SDARE-Bench: Evaluating Large Language Models on Conversational Stigma Detection and Response in Dyadic and Group Dialogue
2026-09-01 • Computation and Language
Computation and Language
AI summaryⓘ
The authors studied how well large language models (LLMs) can recognize and respond to stigma, which is negative judgment about people or groups. They created a new test called SDARE-Bench, which uses realistic conversations instead of simple questions. Their tests with 8 different LLMs showed that these models often fail to notice stigma, especially in group conversations, and tend to give worse advice when stigma is involved. The authors also found that stigma was even more common when the models faced simulated group pressure. Overall, they highlight that dealing with stigma is a weak spot for LLM safety in complex social conversations.
Large Language ModelsStigma DetectionConversational ContextGroup DialogueOpen-ended Response GenerationBenchmarkingSafety VulnerabilitiesSocial JudgementDyadic QueriesGroup Pressure
Authors
Stephanie Fong, Yiwen Jiang, Zimu Wang, Hongxi Yang, Yaling Shen, Hiu Weh Naomi Chow, Heung Ying Lai, Xiangyu Zhao, Qingyang Xu, Zhongxing Xu, Jiahe Liu, Guilherme C. Oliveira, Vincent Lee, Zongyuan Ge, Dominic Dwyer
Abstract
Large Language Models (LLMs) are increasingly used in advice seeking and decision making that may affect social judgements. Despite stigma's profound effects on people and communities, benchmarks remain scarce. Existing general-domain evaluations typically rely on static prompts and fixed-format tasks, overlooking conversational contexts and audience effects in everyday communication. To address these gaps, we introduce SDARE-Bench, the first scenario-based benchmark evaluating both stigma detection and open-ended response generation in LLMs, comprising 1,138 dyadic queries and 1,388 group dialogue. Empirical results across 8 LLMs consistently demonstrate poor identification of stigma components, especially in group dialogues. In open-ended response generation, stigma expression was substantially higher in group settings than in dyadic, with weaker resistance to stigma and more unrealistic advice. Responses were evaluated using a classifier trained on 1,392 human annotated responses. In constructed group pressure settings, stigma expression rates further increased to a striking average of 97.5%. Our findings identify stigma response as a recurring LLM safety vulnerability, especially in socially complex conversational contexts.