Introduction

It has standardly been assumed in generative grammar that the distribution of reflexive pronouns is determined by syntactic principles that govern the conditions under which two NPs can be coreferential. Recently, Reinhart and Reuland (1993) have argued against such an approach in favor of a return to the perspective of traditional grammar (e.g., Jespersen 1933) in which the distribution of reflexive elements is determined by properties of reflexive predicates and not by properties of the reflexive pronouns themselves. In this paper I will show that a complete account of the phenomena associated with reflexive elements in language requires principles about both the syntactic conditions licensing anaphora as well as the semantic and lexical conditions licensing reflexive predicates. In order to see how this theory works, we will pay particular attention to languages with "antilocal anaphors," i.e., anaphors which have been claimed to resist local binding. It is in precisely these cases where it becomes apparent that both a syntactic binding theory as well as a theory of reflexive predicates are required.

Antilocal Anaphors

Consider the difference between Dutch and Chinese, illustrated in (1-2):

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

b. Zhangsan renwei Lisi piping ziji
Zhangsan thinks Lisi critized self
'Zhangsan thinks that Lisi criticized him/himself'

c. Lisi piping taziji
Lisi criticized himself
'Lisi criticized himself'

d. Zhangsan renwei Lisi piping taziji
Zhangsan thinks Lisi criticized himself
'Zhangsan thinks that Lisi criticized himself'

(2) Dutch (Everaert 1986)
a. *Max haat zich
Max hates self
'Max hates himself'

b. Max hoorde mij over zich praten
Max heard me about self talk
'Max heard me talk about him

c. Max haat zichzelf
Max hates selfself
'Max hates himself'

d. *Max hoorde mij over zichzelf praten
Max heard me about selfself talk
'Max heard me talk about him'

In Chinese, there are two anaphors. The morphologically complex anaphor, taziji, allows only local binding, while the morphologically simplex anaphor, ziji, can be bound locally, or across a clause boundary. Dutch also has two anaphors. The morphologically complex anaphor, zichzelf, behaves like the corresponding Chinese anaphor in allowing only a local antecedent. However, the Dutch morphologically simplex anaphor, zich, permits long-distance binding but differs from the corresponding Chinese anaphor in that it does not permit local binding.

The difference between morphologically simplex and morphologically complex reflexives in terms of potential binding domains has been related to the theory of movement (Pica 1987; Batistella 1989; Cole, Hermon and Sung 1990; Cole and Sung 1994). In particular, it has been claimed that because morphologically simplex anaphors are heads, they may move at LF through INFL to Comp and then into the next clause, where they can be locally bound. This is illustrated in (3), two possible LF representations of (1b):[1]

(3) a. [Zhangsan [zhidao [Lisi INFL-ziji(i) [piping t(i) ]]]]]

b. [Zhangsan INFL-ziji(i) [zhidao [Lisi INFL-t(i) [piping t(i)]]]]

In (3a), the anaphor moves to the INFL of its own clause where it can be locally bound only by the subject of that clause. In (3b), the anaphor moves out of its own clause to the matrix INFL, where it can be locally bound only by the matrix subject. Morphologically complex anaphors, on the other hand, are taken to be XPs and thus cannot undergo head-movement out of the clause. As a consequence, potential antecedents for these forms must be within the clause containing the anaphor on the surface. In both cases, the binding theory applies to the LF representation and no changes to the binding theory itself are required to explain the difference between long-distance and local anaphors.

I will assume that this analysis is basically correct. What it fails to explain, however, is the existence of anaphors like Dutch zich which appear to disallow local binding. Various attempts to bring these anaphors into line with the head-movement theory have been made (see, inter alia, Hestvik 1990) with varying degrees of success. What I will show in this paper is that the binding properties of these apparent antilocal reflexives are no different from those of other long-distance reflexives. The apparent antilocality effect is epiphenomenal and falls out from an interaction between the semantics of reflexivity and the lexical representations (i.e., argument-structures/theta-grids) associated with reflexive predicates. In developing this analysis, we will see that a complete account of the phenomena associated with reflexivity requires independent principles in the syntactic, semantic and lexical components, none of which can be reduced to the others.

Predicate-centered Binding Theory

Motivations: Locally bound antilocal reflexives

We stated above that Dutch zich does not allow local binding (2a). This observation is not quite accurate, however. Putting aside the question of the exact definition of binding domain, we find examples where the simplex anaphor patterns with the complex anaphor zichzelf and not with the pronominal hem.

(4) a. Max legt het boek achter zich
Max put the book behind self
'Max put the book behind him'

b. Max legt het boek achter zichzelf
Max put the book behind selfself
'Max put the book behind him'

c. *Max legt het boek achter hem
Max put the book behind him
'Max put the book behind him'

If the simplex anaphor had an antilocality requirement, then we would expect it to pattern like the pronoun in (4) and not like the morphologically complex anaphor. Since it does not, we can conclude that the restriction against local binding of zich is different from the restriction against local binding of pronouns, which are truly antilocal.

