Unlearn Dataset Bias in Natural Language Inference by Fitting the Residual

TL;DR

This paper introduces DRiFt, a residual fitting method to reduce dataset bias in natural language inference models, improving their robustness on challenge datasets by focusing on examples not explained by biased features.

Contribution

The paper formalizes dataset bias under distribution shift and proposes a simple residual fitting algorithm, DRiFt, to debias NLI models effectively.

Findings

Debiased models outperform baselines on challenge test sets.

DRiFt improves robustness while maintaining original performance.

Applicable to multiple high-performing NLI models.

Abstract

Statistical natural language inference (NLI) models are susceptible to learning dataset bias: superficial cues that happen to associate with the label on a particular dataset, but are not useful in general, e.g., negation words indicate contradiction. As exposed by several recent challenge datasets, these models perform poorly when such association is absent, e.g., predicting that "I love dogs" contradicts "I don't love cats". Our goal is to design learning algorithms that guard against known dataset bias. We formalize the concept of dataset bias under the framework of distribution shift and present a simple debiasing algorithm based on residual fitting, which we call DRiFt. We first learn a biased model that only uses features that are known to relate to dataset bias. Then, we train a debiased model that fits to the residual of the biased model, focusing on examples that cannot be predicted well by biased features only. We use DRiFt to train three high-performing NLI models on two benchmark datasets, SNLI and MNLI. Our debiased models achieve significant gains over baseline models on two challenge test sets, while maintaining reasonable performance on the original test sets.