论文检索

Prommy Sultana Hossain, Chahat Raj, Ziwei Zhu, Jessica Lin, Emanuela Marasco

Large Language Models (LLMs) such as GPT and LLaMA excel in natural language tasks, e.g., text generation and machine translation. However, inherent biases from training on vast Internet datasets potentially amplify harmful stereotypes—widely held, oversimplified, and often inaccurate generalizations about groups of people. Our contribution introduces a novel, systematic, and architecture-aware method to identify and mitigate stereotypical bias in decoder-only transformer models. This interpretable approach operates without gradient access or retraining from scratch. We first evaluate bias and then apply a bias localization mechanism that correlates internal activations with a newly defined Context Influence (CI) Score. Our method pinpoints specific attention heads that consistently align with biased shifts in model predictions. To mitigate this, we introduce a soft pruning strategy that scales attention head parameters based on their correlation strength, followed by lightweight fine-tuning to maintain fluent text generation. Experiments across five models demonstrate our approach reduces bias by up to 37% on BBQ, 32% on StereoSet, and 33% on CrowS-Pairs while simultaneously improving reasoning performance on MMLU by up to 10%.

Srikant Panda, Amit Agarwal, Hitesh Laxmichand Patel

Large Language Models (LLMs) are increasingly deployed across diverse domains but often exhibit disparities in how they handle real life queries. To systematically investigate these effects with various disability context, we introduce AccessEval, a large-scale benchmark evaluating total 21 close & open source LLMs across six real-world domains and nine disability types using paired Neutral and Disability-Aware Queries. We evaluated model outputs with metrics for factual accuracy, sentiment, and social perception.Our analysis reveals that responses to disability-aware queries tend to have higher factual error, more negative tone, and increased stereotyping with social perception compared to neutral queries. These effects show notable variation by domain and disability type. Disabilities affecting hearing, speech and mobility are disproportionately impacted. These disparities reveal persistent forms of ableism, highlighting the need for more comprehensive and nuanced assessment.We further argue that framing bias in terms of model performance within real-world decision making helps to better link model behaviors to the potential harms users may face. This approach guides the development of more effective and tailored fairness interventions. AccessEval, therefore, serves as a crucial tool for advancing equitable and inclusive language technologies.

Orfeas Menis Mastromichalakis, Giorgos Filandrianos, Maria Symeonaki, Giorgos Stamou

Machine Translation (MT) systems frequently encounter gender-ambiguous occupational terms, where they must assign gender without explicit contextual cues. While individual translations in such cases may not be inherently biased, systematic patterns—such as consistently translating certain professions with specific genders—can emerge, reflecting and perpetuating societal stereotypes. This ambiguity challenges traditional instance-level single-answer evaluation approaches, as no single gold standard translation exists. To address this, we introduce GRAPE, a probability-based metric designed to evaluate gender bias by analyzing aggregated model responses. Alongside this, we present GAMBIT, a benchmarking dataset in English with gender-ambiguous occupational terms. Using GRAPE, we evaluate several MT systems and examine whether their gendered translations in Greek and French align with or diverge from societal stereotypes, real-world occupational gender distributions, and normative standards.

Jacqueline Rowe, Mateusz Klimaszewski, Liane Guillou, Shannon Vallor, Alexandra Birch

Large language models increasingly support multiple languages, yet most benchmarks for gender bias remain English-centric. We introduce EuroGEST, a dataset designed to measure gender-stereotypical reasoning in LLMs across English and 29 European languages. EuroGEST builds on an existing expert-informed benchmark covering 16 gender stereotypes, expanded in this work using translation tools, quality estimation metrics, and morphological heuristics. Human evaluations confirm that our data generation method results in high accuracy of both translations and gender labels across languages. We use EuroGEST to evaluate 24 multilingual language models from six model families, demonstrating 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. We also show that larger models encode gendered stereotypes more strongly and that instruction finetuned models continue to exhibit gendered stereotypes. Our work highlights the need for more multilingual studies of fairness in LLMs and offers scalable methods and resources to audit gender bias across languages.

Anthony Dubreuil, Antoine Gourru, Christine Largeron, Amine Trabelsi

Large Language Models inherit stereotypes from their pretraining data, leading to biased behavior toward certain social groups in many Natural Language Processing tasks, such as hateful speech detection or sentiment analysis. Surprisingly, the evaluation of this kind of bias in stance detection methods has been largely overlooked by the community. Stance Detection involves labeling a statement as being against, in favor, or neutral towards a specific target and is among the most sensitive NLP tasks, as it often relates to political leanings. In this paper, we focus on the bias of Large Language Models when performing stance detection in a zero-shot setting. We automatically annotate posts in pre-existing stance detection datasets with two attributes: dialect or vernacular of a specific group and text complexity/readability, to investigate whether these attributes influence the model’s stance detection decisions. Our results show that LLMs exhibit significant stereotypes in stance detection tasks, such as incorrectly associating pro-marijuana views with low text complexity and African American dialect with opposition to Donald Trump.

Hongyu Chen, Neele Falk, Michael Roth, Agnieszka Falenska

In NLP, language–gender associations are commonly grounded in the author’s gender identity, inferred from their language use. However, this identity-based framing risks reinforcing stereotypes and marginalizing individuals who do not conform to normative language–gender associations. To address this, we operationalize the language–gender association as a perceived gender expression of language, focusing on how such expression is externally interpreted by humans, independent of the author’s gender identity. We present the first dataset of itskind: 5,100 human annotations of perceived gendered style—human-written texts rated on a five-point scale from very feminine to verymasculine. While perception is inherently subjective, our analysis identifies textual features associated with higher agreement among annotators: formal expressions and lower emotional intensity. Moreover, annotator demographics influence their perception: women annotators are more likely to label texts as feminine, and men and non-binary annotators as masculine. Finally, feature analysis reveals that the text’s perceived gendered style is shaped by both affective and function words, partially overlapping with known patterns of language variation across gender identities. Our findings lay the groundwork for operationalizing gendered style through human annotation, while also highlighting annotators’ subjective judgments as meaningful signals to understand perception-based concepts.

Aida Mostafazadeh Davani, Sunipa Dev, Héctor Pérez-Urbina, Vinodkumar Prabhakaran

Societal stereotypes are at the center of a myriad of responsible AI interventions targeted at reducing the generation and propagation of potentially harmful outcomes. While these efforts are much needed, they tend to be fragmented and often address different parts of the issue without adopting a unified or holistic approach to social stereotypes and how they impact various parts of the machine learning pipeline. As a result, current interventions fail to capitalize on the underlying mechanisms that are common across different types of stereotypes, and to anchor on particular aspects that are relevant in certain cases. In this paper, we draw on social psychological research and build on NLP data and methods, to propose a unified framework to operationalize stereotypes in generative AI evaluations. Our framework identifies key components of stereotypes that are crucial in AI evaluation, including the target group, associated attribute, relationship characteristics, perceiving group, and context. We also provide considerations and recommendations for its responsible use.

Junhyuk Choi, Ro-hoon Oh, Jihwan Seol, Bugeun Kim

We introduce VoiceBBQ, a spoken extension of the BBQ (Bias Benchmark for Question answering) - a dataset that measures social bias by presenting ambiguous or disambiguated contexts followed by questions that may elicit stereotypical responses. Due to the nature of speech modality, social bias in Spoken Language Models (SLMs) can emerge from two distinct sources: 1) content aspect and 2) acoustic aspect. The dataset converts every BBQ context into controlled voice conditions, enabling per-axis accuracy, bias, and consistency scores that remain comparable to the original text benchmark. Using VoiceBBQ, we evaluate two SLMs—LLaMA-Omni and Qwen2-Audio—and observe architectural contrasts: LLaMA-Omni retains strong acoustic sensitivity, amplifying gender and accent bias, whereas Qwen2-Audio substantially dampens these cues while preserving content fidelity. VoiceBBQ thus provides a compact, drop-in testbed for jointly diagnosing content and acoustic bias across spoken language models.

Hannah Cyberey, Yangfeng Ji, David Evans

Large language models (LLMs) are known to perpetuate stereotypes and exhibit biases. Various strategies have been proposed to mitigate these biases, but most work studies biases as a black-box problem without considering how concepts are represented within the model. We adapt techniques from representation engineering to study how the concept of “gender” is represented within LLMs. We introduce a new method that extracts concept representations via probability weighting without labeled data and efficiently selects a steering vector for measuring and manipulating the model’s representation. We develop a projection-based method that enables precise steering of model predictions and demonstrate its effectiveness in mitigating gender bias in LLMs and show that it also generalizes to racial bias.

Guido Ivetta, Marcos J Gomez, Sofía Martinelli, Pietro Palombini, M Emilia Echeveste, Nair Carolina Mazzeo, Beatriz Busaniche, Luciana Benotti

Most resources for evaluating social biases in Large Language Models are developed without co-design from the communities affected by these biases, and rarely involve participatory approaches. We introduce HESEIA, a dataset of 46,499 sentences created in a professional development course. The course involved 370 high-school teachers and 5,370 students from 189 Latin-American schools. Unlike existing benchmarks, HESEIA captures intersectional biases across multiple demographic axes and school subjects. It reflects local contexts through the lived experience and pedagogical expertise of educators. Teachers used minimal pairs to create sentences that express stereotypes relevant to their school subjects and communities. We show the dataset diversity in term of demographic axes represented and also in terms of the knowledge areas included. We demonstrate that the dataset contains more stereotypes unrecognized by current LLMs than previous datasets. HESEIA is available to support bias assessments grounded in educational communities.

