Long Distance Anaphora

There are a number of persistent anomalies in binding theory. One is the lack of an integrated view of long-distance anaphora. Anaphors generally require an antecedent, but languages show striking differences as to where such antecedents may occur. This volume is a collection of original articles by distinguished contributors on the nature of anaphoric systems in a wide variety of genetically and structurally different languages. It examines the general laws underlying the apparent diversity of data from the perspective of current linguistic theory. There is a surprising degree of convergence in the analyses proposed. A substantive introduction summarizes and discusses the main results, providing an integrative picture of individual and common results. This is the first representative collection of articles on this important topic. It is both conceptually coherent and of fundamental theoretical importance.

Long-distance anaphora

Long-distance anaphora Edited by

Jan Koster and

Eric Reuland Department of Linguistics, University of Groningen

The right of Ihl! Un;)'ersity oj Cambridge to print and sell all malll/er of books

was granted by Henry VIII in 1534.

The Vl/hers;ly has printed alld pi/blished contillllously sillce 1584.

CAMBRIDGE UNIVERSITY PRESS Cambridge New York Port Chester Melbourne Sydney

Published by the Press Syndicate of the University of Cambridge The Pitt Building, Trumpington Street, Cambridge CB2 lRP 40 West 20th Street, New York, NY lOOll, USA 10 Stamford Road, Oakleigh, Melbourne 3166, Australia

© Cambridge University Press 1991 First published 1991 Printed in Great Britain at the University Press, Cambridge

British Library cataloguing ill publication data Long-distance anaphora 1. Anaphora. I. Koster, Jan. II. Reuland, Eric J .

Library of Congress cataloguing in publication data Long-distance anaphora / edited by Jan Koster and Eric Reuland. p. cm. Most of the articles originated as contributions to a workshop organized by the Dept. of General Linguistics of Groningen University, June 18-20, 1987. ISBN 052139111 3 (hardback). ISBN 0 521 400007 (paperback) 1. Anaphora (Linguistics) 2. Government-binding theory (Linguistics). I. Koster, Jan. II. Reuland, Eric J. III. Rijksuniversiteit Groningen. Instituut voor Algemene Taalwetenschap. P299.A5L661991 415-dc20 90-39990 CIP ISBN 0 521391113 hardback ISBN 0 521 400007 paperback

FP

Contents

List of contributors Preface

pageix xi

1.

Long-distance anaphora: an overview Eric Reuland and Jan Koster

2.

Containment and connectedness anaphors Lars Hellan

27

3.

Long-distance reflexives and the typology ofNPs Hoskuldur Thniinsson

49

4.

Contextual determination of the anaphor/pronominal distinction Martin Everaert

77

5.

On the interaction between antecedent-government and binding: the case of long-distance reflexivization Pierre Pica

119

6.

Binding in Polish Ewa Reinders-Machowska

137

7.

Anaphors in binary trees: an analysis of Czech reflexives Jindfich Toman

151

8.

Latin long-distance anaphora Elena Benedicto

171

9.

Prepositions, binding and a-marking Alessandra Giorgi

185

Locality, parameters and some issues in Italian syntax M. Rita Manzini

209

10.

1

vii

viii

Contents

11.

Long-distance binding in Finnish Marlies van Steenbergen

231

The primacy condition of anaphora and pronominal variable binding

245

12.

E. Kiss Katalin 13.

The local nature of the long-distance reflexive in Chinese C.-T. James Huang and C.-c. Jane Tang

263

14.

Anaphors and logophors: an argument structure perspective Tanya Reinhart and Eric Reuland

283

References

323 335

Index

Contributors

Elena Benedicto Universita Autonoma de Barcelona Martin Everaert Institute for Language and Speech, University of Utrecht Alessandra Giorgi IRST, Povo Trento Jan Koster Department of Linguistics, University of Groningen Lars Hellan Department of Linguistics, University of Trondheim C.-T. James Huang Department of Modern Languages and Linguistics, Cornell University

E. Kiss Katalin Linguistic Institute, Hungarian Academy of Sciences M. Rita Manzini Department of Italian, University College London Pierre Pica Department of French, The Pennsylvania State University Ewa Reinders-Machowska Eindhoven Tanya Reinhart Department of Poetics and Comparative Literature, Tel Aviv University Eric Reuland Department of Linguistics, University of Groningen

ix

x

Contributors

Marlies van Steenbergen Dr Neher Laboratory, PTT Telecom C.-c. Jane Tang Department of Modern Languages and Linguistics, Cornell University Jindfich Toman Department of Slavic Languages, University of Michigan Hoskuldur Thniinsson Institute of Linguistics, University of Iceland

Preface

This volume is a collection of original articles on the nature of the anaphoric systems in a variety of languages and from a number of different perspectives. The aim of the editors is to provide a new impetus to the study of long-distance anaphors, a phenomenon with ramifications that are rather puzzling from the perspective of the binding theory in its canonical form (as in Chomsky (1981)). All of the contributions are concerned with extending that theory in a manner that is as restrictive as possible. Some of the solutions solidly remain within the domain of the structural binding theory. In other cases, developing a restrictive theory required recognizing different components in binding, and assigning a specific role to pragmatic factors. Many of the articles converge, however, in important respects, leading to considerable simplification of the overall picture. The patterns found in the languages covered are so consistent that they cannot be accidental (although one would certainly wish to investigate a larger number of languages). Most of the articles originated as contributions to a workshop on long-distance anaphora organized by the Department of Linguistics of Groningen University, 18-20 June 1987. The editors added a first chapter in which the main results of the volume are put together. The original plans were for this chapter to be written by the editors together with Tanya Reinhart, who participated in the workshop as a discussant. At a later stage, it turned out that part of the material collected warranted a separate contribution by Reinhart and Reuland, which is included in chapter 14. Once more, the editors found out that editing a book takes more time than one hopes it will do. A number of minor delays add up, and all of a sudden one finds oneself about a year behind schedule. We are grateful to the authors and to Cambridge University Press for their patience. The workshop from which this volume originated would have been impossible

xi

xii

Preface

without the financial support of the Netherlands Organization for the Advancement of Research (NWO) and of the Faculty of Letters of Groningen University. This support is hereby gratefully acknowledged. We are very much indebted to Hennie Zondervan for the excellent way in which she helped us to organize this workshop. We would also like to thank Liesbeth van der Velden for her help in preparing the references. Eric Reuland Jan Koster

1 Long-distance anaphora: an overview Eric Reuland and Jan Koster

1 Introduction One of the major foci of linguistic research during the last decade has been the development of a theory of binding. Despite a reasonable amount of consensus on major issues, there are a number of persistent anomalies. Especially, an integrated view of so-called long-distance anaphors is lacking. The present book sets out to contribute to the development of such a view. In the individual chapters a number of important issues in the theory of local and long-distance anaphors are analysed. The purpose of this overview is to summarize and interpret the results. In section 2 we provide the necessary background. Section 3 summarizes the individual contributions and puts them into context. Section 4 presents an overview of the facts reported. Section 5 discusses a major result of the book as a whole: the existence of just two main classes of A-anaphors.

2 Binding theory and its parameters The starting point of most current discussions of anaphora is the binding theory (BT) developed by Chomsky in a series of works from 1973 on. (1) gives the formulation in Chomsky (1981). (1) A. B. C.

An anaphor is bound in its governing category. A pronominal is free in its governing category. An R-expression is free.

The definition of 'governing category' is given in (2). (2)

b is a governing category for a if and only if b is the minimal category containing a, a governor of a, and a SUBJECT (accessible to a). 1

1

2

Eric Reuland and Jan Koster

As it has been argued that condition C is not part of the grammar (Chomsky (1982), Reinhart (1983a, 1986)), and none of the contributions in this book discusses it, we will further ignore it. It has been noted for quite some time that for anaphors in many languages condition A does not hold as stated. This is in particular true for long-distance anaphors. There are some well-known discussions in the literature of non-clausebounded reflexives (NCBR) in Icelandic (Thrainsson (1976a,b), Maling (1982) and others), and the Italian anaphor proprio (Giorgi (1984)). This led to a number of proposals to modify the binding theory, for instance Koster (1985, 1987), Manzini & Wexler (1987), and Wexler & Manzini (1987). These theories express two claims: (1) languages may differ in the binding domains of their anaphors; (2) within one language different anaphors may have different binding domains. The common assumption in these studies is that all anaphors share the same type of binding relation. The differences are captured by parametrizing the opacity factor defining the governing category, as in (3). (3)

b is a governing category for a if and only if b is the minimal category containing a, a governor of a, and F (F an opacity factor).

The generally accepted definition of binding is given in (4). (4) a binds b iff a and bare coindexed and a c-commands b. F may assume values such as (accessible) SUBJECT, Tense, Agr, or Compo These opacity factors are taken from a universal set, with particular anaphors differing in the value selected. This choice is represented in the lexical entry. Similarly, languages may differ in the opacity factors they make available. Anaphors with an opacity factor beyond the SUBJECT are classified as long-distance anaphors. So far, however, opacity factors have been represented as arbitrary features; hence, no principled restrictions on the set of opacity factors have been developed. In the absence of such restrictions the theory predicts virtually unlimited possibilities for anaphors to differ. In the survey of the languages discussed in the volume which is presented in section 3, we will see to what extent this prediction is borne out. Condition B raises a question of a different kind, namely, why it is so constant across languages. The original formulation in Chomsky (1981) left some empirical gaps. But Huang's (1982) modification that the opacity factor for pronominals is SUBJECT, rather than accessible SUBJECT, comes close to being descriptively correct. It is presently quite unclear why the opacity factor for pronominals does not vary, and why the complementarity between pronominals and anaphors, which

Long-distance anaphora: an overview

3

is generally quite striking, breaks down in some constructions, especially in languages with long-distance anaphors. Since complementarity facts will bear on the assessment of the anaphoric system in a language, they are included in the survey below.

3 The contributions and their implications Many of the contributions contain features that cannot be easily accommodated in a binding theory modelled on (1-4). However, they are often quite compatible with each other. 3.1 Argument structure and binding One of the important issues in current theory is the relation between argument structure and binding. This issue is addressed in a number of contributions. In chapter 2 Hellan discusses the anaphoric systems of a number of languages, focussing on Norwegian and Icelandic, but also including Italian, Dutch, Japanese and Chinese. He proposes that there are two types of conditions on anaphoric relations: containment conditions and connectedness conditions. Containment conditions are in effect conditions on the binding relation. They can be viewed as generalizations of the c-command requirement in (4). They express that the antecedent a must bear a certain prominence relation to the maximal constituent which contains the anaphor b but does not contain a (max(b)). In the case of c-command this relation is purely structural (sister of). The cases discussed by Hellan involve narrower requirements, for instance, that max(b) must be predicated of a, that max(b) is an argument of the same verb a is an argument of, or the (negative) requirement that max(b) does not contain a Tensed S containing b, etc. Connectedness conditions affect a different part of the binding theory. Hellan argues that for local binding relations binder and bindee must be connected by being co-arguments of a lexical head. This makes the structural notion of a governing category irrelevant for local binding. This modification ties in with other proposals. Everaert establishes that there is a direct connection between a-theory and the selection of anaphors. Kiss shows that anaphor binding must be sensitive to the hierarchy of thematic roles. In a different way, also Giorgi shows that the a-assigning properties of the head governing the anaphor are crucial. A synthesis of these findings is attempted by Reinhart and Reuland, who propose that local binding involves the manipulation of the thematic grid ofthe predicate of which binder and bindee are arguments. Hellan also discusses discourse-dependent anaphora, involving the condition

4

Eric Reuland and Jan Koster

that max(b) must be in the perspective of a. b is in the perspective of a if a is the person from whose point of view the report containing b is made. This instantiates the more general phenomenon of logophoricity: the element binding b must bear one of a number of specific roles in the discourse structure (see Sells (1987) for an analysis of the notion of logophoricity, and Reinhart & Reuland for further discussion in the present book). Logophoric binding does not require that the antecedent is structurally represented in the sentence, so the anaphoric element is allowed to be free (as discussed by Thniinsson). 2 It seems that logophoric binding relations fall outside the domain of the structural binding theory, but their existence is important for determining the number of binding domains that have to be distinguished and the nature of their difference. Standard binding theory just takes it for granted that there is a distinction between anaphors and non-anaphors. But Thrainsson, in chapter 3, provides a detailed analysis of the referential properties of NPs introducing much finer distinctions, which standard binding theory is unable to deal with. He provides a typology of reflexives based on comparison of their behaviour in the group of Scandinavian languages. This leads to a theory in which the anaphoric character of an NP can be accounted for in terms of specific lexical properties. Like Hellan, Thrainsson shows that structural requirements represent necessary, but not always sufficient, conditions on anaphoric binding. By careful consideration of the data he leaves no doubt that in certain cases of LD-reflexives the binding requirements cannot be stated in purely structural terms, and that some need not even be syntactically bound at all. The characteristic of such reflexives is that they are incapable of independent reference for purely lexical reasons. Instances of reflexives that are bound from outside a finite clause, and those that , are not syntactically bound at all, form a natural class. This leads to a typology in which the feature [+/- independent reference] is a separate parameter. In the binding theory of (1-4), the complementarity between pronominals and anaphors is accounted for in a largely stipulative manner. One of Everaert's major insights is that complementarity phenomena are essential for our understanding of how binding works. In his contribution in chapter 4 he discusses facts from a wide range of languages while focussing on a comparison of Dutch and Frisian. He notes that a coherent notion of long-distance anaphora has not been established so far, and starts from the null hypothesis, namely that one binding theory suffices for non-Iogophoric anaphors. The distinction between local and non-local anaphors is represented in the requirement that certain anaphors must be free in their minimum governing category, and in the conditions on complementarity between anaphors and pronominals. Of special importance is the fact that the non-local anaphor zich in Dutch may be locally bound in the same environments in which

Long-distance anaphora: an overview

5

the pronominal 'm may be locally bound in Frisian. His finding that the relevant conditions are stated in terms of 9-government constitutes a major step forward. It provides a crucial link between binding theory and argument structure. These results are quite compatible with much of Hellan's approach, and can be viewed as filling in the connectedness module. Although Everaert's contribution itself still uses the notion of a governing category, it effectively contains the considerations needed to abolish it as an independent theoretical concept. His contribution contains an illuminating discussion of reciprocals. The relation between binding and argument structure is also the topic of Kiss's contribution in chapter 12. Kiss discusses anaphora and variable binding in Hungarian, comparing these with their counterparts in English. Kiss argues that due to the flat structure of Hungarian all arguments in a clause c-command each other. Yet, there are binding asymmetries mirroring those of English. Kiss concludes that in general the binding hierarchy is only in part a reflection of formal properties of the structure, such as c-command and precedence. Relative prominence of roles in the thematic lexical argument structure must be another important factor. She argues that this factor is operative not only in Hungarian, but also more generally. The asymmetries among PP complements in English (with NP versus about NP) cannot be explained in terms of differences in c-command, but must be stated in terms of such a hierarchy. Giorgi's contribution in chapter 9 analyses anaphora in PPs, focussing on the long-distance anaphor proprio and the local anaphors se and se stesso. 3 Proprio can be clause bound, or LD-bound. In the latter case it is subject oriented. Se is always subject oriented. It cannot be governed by a verb, but must be governed by a preposition; se stesso is clause bound, but without further restrictions. Proprio is like se when LD-bound, but like se stesso when clause bound. When subject bound these anaphors are not in complementary distribution with pronominals. The properties of se are reminiscent of Dutch zich. This suggests a connection with the theory of zich developed by Everaert (see Reinhart & Reuland for further discussion) . Discussing anaphora in structures of the form [NP XP], Giorgi makes the following important observation. If the XP contains a subject-oriented anaphor and it is an AP, the NP is a possible antecedent, but if the XP is a PP it is not. On the basis of an extensive discussion of 9-marking properties of APs and PPs, she argues that the effect is due to a difference in the following respect: APs may assign an external 9-role, PPs cannot. Hence, when construed with a PP, the NP cannot count as prominent in the relevant sense. The relation between the NP and the PP cannot be the formal predication relation. Interpreted along these lines her contribution provides further support for the 9-related nature of anaphora. There is also

