DIMENSIONS OF REFLEXIVITY by Jeffrey L. Lidz A ... - CiteSeerX

4 downloads 0 Views 243KB Size Report
questions to ask my Kannada informants when I still had no idea what a verbal ...... exhibited by 'bank' in English could also be true across language families, but ...... Descriptively speaking, languages vary with respect to the extent that they.
DIMENSIONS OF REFLEXIVITY

by Jeffrey L. Lidz

A dissertation submitted to the Faculty of the University of Delaware in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Linguistics

Summer 1996

Copyright 1996 Jeffrey L. Lidz All Rights Reserved

DIMENSIONS OF REFLEXIVITY

by Jeffrey L. Lidz

Approved:

__________________________________________________________ William Frawley, Ph.D. Chair of the Department of Linguistics

Approved:

__________________________________________________________ John C. Cavanaugh, Ph.D. Interim Associate Provost for Graduate Studies

I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy. Signed:

__________________________________________________________ Peter Cole, Ph.D. Professor in charge of dissertation I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy.

Signed:

__________________________________________________________ Luigi Burzio, Ph.D. Member of dissertation committee I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy.

Signed:

__________________________________________________________ Robert Frank, Ph.D. Member of dissertation committee I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy.

Signed:

__________________________________________________________ William Frawley, Ph.D. Member of dissertation committee I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy.

Signed:

__________________________________________________________ Gabriella Hermon, Ph.D. Member of dissertation committee I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy.

Signed:

__________________________________________________________ William Idsardi, Ph.D. Member of dissertation committee

Even if the person is, for special speculative purposes, momentarily concentrating on the Problem of the Self, he has failed and knows that he has failed to catch more than the flying coat-tails of that which he was pursuing. Gilbert Ryle - "The Concept of Mind"

v

ACKNOWLEDGMENTS There came a point in writing this thesis that the longer I worked on it, the shorter it got. I thought it best to stop before the whole thing disappeared. My committee played an important and valuable role in both the development and subsequent shrinking of the dissertation and I take this opportunity to thank them for all they have done. Everyone on the committee knows an argument when they see one and also when they don't. Peter Cole, my advisor, deserves thanks for pushing me to make every step of every argument as clear as possible. I thank Peter for teaching me how to do syntax. I hope that some of his respect for language has rubbed off on me. Bob Frank also deserves a good deal of credit for teaching me how to do syntax. He can see through to the core of an argument faster than anyone I know, so arguing with him can be the most frustrating experience of your life. But if you convince him, you know the argument works. Gaby Hermon rarely lets an assumption go unquestioned. Discussing this work with her has forced me to clarify exactly what I mean in many places. Luigi Burzio has had an enormous influence on my thinking about reflexivity. His course on soft-constraints in syntax was one of the highlights of my graduate career. Although there is little in this thesis that directly reflects Luigi's ideas, I think the careful reader will be able to trace many of my ideas back to his. Bill Idsardi, although not a syntactician, has been influential in the development of this thesis. Whenever a conceptual question about the organization of the grammar arose, I could count on Bill to help me see the answer. Bill Frawley also played a very important role in the development of the ideas presented here. He helped me figure out what questions to ask my Kannada informants when I still had no idea what a verbal reflexive was. The insights in Chapter 2 could not have arisen without his input.

vi

vii Beyond my committee, several of the other students here played an important role in the development of these ideas. An afternoon with Tom Purnell, Abdellatif Alghadi and Tonia Bleam brought the analysis in Chapter 4 to light. It took two phonologists to help me see the truth. My classmates, Steve Hoskins, Thomas Klein, Li Hao and Christina Tortora deserve thanks for making my time here a challenging and invigorating experience. Other students who should be thanked for being around include Dave Schneider, Sue Scott, Nancy Goss, Eynat Gutman, Chengchi Wang, Norhaida Aman, Ken Hyde and Bert Xue. I should also thank my teachers at Northwestern, Beth Levin and Brad Pritchett, for encouraging me to pursue lingusitics. Beth has mainainted an open ear throughout my graduate career and has offered many helpful suggestions along the way. My Kannada informants Ananda Murthy, Narahari Pujar and Raama Brasad were extremely generous in offering their time and knowledge to me over the past several years. I also thank Chandra Shekar for help with some of the more complicated judgments. I have discussed this work with several people outside of Delaware over the past couple of years. I thank them all for their interest. These include (but may not be limited to): Alice Davison, Bob Friedin, Jim Gair, Susan Herring, Ray Jackendoff, Barbara Lust, Eric Pederson, John Paolillo, Len Talmy, Angela Terrell, Juan Uriagereka, Caroline Wiltshire, and K.S. Yadurajan. My family deserves special credit for putting up with my obsession for these last years. They have been a constant source of encouragement and support. My

viii father may now know more about linguistics and cognitive science than anyone without a Ph.D. Discussions with Christina Tortora ranging from circumfixes to armageddon made graduate school what it was. Her comments on an earlier version of this thesis have undoubtedly improved this version. She is a comrade and an inspiration. I could not have made it through without her. Finally, Tonia Bleam has given more than anyone to this dissertation. Her tremendous influence on this work is insignificant, however, when compared with her influence on my life. I thank her for what we have. This work was supported in part by a University of Delaware Graduate Fellowship and in part by Grant #BNS-9121167 from the National Science Foundation.

TABLE OF CONTENTS ACKNOWLEDGMENTS............................................................................................. vi TABLE OF CONTENTS .............................................................................................. ix ABSTRACT .................................................................................................................. xi 1. Introduction ............................................................................................................. 13 2. Understanding Verbal Reflexives............................................................................ 17 2.1 Introduction ...................................................................................................... 17 2.2 Semantic Properties of Verbal Reflexives........................................................ 24 2.3 The Mismatch Hypothesis................................................................................ 34 2.3.1 Multi-dimensional argument-structure..................................................... 34 2.3.2 The verbal reflexive is a mismatch morpheme......................................... 38 2.3.2.1 The argument-structure of external causation ...................................... 39 2.3.2.2 The argument-structure of reflexivity .................................................. 43 2.4 On Argument Absorption ................................................................................. 51 2.4.1 Motivations for the Argument Absorption Hypothesis ............................ 52 2.4.2 Against the Argument Absorption Hypothesis......................................... 57 2.5 Alternative Analyses ........................................................................................ 60 2.5.1 The Surface Intransitive Hypothesis ........................................................ 61 2.5.2 The Reflexive Predicate Hypothesis ........................................................ 68 2.5.2.1 Decausatives and the RPH.................................................................... 72 2.5.2.2 Possession and the RPH ....................................................................... 76 2.5.2.2.1 LF Noun Incorporation................................................................. 77 2.5.2.2.2 Possessor Ascension..................................................................... 81 2.5.3 Conclusion................................................................................................ 85 2.6 Consequences of the Mismatch Hypothesis ..................................................... 86 3. Cross-linguistic Variation of Reflexive Morphemes............................................... 91 3.1 Introduction ...................................................................................................... 91 3.2 Connecting Antipassives and Reflexives ......................................................... 91 3.2.1 The semantics of antipassives .................................................................. 95 3.2.2 Boundedness in argument structure.......................................................... 97 3.2.3 The argument structure of antipassives .................................................... 99 3.3 Connecting Middles and Reflexives............................................................... 104 3.4 Connecting Impersonals and Reflexives ........................................................ 107 3.5 Conclusions .................................................................................................... 109 4. Antilocal Reflexives: An Epiphenomenon............................................................ 111 4.1 Introduction .................................................................................................... 111 4.2 Locally bound antilocal reflexives ................................................................. 115

ix

x 4.3 Hestvik (1990): An analysis based on antilocality ........................................ 119 4.4 Anticoargumenthood instead of antilocality .................................................. 121 4.4.1 Semantic reflexivity and the coargument restriction.............................. 125 4.4.1.1 Kannada.............................................................................................. 125 4.4.1.2 Generalizing to Dutch and Norwegian ............................................... 128 4.4.1.2.1 Zich as a verbal reflexive............................................................ 129 4.4.1.2.2 Seg as a verbal reflexive ............................................................. 135 4.4.1.2.3 Deriving the coargument restriction in Dutch and Norwegian .. 137 4.4.1.2.4 The syntactic representation of zich/seg..................................... 145 4.4.1.3 Malayalam .......................................................................................... 150 4.5 Near-reflexives ............................................................................................... 152 4.6 Conclusions and Consequences...................................................................... 159 References .................................................................................................................. 170

ABSTRACT The study of reflexivization has maintained a central position in generative grammar since its inception. Little attention has been paid, however, to the languages that mark reflexivity directly on the verb and not on one of its arguments. This thesis explores such languages, revealing that their analysis is independent of the theory of syntactic binding. The independence of such lexical reflexivity from the expression of coreference relations among NPs has important consequences for binding theory, however. In particular, the analysis of verbal reflexives leads to a significant new understanding of antilocal anaphors, i.e., anaphors which resist binding by a local antecedent. Chapter 2 argues that verbal reflexives are the morphosyntactic instantiation of a mismatch between tiers in argument-structure. As such, these morphemes do not act as argument absorbers. They also do not have a unitary semantic or syntactic representation. This analysis is based on the cross-linguistically stable uses of verbal reflexives in inchoative and certain possessive constructions. Chapter 3 extends the analysis of Chapter 2 to cross-linguistically variable properties of verbal reflexives. Verbal reflexives are often used in antipassive, middle and impersonal constructions. I show how such uses are consistent with the analysis of Chapter 2 and not with any of the other potential analyses of verbal reflexives. Chapter 4 turns to an outstanding problem for the binding theory. Antilocal anaphors have never been satisfactorily understood. I show that the behavior of antilocal anaphors is not due to parameterization of the binding theory or

xi

xii to the featural content of the anaphors but rather to an interaction between the semantic content of the anaphors and Condition R, a principle of UG which forces semantic reflexivity to be expressed in the argument-structure. The binding theory is constant for all languages. What varies across languages is the semantic content of the anaphors. Only those anaphors which require complete identity with their antecedents are antilocal, their antilocality arising from Condition R. The other anaphors, which I call "near-reflexives" are not antilocal because they do not give rise to semantic reflexivity and thus do not interact with Condition R.

13

Chapter 1 INTRODUCTION A tremendous body of research in generative grammar has grown up around the questions associated with reflexivity and coreference. These questions are among the most interesting in language because they reside squarely at the interfaces between components. While the interpretation of anaphora is clearly an issue for semantics, syntactic conditions also play a role in determining the domain and potential antecedents for anaphoric forms. In addition to these considerations, recent work by Reinhart and Reuland (1993) has claimed that reflexivity has more to do with predicates than with the resolution of coreference. This brings information, possibly lexical, about the interpretation of verbs and their arguments into the mix. This thesis is an extensive examination of the behavior of reflexive elements, and can thus be taken as a case study investigating the organization of the language faculty. Since reflexivity and coreference are relevant to so many domains of grammar, tracking the phenomena associated with reflexivity has the potential to yield important insights about the representations within these domains as well as the interactions found among them. Human languages make use of two mechanisms for the expression of reflexivity. Some languages use a verbal affix while others use a special anaphoric pronoun. These mechanisms are not mutually exclusive, however. Many languages have both. The research in this thesis is driven by the question of how many systems are needed to account for this variability. Is the system governing the distribution of verbal reflexives completely independent of the system governing the distribution of nominal reflexives? If not, to what extent do these systems overlap or interact? I propose that the distribution of verbal reflexives is determined primarily by argument-structure

14

representations while the distribution of nominal reflexives is determined primarily by syntactic representations. We do find, however, certain cases of overlap. In these cases, we find that the existence of two systems is precisely what is required to achieve a level of explanatory adequacy. Chapter 2 begins with the observation that there is a systematic crosslinguistic ambiguity in verbal reflexives. The form used to express reflexivity is also used with certain inchoative verbs and optionally when the subject and possessor of the object are coreferent. We might expect this ambiguity to be due to the meaning of the morpheme, the syntactic structures that occur in such constructions or to argument structure representations. We find, interestingly, that there is not a unique semantic or syntactic representation associated with these morphemes. There is, however, a unified argument-structure representation which is compatible with a range of semantic and syntactic structures. In this way, verbal reflexives provide compelling evidence for the existence of argument-structure as an independent level of representation. The idea that verbal reflexives arise due to properties of argument-structure is not new. However, previous analyses along these lines have all taken the position that verbal reflexives are argument absorbing operators. This view is empirically inferior to a view in which verbal reflexives are the morphosyntactic reflex of a particular argumentstructure representation. Verbal reflexives occur whenever the most prominent element on the aspectual tier of the argument-structure representation is not linked to something on the thematic tier. These morphemes, as such, do not have semantic content but rather simply indicate that a particular structure (compatible with multiple meanings) has arisen. While it is true that the three uses of verbal reflexives discussed in Chapter 2 are cross-linguistically stable, there are many other uses of these forms which exist in some languages but not others. Chapter 3 examines the use of verbal reflexives in antipassive constructions, middle constructions and impersonal constructions. The theory developed in Chapter 2 can explain why verbal reflexives can be extended to these uses.

15

The antipassive morpheme occurs whenever the least prominent element on the aspectual tier is unlinked to an element on the thematic tier. In the languages that use the same morpheme for antipassive and the constructions discussed in Chapter 2, this morpheme has been generalized to indicate that there is an unlinked element somewhere on the aspectual tier. The argument-structure representations of middle and impersonal constructions also share properties with the representation giving rise to the verbal reflexive. The fact that some languages use the same morpheme for all of them is due to these shared properties. Interestingly, all of the languages which use the verbal reflexive in impersonal constructions also use it in middle constructions. Not all languages which use the verbal reflexive in middles, however, also use it in impersonals. This implicational hierarchy can be explained by properties of the argument-structure representations associated with each construction. The argument-structure for the verbal reflexive is a subpart of the argument-structure for middles. Similarly, the argumentstructure for middles is a subpart of the argument-structure for impersonals. Chapter 4 turns away from argument-structure and verbal reflexives and examines the cross-linguistic variability found among reflexive pronouns. The head movement theory of Cole, Hermon and Sung (1990) and Cole and Sung (1994) in concert with the binding theory (Chomsky 1981; 1986b) can explain the majority of facts associated with nominal reflexives. One problem which has been unresolved within the head-movement theory of reflexives concerns the existence of anaphors which resist binding by a local antecedent. I show that such anaphors are not subject to different binding conditions from other anaphors. Instead, their antilocality is due to an interaction between syntactic, semantic and argument-structure representations. I propose a principle of UG which states that semantic reflexivity must be lexically expressed. The antilocality of certain anaphors is due to an inability to express reflexivity in the argument-structure representation of particular verbs. A consequence of this analysis is that any locally bound anaphor must not express semantic reflexivity. I show that this

16

prediction is borne out. Locally bound anaphors actually express a relation of "nearreflexivity" which holds between two NPs when they are referentially dependent but not necessarily identical in reference. Through the close examination of reflexivity and related phenomena, the following are uncovered: (1) argument-structure representations of a certain sort are required in the theory of grammar, (2) certain morphemes are the morphosyntactic instantiation of particular argument-structure configurations, (3) the interface between argument-structure representations and the syntactic and semantic components is crucial to explaining certain facts about nominal reflexives.

17

CHAPTER 2 Understanding Verbal Reflexives

2.1 Introduction One of the fundamental insights of generative grammar is that the mapping from form to meaning is not direct. The meaning of a phrase is not simply the concatenation of the meanings of its parts. Instead, the meaning of some string of phones (form) is determined by relating that form to various levels of linguistic analysis. This chapter concerns the form-meaning relationship associated with verbal reflexives. These morphemes are significant from this perspective because their analysis requires a level of representation of argument-structure mediating between syntax and lexical-semantics. That is, there is not one syntactic structure that is associated with verbal reflexives. Nor is there one semantic structure responsible for their distribution. The central property of verbal reflexives is that they are often used in constructions that do not indicate semantic reflexivity.1 In many languages, the same morpheme used to indicate that a predicate is reflexive is also used in decausative constructions.2 This 'homophony' is illustrated below in Diyari (Pama-Nyungan),

1

The traditional term 'verbal reflexive' for such morphemes is unfortunate and somewhat misleading. I use the term here only in deference to tradition although the potential for confusion is great.

18 Finnish (Finno-Ugric), Imbabura (Quechua), Kannada (Dravidian), Lithuanian (Baltic), Russian (Slavic), Spanish (Romance), Yakut (Turkic) and Yavapai (Yuman). (1)

Diyari (Austin 1981) a. ngani muduwa-tadi-yi 1SGS scratch-REFL-PRES 'I scratch myself' b.

(2)

Finnish (Sells, et al. 1987; Geniusˇiene 1987) a. jussi puolusta-utu-i john defend-REFL-PST 'John defended himself' b.

(3)

2

ovi ava-utu door-NOM open-REFL 'the door opens'

Imbabura Quechua (Cole 1982) a. ispiju-pi riku-ri-rka-ni mirror-in see-REFL-PST-1 'I saw myself in the mirror' b.

(4)

nawu mana ngandawalka-tadi-na wara-yi 3SGNFS door-ABS close-REFL-PART AUX-PRES 'the door got closed'

pungu-kuna-ka paska-ri-rka door-PL-TOP open-REFL-PST.3 'the doors opened'

Kannada (Amritavalli 1984) a. avan-u tann-annu hoDe-du-koND-a he-NOM self-ACC hit-PP-REFL.PST-3SM 'He hit himself'

Geniusˇiene (1987) investigates reflexive morphology in about 50 languages. In all of the languages in her sample, this generalization holds. Further, in all of the languages in Haspelmath's (1993) survey of causative/inchoative alternations which have productive decausative morphemes, the decausative doubles as a reflexive.

19 b.

(5)

Lithuanian (Geniusˇiene 1987) a. on-a grazˇina-si ann-NOM adorns-REFL 'Ann adorns herself' b.

(6)

dver' zakryla-s' door-NOM closed-REFL 'the door closed'

Spanish (García 1975) a. él se lava he REFL wash.3S.PRES 'He washes himself' b.

(8)

dur-ys at-si-dare door-NOM.PL PERF-REFL-close 'the doors closed'

Russian (Babby 1975) a. on zastrelil-sja he-NOM shot-REFL 'He shot himself' b.

(7)

baagil-u mucˇcˇ-i-koND-itu door-NOM close-PP-REFL.PST-3SN 'the door closed'

la puerta se abrió the door REFL open.3S.PST 'The door opened'

Yakut (Geniusˇiene 1987) a. kini sime-n-er she-ABS dress.up-REFL-3SG.PRES 'She dressed herself up' b.

mas tohu-n-na stick-ABS break-REFL-3SG.PST 'The stick broke'

20 (9)

Yavapai (Kendall 1976) a. hmañ-c kwe-wiv-v-i child-SUBJ thing-clothe-REFL-TNS 'The child dressed himself' b.

?wa:ta?ami-c skwiñ-v-km door-SUBJ lock-REFL-INC 'The door locked' Another case where we find considerable crosslinguistic overlap in

reflexive constructions is in certain cases of possession, typically involving but not limited to inalienable possession. Here, the reflexive morpheme optionally surfaces on the verb when coreference obtains between the subject and the possessor of the object.3 This is illustrated below in Fula (West Atlantic), Kannada, Lithuanian, Spanish, Russian, and Udmurt (Finno-Ugric). (10) Fula (Geniusˇiene 1987) a. O hett-ii fedenndu makko he cut-ACT.PERF finger his 'He cut his finger' b.

O hett-ike fedenndu he cut-REFL.PERF finger 'He cut his finger'

(11) Kannada a. hari-yu tann-a angi-yannu hari-d-a hari-NOM self-GEN shirt-ACC tear-PST-3SM 'Hari tore his shirt' b.

3

hari-yu tann-a angi-yannu hari-du-koND-a hari-NOM self-GEN shirt-ACC tear-PP-REFL.PST-3SM 'Hari tore his shirt'

In Genuisˇiene's sample, the use of reflexive morphology with these cases of possession holds in 47 of the languages, while it is unknown whether it is used in the remaining languages in this construction.

21 (12) Lithuanian (Geniusˇiene 1987) a. vaik-as su-zˇeide koj-a child-NOM PERF-hurt foot-ACC.SG 'The child hurt his foot' b.

vaik-as su-si-zˇeide koj-a child-NOM PERF-REFL-hurt foot-ACC.SG 'The child hurt his foot'

(13) Spanish (García 1975) a. juan rompió el brazo Juan broke the arm 'Juan broke his arm' b.

juan se rompió el brazo Juan REFL broke the arm 'Juan broke his arm'

(14) Russian4 (Geniusˇiene 1987) a. on rasstegnul pal'to he unbuttoned coat 'He unbuttoned his coat' b.

on rasstegnul-sja pal'to he unbuttoned-REFL coat 'He unbuttoned his coat'

(15) Udmurt (Geniusˇiene 1987) a. Anaj ki-ze sut-iz mother hand-ACC.POSS burn-PST.3S 'Mother burnt her hand' b.

Anaj ki-ze sut-isk-iz mother hand-ACC.POSS burn-REFL-PST.3S 'Mother burnt her hand' While there are many more uses of reflexive morphemes in the world's

languages, these two seem to exist in all of the languages with verbal reflexives, while

4

In Russian the object is typically expressed in instrumental case in this type of sentence. See Gerritsen (1990).

22 other uses of these morphemes only exist in some. For example, the reflexive element is used in middle, impersonal, antipassive and other constructions in a subset of the languages with affixal reflexives, but many of the languages with affixal reflexives have other ways of marking these constructions or else simply do not have them. Thus, the ambiguity of these elements between reflexive, decausative and possessive should be treated as the primary fact to be explained. The other uses will be discussed in Chapter 3. One possible approach to the ambiguity of the reflexive morpheme in these languages is to attribute it to homophony. Under this account, the relation between the reflexive element and the decausative element is the same as the relation between the two meanings of 'bank' in English, where we find two meanings associated with one phonological form. This view, however, is obviously flawed. The fact that reflexive elements show the same "homophony" across such diverse languages could not be due to the accidental connection of one phonological shape with several disparate meanings. It is difficult to imagine that the homophony exhibited by 'bank' in English could also be true across language families, but this is exactly what saying that the decausative and reflexive morphemes were homophonous is like. Clearly, we should look for a better explanation. Alternative analyses will generally make use of the idea of constructional homonymity (Chomsky 1957). In order to unify apparently divergent uses of one morpheme we can say that in one domain of grammar, each use is treated as equivalent, while in other domains, each has a different structure. The ambiguity results from a mismatch between the vocabularies of domains. In the case of reflexive morphology, three possibilities present themselves. First, it is possible that this

23 morpheme has one meaning but multiple syntactic realizations. On this account, all of the sentences given above share some element of meaning, but do not share syntactic structure. Second, we might think that there is one syntactic representation that gives rise to this morpheme though no unified semantic representation. Third, the reflexive morpheme could represent some property of the syntax-semantics interface which cannot be able to be unified in either of the other domains. This third approach is the one to take in order to understand reflexive morphology. The reflexive morpheme arises as a consequence of a particular argumentstructure configuration. This configuration may have multiple interpretations in both the semantics and the syntax. More than one meaning can be associated with reflexive morphology, and more than one syntactic structure can be associated with reflexive morphology. Thus, it seems that the only way to unify the various uses of this morpheme is at the interface between these two components. Assuming a level of argument-structure at the interface between lexical-semantics and syntax, we have a constructional homonymity between a representation in argument-structure and the representations of lexical-semantic and syntactic structure. I assume a two-tiered approach to argument-structure in which prominence relations among semantic participants are represented hierarchically (Grimshaw 1990). The thematic tier represents prominence relations among elements receiving a thematic role while the aspectual tier represents prominence relations among participants in an event-structure representation (Pustejovsky 1991). These tiers are projected from lexical-semantic representations and play a role in determining syntactic projection of arguments. Verbal reflexives occur whenever the most prominent element on the aspectual tier is unlinked to an element on the thematic tier.

24 Since this argument-structure configuration is compatible with a range of lexicalsemantic representations as well as a range of syntactic structures, we can conclude (a) that verbal reflexives do not have semantic or syntactic content and (b) that argumentstructure representations are required to provide an interface between lexical semantics and syntactic structure. The chapter is organized as follows. Section 2.2 discusses some semantic properties of verbal reflexives, showing that the verbal reflexive occurs with multiple semantic interpretations. Sections 2.3 and 2.4 cover two argument-structure analyses of these morphemes. The first holds that reflexive morphemes arise as a consequence of a mismatch between tiers in argument-structure. This position contrasts with and is empirically superior to the hypothesis that verbal reflexives are functors which absorb an argument from the argument-structure representation of a verb. Section 2.5 considers two possible syntactic approaches to verbal reflexives and shows that neither can account for the full range of facts surrounding the above examples. Various consequences of the conclusion that the presence of a verbal reflexive indicates a mismatch between the tiers of argument-structure will be discussed in Section 2.6. 2.2 Semantic Properties of Verbal Reflexives Before we can provide a complete analysis of verbal reflexives, we must examine the semantics of the constructions in question in closer detail. Let us begin with the cases in which the reflexive marker is optional. By understanding the semantic distinctions between the sentences with reflexive marking and those without it, we can gain insight into the sentences that require the reflexive marker. Further, an understanding of the semantic factors associated with reflexive morphemes will point the way towards the final analysis of these elements.

25 As noted above, the reflexive morpheme is optionally present when the subject is coreferent with the possessor of the object. The Kannada sentences in (11) illustrating this are repeated here (the Kannada verbal reflexive is koLL- or koND-):5 (11)

a.

raaju tann-a angiy-annu hari-d-a raaju-NOM self-GEN shirt-ACC tear-PST-3SM 'Raaju tore his shirt'

b.

raaju tann-a angiy-annu hari-du-koND-a raaju-NOM self-GEN shirt-ACC tear-PP-REFL-PST-3SM 'Raaju got his shirt torn'

In (11), we find an interesting distinction in meaning. The sentence in (11a) has an agentive interpretation under which Raaju purposely and actively tore his shirt. In (11b), on the other hand, we lose that interpretation. Here we have only the reading in which Raaju's shirt became torn due to an external cause. For example, (11b) would be used when Raaju's shirt got caught on a nail and then tore as a result. Viewing the verbal reflexive as a mark of the lack of agentivity in these cases, however, would be a mistake since the opposite patterning also occurs. In (16), we see a similar pair of sentences in which the verbal reflexive (16b) occurs on the agentive interpretation and the lack of a verbal reflexive (16a) indicates the accidental occurrence of the event:

5

The majority of the examples in this thesis will be from Kannada because it is the language that I have done the most informant work on. (See Sridhar 1990 for a guide to the transcription.) Where possible I will show the cross-linguistic similarities. Because the semantics of the constructions in question is somewhat slippery, I do not know whether the facts uncovered here in Kannada hold in all of the languages in question. However, as will be shown, the Romance reflexive clitics do share semantic properties with the Kannada verbal reflexive. This should be taken as preliminary evidence that the semantics is shared across all languages with these morphemes, though in depth analysis of each language is still required.

26 (16)

a.

hari-yu tann-a kal-annu mur-id-a hari-NOM self-GEN leg-ACC break-PST-3SM 'Hari broke his leg (accidentally)'

b.

hari-yu tann-a kal-annu mur-idu-koND-a hari-NOM self-GEN leg-ACC break-PP-REFL.PST-3SM 'Hari broke his leg (deliberately)'

In (17), both sentences can be interpreted agentively, though there is a difference in interpretation of the predicate itself. (17)

a.

Hari-yu tann-a tale-yannu eTT-id-a hari-NOM self-GEN head-ACC lift-PST-3SM 'Hari lifted his head'

b.

Hari-yu tann-a tale-yannu eTT-i-koND-a hari-NOM self-GEN head-ACC lift-PP-REFL.PST-3SM 'Hari lifted his detached head'

In (17b), we only find the reading in which Hari's head is disembodied. That is, (17a) can be said when Hari wakes up and looks around, but (17b) can only be said if Hari's head is disconnected from his body and he lifts it with his hands. A similar example involves bodily action. The presence of the verbal reflexive can distinguish between indirect and direct bodily action: (18)

a.

hari-yu kannu-gaL-annu tere-d-a hari-NOM eye-PL-ACC open-PST-3SM 'Hari opened his eyes'

b.

hari-yu kannu-gaL-annu tere-du-koND-a hari-NOM eye-PL-ACC open-PP-REFL.PST-3SM 'Hari opened his eyes'

Semantically, the difference in interpretation between (18a) and (18b) lies in the way that the eyes were opened. In (18a), the eyes open via their own internal properties, while in (18b), some additional factor is involved in the opening.

