Unequal Voices: How LLMs Construct Constrained Queer Narratives

2025/07/21 by Ghosal, Atreya, Gupta, Ashim, Srikumar, Vivek · 2 citations
#Artificial Intelligence (cs.AI) #Computers and Society (cs.CY) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2507.15585

Abstract

One way social groups are marginalized in discourse is that the narratives told about them often default to a narrow, stereotyped range of topics. In contrast, default groups are allowed the full complexity of human existence. We describe the constrained representations of queer people in LLM generations in terms of harmful representations, narrow representations, and discursive othering and formulate hypotheses to test for these phenomena. Our results show that LLMs are significantly limited in their portrayals of queer personas.

Citations

• Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs
• CoMPosT: Characterizing and Evaluating Caricature in LLM Simulations
• Sensitivity, Performance, Robustness: Deconstructing the Effect of Sociodemographic Prompting
• SeeGULL: A Stereotype Benchmark with Broad Geo-Cultural Coverage Leveraging Generative Models
• Trans broken arm syndrome: A mixed-methods exploration of gender-related medical misattribution and invasive questioning
• Language (Technology) is Power: A Critical Survey of "Bias" in NLP
• Fairness in representation: quantifying stereotyping as a representational harm
• The Rani of Sirmur: An Essay in Reading the Archives

Cited by

• LLM Harms: A Taxonomy and Discussion
• Do Language Models Know Their Slang? Queer Slang Understanding in User-Generated Content

Related

• Queer People are People First: Deconstructing Sexual Identity Stereotypes in Large Language Models
• Constructing (non-)normative identities in written lesbian discourse: A diachronic study
• LLMs Reproduce Stereotypes of Sexual and Gender Minorities
• Queer Bias in Natural Language Processing
• QueerBench: Quantifying Discrimination in Language Models Toward Queer Identities