. 33 3.2 Indirect Passives as Causatives The relevant examples for this next section are (25) and (28a), repeated below: C (25) a Johnj bei a Paulj touh-jo keuihj/.j ga che C L John BEI C L Paul steal-PFV 3p C L car 'John's own car was stolen on him by Paul.' C (28a) a Johnj BEI a Paulj touh-jo keuh\\ ga che C L John let C L Paul steal-PFY 3p C L car 'Johnj let Paulj steal hisk car.' Let's first consider (25), where John is the possessor of ga che 'the car.' Semantically, the N P preceding BEI, John, does not have the interpretation of being a direct causer of any action. There is no sense in which John is an agent who caused the action of Paul's stealing of his own car. Rather, John is interpreted as the person who experienced Paul's action of stealing; he is the affectee. However, turning to (28a), where the subject does not possess the object, there is a sense in which John permitted Paul to steal someone else's car. This sentence can certainly be interpreted as a permissive causative; it cannot be interpreted as a passive due to the lack of a possessor-relation. Therefore, the answer to the question whether Indirect passives can be analyzed as causatives is both yes and no. Indirect passives are ambiguous. None of the other Cantonese passives described in (22) have ambiguous interpretations other than the Indirect Passive. They are comparable to English 'get/have' passives which can have both a passive and causative meaning. E (29) John got his car stolen by Paul. - ^ M E A N I N G 1: John's car was stolen by Paul Passive 2: John made Paul steal his car Causative 34 In fact (25), repeated below, can also have a causative meaning, although the subject possesses the object. A possessor-relation does not necessarily block a causative interpretation, but it is crucial in maintaining a passive meaning. (25) a Johnj bei a Paulj touh - jo keuihj ga che CL John BEI C L Paul steal-PFV 3p C L car (i) 'John's own car was stolen on him by Paul.' (ii) 'Johnj let Paul steal hisj car.' The difference in interpretation results from whether the subject is construed as the agent or patient of the sentence. This point can be demonstrated by the insertion of an agent-oriented adverb such as gouyi 'intentionally.' 1 4 C (30) a) a Johnj gouyi bei a Paulj touh - jo keuihj ga che causative only C L John intentionally let C L Paul steal-FFV 3p C L car 'John intentionally had his car stolen by Paul.' b) a Johns bei a Paulj gouyi touh - jo keuihj ga che C L John BEI C L Paul intentionally steal-PFV 3p C L car 'John had his car stolen on him intentionally by Paul.' passive O R O R 'John let Paul intentionally steal his car.' causative When it is modified by the agent-oriented adverb, John in (30a) is definitely interpreted as the initiator, or causer of the stealing action performed by Paul. The example in (30b) is slightly more complicated. When the adverb modifies the lower subject Paul, Paul is construed as the agent of the verb 'steal'. However, John can be construed as either the experiencer or causer of the action, depending on whether John is a derived subject or an underived (true) subject. If John is a derived subject (i.e. such as that in a passive) then it is construed as an experiencer/patient, having no volitional control over Paul's action of 'stealing'. However, if John is a true subject (i.e. such as in a causative) it can still be 141 am grateful to Hamida Demirdache who reminded me of the importance of this test. 35 construed as the indirect causer/agent of the action. The agent-oriented adverb is able to modify Paul because it is the internal subject of the lower verb phrase. The same BEI-construction may have two different interpretations, causative or passive, depending what type of role the N P in [spec, IP] plays. This ambiguity is not confined to Cantonese alone. To further support my earlier suggestion that Cantonese B E I has subsumed the functions of colloquial Mandarin jiao and rang Chappell (1986) makes the following comment regarding Mandarin. A possible linguistic motivation behind the rang and jiao passives retaining their 'traditional' meaning and not developing a translatese form, as did the bei passive, is their use elsewhere as both causative verbs and independent lexical verbs. Bei does not have any other grammatical uses apart from that of passive exponent, nor can it be used as an independently occurring verb.. .so that the problem of ambiguity does not arise. .. .Since the rang and jiao passives are restricted to an adversative interpretation, this aids in the maintenance of distinct semantic structures for rang and jiao passive and causative constructions...The only potential for ambiguity with the passive lies with intentional speech-act causatives...when the latter contain semantically transitive verbs in their predicate but the direct object denoting the undergoer of the action has been ellipsed: (38) Ta rang wo piping - le yixia. 3sg R A N G I criticize-COM one:CL Causative: (i) 'He had me criticize (somebody).' Passive: (ii) 'He was criticized by me.' Chappell (1986:1037-1038) To better understand Chappell's observation, consider the bracketed structures of (31) which are reproductions of her example #38, with the trace and ellipsed object filled in. M (31) b) [ipTaj rang [vpwo [vppiping-le [Np(taj) yixia]]]] Causative a) [ipTa rang [vpwo [vppiping-le [Nptj yixia]]]] Passive t I Chappell notes that there is an ambiguity with sentences such as her #38 because the sentence can either be interpreted as a causative with an ellipsed object N P , or a passive with a moved object N P . If Cantonese B E I has all the properties of Mandarin bei, rang and jiao, then this ambiguity of causative/passive alternation in Indirect passives is expected. 36 The binding properties of sentences involving BEI which have a causative meaning are also different from those with a passive meaning. C (32) a) myu-go yunj dou BEI a Paul touh-gwo tsigeij bo che Passive Meaning every person all BEI C L Paul steal-WW self 's car 'Every person was affected by Paul's having stolen their own cars.' b) myu-go yunj dou BEI a Paulj touh tsigei*j/j bo che Causative Meaning every person all BEI C L Paul steal self 's car 'Every person has let Paul steal his own car.' c) myu-go yunj dou BEI a Paulj touh keuihj/*j/k bo che Causative Meaning every person all BEI C L Paul steal 3p 's car 'Every person has let Paul steal their cars.' (32b) shows that once the sentence acquires a causative interpretation the N P in subject position can no longer bind a reflexive appearing before the object. Unlike (26) from above, the subject myu-go yun 'every person' is not longer able to bind tsgei 'self in (32b). This serves to show that somehow, in these constructions, the structural subject N P is no longer in the binding domain of the reflexive and thus cannot serve as a potential antecedent for it. (32c) reinforces the inference that the [spec, IP] position is no longer contained within the minimal domain of the reflexive. In this example, the subject is able to serve as an antecedent for the third person pronoun, keuih. If the subject myu-go yun is within the minimal domain, Principle B would be violated if the subject was co-indexed with the pronoun in Object position. However, no such violation arises. Principle B also rules out the co-indexation of Paul with the object keuih bo che 'his/her car' because Paul is contained within the minimal domain of the pronoun. Notice that it is possible for the pronoun to be bound by some other antecedent N P (index k) that is not mentioned in (32c), but it cannot be bound by Paul. 37 From these observations, I propose that the expressions of passive meaning have only one governing domain (i.e. is mono-clausal) while those of causative meaning have two (i.e. bi-clausal). A explication of the structures of the BEI-construction will clarify matters. This will be pursued in Chapter 6. 3.3 Indirect Passives as Possessor-Raising The question remains as to how to deal with the apparent 'third' argument in Indirect passives structures. For instances where this construction is a causative the problem is trivial. Causative structures typically introduce another argument into the predicate and assign it the Causer theta role. However, passives cannot. Recall the obligatory possessor-relation between the subject and object of passive-type sentences. These constructions can be analyzed as instances of possessor-raising.15 Thus, there is not actually an extra argument in these types of structures, but rather the possessor is raised out of the base object position. Compare (33a) and (33b): C (33) a) [a Johnj] bei a Paul touh - jo [ t, ga che] C L John BEI C L Paul steal-VVV Clear 'John had his car stolen on him by Paul.' b) [a Johnga che] bei a Paul touh-jo C L John C L car BEI C L Paul steal-VVV 'John's car was stolen by Paul.' (33b) has virtually the same structure as (33a), except that the entire N P has been fronted rather than the possessor alone. Under this analysis, the apparent 'third' N P in Indirect passives is no longer a problem. The N P in subject position is not an independent argument in itself, requiring an independent theta role, but is actually the possessor raised out of the lower object N P . In accordance with possessor-raising structures we may argue that the 38 possessor receives its theta role directly from the object. I will be adopting this view for the rest of the paper. However, the contrast between speaker preference for (22b) over (25), repeated below, still remains to be explained even if we assume that possessor-raising is manifested in Indirect passives. (22b) a John bei a Paul touh-jo ga che. C L John BEI C L Paul steal PFV C L car 'John had his car stolen on him by Paul.' (25) a John; bei a Paulj touh-jo keuihj/.j ga che c.f. (33a) C L John BEI C L Paul steal-PFY 3p C L car 'John's own car was stolen on him by Paul.' Under the possessor-raising hypothesis, the overt possessive pronoun of (25) occupies the exact position where the trace of the raised possessor should be. It should not be possible for a pronoun to occupy this position if the trace is there already. At this point I do not have a full explanation for this phenomenon but one possible avenue of pursuit might be along the lines of 'clitic doubling' or 'clitic-reference' such as in the cases of Macedonia presented by Spencer (1991): As clitic pronouns develop a stronger attachment for a particular grammatical category, they begin to look more like (agreement) affixes than free standing syntactic constituents...In Macedonian this impression is strengthened by the fact that a definite direct object, or any indirect object, must be referenced ('reduplicated' in the descriptive literature's terminology) by a clitic pronoun agreeing for person , number and, in the case of 3sg, NPs, gender...Examples are given in 9.29 - 9.32: 9.29 M i ja dadoa smetka-ta 1 sg.-DAT 3sg.F-ACC gave bill-AKT-F 'He gave me the bill.' 9.30 Dajte mu ja kosula-ta give 3sg. M - D A T 3sg.F-ACC shirt-ART-F 'Give him the shirt.' 9.31 Nemu mu go dadov. To-him 3sg.M-DAT 3sg.M/N-ACC I-gave 'It was to him that I gave it.' I am indebted to Michael Rochemont who suggested this to me. 39 9.32 Mu go dadov pismence -to nemu a ne nejze 3sg.M-DAT 3sg.M/N I-gave note - A R T - N to-him and not to-her T gave the note to him, not to her.' Spencer(1991:359-360) Spencer's examples all show that a clitic is reduplicated for a definite direct object and any indirect object. In his 9.29 and 9.30 the clitic ja is a reduplicant for the feminine direct object (definite reference) nouns 'bil l ' and 'note'; in 9.31 and 9.32 the clitic mu is the reduplicant for the indirect object 'to him'. Although the exact mechanisms of this strategy have to be worked out, we can pursue an analysis of the Cantonese examples along these lines. Since I have argued that the Indirect Passive involves possessor-raising, the trace of the raised possessor naturally shares the exact same ^-features as the possessor itself. Therefore, whatever pronoun clitic is generated in that position will have the same ^-features as the raised possessor as well. The possessor was originally part of the D P of the direct object of the verb, and after it was raised it is possible that a clitic pronoun might be generated to reference the possessor N P . Passives of this sort are also associated with a definiteness effect as exemplified by the insertion of a demonstrative ni 'this' before an indefinite N P go yun 'a person': C (34) a) ni go yun bei a John touh-jo ga che this C L person BEI C L John steal-PFV C L car 'This person had his car stolen on him by John.' b)*go yun bei a John touh-jo ga che C L person BEI C L John steal-WY C L car ' Some person had his car stolen on him by John.' (34) shows that an indefinite N P is ungrammatical in these structures. However, the insertion of a demonstrative, giving definite reference, saves the sentence. These facts all point to the possibility of analyzing the overt possessor in (25) as a kind of 'referential clitic' in the spirit of Spencer. 40 To dissolve the ambiguous meanings associated with the Indirect passive, I will term the constructions with a passive interpretation as 'BEI-passive' and the ones with a causative meaning as 'BEI-causative.' In the following section we will further explore the properties of these two types of BEI. Chapter 4: BEI: Passive versus Causative Having teased apart the two distinct meanings associated with BEI we may now ask how are the passive and causative related in Cantonese? Are they derived from the same structures? From the above examples, we have already witnessed that BEI-causative and BEI-passive constructions are different in both their binding properties and theta grids. What causes this difference? 4.1 What is the Cantonese 'passive'? Let's refer back to Example (22), repeated below for convenience. C (22) a) a Paul touh-jo a John ga che Active C L Paul steal-PFV C L John 's car 'Paul stole John's car.' b) a John ga che bei a Paul touh-jo. Regular C L John's car BEI C L Paul steal PFV Passive 'John's car was stolen by Paul.' c) a John bei a Paul touh-jo ga che. Indirect C L John BEI C L Paul steal PFV C L car Passive 'John had his car stolen on him by Paul.' In the regular passive of (22b) the whole object N P has moved up, out of its base position, to an A-position (presumably [spec, IP]). We speculate that the N P has moved up to an A -position because it can serve as an antecedent for a local anaphor, as shown in (26c). For (22b) I have suggested that these might be instances of possessor-raising. In these cases, only the possessor of the object N P has raised out of the object base position, up to an argument-41 position (presumably [spec, IP] again). Not only can it serve as an antecedent to an anaphor in object position, as shown by (26c), but in the following cases as well: C (35) a) Johnj bei tsgeij go tsey touh-jo bo che John BEI self C L son steal-PFV C L car 'Johnj had his; car stolen on him by self Sj son.' b) myu go yun; bei tsgeij go tsey touh-jo bo che every person BEI self C L son steal-FFV C L car 'Every person, had hisj car stolen on him by self Sj son.' In comparison with the active sentence of (22a), we observe that something must move out of the canonical object position of the verb phrase in a BEI-passive construction. This 'something' can either be the whole object N P or its possessor alone. A BEI-passive construction where neither the object N P nor its possessor is fronted is ungrammatical. Compare (36c) with (36b). C (36) a) a Paul tek laahn-jo ga che C L Paul kick damage-PFV C L car 'Paul damaged the car.' (by kicking) b) gachebei a Paul tek laahn-jo C L car BEI C L Paul kick damage-PFV 'The car was damaged by Paul's kicking of it.' c) * bei a Paul tek laahn-jo ga che 1 6 BEI C L Paul kick damage-PFY C L car 'The car was damaged by Paul's kicking of it.' What could be forcing this kind of movement? I suggest that, similar to some previous analyses of English (Jaeggli, 1986; Baker, Johnson & Roberts, 1989), the passive function in Cantonese suppresses the verb's ability to assign Accusative case ( A C C ) . In a passive, if the object cannot receive abstract Case in its base position from the verb, then it must move to a See the Appendix 1 for a discussion of exceptional circumstances where the following is acceptable: \"bei a Paul tek laahn-jo ga che\" BEI CL Paul kick damage-WV CL car 'The car was damaged by Paul's kicking of it.' 42 Case position in order to obtain Case. Assuming we have a standard functional projection of 1 7 IP above V P then, the closest landing site where the N P can get structural Case would be [spec, IP]. However, recall in Section 3.0 where Adjectival passives were briefly mentioned. One might argue that these types of passives would pose a counterexample to our claim since the object still remains in-situ in these cases. Compare the Adjectival passive of (22e) and (36c) above, repeated as (37a) and (37b) respectively: C (37) a) bei Paul tek laahn-jo geche mo yuhn wuhyew. (22e) Adjectival BEI Paul kick damage- P F V L P car none people will want Passive 'No one will want the car that was damaged by Paul's kicking of it.' b) *bei a Paul tek laahn-jo ga che (36c) BEI C L Paul kick damage-PFV C L car 'The car was damaged by Paul's kicking of it.' c) * bei Paul tek laahn - jo ga che mo yuhn wuh yew. (36c) as an BEI Paul kick damage- P F V C L car none people will want Adjectival * ' N o one will want the car was damaged by Paul's kicking of it.' Passive Upon closer inspection, the Adjectival passive of (37a) does not pose a counterexample to our claim, but rather supports it. Notice that the linking particle ge in (37a) is different from the noun classifier ga in (37b, c). Although phonologically similar, the substitution of one particle for the other changes the grammaticality of the sentence. Compare (37a) and (37c). The linking particle of (37a) serves the function of a relativizer, linking the head noun che 'car' with the phrase that modifies it. It is possible that this particle has inherent abstract Case and is able to directly Case mark the following NP. However, the noun classifier in (37b, c) lacks such an ability, so substitution of this classifier in (37c) yields an ungrammatical sentence because the object N P cannot get Case. I use IP in a neutral way and not to specifically denote an Infl. phrase, a Tense phrase or an AGR phrase. 