Ranking Multiple Dialogue States by Corpus Statistics to Improve Discourse Understanding in Spoken Dialogue Systems

Abstract

Ryuichiro HIGASHINAKA†a) , Nonmember and Mikio NAKANO † †b) , Member SUMMARY This paper discusses the discourse understanding process in spoken dialogue systems. This process enables a system to understand user utterances from the context of a dialogue. Ambiguity in user utterances caused by multiple speech recognition hypotheses and parsing results sometimes makes it difficult for a system to decide on a single interpretation of a user intention. As a solution, the idea of retaining possible interpretations as multiple dialogue states and resolving the ambiguity using succeeding user utterances has been proposed. Although this approach has proven to improve discourse understanding accuracy, carefully created hand-crafted rules are necessary in order to accurately rank the dialogue states. This paper proposes automatically ranking multiple dialogue states using statistical information obtained from dialogue corpora. The experimental results in the train ticket reservation and weather information service domains show that the statistical information can significantly improve the ranking accuracy of dialogue states as well as the slot accuracy and the concept error rate of the top-ranked dialogue states.

Selecting Help Messages by Using Robust Grammar Verification for Handling Out-of-Grammar Utterances in Spoken Dialogue Systems

“…There have been several studies on identifying the user's intentions as slot values by using history information and treating multiple candidates derived from ambiguous ASR results. In such studies, the most appropriate slot value was selected after maintaining multiple candidates using rules for tree structures that maintain histories [17], a particle filter [18], or statistics on dialogue acts and dialogue state up-dates [19]. History information has also been used to estimate confidence measures for slot values in a current user utterance [20], [21].…”

Expansion of WFST-Based Dialog Management for Handling Multiple ASR Hypotheses

“…Figure 1 shows the WFSTDM system. Focused on statistical models for dialog systems, thus far a statistical model of a system action sequence for dialog strategy [7] and that of user concept sequence for understanding [8] [9] [10] were investigated independently. We constructed a statistical model for dialog scenario obtained from the tag sequence of both the clerk and customer sides [11].…”