Presentation 2003/12/12
Out-of-Domain Utterance Detection based on Confidence Measures from Multiple Topic Classification
Ian R. LANE, Tatsuya KAWAHARA, Tomoko MATSUI, Satoshi NAKAMURA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) One significant problem for spoken language systems is how to cope with users' OOD (out-of-domain) utterances which cannot be handled by the back-end system. In this paper, we propose a novel OOD detection framework, which makes use of classification confidence scores of multiple topics and trains a linear discriminant in-domain verifier using GPD. Training is based on deleted interpolation of the in-domain data, and thus does not require actual OOD data, providing high portability. Three topic classification schemes of word N-gram models, LSA and SVM are evaluated, and SVM is shown to have the greatest discriminative ability. In an OOD detection task, the proposed approach achieves an absolute reduction in EER of 6.5 points compared to a baseline method based on a simple combination of multiple-topic classifications. Furthermore, comparison with a system trained using OOD data demonstrates that the proposed training scheme realizes comparable performance while requiring no knowledge of the OOD data set.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Out-of-domain utterance detection / topic classification / confidence measures / in-domain verification
Paper # NLC2003-96
Date of Issue

Conference Information
Committee NLC
Conference Date 2003/12/12(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Natural Language Understanding and Models of Communication (NLC)
Language ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Out-of-Domain Utterance Detection based on Confidence Measures from Multiple Topic Classification
Sub Title (in English)
Keyword(1) Out-of-domain utterance detection
Keyword(2) topic classification
Keyword(3) confidence measures
Keyword(4) in-domain verification
1st Author's Name Ian R. LANE
1st Author's Affiliation Graduate School of Informatics, Kyoto University:ATR Spoken Language Translation Laboratories()
2nd Author's Name Tatsuya KAWAHARA
2nd Author's Affiliation Graduate School of Informatics, Kyoto University:ATR Spoken Language Translation Laboratories
3rd Author's Name Tomoko MATSUI
3rd Author's Affiliation ATR Spoken Language Translation Laboratories
4th Author's Name Satoshi NAKAMURA
4th Author's Affiliation The Institute of Statistical Mathematics
Date 2003/12/12
Paper # NLC2003-96
Volume (vol) vol.103
Number (no) 518
Page pp.pp.-
#Pages 6
Date of Issue