Talk by Massimo Poesio

Defeasible Reasoning with Underspecified Representations

\ Many see underspecified representations as a tool that will allow us to formalize sentence disambiguation as a defeasible reasoning process. Yet, the existing theories of defeasible reasoning are built on the same assumptions about inference that characterize more traditional forms of reasoning: we can conclude that P is true if it can be derived from other wffs already known to be true. But the only way proposed in the literature to make the initial, underspecified interpretation of a sentence definitely true or false is to give underspecified languages a `disjunctive' semantics which doesn't really capture our intuitions about ambiguity. What we can do, however, is to `go meta', that is, to model the disambiguation process as defeasible inference about what has been uttered, rather than about the content of the utterance. The approach to undespecification that results from this shift in perspective is also more general than existing approaches, in that it can also be used to model syntactic underspecification.

Back to AcTen Program


Paul Dekker, November 2, 1995