43 Examples involving apparent double objects in passives also support my proposal that the Object N P moves for Case reasons. Ditransative verbs participating in passive structures were very hard to find because none of them seem to satisfy the adversity requirement, that was mentioned earlier, on Cantonese passives. However, there are some examples involving pai 'to send (a person)' and gsei 'to mail' . The former has already been exemplified in (15b), repeated below as (38a); geei 'to mail' will be presented in (38c): C (38) a) howtseun bei howgouk p a i - j o heui toywan school-principle BEI school-administration send-FFV to Taiwan 'The principle was sent to Taiwan by the administration.' *b) heui toywan bei howgouk pai howtseun to Taiwan BEI school-administration send school-principle 'To Taiwan was sent the principle by the administration.' *c) howtseun bei howgouk p a i - j o toywan school-principle BEI school-administration send-PVW Taiwan 'The principle sent to Taiwan by the administration.' d) bougwo bei a John mh-gokyi gasi-jo heui toywan parcel BEI C L John not-notice mail - P F V to Taiwan 'The parcel was mailed to Taiwan accidentally by John.' *e) heui toywan bei a John mh-gokyi gasi-jo bougwo to Taiwan BEI C L John not-notice mail - P F V parcel 'To Taiwan the parcel was mailed accidentally by John.' f) bougwo bei a John mh-gokyi gaei-jo bey Mary parcel BEI C L John not-notice mail - P F V for Mary 'The parcel was mailed to Mary accidentally by John.' g) bey Mary bei a John mh-gokyi gasi-jo bougwo for Mary BEI C L John not-notice mail - P F V parcel 'Mary was mailed the parcel accidentally by John.' h) bougwo bei a John mh-gokyi gasi-jo Mary parcel BEI C L John not-notice mail - P F V Mary 'The parcel was mailed to Mary accidentally by John.' 44 These constructions all involve three NPs. In (38a) they are: howtseurj 'principal', howgouk 'school administration and toywan 'Taiwan.'; in (38d) they are: John, bougwo 'parcel' and toywan 'Taiwan'; in (38f) they are: John, bougwo 'parcel' and Mary. Notice that only the direct object can be moved but not the indirect object. These so-called ditransitive verbs are also composed of more than just the bare verb itself. There is a 'coverb' 1 8 before each indirect object: heui 'to' and bey 'for'. When these 'coverbs' are omitted the sentence becomes ungrammatical, (38c, h). I argue that these coverbs serve to Case mark the indirect object. If they are absent, then the indirect object N P cannot receive Case. If the indirect object moves to [spec, IP] to get Case then the direct object will be stranded without Case. I propose that the reason for movement of the direct object is to obtain Case, but since the indirect objects already have a way of satisfying case in their base positions, they are never moved in passives. Under this analysis, there are no true double object constructions in Cantonese, where the verb can directly assign Case to two NPs. In constructions involving 3 NPs I argue that a coverb must be present as a Case assignor because the verb can directly assign Case to only 1 N P . Therefore, although these so-called ditransitive verbs involve 3 NPs, and 3 theta roles, the verb must subcategorize for a coverb phrase in order for Case to be satisfied on the indirect object. Returning to the Cantonese passive, one notices that the verb's ability to discharge the external theta role is not suppressed (i.e. it does not reduce the number of arguments in the verb's theta grid). There is always an overt agent phrase following BEI. foLrrSecdon TilC0VCrbS 3 5 P r e p ° S i t i 0 n S - T h d r C a t e § o r i a l s t a t u s i s - \" dea r but more discussion will 45 I have also argued that the Cantonese passive does not introduce another theta role. Even for Indirect passives I have argued that the N P in [spec, IP] position is not an extra argument, but rather a raised possessor19. 4.1.1 The Function of BEI We have already seen an example which shows that the BEI-phrase cannot be an adjunct. Example (35) reveals that a local reflexive N P following BEI can be bound by the matrix subject. This means that both the anaphor and its antecedent must be in the same governing domain and the subject is in an A-position which c-commands the anaphor. It is not possible for an anaphor in an adjunct such as the prepositional \"by+Agent\" phrase in English passives to be bound. E (39) * [ipMaryj was [vphit [P Pby herself;.]] Example (40) shows that the N P following BEI can also serve as an antecedent for binding. C (40) a Johnj bei a Paulj touh-jo tsgei*j/j ga che Causative only C L John let C L Paul steal-PFV self C L car (no passive possible) 'John let Paulj steal hisj own car.' In (40) it is not possible for John to bind the anaphor. Principle A would predict this if John is not contained within the governing domain of the anaphor. Paul is the closer governor. Again, because the subject and object do not hold a possessor-relation no passive interpretation is possible. Now, consider again (22b), the Regular passive. (22b) a John ga che bei a Paul touh-jo. Regular CL John's car BEI C L Paul steal PFV Passive 'John's car was stolen by Paul.' These cases are somewhat problematic in that it is not clear how the object in base position receives Case once the possessor is raised. Perhaps the raised possessor is in a chain relation with its object. However, this i a general problem for all analyses of possessor-raising which is beyond the scope of this paper. 46 Assuming the previous hypothesis regarding the function of the Cantonese passive, the N P John is fine because it has both Case and a theta role. How does the N P Paul get Case then? It can no longer move to [spec, IP] for structural Case because this position is occupied by the object already. According to Chomsky (1995) the verb is not able to assign Case directly to its external argument because sisterhood is a necessity of direct Case assignment. In consideration of these factors, the logical subject Paul should be illicit since it cannot satisfy the Case Filter. However, I suggest that this N P is saved by the presence of BEI. BEI acts as an exceptional Case assignor to this N P , performing the same function as certain functional prepositions in English. E (41) a) For John to leave would be a mistake, b)* John to leave would be a mistake, c) He is envious of Susan. d)* He is envious Susan For in (41a) and of in (41c) contributes no identifiable semantics to the sentences above except that they serve to Case mark the NPs John and Susan respectively. The omission of these prepositions as Case assignors in (41 b, d) yields ungrammatical sentences because the NPs cannot get Case from their respective infinitival and adjectival predicates. Another similarity shared between the functional uses of these prepositions and BEI-passive is that they do not introduce an extra argument into the predicate. Along similar lines, would it be possible for us to suggest that BEI was a preposition? 4.1.2 BEI-passive as a Preposition Several previous analyses based on Mandarin (Cheng 1988; LaPolla 1988; L i 1990; Ting 1993; L i 1994) have argued against treating BEI as a preposition. There are two main arguments against this proposal. First, phrases introduced by prepositions are usually adjuncts 47 and so are optional. However, (41) readily shows that not all phrases introduced by prepositions are adjuncts. The prepositions in (41) are certainly obligatory for Case reasons. The second objection to analyzing BEI as a preposition involves the fact that Chinese disallows preposition stranding. Consider the following: ( 4 2 ) English a) John, I like to talk to. Cantonese b) ngohjung-yi tuhng a John king-gey. I like with C L John have-conversation c) *a Johnj, ngohjung-yi tuhng tj king-gey C L John I like with have-conversation d) a Johnj, ngoh jung-yi tuhng keuihj king-gey C L John I like with 3p have-conversation ' A s for John, I like to talk to him.' Mandarin e) John bei pian le John BEI cheat P F V 'John was cheated.' (42a) shows a stranded preposition in an object topicalized structure. The same configuration given in (41c) for Cantonese is ungrammatical. If we want to maintain the topicalized object in (41c), we must re-inserting a pronoun N P after the preposition so that it will not be stranded. Based on these facts, it has previously been argued that a structure such as (41e) in Mandarin, where the agent N P has been omitted, should be ungrammatical if BEI were a preposition. In (41e), the 'prepositional' BEI would be stranded without an overt agent N P . However, this property will never be problematic in Cantonese because we have seen that, for independent reasons, the agent phrase can never be omitted. Therefore, there is still a possibility we can maintain that BEI is a preposition in Cantonese. 4.1.3 BEI-passive as a Verb A . L i (1990) considers the possibility of Mandarin BEI as a verb. She concludes that this proposal is not tenable for five different reasons. Firstly, unlike most verbs, BEI in BEI-48 constructions cannot take any aspectual marker, such as the perfective. Secondly, it cannot participate in the common A-not-A question forms. It also cannot stand alone in a predicate or be a simple answer to a question as a simple verb can. Finally, it does not seem to introduce any argument, unless it is possible to have canonical (underived) avalent verbs (i.e. verbs which have no arguments). The NPs both preceding and following BEI are independently introduced by the verb in the V P , and not by BEI. (43) exemplifies these properties using the verb chi 'eat' in Mandarin and its correspondant, sik 'eat', in Cantonese. The example will show that lexical entries which are verbal, like chi, can participate in the tests mentioned by A . L i . (43) a) taking of an aspect marker M ta chi - le fan C keuih s i k - j o faan 3p eaf-PFV rice 3p eat-PFV rice 'He has eaten rice.' 'He has eaten rice.' b) participate in the in A-not-A question form M ta chi-bu-chi fan? C keuih sik-mh-sik faan? 3p eat-not-eat rice 3p eat-not-eat rice ' W i l l he eat rice?' 'Wi l l he eat rice?' c) stand alone in a predicate as a simple verb M ta chi fan C keuih sik faan 3p eat rice 3p eat rice 'He eats rice.' 'He eats rice.' d) can be a simple answer to a question M A i t a chi-bu-chi fan? B: chi! C . A : keuih sik-mh-sik faan? B_i sik! 3p eat-not-eat rice eat 3p eat-not-eat rice eat ' W i l l he eat rice?' 'Sure, eat!' ' W i l l he eat rice?' 'Sure, eat!' e ) introduce an argument M ta chi fan chi C keuih sik faan sik 3p eat rice 3p eat rice »/~^. 'He eats rice.' ta fan 'He eats rice.' keuih faan A . Li 's arguments against treating Mandarin BEI as a verb carry over into Cantonese. (44) summarizes these 5 objections as applied to Cantonese BEI: 49 C (44) a) it cannot take an aspect marker like most verbs (i) * ga che b e i - j o a Paul touh-jo C L car BEI -PFV C L Paul steal-VVW —>intended meaning: 'The car had been stolen by Paul.' b) even verbs which cannot take an aspect marker (like the copula) can participate in the in A-not-A question form; BEI cannot (ii) * ga che bei - mh - bei a Paul touh? C L car BEI- not -BEI C L Paul steal —•no meaning can be assigned. c) it cannot stand alone in a predicate as a simple verb (iii) * keuih bei ngoh 3p BEI l p —»no meaning can be assigned d) it cannot be a simple answer to a question (iv) A : heuy-mh-heuy ga che bei a Paul touh-jo? B: *bei is -not- is C L car BEI C L Paul steal-TT-V BEI 'Was it by Paul that the car was stolen?' 'Was.' e) it does not introduce any argument; unless it is possible to have avalent verbs (v) ga che bei a Paul touh - jo C L car BEI C L Paul steal-PFV 'The car was stolen by Paul.' —>in (v) both NPs ga che 'car' and Paul are introduced by the touh 'steal' and not BEI Regarding (44e) it might be possible to argue that BEI is a verb which introduces a Theme argument into Indirect passive constructions . That is, for an example such as (45a), below, John may be analyzable as a second theme which is introduced by BEI: C (45) a) a John bei a Paul da do tsek gou si - io weyzi C L John BEI C L Paul hit till C L dog die-WV to-the-extent-that 'John had his dog hit to death by Paul.' *b) a John do si - jo weyzi bei a Paul da tsek gou C L John till die-VFW to-the-extent-that BEI C L Paul hit C L dog 'John till his death had his dog hit by Paul.' ' Much thanks to M. Rochemont for reminding me of this possibilty 50 The resultative complement sei-jo weyzi 'until dead' is theme-oriented only. Although it can modify the theme gou 'dog' as in (45a), it cannot modify John in (45b), irregardless of whether the sentence has a passive or causative interpretation. This test can be used to show that John can be either an Agent or an Experiencer, but not a Theme. 4.1.4 BEI-Causative as a Verb From (44) we cannot definitively conclude that all occurrences of BEI are non-verbal. Consider the instances of BEI used to derive a causative meaning. BEI-causative can perform functions (a-e) mentioned in (44 ). This is illustrated in (46) below. C (46) a) take an aspect marker John bei- jo keuih cheuk gaai. John let - P F V him go-out street 'John has let him go outside.' b) participate in a A-not-A question form John bei-mh-bei keuih cheuk gaai? John let-not-let him go-out street 'Is John letting him go outside?' c) to stand alone in a predicate keuih bei ngoh 3p let l p 'He/she lets me.' d) answer to a simple question A : bei keuih heui - mah? B: bei! let 3p go V - P R T let! ' Letting him/her go out?' ' O f course!' e) introduces the Causer theta role John bei a Paul touh-jo Mary ga che John let C L Paul steal-FFV Mary's car 'John let Paul steal Mary's car.' In light of these facts, it might be plausible to analyze BEI-causative as a verb. Actually the categorical status of BEI has been hotly debated precisely due to its dual characteristics of 51 having prepositional and verbal properties, depending on the environment it is used in and what its semantics are. However, this categorial ambivalence is not isolated to B E I alone. 4.1.5 The Ambiguity of Categories Matthews & Y i p (1994) present an insightful comment regarding the ambiguity between prepositions and verbs in Cantonese. A s result of this ambiguity they term these items 'coverbs' instead for neutrality and so that they can analyze this group under the main category of verbs, with prepositions as a subclass. \"Whether prepositions exist in Chinese is also an open question. Functionally, the role of prepositions in expressing relationships between noun phrases is played by two different types of word: 'coverbs' and 'localizers'.. .The question thus arises whether the coverbs are functioning as prepositions or merely as verbs.. .Even in those contexts where they correspond to prepositions in English, these words behave like verbs in many respects...in so far as prepositions exist in Cantonese, they are a subclass of verbs which may be used as prepositions. Alternatively, the coverbs may be regarded as verbs in all their manifestations, in which case their characteristic use as 'coverbs' invariably involves a serial verb construction.