Using Natural Sentences for Understanding Biases in Language Models

Abstract: Evaluation of biases in LLMs is often limited to synthetically generated datasets. This dependence traces back to the need for a prompt-style dataset to trigger specific behaviors of LLMs. In this paper, we address this gap by creating a prompt dataset with respect to occupations collected from real-world natural sentences present in Wikipedia. We aim to understand the differences between using template-based prompts and natural sentence prompts when studying gender-occupation biases in LLMs. We find bias evaluations are very sensitive to the design choices of template prompts, and we propose using natural sentence prompts for systematic evaluations to step away from design choices that could introduce bias in the observations.