A Statistical Approach to Expandable Spoken Dialog Systems using WFSTs

We have proposed an efficient approach to manage a dialog system using a weighted finite-state transducer (WFST) in which users¿ concept and system¿s action tags are input and output of the transducer, respectively. A WFST for dialog management was automatically created using a corpus annotated with inter-change format (IF) consisting of dialog acts and argument which is an interlingua for machine translation. A word-to-concept WFST for spoken language understanding (SLU) was created using the same corpus. The scenario and SLU WFSTs acquired from the corpus were composed together and then optimized. We have confirmed the WFST automatically trained using the annotated corpus can manage dialog reasonably.