\" Matthews & Yip (Y&P 1994:60) Most of these items termed as 'coverbs' behave in a similar fashion to BEI . They show both prepositional and verbal properties and adopt a different semantics depending on how they are used. Take for example, hai 'at'. C (47) a) keuih yihga hai guownsi b)* guownsi; keuih yihga hai tj 3p now at company company 3p now at 'He is at the company now.' 'The company, he is there now.' c) keuih yihga hai- gan guownsi d) keuih yihga hai-mh-hai guownsi, ah? 3p now be at - P R O G company 3p now at - not- at company, P R T 'He is at the company right now.' 'Is he at the company now?' The 'coverb' hai 'at' cannot be stranded as (47b) shows, but it can also take a verbal aspect marker, as in (47c), and participate in an A-not -A configuration, as in (47d). A 'coverb' like ging has both the meanings of ' v i a ' (preposition) and 'to pass by' (verb) depending on its context. On the other hand there are items like lihng 'make' which are considered verbs but 52 do not participate in any of the tests. Lihng is analyzed ( M & Y ) as participating in a causative structure similar to BEI-causative where it introduces the causer argument. C (48) a) a John lihng a Paul tsi dou C L John make C L Paul late arrive 'John caused Paul to arrive late.' b) a John lihng tsi dou ge heuy a Paul C L John make late arrive L P is C L Paul 'It was Paul who John made arrive late.' c) * a John l ihng- jo / gan / g w o / h o i a Paul tsi dou C L John make - P F V / P R O G / E X P / H A B C L Paul late arrive 'John caused/ is causing/ has caused/ usually cause Paul to arrive late.' d) * a John lihng -mh- lihng a Paul tsi dou, ah? C L John make-not-make C L Paul late arrive, P R T ' D i d or did not John make Paul late?' In (48) tsi dou 'arrive late' is a one-place predicate, but there are two arguments in (48a): a causer and a patient. The verb lihng serves to introduce the causer John. Notice that it can be stranded without an N P complement as in (48b)21, but it cannot take any aspectual marker, (48c), or can it participate in A-not-A structures, (48d). The ambiguous qualities of these types of items makes it very difficult to determine their categorical status. The question of whether prepositions exist in Cantonese still remains open. So is the possibility of whether we can claim BEI-passive is a pure preposition. Alternatively, we could posit the existence of an entirely new category which I will term \"neutrals\". A l l the problematic 'coverbs' considered in this section would belong to this category, as would BEI. These neutrals are have both verbal and prepositional properties but 1 The verb lihng in (48b) subcategorizes for a CP and not an NP. (48b) [,Pa John [vplihng tsi dou [CP ge[VP heuy a Paul]]]] CLJohn make late arrive LP is CL Paul 'It was Paul who John made arrive late.' 53 are not entirely one or the other. However, their behaviour remains constant within a certain structure. That is, in one given structure it will act either as a preposition or a verb, but not both. The ability of neutrals to have dual functions could derive from the fact that they possess a feature matrix that contains both verbal and prepositional properties. A n alternative explanation is that they are not specified with any features at all but acquire their different interpretations in virtue of the structure that they generate or participate in. I will explore this latter option further in the discussion with BEI in Chapter 6. 4.2 What is the Cantonese 'causative'? In exploring the BEI-passive we have already partly answered the above question. While the BEI-Causative constructions behave differently than BEI-passive in regards to binding (see examples 25-28, 30, 32, 35), agent-oriented pronoun distribution (30, 45) and properties of BEI (section 4.1), they do share several characteristics. First, the subject N P can serve as an antecedent to an anaphor in a BEI-construction with either a causative or a passive interpretation. C (49) a) Johnj bei a Paul hoy dou tsgeij porchan passive only John BEI C L Paul ruin until self bankrupt interpretation 'Johnj was ruined by Paul to the extent that he went bankrupt.' b) a Johnj bei a Paulj hoy dou tsgeij porchan causative only C L John BEI C L Paul ruin until self bankrupt interpretation 'John let Paul ruin himself into bankruptcy.' c) a Johnj bei a Paulj hoy dou keuihj porchan causative only C L John BEI C L Paul ruin until 3p bankrupt interpretation 'John let Paul ruin him till he was bankrupt.' As shown by (27), elements in A'-positions cannot bind an anaphor. The binding in (49a, c) suggest that the subject of the BEI-phrase in both passives and causatives the occupies an A -position. The causative in (49b) also shows that the N P following the BEI-phrase can serve 54 as an antecedent to an anaphor in object position. Thus, this position must also be an A -position. We have already observed that the N P following BEI cannot be omitted in any Cantonese BEI-passive construction. This generalization holds true even for the causative interpretation of the Indirect passive: C (50) *a Johnj bei a ^a^Ik touh - jo ga chej C L John let steal - P F V car 'Johnj let e steal the carj.' (50) shows that the causee N P after BEI cannot be omitted. Since John is not the possessor of the object, it must be an independent argument rather than a raised possessor. Where (50) is interpreted with causative semantics, there are three separate arguments and theta roles: Causer, Causee and Theme. The Causee and the Theme roles are related to the lower verb touh 'steal', while the Causer role is introduced by BEI. We can surmise that because the object can remain in-situ, Case is properly satisfied on the object N P . Thus it seems that the lower verb has not lost its ability to assign Accusative case, allowing the object to remain in-situ. The subject John preceding BEI may be argued to receive structural case in [spec, IP]. Thus again, it is BEI which seems to be case-marking the N P which follows it. However, unlike BEI-passive, it is possible for an aspectual marker to intervene between BEI and the Causee N P . BEI-causative can be modified by an aspectual affix. This has already been shown in (46a), repeated below: C (46) a) John bei- jo keuih cheuk gaai. John let -PFV him go-out street 'John has let him go outside.' The intervention by other modifiers, like a time adverbial, is still not possible, as exhibited by (10b) and (51). I will return to this in a later section. 55 C (51) * John bei kumyak keuih cheuk gaai. John let yesterday him go-out street 'John let him go outside yesterday.' Causation augments the theta grid of the verb by the introduction of a Causer. Example (30) above has already demonstrated that the subject N P John does have volition in these types of sentences, and is in fact the Causer of the event rather than a Patient. The introduction of theta roles are usually accomplished by lexical categories such as verbs. Verbs are also capable of assigning Case to NPs. From the examples, BEI-causative is seen to be consistent with other causative constructions in being able to introduce an extra theta role. For all the examples we have considered so far, BEI-causative behaves like a lexical causative verb in its semantic interpretation and ability to participate in tests for verbs. The chart below summarizes the properties of BEI-passive and BEI-causative. Table 1 Passive BEI Causative BEI Example Subject NP can bind a possessive pronoun in Object position 25 Subject NP can bind a local self-anaphor in Object position X 26c Subject NP must be co-referential with Object NP X 22b NP after BEI can bind a possessive pronoun in Object position X X 32c NP after BEI can bind a local self-anaphor in Object position X ^ 32b, 40 Subject NP can be analyzed as raised possessor •/ X 33a NP after BEI can be modified by agent-oriented adverb 30b Subject NP can be modified by agent-oriented adverb X 30a Object NP (all or possessor only) must move out of base-position ^ X 33, 36 Phrase has 3 identifiable theta roles X 22b BEI-phrase is an optional adjunct X X 35 Participates in tests for verbs (i.e. aspect, A-not-A...etc ) X 44, 46 Cannot be stranded without a following NP X 13c BEI + NP can be separated by an Aspect marker X •/ 44a, 46a In light of these facts, it might be plausible to analyze BEI-causative as a verb since it can behave like one, unlike BEI-passive. However, even i f we conclude that BEI-causative is verbal we cannot propose the same for BEI-passive. As exemplified by (44) it clearly does 56 not behave like a verb. We also mentioned earlier that there is no objection to analyzing BEI-passive as prepositional in Cantonese due to the fact that the Agent is never omitted (except with the use of beih). But on the other hand, we cannot argue that BEI-causative is a functional preposition either because is needs to be able to retain the functions of a verb, as seen in (46). The categorical status of BEI has been hotly debated precisely due to its dual characteristics of having prepositional and verbal properties, depending on the environment it is used in and what its semantics are. One possible solution to this dilemma might be to argue that there are two different BEIs; one is verbal (corresponding to BEI-causative), the other is prepositional (corresponding to BEI-passive). I will consider these possibilities as I examine the tree structures of the BEI-construction in the following sections. Chapter 5; Previous Analyses of the BEI-construction Although there is currently no published analyses of Cantonese BEI-constructions, it would be helpful to compare how the BEI-construction is dealt with in Mandarin. In this chapter I will briefly review two analyses of BEI in Mandarin. A . Y . H . L i (1990) proposes the Mandarin BEI-construction is a passive and that it involves NP-movement. Her analysis was based on two assumptions about Chinese: 1) Chinese is head-final in D-structure; and 2) Case assignment is to the right of the verb, while theta role assignment is to the left. Using her tests mentioned in Example 44, above, she concludes that BEI is neither a preposition or a verb. Rather, it is a kind of identifier which signals a passive. She treats the BEI+NP phrase as an argument. The following illustrates how her system works. [+Case] .... • '\\ M (52) a) [sSubNP [VP ObjNP V ]] = [ sSubNP [VP V ObjNP ]] [+Case] 57 [+Case] b) [ S [ V P bei NP [ypObjNP V ]] = [§ ObjNP [VP bei N P [VP V ]] e f+CaseJ ^ * [ s e [ V P V ObjNP]] c) [s *[vP bei N P [ V P ObjNP 1 ObjNP2 V ]] = [s ObjNPl [V>e /NP[ V P V pro ObjNP2]] 9 — \\^f[+Case] (52a) is an example of an active sentence, while (52b, c) are passive Z>e/-sentences. A . L i argues that all the N P arguments must be generated to the left of the verb for proper theta role assignment. In example (52a), the Agent 0-role is being assigned to what she terms the \"subject\" position. The subject N P is also base-generated in this position. I assume the \"subject\" position means [spec, IP] in G B theory. In this position the N P can receive structural Case. However, the object N P must move to the right of the verb in order to get Case. In (52b), a regular passive sentence, A . L i argues that the Agent 9-role is being assigned directly to the bei+NV phrase. Because the 0-role is not being assigned to [spec, IP], it becomes a non-theta position. Therefore, the object has an option to move into this position. According to A . Li 's analysis, the object is forced move into the position because there are no expletives in Mandarin. If the object does not move into sentence initial position the Extended Projection Principle would be violated. A . L i proposes the (52c) represents a \"source double object structure.\" These cases look very much like the Indirect passives observed in Cantonese. NP1 is the possessor while NP2 is the object possessed. Because 2 Case positions are available both NPs can move to satisfy its Case requirement. To ensure that it is NP1 which raises and not NP2, A . L i appeals to the following structure (her examples #46 and #47), where pro is in the specifier of NP2: 58 #46) a) Ta qiang wo qian he rob me money 'He robbed me of my money.' b) VP NP2 V wo NP1 V pro NPO qiang I qian #47) Zhangsan bei wo qiang le [pro qian] Zhangsan by me rob ASP money 'Zhangsan was affected by the fact that I robbed him of his money.' A. L i (1990: 179-180) Under the structures proposed by A. Li, NP2 cannot move to sentence initial position because then pro could not be identified. The sentence would thus be ungrammatical. If, however, NP1 moves, then pro would be identified with NP1, whereby deriving a possessor-relation. A. Li's analysis faces several problems. First, the dichotomy of Case and 0-role assignment is surprising. Why must 6-roles be assigned in one direction while Case assignment is done in the opposite direction? In current theories, such as GB theory, direct Case assignment correlates with theta role assignment. In order for object NPs in her system to satisfy Case, downward movement is forced. Not only is downward movement problematic, but it is unclear what and where the postverbal position is. Perhaps we can assume that while the tree structure is being built up, an empty position is specifically generated after the verb in order to allow for movement of the object later. However, (52b) shows that the object does not necessarily need to move into this position. Generating a position that may or may not be used later in the derivation runs counter to economy principles, which create positions only if they are required by the structure. If the postverbal position is not generated as the structure is being built up, but only comes into existence later 59 when the object moves there, the 'extension principle' (Chomsky 1995:327) would be violated. This principle guarantees cyclicity in a derivation because it prohibits the returning back to an earlier stage of a derivation and building more structure. Second, it is unclear how the generation of pro in the specifier position of NP2 is justified. How come there is no pro in the specifier of NP1? A final less serious note, although she concludes that Mandarin BEI is not a verb, she nonetheless labels the [bei+NP] phrase as a VP (example 52). Additionally, A. Li's analysis says nothing about the causative type structures that can be generated with Cantonese BEI. In light of this, her proposal cannot adequately deal with the BEI-patterns we observe in Cantonese. Next, let's consider an analysis by Ting (1993). He agrees with A. Li that the Mandarin BEI-construction is passive but proposes that BEI is a verb. It assigns a theta role to the matrix subject which is based generated and not NP-moved. He argues that the BEI-construction is bi-clausal because of the following example (his example #6): #6) Zhangsan; bei wo da - le taj yixia Zhangsan BEI / hit -ASP he once 'Zhangsan was hit once by me.' Ting (1993:243) Ting argues that binding of the pronoun would be impossible because of Principle B of the Binding Theory if the BEI-construction was mono-clausal. However, his claim regarding the grammaticality of #6 is questionable. A. Li gives a similar sentence to argue that the BEI-construction is mono-clausal (A. Li's #50): #50) *Taj bei wo qiang le taj. he by me rob ASP he 'He was affected by the fact that I robbed him.' A. L i (1990:180) 60 From the Cantonese example (32a) above, it was shown that the subject NP can bind an anaphor in object position. This type of behavior also sheds doubt into Ting's claim that the BEI-passive construction is bi-clausal. Putting the aside the question of grammaticality judgments, Ting's analysis also hinges on the proposal that the BEI-phrase selects for a CP complement. A null operator sits in the head of the CP and A'-binds any arguments (overt or covert) in the object position. The structures I posit for the ^/-construction in Chinese are given in (27). (27) a. Non-gapped direct: Zhangsan; bei [Cp 0; [wo da - le ta;]]] BEI I hit-ASP he 'Zhangsan was hit by me.' b. Gapped direct: Zhangsan; bei [cp Oj [wo da - le />roj]]] BEI I hit-ASP 'Zhangsan was hit by me.' c. Indirect (with retained objects): Zhangsanj bei [cp Oj [wo da - le ta; haizi]]] BEI I hit-ASP he child 'Zhangsan was affected by my