Document-Level Event Argument Extraction by Leveraging Redundant Information and Closed Boundary Loss

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3041 - 3052 July 10-15, 2022 ©2022 Association for Computational Linguistics Document-Level Event Argument Extraction by Leveraging Redundant Information and Closed Boundary Loss Hanzhang Zhou1,3, Kezhi Mao2,3∗ 1Institute of Catastrophe Risk Management, Interdisciplinary Graduate Programme, Nanyang Technological University, Singapore 2School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 3Future Resilient Systems Programme, Singapore-ETH Centre, CREATE campus, Singapore hanzhang001@e.ntu.edu.sg ekzmao@ntu.edu.sg Abstract In document-level event argument extraction, an argument is likely to appear multiple times in different expressions in the document. The redundancy of arguments underlying multiple sentences is beneficial but is often overlooked. In addition, in event argument extraction, most entities are regarded as class “others", i.e. Uni- versum class, which is defined as a collection of samples that do not belong to any class of inter- est. Universum class is composed of heteroge- neous entities without typical common features. Classifiers trained by cross entropy loss could easily misclassify the Universum class because of their open decision boundary. In this paper, to make use of redundant event information un- derlying a document, we build an entity corefer- ence graph with the graph2token module to pro- duce a comprehensive and coreference-aware representation for every entity and then build an entity summary graph to merge the multiple extraction results. To better classify Universum class, we propose a new loss function to build classifiers with closed boundaries. Experimen- tal results show that our model outperforms the previous state-of-the-art models by 3.35% in F1-score. 1 Introduction Event argument extraction (EAE) is a crucial sub- task of event extraction (EE), aiming to identify the arguments of a given event and recognize the specific roles they play. Previous works are mostly focused on sentence-level EE (Liao and Grishman, 2010; Nguyen et al., 2016; Liu et al., 2018; Yang et al., 2019b; Du and Cardie, 2020b; Wei et al., 2021; Wang et al., 2021; Lyu et al., 2021). How- ever, events are often described in the form of doc- uments in the real world. Document-level event ex- traction has received consideration in recent years. Research on document-level event extraction has been focused on tackling challenges such as *Corresponding author. Figure 1: An example of redundant event information in the document-level event argument extraction. arguments-scattering and multiple-events (Zheng et al., 2019; Du and Cardie, 2020a; Du et al., 2021; Lou et al., 2021; Li et al., 2021; Huang and Peng, 2021; Xu et al., 2021; Yang et al., 2021; Ahmad et al., 2021; Ebner et al., 2020). The benefit of re- dundant event information in a document is largely neglected. We believe that the redundant event in- formation in a document can be used to improve event extraction, as illustrated in the example in Figure 1. The upper part of Figure 1 shows seven simplified sentences selected from a document in the MUC-4 dataset. All entities marked in blue are the same entity "soldiers", which appears in differ- …

Content truncated to 500 words