EXPLAIN, EDIT, GENERATE: Rationale-Sensitive Counterfactual Data Augmentation for Multi-hop Fact Verification

Abstract

Automatic multi-hop fact verification task has gained significant attention in recent years. De spite impressive results, these well-designed models perform poorly on out-of-domain data One possible solution is to augment the training data with counterfactuals, which are generated by minimally altering the causal features of the original data. However, current counterfac tual data augmentation techniques fail to handle multi-hop fact verification due to their incapa bility to preserve the complex logical relation ships within multiple correlated texts. In this paper, we overcome this limitation by devel oping a rationale-sensitive method to generate linguistically diverse and label-flipping counter factuals while preserving logical relationships In specific, the diverse and fluent counterfactu als are generated via an Explain-Edit-Generate architecture. Moreover, the checking and fil tering modules are proposed to regularize the counterfactual data with logical relations and flipped labels. Experimental results show that the proposed approach outperforms the SOTA baselines and can generate linguistically di verse counterfactual data without disrupting their logical relationships.