hitting his child.' .. .1 propose that the null operator, which forms a SPEC-head agreement chain with C in the predicate, agrees with the subject and that the predication relation between the matrix subject and the CP complement of bei is therefore licensed.. .The hypothesis that the null operator ensures the predication relation between the matrix subject and the CP complement of bei is supported by the obligatory coindexation relation that holds between the matrix subject and the element in the embedded clause taken by bei. Ting(1993: 245, 247) Ting's analysis is attractive for a couple of reasons. Firstly, it correctly predicts the possessor-relation in Indirect passives. Secondly, his structures can be easily modified to generate a causative BEI-sentence. All that needs to be done is to say that the verb bei in sentences with causative semantics specifically assigns a Causer 9-role and no null operator exists. The BEI-construction is still bi-clausal, but whether it selects for a CP or an IP is not crucial. However, this type of analysis faces one serious problem if it is carried over into 61 Cantonese. A s pointed out earlier, example (32a) shows that the subject N P can serve as an antecedent to an anaphor in object position. If we adopt Ting's structure, the subject N P and the null operator would both be co-indexed with the anaphor. However, operators in A ' -positions cannot bind anaphors. This was exhibited in my earlier example (27). Therefore, Ting's analysis cannot be carried over into Cantonese either because it is dependent on an A ' -binding approach. In the next chapter I will discuss possible tree structures for the BEI-construction in order to arrive at an analysis which will handle the range of phenomenon we have observed in Cantonese. First, we need to consider the range of positions where BEI could be base-generated. Chapter 6: The Structure of the BEI-construction 6.1 The Interaction of Time adverb ials, Instrumental and Patient: Review of L i (1990) L i (1990) attempts to manipulate the positions of time adverbs, instrumentals and patients with respect to each other to argue that BEI resides outside of the V P . The following are her examples for Mandarin (I have corrected some typos) taken from her book Order and Constituency in Mandarin Chinese (pp. 169-170). She uses the BA-construction which preposes the object N P before the verb and allows it to have greater freedom of movement in the phrase as compared to a normal V+NP structure: (32) a. Tufei zai natian yong dao henhende ba tade erzi sha le bandit at that day with knife cruelly B A his son kill ASP 'The bandit killed his son with a knife cruelly on that day.' b. Tufei zai natian henhende yong dao ba tade erzi sha le at that day cruelly with knife B A his son c. Tufei zai natian ba tade erzi henhende yong dao sha le at that day B A his son cruelly with knife d. Tufei zai natian yong dao ba tade erzi henhende sha le at that day with knife B A his son cruelly 62 e. Tufei zai natian henhende ba tade erzi yong dao sha le at that day cruelly B A his son with knife f. Tufei zai natian yong dao henhende ba tade erzi sha le at that day with knife cruelly B A his son g. T u f e i yong dao zai natian henhende ba tade erzi sha le with knife at that day cruelly B A his son h. * Tufei henhende zai natian yong dao ba tade erzi sha le cruelly at that day with knife B A his son The examples in (32) show that the manner adverbial 'cruelly', the instrument 'with a knife', and the patient 'son' do not have fixed word order with respect to each other. However, the time expression 'that day' does not have the same freedom of word order. It must precede the manner adverbial, instrument, and patient. This difference is position can be straightforwardly captured by the generalization that time adverbials occur outside the V P but the others occur within the V P . . .Similarly, a bei phrase has a fixed order with respect to a patient, an instrument, and a manner adverbial: (33) a. Ta bei tufei yong dao henhende ba tade erzi sha le he by bandit with knife cruelly B A his son kill ASP 'He was affected by the fact that his son was killed by a bandit cruelly with a knife.' b. Ta bei tufei ba erzi henhende yong dao sha le by B A son cruelly with knife c. Ta bei tufei henhende ba erzi yong dao sha le by cruelly B A son with knife d. *Ta yong dao bei tufei ba erzi sha le with knife by bandit B A son e *Ta ba erzi bei tufei yong dao sha le BA son by bandit with knife f. *Ta henhende bei tufei ba erzi sha le cruelly by bandit B A son The sentences in (33) show that [bei NP] occurs outside the VP. It does not occur after the patient NP (ba NP), instrument, or manner adverbial. If [bei NP] occurs outside the V P , we expect it to have free word ordering with respect to time adverbials, which is the case: (33) g. Ta zai natian bei ren sha le he at that day by man kill ASP 'He was killed on that day.' h. Ta bei ren zai natian sha le he by man at that day kill ASP 'He was killed on that day.' (A. L i 1990: 169-170) If we adopt A . L i ' s approach into Cantonese, we expect to derive the same set of results. Compare (48a-h) with her (32a-h) example. (I have omitted a corresponding (48f) because L i has repeated her (32a) as (32f) again). 63 C (53) kumyuk keuih gouyi your} hup ba ji tseurj soumai yesterday 3p intentionally use box B A CL gun hide 'Yesterday, he intentionally used a box to hide the gun. 'Yesterday, he used a box intentionally to hide the gun. 'Yesterday, he intentionally hid the gun (by) using a box. 'Yesterday, he hid the gun intentionally (by) using a box a) keuih kumyuk your) hup gouyi ba ji tseurj sou - mai SUBJ. TIME INSTR. M A N N . OBJ. V b) keuih kumyuk gouyi your) hup ba ji tseurj sou SUBJ. TIME M A N N . INSTR. OBJ. V mai ??c) keuih kumyuk baji tseurj gouyi your) hup sou - mai SUBJ. TIME OBJ. M A N N . INSTR. V d) keuih kumyuk your) hup baji tseurj gouyi sou - mai SUBJ. TIME INSTR. OBJ. M A N N . V e) keuih kumyuk gouyi baji tseurj your) hup sou - mai SUBJ. TIME M A N N . OBJ. INSTR. V Sg) keuih youn hup kumyuk gouyi baji tseun sou SUBJ. INSTR. TIME M A N N . OBJ. V mai •Sh) keuih gouyi kumyuk yourj hup baji tseurj sou - mai SUBJ. M A N N . TIME INSTR. OBJ. V i) keuih gouyi youn hup kumyuk baji tseurj sou - mai SUBJ. M A N N . INSTR. TIME OBJ. V j) keuih gouyi your) hup baji tseurj kumyuk sou - mai SUBJ. M A N N . INSTR. OBJ. TIME V *k) keuih gouyi your) hup baji tseurj sou SUBJ. M A N N . INSTR. OBJ. V mai kumyuk TIME *1) kumyuk keuih yourj hup baji tseurj sou - mai gouyi TIME SUBJ. INSTR. OBJ. V M A N N . \"m) kumyuk keuih gouyi baji tseurj sou - mai yourj hup TIME SUBJ. M A N N . OBJ. V INSTR. *n) gouyi keuih baji tseurj youq hup sou - mai M A N N . SUBJ. OBJ. INSTR. V *o) baji tseug keuih gouyi youg hup sou OBJ. M A N N . INSTR. V mai TIME = time adverbial INSTR. = instrumental M A N N . = manner adverb OBJ. = object T H M = theme SUBJ. = subject 64 As expected, (53a-e), like Li's (32a-f) are fine, with the time adverbial, manner adverb and instrument having free word order related to each other. The only exception is (53c) which sounded very unnatural to my consultants, perhaps because the object has been displaced too far from the verb. However, (53g) and (53h) are also acceptable, contrary to Li's claim. In fact, as (53i, j) shows, the time adverbial can be placed directly before the verb. (53k, 1, m) illustrate that the only restriction seems to be that these modifiers must precede the verb. Because of the use of the BA-construction, the object must obligatorily precede the verb. In normal sentence patterns where the object follows the verb, tests to show whether the modifiers can precede the object would be irrelevant because the sentence will always be ungrammatical with these modifiers coming after the verb. However, speakers have noted that it is more 'natural' to have the time adverbial out in the front of the sentence, before the subject, rather than after it. In fact, this position seems reserved for the time adverbial only. (53n, o) show that the instrument, manner adverb, and the preposed object cannot precede the subject. From (53) we can speculate that time adverbials can occur freely anywhere in the sentence as long as it precedes the verb. Manner and instrument are restricted to the domain between IP and V. However, we do not know at this point whether they occur outside or inside the verb phrase. Tests involving a modal wuh 'will' may help to clarify this situation. I will be treating the modal as a kind of inflection which occupies the head of IP.2 2 C (54) a) tingyuk keuih wuh gouyi yourj ssk da- laahn ga che tomorrow 3p will intentionally use stone hit-damage C L car 'Tomorrow, he will intentionally use a stone to damage the car.' b) keuih tingyuk wuh gouyi youn sek da- laahn ga che SUBJ. TIME MOD M A N N . INSTR. V Alternatively, it can head its own Modal Phrase (ModP) inside IP. 65 c) keuih wuh tingyuk gouyi yourj ssk da- laahn ga che SUBJ. M O D T I M E M A N N . INSTR. V d) keuih tingyuk wuh youn ssk gouyi da - laahn ga che SUBJ . T I M E M O D INSTR. M A N N . V e) * keuih yourj sek wuh gouyi da-laahn ga che SUBJ . INSTR. M O D M A N N . V f) ? keuih gouyi wuh yourj ssk da-laahn ga che SUBJ . M A N N . M O D INSTR. V (54a, b, c) demonstrate that the time adverbial can occur freely before and after the modal as predicted. Comparison of (54c) with (54d) shows that manner and instrument have no fixed word order with respect to each other when they occur after the modal. However, when the instrument precedes the modal the sentence becomes ungrammatical, as in (54e). Speakers are uncertain of their judgments for (54f), where the manner adverb precedes the modal. Some of the time it sounds acceptable, while at other times it is not. A possible intended meaning would be something like, \"He is intentionally planning to use a stone to damage the car\" where the emphasis would be on the 'intention' of the event. However, there is no question that the sentences where the manner and instrument are located after the modal are acceptable. From this, we may speculate that manner and instrument occur somewhere below the IP but before V . A . Li 's suggestion that they are confined within V P (or vP) may be correct. This is logical since manner and instrument are predicative modifiers. Because time adverbials, being sentential modifiers, seem to have a relatively free distribution they are not good tests for structural location of items within a phrase. As a side note, negation always precedes the modal: C (55) a) tingyuk keuih mh-wuh gouyi yourj ssk da- laahn ga che tomorrow 3p not-will intentionally use stone hit-damage C L car 'Tomorrow, he will not intentionally use a stone to damage the car.' 66 b) keuih tingyuk mh - wuh gouyi yourj ssk da- laahn ga che SUBJ . T I M E N E G M O D M A N N . INSTR. V c) keuih mh - wuh tingyuk gouyi yourj ssk da- laahn ga che SUBJ . N E G M O D T I M E M A N N . INSTR. V d) keuih tingyuk mh - wuh youn ssk gouyi da - laahn ga che SUBJ . T I M E N E G M O D INSTR. M A N N . V We shall see in the ensuing discussion that manner adverbs may venture out of the verb phrase depending on certain circumstances. In sum, the previous discussion has identified the fact that the adverbials of instrument and manner generally occur in a position which is below the IP projection but above the V -head. Time adverbials appear to have a relative freedom in word order. Where they precisely are base-generated are indeterminate at this point. 6.1.1 BEI and the Interaction with Modifiers Assuming that manner and instrument cannot occur higher than IP, we can use their interaction with the BEI phrase to determine where BEI might be generated. C (56) a) keuih wuh gouyi sik-sei di toen Active 3p will intentionally eat-all C L candy 'He/she will intentionally eat up all the candy.' b) di toen wuh BEI keuih gouyi sik-sei Passive T H M M O D A G T M A N N . V 'The candy was all eaten up by him.' c) * di toen BEI keuih wuh gouyi sik-sei T H M A G T M O D M A N N . V d) * di toen BEI wuh keuih gouyi sik-sei T H M M O D A G T M A N N . V e) * di toen wuh gouyi BEI keuih sik-sei T H M M O D M A N N . A G T V f) * di toen gouyi wuh BEI keuih sik-sei T H M M A N N . M O D A G T V 67 The passive of (56a) is rendered by (56b). The sentence is fine when the modal precedes the BEI-phrase. However, if the modal intervenes between BEI and the agent N P , or if it follows the BEI-phrase, the sentence is uninterpretable, (56c, d). Sentences (56e, f) are also ungrammatical, as should be expected. We have already seen from (30) that an agent-oriented adverb such as gouyi 'intentionally' cannot precede the BEI phrase in a passive interpretation. The reason behind this is perhaps because the Agent N P always follows BEI. This adverb might require a local c-commanded relation with its controller (the Agent) in order to be licensed. If we substitute the examples with an instrumental modifier we would achieve the same results in grammaticality. A n instrument cannot precede the BEI-phrase presumably because it is too far from the verb to be locally controlled by it, and thus cannot modify the verb. C (57) a) keuih wuh yoerj geuk tek-laahn ga che Active 3p will use foot kick damage C L car 'He/she will use his/her foot to kick the car until it is damaged.' b) ga che wuh BEI keuih yoerj geuk tek-laahn Passive T H M M O D A G T M A N N . V 'The car will be damaged by him through kicking.' c) * ga che BEI keuih wuh yoerj geuk tek-laahn T H M A G T M O D M A N N . V d) * ga che BEI wuh keuih yoen geuk tek-laahn T H M M O D A G T M A N N . V e) * ga che wuh yoen geuk BEI keuih tek-laahn T H M M O D M A N N . A G T V f) * ga che yoerj geuk wuh BEI keuih tek-laahn T H M M A N N . M O D A G T V From (56) and (57) we can tentatively conclude that BEI-passive must be generated below IP but before the sites for the manner and instrument modifiers. Assuming that these modifiers 68 are confined within V P (or vP), BEI would occupy a position just above the verb phrase, but below the IP. Turning to BEI-causative, it seems to interact with predicative modifiers and the modal in the same way as BEI-passive does. However, in these structures it is possible to have the manner adverb precede the BEI phrase because there is a volitional agent N P that can c-command and locally control it. However, the semantics of the sentence would be altered (compare (58a) and (58c). If the manner adverb can precede BEI, and BEI precedes the verb phrase, then this seems to run counter to Li 's claim that manner adverbs are restricted to the verb phrase. This is not necessarily the case if we postulate that BEI-causative is a verb and the manner adverb is generated within the BEI-phrase. This property supports the suggestion that BEI-causative is actually a verb. C (58)23 a) ngoh wuh bei keuihj gouyi tseetmor keuih tsgeij Causative lp will let 3p intentionally depress him self A G T i M O D A G T 2 M A N N . V T H M T will let him feel sorry for himself intentionally.' (i.e. he is making himself depressed by his own free will) b) * ngoh bei keuihj wuh gouyi tseetmor keuih tsgeij A G T i A G T 2 M O D M A N N . V T H M c) ngoh wuh gouyi bei keuihj tseetmor keuih tsgeij A G T i M O D M A N N . A G T 2 V T H M T will intentionally let him feel sorry for h imsel f (i.e. I want him to continue to be depressed) d) ngoh gouyi wuh bei keuihj tseetmor keuih tsgeij A G T i M A N N . M O D A G T 2 V T H M There are no corresponding examples in Mandarin because the Mandarin BEI-construction has no causative interpretations associated with it. 69 From (58b) we maintain that BEI-causative, like BEI-passive cannot precede the modal, would suggest that both BEIs may be generated in similar structures. We turn to a formulation of the trees for BEI in the next section. This 6.2 The BEI tree We have so far determined that the BEI-passive involves a verb and its object2 4. Thus we minimally need a structure such as Figure (1.1). At this point, I am making an assumption that has yet to be confirmed that only transitive verbs can participate in a passive construction. If so then I adopt Chomsky (1995) and posit a structure such as (Fig. 1.2) where there is a light verb v projection. Chomsky notes that this projection houses a verbal affixal feature. To make this more concrete suppose that it is a 'transitivizing' feature which allows 9^ the verb to assign A C C case . Therefore, if only transitive verbs can participate in the passive construction then the basic structure must minimally contain the light v projection. If the structure of (Fig. 1.2) is standardly adopted in my analysis, it makes a very strong • 26 prediction that only transitive verbs can participate in a BEI-construction. Unaccusatives have no underived subject; intransitives have no underived object; ditransitives (if they exist) have too many arguments, (see the end of Section 4.1 for a discussion of double object constructions). So far I have found no examples which disproves my claim. C (59) a) [ipkeuihj [ypsei -jo tj]] b)*[ipkeuihj BEI [vpsei - jo tj]] Unaccusative 3p die-PFV 3p BEI die-PFV 'He/she died.' *'He/she was died.' 2 4 i.e. the transitive class of verbs. 