Binding Theory Semantic Interpretation and Context

CatWPL 511, 1996 81-112 Binding Theory, Semantic Interpretation and Context* Josep Macii Department of Linguistics and Philosophy. MIT, 20D-219. 77 Massachusetts Ave. Cambridge, MA 02139, USA e-mail: [email protected] Received: January 1st 1996 Accepted: July 6th 1996 Abstract There are several well-known counter-examples to the Binding Theory of Chomsky (1981, 1993) which have to do with Principles (B) and (C). These counter-examples have different degrees of acceptability. We can account for these counter-examples and explain their different degrees of acceptability by (i) Distinguishing different types of sentences among these examples; different kinds of sentences exemplify different phenomena and require different explanations. (ii) Viewing Binding Theory as a semantic theory about how to semantically interpret certain structural relationships between NPs. (iii) Modifying principies (B) and (C) so that they are not about the relation of having the same value, but rather about the relation of it being presupposed to have the same value in a given context. (iv) Identifying how the Binding Principles interact with other sources of semantic information in order to yield the interpretation for a sentence. These other semantic factors rnight provide information that conf¡icts with (B) or (C). The data show that when these other semantic factors determine some interpretation for a sentence in a clear enough way the speakers judge such an interpretation as at least partially acceptable. Key words: formal semantics, binding theory, semantic interpretation. Resum. Teoria del Lligam, interpretació semdntica i context Hi ha tota una sbrie de coneguts contraexemples a la Teoria del Lligam de Chomsky (1981,1993) que tenen a veure amb els principis (B) i (C). Aquests contraexemples posseeixen diferents graus d'acceptabilitat. Podem explicar aquests exemples i explicar a més per qui: unes frases són jutjades més acceptables gramaticalment que altres si fem * I would like to thank Manuel Garcia-Carpintero, Howard Lasnik, Joe Lau, Alec Marantz, Susanna Pi, Gemma Rigau, Joana Rosselló Asensi, Robert Stalnaker and an anonymous CWPL reviewer for comments, discussions and other assistance. Special thanks are due to James Higginbotham for encouraging me to work on this topic, and to Irene Heim both for discussions and for the influence that her ideas on this topic have had on my own. I also benefited from the comments and reactions of the audiences at the I Incontro Italo-Spagnolo di Filosofia del Linguaggio e Filosofia della Mente (Universith di Venezia, June 3, 1994), and the Fourth Annual Harvard-MIT Graduate Philosophy Conference (Harvard University, March 10, 1996) where I presented most of the ideas in this paper. 82 CatWPL 511, 1996 Josep Mac2 el següent: (i) Distingir diferents tipus de frases dintre d'aquests exemples; diferents frases exemplifiquen diferents fenbmens i requereixen diferents explicacions. (ii) Considerar la Teoria del Lligam com una teoria semintica sobre com interpretar seminticament determinades relacions estructurals entre SNs. (iii) Modificar els principis (B) i (C) de forma que no tractin sobre la relació x té el mateix valor que y, sinó sobre la relació es pressuposa (respecte a cert context) que x i y tenen el mateix valor. (iv) Identificar com els principis de la Teoria del Lligam interaccionen amb altres fonts d'informació semantica a fi de produir la interpretació d'una oració. Aquests altres factors semintics poden proporcionar informació que estigui en conflicte amb (B) o (C). Les dades mostren que quan aquests altres factors semintics determinen una interpretació per a una oració de forma suficientment clara els parlants consideren aquesta interpretació com a, si més no, parcialment acceptable. Paraules clau: semintica formal, teoria del lligam, interpretació semintica. Table of Contents 1. Standard Binding Theory 6. Some Comments 2. Counter-examples to Standard 7. Two-stages in the Interpretation of an Binding Theory Principles (B) and (C) Utterance 3. An Altemative Proposa1 8. Conclusion 4. The Disjointness Conditions and References Other Sources of Semantic Information 5. Identity and the Disjointness Conditions 1. Standard Binding Theory 1 .l. Principles (B) and (C) Consider the following examples, where two NPs in a sentence have the same index if and only if either they both refer to the sarne individual or one of them is a quantificational NP (like, e.g, Every boy, or Somebody) and the other is bound to it (in the same sense that a logical variable is bound to a quantifier): (1) a. Kathrini saw herj b. *Kathrini saw heri c. Shei saw heri father d. Danieli says that hei is tired e. Danieli says that hej is tired f. He, says that Jasonj is, tired g. *Hei says that Jasoni is tired h. His, father says that Jason, is tired i. "(His father)i says that Richardi is tired j. (His father)i says that hei is tired Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 83 k. Hei saw Robertj 1. *Hei saw Roberti m. His, mother saw Robert, n. *Everybodyi saw himi o. Everybodyi saw himj p. Every thief, fears that he, gets caught k. In Scarsdale, every single boyils mother sends the twerpi off to summer camp r. *In Scarsdale, every boy, thought the twerp, would hate summer camp We can explain all these data by appealing to conditions such as Chomsky's Principles (B) and (C) of the Binding heo or^' below (we will not go here, though, into the detail of discussing how each sentence above is accounted for by those principles): Principle (B): Pronominals must be A-free in their Governing Category. Principle (C): R-expressions must be A-free. A succinct explanation of the terminology involved in the formulation of Principles (B) and (C): The class of Pronominals includes the non-reflexive pronouns (both possessive and non-possessive); R-expressions include proper names, epithets, definite descriptions, quantificational NPs, and traces of whmovement and Quantifier Raising. Node X binds node Y if they are co-indexed and X c-commands Y. To be A-free is not to be bound by anything in an A-position (complement and subject positions are A-positions, adjoined positions and the specifier of CP are non A-positions). For the purposes of the present discussion we can set aside some difficulties arising when trying to specify the notion of Governing Category, and we can take the Governing Category (GC) of a node X to be the smallest maximal projection containing both a subject and a lexical category that governs X -or, more simply: the smallest NP or S that contains X. Different sets of conditions can be formulated that, even if they are not equivalent to (B) and (C), are similar to them and are intended to account for the same kind of phenomena. We will call Disjointness Conditions any such set of conditions. We will call Standard Binding Theory the theory we are presenting in this section and which includes Principles (B) and (C). 