Improving the Out-Of-Distribution Generalization Capability of Language Models: Counterfactually-Augmented Data is not Enough

TL;DR

This paper identifies the Myopia Phenomenon limiting CAD's effectiveness in OOD generalization and proposes constraints to enable models to extract more comprehensive causal features, improving performance on Sentiment Analysis and NLI tasks.

Contribution

The paper introduces two constraints based on CAD's structural properties to mitigate Myopia Phenomenon and enhance OOD generalization in language models.

Findings

Improved OOD performance on Sentiment Analysis and NLI tasks.

Constraints help models utilize more complete causal features.

Method outperforms baseline approaches.

Abstract

Counterfactually-Augmented Data (CAD) has the potential to improve language models' Out-Of-Distribution (OOD) generalization capability, as CAD induces language models to exploit causal features and exclude spurious correlations. However, the empirical results of OOD generalization on CAD are not as efficient as expected. In this paper, we attribute the inefficiency to Myopia Phenomenon caused by CAD: language models only focus on causal features that are edited in the augmentation and exclude other non-edited causal features. As a result, the potential of CAD is not fully exploited. Based on the structural properties of CAD, we design two additional constraints to help language models extract more complete causal features contained in CAD, thus improving the OOD generalization capability. We evaluate our method on two tasks: Sentiment Analysis and Natural Language Inference, and the experimental results demonstrate that our method could unlock CAD's potential and improve language models' OOD generalization capability.