A Joint Model for Graph-based Chinese Dependency Parsing

A Joint Model for Graph-based Chinese Dependency Parsing Xingchen Li, Mingtong Liu, Yujie Zhang†, Jinan Xu, Yufeng Chen School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China †yjzhang@bjtu.edu.cn Abstract In Chinese dependency parsing, the joint model of word segmentation, POS tagging and depen- dency parsing has become the mainstream framework because it can eliminate error propagation and share knowledge, where the transition-based model with feature templates maintains the best performance. Recently, the graph-based joint model (Yan et al., 2019) on word segmenta- tion and dependency parsing has achieved better performance, demonstrating the advantages of the graph-based models. However, this work can not provide POS information for downstream tasks, and the POS tagging task was proved to be helpful to the dependency parsing accord- ing to the research of the transition-based model. Therefore, we propose a graph-based joint model for Chinese word segmentation, POS tagging and dependency parsing. We designed a charater-level POS tagging task, and then train it jointly with the model of Yan et al. (2019). We adopt two methods of joint POS tagging task, one is by sharing parameters, the other is by using tag attention mechanism, which enables the three tasks to better share intermediate information and improve each other’s performance. The experimental results on the Penn Chinese treebank (CTB5) show that our proposed joint model improved by 0.38% on dependency parsing than the model of Yan et al. (2019). Compared with the best transition-based joint model, our model improved by 0.18%, 0.35% and 5.99% respectively in terms of word segmentation, POS tagging and dependency parsing. 1 Introduction Chinese word segmentation, part-of-speech (POS) tagging and dependency parsing are three fundamen- tal tasks for Chinese natural language processing, whose accuracy obviously affects downstream tasks such as semantic comprehension, machine translation and question-answering. The traditional method is usually following pipline way: word segmentation, POS tagging and dependency parsing. However, there are two problems of the pipline way, one is error propagation:incorrect word segmentation directly affects POS tagging and dependency parsing, another is information sharing: the tree tasks are strongly related, the label information of one task can help others, but the pipline way cannot exploit the correla- tions among the three tasks. Using joint model for Chinese word segmentation, POS tagging and dependency parsing is a solution to these two problems. The previous joint models (Hatori et al., 2012; Zhang et al., 2014; Kurita et al., 2017) mainly adopted a transition-based framework to integrate the three tasks. Based on the standard sequential shift-reduce transitions, they design some extra actions for word segmentation and POS tag- ging. Although these transition-based models maintained the best performance of word segmentation, POS tagging and dependency parsing, its local decision problem led to the low precision of long-distance dependency parsing, which limited the precision of dependency parsing. Different from the transition-based framework, the graph-based framework has the ability to make global decisions. Before the advent of neural network, the graph-based framework was rarely applied to the joint model due to its …

Content truncated to 500 words