1.2. Some Comments Notice that we are not using indexes simply as part of some provisional notational convention to help us indicate what the reading of a sentence that we want to consider is. Rather, in the way we are using them, indexes are an essential part of the syntactic representation of a sentence (in the same way that each node in a tree is part of that representation). This is so because the Binding Principles are 1. In, for instance, Chomsky (1981:188). 84 CatWPL 511, 1996 Josep Macih formulated in terms of the relation of binding, which, in turn, appeals to the relation of snmeness of indexes. Also notice that in order to account for the data in (1) we need to appeal not only to the Binding Principles but also to a principle such as {{Referential NPs with the same index must refer to the same individual, referential NPs with different indexes must refer to different individuals; a NP is interpreted as bound by a quantificational NP if and only if they have the same index,,. We mentioned this principle above as in passing, but notice that it is crucial. Without it, Binding Theory would not explain, for instance, why (2a) is unacceptable if Begonya is coreferential with her: Even if the syntactic representation (2b)2 is ruled out by Principle (B), (2c) is not. If it were possible for two NPs with different indexes to have the same referent, then (2c) could give rise to an interpretation that (2a) does not have. (2) a. Begonya likes her b. Begonyai likes heri c. Begonya, likes herj Thls discussion shows also another point: a theory that tries to explain why (2a) cannot have a coreferential reading has to include, some way or other, a semantic component. The explanation can not be given in purely syntactic terms. It rnight be, as it is the case with Standard Binding Theory, that the bulk of the explanation is carried out by purely syntactic principles, and that the semantic component is straightforward. Still the semantic part, even if straightforward, is essential to explain what we want to explain. It is possible to have a theory where the bulk of the explanation is at the semantic level. We can, for instance, formulate disjointness conditions which are completely analogous to Standard Binding Theory principles (B) and (C) but that operate at the semantic level (that is, they are conditions that directly constrain which interpretations are possible for a given LF-syntactic structure). Chomsky (1993) offers one set of such a semantic kind of disjointness c~nditions.~ Principles (B) and (C) are formulated there, roughly, in the following way: (B): If a is a pronominal, interpret it as distinct in reference from every ccommanding phrase in its GC. (C): If a is an r-expression, interpret it as distinct in reference from every ccommanding phrase. (In order to account for cases like (1n)-(lr) where there is no co-reference involved, these principles should, of course, be modified -or at least the term ccdistinct reference,, should be re-defined in such a way that it also applies to a relation 2. (2b) should be regarded as a partia1 specification of a tree. We do not explicitly specify the whole tree for simplicity of exposition. Analogously for (2c). 3. There are several examples of earlier semantic formulations of disjointness conditions, which include the non-coreference rule of Reinhart (1983), and the Rule of Interpretation of Chomsky (1973). Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 85 such as the one between Everybody and hirn in (ln). With such a modification in place, this theory makes the same predictions as Standard Binding Theory with respect to what interpretations the different sentences can have). 2. Counter-examples to Standard Binding Theory Principles (B) and (C) It has been observed in the literature that there are severa1 different kinds of sentences that are counter-examples to the Standard Binding Theory Principles (B) and (C). In (3) we have some examples of sentences that pose a problem for principle (C):4 (3) a. Who is that woman over there? She is professor Rigau. b. Everyone has finally realized that Oscar is incompetent. Even he has finally realized that Oscar is incompetent. c. Mary, Betty and John have one thing in common: Mary admires John, Betty admires John, and he himselfalso admires John. d. What do you mean Oscar loves no one? He loves Oscar. e. The logic tutor while trying to explain the law of Universal Instantiation to a student tells him: Look fathead. If everyone loves Oscar, then certainly he himselfmust love Oscar. f. I think that woman talking on TV is Zelda. She says the sarne things that Zelda says in her book. Each of these sentences can be interpreted so that the two italicized NPs have the same referent. If they are interpreted as having the same referent, then the two NPs must have the same index in the syntactic structure of the sentence. In each of these sentences, though, one NP c-commands the other, and, furthermore, the c-commanded NP is an R-expression. The syntactic structure, so, is undesirably ruled out by Principle (C) of the Standard Binding Theory. Analogously, the sentences in (4) pose a problem for the Principle (B) of the Standard Binding Theory (4) a. Mary, Betty and John have one thing in common: Mary admires John, Betty admires him, and John himselfalso admires hirn. b. At a Halloween party someone says: I think that the man with the devil costume is Joe. It is suspicious that he knows him so well. c. What do you mean no one loves Oscar? Oscar loves hirn. d. What do you mean no one loves Oscar? He loves hirn. Some of these sentences (like (3a) or (4b)) are one hundred per cent acceptable, others are not perfect but are good enough as to pose a problem for a theory according to which they are simply ruled out. 4. Even if I do not specify the source of each of these sentences, most of them are adapted or directly taken from the references mentioned at the end; the same is true of the sentences in (4). 86 CatWPL 511. 1996 Josep Macii Heim (1993) offers some examples involving quantification which she claims are marginally acceptable and, to the extent that they are acceptable, also pose a problem for Principle (B), as for instance (5) a. The logic tutor while trying to explain the law of Universal Instantiation tells his student: Look, if everyone loves Oscar then it surely follows that Oscar himself must love Oscar. And, of course, this doesn't just hold for Oscar, but for any arbitrary man: if everyone loves a man, then that man himself loves him. b. Somebody said that what he had in common with his siblings was that his sister admired him, his brother admired him, and he himself also admired him. 3. An alternative proposa1 3.1. Other Proposals There are severa1 ways in which we could try to dea1 with these problematic examples: One possibility would be to claim that the disjointness conditions should not be constraints on the relation of 'co-indexation' or of 'co-reference' but rather they should constrain some other relation, and that when this is done the disjointness conditions will allow just those interpretations that the sentences can actually have. This is the approach taken in Higginbotham (1992). Of course the difficult task when opting for this view is to specify a suitable relation in terms of which we can