An AMR-based Link Prediction Approach for Document-level Event Argument Extraction

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 12876–12889 July 9-14, 2023 ©2023 Association for Computational Linguistics An AMR-based Link Prediction Approach for Document-level Event Argument Extraction Yuqing Yang1∗†, Qipeng Guo2†, Xiangkun Hu2, Yue Zhang3, Xipeng Qiu1‡, Zheng Zhang2 1School of Computer Science, Fudan University 2Amazon AWS AI, 3School of Engineering, Westlake University yuqingyang21@m.fudan.edu.cn, {gqipeng, xiangkhu, zhaz}@amazon.com xpqiu@fudan.edu.cn, zhangyue@westlake.edu.cn Abstract Recent works have introduced Abstract Mean- ing Representation (AMR) for Document-level Event Argument Extraction (Doc-level EAE), since AMR provides a useful interpretation of complex semantic structures and helps to cap- ture long-distance dependency. However, in these works AMR is used only implicitly, for in- stance, as additional features or training signals. Motivated by the fact that all event structures can be inferred from AMR, this work refor- mulates EAE as a link prediction problem on AMR graphs. Since AMR is a generic structure and does not perfectly suit EAE, we propose a novel graph structure, Tailored AMR Graph (TAG), which compresses less informative subgraphs and edge types, integrates span information, and highlights surrounding events in the same document. With TAG, we further propose a novel method using graph neural networks as a link prediction model to find event arguments. Our extensive experiments on WikiEvents and RAMS show that this simpler approach outper- forms the state-of-the-art models by 3.63pt and 2.33pt F1, respectively, and do so with reduced 56% inference time. The code is available at https://github.com/ayyyq/TARA. 1 Introduction Event Argument Extraction (EAE) is a long- standing information extraction task to extract event structures composed of arguments from un- structured text (Xiang and Wang, 2019). Event structures can serve as an intermediate semantic representation and be further used for improving downstream tasks, including machine reading com- prehension (Han et al., 2021), question answer- ing (Costa et al., 2020), dialog system (Zhang Lab.∗Work done during internship at Amazon Shanghai AI †Equal contribution. ‡Corresponding author. Zabiullah Mujahid , the spokesperson for the Taliban , claimed they were responsible for the attack via Twitter and e - mail . Zabiullah Mujahid Taliban attack-01 responsible-01 spokesperson Twitter claim-01 Zabiullah Mujahid claim-01 Taliban claim-01 Communicator No link Figure 1: Treating EAE as a link prediction problem, where the green box means the event trigger and blue cir- cles are non-trigger nodes. The highlighted text means they are captured by the graph. We parse the document to a tailored AMR graph and apply a GNN model to pre- dict edges between the trigger and other nodes. In this example, the model predicts one argument, “Zabiullah Mujahid”, with the role of “Communicator”. et al., 2020), and recommendation system (Li et al., 2020). Despite the large performance boost by Pre-trained Language Models (PLMs), extracting complex event structures across sentences is still challenging (Ebner et al., 2020). In real-world text, event structures are usually distributed in multiple sentences (Li et al., 2021). To capture cross-sentence and multi-hop structures, Xu et al. (2022) introduces Abstract Meaning Rep- resentation (AMR) graphs to assist …

Content truncated to 500 words