Balancing out Bias: Achieving Fairness Through Balanced Training

TL;DR

This paper proposes a balanced training method with a gated model to reduce demographic bias in NLP tasks, achieving fairer outcomes by directly addressing correlations between demographics and language.

Contribution

It introduces a novel balanced training objective and a gated model that incorporates protected attributes, improving bias mitigation in NLP systems.

Findings

Outperforms existing bias mitigation techniques when combined with balanced training.

Effectively reduces disparities in error rates across demographic groups.

Enhances fairness in NLP predictions through demographic input perturbation.

Abstract

Group bias in natural language processing tasks manifests as disparities in system error rates across texts authorized by different demographic groups, typically disadvantaging minority groups. Dataset balancing has been shown to be effective at mitigating bias, however existing approaches do not directly account for correlations between author demographics and linguistic variables, limiting their effectiveness. To achieve Equal Opportunity fairness, such as equal job opportunity without regard to demographics, this paper introduces a simple, but highly effective, objective for countering bias using balanced training. We extend the method in the form of a gated model, which incorporates protected attributes as input, and show that it is effective at reducing bias in predictions through demographic input perturbation, outperforming all other bias mitigation techniques when combined with balanced training.