Surfacing Subtle Stereotypes: A Multilingual, Debate-Oriented Evaluation of Modern LLMs

• URL: http://arxiv.org/abs/2511.01187v1
• Date: Mon, 03 Nov 2025 03:25:40 GMT
• Title: Surfacing Subtle Stereotypes: A Multilingual, Debate-Oriented Evaluation of Modern LLMs
• Authors: Muhammed Saeed, Muhammad Abdul-mageed, Shady Shehata,
• Abstract summary: We introduce DebateBias-8K, a new multilingual, debate-style benchmark to reveal how narrative bias appears in realistic generative settings.<n>Our dataset includes 8,400 structured debate prompts spanning four sensitive domains: women's rights, socioeconomic development, terrorism, and religion, across seven languages.<n>Results show that all models reproduce entrenched stereotypes despite safety alignment.
• Score: 32.12545369011503
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) are widely deployed for open-ended communication, yet most bias evaluations still rely on English, classification-style tasks. We introduce DebateBias-8K, a new multilingual, debate-style benchmark designed to reveal how narrative bias appears in realistic generative settings. Our dataset includes 8,400 structured debate prompts spanning four sensitive domains: women's rights, socioeconomic development, terrorism, and religion, across seven languages ranging from high-resource (English, Chinese) to low-resource (Swahili, Nigerian Pidgin). Using four flagship models (GPT-4o, Claude 3, DeepSeek, and LLaMA 3), we generate and automatically classify over 100,000 responses. Results show that all models reproduce entrenched stereotypes despite safety alignment: Arabs are overwhelmingly linked to terrorism and religion (>=95%), Africans to socioeconomic "backwardness" (up to <=77%), and Western groups are consistently framed as modern or progressive. Biases grow sharply in lower-resource languages, revealing that alignment trained primarily in English does not generalize globally. Our findings highlight a persistent divide in multilingual fairness: current alignment methods reduce explicit toxicity but fail to prevent biased outputs in open-ended contexts. We release our DebateBias-8K benchmark and analysis framework to support the next generation of multilingual bias evaluation and safer, culturally inclusive model alignment.