6

Eric Reuland and Jan Koster

another possible interpretation: APs exhibit agreement, and PPs do not. If subjectoriented anaphors required an Agr, the same pattern would follow. Establishing which is the correct factor will be important for a proper understanding of the Italian system of anaphors. 3.2 Binding and movement Standard binding theory has little to say on the nature of the opacity factors. It essentially stipulates that certain categories are opaque for binding. In recent years, movement theory has become increasingly articulated. Building on earlier proposals (Lebeaux (1983), Chomsky (1986a)), Manzini, Pica, and Huang & Tang set out to derive a number of properties of binding from the theory of movement discussed in Chomsky (1986b). These proposals deal with the mechanics of binding, i.e. the question of how coindexing is brought about. The discussion in Hellan, Thniinsson, Everaert, Kiss, and Giorgi involves higher-level issues. It is an important question to what extent the various higher and lower-level analyses are compatible. Faltz (1977) and Pica (1985, 1987) have observed that there are systematic differences between simplex (mono-morphemic) and complex anaphors. Anaphors of the former type may be non-local, those of the latter type are always local. In his earlier work Pica attempted to give a full explanation of binding in terms of movement. In his present contribution (chapter 5), Pica argues that a full reduction cannot be achieved, but that the differences between local and non-local binding follow from the way in which the theory of movement proposed in Chomsky (1986b) interacts with the binding theory proper. Cross-linguistic variation between reflexives can be reduced to differences in their lexical properties. No reference to parameters is necessary. Central is the idea that reflexives are subject to cyclic movement from Infl to Infl through Compo Mono-morphemic anaphors like sig are full-argument NPs; hence the term 'argument anaphors'. If at LF they move up to a position governed by their prospective antecedent, intermediate traces may delete. So, this movement will not be generally clause bounded. In the case of complex anaphors like himself (adjunct reflexives in his terms), the specifier him moves up. Him is an adjunct, hence intermediate traces may not delete. Therefore, movement will be clause bounded. Similar considerations apply to clitic reflexives, which are not arguments either. a-theory and argument structure are only indirectly involved. It is not clear how Pica's approach is to be reconciled with the facts and insights put forward by Hellan, Everaert, and Kiss. In chapter 10 Manzini investigates the hypothesis that the binding conditions reflect conditions on movement, and that all locality conditions can be ~nified under the notion government. In particular she investigates whether the natural

Long-distance anaphora: an overview

7

minimal binding category for an anaphor is the first maximal projection containing it (see also Koster (1987». She argues that binding of anaphors in verb complement position can be brought under that conception, but for binding of anaphors contained in NPs presence or absence of a subject is relevant. Also, the fact that anaphors may differ as to their binding domain shows that opacity factors must enter into domain specifications to express the parametrization (see (4». Her analysis is based on a detailed discussion of inalienable possession and reciprocals in Italian. She shows that when l'uno l'altro 'each other' is part of an NP, the opacity conditions on l'altro are subject based, those on l'uno barrier based. She then argues that l'altro is an A-anaphor, while l'uno is an A-bar anaphor. Pica and Manzini converge in the conclusion that a full reduction of binding theory to principles of movement has not been achieved. As in the case of Pica, it is still to be determined to what extent Manzini's approach ties in with the results obtained by Hellan, Everaert, and Kiss. This is especially relevant, since these authors have nothing to say about A'-anaphors. However, if it is in general true that A'-anaphors are sensitive to barriers and A-anaphors to other conditions, this raises the question to what extent the theory of the two should be unified. As we noted earlier, most theories have nothing specific to say on the question of what makes an anaphor into an anaphor. To our knowledge, Bouchard (1984) has been the first to claim that NPs must have I Rather than attributing this circumstance in (12) to the lack of perspective command between Jon and sig, one might see it as an effect of a connectedness condition on sig, since sig in (12), or REFL in (14), is not inside of an argument of the verb taking the binder as argument. What such a proposal leaves unexplained is how (12) can be good with H araldur as binder. True enough, sig is then inside an argument C of the verb of which Haraldur is external argument, namely the bracketed constituent in (12), but the 'argument-of' chain stops at this argument C: the verb heading this argument (i.e. komi) has no argument-taking relation to what follows, i.e. to the nema-clause containing sig. So, if this alternative approach can hope to dispense with the perspective-command restriction on sig, this can happen only at the cost of introducing another containment condition: that sig be contained inside an argument of the same verb as the binder is an argument of. Let us call this putative relation argument command between the binder and sig. (4), then, clearly holds true also on this analysis. Further examples show that argument command cannot generally replace perspective command. Consider (15): (15)

Joni heldur a5 Mariuj hafi veri5 sagt a5 pu tala5ir urn sigi/*j Jon thinks that Maria has been told that you talked about REFL

36

Lars H ellan

In this example, Manu cannot bind sig, even though the verb segja 'tell' takes the ao-clause as argument, and Manu hence argument commands sig. What is missing in this case is arguably perspective command between Manu and sig: Manu, as a passivized NP, does not carry a role which is naturally construed as corresponding to a perspective holder with regard to the ao-clause. This indicates that perspective command is the necessary condition on NCBR in Icelandic, not argument command. Still further examples seem to indicate that both types of command may in fact playa role in Icelandic NCBR. Maling (1984) notes the contrast between (16a) and (16b) (mentioned also in Sells (1987»: (16)

a.

*Olafuri hefur ekki enn fundiD vinnu sem Seri likar Olafur has not yet found a job that REFL likes b. Joni segir ao Olafurj hafi ekki enn fundiD vinnu sem Seri/j likar Jon says that Olafur has not yet found a job that REFL likes

In (16a), the reflexive is inside a relative clause. Although this clause is part of an NP serving as co-argument with the putative binder, the binding relation is impossible, confirming the position that argument command is neither necessary nor sufficient in order for NCBR to obtain. In (16b),Jon is a possible binder, again as we would expect, since Jon here perspective commands the reflexive. However, in this case, the reflexive can be found by OZajur as well. This contrasts with the illformedness of a binding between Jon and sig in (12). The only difference between these cases is that in (16b), argument command after all does obtain between OZajur and the reflexive, while it does not in (12) between Jon and sig. Thus, it seems that once a certain domain is perspective commanded by some item X, certain operations can be licensed by argument command within this domain, even if the argument commander is not X. This may then indicate that argument command does have a function in Icelandic NCBR, but only a secondary function compared with perspective command. And with regard to the principle (4), since both these command relations are containment relations, (4) is supported. Another circumstance which might seem to suggest that Icelandic NCBR is a connectedness phenomenon is the frequent use of subjunctive mood on verb forms interconnecting binder and bindee. The constellation 'V ... V[ +subj]' might be seen as a manifestation of a government relationship between the two verbs (or between the first verb and the VP headed by the second), and on that view, NCBRs in Icelandic would seem to be connected in exactly the sense (5), once we count any governee as an argument of the governor. 2 However, it appears that subjunctive mood is neither necessary nor sufficient in order for NCBR to occur. That it is not necessary is argued in Sigurosson (1986a); he adduces

Containment and connectedness anaphors

37

examples which look quite like those above, except that no subjunctive occurs. That subjunctive alone is not sufficient is shown by examples like (17), where the subjunctive is induced by nema. (From Thniinsson (1976a); cf. (12).) (17)

*J6ni kemur ekki nema Maria kyssi sigi Jon comes not unless Mary kisses REFL

In (17), in accordance with what we have already said, neither perspective command nor argument command obtains; and in the examples used by Sigurasson, perspective command does obtain, serving as the licenser, according to Sigurasson. Thus, the presence or absence of subjunctive has no effects independently of the factors discussed above, and so is not critical to the use of NCBR. Rather, as indicated earlier, the presence of subjunctive should be seen at least partly as a reflex of the perspective-command relation, and is relevant for NCBR primarily in virtue of being a means by which perspective command has been grammaticalized. This concludes our demonstration that (4) holds both for Norwegian and Icelandic long-distance anaphors. The licensing containment conditions for these anaphors are all phrased in terms of what we have called 'command' relations, which is to say that the binder must have a superiority status with regard to the anaphor, identifiable on either syntactic or semantic grounds. The phenomenon of longdistance anaphora thus seems to highlight in a very clear-cut fashion the existence of these command relations, or what we may call the command module. These command relations interact with connectedness conditions, or the argument . module, in the cases where the LDA-element occurs together with selvlsjrilfan, and the simple insight stated in (4) is that LDA is what results when this extra element is omitted, i.e. when the command relations are the sole domain-defining factors in the anaphors. This having been established for Norwegian and Icelandic, a question is still whether the only way in which an LDA can emerge, cross-linguistically speaking, is through its constituent element(s) being associated with command-type conditions only. Our conjecture, as expressed in (4), is in the affirmative, but limitations of space and time preclude an investigation of this in the present chapter. We confine ourselves to just reviewing data from a few languages in the perspective of (4). 2.3 Other long-distance anaphors

2.3.1 Italian proprio The possessive reflexive proprio in Italian undergoes NCBR, as shown in example (18) (from Giorgi (1984)):

38

Lars H ellan

(18)

Gianni j ritiene che Osvaldoi sia convinto che quella casa Gianni believes that Osvaldo is persuaded that that house appartenga ancora alla propriai/j famiglia belongs still to REFL's family

The main restriction on proprio as an NeBR, according to Giorgi, is that it be in the P-domain of its binder, where for X to be in the P-domain of Y means that X is contained in some item Z such that Y and Z get their O-roles from the same O-roleassigner, and Y's O-role ranges higher than Z's O-role on the hierarchy '(1) Agent, (2) Experiencer, (3) Theme'. Using a term from Hellan (l986a, 1988) we will say that in such a case, Y role commands X. The relevance of this command relation has been demonstrated for more locally bound anaphors in languages such as Hungarian (Kiss (in preparation» and Norwegian (Hellan (1986a, 1988». It is distinct from the command relations established so far, and appears to be the only one in the family relevant for proprio as an LDA. First, in contrast to predication command, it does not entail a c-command relation between binder and bindee, and thus allows (19), a construction whose counterpart with seg or sig is ill-formed in Scandinavian (for seg, the relevant example would have to have an infinitival subject clause): (19)

La salute di quelli che amano la propriai moglie preoccupa molto Osvaldoi a lot Osvaldo the health of those who love REFL's wife worries

Secondly, the role-command relation entails argument command, and so the rolecommand requirement suffices to rule out (20), which exemplifies constellation (14).

(20) *Osvaldoi ritorna III patria prima che il fisco Osvaldo returned to his country before the public treasury sequestrasse il proprioi patrimonio sequestered REFL's estate Role command is presumably entailed by perspective command, since perspective holders are likely to be higher in the O-hierarchy than the material within the perspective. However, the entailment clearly does not go the other way - the higher one among two given roles need not be a perspective holder. In sum, then, role command is an additional member to the set of command relations. Since in all of the well-formed examples considered, the binder could be interpreted as a perspective holder, and in (20) not, a possible alternative analysis might be that proprio be perspective commanded by its binder; this question cannot be decided here, however, so we stick to Giorgi's analysis for the present.

Containment and connectedness anaphors

39

(4A) thus being true of proprio, the question is whether role command is associated with proprio as an essential condition in the way predicted by (4B). The literature does not give any explicit answer so far, so we leave this point open. 3

2.3.2 Japanese zibun The behaviour of the Japanese reflexive zibun as an LDA is discussed in a number of works (e.g. Kuno (1973), Kuno & Kaburaki (1977), Kuroda (1965), Inoue (1976), Farmer (1980), and many others). It can be bound out of clauses, as exemplified in (21) (from Inoue (1976: 163», hence it allows NCBR: (21)

Tarooi ga Hanako ni zibuni Amerika e itta koto Taroo Hanako REFL America to go-PAST that 'Taro did not tell Hanako that he had been to the States'

0

hanasanakatta tell-NEG-PAST

It is commonly assumed (see, e.g., all of the works cited, and Sells (1987» that perspective command plays an essential licensing role for NCBR in Japanese. A perspective-command requirement licenses zibun in (22) (an example similar to (19) in Italian above), the well-formedness of which shows that predication command is not a necessary condition (Inoue (1976: 171»: (22)

Zibuni ga Mary ni karakawareta koto ga Johni 0 zetuboo e oiyatta SELF Mary was-made-fun-of that John desperation to drove 'that self (= he) was made fun of by Mary drove John to desperation'

Perspective command furthermore suffices to rule out (23) (instantiating constellation (14», rendering 'argument command' just as unnecessary as it was for proprio in Italian (Kuroda (1965»: (23)

*Takasii wa [Yosiko ga mizu 0 zibuni no ue ni kobosita toki] nurete simatta Takasi Yosiko water REFL 's on spilled when wet got 'Takasi got wet when Yosiko spilled water on him'

Reportedly (Kuroda (1965); see also Sells (1987», substituting node 'because' for toki 'when' in (23) yields a well-formed result. Assuming that this item turns the adverbial clause into a perspective domain for Takasi, the requirement of perspective command then licenses also this new construction. This time, a role-command requirement would not have such a licensing function, and this will be an indication that it is perspective command and not role command which is the crucial notion for zibun as an LDA. As with Italian, we are now hypothesizing that it is just a single command relation which is relevant for zibun as an LDA, as opposed to the situation in Icelandic. This may be an oversimplified picture, due to the incompleteness of the

40

Lars H ellan

data chosen. Also, if a single relation is to be selected, one might hope for something more tangible than perspective command, which is probably the most elusive of those considered. Let us still, for the present, assume that perspective command is the licensing-containment condition for zibun as an LDA, and role command for

proprio. (4A) being true of zibun, the next question is whether (4B) holds, i.e., whether perspective command acts as a condition also on more locally bound occurrences of zibun. An example like (24) (from Farmer (1981: 178)) indicates that some condition within the family we have discussed holds, but it does not determine whether it is predication command, role command or perspective command: (24)

Tarooi wa kodomoj 0 zibuni/*j no kutu de butta REFL 's shoe with hit Taroo child 'Taroo hit the child with self's shoe'

(25a) lets through the same possibilities, whereas the relative acceptability of the 'j' -indexed version of (25b) is difficult to view as licensed by predication command; both of the other relations obtain, however (from Inoue (1976: 148)): (25)

a. Oyazii wa bokuj ni zibuni/j no kuruma 0 Dad me REFL 's car 'Dad made me wash self's car' b. Bokui wa oyazij ni zibuni/(j) no kuruma 0 1 Dad REFL 's car 'I was-caused-to-wash self's car by Dad'

arawaseta made wash arawaserareta was made wash

At the present point we are not able to press these considerations any further, but there is clearly a possibility that (4B), and thus (4) as a whole, will turn out to be true. (See some further remarks below.) 2.3.3 Chinese ziji As a final example of anaphors with clear long-distance properties, we mention the non-contrastive variety of the Chinese reflexive ziji, basing our discussion on that in Xu (forthcoming). Examples like the following show that ziji can attain NCBR, (b) indicating that only subjects can serve as binders.

(26)

a. Lao Zhangi ting shuo Xiao Iij rna ZlJli/j Lao Zhang hear say Xiao Li swear REFL 'Lao Zhang heard Xiao Li swore at himself/him' b. Johni xiangxin Billj dui Samk shuo zijii/j/*k taoyan Mary John believe Bill to Sam say REFL hate Mary 'John believes that Bill said to Sam that he hated Mary'

Containment and connectedness anaphors

41

According to Xu, the requirement that the binder be a subject holds generally, i.e. also when binder and bindee are inside the same minimal clause. In our terms, these observations will indicate that the containment condition of predication command is the one operative in Chinese NCBR, and, in accordance with (4), that this condition is inherently associated with ziji. Provisionally, then, we interpret the Chinese data too as potentially confirming hypothesis (4).

2.3.4 A possible weaker version of(4), and criteria of being a long-distance anaphor With the discussion of proprio, zibun and ziji in mind, there might be cases of LDAs conforming to (4A), but where the containment condition(s) obeyed by the relevant anaphoric element in local uses is/are different from the one(s) in operation in long-distance uses. For instance, perspective command could be the relevant notion for zibun as an LDA, but role command the one governing locally bound occurrences. If we assume that the command relations constitute a natural class, a weaker, but still interesting version of (4B) would be to say that, if (4A) holds, then some command relation must also govern the locally bound occurrences of the anaphoric element in question. It might in turn be possible to establish a ranking among the possible command relations. Concerning (4A), the question will arise as to exactly what the criterion should be for a binding relation to count as 'long distance'. Consider the Dutch anaphor zich. Apparently, this anaphor allows long-distance binding, judging from the possibility of (27); on the other hand, although putatively long-distance binding of zich must have a subject as its binder, zich does not induce any predication requirement in locally bound occurrences, like in the anaphor zich zelf, as evidenced by the well-formedness of (28): (27)

dat Mariej Piet voor zichj liet werken that Mary Peter for her let work 'that Mary let Peter work for her'

(28)

Ik vroeg Pietj over zichzelfj I asked Peter about REFL

The question is whether zich should be counted as an LDA, and perhaps satisfying (4A), but then refuting (4B) (even in its weaker version). Our proposal is that zich should not count as an LDA in the first place. As witnessed by the ill-formed examples in (29), zich is peculiarly restricted in its binding domain; even in a language with only moderate use of LDA like Norwegian (where Tensed S cannot be crossed), the counterparts of all of these examples are well-formed:

42

Lars H ellan

(29)

a. *Jan; liet Marie zich; ophangen Jan let Mary him hang 'Jan let Mary hang him' b. *Jan; liet Marie zich; respecteren Jan let Mary him respect 'Jan let Mary respect him' c. *Marie; vroeg ons (om) PRO over zich; te praten Marie asked us (COMP) about REFL to talk d. *Jan; vroeg mij Marie voor zich; te laten werken Jan asked me Mary for him to let work 'Jan asked me to let Mary work for him' e. *dat Jan; het schot op zich; betreurde that Jan the shot at REFL regretted 'that Jan regretted the shot at himself' f. *dat Jan; Marie verliefd op zich; achtte that Jan Marie in love with REFL considered 'that Jan considered Marie in love with himself'

Clearly, zich lacks the basic freedom to occur long-distance bound which has been manifest in all the other cases considered so far - it is not 'essentially non-local', in the terms from 1.1, and so does not naturally fall within the group constituting the clear cases of LDA. And if zich is not an LDA, (28) ceases to be a counterexample to (4B). What is zich then? We propose that it is a connectedness anaphor with a binding domain extending beyond co-argumenthood, as our definition (5) renders possible. This is in line with analyses presented in Everaert (l986a) and Koster (1987), a basic point of which is that there must be some kind of uninterrupted connection, describable in terms of government, extending from zich to its binder. Thus, in Koster (1987), on the assumption that 'regret' and 'consider' do not govern their complement, (2ge,f) are accounted for, and on the assumption that complementizers interrupt the relevant connection and that infinitives can only be governed to the left in Dutch, (29c) is accounted for. While (29a, b) require more specific stipulations and (29d) remains unaccounted for, the picture emerging is that zich must be 'connected' to its binder, in the sense of the notion of 'connectedness' stated above. Clearly, now, on the basis of this line of analysis and examples like (27) looked at in isolation, zich might seem like an LDA subject to connectedness conditions, in strict violation of (4A). By the reasonable requirement on the status as LDA just proposed, however, zich cannot count as an LDA, thus saving (4A); (4B) is left

Containment and connectedness anaphors

43

unaffected in turn. 4 (Notice that although the definition of LDA plays a crucial role here, we are not providing a definition which renders (4) automatically true: we are just suggesting a limitation on the empirical range of the notion LDA, about which (4) still makes an empirical claim.) 2.4 Concluding remarks We have seen that all ofthe LDAs seg, sig, proprio, zibun and ziji obey containment conditions when long-distance bound: seg, sig and ziji must all be predication commanded by their binder; sig and zibun must be perspective commanded; proprio must be role commanded. Among these anaphors, seg and sig are those on which hypothesis (4), repeated below, is based. (4) A. B.

