Peter Bosch - Computing Definites with Unarticulated Arguments

Consider the sequences (1a,b) and (2a,b), each of which is perceived as a coherent discourse:

(1)    a.    A Lotus 2-Eleven was offered on ebay.    b. The price was ridiculous.
(2)    a.    A Lotus 2-Eleven was just buzzing by.     b. The sound was terrible.

The feeling of coherence is gone if we swap the two (b)-sentences. Conventional wisdom about “bridging”, associative anaphora, or accommodation cannot explain the resulting incoherence – which indicates that it doesn’t suffice as an explanation for the NP definiteness in the (b)-sentences either. One problem is that, although price is usually taken to denote a functional concept, sound normally is not. Yet, in (1) and (2) both seem to denote functional concepts, the arguments remaining ‘unarticulated’ in the (b) sentence. Another problem is that the unarticulated argument cannot simply be the denotation of Lotus 2-Eleven – otherwise also the sequences (1a)+(2b) and (2a)+(1b) should be equally coherent. Apparently we require different concepts as denotations for Lotus 2-Eleven in (1) and (2), one with a ‘price’ attribute and no ‘sound’ attribute and another one with a ‘sound’ attribute and no ‘price attribute.
This paper presents a conception of the semantics-pragmatics interface that pretends to account for these observations. The idea comes close to Jerry Fodor’s disquotational view as far as lexical semantics is concerned by shifting much of the burden of lexical semantic explanation to the conceptual system. And it builds upon a Kaplan-type semantics in its treatment of context dependence, but extends the Kaplan account to subsentential constituents: Instead of evaluating sentential expressions at a context, we propose to build sentence denotations (truth conditions) compositionally and incrementally from denotations of smaller constituents, ultimately denotations of words, that are assigned to expressions relative to the context. This architecture uses lexical representations that are vastly underspecified, containing only some constraints over possible denotations (including concepts) and their arguments. All additional information that is required in language comprehension is drawn in from other sources: The conceptual system (conceived of as in description logics, but with additional options of cognitively realistic dynamic subsumption) as well as any discourse or external (possi-bly perceptual) information that is antecedently or concurrently available.
The motivation for this architecture comes from an attempt to use established methods of formal syntax and semantics in a cognitively realistic theory of utterance comprehension. In particular, we are trying to come to grips on the one hand with phenomena of context-dependent linguistic productivity as they have been widely observed in lexical semantics and, on the other hand, with experimental evidence showing that “pragmatic” information is being used by the comprehender immediately as it becomes available and long before the end of a clause or sentence is reached.
This effort has lead to the notion of “contextual concepts”, which are the denotations of occurrences of unsaturated expressions (“predicates”) and are truth functions (defined, however, only for arguments in the intended current context). Contextual concepts are instances of generic types in the knowledge representation and are subject to productive formal operations on the conceptual representation. They are computed on the fly in the course of utterance comprehension.

References

Bosch, Peter (to app.) Productivity, Polysemy, and Predicate Identity. ten Cate & Zeevat (eds.) Proc. Sixth Tbilisi Symposiumon Language, Logic & Computation. Springer.

Löbner, Sebastian (1998): Definite Associative Anaphora, unpubl. 1998/2003