权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

音声対話における言語の解析と生成に関する研究

口语对话中的语言分析与生成研究

基本信息

批准号：
05241102
负责人：
田中穂積
金额：
$ 54.21万
依托单位：
Tokyo Institute of Technology
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research on Priority Areas
财政年份：
1993
资助国家：
日本
起止时间：
1993 至 1995
项目状态：
已结题

项目摘要

本年度は最終年度であるため,各分担研究項目の相互の関連を考慮しつつ研究のまとめを行った.「対話文理解のためのシソ-ラスの構築と対話文の解析」では,試作したシソ-ラスの不備を検討し上位・下位関係シソ-ラスに部分・全体関係のシソ-ラスを融合させ拡張を施した(鶴丸).対話文の頑健な解析を行うために,コストと報酬という概念を導入した不適格文処理の方法を確立するとともに、実験によりその有効性を実証した(松本).また,対話文に含まれる割り込み現象と割り込みからの復帰について分析し,プランに基づく対処法を提案した(田中).「対話文の言語学的ならびに認知科学的研究」では,制約に基づく文法理論の考え方に従い,日本語の音韻理論の構築を行った.実例として,動詞の活用形を単一化により記述可能なことを示すとともに、日本語の使役形をとりあげ、提案した音韻理論の有効性を示した(郡司).また対話理解のために,言語情報の他に,音声情報,映像情報が果たす役割を認知科学的観点から検討し,実験により,音声情報と対話相手の映像情報が対話を円滑に進める重要な要因であることを明らかにした(阿部).「対話文の生成」では,対話文生成のついて前年度までの研究成果をさらに洗練した.特に,システム全体を動的なネットワークで表現し,対話者がそれぞれこのネットワークの構成に関与するとともに,そこに,対話者の心理状態を反映させるモデルを構築し,その有効性を明らかにした(岡田).「言語情報と音声情報の統合処理」では,前年度までに提案した,異音のモデルを統語モデルに組み込む新しい統合処理のアルゴリズムを洗練し,これを実装しその有効性を確認した.また,このアルゴリズムに異音のバイグラムの情報を組み込む新しい方法を提案した.これによりテストセットパ-プレクシチィの顕著な減少が見られることを見いだした(田中).

This year is the final year, and the interrelationships between the research projects are considered. "The construction and analysis of the corresponding text understanding", try to make the best of the situation, discuss the superior relationship, the subordinate relationship, the partial relationship, the overall relationship, and the integration of the situation (Tsurumaru). The concept of salary is introduced into the text processing method. (Tanaka). "Research on Linguistics and Cognitive Science of Japanese Language" is a review of the basic grammar theory of Japanese language, and the construction of phonological theory of Japanese language. For example, if the flexible forms of verbs are simplified, it is possible to describe them, the causative forms of Japanese are simplified, and the validity of phonological theory is proposed (County Office). The important reason for speech comprehension is that speech information is related to other information, audio information, and image information. The research results of the previous year are now being refined. In particular, the speech of the whole dynamic life of the speech, the speech of the speaker is related to the composition of the speech, the speech of the psychological state of the speaker is reflected in the speech of the speech, the speech of the speaker is reflected in the speech of the speech, the speech of the speaker is reflected in the speech of the speech, the speech of the speaker is reflected in the speech of the speech, the speech of the speaker is reflected in the speech of the speech of the speech, the speech of the speaker is reflected in the speech of the speech, the speech of the speech of the speaker is reflected in the speech of the speech of the speech, the speech of the speaker is reflected in the speech of the speech of the speech, the speech of the speech of the speaker is reflected in the speech of the speech, the speech of the speaker is reflected in the speech of the speech, the speech of the speaker is reflected in the speech of the speech, the speech of the "Integrated processing of speech information and sound information" was proposed in the previous year, and the effectiveness of the integration of sound information and sound information was confirmed. This is a new way to organize information about different sounds in different languages. The number of visitors to the site is decreasing.