Abstract: Counterfactual Data Augmentation (CDA) has proven effective in enhancing robustness and interpretability of Natural Language Processing (NLP) models. However, existing CDA methods often rely ...