共起情報に基づく呼応関係自動抽出法の検討(抽出(1))  [in Japanese] Examination of Method for Extracting KOOU Relations based on Co-occurrence Information  [in Japanese]

本稿では,文を理解する際に役立ち得る呼応関係を頻度情報のみに基づく類似尺度による自動抽出法を検討する.呼応関係とは,一文において,副詞や係助詞が呼びかけとなり,文末がその呼びかけに応答する形式をとる,語間に存在する関係の一つである.この呼応関係を知ることによって,文を文末まで読むまたは聞くことなく,文を理解することが可能となる.呼応関係は呼要素と応要素からなる.本研究では,コーパス中の各文に現れる副詞や係助詞を呼要素の候補,その呼要素の候補より後方に続く形態素を単位としたngramを応要素の候補とし,いくつかの共起情報に基づく類似尺度を呼応関係の自動抽出に適用した.その結果を比較し,・呼応関係抽出法として有効的であった尺度を示す.

This paper examines an automatic extraction method by the similar measure only based on frequency information in corpora for a certain concord relation which can be useful in case he understands a sentence. The concord relation that we extract is called KOOU relation, which is particular to Japanese. The KOOU relation is a relation that KO element calls out OU element which responds to its call in a sentence. KO elements are kakari particle or adverb in many cases. OU elements are phrase or string nearby the sentence end. In this study, we consider that all ngrams made of the morphemes which continue the candidate of KO element appearing in each sentence as candidates of OU element, and apply some similar measures based on co-occurrence information to extract KOOU relation. From a comparison of the results, we show the effective measures as KOOU relation extraction method in our experiment.