Chinese Event Coreference Resolution: An Unsupervised Probabilistic Model Rivaling Supervised Resolvers

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1097–1107, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Chinese Event Coreference Resolution: An Unsupervised Probabilistic Model Rivaling Supervised Resolvers Chen Chen and Vincent Ng Human Language Technology Research Institute University of Texas at Dallas Richardson, TX 75083-0688 {yzcchen,vince}@hlt.utdallas.edu Abstract Recent work has successfully leveraged the semantic information extracted from lexi- cal knowledge bases such as WordNet and FrameNet to improve English event corefer- ence resolvers. The lack of comparable re- sources in other languages, however, has made the design of high-performance non-English event coreference resolvers, particularly those employing unsupervised models, very diffi- cult. We propose a generative model for the under-studied task of Chinese event corefer- ence resolution that rivals its supervised coun- terparts in performance when evaluated on the ACE 2005 corpus. 1 Introduction Event coreference resolution is the task of deter- mining which event mentions in a text refer to the same real-world event. Compared to entity corefer- ence, event coreference is not only much less stud- ied, but it is arguably more challenging. Recall that for two event mentions to be coreferent, both their triggers (i.e., the words realizing the occurrence of the events) and their corresponding arguments (e.g., the times, places, and people involved in them) have to be compatible. However, identifying potential ar- guments (which is typically performed by an entity extraction system), linking arguments to their event mentions (which is typically performed by an event extraction system), and determining the compatibil- ity between two event arguments (which is the job of an entity coreference resolver), are all non-trivial tasks. In other words, end-to-end event coreference resolution is complicated in part by the fact that an event coreference resolver has to rely on the noisy outputs produced by its upstream components in the standard information extraction (IE) pipeline. In this paper, we examine Chinese event coref- erence resolution. While English event coreference is under-investigated, Chinese event coreference is much less studied than English event coreference. In terms of task definition, there is no difference be- tween English and Chinese event coreference. How- ever, the design of high-performance Chinese event coreference resolvers is complicated in part by the lack of large-scale lexical knowledge bases. Re- cent work by Bejan and Harabagiu (2010; 2014) has shown that the semantic information extracted from WordNet (Fellbaum, 1998) and FrameNet (Baker et al., 1998) significantly contributed to the perfor- mance of their English event coreference resolver. While the lack of comparable lexical knowledge bases in Chinese can be mitigated in part by the use of event coreference annotated data, we focus on a challenging version of the task --- unsupervised Chi- nese event coreference resolution. Specifically, our goal is to learn an event coreference model without using data annotated with event coreference links. When evaluated on the Chinese portion of the ACE 2005 corpus, our unsupervised probabilistic model for event coreference resolution rivals its state-of- the-art supervised counterpart …

Content truncated to 500 words