2026-05-20T04:14:40Zhttps://keep.lib.asu.edu/oai/request

oai:keep.lib.asu.edu:node-1985282025-02-13T23:39:23Zoai_pmh:alloai_pmh:repo_items

198528 https://hdl.handle.net/2286/R.2.N.198528 http://rightsstatements.org/vocab/InC/1.0/ http://creativecommons.org/licenses/by-nc-sa/4.0 2024-12 27 pages Tint, Joshua Baral, Chitta Senanayake, Ransalu Barrett, The Honors College Computer Science and Engineering Program Text Language models have integrated themselves into many aspects of digital life, shaping everything from casual conversations to critical support systems. This thesis investigates how large language models (LLMs) respond to LGBTQ+ slang and heteronormative language. While LLMs have the potential to provide inclusive digital support, biases in their training data often result in misinterpretation of queer language, reinforcing stereotypes and marginalizing LGBTQ+ communities. Through a series of experiments, the study evaluates factual accuracy, emotional content, and the impact of queer slang on responses from models including GPT-3.5, GPT-4o, Llama2, Llama3, Gemma, and Mistral. The findings reveal that heteronormative prompts can trigger safety mechanisms, leading to neutral or corrective responses, while LGBTQ+ slang elicits more negative emotions, indicating a need for improved inclusivity in LLMs. These insights highlight the importance of enhancing LLM fairness and emotional responsiveness to reflect diverse linguistic identities. NLP Natural Language Processing Queer Studies Large Language Models Sentiment Analysis Understanding Responses to LGBTQ+ Language in Large Language Models