27 These illustrations suggest that the verbal reflexive has less to do with agentivity than with an external role in the event (cf. Levin and Rappaport-Hovav 1995). In (11), the verbal reflexive is associated with an external factor causing the tearing, while in (16b-18b), the event is portrayed from the point of view of the body part, thus backgrounding the cause of the event. In other words, the cause of the event is external to the predication itself. When the cause of the event is independent of the predicate in this sense, the verbal reflexive is required. That is why the verbal reflexive is not possible with the interpretation in (17a): normally, when one lifts one’s head, the head can be viewed as both the lifter and the liftee (i.e., the muscles in the head/neck are what actually lifts it), so there is no explicit external cause. As evidence that the verbal reflexive occurs on the externally caused reading of the above predicates, we find that an instrumental adjunct is possible with the (b) examples and not with the (a) examples in (16-18). In the (a) examples, since the lack of the reflexive morpheme indicates that the event was internally caused, the instrument is not licensed. This fact is illustrated in (19-21) in Kannada, and in (2224) in Spanish. (19)

(20)

a.

*hari-yu tann-a kal-annu kooli-ninda mur-id-a hari-NOM self-GEN leg-ACC stick-INSTR break-PST-3SM 'Hari broke his leg with a stick'

b.

hari-yu tann-a kal-annu kooli-ninda mur-idu-koND-a hari-NOM self-GEN leg-ACC stick-INSTR break-PP-REFL.PST-3SM 'Hari broke his leg with a stick'

a.

*hari-yu tann-a tale-yannu kai-gaL-inda eTT-id-a hari-NOM self-GEN head-ACC hand-PL-INSTR lift-PST-3SM 'Hari lifted his head with his hands'

28

(21)

(22)

(23)

(24)

b.

hari-yu tann-a tale-yannu kai-gaL-inda eTT-i-koND-a hari-NOM self-GEN head-ACC hand-PL-INSTR lift-PP-REFL.PST3SM 'Hari lifted his head with his hands'

a.

*Hari-yu kannu-gaL-annu kai-gaL-inda tere-d-a Hari-NOM eye-PL-ACC hand-PL-INSTR open-PST-3SM 'Hari opened his eyes with his hands'

b.

Hari-yu kannu-gaL-annu kai-gaL-inda tere-du-koND-a Hari-NOM eye-PL-ACC hand-PL-INSTR open-PP-REFL.PST-3SM 'Hari opened his eyes with his hands'

a.

Juan rompió la pierna (*con un bastón) Juan broke the leg with a stick 'Juan broke his leg with a stick'

b.

Juan se rompió la pierna (con un bastón) Juan REFL broke the leg with a stick 'Juan broke his leg with a stick'

a.

Juan levantó la cabeza (*con las manos) Juan lifted the head with the hands 'Juan lifted his head with his hands'

b.

Juan se levantó la cabeza (con las manos) Juan REFL lifted the head with the hands 'Juan lifted his head with his hands'

a.

Maria abrió los ojos (*con las manos) Maria opened the eyes with the hands 'Maria opened her eyes with her hands'

b.

Maria se abrió los ojos (con las manos) Maria REFL opened the eyes with the hands 'Maria opened her eyes with her hands'

Levin and Rappaport-Hovav (1995) describe the difference between internally and externally caused eventualities in the following way: the eventuality associated with a verb is internally caused if some property of the argument undergoing a change is responsible for bringing about that change. The eventuality

29 associated with a verb is externally caused if some external mediation is required for the event to take place.6 For example the eventuality associated with the English verb blush is internally caused because it can only come about through the properties of the person blushing. In contrast the eventuality associated with a verb like break is externally caused because a breaking event requires some external mediation in order to occur (although it is true that an entity must have certain properties in order to be breakable). Notice that, in English, externally caused verbs can occur with the phrase by itself on the 'without outside help' reading, but internally caused verbs cannot: (25)

a.

the vase broke by itself

b.

*Mary blushed by herself

In (25a), the theme is identified with the external cause and thus the anaphor is allowed. However, in (25b), since there is no external cause, the anaphor is not 6

Acutally, Levin and Rappaport-Hovav introduce this distinction only in the domain of monadic predicates. Thus, their definition is somewhat simpler. They say, "With an intransitive verb describing an internally caused eventuality, some property inherent to the argument of the verb is 'responsible' for bringing about the eventuality." (p. 91) And, "Unlike internally caused verbs, externally caused verbs by their very nature imply the existence of an 'external cause' with immediate control over bringing about the eventuality described by the verb: an agent, an instrument, a natural force, or a circumstance." (p. 92) Extending these ideas to transitive verbs requires a statement of which argument counts as relevant with respect to the distinction. We have opted above for the claim that the internal/external cause distinction is determined with respect to the entity undergoing a change, but it is possible that each argument be evaluated independently with respect to this distinction. Consider the following: (i) John opened Bill's eyes Here the sentence is externally caused if external cause is evaluated only with respect to the element undergoing a change (in this case, the eyes). It is internally caused if an external cause must be external to the entire predication . This issue is addressed below.

30 licensed. Note also that whether an event is internally or externally caused is a property of the relationship between the grammar and the human conceptual system, and not of the world itself. Some events which could be conceptualized as either internally or externally caused will vary crosslinguistically with respect to their grammatical encoding (cf. Haspelmath 1993). Returning to the Kannada examples above, we can see that it is the notion of responsibility that underlies the internal/external cause distinction. Internally driven eventualities that unfold because of natural properties of an entity are internally caused, while eventualities that unfold because of outside factors are externally caused. The entity that is responsible for the event's taking place is the cause, whether internal or external. Similar evidence for the verbal reflexive being a mark of external causation can be seen in the intransitive uses of causative verbs. (26)

a.

Naan-u baagil-annu mucˇcˇ-i-de I-NOM door-ACC close-PST-1S 'I closed the door'

b.

baagil-u mucˇcˇ-i-tu door-NOM close-PST-3SN 'The door closed'

c.

baagil-u mucˇcˇ-i-koND-itu door-NOM close-PP-REFL.PST-3SN 'The door closed'

Again we are faced with the question of the semantic distinction between sentences like (26b) and (26c). And again it seems as though it is the distinction between externally and internally caused eventualities that can distinguish the two sentences. Consider the following Kannada construction in which a cause adjunct appears at the beginning of the sentence in dative case (Giridhar 1987):

31 (27)

a.

divya-la celuv-ige divaakara mana soot-a divya-GEN beauty-DAT divakara-NOM mind lose.PST-3SM 'Divakara lost his mind because of Divya's beauty'

b.

caL-ige avan-age jvara bant-u cold-DAT he-DAT fever-NOM come-PST-3SN 'He got a fever because of the cold'

When the noncausative (intransitive) variant of a causative verb is used with a cause adjunct, only the reflexive version is possible. (28)

a.

gaaL-ige baagil-u mucˇcˇ-i-koND-itu wind-DAT door-NOM close-PP-REFL.PST-3SN 'the door closed because of the wind'

b.

* gaaL-ige baagil-u mucˇcˇ-i-tu wind-DAT door-NOM close-PST-3SN

Since the cause adjunct is only possible with these verbs when the reflexive morpheme is present, we can see again that the reflexive morpheme is associated with an external role in the event. That is, the reflexive morpheme is present only when we have an external cause interpretation. Labelle (1992) also shows a connection between the reflexive morpheme in French and external causation. Consider the following. (29)

a.

Le ballon gonfle (*de gaz carbonique) depuis cinq minutes the balloon inflates (*with gas carbonic) after five minutes “the balloon inflates (*with gas) after five minutes”

b.

Le ballon se gonfle (de gaz carbonique) depuis cinq minutes the balloon REFL inflates (with gas carbonic) after five minutes “the ballon inflates (with gas) after five minutes”

In this case we have a decausative intransitive that can occur either with or without the reflexive clitic. Crucially, when the entity responsible for the unfolding of the event is present in an adjunct position, only the variant with the reflexive clitic is possible,

32 suggesting again a connection between external causation and reflexive morphology. Just as the instrument was licensed only when the reflexive-marker was present in (1924), here again, the external cause is licensed only in the presence of the reflexivemarker. Thus, we have strong support for establishing a connection between reflexive morphemes and external causation. Having established that the reflexive morpheme indicates external causation in two non-reflexive uses (at least in Kannada and Romance), we can ask whether even normal reflexivity is an instance of external causation. If so, then what has been called reflexivity in the past is actually a kind of external causation. If, on the other hand, reflexivity cannot be reduced to external causation, we must look to other hypotheses in order to unify the uses of reflexive morphemes. Consider an event that is prototypically reflexive, like (30): (30)

Hari-yu tann-annu hode-du-koND-a hari-NOM self-ACC hit-PP-REFL.PST-3SM 'hari hit himself'

Without a proper characterization of external causation, it is hard to say whether this event is internally or externally caused. We said above that an eventuality is externally caused if the entity undergoing a change is distinct from the entity that is responsible for bringing about the event. We also said that an event is internally caused if the entity undergoing a change is responsible for bringing about the event. The problem with (30) is that there is no entity undergoing a change so it would seem that asking whether the eventuality described by the sentence is externally caused is not a well-formed question. Further, not all externally caused eventualities allow the reflexive morpheme. Consider the following:

33 (31)

*raaju hari-ya kannu-gaL-annu tere-du-koND-a raaju-NOM hari-GEN eye-PL-ACC open-PP-REFL.PST-3SM 'Raaju opened Hari's eyes'

Here, the entity undergoing a change is not responsible for bringing about the event, so the event should count as externally caused. But the reflexive morpheme is not allowed, casting some doubt that the reflexive morpheme only indicates external causation. We could, however, define external/internal causation not relative to the argument undergoing a change, but rather with respect to both arguments. On this view, if neither argument is responsible for bringing about the event, then the event is externally caused by default. If either argument is responsible for bringing about the event, then it is internally caused. So, in (31), the reflexive morpheme is blocked because the subject is responsible for bringing about the event. The problem with this account of external causation can be seen in a sentence like (32), in which neither argument is responsible for bringing about the event, but still, the reflexive morpheme is not allowed. (32)

*Raaju Hariyannu hoDe-du-koND-u-bitt-a Raaju Hari-ACC hit-PP-REFL-PP-INCIDENTAL.PST-3SM ‘Raaju inadvertently hit Hari’ These considerations make it difficult to maintain the idea that reflexive

morphemes are only external-causation morphemes. On either definition of external causation, we find examples of externally caused eventualities that do not license the reflexive morpheme and instances of the verbal reflexive in the absence of external causation. We therefore discount the possibility that the use of the reflexive morpheme is dependent upon this single semantic factor. However, the fact that the reflexive morpheme does correlate with our intuitive notion of external causation in

34 many of the non-reflexive examples above suggests that there is some generalization to be captured

2.3 The Mismatch Hypothesis In this section I will argue that reflexive-marking arises as a result of a mismatch between the thematic and aspectual tiers of argument-structure. Before we can see how this works, we must first examine argument-structure representations. 2.3.1 Multi-dimensional argument-structure Grimshaw (1990), Levin and Rappaport (1986; 1988), Williams (1981) and Zubizaretta (1987), among others, posit an independent level of representation which interfaces with representations in lexical conceptual structure (Jackendoff 1983; 1990; Hale and Keyser 1986) and with syntactic representations. Standardly, argument-structure representations, which are derived from lexical-semantic representations, consist of an ordered set of elements which map into D-structure positions. Grimshaw (1990) also treats argument-structure representations as structured representations over which relations of prominence are defined. The basic idea is that each verb has a lexical-semantic representation which includes at least the participants in the eventualities described by the verb. Some of these participants are realized as syntactic arguments and are thus represented in the level of argumentstructure. The character of the argument-structure, in combination with other syntactic properties, then determines the syntactic position of arguments. Argument-structure consists of two tiers: a thematic tier, which represents elements projected from

35 conceptual structure (Jackendoff 1983; 1990), and an aspectual tier which represents elements projected from event structure (Pustejovsky 1991).7 As Grimshaw notes, assuming that the elements on the thematic tier are the only elements in argument-structure is problematic when handling psych-verbs. Consider a pair like (33), in which both verbs project a theme and an experiencer. (33)

a.

John fears the wind

b.

The wind frightens John

If thematic structure were all that were relevant to determining syntactic position, then we should not expect to find pairs like these in which one verb projects the experiencer to subject position, like (33a), and the other projects the experiencer to object position, like (33b). Grimshaw notes, however, that there is a difference between fear and frighten with respect to aspectual type. While fear is stative, frighten is causative . That is, a reasonable paraphrase of (33b) is: (34)

The wind causes John to experience fear.

If we take the inability to take the progressive as indicative of stative verbs, then stativity does distinguish the two verbs. (35)

a.

*John is fearing the wind

b.

The wind is frightening John

The difference between the realization of the arguments in the two sentences is clear. The cause argument of causative predicates is always realized as a subject. So, because the theme is also a cause in (33b), it is realized as the subject. In 7

Jackendoff (1987, 1990) and Li (1995) have similar two-tiered argument structure representations with different formal properties. Zaenen and Goldberg (1993) argue against Grimshaw's approach.

36 (33a), there is no cause and thus the experiencer is realized as the subject and the theme as the object. However, merely stating that cause arguments are always subjects does not explain the phenomenon. To understand the phenomenon more thoroughly, Grimshaw posits a second, aspectual, dimension of argument-structure which represents the prominence of elements with respect to their participation in event structure. Drawing on work by Pustejovsky (1991), Grimshaw argues that each verb has an event structure associated with it. It is the event structure which determines the aspectual analysis of the clause in which the verb appears by breaking an event down into aspectual subparts. For example, an accomplishment, like break, can be seen as a transition between two sub-events (cf. Dowty 1979). The first sub-event is an activity (hitting, perhaps) and the second sub-event is the resultant state (material separation, perhaps). So, break has the following event structure: (36)

[[1 ACT ON 2] [3 CHANGE]]. This representation is composed of two subevents, the first representing

the action and the second representing the material separation of the affected object.8 More generally, accomplishments are composed of two sub-events: an activity and a resultant state, as in (37).

8

Note that the predicates used here are not intended to be primitives of event structure, but are merely instructive as to the character of this level of representation. Whether the details of the action are represented at this level of representation is an open question. It could simply be that rough grained details of the event are represented here (as above) with more details expressed in LCS or other dimensions of lexical representation. Alternatively, details about the nature of the action could be represented here. This is an empirical issue, of course, but is not immediately relevant to this thesis.

37 (37)

event activity

state

States (like know) and activities (like walk) can be seen as composed of only one subevent since no transition is involved. It is the level of event structure which underlies the projection of the aspectual tier in argument-structure. As for the thematic tier, this representation is projected from LCS.9 For a verb like break, we have the following LCS. (38)

[CAUSE [x], [BE [y] , [BROKEN]]] Having set out the basics of the representations that underlie argument-

structure, let us now consider how the argument-structure representation of break is derived. On the thematic tier (a projection of LCS), we have the first argument of CAUSE and the first argument of BE. Since BE is embedded within CAUSE, the first argument of CAUSE projects to a higher position in argument-structure. That is, prominence on the thematic tier is directly derivable from prominence within LCS. On the aspectual tier, we have three arguments. In the first sub-event, ACT-ON, there are two arguments and in the second subevent, CHANGE, there is one argument. Thus, three arguments project to the aspectual tier. Elements in the first subevent are more prominent than elements in the second subevent. Among the elements within the first subevent, the first argument is more prominent than the second. In this example, the second argument of ACT-ON is coreferential with the argument of CHANGE. That these aspectual elements correspond to only one entity is expressed

9

Note that Grimshaw states that the thematic tier should be derivable from LCS but gives no explicit theory of how this works. In what follows, the rules for deriving the thematic tier from LCS will be made clear.

38 by linking them both to the same element on the thematic tier. That is, both 2 and 3 are linked to y. The argument-structure for break is: (39)

(x (y))

thematic tier

(1 (2 (3)))

aspectual tier

2.3.2 The verbal reflexive is a mismatch morpheme Assuming that the empirical basis for these representations is real, the question we have to ask is whether such representations give us the tools to predict the behavior of verbal reflexives. We expect that these representations can give us some insight into why the same element is used to indicate external causation in some cases and reflexivity in others. That is, given that reflexive morphemes are used in both reflexive and externally caused eventualities, what is the nature of the lexical-semantic and lexical-syntactic representations such that they are treated as equivalent in the morpho-syntax? In order to answer this question, we have essentially to view the morpho-syntax as missing a distinction that is present at the lexical-semantic level. Thus, we can examine the possibilities for making the lexical-syntactic representations of these eventualities the same. Our task, then, is to show that one property of argument-structure representation can account for all of the uses the verbal reflexive.10

10

Note that showing that the verbal reflexive is best characterized by a particular configuration in a two-tiered argument structure also provides evidence that argument structures have this two-diemsional character.

39 2.3.2.1 The argument-structure of external causation Consider first the representation for (40). Here, as discussed above, we have an externally caused reading. The reflexive morpheme is only compatible with an interpretation in which the eyes open via some external mediation. (40)

Hari-yu kannu-gal-annu tere-du-koND-a hari-NOM eye-PL-ACC open-PP-REFL.PST-3SM 'Hari opened his eyes (with some instrument)' Let us consider the LCS and Event-Structure representations for this

sentence. In LCS, open has the following representation (like break above): (41)

[CAUSE [x], [BE [y], [OPEN]]]

The Event-structure representation is: (42)

[1 ACT-ON 2] [3 CHANGE]

This gives us the argument-structure representation in (43). (43)

(x (y)) (1 (2 (3)))

thematic tier (x=Hari, y=eyes) aspectual tier

Let us now consider what the linkings between tiers are in (43) and also how they project into the syntax. Hari and eyes correspond to x and y, respectively. We know this because it is the eyes that are opening and Hari that is causing the opening. As for the event structure, recall that we have an interpretation in which something happened to Hari such that his eyes opened. Essentially, we know that the instrument is acting on Hari causing his eyes to open. So, we can view 1 as corresponding to the instrument because it is the thing that is responsible for the opening. 2, then, must be Hari, since he is acted upon, and 3 the eyes, since they are

40 the changed entity.11 So, here, x corresponds to 2 and y corresponds to 3. Thus, we have an extra, unlinked, argument on the aspectual tier.12 The difference between this representation and the representation for (44), a sentence without the reflexive morpheme, lies in the aspectual tier. Consider how the argument-structure of (44) is derived. (44)

hari kannu-gaL-annu tere-d-a Hari-NOM eye-PL-ACC open-PST-3SM ‘Hari opened his eyes’

Again we have two elements on the thematic tier and three on the aspectual tier. However, the linkings between tiers are different in this case. When the eyes are opening by their own internal properties, we can view the first subevent as Hari acting on his eyes and the resultant state as the eyes being open. Thus, the first element on the aspectual tier links to the first element on the thematic tier and the second and third elements on the aspectual tier each link to the second element on the thematic tier. This gives us a configuration like (45). 11

A question arises as to whether the instrument in the syntax actually fills the role of the unlinked aspectual element or whether this connection is achieved through a pragmatic inference. That is, instruments are also licensed in cases without unlinked elements, as in John opened the door with a hammer, suggesting that instruments do not arise solely from unlinked aspectual elements. In such sentences, the instrument is associated in some intuitive sense with the actor on the aspectual tier but it is the subject (John) which actually fills that role. This would suggest that the availability of the instrument in (40) above may not be directly dependent on the unlinked aspectual element. On the other hand, if the unlinked aspectual element does not directly license the instrument, then it is a mystery why an instrument is not available when we have coreference between the subject and the possessor of the direct object and no verbal reflexive. We must conclude, then, that there is more than one way to license an instrument, a not surpising result. 12

These representations suggest that the descriptive term 'external cause' might best be thought of as 'external act-er on-er.'

41 (45)

(x

(y))

(1 (2 (3)))

thematic tier (x=Hari, y=eyes) aspectual tier

Note that while in (43) we had an unlinked element on the aspectual tier, the semantics of (44) gives rise to an argument-structure without an unlinked element.13 The contrast illustrated by (40/43) and (44/45) gives us a possible explanation for the presence of the verbal reflexive. It occurs in the sentences with a mismatch between the two tiers, i.e., when there is an unlinked element at the top of the aspectual tier. Given these representations, we should look to the other sentences with the reflexive morpheme. If these sentences have a similar argument-structure configuration, then this argument-structure representation is responsible for the "reflexive" morpheme. We will call the hypothesis that the verbal reflexive occurs only when the most prominent element on the aspectual tier is unlinked the Mismatch Hypothesis. Consider (46): 13

One might think that a possible interpretation of the aspectual tier is one in which 1 represents the external cause and both 2 and 3 represent the eyes. That is, we might think that it is the eyes that are acted upon. A similar example, however, shows that the element undergoing a change need not be acted upon in sentences licensing the verbal reflexive. Consider the following: (i)

Hari tanna naaligey-annu chaach-i-(koND)-a hari self-GEN tongue-ACC stick.out-PP-REFL.PST-3SM 'hari stuck out his tongue'

Here, the reflexive marking is possible only on the interpretation that Hari was provoked into sticking his tongue out. While it is the tongue that gets stuck out, it is Hari that is acted upon. Nothing is done to the tongue in order to cause it to come out. Thus, in the externally caused sentences in which the subject is coreferent with the possessor of the object, we can conclude that it is the subject and not the object which corresponds to 2 on the aspectual tier.

42 (46)

bagil-u much-i-koND-itu door-NOM close-PP-REFL.PST-3SN 'the door closed' (externally caused)

Let us assume that the argument-structure for transitive close is like break above and that intransitive verbs which alternate with transitives (causatives) are derived from the causative/transitive use (Levin and Rappaport-Hovav 1995).14 For a sentence like (47), we have the representation in (48). (47)

(48)

hari bagil-annu much-id-a hari door-ACC close-PST-3SM ‘Hari closed the door’ (x

(y))

(1 (2 (3)))

thematic tier (x=Hari, y=door) aspectual tier

Here, x is linked to 1 and y to 2 and 3. Given this representation, we know that one argument must be suppressed in order to derive the intransitive use. If we suppress the most prominent element on the thematic tier, then we will have an unlinked argument on the aspectual tier, just as in (45) above. So, the representation for (46) is (49). (49)

(y) (1 (2 (3)))

thematic tier (y=door) aspectual tier

Recall the observation that sentences with the reflexive clitic also optionally allow the presence of an adjunct corresponding to the external cause. If such adjuncts are licensed by the unlinked element on the aspectual tier, then we can understand the connection between the reflexive clitic, external causation and the 14

The possible subjects of the intrasitive variants of alternating verbs are a subset of the possible objects of transitive variants, suggesting that the transitive variant is underlying. See Levin and Rappaport-Hovav (1995) for extensive discussion.

43 presence of an adjunct. Since the reflexive indicates that there is an unlinked argument on the aspectual tier, the licensing of the adjunct is straightforward. The adjunct simply corresponds to the unlinked element. Because this element was not projected into an argument position, it can only be represented by an adjunct.15 At this point we have shown that the two externally caused uses of reflexive-markers can be captured by a two tiered argument-structure representation in which the most prominent element on the aspectual tier is not linked to an argument in the syntax. We have also shown how the dependence of certain adjuncts on the verbal reflexive falls out from such an account. What remains is to show that a similar mismatch is involved in the representation of canonical reflexivity. 2.3.2.2

The argument-structure of reflexivity What is crucial in understanding reflexivity is the coreference between the

two arguments. By definition, in a reflexive structure two arguments are linked to each other. This linking takes place through the aspectual tier so that the two arguments on the thematic tier are linked to one element on the aspectual tier.16 Coreference gives us a representation like (50b) for (50a).

15

See Grimshaw and Vikner (1995) for additional examples of elements on the aspectual tier licensing adjuncts. 16

In Chapter 4, I will argue that the grammar prefers reflexivity to be expressed lexically (i.e., not syntactically) whenever possible. The lexical expression of reflexivity is realized through the colinking of arguments illustrated in (50b). This preference is ultimately explained by a principle of economy limiting the syntactic operations. Its conseqence is that whenever two elements on the thematic tier are coreferential, they must be linked to the same element on the aspectual tier.

44 (50)

a.

Hari-yu (tann-annu) hoDe-du-koNDa Hari-NOM (self-ACC) hit-PP-REFL.PST-3SM ‘Hari hit himself’

b.

(x (y))

thematic tier (x=Hari, y=self)

(1 (2))

aspectual tier17

Since the coreference of x and y is realized by linking them to the same argument (2) on the aspectual tier, the most prominent argument on that tier remains unlinked. This representation captures the fact that the object is optional in sentences like (50a). That is, the thematic tier of (50b) can be interpreted as projecting two syntactic arguments or one. The optionality of the argument can be taken as preliminary evidence that the structure in (50b) can project either one or two arguments into the syntax and that no matter how many arguments are projected, the colinking of x and y results in a mismatch between the thematic and aspectual tiers.18 Finally, this representation requires that the coindexation of x and y be realized by linking both arguments to y's link. This follows straightforwardly from Grimshaw's system in which all linking begins with the least prominent element on the thematic tier. In this case, we first link y to 2, then, to realize the coreference, link x to 2. This leaves 1 unlinked, licensing the verbal reflexive. We can conclude, then, that the 'meaning' of the verbal reflexive is not reflexivity at all, but rather any meaning that is compatible with an argument-structure 17

Note that hit is an activity verb and has only one subevent. This has no effect on the analysis of verbal reflexives. 18

The intuition here is that when two thematic elements link to one aspectual element, there is an ambiguity with respect to how many NPs in the syntax correspond to these two thematic elements. In some languages, like Italian, a configuration like (50b) only allows one argument to project into the syntax while Kannada optionally projects two. More on this below.

45 representation in which the most prominent argument on the aspectual tier is not linked to an argument position in the syntax. This is given in (51): (51)

(x (y...)) (1... n ...)

thematic tier aspectual tier

Here, the most prominent element on the thematic tier links to something that is not the most prominent element on the aspectual tier. The most prominent element on the aspectual tier is not linked to anything on the thematic tier and thus not to an argument position in the syntax. The representation in (51) predicts that not any sort of unlinking is sufficient to license the reflexive morpheme but that if there were a configuration in which the most prominent element on the aspectual tier linked to a non-prominent element on the thematic tier, the reflexive morpheme could not be licensed. Such a case is readily available. Recall that the way Grimshaw captures the distinction between fear and frighten is to posit an asymmetry in prominence between the thematic and aspectual tiers, where the most prominent element on the aspectual tier is linked to a non-prominent element on the thematic tier and that prominence on the aspectual tier is more important to determining subject position than prominence on the thematic tier. Further, Belletti and Rizzi (1988) show that verbs in the frighten sub-class of psych-verbs do not allow reflexive cliticization. (52)

a.

questo preoccupa Gianni this worries Gianni 'This worries Gianni'

b.

*Gianni si preoccupa Gianni REFL worries 'Gianni worries himself'

46 This fact follows directly from the analysis of reflexive morphemes presented here. Consider the argument-structure representation of preoccupare, a verb in the frighten subclass in Italian: (53)

(x (y))

thematic tier (x=Gianni, y=this)

(1 (2))

aspectual tier

As we saw above, in Grimshaw’s analysis 1 is linked to y and 2 is linked to x. So, if x and y were colinked, they would still be linked to the most prominent element on the aspectual tier, assuming, as she argues we should, that linking begins with the least prominent element on the thematic tier. Since reflexive morphemes are licensed only when the most prominent element on the aspectual tier is unlinked, then it follows that verbs in the frighten subclass of psych-verbs do not allow reflexive cliticization in Italian. In other words, if we marked x and y as coreferential on the thematic tier, it would be the least prominent element on the aspectual tier that is unlinked, not the most prominent, so the conditions licensing the reflexive clitic would not be met. (54)

(x (y))

thematic tier (x=Gianni, y=this)

(1 (2))

aspectual tier

Note that this representation differs from the one that arises when preoccupare is used as an inchoative, as in (55). (55)

Gianni si preoccupa di questo Gianni REFL worries of this 'Gianni gets worried from this'

Here, we delete the subject θ-role (in this case, the theme, questo), giving us the representation in (56), in which the most prominent element on the aspectual tier is unlinked, licensing the reflexive clitic.

