In-Depth Look at Word Filling Societal Bias Measures

TL;DR

This paper critically examines two popular societal bias measures in language models, revealing their flaws and proposing improved testing protocols, along with introducing a new Slovak gender bias dataset.

Contribution

It analyzes the validity of StereoSet and CrowS-Pairs bias measures, identifies their issues, and proposes a new testing protocol and a Slovak gender bias dataset.

Findings

StereoSet and CrowS-Pairs produce illogical results with proper controls

Current bias measures may be unreliable for societal bias evaluation

Introduces a new gender bias dataset for Slovak language

Abstract

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.