Long-distance anaphors (LDAs) obey containment conditions exclusively. The elements of an LDA obey the same containment conditions, or a subset of them, when they are elements of a more locally bound anaphor as they do in long-distance binding.

The results mentioned lend initial plausibility to (4A) as a universal hypothesis, which in turn shows the significance of the command module. (4B) holds for the Scandinavian reflexives, and our cursory survey of the other anaphors indicates that this proposition too may hold universally, but this is much less clear than in the case of (4A). (4) yields one expectation which is fulfilled: the fewer elements which serve as constituents in an LDA, the greater is the possibility that the conditions associated with the anaphor will be purely of one type, here the containment type. From this perspective, the optimal LDA will consist of one element only; and this holds true of all the LDAs we have considered. Non-LDAs, on the other hand, are often complex, like seg seiv, with selv inducing the connectedness conditions. 5

3 Differences between containment and connectedness anaphors in certain interpretive processes We now address the distinction between containment and connectedness conditions from a different angle. Returning to Norwegian, we observe two contrasts between the connectedness anaphor seg selv and the containment anaphors seg and sin, both in the domain of semantic interpretation. We will propose a unifying principle for these contrasts phrased crucially in terms of the containment/connectedness distinction.

44

Lars H etlan

3.1 VP-anaphora It appears that in VP-anaphora constructions involving containment anaphors, like those in (30), it is possible to interpret the VP-anaphor both with a 'sloppy' and a 'strict identity' reading concerning the reference of the reflexive. (30)

a.

b.

Johni snakker ofte om bilen sinh og det gjj2lr de pa John talks often about car his and it do those at verkstedet ogsa the garage too 'John often talks about his car, and so do the people at the garage' Johni hadde hj2lrt meg snakke nedsettende om segi, og det John had heard me talk depreciatorily about him and it hadde de som stod rundt ogsa had those who stood around too 'John had heard me talking depreciatorily about him, and so had those who were standing around'

That is, in (30a), both a reading on which the people at the garage talk about John's car (,strict identity'), and one where they talk about their own car ('sloppy identity'), are possible. In (30b), the people standing around could have heard me talking depreciatorily about John ('strict identity') or about themselves ('sloppy identity'). On the other hand, if seg selv is used in such a construction, like in (31), a 'sloppy identity' reading is highly preferred. (31)

Joni skryter av seg selvi, og det gjj2lr Marit ogsa Jon boasts of himself, and it does Marit too

The way in which the containment-connectedness distinction will be applied in the following analysis of this contrast is tied to a particular formal analysis of binding and VP-anaphora; but the distinction could probably be construed as relevant also with other formal construals. We assume that the difference 'strict''sloppy' identity in such constructions is to be analysed as follows: in a given structure, occurrences of reflexives and pronominals are annotated for their binding relations, and are subject to rules checking whether these relations are admissible. The interpretation of constructions with VP-anaphoraldeletion involves a process of copying the VP from the first conjunct onto the emptyianaphoric VP of the second conjunct. The rules of assigning and checking binding relations can take place either before or after this copying. If they apply before, the reflexive is marked as being bound by an NP in the first conjunct; this means that the 'copied' instance of the reflexive, i.e. the occurrence contained in the copied VP, will be marked as having the NP in the first conjunct as its binder. This yields the 'strict

Containment and connectedness anaphors

45

identity' reading. If the rules for assigning and checking binding relations apply after the copying of the VP, the reflexive inside the copied VP gets assigned an NP in the second conjunct as its binder - this yields the 'sloppy identity' reading. (See Hellan (1988) for details.) Apparently, now, only the latter option is open in a case like (31), since it allows only a 'sloppy identity' reading. The question is why. One imaginable possibility might have been to ascribe seg selv some sort of inherent variable status, while seg/ sin could have a freer status, and pin the possibilities of readings on this difference. Regardless how such a proposal were to be made explicit, however, it would seem to be on the wrong track, since sin can obtain in a 'strict identity' construction even when it is bound by an NP like 'each participant', the prima facie kind of construction where an anaphor would be expected to behave like a 'variable':6 (32)

Hver deltageri var forhindret each participant was prevented naboen hans matte gjpre det neighbour his had to do it

fra a hente premien sini, sa from fetching trophy his, so istedet instead

The approach sketched above provides another possibility of analysis, and this will be our proposal. We will assume that when a binding relation is connected, it is resistant to being 'interrupted' in the way which happens when the copying process removes the reflexive without bringing along the binder. A connected reflexive thus is connected to its binder in a very concrete sense: the bindee cannot be displaced without its binder following it. No such requirement applies to containment anaphors, hence the differences in the interpretation of (30) and (31). This hypothesis has to face the fact that connected anaphors can easily be 'moved away' from their binder in a very overt fashion, like in (33): (33)

a. b.

Seg selvi himself Beundre admire

jeg at vet know I that seg selvi vet himself know

J oni beundrer Jon admires jeg at Joni gjpr I that Jon does

Apparently, the existence of a trace, or the reconstruction process/relation thereby encoded, makes this case count as a non-intrusion in the respect in question. Consequently, the ban against displacement of connected anaphors relative to their binder holds only with regard to 'non-visible' displacement. Once the ban is associated with connected anaphors in general, it follows that also 'non-local' connected anaphors, such as Dutch zich, should be ineligible for strictly read VP-anaphora. According to Everaert in chapter 4, this prediction is

46

Lars H ellan

borne out; but further investigations at a cross-linguistic level are obviously needed to assess fully the validity of our hypothesis. 3.2 Speaker construal An analogous situation to the one now described may be seen to arise in cases where a given part of a structure is subject to construal on the part of the speaker rather than by (the referent of) an NP in the structure? In (34), the part sin egen kone 'his own wife', with the reflexive sin, is most reasonably read as a description supplied by the speaker, not as one representing Jon's or Peter's construal of the situation (that is, Jon observes Peter's interest in a certain woman, without knowing that they are in fact married, and interprets what he sees as a wish on Peter's part to marry the woman); (34)

Jon tror at Peterj viI gifte seg med sinj egen kone Jon thinks that Peter will marry his own wife

Sin is here a containment reflexive. Consider now what happens when a connected reflexive is involved, as in (35): (35)

Jon tror at Peterj snakker om seg selVj Jon thinks that Peter talks about himself

Here it appears very difficult to interpret the description seg selv as supplied by the speaker and not by Jon. (That is, it is difficult to apply (35) to a situation where Jon hears that Peter speaks about somebody, and only the speaker is aware that this somebody is Peter himself.) Various explanations could be imagined for this, but one possibility is that the exclusive speaker construal of seg selv in (35) is blocked by the connected binding relation in the same way as the copying process mentioned above is. This means that 'exclusive speaker construal' must be understood as a process at the interpretive level which somehow 'cuts loose' some expression and assigns it a higher scope than its syntactic position would indicate. Various conceptions of 'raising at LF' suggest themselves as implementations of this idea. For our purposes, it is enough to point to the 'invisibility' of such a process as a property it would share with the copying assumed in VP-anaphora constructions (which could also be taken to be an LF-process), and state the essential characteristic of connected binding relations with regard to both processes: that they cannot displace the anaphor from the domain of its binder. If correct, this analysis lends further support to the distinction between connectedness and containment anaphors, and entails that it be represented at the level (e.g. LF) where scope relations are represented.

Containment and connectedness anaphors

47

4 Concluding remarks The purpose of this chapter has been to provide motivation for drawing a distinction between containment and connectedness conditions on anaphors. While the latter fit into models largely based on locality notions of the type 'govern' or 'argument of', such as the model of Chomsky (1986), the containment conditions represent an entirely different paradigm of principles: their common denominator, apart from 'containment', is the superiority notion command, whose instances comprise predication command, role command, perspective command, c-command, perhaps argument command, and possibly others. We propose that these notions be regarded as constituting a module on their own in the grammar, namely the command module. As has been pointed out, this module is not restricted in its application to long-distance anaphora, but is observable in local anaphora as well; and the main hypothesis of section 1 of this chapter is that long-distance potential of an anaphor presupposes (1) that it obeys command-relational conditions only, and (2) that its constituting element(s) is/are subject to the same commandrelational conditions when it/they constitute(s) locally bound anaphors. Apart from the cross-linguistic investigation of this hypothesis, another topic for further research will be in what other areas of the grammar the command module is operative. The proposals of section 2, likewise, are open for investigation, both cross-linguistically and with respect to the nature of LF.

Acknowledgements This paper has benefited from comments by Lars Johnsen, Anneliese Pitz, Tor Afarli, and the audience at the Workshop on Long-Distance Anaphora at Groningen in 1987.

Notes 1. See Hellan (1988) for a treatment of Norwegian non-argument reflexives. 2. This is the idea of, e.g., Koster (1987). 3. In a full discussion of anaphors in Italian, also the reflexives se and se stesso must be included. See Giorgi (1984 and 1986), the latter addressing facts of NP-internal binding with possible relevance for the proposition (4B). 4. What is not yet accounted for is why the 'stretched' connectedness-type zich apparently only takes a subject as binder. Whether this will follow in any way from principles concerning the connectedness group of relations is a matter we cannot go into here. 5. Given that self in English induces connectedness conditions as well, and is the only reflexive element in English, it follows that English has no long-distance anaphors. English and other reflexives have been noted to display certain long-distance effects

48

Lars H ellan

when they appear as parts of subjects in embedded clauses (see, e.g., Chomsky (1981), Giorgi (1984), and many others). I believe this to be a quite different phenomenon from the cases under discussion, and nothing is said about it here. 6. This example is due to Elisabet Engdahl, personal communication. 7. I take it that this distinction accounts for many of the classical 'de dicto/de re' ambiguities.

3 Long-distance reflexives and the typology of NPs Hoskuldur Thrainsson

1 Introduction Part of the importance of linguistic theories, as well as other scientific theories, is that they tell you which facts are interesting, why they are interesting, and where to look for other interesting facts. The standard binding theory (BT) of Chomsky (1981) has been extremely important in this respect since it has sparked a great deal of cross-linguistic research into the binding properties of pronouns, reflexives, and other NPs, overt and non-overt. As a result, we have learned a lot about the nature of NPs in various languages, what their similarities are and how they differ, both within a given language and cross-linguistically. It is well known that one does not have to look very hard or very far to find, say, reflexives that do not obey the same restrictions as reflexives in English. The socalled long-distance reflexives found in various languages are a case in point (see, e.g., Thniinsson (1976a, c), Napoli (1979), Yang (1983), Hellberg (1984), von Bremen (1984), Giorgi (1984), Maling (1984), Anderson (1986), Rognvaldsson (1986), Everaert (1986a), Barnes (1984, 1986), Sigurosson (1986a), Sportiche (1986), Sells (1987), Manzini & Wexler (1987), Wexler & Manzini (1987), Pica (1987), Koster (1987), Kuno (1987), and references cited there). The question is, however, what the existence of such reflexives implies for BT. There are various possibilities, and different suggestions can be found in the literature cited above. In the present chapter I would like to argue for the following claims among others: (1) a.

b.

Some of the 'long-distance reflexives' described in the literature can be accounted for by assuming parametric variation in the definition of 'binding domain', with minimal changes in the standard binding principles (cf. e.g., Yang (1983), Anderson (1986); see also Harbert (1986). There is a class of 'extra long-distance (logophoric?) reflexives' that need not be syntactically bound at all and thus do not obey principle A of

49

50

Hoskuldur Thrainsson

c.

standard BT nor any extension of such a syntactic binding principle (cf., e.g., Maling (1984), and Sigurosson (1986a». The standard classification of NPs in terms of the two binary features [± anaphoric] and [± pronominal] leaves a few existent categories ofNPs unaccounted for. By assuming the basic lexical feature [± independent reference] (to be explained below) and the (more BT-internal) feature [± R-expression] (cf. Lasnik (1986», it becomes possible to account for the syntactic behaviour of (or explain the existence of) 'truly long-distance reflexives', and such an account also makes some interesting predictions about the possible existence of other NP-types, which seem to be borne out.

The organization of this chapter is as follows: in section 2, I review the main characteristics of standard BT (Chomsky (1981», give a few examples of longdistance reflexives that are not immediately accounted for within it, describe briefly a possible extension of BT along the lines of Anderson (1986) and argue that although it - or other syntactic accounts - will work for a number of cases, there is a residue of cases that does not seem to lend itself to a syntactic account in terms of an extension of binding domains or other similar syntactic approaches (say, movement at LF). In section 3, I argue that we need to account for the fact that (standard BT) anaphors and long-distance reflexives of all types have something in common and this something is the lack of capacity for independent or deictic reference (i.e. they are all referentially dependent in some sense) rather than, say, the requirement that they must all be interpreted as bound variables. In section 4, I present a new classification of NPs, first schematically and then in terms of four binary features. Because of the mutual incompatibility of certain feature combinations, it turns out, however, that this system only predicts the existence of eight categories of NPs and it is argued that this is not too many since they all exist (although not necessarily all within a given language). Finally, section 5 is a conclusion.

2 Standard BT classification of NPs and some extensions 2.1 The standard BT Let us assume something like the following as the standard BT: Table 3.1. Standard BT - feature representation Feature complex

Categories

Some examples

A.

(bound) anaphors NP-trace

himself, each other Johnj was killed ej

[+an, -prJ

Long-distance reflexives and the typology of NPs

B.

[-an, +pr]

51

pronominals

he (she, it)

pro

pro leemos muchos libros (Sp.) '(We) read many books' John, the bastard Whoi did you see ei? John told Peter [PRO to go]

c.

[-an, -prj

lexical NPs wh-trace (variables)

D.

[+an, +pr]

PRO

(2) The standard BT principles: A. A [+anaphoric] NP must be bound in its governing category (or some such well-defined local domain). B. A [+pronominal] NP must be free in its governing category (or some such well-defined local domain). C. A [-anaphoric, -pronomimal] NP must be free. If we concentrate on the overt NPs, this classification could be represented schematically as in (3): _ _ NPs

(3)

must be free .

~

--

must be bound

I

free in gov cat

free in any domain

bound in gov cat

pronominals

lexical NPs

anaphors

2.2 A first revision - the syntactically defined domain extended It is not immediately obvious how, say, reflexives in the Scandinavian languages would fit into this classification since they can occur in infinitival clauses and be bound by the matrix subject whereas their English counterparts have to be bound within the infinitival clause in such structures (see, e.g., Anward (1974), Thrainsson (1979: 289ff.), Vikner (1985), Hellan (1986a), Anderson (1986»: (4)

(5)

a. b. c. a.

