UTokyo Repository 東京大学

UTokyo Repository >
124 情報理工学系研究科 >
40 電子情報学専攻 >
1244025 修士論文(電子情報学専攻) >

このページ(論文)をリンクする場合は次のURLを使用してください: http://hdl.handle.net/2261/51736

タイトル: イベント共参照関係を利用した因果関係知識の獲得
その他のタイトル: Extracting Causal Relation using Event Coreference
著者: 田中, 翔平
著者(別言語): Tanaka, Shohei
発行日: 2012年3月22日
抄録: Causal relations are essential knowledge for interpreting discourse structure of text. This paper presents a method that extracts causal relations of lexical patterns in the form of quasi Horn clauses: for example, "A acquires B" AND "B is located in X"→"the acquisition gives A operations in X". The input of the method is a relation tuple consisting of two entities and a verb, e.g., (A, acquire, B). The method finds coreferencial expressions of the given relation that mention the same event. We use nominal forms of verbs included in FrameNet for finding the coreference expression. If a nominal form of a relation occurs as the subject in the dependency tree of a sentence, the sentence is likely to describe what is caused by the event. Then the method uses several NLP techniques (part-of-speech tagging, coreference resolution, dependency parsing and named entity recognition) in order to build a lexical pattern containing the entities (A and B). However, such rules are so specific that computers cannot reuse the knowledge of causality relation: for example, "the acquisition gives A operations in Nevada". Therefore, the proposed method generalize the rules by introducing a variable X and estimating the relation between X and A or between X and B. For evaluation, we asked human annotators to judge the correctness of the causal-relation rules. The result shows that the proposed method precisely extracts the causal-relation rules.
内容記述: 報告番号: ; 学位授与年月日: 2012-03-22 ; 学位の種別: 修士 ; 学位の種類: 修士(情報理工学) ; 学位記番号: ; 研究科・専攻: 情報理工学系研究科電子情報学専攻
URI: http://hdl.handle.net/2261/51736
出現カテゴリ:025 修士論文
1244025 修士論文(電子情報学専攻)


ファイル 記述 サイズフォーマット
48106422.pdf745.57 kBAdobe PDF見る/開く



Valid XHTML 1.0! DSpace Software Copyright © 2002-2010  Duraspace - ご意見をお寄せください