47 (56)

(x) (1 (2)) Belletti and Rizzi (1988) also show that the reflexive clitic in Italian is not

possible with derived subjects, as in passive or raising constructions. (57)

a.

*Gianni si è stato affidato Gianni REFL is been entrusted 'Gianni was entrusted to himself'

b.

*Gianni si sembra simpatico Gianni REFL seems nice 'Gianni seems to himself to be nice'

Similarly, Kannada disallows the verbal reflexive in passive and raising constructions. (58)

a.

rama tann-inda vancis-al-paTT-a rama self-INSTR deceive-INF-PASS.PST-3SM 'Rama was deceived by himself'

b.

*rama tann-inda vancis-koLL-al-paTT-a rama self-INSTR deceive-INF-PASS-PP-REFL.PST-3SM 'Rama was deceived by himself'

c.

hari santooshaag-iruw-aage kaNis-utt-aane hari happy-be-PRED seem-PRES-3SM 'Hari seems to be happy'

d.

*hari tan-age santooshaag-iruw-aage kaNis-koLL-utt-aane hari self-DAT happy-be-PRED seem-REFL-PST-3SM 'Hari seems to himself to be happy'

These facts also fall out naturally from the analysis presented here. Raising verbs, like seem, will allow either one or two arguments. In (58c), for the verb seem, we have only one argument on the thematic tier (the predicate happy ), while in (58d) we have the additional argument represented by the reflexive pronoun. The argument-structure of (58d) is illustrated in (59).

48 (59)

(x (y))

thematic tier (x=happy, y=himself)

(1 (2))

aspectual tier (1=happy, 2=himself)

The fact that move-α has applied to the argument of happy (i.e., Hari) has no effect on the argument-structure whatsoever. No mismatch arises and the verbal reflexive is not licensed. As for the passive cases, the theory under discussion also makes the right predictions. In order to see how this works, we should first consider the nature of the element that is suppressed in a passive. The distinction we will need to draw in these cases is between an element in argument structure that is suppressed and one that is deleted. Roeper (1987) shows that the 'suppressed' argument in passives may surface in a by-phrase, but the 'suppressed' argument in decausatives may not. (60)

a.

The ship was sunk (by the navy)

b.

the ship sank (*by the navy)

We will take this evidence to indicate that the argument represented by the by-phrase in a passive is still available to the argument-structure, even though it is suppressed, while the same argument is simply missing from the thematic tier of the inchoative (cf. Grimshaw 1990, ch. 4; Zubizaretta 1987, ch. 3). The argumentstructure for (60a) is (61) and the argument-structure for (60b) is (62). (61)

(62)

(x-ø (y))

thematic tier (x=navy, y=ship)

(1

aspectual tier (1=navy, 2=ship)

(2)) (y)

(1 (2))

thematic tier (y=ship) aspectual tier (1=navy, 2=ship)

The suppressed argument in (61) is still available to for linking with the most prominent element on the aspectual tier, even though it is marked as suppressed.

49 The suppression simply indicates that it may not fill an argument position in the syntax. In (62), on the other hand, the argument on the thematic tier which would have represented "navy" has been deleted, leaving the most prominent element on the aspectual tier unlinked. Returning to (58b) above, the question is what blocks the reflexivemarking in passives. First, recall that when an element is marked in argument structure as suppressed, that means that it may not be linked to an argument position in the syntax. Second, recall that in active reflexive sentences, the reflexive-marking arises because both elements on the thematic tier are linked to the same element on the aspectual tier, leaving an unlinked element on the aspectual tier. So, if we were to double-link an element that has been suppressed, we would violate the condition that such an element may not occur in an argument position. Thus, reflexive-marking is not possible in passives. The potential representations for a reflexive-marked passive are given in (63). (63)

a.

b.

(x-ø (y))

thematic tier

(1

aspectual tier

(2))

(x-ø (y)) (1

(2))

In (63a), to express the reflexivity of the predicate, we link x and y to the same aspectual element. This leaves the most prominent aspectual element unlinked and should give rise to the reflexive morpheme. However, x is marked as suppressed (by the lexical rule of passive). By linking x and y together, x projects into a syntactic argument position, violating the condition imposed by the passive rule that x not be

50 expressed in such a position. (63a) is ruled out, blocking reflexive-marking on passives. In (63b), on the other hand, the reflexivity is not expressed in the argument structure representation so an unlinked element on the aspectual tier does not arise. (63b) does not license the reflexive morpheme but does satisfy the condition imposed by the passive rule. Finally, the discussion of (60) leads to a question of the difference between the decausatives that allow the verbal reflexive and those that don't. That is, if decausatives are different from passives in deleting rather than suppressing the most prominent thematic element, then what differentiates the two types of decausative, given in (64), from each other? (64)

a.

baagil-u mucˇcˇ-i-koND-itu door-NOM close-PP-REFL.PST-3SN 'The door closed'

b.

baagil-u mucˇcˇ-i-tu door-NOM close-PST-3SN 'The door closed'

As we noted above, the reflexive-marked decausatives in Kannada license causal adjuncts while the non-reflexive-marked decausatives do not, as illustrated in (65): (65)

a.

gaal-ige baagil-u mucˇcˇ-i-koND-itu wind-DAT door-NOM close-PP-REFL.PST-3SN 'Because of the wind, the door closed'

b.

*gaal-ige baagil-u mucˇcˇ-i-tu wind-DAT door-NOM close-PST-3SN 'Because of the wind, the door closed'

We also claimed that these adjuncts were licensed by the presence of an unlinked element on the aspectual tier, explaining the correlation between the the adjuncts and

51 the reflexive morphology. Thus, the non-reflexive-marked decausatives must not have this unlinked element on the aspectual tier. We can capture the distinction through the ordering of operations within the argument structure. Consider first the derivation of (64a). Here we project the argument structure of the transitive variant, delete the most prominent thematic element, and then establish the linkings between the tiers:19 (66)

(x (y))



(1 (2 (3)))

(y)



(1 (2 (3)))

(y) (1 (2 (3)))

Here we have an unlinked element at the top of the aspectual tier, licensing the verbal reflexive. Now consider the derivation of (64b). Here, we establish the linkings between the tiers and then delete the most prominent thematic element. Because this element has already been linked to something on the aspectual tier, that aspectual element is also deleted. (67)

(x (y))



(1 (2 (3)))

(x (y)) (1 (2 (3)))



(y) (2 (3))

2.4 On Argument Absorption In the previous sections, I developed an analysis which treats verbal reflexives as the morpho-syntactic instantiation of a particular argument-structure representation. Previous researchers have claimed that verbal reflexives (particularly,

19

Alternatively, we could say that the deletion of the agent takes place within the LCS and so it never exists on the thematic tier.

52 the Romance reflexive clitics) are functors which operate on argument-structures (Grimshaw 1982, 1990; Wehrli 1986). On this view, the reflexive element absorbs one of the arguments from the argument-structure representation. The basic point of contention between this approach and the Mismatch Hypothesis advocated above lies in the nature of operations on the argument-structure. For the absorption approach, it is a particular morpheme which is responsible for an operation over a given argumentstructure representation. Under the Mismatch Hypothesis, operations on argumentstructure apply freely; the morpheme signals the resulting structure. There are empirical reasons to adopt the Mismatch Hypothesis over an absorption approach. Before we can see the empirical differences between these analyses, we must examine the motivations for the argument absorption hypothesis. 2.4.1 Motivations for the Argument Absorption Hypothesis Many languages treat verbs with reflexive morphemes as though they are intransitive. It is common for a language to show formal distinctions between transitive and intransitive verbs. In such languages, reflexive-marked verbs often pattern with the intransitives and not the transitives. In French, for example, the realization of the arguments of a predicate embedded under a causative verb depends upon the transitivity of that predicate (Kayne 1975; Aissen 1974). When the embedded verb is transitive, the causee is realized in dative case, introduced by the preposition à. On the other hand, when the embedded verb is intransitive, the causee is realized in accusative case, following the verb directly without a preposition. This is illustrated in (68). (68)

a.

J'ai fait partir Jean I have made leave Jean 'I made Jean leave'

53 b.

*J'ai fait partir à Jean

c.

Il fera boire un peu de vin à son enfant He will.make drink a bit of wine to his child 'He will make his child drink a little wine'

d.

*Il fera boire un peu de vin son enfant

When the lower verb is transitive, but its object is realized as a reflexive clitic, that verb behaves as though it were intransitive with respect to the causative construction, i.e., the causee is not introduced by a preposition. (69)

a.

La crainte du scandale a fait se tuer le juge the fear of scandal has made REFL kill the judge 'the fear of scandal made the judge kill himself'

b.

*La crainte du scandale a fait se tuer au juge

In contrast, when the lower verb is transitive and its object is realized as a pronominal clitic, the verb still patterns like a transitive. The causee must be introduced by a preposition. Thus, it is not simply cliticization that causes the intransitive behavior in (69), but the reflexive-marking: (70)

a.

La crainte du scandale l’a fait tuer au juge the fear of scandal him.has made kill to.the judge 'the fear of scandal made the judge kill him'

b.

*La crainte du scandale l’a fait tuer le juge

A similar contrast between reflexive clitics and non-reflexive clitics in French comes from NP extraposition structures (Kayne 1975). Again we see that verbs marked with reflexive clitics behave as though they are intransitive while verbs that are marked with non-reflexive clitics behave as though they are transitive. Extraposition in French takes an indefinite subject and puts it in postverbal position,

54 filling the subject position with the impersonal il. NP extraposition is possible with intransitive verbs and not transitive verbs. This is illustrated in (71). (71)

a.

Un train passe a train passes 'A train passes'

b.

Il passe un train it passes a train 'A train passes'

c.

Quelqu’un lave une voiture someone washes a car 'Someone washes a car'

d.

*Il lave une voiture quelqu’un it washes a car someone 'Someone washes a car'

In this case also, verbs with reflexive clitics pattern with intransitives in allowing extraposition.20 Verbs with non-reflexive clitics pattern with transitives in disallowing extraposition. The contrast is illustrated in (72). (72)

a.

*Il la lave quelqu’un it it washes someone 'Someone washes it'

b.

Il se lave quelqu’un it REFL washes someone 'Someone washes himself'

Finally, it is well known that verbs marked with the reflexive clitic in Italian behave as though they are unaccusative with respect to auxiliary selection

20

Burzio (1981) argues that embedding under causative and extraposition are actually tests for unaccusativity and not simple intransitivity, but the point stands: verbs with reflexive-clitics behave as though they are monadic and not dyadic.

55 (Burzio 1981; Rosen 1981). In Italian, transitive and unergative verbs select the auxiliary HAVE, while unaccusative verbs select BE. This is illustrated in (73). (73)

a.

Gianni ha accusato Maria Gianni has accused Maria 'Gianni accused Maria'

b.

Gianni ha telefonato Gianni has telephoned 'Gianni telephoned'

c.

Gianni è caduto Gianni is fallen 'Gianni fell'

Once again, typically transitive verbs occurring with the reflexive clitic pattern with unaccusatives in selecting BE and not HAVE, while the same verbs with the non-reflexive clitic continue to select HAVE and not BE. (74)

a.

Maria si è accusata Maria REFL is accused 'Maria accused herself'

b.

Gianni la ha accusata Gianni her has accused 'Gianni accused her'

The fact that dyadic verbs marked with the reflexive clitic behave as though they are monadic with respect to these tests suggests that the reflexive clitic does not represent an argument like other pronominal clitics do, but rather is a lexical affix that absorbs an argument from the argument grid of the verb to which it attaches. This is the idea put forth by Grimshaw (1982) and Wehrli (1986). Supporting evidence for the view that reflexive clitics absorb an argument comes from the fact that many verbs that participate in the causative/inchoative alternation use the reflexive clitic in the inchoative variant. If we view the inchoative as a lexical

56 absorption of one of the arguments, then the reflexive clitic is responsible for the absorption. The alternation is illustrated here in Italian. (75)

a.

Gianni rompe il vetro Gianni breaks the glass 'Gianni breaks the glass'

b.

Il vetro si rompe the glass REFL breaks 'the glass breaks'

Arguments for the monadic status of reflexive-marked verbs can also be found in unrelated languages. In Yidiny, for example, the adicity of a verb can be determined on the basis of the case of its arguments (Dixon 1977). A dyadic verb shows ergative case on the subject and absolutive case on the object. A monadic verb shows absolutive case on the subject. (76)

a.

gudaaga gadang dog-(ABS) comes 'the dog is coming'

b.

waguda-nggu gudaaga bundaan man-ERG dog-(ABS) hit 'the man hit the dog'

When the verbal reflexive is added to the verb, we have the case-marking associated with monadic predicates: the subject in (77) is absolutive. (77)

nayu bambii-dinu I-(ABS) cover-REFL 'I covered myself' At this point, we have seen three arguments from Romance that when

verbs which are typically transitive occur with reflexive clitics, they behave as though they are intransitive. This contrasts with the situation found when the same verbs occur with non-reflexive clitics. In these cases, the verb maintains its transitivity.

57 Similarly, in Yidiny, reflexive morphology gives rise to the case pattern of intransitive verbs. The contrast has led several researchers to the conclusion that reflexive morphemes act as functors, altering the argument-structure of the verb to which they attach. We will call this the Argument Absorption Hypothesis. 2.4.2 Against the Argument Absorption Hypothesis There are several reasons to doubt that reflexive-marking arises as a result of absorption of an argument. First, as noted in section 4.1, many languages with verbal reflexives maintain their transitivity when the verbal reflexive occurs. (78) Kannada hari (tann-annu) hoDe-du-koND-a hari (self-ACC) hit-PP-REFL.PST-3SM 'Hari hit himself' (79) Padovano (Christina Tortora, personal communication) gianni se varda lu gianni REFL saw him 'Gianni saw himself' (80) Spanish juan se vió (a si mismo) juan REFL saw (to self same) 'Juan saw himself' (81) Yavapai hmañ-c yem lap-v-kñ child-SUBJ self strike-REFL-CMP 'The child struck himself' Here, the reflexive-marker could not be absorbing an argument because the "absorbed" argument is present in the syntax.21 Of course, we could argue that the

21

Several people have suggested that the pronouns present in (78-81) are actually absorbed NPs realized as chomeurs of some sort. It is important to see however that

58 reflexive marker in these languages is different from the ones in which transitivity is reduced. This approach, however, would not be satisfactory as it does not explain the similarities in constructions across languages, i.e., the fact that they all participate in the inchoative and possessive paradigms discussed above. Further evidence against the view that the reflexive clitic absorbs an argument comes from additional constructions in which the reflexive clitic is used without altering the transitivity of the verb, as in (82): (82)

a.

Maria abrió los ojos Maria opened the eyes 'Maria opened her eyes'

b.

Maria se abrió los ojos Maria REFL opened the eyes 'Maria opened her eyes'

One way for the Argument Absorption Hypothesis to handle sentences like (82) is to say that the reflexive clitic in (82b) absorbs a benefactive argument. On this account, open is optionally triadic and the presence of the reflexive clitic indicates an absorption of the third argument. To determine the viability of this account, we must reconsider the data involving the presence of an instrument in these sentences. As discussed in Section 2.2, the difference in interpretation between (82a) and (82b) lies in the way that the eyes are opened. In (82a), the eyes open via their own internal properties, while in (82b), some additional factor is involved in the opening. Recall that an instrumental adjunct is possible with (82b) and not (82a). In (82a), since the lack of the reflexive clitic indicates that the eyes opened naturally, the instrument is not licensed. This fact is repeated here: the these NPs show the surface case properties of direct objects and not chomeurs. In Kannada, for example, we find no other instances of accusative marked chomeurs.

59 (24)

a. ?? Maria abrió los ojos con las manos Maria opened the eyes with the hands 'Maria opened her eyes with her hands' b.

Maria se abrió los ojos con las manos Maria REFL opened the eyes with the hands 'Maria opened her eyes with her hands'

It is not clear what the presence or absence of an optional beneficiary has to do with instrument. While we could stipulate that Maria is more affected by her eyes opening when there is an instrument present than when they open naturally, thus licensing the instrument only when the benefactive argument is present (but suppressed), this would not be explanatory. That is, such a stipulation requires that this third θ-role can never be realized without the appearance of the reflexive clitic, the very element that was supposed to absorb it. We would prefer an account of the relationship of the instrument to the clitic to fall out from our theory of reflexive elements in general, as it does in the Mismatch Hypothesis. Further evidence against viewing the verbal reflexive as an argument absorber comes from Kannada. As we have noted, in Kannada when the verbal reflexive arises due to coreference between the subject and object, the expression of the reflexive object pronoun is optional. Similarly, when there is coreference between a subject and an indirect object, the verbal reflexive arises and the nominal reflexive is optional: (83)

hari (tan-age) pustaka-vannu koTTu-koND-a hari (self-DAT) book-ACC give.PP-REFL.PST-3SM 'Hari gave himself a book'

However, in the externally caused cases such as (84), expression of a dative reflexive pronoun is not possible, suggesting that no argument has been absorbed.

60 (84)

hari (*tan-age) kannu-gaL-annu tere-du-koND-a hari (*self-DAT) eye-PL-ACC open-PP-REFL.PST-3SM 'Hari opened his eyes' Under the Argument Absorption Hypothesis, all uses of the verbal

reflexive are due to the absorption of an argument. Assuming that this hypothesis can explain the optional occurrence of reflexive pronouns in sentences like (83), it will still require additional mechanisms to explain the difference between (83) and (84). Under the Mismatch Hypothesis, however, these facts are explained straightforwardly. In (83), the reflexive marking arises because the thematic elements representing the subject and indirect object are linked to the same aspectual element, i.e., due to standard reflexivity. In (84), on the other hand, the reflexive marking arises because of the argument-structure associated with external causation. Because there is no coreference between the subject and another argument, no reflexive pronoun can occur. Because the verbal reflexive arises only when the most prominent element on the aspectual tier is unlinked, several different argument-structures can give rise to the verbal reflexive. Some of these will be compatible with the expression of a pronominal reflexive while others will not. From the perspective of the argument absorption hypothesis, however, the difference remains a mystery.

2.5 Alternative Analyses In the previous sections, we saw how the distribution of verbal reflexives depends upon a particular argument-structure configuration. An interesting consequence of this view is that a level of argument-structure is required to provide an

61 interface between lexical semantic representations and syntactic representations. One might think that it would be preferable to do away with this level of representation altogether. That is, a purely syntactic account that does not require a level of argument-structure might be preferred to the account given here for simplicity's sake since the Mismatch Hypothesis requires an additional level of representation. In this section, we will consider two purely syntactic analyses of verbal reflexives, showing that they cannot explain the full range of facts accounted for by the Mismatch Hypothesis. The first, a variant of which was argued to be responsible for reflexive morphology in Russian by Babby and Brecht (1975) and Babby (1975), claims that the reflexive morpheme surfaces whenever the direct object position has been "emptied" by some transformational rule. The second, an interpretation and expansion of the system argued by Reinhart and Reuland (1993) to explain the distribution of reflexive nominals in Dutch and English, claims that reflexive morphemes arise when two arguments of a predicate are coindexed. Both of these hypotheses will be shown to be untenable. 2.5.1 The Surface Intransitive Hypothesis Babby and Brecht (1975) and Babby (1975) argue that the Russian reflexive marker -sja is introduced transformationally whenever the direct object is either moved or deleted by syntactic rules. On this account, the reflexive marker is not identified with one sentence type (like passive) or one meaning (like reflexive), but rather only signals the surface intransitivity of an underlyingly transitive verb. In this sense, we have a constructional homonymity between syntactic representation and semantic representation. That is, the morpheme is an indicator of a syntactic

62 configuration and can occur with a range of meanings. We will call this the Surface Intransitive Hypothesis (SIH).22 Evidence for the SIH comes from the causative/inchoative alternation discussed above. As we have seen, many languages mark decausative verbs with the reflexive morpheme. Further, many of the languages in (1-9) do not allow the reflexive pronoun to surface when there is a reflexive marker on the verb. In Russian, for example, we find both a verbal reflexive and a reflexive pronoun. The two can never cooccur (Babby 1975): (85)

a.

On zastrelil-sja he shot-refl 'He shot himself'

b.

On zastrelil sebja he shot self 'He shot himself'

c.

*On zastrelil-sja sebja

Since the expression of the object is blocked by the verbal reflexive, we have evidence that verbs marked with the verbal reflexive are intransitive. Given the general cross-linguistic stability of the uses of reflexive morphology discussed above, we would expect that the SIH could be carried over to all of the languages that exhibit this morpheme. The SIH predicts that in all languages that exhibit this morpheme, verbs with this morpheme attached will behave as though they are intransitive. While it is common for reflexive verbs to be treated as

22

This hypothesis shares the basic intuition of the Argument Absorption Hypothesis: that reflexive-marked verbs are intransitive on the surface but transitive at some other, deeper, level. Many of the arguments raised against the AAH will hold against the SIH as well.

63 intransitive, in many languages reflexive verbs are transitive. For example, in Kannada, Padovano, Spanish and Yavapai, the direct object may occur when the reflexive marker is present on the verb:23 (86) Kannada hari (tann-annu) hoDe-du-koND-a hari (self-ACC) hit-PP-REFL.PST-3SM 'Hari hit himself' (87) Padovano (Christina Tortora, personal communication) Gianni se varda lu Gianni REFL saw him 'Gianni saw himself' (88) Spanish juan se vió (a si mismo) juan REFL saw (to self same) 'Juan saw himself' (89) Yavapai hmañ-c yem lap-v-kñ child-SUBJ self strike-REFL-CMP 'The child struck himself' While it is often the case that verbs marked with reflexive morphemes behave as though they are intransitive, detransitivization does not seem to be a necessary condition on licensing reflexive morphemes in all languages. Further, when the subject is coreferential with the possessor of the direct object, the verbal reflexive may occur even though the direct object has not been

23

Several readers have suggested that the possibility of having the verbal reflexive cooccur with a pronoun is correlated with clitic doubling in general. Kannada, however, doesn't have clitics or anything like clitic doubling in other constructions, making it unclear how to pursue this stragegy.

64 moved or deleted -- even in languages that generally treat reflexive verbs as intransitive.24 (90) Russian a. on rasstegnul pal'to he unbuttoned coat 'He unbuttoned his coat' b.

on rasstegnul-sja pal'to he unbuttoned-REFL coat 'He unbuttoned his coat'

(91) Kannada a. hari angiy-annu hari-d-a hari shirt-ACC tear-PST-3SM 'Hari tore his shirt' a.

hari angiy-annu hari-du-koND-a hari shirt-ACC tear-PP-REFL.PST-3SM 'Hari tore his shirt'

(92) Italian (Burzio 1994) a. Gianni apre gli occhi gianni opens the eyes 'Gianni opens his eyes'

24

It is important to see here that the reflexive marking does not represent a deleted indirect object. While an indirect object coreferent with the subject is possible in some sentences with the verbal reflexive (in Kannada), it is not possible in sentences like (91). (i)

Hari (tan-age) pustaka-vannu koTTu-koND-a Hari (self-DAT) book-ACC give.PP-REFL.PST-3SM 'Hari gave himself a book'

(ii)

Hari (*tan-age) angiy-annu hari-du-koND-a hari (*self-DAT) shirt-ACC tear-PP-REFL.PST-3SM 'Hari tore his shirt'

65 b.

Gianni si apre gli occhi gianni REFL opens the eyes 'Gianni opens his eyes'

The presence of the reflexive marker cannot be dependent on derived intransitivity. There are transitive verbs that occur with the reflexive marker. A second prediction of the SIH is that the reflexive morpheme should always surface when the direct object has been emptied. So, if there were examples of detransitivized verbs that did not allow the reflexive marker, we would have evidence against the SIH. This evidence is readily available. First, in languages like Kannada, verbs that alternate between a transitive and an intransitive use allow the reflexive marker optionally. That is, we have examples of 'detransitive' verbs that do not have the reflexive marker. (93)

a.

hari-yu baagil-annu mucˇcˇ-id-a hari-NOM door-ACC close-PST-3SM 'Hari closed the door'

b.

baagil-u mucˇcˇ-i-tu door-NOM close-PST-3SN 'The door closed'

c.

baagil-u mucˇcˇ-i-koND-itu door-NOM close-PP-REFL.PST-3SM 'The door closed'

Second, in some languages that have reflexive markers there are verbs that alternate between a transitive and an intransitive use which do not allow the reflexive marker in the intransitive use. If the reflexive marker were an indication that the direct object position had been emptied or deleted, we would expect it to occur on all verbs that had a direct object moved or deleted. Consider the following Italian examples (Burzio 1986):

66 (94)

(95)

a.

L'artiglieria affondò due navi the.artillery sink.PST.3S two ships 'The artillery sank two ships'

b.

due navi affondarono two ships sink.PST.3PL 'Two ships sank'

c.

*due navi si affondarono two ships REFL sink.PST.3PL

a.

Giovanni rompe il vetro giovanni break.3S the glass 'Giovanni breaks the glass'

b.