Peturi Peteri Peteri Annai Anne b. Annei Anne

baa Jensj urn [PROj aa raka sigi/j] (Icel.) bad Jensj om [PRO j at barbere sigi/j] (Dan.) asked Jensj [PROj to shave himself*i/j] telur [pig hafa svikiO sigi] believes you (Acc) have (Inf) betrayed self hprte [mig snakke med dig om sigi] heard me talk (Inf) to you about self

52

Hoskuldur Thrdinsson

This shows that the binding domain for reflexives in the Scandinavian languages is larger than in English, perhaps something like the minimal tensed clause. Note also that intervening subjects of the infinitives do not seem to have any effect, not even overt ones like the /Jig 'you' in (5a) (assuming an exceptional Case-marking analysis for Accusative-with-Infinitive structures like this) and the mig 'me' in (5b) (if mig is the subject there). Similar facts have been reported for Gothic, for instance, by Harbert in a series of interesting papers (1981, 1982, 1983, 1986). Given what we have found so far, it might not seem so difficult to give a syntactic definition of the extended binding domain in the Scandinavian languages, for instance. There is an added complication, however. What we typically find is that 'semi-long-distance reflexives' of this type are bound by subjects and they cannot be bound by objects. This is illustrated by examples from Danish and Icelandic below (cf. Vikner (1985), Thrainsson (1979: 288ff.)): (6)

a. *Eg; lofaoi 6nnuj [PROj ao kyssa sig j] (Ice!.) b. *Jegj lovede Annej [PROj at kysse sig j] (Dan.) I promised Anne to kiss self

Several attempts have been made to account for facts of this sort in the spirit of standard BT. Since the main objective of this chapter is not to account for the syntactic binding of semi-long-distance anaphors of this type, I will just briefly review here the interesting proposal of Anderson (1986) (for a valuable discussion of other recent approaches and a different proposal see Harbert (1986)). Anderson argues that there is a parametric variation in principle A of BT of roughly the following kind: I (7) Principle A: A [+anaphoric] NP must be 1. bound in its governing category, or 2. bound by a superordinate subject within its anaphoric domain. 2 Al is obviously the standard principle A of BT, and Anderson refers to NPs obeying Al as anaphors, whereas he uses the term reflexive pronouns to refer to NPs obeying A2. The idea is, then, that languages can choose between Al and A2, and English uses Al for reflexives and reciprocals whereas the Scandinavian languages (and Latin etc.) would use A2 for (at least some of) their reflexives 3 but Scandinavian reciprocals would obey Al like their English counterparts, as witnessed by Icelandic examples like those in (8) (for further discussion see Vikner (1985), Anderson (1986), Hellan (1986a), and Hellan & Christensen (1986a)):

Long-distance reflexives and the typology of NPs (8)

53

a. Peiri rokuau hvorn annani4 they shaved each other b. Eg sendi peimi gallabuxur ii hvorn annani I sent them bluejeans for each other c. Peiri lofuau mer [PRO i aa raka hvorn annani] they promised me to shave each other d. *Peiri skipuau mer; [PRO; aa raka hvorn annani] they ordered me to shave each other e. *Peiri telja [aa eg hati hvorn annani] they believe that I hate (sbjnct) each other

Now recall that principle B of standard BT (pronominals must be free in their governing category) is in a sense the inverse of principle A. Given the variation of A expressed in (7), Anderson suggests that we should have a similar variation for principle B, namely:5 (9) Principle B: A [+pronominal] NP must be 1. free in its governing category, or 2. subject free (( = not bound by a superordinate subject) in its anaphoric domain. The assumption would then be that English chooses B1 and the following would seem to suggest that Icelandic chooses B2: (10)

(11)

a. *Haralduri syndi mer fot ii hanni Harold showed me clothes for him b. Eg syndi Haraldii fot ii hanni I showed Harold clothes for him Peturi baa Jens; [PRO; aa raka hann*ii*;] Peter asked Jens to shave him

In this rather ingenious way, Anderson tries to express the well-known (partial) complementary distribution of pronouns and reflexives. The contrast between (11) for Icelandic and (12) for the Mainland Scandinavian languages would then suggest that these choose B1 rather than B2 (cf. Hellan (1983), Vikner (1985), Anward (1974)):

(12)

a. Joni bad oss; [PRO; hjelpe hami] (Norw.) John asked us (to) help him b. Susani bad mig; om [PRO; at ringe til hendei] (Dan.) Susan asked me to call her

54

Hoskuldur Thrtiinsson c. Honj bad migj [PROj klippa henneil (Swed.) she asked me cut her 'She asked me to cut her hair'

The following would also be consistent with Bl (again, see Hellan (1983) and Vikner (1985)): (13)

a. *Vi fortalte we told b. *Vi fortalte we told

Jonj om hamj (Norw.) John about him Annej om hendej (Dan.) Anne about her

Note, however, that if a given language uses both Al and A2, as the Scandinavian languages seem to do, there is nothing to prevent it from choosing both Bl and B2, which could mean, for instance, that within clauses (i.e. inside their governing categories) pronominals had to be completely free whereas within the larger anaphodc domain (say, the minimal tensed clause) they would only have to be subject free. While we have omitted certain details in this discussion of binding in the Scandinavian languages,6 it should be clear from the examples given how Anderson's approach allows for an extension of the standard BT while still keeping its spirit. Note, for instance, that it is still true under this approach that (all types of) reflexives and reciprocals (i.e. [+anaphoric] NPs) must be bound in some syntactically defined domain whereas pronominals ([ + pronominal] NPs) must be free in some (corresponding) domain. But now we will look at some reflexive-like elements that seem more problematic for this standard classification of NPs. 2.3 Truly long-distance reflexives, bound and unbound There are also languages where a reflexive inside a finite clause can be bound by a subject outside the clause. Latin, Icelandic and Faroese are a case in point (cf. Milner (1978), Thniinsson (1976a, c), Anderson (1986: 84-6), Barnes (1986), Kuno (1987: 136ff.)), whereas such constructions are normally not found in the Mainland Scandinavian languages (but see Barnes (1984), Hellberg (1984)): (14)

a. Orat te paterj rut ad sej venias] (Latin) asks you father that to self come (sbjnct) 'The father (your father) asks that you come to him' b. Ariovistusj ad Caesaremj legatos mittit rut ex suis j ambassadors sent that of self's Ariovistus to Caesar legatis aliquem ad sej mitteret] ambassadors somebody to self sent (sbjnct)

Long-distance reflexives and the typology of NPs

55

'A. sent ambassadors to C., in order that he (C.) would send his [refl] (C.'s) ambassadors to him [refl] (A.)' (15) a. J6ni sagoi [ao eg hefOi svikio sigi] (Icel.) b. *Jensi sagde [at jeg havde svigtet sigiJ (Dan.) c. *Johni said [that I had betrayed himselfi] (16) Guoruni skilti [at Martin hevoi skrivao eitt brl£v til sini] (Far.) Gudrun understood that Martin had written a letter to self 'Gudrun understood that Martin had written a letter to her' It should be made clear that this phenomenon is truly long distance, since the antecedent (or binder) of the reflexive pronoun can be arbitrarily far away (cf. Thniinsson (1976a: 226»: (17) J6ni segir [ao Maria telji [ao Haraldur vilji [ao Billi heimsl£ki sigiJ]] John says that Mary believes that Harold wants that Bill visits self Similar facts have been reported for Italian by Napoli (1979), for instance, and truly long-distance reflexives exist in several other languages as well (cf. Yang (1983), Kuno (1987), etc.; see also Giorgi (1984». At first it might seem that we have here a somewhat similar syntactic phenomenon to that observed in section 2.2 - i.e. some sort of syntactically definable extension of the binding domain for reflexives. It would, however, seem that one would not want to have principle B of the BT apply to this extended domain since pronominals do not have to be (subject) free in it, as can be seen from the comparison of (15a) and (18). In other words, we do not have complementary distribution of pronouns and reflexives within this domain: 7 (18)

J6ni sagoi [ao eg hefOi svikiO hanni] John said that I had betrayed him

Note, on the other hand, that here again we find that objects do not seem to be acceptable antecedents, at least not in Icelandic: (19)

*Eg sagoi J6nii [ao ]:nl hefOir svikio sigiJ I told John that you had betrayed self

It has turned out, however, that it is quite difficult to come up with a convincing syntactic account and it is probably worthwhile to review the main properties of this kind of long-distance reflexives here. First, observe that for most speakers of Icelandic there is a correlation between the long-distance reflexive (LDR) and subjunctives. Typically, verbs of saying, thinking etc. take subjunctive complements and it is in such complements that we find the LDR. A (semi-)factive verb

56

Hoskuldur Thrainsson

like vita 'know', on the other hand, takes an indicative complement and then most speakers reject the LDR:8 (20)

*J6ni veit [ao pu hefur svikio sigil John knows that you have betrayed self

There are, however, a few verbs that can take either a subjunctive or an indicative complement. In such cases the indicative seems to imply the speaker's presupposition that the complement is true whereas the subjunctive implies a report from the subject's point of view in some sense and the speaker does not commit himself with respect to its truth (cf. Thrainsson (1976a), Maling (1984) and most extensively Sigurosson (1986a). See also Sells (1987». Hence the speaker can deny the content of a subjunctive complement but not that of an indicative one without creating a contradiction: (21)

a. J6n heyroi [ao Maria hefOi komia] en John heard that Mary had (sbjnct) come but b. J6n heyroi [ao Maria hafOi komii'll *en John heard that Mary had (ind) come but

hUn she hUn she

hafOi had hafOi had

ekki not ekki not

komii'l come komii'l come

In cases like these, the LDR goes together with the subjunctive: 9 (22)

J6ni heyroi [ao eg hefOiI*hafOi svikio sigil John heard that I had (sbjnctlind) betrayed self

It is not the case, however, that LDRs are mechanically conditioned by the subjunctive. We do not find them in adverbial clauses that contain subjunctives signalling unfulfilled conditions, for instance (23a), nor in adverbial clauses that contain sUbjunctives because the conjunction in question requires the subjunctive (23b) (cf. Thrainsson (1976a»: (23)

a. *J6ni John b. *J6ni John

yroi glaour [ef pu hjalpaoir seri] would be glad if you helped (sbjnct) self lykur pessu ekki [nema pu hjalpir Seri] finishes this not unless you help (sbjnct) self

But if sentences containing adverbial clauses are embedded under verbs of saying, for instance, the LDR can occur inside the adverbial clause (cf. Thrainsson (1976a: 230-1»: (24)

a. J6ni sagoi [ao hanni yroi glaour [ef pu hjalpaoir ser]] John said that he would be glad if you helped self

Long-distance reflexives and the typology of NPs

57

b. Joni segir [ao hann Ijtiki pessu ekki [nema pti hjalpir ser]] finishes this not unless you help self John says that he Second, in the light of some recent proposals suggesting some sort of movement analysis (movement in LF) to account for apparent long-distance phenomena involving anaphors (see Lebeaux (1983), Chomsky (1986a), and especially Pica (1987) (who discusses Icelandic)), it is perhaps particularly interesting to note that LDRs seem to occur freely in syntactic islands, as long as these islands are embedded under verbs of saying or believing, for instance. The subjects of such verbs can thus be the antecedents of LDRs that occur inside islands. We have actually already seen an indication of this in (24), since adverbial clauses are normally islands. This is illustrated further in (25-26), where it is shown that relative clauses and complex NP-clauses are extraction islands even when embedded under verbs of saying but in such contexts we can get LDRs inside them (see also Thrainsson (1976a: section 3.2)): (25)

a. Jon segir [ao pti hafir baria konuna sem hafi John says that you have (sbjnct) hit the woman that has (sbjnct) svikio pig]] betrayed you b. *Hverni segir Jon [ao pti hafir bario konuna [sem hafi svikio ti]]? who says John that you have hit the woman that has betrayed 'Who does John say that .. .' c. Joni segir [ao pti hafir bario konuna sigil] [sem hafi svikia John says that you have hit the woman that has betrayed self 'John says that you hit the woman that betrayed him (refl), (26) a. Jon segir [ao Maria trtii ekki peirri fullyroingu John says that Mary believes (sbjnct) not that claim [ao konan hafi svikio pig]] that the woman has (sbjnct) betrayed you 'John says that Mary doesn't believe the claim that the woman betrayed you' b. *I>igi segir Jon [ao Maria trtii ekki peirri fullyroingu [ao you says John that Mary believes not that claim that konan hafi svikio til] the woman has betrayed c. Joni segir [ao Marfa trtii ekki peirri fullyroingu [ao John says that Mary believes not that claim that sigi]] konan hafi svikio the woman has betrayed self

58

Hoskuldur Thrainsson 'John says that Mary doesn't believe the claim that the woman has betrayed him [refl],

Thirdly, as pointed out by Maling (1984), Hellberg (1984) and Barnes (1986), and discussed more extensively by Sigurosson (1986a), it is not the case that all long-distance reflexives in languages like Icelandic and Faroese need to be syntactically bound. Consider the following examples, for instance: Sigvaldii neitaoi jwi, ao petta va:ri vilji pj6oarinnar. Ao rninnsta kosti Sigvaldi denied it that this was will the nation's at least v::eri pao ekki sinni vilji (Icel.) was it not self's will 'Sigvaldi denied that this was the nation's will. At least it was not his [refl] will [he said]' ta iO hanni (28) ... hanni vildi ikki leypa fnt sinarii abyrgd, he would not run from self's responsibility now that he var komin soleiOis fyri vio Sigrid. was come so for with Sigrid Hon hevoi meiri krav upp a segi enn hin (Far.) she had more demand up on self than the other 'He would not run from his responsibility now that he had got into this situation with Sigrid. She had more right to him [refl] than the other [girl]' (29) Maria var alltaf svo andstyggileg. I>egar Olafurj k::emi segoi hUn Mary was always so nasty when Olaf came said she seri/*j areiOanlega ao fara ... (lcel.) himself certainly to leave 'Mary was always so nasty. When Olaf came, she would certainly tell himself [the person whose thoughts are being presented - not Olaf] to leave ... '

(27)

It seems very unlikely that the concept of syntactic binding can be extended in a meaningful way to cover intersentential reflexives of this sort, and they are not limited to Icelandic or Faroese. Similar examples can be found with Japanese zibun (cf. Sells (1987: 455)), for instance. The example in (29) is perhaps particularly interesting, however, since the antecedent is nowhere to be seen in the immediately preceding sentence. Although examples of this sort are not discussed in Anderson's paper (1986), he has later studied them in some detail and argued that they show the necessity of setting up an additional category, namely that of logophoric pronouns (class lectures 1985). This had also been argued by Maling (1984), with reference to West-African logophoric pronouns of the sort discussed

Long-distance reflexives and the typology of NPs

59

by Clements (1975), for instance. Sigurosson (1986a) discusses the (semantic! pragmatic) aspect of these Icelandic long-distance reflexives in some detail. Other discussions of 'logophoricity' in this (or a similar) sense include Kuno (1987: 136ff.) and Sells (1987), both dealing with a number of languages, including Icelandic. (See also note 12 below.) But it is important to note that the semantic conditions for these syntactically unbound cases oflong-distance reflexives in Icelandic (and Faroese) seem to be the same as those for the ones where a reflexive inside a finite (subjunctive) clause is syntactically bound by the subject of a higher clause in the same sentence. This is shown in some detail in Sigurosson (1986a) and it indicates that we do not want a special account of the syntactically unbound long-distance reflexives in these languages. What we need is rather an account that takes care of both the more familiar instances of reflexives inside finite (subjunctive) clauses bound by (subject) antecedents in a higher clause and the intersentential, unbound reflexives just observed. That seems to make any attempt to extend the syntactic binding domain beyond finite-clause boundaries in languages like Icelandic and Faroese, for instance, a dubious enterprise. That does not necessarily mean, however, that there cannot be languages that allow syntactic binding domains to be extended to a larger domain than the minimal finite clause but do not allow intersentential relations between reflexives and their antecedents. Rather than concentrating on what is special about the semantics of the longdistance reflexives in languages like Icelandic and Faroese, as many of the abovementioned studies have, I will in the following sections try to determine what, if anything, all types of anaphors and reflexives have in common and how they differ syntactically from pronominals and lexical NPs. That will lead to a new classification or typology of NPs, but the basic principles of BT will require only minimal modifications if we keep in mind that they are to be interpreted as necessary rather than sufficient syntactic conditions.

