Robust interpretation for spoken dialogue systems

Abstract

Spoken dialogue systems must allow for robust and efficient interpretation of user utterances. This can be achieved by using shallow and partial interpretation. Partial interpretation is feasible together with a dialogue manager which provides information to guide the analysis. In this paper we present results on developing interfaces for information retrieval applications utilizing partial and information directed interpretation with unification-based formalisms, traditionally used for deep and complete analysis. The major advantage with our approach is that the time to develop the interpretation modules is reduced. Furthermore, the system will be fairly robust as large parts of the knowledge bases containing knowledge on ways in which a user can express a domain concept can be generated automatically or semi-automatically.