@techreport{oai:ipsj.ixsq.nii.ac.jp:00048271,
 author = {渋木, 英潔 and 荒木, 健治 and 栃内, 香次 and Hideyuki, Shibuki and Kenji, Araki and Koji, Tochinai},
 issue = {76(2003-NL-156)},
 month = {Jul},
 note = {本稿では構文解析が行われた文に対して深層格を推測する手法について提案する．提案手法は，以下の二段階で推測を行う．第一段階では，単語概念がもつ深層格に対する制約と選好を基に，推測される可能性がない深層格を排除した後，フィルモアの一文一格の原理に違反しないように仮説を立てて推測する．第二段階では，第一段階の結果から，深層格と一意に対応する表層格の規則を獲得し，その規則に従って再解析を行う．一文一格の原理が満たされた文を用いた場合と，一文一格の原理が満たされておらず未知語を含む文を用いた場合の二通りの条件で実験を行った．一文一格の原理が満たされた文に対しては，確率的に最大の深層格を推測する手法の正解率61.8%から66.0%に向上した．また，未知語を含む文に対しては，28.4%から50.1%に向上した．, We propose a method to infer deep cases of sentences analyzed syntactically.  The proposal method is carried out with the following two stages. In the first stage, plausible deep cases are focused on using the constraint and preference of deep cases of each concept, and then a plausible one is inferred based on a hypothesis that satisfies Fillmore's one-instance-per-clause principle. In the second stage, rules that correspond a surface case to a deep case are acquired from a result of the first stage, and then deep cases are fixed based on the rules. We conducted two experiments under the condition that satisfies the one-instance-per-clause principle and condition that includes unknown words. In the condition that satisfies the one-instance-per-clause principle, we confirmed that the accuracy rate was increased from 61.8% to 66.0% in comparison with inference by simple maximum probability. In the condition that includes unknown words, the accuracy rate was increased from 28.4% to 50.1%.},
 title = {一文一格の原理と深層格選好に基づいた日本語深層格規則の自動獲得手法},
 year = {2003}
}