Document-level Event Extraction via Parallel Prediction Networks

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 6298–6308 August 1–6, 2021. ©2021 Association for Computational Linguistics 6298 Document-level Event Extraction via Parallel Prediction Networks Hang Yang1,2, Dianbo Sui1,2, Yubo Chen1,2, Kang Liu1,2, Jun Zhao1,2, Taifeng Wang3 1National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China 2School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, 100049, China 3Ant Group, Hangzhou, 310013, China {hang.yang, dianbo.sui, yubo.chen, kliu, jzhao}@nlpr.ia.ac.cn , taifeng.wang@antgroup.com Abstract Document-level event extraction (DEE) is indispensable when events are described throughout a document. We argue that sentence-level extractors are ill-suited to the DEE task where event arguments always scat- ter across sentences and multiple events may co-exist in a document. It is a challenging task because it requires a holistic understanding of the document and an aggregated ability to as- semble arguments across multiple sentences. In this paper, we propose an end-to-end model, which can extract structured events from a document in a parallel manner. Specifically, we first introduce a document-level encoder to obtain the document-aware representations. Then, a multi-granularity non-autoregressive decoder is used to generate events in parallel. Finally, to train the entire model, a matching loss function is proposed, which can bootstrap a global optimization. The empirical results on the widely used DEE dataset show that our ap- proach significantly outperforms current state- of-the-art methods in the challenging DEE task. Code will be available at https:// github.com/HangYang-NLP/DE-PPN. 1 Introduction The goal of event extraction (EE) is to identify events of a pre-specified type along with corre- sponding arguments from plain texts. A great num- ber of previous studies (Ahn, 2006; Ji and Grish- man, 2008; Liao and Grishman, 2010; Hong et al., 2011; Li et al., 2013; Chen et al., 2015; Nguyen et al., 2016; Yang and Mitchell, 2016; Chen et al., 2017; Huang et al., 2018; Yang et al., 2019; Liu et al., 2020) focus on the sentence-level EE (SEE), while most of these works are based on the ACE evaluation (Doddington et al., 2004). 1 However, these SEE-based methods make predictions within 1https://www.ldc.upenn.edu/ collaborations/past-projects/ace [S3] On November 1, 2018, Shenzhen 007 Co., Ltd. received a notice that the corporate shareholder Shanghai Fukong Co., Ltd and the actual controller Jing Yan were judicial frozen. [S7] The corporate shareholder holds 150000 shares of the company. The 10000 shares were frozen by the Shenzhen Intermediate Peoples Court from October 30, 2018 to October 30, 2019. [S8] The controller of the company holds 310000 shares of the company. The 20000 shares were frozen by the Shenzhen Inter - mediate People's Court on November 1, 2018. Shanghai Fukong Co., Ltd Jing Yan 10000 shares 20000 shares Shenzhen Intermediate People's Court October 30, 2018 October 30, 2019 November 1, 2018 \ Figure 1: An example of a document contains two Eq- uity Freeze type events: Event-1 and Event-2. Words in bold-faced are arguments that scatter across multiple sentences. a …

Content truncated to 500 words