A Framework Based on Graphical Models with Logic for Chinese Named Entity Recognition

A Framework Based on Graphical Models with Logic for Chinese Named Entity Recognition ∗ Xiaofeng YU Wai LAM Shing-Kit CHAN Information Systems Laboratory Department of Systems Engineering & Engineering Management The Chinese University of Hong Kong Shatin, N.T., Hong Kong {xfyu,wlam,skchan}@se.cuhk.edu.hk Abstract Chinese named entity recognition (NER) has re- cently been viewed as a classification or sequence labeling problem, and many approaches have been proposed. However, they tend to address this problem without considering linguistic informa- tion in Chinese NEs. We propose a new framework based on probabilistic graphical models with first- order logic for Chinese NER. First, we use Condi- tional Random Fields (CRFs), a standard and the- oretically well-founded machine learning method based on undirected graphical models as a base system. Second, we introduce various types of domain knowledge into Markov Logic Networks (MLNs), an effective combination of first-order logic and probabilistic graphical models for vali- dation and error correction of entities. Experimen- tal results show that our framework of probabilis- tic graphical models with first-order logic signifi- cantly outperforms the state-of-the-art models for solving this task. 1 Introduction Named entity recognition (NER) is the task of identifying and classifying phrases that denote certain types of named entities (NEs), such as person names (PERs), locations (LOCs) and organizations (ORGs) in text documents. It is a well-established task in the NLP and data mining com- munities and is regarded as crucial technology for many higher-level applications, such as information extraction, question answering, information retrieval and knowledge management. The NER problem has generated much in- terest and great progress has been made, as evidenced by its inclusion as an understanding task to be evaluated in the ∗The work described in this paper is substantially supported by grants from the Research Grant Council of the Hong Kong Special Administrative Region, China (Project Nos: CUHK 4179/03E and CUHK4193/04E) and the Direct Grant of the Faculty of En- gineering, CUHK (Project Codes: 2050363 and 2050391). This work is also affiliated with the Microsoft-CUHK Joint Laboratory for Human-centric Computing and Interface Technologies. Message Understanding Conference (MUC), the Multilin- gual Entity Task (MET) evaluations, and the Conference on Computational Natural Language Learning (CoNLL). Compared to European-language NER, Chinese NER seems to be more difficult (Yu et al., 2006). Recent ap- proaches to Chinese NER are a shift away from manu- ally constructed rules or finite state patterns towards ma- chine learning or statistical methods. However, rule- based NER systems lack robustness and portability. Sta- tistical methods often suffer from the problem of data sparsity, and machine learning approaches (e.g., Hidden Markov Models (HMMs) (Bikel et al., 1999; Zhou and Su, 2002), Support Vector Machines (SVMs) (Isozaki and Kazawa, 2002), Maximum Entropy (MaxEnt) (Borthwick, 1999; Chieu and Ng, 2003), Transformation-based Learn- ing (TBL) (Brill, 1995) or variants of them) might be un- satisfactory to learn linguistic information in Chinese NEs. Current state-of-the-art models often view Chinese NER as a classification or sequence labeling problem without con- sidering the linguistic and structural information in Chinese NEs. They …

Content truncated to 500 words