3 The common property of 'anaphors' in the extended sense 3.1 Bound-variable reading? Contrasts like the following have frequently been observed: (30) a. Johni shaved himself, and Peterj did too (= 'Peterj shaved himselfj, 'Peter shaved John') b. Johni saw hisi book and Peterj did too (= 'Peter saw his own book', or: 'Peter saw John's book') c. Johni said [that you had betrayed himi] and Peterj did too

*

60

H oskuldur Thrainsson (= 'Peter said that you had betrayed John', or: 'Peterj said that you had betrayed him j')

On the basis of examples of this sort, it has been claimed that anaphors like himself (normally) receive a bound-variable reading (at least in these 'VP-deletion' contexts)lO and hence we only get the 'sloppy identity' interpretation of sentences like (30a), whereas non-anaphors like the possessive his in (30b) and the pronominal him in (30c) may either be interpreted as bound variables with respect to their antecedent or be coreferential with it.ll Hence (30b, c) are ambiguous. Given this, one might expect the long-distance reflexives discussed in the preceding section to pattern with (local) anaphors in this respect. The sentences in (31), however, show that this is not the case for the Icelandic long-distance sig, although locally bound sig seems to behave like himself here: (31)

a. Joni rakaoi sigi og Peturj geroi pao lika John shaved himself and Peter did so too (=1= 'Peter shaved John') b. Joni sagoi [ao pu hemir svikio hannil og Peturj geroi pao lika John said that you had betrayed him and Peter did so too (= 'Peter said that you had betrayed John', or: 'Peterj said that you had betrayed himj') c. Joni sagoi [ao pu hemir svikiO sigi] og Peturi geroi pao lika John said that you had betrayed self and Peter did so too (= 'Peterj said that you had betrayed himj', or: 'Peter said that you had betrayed John')

In sentences like (31c), most speakers seem to find it rather easy to get both the sloppy and strict readings. Interestingly, however, it seems to be more difficult (if not impossible) to get the sloppy (i.e. bound-variable) reading in (32), where the antecedent does not c-command the long-distance reflexive: (32)

Skooun J onsi er [ao pu hafir svikiO sigi] og pao er skooun Peturs lika opinion John's is that you have betrayed self and that is opinion Peter's too 'John's opinion is that you have betrayed him and that is Peter's opinion too' (= 'It is Peter's opinion that you have betrayed John')

It is at least more difficult to get the bound-variable reading here and if it were indeed impossible it would be consistent with Reinhart's claim that 'the antecedent must c-command the pronoun for the sloppy identity interpretation to be obtained' ((1983c: 63) - see also Reinhart (1986: 125ff.)). But the available interpretations of (31c) and (32) show in any case that the long-distance reflexives in Icelandic are not

Long-distance reflexives and the typology of NPs

61

always interpreted as bound variables. 12 One should perhaps point out in this connection, however, that the long-distance reflexives still differ from pronominals in various respects, one being that they do not allow split antecedents: (33)

a.

Jani John b. *Jani John

sagoi told sagoi told

Mariuj Mary Mariuj Mary

[ao that [ao that

pu you pu you

hefOir had hefOir had

svikio betrayed svikio betrayed

paui+j] them sigi+j] self13

With coordinated antecedents, on the other hand, both pronominals and longdistance reflexives are possible, of course: (34)

a

Jani John b. J ani John

og and og and

Mariaj Mary Mariaj Mary

siigou said siigou said

[ao that [ao that

pu you pu you

hefOir had hefOir had

svikio betrayed svikiO betrayed

paui+j] them sigi + j] self

3.2 Incapable of 'independent' or 'deictic' reference? In his discussion of the BT classification of NPs, Chomsky (1981: 188) says that 'intuitively, anaphors are NPs that have no capacity for "inherent reference"'. This property of BT anaphors has frequently been noted and sometimes even used as a defining characteristic of anaphors (cf., e.g., Reinhart (1983c: 70), Giorgi (1984: 309)). If we take 'capacity for independent reference' to mean the capability of 'picking up a definite referent in the world, or [freely] in the previous discourse' (cf. Giorgi (1984: 309)), it should be clear that syntactically bound anaphors of the type Chomsky (1981) was discussing will never be able to do this since they will always be syntactically bound by some (c-commanding, coreferential) antecedent. And it is well known there is a clear contrast in this respect between pronominals and lexical NPs (R-expressions) on the one hand and anaphors like the English reflexive on the other. This can be illustrated by using a diagnostic test ofthe following type (cf. Hankamer & Sag (1976)): (35)

a. [Somebody in the audience gets up and leaves] Speaker: He is weird OR Where is the bastard going? OR Where is John going? OR I guess his patience ran out

62

Hoskuldur Thrainsson b. [Same situation] Speaker: *1 like himself (vs. I like him)

It is perhaps not a priori clear, on the other hand, how the long-distance reflexives would pattern in this respect, especially if we' consider the ones that are not syntactically bound at all. Recall, however, that the long-distance reflexives differ from pronominals in that they cannot just have any old syntactic constituent as an antecedent - if there is one, it seems to have to be the subject (subject of S or of NP) in Icelandic, for instance (cf. the discussion above). But what about cases like (29), repeated here for ease of reference? (29)

Maria var alltaf svo andstyggileg. Pegar Olafurj kfemi segoi hUn Mary was always so nasty when Olaf came said she seri/*j areioanlega ao fara ... himself certainly to leave 'Mary was always so nasty. When Olaf came, she would certainly tell himself [the person whose thoughts are being presented - not Olaf] to leave .. .'

Here there is no syntactic antecedent in sight for the reflexive. One might think, therefore, that cases of this type could be interpreted as instances of 'independent reference' . I want to argue, however, that cases like (29) are not instances of independent reference and there is a clear difference between pronominals and long-distance reflexives of this type with respect to the 'independence' of their reference (cf. also Sigurosson (1986a: 14)). Thus it is impossible to begin a book or a short story or a chapter or any such section of a narrative by something like (29). We need an introduction of some kind - an introduction where it is made clear whose thoughts are being presented. The use of pronominals needs no introduction of that sort. Hence (36) could very well be the beginning of a novel or of a smaller section of some narrative, and the pronominal hann has independent reference in the sense that it needs no previous mention but will become clear from the situation described. Given this introduction, we can then go on as in (29): (36)

Hanni la einn i myrkrinu og hugsaoi. Maria var alltaf ... 'He lay alone in the dark, thinking. Mary was always .. .'

This shows, then, that even the inter sentential reflexives cannot have independent or deictic reference in the sense that pronominals can. It is this property which seems to be common to all 'anaphors' in the extended sense - i.e. the NPs that are

Long-distance reflexives and the typology of NPs

63

not of the lexical NP or the personal-pronoun type - be they locally bound, longdistance bound or syntactically unbound. This must be made explicit in our classification of NPs. 4 Towards a new classification of NPs 4.1 Schematically What we have found so far, then, is that not all 'anaphors' in the extended sense, taking 'anaphors' now to mean the NPs not capable of independent reference, need to be syntactically bound. But it is also important to recall too that not all NPs that are capable of independent reference have to be free in a particular domain. More specifically, non-reflexive possessive pronouns need not be free at all in languages like English and German, for instance: (37)

a. Johni saw hisilj book b. Hansi sah seinilj Buch

The same is true for the 3rd person plural possessive in Danish: (38)

Dei sa deresilj bog they saw their book

These possessive pronouns are, however, like regular pronominals in that they have the capability of independent or deictic reference (cf. (35a) above). 1st and 2nd person personal pronouns in languages that do not have 1st and 2nd person reflexive pronouns behave the same way as the English (and German) possessive pronouns with respect to binding. In other words, they need not be free at all and in that respect they contrast with their English counterparts, for instance. This is illustrated in (39), with examples from Icelandic, Danish, and English: (39)

a. Egi rakaoi migi (IceI.) b. Jegi barberede migi (Dan.) c. *Ii shaved mei

Given this, we could present the observed types schematically, as shown in (40): (40)

_________ NPs-------capable of indep. ref.

~ need not be

must be free in some domain

Domain varies:

free in any domain

not capable of indep. ref.

--------------

must be synt. bound in some domain

Domain varies:

need not be synt. bound in any domain

64

Hoskuldur Thrainsson Examples: 2: his la: he Ib: Icel. hann Ger. sein Icel. eg lc: the bastard

3a: himself 3b: Dan. sig 3c: Ital. se (?)

4: Icel. & Far. sig. & more LDRs

Domains la and 3a:

'Governing category' (or some such welldefined local domain)

Domains Ib and 3b:

Minimal tensed clause (or something like that)

'Domain' lc:

Any domain

Domain 3c:

Minimal indicative clause (or some such definition) - see, e.g., Anderson (1986), Yang (1983), Giorgi (1984, and references cited there)

We have already presented some evidence for the classification given in (40), both with respect to the four basic types of NPs assumed there and the different anaphoric domains. As pointed out in section 2.2, there are several languages where the anaphoric domain for reflexive-like elements is not simply the governing category but rather an extended domain like the minimal tensed clause (domains Ib and 3b in (40)), even though something like governing category may seem the correct domain for reciprocals in these languages (cf., e.g., (8) above). It is also conceivable that rules of anaphora need to refer to a larger domain like the 'minimal indicative clause' (3c), although we saw in 2.3 above that such an extension would not do for Icelandic or Faroese, since the long-distance reflexives in these languages need not be syntactically bound at all. Sentences of the type given in (41) have been presented as arguments for the claim that something like the minimal indicative clause is the relevant domain for (the stressed) se in Italian (cf. Napoli (1979), Yang (1983) - compare also Giorgi (1984)): (41)

a. Gianni; e molto contento di sei/*j Gianni is very happy with self b. La signora; mij dice [PRO j di giacere presso di Sei/*k] the woman orders me to lie near self c. La signora; dice [che io giaccia presso di sei/*j] the woman orders that I lie (sbjnct) near self d. *La signora; dice [che io giaccio presso di sei/j] the woman says that I lie (ind) near self

Long-distance reflexives and the typology of NPs

65

For all I know, however, it may very well be that Italian se is like its Icelandic and Faroese counterparts in that it need not be syntactically bound at all, and that may be true of all 'logophoric' reflexives. It seems to hold for the logophoric ye in Ewe, for instance (cf. Clements (1975». If that is true, then we may not need a syntactically defined anaphoric domain like 3c at all. Now observe that we really should be more explicit about what it means that certain NPs 'need not be free in any domain' or 'need not be syntactically bound in any domain'. What is the relationship of these NPs to other types of NPs? Could they observe certain BT principles but not others? In particular, how could we specify them in terms of features like [±anaphoric] and [±pronominal]? Is the classification suggested here compatible with feature specifications of that sort? In the following subsection I shall show that it is indeed advantageous to continue to use binary feature specifications of that sort, and if we add two more binary features, we will arrive at a quite explicit and interesting typological classification of NPs.

4.2 A (partially) new classification in terms of features First, recall that lexical NPs (or R-expressions) are specified as [-an, -prj in standard BT. Since this seems to be a sort of a default value, it is perhaps somewhat strange that principle C refers specifically to this feature combination. More importantly, however, Lasnik (1986) has argued quite convincingly that we need to be able to express the fact that the so-called pronominal epithets (the bastard, etc.) do have certain things in common with names (John, etc.), as assumed in standard BT, but also behave like pronominals in certain ways. He suggests, therefore, that we need the binary BT-feature [±R-expression], and 'pure' R-expressions, like names, have the feature specification [-an, -pr, + R] but pronominal epithets are [-an, +pr, +R]. Given this, let us now assume that the property of being capable - or not capable - of independent reference in the sense discussed earlier can be expressed by a binary feature too, namely [±ind refJ.lfwe take [+anaphoric] to mean, as before, that the NP in question must be syntactically bound (in some domain), it is obvious that the feature specifications [+ind ref, +an] are incompatible. We can also assume that the feature specifications [-ind ref, + R] are incompatible, since R-expressions (= referential expressions) must be capable of independent reference (by definition, one would think). This will then give us the following possible classes of NPs (where a * indicates an impossible specification):

66

Hoskuldur Thrdinsson

(42)

NPs [-ind ref]

[+ind ref]

~ [-R]

~ [-R]

*[+R]

[+R]

~ [-an]

*[ + an)

/ [-pr] \

[+pr] 1.

2.

A [-an]

*[ + an]

~ [-pr]

[+pr] 3.

4.

~ [-an]

[+an]

A

[+pr] 5.

[-pr] 6.

~ [-pr]

[+pr] 7.

8.

Now let us assume the version of BT principles A and B suggested by Anderson (1986, cf. section 2.2 above; see also the discussion in Yang (1983), Mohanan (1982), and Harbert (1986) for other possible parametrizations of BT) and the version of principle C suggested by Lasnik (1986). That gives the following BT: (43) A revised BT: A A [+anaphoric] NP must be 1. bound in its governing category, or 2. bound by a superordinate subject within its anaphoric domain. B A [+pronominal] NP must be 1. free in its governing category, or 2. subject free (= not bound by a superordinate subject) in its anaphoric domain. C A [ + R-expression] NP must be free. 14 The categories defined by this BT and the feature system shown in (42) will then include the ones given in Table 3.2: Table 3.2. Revised BT - feature representation Feature complex 1. 2.

3. 4. 5. 6. 7. 8.

[+indref, [+indref, [+ind ref, [+ind ref, [-ind ref, [-ind ref, [-ind ref, [-ind ref,

+R, +R, - R, - R, - R, - R, - R, - R,

-an, -an, -an, -an, +an, +an, -an, -an,

+pr] -prJ +pr] -prJ +pr] -prJ +pr] -prJ

Categories

Some examples

pron epith true R-expr pronominals poss prons/l st, 2nd pers prons PRO(?), etc. anaphors pran LDRs(?) (logoph) LDRs

the bastard John he his; sein; eg ham selv himself aapa!J; !aan (?) sig

Long-distance reflexives and the typology of NPs

67

A few of the claims and suggestions made in Table 3.2 need further clarification or discussion. We will proceed in the order given in the table. First, we should say something about the distinction between categories 1 and 2, i.e. pronominal epithets and true R-expressions. For one thing, it has been observed that pronominal epithets can have antecedents like pronouns, as long as they are not syntactically bound by them. Names do not share this property (cf. Lasnik (1986)): (44) a. Johni promised me to come to my talk but the bastardi never came b. ?Johni promised me to come to my talk but Johni never came Thus apart from condition C, there seems to be some sort of a repetition constraint on names that does not hold for pronominal epithets. This difference is not accounted for by the binding theory - it is mainly shown here as an argument for the claim that these types of NPs differ in some respects at least. More importantly, however, there are languages that do not seem to observe condition C (cf. note 14). If these languages do observe condition B of BT, we would expect that pronominal epithets would have to be free in their governing category but names would not have to be, given the feature specifications assumed here. According to Lasnik (1986), this seems to be what we find in Thai: (45)

a. C:l:lni ch:l:lp C:l:lni John likes John b. C:l:lni khit w;a C:l:lni chalaat John thinks that John is smart c. *C:l:lni ch:l:lp ?iiyba~ John likes the nut d. C:l:ln khit wlia ?iiyb;a chalaat John thinks that the nut is smart

The feature specification of pronominals (category 3) needs no further comment, but category 4 is the possessive pronouns and personal pronouns that need not be free in any domain (cf. (37), (38), and (39) above) and these receive here the default specification [ - R, - an, - pr]. That is meant to imply that they are not subject to any of the rules that refer to the plus-values for these features. That seems to be exactly the effect we want. Turning now to category 5 in Table 3.2, the obvious suggestion would seem to be PRO, in accordance with standard BT. This is somewhat questionable, however. First, there has been some controversy as to whether this is the right way to account for the distribution of PRO (cf. Manzini (1983), Bouchard (1985)). In addition, for the PRO-theorem to follow from BT, it is crucial that the anaphoric

68

H oskuldur Thniinsson

domain relevant for [+anaphoric] NPs and [+pronominal] NPs be governing category, as in standard BT, but we have seen ample evidence above that this is not always the case. Consequently, we have parametrized principles A and B of our BT so it is no longer clear that the PRO-theorem would follow. Instead, new possibilities are predicted for the specification [+an, +pr]. It should, for instance, be possible to find an NP that was so specified where [+anaphoric] implied that it had to be bound in its governing category (condition Al in (43) above) but [+pronominal] meant that it had to be subject free in its anaphoric domain (condition B2 above). Something like this seems to be a correct statement of the conditions on Danish ham/hende selv 'him/her self', for instance, as indicated by the following (cf. Vikner (1985»: (46)

a. Susani fortalte Annej om hende selv*ilj Susan told Anne about her self b. *Susani bad Annej om [PROj at ringe til hende selvilj] Susan asked Anne to ring to her self

As the reader can undoubtedly figure out, there should be other possibilities for category 5 too, given the parametrized versions of conditions A and B and the varying domains for pronominals and anaphors. It should also be mentioned here that Mohanan (1982) argues that there are 'pronominal anaphors' in Malayalam, i.e. overt NPs that must be specified as [+an, +pr], and the reason is obviously the same as the one we have found here - i.e. that the domains where anaphors must be bound and pronominals must be free do not always coincide. We will return to Malayalam in our discussion of category 7 below. Nothing further needs to be said about category 6, except that the reader should remember that we are assuming possible variations in the anaphoric domain, so this category will not only include 'regular anaphors' bound in their governing category but also the reflexive pronouns of Anderson (1986), for instance, i.e. NPs obeying condition A2 of (43) above, and possibly other 'long-distance' reflexives, as long as they need to be bound within a syntactically specifiable domain. The feature specification for category 7 in Table 3.2 above would mean that we had an NP that need not be syntactically bound in any domain (that is what we take [-anaphoric] to mean) but had to be free in the domain relevant for pronominals in that language. This seems an entirely plausible possibility. Thus it has been reported in the literature that the 'anaphor' aaparj in Marathi is a long-distance anaphor in that it can occur in a finite clause with an antecedent in the matrix clause but it cannot have its antecedent in the same clause (cf. Wali (1979: 405-6); see also von Bremen (1984: 203, 212»:

Long-distance reflexives and the typology of NPs (47)

a.