formulate the disjointness conditions. Higginbotham introduces the relation of common reference, and formulates the disjointness conditions in terms of it. This relation is defined in terms of the intentions of the speaker. Another possible approach is to accept principles (B) and (C) as they are in Standard Binding Theory, but to modify the semantic part of the theory. Recall that the principle on the interpretation of indexes requires that two NPs get assigned the same individual if and only if they have the same index (and it also constrains how to interpret those NPs that will have an interpretation analogous to bound variables). One possible modification to the principle is to require only that if two NPs have the same index then they get assigned the same individual, but allowing that NPs with different indexes might also get assigned the same individual. Fiengo & May (1994) takes this kind of approach. The difficult part for this approach is not to explain, for instance, why (3b) can have a co-referential interpretation, but rather why (2a) can not usually have it. Fiengo & May (1994) tries to solve this by appealing to pragmatic principles. Another possibility is to think that in addition to disjointness conditions, we should also have some other principle which will allow us to make the desired distinction between, on the one hand, sentences like the ones in (3)-(4) and, on the other hand, sentences which are bad and should be ruled out. This is, very roughly, the line followed both by Tanya Reinhart (in, for instance, Reinhart (1983a,b) and Grodzinsky & Reinhart (1993)) and by Irene Heim (in Heim (1993)). The main difference between the two is this: Reinhart distinguishes, among the cases to which Standard Binding Theory applies, between the cases where there is binding Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 87 and cases where there is merely co-reference, and proposes that the disjointness conditions should only constrain cases involving binding, whereas cases involving co-reference should be accounted for by another sort of principle (her Coreference Rule), which is les~ restrictive than principles (B) and (C) of the Standard Bindng Theory. Heim, on the other hand, proposes that the disjointness conditions apply to both the cases involving binding and the cases involving co-reference; the cases involving co-reference, though, will be in addition subject to another principle (her Co-determination Kule), which can redeem some sentences that the binding principles would rule out. I think that none of these proposals is entirely satisfactory, though each one provides valuable insight into the phenomena. I will not comment on these or other theories here, though. I would rather like to propose here another way of looking at the phenomena we have been discussing. Of course, a complete defense of the proposa1 I will put forward would require a careful discussion of the alternatives. I intent to present such a discussion elsewhere. 3.2. Overview An adequate theory should not only account for the contrast between the sentences which are not acceptable and those which are at least partially acceptable but it should also give an account of why the sentences in (3)-(5) have different degrees of acceptability, and specially, it should account for the difference between fully grammatical sentences like (3a), (3f) or (4b) and sentences which are regarded as grammatically awkward and only partially acceptable -like, for instance, (3c) or (4d). The proposa1 defended here does not consist in trying to find a set of disjointness conditions with which none of the sentences in (3)-(5) is in conflict, but rather the disjointness conditions will be part of a more general account of what makes us regard some sentence (with some specific interpretation) as acceptable. This general account will explain why these sentences are acceptable even if some of them will be in conflict with the disjointness conditions we will provide. In the present proposa1 I attempt to explore the idea in Chomsky (1993) of regarding the disjointness conditions as semantic principles about how to interpret the NPs in a ~entence.~ In surnmary, the present proposa1 is the following: The sentences that are fully acceptable can be divided in two groups: identity sentences, and non-identity sentences. I formulate in section 3.3. the disjointness conditions (in semantic terms) so that they do not constrain the relation of co-reference but rather the relation of presupposed co-reference6 -in this I follow Postal (1970) and Heim (1993). These disjointness conditions are then no longer in conflict with fully acceptable non-identity sentences like (3f) or (4d) (section 5.1.). On the other hand, I argue in section 5.2 that identity sentences like (the second clause of) (3a) 5. All the main ideas put forward in the following pages, though, could be incorporated, with some adjustments, into an account using a 'syntactic version' of the disjointness conditions. 6. Actually, once we take also into consideration quantificational NPs, the relation being constrained is that of presupposed co-valueness. These notions are explained in the next section (section 3.3). 88 CatWPL 511, 1996 Josep Macia are not constrained by the disjointness conditions, since one of the NPs is not an argument but a predicate. One of the main contentions of this article is about what explains the (partial) acceptability of sentences like (3c): when interpreting an utterance of a sentence like (3c) the speakers use severa1 sources of information, the disjointness conditions being just one of them. If the other sources determine an interpretation for the NPs in a clear enough way, the sentence will be able to be understood in accordance with that interpretation even if it conflicts with the information provided by the disjointness conditions. The existence of this conflict will explain that the sentences are not regarded as completely good (sections 4 and 6). Finally, in section 7 I argue that we must distinguish two different levels in the interpretation of a sentence: one where only the lexical items and the grammatical structure of the sentence is taken into account, and a second level where contextual information is introduced so as to produce the full interpretation of the sentence. 3.3. Disjointness Conditions In this section I propose a different formulation of the disjointness conditions. First I state the principles, then I explain the notions involved in their formulation. I think that we should understand the disjointness conditions in the following way (I will refer to (B)' and (C)' below as the Disjointness Conditions): (B : If a sentence whose LF representation is . . . a.. .D.. . (where P is a pronominal, and a is an NP that c-cornmands /3 in its GC) is used in a context C, it is not presupposed in C that a and P are co-valued. (C): If a sentence whose LF representation is . . .a.. ./3... (where P is an Rexpression, and a is an NP that c-commands P) is used in a context C, it is not presupposed in C that a and P are co-valued. Let's explain what we mean by the notions of 'being presupposed', 'context' and 'being covalued' that we use in stating the Disjointness Conditions. 