Weicheng Ma, John J. Guerrerio, Soroush Vosoughi

Research on stereotypes in large language models (LLMs) has largely focused on English-speaking contexts, due to the lack of datasets in other languages and the high cost of manual annotation in underrepresented cultures. To address this gap, we introduce a cost-efficient human-LLM collaborative annotation framework and apply it to construct EspanStereo, a Spanish-language stereotype dataset spanning multiple Spanish-speaking countries across Europe and Latin America. EspanStereo captures both well-documented stereotypes from prior literature and culturally specific biases absent from English-centric resources. Using LLMs to generate candidate stereotypes and in-culture annotators to validate them, we demonstrate the framework’s effectiveness in identifying nuanced, region-specific biases. Our evaluation of Spanish-supporting LLMs using EspanStereo reveals significant variation in stereotypical behavior across countries, highlighting the need for more culturally grounded assessments. Beyond Spanish, our framework is adaptable to other languages and regions, offering a scalable path toward multilingual stereotype benchmarks. This work broadens the scope of stereotype analysis in LLMs and lays the groundwork for comprehensive cross-cultural bias evaluation.

Alberto Testoni, Barbara Plank, Raquel Fernández

Ambiguity resolution is key to effective communication. While humans effortlessly address ambiguity through conversational grounding strategies, the extent to which current language models can emulate these strategies remains unclear. In this work, we examine referential ambiguity in image-based question answering by introducing RAcQUEt, a carefully curated dataset targeting distinct aspects of ambiguity. Through a series of evaluations, we reveal significant limitations and problems of overconfidence of state-of-the-art large multimodal language models in addressing ambiguity in their responses. The overconfidence issue becomes particularly relevant for RAcQUEt-BIAS, a subset designed to analyze a critical yet underexplored problem: failing to address ambiguity leads to stereotypical, socially biased responses. Our results underscore the urgency of equipping models with robust strategies to deal with uncertainty without resorting to undesirable stereotypes.

Mahdi Zakizadeh, Mohammad Taher Pilehvar

Accurately measuring gender stereotypical bias in language models is a complex task with many hidden aspects. Current benchmarks have underestimated this multifaceted challenge and failed to capture the full extent of the problem. This paper examines the inconsistencies between intrinsic stereotype benchmarks. We propose that currently available benchmarks each capture only partial facets of gender stereotypes, and when considered in isolation, they provide just a fragmented view of the broader landscape of bias in language models. Using StereoSet and CrowS-Pairs as case studies, we investigated how data distribution affects benchmark results. By applying a framework from social psychology to balance the data of these benchmarks across various components of gender stereotypes, we demonstrated that even simple balancing techniques can significantly improve the correlation between different measurement approaches. Our findings underscore the complexity of gender stereotyping in language models and point to new directions for developing more refined techniques to detect and reduce bias.

Vera Neplenbroek, Arianna Bisazza, Raquel Fernández

Generative Large Language Models (LLMs) infer user’s demographic information from subtle cues in the conversation — a phenomenon called implicit personalization. Prior work has shown that such inferences can lead to lower quality responses for users assumed to be from minority groups, even when no demographic information is explicitly provided. In this work, we systematically explore how LLMs respond to stereotypical cues using controlled synthetic conversations, by analyzing the models’ latent user representations through both model internals and generated answers to targeted user questions. Our findings reveal that LLMs do infer demographic attributes based on these stereotypical signals, which for a number of groups even persists when the user explicitly identifies with a different demographic group. Finally, we show that this form of stereotype-driven implicit personalization can be effectively mitigated by intervening on the model’s internal representations using a trained linear probe to steer them toward the explicitly stated identity. Our results highlight the need for greater transparency and control in how LLMs represent user identity.

Nicholas Deas, Kathleen McKeown

We introduce and study artificial impressions–patterns in LLMs’ internal representations of prompts that resemble human impressions and stereotypes based on language. We fit linear probes on generated prompts to predict impressions according to the two-dimensional Stereotype Content Model (SCM). Using these probes, we study the relationship between impressions and downstream model behavior as well as prompt features that may inform such impressions. We find that LLMs inconsistently report impressions when prompted, but also that impressions are more consistently linearly decodable from their hidden representations. Additionally, we show that artificial impressions of prompts are predictive of the quality and use of hedging in model responses. We also investigate how particular content, stylistic, and dialectal features in prompts impact LLM impressions.

