Chinese Dependency Parsing Based on An Improved Model of MST

摘要

In this paper, a Chinese dependency parsing method was presented based on improved Maximum Spanning Tree algorithm. Within this method, Conditional Random Field (CRF) is adopted to establish sequence labeling model. Recognizing POS of head node is employed to modify the weights of directed edges in the MST model. Comparative experiments on CoNLL 2009 data set show that the new method shows better performance than current Chinese dependency methods, with precision reaching to 85.45%.