Multi-Lingual Coreference Resolution With Syntactic Features

Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP), pages 660–667, Vancouver, October 2005. c⃝2005 Association for Computational Linguistics Multi-Lingual Coreference Resolution With Syntactic Features Xiaoqiang Luo and Imed Zitouni 1101 Kitchawan Road IBM T.J. Watson Research Center Yorktown Heights, NY 10598, U.S.A. {xiaoluo, izitouni}@us.ibm.com Abstract In this paper, we study the impact of a group of features extracted automatically from machine-generated parse trees on coreference resolution. One focus is on designing syn- tactic features using the binding theory as the guideline to improve pronoun resolution, al- though linguistic phenomenon such as apposi- tion is also modeled. These features are ap- plied to the Arabic, Chinese and English coref- erence resolution systems and their effective- ness is evaluated on data from the Automatic Content Extraction (ACE) task. The syntactic features improve the Arabic and English sys- tems significantly, but play a limited role in the Chinese one. Detailed analyses are done to un- derstand the syntactic features’ impact on the three coreference systems. 1 Introduction A coreference resolution system aims to group together mentions referring to the same entity, where a mention is an instance of reference to an object, and the collection of mentions referring to the same object in a document form an entity. In the following example: (I) “John believes himself to be the best student.” mentions are underlined. The three mentions “John”, “himself”, “the best student” are of type name, pronoun 1, and nominal, respectively. They form an entity since they all refer to the same person. Syntactic information plays an important role in corefer- ence resolution. For example, the binding theory (Haege- man, 1994; Beatrice and Kroch, 2000) provides a good account of the constraints on the antecedent of English pronouns. The theory relies on syntactic parse trees to de- termine the governing category which defines the scope 1“Pronoun” in this paper refers to both anaphor and normal pronoun. of binding constraints. We will use the theory as a guide- line to help us design features in a machine learning framework. Previous pronoun resolution work (Hobbs, 1976; Lappin and Leass, 1994; Ge et al., 1998; Stuckardt, 2001) explic- itly utilized syntactic information before. But there are unique challenges in this study: (1) Syntactic informa- tion is extracted from parse trees automatically generated. This is possible because of the availability of statistical parsers, which can be trained on human-annotated tree- banks (Marcus et al., 1993; Xia et al., 2000; Maamouri and Bies, 2004) for multiple languages; (2) The bind- ing theory is used as a guideline and syntactic structures are encoded as features in a maximum entropy corefer- ence system; (3) The syntactic features are evaluated on three languages: Arabic, Chinese and English (one goal is to see if features motivated by the English language can help coreference resolution in other languages). All con- trastive experiments are done on publicly-available data; (4) Our coreference system resolves coreferential rela- tionships among all the annotated mentions, not just for pronouns. Using …

Content truncated to 500 words