Coreference Resolution with World Knowledge

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 814–824, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Coreference Resolution with World Knowledge Altaf Rahman and Vincent Ng Human Language Technology Research Institute University of Texas at Dallas Richardson, TX 75083-0688 {altaf,vince}@hlt.utdallas.edu Abstract While world knowledge has been shown to improve learning-based coreference resolvers, the improvements were typically obtained by incorporating world knowledge into a fairly weak baseline resolver. Hence, it is not clear whether these benefits can carry over to a stronger baseline. Moreover, since there has been no attempt to apply different sources of world knowledge in combination to corefer- ence resolution, it is not clear whether they of- fer complementary benefits to a resolver. We systematically compare commonly-used and under-investigated sources of world knowl- edge for coreference resolution by applying them to two learning-based coreference mod- els and evaluating them on documents anno- tated with two different annotation schemes. 1 Introduction Noun phrase (NP) coreference resolution is the task of determining which NPs in a text or dialogue refer to the same real-world entity. The difficulty of the task stems in part from its reliance on world knowl- edge (Charniak, 1972). To exemplify, consider the following text fragment. Martha Stewart is hoping people don’t run out on her. The celebrity indicted on charges stemming from . . . Having the (world) knowledge that Martha Stewart is a celebrity would be helpful for establishing the coreference relation between the two NPs. One may argue that employing heuristics such as subject pref- erence or syntactic parallelism (which prefers re- solving an NP to a candidate antecedent that has the same grammatical role) in this example would also allow us to correctly resolve the celebrity (Mitkov, 2002), thereby obviating the need for world knowl- edge. However, since these heuristics are not per- fect, complementing them with world knowledge would be an important step towards bringing coref- erence systems to the next level of performance. Despite the usefulness of world knowledge for coreference resolution, early learning-based coref- erence resolvers have relied mostly on morpho- syntactic features (e.g., Soon et al. (2001), Ng and Cardie (2002), Yang et al. (2003)). With recent ad- vances in lexical semantics research and the devel- opment of large-scale knowledge bases, researchers have begun to employ world knowledge for corefer- ence resolution. World knowledge is extracted pri- marily from three data sources, web-based encyclo- pedia (e.g., Ponzetto and Strube (2006), Uryupina et al. (2011)), unannotated data (e.g., Daum´e III and Marcu (2005), Ng (2007)), and coreference- annotated data (e.g., Bengtson and Roth (2008)). While each of these three sources of world knowl- edge has been shown to improve coreference resolu- tion, the improvements were typically obtained by incorporating world knowledge (as features) into a baseline resolver composed of a rather weak coref- erence model (i.e., the mention-pair model) and a small set of features (i.e., the 12 features adopted by Soon et al.’s (2001) knowledge-lean approach). As a result, some questions …

Content truncated to 500 words