Related papers

• SAFARI: A Community-Engaged Approach and Dataset of Stereotype Resources in the Sub-Saharan African Context [10.43559852429736]
  Stereotype repositories are critical to assess generative AI model safety, but currently lack adequate global coverage.<n>This work introduces a multilingual stereotype resource covering four sub-Saharan African countries that are severely underrepresented in NLP resources: Ghana, Kenya, Nigeria, and South Africa.
  arXiv  Detail & Related papers  (2026-02-25T20:56:27Z)
• Evaluating Modern Large Language Models on Low-Resource and Morphologically Rich Languages:A Cross-Lingual Benchmark Across Cantonese, Japanese, and Turkish [12.286855282078305]
  GPT-4o, GPT-4, Claude3.5Sonnet, LLaMA3.1, MistralLarge2, LLaMA-2Chat13B, and Mistral7BInstruct are evaluated.<n>Our benchmark spans four diverse tasks: open-domain question answering, document summarization, English-to-X translation, and culturally grounded dialogue.
  arXiv  Detail & Related papers  (2025-11-05T22:09:53Z)
• I Am Aligned, But With Whom? MENA Values Benchmark for Evaluating Cultural Alignment and Multilingual Bias in LLMs [5.060243371992739]
  We introduce MENAValues, a novel benchmark designed to evaluate the cultural alignment and multilingual biases of large language models (LLMs)<n> Drawing from large-scale, authoritative human surveys, we curate a structured dataset that captures the sociocultural landscape of MENA with population-level response distributions from 16 countries.<n>Our analysis reveals three critical phenomena: "Cross-Lingual Value Shifts" where identical questions yield drastically different responses based on language, "Reasoning-Induced Degradation" where prompting models to explain their reasoning worsens cultural alignment, and "Logit Leakage" where models refuse sensitive questions while internal probabilities reveal strong hidden
  arXiv  Detail & Related papers  (2025-10-15T05:10:57Z)
• MMA-ASIA: A Multilingual and Multimodal Alignment Framework for Culturally-Grounded Evaluation [91.22008265721952]
  MMA-ASIA centers on a human-curated, multilingual, and multimodally aligned benchmark covering 8 Asian countries and 10 languages.<n>This is the first dataset aligned at the input level across three modalities: text, image (visual question answering), and speech.<n>We propose a five-dimensional evaluation protocol that measures: (i) cultural-awareness disparities across countries, (ii) cross-lingual consistency, (iii) cross-modal consistency, (iv) cultural knowledge generalization, and (v) grounding validity.
  arXiv  Detail & Related papers  (2025-10-07T14:12:12Z)
• SESGO: Spanish Evaluation of Stereotypical Generative Outputs [1.1549572298362782]
  This paper addresses the critical gap in evaluating bias in multilingual Large Language Models (LLMs)<n>Current evaluations remain predominantly US-English-centric, leaving potential harms in other linguistic and cultural contexts largely underexamined.<n>We introduce a novel, culturally-grounded framework for detecting social biases in instruction-tuned LLMs.
  arXiv  Detail & Related papers  (2025-09-03T14:04:51Z)
• PakBBQ: A Culturally Adapted Bias Benchmark for QA [3.4455728937232597]
  We introduce PakBBQ, a culturally and regionally adapted extension of the original Bias Benchmark for Question Answering dataset.<n> PakBBQ comprises over 214 templates, 17180 QA pairs across 8 categories in both English and Urdu, covering eight bias dimensions including age, disability, appearance, gender, socio-economic status, religious, regional affiliation, and language formality that are relevant in Pakistan.
  arXiv  Detail & Related papers  (2025-08-13T20:42:44Z)
• EuroGEST: Investigating gender stereotypes in multilingual language models [58.871032460235575]
  We introduce EuroGEST, a dataset designed to measure gender-stereotypical reasoning in LLMs across English and 29 European languages.<n>We show that the strongest stereotypes in all models across all languages are that women are 'beautiful', 'empathetic' and 'neat' and men are 'leaders','strong, tough' and 'professional'
  arXiv  Detail & Related papers  (2025-06-04T11:58:18Z)
• Spoken Stereoset: On Evaluating Social Bias Toward Speaker in Speech Large Language Models [50.40276881893513]
  This study introduces Spoken Stereoset, a dataset specifically designed to evaluate social biases in Speech Large Language Models (SLLMs) By examining how different models respond to speech from diverse demographic groups, we aim to identify these biases. The findings indicate that while most models show minimal bias, some still exhibit slightly stereotypical or anti-stereotypical tendencies.
  arXiv  Detail & Related papers  (2024-08-14T16:55:06Z)
• White Men Lead, Black Women Help? Benchmarking and Mitigating Language Agency Social Biases in LLMs [58.27353205269664]
  Social biases can manifest in language agency in Large Language Model (LLM)-generated content.<n>We introduce the Language Agency Bias Evaluation benchmark, which comprehensively evaluates biases in LLMs.<n>Using LABE, we unveil language agency social biases in 3 recent LLMs: ChatGPT, Llama3, and Mistral.
  arXiv  Detail & Related papers  (2024-04-16T12:27:54Z)
• SeeGULL Multilingual: a Dataset of Geo-Culturally Situated Stereotypes [18.991295993710224]
  SeeGULL is a global-scale multilingual dataset of social stereotypes, spanning 20 languages, with human annotations across 23 regions, and demonstrate its utility in identifying gaps in model evaluations.
  arXiv  Detail & Related papers  (2024-03-08T22:09:58Z)
• CBBQ: A Chinese Bias Benchmark Dataset Curated with Human-AI Collaboration for Large Language Models [52.25049362267279]
  We present a Chinese Bias Benchmark dataset that consists of over 100K questions jointly constructed by human experts and generative language models. The testing instances in the dataset are automatically derived from 3K+ high-quality templates manually authored with stringent quality control. Extensive experiments demonstrate the effectiveness of the dataset in detecting model bias, with all 10 publicly available Chinese large language models exhibiting strong bias in certain categories.
  arXiv  Detail & Related papers  (2023-06-28T14:14:44Z)
• Towards Understanding and Mitigating Social Biases in Language Models [107.82654101403264]
  Large-scale pretrained language models (LMs) can be potentially dangerous in manifesting undesirable representational biases. We propose steps towards mitigating social biases during text generation. Our empirical results and human evaluation demonstrate effectiveness in mitigating bias while retaining crucial contextual information.
  arXiv  Detail & Related papers  (2021-06-24T17:52:43Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.