CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Models

TL;DR

CrowS-Pairs is a benchmark dataset designed to measure social biases in masked language models by comparing stereotypical and less stereotypical sentences across multiple bias categories, revealing prevalent biases in current models.

Contribution

Introduces CrowS-Pairs, a new dataset with 1508 examples to evaluate social biases in masked language models across nine bias types, focusing on disadvantaged groups.

Findings

All evaluated MLMs favor stereotypical sentences across categories.

CrowS-Pairs can serve as a benchmark for bias evaluation.

The dataset highlights the pervasiveness of social biases in language models.

Abstract

Pretrained language models, especially masked language models (MLMs) have seen success across many NLP tasks. However, there is ample evidence that they use the cultural biases that are undoubtedly present in the corpora they are trained on, implicitly creating harm with biased representations. To measure some forms of social bias in language models against protected demographic groups in the US, we introduce the Crowdsourced Stereotype Pairs benchmark (CrowS-Pairs). CrowS-Pairs has 1508 examples that cover stereotypes dealing with nine types of bias, like race, religion, and age. In CrowS-Pairs a model is presented with two sentences: one that is more stereotyping and another that is less stereotyping. The data focuses on stereotypes about historically disadvantaged groups and contrasts them with advantaged groups. We find that all three of the widely-used MLMs we evaluate substantially favor sentences that express stereotypes in every category in CrowS-Pairs. As work on building less biased models advances, this dataset can be used as a benchmark to evaluate progress.