Transition-based dependency parser with postponed determinations for Japanese sentences

Abstract

Dependency analysis is recognized as a basic part of natural language processing. Transition-based dependency parsers transition between states in a deterministic way using local state information. Because most transition-based parsers are incremental, once they add an incorrect dependency between lexical tokens, this is never revised. Therefore, an incorrect final state is reached. To avoid the addition of incorrect dependencies, this paper reports the progress of research to develop a novel transition-based parser that postpones the determination of whether to add a dependency between a certain pair of lexical tokens.