Counter-GAP: Counterfactual Bias Evaluation through Gendered Ambiguous Pronouns

TL;DR

This paper introduces Counter-GAP, a novel dataset and evaluation method for gender bias in coreference resolution, highlighting bias inconsistencies in language models and proposing effective mitigation strategies.

Contribution

The work presents a new counterfactual data collection approach, a bias measurement metric addressing bias cancellation, and insights into bias mitigation effectiveness.

Findings

Language models show greater inconsistency across genders than within.

Counterfactual augmentation reduces gender bias more effectively than anonymization.

Counter-GAP dataset enables more reliable bias evaluation.

Abstract

Bias-measuring datasets play a critical role in detecting biased behavior of language models and in evaluating progress of bias mitigation methods. In this work, we focus on evaluating gender bias through coreference resolution, where previous datasets are either hand-crafted or fail to reliably measure an explicitly defined bias. To overcome these shortcomings, we propose a novel method to collect diverse, natural, and minimally distant text pairs via counterfactual generation, and construct Counter-GAP, an annotated dataset consisting of 4008 instances grouped into 1002 quadruples. We further identify a bias cancellation problem in previous group-level metrics on Counter-GAP, and propose to use the difference between inconsistency across genders and within genders to measure bias at a quadruple level. Our results show that four pre-trained language models are significantly more inconsistent across different gender groups than within each group, and that a name-based counterfactual data augmentation method is more effective to mitigate such bias than an anonymization-based method.