WEC: Deriving a Large-scale Cross-document Event Coreference dataset from Wikipedia

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2498–2510 June 6–11, 2021. ©2021 Association for Computational Linguistics 2498 WEC: Deriving a Large-scale Cross-document Event Coreference dataset from Wikipedia Alon Eirew1,2 Arie Cattan1 Ido Dagan1 1Bar Ilan University, Ramat-Gan, Israel 2Intel Labs, Israel alon.eirew@intel.com arie.cattan@gmail.com dagan@cs.biu.ac.il Abstract Cross-document event coreference resolution is a foundational task for NLP applications in- volving multi-text processing. However, ex- isting corpora for this task are scarce and rel- atively small, while annotating only modest- size clusters of documents belonging to the same topic. To complement these resources and enhance future research, we present Wikipedia Event Coreference (WEC), an ef- ficient methodology for gathering a large- scale dataset for cross-document event coref- erence from Wikipedia, where coreference links are not restricted within predefined top- ics. We apply this methodology to the English Wikipedia and extract our large-scale WEC- Eng dataset. Notably, our dataset creation method is generic and can be applied with relatively little effort to other Wikipedia lan- guages. To set baseline results, we develop an algorithm that adapts components of state- of-the-art models for within-document coref- erence resolution to the cross-document set- ting. Our model is suitably efficient and out- performs previously published state-of-the-art results for the task. 1 Introduction Cross-Document (CD) Event Coreference resolu- tion is the task of identifying clusters of text men- tions, across multiple texts, that refer to the same event. Successful identification of such corefer- ring mentions is beneficial for a broad range of applications at the multi-text level, which are gain- ing increasing interest and need to match and inte- grate information across documents, such as multi- document summarization (Falke et al., 2017; Liao et al., 2018), multi-hop question answering (Dhin- gra et al., 2018; Wang et al., 2019) and Knowledge Base Population (KBP) (Lin et al., 2020). Unfortunately, rather few datasets of reason- able scale exist for CD event coreference. No- table datasets include ECB+ (Cybulska and Vossen, 2014), MEANTIME (Minard et al., 2016) and the Gun Violence Corpus (GVC) (Vossen et al., 2018) (described in Section 2), where recent work has been evaluated solely on ECB+. When addressed in a direct manner, manual CD coreference anno- tation is very hard due to its worst-case quadratic complexity, where each mention may need to be compared to all other mentions in all documents. Indeed, ECB+ contains less than 7000 event men- tions in total (train, dev, and test sets). Further, ef- fective corpora for CD event coreference are avail- able mostly for English, limiting research oppor- tunities for other languages. Partly as a result of this data scarcity, rather little effort was invested in this field in recent years, compared to dramatic recent progress in modeling within-document coref- erence. Furthermore, most existing cross-document coreference datasets are restricted in their scope by two inter-related characteristics. First, these datasets annotate sets of documents, where the documents in each set all describe the same topic, mostly a news event (consider …

Content truncated to 500 words