This Prompt is Measuring <MASK>: Evaluating Bias Evaluation in Language Models

TL;DR

This paper critically examines bias evaluation methods in language models, revealing ambiguities and gaps in current practices, and offers guidance to improve the measurement of social biases in NLP.

Contribution

It introduces a taxonomy for bias measurement in language models and analyzes 90 bias tests to identify conceptual and operational issues.

Findings

Many bias tests have unstated or ambiguous assumptions.

Core bias types are under-researched and not fully captured.

Guidance is provided to expand bias measurement scope.

Abstract

Bias research in NLP seeks to analyse models for social biases, thus helping NLP practitioners uncover, measure, and mitigate social harms. We analyse the body of work that uses prompts and templates to assess bias in language models. We draw on a measurement modelling framework to create a taxonomy of attributes that capture what a bias test aims to measure and how that measurement is carried out. By applying this taxonomy to 90 bias tests, we illustrate qualitatively and quantitatively that core aspects of bias test conceptualisations and operationalisations are frequently unstated or ambiguous, carry implicit assumptions, or be mismatched. Our analysis illuminates the scope of possible bias types the field is able to measure, and reveals types that are as yet under-researched. We offer guidance to enable the community to explore a wider section of the possible bias space, and to better close the gap between desired outcomes and experimental design, both for bias and for evaluating language models more broadly.