'being covalued with': NPs can have different sorts of semantic value. Corresponding to these different sorts of semantic value, there are different possible relations among NPs in a sentence. If two NPs refer to the same individual, we would say that they are on the relation of common reference. If two NPs are in the same relation that his and Every boy are in (6) when we understand the sentence so that every boy loves his own mother -where one NP acts like a variable bound by another NP-, we would say that the latter links7 the former. (6) Every boy loves his mother. 7. I borrow the tems alinking, and ucommon reference, from Higginbotham (1983,1992). My use of them is not exactly his. Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 89 Our characterization of the distinction between the two sorts of relation is quite uncommitted, but it suffices for the present purposes of defining the relation of 'being covalued', since, as we will presently see, the two former relations get combined in the definition of the latter.* We define 'to be covalued with' as the transitive closure of the relation 'to link, to be linked to, or to be in common reference ~ith'.~ That is, NP a is co-valued with NP ,6 iff a belongs to any set s which is such that: (I) ,8 belongs to s and (11) if x belongs to s then any y such that x links y or y links x or x is in common reference with y also belongs to s. So, for instance, if in (7) we were to interpret both he and him as linked to Every boy, then he and him would be co-valued (even though the two pronouns are not themselves in a linking relation or in a common reference relation), and so such an interpretation would violate the Disjointness Conditions: (7) Every boy thinks that he loves him. 'Presupposition' and 'context': The notions of 'presupposition' and 'context' that we use in stating our disjointness conditions are the same as in Stalnaker (1973,1974): linguistic communication always takes place on the basis of a background of common beliefs and assumptions, or context. Using the possible worlds framework we can identify a context in which some instance of linguistic communication takes place with a set of possible worlds: those worlds that as far as the participants in the conversation can tell could be the actual world. The context set consists of those worlds that could be the actual world according to what the participants in the conversation believe, and believe that the others believe, and believe that the others believe that they believe and so on. (If we wanted to make this characterization more precise we should take into consideration that what is relevant is not only what the participants believe but also what they pretend to believe). A proposition p is presupposed in a context C if p is true in each world in C. A sentence s carries the presupposition that p if it would be infelicitous to assert s in any context where it is not presupposed that p. The individual that a referential NP picks up at a world in the context w, is the individual that the NP would refer to if w were the actual world. A sentence is true in a world in the context w, if the sentence would be true if w were the actual world. Let's consider one example. Suppose I utter the sentence He is un artist while pointing to a man who is in front of us. Since all the participants in the conversa8. For our present purposes, for instance, we do not need to characterize the two relations in a way which is precise enough as to decide whether in (i) (when it is interpreted as meaning that Albert loves his own brother) Albert and his are in common reference, one links the other, or the sentence is ambiguous with respect to which of the two sort of relations holds (i) Albert loves his brother 9. This notion of 'co-valueness' is a very close relative of the notion of xcodeterminationx in Heim (1993). 96 CatWPL 511, 1996 Josep Macih We cannot understand the last clause so that he and him both refer to that individual that we have already established that is both Joe and the man with the devil costume.16 5.2. Identity Sentences Regarding sentences with a so called equative use of the copula like for instance the last part of (14) (which is the same as (3a)) (14) Who is that woman over there? She is professor Rigau. one might think that the explanation for why they are completely good sentences is the same that we just gave for (12). However this is not really the case. 5.2.1. A Problem. It is true that usually when we utter a sentence such as the last clause of (14) we are not presupposing that the two NPs refer to the same individual. Nevertheless it still might be perfectly fine to utter an identity sentence in a situation where according to what the participants assume, the two NPs will refer to same individual. For instance, if Diana, Pilar and Colin are in a room and they all know each other (and know that they know each other, etc), Diana's utterance of (15) is still good: (15) He is Colin. It might be a silly and pointless utterance, but it is still perfectly good from a grammatical point of view. 5.2.2. Heim's Reply. One reply given to this objection by severa1 authors (in Heim (1988,1993), Lasnik (1990), and in some way also in Higginbotham (1992)) 16. We can make the sentence good, or at least much better, by changing the tense of rhe discourse to past, and by replacing ((It's suspicious thab in the last clause by an expression such as eThis is why~, c<This explains thatn or ((No wonder then thatn. I do not think this poses a problem for the claim that what makes the last part of (8) good is that we are not presupposing that the individuals that he and him refer to are the same. The use of past tense and of expressions like (~This is the reason why ... )) facilitates interpreting the sentence with respect to the context as it was before the previous sentence had been uttered. We can also see this in a text like (i): (i) A: I wonder why Tom did not come to the party with his wife. B: Tom has never been married A: This explains why he did not come with his wife. The use of his wife in the last clause requires that it is not presupposed that Tom is not married. This, though, is exactly the information that has been introduced in the context by B. Nevertheless, the use of athis explains why~ makes easier to understand that what follows does not take for granted the information that this (in dhis explains why~) refers to. More could be said about how expressions iiie &is is the reason why ... > affect what the participants will be able to understand is the context they should use to evaluate a particular utterance. Examining this any further, though, is beyond the scope of this paper. Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 97 consists in claiming that by the very act of uttering a sentence that asserts certain identity the speaker indicates that we are not presupposing that identity (and so that the hearers should assume that the identity was not presupposed, even if they thought it was).17 If the context is assumed to have been such that it was not presupposed that the two individuals were the same, then the identity sentence is not in conflict with our disjointness conditions. The reason why the speaker indicates we are not presupposing the identity when uttering the identity sentence is that it is a general principle governing conversation that what it is asserted cannot be presupposed. This principle follows from Grice's Cooperative Principle, and, in particular, from the maxim of el atio n. l8 5.2.3. Objection-1 to Heim's Reply. I do not think we should be satisfied with this explanation of why identity statements are perfectly good sentences even when the identity is already presupposed. One way of realizing that this explanation cannot be satisfactory is by noticing the following: The principle that what is said can not be presupposed is just a principle about rational interchange of information. As with any other principle derived from the Cooperative Principle, it is possible, at the price of acting silly, not to act in accordance with the principle. I can say to you <<This man is Higginbothaml,, and you might reply ccYes certainly, he is Higginbothamn, and I might go on to say <<He is Higginbotham. He is Higginbotham>>. My utterance would be silly, but completely acceptable from a gramatical point of view. The reason why we feel that my utterance would be silly is that it would violate the maxim of Relation ('Be relevant'). So, if in the situation described we would realize that the maxim of Relation is not operating, we should not expect the principle ccwhat is said can not be presupposed>> (that gets its justification from that maxim) to apply either. Nevertheless, the sentence is good. So, it is not the operation of the principle ccwhat is said can not be presupposed), that explains why in the situation described the last utterance of <<He is Higginbothamn is still completely correct. 5.2.4. Objection-2 to Heim's Reply. The reasoning just given shows that it cannot be that the explanation for why identity sentences are correct in contexts were the identity is already presupposed makes essential use of the maxim of Relation, because there are contexts where the maxim of Relation is violated but where an identity sentence stating some identity that is already presupposed is still acceptable. Now, a more direct way of objecting to what we have termed Heim's Reply is to indicate where exactly there is a problem in the argumentation given as part of the reply. In order to do so, it will help to make more explicit what the argument that appeals to the principle 'what is said can not be presupposed' and that I arn objecting to is. 17. This process of reacting to an utterance of the speaker by regarding the context as having been different from what it actually was is what David Lewis calls accommodation. See Lewis (1979). 18. See essay 2 in Grice (1989). 98 CatWPL 511, 1996 Josep Macih The argument can be formulated in the following way: Let NP1 and NP2 be two referential noun phrases, and C a context where 'NP1 is NP2' is uttered and where the referents of NPI and of NP2 are presupposed to be the same; let 'val(X,w)' denote the individual that the referential expression X picks up in the world w, then, by assumption, we have (I) and (11): (I) 'NP1 is NP2' is uttered in C (11) NP1, NP2 and C are such that: VWE C (val(NP1,w) = val(NP2,w)) Given (11) we have that what 'NP1 is NP2' says is true in each world in C, and so that (111) holds: l9 (111) What 'NP1 is NP2' (when uttered in C) says is already presupposed in C. By Grice's Cooperation Principle and, in particular, by the maxim of Relation, we have (IV): (IV) When the speaker utters 'NP1 is NP2' in C, she is not saying something that is already presupposed in C. So, there is a conflict between (I), (111) and (IV). It is a general fact that conflicts involving the Cooperation Principle can be solved by accommodating (i.e., by assuming that the context was different from what it actually was). In the present situation the conflict is resolved by assuming (V): (V) The context where 'NP1 is NP2' was uttered was not C, but rather a context C' which is such that 3w~ Ct(val(NPl,w) # val(NP2,w)) Finally, given (V) we have (VI): (VI) The context where 'NP1 is NP2' is taken to have been uttered is such that there is no conflict between the Disjointness Conditions and the interpretation of the utterance in that context. Now that we have the argument laid out in detail we can see that there is a problem in (111). (111) assumes that the sentence 'NP1 is NP2' has an interpretation before the conclusion (VI) which establishes that the sentence is not in a conflict with the Disjointness Conditions has been reached. Premise (111) could be analyzed into two sub-premises (for concreteness we assume that what a sentence says is a proposition, that a proposition is a set a possible worlds, and that identity statements express the diagonal proposition in the sense of Stalnaker (1979) -these assumptions are not at all essential to the point I want to make, th~u~h):~~ 19. Recall that a context C presupposes that p if for each w 'C, p is true in w. 20. We could take (IIIa) and (IIIb) simply to be (111) a. 'NP1 is NP2' when uttered in C says that p. b. C presupposes that p. Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 99 (111) a. An utterance of 'NP1 is NP2' in C expresses the proposition: ( WE C:val(NPl ,w) = val(NP2,w)J b. C presupposes the proposition {WE C:val(NPl,w) = val(NP2,w)j (since from (11) it follows that C = (WE C:val(NPl,w) = val(NP2,w)), and so, in particular, C is included in {w~C:val(NPl,w) = val(NP2,w)J). (IIIa) is completely unjustified, unless we regard the problem we are trying to solve as already solved -that is, unless we assume that there is no problem in establishing that, given certain context, the identity sentence has certain interpretation even though the Disjointness Conditions prescribe that the sentence does not have that interpretation given that context. (Notice, incidently, that disjointness conditions are not taken at all into consideration in the argument (I)-(V) that establishes conclusion (VI)). In particular, under the assumptions accepted by Heim (where the disjointness conditions are syntactic constrains ruling out certain syntactic structures) (IIIa) is certainly not true. Any syntactic representation of 'NP1 is NP2' that would give rise in C to the interpretation that (IIIa) attributes to the sentence would already be ruled out at the syntactic level by the Principles (B) or (C) of the Binding heo or^,^' and so would not express any proposition. 