Event Causality Identification via Generation of Important Context Words

Proceedings of the 11th Joint Conference on Lexical and Computational Semantics, pages 323 - 330 July 14-15, 2022 ©2022 Association for Computational Linguistics Event Causality Identification via Generation of Important Context Words Hieu Man1, Minh Van Nguyen2, and Thien Huu Nguyen2 1 VinAI Research, Vietnam 2 Dept. of Computer and Information Science, University of Oregon, Eugene, OR, USA v.hieumdt@vinai.io, {minhnv,thien}@cs.uoregon.edu Abstract An important problem of Information Extrac- tion involves Event Causality Identification (ECI) that seeks to identify causal relation be- tween pairs of event mentions. Prior mod- els for ECI have mainly solved the problem using the classification framework that does not explore prediction/generation of important context words from input sentences for causal recognition. In this work, we consider the words along the dependency path between the two event mentions in the dependency tree as the important context words for ECI. We in- troduce dependency path generation as a com- plementary task for ECI, which can be solved jointly with causal label prediction to improve the performance. To facilitate the multi-task learning, we cast ECI into a generation prob- lem that aims to generate both causal relation and dependency path words from input sen- tence. In addition, we propose to use the REIN- FORCE algorithm to train our generative model where novel reward functions are designed to capture both causal prediction accuracy and generation quality. The experiments on two benchmark datasets demonstrate state-of-the- art performance of the proposed model for ECI. 1 Introduction In Information Extraction (IE), Event Causality Identification (ECI) aims to predict causal relation between a pair of events mentioned in text. For in- stance, in the sentence “Massive fires cause major damages in the downtown area.”, an ECI system needs to realize the causal relation between the two events triggered by “fires” and “damages” (called event mentions), i.e., “fires” cause −−−→ “damages”. ECI is an important problem with many applications in NLP (Hashimoto, 2019; Berant et al., 2014). Compared to the feature-based methods (Do et al., 2011; Ning et al., 2018), recent deep learn- ing models have demonstrated their state-of-the-art performance for ECI (Kadowaki et al., 2019; Liu et al., 2020; Zuo et al., 2021). As such, prior work has mainly treated ECI as a classification problem where the only output from the models is a label to indicate causal or non-causal relation between input events. A major issue with this classification formulation is that current ECI models do not out- put important contexts for causal prediction of two event mentions. In this work, important contexts refer to the words in the input sentence that are critical to reveal the causal relation between two given event mentions (e.g., the words “caused” and “by” in our example). This limitation of current ECI models is undesirable as we expect that including important context words as a part of the outputs for ECI models can improve the training signals for the models. In particular, motivated by rela- tion exaction models in IE (Zhang et al., 2018), we use the words …

Content truncated to 500 words