Handling Multiword Expressions in Causality Estimation

Previous studies on causality estimation mainly aquire causal event pairs from a large corpus based on lexico-syntactic patterns and coreference relations, and estimate causality by a statistical method. However, most of the previous studies assume event pairs can be represented by a pair of single words, therefore they cannot estimate multiword causality correctly (e.g.“tired”-“give up”) . In this paper, we create a list of multiword expressions and extend an existing method. Our evaluation demonstrates that the proper treatment of multiword expression events is effective and the proposed method outperforms the state-of-the-art causality estimation model.