Presentation | 2010-12-20 Evaluation of spotting algorithm constrained by keyword co-occurrence for dialogue systems Aki KATO, Yoshihiko NANKAKU, Akinobu LEE, Keiichi TOKUDA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Question-answering dialogue system often choose a response sentence based on recognized keywords in a user's utterance. In such case, a robust utterance understanding can be achieved by recognizing only the keywords, skipping irrelevant part of speech, rather than decoding the whole input speech. Also, since the intention of an utterance can be expressed as combination of keywords (keyword set) rather than a set of single keyword, it is desirable to extract the keywords as keyword sets. In this study, we propose an algorithm which directly applies the keyword set constraints by consulting their co-occurrences during search using a large vocabulary garbage model. By applying the constraints dynamically while search, it suppresses unnecessary hypotheses and thus expected to perform more efficiently and result in more robust intention detection. The proposed keyword-set spotting algorithm is implemented on large vocabulary continuous speech recognition decoder "Julius" fully on both passes. We evaluated the performance of the proposed method. It was confirmed that the recognition rate of keywords by spotting was superior to the dictation-based method. In addition, keyword spotting constrained by co-occurrences improved the keyword extraction rate by about 12.5 % relatively at the maximum. In this paper, we report the evaluation results in a small task of 150 keywords and the Takemaru task. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | dialogue system / keyword / co-occurence constraints / spotting / garbage model |
Paper # | NLC2010-16,SP2010-89 |
Date of Issue |
Conference Information | |
Committee | NLC |
---|---|
Conference Date | 2010/12/13(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 | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Evaluation of spotting algorithm constrained by keyword co-occurrence for dialogue systems |
Sub Title (in English) | |
Keyword(1) | dialogue system |
Keyword(2) | keyword |
Keyword(3) | co-occurence constraints |
Keyword(4) | spotting |
Keyword(5) | garbage model |
1st Author's Name | Aki KATO |
1st Author's Affiliation | Department of Computer Science and Engineering, Nagoya Institute of Technology() |
2nd Author's Name | Yoshihiko NANKAKU |
2nd Author's Affiliation | Department of Computer Science and Engineering, Nagoya Institute of Technology |
3rd Author's Name | Akinobu LEE |
3rd Author's Affiliation | Department of Computer Science and Engineering, Nagoya Institute of Technology |
4th Author's Name | Keiichi TOKUDA |
4th Author's Affiliation | Department of Computer Science and Engineering, Nagoya Institute of Technology |
Date | 2010-12-20 |
Paper # | NLC2010-16,SP2010-89 |
Volume (vol) | vol.110 |
Number (no) | 356 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |