Do Multilingual Large Language Models Mitigate Stereotype Bias?

TL;DR

This study systematically compares monolingual and multilingual large language models, finding that multilingual training reduces stereotype bias and improves prediction accuracy across multiple languages.

Contribution

It provides a comprehensive evaluation of bias mitigation in multilingual LLMs using a controlled experimental setup with six models.

Findings

Multilingual models exhibit less stereotype bias than monolingual models.

Multilingual training improves prediction accuracy across languages.

Bias benchmarks translated and verified for consistency.

Abstract

While preliminary findings indicate that multilingual LLMs exhibit reduced bias compared to monolingual ones, a comprehensive understanding of the effect of multilingual training on bias mitigation, is lacking. This study addresses this gap by systematically training six LLMs of identical size (2.6B parameters) and architecture: five monolingual models (English, German, French, Italian, and Spanish) and one multilingual model trained on an equal distribution of data across these languages, all using publicly available data. To ensure robust evaluation, standard bias benchmarks were automatically translated into the five target languages and verified for both translation quality and bias preservation by human annotators. Our results consistently demonstrate that multilingual training effectively mitigates bias. Moreover, we observe that multilingual models achieve not only lower bias but also superior prediction accuracy when compared to monolingual models with the same amount of training data, model architecture, and size.