5.2.5. Another Reply. Maybe behind what I called Heim's reply lurks another argument that does not appeal to the principle 'what is asserted can not be presupposed'. It is the following: Suppose that a context C is such that a use of the pronoun he would naturally pick up Collin in each world in C, and so would a use of the NP Collin. If someone utters He is Collin in C, what would be the natural way of interpreting the two NPs conflicts with what the Disjointness Conditions prescribe. The hearers, assuming that the speaker is rational and is trying to abide by the Cooperative Principle, will try to find a way of escaping the conclusion that the speaker's utterance is not interpretable. There is one way of doing so: to accommodate, that is, to assume that the context was not C but rather C', where C' is such that for some world in C' the two NPs pick up different individuals. So the hearers will accommodate, and if accommodation takes place then there is no problem in interpreting the sentence in accordance with the disjointness conditions. So, the fact that accommodation will take place explains that an utterance of He is Colin is grammatical even if uttered in a context where the identity was already presupposed. I think, though, that this argumentation is not satisfactory. If the reasoning in the previous paragraph were correct then accommodation should also be an option when trying to interpret other sentences that also involve a conflict with the disjointness conditions. Consider the contrast between (16) and (17): (16) This woman is Pilar. She is Pilar. 21. Heim (1993) focuses her discussion only on phenomena related to principle (B). What she says, though, is easily and naturally extendable to phenomena related to principle (C) as well. 100 CatWPL 511, 1996 Josep Macil (17) This woman is Pilar. She knows Pilar. (17) is quite bad (if She is taken to refer to Pilar), whereas (16) is fully acceptable. If accornrnodation is possible when interpreting the second clause of (16), then it should also be possible when interpreting the second clause of (17); and if the fact that accornmodation is possible is what explains that (16) is good then (17) should be predicted to be also perfectly good; but (17) is not good. Similarly, we cannot explain why identity sentences of the same form as (15) are completely good by appealing to an explanation like the one we gave in section 4 to account for the acceptability of sentences like (4a) or (4~).~~ Sentences like (4a) or (4c) are only partially acceptable. As pointed out in section 4.2, it is precisely the fact that the interpretation of these sentences is in conflict with what the Disjointness Conditions prescribe that explains why they are only partially acceptable. Identity sentences even when uttered in contexts as the one described for (15) are perfectly good, though. So, it cannot be that the interpretation of identity sentences involves a resolution of a conflict similar to the one involved in the interpretation of (4a) or (4c). So, the explanation of the complete acceptability of identity sentences can not be along the same lines as the one we gave for (4a) or (4c). 5.2.6. A feasible solution. I think that a promising way of trying to account for the complete grammaticality of identity sentences is in terms of the special syntactic character of sentences containing the verb to be. Severa1 works have pointed out and tried to explain the special syntactic properties of the so called copular senten ce^.^^ In particular Heggie (1988) argues that the predicate of copular sentences is not the verb to be, but rather one of the NPS.~~ That means that there is only one argument in identity sentences: the NP that is not the predicate.25 If we assume that the Disjointness Conditions apply only to arguments, then the Disjointness Conditions do not provide any information about how copular sentences 22. Recall that the explanation was that the interpretation of the sentence involves a conflict between the Disjointness Conditions and other kinds of information. In sentences like (4a) and (4c) the infonnation opposing the Disjointness Conditions is strong enough as to prevail and make the sentence interpretable. 23 See, for instance, Longobardi (1985), Rappaport (1987), Heggie (1988) and Moro (1991). 24 According to Heggie, this claim requires some qualifications. 25. Argument andpredicate are technical notions in linguistic theory. They are usually characterized in terms of the so called Theta-theory. One alternative way of characterizing the notions of argument andpredicate is in terms of the type of semantic values they can have. We will not commit ourselves here to any particular characterization of the two notions. Among many other altemative ways, one of the simplest characterizations would be the following: We define a predicate of degree-n inductively as follows: A predicate of degree-1 is an expression whose semantic value is a function from the set of individuals to the set of truth values; a predicate of degree-n+l is an expression whose semantic value is a function from the set of individuals to the set of predicates of degree-n. X is apredicate if, for some n, X is a predicate of degree-n. X is an argument, if X can be functionally combined with a predicate of degree-1 to yield a truth value. (Notice that under this definition not only expressions whose semantic value is an individual can be arguments; if the semantic value of a quantificational expression is a function from predicates of degree-1 to truth values, then quantificational expressions are also arguments). Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 101 like (15) should be interpreted. The interpretation of identity sentences does not, then, involve any conflict with the Disjointness Conditions. Discussing the different arguments that show the special character of copular sentences and that try to prove that the predicate of copular sentences is one of the NPs, as well as examining the possible evidence against this view is by itself a lengthy topic. The reader is referred to the discussion in Heggie (1988) and the other works mentioned in footnote 23. Here we will restrict ourselves to mentioning one kind of data that gives plausibility to the view, without entering in any further details. In Catalan the clitic el (L') corresponds to an argument position; the clitic ho can correspond to predicates but cannot correspond to any personal NP in an argument position. This is illustrated in (18). In (18b) the clitic el stands for the argument el Joan in (18a), while it is not possible for ho to stand for that argument:26 (18) a. Aquell home estima el Joan. that man loves (the) John b. Aquell home l'/*ho estima. that man himlit loves In (19b) the clitic ho stands for the predicate molt felig 'very happy' in (19a), whereas el can not stand for that predicate (19) a. El Joan és molt feliq. (the) John is very happy b. El Joan ho/*ll és. (the) John ithim is In contrast with (18a-b), the argument clitic el in (20b) can not be made to stand for the NP el Joan in the identity sentence (20a), but the predicate clitic ho can:27 (20) a. Aquest home és el Joan. this man is (the) John b. Aquest home *l'/ho és. this man hidit is 26. Heggie(1987) and Longobardi(l985) describe similar facts to (18)-(20) for French and Italian, respectively. 