2 5 Due to the interaction with passive one might argue that this 'transitivizing' feature is deactivated, or the whole vP node is eliminated (for economy reasons) because passive suppresses the verb's ability to assign Accusative case anyways. I leave this objection as an open question for the time being. 2 6 1 am using this term to denote verbs which can directly assign Case to maximum one NP. 2 7 1 am using this term to denote verbs which can directly assign Case to 2 NPs without the necessity of a functional preposition or a coverb. 70 c) [ipkeuih [vphaam]] b)*[ip BEI [vpkeuih haam]] Intransitive 3p cry 3p cry ' He/she is crying.' * ' He/she was cried.' d) ? [IP a Paul [vpsoorj buun su a John]] C L Paul give C L book C L John 'Paul gave John a book.' e) [IP a Paul [vpsoorj buun su bey a John]] C L Paw/ give C L oooA: to C L John 'Paul gave a book to John.' f) * [IP buun su BEI [Vp a Paul soon a John]] Di transitive C L ooo& BEI C L Paw/ give C L John 'The book was given to John by Paul.' Adopting a Hale & Keyser (1993) system in which lexical relations are associated with the generation of argument structure, Chomsky argues that the v-VP configuration can be interpretted as the structure whereby the external theta role is licensed 2 8. \"The external role is a property of the v-VP configuration, and a specifier bearing this role is therefore a necessary part of the configuration; a transitive verb assigns an external 9-role by definition.\" Chomsky (1995: 316) 2 8 A similar proposal was explored in Kratzer (1993). She argues that the external theta role is not directly assigned by the lexical verb head but rather by an inflectional Voice projection generated above VP. The first move was to sever the external argument from its verb. The second move was to let the inflectional head that is responsible for accusative case introduce the external argument...The fact that the external argument must be realized outside of the V P follows from the Realization Principle and the fact that external arguments are introduced by inflectional heads. (55) The Realization Principle Arguments of a lexical item must be realized within the projection of that item. (56) VoiceP external Voice' argument Voice V P NP V V Kratzer (1994:124) Her idea of a Voice projection parallels Chomsky's light verb projection, which is an inflectional (functional) projection. Combined with the suggestion that v contains a 'transitivizing' feature which allows the verb to assign Accusative case once adjoined to v we are able to realize Kratzer's suggestion in a Minimalist framework. 71 Extending the v-node to include the specifier position we derive (Fig. 1.3), where the subject is generated. We can argue that such a structure is automatically generated when Selects chooses transitive verbs from the Numeration and CHL maps them into a derivation. If passive in Cantonese is syntactic then even in passive structures the v-node is still present, and the v-VP configuration necessarily licenses the appearance of the external theta role. Suppression of the external theta role is accomplished at the lexical level. If passive were syntactic then the [spec, vP] position would already be present and cannot be suppressed in the syntax. The problem specific to Cantonese then, would not be how to get rid of the agent in a passive, but rather, how to keep it. A s a reflex of the v-VP configuration, which is naturally part of the lexical structure of all transitive verbs (if Chomsky is correct in his generalization) we have a means of ensuring that the position of the external theta role is preserved. This requirement is necessary due to the obligatory-Agent constraint on the BEI-construction. Figure 1: Preliminary Trees 1.1) X \\ L 2 ) v ' L 3 ) v P 1.4) IP V Object v V P Subject v' spec V V Object v V P I ? V Object Figure (1.4), with IP used neutrally, would denote the top of the tree structure. However, the question is what is the complement of I? Is it vP? Continuing from (Fig. 1.3) we have several choices. I will first consider two: (i) BEI, either causative/verb or passive/preposition, adjoins to vP; vP is the complement of I; (ii) there is another projection above vP containing 72 BEI, which is responsible for either a causative or passive interpretation; this projection is the complement of I. 6.2.1 Proposal 1 Several previous proposals have aimed at analyzing Mandarin BEI-passive as participating in a serial verb configuration (Chu 1973; Cheng 1988; Matthews & Y i p 1994). BEI is the first member of a the serial construction (called the 'coverb'). BEI and its complement N P are adjoined to the verbal projection. For passive interpretations, BEI+NP would be a prepositional projection; for causative, BEI+NP is a verbal projection. These structures are exemplified below. Figure 2: Proposal 1: Passive BEI Figure 3: Proposal 1: Causative BEI IP IP Although the BEI-phrase is situated within vP, these structures do not necessarily pose a difficulty regarding the interaction of BEI with manner and instrumental adverb phrases. We can bypass this hazard by stipulating that these modifiers are generated no higher than v'. However, I argue that Figures 2 and 3 could not be the correct structures for BEI because of two real problems. 73 Firstly, it is undesirable to stipulate two BEIs which differ in category but not in structural positions or functions. These two BEIs seem to serve the same function of Case marking the N P that follows it. Although it is tempting to segregate BEI-passive and BEI-causative because of their differences in interpretation and the properties listed in Table 1, it is apparent from the discussion in Section 4.1.5 that there is no concrete evidence that prepositions and verbs are separate categories. There are other items which behave much like BEI does. We would be led to stipulate two homophonous expressions, corresponding to the appropriate category, for these items as well. Secondly, and more seriously, (60) below demonstrates that several binding facts do not obtain if we adopt the structures in Figures 2 and 3. C (60) a) a Johnj bei a Paulj touh-jo gache Passive C L John BEI C L Paul steal-PFW C L car 'John's own car was stolen on him by Paul.' b) a Johnj bei a Paulj touh-jo tsgei*;/j ga che Causative only C L John let C L Paul steal-PFV self C L car (no passive possible) 'John let Paulj steal hisj own car.' c) a Johnj bei a Paulj touh-jo keuihj/*j ga che Causative only C L John let C L Paul steal-PFV 3p C L car (no passive possible) 'Johnj let Paulj steal hisj car.' A s previously mentioned, pronouns and anaphors are governed by different principles of the Binding Theory and must be c-commanded by its antecedent. Binding Theory (A) An anaphor is bound in its governing category. (B) A pronoun is free in its governing category. (C) An R-expression is free everywhere. Chomsky (1981:188) Paul in (60b) should be able to serve as an antecedent for the reflexive in object position because it is within the governing category of the reflexive, and it is the closest available 74 subject N P . However, Figure 3 does not predict this because Paul does not c-command the reflexive and thus cannot serve as a potential antecedent. The N P in [spec, IP] cannot serve as a potential antecedent to the reflexive because it is not contained within its minimal domain. (60c) shows that the causative interpretation must be bi-clausal in order to allow for the co-indexing of the pronoun with the subject in [spec, IP], else a Principle B violation would arise. If causatives are bi-clausal then (60b) is predicted to be ungrammatical because of a Principle A violation: the anaphor is not bound within its minimal domain, the lower IP. However, this is not the case as witnessed by the grammaticality of (60b). For BEI-causative, at least, the structure presented in Figure 3 cannot be correct. Since the vP adjunction analysis for BEI as separate PP and V P categories cannot bypass this complication, let us consider other alternatives. , 6.2.2 Proposal 2 Recall from the earlier discussion that there were two possibilities for the expansion of Figure (1.3). The second option is by postulating another projection over vP which is responsible for the difference in interpretations for the arguments of vP. This maximal projection would be the sister of I. The problems encountered in Proposal 1 can be overcome if we postulate trees of the following configuration: Figure 4: Proposal 2: Passive BEI Figure 5: Proposal 2: Causative BEI If B E I participated in the structures given above, the binding relations discussed earlier would obtain. It would no longer be problematic for Paul in (60b) to bind the self-anaphor because the position of the Causee now c-commands the object position. However, we are still left with the undesirable effect of having to postulate two different BEIs, which occupy similar syntactic structures. We would also have to argue that the prepositional B E I is somehow able to produce the effect of a passive by suppression of the Case assigning ability of the verb. Prepositions in general assign Case, but are not known to suppress it. In an effort to unify the two BEIs, let's consider another option. 6.2.3 Pvitter& Rosen (19931 on 'have ' Another possible solution may be to treat B E I much in the same way that Ritter and Rosen (1993) treat English 'have', which also has dual characteristics. Consider (61): E (61) John had Paul steal his car. - » 2 meanings (i) 'John' is the causer. (ii) 'John is the affectee Ritter and Rosen (1993) attribute this variability in meaning due to have's lack of any semantic content. \". . .the interpretation of have's argument as causer or experiencer comes from the role it plays in the event.. .The action of a causer marks the beginning point of an event.. .An argument that is interpreted as an experiencer can extend the event beyond the termination of action denoted by the predicate.. .Have adds an extra participant to the event or state denoted by its complement.. .Due to its lack of semantic specification, have cannot project into the syntax as an independent verb because it lacks the capacity to assign a thematic interpretation to its argument(s). Projection of an unspecified verb would result in a violation of the Principle of Full Interpretation (Chomsky 1986) because its arguments could not be interpreted at LF . The argument structure operation which combines the argument array of have with that of an independent predicate salvages the structure by providing a mechanism for interpreting the argument(s) of have.. .either the causer or experiencer...interpretations arise because adding have (and its argument) has the consequence of extending the duration of the event denoted by the predicate. There are two ways in which an event can be extended to include a new participant outside 76 the core event - the beginning point can be pushed back, or the endpoint can be opened up by including a non-delimited state.\" Ritter and Rosen2 9 (1993: 523-524) Essentially, how their system works, is whenever have is selected, it must be combined with another predicate to form a 'complex predicate.' Otherwise, the argument introduced by have cannot be fully interpreted. Whether an Experiencer or Causer interpretation is formed depends partly on the nature of the predicate that have combines with (i.e. whether it has causative/affected semantics) and partly on the way the event is extended (i.e. backwards or forwards). Ritter and Rosen's proposal is also attractive in that they mention an interesting correlation between have and Experiencer interpretations: \"the experiencer interpretation imposes additional constraints...there must be a coreferential ethical dative or a relation of possession between the subject and some argument of the complement\" (R&R: 526) This correlation has already been observed in Cantonese Indirect passives. If R & R ' s arguments for have can be carried over into an analysis for Cantonese BEI we would be able to account for its dual characteristics and also predict the possessor-relationship. However, this solution is not tenable for three different reasons. Firstly, BEI-passive does not introduce any extra argument that can be combined with a following predicate. This is most apparent in the cases involving Regular passives, even if one does want to contend that the possessor in Indirect passives is a separate argument on its own. Secondly, have in English is categorically a verb. It is still indeterminate at this point whether BEI is a verb or a preposition. (44) and (46) above show that it has properties of both. 77 Lastly, R & R point out that complex predicates can only denote one single event. Because of its impoverished semantic content, have does not have any event specification of its own to contribute to the proposition when it is combined with another predicate. Therefore, it should not be possible to negate the event denoted by have while confirming the event denoted by the predicate combined with have because they are essentially the same event. This contrasts with a true causative verb, such as make which has its own event specification, and can be negated. This is shown in (62). Contrast (62a) with have and (62b) with make. E (62) a) * Paul didn't have John go outside but he sneakily went outside anyways, b) v^Paul didn't make John go outside but he sneakily went outside anyways. If BEI was like have in needing to be combined with another predicate in order for its argument to be fully interpretable, we would expect that to get the same results as (62). That is, the event of BEI cannot be negated while the event of the lower predicate is confirmed. C (63) Sa Paul mo bei a John tseuk gaai butgwo keuih touhtouh-dey tseuk-jo heui C L Paul not BEI C L John go-out street but he sneakily go-out-PFV go 'Paul didn't let John go outside but he sneakily went outside anyways.' (63) shows that this prediction is not borne out. Thus, we cannot provide a parallel treatment for BEI as for English have. Cheng (1988) provides an analysis of BEI-constructions in Mandarin along similar lines as Ritter and Rosen. She treats Mandarin BEI as the first member of a serial verb construction. BEI combines with another predicate and their argument structures are combined. I propose to treat the bei-construction as a case of serial verb constructions in Mandarin...the characteristics shared by both the lexical and the syntactic serialization was that the argument structure of serialized verbs are \"combined\". Following Li , Y.-F. (1988), 2 9 Henceforth R&R 78 I assume an identification mechanism which pairs up arguments. That is, if V I has <1, 2> and V2 has <1>, then if <2> of V I is identified with <1> of V2, then V I <2> and V2 <1> are assigned the same N P . . .1 propose to treat bei as a verb. It takes two arguments, one external (patient) and one internal (agent). When it occurs in the serial verb constructions, its theta-roles are identified with the other verb in a sentence. The identification essentially internalizes the agent theta-role of the other verb while externalized its internal theta-role. Cheng (1988:18-19) The central idea behind the use of Cheng's identification mechanism is to combine the arguments of BEI with the arguments of a following predicate in such a way that the arguments switch roles. Example (64) illustrates this procedure. patient agent 1 2 1 2 M (64) John BEI Paul da - le BEI da John BEI Paul hit-PFV I 1 1 I 'John was hit by Paul.' In a serial verb structure the arguments of both predicates are combined and only 2 theta roles are assigned to 2 available NPs which are identified with each other. However, what happens in the Mandarin cases where the Agent phrase can be dropped? Cheng suggests that in cases with no overt agent, the BEI+VP sequence is reanalyzed as a single verb instead of a serial verb. ...when the NP following bei is null, reanalysis takes place and [bei+V] is reanalyzed as a verb.. .If [bei+V] is derived straight from the lexicon, there is no reason why the agent role disappears. On the other hand, with reanalysis, it is possible that since the NP following bei is null, reanalysis forces identification to discharge both theta-roles. Cheng (1988: 20) patient agent M ( 6 5 ) John BEI da-le BEI + da - » BEI da John BEI /zzr-PFV 'John was hit (by someone).' Since Cheng argues that no movement actually takes place in BEI-constructions, the null N P after BEI cannot be an N P trace or a wh-trace. The null N P also cannot be P R O because the position after BEI is governed. Therefore, it must be pro which takes a discourse antecedent. 