Dependency Parsing of Japanese Spoken MonologueBased on Clause-Starts Detection

摘要

A dependency parsing method based on sentence segmentationinto 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 asentence can not be segmented into complete clauses, in the pastresearch, a unit sandwiched between two clause-end boundaries(clause boundary unit) was adopted as an approximate unit ofthe complete clause. There is the problem that the dependencystructure of the clause boundary unit is not necessarily closed.This paper proposes a method for dependency parsing based onsentence segmentation into units which correspond to clausesand whose dependency structure is completely closed (clausefragment). Our method identifies such the unit by re-dividinga clause boundary unit at clause-start boundaries. As the resultsof the experiment show, we confirmed the improvement of thedependency parsing accuracy by utilizing the clause fragmentas a parsing unit.