27. According to Fabra (1956), the standard normative text for the Catalan language, the judgements for (20b) should rather be (i). Aquest home lt/*ho és. I have collected the linguistic judgements for these sentences from six Catalan speakers. None agrees with what Fabra (1956) prescribes; five of tbem agree with the data as I presented above; one of them -maybe under the normative influence of Fabra (1956)- judges that the NP in (20a) can pronominalize as el or ho. 102 CatWPL 511, 1996 Josep Macih The data (18)-(20) suggest that the NP el Joan in (20a) is not an argument but a predicate. If the Disjointness Conditions apply only to argument NPs, then the Disjointness Conditions are irrelevant for the interpretation of (20a). In particular, there would be no conflict with the Disjointness Conditions if we interpreted the two NPs in (20a) in such a way that they determine the same individual in each world in the context.28 6. Some comments 6.1. The Disjointness Conditions and the Cooperative Principle That our Disjointness Conditions (B)' and (C)' be concerned with what it is presupposed to be the case, rather than what is actually the case is what allows our proposa1 to dea1 with examples like (12). It could seem, though, that this very same feature gives rise to wrong predictions regarding the use of sentences like (21) in certain contexts. (21) He loves him. Usually when someone utters (21) the context will contain information about who each NP refers to (say, we are talking about John's feelings for Paul, or we are talking about who loves Clinton and at that point Henry comes into the room, etc). If this contextual information implies that He and him refer to different individuals (and so that, for each world in the context, the individual that He picks up is distinct from the individual that him picks up), then we will interpret the sentence with the two NPs referring to two different individuals. This is in agreement with what the Disjointness Conditions prescribe (the Disjointness Conditions prescribe that it not be presupposed that the NPs refer to the same individual, and the contextual information agrees with that since this information will determine that the individuals that the two NPs refer to are presupposed not to be the ~ame).~~ If we hear someone utter (21) and we have to interpret it without being able to use any previous context (say we have just joined an ongoing conversation), though, we will also think that the individuals referred to by He and him are distinct, not just that it is not presupposed that they are the same. This rnight seem to be a problem for the formulation of the disjointness conditions that I am defending for the following reason: If (21) is uttered in a context that contains no information about who the referent of the two NPs is, then it seems that the hearers, in order to interpret the NPs, will have to rely solely on the lexical information and the information 28. I use the word <<detennine)> in an ambiguous way. The two NPs will 'determine' the individual in a different way, since, presumably, they will have a different type of semantic value. If the semantic value of the argument NP Aquest home 'this man' is an individual, then the semantic value of the predicate NP el Joan 'John' might, for instance, be a set of individuals -actually, a singleton set, which also determines an individual. 29. Compare: (I) Not to presuppose that p; (II) to presuppose that not-p. (I) does not imply (11), though (11) implies (I) (assuming consistency), so to satisfy (11) is one way of satisfying (I). Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 103 provided by the Disjointness Conditions. The lexical information is just that the individuals the NPs refer to are male and are not the speaker or the hearers though they are somewhat salient at that point in the conversation. The information provided by the Disjointness conditions is that either (I) it is an open question whether the two NPs refer to the same individual &e.. in some worlds in the context thev \. pick up the same individual, but in some other worlds they do not), or (11) the two NPs refer to two different individuals (i.e., in each world in the context they pick up different individuals). The Disjointness Conditions do not determine which of (I) or (111) is the case. So, using solely the Disjointness Conditions (and the lexical information) we could not conclude (11). But, as pointed out above, (11) is what we do conclude when we hear (21) in a context that does not provide any information about the reference of the NPs. This seems to suggest that there is a problem for my formulation of the disjointness conditions and that they should be amended. If what we conclude when we hear (21) in a context that does not provide information on the reference of the NPs is that they refer to different individuals, and it seems that all the information we can use in such a context is provided by the disjointness conditions (and lexical semantics), then it seems that what the disjointness conditions should prescribe is that (when the relevant syntactic relation obtains) the interpretation of the two NPs must be presupposed to be distinct, not just not presupposed to be the same. (Adopting this modification would, of course, be problematic in other respects since, for instance, we would no longer be able to account for sentences like (12)). I think, though, that the suggested objection above overlooks one kind of information that will be available to the hearers when (21) is uttered even if the context does not include any information about who the NPs refer to: that the speaker is conforming to Grice's Cooperative Principle. Barring information to the contrary, we will always assume that the Cooperative Principle and in particular, the Gricean maxims, are respected. And we will rely on this assumption when trying to interpret what the speaker says. If we hear (21) in a context that does not contain information about who the referent of the two pronouns is, but that, otherwise, is a non-special context, we will assume that the speaker knows who she is talking about. So she knows who the person that she is referring to by He is, and who the person that she is referring to by him is, and so she knows whether they are the same or not. If she knew she was talking about one single person, then she would not talk as if the question whether there is one person or two was open, since doing so would go against the maxim of Quantity ('be as informative as is required'). So, since she is treating the question as open, we can conclude that she is not talking about one single verson but two. Of course there can be special contexts (as, for instance, in (12)) in which it is clear that the speaker is unsure about the identity of the individuals she is talking about, or where she has some good reason to act as if she is unsure. This is not, though, what is usually the case. Barring information to the contrary, we will assume that the conversation does not take place in any of these special contexts, and so we will assume that the speaker knows who she is talking about, and that she is open about it. So, assuming Disjointness Condition (B)' and the Gricean 104 CatWPL 511, 1996 Josep Macil maxims, it follows, as desired, that in non-special contexts the hearers will assume that when the speaker utters (21) she is talking about two individuals. Notice that in a special context where it is reasonable to believe that Gricean maxims do not apply, like, for instance, when an oracle says something, or in the statement of some puzzle in, say, the Sunday edition of a newspaper, we would not conclude from the use of (21.) that the individual referred to by He is not the individual referred to by him, but rather only that it might or it might not be the same. 6.2. Sources of information for the interpretation of NPs As we have pointed out, there are severa1 sources of information that are used in interpreting the NPs in a sentence, besides the information provided by the Disjointness Conditions on the basis of the structural relation among the NPs. In this subsection we will exarnine some of the ways of generating the information that, as we have seen, in sentences like the ones in (3)-(5) can conflict with and prevail over the information provided by the Disjointness Conditions. 