Capturing the Content of a Document through Complex Event Identification

Proceedings of the 11th Joint Conference on Lexical and Computational Semantics, pages 331 - 340 July 14-15, 2022 ©2022 Association for Computational Linguistics Capturing the Content of a Document through Complex Event Identification Zheng Qi, Elior Sulem, Haoyu Wang, Xiaodong Yu, Dan Roth Department of Computer and Information Science, University of Pennsylvania {issacqzh, eliors, why16gzl, xdyu, danroth}@seas.upenn.edu Abstract Granular events, instantiated in a document by predicates, can usually be grouped into more general events, called complex events. Together, they capture the major content of the document. Recent work grouped granular events by defin- ing event regions, filtering out sentences that are irrelevant to the main content. However, this approach assumes that a given complex event is always described in consecutive sen- tences, which does not always hold in practice. In this paper, we introduce the task of complex event identification. We address this task as a pipeline, first predicting whether two granu- lar events mentioned in the text belong to the same complex event, independently of their po- sition in the text, and then using this to cluster them into complex events. Due to the difficulty of predicting whether two granular events be- long to the same complex event in isolation, we propose a context-augmented representa- tion learning approach CONTEXTRL that adds additional context to better model the pairwise relation between granular events. We show that our approach outperforms strong baselines on the complex event identification task and fur- ther present a promising case study exploring the effectiveness of using complex events as in- put for document-level argument extraction.1. 1 Introduction Event extraction aims to identify event predicates and arguments from text and then identify their types and roles respectively, helping humans to easily understand the events. It has attracted con- siderable interest in the last few years (Chen et al., 2015; Nguyen et al., 2016; Sha et al., 2018; Lin et al., 2020; Ebner et al., 2020; Chen et al., 2020b; Li et al., 2021) due to the vast amounts of unstruc- tured text available in domains like e-commerce, healthcare and industry. However, considering each 1The data and code are available at http://cogcomp. org/page/publication_view/978 Five protesters were (e2:killed) when they were (e3:attacked) by an armed group. The armed group (e5:attacked) the demonstrators who have for days been staging their (e7:protest) against the military government. Many protesters are supporters of an ultraconservative Islamist candidate in (e8:elections) who was expelled from the election (e10:race) when it was (e11:discovered) that his mother held dual Egyptian-U.S citizenship. The (e13:attack) on Wednesday (e14:wounded) at least 50 protesters, and the attackers (e15:used) stones, sticks and Molotov cocktails. e2: killed e3: attacked e5: attacked e7: protest e13: attack e14: wounded e15: used ce1 e8: elections e10: race e11: discovered ce2 Figure 1: An example of complex events (ce1 and ce2) described in the document. For clarity, not all event mentions are shown in the figure. granular event instantiated in the document by a predicate in isolation is not sufficient for under- standing the entire context of …

Content truncated to 500 words