69

Minilaai vaatta [ki aapaJ;li cuk keli] Mini thinks that self mistake made

b. *Mininei aaplayaalaai ba' •• ,'Tn).9 In the case of (28) this set should include 'TJ and 'T2 but not 'T3' Languages differ in their choice of potential governing categories. In a configuration [T2 ... [TJ ... anaphor/pronominal ... ] ... ], both 'TJ and 'T2 are potential governing categories only: - if 'TJ is a small clause complement for Dutch; - if'TJ is a small clause or infinitival complement for Norwegian; - if'TJ is a small clause, infinitival or subjunctive complement for Icelandic. We will not be concerned here with the question of how the choice for a governing category is made and why languages differ in their choices. We simply assume that it is possible to get the differences in the range of potential governing categories without making reference to the nature of the anaphoric or pronominal element, as I have shown in Everaert (1986a). We are now in a position to present our revision of the BT.

5.3 A revised binding theory Suppose we limit the class of elements subject to the BT to those lexical elements that do not denote by virtue of their inherent properties, i.e. only to anaphors and pronominals and not R-expressions. JO These elements, for which we will use the term pronoun, will be specified for

(0)

V NPj

(OJ)

On this view, the O-criterion requires that all O-roles in a O-grid must be assigned an index. This also means that reference is attached directly to the O-role as an

Contextual determination of the anaphorlpronominal distinction

99

index (cf. Williams (1985)), and, therefore, it comes as no surprise that a a-role can also be independently indexed, for instance by a rule like (57). This raises the question whether or not a syntactic position can be projected in these cases. Rizzi (1986a) argues that a syntactic position cannot be projected if its corresponding arole is lexically saturated, i.e. syntactically invisible. The reverse, though, does not hold: syntactically visible a-roles are not necessarily projected (cf. 58b). Now suppose, contrary to Rizzi (1986a), that a syntactic position can be projected if the verb is specified to do so, even though the a-role is 'inherently' indexed. However, because the a-role is already lexically indexed, the lexical elements occupying the projected syntactic positions may not be referential. These lexical elements would then derive their (quasi-)referential status from the inherently indexed a-role. Possible candidates are quasi-arguments like idiomatic phrases (cf. Coopmans & Everaert 1988), expletives or, for that matter, (clitic) reflexive pronouns. I will use the term 'a-linking' for these cases of a-assignment. Now that we have clarified the general outline of our theory of a-government, we are in a position to define the notions a-marking and a-linking as they are used in the CBPs. I will define a-government as in (60) and a-markingla-linking as in (61) (cf. Coopmans & Everaert (1988)): (60)

(61)

a a-governs 13 iff a is a zero-level category that (i) a-marks 13 or (ii) a-links 13. a. 13 is a-marked by a iff a position in the a-grid of a is assigned an index by 13. b. 13 is a-linked to a iff a, 13 are sisters and a position in the a-grid of a is coindexed with 13, and a and 13 are dominated by the same lexical projections.

6.3 (Inherently) reflexive verbs With the theory of a-government in mind, we are able to say something more about the class of inherently reflexive verbs. In our view, inherently reflexive verbs are lexically specified as having an internal thematic role coindexed with the subject thematic role, while the verb is, at the same time, specified to project a syntactic position. As a result, the object position is necessarily occupied by an anaphor, zich in Dutch, the only phrase that can be a-linked to the lexically saturated a-role in these cases: (62) a. schamen: (ai> ai) b. Jani schaamt zichi 'Jan is ashamed'

100

Martin Everaert

A similar process might be at work in the English examples in (63) but in these cases no syntactic argument is projected (cf. Zubizarreta (1987)). (63) a. wash, shave, etc.: NPace> NPdela!

The lexical hierarchy of arguments, encoded in the a-grid, is both functionally and structurally independent of the phrase structure argument hierarchy expressible in terms of c-command. While the phrase structure argument hierarchy is the expression of head-complement and specifier-head relations (cf. Speas (1986)),

The primacy condition of anaphora and pronominal variable binding

251

the lexical argument hierarchy reflects the relative prominence of coarguments. The fact that in the case of inquire, for example, the D-structure hierarchy of the three arguments is identical to their lexical hierarchy is a mere accident, which may not happen if two of the arguments are VP-internal arguments of the same category. A VP containing two NPs simply cannot be assigned a phrase structure in which the argument closer to the verb is hierarchically more prominent than the argument farther from the verb - as noted by Barss & Lasnik (1986). In the case of two PPs, the prominence of the PP next to the verb can only be achieved if its preposition is reanalysed as part of the verb. We claim that the licensing condition of anaphora has to make reference to the lexical hierarchy of arguments. The Hungarian data in (3-8) could, in fact, be accounted for on the basis of the lexical hierarchy of arguments alone, by means of the following simple primacy condition: (23) An anaphor must have an antecedent that precedes it in the lexical argument hierarchy. The bidirectionality of the anaphoric relation between an accusative NP and a dative NP could be derived from the - intuitively plausible - assumption that in a a-grid containing both a beneficiary argument, realized as NP dat> and a theme argument, realized as NPacc , either the beneficiary or the theme can be viewed as the more active participant of the action. Under this view, a verb with a beneficiary and a theme argument can be associated with two a-grids, which order the same arguments differently, as follows:

(24) megmutat, +V - N (i) (1, 2, 3, E) agent (1), theme (2), beneficiary (3) (ii) (1, 2, 3, E) agent (1), beneficiary (2), theme (3) In (8a), in which the theme argument binds the beneficiary argument, the primacy condition of anaphora is interpreted on a-grid (i), while in (8b), in which the beneficiary argument binds the theme argument, the primacy condition of anaphora is interpreted on a-grid (ii). In English, the two a-grids are realized categorially in different ways: (25)

show, + V - N (i) (1,2, 3, E) agent (1), theme (2), beneficiary (3)

II

~

NP NP toNP

252

E. Kiss Katalin (ii)

(1, 2,

I

3,

E) agent (1), beneficiary (2), theme (3)

I

NP NP NP If the a-grid in (i) is mapped onto syntactic structure, the primacy condition in (23) allows the theme argument to bind the beneficiary argument, as in (lla). If the agrid in (ii) is mapped onto a syntactic structure, the primacy condition in (23) allows the beneficiary argument to bind the theme argument, as in (lIb). The proposed primacy condition of anaphora could also predict the ungrammaticality of the anaphoric relation between a passive subject and a by-phrase:

(26) *John was shot by himself It is also possible, however, that in (26) the locality condition of anaphora is violated, the subject and the by-phrase not being coarguments (cf. Jaeggli (1986)). The tentative version of the primacy condition of anaphora in (23) also accounts for the distribution of grammaticality in the English examples in (9-11). The facts of English anaphora, however, are more complex than indicated in (9-11). The following range of data has to be explained: (27) a. b. c. d.

I *1 *1 *1

talked with the girls about each other talked about each other with the girls talked about the girls with each other talked with each other about the girls

The tentative primacy condition in (23), making reference only to the lexical prominence relation of the arguments involved, can only rule out sentences (27 c, d). The ungrammaticality of (27b) is the consequence either of an additional S-structure precedence requirement, or of an additional c-command requirement. The Hungarian data argue against an explanation in terms of S-structure precedence, since the Hungarian equivalent of (27b), with the anaphor preceding the antecedent, is grammatical: (28) . Egymasr61 beszelgettem a lanyokkal each-other-about talked-I the girls-with 'About each other, I talked with the girls' Since in (28) the anaphor has been focussed, i.e. it has been preposed into the specifier position of the VP (cf. E. Kiss (1987b)), the antecedent neither precedes the anaphor nor c-commands it at S-structure: (28')

[8 [VP Egymasr6li [" beszelgettem pro a lanyokkal tim

The primacy condition of anapllOra and pronominal variable binding

253

However, the antecedent does c-command the trace of the anaphor. Consequently, the primacy condition, which, together with condition (23), can predict both the grammaticality of the Hungarian (28) and the ungrammaticality of the English (27b), is the requirement that an anaphor or its trace be c-commanded by its antecedent. In the grammatical (27a), the NP the girls c-commands the anaphor because the preposition is reanalysed as part of the verb. In (27b) the c-command condition of anaphora cannot be satisfied because the preposition of the intended antecedent is not adjacent to the verb, therefore its reanalysis is impossible. That reanalysis is a precondition of anaphora in such constructions is demonstrated in the literature by the following contrast: (29) a. Who did you talk with about himself? b. *With whom did you talk about himself? In view of these observations, we replace the tentative primacy condition of anaphora proposed in (23) with the following - still tentative - condition: (30) An anaphor must have an antecedent that both c-commands it or its trace at S-structure, and precedes it in the lexical argument hierarchy. The primacy condition of anaphora in (30) is still only capable of handling anaphoric relations between coarguments. Its validity, however, can also be extended to non-coarguments if the term 'precedence in the lexical argument hierarchy' is given a broader interpretation, along the following lines: (31) a precedes b in the lexical argument hierarchy iff a is ordered prior to b or to the constituent containing b in a O-grid Consider the types of example that have to be accounted for: (32)

a. Janos es Mari nem ismerik egymas sziileit John and Mary not know each-other's parents-ACC 'John and Mary don't know each other's parents' b. ?Egymas sziileit nem ismeri Janos es Mad 'Each-others' parents, John and Mary don't know' c. ?Janost es Marit nem ismerik egymas sziilei 'John and Mary, each other's parents don't know' d. *Egymas sziilei nem ismerik Janost es Marit 'Each other's parents don't know John and Mary' (33) a. Elbeszelgettem Janossal es Marival egymas sziileir6l 'I conversed with John and Mary about each other's parents'

254

E. Kiss Katalin b. ?Elbeszelgettem egymas sziileir6l Janossal es Marival '1 conversed about each other's parents with John and Mary' c. ?Elbeszelgettem Janosr6l es Marir6l egymas sziileivel '1 conversed about John and Mary with each other's parents' d. *Elbeszelgettem egymas sziileivel Janosr6l es Marir6l 'I conversed with each other's parents about John and Mary'

In English, such anaphoric constructions involving two VP-internal arguments of the same category display a similar distribution of grammaticality except that the equivalent of (33b) is ungrammatical: (34) a. b. c. d.

I talked with John and Mary about each other's parents *1 talked about each other's parents with John and Mary ??I talked about John and Mary with each other's parents *1 talked with each other's parents about John and Mary

What the Hungarian data demonstrate is that the antecedent must be more prominent than the anaphor in terms of at least one of two primacy principles: the antecedent must precede the anaphor in the lexical argument hierarchy (as in (32a, b) and (33a, b», or, it must precede the anaphor at S-structure (as in (32a, c) and (33a, c». If neither of the two conditions is satisfied (as in (32d) and (33d», the sentence is out. As for English, the distribution of grammaticality in (34a, c, and d) also supports this generalization. The ungrammaticality of (34b) is a consequence of the violation of the c-command condition of anaphora. The analysis of the primacy factors influencing anaphora has led to the identification of three primacy principles: lexical prominence, S-structure c-command, and S-structure precedence. Their interaction can be summarized in the form of the following complex primacy condition of anaphora: (35) The primacy condition of anaphora An antecedent can bind an anaphor iff it c-commands the anaphor or its trace at S-structure, and at least one of (i) and (ii) also holds: (i) it precedes the anaphor in the lexical argument hierarchy (ii) it precedes the anaphor at S-structure The proposed primacy condition of anaphora is too powerful in so far as it also licenses sentences of the following type in Hungarian: (36) *Nem szereti egymas Janost es Marit not like each-other-NOM John-ACC and Mary-ACC 'Each other don't like John and Mary'

The primacy condition of anaphora and pronominal variable binding

255

The English equivalent of (36) would be ruled out as a violation of the c-command requirement of the primacy condition of anaphora. In the flat propositional component of the Hungarian sentence, in which all the arguments c-command each other, however, the antecedent in (36) both c-commands and precedes the anaphor. To account for (36), we cannot resort to binding principle C, because binding principle C cannot rule out (36) without also ruling out the grammatical (37): (37)

Nem szereti egymast Janos es Mari not like each-other-ACC John-NOM and Mary-NOM 'John and Mary don't like each other'

The application of binding principle C in the case of (37) and (36) is obviously blocked by a prohibition against referential circularity, which does not allow a constituent a to be the antecedent of a constituent b, and at the same time b to be the antecedent of a. Such a prohibition falls out from various theories of binding, e.g. from Higginbotham (1983), or Brody (1981). What we need in order to rule out (36) is a lexical structure equivalent of binding principle C, such as: (38)

An anaphor cannot precede its antecedent in the lexical argument hierarchy.

3 Pronominal variable binding The primacy notion of binding, requiring that the binder c-command the bound element, is also too powerful in licensing pronominal variable binding - especially in non-configurational languages. Consider, for instance, the Hungarian sentences in (39-40). The operator c-commands the pronominal in every example, as required; nevertheless, the sentences differ sharply in grammaticality. (39)

Mindenki; szereti az pro; anyjat7 everybody loves the mother-his-ACC 'Everybody loves his mother' mindenki; szereti b. Az pro; anyjat the mother-his-ACC everybody loves 'His mother, everybody loves' c. Mindenkit; szeret az pro; anyja everybody-ACC loves the mother-his-NOM 'Everybody, his mother loves' d. *Az pro; anyja mindenkit; szeret the mother-his-NOM everybody-ACC loves 'His mother loves everybody'

a.

256

E. Kiss Katalin

(40)

Kivel i beszelgettel a proi felesegerol? who-with talked-you the wife-his-about 'With whom did you talk about his wife?' kivel i beszelgettel? b. A proi felesegerol the wife-his-about who-with talked-you 'About his wife, with whom did you talk?' c. Kiroli beszelgettel a proi felesegevel? who-about talked-you the wife-his-with 'About whom did you talk with his wife?' d. *A Proi felesegevel kirolj beszelgettel? the wife-his-with who-about talked-you 'With his wife, about whom did you talk?' a.

If the c-command of the pronominal by the operator were the sole primacy condition of pronominal variable binding, as is generally assumed, it would remain unclear why (39d) and (40d) are ungrammatical. English sentences in which the antecedent and the pronominal are contained in arguments mutually c-commanding each other raise the same problem: (41) a. b. c. d.

I heard from each mani about his i wife ?I heard about hisj wife from each manj ?I heard about each mani from hisi wife *1 heard from his i wife about each manj

We suspect on the basis of (39-41) that the relative surface order of the operator and the anaphor is among the primacy factors that license binding. Therefore, in English we test pronominal variable binding in echo questions, in which the order of the wh-phrase and the pronominal is not invariant. (42)

a. b. c. d.

You talked with whomj about his i wife? ?You talked about hisj wife with whomj? ?You talked about whomj with hisj wife? *You talked with his i wife about whomi?

In English it is not only the ungrammaticality of (41d) and (42d) that has to be accounted for, but also the more or less (for some speakers more, for others less) marginal nature of (41b, c) and (42b, c). According to Barss & Lasnik (1986), the primacy condition of pronominal variable binding is that the pronominal be in the domain of the operator, i.e. it be both c-commanded and preceded by the operator. This solution not only rules out

The primacy condition of anaphora and pronominal variable binding

257

the ungrammatical (d) sentences in (39-42), but also excludes the grammatical (b) sentences. Besides, neither can it account for the somewhat marginal character of the English (41c) and (42c). Mankz (1986: 1l2), aiming to account for the distribution of grammaticality in (39), suggests the following rule of bound variable interpretation for Hungarian: (43)

(44)

A pronoun can be interpreted as a bound variable iff it does not both cyclicc-command its antecedent at lexical structure and precede it at phrase structure. a cyclic-c-commands b iff b. a c-commands b or b. if c is the minimal cyclic node (NP or S-bar) that dominates a but is not immediately dominated by another cyclic node, then c c-commands b.

In Mankz's interpretation, lexical structure is a representation of the sentence the articulation of which corresponds to the articulation of an English-type, configurational phrase structure. In this framework, the lexical structure assigned to (40a) and (40b) would be one of the three theoretically possible configurational constructions shown in (45a-c). (45)

a.

~

~

NP,

proj (45)

1 T6

beszelgettelj

kiveli

a Proi feleseger5l

c.

b.

