MPO: Multilingual Safety Alignment via Reward Gap Optimization
This addresses safety alignment for multilingual AI applications, but it is incremental as it builds on existing preference learning methods.
The paper tackled the problem of multilingual safety alignment in large language models, which existing monolingual methods struggle with due to noisy data, by introducing MPO to minimize reward gaps between English and target languages, achieving improved safety across three LLMs without degrading utility.
Large language models (LLMs) have become increasingly central to AI applications worldwide, necessitating robust multilingual safety alignment to ensure secure deployment across diverse linguistic contexts. Existing preference learning methods for safety alignment, such as RLHF and DPO, are primarily monolingual and struggle with noisy multilingual data. To address these limitations, we introduce Multilingual reward gaP Optimization (MPO), a novel approach that leverages the well-aligned safety capabilities of the dominant language (English) to improve safety alignment across multiple languages. MPO directly minimizes the reward gap difference between the dominant language and target languages, effectively transferring safety capabilities while preserving the original strengths of the dominant language. Extensive experiments on three LLMs, LLaMA-3.1, Gemma-2 and Qwen2.5, validate MPO's efficacy in multilingual safety alignment without degrading general multilingual utility.