Introducing Utterance Verification in Spoken Dialogue System to Improve Dynamic Help Generation for Novice Users
Kazunori Komatani, Yuichiro Fukubayashi, Tetsuya Ogata, Hiroshi G. Okuno · 2007
A method is presented that helps novice users understand the language expressions that a system can accept, even from unacceptable utterances made that may contain automatic speech recognition errors.We have developed a method that dynamically generates help messages, which can avoid further unacceptable utterances from being made, by estimating a users' knowledge from their utterances.To improve the accuracy of the estimation, we developed a method to estimate a user's knowledge from utterance verification results.This method estimates whether a user knows an utterance pattern that the system considers acceptable, and suppresses useless help messages from being generated.