Reversible discourse processing

In this paper we describe a reversible framework for discourse processing, a framework in which contextual constraints can be speci ed independently of the processing algorithm. We take as a starting point Scha and Polanyi's discourse grammar ([Scha/Polanyi 1988]) which is in essence a uni cation based grammar extended with arbitrary relations. Although such a framework can be shown to have many non properties (e.g. non monotonicity and non declarativity), we argue that it is inherently nitely reversible in the sense of [Dymetman 1991] that is, there is a parsing/generation program for this grammar such that for any string/semantics, the program enumerates all associated semantics/strings and terminates. We show by means of examples that one and the same speci cation can be used both for analysis and for generation. ii