This paper discusses the establishment of semantic classification rules for Japanese subordinate clauses based on a reconstruction of the Tori-Bank interclausal semantic classification system for Japanese to English machine translation, which is achieved by analyzing sentence patterns within Japanese-English corpora. As a conflict analysis of semantic classification tagging for subordinate clause annotation in the Balanced Corpus of Contemporary Written Japanese has revealed problems in the cross-application of this system to Japanese subordinate clauses, this paper attempts to reconstruct this classification system from a Japanese linguistics perspective. As context and intention affect the type of an adverbial clause, semantic classification rules can maintain the accuracy of clause annotation by proposing coarse-grained classification. The paper also summarizes the interclausal keyword problems inhibiting subordinate clause annotation (overlapping interclausal keywords between multiple clauses) and proposes improvements to reliability. While there are a number of challenging problems in clause boundary recognition and semantic classification of clauses, this research points to the potential for stable rules for the classification of Japanese subordinate clauses through reconstructing this system.
日本語従属節の意味分類基準策定について : 「鳥バンク」節間意味分類体系再構築の提案
papers1507.pdf
(866.75KB)
概要
