In-Depth Look at Word Filling Societal Bias Measures
This work addresses methodological flaws in bias evaluation for language models, which is crucial for researchers and practitioners in AI ethics, though it is incremental as it critiques existing measures rather than introducing a new paradigm.
The paper analyzed the validity of two popular societal bias measures, StereoSet and CrowS-Pairs, showing they produce unexpected and illogical results when control groups are used, and proposed an improved testing protocol while introducing a new gender bias dataset for Slovak.
Many measures of societal bias in language models have been proposed in recent years. A popular approach is to use a set of word filling prompts to evaluate the behavior of the language models. In this work, we analyze the validity of two such measures -- StereoSet and CrowS-Pairs. We show that these measures produce unexpected and illogical results when appropriate control group samples are constructed. Based on this, we believe that they are problematic and using them in the future should be reconsidered. We propose a way forward with an improved testing protocol. Finally, we also introduce a new gender bias dataset for Slovak.