抄録
類似した文書群に共通するイベント系列により文書群を組織化する可能性について論じる。そのために、文の格構造の汎化を要素としてもつ極大類比を導入し、汎化コスト条件を用いた極大類比の発見的検出手法を示す。あわせて、非類似文書とのネガティブマッチングにより、可能な類比を抑制する効率の良い方法も示す。意味処理の困難さに起因する問題点についても触れ、それに関する今後の計画も述べる。
Given two or more similar documents in the form of texts, we present a notion of maximal analogies representing maximal sequences consisting of pairs of similar events in the documents. They are required to satisfy certain cost condition so that meaningless similarities between documents are never concluded. A bottom-up search procedure to find a maximal analogy satisfying the cost condition is also presented. In addition to the set of similar documents, we suppose another set of dissimilar ones. Then a maximal analogy is furthermore tested for their appropriateness so as not to explain the latter ones. The test can be performed by an effective subsumption check procedure.
