Experimental Research on Stereotype Expression in Freely Generated Language: A Systematic Literature Review

Social stereotypes spread through biased language. Experiments in which participants freely generate language allow researchers to causally test how stereotypes affect language use and capture multiple linguistic variables simultaneously, offering high ecological validity. Despite this potential, it remains unclear how and to which extent this paradigm has been used to study linguistic biases. To address this gap, we conducted a systematic literature review ( k = 106 experiments) and found that most freely generated language experiments examined only one linguistic variable, leaving the interplay between linguistic variables largely underexplored. To advance our understanding of stereotype expression in everyday language, we recommend experiments to measure multiple linguistic variables simultaneously and in relation to each other. This knowledge is needed to address harmful stereotype communication in the real world through automated detection tools, communication interventions, and the fine-tuning of Large Language Models.