Event Coreference Resolution with their Paraphrases and Argument-aware Embeddings

Proceedings of the 28th International Conference on Computational Linguistics, pages 3084–3094 Barcelona, Spain (Online), December 8-13, 2020 3084 Event Coreference Resolution with their Paraphrases and Argument-aware Embeddings Yutao Zeng, Xiaolong Jin, Saiping Guan, Jiafeng Guo, and Xueqi Cheng School of Computer Science and Technology, University of Chinese Academy of Sciences; CAS Key Laboratory of Network Data Science and Technology, Institute of Computing Technology, Chinese Academy of Sciences {zengyutao18s,jinxiaolong,guansaiping,guojiafeng,cxq}@ict.ac.cn Abstract Event coreference resolution aims to classify all event mentions that refer to the same real-world event into the same group, which is necessary to information aggregation and many downstream applications. To resolve event coreference, existing methods usually calculate the similarities between event mentions and between specific kinds of event arguments. However, they fail to accurately identify paraphrase relations between events and may suffer from error propagation while extracting event components (i.e., event mentions and their arguments). Therefore, we pro- pose a new model based on Event-specific Paraphrases and Argument-aware Semantic Embed- dings, thus called EPASE, for event coreference resolution. EPASE recognizes deep paraphrase relations in an event-specific context of sentences and can cover event paraphrases of more situa- tions, bringing about a better generalization. Additionally, the embeddings of argument roles are encoded into event embedding without relying on a fixed number and type of arguments, which results in the better scalability of EPASE. Experiments on both within- and cross-document event coreference demonstrate its consistent and significant superiority compared to existing methods. 1 Introduction Event coreference resolution clusters event mentions referring to the same real-world event, no matter within a single document (denoted as WD) or across multiple documents (denoted as CD). It is vital for information aggregation and can further benefit many downstream natural language processing applica- tions, including contradiction detection (De Marneffe et al., 2008), text mining (Ferracane et al., 2016) and question answering (Khashabi et al., 2018; Welbl et al., 2018). Usually, each event consists of an event mention and a few arguments. Take the following two sen- tences as an example, where event mentions are marked in bold, and the subscripts indicates their IDs: (1) Perennial party girl Tara Reid checkedm1 herself intom1 Promises Treatment Center, her repre- sentative toldm2 The Washington Post. (2) The original trainwreck Tara Reid’s publicist confirmedm3 that the actress Reid was admitted intom4 Promises Treatment Center in Los Angeles and it was her decisionm5. these five event mentions in the sentences can be clustered into two sets {m1,m4, m5} and {m2, m3}. Each set constitutes an event coreference chain and the event mentions in it are all coreferential. An intuitive and effective way to resolve event coreference is to calculate the similarities between event mentions and between their arguments. Many methods (Yang et al., 2015; Choubey and Huang, 2017; Barhom et al., 2019) have adopted these similarities as important features to train classifiers. Essentially, this means is to identify the paraphrase relation between events (i.e., one event can be viewed as a paraphrase of another event) by modeling the similarity between event components …

Content truncated to 500 words