Il vetro si rompe the glass REFL break.3S 'The glass breaks'

Here, we have two verbs that participate in the causative/inchoative alternation, but only one of them allows the reflexive marker. If the reflexive marker were an indication that an underlyingly transitive verb were intransitive on the surface, we would expect the reflexive marker to be possible in both (94c) and (95b). Along the same lines, an argument against the SIH comes from antipassive constructions in which a transitive verb marked with a certain affix behaves like an intransitive, its direct object surfacing with an oblique case. As we will see in Chapter 3, many languages use the verbal reflexive in antipassive constructions. One such language is Diyari (Austin 1981): (96)

a.

ngaTu nana wila kalka-yi 1SG.ERG 3SG.F.ABS woman-(ABS) wait.for-PRES 'I wait for the woman'

b.

ngani kalka-tadi-yi nangkangu wila-ni 1SG.ABS wait.for-AP-PRES 3SG.F.LOC woman-LOC 'I wait for the woman'

67 c.

ngani muduwa-tadi-yi 1S.ABS scratch-REFL-PRES 'I scratch myself'

In (96a), we see the verb wait for in its typical, transitive use. The subject is marked ergative and the object absolutive. In (96b), the antipassive morpheme tadi has been affixed to the verb and the subject is marked with absolutive case, the normal case for subjects of intransitive verbs, and the object is in locative. In (96c), we see that the verbal reflexive is the same as the antipassive. Under the SIH, languages like Diyari are predicted to exist. In fact, the SIH predicts that any language with a verbal reflexive and an antipassive construction will use the same morpheme for both since both involve the D-structure object surfacing in some other position. This prediction is not borne out. There are also languages like Kalkatungu in which the verbal reflexive is not the same morpheme as the antipassive marker (Blake 1994): (97)

a.

marapai-thu rumpa-mi ithirr matyamirla-thu woman-ERG grind-FUT seed-(NOM) grindstone-ERG 'the woman will grind the seed with the grindstone'

b.

marapai rumpa-yi-mi ithirr-ku matyamirla-thu woman-(NOM) grind-AP-FUT seed-DAT grindstone-ERG 'the woman will grind seed with the grindstone'

c.

marapai karri-ti-mi thupu-ngku woman.NOM wash-REFL-FUT soap-ERG 'the woman will wash herself with soap'

In Chapter 3, I show how the existence of languages like Diyari and languages like Kalkatungu is able to be explained under the Mismatch Hypothesis. Finally, some verbs which take clausal complements allow the reflexive marker when the matrix subject is coreferential with the embedded subject. Here

68 again we find a case where the reflexive morpheme is possible even though the verb has maintained its transitivity. This is illustrated in Kannada: (98)

a.

hari tann-annu punyavantan-endu namb-utt-aane hari self-ACC wealthy-PRED believe-PRES-3SM 'Hari believes himself wealthy'

b.

hari tann-annu punyavantan-endu nambi-koLL-utt-aane hari self-ACC wealthy-PRED believe-REFL-PRES-3SM 'Hari believes himself wealthy'

The presence of the reflexive morpheme cannot be attributed crosslinguistically to the deletion or movement of the direct object. We have seen cases where the reflexive may occur even though the object is present, and we have seen cases where the movement or deletion of the object does not license the reflexive morpheme. Surface detransitivization is neither necessary nor sufficient to license reflexive morphology crosslinguistically. 2.5.2 The Reflexive Predicate Hypothesis A second possible syntactic approach to the multiple uses of reflexive morphemes might be created along the lines of Reinhart and Reuland's (1993) proposal that reflexivity is a property of predicates and not of arguments. On this view, reflexive morphology occurs whenever a predicate is reflexive. We will call this the Reflexive Predicate Hypothesis (RPH). Note that this view does not involve constructional homonymity. Here the mapping from form to meaning is direct. The mapping is captured in this theory by the following two binding conditions: (99) Condition A: a reflexive-marked predicate is reflexive (100) Condition B: a reflexive predicate is reflexive-marked (Reinhart and Reuland's (12") and (12'))

69 The mapping from form to meaning is direct here because there is a oneto-one correspondence between a semantic property (reflexive) and a formal property (reflexive-marked). Under this theory all reflexive-marked predicates should share a component of meaning. However, we argued above that there is not one unified semantics of reflexive-marked verbs (where reflexive-marked means the predicate is morphologically marked with a particular morpheme). Another possibility, then, is that "reflexive" is syntactic and not semantic. On this view the coreference between the arguments must be part of the syntactic representation. Under this interpretation of the RPH, the mapping is not from form to meaning but from morphology to syntax. That is, reflexive marking is a morpho-syntactic property and reflexive is a syntacticosemantic property. Even on this interpretation, however, the RPH cannot be maintained. Before we show why, let us examine the RPH itself. There are three components to the RPH: coargumenthood, reflexive and reflexive-marked. Two NPs are coarguments if they are assigned a θ-role or case by the same predicate. A predicate is reflexive if it has coindexed coarguments. Reflexive-marking depends on the morphological properties of the arguments of a predicate. Reinhart and Reuland distinguish between SELF anaphors, which are morphologically complex, like English himself, and SE anaphors, which are morphologically simplex, like Dutch zich. A predicate is reflexive-marked if one of its arguments is a SELF anaphor. The binding conditions under the RPH amount to saying that if two arguments of a predicate are coindexed, then one of the arguments must be a SELF anaphor because a SELF anaphor serves to reflexive-mark its predicate. Further, if one of the arguments of a predicate is a SELF anaphor, then it must be coindexed with

70 one of the other arguments. If a predicate is lexically reflexive, then it may have two coindexed arguments where neither is a SELF anaphor because a lexically reflexive predicate is already reflexive-marked. In the languages under discussion in this chapter, however, a predicate may be reflexive-marked morphologically. While Reinhart and Reuland allow a predicate to be reflexive-marked either through one of its arguments or lexically, these languages provide evidence for a third type of reflexive-marking.25 In order to see the extent to which languages that mark their predicates as reflexive with a morpheme can be brought under the RPH, let us examine Kannada as an example of one of these languages. First, it should be noted that the Kannada reflexive pronoun tannu is monomorphemic (Sridhar 1990, Amritavalli 1991) and thus qualifies as a SE anaphor in Reinhart and Reuland's system. It does not reflexivemark its predicate. Thus, we should not expect to see tannu locally bound. This is essentially true, as illustrated in (101) and (102). (101) *avanui tann-annui hoDe-d-a He self-ACC hit-PST-3SM (102)

raamui shyaamuk tann-annui/*k priitis-utt-aane anta namb-utt-aane Raamu Shyamu self-ACC love-PRES-3SM that believe-PRES-3SM 'Raamu believes that Shyamu loves self' (self=Raamu only)

However, Kannada does not have a SELF anaphor. What it does have, as we have seen, is a verbal reflexive, koL (koND in past tense), which is added to the past participle form of the verb and is inflected for tense and agreement. The verbal

25

Morphological reflexive-marking may be a subcase of lexical reflexivity. See Reinhart and Reuland (1993, p. 662).

71 reflexive makes it obligatory for the reflexive pronoun, tannu, to have a local antecedent. This is illustrated in (103) - (106). (103) shyaamuj raamui tann-annui/*j hoDe-du-koND-a anta heeL-id-a Shyamu Raamu self-ACC hit-PP-REFL.PST-3SM that say-PST-3SM 'Shyamu said that Raamu hit himself' (104) shyamuj raamui tann-annu*i/j hoDe-d-a anta heeL-id-a Shyamu Raamu self-ACC hit-PST-3SM that say-PST-3SM 'Shyamu said that Raamu hit him' (105) raamui tann-annui hogal-i-koND-a Raamu self-ACC praise-PP-REFL.PST-3SM 'Raamu praised himself'

(*hogalida) (*praise-pst-3sm)

(106) raamui tan-agei kaagada bare-du-koND-a Raamu self-DAT letter write-PP-REFL.PST-3SM 'Raamu wrote himself a letter'

(*barda) (*write.pst-3sm)

The most obvious way to interpret koL in Reinhart and Reuland's system is to say that it is a reflexive-marker. So, rather than having a predicate reflexive-marked by its arguments, we can say that koL serves to morphologically reflexive-mark the verb that it is attached to. Recall that Reinhart and Reuland provide two ways for a predicate to be reflexive-marked. Predicates can be reflexivemarked by a SELF anaphor in an argument position, or certain predicates can be lexically reflexive, and thus not require overt reflexive-marking. So, if we follow the idea of reflexive-marking as something that can be marked on the predicate itself and not mediated through one of its arguments, as in the case of lexically reflexive predicates, then we can interpret the verbal reflexive koL in Kannada as a morphological reflexive-marker. That is, while reflexive-marking on the predicate (not mediated through a SELF anaphor) is limited in Reinhart and Reuland's discussion to verbs that are lexically reflexive, in Kannada any verb can be reflexive-

72 marked morphologically via the verbal reflexive. KoL reflexive-marks the predicate it affixes to and therefore allows tannu to be locally bound. Now, if we view koL (and all verbal reflexives) as a morphological reflexive-marker, then the question we have to ask of the RPH is whether all of the uses of this morpheme can be attributed to its role as a reflexive marker. If they cannot, then we have evidence against the RPH. 2.5.2.1 Decausatives and the RPH Consider first the case of decausative predicates. In order for the connection between reflexive morphology and decausative verbs to be brought under the RPH, we would have to say that these verbs are unaccusative (cf. Rosen 1981, Burzio 1981; Grimshaw 1990). Movement of the underlying object to the surface subject position makes the two argument positions of the predicate coreferential. Thus, the predicate is reflexive and reflexive-marking is required by principle B. We encounter three problems, however, when we consider the full range of facts associated with decausative and other unaccusative predicates. First, as pointed out above, in Kannada (ex. 26) reflexive-marking on decausative verbs is optional. If the reflexive-marking arises because of the movement of the object to subject position, it should be obligatory. To get around this objection, we might say that only the decausative verbs with the verbal reflexive are unaccusative and that the non-reflexive-marked decausatives are really unergative. This approach, however, is not tenable. A test for unaccusativity in Kannada is the inability to be embedded under causative (cf. Tirumalesh 1976, Chandrashekar 1994). In (107a), we have a transitive verb embedded under a causative. In (107b) we have an unergative verb embedded

73 under causative. In (107c) we see that an unaccusative verb cannot be embedded under a causative. (107) a.

hari raajuv-inda niitu-vannu hogaL-is-id-a hari raaju-INSTR niitu-ACC praise-CAUS-PST-3SM 'Hari made Raaju praise Niitu'

b.

akka ravi-yannu malag-is-id-aLu older.sister ravi-ACC sleep-CAUS-PST-3SM 'Older sister made Ravi sleep'

c.

*hari elea-gaL-annu biddav-is-id-a hari leaf-PL-ACC fall-CAUS-PST-3SM 'Hari caused the leaves to fall'

For the decausatives, neither the reflexive-marked nor the non-reflexive marked ones can be embedded under causative. (108) a.

b.

*hari kiTiki-yannu tered-is-id-a hari window-ACC open-CAUSE-PST-3SM 'Hari caused the window to open' *hari kiTiki-yannu tere-du-koLL-is-id-a hari window-ACC open-PP-REFL-CAUS-PST-3SM 'Hari caused the window to open'

We can conclude, then, that both classes of decausatives are unaccusative. Similarly, if we turn our attention to the Romance languages, we can see that the claim that non-reflexive-marked decausatives are unergative cannot be maintained. As noted above (ex. 94 and 95), in Italian some decausatives allow the reflexive clitic while others do not. Unaccusatives in Italian take the auxiliary essere (BE), while unergatives take the auxiliary avere (HAVE) (Perlmutter 1978; Burzio 1981). (109) a.

Gianni ha telefonnato Gianni has telephoned 'Gianni has telephoned'

(unergative)

74 b.

Gianni è arrivato Gianni is arrived 'Gianni has arrved'

(unaccusative)

But the decausatives that do not take the reflexive clitic in Italian do take the auxiliary BE, attesting to their unaccusative status. (110) due navi nemiche sono affondate two ships enemy are sunk 'Two enemy ships have sunk' Thus, we cannot say that some decausatives are unaccusative and others are unergative. The RPH cannot explain why some decausatives allow reflexive-marking and others do not. The second problem is that there are many unaccusative verbs which are not decausative and also do not take the reflexive clitic. This is illustrated here in Kannada and Italian: (111) a.

marad-inda elea-gaL-u bidda-vu tree-INSTR leaf-PL-NOM fall.PST-3PLN 'The leaves fell from the tree'

(112) a.

Lo sposo è svenuto the groom is fainted 'The groom fainted'

b.

Gianni è tornato Gianni is returned 'Gianni returned'

If the reflexive clitic arose because the movement of the object of an unaccusative predicate to subject position made the predicate reflexive, we would expect all unaccusatives to be reflexive-marked. But as we can see, they are not.26 26

Rosen (1981), in a framework with substantially different assumptions, also proposed that reflexives and unaccusatives were alike in that the subject and object were connected. Her theory did not predict that all unaccusatives be reflexive-marked,

75 The third problem is that if movement of the object to subject position made a predicate reflexive, we would expect reflexive marking to be obligatory in passive constructions. In fact, verbal reflexives are impossible with passives in Kannada and Italian, suggesting again that the RPH cannot explain the distribution of verbal reflexives: (113) a.

b.

rama tann-inda vancis-al-paTT-a rama self-INSTR deceive-INF-PASS.PST-3SM 'Rama was deceived by himself' *rama tann-inda vancis-koLL-al-paTT-a rama self-INSTR deceive-INF-PASS-PP-REFL.PST-3SM 'Rama was deceived by himself'

(114) *Gianni si è stato affidato Gianni REFL is been entrusted 'Gianni was entrusted to himself' The RPH does not predict the distribution of reflexive morphemes with decausative verbs. Because the RPH requires a predicate to be reflexive in order to license reflexive-marking, it follows that reflexive-marked decausatives are reflexive. The way to view decausatives as reflexive is to say that they are unaccusatives and that unaccusatives are reflexive due to the movement of object to subject. But reflexive marking is not required on all decausatives or on unaccusatives in general. So, the RPH makes the wrong predictions with respect to decausative verbs. To get

however, because she differentiated two types of unaccusatives, plain and retroherent. In a retroherent unaccusative the surface subject retained its status as a direct object (it headed two arcs) while in a plain unaccusative the subject lost its direct object status. It was only the retroherent unaccusatives which were reflexive-marked, reflexivemarking being due to one NP being associated with two grammatical relations. The possibility that there are two types of unaccusatives which are differentiated in this way is not an option under the theoretical assumptions of the RPH.

76 around this objection, we could always say that reflexive-marking on decausatives is lexical and therefore idiosyncratic. However, such a tack makes the theory untestable and therefore unsatisfactory. 2.5.2.2 Possession and the RPH Next consider the cases of reflexive-marking involving possession. Recall that the verb may optionally be reflexive-marked when the subject is coreferential with the possessor of the direct object, as in (11), repeated here: (11)

a.

hari tann-a angi-yannu hari-du-koND-a hari self-GEN shirt-ACC tear-PP-REFL.PST-3SM 'Hari tore his shirt'

b.

hari tann-a angi-yannu hari-d-a hari self-GEN shirt-ACC tear-PST-3SM 'Hari tore his shirt'

The question here for the RPH is whether we can view sentences like (11a) as reflexive since they are reflexive-marked. The RPH predicts that since (11a) is reflexive marked, it is reflexive. But if a reflexive verb is one with two coindexed arguments, then (11a) is not reflexive since the two coreferential elements (the subject and the possessor of the object) are not arguments of the same predicate. So, in order to maintain the RPH, we would have to show that the possessor of the object is an argument of the verb at the point in the derivation when the reflexivity of predicates is determined, presumably LF. Two possibilities present themselves. First, we might say that there is an operation of LF noun incorporation (Sadock 1980, 1991; Baker 1991) and that the stranded possessor is viewed as an argument of the complex predicate formed as a consequence of the incorporation. A second hypothesis is that there is an LF

77 possessor ascension operation which takes the possessor of the object and makes it an oblique argument of the verb.27 Both of these alternatives will be shown below to be untenable and so the RPH cannot account for the possessive reflexive facts. 2.5.2.2.1 LF Noun Incorporation One way in which sentences like (11a) can be viewed as reflexive is if there is an LF operation raising the object into the verb and stranding the possessor. As a consequence of the raising, a complex predicate of the N and the V is formed. Since the possessor is an argument of the N (before raising), when the N raises into the V, the V inherits the arguments of N. The possessor thus becomes an argument of the main verb (cf. Baker 1991 p.101). Since the possessor (now an argument) is coreferential with the subject, the verb forms a reflexive predicate and reflexivemarking is required. Reflexive-marking, which reflects the movement, is optional under these conditions because LF Noun Incorporation is optional. The incorporation is illustrated below. (115) a.

s-structure:

VP NP NP

V N

tanna angiyannu haridukoNDa

27

An operation of Possessor Ascension would violate the Projection Principle (Chomsky 1981), which prohibits altering lexically specified information (like argument-structure) in the course of a derivation. Similarly, the Projection Principle will be violated if noun incorporation makes an argument of the noun into an argument of the verb. This alone suggests that neither option is viable for the RPH. We will ignore this issue for the moment, giving the RPH the benefit of the doubt.

78 b.

LF:

VP NP NP tanna

V N t

N

V

angiyannu haridukoNDa

Let us call this hypothesis the LF Noun Incorporation Hypothesis (LFNIH). In order to test the LFNIH, we should look to the nature of noun incorporation in general. Fortunately, Kannada exhibits noun incorporation of inanimate direct objects on the surface. If the semantics of overt noun incorporation structures is the same as the semantics of reflexive-marked sentences like (11a), then we have evidence for the LFNIH and thus for the RPH. However, if the semantics of the two constructions is different, then we have evidence against the LFNIH and thus should look for other alternatives to save the RPH. If the meaning of a sentence is determined on the basis of its structure at LF, as is standardly assumed, then the only difference between overt noun incorporation and LF noun incorporation is the point in the derivation at which the incorporation takes place. At LF, the two have the same structure and thus should give rise to the same interpretation.

2.5.2.2.1.1 Properties of Overt Noun Incorporation in Kannada Accusative case-marking obligatorily occurs on direct object NPs in Kannada if the DO is animate, but only optionally if the DO is inanimate. (116) a.

b.

naageendra visˇalaakSiy-annu maduveyaa-d-a Naageendra Vishalaksi-ACC marry-PST-3SM 'Nagendra married Vishalakshi' *naageendra visˇalaakSi maduveyaa-d-a

79 c.

hari pustakav-annu oodi-d-a hari book-ACC read-PST-3SM 'hari read the book'

d.

hari pustaka oodi-d-a hari book read-PST-3SM 'hari read a/the book'

The interpretation of non-casemarked DOs is broader than that of case-marked DOs. (116d) can mean either "Hari read a/the book" or "Hari engaged in book-reading," but (116c) can only mean "Hari read the book" and cannot mean "Hari engaged in bookreading." In other words, overt case-marking limits the interpretation of the sentence. The DO must be both referential and definite when it is case-marked, but can also be non-referential or indefinite when it is not case-marked. Let us suppose that the non-referential reading is due to the incorporation of the N into V. On this view, the non-referential reading can be attributed to the incorporated element's sub-word status (Postal 1969; see also Hopper and Thompson 1984, p. 711; Sapir 1911, p. 267). Several facts support this hypothesis. First, an inanimate DO may be non-case-marked only if it is immediately left-adjacent to the V. If the DO is moved to another position in the sentence, the accusative case-marking is obligatory. (117) a.

b.

chatriy-annu raama aafiisin-alli biTTubiTT-a umbrella-ACC rama office-LOC leave.PST-3SM 'Rama left the umbrella in the office' *chatri ramma aafiisin-alli biTTubiTT-a

The adjacency restriction suggests that non-casemarked DOs may be incorporated into the V. If the non-referential reading is due to the incorporation, and incorporation is

80 only possible from an adjacent position, then we correctly predict that only the referential reading is possible in (117a). Similarly, if the DO is marked with an additional clitic/affix, like plural, inclusive, etc., then the case-marker is obligatory. If only bare stems can be incorporated, then we correctly predict that only the referential interpretation will be possible in (118a). (118) a.

b.

hari pustaka-gaL-annu bare-d-a hari book-PL-ACC write-PST-3SM 'Hari wrote the books' *hari pustaka-gaL bare-d-a

Finally, if a non-casemarked DO is modified by an adjective, only the first reading is possible. This follows again if incorporation does not allow modifier stranding. (119) hari haLe pustaka koND-a hari old book buy.PST-3SM 'Hari bought the/a old book' '*Hari engaged in old book buying' (120) hari haLe baTTe oge-d-a hari old clothing wash-PST-3SM 'Hari washed the old clothing' '*Hari engaged in old clothing washing'

2.5.2.2.1.2 Is there LF Noun Incorporation in Kannada? Assuming that (115) is the structure of noun incorporation in Kannada and that the facts given in the previous section are a consequence of this structure, the question we have to ask is whether there is LF noun incorporation in Kannada. Recall that the RPH claims that reflexive marking is only possible if there are two

81 coreferential arguments. The hypothesis under investigation is that there is LF noun incorporation which causes the possessor to count as an argument of the main verb. So, the LFNIH predicts that in reflexive-marked sentences like (11a) the DO can be interpreted non-referentially. If semantic interpretation is determined on the basis of the LF representation, as is standardly assumed, then the semantic component cannot distinguish s-structure incorporation from LF incorporation. Both give the same input to the interpretive component of the grammar. Thus, sentences with LF incorporation should behave exactly as those with s-structure incorporation as far as semantics is concerned. The prediction is not borne out. The DO in (11a) cannot be interpreted non-referentially. If the non-referential interpretation is due to incorporation of the head N into the V, then the lack of this interpretation indicates that incorporation has not taken place. (11)

a.

hari tann-a angi-yannu hari-du-koND-a hari self-GEN shirt-ACC tear-PP-REFL.PST-3SM 'Hari tore his shirt' *'Hari engaged in self's shirt tearing'

The LFNIH cannot be maintained as a way to salvage the RPH. 2.5.2.2.2 Possessor Ascension A second way to explain sentences like (11a) under the RPH would be to say that there is an LF Possessor Ascension construction. On this view, the possessor raises at LF to become an oblique argument of the main verb. Since the possessor (now an argument) is coreferential with the subject, the predicate is reflexive and thus must be reflexive marked. The optionality of reflexive marking under these conditions

82 is due to the optionality of the Possessor Ascension operation. Let us call this hypothesis the LF Possessor Ascension Hypothesis (LFPAH). Again, we do find possessor ascension in Kannada. Consider the following pair (Amritavalli 1984; 1990). (121) a.

amma mak-kal-a kai ToLe-d-aLu mother child-PL-GEN hand wash-PST-3SF 'Mother washed the children's hands'

b.

amma mak-kal-ige kai ToLe-d-aLu mother child-PL-DAT hand wash-PST-3SF 'Mother washed the children's hands'

Here, the possessor may occur in genitive (121a) or in dative (121b). When the possessor is in dative case, we could say that it has raised into an argument position. Interestingly, when the possessor is a reflexive pronoun, it may not occur in dative case: (122) a.

amma tanna kai ToLe-du-koND-aLu mother self-GEN hand wash-PP-REFL.PST-3SF 'Mother washed her hands'

b.

*amma tan-age kai ToLe-du-koND-aLu mother self-DAT hand wash-PP-REFL.PST-3SF 'Mother washed her hands'

The fact that the possessor may not raise overtly when it is a reflexive pronoun casts doubt on the possibility that it raises at LF to become a coargument. Putting this aside, there is nothing in the LFPAH which can predict the interpretive differences between the sentences in which LF possessor ascension has applied and those in which it has not. Recall that the sentences with coreference between the subject and the possessor of the object and with reflexive-marking have an external cause interpretation while those without reflexive-marking do not. If the LFPAH is

83 correct, we would expect this interpretive difference to be related to the syntactic structure of the sentence. But the interpretation of non-reflexive sentences with a dative possessor does not correspond to the interpretation of the reflexive-marked sentences with a genitive possessor. Consider the following sentences: (123) a.

b.

(124) a.

b.

hari raaju-ge kannu-gaL-annu tere-d-a hari raaju-DAT eye-PL-ACC open-PST-3SM 'hari opened raaju's eyes' hari tann-a kannu-gaL-annu tere-du-koND-a hari self-GEN eye-PL-ACC open-PP-REFL.PST-3SM 'hari opened his eyes' hari raaju-ge angiy-annu hari-d-a hari raaju-DAT shirt-ACC tear-PST-3SM 'hari tore raaju's shirt' hari tann-a angiy-annu hari-du-koND-a hari self-GEN shirt-ACC tear-PP-REFL.PST-3SM 'hari tore his shirt'

In (123a-b), the interpretations coincide. In both, the event is portrayed as though the eyes and the opener were separate. In (123a), we get this interpretation naturally because Hari cannot control Raaju's eyes internally. In (124a-b), however, the interpretations are different. In (124a) we have an agentive interpretation in which Hari deliberately tore Raaju's shirt. But in (124b) we only have the interpretation in which the shirt was torn accidentally via some external mediation like getting caught on a nail. If the reflexive marking arises as a result of the possessor raising into an oblique argument position (making the predicate reflexive), then the fact that we get an external cause interpretation when the verb is reflexive-marked should fall out from the position of the possessor. But this position does not give rise to parallel

84 interpretations in reflexive and non-reflexive sentences. In particular, the role of the subject is different in (124a) and (124b). The connection between external causation and reflexive-marking in possessive constructions would seem to be unrelated to the part of the theory that gives rise to the reflexive-marking in the first place. In order to account for the reflexive-marking of the verb under coreference between the subject and possessor of the object, we would have to adopt the LFPAH, and add an additional stipulation to the effect that reflexive-marking adds external causation to the meaning of the sentence and a further stipulation that blocks reflexive pronouns from raising in the overt syntax. Recall from section 2 that an instrumental cause adjunct is only licensed in sentences like (19-21), when the verbal reflexive is present. Under the LFPAH, since the presence of reflexive-marking is dependent on the possessor ascension, and since the instrument is apparently dependent upon the reflexive marking, we must conclude that the reason the instrument is only licensed when the reflexive-marking is present has something to do with the possessor ascension. But there is no theory that can relate the presence of an adjunct to whether or not a rule of possessor raising has applied. It is well known that languages exhibit argument-adjunct asymmetries when it comes to interrogation. Arguments may be questioned more easily than adjuncts. One crucial component of the LFPAH is that the position of a dative possessor is an argument position. It is this fact which allows the reflexive-marked predicate to be considered reflexive.

85 If the position of a dative possessor is an argument position, it should be able to be questioned, just as a subject or an indirect object with dative case can be questioned, but it cannot: (125) a.

sudhaama yaar-ige naaLe kaalceND-annu koD-al-idd-aane sudhama who-DAT tomorrow football-ACC give-INF-BE.NPST -3SM 'Who is Sudhama going to give the football to tomorrow?'

b.

yaar-ige Tayfaayd band-id-e who-DAT typhoid come-be.NPST-3SN 'Who has typhoid?'

c.

*amma yaar-ige kai-gaL-annu ToLe-d-aLu mother who-DAT hand-PL-ACC wash-PST-3SF 'Whose hands did mother wash?'

In (125a) we see that indirect objects can be questioned. In (125b) we see that dative subjects can be questioned. In (125c), however, we see that a dative possessor cannot be questioned. If possessor ascension moves the possessor into the position of an indirect object, we would expect to be able to question it. Since we cannot, we can conclude that the position of the dative possessor is not an argument position. So, even if there were an LF possessor ascension operation in the reflexive-marked sentences in question, this operation would not be to an argument position and thus would not make the predicate reflexive and would not license the reflexive-marking. 2.5.3 Conclusion In this section, I have argued against two strictly syntactic accounts of reflexive morphology. The hypothesis that verbal reflexives indicate that an underlyingly transitive verb is intransitive on the surface (the SIH) cannot account for the fact that many verbal reflexives occur with transitive verbs and that many

86 detransitive verbs do not occur with verbal reflexives. Nor can treating verbal reflexives as an indicator that a predicate is reflexive (the RPH) account for the distribution of reflexive-marking on detransitive verbs. The RPH also cannot predict the appearance of reflexive-marking when the subject and possessor of the object are coreferential. We must discount the general hypothesis that reflexive-marking is determined on the basis of syntactic structure alone.

2.6

Consequences of the Mismatch Hypothesis The distribution of verbal reflexives is not governed by uniquely syntactic

or semantic conditions. Instead, these morphemes are the morphosyntactic instantiations of a mismatch between the thematic and aspectual tiers in argumentstructure. A consequence of this view is that reflexive morphology arises from neither strictly semantic nor strictly syntactic properties. The verbal reflexive reflects a correspondence between lexical-syntactic structure and surface syntactic structure. This view captures the fact that there is not one semantic structure that gives rise to reflexive morphemes. Similarly, the fact that reflexive morphology can occur with a range of syntactic structures is predicted by this theory. In this way, the theory provides evidence for a dimension of representation that is independent of both semantic and syntactic structure and which interacts with both of these other levels of representation. This view also makes available the possibility that other morphemes do not have semantic content but rather represent that correspondence from one domain of linguistic structure to another. In this way, we can view the theory of reflexive

87 morphemes presented here as shedding light on the questions posed by Carter (1976) and Talmy (1985) as to what is a possible lexical item. From this perspective, we have found that lexical elements can express correspondences between any dimensions of linguistic structure in addition to expressing correspondences between meaning and phonological structure. Since the reflexive morpheme provides a mapping from form to structure and not directly from form to meaning, we have evidence for the possibility in general that lexical entries can refer to complex linguistic structures and are not limited to 'pure' semantics. Further, since the licensing of reflexive morphology is not due to strictly syntactic factors, we have evidence that the theory of reflexive morphology is independent of the theory of anaphora, which is syntactic. The interaction of the argument-structure of reflexivity with the syntactic theory of anaphora will be taken up in Chapter 4. One area where we can see that argument-structure representations are independent of syntactic representations is the transitivity of reflexive sentences. As noted above, languages vary with respect to whether they treat reflexive verbs as transitive or intransitive, i.e., whether they allow the reflexive pronoun to cooccur with the verbal reflexive. Languages like Russian systematically treat reflexive verbs as intransitive, while languages like Kannada appear to allow reflexive verbs to be transitive or intransitive. (126) a.

On zastrelil-sja he shot-REFL 'He shot himself'

b.

On zastrelil sebja he shot self 'He shot himself'

88 c. (127) a.

b.

*On zastrelil-sja sebja hari hoDe-du-koND-a hari hit-PP-REFL.PST-3SM 'hari hit himself' hari tann-annu hoDe-du-koND-a hari self-ACC hit-PP-REFL.PST-3SM 'Hari hit himself'

If the representation for verbal reflexives is universal, as I am proposing, then the expression of arguments must be independent of the representation that gives rise to reflexive morphology. The representation in (50), repeated here, can be interpreted by a language as projecting two arguments or one. (50)

a.