79 Her analysis faces several problems. First, in order for her identification mechanism to work, there must be two different thematic hierarchies: one for BEI and one for the second predicate. While BEI has an external 9-role of Patient, and an internal 9-role of Agent, the exact opposite is true for a predicate such as da 'hit', in order for the arguments to be mapped properly with each other. However, in accordance with Theta Theory, a 9-role such as Patient is usually assigned to an internal argument, while the role of Agent is assigned to an external argument, not the other way around. Second, her reanalysis approach is an ad hoc method of getting rid of the external role of the second predicate when the N P following BEI is null. Consider (66): patient agent 1 2 1 2 M ( 6 6 ) John BEI da - le BEI da John BEI tof-PFV 1 1 1 I ' John was hit.' BEI + da -> BEI da Let's suppose that in the lexicon the serialization of BEI and da has already occurred and its arguments are combined as in (64). Reanalysis of this serialized structure must take place once the null argument of BEI is identified with the external role of the second predicate. It is unclear whether the external role of the second predicate must be null as well or if BEI with a null argument can combine with any other predicate. However, why must reanalysis take place in cases with no overt Agent, and how does reanalysis force the discharge of the Agent theta role? If [bei+V] were reanalyzed as a single verb, its structure would preclude any argument, null or overt from intervening between BEI and the second verb. Where then is the Agent 9-role discharged? 80 Third, what mechanism constrains the serialization of BEI? Can it be combined with any predicate or are there licensing conditions? That is, there is no discussion about which types of predicates are eligible to be combined with BEI. Lastly, this type of analysis faces the same problem that Ritter and Rosen's analysis has. That is, if the arguments of BEI and another predicate are combined we would expect that there is only one event. We have already seen that this is not the case with the BEI-construction. The negate-event-l/confirm-event-2 test in (63) shows that the events in a BEI-construction seem to be independent of each other. In addition to these problems, Cheng's analysis has nothing to contribute to the dilemma regarding BEI-constructions which have a causative type meaning. Nonetheless, it would still be desirable to find a unified account for the two seemingly disparate BEIs. We could resort to the earlier suggestion that it is a \"neutral\", having dual-functions, or we can explore the possibility that something else is affecting the behaviour of BEI. I will start with the latter possibility first in the next Section. 6.2.4 Proposal 3: Functional versus Lexical Differences It is clearly desirable, in light of the binding facts, to preserve a configuration similar to Figures 4 and 5, rather than 2 and 3. Before developing another approach, let's reconsider what it is in general that we want the passive and causative BEI structures to accomplish. 81 Figure 6: Proposal 3: General BEI-passive F igure 7: Proposal 3: General BEI-causative In a passive, I propose that the verb is altered so that it is no longer able to assign Accusative case, forcing movement of the object (or possessor-raising) to satisfy Case. This is exemplified by Figure 6 where the object of the lower V is raised to [spec, IP]. Because no new theta roles or arguments are introduced into the predicate and because the Accusative case assignment property of the verb itself has been altered, I propose that the formation of the passive is a functional process and so a functional projection is projected. The verb does not suddenly lose its ability to assign Accusative case. Case is being suppressed by a higher functional projection which alters the lower verb's Case assigning abilities. In a causative, the main verb is not altered. The verb retains the ability to assign Case to its object complement, so no movement is necessary. A n extra Causer theta role is introduced into the predicate, either by BEI or some other lexical head. The subject of this higher projection (Causer) later raises to [spec, IP]. It is typical of lexical projections, such as 8 2 V P 3 0 , to be able to introduce another argument31, but not of functional heads, such as IP. I argue that this construction, being causative, and being able to augment the theta grid, is lexical. We may be able to use this Functional / Lexical distinction to obtain the desired effects for the BEI-construction. Consider Figures 8 and 9: Figure 8: Proposal 3: 0-passive Figure 9: Proposal 3: 0-causative IP IP In Figures (8) and (9) the projections which dominate vP for passive and causative interpretations are functional and lexical projections, respectively. To form a passive construction, a functional phrase (FP), containing a null functional head, 0 , is generated above vP. The passive function licenses the appearance of this FP. The phonologically null head of FP houses the semantic features of [passive]. It is this functional projection which suppresses the Accusative case assigning ability of the main 3 0 It is the head of lexical projections which introduces the theta role. 3 1 See Bresnan (1972), Emonds (1985), and Larson (1988) 83 verb. This line of argumentation is desirable because we no longer have to stipulate that BEI was a preposition which can suppress Case. The head of FP strictly selects for a transitive verbal projection. This selectional restriction obtains the generalization that an N P always follows BEI. A s discussed earlier, the N P would be the external argument of the v-VP configuration, which is always the case in transitive verbs. Along the same lines, a lexical projection, containing a null lexical head, is generated over vP to form a causative construction. The extra 9-role is introduced by the phonologically null head 0 of this lexical projection. The lexical head also selects for a sentential complement (IP) resulting in the bi-clausal structure effect that we observe with the binding of anaphors. The selectional restrictions of the respective functional and lexical heads make certain predictions regarding the rest of the structure. I will return to this point later. Where does BEI fit in on all this? Dechaine (1993, and p.c.) suggests that in terms of the categorial features [ocNominal, aReferential, aFunctional], which have a unique specification for verbs, nouns, adjectives, etc., prepositions are syntactically transparent. That is, they are not specified for any of these features, if we adopt a privative system. This lack of specification may explain the ability of prepositions to alternate between being able to assign a theta role or not. While lexical prepositions exist there could be another class of 'grammatical prepositions' which have all the syntactic properties of the Lexical head P, but no thematic content (Williams, 1989a), (Dechaine, 1993:65). A Lexical head, such as V or N , can introduce additional argument structure while Functional heads, such as Comp or Tense, cannot. Because prepositions are not positively specified for any category they may 84 have the ability to be selected for either for a Lexical or Functional projection depending on the type of construction they are in. Rochemont (p.c.) also suggests that perhaps such 'grammatical prepositions' may also lack any semantic content but derive their meaning in virtue of the configuration they are found in. Abney (1987) also notes \"Words with no independent semantic content belong to the set of functional categories.\" Following along the lines of these suggestions, I propose that BEI is a category-neutral element (negatively specified for category features [-Nominal, -Referential]) and lacks semantic content. Therefore, BEI is a unmarked, functional element able to assign case like a functional preposition such as for or of. BEI acquires its meaning in virtue of the structure it participates it. Therefore, when it adjoins to a functional projection that has passive semantics, it assumes a passive interpretation; when adjoined to a lexical projection with causative semantics, it assumes a causative 'let' meaning. It is not BEI which affects the predicates of these constructions, but rather the FP and L P themselves. The same functional element BEI is selected for both causative and passive constructions because it is not positively specified for the Lexical/Functional distinction. A question for consideration is why must BEI selected for these two constructions versus something else? It could be the case that there are not other competing functional elements which are negatively specified the way BEI is. Another question we are faced with would be why do we need BEI if the work is being done by FP and LP? There are two motivations. Firstly, following Chomsky (1995) it is plausible that BEI is a kind of \"identifier\". For satisfaction of the PF component, BEI is selected for full interpretation of the null causative and passives morphemes. This is the overt phonological 'signal' of a passive or causative construction. Secondly, it has been 85 assumed that for independently motivated reasons, there is always an external 6-role in the lower V P . A s discussed earlier, the N P in [spec, vP] would have no way of getting case through movement since in both structures [spec, IP] is occupied. BEI is required in order for the subject N P to receive Case. BEI is adjoined in a position which can assign Case to the subject N P of the lower verb. A n alternative route that can be pursued is that there is a single BEI, specified as [-Nominal, -Referential]. BEI itself is capable of generating either a functional or lexical projection as the structure requires. We can dispense with the null F and L heads and replace them with BEI instead, thereby simplifying the system. The type of projection generated by BEI would depend on the syntactic structure. If a passive structure is called for then BEI would be selected to extend a functional projection; if a causative structure was needed then BEI would be selected to form a causative projection. This analysis is not equivalent to the one proposed earlier in which there were two different BEIs, a preposition and a verb. There is still only one element BEI, but because it is unspecified for either a functional or lexical distinction it can take on the properties and functions of both without any conflict. Because BEI must necessarily participate in a lexical projection, we can no longer specify that it is only a functional preposition. Instead, under this alternative, BEI is a neutral, underspecified for either distinction. The respective structures would look like Figures 10 and 11: 86 Figure 10: Proposal 3: BEI-passive F igure 11: Proposal 3: BEI-causative IP IP Object The advantage with this proposal is that there is no further need to postulate a separate F or L head that is invisible on the surface. BEI is the head of either the functional or lexical phrase, depending on which structure is generated. However, a more detailed investigation into how the mechanisms of this approach can be implemented is required before the proposal can be adopted. At present, both approaches are equally attractive. We no longer need to postulate that the BEI which participates in a passive construction is different from the BEI which participates in a causative construction. For the rest of the paper I will continue to use the null functional and lexical heads for ease of exposition. Also, in Chapter 6,1 will discuss whether my analysis can be applied to the Mandarin BEI-constructions as well. By maintaining that a separate Functional or Lexical projection is responsible for the passive and causative interpretations we can maintain that these functions are the equivalent in both dialects except for the form of the element which the respective projections select for. This item is BEI in Cantonese but can be BEI, R A N G or JIAO on Mandarin. 87 6.2.5 Consequences of Proposal 3 We can now address some of the idiosyncrasies of the BEI-construction. Firstly, recall from (46a) that an aspectual marker can intervene between BEI and the following N P in a causative, but not in a passive. This consequence follows in virtue of the structures proposed. Because the causative is a lexical projection it is not impossible for it to also project a functional extension of itself. This functional projection might be similar to the light verb projection. Suppose that L P is able to project a functional Aspect phrase which houses the aspectual markers. This explains why it is possible for BEI-causative to be modified by an aspectual marker, but not BEI-passive. The aspect marker is not actually affixed to BEI but to the head of LP. The speaker gets a sense that the marker is attached to BEI because the head is phonologically null. Normally the lexical head alone would raise to adjoin to the affix and check off its features. However, being adjoined to the L-head, the whole complex L-BEI may be raised and adjoined to the aspectual phrase, much like v - V to T movement in Chomsky (1995). This type of adjunction may yield a structure such as Figure (12)32. 3 2 If the Aspect selects only for a lexical projection (such as VP) then it could not be projected above another functional projection. The following type of structure could be prevented: Figure 13: Aspect and BEI-passive AspP 88 Figure 12: Aspect and BEI-causative AspP LP t; This configuration mirrors the v-VP configuration. Abstractly speaking, it might be possible to propose that the Aspect projection above L P is equivalent to the light verb projection above V P . That is the Asp-LP configuration is like the v-VP configuration. This step is desirable because it would make the lexical projection on par with the verbal projection and we could explain the availability of the extra external theta role position in the same way. O n the other hand, the passive FP could never participate in such a structure because it dominates another functional projection (FP-vP configuration). Therefore, it is never predicted to be able to assign an external argument 9-role. Although the structures presented in Figures 8 and 9 adequately account for the range of data so far, they do not predict any constituency between BEI+NP in a passive construction. However, the BEI-phrase does seem to behave like a unit. Although it is difficult to find tests to isolate the constituent, if it exists, there is some evidence in question-answer forms. C (67) a) A : neihbei beenyuk go da ga? B: bei a Paul! 2p BEI which one C L hit Q - P R T BEI C L Paul ' Who were you beat up by?' ' By Paul! ' b) A : juebo hey-mey bei yun touh-jo? B: hey! bei a Paul touhge! jewels is - not BEI person steal-PFV yesl BEI C L Paul steal P R T 'Were the jewels stolen by someone?' 'Yes! By Paul.' 8 9 (67a) shows that in answer to a question it is possible to respond only with BEI + Agent N P , but not BEI alone or the Agent alone. This constituency is not predicted by the current tree structures. 6.3 A Revision: Proposal 4 Although Proposal 3 can adequately handle the problems encountered in the other proposals, there is another possible structure. These are given in Figures (14) and (15): Figure 14: Proposal 4: 0-passive Figure 15: Proposal 4: 0-causative IP IP The main difference between Figures (8), (9) and Figures (14), (15) is the adjunction site of BEI. In (Fig. 14) and (Fig. 15) BEI is directly attached to the lower subject N P where it can assign Case directly. Because the mechanism of attachment is adjunction, BEI does not change the category of the maximal projection it is adjoined to. This is crucial because we want to maintain the c-command relation between [spec, vP] and the complement position so that we do not run into binding problems like (61b). 