Presentation 1996/12/13
Task adaptation of a stochastic language model for dialogue speech recognition
Akinori Ito, Masaki Kohda,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) A stochastic language model (SLM) is indispensable for continuous speech recognition. Generally,large corpus of the task domain is required to make a good SLM. When making a SLM for a specific task domain, it is ideal to obtain large number of sentences of the domain. But it takes large time and effort to collect linguistic data of a specific domain, especially of a spoken dialog domain. In this paper, we investigated possibility of making a good N-gram SLM using small corpus of a specific domain with task independent large corpus. Sightseeing information dialog task was chosen for the specific task, and we examined several kinds of corpora for task independent corpus. We carried out experiments to measure perplexity of the adapted N-gram model. From the experiments, it is found that the adaptation improved perplexity of the model when the task domain of the small and large corpora are similar. The results also showed that the coherence of morphemic analysis of the small and large corpora greatly affects the perplexity of the adapted model.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) continuous speech recognition / stochastic language model / N-gram / task adaptation
Paper # NLC96-50,SP96-81
Date of Issue

Conference Information
Committee NLC
Conference Date 1996/12/13(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Vice Chair

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) Task adaptation of a stochastic language model for dialogue speech recognition
Sub Title (in English)
Keyword(1) continuous speech recognition
Keyword(2) stochastic language model
Keyword(3) N-gram
Keyword(4) task adaptation
1st Author's Name Akinori Ito
1st Author's Affiliation Faculty of Engineering, Yamagata University()
2nd Author's Name Masaki Kohda
2nd Author's Affiliation Faculty of Engineering, Yamagata University
Date 1996/12/13
Paper # NLC96-50,SP96-81
Volume (vol) vol.96
Number (no) 420
Page pp.pp.-
#Pages 8
Date of Issue