Pica/Snyder:1 ON THE SYNTAX AND SEMANTICS OF LOCAL ANAPHORS IN FRENCH AND ENGLISH PIERRE PICA WILLIAM SNYDER UQAM & C.N.R.S MIT INTRODUCTION* 1. Local anaphors in French and in English have a variety of lexical and morhpological properties that are unexpected on conventional approaches to Binding Theory. We propose an account in which these characteristics follow from the fact that reflexives are essentially pronominals. On our approach, Principle B is the core of Binding Theory, and the surprising properties of local anaphors derive from the need for local anaphors to escape Principle B. We argue that all local anaphors are bimorphemic (sometimes in contrast to surface appearances), and that this structure is related to a semantics of ‘partition’ by which local anaphors escape Principle B. In support of these views we present a variety of facts that follow in a natural way from our syntactic and semantic proposals, but which to our knowledge have not previously been explained within the framework of generative grammar. Our theory is shown to have a modular character, comprising distinct contributions of Principle B, the morphological structure of local anaphors, and the broader architecture of human cognition. * Portions of this research have been presented in talks at CUNY Graduate Center, McGill, and the Universities of Stuttgart and Geneva. Special thanks to A. Belletti, H.-N. Castañeda, N. Chomsky, A.-M. Di Sciullo, R. Fiengo, J. Greenberg, J. Gruber, A. Hestvik, R. Kayne, H. Koopman, H. Haider, I. Heim, H. Lasnik, L. Rizzi, J. Rooryck, K. Safir, S. Vikner, and the audience at LSRL. All errors remain our responsibility. This work was partly supported by the Social Sciences and Humanities Research Council of Canada Grant n° 411-92-0012 (P. Pica). W. Snyder was supported by an NSF Fellowship in Linguistics, an NSF Research and Training Grant for Linguistics & Cognitive Science, and the McDonnell-Pew Center for Cognitive Neuroscience at MIT. Pica/Snyder:2 ON THE LACK OF SAUSSUREAN ARBITRARINESS IN LOCAL ANAPHORS 2. Standard Binding Theory assumes a lexical distinction between pronominals and anaphors. On this traditional view, pronominals are subject to a disjointness constraint (Principle B), while anaphors are exempt from Principle B but subject to their own locality requirement (Principle A). Such an approach crucially assumes a lexically encoded categorial distinction between pronominals and anaphors. It is remarkable, however, that if languages provide elements marked as anaphors in the lexicon, and if these elements are exempt from Principle B, that anaphors should be subject to the variety of severe restrictions observed in natural languages. For example, there are few, if any, languages with an anaphor completely unrelated, in its phonology and morphology, to either the pronominal paradigm or the agreement paradigm. This pattern is a source of puzzlement on standard approaches, because the inventory of anaphors in the lexicon should be subject to Saussurean arbitrariness. Restricting ourselves to strictly local anaphors, the lack of arbitrariness is illustrated by French, where the following elements all involve, in one way or another, a pronominal or agreement element: (a) the so-called complex anaphor, lui-même; (b) the so-called pronominal clitics such as nous and vous; and (c) the reflexive clitic se. The complex anaphor is formed from the pronominal lui. The ‘pronominal’ reflexive clitics are identical in form to the corresponding pronominals. The reflexive clitic se is also related to the pronominal and agreement systems, as suggested, for example, by its etymological relation to the so-called impersonal si of Italian, which is often taken as involving a pronominal use of si akin to English one. A more direct relation of se to the pronominal paradigm is the phonological (and etymological) relationship of se to the third person singular possessive pronouns son, sa, and ses (see Benveniste (1969)). Furthermore, across languages, the antecedent of an anaphor is often subject to constraints that have been stated in terms of animacy, agenthood, awareness, or point of view. (On the role of awareness in long distance anaphors, see among others Kuno (1987); on point of view, see Cantrall (1974).) These constraints, which have generally been formulated in pragmatic terms, are likewise a source of puzzlement on the standard view, because they do not derive in an obvious way from known principles of the grammar. For example, 1 illustrates the well-known animacy constraint on the French clitic se in its reflexive use: Pica/Snyder:3 (1) a. Jean se lave. ‘John washes himself.’ b. * La table se lave. ‘the table washes itself.’ The examples in 2 illustrate what we take to be a less readily observed property, namely that the reflexive clitic must have an agentive antecedent1: (2) a. Jean se frappe. (agentive) ‘John hits himself.’ b. ??Jean s’apprécie. (non-agentive) ‘John likes himself.’ c. Jean se détache de la chaise. (agentive) ‘John is untying himself from the chair.’ d. ?? Jean se connaît. (non-agentive) ‘John knows himself.’ The agentiveness constraint observed in 2b,d is reminiscent of the awareness constraint discussed by Kuno and others for long-distance anaphora (and local anaphora involving anaphors in oblique positions), but, to our knowledge, has not previously been noted for local anaphoric relations (between a subject and a direct object). 1 This constraint can be superseded, however, under certain conditions, notably when focus is applied to either the surface subject (Jean) or the predicate (s’apprécie) in 2b. (Addition of the modifier beaucoup also renders 2b acceptable, as will be discussed.) One explanation for the effect of focus would be that focus introduces a partition of the same type discussed in Section 3. This would account for the effect of focus in (i), where underlining indicates “coreference” and capitalization indicates a pitch accent. Yet, focus has more dramatic effects on possible binding relations than would be expected under a straightforward partition account, as illustrated by the example in (ii). (i) John thinks M ary likes H E R /* h er . (ii) J ohn thinks Mary likes H IMSEL F /* h imself . Pica/Snyder:4 THE ROLE OF PARTITION IN LOCAL ANAPHORA 3. All of the above observations run counter to the predictions of the standard binding theory, and cast doubt on its accuracy. As an alternative, we propose, first, that there is no Principle A as such, and that Principle B, which is the core of binding theory, applies to both pronominals and anaphors. On this view, an anaphor can be understood as a type of pronominal whose internal structure allows it both to approximate coreference with a local antecedent, and at the same time to escape Principle B. We will argue that local anaphors satisfy these conflicting requirements through a semantics of ‘partition,’ in which the anaphor is interpreted as related to, but distinct from, its antecedent. For example, in the case of the English anaphor himself, the morpheme him is interpreted as the same individual as the antecedent. Yet, the semantics of the morpheme -self causes the anaphor as a whole to be interpreted as an individual related to, but distinct from, the individual denoted by him. On our view, the necessity of creating a partition accounts for the lack of Saussurean arbitrariness in the form of local anaphors, including both the tendency for local anaphors to be morphologically complex, and the tendency for anaphors to bear a visible similarity to elements from the pronominal and agreement paradigms. We have identified two principal types of partition that are compatible with local anaphora. One type of partition appears to divide an individual into an agentive aspect (the individual understood as the psychological agent of a reflexive action) and a physical aspect (the individual understood as the physical patient of the reflexive action). A second type of partition appears to distinguish the individual, understood as a conscious observer or possessor of knowledge, from the same individual understood as the object of knowledge. (We will present evidence that in the second type of partition, the object of knowledge is generally understood psychologically rather than physically.) These two types of partition are summarized in 3.2 2 We do not, however, intend 3 as an exhaustive list. The task of characterizing the class of semantic partitions compatible with local anaphora is an on-going research project. A possible third type, in addition to 3a,b, is the partition between an individual understood as a source and that individual understood as a goal. Such a partition may be responsible for the acceptability of examples such as (ia,b). Pica/Snyder:5 (3) a. Agent (Psychological) vs. Patient (Physical) b. Knower (Psychological) vs. Object of Knowledge (Psychological) Thus in 4, the NP John is understood psychologically, as the agent of hitting, while the NP himself is understood physically, as the patient. (4) John hit himself. In 5, the predicate likes is non-agentive, and so the anaphor himself 5b cannot be dependent on the type of partition in 3a. Instead, we argue, Principle B is satisfied in 5b by the partition of John into ‘knower’ and ‘object of knowledge’ 3b. We take the English verb likes to be ambiguous between a reading in which the subject’s positive attitude towards the object is based on knowledge about the object, and a reading in which the reason for the subject’s positive attitude towards the object is left vague. When the object is non-reflexive, both readings are possible, as in 5a. (5) a. John likes Mary, but he doesn’t realize it. b. John likes himself (?* but he doesn’t realize it). When the object is a reflexive, however, the conscious-knowledge reading is obligatory, as illustrated by the anomalous nature of the parenthetical clause in 5b. This follows because the reflexive, with a non-agentive antecedent, depends on a partition of the type in 3b. In French, the conscious-knowledge reading of English like can be unambiguously expressed by the predicate apprécier beaucoup, as illustrated in 6a. (6) a. ?* Jean apprécie beaucoup Marie, mais il ne le sait pas. ‘John likes Mary, but he doesn’t know it.’ (i) a. John talks to himself. b. John sent a letter to himself. An interesting possibility is that the relevant types of partition can be expressed entirely in a more restricted vocabulary of “thematic roles,” such as those assumed in Hale & Keyser (1993) or Gruber (forthcoming), but this remains a direction for future research. Pica/Snyder:6 b. Jean s’apprécie beaucoup. ‘John likes (what he knows about) himself.’ (Conscious knowledge implied) As expected, the reflexive (6b) is fully grammatical with this predicate. French provides several other predicates corresponding roughly to English like. The verb aimer bien is ambiguous between the conscious-knowledge reading of like and the reading in which conscious knowledge is not implied, while the verb aimer alone preferentially corresponds to the latter reading only. As expected, 7a is fully acceptable, while 7b is anomalous. (7) a. Jean s’aime bien. ‘John likes (what he knows about) himself.’ (Conscious knowledge implied on the relevant reading) b. ?? Jean s’aime. ‘John likes himself.’ (Conscious knowledge not implied) It should be noted that the judgements in 6 and 7 are somewhat delicate, and are sensitive both to individual variation in lexical semantics and to possible effects of focus. Nonetheless, the same patterns can be observed with a variety of non-agentive verbs, and are completely unexpected on the standard binding theory, where similar effects have generally been taken as falling outside of the grammar. For example, the sentences in 8, involving two senses of the verb know, are exactly parallel to the observations in 6 and 7. (8) a. John knows Mary, but he’s momentarily forgotten. b. John knows himself (?* but he’s momentarily forgotten). c . Jean se connaît ?? (bien). ‘John knows himself.’ In 8c, connaître bien is again ambiguous as to whether conscious knowledge is implied, but connaître alone does not clearly imply conscious knowledge on the part of the subject. Pica/Snyder:7 Interestingly, the precise set of semantic partitions with which local anaphora are compatible is considerably more restricted than the full set of partitions that one could, in principle, impose on an individual. An example is given in 9. (9) a. John hit his body. b. John hit himself. (physical) c. John likes his body. d. John likes himself. (psychological/??physical) Examples 9a and 9b are at least approximately equivalent, as we expect if the partition in 9b=4 is between John understood as a psychological agent, and John understood as a physical patient. We might reasonably expect a partition into psychological and physical aspects to enable 9d to have approximately the same interpretation as 9c, but this expectation is at most partially borne out. The English- speaking informants whom we have consulted find 9d relatively odd on an interpretation in which himself is understood physically. Thus, we specify in 3b that the relevant partition is between an individual as ‘knower’ and an individual as ‘object of knowledge,’ in both cases understood psychologically. We propose that the restricted set of partitions found with local anaphora may reflect significant properties of human cognitive architecture, potentially of relevance to much more than binding theory, and for the present we content ourselves with a description rather than an explanation. For present purposes, we simply take the anomalous character of the ‘physical’ reading in 9d as another indication that local anaphora are restricted in ways that are unexpected on the standard binding theory, but that are compatible with a partition analysis. Further support for the role of partition in local anaphora comes from the incompatibility of certain experiencer-object verbs with the reflexive. For example, the contrast in acceptability of 10a,b versus 11a,b corresponds to the degree to which the surface subject of these predicates is understood as an object of (the experiencer’s) knowledge. This interpretation of the evidence is supported by the contrast in acceptability of 10c versus 11c, on the reading in which it is the information that Mary knows about John which bothers, annoys, frightens, or worries her. Pica/Snyder:8 (10) a. ?* John bothers himself. b. ?? John annoys himself. c. The more Mary learns about John, the more he ?*bothers/??annoys her. (11) a. ? John frightens himself. b. ? John worries himself. c. ? The more Mary learns about John, the more he frightens/worries her. While the precise degree of ungrammaticality of the examples in 10 and 11 is difficult to assess, and undoubtedly varies across speakers, we expect that the relative grammaticality of examples such as 10a,b and 11a,b will be correlated with the grammaticality of corresponding examples of the form in 10c and 11c. Again, the facts in English are highly similar to those of French. In 12a,b we find a contrast between the approximate equivalents of annoy and worry, in which the former more strongly resists the reflexive. (Here, we are specifically evaluating 12a,b on the readings in which they are not inherent reflexives; we are thus excluding the alternate reading, ‘John is becoming annoyed/worried.’) This contrast is paralleled by the contrast in acceptability of the two predicates in 12c. (12) a. ?* Jean s’ennuie. ‘John annoys himself.’ b. ?? Jean s’inquiète. ‘John worries himself.’ c. Le plus de choses Marie apprend sur Jean, le plus cela ?? l’inquiète / ?* l’ennuie. ‘The more Marie learns about Jean, the more this worries/annoy her.’ The evidence from experiencer-object predicates as in 10, 11, and 12 is especially interesting because in 11a,b the anaphor himself expresses the ‘knower’ rather than the object of knowledge. This suggests that the semantics of -self is relatively general, making reference to a contextually available partition, but allowing the interpretive component to fill in the details. Below we will propose a modular view of local anaphora, comprising independent roles for a disjointness condition (Principle B), the lexical semantics of -self (or its approximate equivalents in other languages), and the varieties of partition allowed by the cognitive architecture. Pica/Snyder:9 THE FORMAL SEMANTICS OF PARTITION 4. We find it interesting to note that in the preceding discussion, philosophical notions such as the de se /de re distinction have played a tacit role (cf. Lewis (1970), Castañeda (1967), (1975), and Quine (1955) among others). We suggest not only that such philosophical notions may be relevant to the proper understanding of linguistic phenomena, but also that many of the philosophical questions, including questions about the concepts of mind and body, about the concepts of person and identity (cf. Quine (1955), Lewis (1970), and Castañeda (1967), (1975) among others) may have an unrecognized source in the phenomena of natural language (the semantics and syntax of partition in particular). In particular, there exists an interesting and extensive philosophical literature on the concept of ‘self,’ which is suggestive in light of our approach to local anaphora. A widespread observation in this literature is that the term ‘self,’ which originates as a nominalization of the -self from local anaphors, is invoked precisely when there is a real or apparent contradiction between two roles that are performed by a single individual. This is surprising on a view in which the concept of ‘self’ expresses something more like integral/defining/persistent properties of an individual, for example. Yet, the observation is precisely what we should expect if the noun self derives its meaning from the semantics of the reflexive morpheme -self, and if -self has the semantic function of specifying some aspect of an individual that is related to, but distinct from, a second aspect of the same individual. In this connection it is also interesting to note that a remarkable number of languages use, in the formation of local anaphors, a morpheme that ordinarily denotes a body-part: Hebrew ‘bone’, Hausa ‘head’, Basque ‘head’, and so forth. This strikes us as natural on an account in which many uses of local anaphora depend on a partition between psychological and physical aspects of an individual, because a body-part relates both to the physical body and, as an inalienable possession, to a psychological possessor. We can represent the semantic contribution of the English morpheme -self as in 13, where we assume an extensional semantics for ease of exposition. (13) || -self || = f: De -> De For any x an element of the domain of individuals, where x denotes a contextually salient ‘aspect’ of a person or object, f(x) denotes a salient, distinct aspect of the same person or object. Pica/Snyder:10 The important points about the characterization in 13 are first that -self requires an individual-type argument corresponding to the morpheme him in himself, and second that the result of combining -self with its argument is an element, drawn from the domain of individuals, which denotes an ‘aspect’ of a person or object.3 A further point is that the denotation of -self does not directly determine the type of partition that allows the anaphor to satisfy Principle B. Instead, the denotation in 13 makes reference to the partition, if any, that is made contextually salient by the semantics of the other lexical items in the sentence. We propose a slightly different analysis of the French local anaphor se. In the case of se, we are again led to a bipartite structure analogous to that of English himself. We take the surface form se to correspond to an underlying structure of the form [se ø], where se corresponds to him and ø corresponds to -self. In contrast to English, however, the null counterpart of -self, on our account, has a semantics closer to that given by Heim, Lasnik, and May (1991) for the English morpheme other. 4 We represent the proposed meaning as in 14. (14) || ø || = f: De -> D<et,t> For any x an element of the domain of individuals, and any predicate g mapping individuals to truth values, f(x)(g) is true if and only if, for every y in the domain of individuals, distinct from x, and in a discourse-supplied context set containing x, g(y) is True. For present purposes, an important characteristic of 14 is that the semantics of the element ø does not make reference to ‘aspects’ of a single individual, but is instead stated in the more general terms of a partition on a set. In the case 3 Here we allow the antecedent of a local anaphor to be an inanimate object, although in many cases this will be ruled out by the nature of the available types of partition. For example, both of the partitions in 3 would seem to be incompatible with an inanimate object. Yet, there may exist other available types of partition that would not impose this constraint. Alternatively, it may turn out that apparent anaphors such as English itself have a status distinct from “true” reflexives, as might be inferred from the fact that itself was first attested in English more recently than the other reflexive forms, and also by the animacy constraint on reflexives in French. 4 The interpretation we propose is also reminiscent of the French word partie, which in its various uses can denote a person, a (sub)set of individuals, or a body-part, as suggested by the following expressions: la partie adverse ‘the opponent’, les parties en présence ‘the parties present’, d’autre part ‘on the other hand’, à part (exclusion) ‘separately’.