Graph-based Semi-Supervised Model for Joint Chinese Word Segmentation and Part-of-Speech Tagging

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 770–779, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Graph-based Semi-Supervised Model for Joint Chinese Word Segmentation and Part-of-Speech Tagging Xiaodong Zeng† Derek F. Wong† Lidia S. Chao† Isabel Trancoso‡ †Department of Computer and Information Science, University of Macau ‡INESC-ID / Instituto Superior T´ecnico, Lisboa, Portugal nlp2ct.samuel@gmail.com, {derekfw, lidiasc}@umac.mo, isabel.trancoso@inesc-id.pt Abstract This paper introduces a graph-based semi- supervised joint model of Chinese word segmentation and part-of-speech tagging. The proposed approach is based on a graph-based label propagation technique. One constructs a nearest-neighbor simi- larity graph over all trigrams of labeled and unlabeled data for propagating syn- tactic information, i.e., label distribution- s. The derived label distributions are re- garded as virtual evidences to regular- ize the learning of linear conditional ran- dom fields (CRFs) on unlabeled data. An inductive character-based joint model is obtained eventually. Empirical results on Chinese tree bank (CTB-7) and Microsoft Research corpora (MSR) reveal that the proposed model can yield better result- s than the supervised baselines and other competitive semi-supervised CRFs in this task. 1 Introduction Word segmentation and part-of-speech (POS) tag- ging are two critical and necessary initial proce- dures with respect to the majority of high-level Chinese language processing tasks such as syn- tax parsing, information extraction and machine translation. The traditional way of segmentation and tagging is performed in a pipeline approach, first segmenting a sentence into words, and then assigning each word a POS tag. The pipeline ap- proach is very simple to implement, but frequently causes error propagation, given that wrong seg- mentations in the earlier stage harm the subse- quent POS tagging (Ng and Low, 2004). The join- t approaches of word segmentation and POS tag- ging (joint S&T) are proposed to resolve these t- wo tasks simultaneously. They effectively allevi- ate the error propagation, because segmentation and tagging have strong interaction, given that most segmentation ambiguities cannot be resolved without considering the surrounding grammatical constructions encoded in a POS sequence (Qian and Liu, 2012). In the past years, several proposed supervised joint models (Ng and Low, 2004; Zhang and Clark, 2008; Jiang et al., 2009; Zhang and Clark, 2010) achieved reasonably accurate results, but the outstanding problem among these models is that they rely heavily on a large amount of labeled data, i.e., segmented texts with POS tags. However, the production of such labeled data is extremely time- consuming and expensive (Jiao et al., 2006; Jiang et al., 2009). Therefore, semi-supervised join- t S&T appears to be a natural solution for easily in- corporating accessible unlabeled data to improve the joint S&T model. This study focuses on using a graph-based label propagation method to build a semi-supervised joint S&T model. Graph-based label propagation methods have recently shown they can outperform the state-of-the-art in sever- al natural language processing (NLP) tasks, e.g., POS tagging (Subramanya et al., 2010), knowl- edge acquisition (Talukdar et al., 2008), shallow semantic parsing for unknown predicate (Das and Smith, …

Content truncated to 500 words