Page 393 - Encyclopedia of Philosophy of Language

which the heterogeneity of this phenomenon can be reduced, cf. Cooper 1979; Hausser 1979; Kamp 1981; Heim 1982; Kempson 1988; Heim 1990; Chierchia 1995; Kamp and Reyle 1993; van Eijk and Kamp 1997).
This proliferation of ambiguities is across the board. All anaphoric expressions depend for their interpretation on some concept of context of utterance giving rise to a range of truth-theoretically discrete types of meaning. Examples (8)-(17) display an array of VP, nominal, and demonstrative anaphoric depen- dencies, having in common only this property of dependency on their immediate surrounding context for assignment of some interpretation:
John likes Mary, but I don't. (8)
Everyone worries about their logic paper, (9) except Marcelo, who never does.
John kissed everyone that Sue didn't. (10)
Don't. (11)
Most students like Maths, but I know at least (12) one who does not
One will do. (13)
Jo telephoned a journalist every time Sue (14) interviewed one.
That man is a nuisance. (15)
She made a cake for John and that bastard never (16) thanked her.
Every time she had a coke, she knew that later, (17) that week she'd have a headache.
The phenomenon is not even restricted to explicitly anaphoric processes. It can apply to tense construal (Partee 1984b), to the interpretation of adjectives (Klein 1980), adverbs (Stump 1985), and so on. Indeed natural-language expressions, both simple and complex, are invariably construed relative to some unfolding concept of context. We seem faced with a multiplicity of ambiguities far beyond what any lexical sources of ambiguity would lead us to expect.
This phenomenon is additionally problematic for the Montague program, because it conflicts with a central cornerstone of the program—the com- positionality principle. According to this principle, the meaning of any compound expression is recursively defined as a function of the meaning of its elementary parts. Indeed the force of model-theoretic semantics lies in large measure in the substance it provides to the claim that the truth-theoretic content for sentence- sized expressions is a function of the way in which the content of lexical elements combines together to yield a truth-evaluable whole. But as Kamp pointed out in 1981, it is not obvious in what sense the sentence He was sick in (2) has a truth-evaluable content as a sentence—rather the truth-evaluable content that it has is dependent on the evaluation of the previous
sentence John came in and the relation between the two.
The solution Kamp proposed to these problems became the first model to blur the syntax-semantics dichotomy, and this analysis was the first dynamic model of semantic evaluation (Kamp 1981). He defined an intermediate level of characterization called 'discourse representation structure' (DRS), a structure defined by an algorithm whose role was to assign values to anaphoric expressions as part of the mapping from syntactic construct to DRS. The DRS so assigned to a sentence was then subject to model-theoretic evaluation. So rather than interpreting the sentence John came in, and only then, independently inter- preting He was sick, the algorithm projects a DRS for the first sentence (I), and extends this by the infor- mation provided by the second sentence to create the new DRS II:
Semantics versus Syntax: Perspectives on Natural Language Content
A DRS is a partial model containing some nonempty set of entities called discourse referents, and predicates on those entities. Truth is then defined for a DRS in terms of its embeddability in some total model, and not as a property of sentences directly. Despite the model-theoretic status of these mini-models, the internal structure of such DRS'S has played an impor- tant role in the way the theory has developed. There is for example a stated locality condition on the identi- fication of a discourse referent for pronouns that they be identified from some 'higher' DRS. This structural condition on pronominal identification led in its turn to a rule moving elements from some subordinate box to some higher DRS. For example the two readings of (18) according as the indefinite is or is not internal to the conditional are distinguished as the discourse referent for the indefinite is or is not moved to the 'top box':
If a friend of mine conies to see me, let her in. (18)
Such locality conditions and/or movement processes critically invoke properties of the DRSqua rep- resentation in ways not naturally reducible to some model-theoretic image. Thus we get the first blurring of semantic and syntactic constructs—constructs defined in semantic terms but manipulated in terms of their configurational properties. There is also unclar- ity in the status of such intermediate representations. Are they a set of representations internal to the natu-
371