Incremental Joint POS Tagging and Dependency Parsing in Chinese

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1216–1224, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Incremental Joint POS Tagging and Dependency Parsing in Chinese Jun Hatori1 Takuya Matsuzaki1 Yusuke Miyao3 Jun’ichi Tsujii4 1University of Tokyo / 7-3-1 Hongo, Bunkyo, Tokyo, Japan 2National Institute of Informatics / 2-1-2 Hitotsubashi, Chiyoda, Tokyo, Japan 3Microsoft Research Asia / 5 Danling Street, Haidian District, Beijing, P.R. China {hatori,matuzaki}@is.s.u-tokyo.ac.jp yusuke@nii.ac.jp jtsujii@microsoft.com Abstract We address the problem of joint part-of-speech (POS) tagging and dependency parsing in Chi- nese. In Chinese, some POS tags are often hard to disambiguate without considering long- range syntactic information. Also, the traditional pipeline approach to POS tagging and depen- dency parsing may suffer from the problem of error propagation. In this paper, we propose the first incremental approach to the task of joint POS tagging and dependency parsing, which is built upon a shift-reduce parsing framework with dy- namic programming. Although the incremental approach encounters difficulties with underspeci- fied POS tags of look-ahead words, we overcome this issue by introducing so-called delayed fea- tures. Our joint approach achieved substantial improvements over the pipeline and baseline sys- tems in both POS tagging and dependency pars- ing task, achieving the new state-of-the-art per- formance on this joint task. 1 Introduction The tasks of part-of-speech (POS) tagging and de- pendency parsing have been widely investigated since the early stages of NLP research. Among mainstream approaches to dependency parsing, an incremental parsing framework is commonly used (e.g. Nivre (2008); Huang and Sagae (2010)), mainly because it achieves state-of-the-art ac- curacy while retaining linear-time computational complexity, and is also considered to reflect how humans process natural language sentences (Fra- zier and Rayner, 1982). However, although some of the Chinese POS tags require long-range syntactic information in order to be disambiguated, to the extent of our knowledge, none of the previous approaches have addressed the joint modeling of these two tasks in an incremental framework. Also, since POS tag- ging is a preliminary step for dependency parsing, the traditional pipeline approach may suffer from the problem of error propagation. In the example sentence in Figure 1, „ has POS ambiguity between DEG (a genitive marker), which connect two noun phrases as “’s” in En- glish, and DEC (a complementizer), which intro- duces a relative clause. Since both can take the form of NP-„-NP (NP: noun phrase), it is hard to distinguish these two tags only by considering local context. Based only on local context, a stan- dard n-gram tagger is likely to assign the wrong tag DEG to „, which inevitably makes the fol- lowing parsing step fail to process the sentence correctly. However, knowing that the NP preced- ing „ is the object of Ÿ/VV (VV: verb), we can assume that „ is a complementizer because „/DEG is unlikely to follow a verb phrase. In this paper, we propose the first incremental approach to the task of joint POS tagging and de- pendency parsing. Given a segmented sentence, …

Content truncated to 500 words