A Two-Stream AMR-enhanced Model for Document-level Event Argument Extraction

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5025 - 5036 July 10-15, 2022 ©2022 Association for Computational Linguistics A Two-Stream AMR-enhanced Model for Document-level Event Argument Extraction Runxin Xu1∗, Peiyi Wang1∗, Tianyu Liu3, Shuang Zeng1,2, Baobao Chang1†, Zhifang Sui1 1Key Laboratory of Computational Linguistics, Peking University, MOE, China 2School of Software and Microelectronics, Peking University, China 3Tencent Cloud Xiaowei {runxinxu,wangpeiyi9979}@gmail.com, rogertyliu@tecent.com {zengs,chbb,szf}@pku.edu.cn Abstract Most previous studies aim at extracting events from a single sentence, while document-level event extraction still remains under-explored. In this paper, we focus on extracting event arguments from an entire document, which mainly faces two critical problems: a) the long- distance dependency between trigger and argu- ments over sentences; b) the distracting con- text towards an event in the document. To ad- dress these issues, we propose a Two-Stream Abstract meaning Representation enhanced ex- traction model (TSAR). TSAR encodes the doc- ument from different perspectives by a two- stream encoding module, to utilize local and global information and lower the impact of dis- tracting context. Besides, TSAR introduces an AMR-guided interaction module to capture both intra-sentential and inter-sentential fea- tures, based on the locally and globally con- structed AMR semantic graphs. An auxiliary boundary loss is introduced to enhance the boundary information for text spans explicitly. Extensive experiments illustrate that TSAR out- performs previous state-of-the-art by a large margin, with 2.54 F1 and 5.13 F1 performance gain on the public RAMS and WikiEvents datasets respectively, showing the superiority in the cross-sentence arguments extraction. We re- lease our code in https://github.com/ PKUnlp-icler/TSAR. 1 Introduction Event Argument Extraction (EAE) aims at identify- ing the entities that serve as event arguments, and predicting the roles they play in the event, which is a key step for Event Extraction (EE). It helps to transform the unstructured text into structured event knowledge that can be further utilized in rec- ommendation systems (Li et al., 2020), dialogue systems (Zhang et al., 2020a), and so on. Most previous studies assume that the events are only ∗Equal contribution. †Corresponding authors. Vehicle Transport artifact origin importer destination … [1] A ship carrying half a million barrels of oil that was pumped in the U.S. docked at a terminal owned by Venezuela last week, according to oil data research firm ClipperData. [2] The shipment was sent to a facility located on Dutch Island of Curacao in Caribbean. [3] The fact that Venezuela is import- ing American oil is raising eyebrows because Venezuela has 298 billion barrels of oil reserves, according to the Energy Information Administration. [4] That 's more than Saudi Arabia, Russia or Iran and eight times the reserves of the United States. Figure 1: A document from RAMS dataset (Ebner et al., 2020). A transport event is triggered by shipment, with five event arguments scattering across the document. expressed by a single sentence and hence focus on sentence-level extraction (Chen et al., 2015; Liu et al., 2018; Zhou et al., 2021). However, …

Content truncated to 500 words