How Gender Debiasing Affects Internal Model Representations, and Why It Matters
This addresses the problem of understanding and mitigating gender bias in NLP systems for more informed deployment, though it is incremental in linking bias types.
The paper investigates the relationship between extrinsic and intrinsic gender bias in NLP models by debiasing during fine-tuning and measuring effects on internal representations, showing that an information-theoretic probing metric better indicates debiasing than standard metrics and exposes superficial cases.
Common studies of gender bias in NLP focus either on extrinsic bias measured by model performance on a downstream task or on intrinsic bias found in models' internal representations. However, the relationship between extrinsic and intrinsic bias is relatively unknown. In this work, we illuminate this relationship by measuring both quantities together: we debias a model during downstream fine-tuning, which reduces extrinsic bias, and measure the effect on intrinsic bias, which is operationalized as bias extractability with information-theoretic probing. Through experiments on two tasks and multiple bias metrics, we show that our intrinsic bias metric is a better indicator of debiasing than (a contextual adaptation of) the standard WEAT metric, and can also expose cases of superficial debiasing. Our framework provides a comprehensive perspective on bias in NLP models, which can be applied to deploy NLP systems in a more informed manner. Our code and model checkpoints are publicly available.