(x (y))

thematic tier

(1 (2))

aspectual tier

In a language like Russian, this structure is interpreted as only projecting one argument into the syntax, while in a language like Kannada, it is interpreted as projecting two. Presumably, the way a language interprets these structures will have to do with the interface between syntactic representations and argument-structure representations. The examination of this interface will be left for future research. The morphological inventory of a language will play a role in the interpretation of the structure in (51), repeated here. (51)

(x (y...)) (1... n ...)

thematic tier aspectual tier

In West Greenlandic (Sadock 1980), we can find evidence that this structure plays a role in the expression of arguments, but there is no morpheme that corresponds to this structure. In this language, transitive verbs show agreement with both the subject and

89 the object, while intransitive verbs only agree with the subject. Interestingly, a sentence like (128), with intransitive agreement morphology, shows a systematic ambiguity between an inchoative/decausative interpretation and a reflexive interpretation. (128) piniartoq toquppoq hunter-(ABS) kill-INDIC-3SG 'The hunter got killed' OR 'The hunter killed himself' West Greenlandic is therefore a language that shows the same ambiguity between reflexive and inchoative that was discussed above, though without the accompanying morpheme. We have evidence for the argument-structure representation in (51) in West Greenlandic without a morpheme that instantiates that structure. Interestingly, the expression of an anaphoric object is possible in sentences like (128) in West Greenlandic, but only in an oblique case. The verb retains its intransitive status here even though the object is expressed. (cf. Postal 1977). (129) Imminut takuvoq self-ALL see-INDIC-3SG 'He saw himself' The structure in (51) can be play an explanatory role in the theory of a given language even if there is no morpheme that represents it. Similarly, we expect to find languages in which the structure in (51) does not play any role at all. English appears to be such a language. In English, we have neither a decausative morpheme nor a verbal reflexive and we have no evidence that reflexives and decausatives are treated as equivalent with respect to any other linguistic phenomena. Because the availability of the structure in (51) is determined by the lexicon (i.e., by whether a morpheme instantiating that structure exists in the morphological inventory), the

90 limited variation found with respect to reflexive morphology can essentially be reduced to variation in the lexicon. Some further cross-linguistic variation with respect to reflexive morphemes will be discussed in the next chapter.

Chapter 3 CROSS-LINGUISTIC VARIATION OF REFLEXIVE MORPHEMES

2.1 Introduction Verbal reflexives are best understood as indicating that the most prominent element on the aspectual tier of argument structure is unlinked to something on the thematic tier of argument structure. Crucial to this analysis are certain crosslinguistically stable properties of verbal reflexives. In addition to these stable properties, there is some variation with respect to the uses of verbal reflexives. Many languages use the same morpheme for antipassive constructions as they use for reflexive constructions. Middles are often represented by the same morpheme as reflexives, and impersonal sentences commonly mark the verb with the reflexivemarker. In order for the theory presented in the previous chapter to be explanatorily adequate, it should be able to predict the cross-linguistically variable properties of verbal reflexives as well as it can predict the cross-linguistically stable ones. In this chapter, I show how the Mismatch Hypothesis can be extended to account for the close connection between reflexives and antipassives, middles, and impersonals. 2.2 Connecting Antipassives and Reflexives This section considers the relationship between antipassive morphemes (Silverstein 1976) and verbal reflexives. In an antipassive construction, an NP that normally appears as a direct object occurs as an oblique NP. This "demotion" is

91

92 typically accompanied by some morphological change to the verb. In languages that distinguish transitive and intransitive verbs on the basis of the case-marking of the subject or agreement on the verb, antipassives pattern with intransitives (Dixon 1994). Antipassive is illustrated in the (b) sentences below in Diyari, Kalkatungu, Lithuanian, West Greenlandic, Udmurt and Yidiny. (130) Diyari (Austin 1981) a. ngaTu nana wila kalka-yi 1SG.ERG 3SG.F.ABS woman-(ABS) wait.for-PRES 'I wait for the woman' b.

ngani kalka-tadi-yi nangkangu wila-ni 1SG.ABS wait.for-AP-PRES 3SG.F.LOC woman-LOC 'I wait for the woman'

(131) Kalkatungu (Blake 1994) a. marapai-thu rumpa-mi ithirr matyamirla-thu woman-ERG grind-FUT seed-(NOM) grindstone-ERG 'the woman will grind the seed with the grindstone' b.

marapai rumpa-yi-mi ithirr-ku matyamirla-thu woman-(NOM) grind-AP-FUT seed-DAT grindstone-ERG 'the woman will grind seed with the grindstone'

(132) Lithuanian (Geniusˇiene 1987) a. Petr-as svaido akmen-is peter-NOM throws stone-ACC.PL 'Peter is throwing stones' b.

petr-as svaido-si akmen-imis peter-NOM throws-AP stone-INSTR.PL 'Peter is throwing stones'

(133) West Greenlandic (Bittner 1987) a. Jaaku-p ujarak tigu-a-a Jacob-ERG stone-(ABS) take-TR.INDIC-3SGE/3SGA 'Jacob took stone' b.

Jaaku

ujaqqa-mik tigu-si-vu-q

93 jacob-(ABS) stone-INSTR take-AP-INTR.INDIC-3SGA 'Jacob took stone' (134) Udmurt (Geniusˇienne 1987) a. nil gu kop-a girl-(NOM) hole-(ABS) dig-PRES.3SG 'the girl is digging a hole' b.

nil gu-in kopa-sk-e girl-(NOM) hole-LOC dig-AP-PRES.3SG 'the girl is digging a hole'

(135) Yidiny (Dixon 1977) a. waguda-nggu gudaaga wawaa-l man-ERG dog-(ABS) see-PST 'the man saw the dog' b.

waguuda wawaa-diny-u guda-gala man-(ABS) see-AP-PST dog-DAT 'the man saw the dog'

What is interesting about languages with antipassive constructions is that it is extremely common for languages with both reflexive and antipassive verbal morphology to use the same morpheme in both cases. Of the six languages mentioned above, four show this pattern (Diyari, Yidiny, Lithuanian and Udmurt):28 (136) Diyari ngani muduwa-tadi-yi 1S scratch-REFL-PRES 'I scratch myself' (137) Lithuanian vaik-as su-si-zˇeide child-NOM PERF-REFL-hurt 28

In the Australian aboriginal languages, only the Pama-Nyungan languages have antipassive morphemes. Of the 13 that do have antipassive, the majority use the same morpheme for antipassive and reflexive. Some have an antipassive morpheme but no reflexive marker. Warungu and Kalkatungu and Dyirbal use a different morpheme for each. (Angela Terrell, personal communication)

94 'the child hurt himself' (138) Udmurt Anaj disa-sk-e mother-(NOM) dress-REFL-PRES.3SG 'Mother dresses herself' (139) Yidiny ngayu banggaal-da gundaa-diny-u I-(ABS) axe-INSTR cut-REFL-PST 'I cut myself with an axe' West Greenlandic has no explicit reflexive verbal morphology. Reflexivity is indicated on the verb only by using the intransitive agreement paradigm in which the verb agrees only with the subject: (140) piniartoq (immi-nut) toq-up-poq hunter-(ABS) (self-ALL) kill-INDIC-3SG 'The hunter killed himself' Kalkatungu does have a verbal reflexive, but it is not the same morpheme as the antipassive marker: (141) a.

b.

marapai karri-ti-mi thupu-ngku woman.NOM wash-REFL-FUT soap-ERG 'the woman will wash herself with soap marapai rumpa-yi-mi ithirr-ku matyamirla-thu woman-(NOM) grind-AP-FUT seed-DAT grindstone-ERG 'the woman will grind seed with the grindstone'

What properties, then do reflexive and antipassive constructions share such that they are treated as the same by unrelated languages like Lithuanian and Diyari? Further, given this similarity, what enables languages that have both functions to dissociate the two? I will argue here that, like reflexive constructions, the antipassive construction represents a mismatch between the thematic and aspectual tiers in argument structure, but the mismatch represented by antipassive is not the

95 same as that represented by reflexive. Thus, languages with one morpheme subsuming both functions have opted for one form to indicate mismatch in general. In languages with different forms, the reflexive form signals the kind of mismatch discussed in the previous chapter, while the antipassive form signals the kind of mismatch to be discussed below. Finally, many languages have no antipassive construction at all but still have a reflexive morpheme with the properties discussed in the previous chapter. 2.2.1 The semantics of antipassives Bittner (1987) argues that antipassive in West Greenlandic affects more than just the transitivity of the verb it applies to. In particular, she shows that antipassive has aspectual consequences, such as changing a perfective verb into an imperfective one, or adding an inceptive or frequentive components to the interpretation. For example, in (142), the phrase "every day" is only possible with the antipassive and not with the simple transitive, indicating that the antipassive supplies a frequentive interpretation. (142) a.

b.

*ullut tamaasa jaaku malig-pa-a days all jacob-(ABS) follow-TR.INDIC-3SG.ERG/3SG.ABS 'He followed Jacob every day' ullut tamaasa jaaku-mik malig-si-pu-q days all jacob-INSTR follow-AP-INTR.INDIC-3SG.ABS 'He followed Jacob every day'

Bittner also shows that some verbs which can be interpreted as either accomplishments or activities (Vendler 1957) in their transitive use can only be interpreted as activities in their antipassive use:

96 (143) a.

b.

jaaku-p illu taa-nna sana-pa-a Jacob-ERG house-(ABS) this-SG.ABS build-TR.INDIC3SG.ERG/3SG.ABS 'Jacob built/is building this house' (may but need not have finished) jaaku illu-mik taa-ssuminnga sana-ø-pu-q Jacob-(ABS) house-INSTR this-SG.INSTR build-AP-INTR.INDIC3SG.ABS 'Jacob is building this house' (has not finished it yet)

Similarly, in Lithuanian, the difference between (132a) and (132b), repeated here, lies in aspectual interpretation: (132) a.

b.

Petr-as svaido akmen-is peter-NOM throws stone-ACC.PL 'Peter is throwing stones' petr-as svaido-si akmen-imis peter-NOM throws-AP stone-INSTR.PL 'Peter is throwing stones'

In (132a) we have the interpretation in which Peter is in the process of throwing stones now, while (132b) describes a property of Peter, i.e., that he generally throws stones. Kalkatungu antipassives also have aspectual consequences, signalling ongoing or uncompleted activity or characteristic activity (Blake 1987). In (131), repeated here, the transitive indicates that there is a specified amount of seed to be ground, while the antipassive indicates that the woman will engage in seed-grinding: (131) a.

b.

marapai-thu rumpa-mi ithirr matyamirla-thu woman-ERG grind-FUT seed-(NOM) grindstone-ERG 'the woman will grind the seed with the grindstone' marapai rumpa-yi-mi ithirr-ku matyamirla-thu woman-(NOM) grind-AP-FUT seed-DAT grindstone-ERG 'the woman will grind seed with the grindstone'

97 What these interpretations have in common is the unboundedness (see Tenny 1987; Jackendoff 1991) of the events portrayed by the sentences. On this view, what antipassive morphemes do is unbound an event. Given the theory of argument structure discussed in the previous chapter, we expect unbounding operations to be operations on argument structure, specifically at the aspectual level. 2.2.2 Boundedness in argument structure Tenny (1987) shows that a verb's direct internal argument determines the aspectual properties of that verb. This argument 'measures out' the event described by the verb: the internal argument delimits or bounds the event and the termination of the event corresponds to the termination or consumption of the referent of the direct internal argument. So, in a sentence like (144), the end of the performing event corresponds to the end of the play. (144) The troupe performed a play One way that Tenny illustrates this idea is through time adverbials like 'halfway'. So, in both of the verb phrases in (145), we see that halfway through the play is halfway through the performance. (145) a. b.

The troupe performed a play halfway The troupe performed half a play

In the two dimensional theory of argument structure described in the previous chapter, the element that measures out the event is the element that is linked to the least prominent argument on both tiers. To illustrate, let us consider the semantics of a verb like perform. In LCS, we can assume that there are two open positions that project into the thematic tier: one for 'the troupe' and one for 'the play'. The event structure for perform involves two subevents. As Pustejovsky (1991) ,

98 among others, shows, adverbials like almost can distinguish verbs with two subevents from those with only one subevent. (146) a b.

The troupe almost performed a play John almost ran

(146a) is ambiguous between a reading in which the troupe almost began the performance and a reading in which the troupe almost finished the performance. The first reading arises if the adverbial modifies the first subevent and the second reading if the adverbial modifies the second subevent. In contrast, (146b) is not ambiguous; it only means that John almost began to run and thus has only one subevent. Thus, the event structure for perform is something like:29 (147) ((1 DO-SOMETHING) (2 EXIST)) The first subevent indicates that some action took place, while the second subevent indicates that the play has been completed. This event structure representation projects two arguments onto the aspectual tier. We thus have the full argument structure representation given in (148). (148)

(x (y))

thematic tier (x=troupe, y=play)

(1 (2))

aspectual tier (1=troupe, 2=play)

Because the play is linked to the least prominent element on both tiers, it is the measure of the event. More generally, it is the argument that is linked to the least prominent element on both the aspectual and thematic tiers that measures out the

29

The precise details of the event structure may be different from (147), but any changes will not affect the analysis being presented. What is important is simply that the element associated with the performer is the most prominent aspectual element and that the element associated with the play is the least prominent aspectual element.

99 event. If antipassive involves the unbounding of an event, then we expect the representation of antipassives to involve breaking the connection between the least prominent elements on the two tiers. 2.2.3 The argument structure of antipassives Returning to antipassive, we can consider antipassive to be an operation just like passive in that it marks an argument as suppressed. However, while passive suppresses an external argument, antipassive supresses an internal argument. By suppressing the internal argument, we no longer have an element linked to the least prominent element on both tiers, and the event consequently loses its delimiter. For this reason, we get an unbounded interpretation when we have the antipassive morpheme. Consider the representation of the Kalkatungu sentence in (131), repeated here. (131) a.

b.

marapai-thu rumpa-mi ithirr matyamirla-thu woman-ERG grind-FUT seed-(NOM) grindstone-ERG 'the woman will grind the seed with the grindstone' marapai rumpa-yi-mi ithirr-ku matyamirla-thu woman-(NOM) grind-AP-FUT seed-DAT grindstone-ERG 'the woman will grind seed with the grindstone'

Here, the LCS representation of "grind" is (149). (149) [CAUSE [x], [BE [y], [GROUND]] The event structure of "grind" will be: (150) ((1 act-on 2) (3 change)) We thus have the argument structure given in (151).

100 (151)

(x

(y))

(1 (2 (3)))

thematic tier (x=woman, y=seed) aspectual tier (1=woman, 2=seed)

There is a bounded interpretation here because the argument associated with "seed" is the least prominent element on both tiers of argument structure. Applying the antipassive operation to the structure in (151), we get the argument structure in (152), which corresponds to the sentence (131b): (152)

(x

(y-ø))

(1 (2 (3))) The suppression of y on the thematic tier causes that element to be unassociated with the internal argument position in the syntax, and so it must be realized as an oblique. Further, by delinking y from 2, the NP associated with those elements is no longer associated with the least prominent argument in both tiers and thus can no longer delimit the event. Thus the argument structure operation that suppresses the least prominent thematic element accounts for both properties of the antipassive construction: the verb is intransitive and the event is interpreted as unbounded. In general, then, we can say that the antipassive morpheme arises when an internal argument has been suppressed, resulting in an unlinked element at the bottom (right edge) of the aspectual tier. In this way, antipassive is like reflexive in that it indicates that there is an unlinked element on the aspectual tier. The two are different in terms of the prominence of of the unlinked element. It is the most prominent element that is unlinked in reflexive constructions while it is the least prominent element that is unlinked in antipassive constructions. If a language uses the same morpheme for both antipassive and reflexive, like Lithuanian or Diyari, it is because that morpheme indicates mismatch in general. In contrast, if a language uses one

101 morpheme for antipassive and another for reflexive, like Kalkatungu, it is because there is one morpheme for each kind of mismatch. Some languages, like Kannada, will have one morpheme for reflexive and none for antipassive, marking only when the most prominent aspectual element is unlinked. Finally, some languages, like K'ekchi, will have an antipassive morpheme but no verbal reflexive, marking only whent he least prominent aspectual element is unlinked. In this way, the variation among languages is reduced to lexical differences. That is, some languages will have mismatch morphemes and others will not. Among the ones that do have mismatch morphemes, some will have generalized mismatch morphemes which indicate any kind of mismatch, while other languages will have either one or two mismatch indicators. It is simply the morphological inventory of a language which determines whether reflexive and antipassive are marked in the same way. Note that the account here differs from that in Postal (1977), where it is argued that reflexives and antipassives are the same in that they involved detransitivization. As we saw in the previous chapter, not all reflexive verbs are intransitive. Further, by treating antipassive and reflexive as the same, Postal predicts that the semantics of the two should be the same.30 If the unbounded interpretation discussed above is due to the antipassive operation, then we would expect reflexives to be unbounded in interpretation as well. Of course, not all reflexive sentences are unbounded, so Postal's account would require additional stipulations in order to account for the lack of unbounded interpretation in some reflexive sentences. In my account of the close relationship between antipassive and reflexive, the

30

This point depends on Postal having a theory of semantics which is sensitive to syntactic structure.

102 unboundedness of antipassive and its similarity to reflexive are captured without requiring the two constructions to be due to exactly the same structure. Finally, Postal's account, by equating antipassive and reflexive, cannot explain the existence of languages which have two different morphemes for antipassive and reflexive. Further support for this analysis of the relationship between antipassive and reflexive comes from dative subjects. Interestingly, languages like Kannada (which have one form for reflexive and none for antipassive) do not allow the verbal reflexive to occur on verbs that take a dative subject (Bhat 1991). (153) a.

avan-ige beva-t-itu he-DAT sweat-PST-3SN 'He sweated'

b.

*avan-ige beva-tu-koND-itu he-DAT sweat-PP-REFL-3SN

c.

avan-u beva-t-a he-NOM sweat-PST-3SM 'He sweated'

d.

avan-u beva-tu-koND-a he-NOM sweat-PP-REFL-3SM 'He sweated'

(154) a.

b.

Hari-ge tann-a mak-kaL-annu serit-idd-aru Hari-DAT self-GEN child-PL-ACC fond-be.PST-3PL 'Hari was fond of his children' *Hari-ge tann-a mak-kaL-annu seri-koND-idd-aru Hari-DAT self-GEN child-PL-ACC fond.PP-REFL.PP-be.PST-3PL

103 In contrast, languages like Lithuanian (which have one morpheme indicating both reflexive and antipassive)31 seem to require the reflexive marker on verbs taking a dative subject. (155) a.

asˇ ne dirbu I-NOM not work 'I don't work'

b.

*asˇ ne se-dirbu I-NOM not REFL-work

c.

man ne se-dirba I-DAT not REFL-work 'I don't feel like working'

d.

*man ne dirba I-DAT not work

If the dative subject construction arises due to a mismatch in argument structure that is different from the one that gives rise to the reflexive marker in languages like Kannada, then we can understand why the reflexive marker is used in languages like Lithuanian. Since the reflexive marker simply means mismatch in Lithuanian, we expect it to arise with all mismatches and thus with the dative-subject construction. But if the dative subject construction involves a mismatch that does not involve the most prominent aspectual element being unlinked, then the reflexive marker will not occur in this construction in Kannada because such marking only arises in Kannada when the most prominent aspectual element is unlinked.

31

Because most of the languages with antipassive constructions are ergative, it is difficult to find languages with both antipassive and dative subjects.

2.3 Connecting Middles and Reflexives It is well known that many languages use the same morpheme to indicate middle and reflexive. Languages that show this property include Tatar, Italian and Polish: (156) Tatar (Geniusˇiene 1987) a. ul jasˇer-el-de he.ABS hide-REFL-PST.3SG 'He hides himself' b.

kran bory-l-a tap-(ABS) turn-REFL-PRES.3SG 'the tap turns'

(157) Italian (Napoli 1979) a. Maria si guarda Maria REFL watches 'Maria watches herself' b.

La pellice si vendono bene d'autunno the furs REFL sell well in-autumn 'The furs sell well in the autumn'

(158) Polish a. Jan bija sie Jan beats REFL 'Jan beats himself' b.

Ksiazk-a czyta sie przyjemnie book-NOM reads REFL pleasantly 'The book reads pleasantly'

Why do some languages that use the reflexive morpheme for middles and some not (like Kannada and Lithuanian)? If I am correct in drawing the conclusion that the distribution of reflexive morphemes is universally governed by a mismatch in argument structure, then the following question arises: how do we characterize the

104

105 conditions on middle markers and reflexive markers with enough similarity to capture the langauges that treat the two as the same but with enough difference to capture the languages that treat them differently. The possibility that the middle morpheme represents a different kind of mismatch from that represented by the reflexive (as claimed for antipassive) does not seem tenable since Lithuanian (which, as we argued above, uses the reflexive marker to indicate mismatch in general) does not use the reflexive for middles. This raises the possibility that the middle is a subcase of the reflexive mismatch. On this view, middles do arise when the most prominent element on the aspectual tier is unlinked, but there is additional information in middles that distinguishes them from the other meanings that give rise to this mismatch. Let us assume, then, that middles give rise to a mismatch of the same sort that arises in reflexives and inchoatives (see Hale and Keyser 1986). It seems clear that a sentence like (159a) shares a good deal of lexical structure with a sentences like (159b). (159) a. b.

The vase broke (*by John) This vase breaks easily (*by John)

In both cases, we have a surface subject which would be the object of the transitive variant of the verb. Hale and Keyser (1986) show that both decausatives and middles require 'change of state' to be part of the lexical entries to which they apply. The question is simply what element of meaning the middle has that the decausative does not. From Hale and Keyser's perspective, the two are the same. However, since some languages treat decausatives and middles differently from a morphological perspective, there must be some difference.

106 The difference lies in the 'potential' or 'generic' interpretation involved in the middles. A middle does not portray an event that has already taken place but rather indicates the potentiality for an event to take place. We must assume that this potentiality/genericity is part of the representation that is relevant to the projection of arguments in to the syntax. In the case of languages that use the same element for reflexives and middles, the representation of potentiality is ignored. In languages that have a different way to mark middles and reflexives, the reflexive indicates that there is an unlinked element somewhere on the aspectual tier, while the middle indicates that there is an unlinked element on the aspectual tier and that potentiality is also represented. In this way, the middle is simply a subcase of the reflexive. Some languages will mark this subcase while other will not. The representations for reflexives and middles are given in (160a) and (160b), respectively: (160) a.

(x (y...)) (1... n ...)

b.

(x (y...)) (1... n ...) [generic]

thematic tier aspectual tier thematic tier aspectual tier additional info. tier

Where the [generic] feature is represented remains to be discovered, but for the moment, let us assume that there is an additional tier in the lexical representation in which that feature is relevant. What is important here is simply that the [generic] feature is the part of the representation that distinguishes middles from reflexives and decausatives.

107

2.4 Connecting Impersonals and Reflexives In addition to being used for antipassives and middles, reflexive morphemes are often used in impersonal constructions. Languages like this include Italian and Tatar. (161) Italian a. Maria si guarda Maria REFL watches 'Maria watches herself' b.

(162) Tatar a.

b.

Domani si discuterà le due proposte di legge tomorrow REFL discuss.FUT.3SG the two bills of law 'Tomorrow they will discuss two bills'

ul jasˇer-el-de he.ABS hide-REFL-PST.3SG 'He hides himself' aktanys avyly-na kaisy jul-dan bar-yl-a aktanys village-AD what road-ABL go-REFL-PRES.3SG 'By what road does one go to Aktanys village?'

Geniusˇiene (1987) reports an implicational hierarchy in which languages that use the reflexive marker for impersonals also use it for middles and languages that use the reflexive marker for middles also use it for decausatives (as we have seen, decausative is a universal use of the reflexive). Thus, in trying to understand what gives rise to the difference between languages like Italian, which use the reflexive for impersonals, and languages like Kannada which do not, we must incorporate the analysis for middles. Since no language uses the reflexive marker for impersonals but not middles, impersonals must be a subcase of middles. On this view, then, impersonals also represent a mismatch between tiers in argument structure, but have

108 some additional component in the representation which enables them to be dissociated from reflexives. Since we have argued above that some representation of potentiality is necessary to distinguish middles from reflexives, impersonals must also include the representation of potentiality that middles include as well as some further meaning component. (see Cardinaletti and Starke 1994 for discussion of the generic properties of impersonals) As a preliminary hypothesis, let us suppose that what differentiates impersonals from middles is the connection of the 'potential/generic' meaning component to the unlinked element on the aspectual tier. Since it is the unlinked aspectual element which essentially represents the suppressed agent in decausatives, middles and impersonals, and since what makes an impersonal an impersonal is the fact that the agent is generic, linking the generic element responsible for middles to the unlinked aspectual element allows us to see how to represent impersonals. In (163) are the argument structure for reflexives and decausatives (a), middles (b), and impersonals (c). (163) a.

(x (y...)) (1... n ...)

b.

(x (y...)) (1... n ...) [generic]

c.

(x (y...)) (1... n ...) [generic]

thematic tier aspectual tier thematic tier aspectual tier additional info. tier thematic tier aspectual tier additional info. tier

109 The observed implicational relationship is captured in these representations because each representation shares a formal component with the others. The representation of impersonals includes the representation of middles as a subpart and the representation of middles includes the representation of reflexives and decausatives as a subpart. Because all of these representations have an unlinked element on the aspectual tier, some languages will have one morpheme that is used in all three cases (e.g., Tatar, Italian, Spanish). The morphological component of these languages ignores the information that distinguishes middles and impersonals from reflexives and decausatives. Since such information is represented, however, we expect to find languages that are morphologically sensitive to it. Thus, we expect to find languages with one morpheme for (a) and (b) but not (c) (e.g., Chuvash, Polish, Udmurt) since (c) has a formal component that is missing from (a) and (b). We also expect to find languages that have one morpheme for (a), but a different way of encoding each of (b) and (c) (e.g., Kannada, Lithuanian, Selkup), since (b) and (c) are each more specific than (a).

2.5 Conclusions In this chapter we have seen how the theory of reflexive morphemes presented in the previous chapter can be extended to handle certain cross-linguistically variable properties of verbal reflexives. I have argued that antipassives represent a different kind of argument structure mismatch from reflexives. Languages that treat the two as the same simply have one morpheme to indicate mismatch. Middles and impersonals, on the other hand, involve the same mismatch as reflexives. However, there

110

is additional information in middles and impersonals which allows languages to distinguish them from reflexives. The analysis essentially reduces the cross-linguistic variation in reflexive morphology to lexical differences. Different morphemes represent different kinds of structures. The range of uses of a reflexive morpheme in a language is dependent only upon what structures have been lexicalized.

111

Chapter 4 ANTILOCAL REFLEXIVES: AN EPIPHENOMENON

2.1 Introduction A great deal of research in generative grammar has been dedicated to determining the appropriate syntactic conditions on anaphora. One problem which has been unresolved in this research is the existence of anaphors which cannot be locally bound. We know that such elements are anaphors because they require a c-commanding antecedent. Unlike most anaphors, however, they do not permit their antecedents to occur within the local domain. The contrast between languages with antilocal anaphors and those without is illustrated in (164-165) below: (164) Dutch (Everaert 1986) a. *Max haat zich Max hates self 'Max hates himself' b.

hij hoorde mij over zich praten he heard me about self talk 'He heard me talk about him(self)'

c.

Max haat zichzelf Max hates selfself 'Max hates himself'

d.

*hij hoorde mij over zichzelf praten he heard me about self-self talk 'He heard me talk about him'

(165) Chinese (Cole, Hermon and Sung 1990) a. Lisi piping ziji Lisi criticized self 'Lisi criticized himself'

112 b.

Zhangsan renwei Lisi piping ziji Zhangsan thinks Lisi criticized self 'Zhangsan thinks Lisi criticized him/himself'

c.

Lisi piping ta-ziji Lisi criticized him-self 'Lisi criticized himself'

d.

*Zhangsan renwei wo piping ta-ziji Zhangsan thinks I criticized him-self 'Zhangsan thinks I criticized him'

In Dutch we see that local binding of the simplex anaphor zich is not possible while longdistance binding is possible. In Chinese, the simplex anaphor ziji may be locally bound as well as long-distance bound. In both languages, the complex anaphor only permits local binding. An important observation with respect to the form of reflexive pronouns was made by Faltz (1977) who observed that polymorphemic reflexives may not be longdistance bound while long-distance reflexives are always monomorphemic. Pica (1987) proposed that the determination of domain is related in some way to the morphological form of the reflexive. In particular, Pica proposed that morphologically simplex reflexives are LF clitics, which move to INFL at LF (cf. Battistella 1989, Cole, Hermon and Sung 1990; Cole and Sung 1994). Because they are heads, these reflexives may move through INFL to Comp and then into the next clause, explaining their possible

113

long-distance interpretation.32 A sentence like (166) has the two possible LF representations in (167).33 (166) Lisi zhidao Wangwu xihuan ziji Lisi think Wangwu like self 'Lisi thinks Wangwu likes him/himself' (167) a. b.

