Document-level Event Factuality Identification via Machine Reading Comprehension Frameworks with Transfer Learning

Proceedings of the 29th International Conference on Computational Linguistics, pages 2622–2632 October 12–17, 2022. 2622 Document-level Event Factuality Identification via Machine Reading Comprehension Frameworks with Transfer Learning Zhong Qian1∗, Heng Zhang1, Peifeng Li1,2, Qiaoming Zhu1,2 and Guodong Zhou1,2 1School of Computer Science and Technology, Soochow University, Suzhou, China 2AI Research Institute, Soochow University, Suzhou, China qianzhong@suda.edu.cn, 20204227055@stu.suda.edu.cn, {pfli, qmzhu, gdzhou}@suda.edu.cn Abstract Document-level Event Factuality Identification (DEFI) predicts the factuality of a specific event based on a document from which the event can be derived, which is a fundamental and crucial task in Natural Language Processing (NLP). However, most previous studies only consid- ered sentence-level task and did not adopt document-level knowledge. Moreover, they modelled DEFI as a typical text classification task depending on annotated information heav- ily, and limited to the task-specific corpus only, which resulted in data scarcity. To tackle these issues, we propose a new framework formulat- ing DEFI as Machine Reading Comprehension (MRC) tasks considering both Span-Extraction (Ext) and Multiple-Choice (Mch). Our model does not employ any other explicit annotated in- formation, and utilizes Transfer Learning (TL) to extract knowledge from universal large-scale MRC corpora for cross-domain data augmenta- tion. The empirical results on DLEFM corpus demonstrate that the proposed model outper- forms several state-of-the-arts. 1 Introduction This paper focuses on Document-level Event Factu- ality Identification (DEFI) task, which is defined as identifying the factuality of a specific event based on a document from which the event is derived. As a sub-task in Event Factuality Identification (EFI), different from Sentence-level Event Factu- ality Identification (SEFI) focusing on just a sin- gle sentence, DEFI requires comprehensive under- standing documents with regard to events. Figure 1 illustrates an example of document- level event factuality. The current event is E1, i.e., “Barack Obama joins Joe Biden’s cabinet”, and the sentences S2-S6 contain the event men- tions referring to E1. From Figure 1 we can know that: 1) Factuality of event mentions vary among sentences. S4, S5 and S6 negate E1 by the nega- tive cues “not” and “denied”, and commit to E1 as Event (E1): Barack Obama joins Joe Biden’s cabinet. Document: (S1) “I will help him in any ways that I can," Obama, said of his former vice president, in a new interview with CBS Sunday Morning’s Gayle King. (S2) With the victory of former US Vice President Biden in the presidential election, there are speculations about whether Obama will return to the White House and serve in Biden’s cabinet. (S3) King also asked if he would join Biden’s cabinet. (S4) “I’m not planning to suddenly work on the White House staff or some- thing,” said Obama. (S5) He also jokingly responded, "There are probably some things I would not be doing, because Michelle would leave me. ... ...”(S6) Although Obama denied that he would take a position in Biden’s cabinet, Susan Rice and Michèle Flournoy were among Obama administration veterans reportedly being prob- ably considered for key posts under Biden. ... ... (S7) When Barack Obama was elected …

Content truncated to 500 words