Controlling Bias Exposure for Fair Interpretable Predictions

TL;DR

This paper introduces a novel debiasing algorithm for NLP models that balances bias mitigation and task performance by controlling the use of sensitive information rather than eliminating it entirely.

Contribution

The work proposes a new debiasing method that adjusts the model's reliance on sensitive attributes based on their relevance, providing more nuanced bias control.

Findings

Achieves a better trade-off between debiasing and task accuracy.

Produces debiased rationales as evidence.

Effective on multiple NLP tasks influenced by gender and race.

Abstract

Recent work on reducing bias in NLP models usually focuses on protecting or isolating information related to a sensitive attribute (like gender or race). However, when sensitive information is semantically entangled with the task information of the input, e.g., gender information is predictive for a profession, a fair trade-off between task performance and bias mitigation is difficult to achieve. Existing approaches perform this trade-off by eliminating bias information from the latent space, lacking control over how much bias is necessarily required to be removed. We argue that a favorable debiasing method should use sensitive information 'fairly', rather than blindly eliminating it (Caliskan et al., 2017; Sun et al., 2019; Bogen et al., 2020). In this work, we provide a novel debiasing algorithm by adjusting the predictive model's belief to (1) ignore the sensitive information if it is not useful for the task; (2) use sensitive information minimally as necessary for the prediction (while also incurring a penalty). Experimental results on two text classification tasks (influenced by gender) and an open-ended generation task (influenced by race) indicate that our model achieves a desirable trade-off between debiasing and task performance along with producing debiased rationales as evidence.