6.2.1. Presuppositions. The most common way of generating such information is by making it clear that the clause in which certain NP X appears expresses some property which is the same as certain property that has been introduced before. The property involves certain individual or certain pattern of linking. So X must refer to that individual or be subject to that particular pattern of linking. One way of having this effect is by repeating severa1 times the attribution of a property, so as to create a pattern that will make the listener expect that the next attribution of a property will fit the same pattern (this is illustrated, for instance, in (3c)). Another way of having this effect is by having mentioned the application of the property as a general case and then making clear that we are considering a particular application of that general case (this is illustrated in (3e) and (5a)). One specially good way of having this effect, though, is by using some device which introduces a presupposition. We have already cornmented (with respect to (3c)) on the effect that the presence of also can have. The word even (that appears in (3b)) has a similar effect to that of also. The sentences in (3)-(5) illustrate, though, other ways of introducing presuppositions besides including some specific word. Consider, for instance, (22): (22) What do you mean no one loves Oscar? He loves him. P.51 He clearly refers to Oscar, which is the only salient individual when the pronoun is uttered. In a more common discourse hirn would be taken as referring to some other individual different from Oscar (because of Disjointness Condition (B)' and the Cooperation Principle). But a sequence of the form What do you mean a? P carries the presupposition that P implies no-a. In the case of (22) this means that He loves hirn implies that it is not the case that no one loves Oscar. The most simple and likely way for He loves hirn to have that implication is if He loves hirn itself is an attribution of the property of loving Oscar. If this is so, then hirn must refer to Oscar, Disjointness Condition (B)' notwithstanding. Binding Theory, Semantic Interpretation and Context CatWPL 511, 1996 105 Another way of introducing presuppositions is by stressing some word. Whatever exactly the presupposition induced by stressing an NP is, it includes that the property being attributed to the individual determined by the NP, has already been considered. So in (23) (23) What do you mean John loves no one? He loves JOHN. í5.71 the stress on John carries the presupposition that the property of being loved by the subject of the last clause has already being considered. There are two properties that have been considered in the first clause: the property of being loved by John and the property of loving no one. The latter, though, cannot be the one that the stressing of John carries the presupposition about since in the last clause -even before determining who the NPs refer toit is clear that we are not attributing the property of loving no one. So the presupposition brought about by stressing John, is that, in the last clause, we are attributing the property of being loved by John; that means that He must refer to John, Disjointness Condition (C)' notwithstanding. There is also, of course, the presupposition brought about by the structure What do you mean a? P which adds its effect to the stressing of John. 6.2.2. Demonstration. Another way of generating information that can override the information provided by the Disjointness Conditions is by using an NP demonstratively. (24a,b) illustrate this: (24) a. He refuted HIM [points to the person] b. A: John saw Peter. B: No, John saw HIM [points to John] In, for instance, (24a) He refers to whoever is most salient individual at that point in the conversation; because of Disjointness Condition (B)', without demonstration him would be assumed not to refer to the same individual; the demonstration, though, forces it to refer to that same individual. Using an NP demonstratively is a very clear way of indicating what the NP refers to; this is why the intended reference can be communicated even if the Disjointness Conditions are providing opposing information. Notice that most of the sentences considered in the previous subsection involved severa1 devices that together were able to quite successfully override (B)' or (C)'. (24a) shows that pointing is effective enough as to have that effect on its own. Still the combination of demonstration with other devices, as in (24b), makes the overruling of (B)' clearer and the sentence better. 6.2.3. Pronominals versus Referential Expressions. A pronoun is very strongly dependent on the context to determine what it refers to. This is not so for proper names and definite descriptions. They are to an extend dependent on the context (there are many people named John and many presidents), but not as much as pronouns are. Proper narnes and definite descriptions are able, to a good extend, to determine their referent 'on their own'. 112 CatWPL 511, 1996 Josep Macia Longobardi, G. (1985). CSU alcune propietti della sintassi e della forma logica delle frasi copulari)>. In: Sintassi e mo$ologia della lingua Italiana dfuso:teorie ed applicazioni descriptive, Atti del XVII Congresso Intemazionale di Studi della SLI. Urbino: Sept. 1983. Roma: Bulzoni, p. 21 1-213. Moro, A. (1991), ccThe Anomaly of Copular Sentences)). The University of Venice Working Papers in Linguistics 8, Centro Linguistico Interfacoltti, Universith di Venezia. Postal, P. (1970). <<On coreferential complement subject deletion,). Linguistic Inquiry 1: 439500. Reinhart, T. (1983a). Anaphora and semantic interpretation. London: Croom Helm. - (1983b). doreference and Bound Anaphora: A Restatement of the Anaphora Questions)>, Linguistics and Philosophy 6: 47-88. Schutze, C. (1995). The empirical base of linguistics: Grammaticality judgements and linguistic methodology. Chicago: University of Chicago Press. Stalnaker, R. (1973). c<Presuppositions>). Journal of Philosophical Logic 2: 77-96. - (1974). c<Pragmatic Presuppositions>>. In: M.Munitz; P.Unger (eds.). Semantics and Philosophy. New York: New York University Press. - (1979). c<Assertion>>. In: P.Cole (ed.) Syntax and Semantics 9: Pragmatics, p. 315-332.