[Lisi [zhidao [Wangwu INFL-zijii [xihuan ti]]]] [Lisi INFL-zijii [Wangwu INFL-ti [xihuan ti]]]

In (167a), ziji is locally bound only by Wangwu, while in (167b), ziji is locally bound only by Lisi. The long distance interpretation in (167b) is really an instance of local binding at LF. Under this account, the binding theory remains constant in enforcing the antecedent to be locally related to the reflexive. What varies from language to language is whether the reflexive moves. Whether the reflexive moves is, in turn, explained by the morphological form of the reflexive. The details of how the morphological form of the reflexive relates to the possibility of movement are irrelevant to the task at hand, however. What is relevant is that the antilocality of anaphors like Dutch zich is not obviously explained by this approach. This chapter argues that the antilocality of anaphors like Dutch zich is epiphenomenal. These elements may, in fact, be locally bound. In many cases where

32

One possible objection to this approach is that 'real' reflexive clitics can never be long-distance bound. However, there are two responses to this objection. First, given the theory of verbal reflexives developed in Chapter 2 in which verbal reflexives (including reflexive clitics) are not nominal anaphors but rather indicate a mismatch between tiers in argument structure, we have no reason to expect 'reflexive' clitics to behave like anaphors. Second, Kayne (1989) shows that pronominal clitics in Romance can overtly undergo long head-movement of this sort, providing support for the idea that clitic pronouns can move across clause boundaries. 33

These representations are simplified so as to show only the traces in the base position and in INFL. In order to satisfy the Head-Movement Constraint (Chomsky 1986a), movement must actually be through all heads between the base position and the landing site.

114

local binding obtains, however, independent principles relating to the interface of syntactic, semantic and argument structure representations rule the sentence out. Such restrictions are predicted to hold in all languages. What I will show is that the semantic representation of the antilocal reflexives differs from that of the long-distance reflexives which can be locally bound. Accordingly, the existence of two types of long-distance anaphor, local and antilocal, does not have to be stipulated but rather falls out from independent principles of grammar. The chapter is organized as follows. Section 2.2 shows that so-called antilocal reflexives may be locally bound in certain contexts, placing in doubt the descriptive generalization that these elements have an antilocality requirement. I show that the correct descriptive generalization is that these elements may not be bound by a coargument. Section 2.3 examines Hestvik's (1990) analysis of antilocal reflexives, showing that it misses the descriptive generalization and consequently provides an inaccurate analysis. In Section 2.4, we find that the restriction on coargument antecedents falls out from an interaction of indepenent principles of grammar concerning the lexical expresssion of reflexivity. In particular, I propose a principle of UG stating that semantic reflexivity must be lexically expressed. The coargument restriction is thus due to the absence of lexical reflexivity in the presence of semantic reflexivity. This theory then predicts that the coargument restriction should hold in all languages. I show, however, that there are two types of reflexivity and that those anaphors which may be locally bound do not occur in sentences expressing semantic reflexivity.

115

2.2 Locally bound antilocal reflexives Antilocal reflexives are elements that have been characterized in the past as requiring a binder outside of the local domain (Bhat 1981; Everaert 1986, 1991; Hestvik 1990). Such forms include Kannada tannu, Dutch zich and Norwegian seg. In this section, I will show that this characterization is inaccurate and, as a consequence, any theory based on this characterization inherits the inaccuracy. For concreteness, let us assume the Binding Principles given in (168), with the definition of local domain given in (169): (168) Binding Principles (Chomsky 1986b, p. 166) a. an anaphor is bound in a local domain b. a pronominal is free in a local domain c. an R-expression is free (in the domain of the head of its chain) (169) Local Domain = Complete Functional Complex Complete Functional Complex is "the category in which all the grammatical functions associated with its head are satisfied" (Chomsky 1986b, p. 169) Let us first consider the facts associated with Dutch zich. (170) *Jan verraste zich Jan surprised self 'Jan surprised himself' (171) hij hoorde mij over zich praten he heard me about self talk 'He heard me talk about him(self)' In (170), the CFC is the entire clause and zich is not allowed. In contrast, the minimal CFC for zich in (171) is the embedded clause. In this sentence, only the long-distance interpretation is possible. These facts might lead us to believe that local binding of zich is not allowed. In (172), however, the minimal CFC for the anaphor is also the clause, and local binding is licensed.

116 (172) Max legt het boek achter zich Max put the book behind self 'Max put the book behind him(self)' The PP headed by achter does not have a subject, so it does not qualify as a CFC. The VP also does not have a subject, so it also does not qualify as a CFC. The clause, however, does have a subject and therefore constitutes a CFC; it is the only category (IP) containing the anaphor in which all of the grammatical functions of a head are satisfied. Thus, local binding is possible in (172), and the claim that Dutch anaphors are antilocal is simply false.34 Norwegian has also been taken to have an antilocal reflexive (Hestvik 1990; Reinhart and Reuland 1993). However, we find the same situation here as in Dutch. The Norwegian simplex reflexive, seg, may be bound within the local domain in some cases. Consider the following (Hellan 1988): (173) a.

34

*Jon fortalte om seg Jon told about self 'Jon told about himself'

b.

*Jon foraktet seg Jon despises self 'Jon despises himself'

c.

Jon legger boken bak seg Jon puts book behind self 'Jon puts the book behind himself'

d.

Jon traff noen venner av seg Jon met some friends of self 'Jon met some friends of his'

e.

Jon fortalte om seg og sin nabo Jon told about self and self's neighbor 'Jon told about himself and his neighbor'

Assuming that the subject is generated internal to VP does not change this conclusion. That is, if VP counts as a CFC (because of the subject trace in [Spec, VP]), then zich is locally bound within that VP.

117

In (173a-b), seg does not allow local binding, while in (173c-e), seg may be locally bound. In (173c), seg is governed by the preposition bak. This PP, however, does not have a subject and therefore does not constitute a CFC. The only category containing both the anaphor and a subject is the entire clause. Thus, the local domain is the entire clause and seg is locally bound within it. It is important to see here that the complex anaphor, seg selv, is also possible in this context and that the pronoun, ham, is not (Hestvik 1990): (174) a.

Jon legger boken bak seg selv Jon puts book behind self self 'Jon puts the book behind himself'

b.

*Jon legger boken bak ham Jon puts book behind him 'Jon puts the book behind himself'

Thus, any attempts to say that the PP does constitute a CFC here will run into problems. That is, one might try to explain the grammaticality of (173c) by claiming that the PP headed by bak is a CFC, making the antecedent of seg outside of the local domain. However, in long-distance binding contexts in general, pronouns are possible and complex reflexives are not. (174) shows the opposite pattern, demonstrating that the object of the preposition and the matrix subject are in the same local domain.35 Similarly,

35

Hestvik explains these examples in a somewhat complicated fashion. He claims that the PP is a CFC for elements which must be free at S-structure (pronouns and seg) but not for elements which must be bound at S-structure (segselv). The matrix S is the CFC for the elements which must be bound at S-structure. As a consequence, seg cannot be bound within the PP. At LF, seg raises to INFL where it can be bound by the subject (since it must be bound at LF). The complex anaphor segselv requires S-structure binding and so may be bound by the subject without any complications. The pronoun, ham, (which must be locally free at S-structure and LF) also moves to INFL at LF (since all X° elements do so) where it is illicitly locally bound by the subject. While this account does capture the facts, it requires that the definition of CFC depends on the element being bound (cf. Chomsky 1986; Wexler and Manzini 1987) and that all X° anaphors and pronouns move to INFL at LF. We will examine this account more closely below.

118

in (173d) seg is contained within a PP without a subject and an NP without a subject (i.e., the NP headed by venner). The smallest CFC containing the anaphor is the clause. The anaphor, seg, is therefore locally bound in (173d) as well. The sentence in (173e) makes a minimal pair with the sentence in (173a). In both of these sentences, seg is locally bound. This configuration is ungrammatical in (173a), but grammatical in (173e). In Kannada, antilocality is also only apparent. The simplex anaphor tannu may be locally bound in some situations (Amritavalli 1991): (175) a.

Hari tann-a mak-kaL-annu hoDe-d-a hari self-GEN child-PL-ACC hit-PST-3SM 'Hari hit his (own) children'

b.

raama tann-a yedur-ige ondu kaaLavanna kaND-a raama self-GEN front-DAT one buffalo saw-3SM 'Raama saw a buffalo in front of him(self)'

c.

*Hari tann-annu hoDe-d-a Hari self-ACC hit-PST-3SM 'Hari hit himself'

In (175a), the anaphor is the possessor of the object. Here, there is no potential antecedent inside the NP, so that category cannot be considered a CFC for the anaphor. The only CFC that contains the anaphor is the whole sentence. Thus, tanna is locally bound in (175a). In (175b) the anaphor is the object of a preposition. The PP containing the anaphor, however, does not have a subject and cannot be considered a CFC. The only CFC containing the anaphor is the whole sentence and the anaphor is bound within it. In (175c), the anaphor may not be bound within the clause, even though the clause is the smallest CFC containing it. Thus, it is not the binding theory (at least as it is stated above) which rules out (175c). The restriction on potential antecedents for the Kannada anaphor is not that it cannot be bound within the local domain.

119 2.3 Hestvik (1990): An analysis based on antilocality Hestvik (1990) develops a binding theory in which anaphors may be marked with a feature indicating whether they are to be bound or free at S-structure or at LF. If an element has the feature [+bs], then it may be locally bound at S-structure. If it has the feature [-bs], then it must be locally free at S-structure. The additional feature for LF binding allows Hestvik to explain the behavior of long-distance reflexives. If an anaphor has the feature [+bL], then it must be locally bound at LF, while if it has the feature [-bL], it must be locally free at LF. Hestvik follows Pica (1987) and others in assuming that X˚ reflexives move to INFL at LF. So, if an X˚ reflexive moves out of its own clause at LF and has the feature [+bL], then it will have the appearance of being long-distance bound, exactly as we saw above in Chinese in (167). To illustrate, consider the following: (176) a.

b.

hij hoorde Jan over zich praten he heard Jan about self talk 'He heard Jan talk about him' Zhangsan renwei Lisi piping ziji Zhangsan thinks Lisi criticized self 'Zhangsan thinks Lisi criticized him/himself'

The Dutch sentence in (176a) allows only long-distance binding. This follows from Hestvik's system because the anaphor, zich, is marked with the features [-bs] and [+bL]. Because it is marked [-bs], it cannot be bound by Jan, the only NP which locally ccommands the base position at S-structure. The anaphor moves at LF to the matrix INFL, however, where it can be locally bound by the matrix subject, as illustrated in (177) (abstracting away from V2): (177) [hij hoorde-zichi [Jan INFL-ti over ti praten] The Chinese sentence in (176b) illustrates how the system handles an anaphor with the features [+bs] and [+bL]. Because ziji has the feature [+bs], it may be locally bound at Sstructure. Thus, the embedded subject (Lisi) may be the antecedent for the anaphor.

120

Because ziji also has the feature [+bL], it may optionally move to the matrix INFL at LF where it can be locally bound by the matrix subject, as illustrated in (178): (178) [Lisi INFL-zijii [Wangwu INFL-ti xihuan ti]] As we have seen, Hestvik gives the antilocal reflexives, like Norwegian seg, the features [-bs] and [+bL]. As a consequence, these elements must not be locally bound at S-structure and must be locally bound at LF. Since the feature [-bs] disallows Sstructure binding, the anaphor must move out of its own clause at LF, where it can be locally bound, in accordance with the feature [+bL]. There are two problems with this approach to distinguishing antilocal reflexives from other reflexives. First, it is purely classificatory and has no explanatory value. By positing the existence of a feature distinguishing long-distance reflexives which can be locally bound from those which cannot, Hestvik merely stipulates and does not derive the distinction between the two types of long-distance anaphor. A satisfactory account of these facts will derive the difference between Chinese and Dutch from something other than a stipulation meant only to capture that difference. Second, Hestvik's approach does not make the right generalization. It is not possible to state the binding conditions for 'antilocal' reflexives only in terms of the level of representation at which the binding principles apply. The examples in (172-175) make this second point clear. As we saw in the previous section, the anaphors in Dutch, Norwegian and Kannada can be locally bound at S-structure in certain cases. The claim that these anaphors must be free at S-structure and bound at LF is false. There are both grammatical and ungrammatical examples of these anaphors being bound at S-structure. The appropriate descriptive generalization of this type of anaphor is that it may be locally bound unless its antecedent is an argument of the same predicate as the anaphor itself (Reinhart and Reuland 1993). We will call this the coargument restriction.

121

One possible way for Hestvik to capture the coargument restriction would be to say that the local domain should be characterized by coargumenthood and not by CFC. This solution will only work if we are willing to accept a parameterized theory of binding domains which allows for one domain in Dutch, Norwegian and Kannada but another domain for other languages which do not obey the coargument restriction. That is, we could allow not only for parameterization in terms of the level at which binding applies but also in terms of the binding domain. Such a move deviates from the goal of explaining anaphora with a unified set of principles for all languages. Moreover, the partially overlapping distribution of the simplex and complex anaphors is problematic for such a tack (see (174) ff.). To handle these facts, we would also have to follow Hestvik in parameterizing the binding domain within a particular language for the different anaphors (see footnote 35). For example, we would have to say that the coargument domain was the local domain for seg but the CFC was the local domain for segselv.36 Clearly, such a move is undesirable, taking us even farther away from a unified theory. If we can capture the difference between languages like Dutch and languages like Chinese without parameterizing the binding domains across languages and across anaphors, then we will have a better theory. Such an account of the coargument restriction is developed below. 2.4 Anticoargumenthood instead of antilocality The data in section 2.2 force upon us the question of the correct characterization of the 'antilocal' reflexives. Only when we understand the descriptive generalization can explanation be possible. Reinhart and Reuland (1993) capture the generalization illustrated above in Dutch, Norwegian and Kannada: the reflexives in these

36

This, in fact, is the analysis proposed by Dalrymple (1990).

122

languages may not be bound by a coargument.37, 38 In all of the examples listed above in which an anaphor may not be locally bound, that anaphor is an argument of the same predicate as its antecedent. In all of the examples in which the anaphor may be locally bound, it is not a coargument of its antecedent. Reinhart and Reuland (1993) take this evidence to mean that coargumenthood is the defining property of binding theory. So, rather than stating the binding conditions in terms of a domain for finding an antecedent, they state the binding conditions in terms of predicates. The Reinhart and Reuland binding conditions are given in (179): (179) A. A reflexive-marked predicate is reflexive B. A reflexive predicate is reflexive-marked (Reinhart and Reuland's 12"-12') In this theory, a predicate is reflexive iff two of its arguments are coindexed. A predicate is reflexive-marked iff it is lexically reflexive or one of its arguments is a SELF anaphor.39 A SELF anaphor is a morphologically complex anaphor, which contrasts with the morphologically simplex type found in Dutch, Norwegian and

37

There are some apparent counterexamples to this generalization in the case of 'inherently' reflexive predicates in Dutch and Norwegian and in the case of sentences with the verbal reflexive in Kannada. These will be discussed below. 38

This insight has its roots in Hellan (1988) and Everaert (1986). Dalrymple (1990) incorporates this insight into an LFG analysis. 39

The inclusion of lexical reflexivity in this definition of reflexive-marking is designed to explain apparent counterexamples to the coargument restriction. In the following sentences, the anticoargument anaphors may be bound by a coargument. These cases involve lexical reflexivity, however, and are captured by Reinhart and Reuland's theory: (i)

Max wast zich Max washes self 'Max washes himself'

(ii)

Jon wasket seg Jon washed self 'Jon washed himself'

123

Kannada. The simplex anaphors in these languages, according to (179), may not be coindexed with a coargument because the resulting reflexive predicate would not be reflexive-marked, in violation of condition B. One problem with this characterization of the binding theory, noted for German by Reinhart and Reuland, is that not all simplex anaphors resist binding by a coargument: (180) a. Icelandic (Thrainsson 1991) Jón varthi sig Jon defends self 'Jon defended himself' b. Lithuanian (Genius˘ iene 1987) Jis gerbia save He respects self 'He respects himself' c. Chinese (Huang and Tang 1991) Lisi piping ziji Lisi criticize self 'Lisi criticized himself' d. Hindi (Davison 1995) joon apnee-par viSvaas kar-taa hai john self-on belief do-impf is 'John believes in himself' f. Russian (Babby 1975) on zastrelil sebja he shot self 'He shot himself' In order to maintain their theory, Reinhart and Reuland would have to say either that some simplex anaphors are ambiguous between being SE anaphors and SELF anaphors or that all verbs can be inherently reflexive in these languages. The first solution merely restates the problem and also makes the theory unfalsifiable. If morphological complexity does not determine whether an anaphor is simplex, then only by looking at an

124

element's behavior can we determine which part of the theory applies to it.40 This type of post-hoc analysis is generally to be avoided. The alternative solution has the same trouble. By saying that any verb can be lexically reflexive, we take away any independent criterion for determining lexical reflexivity. In Dutch, for example, the lexical reflexivity of a verb can be determined independently of its behavior with respect to the binding of zich. In particular, when such verbs are nominalized, they can be interpreted as reflexive, while the nominalizations of other verbs cannot be interpreted in this way (Everaert 1986): (181) a.

b.

(182) a.

b.

*Max haat zich Max hates self haten is niet gezond hating is not healthy 'Hating (only someone else) is unhealthy' Max wast zich Max washes self wassen is gezond washing is healthy 'Washing (oneself) is healthy'

If we say that any verb may be inherently reflexive, then all such tests evaporate. We can conclude, then, that Reinhart and Reuland's binding theory does not accurately distinguish the set of elements that can be bound by a coargument from the set of elements that cannot. While it is true that morphologically complex anaphors must be locally bound, as Reinhart and Reuland's theory predicts, some morphologically simplex anaphors may also be locally bound, while other morphologically simplex elements may 40

Reinhart and Reuland suggest (fn. 16) that the ability to bear stress may distinguish simplex SE anaphors from simplex SELF anaphors. This view has the curious consequence that the Romance reflexive clitics (which may not be stressed) should be obligatorily long-distance, when in fact the opposite is true. These clitics are obligatorily local. If, however, the Romance reflexive clitics are really verbal reflexives (as argued in Chapter 2), then this problem does not arise.

125

not. In the next section, I will show how the theory of verbal reflexives interacts with the binding theory to predict the behavior of so-called antilocal reflexives. The coargument restriction on simplex anaphors in languages like Dutch is an emergent property of this interaction and is not explicitly stated in the grammar. 2.4.1 Semantic reflexivity and the coargument restriction In this section I will show that the existence of anticoargument reflexives is not due to the binding theory, but to an interaction of the theory of anaphora and the theory of verbal reflexives. In order to see how this works we must first examine the interactions between the reflexive pronoun and the verbal reflexive in Kannada. We will then see how the account given for Kannada can be extended to explain the other languages with antilocal reflexives as well as the languages without antilocal reflexives. 2.4.1.1 Kannada We argued in previous chapters that the verbal reflexive, koL, arises in Kannada as the morphological instantiation of a particular argument structure. One ancillary assumption of this analysis is that coreference of arguments can be represented in the argument structure. Coreference at this level gives rise to koL in sentences like (183): (183) a.

b.

hari tann-annu hogaL-i-koND-a hari self-ACC praise-PP-REFL.PST-3SM 'Hari praised himself' (x (y)) (1 (2))

As a consequence of linking the two thematic arguments to the same aspectual element (to indicate coreference), the most prominent aspectual element remains unlinked, licensing the verbal reflexive. It is important to see that the anaphor tannu is locally

126

bound in (183). That is, the coargument restriction applies only in the absence of the verbal reflexive, as illustrated in (184). (184) *hari tann-annu hogaL-id-a hari self-ACC praise-PST-3SM 'Hari praised himself' If we want to maintain the assumption that there is one binding theory for all languages and all anaphors, then we must attribute the ungrammaticality of (184) to the absence of the verbal reflexive rather than to a parameterized binding theory. Since we have seen in Chapter 2 that the verbal reflexive does not indicate semantic reflexivity but rather indicates a mismatch between the tiers of argument structure, we cannot directly attribute the ungrammaticality of (184) to the lack of the verbal reflexive. That is, since the distribution of the verbal reflexive is independent of semantic reflexivity, we cannot simply state that the verbal reflexive is required whenever semantic reflexivity obtains. We do know, however, that the expression of reflexivity in the argument structure will give rise to the mismatch responsible for the verbal reflexive (as illustrated in (183b)), and so the ungrammaticality of (184) must be attributed to not expressing the coreference between the arguments in the argument structure. The general conclusion we must draw, then, is that coreference of coarguments must always be expressed in the argumentstructure in Kannada. That is, whenever two arguments are interpreted as coreferential, this fact must be indicated in the argument-structure representation.41 For concreteness, we will assume that this restriction is stated as an explicit condition (Condition R) at the interface between the semantic component and the argument structure component, as in (185): 41

This characterization of the facts is reminiscent of the account of Reinhart and Reuland (1993). The difference between this account and theirs concerns the role of reflexive-marking. For Reinhart and Reuland, what is crucial is a one-to-one mapping between semantic reflexivity and reflexive marking. In my account, however, there is no such strict mapping. The form that arises with lexical reflexivity does not have reflexivity as its core meaning, but rather only as one of a set of possible meanings.

127

(185) Condition R: P(x, x) ↔

(x (y)) (…n…)

sem. str.

arg. str.

This correspondence rule has the property of avoiding the syntax in establishing reflexivity. In order to provide a mapping between the syntax and the semantics when an anaphor is present in the syntax and the predicate is reflexive in the semantics, we would have to find the anaphor in the syntactic representation, determine the binding domain for that anaphor and then find out whether a c-commanding antecedent was present within that domain. In contrast, if reflexivity is established lexically, the correspondence between the semantics and the syntax simply involves finding the lexical entries for each element. Establishing reflexivity in the argument structure reduces the complexity of the syntax-semantics interface, effectively building a principle of least effort into the interface conditions. The syntactic consequences of this condition will be discussed below. This type of economy condition is somewhat different from the kind of economy condition advocated by Chomsky (1993; 1995). For Chomsky, economy is an extremely local consideration, dealing only with individual steps in the syntactic derivation. The economy condition I am proposing here considers economy at the interfaces. The rule forcing coreference to be expressed lexically (i.e., in the argumentstructure) is potentially due to an economy condition of a more global sort, dealing with economy across components rather than within them. That is, the condition on the interface of semantics and argument-structure exists because it simplifies the interface between the semantics and the syntax. Note that Condition R has nothing to say about sentences in which the anaphor is not a coargument of its antecedent. In a sentence like (186), the anaphor may

128

be locally bound in the absence of a verbal reflexive because coreference of elements which are not coarguments cannot be represented in the argument-structure: (186) a.

raama tann-a yedur-ige ondu kaaLa-vanna kaND-a raama self-GEN front-DAT one buffalo-ACC see.PST-3SM 'Raama saw a buffalo in front of him(self)'

The coargument restriction in Kannada is therefore not due to the binding principles, but rather to Condition R requiring semantic reflexivity to be lexically represented. In this way, we can maintain the standard binding principles to explain the distribution of all anaphors.42 Anaphors which obey the coargument restriction are subject to the same binding conditions as all other anaphors. Independent factors concerning the interfaces between components give rise to the apparent antilocality effect. In the following sections, I will discuss how this explanation can be carried over to the other languages illustrating the coargument restriction and also those that do not. 2.4.1.2 Generalizing to Dutch and Norwegian In the previous section we saw that the coargument restriction in Kannada is due to Condition R, which forces semantically reflexive predicates to be lexically reflexive. The reason that tannu cannot be bound by a coargument in the absence of the verbal reflexive is that whenever it is bound by a coargument, the argument-structure represents this coreference and consequently the verbal reflexive occurs. The coargument restriction is thus an epiphenomenon of this interaction. In this section, we will see how this analysis can be extended to explain the coargument restriction in Dutch and Norwegian. In order to do so, we must first take a digression into the status of the antilocal anaphors in these languages.

42

The point here is that the same binding principles apply to all anaphors, but in some cases computing their application is averted by establishing coreference in the argumentstructure representation. Lexical reflexivity ensures local binding and so computing the binding domain for the anaphor is unneccesary.

129 2.4.1.2.1 Zich as a verbal reflexive The Dutch simplex anaphor zich resists binding by a coargument except when the predicate is lexically reflexive (cf. footnote 39). This is illustrated in (187), where (187a-b) represent lexical reflexivity and (187c) does not: (187) a.

Max schaamt zich Max shames self 'Max is ashamed'

b.

zij heeft zich gewassen she has self washed 'She has washed herself'

c.

*zij hatt zich she hates self 'She hates herself'

Interestingly, zich does not always behave like an argument. For example, there is no non-reflexive counterpart to (187a): (188) *Max schammt Jan Max shames Jan Along similar lines, Everaert (1986) shows that zich occurs with certain verbs as the consequence of particular morphological operations. In these cases, the reflexive verb is related to an agentive verb by a productive morphological operation: (189) a.

b.

(190) a.

b.

hij eet he eats hij overeet zich he overeats self 'He overeats' Hij spreekt he speaks hij verspreekt zich he misspeaks self 'He misspeaks'

130

According to Everaert, there is a semantic effect corresponding to the morphological change. The (b) cases have an interpretation in which "an involuntary effect … has taken hold of the agentive subject" (p. 52). This characterization is reminiscent of the discussion of external causation and verbal reflexives in Chapter 2, suggesting that zich may actually be a verbal reflexive in some sentences. It is important to see here that these verbs have only one semantic argument. The anaphor, zich, should not be viewed as the object in these cases. In (189b), for example, zich is not the object of the overeating. That is, the sentence does not mean that the subject ingested himself. Typically, the original verb undergoing such morphological operations is intransitive. If, however, the original is transitive, the direct object is turned into a prepositional object, similar to the antipassives discussed in Chapter 3.43 (191) a.

b.

*eva overat zich de appels eva overate self the apples

c.

eva overat zich ann de appels eva overate self on the apples 'Eva overate on the apples'

(192) a.

b.

43

eva at de appel eva ate the apple 'Eva ate the apple'

jan tilt de kist jan lifts the trunk 'Jan lifts the trunk' *jan vertilt zich de kist jan lifts self the trunk

This morphological operation should be viewed differently from antipassive, however, since antipassive morphemes typically apply only to transitive verbs, while this operation in Dutch applies both to transitive and intransitive verbs. Further, the semantics of antipassive involves unbounding of the event, while no such unbounding appears to be present here.

131 c.

jan vertilt zich aan de kist jan lifts self on the trunk 'Jan lifts the trunk'

The simplex 'anaphor,' like verbal reflexives in general, also arises with some decausative inchoative verbs (Everaert 1986): (193) a.

hij verspreidde het gerucht he spread the rumor 'He spread the rumor'

b.

het gerucht verspreidde zich the rumor spread self 'the rumor spread'

c.

*het gerucht verspreidde the rumor spread 'The rumor spread'

(194) a.

de chemicus verbond de zuurstof met stikstof the chemist combined the oxygen with nitrogen 'The chemist combined the oxygen with nitrogen'

b.

de zuurstof heeft zich met stikstof verbonden the oxygen has self with nitrogen combined 'The oxygen combined with nitrogen'

c.

*de zuurstof heeft met stikstof verbonden the oxygen has with nitrogen combined 'The oxygen combined with nitrogen'

(195) a.

ik los de suiker op I dissolve the sugar up 'I dissolved the sugar'

b.

de suiker lost op the sugar dissolves up 'The sugar dissolves'

c.

de suiker lost zich op in het water the sugar dissolves self up in the water 'The sugar dissolves in the water'

Also like verbal reflexives, zich may occur with the externally caused interpretation found in the possessive constructions discussed in Chapter 2. In (196), for

132

example, the only possible interpretation is one in which the eyes are opened with some instrument and not in the natural, internally caused, way (Eric Reuland, personal communication): (196) Hij opende zich de ogen He opened REFL the eyes 'He opened his eyes (with some instrument)' Everaert (1986) also shows that the verbs that take zich behave like unaccusative verbs with respect to various tests in Dutch. First, unaccusative verbs in Dutch disallow -er affixation, while unergatives allow this morphological operation: (197) a.

b.

hij is gevallen he has fallen 'He fell' *de valler the faller

(198) a.

zij eet she eats 'She eats'

b.

de eter 'the eater'

The verbs that take zich also disallow -er affixation, suggesting that they are unaccusative. (199) a.

b.

