@techreport{oai:ipsj.ixsq.nii.ac.jp:00194262,
 author = {陰山, 宗一 and 井上, 一成 and 奥村, 紀之 and Soichi, Kageyama and Kazunari, Inoue and Noriyuki, Okumura},
 issue = {9},
 month = {Jan},
 note = {小説などの物語文章に登場する人物を的確に抽出したい場合，ゼロ代名詞も含む代名詞がどの人物を指すのかを正しく解析する必要がある．また，同姓の登場人物がいる場合，苗字のみの記述からどの人物を指すのかを同定することも必要となる．本研究では KNP を用いて照応解析を行い，照応解析を複数ステップで実施した場合の傾向を調査している．さらに，KNP では抽出不可能であった人物名を抽出するため，KNP の固有表現モデルの再学習を実施した．解析対象の小説自身をモデルの学習に使用した場合では，小説特有の表現に由来するミスを軽減できた．一方，解析対象となる小説の著者以外の複数著者による小説を学習データとした場合でも，人物名抽出が適切に行えることを確認した．, When accurately extracting the person' s name in a story such as a novel, it is necessary to correctly analyze a person indicated by a pronoun including a zeroâĂŞ-pronoun. Furthermore, when there is a character of the same family name, it is necessary to analyze which person is indicated from the description of only the family name. In this research, the anaphoric analysis is performed using KNP, and the tendency when anaphora analysis is performed in multi-steps is investigated. Besides, to extract person names that could not be extracted were constructed the named entity model of KNP. When learning a model with a novel to be analyzed, it was possible to reduce the mistakes caused by a novel specific expression. On the other hand, we confirmed that even if novels by multiple authors who are not the authors of the novel to be analyzed are used as learning data, person name extraction can be performed appropriately.},
 title = {KNPを用いた複数ステップの照応解析による日本語ゼロ代名詞の先行詞同定},
 year = {2019}
}