Acquiring and Generalizing Causal Inference Rules from Deverbal Noun Constructions

This paper presents a novel approach for inducing causal rules by using deverbal nouns as a clue for finding causal relations. We collect verbs and their deverbal forms from FrameNet, and extract pairs of sentences in which event verbs and their corresponding deverbal forms co-occur in documents. The most challenging part of this work is to generalize an instance of causal relation into a rule. This paper proposes a method to generalize and constrain causal rules so that the obtained rules have the high chance of applicability and reusability. In order to find a suitable constraint for a causal rule, we utilize relation instances extracted by an open-information extractor, and build a classifier to choose the most suitable constraint. We demonstrate that deverbal nouns provide a good clue for causal relations and that the proposed method can induce causal rules from deverbal noun constructions.