(200) a.

b.

zij overeet zich she overeats self 'She overeats' *de (zich) overeter the (self) overeater zij vergist zich she mistakes self 'She is mistaken' *de (zich) vergisser the(self) mistaker

133

Second, unaccusative verbs disallow impersonal passivization while unergatives allow it (Perlmutter 1978): (201) a.

b.

(202) a.

b.

hij rent he runs er wordt gerend there is run 'People are running' hij valt he falls *er wordt gevallen there is fallen

Verbs which take zich also disallow impersonal passivization, providing further evidence for their unaccusative status: (203) a.

b.

*er werd (zich) vergist there was self mistaken *er werd zich overeten there was self overeaten

If these verbs that take zich are unaccusative, then the NP that occurs in subject position in sentences like (200), and notzich, is the D-structure object. If zich is not the object, then it is not an argument. So, if zich is not an argument in these cases, it seems clear that it is not a reflexive pronoun licensed by inherent reflexivity either. However, we do find a strong similarity between the 'inherent' uses of zich here and the verbal reflexives discussed in chapter 2. The deagentive nature of zich in sentences like (189b) as well as its use in decausatives and externally caused possessives suggest that this is the correct characterization. Not all uses of zich can be characterized as verbal reflexives, however. In these cases, such as (171) and (172) where it is long-distance bound or locally bound by a noncoargument, zich is an anaphor. We can therefore conclude that zich is ambiguous between a verbal reflexive and an anaphor.

134

Further support for the idea that zich is a verbal reflexive can be found in comparative deletion constructions and the strict/sloppy identity contrast.44 Consider the following: (204) John defended himself better than Peter This sentence has three possible interpretations, given in (205): (205) a.

John defended himself better than Peter defended himself

b.

John defended himself better than Peter defended him

c.

John defended himself better than he defended Peter

The first reading (sloppy) illustrates a bound variable interpretation. The 'missing VP' is treated as though its object is dependent on the embedded subject for its interpretation, that is, as though the eventuality denoted by the missing VP were a 'self-defending' event. The second reading (strict) illustrates a coreference interpretation. The missing VP is treated as though its object were dependent on the matrix subject for interpretation, that is, as though the eventuality denoted by the missing VP were a 'John-defending' event. The third interpretation is irrelevant to the issue at hand. In sentences with verbal reflexives, only the sloppy interpretation is possible while sentences with nominal reflexives behave like (204) in allowing the full range of interpretations. The availability of only the sloppy interpretation with verbal reflexives is illustrated below in Finnish (206a) and Serbo-Croatian (206b) (Sells, Zaenen and Zec 1987):45

44

See Sag (1976) for an early treatment of the strict-sloppy identity distinction.

45

The availability of only the sloppy reading with verbal reflexives cannot be illustrated in Kannada because comparative constructions of this sort require that the whole VP be repeated, as in (i): (i)

hari defend-pp-refl.pst-3sm more raaju defend-pp-refl.pst-3sm 'Hari defended himself better than Raaju defended himself'

135 (206) a.

b.

Jussi puolusta-utu-i paremmin kuin Pekka John defended-REFL-PST better than Peter 'John defended himself better than Peter defended himself' *'John defended himself better than Peter defended him' Peter se branio bolje nego Ana Peter REFL defended better than Ana 'Peter defended himself better than Ana defended herself' *'Peter defended himself better than Ana defended him'

In Dutch, locally bound zich behaves like a verbal reflexive in allowing only the sloppy interpretation. In contrast, zichzelf behaves like a nominal reflexive in allowing both the strict and the sloppy interpretations (Sells, Zaenen and Zec 1987). (207) a.

b.

zij verdedidge zich beter dan Peter she defended REFL better than Peter 'She defended herself better than Peter defended himself' *'She defended herself better than Peter defended her' zij verdedidgde zich-zelf beter dan hij she defended self-self better than he 'She defended herself better than he defended himself' 'She defended herself better than he defended her'

We can conclude that the Dutch morpheme zich, when 'bound' by a coargument, has properties of a verbal reflexive and not of a reflexive pronoun. 2.4.1.2.2 Seg as a verbal reflexive Similar arguments can be raised for Norwegian seg. First, seg participates in the causative/inchoative alternation with certain verbs (Hellan 1988): (208) a.

Jon apnet doren Jon opened door 'Jon opened the door'

b.

doren apnet seg langsomt door opened REFL slowly 'The door opened slowly'

c.

Jon spredte mengden Jon dispersed crowd 'Jon dispersed the crowd'

136 d.

Mengden spredte seg crowd dispersed REFL 'The crowd dispersed'

Second, the 'inherently reflexive' verbs occurring with seg behave like unaccusatives (and not transitives) with respect to disallowing -er affixation (Hellan 1988): (209) a.

*en seg-skammer a self-shamer

b.

*en skammer seg a shamer self

Third, inherently reflexive verbs occurring with seg behave like unaccusatives in that they disallow impersonal passivization (Hestvik 1990): (210) a.

b.

*det ble ankommet there was arrived *dat ble vasket seg (av mannen) there was washed self (by the.man)

Finally, seg behaves like a verbal reflexive in comparative deletion constructions in allowing only the sloppy interpretation (Hellan 1988; Hestvik 1990): (211) a.

Jon vasket seg bedre enn Per Jon washed REFL better than Peter 'Jon washed himself better than Peter washed himself' *'Jon washed himself better than Peter washed him'

b.

Jon wasket segselv bedre enn Per Jon washed himself better than Peter 'Jon washed himself better than Peter washed himself' 'Jon washed himself better than Peter washed him'

Interestingly, when long-distance bound, seg behaves like a reflexive pronoun in allowing both the strict and the sloppy interpretations, suggesting again that seg is ambiguous between a reflexive pronoun and a verbal reflexive (Hellan 1988):

137 (212) John hadde hort meg snakke nedsettende om seg, og det hadde John had heard me talk depreciatorily about self, and it had de som stod rundt ogs_ those who stood around too 'John heard me talking depreciatorily about him, and those who were standing around also heard me talking that way about him/them' 2.4.1.2.3 Deriving the coargument restriction in Dutch and Norwegian As we have seen, zich/seg may be bound by a coargument only when the verb is inherently reflexive. These cases, like the inchoatives, represent the use of zich/seg as a verbal reflexive. For such verbs, because they are inherently reflexive, the two arguments are given by the lexical entry as coreferent. When the lexical semantic representation projects this information to the argument structure, these two arguments are linked together, giving us an unlinked element on the aspectual tier, thereby licensing the verbal reflexive. In sentences like (213a), then, zich arises not because it is a bound anaphor but because it represents an unlinked element on the aspectual tier, just as it does in (214): (213) a.

b.

Max wast zich Max washes self 'Max washes himself' (x (y)) (1 (2))

(214) a.

b.

het gerucht verspreidde zich the rumor spread self 'the rumor spread' (y) ((1 (2))(3))

This contrasts with (215a), in which zich is an anaphor because it is not an argument of the same predicate as its antecedent. That is, in (215b), we cannot mark any two arguments of the verb legt 'put' as coreferent, so no mismatch arises:

138 (215) a.

b.

Max legt het boek achter zich Max put the book behind self 'Max put the book behind him(self)' (x (y (z)))

(x=agent, y=theme, z=locative)

((1 (2))(3)) If zich is a verbal reflexive in sentences like (213a), then we should ask what blocks its occurrence in (216), especially since such a sentence is grammatical in Kannada with the verbal reflexive: (216) *Max haat zich Max hates self Since zich is ambiguous between a verbal reflexive and an anaphor, we have two possibilities to examine. First, zich could be a verbal reflexive. In this case, (216) would have an argument structure representation like (213b) in which the two arguments are colinked lexically giving rise to the mismatch licensing the verbal reflexive. Alternatively, zich could be an anaphor. In this case, the argument structure representation of (216) would have no two elements colinked and the reference of zich would be resolved through the binding conditions. The argument structure of (216) where zich is an anaphor is given in (217): (217) (x (y))

(x=Max, y=zich)

(1 (2)) Both options need to be ruled out in order to block (216) from being grammatical. Let us consider the semantic representation for (216). Because the two arguments of the predicate are coreferential (whether zich represents a verbal reflexive or an anaphor), the semantic representation will include at least the following information: (218) hate (Max, Max) Recall from our discussion of Kannada that there is a rule forcing semantic reflexivity to be lexically expressed. This rule is repeated here:

139 (185) Condition R: P(x, x) ↔

(x (y)) (…n…)

sem. str.

arg. str.

The simplest assumption is that Condition R is not specific to Kannada, but rather is part of UG and can therefore be found in all languages. Returning now to (216), we have a sentence in which the intended interpretation has a reflexive predicate. Condition R applies, forcing reflexivity to be established in the lexical representation. However, the possible lexical representations for the verb hate do not include one with colinked arguments as given in the argument structure portion of Condition R. That is, hate projects an argument structure in which the two elements on the thematic tier are necessarily linked to different elements on the aspectual tier, as in (219): (219) hate:

(x (y)) (1 (2))

Because this representation is not compatible with the requirements of Condition R, the sentence (216) is ungrammatical. The semantic representation of the sentence requires an argument structure representation with colinked elements on the thematic tier, but the argument structure of this particular verb cannot satisfy this requirement. If zich is a verbal reflexive, then lexical properties of hate are violated. On the other hand, if zich is an anaphor in (216), then Condition R is violated. That is, if zich is an anaphor, then the argument structure in (217) is used. This argument structure, however, violates the Condition R because a semantically reflexive predicate does not allow an argument structure without two colinked elements. As a consequence, (216) is ungrammatical no matter what the interpretation of zich is. The basic insight here is simply that zich is interpreted as a verbal reflexive whenever it is in the appropriate syntactic context, although this insight is not explicitly stated as such in the grammar. Instead, the

140

preference to use zich as a verbal reflexive is due to a conspiracy of independent principles. The preceding analysis of (216) brings up the question of why sentences with reflexive pronouns are ever possible in coargument contexts. That is, if reflexivity must be established lexically, then why are complex anaphors allowed in such contexts as (220): (220) Max haat zichzelf Max hates selfself 'Max hates himself' Reinhart and Reuland (1993) explain the existence of such sentences by claiming that complex anaphors reflexivize a predicate. However, Reuland (1995) provides evidence that these sentences do not contain semantically reflexive predicates at all. The zichzelf anaphor does not establish identical reference with the subject but rather a relation which allows a coreference interpretation but may be distinguished from it. As evidence, Reuland presents a modified Madame Tussaud context (Jackendoff 1992) in which a famous person walks into the wax museum and looks into a mirror. The story continues with one of the following: (221) a.

b.

ze zag zich in een griezelige hoek staan she saw REFL in a creepy corner stand 'She saw herself (=reflection) in a creepy corner stand' ze zag zichzelf in een griezelige hoek staan she saw selfself in a creepy corner stand 'She saw herself (=statue of herself) in a creepy corner stand'

Only the complex anaphor allows the interpretation that the famous person saw a statue of herself, indicating that zichzelf does not necessarily involve identical reference while the simplex form zich does. That is, X-SELF anaphors receive the interpretation of an entity related to and similar to the subject, but not necessarily identical to the subject. In most cases, however, the interpretation of the X-SELF anaphor will be indistinguishable from the interpretation of complete identity. That is, the interpretation of a sentence with

141

an X-SELF anaphor is extensionally equivalent in most cases to a sentence with a reflexive predicate although semantically distinct. This distinction will be addressed further below. Having given an analysis for the ungrammaticality of (216), the availability of zich in (200), repeated here, is easily explained. (200) zij vergist zich she mistakes self 'She is mistaken' Here we have a semantically reflexive predicate. Condition R forces the reflexivity to be expressed lexically. Because the predicate is lexically reflexive, this condition can be met.46 A similar analysis is available for (213), repeated here: (213) Max wast zich Max washes self 'Max washes himself' Again, semantic reflexivity triggers Condition R. There is a difference between (200) and (213) at this point. The verb in (200) is obligatorily reflexive; it does not allow any NP in object position. In contrast, the verb in (213) can be nonreflexive in other sentences.

46

(222) a.

*zij vergist max she mistakes max

b.

max wast otto Max washes Otto 'Max washes Otto'

Alternatively, the verb in (200) could have a decausative-like argument structure. On this view, the interface condition requiring reflexive predicates to be lexically reflexive does not apply since there is no reflexive predicate in the sentence. On either view, the possibility that zich is an anaphor simply doesn't arise since the presence of zich is due to the argument structure representation of the verb.

142

In order to account for both the reflexive and nonreflexive uses of the verb wash we will have to say that the argument structure representation associated with this verb is lexically underspecified. That is, while hate requires a non-reflexive argument structure and mistake requires a reflexive argument structure, wash allows either. Since wash does allow a reflexive argument structure, it is compatible with Condition R which requires reflexivity to be expressed lexically and therefore allows the verbal reflexive, zich, to be present.47 Additional support for the idea that Condition R plays an explanatory role in the distribution of zich/seg comes from sentences with embedded clauses containing these forms. Consider the following Norwegian sentences (Everaert 1986; Hestvik 1990): (223) a.

47

Joni lot Olaj vaske segi/j Jon made Ola wash self 'Jon made Ola wash herself/him'

b.

Jon vasket seg Jon washed REFL 'Jon washed himself'

c.

Johni bad Maritj se pF segi/*j John made Mary look at self 'John made Mary look at him'

One might ask at this point what property of wash allows it to be underspecified in its lexical representation, i.e., why it can be lexically reflexive. The answer to this question is not obvious but it is worth noting that there is a set of verbs (wash, behave, shame, mistake, dress, shave, etc.) which cross-linguistically allow the inherently reflexive interpretation. See Jackendoff (1990, pp. 64-68) for discussion of the semantic representations of these verbs. Note further that Kannada differs from Dutch in allowing any verb to be lexically reflexive. Luigi Burzio (personal communication) has suggested that this difference may derive from the fact that Kannada has distinct lexical items for the verbal reflexive and the anaphor while Dutch uses the same lexical item for both. Malayalam is the most extreme case, in that it never allows reflexivity to be lexically expressed. This question will be discussed further below.

143 d.

*John sF pF seg John looked at self 'John looked at himself'

What is important here is the generalization that if seg allows coargument binding in a simple sentence, then it allows both a local and long-distance interpretation in a biclausal sentence. (223b) is grammatical with seg locally bound and (223a) is grammatical with seg locally or long-distance bound. On the other hand, if seg does not allow local binding in a simple sentence, then it does not allow it in the long-distance context either, as illustrated in (223d) and (223c). We saw above that zich/seg must be a verbal reflexive whenever it can be. This generalization, however, is not stated as such in the grammar. Rather, a reflexive predicate in the semantics forces an argument structure in which reflexivity is expressed. The expression of reflexivity in the argument structure in turn forces the verbal reflexive to be present. These conditions are what allow seg to be present in (223b) but not (223d) above. The availability of long-distance binding in (223a) and (223c) supports this account. In a sentence with a long-distance anaphor, there are no reflexive predicates because the two coreferential elements are not coarguments. Since there are no reflexive predicates, Condition R, forcing the lexical expression of reflexivity, does not apply and so the anaphor variant of seg is available. The choice of verbal reflexive seg over anaphor seg is therefore sensitive to the semantics of the sentence involved. In other words, the generalization that the verbal reflexive is used whenever possible is not a generalization about choosing between lexical items but rather falls out from an interaction of independent properties of the grammar concerning the interface between semantic representations and argument structure representations. The idea that the choice between the anaphor and verbal reflexive variants of zich/seg is due to interactions between semantics and argument structure gains further support from sentences with embedded lexically reflexive predicates, as in (224) (Everaert 1986):

144 (224) a. Dutch *Jan zag mij zich verslikken Jan saw me self choke 'Jan saw me choke him(self)' b. Norwegian *Ola lot oss skamme seg Ola made us shame self 'Ola made us shame her(self)' Here, although the syntactic context would allow long-distance binding of the anaphor, the embedded verbs are specified lexically as being reflexive. That is, the argument structure representations for the embedded verbs in (224) require the two thematic elements to be colinked, as in (225): (225) (x (y)) (1 (2)) This lexical structure blocks the possibility that zich/seg is an anaphor referentially dependent on the matrix subject in (224). If it were an anaphor there, then the two thematic elements in (225) would not be colinked. But this colinking is required by lexical properties of the verb. The long-distance binding option in (224) is thus blocked by the argument structure requirements of the embedded verb. An interesting difference between Dutch and Norwegian arises when predicates which permit but do not require inherent reflexivity are embedded. Consider the following: (226) a. Norwegian Joni lot Olaj vaske segi/j Jon made Ola wash self 'Jon made Ola wash herself/him' b. Dutch Jani laat Pietj zich*i/j wassen Jan lets Peit self wash 'Jan lets Piet wash himself'

145

In Norwegian, long-distance binding is possible over verbs like wash which permit but do not require lexical reflexivity. In Dutch, on the other hand, long-distance binding is not possible over such verbs. In discussing such verbs in Norwegian above, we saw that their lexical representations were underspecified for argument structure. That is, they allow their two arguments to be either lexically colinked or not. In order to explain the difference between Dutch and Norwegian with respect to these verbs, we will have to say that the reflexive argument-structure representation is preferred over the non-reflexive argument-structure for verbs like wash in Dutch but not in Norwegian. In Dutch, if the verb wash is in a syntactic context in which it can have the reflexive argument structure, it must have this argument structure. In other words, the argument-structure to syntax interface takes precedence over the argument-structure to semantics interface in these cases in Dutch. The possibility of treating zich as a verbal reflexive with wash outranks the fact that the long-distance interpretation is compatible with a different possible argument structure representation of the verb. Simply put, wash in Norwegian has a lexical representation which is compatible with a reflexive or nonreflexive argument structure while the same verb in Dutch requires lexical reflexivity whenever it can.48 2.4.1.2.4 The syntactic representation of zich/seg Up to this point, we have seen that zich/seg is ambiguous between a verbal reflexive and an anaphor. We have not, however, explored the syntactic representation of these morphemes. That is, in the development of the above analysis, we have treated sentences with zich/seg as a verbal reflexive differently from sentences with zich/seg as

48

The difference in the treatment of wash in Dutch and Norwegian suggests interacting violable constraints of the sort argued for by Prince and Smolensky (1993) and others. Note, however, that we have found evidence for such interacting constraints only at the interfaces between components and not within components. We have seen no evidence of violable constraints within the syntax, semantics or argument structure. Rather, where these components interact we find the possibility of interacting constraints.

146

an anaphor with respect to argument-structure and semantic representations, but we have not addressed the syntactic representation of these forms. The question that arises is whether the syntax of sentences with the verbal reflexive is really equivalent to the syntax of sentences with the anaphor. At issue is whether we can derive the homophony of these elements from independent principles of grammar. Let us suppose that zich/seg has the same categorial representation whether it is a verbal reflexive or an anaphor. We have no reason to believe this supposition to be false. Maintaining it allows us to understand a piece of the homophony puzzle. If the two zichs have one syntactic representation, then we can begin to understand why the ambiguity exists. One reason to believe that the syntactic representations of these elements is equivalent concerns their clitic status. Interestingly, Dutch zich behaves like a clitic in several respects (Everaert 1986). First, zich behaves like a clitic in that it occurs in a fixed position with respect to the verb. Nothing may intervene between the verb (in second position) and zich: (227) a.

b.

zij heeft zich met opzet niet verdedigd she has self on purpose not defended 'She has purposely not defended herself' *zij heeft met opzet zich niet verdedigd she has on purpose self not defended

Second, like other clitics, zich may not be topicalized: (228) a.

b.

*'ri gaf Jan ti het book her gave Jan the book 'Her, Jan gave the book' *zichi wast Jan ti nooit REFL washes Jan never 'Himself, Jan never washes'

This prohibition also applies to prepositionally governed clitics, as illustrated in (229a). Note that even when zich is long-distance bound, it may not occur in the topicalized position, as in (229b):

147

(229) a.

b.

*van 'mi houdt Jan ti P him loves Jan 'Him, Jan loves' *voor zichi liet Jan Piet ti werken for self let Jan Piet work 'For him(self), Jan let Piet work'

Also, a PP may move out of its canonical position to the left of the verb only if it contains a nonclitic pronoun (Everaert 1986): (230) a.

Zij liet mij een huis voor zich bouwen She let me a house for self build 'She had me build a house for her'

b.

Zij liet mij een huis voor haar bouwen She let me a house for her build 'She had me build a house for her'

c.

*Zij liet mij voor zich een huis bouwen She let me for self a house build 'She had me build a house for her'

d.

Zij liet mij voor haar een huis bouwen She let me for her a house build 'She had me build a house for her'

The anaphor zich behaves like a clitic in blocking the PP movement in (230c) even though it is long-distance bound. Finally, zich, like other clitics, cannot be coordinated. (231) a.

b.

*zij verdedidge zich en haar vrienden she defended self and her friends zij verdedigde zichzelf en haar vrienden she defended selfself and her friends

All the above criteria suggest that zich is a clitic (see Everaert 1986; Kayne 1975; Zwicky 1977). Its clitic status may be what enables it to behave in some ways like an anaphor (it can be long-distance bound) and in other ways like a verbal reflexive

148

(external causation, inchoatives, lexical reflexives).49 Because it is a clitic dependent on a verb, it is ambiguous between a verbal element and a nominal element and so may be handled either by the system of argument structure developed in Chapter 2 or by the syntax. In general, verbal reflexives are contiguous with the verb, either as affixes or clitics. On the other hand, the fact that zich is a clitic even when it is an anaphor demonstrates that being a clitic is not a sufficient condition for being a verbal reflexive. That is, some clitic reflexives are not verbal reflexives. We can then ask whether being a clitic is a necessary condition for being a verbal reflexive. It is possible that verbal reflexives do not have a morpho-syntactic definition, despite the traditional terminology. The fact that a particular morpheme is a verbal reflexive does not necessarily entail a particular morpho-syntactic relationship between that morpheme and the verb. A morpheme's status as a verbal reflexive entails only that that morpheme signals a particular argument structure configuration, as argued in Chapter 2. On the other hand, I know of no verbal reflexives which are not either affixes or clitics. There is nothing, however, in the theory of verbal reflexives which requires this result.50 Whether this agnosticism is desirable remains to be seen. Note that there are other verbal elements which are not necessarily contiguous with the verb, as illustrated by phrasal verbs such as look up. Such predicates have a single argument-structure representation but project two syntactic positions, one for the verb and one for the preposition. These two verbal elements may be separated by other syntactic material: (232) a.

John looked up the information

49

The fact that the Romance reflexive clitics do not allow long-distance binding is irrelevant here since the anaphor in these languages is a different form than the verbal reflexive. 50

One possibility is that independent NPs must have case features and/or a thematic interpretation, so that if a verbal reflexive had such properties, it would be interpreted as a bearing a theta-role. Since verbal reflexives do not bear theta-roles, we might derive the fact that verbal reflexives cannot be full NPs.

149

b.

John looked the information up

Similarly, sentences with a verbal reflexive may allow the independent syntactic projection of the verb and the element marking the mismatch in the argument structure. If it is true that zich has the same syntactic status when it is an anaphor as when it is a verbal reflexive, then what is the difference between the two uses? We can answer this question by examining the syntactic properties of anaphors in general. What makes an anaphor an anaphor is its referential and syntactic dependence on some other NP. Let us assume that this dependence is a chain relation (cf. Chomsky 1973; Reinhart and Reuland 1993). That is, when an anaphor is dependent on some antecedent, then it is part of the chain headed by its antecedent. A verbal reflexive, on the other hand, is not in a chain relation with any other element. Verbal reflexives, being nonreferential (more accurately, a-referential), are not referentially dependent on other NPs. The syntactic difference between zich as a verbal reflexive and zich as an anaphor, then, lies in whether it is part of a chain with another NP. When zich is a verbal reflexive, it is not part of a chain with another NP. When it is an anaphor, it is part of a chain with another NP. So, although the categorial representation of zich is constant, the properties of the syntactic chain containing it determine which variant is used. We can thus derive Condition R requiring semantic reflexivity to be lexically expressed from a general syntactic economy condition avoiding the operation of forming a chain between two NPs. That is, connecting two elements with a chain is more costly than leaving them independent. So, if reflexivity is not expressed lexically, then a chain must be formed to be able to relate zich to an antecedent. In terms of the previous discussion of interface conditions, we see that the semantic component prefers to interpret one-membered chains over multiplemembered chains. Thus, the choice of verbal reflexive zich over anaphor zich derives ultimately from the interfaces of the semantic component and the syntactic and argumentstructure components.

150

Lexically, there is one zich, with the syntactic category D. This form is polysemous between an anaphor and a verbal reflexive. When it is an anaphor, its lexical representation includes whatever information requires a form to be subject to principle A, most likely the lack of any referential properties. On the other hand, when it is a verbal reflexive its lexical representation indicates that it is used whenever there is a mismatch between the two tiers in argument structure. These two options are not independent lexical entries, as for example with the English word bank. Instead, they are one lexical entry with independent properties, much as the English word 'book' is polysemous between an abstract object and a physical object (Jackendoff 1983). One way we might think of these two possibilities for zich is as repair strategies for a nominal form with no referential properties. One repair strategy is to put the element into a syntactic chain dependency so that it can inherit the referential properties of the head of the chain. The other repair strategy is to treat it as a verbal reflexive, sensitive to argument structure properties and thereby indifferent to reference. Note that the first of these strategies requires more syntactic information to be included in the final representation (the chain), but no additional lexical information. The alternative strategy adds additional lexical information (the argument structure) but no additional syntactic information. In this way, we have a trade-off between syntactic complexity and lexical complexity. The more complex the syntactic representation is, the less complex the lexical representation will be. Similarly, more complexity in the lexical representation entails less complexity in the syntactic representation. 2.4.1.3 Malayalam Returning now to the nature of the coargument restriction, the theory developed so far makes an interesting prediction. If a language never allows reflexivity to be lexically expressed, then it should always obey the coargument restriction, even in the context of apparent inherent reflexivity. This prediction follows from the above

151

analysis because of the interaction of particular lexical representations with Condition R forcing semantic reflexivity to be lexically expressed. Kannada allows reflexivity to be lexically expressed with any verb. As a consequence the coargument restriction on tannu is obviated by the presence of the verbal reflexive. The verbal reflexive is possible with all verbs because any verb can be lexically reflexive. Coargument-bound tannu is therefore licensed with all such verbs. Dutch and Norwegian obey the coargument restriction with all verbs except those that are lexically reflexive. The verbs that are lexically reflexive involve the use of zich/seg as a verbal reflexive. The coargument restriction, in turn, derives from Condition R forcing lexical reflexivity and the particular lexical items of these languages blocking such lexical reflexivity. Since fewer verbs may be lexically reflexive in these languages, we find more instances of verbs enforcing the coargument restriction. The prediction is thus that a language might exist which has no predicates that allow lexical reflexivity. Such a language would enforce the coargument restriction across the board. Malayalam is such a language. In Malayalam, the anaphor, tan, may not be bound by a coargument (Mohanan 1982): (233) *moohan tan-ne aaraaDhik'k'uNNu mohan self-ACC worships 'Mohan worships himself' Even predicates which are lexically reflexive in other languages disallow local binding of tan in Malayalam (K.A. Jayaseelan, personal communication). (234) a.

b.

*raaman tan-ne kshauram ceytu raaman self-ACC shaving did 'Raman shaved himself' raaman tan-ne-tanne kshauram ceytu raaman self-ACC-EMPH shaving did 'Raaman shaved himself'

These facts suggest that no verb may be lexically reflexive and as a consequence, tan may never be locally bound. Its antilocality is not due to a modified binding theory.

