Resolving Event Coreference with Supervised Representation Learning and Clustering-Oriented Regularization

Proceedings of the 7th Joint Conference on Lexical and Computational Semantics (*SEM), pages 1–10 New Orleans, June 5-6, 2018. c⃝2018 Association for Computational Linguistics Resolving Event Coreference with Supervised Representation Learning and Clustering-Oriented Regularization Kian Kenyon-Dean School of Computer Science McGill University kian.kenyon-dean @mail.mcgill.ca Jackie Chi Kit Cheung School of Computer Science McGill University jcheung@cs.mcgill.ca Doina Precup School of Computer Science McGill University dprecup@cs.mcgill.ca Abstract We present an approach to event coreference resolution by developing a general framework for clustering that uses supervised representa- tion learning. We propose a neural network architecture with novel Clustering-Oriented Regularization (CORE) terms in the objective function. These terms encourage the model to create embeddings of event mentions that are amenable to clustering. We then use agglom- erative clustering on these embeddings to build event coreference chains. For both within- and cross-document coreference on the ECB+ corpus, our model obtains better results than models that require significantly more pre- annotated information. This work provides in- sight and motivating results for a new general approach to solving coreference and clustering problems with representation learning. 1 Introduction Event coreference resolution is the task of deter- mining which event mentions expressed in lan- guage refer to the same real-world event in- stances. The ability to resolve event coreference has improved the quality of downstream tasks such as automatic text summarization (Vander- wende et al., 2004), questioning-answering (Be- rant et al., 2014), headline generation (Sun et al., 2015), and text-mining in the medical domain (Ferracane et al., 2016). Event mentions are comprised of an action component (or, head) and surrounding arguments. Consider the following passages, drawn from two different documents; the heads of the event men- tions are in boldface and the subscripts indicate mention IDs: (1) The president’s speechm1 shockedm2 the au- dience. He announcedm3 several new con- troversial policies. (2) The policies proposedm4 by the president will not surprisem5 those who followedm6 his campaignm7. In this example, m1, m3, and m4 form a chain of coreferent event mentions (underlined), because they refer to the same real-world event in which the president gave a speech. The other four are singletons, meaning that they all refer to separate events and do not corefer with any other mention. This work investigates how to learn useful rep- resentations of event mentions. Event mentions are complex objects, and both the event mention heads and the surrounding arguments are impor- tant for the event coreference resolution task. In our example above, the head words of mentions m2, shocked, and m5, surprise, are lexically sim- ilar, but the event mentions do not corefer. This task therefore necessitates a model that can cap- ture the distributional relationships between event mentions and their surrounding contexts. We hypothesize that prior knowledge about the task itself can be usefully encoded into the rep- resentation learning objective. For our task, this prior means that the embeddings of corefential event mentions should have similar embeddings to each other (a “natural clustering”, using the termi- nology of Bengio et al. (2013)). …

Content truncated to 500 words