This research breaks down language barriers in healthcare by efficiently adapting medical LLMs to 50 diverse languages, including low-resource ones.

Key Innovations:

• Created high-quality multilingual medical datasets to address data scarcity
• Developed a novel Mixture of Language Family Experts approach for efficient scaling
• Analyzed internal information flow of LLMs to improve cross-lingual generalization
• Significantly enhanced accessibility of AI healthcare tools for underserved linguistic communities

Why It Matters: This work directly addresses global healthcare inequities by making medical AI accessible across language barriers, potentially improving healthcare outcomes for billions of people worldwide who don't speak dominant languages.

Efficiently Democratizing Medical LLMs for 50 Languages via a Mixture of Language Family Experts