Study on response generation for spoken dialogue systems based on machine learning and statistical models

2015 Fiscal Year Final Research Report

• Project/Area Number: 24500115
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Media informatics/Database
• Research Institution: Nara Institute of Science and Technology
• Principal Investigator: Kawanami Hiromichi 奈良先端科学技術大学院大学, 情報科学研究科, 助教 (80335489)
• Co-Investigator(Kenkyū-buntansha): SHIKANO Kiyohiro 奈良先端科学技術大学院大学, 情報科学研究科, 教授 (00263426) ; SARUWATARI Hiroshi 東京大学, 情報理工学系研究科, 教授 (30324974)
• Project Period (FY): 2012-04-01 – 2016-03-31
• Keywords: 音声インタフェイス / 音声対話システム / 情報検索

Outline of Final Research Achievements

Speech-oriented information guidance systems have put into practice. Many of them introduce an example-based response generation technique. However,improvement of response accuracy and variety of response text expression are expected for a more useful interface. In this study, we propose machine learning approach to response selection and statistical machine translation approach to produce variation of response expression.
The proposed machine learning classifier which introduces two-stage classification using outputs of first stage classifiers, SVM, pboost etc. overcomes a conventional example-based method. On the other hand, it remains precise analysis to translate/convert a question to an answer because of difficulty of association between question and a answer.
In addition, we developed a prototype of a chat robot for upcoming demand on a personal spoken dialogue system.

Free Research Field

音声情報処理