Social Bias Probing: Fairness Benchmarking for Language Models
This addresses fairness benchmarking for language models, providing a more comprehensive tool for researchers and practitioners, though it is incremental in expanding beyond existing binary methods.
The paper tackles the problem of evaluating social biases in language models by proposing a novel framework that assesses disparate treatment, and it results in the creation of SoFa, a large-scale benchmark revealing more nuanced biases, with findings showing that identities expressing different religions lead to the most pronounced disparate treatments across models.
While the impact of social biases in language models has been recognized, prior methods for bias evaluation have been limited to binary association tests on small datasets, limiting our understanding of bias complexities. This paper proposes a novel framework for probing language models for social biases by assessing disparate treatment, which involves treating individuals differently according to their affiliation with a sensitive demographic group. We curate SoFa, a large-scale benchmark designed to address the limitations of existing fairness collections. SoFa expands the analysis beyond the binary comparison of stereotypical versus anti-stereotypical identities to include a diverse range of identities and stereotypes. Comparing our methodology with existing benchmarks, we reveal that biases within language models are more nuanced than acknowledged, indicating a broader scope of encoded biases than previously recognized. Benchmarking LMs on SoFa, we expose how identities expressing different religions lead to the most pronounced disparate treatments across all models. Finally, our findings indicate that real-life adversities faced by various groups such as women and people with disabilities are mirrored in the behavior of these models.