Joint Chinese word segmentation and POS tagging system with undirected graphical models

For Chinese Part-Of-Speech(POS) tagging, word segmentation is a preliminary step. To reduce accumulated errors between two steps and improve the segmentation performance by utilizing POS information, segmentation and POS tagging can be performed simultaneously. In this paper, a joint segmentation and POS tagging system is proposed based on undirected graphical models which can make full use of the dependencies between the two stages. In the joint system, segmenting and tagging are viewed as the sequence labeling; moreover any connected sub-graph can be viewed as a certain dependency which can be used to find the final opinion labeling. The joint model achieves high performances with 97.19% in segmentation precision and 95.34% in POS tagging precision, which are the state-of-art performances for Chinese word segmentation and tagging on 1998-year People's Daily corpus.