Event Causality Extraction with Event Argument Correlations

Proceedings of the 29th International Conference on Computational Linguistics, pages 2300–2312 October 12–17, 2022. 2300 Event Causality Extraction with Event Argument Correlations Shiyao Cui1,2 Jiawei Sheng1,2 Xin Cong1,2 QuanGang Li1,2∗Tingwen Liu1,2 Jinqiao Shi3,1 1Institute of Information Engineering, Chinese Academy of Sciences. Beijing, China 2School of Cyber Security, University of Chinese Academy of Sciences. Beijing, China 3Beijing University of Posts and Telecommunications. Beijing, China {cuishiyao, shengjiawei, congxin, liquangang}@iie.ac.cn liutingwen@iie.ac.cn shijinqiao@bupt.edu.cn Abstract Event Causality Identification (ECI), which aims to detect whether a causality relation ex- ists between two given textual events, is an im- portant task for event causality understanding. However, the ECI task ignores crucial event structure and cause-effect causality component information, making it struggle for downstream applications. In this paper, we explore a novel task, namely Event Causality Extraction (ECE), aiming to extract the cause-effect event causal- ity pairs with their structured event informa- tion from plain texts. The ECE task is more challenging since each event can contain mul- tiple event arguments, posing fine-grained cor- relations between events to decide the cause- effect event pair. Hence, we propose a method with a dual grid tagging scheme to capture the intra- and inter-event argument correlations for ECE. Further, we devise a event type-enhanced model architecture to realize the dual grid tag- ging scheme. Experiments demonstrate the ef- fectiveness of our method, and extensive analy- ses point out several future directions for ECE. 1 Introduction Event causality (Liu et al., 2020; Cao et al., 2021) denotes an explicit causal relation between two events, constituting a specific cause-effect event pair. As shown in Figure 1, a causal relation exists between a Price Rise event (The worldwide rise of oil prices) and a Cost Rise event (in- creases the cost of international shipping industry). Understanding such event causality could facilitate various downstream applications including event forecasting (Hashimoto et al., 2014), intelligent search (Rudnik et al., 2019) and question answer- ing (Costa et al., 2020), which is important for natural language understanding. In recent years, it has aroused the research in- terest for Event Causality Identification (ECI) (Liu et al., 2020; Cao et al., 2021; Zuo et al., 2021a,b, ∗Corresponding Author The worldwide rise of oil prices increases the cost of international shipping industry and stimulates the demand for new energy such as Ammonia fuel. No. Event Type Event Roles Product Region Industry (1) Cause Price Rise oil worldwide None Effect Cost Rise None international shipping industry (2) Cause Price Rise oil worldwide None Effect Demand Rise new energy None Ammonia fuel Causality Component Figure 1: Illustration for ECE which takes the raw text as input, and outputs the structured event causality pair. 2020; Tran Phu and Nguyen, 2021), which aims to detect whether the causality exists between two given events. Despite of its success, there exist two issues that the ECI task fails to address. 1) Event Structure Missing, where each event in ECI is only expressed using a word or phrase which re- flects its occurrence, but ignores the explicit event type …

Content truncated to 500 words