Representing Conversation Acts in a Uni ed Semantic/Pragmatic Framework

We would like to argue for a uni ed formal theory to represent and allow automated reasoning about both the semantic and the pragmatic aspects of conversational context, particularly the common ground between conversational participants. Central to this theory is the hypothesis that speech act occurrences are recorded in the common ground and can serve as the basis for a model of language processing in context that reconciles in a general and systematic fashion the di erences between the theories of the common ground adopted in current theories of discourse processing which focus more narrowly on sub-issues such as reference resolution, intention recognition, and dialogue management. One example of poor t between existing theories of context is the contrast between, on the one hand, linguistically motivated theories developed to account for the semantics of anaphora (e.g., (Kamp & Reyle 1993)); and on the other hand, the models of context proposed for intention recognition and dialogue management, whose emphasis is on capturing the effects of speech acts on the beliefs, intentions, and obligations of the participating agents (Allen 1983; Carberry 1990; Cohen & Levesque 1990; Perrault 1990; Traum& Allen 1994). These traditions resulted in very detailed proposals about context and context update; but the resulting models of context di er signi cantly. It is not possible to simply adopt one or the other model. While the linguistically motivated theories of context integrate well with current theories of semantic interpretation, their relation with current work on planning and plan recognition is less clear; the opposite is true of theories of context based on actions and their e ects. We desire a uni ed model for several reasons. For one thing, we wish to develop theories of the interpretation of, discourse phenomena, such as de nite descriptions, which requires information of both a semantic