A

r

NP 1

V

~

NP 1

VP

NP2

NP '

VP

vA NP2

NP 3

NP3 , a pro felesegerOl, c-commands NP2 , kivel, in each structure. Consequently, in (40b) pro not only precedes its binder at S-structure but also cyclic-c-commands it at lexical structure. The sentence is, nevertheless, grammatical, contrary to Mar-

258

E. Kiss Katalin

acz's prediction. If Maracz's primacy condition were extended to English, it would run into similar problems, e.g. in the constructions represented by (41) and (42). The primacy condition of pronominal variable interpretation to be proposed in this chapter retains Maracz's reference to S-structure precedence; however, it discards the reference to lexical structure c-command (since it cannot properly order two VP-internal NPs or PPs). It introduces the notion 'precedence in the lexical argument hierarchy' instead. According to the evidence of (39-42), the requirement of S-structure precedence and the requirement of precedence in the lexical argument hierarchy interact in the following way: The most felicitous cases of pronominal variable binding are those in which the operator precedes the pronominal both at S-structure and in the lexical argument hierarchy, i.e. the (a) sentences in (39-42). The examples in which only one of the two primacy principles is observed, i.e. the (b) sentences, in which the operator only precedes the pronominal in the lexical argument hierarchy, and the (c) sentences, in which the operator only precedes the pronominal at S-structure, are also grammatical, but - especially in English - are somewhat more marginal than the (a) sentences. If both primacy principles are violated, as in the (d) sentences, the bound variable interpretation is impossible. Apparently, English speakers may attribute different significance to the precedence and the lexical prominence conditions. Certain speakers find (41b) and (42b) perfectly grammatical, while barely accepting (41c) and (42c), or not accepting them at all. For these speakers S-structure precedence plays little or no role in licensing pronominal variable interpretation. There are also speakers for whom (41c) and (42c) are perfectly grammatical, and (41b) and (42b) are marginal. These speakers obviously value the S-structure precedence condition higher than the lexical prominence condition. In addition to S-structure precedence and precedence in the lexical argument hierarchy, S-structure c-command also seems to be involved in licensing pronominal variable binding. It is needed to prevent the bound variable interpretation of the pronominal in such sentences as: (46)

*1 discussed the rumours about each manj with hisj wife

The role of c-command, however, is not straightforward, as the operator phrase in (41) and (42), embedded under a PP, does not, strictly speaking, c-command the pronominal either. Although in the (a) and (c) sentences, the P could be assumed to be reanalysed as part of the verb, in which case the c-command condition would be satisfied, in the (b) sentences, which represent for most informants the same degree of grammaticality, reanalysis is impossible. It does not seem to be unreasonable, however, to assume that the feature 'universally quantified' in (41), or the

The primacy condition of anaphora and pronominal variable binding

259

wh-feature in (42), can percolate up to the PP-node containing the operator, yielding an operator phrase that c-commands the pronoun. The assumption that an operator feature can percolate up across one phrase boundary is, in all probability, independently needed in English to account for such cases as: (47) Everyone/s mother thinks hej is a genius Our tentative claim that the operator feature can percolate up across one maximal projection is based on the distribution of grammaticality in the following sentences: (48) (49) a. b. c.

*1 talked with the wife of each manj about hisj mother 1 talked with each man/s wife about hisj mother 1 talked about each man/s wife with hisj mother *1 talked about hisj mother with each manj's wife

Let us assume that the preposition immediately following the verb is reanalysed as part of the verb in (48) and (49). Let us, crucially, also assume that the feature 'universally quantified' percolates up from the NP minimally containing it to the next higher maximal projection. Then in (48) the resulting quantified PP (of each man) does not c-command the pronominal, while in (49a, b), the resulting quantified NP (each man's wife) does c-command it. In (49c) the preposition of the PP containing the operator cannot reanalyse, therefore the NP to which the operator feature has percolated up (each man's wife) does not c-command the pronominal. That is, the distribution of grammaticality in (48) and (49) falls out. The c-command requirement of pronominal variable binding has to be extended so as to allow the operator phrase to c-command, instead of the pronominal, the trace of the pronominal (or of the phrase containing the pronominal) - as the (a) sentences in (50) and (51) below indicate. (The extended c-command condition is satisfied also in the ungrammatical (SOb) and (51 b); they are ruled out because they observe neither the S-structure precedence requirement, nor the lexical prominence requirement of binding.) (50) a. b. (51) a. b.

?About hisj wifej, 1 talked with each manj tj for hours *With hisj wifej, 1 talked with each manj tj for hours ?About which ofhisj wife's friends j did you talk with each manj tj? *With which ofhisj wife's friends j did you talk about each manj tj?

The three primacy principles involved in licensing pronominal variable interpretation are not only identical to those involved in licensing anaphora, but they also interact in the same way. That is:

260

E. Kiss Katalin

(52) The primacy condition of pronominal variable binding A pronominal can be interpreted as a variable bound by an operator iff the operator phrase c-commands the pronominal or its trace at S-structure, and at least one of (i) and (ii) also holds: (i) the operator phrase precedes the pronominal in the lexical argument hierarchy (ii) the operator phrase precedes the pronominal at S-structure While the primacy condition of pronominal variable binding in (52) correctly predicts the binding possibilities in all the sentence types we have discussed, it seemingly breaks down in interrogative double object constructions in which the direct object functions as the wh-operator, and the indirect object contains the pronominal. Consider the following example, taken from Barss & Lasnik (1986): (53) *Which lioni did you show itsi trainer ti? The problem with (53) is that the intended but unavailable interpretation of the sentence, in which the wh-phrase binder of the pronoun is understood as the direct object, is predicted to be possible by principle (52). The wh-operator both ccommands the pronoun and precedes it; i.e. two of the three primacy requirements of pronominal variable interpretation are satisfied. Nevertheless, the intended interpretation does not arise. Notice, however, that the pronoun can be understood as a variable bound by the wh-operator in (53) - if the wh-phrase is interpreted as the indirect object, i.e. if the sentence is analysed as follows: (54) Which lioni did you show ti itsi trainer? In view of this fact, the question is why the former of the two theoretically

possible interpretations of the sentence ((53) and (54» never arises. We suggest the following explanation: In structure (54) all the three primacy requirements of bound pronominal interpretation are observed: the wh-trace c-commands the pronoun, the wh-phrase precedes the pronoun, and the wh-argument is also more prominent than the pronoun in the lexical argument hierarchy. In (53), on the other hand, only two elements of the primacy condition are satisfied - the c-command requirement, and the S-structure precedence requirement - so if the bound pronominal interpretation which is theoretically possible in (53) could be evoked, it would be somewhat marginal, representing the same degree of grammaticality as, for example, (41c) and (42c). Apparently, if English speakers have to choose between a marginal and a fully grammatical interpretation in the case of structural homonymy, they invariably choose the fully grammatical interpretation.

The primacy condition of anaphora and pronominal variable binding

261

4 Summary and conclusions

The analysis of anaphora and pronominal variable binding in English and Hungarian has revealed that anaphoric binding and pronominal variable binding are licensed by the same primacy condition. This condition requires the prominence of the binder over the bound element in terms of at least two ofthree primacy factors: S-structure c-command, lexical (thematic) prominence, and S-structure precedence. The proposed primacy condition of binding is inapplicable in the case of binding principles Band C, as the following examples show: (55)

(56)

a. *Beszeltem Janossal; rola; spoke-I John-with about-him 'I spoke with John about him' b. *Beszeltem J anosrol; vele; spoke-I John-about with-him 'I spoke about John with him' a. *Beszeltem Janos; anyjarol vele; spoke-I John's mother-about he-with 'I spoke about John's mother with him' b. *Beszeltem Janos; anyjaval rola; spoke-I John's mother-with about-him 'I spoke with John's mother about him'

If binding principles Band C were sensitive to lexical prominence, i.e. if disjoint reference only arose if the NP c-commanding the pronominal in (55) or the Rexpression Janos in (56) also preceded the pronominal or Janos in the lexical argument hierarchy, then (55b) and (56b) would not display any disjoint reference effects. The fact that anaphoric binding and pronominal variable binding are subject to the same primacy condition, while anaphoric binding and disjoint reference, i.e. binding principle A on the one hand, and binding principles Band C on the other hand, are subject to different primacy conditions, suggests that anaphoric binding and pronominal variable binding may form a more natural class than anaphoric binding and disjoint reference; that is, a reorganization of binding theory perhaps along the lines proposed in Reinhart (1983a) - might be desirable.

Notes 1. Barss (1986) proposes a version of binding principle A which combines the primacy

262

2.

3. 4.

5.

6. 7.

E. Kiss Katalin

requirement and the locality requirement of anaphora into a single notion: the notion of minimal chain-accessibility. According to Mohanan (1983-4), lexical structure is distinct from phrase structure (or configurational structure) in configurational and non-configurational languages alike. While lexical structure is the representation of grammatical functions, configurational structure is the representation of dominance and precedence relations. Mohanan claims that anaphora is interpreted at lexical structure in both language types. Delative Case represents a source; it may be rendered in English by prepositions such as of, from or about. A thematic hierarchy similar to (16) also plays a role in the anaphora theories of Hellan (1986b) and Kiparsky (1987), and also in the theory of long-distance anaphora of Giorgi (1984). It has also been claimed that while himself in (17a) is a beneficiary, to himself in (17b) is a goal. According to this view, the anaphor cannot represent the same, beneficiary, O-role in both cases because, if a beneficiary could alternatively be realized either as an NP or as a to-NP, then it could not be explained why the beneficiary in, for example, I gave him a headache has no to-NP variant. The morphological realization of O-roles, however, is known to involve a lot of idiosyncrasies, therefore it would not be illegitimate to include a constraint in the lexical entry associated with the idiom give somebody a headache to the effect that the beneficiary argument can only be realized as a bare NP. E corresponds to a 'hidden' argument place for events in the semantic theory of Higginbotham (1985). Hungarian is a pro-drop language. The subject pronoun also drops in the NP, where it is coindexed with an Agr-marker on the head noun.

13 The local nature of the long-distance reflexive in Chinese C.-T. James Huang and C.-C. Jane Tang

1 Introduction Chinese has two reflexive forms: the bare reflexive having the invariant form ziji 'self' and the compound reflexive having the form of a pronoun + ziji sequence, as in taziji 'himself/herself', niziji 'yourself', etc. As described in Y.-H. Huang (1984), Tang (1989), and Wang & Stillings (1984), these elements exhibit distributional and referential properties that are of considerable interest to linguistic theory. As far as their reference is concerned, one property of the reflexives is that the bare reflexive, though not the compound reflexive, exhibits possibilities of having a long-distance antecedent apparently outside of its governing category. Thus in (Ia) ziji may have either Zhangsan or Lisi as its antecedent, but in (Ib) taziji must be locally bound by Lisi: (1)

a.

b.

Zhangsani renwei [Lisij hai-Ie zijiilil Zhangsan think Lisi hurt-ASP self 'Zhangsani thought that Lisi j hurt himselfilj ' Zhangsani renwei [Lisi j hai-Ie ta-ziji*ilj] Zhangsan think Lisi hurt-ASP himself 'Zhangsani thought that Lisij hurt himself*ilj'

Long-distance binding with ziji is, however, restricted by a condition that requires the remote antecedent to agree in person and number features with all closer potential antecedents. In particular, a remote NP can antecede ziji only if it agrees with the local NP in the governing category of ziji. Thus, although ziji may have Zhangsan as its antecedent in (Ia), where it agrees with Lisi in person and number, long-distance binding is blocked in examples like (2), where the remote NP differs from the local NP either in person or in number, or both. In all these cases, ziji must be bound by the local NP: I

263

264

(2)

C.-T. James Huang and C.-C. Jane Tang

a.

b.

c.

d.

Zhangsani renwei [woi hai-Ie ziji*ili] Zhangsan think I hurt-ASP self 'Zhangsan thought that I hurt myself' Nii renwei [Zhangsani dui ziji*ili meiyou xinxin] you think Zhangsan to self not-have confidence 'You think that Zhangsan has no confidence in himself' WOi renwei [womeni yinggai dui ziji*ili you xinxin] should to self have confidence I think we 'I think we should have confidence in ourselves' Womeni renwei [tai dui ziji*ili meiyou xinxin] we think he to self not-have confidence 'We think that he has no confidence in himself'

In the following sentence, long-distance binding of ziji is also blocked, in spite of the fact that the remote NP agrees with the most local NP: (3)

Zhangsani shuo [woi zhidao [Lisi k chang piping ziji*iI*i/kll Zhangsan say I know Lisi often criticize self 'Zhangsan said that I feel that Lisi always criticized himself'

This is because of the intervening NP wo '1', which agrees with neither Zhangsan nor Lisi. If wo is replaced by Wangwu, as in (4), long-distance binding is again allowed: 2 (4) Zhangsani shuo [Wangwui zhidao [Lisik chang piping zijiifi/k]] 'Zhangsani said that Wangwui knew that Lisi k often criticized selfifi/k' The purpose of this chapter is to consider how these facts regarding the longdistance reflexive may be best explained. In particular, we will be concerned with (a) why only the bare reflexive may exhibit long-distance binding, and (b) why long-distance ziji is subject to the strict requirement of agreement just described. In section 2 we briefly indicate general conditions on what may qualify as a 'potential antecedent of the Chinese reflexives, and in section 3 we review two recent accounts of long-distance ziji. The discussion in these two sections will help to crystallize the nature of the problems we are dealing with, which will then lead to our proposal in section 4, followed by a brief conclusion in section 5. It will be our claim that the phenomena observed are best explained if we assume that the bare reflexive is an anaphor in two ways, since it lacks not only reference, but also intrinsic features normally associated with pronouns, and that as such it needs to receive two indices under binding theory, first at S-structure and again at LF. Other principles of grammar will combine to derive the facts to be explained.

The local nature of the long-distance reflexive in Chinese

265

2 Potential binders of the reflexive Before discussing the issue of long-distance binding, we must identify a few conditions on what, in general, may qualify as a binder, local or remote. First, the Chinese reflexive ziji can be bound only to a subject, but not in general to an object (see C.-T. J. Huang (1982». (5)

WO i gaosu Lisi j zijiiJ*j de fenshu 1 tell Lisi self's grade 'I told Lisi my own grade'

Second, only an animate, not an inanimate, NP can antecede ziji: (6)

Wo bu xiaoxin dapo-Ie ziji de yanjing 1 not careful break-ASP self's glasses 'Not being careful, 1 broke my own glasses' (7) *Yanjing diao-dao dishang, dapo-Ie Zl)l glasses drop-to floor break-ASP self 'The glasses dropped on the floor, and broke themselves' Generally, any c-commanding animate subject NP in the governing category of a reflexive may be its antecedent. There are two situations under which a non-ccommanding subject may be an antecedent. The first situation arises when a sentence contains a 'psychological' verb: (8) [Zijii de xiaohai mei de jiang de xiaoxi] shi Lisii hen nanguo self's child not get prize DE news make Lisi very sad 'The news that his own child did not get a prize made Lisi sad' The special status of psych-sentences is, of course, well known since Postal (197l), and recent work by Giorgi (1984) and others has revived an interest in such constructions (cf. Pesetsky (1987b), Belletti & Rizzi (1988), etc.). We will not discuss in this chapter how psych-sentences are to be analysed, but simply note that, in Chinese too, the experiencer argument may bind a reflexive even though the element is otherwise subject oriented. The other situation where an NP that is not a c-commanding animate subject may antecede ziji is when it appears as the 'most prominent' animate subject NP within an inanimate NP that c-commands ziji. This is illustrated in (9), to be compared with (10): (9)

[Woi de jiaoao]j hai-Ie Zl)liJ*j 1 's pride hurt-ASP self 'My pride hurt myself'

266

G.-T. James Huang and C.-C. Jane Tang

(10) [WOj de meimei]j hai-Ie ZlJhi/j I 's sister hurt-ASP self 'My sister hurt herself' Clearly, the fact that the non-c-commanding wo 'I' can be the antecedent of ziji is related to the fact that the c-commanding NP wo dejiaoao 'my pride' is inanimate, hence not a potential antecedent. To capture this intuition, Tang (1989) proposed the notion of 'sub-command': (11)

f3 sub-commands ex iff f3 is contained in an NP that c-commands ex or that sub-commands ex, and any argument containing f3 is in subject position.

The condition under which the c-command requirement may be relaxed is stated as in (12).3 (12)

A reflexive ex may take an NP f3 as its binder if a. f3 sub-commands ex, and b. there is no NP 'Y, 'Y a potential binder for ex, such that 'Y is closer to ex than f3 is.

