Whose Morality Do They Speak? Unraveling Cultural Bias in Multilingual Language Models

2024-12-25

AI Summary

Multilingual large language models (LLMs) exhibit English-centric bias in moral judgment, undermining cross-cultural fairness and ethical reliability. Method: We conduct the first systematic, multilingual evaluation of moral foundations—Care, Equality, Fairness, Loyalty, Authority, and Purity—across eight languages using an adapted multilingual Moral Foundations Questionnaire (MFQ-2), applied to GPT-3.5-Turbo, GPT-4o-mini, Llama 3.1, and MistralNeMo. Contribution/Results: We find that most models strongly align with English cultural norms, with only marginal evidence of local cultural adaptation; model moral preferences are directly shaped by the cultural composition of their training data, refuting the “moral universality” hypothesis. This work provides the first empirical demonstration of cultural hegemony in multilingual LLMs, establishing a critical benchmark and methodological foundation for developing fair, interpretable, and culturally grounded AI ethics frameworks.

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: Accountability, Transparency, and Ethics for personalization
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Large language models (LLMs) have become integral tools in diverse domains, yet their moral reasoning capabilities across cultural and linguistic contexts remain underexplored. This study investigates whether multilingual LLMs, such as GPT-3.5-Turbo, GPT-4o-mini, Llama 3.1, and MistralNeMo, reflect culturally specific moral values or impose dominant moral norms, particularly those rooted in English. Using the updated Moral Foundations Questionnaire (MFQ-2) in eight languages, Arabic, Farsi, English, Spanish, Japanese, Chinese, French, and Russian, the study analyzes the models' adherence to six core moral foundations: care, equality, proportionality, loyalty, authority, and purity. The results reveal significant cultural and linguistic variability, challenging the assumption of universal moral consistency in LLMs. Although some models demonstrate adaptability to diverse contexts, others exhibit biases influenced by the composition of the training data. These findings underscore the need for culturally inclusive model development to improve fairness and trust in multilingual AI systems.

Problem

• Cultural Bias
• Ethical Judgement
• Language Models

Innovation

• Multilingual Language Models
• Cultural Bias in Ethics
• Moral Judgment Variation