Joint Model of Korean Part-of-Speech Tagging and Dependency Parsing with Partial Tagged Corpus

Abstract

Abstract-Most recent studies on part-of-speech (POS) tagging and dependency parsing employ a pipelined model design. However, pipelined structures may decrease performance on account of error propagations. Furthermore, syntactic information is required to improve POS tagging performance. In this paper, we propose a joint model of POS tagging and dependency parsing for the Korean language. Our joint model analyzes the maximum score dependency tree with POS tagging using the graph-based CKY parsing method. We present an effective application method for an additional POS tagged corpus and evaluate the method for POS tagging and dependency parsing. The results show that our model improves accuracy by approximately 2.3% and 1.7% more than a pipelined structure using a hidden Markov model and graph-based dependency parsing, respectively.