Knowledge-Enriched Event Causality Identification via Latent Structure Induction Networks

Knowledge-Enriched Event Causality Identification via Latent Structure Induction Networks Pengfei Cao1,2, Xinyu Zuo1,2, Yubo Chen1,2, Kang Liu1,2, Jun Zhao1,2, Yuguang Chen3 and Weihua Peng3 1National Laboratory of Pattern Recognition, Institute of Automation, CAS, Beijing, China 2School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China 3Beijing Baidu Netcom Science Technology Co., Ltd {pengfei.cao, xinyu.zuo, yubo.chen, kliu, jzhao}@nlpr.ia.ac.cn, {chenyuguang, pengweihua}@baidu.com Abstract Identifying causal relations of events is an im- portant task in natural language processing area. However, the task is very challenging, because event causality is usually expressed in diverse forms that often lack explicit causal clues. Existing methods cannot handle well the problem, especially in the condition of lacking training data. Nonetheless, humans can make a correct judgement based on their background knowledge, including descriptive knowledge and relational knowledge. Inspired by it, we propose a novel Latent Structure Induction Network (LSIN) to incorporate the external structural knowledge into this task. Specifically, to make use of the descriptive knowledge, we devise a Descriptive Graph Induction module to obtain and encode the graph-structured descriptive knowledge. To leverage the relational knowledge, we propose a Relational Graph Induction module which is able to automatically learn a reasoning struc- ture for event causality reasoning. Experi- mental results on two widely used datasets in- dicate that our approach significantly outper- forms previous state-of-the-art methods. 1 Introduction Event causality identification (ECI) aims to iden- tify causal relation of events in texts. For exam- ple, in the sentence “The earthquake generated a tsunami.”, an ECI model should be able to identify a causal relationship that holds between the two mentioned events, i.e., earthquake cause −−−→ tsunami. ECI is an important task in natural language pro- cessing (NLP) area and can support many NLP ap- plications, such as machine reading comprehension (Berant et al., 2014), process extraction (Thalap- pillil Scaria et al., 2013) and future event prediction (Radinsky et al., 2012; Hashimoto et al., 2014). Identifying event causal relation is inherently Global warming worsened, and tsunami strengthened. sea-level rising AtLocation CapableOf leveraging external structural knowledge global warming IsA heating glacier melting temperature change IsA greenhouse gas CreatedBy Causes CapableOf IsA death Causes tsunami ocean destroy house tidal wave AtLocation Figure 1: An example of leveraging the external struc- tural knowledge for ECI task. The dashed arrow indi- cates a missing link in the knowledge base. challenging, because event causality is usually ex- pressed in diverse forms that often lack explicit clues indicating its existence. For example in Fig- ure 1, the sentence has no explicit clue indicat- ing the causal relation between “global warming” and “tsunami”. In this scenario, models can re- sort to a large amount of labeled data to learn di- verse causal expressions. However, existing ECI datasets are very small. For example, the largest dataset EventStoryLine (Caselli and Vossen, 2017) only contains 258 documents, which is not suffi- cient to train neural network models (Liu et al., 2020). Consequently, models cannot thoroughly understand the text and possibly make a wrong pre- diction. Nonetheless, humans …

Content truncated to 500 words