Abstract.
Spoken language dialogue systems are an important step on the way to the ideal human-computer interface, and the semantic analysis of spontaneous speech plays a fundamental role in any such system. This article presents the development of a semantic analyzer for unconstrained speech, independent of the application domain and the language spoken. A case grammar formalism is used for knowledge representation, and the parsing is based on a hidden Markov model trained from annotated dialogue corpora. After describing the architecture of the analyzer, together with details of its components, a series of experiments towards integrating it in a Romanian dialogue system for classes timetable information retrieval are presented. Evaluations carried out during these experiments have shown performance figures close to the best previously reported in the literature.
Work done while at Politehnica University of Timi?oara.