Page 395 - Encyclopedia of Philosophy of Language

straints precluding dependency into a relative clause for example (cf. Morgan 1973). Thus bare argument fragments can be interpreted in context as expressing whole propositions, reconstructing complex prep- ositional structure from some antecedent source:
Joan wants Sue to visit Bill in hospital. And Mary too.
(23)
The elliptical And Mary too, which can be uttered by some other speaker, can be reconstructed either as (a) 'Mary wants Sue to visit Bill in hospital,' or (b) 'Joan wants Mary to visit Bill in hospital,' or as (c) 'Joan wants Sue to visit Mary in hospital.' However, just as wh- questions cannot license binding of the wh- expression into a relative clause, so these fragments cannot be reconstructed as replacing some expression inside a relative clause:
Joan visited the woman who likes Bill, in hospital. (24) And Mary too.
The fragment in (24) can be construed as 'Mary visited the woman who likes Bill in hospital,' and as 'Joan visited Mary in hospital.' But it cannot be interpreted as 'Joan visited the woman who likes Mary, in hospi- tal.' This island-constraint phenomenon is generally taken as definitive of a grammatical phenomenon. But there is independent reason to consider that the interpretation of fragments is a pragmatic process, sensitive to general on-line constraints on utterance interpretation—just as is anaphora. If we are not to abandon all hope of retaining the concept of grammar as some encapsulated language-specific faculty, we have to state grammar-internal processes in such a way that they are able to be implemented over prag- matically induced configurations. We need to define utterance interpretation as a structure-building pro- cess from the under-determining input provided by grammar-internal principles, and construe all con- figurational constraints on interpretation as con- straints on that structure-building process. We thus arrive at the conclusion that processes of interpret- ation need to bejust as syntactic as the configurations familiar from syntactic theory. The boundaries between syntax and interpretation need to be blurred yet further.
The exploration of such processes of interpretation is taking several routes. There is Discourse Rep- resentation Theory, with its intermediate hybrid rep- resentations, part semantic, part syntactic (Kamp and Reyle 1993). There is so-called 'structured semantics,' in which structure is superimposed on the model itself (cf. Cresswell 1985). And there is a proof-theoretic route whereby utterance interpretation is char- acterized as an inferential process of syntactically building a proof structure. This was first proposed by Sperber and Wilson (1986) as one aspect of their over- all theory of utterance interpretation. Here a formal model of this option is outlined (Gabbay and Kemp-
son 1992; Kempson and Gabbay in press), showing how it predicts directly the interaction between fam- iliar syntactic constraints and processes of general reasoning without abandoning the encapsulation of the language faculty as an independent input to pro- cesses of general reasoning. (This model has been fur- ther developed by Kempson, Gabbay and Meyer- Viol; cf. Kempson, et al. in press.)
Suppose we assume that stored in the lexicon for each lexical item is a specification of its contribution to utterance interpretation. In the simple cases, this takes the form of a pair—a conceptual expression, and a specification of its logical type (expressed as in Montague semantics in terms of the two primitives: e, an individual denoting expression; and t, a truth- bearing entity; and combinatorial functions on these). A verb such as swim for example, expresses the two- place relation swim', an expression of logical type (e, (e, / » . Expressing types as propositions and taking the corresponding logical expression as a matching label for its twinned type, we can view this lexically stored information as labeled premises in a logic— premises which will combine together to deduce the proposition expressed by an uttered sentence, as con- clusion. Thus in a sentence such as John loves Mary, we have three words:
John, loves, Mary
yielding from the lexicon three premises:
John' :e
love' :e• •(e^t) Mary':e
Assuming here concepts of subject and tense, infor- mation from the lexicon will lead by two steps of modus ponens to the proposition:
love' (Mary') (John'): t
For every step of modus ponens taken, the infor- mation in the labels builds up, recording the assump- tions used and their mode of combination, the resulting conclusion a well-formed formula of a predi- cate logic labeling the logical type t (the logic assumed is the labeled natural deduction system of Gabbay
1996).
Such a sentence as this displays no obvious under-
determinacy, but the phenomenon of anaphora can be reconstructed, with its under-determining input, and dependency established in context, as a process of natural deduction. An initiating assumption of some metavariable over labels is entered as the premise lexi- cally associated with the pronoun, an assumption which is discharged by identification with some infor- mation independentlyrecoverable from theinference- structure already presented—in effect, the natural deduction moves of the Rule of Assumptions, reiter- ation of a premise from one inference-structure to another, and assumption discharge:
Semantics versus Syntax: Perspectives on Natural Language Content
373