Towards a Probabilistic Version of Bidirectional OT Syntax and Semantics

This paper argues that a purely probabilistic version of bidirectional Optimality-Theoretic syntax and semantics (Blutner 2000, Zeevat 2000) can make some significant contributions to the computational processing of natural language. The ideas outlined in this paper apply to interpretation as well as generation, but particular attention will be given to the question how bidirectionality can be put to use in Natural Language Generation. The paper sketches a number of different possible approaches, but one of them will be described in more detail, at the heart of which lies an empirically oriented notion of vicious ambiguity.