Identifying While Learning for Document Event Causality Identification

Event Causality Identification (ECI) aims to detect whether there exists acausal relation between two events in a document. Existing studies adopt a kindof identifying after learning paradigm, where events' representations are firstlearned and then used for the identification. Furthermore, they mainly focus onthe causality existence, but ignoring causal direction. In this paper, we takecare of the causal direction and propose a new identifying while learning modefor the ECI task. We argue that a few causal relations can be easily identifiedwith high confidence, and the directionality and structure of these identifiedcausalities can be utilized to update events' representations for boosting nextround of causality identification. To this end, this paper designs an*iterative learning and identifying framework*: In each iteration, we constructan event causality graph, on which events' causal structure representations areupdated for boosting causal identification. Experiments on two public datasetsshow that our approach outperforms the state-of-the-art algorithms in bothevaluations for causality existence identification and directionidentification.