Document-Level Event Argument Extraction via Optimal Transport

Findings of the Association for Computational Linguistics: ACL 2022, pages 1648 - 1658 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Document-Level Event Argument Extraction via Optimal Transport Amir Pouran Ben Veyseh1, Minh Van Nguyen1, Franck Dernoncourt2, Bonan Min3, Thien Huu Nguyen1 1Department of Computer and Information Science, University of Oregon, Eugene, OR, USA 2Adobe Research, San Jose, CA, USA 3 Raytheon BBN Technologies, USA {apouranb,minhnv,thien}@cs.uoregon.edu franck.dernoncourt@adobe.com, bonan.min@raytheon.com Abstract Event Argument Extraction (EAE) is one of the sub-tasks of event extraction, aiming to recognize the role of each entity mention to- ward a specific event trigger. Despite the suc- cess of prior works in sentence-level EAE, the document-level setting is less explored. In particular, whereas syntactic structures of sentences have been shown to be effective for sentence-level EAE, prior document-level EAE models totally ignore syntactic structures for documents. Hence, in this work, we study the importance of syntactic structures in document-level EAE. Specifically, we propose to employ Optimal Transport (OT) to induce structures of documents based on sentence- level syntactic structures and tailored to EAE task. Furthermore, we propose a novel regu- larization technique to explicitly constrain the contributions of unrelated context words in the final prediction for EAE. We perform exten- sive experiments on the benchmark document- level EAE dataset RAMS that leads to the state-of-the-art performance. Moreover, our experiments on the ACE 2005 dataset reveals the effectiveness of the proposed model in the sentence-level EAE by establishing new state- of-the-art results. 1 Introduction Event Extraction (EE) is one of the important sub- tasks of Information Extraction (IE). The major goal of EE is to identify events and their engaged entities (i.e., event arguments). To this end, two sub-tasks should be solved: 1) Event Detection (ED): To recognize event triggers (i.e., the words or phrases that clearly specify the occurrence of events) and their event types, 2) Event Argument Extraction (EAE): To identify participants of events (i.e., entities engaged in events) and their argument roles. Compared to ED, the EAE sub-task is rela- tively less explored in IE. Moreover, prior works on EAE are mainly restricted to sentence-level setting where event triggers and arguments are assumed to appear in the same sentences. This is unfortu- nate as a considerable portion of event arguments might not be immediately mentioned in the same sentence as their event triggers. For instance, in the EE dataset of the DARPA AIDA program (phase 1) extraction1, 38% of arguments has been shown to be outside sentences containing the corresponding triggers, i.e., in the document-level context (Ebner et al., 2020). Thus, further research to study the more realistic setting of document-level EAE is extremely needed. To the best of our knowledge, there are two prior works on document-level EAE (Ebner et al., 2020; Chen et al., 2020). Both works employ represen- tation vectors for argument spans obtained from a transformer to compute likelihood scores for ar- gument roles. Unfortunately, those prior works only exploit the sequential order of words in docu- ments to represent the arguments …

Content truncated to 500 words