152

Rather, it is due to the interaction of Condition R and particular lexical items. A further question might be asked as to why Malayalam has no verbs which allow reflexivity to be lexically expressed. Ultimately, this is a question about what is lexicalized by particular languages and its answer may be derivable from other properties of the lexicon. For the moment, however, it is simply a peculiar property of the Malayalam lexicon and not due to any deep differences between the grammar of Malayalam and the grammars of other languages, a welcome result. 2.5 Near-reflexives The theory developed in the previous section to account for the coargument restriction makes a further prediction. If it is true that all semantically reflexive predicates must be lexically reflexive, then local binding of an anaphor must not involve semantic reflexivity. If local binding, in the absence of lexical reflexivity, did involve semantic reflexivity, then Condition R would be violated. In other words, the theory predicts that the coargument restriction should hold in all languages, at least with verbs that are not lexically reflexive. Coargument binding of an anaphor in Chinese, for example, should be ruled out by the same conditions that rule it out in Malayalam and (with particular verbs) in Dutch. Since this generalization does not hold in all languages, then the counterexamples to the coargument restriction must not involve semantic reflexivity. Let us examine in closer detail why this prediction is made. Consider the following: (235) John hit himself On the surface, it seems quite obvious that this sentence is semantically reflexive. The two coarguments of the verb are coreferential. If the sentence is semantically reflexive, and if Condition R is active in all languages, then the verb hit must be lexically reflexive. Since Condition R requires semantic reflexivity to be lexically represented, it follows that

153

this sentence involves lexical reflexivity. However, tests for lexical reflexivity reveal that the verb is not lexically reflexive: (236) a. b.

Hitting is not nice Washing is healthy

In (236a) we find the interpretation that hitting others is not nice but not the interpretation that hitting yourself is not nice. In (236b), on the other hand, we find the interpretation that washing yourself is healthy but not the interpretation that washing others is healthy. Following Everaert (1986), we take these examples to show that wash is lexically reflexive and that hit is not. Lexically reflexive verbs maintain their reflexive interpretation when they are nominalized. Since hit is not a lexically reflexive verb, then it follows that (235) does not involve lexical reflexivity. Thus, it appears that (235) violates the rule forcing semantic reflexivity to be lexically represented. Assuming that Condtion R is correct, we must conclude that sentences like (235) do not involve semantic reflexivity.51 The conclusion that some reflexive sentences do not involve semantic reflexivity is not as far-fetched as it might seem at first. Jackendoff (1992) shows that reflexive pronouns may be interpreted as referring to a representation of their antecedents and not only as strictly identical to their antecedents. This is illustrated through the following example. Ringo Starr goes into Madame Tussaud's wax museum. While walking through the museum, Ringo sees the statue that portrays him and then he trips. If, in tripping, Ringo falls on his statue, we can then say: (237) Ringo fell on himself

51

One possibile solution to this problem is to say that X-SELF anaphors make the predicate lexically reflexive, although the technical implementation of such an approach is far from clear. However, we will see below that lexically reflexive predicates have other properties that verbs taking X-SELF complements do not have, indicating that X-SELF anaphors do not make a predicate lexically reflexive.

154

Here, the referent of himself can be the statue of Ringo. That is, the sentence can mean that Ringo tripped and fell on the statue depicting him. What this illustrates is that the reflexive pronoun does not express total identity with its antecedent. Rather, it can be interpreted as an entity related to and similar to its antecedent. We will call such reflexives 'near-reflexives.' Returning to (235), we say that the anaphor in this case is a near-reflexive. As such, the sentence does not express semantic reflexivity or complete identity between the two arguments. In this case, the use of a near-reflexive results in an interpretation which is extensionally equivalent to semantic reflexivity but which differs in its formal properties. Thus, Condition R is not violated and local binding is possible. Note that near-reflexives are reflexives in the sense of being subject to Condition A of the binding theory. The interpretation of sentences with near-reflexives is different, however, from sentences which are semantically reflexive. Further support for the distinction between near-reflexives and other reflexives comes from other languages. As we have seen, Dutch illustrates this distinction in the difference between zich and zichzelf. The verbal reflexive zich does not allow the statue interpretation while the near-reflexive zichzelf does, as illustrated in (221), repeated here: (221) a.

b.

ze zag zich in een griezelige hoek staan she saw REFL in a creepy corner stand 'She saw herself (=reflection) in a creepy corner stand' ze zag zichzelf in een griezelige hoek staan she saw selfself in a creepy corner stand 'She saw herself (=statue of herself) in a creepy corner stand'

This distinction is also illustrated in Italian, where the verbal reflexive, si, does not allow the statue interpretation, while the near-reflexive, se-stesso, does (Burzio 1994):

155 (238) a.

Ringo si è sputato addosso Ringo REFL is spit upon 'Ringo spat on himself' (himself ≠ statue)

b.

Ringo cadde su se-stesso Ringo fell on self-same 'Ringo fell on himself' (himself = statue)

What the Dutch and Italian examples show is that verbal reflexives allow only an identity interpretation while certain reflexive pronouns allow the near-reflexive interpretation. Since we have already claimed that the verbal reflexives will arise when reflexivity is lexically expressed and that lexical reflexivity is semantic reflexivity, it follows that semantic/lexical reflexivity requires complete identity of the two coreferential arguments and thus that near-reflexives have a semantic representation which differs from semantic reflexivity. For concreteness, we can assign the following representations to semantic reflexivity and near reflexivity, where the value of f(x) is referentially dependent on the value of x, but not semantically equivalent to the value of x (cf. Reuland 1995): (239) a. b.

Semantic reflexivity λx (P(x, x)) Near-reflexivity

λx (P(x, f(x)))

A second difference between verbal reflexives and nominal reflexives concerns the availability of certain interpretations in comparative deletion constructions. As we have seen, reflexive pronouns allow more interpretations under comparative deletion than verbal reflexives. This is illustrated here in Serbo-Croatian (Sells, Zaenen and Zec 1987): (240) a.

Peter se branio bolje nego Ana Peter REFL defended better than Ana 'Peter defended himself better than Ana defended herself' *'Peter defended himself better than Ana defended him'

156 b.

Peter je branio sebe bolje nego Ana Peter AUX defended self better than Ana 'Peter defended himself better than Ana defended herself' 'Peter defended himself better than Ana defended him'

Here, the verbal reflexive allows only the sloppy reading under comparative deletion while the nominal reflexive allows either the strict or sloppy readings. Given that the verbal reflexive arises from semantic/lexical reflexivity in sentences like (240a), we can conclude that sentences which express semantic reflexivity allow only the sloppy reading under comparative deletion. From this we can conclude that (240b) does not express semantic reflexivity but rather near-reflexivity. Returning to the issue at hand, recall that a consequence of Condition R was that the coargument restriction should hold in all languages. The prediction which derives from this consequence is that any locally bound reflexives which do not occur with lexically reflexive predicates must be near-reflexives and must not express semantic reflexivity. To test this prediction, we can use the two tests for near-reflexivity. Consider first the behavior of Russian reflexives. Russian has both a verbal and a nominal reflexive (Babby 1975). The nominal reflexive may be locally or longdistance bound (Timberlake 1979). These facts are given in (241): (241) a.

on zastrelil-sja he shot-refl 'He shot himself

b.

on zastrelil sebja he shot self 'He shot himself'

c.

starik ozivilsja i prosil na kurort sebja pokuda ne otpravljat' old.man enliven and ask to resort self now not send.off 'The old man came to life and asked (someone) not to send him off to a health resort just now' The question for the theory under discussion is why the reflexive pronoun sebja does not obey the coargument restriction, as in (241b). If semantic reflexivity must be lexically expressed and if the lexical expression of reflexivity gives rise to the verbal reflexive,

157

then how is (241b) possible? The theory predicts that (241b) is possible only if this sentence does not express semantic reflexivity but rather near-reflexivity. The tests for near-reflexivity bear this prediction out. First, in comparative deletion constructions, the Russian verbal reflexive allows only the sloppy interpretation (242a) while the Russian nominal reflexive (242b) allows both the strict and sloppy interpretations. 52 (242) a.

Ivan zashchishchal-sja lachshe chem Petr Ivan defended-REFL better than Peter 'Ivan defended himself better than Peter defended himself' *'Ivan defended himself better than Peter defended him'

b.

Ivan zashchishchal sebja lachshe chem Petr Ivan defended self better than Peter 'Ivan defended himself better than Peter defended himself' 'Ivan defended himself better than Peter defended him'

Thus, sentences with the verbal reflexive express semantic reflexivity and sentences with the nominal reflexive express near-reflexivity. Second, in the Madame Tussaud context, the verbal reflexive is not possible while the reflexive pronoun is possible. Here, imagine that Boris Yeltsin goes into the wax museum and upon seeing his statue he becomes enraged. In this context, we find the following possibilities: (243) a.

Yeltsin zastrelil-sja Yeltsin shot-REFL 'Yeltsin shot himself' (himself = Yeltsin, ≠ statue)

b.

Yeltsin zastrelil sebja Yeltsin shot self 'Yeltsin shot himself' (himself = Yeltsin or statue)

Thus, we can conclude that the reflexive pronoun in Russian is a near-reflexive and does not express complete identity with its antecedent.

52

Thanks to Alexander Lehrmann for providing the following Russian judgements.

158

Since the Russian reflexive pronoun is a near-reflexive, the fact that it may be locally bound does not present a counterexample to Condition R. This apparent counterexample actually supports the theory being developed. A similar argument can be made for Chinese. In Chinese, the reflexive pronoun ziji may be locally or long-distance bound: (244) Zhangsan renwei Lisi piping ziji Zhangsan thinks Lisi criticized self 'Zhangsan thinks Lisi criticized him/himself' The possibility of local binding for ziji also presents a potential counterexample to the theory being developed here. Again, however, this counterexample is only an apparent one. The reflexive pronoun ziji displays the properties of a near-reflexive and not of identity. In comparative deletion constructions, ziji allows either a strict or a sloppy interpretation, suggesting that it is a near-reflexive:53 (245) Zhangsan bi Lisi wei ziji bianhu de hao Zhangsan than Lisi for self defend DE well 'Zhangsan defended himself better than Lisi defended himself' 'Zhangsan defended himself better than Lisi defended him' The reflexive pronoun ziji also allows both the statue or the identity interpretation in the Madame Tussaud context, suggesting again that it is a near-reflexive: (246) Mao Tse Tung ba ziji qiangbi le Mao Tse Tung BA self shoot ASP 'Mao Tse Tung shot himself' (himself = statue or Mao) We can therefore conclude that the prediction that locally bound reflexive pronouns do not express semantic reflexivity is borne out in Russian and Chinese. In Russian, semantic reflexivity is realized through the verbal reflexive, as expected, while in Chinese it would appear that no sentence expresses semantic reflexivity.

53

Thanks to Bert Xue for providing the following Chinese judgements.

159 2.6 Conclusions and Consequences In this Chapter I have argued that there are no antilocal reflexives in language. Those elements which have been characterized as antilocal may be locally bound as long as they are not bound by a coargument. The coargument restriction, in turn, is not explicitly stated as such in the grammar but rather arises due to a conspiracy of principles constraining the interaction of lexical, syntactic and semantic representations. The principle that is ultimately responsible for the coargument restriction is Condition R, an interface condition forcing semantic reflexivity to be lexically expressed, i.e., by colinking the coreferential elements in the argument-structure representation, repeated here: (185) Condition R: P(x, x) ↔

(x (y)) (…n…)

sem. str.

arg. str.

Semantic reflexivity obtains whenever two arguments of a predicate are coreferential. When a particular predicate does not allow such colinking in its lexical representation and its two arguments are coreferential, this principle is violated. This violation gives rise to the coargument restriction. Given a verb which does not allow colinking in its argument-structure and an anaphor which does not obey the coargument restriction, we must assume that semantic reflexivity does not hold. Such a case is illustrated by the English example (235), repeated here: (235) John hit himself We have seen that such sentences, in fact, do not express semantic reflexivity but rather express near-reflexivity, a semantic relationship which in most cases is extensionally equivalent to semantic reflexivity but which is formally distinguishable from it. In the

160

absence of semantic reflexivity in sentences like (235), the incompatibility problem between the verb’s inability to colink its arguments and Condition R forcing colinking under semantic reflexivity does not arise. Descriptively speaking, languages vary with respect to the extent that they enforce the coargument restriction. This variation is explained by lexical differences between languages. Kannada, being the least restrictive, has a productive lexical rule of reflexivization. This rule links arguments together in the argument-structure representation. The output of this rule is given in (247): (247) (x (y)) (...n...) As we saw in Chapter 2, this argument-structure gives rise to the verbal reflexive.54 As a consequence of this rule, any verb may be lexically reflexive in Kannada and so coargument binding is impossible only in the absence of the verbal reflexive, i.e., only when (247) does not obtain. Dutch and Norwegian allow only certain predicates to be lexically reflexive. Those verbs that do not allow lexical reflexivity show the coargument restriction, as discussed above. Malayalam has no lexically reflexive predicates and so the coargument restriction is found with all verbs. That is, since (247) is impossible with all verbs in Malayalam, semantic reflexivity is never possible, in accord with Condition R. As a consequence, the coargument restriction obtains across the board for the anaphor tan, which is a true reflexive and not a near-reflexive. Chinese differs in yet another way. Like Malayalam, Chinese does not allow any lexically reflexive predicates. Unlike Malayalam, however, Chinese does not show evidence of the coargument restriction. This difference is attributable to the semantic content of the anaphors found in these

54

It is important to see, however, that the implication goes in one direction only. While all reflexive predicates will give rise to the verbal reflexive, not all instances of the verbal reflexive will correspond to semantic reflexivity. See Chapter 2 for discussion.

161

languages. The Malayalam anaphor tan expresses complete identity with its antecedent, forcing semantic reflexivity and all that follows from that, while the Chinese anaphor ziji expresses near-reflexivity, thereby avoiding Condition R altogether. Interestingly, the conclusion that we come to with respect to Malayalam is exactly the opposite of that reached by Mohanan (1982), who claimed that the antilocality effect on tan was due to its bearing both anaphoric and pronominal features, i.e., that tan was in some sense less reflexive than other anaphors. The conclusion we find here is that tan is actually more reflexive than anaphors which allow local binding. In a local binding context, tan requires complete identity with its antecedent, as opposed to other reflexives which allow near-reflexivity. An important question for further research lies in the nature of nearreflexives. The initial observation appears to be that all morphologically complex reflexives, like English himself and Dutch zichzelf, are near-reflexives. However, the converse is not also true. Not all near-reflexives are morphologically complex, as illustrated by Chinese ziji and Russian sebja. This problem is reminiscent of the one faced by Reinhart and Reuland in their distinction between SELF anaphors and SE anaphors. For Reinhart and Reuland, the SELF anaphors, characterized by morphological complexity, reflexivize a predicate and thus can be locally bound by a coargument. The problem for their approach is that not all SELF anaphors are morphologically complex and so the categories of anaphor that they develop on this basis are not accurately distinguished. Thus, both Reinhart and Reuland and I are faced with the question of accurately characterizing those anaphors which allow local binding. An important contrast between Reinhart and Reuland’s result and my own lies in the role of SELF anaphors. While Reinhart and Reuland assume that SELF anaphors make a predicate reflexive, we have seen evidence that SELF anaphors do not occur in semantically reflexive sentences, casting doubt on this assumption. Instead, what Reinhart and Reuland call SELF anaphors are actually near-reflexives, referentially

162

dependent on but not identical to their antecedents. As we have seen, the distinction between near-reflexives and identity-reflexives is independently motivated by evidence from comparative deletion and the Madame Tussaud context. The distinction between SELF

anaphors and SE anaphors is not as well motivated.

References Aissen, Judith (1974) The Syntax of Causative Constructions Ph.D. Dissertation, Harvard University: Cambridge, MA. Amritavalli, R. (1984) "Anaphorization in Dravidian," Central Institute of English and Foreign Languages Working Papers in Linguistics 1.1:1-31. Amritavalli, R. (1990) "Case Marking Choices for Experiencer Subjects," in Verma, Manindra and K.P. Mohanan, eds. Experiencer Subjects in South Asian Languages CSLI: Stanford. Amritavalli, R. (1991) "Lexical Anaphora in Kannada," manuscript Central Institute of English and Foreign Languages: Hyderabad. Austin, Peter (1981) A Grammar of Diyari, South Australia Cambridge University Press: Cambridge. Babby, Leonard (1975) "A Transformational Analysis of Transitive -sja Verbs in Russian," Lingua 35:297-332. Babby, Leonard and Richard Brecht (1975) "The Syntax of Voice in Russian," Language 51.2:342-367. Baker, Mark (1991) Incorporation: A Theory of Grammatical Function Changing University of Chicago Press: Chicago. Battistella, Edwin (1989) "Chinese Reflexivization: A Movement to Infl Approach," Linguistics 27:987-1012. Belletti, Adriana and Luigi Rizzi (1988) "Psych-verbs and θ-Theory," Natural Language and Linguistic Theory 6:291-352. Bhat, D.N.S. (1981) Pronominalization Deccan College Postgraduate and Research Institute: Pune. Bhat, D.N.S. (1991) Grammatical Relations: The Evidence Against their Necessity and Universality Routledge: London. Bittner, Maria (1987) "On the Semantics of the Greenlandic Antipassive and Related Constructions," International Journal of American Linguistics 53:194-231. Blake, Barry (1987) Australian Aboriginal Grammar Croom Helm: London.

170

171 Blake, Barry (1994) Case Cambridge University Press: Cambridge Burzio, Luigi (1981) Intransitive Verbs and Italian Auxiliaries Ph.D. Dissertation, MIT: Cambridge, MA. Burzio, Luigi (1986) Italian Syntax: A Government-Binding Approach Kluwer: Dordrecht Burzio, Luigi (1994) "Weak Anaphora," manuscript Johns Hopkins University Cardinaletti, Anna and Michel Starke (1994) "The Typology of Structural Deficiency: On the Three Grammatical Classes," manuscript, University of Venice and University of Geneva. Carter, Richard (1976) "Some Constraints on Possible Words," Semantikos 1:27-66. Chandrashekar, Siddaramaiah (1994) The Syntax of Nominative Case in Kannada Ph.D. Dissertation, University of Washington: Seattle. Chomsky, Noam (1957) Syntactic Structures Mouton: The Hague. Chomsky, Noam (1973) "Conditions on Transformations," in S. Anderson and P. Kiparsky (eds.) A Festschrift for Morris Halle. Holt, Reinhart and Winston: New York. Chomsky, Noam (1981) Lectures on Government and Binding Foris: Dordrecht. Chomsky, Noam (1986a) Barriers MIT Press: Cambridge. Chomsky, Noam (1986b) Knowledge of Language: Its Nature, Origin and Use Praeger: New York. Chomsky, Noam (1993) "A Minimalist Program for Linguistic Theory," in Kenneth Hale and S.J. Keyser, eds. The View from Building 20: Essays in Honor of Sylvain Bromberger MIT Press: Cambridge. Chomsky, Noam (1995) The Minimalist Program MIT Press: Cambridge. Cole, Peter (1982) Imbabura Quechua North Holland Publishing Company: Amsterdam. Cole, Peter, Gabriella Hermon and Li-May Sung (1990) "Principles and Parameters of Long-Distance Reflexives," Linguistic Inquiry 21:1-22.

172 Cole, Peter and Li-May Sung (1994) "Head Movement and Long-distance Reflexives," Linguistic Inquiry 25:355-406. Dalrymple, Mary (1990) The Syntax of Anaphoric Binding CSLI: Stanford. Davison, Alice (1995) "Lexical Anaphora in Hindi/Urdu," in K. Wali, K.V. Subbarao, B. Lust and J. Gair (eds.) Lexical Anaphors and Pronouns in Some South Asian Languages: A Principled Typology Mouton: Berlin. Dixon, R.M.W. (1977) A Grammar of Yidiny Cambridge University Press: Cambridge. Dixon, R.M.W. (1994) Ergativity Cambridge University Press: Cambridge. Dowty, David (1979) Word Meaning and Montague Grammar Reidel: Dordrecht. Everaert, Martin (1986) The Syntax of Reflexivization Foris: Dordrecht. Everaert, Martin (1991) "Contextual Determination of the Anaphor/Pronominal Distinction," in J. Koster and E. Reuland (eds.) Long-Distance Anaphora Cambridge University Press: Cambridge. Faltz, Leonard (1977) Reflexivization: A Study in Universal Syntax Ph.D. Dissertation, University of California at Berkeley. García, Erica (1975) The Role of Theory in Linguistic Analysis: The Spanish Pronoun System North Holland Publishing Company: Amsterdam. Geniusˇiene, Emma (1987) The Typology of Reflexives Mouton de Gruyter: Berlin. Gerritsen, Nelleke (1990) Russian Reflexive Verbs: In Search of Unity in Diversity Rodopi:Amsterdam. Giridhar, P.P. (1987) A Case Grammar of Kannada Central Institute of Indian Languages: Mysore. Grimshaw, Jane (1982) "On the Lexical Representation of Romance Reflexive Clitics," in Joan Bresnan, ed. The Mental Representation of Grammatical Relations MIT Press: Cambridge, MA. Grimshaw, Jane (1990) Argument Structure MIT Press: Cambridge, MA. Grimshaw, Jane and Sten Vikner (1995) "Obligatory Adjuncts and the Structure of Events," in Eric Reuland and Werner Abraham, eds. Knowledge and Language II: Lexical and Conceptual Structure Kluwer: Dordrecht.

173 Hale, Kenneth and S.J. Keyser (1986) A View from the Middle Lexicon Project Working Papers 10, Center for Cognitive Science, MIT. Haspelmath, Martin (1993) "More on the Typology of Inchoative/Causative Verb Alternations," in Bernard Comrie and Maria Polinsky, eds. Causatives and Transitivity John Benjamins: Amsterdam. Hellan, Lars (1988) Anaphora in Norwegian and the Theory of Grammar Foris: Dordrecht. Hermon, Gabriella (1984) Syntactic Modularity Foris: Dordrecht. Hestvik, Arild (1990) LF Movement of Pronouns and the Computation of Binding Domains Ph.D. Dissertation, Brandeis University. Hopper, Paul and Sandra Thompson (1984) "The Discourse Basis for Lexical Categories in Universal Grammar," Language 60:703-752. Huang, C.T. James and C.C. Jane Tang (1991) "The Local Nature of the LongDistance Reflexive in Chinese," in J. Koster and E. Reuland (eds.) Long Distance Anaphora Cambridge University Press: Cambridge. Jackendoff, Ray (1983) Semantics and Cognition MIT Press: Cambridge. Jackendoff, Ray (1987) "The Status of Thematic Relations in Linguistic Theory," Linguistic Inquiry 18:369-411. Jackendoff, Ray (1990) Semantic Structures MIT Press: Cambridge. Jackendoff, Ray (1991) "Parts and Boundaries," in B. Levin and S. Pinker (eds.) Lexical and Conceptual Semantics Blackwell: Cambridge. Jackendoff, Ray (1992) "Madame Tussaud Meets the Binding Theory," Natural Language and Linguistic Theory 10:1-31. Jayaseelan, K. A. (1995) "Anaphors as Pronouns," manuscript, Central Institute of English and Foreign Languages, Hyderabad. Kayne, Richard (1975) French Syntax MIT Press: Cambridge, MA. Kayne, Richard (1989) "Null Subjects and Clitic Climbing," in Osvaldo Jaeggli and Kenneth Safir, eds. The Null Subject Parameter Kluwer: Dordrecht. Kendall, Martha (1976) Special Problems in Yavapai Syntax: The Verde Valley Dialect Garland Publishing: New York.

174 Koopman, Hilda and Dominique Sportiche (1988) "Subjects," manuscript, UCLA. Labelle, Marie (1992) "Change of State and Valency," Journal of Linguistics 28:375414. Levin, Beth and Malka Rappaport (1986) "The Formation of Adjectival Passives," Linguistic Inquiry 17:623-661. Levin, Beth and Malka Rappaport (1988) "Non-event -er Nominals: A Probe into Argument Structure," Linguistics 26:1067-1083. Levin, Beth and Malka Rappaport-Hovav (1995) Unaccusativity: At the SyntaxLexical Semantics Interface MIT Press: Cambridge, MA. Li, Yafei (1995) "The Thematic Hierarchy and Causativity," Natural Language and Linguistic Theory 13:255-282. Mohanan, K.P. (1982) "Grammatical Relations and Anaphora in Malayalam," in Alec Marantz and Tim Stowell, eds., MIT Working Papers in Linguistics, Papers in Syntax, 163-190. MIT: Cambridge, MA. Napoli, Donna Jo (1979) "Reflexivization across Clause Boundaries in Italian," Journal of Linguistics 15:1-28. Pederson, Eric (1991) Subtle Semantics: Universals in the Polysemy of Reflexive and Causative Constructions Ph.D Dissertation, University of California, Berkeley. Perlmutter, David (1978) "Impersonal Passives and the Unaccusative Hypothesis," in Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society University of California: Berkeley. Pica, Pierre (1987) "On the Nature of the Reflexivization Cycle," in Proceedings of NELS 17 Univeristy of Massachusetts: Amherst. Postal, Paul (1969) "Anaphoric Islands," in Proceedings of the Fifth Regional Meeting of the Chicago Linguistics Society University of Chicago: Chicago. Postal, Paul (1977) "Antipassive in French," Lingvisticæ Investigationes 1.2:333-374. Prince, Alan and Paul Smolensky (1993) Optimality Theory, manuscript, Rutgers University and University of Colorado. Pustejovsky, James (1991) "The Syntax of Event Structure," in Beth Levin and Steven Pinker, eds. Lexical and Conceptual Semantics Blackwell: Cambridge, MA.

175 Reinhart, Tanya and Eric Reuland (1993) "Reflexivity," Linguistic Inquiry 24:657720. Roeper, Thomas (1987) "Implicit Arguments and the Head Complement Relation," Linguistic Inquiry 18:267-310. Rosen, Carol (1981) The Relational Structure of Reflexive Clauses: Evidence from Italian Ph.D. Dissertation, Harvard University: Cambridge, MA. Reuland, Eric (1995) "Primitives of Binding," paper presented at Generative Linguistics of the Old World Conference. Sadock, Jerrold (1980) "Noun Incorporation in Greenlandic: A Case of Syntactic Word Formation," Language 57:300-319. Sadock, Jerrold (1991) Autolexical Syntax: A Theory of Parallel Grammatical Representations University of Chicago Press: Chicago. Sag, Ivan (1976) Deletion and Logical Form Ph.D Dissertation, MIT: Cambridge, MA. Sapir, Edward (1911) "The Problem of Noun Incorporation in American Languages," American Anthropologist 13:250-282. Sells, Peter, Annie Zaenen and Draga Zec (1987) "Reflexivization Variation: Relations between Syntax, Semantics and Lexical Structure," in Masayo Iida, Stephen Wechsler and Draga Zec, eds. Working Papers in Grammatical Theory and Discourse Structure CSLI: Stanford. Silverstein, Michael (1976) "Hierarchy of Features and Ergativity," in R.M.W Dixon (ed.) Grammatical Categories in Australian Languages Australian Institute of Aboriginal Studies: Canberra. Sridhar, S.N. (1990) Kannada Routledge: London. Talmy, Leonard (1977) "Rubber Sheet Cognition in Language," in Proceedings of Chicago Linguistics Society 13. Talmy, Leonard (1985) "Lexicalization Patterns: Semantic Structure in Lexical Forms," in Timothy Shopen, ed. Language Typology and Syntactic Description 3: Grammatical Categories and the Lexicon Cambridge University Press: Cambridge.

176 Tenny, Carol (1987) Grammaticalizing Aspect and Affectedness Ph.D. Dissertation, MIT. Thrainsson, Hoskuldur (1991) "Long-distance Reflexives and the Typology of NPs," in J. Koster and E. Reuland (eds.) Long Distance Anaphora Cambridge University Press: Cambridge. Tirumalesh (1976) The Passive in English and Kannada: A Cross-linguistic Study M.Litt. Thesis, Central Institute for English and Foreign Languages, Hyderabad. Vendler, Zeno (1957) "Verbs and Times," Philosophical Review 56:143-160. Wehrli, Eric (1986) "On Some Properties of French Clitic se," in Hagit Borer, ed. The Syntax of Pronominal Clitics Academic Press: New York. Williams, Edwin (1981) "Argument Structure and Morphology," Linguistic Review 1:81-114. Zaenen, Annie and Adele Goldberg (1993) "Review of Argument Structure," Language 69:807-817. Zubizaretta, Maria Luisa (1987) Levels of Representation in the Lexicon and Syntax Foris: Dordrecht. Zwicky, Arnold (1977) "On Clitics," Indiana University Linguistics Club: Bloomington.

177