Condition B effects in two simple steps

Floris Roelofsen 0 Anaphora Pronouns Condition B 0 F. Roelofsen (&) ILLC, University of Amsterdam , P.O. Box 94242, 1090 GE Amsterdam, The Netherlands This paper is concerned with constraints on the interpretation of pronominal anaphora, in particular Condition B effects. It aims to contribute to a particular approach, initiated by Reinhart (Anaphora and semantic interpretation, 1983) and further developed elsewhere. It proposes a modification of Reinhart's Interface Rule, and argues that the resulting theory compares favorably with others, while being compatible with independently motivated general hypotheses about the interaction between different interpretive mechanisms. 1 Introduction The interpretation of anaphoric pronouns is constrained by syntactic structure. One particular approach to capturing such constraints was initiated by Reinhart (1983). Her main thesis was that only one type of anaphoric relation, binding, is syntactically encoded and subject to rules of syntactic well-formedness. She assumed that other anaphoric relations, such as coreference, are not syntactically encoded and are therefore not regulated by rules of grammar. Instead, she suggested, there is a pragmatic principle which implies that coreference is ruled out whenever it yields exactly the same interpretation as binding. This approach has been further developed by Heim (1998), Fox (2000), Bu ring (2005b), Reinhart (2006), and Reuland (2008). The aim of the present paper is to evaluate and refine these analyses. In particular, some previously unnoticed problems for the analyses of Buring (2005b) and Reinhart (2006) will be pointed out, and a modification of the latter will be proposed. The resulting theory will be argued to have certain advantages over the analyses of Heim and Bu ring. In particular, it avoids certain controversial assumptions about VP ellipsis that Burings account requires, and it avoids the assumption that anaphoric relations other than binding are syntactically represented and constrained, which is what Heims account requires. The idea behind the proposed account can be seen as a particular instance of the more general hypothesis that hearers minimize interpretive options (Reinhart 2006) or of the closely related idea that rejection of a certain interpretation by one component of the grammar cannot be overruled by other components (Reuland 2008). To the extent that the account is empirically correct, it provides support for both these general hypotheses. 2 Reinharts original approach The main thesis of Reinhart (1983) is that binding relations are encoded in the syntax and subject to grammatical constraints, whereas other kinds of anaphoric relations, such as coreference, are not encoded in the syntax, and therefore not subject to grammatical constraints. Rather, coreference is established contextually, and restrictions on coreference are of a pragmatic nature. In order to discuss Reinharts proposal in more detail, we must first fix some formal terminology and notation, as well as some basic assumptions about the syntax-semantics interface. 2.1 Basic terminology, notation, and assumptions The most straightforward way to encode binding relations in the syntax is by means of indices. Let us assume, then, that pronouns enter a syntactic derivation either with or without an index. If a pronoun comes with an index, we will call that index its binding index, and we will append the index to the pronoun in subscript (e.g., [him1]). Pronouns with a binding index will be treated as bound by some other determiner phrase; pronouns without a binding index will be treated as referential. I will assume that determiner phrases may undergo wh-movement or quantifier raising. If a DP undergoes wh-movement or QR, it receives a binder index n, which is adjoined to it in superscript (e.g., [who]3). It also leaves behind a trace which has that same index n as its binding index (e.g., the trace of [who]3 would be [t3]). I will assume a two-stage Montagovian syntax-semantics interface: logical forms are first compositionally mapped to expressions in some type-theoretical language, and these expressions are in turn interpreted model-theoretically. A pronoun or a trace with a binding index n is interpreted as a variable xn, and a constituent of the form XnY is interpreted as: where X0 is the interpretation of X and Y0 is the interpretation of Y. This composition rule embodies what Heim and Kratzer (1998) call predicate abstraction. By way of illustration, the logical form in (4) is mapped to the type-theoretical expression in (5) (where I use somewhat sloppy notation): [John]1[t1 loves his1 mother] We will say that a determiner phrase X binds a pronoun P iff (i) X c-commands P, (ii) Xs binder index coincides with Ps binding index, and (iii) X does not c-command any other DP that satisfies (i) and (ii). This notion of binding is what Heim and Kratzer (1998) and Bu ring (2005a) call semantic binding and what Reinhart (2006) calls A-binding. Notice that, according to this notion, [John] binds [his] in (4). As for referential pronouns, i.e. pronouns without a binding index, I will make a slightly non-standard assumption. I will assume that the formal language contains, besides standard individual constants and variables, a set of individual proterms. Moreover, formal expressions are interpreted not only relative to a model and an assignment function, but also relative to a resolution function. The interpretation of individual constants is determined by the interpretation function, which is given by the model; variables are interpreted by the assignment function, and proterms are interpreted by the resolution function. If a pronoun does not have a binding index, it is mapped to a proterm, and thus interpreted by the resolution function, which assigns it a contextually salient individual concept.1 Finally, we will say that two referential determiner phrases A and B corefer in a particular utterance context C (which furnishes a particular resolution function) iff A and B denote the same individual in every world that is consistent with the speech participants common assumptions in C. This notion of coreference is often called presupposed coreference (cf. Bu ring 2005a, p. 153). If a referential pronoun P is supposed to be resolved in such a way that it corefers with a determiner phrase A, I will write P A next to the LF in question. For example, I will write: 1 This setup deviates from that of Heim and Kratzer (1998), which is often taken as a starting point. In particular, referential pronouns are not treated as free variables here. Bound and referential pronouns are really dealt with by two separate mechanisms. The main reason for me to take this line here is that it seems to be the most straightforward way of implementing Reinharts distinction. Incidentally, setting things up in this way has the additional advantage of not having to stipulate that no logical form may contain both b (...truncated)