The Development of a Thai Spoken Dialogue System

Abstract

This article summarily reports the development of the first Thai spoken dialogue system (SDS), namely a Thai Interactive Hotel Reservation Agent (TIRA) . In the development, a multi-stage technique of spoken language understanding (SLU), which combined a word spotting technique for concept extraction and a pattern classification technique for goal identification, was proposed. To improve system performance, a novel approach of logical n-gram modeling was developed for concept extraction in order to enhance the SLU robustness. Furthermore, dialogue contextual information was used to help understanding and finally, an error detection mechanism was constructed to prevent unreliable interpretation outputs of the SLU. All the mentioned algorithms were incorporated in the TIRA system and evaluated by real users.