Improving Chinese Word Segmentation and POS Tagging with Semi-supervised Methods Using Large Auto-Analyzed Data

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 309–317, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Improving Chinese Word Segmentation and POS Tagging with Semi-supervised Methods Using Large Auto-Analyzed Data Yiou Wang†, Jun’ichi Kazama†, Yoshimasa Tsuruoka‡†, Wenliang Chen§†, Yujie Zhang∗†, Kentaro Torisawa† †National Institute of Information and Communications Technology (NICT), Japan ‡School of Information Science, JAIST, Japan; ∗Beijing Jiaotong University, China §Human Language Technology, Institute for Infocomm Research, Singapore {wangyiou, kazama, torisawa}@nict.go.jp; tsuruoka@jaist.ac.jp wechen@i2r.a-star.edu.sg;yjzhang@bjtu.edu.cn Abstract This paper presents a simple yet effective semi-supervised method to improve Chi- nese word segmentation and POS tagging. We introduce novel features derived from large auto-analyzed data to enhance a sim- ple pipelined system. The auto-analyzed data are generated from unlabeled data by using a baseline system. We evaluate the usefulness of our approach in a series of experiments on Penn Chinese Treebanks and show that the new features provide substantial performance gains in all exper- iments. Furthermore, the results of our proposed method are superior to the best reported results in the literature. 1 Introduction In Chinese, most language processing starts from word segmentation and part-of-speech (POS) tag- ging. These two steps tokenize a sequence of char- acters without delimiters into words and predict a syntactic label (POS tag) for each segmented word. They are considered indispensable steps for higher-level NLP tasks such as parsing and infor- mation extraction. Although the performance of Chinese word segmentation and POS tagging has been greatly improved over the past years, the task is still challenging. To improve the accuracy of NLP systems, one of the current trends is semi-supervised learning, which utilizes large unlabeled data in supervised learning. Several studies have demonstrated that the use of unlabeled data can improve the perfor- mance of NLP tasks, such as text chunking (Ando and Zhang, 2005), POS tagging and named entity recognition (Suzuki and Isozaki, 2008), and pars- ing (Suzuki et al., 2009; Chen et al., 2009; Koo et al., 2008). Therefore, it is attractive to con- sider adopting semi-supervised learning in Chi- nese word segmentation and POS tagging tasks. In this paper, we present an approach to improve the performance of both segmentation and POS tagging by incorporating large unlabeled data. We first preprocess unlabeled data with our baseline models. We then extract various items of dictio- nary information from the auto-analyzed data. Fi- nally, we generate new features that incorporate the extracted information for both word segmenta- tion and POS tagging. We also perform word clus- tering on the auto-segmented data and use word clusters as features in POS tagging. In addition, we introduce lexicon features by using a cross- validation technique. The use of sub-structures from the auto- annotated data has been presented previously (No- ord, 2007; Chen et al., 2008; Chen et al., 2009). Chen et al. (2009) extracted different types of sub- trees from the auto-parsed data and used them as new features in standard learning methods. They showed this simple method greatly improves the accuracy of …

Content truncated to 500 words