Dependency parsing of Japanese spoken monologue based on clause-starts detection

A dependency parsing method based on sentence segmentation into clauses has been proposed and confirmed to be effective. In this method, dependency parsing is executed in two stages: at the clause level and the sentence level. However, since a sentence can not be segmented into complete clauses, in the past research, a unit sandwiched between two clause-end boundaries (clause boundary unit) was adopted as an approximate unit of the complete clause. There is the problem that the dependency structure of the clause boundary unit is not necessarily closed. This paper proposes a method for dependency parsing based on sentence segmentation into units which correspond to clauses and whose dependency structure is completely closed (clause fragment). Our method identifies such the unit by re-dividing a clause boundary unit at clause-start boundaries. As the results of the experiment show, we confirmed the improvement of the dependency parsing accuracy by utilizing the clause fragment as a parsing unit.