Chinese Coreference Resolution via Ordered Filtering

Proceedings of the Joint Conference on EMNLP and CoNLL: Shared Task, pages 95–99, Jeju Island, Korea, July 13, 2012. c⃝2012 Association for Computational Linguistics Chinese Coreference Resolution via Ordered Filtering∗ Xiaotian Zhang1,2 Chunyang Wu1,2 Hai Zhao1,2† 1Center for Brain-Like Computing and Machine Intelligence, Department of Computer Science and Engineering, Shanghai Jiao Tong University 2MOE-Microsoft Key Laboratory for Intelligent Computing and Intelligent Systems Shanghai Jiao Tong University xtian.zh@gmail.com, chunyang506@sjtu.edu.cn, zhaohai@cs.sjtu.edu.cn Abstract We in this paper present the model for our participation (BCMI) in the CoNLL-2012 Shared Task. This paper describes a pure rule-based method, which assembles dif- ferent filters in a proper order. Different filters handle different situations and the filtering strategies are designed manually. These filters are assigned to different or- dered tiers from general to special cases. We participated in the Chinese and En- glish closed tracks, scored 51.83 and 59.24 respectively. 1 Introduction In this paper, we describes the approaches we u- tilized for our participation in the CoNLL-2012 Shared Task. This year’s shared task targets at modeling coreference resolution for multiple lan- guages. Following (Lee et al., 2011), we extend- s the methodology of deterministic coreference model, using manually designed rules to rec- ognize expressions with corresponding entities. The deterministic coreference model (Raghu- ∗ This work was partially supported by the Na- tional Natural Science Foundation of China (Grant No. 60903119 and Grant No. 61170114), the National Re- search Foundation for the Doctoral Program of Higher E- ducation of China under Grant No. 20110073120022, the National Basic Research Program of China (Grant No. 2009CB320901), the Science and Technology Commission of Shanghai Municipality (Grant No. 09511502400), and the European Union Seventh Framework Program (Grant No. 247619). † Corresponding author. nathan et al., 2010) has shown good perfor- mance in the shared task of CoNLL-2011. This kind of model focuses on filtering with ordered tiers: One filter is applied at one time, from highest to lowest precision. However, compared with learning approaches (Soon et al., 2001), s- ince effective rules are quite heterogeneous in different languages, several filtering methods should be redesigned when different languages are considered. We modified the original Stan- ford English coreference system1 to adapt to the Chinese scenario. For the English participation, we implemented the full strategies and interface of the semantic-based filters which are not ob- tained from the open source toolkit. The rest of this paper is organized as follows: In Section 2, we review the related work; In Sec- tion 3, we describe the detail of our model of handling coreference resolution in Chinese; Ex- periment results are reported in Section 4 and the conclusion is presented in Section 5. 2 Related Work Many existing works have been published on learning relation extractors via supervised (Soon et al., 2001) or unsupervised (Haghighi and K- lein, 2010; Poon and Domingos, 2008) approach- es. For involving semantics, (Rahman and Ng, 2011) proposed a coreference resolution model with world knowledge; By using word associa- tions, (Kobdani et al., 2011) showed its effec- tiveness to coreference …

Content truncated to 500 words