90 The structures of (Fig. 14) and (Fig. 15) make a prediction regarding constituency which (Fig. 8) and (Fig. 9) cannot. Namely, in the revised structure, BEI and the N P are predicted to function together as a constituent. So far, this prediction has been very difficult to test. We cannot use questions as diagnostics because Chinese is an in-situ language, and BEI+NP cannot be topicalized. The question-answer forms given in (67) serve as some evidence. However, for the causative type construction it is more desirable to maintain Figure (9) as the structure because the BEI and Causee N P are separate from each other. This makes it possible for an Aspect marker to intervene, as discussed. (46a) above demonstrates that the perfective aspectual marker -jo can intervene between BEI-causative and the Causee N P . C (46a) John bei- jo keuih cheuk gaai. John let -PFV him go-out street 'John has let him go outside.' The difficulty with a structure like (Fig. 15) is that there is no place for the aspectual marker to be inserted and/or generated between BEI and the N P . This is only problematic for BEI-causatives if we adopt (Fig. 15). In addition, the question-answer form tests used above derive different results when used in an expression with causative meaning: C (68) A L neih bei-mh-bei a Paul leih-a? Bj_ bei! *bei Paul! 2p let not let C L Paul come P R T let let Paul ' W i l l you let Paul come?' 'Sure!' 'Sure let Paul.' From (68) there does not seem to be any constituency involved in BEI-causative constructions. (Fig. 14) seems more desirable for BEI-passive because we want to capture the sense that BEI+Agent N P is a constituent. Although BEI, as previously argued, serves the same function in both Cantonese passive and causative constructions, it is possible that it is 91 generated in different positions for passive and causative. Suppose (Fig. 14) is the correct structure for passive interpretations of the BEI-construction, while (Fig.9) is the correct structure for causative interpretations. Although more data and research is required in this area to clarify the issue, the possibility seems plausible. After all, these structures result in different semantic interpretations. It is logical to assume that a difference in structure will reflect a difference in meaning. Maintaining the same structure for both BEI-causative and BEI-passive will not capture the fact that they are indeed different projections, having different properties and interpretations. Chapter 7: Final Comments This paper has attempted to show that the Cantonese BEI-construction can have causative and passive meanings. Although it behaves both like a passive (preposition) marker and a causative (verb) marker, there is actually no need to stipulate that it are two different homophonous items that have properties which do not fit neatly into any one category. Previous analyses have not been able to satisfactorily conclude the identity of BEI because they have mainly focused on trying to pinpoint its categorial status. The ambiguity arises precisely because BEI is not positively specified for any category and lacks semantic content. I have proposed that it is due to this unique status that BEI is able to participate in both causative and passive constructions. The discussion would benefit i f we can draw parallels between BEI and other 'coverbs' in Cantonese which show similar idiosyncrasies of being both verbal and prepositional. Would it be possible to expand the lexical/function projection hypothesis to account for the difference in behaviour for these items as well? In 92 Chapter 8 I will explore this possibility with the Mandarin items rang and jiao also used for passive and causatives. One of the main aims of this paper was to explain the reason why the agent could not be omitted in a Cantonese passive. The proposal advanced so far make this a reflex of the syntactic structure. That is, if passive in Cantonese is syntactic, and if only transitive verbs can participate in BEI-constructions, then the v-VP configuration that is available would always guarantee that an external argument could be generated. BEI is required to case mark this external argument. Another alternative way to force the appearance of the external argument might be to use the manner component argument as suggested by Hale & Keyser (1993). Chu (1973) and Xiaobo (1991) have argued that a post-verbal element is always required in order to license a passive structure. The post-verbal element can be either an aspect marker or a modifier phrase, or both. For example, (69a) below is ungrammatical, but the same sentence with a verbal modifier is fine, (69b). C (69) a)* ngoh BEI keuih gik b) ngoh BEI keuih gik dou sei weyzi lp BEI 3p anger lp BEI 3p anger until die in-state-of T was angered by him ' T was so angered by him till I felt like dying.' We have already seen that many of the verbs used as examples in this paper are followed by a post-verbal element. Mandarin Cantonese pian- le (8) saksei-jo (11) cheat-FFN kill die P F V s h a s i - l e (10) da sheung (16), (17), (19) kill die-FFV hit injure da shang (18) touh-jo (22) hit injure steal-FFV chi- le (9a), (43 a) hoy dou porchan (49) 93 eat-FFY ruin until bankrupt chuban-le (9d) publish-FFV tek laahn kick damage (36) sik sey eat all (56) A l l these modifiers indicate the result brought on by some action from an agent: sak sei 'to kill something/someone until dead'; da sheung 'to hit something until damaged'; sik sei 'to eat something till all finished'. Even the perfective aspect can be viewed in this manner. Some action was undertaken by an agent to arrive at such a state. We can argue, following Hale and Keyser, that these post-verbal elements are externally oriented manner modifiers identified with the manner component of the verb. Therefore, the manner component relates to an external argument. The appearance of the external argument is forced for proper licensing of the manner component of the verb. Further discussion is needed to develop this possibility. Since this paper started with a discussion of passive in Mandarin, it is only appropriate that it end with one as well. To summarize, I have tried to accomplish the following in regards to the BEI-construction in Cantonese: i) The prediction that the external theta role will always be present in Cantonese \"passives\" because the Cantonese passive is a syntactic operation but suppression of the external theta role happens at the lexical level. ii) The prediction that only transitive verbs can participate in BEI-constructions and the constant presence of the v-VP configuration which licenses the external theta role. iii) The BEI-construction yields not only Passive interpretations but Causative interpretations as well, depending on the type of syntactic structure generated. Chapter 8: A Final Comparison with Mandarin 94 iv) BEI is a category-neutral element devoid of semantic content and underspecified for the Lexical/Functional distinction, so is freely selected for Case reasons in the appropriate structure. v) The Passive or Causative interpretations in BEI-constructions are generated by the respective null Functional (passive) or Lexical (causative) heads. (Alternatively, it is BEI = L-head or BEI = F-head, depending on the structure it participates it.) The next section is devoted to the question of whether these findings can be carried over into an analysis for the Mandarin passive. Because the goal of this paper was not to study the BEI-construction in Mandarin I cannot guarantee an absolute answer until further research is done. However, on a preliminary examination the analysis presented is definitely plausible. The most problematic difference to be overcome would be the obligatory presence of an overt Agent after BEI in Cantonese whereas Mandarin appears to have no such constraint. However, an observation made by E . W . L i (1994) may shed light into this issue. She suggests that although the agent may not always be overtly expressed in Mandarin, there is evidence to suggest that it can be present covertly. The following example #57 is taken from page 45 of her thesis: Fangzi bei p ro i fen gen tazijij le house BEI allot to himself PRT *'The house was allotted to himself by pro.' \"In the above case, NP taziji 'himself is an anaphor, which requires a binder in an A-position according to Binding Theory. This requirement is not satisfied by any lexical category here. The wellformedness of such a sentence in Chinese entails that there is an empty category present to act as a binder. As no NP or wh-trace is involved and PRO has to be ungoverned, this empty category is a pro. However, the above does not hold for English\" (E.W.Li 1994, p.45) Although this is only one example, it opens up the possibility that the agent is always obligatory in Mandarin as well, whether overtly or covertly. We could also apply her suggestion of a covert agent to cases involving the different toned BEI (beih) in Cantonese as 95 well. That is, the appearance of a non-overt agent pro may be licensed by this special form of BEI because it carries a different tone as a phonological signal; pro requires B E I H to license it. Rizzi (1986) imposes two conditions on pro: 1) it must be licensed; 2) it must be identified. Cheng (1989) proposes that \"[Mandarin] pro must be licensed by a distinguished functional head\". She argues that the Aspect phrase serves to license pro in Mandarin, and the discourse context identifies pro. The analysis I have presented in this paper argues that BEI-passive is functional in nature. It is possible that B E I H , is \"the distinguished functional head\" which can license pro as a null Agent argument in these constructions. Following Cheng, the discourse context serves to identify pro. This initiative creates a generalization that Agents are obligatory in all BEI-constructions in both Mandarin and Cantonese. As previously noted by Chappell (1996), Mandarin BEI-constructions are never associated with causative meanings although the items rang 'let' and jiao 'call' can be used in colloquial speech for both the passive voice and the causative. Therefore, these items also seem to behave much in the way that Cantonese BEI does. To show their correlation we can use the same tests as listed in Section 4.1.3 above to see whether rang and jiao are verbal when they are in sentences where it is possible to have a passive meaning. The following sentence is adapted from Chappell's #38 (1996:1038). M (70) a) can it take an aspect marker like most verbs? N O (i) * ta rang I jiao - le wo piping - le yixia 3p R A N G / J I A O - P F V me criticize-PFW once 'He had been criticized by me.' b) even verbs which cannot take an aspect marker (like the copula) can participate in the in A-not-A question form; rang I jiao (passive) cannot (ii) * ta rang bu rang I jiao bu jiao wo piping - le yixia? 3p R A N G not R A N G / JIAO not JIAO me criticize-PFY once 'He was-not-was criticized by me?' 96 c) it cannot stand alone in a predicate as a simple verb (iii)* ta rang I jaio wo 3p R A N G / JIAO lp * 'He was by me.' d) it cannot be a simple answer to a question (iv) A : xi-bu-xi ni rang I jiao ta piping - le yixia a? B : rang] (jiao]) is-not-is 2p R A N G / JIAO 3p criticize-?FV once Q - P R T ' Was it that you were criticized by him?' ' Was!' e) it does not introduce any argument; unless it is possible to have avalent verbs ->in the above sentence the only two arguments are ta 'he' and wo 'me', both introduced by the verb piping 'to criticize', and not by either rang or jiao As the above tests show, these two items do not behave verbally in this type of environment. However, they will successfully pass the same tests when the sentences are associated with a causative interpretation. M (71) a) take an aspect marker (i) ta rang I jiao - le wo piping yixia\" 3p R A N G / J I A O - P F V me criticize once 'He had me criticize (someone).' b) participate in a A-not-A question form (ii)ta rang bu rang I jiao bu jiao wo piping yixia? 3p R A N G not R A N G / JIAO not JIAO me criticize once ' D i d he have me criticize (someone)?' c) to stand alone in a predicate (iii) wo rang ta wo jiao ta lp let 3p lp call 3p T let him.' T call him.' d) answer to a simple question A : rang I jiao ta laih ma? B : rang] jiao] let I call 3p come Q - P R T let] call] 'Are we letting /calling him (to) come? ' O f course!' The Mandarin consultants indicate that this example where there is an Aspect marker attached to either rang or jiao is rather awkward. However, the utterance is acceptable and not entirely ungrammatical. This point is very interesting in light of the fact that all the other tests and examples are perfectly fine. I leave this topic open for further research. 97 e) introduces the Causer theta role (iv) Paul rang John dashang - le Mary. Paul had John hit-till-hurt-PYW Mary 'Paul had John hit Mary till she was hurt.' A point of note is that in (7 Id) rang and jiao mean something other than \"have\". They mean 'let' and 'call' respectively. This behaviour shows that these two items may be independent lexical verbs as well, having a different semantic matrix. Otherwise, for the cases we are considering here, we could apply the analysis suggested for Cantonese to rang 'let' and jiao 'call' as well. Let's suppose that they are also unspecified for the Lexical/Functional distinction and also lack semantic content. In passive structures they acquire a meaning similar to the get-passives in English; in causatives they acquire the meaning of \"have\", in the sense of \"to cause\". A Lexical projection licenses a causative construction; a Functional projection licenses a passive construction; these items acquire their respective interpretations in virtue of the structure they reside in. Figure 16: 0-rang /jiao Passive Figure 17: 0-rang/jiao Causative IP IP V Object 98 The use of the null Functional or Lexical head to generate either a passive or a causative construction in both Mandarin and Cantonese provides for a consistent way in which these constructions are licensed and identified. We dispense with needing to stipulate that BEI or rang and jiao themselves each carry this dual property. However, as discussed earlier with BEI, it is possible to specify that the item itself is in a neutral category and inherently is able operate both as a functional element and a lexical element. Turning back to Mandarin BEI, I propose that since it can only have a passive interpretation it can participate only in the Functional projection. That is, it is already pre-specified as a functional element and thus can only occur in passive structures. It also Case marks the X P following it, whether it is an overt N P or a pro. Figure 18: Mandarin BEI From the above pattern, we could postulate a generalization that the Passive in Chinese is generated by a Functional projection, while the Causative is generated by a Lexical projection. The appropriate elements required to fill the projections are selected according to IP spec Infl \\ , FPpassive 99 their feature specifications and feature compatibility. This paper makes the prediction that only BEI, bei, beih, rang and jiao will be selected for these constructions because they have certain inherent properties that satisfy the selection requirement for a particular structure. The present discussion would benefit from further research into the use of functional and lexical distinctions in other categories in the language. 100 Bibliography Abney, S. (1987) The English Noun Phrase in its Sentential Aspects. Doctoral dissertation, MIT. Aoun, Joseph. (1985) A Grammar of Anaphora. M I T Press. Cambridge. M A . Baker, M . , K.Johnson & I.Roberts. (1989) \"Passive Arguments Raised\", in Linguistic Inquiry, Vol.20: pp.219-252. Bresnan, J. (1972) Theory of Complementation in English Syntax. Doctoral dissertation. MIT Chappell, Hilary. (1986) \"Formal and Colloquial Adversity passives in Standard Chinese.\" In Linguistics 24 :pp. 1025-1052. Cheng, Lisa L.S. (1988) \"Raising Structures and Passive in Mandarin Chinese\", M I T Ms . Chomsky, Noam (1981). Lectures on Government and Binding, Dordrecht, Foris Publications. Chomsky, Noam. (1982) Some Concepts and Consequences of the Theory of Government and Binding. M I T Press, Cambridge, M A . Chomsky, Noam (1986). Barriers. M I T Press, Cambridge, M A . pp. 1-27. Chomsky, Noam (1995). The Minimalist Program. M I T Press. Cambridge. M A . pp.219-394. Chu, Chaucey (1973). \"The passive construction: Chinese and English\" in Journal of Chinese Linguistics, Vol. 3.1: pp.437-470. Cowen, J.R. & Reed, Bradley. (1988) \"Applying an Experimental Approach to the Investigation of Syntactic Change,\" Gurt '88 Synchronic and Diachronic Approaches to Linguistic Variation and Change. Georgetown University Press, U S A . pp.261-267. Dechaine, Rose-Marie (1993). Predicates Across Categories. University of Massachusetts, Boston, pp.25-75. Emonds, Joseph (1985). A Unified Theory of Syntactic Categories, Foris Publications-Dordrecht, Netherlands, pp.67-109 & pp.247-272. Grimshaw, Jane (1990). Argument Structure, M I T Press, Cambridge, M A . Haegeman, Liliane (1993) Introduction to Government and Binding Theory, Blackwell Publishers, Cambridge, M A . 