Social Bias in Multilingual Language Models: A Survey

2025-08-27

AI Summary

Pretrained multilingual large language models (LLMs) perpetuate and amplify societal biases in non-English contexts, yet existing research suffers from limited language coverage, inadequate cultural appropriateness, inconsistent cross-lingual evaluation protocols, and poor generalizability of bias mitigation techniques. To address these gaps, we conduct a systematic literature review and propose the first cross-cultural bias evaluation framework tailored for multilingual settings. Our analysis exposes implicit resource-language bias in mainstream methodologies and identifies critical issues—including cultural mismatch and metric invalidation—during benchmark transfer across languages. Our contributions are threefold: (1) a dual-dimension (language–culture) principle for bias evaluation; (2) a taxonomy of technical adaptation bottlenecks in cross-lingual bias mitigation, along with empirically grounded improvement pathways; and (3) reproducible methodological guidance and empirically validated benchmark recommendations to advance fairness research in multilingual AI.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Ethics, Bias, and Fairness
• Philosophy and Ethics of AI: Bias, Fairness & Equity

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Pretrained multilingual models exhibit the same social bias as models processing English texts. This systematic review analyzes emerging research that extends bias evaluation and mitigation approaches into multilingual and non-English contexts. We examine these studies with respect to linguistic diversity, cultural awareness, and their choice of evaluation metrics and mitigation techniques. Our survey illuminates gaps in the field's dominant methodological design choices (e.g., preference for certain languages, scarcity of multilingual mitigation experiments) while cataloging common issues encountered and solutions implemented in adapting bias benchmarks across languages and cultures. Drawing from the implications of our findings, we chart directions for future research that can reinforce the multilingual bias literature's inclusivity, cross-cultural appropriateness, and alignment with state-of-the-art NLP advancements.

Problem

• Social bias in multilingual language models
• Bias evaluation and mitigation across languages
• Gaps in methodological design and cultural appropriateness

Innovation

• Extends bias evaluation to multilingual contexts
• Analyzes linguistic diversity and cultural awareness
• Proposes inclusive cross-cultural mitigation techniques