Abdullah Hashmat, Muhammad Arham Mirza, Agha Ali Raza

With the widespread adoption of Large Language Models (LLMs) across various applications, it is imperative to ensure their fairness across all user communities. However, most LLMs are trained and evaluated on Western centric data, with little attention paid to low-resource languages and regional contexts. To address this gap, we introduce PakBBQ, a culturally and regionally adapted extension of the original Bias Benchmark for Question Answering (BBQ) dataset. 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. We evaluate multiple multilingual LLMs under both ambiguous and explicitly disambiguated contexts, as well as negative versus non negative question framings. Our experiments reveal (i) an average accuracy gain of 12% with disambiguation, (ii) consistently stronger counter bias behaviors in Urdu than in English, and (iii) marked framing effects that reduce stereotypical responses when questions are posed negatively. These findings highlight the importance of contextualized benchmarks and simple prompt engineering strategies for bias mitigation in low resource settings.

Seorin Kim, Dongyoung Lee, Jaejin Lee

Large language models (LLMs) often exhibit societal biases in their outputs, prompting ethical concerns regarding fairness and harm. In this work, we propose KLAAD (KL-Attention Alignment Debiasing), an attention-based debiasing framework that implicitly aligns attention distributions between stereotypical and anti-stereotypical sentence pairs without directly modifying model weights. KLAAD introduces a composite training objective combining Cross-Entropy, KL divergence, and Triplet losses, guiding the model to consistently attend across biased and unbiased contexts while preserving fluency and coherence. Experimental evaluation of KLAAD demonstrates improved bias mitigation on both the BBQ and BOLD benchmarks, with minimal impact on language modeling quality. The results indicate that attention-level alignment offers a principled solution for mitigating bias in generative language models.

Elle

Reward models (RMs) are central to the alignment of language models (LMs). An RM often serves as a proxy for human preferences to guide downstream LM behavior. However, our understanding of RM behavior is limited. Our work (i) formalizes a framework for measuring the alignment of opinions captured by RMs, (ii) investigates the extent to which RMs demonstrate sociodemographic biases, and (iii) explores the effects of prompting to steer rewards towards the preferences of a target group. We study the subjective and diverse perspectives on controversial topics, which allows us to quantify \textit{RM perspectives} in terms of their opinions, attitudes, and values. We show that RMs are poorly aligned with several demographic groups and can systematically reward harmful stereotypes, and steering alone is not enough to overcome these limitations. Our findings underscore the need for more careful consideration of RM behavior in model alignment during preference learning to prevent the propagation of unwanted social biases in the language technologies that we use.

Ahmet Yavuz Uluslu, Tannon Kew, Tilia Ellendorff, Gerold Schneider, Rico Sennrich

Large language models (LLMs) often achieve high performance in native language identification (NLI) benchmarks by leveraging superficial contextual clues such as names, locations, and cultural stereotypes, rather than the underlying linguistic patterns indicative of native language (L1) influence. To improve robustness, previous work has instructed LLMs to disregard such clues. In this work, we demonstrate that such a strategy is unreliable and model predictions can be easily altered by misleading hints. To address this problem, we introduce an agentic NLI pipeline inspired by forensic linguistics, where specialized agents accumulate and categorize diverse linguistic evidence before an independent final overall assessment. In this final assessment, a goal-aware coordinating agent synthesizes all evidence to make the NLI prediction. On two benchmark datasets, our approach significantly enhances NLI robustness against misleading contextual clues and performance consistency compared to standard prompting methods.

Minh Duc Bui, Carolin Holtermann, Valentin Hofmann, Anne Lauscher, Katharina von der Wense

Dialects represent a significant component of human culture and are found across all regions of the world. In Germany, more than 40% of the population speaks a regional dialect (Adler and Hansen, 2022). However, despite cultural importance, individuals speaking dialects often face negative societal stereotypes. We examine whether such stereotypes are mirrored by large language models (LLMs). We draw on the sociolinguistic literature on dialect perception to analyze traits commonly associated with dialect speakers. Based on these traits, we assess the dialect naming bias and dialect usage bias expressed by LLMs in two tasks: association task and decision task. To assess a model’s dialect usage bias, we construct a novel evaluation corpus that pairs sentences from seven regional German dialects (e.g., Alemannic and Bavarian) with their standard German counterparts. We find that: (1) in the association task, all evaluated LLMs exhibit significant dialect naming and dialect usage bias against German dialect speakers, reflected in negative adjective associations; (2) all models reproduce these dialect naming and dialect usage biases in their decision making; and (3) contrary to prior work showing minimal bias with explicit demographic mentions, we find that explicitly labeling linguistic demographics—German dialect speakers—amplifies bias more than implicit cues like dialect usage.