We might say that the antilocality requirement of zich is that it cannot be bound by a coargument (see Everaert 1986; Reinhart and Reuland 1993). However, as others have observed, the coargument restriction on zich is obviated when the predicate is inherently reflexive:

(5) a. Max wast zich
Max washes self
'Max washes himself'

b. Max schaamt zich
Max shames self
'Max is ashamed'

This observation leads us to the conclusion that the coargument restriction on zich is a fact about predicates and not about binding of the anaphor. That is, zich can be locally bound in some cases and bound by a coargument in some cases. So, assuming that syntactic principles are inviolable, we can conclude that the binding conditions are satisfied in examples like (2a) and that it is some property of the predicate which causes them to be ungrammatical.

Similar conclusions can be reached in unrelated languages. In Kannada, for example, we find an alleged antilocal anaphor which can be locally bound in certain cases (Amritavalli 1991; Lidz 1995):

(6) a. *Hari tann-annu hode-d-a
Hari self-ACC hit-PST-3SM
'Hari hit himself'

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

In (6a), the anaphor tannu cannot be bound by a coargument. However, when the predicate is morphologically marked as reflexive with the morpheme -koND-,[2] coargument binding of tannu is possible, as shown in (6b). The coargument restriction on tannu is therefore dependent on properties of the predicate and not statable simply in terms of syntactic domain, since the relationship between the anaphor and its antecedent is the same in both (6a) and (6b).

Reinhart and Reuland (1993)

Reinhart and Reuland (1993)(henceforth R&R) take sentences like (5) as indicative that an anaphor-centered binding theory should be replaced by a predicate-centered binding theory. That is, while the standard binding conditions are conditions on when anaphora is possible, R&R hold that binding facts are facts about reflexive predicates and not about anaphora at all. They propose the following binding conditions (to replace the standard binding conditions of Chomsky 1981/1986b):[3]

(7) Condition A: A reflexive-marked predicate is reflexive
Condition B: A reflexive predicate is reflexive- marked

These conditions depend on the following definitions:

(8) a. A predicate is reflexive iff it has two coreferential
arguments
b. A predicate is reflexive-marked iff
i) it is lexically reflexive; or
ii) one of its arguments is a SELF anaphor

A SELF anaphor is a morphologically complex anaphor and contrasts with SE anaphors which are morphologically simplex.

The basic intuition behind these conditions is that there is a one-to-one mapping between a particular semantic property (reflexive) and a particular formal property (reflexive-marked), where there are two ways to realize the formal property. A predicate can be reflexive-marked if it is lexically reflexive (i.e., if it has two elements on its theta-grid marked as coreferential) or if one of its arguments is a SELF anaphor.

Let us see how these conditions account for the facts in (2), repeated here:

(2) Dutch
a. *Max haat zich
Max hates self
'Max hates himself'

b. Max hoorde mij over zich praten
Max heard me about self talk
'Max heard me talk about him

c. Max haat zichzelf
Max hates selfself
'Max hates himself'

d. *Max hoorde mij over zichzelf praten
Max heard me about selfself talk
'Max heard me talk about him'

In (2a), we have a predicate with two coreferential arguments, i.e., a reflexive predicate. Condition B therefore applies. However, this predicate is not reflexive-marked. The verb hate is not lexically reflexive and the anaphor is not morphologically complex. Thus, Condition B is violated. In (2c), on the other hand, the predicate is reflexive but now the predicate is also reflexive-marked because the SELF anaphor zichzelf is one of its arguments. So, Condition B is met. Condition A is also met since the reflexive-marked predicate is reflexive. In (2b), we find no reflexive predicates and no reflexive-marked predicates and so both conditions are satisfied trivially by not applying. In (2d), the anaphor zichzelf reflexive-marks the lower predicate. This predicate is not reflexive and so Condition A is violated. Turning now to the examples in (5), here we find reflexive predicates which do not appear to be reflexive-marked (since they do not have a SELF anaphor as arguments). However, these predicates are lexically reflexive (thus reflexive-marked) and so both conditions are satisfied. The predicates are both reflexive and reflexive-marked.

Problems with R&R

While this binding theory works fairly well for these data, there are three problems. First, there are many languages in which a morphologically simplex anaphor can be locally bound (as noted for German by R&R). This problem is not too serious, however, since it may be possible to manipulate the definitions of SELF anaphor and SE anaphor so that all of the languages with simplex anaphors which behave like SELF anaphors can be accounted for. A second problem lies in sentences like (4). For R&R's theory, since there are no reflexive predicates here, the binding theory should not apply. Thus, R&R are forced to resort to an additional stipulation on properties of chains in order to rule out (4c). While this condition does capture the desired result, it fails to note the complementarity between anaphors and pronouns which is the hallmark generalization of the binding theory. It is quite surprising that the complementarity between anaphors and pronouns that follows from R&R's binding theory is maintained even in contexts in which their version of the binding theory does not apply. Finally, and most seriously, R&R's theory predicts that reflexive-marked predicates are semantically uniform since the two types of reflexive-marking are collapsed in the binding conditions. However, predicates which are reflexive-marked by SELF anaphors have different semantic properties than predicates which are lexically reflexive-marked. That is, the one-to-one mapping between the semantic property of being reflexive and the formal property of being reflexive-marked breaks down when we examine the semantic properties of these two types of predicate more closely. It is from the perspective of this third problem that we will be able to see how the theory of reflexive predicates is independent of the theory of anaphora.

