Online error detection of barge-in utterances by using individual users' utterance histories in spoken dialogue system

Authors:
Kazunori Komatani;Hiroshi G. Okuno
Affiliations:
Kyoto University, Yoshida-Hommachi, Sakyo, Kyoto, Japan;Kyoto University, Yoshida-Hommachi, Sakyo, Kyoto, Japan
Venue:
SIGDIAL '10 Proceedings of the 11th Annual Meeting of the Special Interest Group on Discourse and Dialogue
Year:
2010

Citing 5
Cited 0

Learning to predict problematic situations in a spoken dialogue system: experiments with how may I help you?

NAACL 2000 Proceedings of the 1st North American chapter of the Association for Computational Linguistics conference
Flexible mixed-initiative dialogue management using concept-level confidence measures of speech recognizer output

COLING '00 Proceedings of the 18th conference on Computational linguistics - Volume 1
Automatic detection of poor speech recognition at the dialogue level

ACL '99 Proceedings of the 37th annual meeting of the Association for Computational Linguistics on Computational Linguistics
User Modeling in Spoken Dialogue Systems to Generate Flexible Guidance

User Modeling and User-Adapted Interaction
Predicting barge-in utterance errors by using implicitly supervised ASR accuracy and barge-in rate per user

ACLShort '09 Proceedings of the ACL-IJCNLP 2009 Conference Short Papers

Quantified Score

Hi-index	0.00

Visualization

Abstract

We develop a method to detect erroneous interpretation results of user utterances by exploiting utterance histories of individual users in spoken dialogue systems that were deployed for the general public and repeatedly utilized. More specifically, we classify barge-in utterances into correctly and erroneously interpreted ones by using features of individual users' utterance histories such as their barge-in rates and estimated automatic speech recognition (ASR) accuracies. Online detection is enabled by making these features obtainable without any manual annotation or labeling. We experimentally compare classification accuracies for several cases when an ASR confidence measure is used alone or in combination with the features based on the user's utterance history. The error reduction rate was 15% when the utterance history was used.