Measuring South Asian Biases in Large Language Models

2025-05-24

AI Summary

This study addresses the understudied intersectional cultural biases in large language models (LLMs) within underrepresented multilingual regions—particularly South Asia—where biases across gender, religion, and marital status remain unexamined. Method: We systematically evaluate generative biases across ten Indo-Aryan and Dravidian languages, introducing the South Asia–Culturally Anchored Multidimensional Bias Lexicon and a novel quantitative framework for measuring intersectional bias in open-ended generation tasks, coupled with a self-debiasing evaluation paradigm. Leveraging multilingual prompt engineering and cross-lingual content analysis, we assess how LLMs reproduce culturally embedded stigmas—including chastity norms and purdah practices. Contribution/Results: Empirical findings reveal that mainstream LLMs consistently amplify such local cultural stereotypes; moreover, complex self-debiasing prompts significantly outperform simplistic ones. Our work establishes the first reproducible, scalable methodology for culturally adaptive fair alignment of LLMs in linguistically diverse, low-resource settings.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Bias, Fairness & Equity

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Evaluations of Large Language Models (LLMs) often overlook intersectional and culturally specific biases, particularly in underrepresented multilingual regions like South Asia. This work addresses these gaps by conducting a multilingual and intersectional analysis of LLM outputs across 10 Indo-Aryan and Dravidian languages, identifying how cultural stigmas influenced by purdah and patriarchy are reinforced in generative tasks. We construct a culturally grounded bias lexicon capturing previously unexplored intersectional dimensions including gender, religion, marital status, and number of children. We use our lexicon to quantify intersectional bias and the effectiveness of self-debiasing in open-ended generations (e.g., storytelling, hobbies, and to-do lists), where bias manifests subtly and remains largely unexamined in multilingual contexts. Finally, we evaluate two self-debiasing strategies (simple and complex prompts) to measure their effectiveness in reducing culturally specific bias in Indo-Aryan and Dravidian languages. Our approach offers a nuanced lens into cultural bias by introducing a novel bias lexicon and evaluation framework that extends beyond Eurocentric or small-scale multilingual settings.

Problem

• Identifies cultural biases in South Asian LLM outputs across 10 languages
• Analyzes intersectional biases linked to gender, religion, and marital status
• Evaluates self-debiasing strategies for culturally specific bias reduction

Innovation

• Multilingual intersectional analysis of LLM outputs
• Culturally grounded bias lexicon construction
• Evaluation of self-debiasing strategies effectiveness