On the disunity of reflexive-marked predicates

R&R's binding conditions predict that lexically reflexive-marked predicates are semantically equivalent to syntactically reflexive-marked predicates. Since the binding conditions apply at LF and only make reference to reflexive-marking and not to the manner of reflexive-marking, it follows that these predicates are indistinguishable at LF. This prediction is not borne out, however. Consider first the Madame Tussaud context first discussed in Jackendoff (1992). Here, a famous person goes into the wax museum and looks into the mirror. If the speaker describing this situation says (9a), then we get the interpretation that the person saw her own reflection in the mirror. But, if the speaker says (9b), we get either the interpretation that the person saw her own reflection or the interpretation that the person saw a reflection of her statue (Reuland 1995):

(9) a. ze zag zich in een griezelige hoek staan
she saw self in a creepy corner stand
'She saw herself (=reflection, *statue) in a creepy corner stand'

b. ze zag zichzelf in een griezelige hoek staan
she saw selfself in a creepy corner stand
'She saw herself (=reflection or statue) in a creepy corner stand'

The fact that there is an additional reading available in the sentence in which the SELF anaphor occurs is not predicted by R&R's theory.

This set of facts is found also in Kannada. When the verb is lexically reflexive-marked (via the affixal reflexive), the statue interpretation is blocked. When the morphologically complex anaphor is present, however, this interpretation is available.

(10) a. Hari tann-annu nod-i-koND-a
Hari self-ACC see-PP-REFL.PST-3SM
'Hari saw himself (=Hari, *statue)'

b. Hari tann-annu-tanne nod-id-a
Hari self-ACC-self see-PST-3SM
'Hari saw himself (=Hari or statue)'

A second place where R&R's theory makes the wrong prediction is in comparative deletion constructions. Here again, all reflexive-marked predicates are predicted to behave alike but they do not. The lexically reflexive-marked predicate allows only a sloppy interpretation while the syntactically reflexive-marked predicate allows either a strict or sloppy reading (Sells, Zaenen and Zec 1987):[4]

(11) a. zij verdedigde zich beter dan Peter
she defended self better than Peter
'She defended herself better than Peter defended himself'
'*She defended herself better than Peter defended her'

b. zij verdedigde zichzelf beter dan Peter
she defended selfself better than Peter
'She defended herself better than Peter defended himself'
'She defended herself better than Peter defended her'

Given that R&R's theory falsely predicts the semantic equivalence of the two types of reflexive-marked sentence, we need to find a way to capture their generalization that the coargument restriction on anaphors like zich is due to properties of the predicate without making the same false prediction. In order to do this, we must first characterize the difference between the interpretations allowed in the syntactically reflexive-marked sentences and those of the lexically reflexive-marked sentences.

Near Reflexives

What the Madame Tussaud examples illustrate is that SELF anaphors allow an interpretation in which the anaphor is referentially dependent on its antecedent but not necessarily identical with it. That is, the anaphor can pick out an entity from the domain of discourse which is related to and similar to its antecedent. We will call such anaphors "Near-Reflexives." The difference between Near-Reflexive predicates (i.e., predicates which have a Near-Reflexive as an argument) and semantically reflexive predicates is given in (12):

(12) a. lx [P(x,x)] (semantic reflexive)
b. lx [P(x, f(x))] (Near Reflexive)

(12a) indicates that the two arguments of the predicate P are identical. In (12b), the second argument is a function of the first argument. These representations are formally distinct, although in most cases they are extensionally equivalent. That is, the Near-Reflexive function does not prohibit the antecedent and the anaphor from being the same entity in the world, although in the semantic representation they are distinguished. On the other hand, in a semantically reflexive sentence, the anaphor and its antecedent must be completely identical, both in the world and in the semantic representation.

Since lexically reflexive predicates do not allow a non-identity interpretation, it follows that they are semantically reflexive, i.e., that they have the representation (12a). Predicates which are reflexive-marked via a morphologically complex anaphor do permit the non-identity interpretation, and so implicate the semantic representation (12b).

Given that the coargument restriction on anaphors like zich and tannu is obviated in the presence of lexical reflexivity and that lexical reflexivity is semantic reflexivity (i.e., not Near-Reflexivity), I have proposed the following principle (Lidz 1996):

(13) Condition R

lx(P(x,x))	iff	(1=2)
semantic structure		thematic grid

The left side of this formula depicts the semantic representation of reflexivity. The right side depicts the representation of the theta-grid of a lexically reflexive predicate. The Condition states that if a predicate is semantically reflexive, then it must be lexically reflexive. Similarly, if a predicate is lexically reflexive, then it must be semantically reflexive.

Let us now see how Condition R gives us the desired results. Consider the following:

(14) a. *Max haat zich
Max hates self
'Max hates himself'

b. ze zag zich in een griezelige hoek staan
she saw self in a creepy corner stand
'She saw herself (=reflection, *statue) in a creepy corner stand.'