For the notion of relative distance, assume that, other things being equal, a ccommander of ex is closer than a sub-commander is to ex, and a c-commander or sub-commander in the minimal clause containing ex is closer than one outside of the minimal clause, etc. In the case where a sub-commander is contained in a ccommander, the c-commander is closer to ex. A 'potential binder' is any NP that satisfies all conditions of being a binder of ex except that it is not yet coindexed with ex. Given (11-12), in (9) wo 'I' can antecede ziji, for the former sub-commands the latter and there is no potential binder closer to the reflexive. The only NP closer to ziji than wo 'I' is the NP wo de jiaoao 'my pride', which, not being animate, is not a potential binder. In (10), on the other hand, wo 'I' cannot antecede ziji, because, though the former sub-commands the latter, it is contained in the animate NP wo de meimei 'my sister', which is a potential binder and is closer to ziji. (See Tang (1989) for more details and other facts captured by (11-12).) Summarizing, we have seen that the reflexive may take as its antecedent an animate NP that is (a) a c-commanding subject, (b) a sub-commanding subject, or (c) an experiencer. Without attempting an exact formulation ofthe term, we shall refer to coindexing with any of these NPs as an instance of binding (even though the standard definition of binding entails c-command). We have seen examples in which certain NPs are potential local binders. What about potential long-distance binders? There is reason to believe that the con-

The local nature of the long-distance reflexive in Chinese

267

ditions are essentially the same (except for the restrictions noted in section 1). For example, we have seen in section 1 that a c-commanding subject can be a longdistance binder (see (la) and (4)). Furthermore, an experiencer can be a longdistance binder, too, as illustrated below: (13) [Zhangsanj taoyan zijii/j de xiaoxi] shi Lisi j hen nangguo Zhangsan dislike self DE news make Lisi very sad 'The news that Zhangsanj disliked selfi/j made Lisi j very sad' The only exception is that a sub-commander cannot, in general, be a long-distance binder. (14) allows only local binding: (14)

Zhangsanj de xin biaoshi [Lisij hai-Ie ziji'i/j] Zhangsan 's letter indicate Lisi hurt-ASP self 'Zhangsan's letter indicates that Lisi hurt himself'

This may be attributed to the fact that a sub-commander is picked out as a (marked) antecedent only as a 'last resort', when there is no other more accessible NP that can bind a given reflexive (cf. condition (l2b)). In a situation where a subcommander occurs in a remote position, there is already an NP in a local position that is more accessible to the reflexive. Therefore, given (12b), a sub-commander cannot be a long-distance binder. 4

3 Long-distance ziji: previous analyses How does the 'long-distance' ziji fit into an optimal theory of grammar? From what we have seen so far, it is clear that ziji is really not something whose referential properties are unconstrained by principle A of binding theory, nor an element that is free from the locality restriction imposed by the notion of minimal governing categories. The restrictions we saw suggest that in order for ziji to be bound by an NP outside of its governing category, it must first be licensed by an NP in its governing category that agrees in person and number with the remote NP. And each further remote NP may be an antecedent only if ziji can be successively licensed by all lower potential antecedents in the same way. This indicates that there is a sense of strict locality involved here, and that so-called long-distance binding should be described in terms of successive steps of local binding. 'Longdistance' ziji, in other words, should not be admitted by parametrizing the notion of a governing category, as might be suggested along the lines of Yang (1983) and Manzini & Wexler (1987). It would also be inappropriate to simply define ziji as a non-anaphor, as a 'pronominal anaphor' (Wang & Stillings (1984)), or as a bound pronoun (e.g. Sportiche (1986)). For all these proposals simply relax the locality

268

G.-T. James Huang and C.-C. Jane Tang

conditions on the binding of ziji, thus failing to capture the strict locality requirements observed. To capture the relevant restrictions, Tang (1985, 1989) proposed that the bare ziji originates as pro-ziji, a compound reflexive with an empty pro, and that the limited cases of long-distance ziji are derived from the optional feature-copying rule (15) and the iterative reindexing rule (16): (15) Feature-copying rule The pro in a pro-ziji reflexive may transfer its features (such as person and number) to -ziji after the application of binding theory, thus turning ziji into a 'long-distance' reflexive. (16) Reindexing rule Reindex the long-distance reflexive with the potential antecedent of the next-higher governing category. Given (15) and (16), the restrictions noted in section 1 are accounted for. Consider (la) and (lb), above. Since the feature-copying rule applies only to proziji, the compound reflexive in (lb) is unaffected and remains a local reflexive. In (la), on the other hand, pro-ziji can optionally undergo rule (15). If (15) does not take place, the pro-ziji is still a local reflexive and is bound to the local Lisi. If (15) applies, ziji becomes a long-distance reflexive, carrying the person and number features of Lisi (after binding theory has applied). At this time, ziji is reindexed with Zhangsan, under (16). Since ziji agrees with Zhangsan in person and number features, reindexing is allowed, and ziji comes to be bound by a remote antecedent. The blocking effects indicated above also follow straightforwardly. Consider the sentences in (2) and (3) above. When binding theory applies at S-structure, ziji in each of these sentences is bound by its local antecedent. So if the copying rule applies in (2a), ziji must carry the features [1st person, singular] as it is turned into a long-distance reflexive. But this prevents it from taking Zhangsan as its remote antecedent under reindexing. Therefore ziji cannot be turned into a long-distance anaphor. Similarly, since reindexing is required to be successive-cydic, longdistance binding by Zhangsan is also blocked in (3). Although the proposal embodying (15-16) accounts for the relevant facts, it also leaves a number of questions unanswered. For example, it does not explain why the reindexing rule (16) applies only to ziji, nor does it explain why the copying rule can change a local reflexive into a long-distance one. Furthermore, as Battistella (1987) points out correctly, it is not clear why copying should trigger reindexing, nor why reindexing should mimic the effects of binding. In recent years, a new analysis of the reflexive anaphor has aroused considerable

The local nature of the long-distance reflexive in Chinese

269

interest. Inspired by the work of Lebeaux (1983) and Chomsky (1986a), a number of writers (Pica (1987), Battistella (1987), among others) have suggested that certain reflexives may be raised in LF into Infl, in a way analogous to clitic movement in syntax, thereby accounting for their subject orientation. Furthermore, certain such reflexives may move from Infl to Infl (an instance of head-tohead movement, in a way analogous to the phenomenon of 'clitic climbing'), thereby accounting for their long-distance binding possibilities. Under Battistella's approach, for example, the compound reflexive is a full NP, whereas the bare ziji is an NO. Given that Infl-to-Infl movement is a head-to-head movement, he argues that only ziji may undergo this Infl-to-Infl movement. Assuming that no successive-cyclic movement is otherwise available for the compound reflexive, he accounts for the fact that only ziji exhibits long-distance binding. To account for the blocking effects of ziji, he proposes (a) that Inflmovement must go successive-cyclically, and (b) that each trace left by ziji in Infl, as well as the moved ziji itself, must agree in grammatical features with its own local subject, as a general requirement of subject-Infl agreement, construed abstractly in Chinese. Since all traces must be coindexed with the moved ziji, it follows that all local and non-local subjects must agree in person and number, in cases where ziji has a remote antecedent. Attractive as it is, the Infl-movement theory is faced with important difficulties. For one thing, Battistella does not explain why the compound reflexive taziji cannot adjoin successive-cyclically in LF, giving rise also to long-distance binding. (Note incidentally that the Lebeaux-Chomsky proposal in fact assumes that the compound reflexive himself is moved to Infl.) A more serious problem concerns the blocking effects. According to Battistella, the blocking effects follow partially from the fact that Infl is the locus of agreement. Notice, however, that the potential blockers oflong-distance ziji include not only c-commanding local subjects (as we have seen in all the relevant examples), but also local sub-commanders and experiencer non-subjects. For example, in the following sentences, ziji must be bound by its local sub-commander, but not by the matrix subject: (17)

(18)

Zhangsani shuo [WOj de jiaoao hai-Ie ziji*iliJ Zhangsan say I 's pride hurt-ASP self 'Zhangsani said that mYj pride hurt myself*ilj' Zhangsanj shuo [[nij zheyang zuo] dui ziji*ilj bu Ii] Zhangsan say you thus do to self not advantage 'Zhangsanj said that yourj doing this will do yourself*ilj no good'

270

G.-T. James Huang and C.-C. Jane Tang

(19)

Zhangsani shuo [[nij zuo shi de taidu] du ziji*i/j bu hao] Zhangsan say you do work REL attitude to self not good 'Zhangsani said that the attitude with which yOUj work is not good for yourself*i/j'

Battistella suggests that in these cases, the verb of the embedded clause containing

ziji in effect agrees with the sub-commanding NP. However, this way of looking at subject-Inft agreement does not seem seriously entertainable. There is little reason, other than to derive the blocking effects, to say that a matrix verb agrees not with its own subject, but with the subject of its sentential subject (as in (18)) or ofits complex NP subject (as in (19)).5 Furthermore, (20) shows that an experiencer non-subject may block long-distance ziji. (20)

[[[Zhangsani dui zijii/*j/*k mei xinxin de shi] shi WOj hen nanguo Zhangsan to self no confidence's fact make me very sad de xiaoxi] shi Lisi k hen yiwai] DE news make Lisi very surprised 'The news that I was saddened by the fact that Zhangsan had no confidence in himself surprised Lisi'

In this sentence, the matrix predicate shi Lisi henyiwai 'make Lisi very surprised' takes a complex NP subject meaning 'the news that the fact that Zhangsan has no confidence in himself saddened me'. Long-distance binding of ziji by the outermost experiencer Lisi is blocked by the inner experiencer wo 'me'. Here the blocker is not a subject, and it thus looks even more unlikely that the blocking effects have to do with subject-Inft agreement. Rather, the emerging generalization we want to capture is the following; (21) The set of potential blockers of long-distance ziji is exactly the set of its potential local, or less remote, binders. This generalization suggests that the blocking effects should not be treated as an effect of agreement, but as a property of binding. Finally, since in the Inft-movement theory the movement of ziji in LF is a case of head movement, the traces left over by ziji are subject to antecedent-government (see Chomsky (1986b, 1988b)). While this has the consequence that the movement must be successively cyclic, it incorrectly rules out certain acceptable cases of longdistance ziji. In particular, we know from independent evidence that movement of a phrase whose trace needs to be antecedent-governed cannot cross any singular

The local nature of the long-distance reflexive in Chinese

271

barrier. Thus, adjuncts located in adverbial clauses and relative clauses cannot be wh-moved out of these islands. This is true both in the syntax (22) and in LF (23): (22) a. *WhYi did you go home [before John bought the book ti]? b. *WhYi did you like [the man who kicked Bill ti]? (23) a. *Suiran Lisi weishenme mei lai, ni haishi bu shengqi? though Lisi why not come you still not angry '*Though Lisi didn't come why, you weren't angry?' b. *Ni zui zihuan [ta weishenme mai de shu]? you most like he why buy REL book '*You like the book that he bought why?' Furthermore, Huang (1982) argues that A-not-A questions in Chinese exhibit ECP effects. The A-not-A element is an element in Infl, and Infl-movement cannot cross barriers: (24)

a. *Ruguo ta lai-bu-Iai, ni jiu hui shengqi? if he come-not-come you then will angry '*If he comes or not, then you will be angry?' a. *Ni zui xihuan ta mai-bu-mai de shu? you most like he buy-not-buy REL book '*You like the books that he will buy or will not buy?'

These facts lead one to expect that, if ziji undergoes head-to-head movement in LF, no long-distance binding is possible across adjunct clauses or complex NPs. But this prediction is incorrect. In the following sentences, long-distance binding is fully acceptable, suggesting that LF-traces of ziji are not subject to antecedent-government. (25)

(26)

Zhangsani shuo [ruguo Lisi j piping zijii/j], ta jiu bu qu Lisi criticize self he then not go Zhangsan say if 'Zhangsani said that if Lisij criticized himselfi/j, then he won't go' renj] Zhangsani bu xihuan [neixie piping zijii/j de Zhangsan not like those criticize self REL person 'Zhangsani does not like those peoplej who criticize selfi/j'

In view of these problems, we must now look elsewhere for an explanation oflongdistance ziji. 6

272

G.-T. James Huang and C.-C. Jane Tang

4 The locality of long-distance ziji

Although the Infl-movement theory cannot account for the locality restrictions of long-distance ziji in a proper way, one property of the theory that seems to us to be correct is the idea that the locality restrictions are to be expressed by successivecyclic movement of ziii in LF. We shall pursue an explanation along this line, but attempt to derive the locality requirements from other sources. To see how this may be done, consider the following sentences: (27)

(28)

a. b. c. a. b. c.

John knows that Bill likes pictures of himself John knows that, pictures of himself, Bill likes Pictures of himself, John knows that Bill likes John knows that Bill likes these pictures of himself John knows which pictures of himself Bill likes. Which pictures of himself does John think that Bill likes?

These sentences exemplify the so-called 'reconstruction problem' or 'connectivity effect', well known since Higgins (1973) and more recently Barss (1986). In each (a) sentence above, the reflexive must have Bill as its antecedent, but not the remote John. In the (b) sentences, however, himself may have either John or Bill as its antecedent, though only John actually c-commands the reflexive. And in the (c) sentences, either John or Bill may antecede himself, though neither c-commands the latter. The suppression of the c-command requirement in these sentences is dealt with in Barss (1986) in terms of a condition of 'chain accessibility' on binding theory as applied at S-structure. That is, by virtue of its relation to a trace in the minimal c-command domain of John or Bill, himself is defined as being 'chain bound' by John or Bill in the (b) and (c) sentences. What is important here is that in these cases a locality requirement is still maintained for the binding of himself to be possible. Thus, binding by John in the (b) sentences is possible only because himself has been moved to a position where its governing category contains John. In (29), where himself is in the most deeply embedded Comp, only Bill may antecede himself: (29) John knows that Bill wondered which pictures of himself I would buy Similarly, Barss shows that the (c) sentences allow John to antecede himself only because the NP (which) pictures of himself binds a trace in the intermediate Comp, creating a chain configuration in which John is allowed to bind himself as a 'minimally accessible' antecedent. (See Barss (1986: chapter 3) for more details.) From the point of view of D-structure (cf. (27a) and (28a)), then, we may say that a reflexive in its D-structure argument position has only a local antecedent, but may

The local nature of the long-distance reflexive in Chinese

273

pick up a 'long-distance' antecedent as a result of successive-cyclic movement. But from the point of view of S-structure, all 'long-distance' antecedents are in fact local ones, each being a minimally accessible 'chain binder' in the sense of Barss. What we would like to suggest is that the 'long-distance' ziji is essentially the same phenomenon as that illustrated in (27-9), except that it is a phenomenon that occurs in LF rather than in the syntax. That is, from the point of view of Sstructure, ziji in its S-structure argument position has only a local binder, but may pick up a remote antecedent as a result of successive-cyclic movement in LF. From the LF point of view, however, all 'remote antecedents' are local antecedents. This proposal is similar to the Infl-movement or the 'clitic-climbing' theory, but we claim that the LF movement involved is simply A'-movement, more specifically IP-adjunction, perhaps as a case of QR. 7 Thus, the ambiguous readings of (30) are unambiguously represented at LF as in (31), with ziji bound in each case by a local antecedent: (30)

Zhangsan manyuan Lisi chang shuo Wangwu bu xihuan ziji Zhangsan complain Lisi often say Wangwu not like self 'Zhangsan complained that Lisi often said that Wangwu does not like Wangwu/LisilZhangsan' (31) a. Zhangsank manyuan [Lisi j chang shuo [Wangwui bu xihuan zijii]] b. Zhangsank manyuan [Lisij chang shuo [ziji j [Wangwui bu xihuan til]] c. Zhangsank manyuan [zijik [Lisi j chang shuo [tk [Wangwui bu xihuan tk]]]]

Our proposal thus assumes that the reference of ziji can be determined by binding theory applying at LF. This by itself is not a problematic assumption, given the arguments of Aoun (1985) and the discussion in Chomsky (1982: note 11). However, there is well-known evidence that binding theory must also apply at S-structure (see Chomsky (1981, 1982), Barss (1986), among others). For example, the binding possibilities of himself in (27-9), or in (32-3) below, must be determined at S-structure: (32)

a. b. (33) a. b.

John said that Bill criticized himself John said that, himself, Bill criticized John said that Bill likes every picture of himself John said that, every picture of himself, Bill likes

In these cases, himself can take John as its antecedent only as a result of its movement in the syntax. In LF, IP-adjunction of every picture of himself or himself may take place (under QR or as instances of Move u), but this movement does not alter binding possibilities of himself as does movement in the syntax. The same

274

C.-T. James Huang and C.-C. Jane Tang

point holds in Chinese with sentences containing pronoun + ziji. The compound reflexive taziji can have Zhangsan as its antecedent only in (34b): (34) a. Zhangsan shuo Lisi chang piping taziji Zhangsan say Lisi often criticize himself 'Zhangsan said that Lisi often criticized himself' b. Zhangsan shuo, taziji, Lisi chang piping 'Zhangsan said that himself, Lisi often criticized' Given that taziji may be IP-adjoined in the syntax (as in (34b)), nothing seems to prevent the same element in (34a) from being IP-adjoined in LF. The fact that (34a) does not allow Zhangsan to antecede taziji shows that LF-movement does not alter the binding possibilities of taziji. The index of taziji that is licensed (or "1marked, extending Lasnik & Saito's (1984) terminology) by principle A at Sstructure remains in LF wherever taziji goes. Therefore, not only must binding theory apply at S-structure, but the following must also hold, in both Chinese and English: (35) The indices licensed by the binding theory at S-structure cannot be undone inLF. This means that binding theory, if it applies in LF, can affect only NPs whose indices are not already licensed at S-structure with respect to specific binding principles. These considerations, however, contradict the hypothesis that longdistance binding of ziji arises as a result of LF-movement. A sentence like (30) is assigned the reading (31a) at S-structure, and, given (35), LF-movement should not be expected to derive representations like (31b) or (31c). Note that this is a problem not only for the hypothesis we are entertaining, but also for the Inflmovement or clitic-climbing account discussed in the preceding section. Clearly, the difference in referential behaviour between the bare reflexive in Chinese and the compound reflexive in Chinese and English must be tied to their difference in form. Note that the bare reflexive is more 'anaphoric' than the compound reflexive, in that it not only lacks inherent reference as the compound reflexive does, but also contains less 'sense' than the latter: it does not contain '