101 Hale, K . & J. Keyser (1993) \"Argument Strucuture and the Lexical Expression of Syntactic Relations.\" in The View from Building 20, M I T Press, Cambridge, M A Huang, James C T . (1982) Logical Relations in Chinese and the Theory of Grammar. Doctoral dissertation, MIT. Huang, J .C.T. and C . - C . Jane Tang. (1991) \"The local nature of the long-distance reflexive in Chinese\" in Long Distance Anaphora. J. Koster and E. Reuland (eds.) Cambridge University Press, Cambridge, M A . pp.263-282. Jaeggli, Osvaldo A . (1986) \"Passive\" in Linguistic Inquiry, Vol. 17:4, MIT. pp. 587-622. Kayne, Richard. (1994) The Antisymmetry of Syntax, M I T Press, Cambridge, M A . Keenan, Edward L . (1985) \"Passive in the World's Languages\" in Language Typology and Syntactic Description, Vol.1. Timothy Shopen (ed.) pp.243-281. Kratzer, Angelika (1994) \" O n External Arguments\" in UMOP 17: Functional Projections, University of Massachusetts, Boston, pp. 103-130. LaPolla, Randy J. (1988) \"Topicalization and the Question of Lexical Passives in Chinese\" Proceedings of the Third Ohio State Univerisity Conference on Chinese Linguistics (13-14 May 1988), pp. 170-188. Larson, Richard (1988) \" O n the Double-Object Construction\" in Linguistic Inquiry Vol.19. M I T Press, pp. 335-391 L i , Erica Wen. (1994) Passives in Mandarin Chinese, M A Thesis, U B C . L i . Y . H . A . (1990) Order and Constituency in Mandarin Chinese, Kluwer Academic Press, Dordrecht. Chapter 6, pp.153-181. Matthews, Stephen & Virginia Yip. (1994) Cantonese: A Comprehensive Grammar. Routledge, New York. Peyraube, Alain. (1989) \"History of the Passive Constructions in Chinese until the 10th Century\" Journal of Chinese Linguistics 17:2, pp.335-371. Po-Ching, Y . & D . Rimmington (1997) Chinese: A n Essential Grammar, Routledge, London, G B . Speas, Margaret, (1994) \" N u l l Arguments in a theory of Economy Projections\" in UMOP 17: Functional Projections, University of Massachusetts, pp. 179-208. 102 Spencer, Andrew (1991) Morphological Theory. Blackwell Publishers, Oxford, G B . Teng, Jen. (1993) \"A'-binding and Bei-construction in Mandarin Chinese\" in Papers from the Fifth Student Conference in Linguistics, University of Washington, pp.239-254. Tomioka, Satoshi (1994) \" O n the Licensing of Lexical Projections\" in in UMOP 17: Functional Projections, University of Massachusetts, pp.209-226. Xiaobo, Ren. (1991) \"The Post-Verbal constituent in Chinese Passive Forms\" in Journal of East Asian Linguistics 19:2, pp.221-241. Washio, Ryuichi. (1993) \"When Causatives Mean Passive: A Cross-Linguistic Perspective\" Journal of East Asian Linguistics 2, pp.45-90. Williams, E . (1989) \"The anaphoric nature of 0-roles.\" in Linguistic Inquiry, Vol.20. pp.425-456. 103 A P P E N D I X 1: Exceptional \"no-movement\" Cases There are exceptional circumstances where the following is considered acceptable: I) bei a Paul tek laahn-jo ga che BEI C L Paul kick damage-VFW C L car 'The car was damaged by Paul's kicking of it.' One such scenario where this would be possible would be something like the following: suppose one was telling a story about Paul trying to damage someone's car. The conclusion of the story is, \"In the end, Paul still managed to damage the car.\" In this case we would be able to use (I) if the modifiers jungyue dou 'finally still' were inserted before the phrase: C (la) jungyue dou bei a Paul tek laahn-jo ga che finally still BEI C L Paul kick damage-PFY C L car 'In the end, the car was still damaged by Paul.' In these situations, it is possible to omit dou 'still', but omission of jungyue 'finally' would yield an ungrammatical sentence. C (Ic) jungyue bei a Paul tek laahn-jo ga che finally BEI C L Paul kick damage-PFV C L car 'In the end, the car was still damaged by Paul.' (Id)* bei a Paul tek laahn-jo ga che BEI C L Paul kick damage-PFY C L car 'The car was still damaged by Paul.' Notice that because the object remains in its base position this type of example poses a problem for my passive analysis if I were to claim that something from the object position must move out (e.g. the verb cannot assign A C C for instance). However, if these are cases of the permissive causative-type then no problem arises (see section 4.2). Because judgment from speakers conflict regarding the interpretation of these constructions, I have not formally included them as part of the analysis till further research and data can provide more useful comment. The consultants are not sure whether these sentences are definitely passive or 104 definitely causative. However, they commented that these types of sentences have more of a \"causative-feel\" to them. Other examples of phrases, other than subject NPs, which can be inserted before the BEI-phrase are given below. C II) a) kumyuk bei ginchaktsouk - jo go chak yesterday BEI police catch P F V C L thief b) go chak kumyuk bei ginchak tsouk-jo <- preferred C L thief yesterday BEI police catch P F V 'The thief was caught yesterday by the police.' III) jungyue bei ginchak tsouk - jo go chak finally BEI police catch P F V C L thief Tn the end, the thief was caught by the police.' IV) ?? a) heiy guy-sheun bei ginchak tsouk-jo go chak on street up BEI police catch P F V C L thief b) go chak heiy guy -sheurj bei ginchak tsouk - jo <- preferred C L thief on street up BEI police catch P F V 'The thief was caught by the police in the streets.' V ) * youn tseun bei chak dasheurj - jo ginchak use gun BEI thief hit-hurt P F V police 'The police was hurt by the thief using a gun.' VI) * yunwei keuih maahn bei ginchak tsouk - jo go chak because he slow BEI police catch P F V C L thief 'Because he was slow, the thief was caught by the police.' Notice that (II) and (III) are grammatical (although (lib) is preferred over (Ila)) but the rest are not. The marginal case is (IV) where I obtained mixed results from my consultants. They informed me that the phrase is very awkward and that no one would say anything resembling (IVa). (IVb) is definitely the preferred expression. However, they also agreed that the sentence is not as bad as (V) or (VI). Therefore, I have place double question-marks before (IVa), but I will still consider it as being ungrammatical. One could argue that (IV), (V) and (VI) are ungrammatical because the adverbial has been fronted. The examples below show that this objection is not correct: 105 C (VII) a John sak-jo a Mary. C L John kill - P F V C L Mary 'John killed Mary.' (VIII) kumyak a John sak - jo a Mary yesterday C L John kill - P F V C L Mary ' John killed Mary yesterday.' (FX) jungyue a John sak - jo a Mary finally C L John kill - P F V C L Mary Tn the end, John killed Mary.' (X) heiy guy-sheun a John sak- jo a Mary on street- up C L John kill - P F V C L Mary 'John killed Mary in the streets.' (XI) your) tseun a John sak- jo a Mary use gun C L Jo/?« &z'/7 - P F V C L Mary 'John killed Mary by using a gun.' (XII) yunwei keuih hun-sum a John sak- jo a Mary because 3p cruel-hearted C L JO/JA? fo7/ - P F V C L Marry 'John killed Mary because he was cruel-hearted.' As the examples above show, all the adverbs can be fronted before a regular active sentence. Although it has been pointed out that the above expressions are not the preferred utterances of native speakers, they are nonetheless acceptable and not as awkward as (IV), (V) or (VI) above. What, then could be causing the effects in grammaticality in the BEI-constructions? Before I attempt an answer to this question, consider the following examples. These sentences can only be used in a restricted context where both speaker and hearer are aware of the proceedings of the discourse and the details of the discussion. These sentences cannot be utter out of the blue and carry an existential meaning, similar to English, \"There was a. . . \" or \"It so happened that...\" type sentences. C (XIII) kumyak bei a Paulj touh-jo keuih*j/j ga che Ambiguous yesterday BEI C L Paul steal-¥FN his C L car Interpretation 'Yesterday, it happened that Paulj to stole his*j/j car.' 106 (XIV) kumyak bei a Paulj touh-jo tsgeij ga che Clear Causative yesterday BEI C L Paul steal-PYV his C L car Interpretation 'Yesterday, it happened that Paulj stole hisj own car. (XV) jungyue bei a Paulj touh-jo keuih*j/j ga che Ambiguous finally BEI C L Paul steal-PFY his C L car Interpretation 'In the end, it happened that Paulj stole his*j/j car.' (XVI) jungyue bei a Paulj touh - fan tsgeij/*j ga che Clear Causative finally BEI C L Paul steal-back his C L car Interpretation 'In the end, it happened that Paulj stole his*j/j own car back.' The above examples may help to clarify whether these types of expressions where no apparent movement has taken place are really passive or causative in nature. Recall from earlier discussion in Section 3.3, it was noted that Indirect passives with an overt possessor reduplicated before the unraised portion of the direct object were considered awkward by native speakers. However, such utterances were possible as long as the overt possessor is co-referential with the raised possessor N P . Although all the sentences presented in (XIII) to (XVI) above have overt possessors appearing before the direct object they are not awkward sounding expressions. They do not seem to display the kind of awkwardness associated with (25). If this complication is the result of the reduplication of a overt possessor clitic in the place of the trace, then the effect does not show up in the above sentences. This definitely confirms that the examples above are not like the Indirect passive examples presented earlier. I present two possible explanations: (i) these sentences are passives which do not involve object raising (again this would be problematic in my analysis where movement is driven by the need to get Case); (ii) these sentences are only ever causative. Sentences (XIV) and (XVI) contain a bare reflexive which must be bound in its governing domain. A s the examples show Paul is the only available binder. Since this is so, the sentence only has a causative interpretation. In (XVI) I have substituted a different aspect 107 on the verb for semantic reasons. A s only a causative interpretation is possible the sentence does not seem to make much sense if it meant something like \"Finally Paul accomplished the task of stealing his own car\" because the car was Paul's in the first place. For a passive interpretation to be available, the object must share a possessor-relation with an N P in the matrix [spec, IP] position. However, in these examples, there is no visible N P in such a position. One could conclude that this is because nothing has raised out of the object position so there is nothing else available to bind the reflexive. However, even for a causative interpretation, one is led to ask, where is the third theta role assigned to? Paul is assigned the Causee role; ga che 'car' is assigned the Direct Object role; where is Causer being assigned? I propose that this role is assigned to an empty pronominal, that is,pro, sitting in [spec, IP]. Therefore, (XIV) and (XVI) have the following structures: (XlVa) kumyak proj bei a Paul; touh-jo tsgeij/*j ga che Clear Causative yesterday pro BEI C L Paul steal-FFV his C L car s Interpretation 'Yesterday, it happened that Pauls stole hiss own car. (XVIa) jungyue proj bei a Paulj touh - fan tsgeij/*j ga che Clear Causative finally pro BEI C L Paul steal-back his C L car Interpretation T n the end, it happened that Paulj stole his*j/j own car back.' The pro in the above examples cannot bind the reflexive for the same reason that the overt N P in (32b) could not bind the reflexive: it is not in the governing domain of the anaphor. We can apply the ^ro-analysis to (XIII) and (XV) , deriving: C (XHIa) kumyak proj bei a Paulj touh - jo keuih* j/j ga che yesterday pro BEI C L Paul steal-FFV his C L car 'Yesterday, it happened that Paulj managed to steal his*j/j car.' (XVa). jungyue proj bei a Paulj touh - jo keuih*j/j ga che finally pro BEI C L Paul steal-PFY his C L car T n the end, Paul; managed to steal his*j/j car.' 108 In (XHIa) and (XVa) the pro can bind the pronoun in the lower clause, but Paul cannot because it is in the minimal domain of the pronoun (Principle B violation). It must be pointed out that these sentences can only be used if the context is known to both speaker and hearer. For example, (XHIa) and (XVa) can be used in a situation where the discussion was about Paul and his car, and the speaker was pointing out that due to some course of events, it came about that Paul had an opportunity to steal back his own car. That is, the situation must be such that pro is able to get a discourse antecedent. This leads to the question of why can't we posit that there is an invisible pro possessor which is raised in these structures? This would allow us to maintain the hypothesis that there is always movement from the object position. A very obvious answer is that such a structure is never verified by the interpretations. Take for example (XlVa): (XlVa) k u m y a k /7ro j bei a Paulj touh-jo tsgei j ga che yesterday pro BEI C L Paul steal-PFV his C L car 'Yesterday, it happened that Paul; stole hisj own car. It was impossible for my speakers to get the reading where the reflexive is bound by a non-overt possessor. This sentence can only ever be causative. I propose, in fact, that all sentences of this type are causative. None of these are passive expressions where the object has not moved, because there would be no way for the Object N P to get Case. The uncertainty in interpretation for (XIII) and (XV) result from the fact that the sentences contain BEI. Speakers naturally connect the appearance of BEI with the generation of a passive sentence, unless it is clear that such an expression is a causative (i.e three distinct arguments are involved). M y hypothesis is reinforced by examples where there is no question of possessor-raising involved. 109 (XVII) kumyakpro bei a Paul d a - s e i - j o a John yesterday pro BEI C L Paul beat-kill-VYV C L John 'Yesterday, circumstances brought it about that Paul killed John.\" (XVIII) jungyue pro bei a Paul d a - s e i - j o a John finally pro BEI C L Paul beat-kill-VVN C L John 'Circumstances brought it about that Paul finally killed John.\" The N P John is not an inanimate object that can be owned so the complication with possessor-raising is removed. Speakers corroborated that that (XVII) and (XVIII) both have a permissive-causative interpretation. A set of circumstances allowed Paul to carry out the grisly action of killing John. These findings help explain the grammaticality differences observed in (II) to (VI) above. I argue that verbal modifiers like jungyue 'finally' and time adverbials like kumyak 'yesterday' occupy an A'-position, perhaps [spec, CP], when fronted. The other adverbials of instrument, location and reason, on the other hand, must occupy an A-position, when fronted 3 4. However, the only available A-position is [spec, IP] but it must be reserved for the pro if these expressions are all causative. Therefore, if these adverbials are fronted to such a position then there would be no place for pro and the sentence becomes ungrammatical because the third theta role cannot be assigned. M y argument so far, is only tentative. I have been unsuccessful in finding tests which would show whether these adverbials truly occupy A-positions. More studies into this areas is a necessity in order to arrive at a definitive answer. Much thanks to Michael Rochemont for this suggestion. 110 "@en ; edm:hasType "Thesis/Dissertation"@en ; vivo:dateIssued "1997-11"@en ; edm:isShownAt "10.14288/1.0088269"@en ; dcterms:language "eng"@en ; ns0:degreeDiscipline "Linguistics"@en ; edm:provider "Vancouver : University of British Columbia Library"@en ; dcterms:publisher "University of British Columbia"@en ; dcterms:rights "For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use."@en ; ns0:scholarLevel "Graduate"@en ; dcterms:title "Bei and the passive in Cantonese"@en ; dcterms:type "Text"@en ; ns0:identifierURI "http://hdl.handle.net/2429/6542"@en .