CURE-Med: Curriculum-Informed Reinforcement Learning for Multilingual Medical Reasoning

2026-01-19

AI Summary

This study addresses the poor performance of large language models in medical reasoning tasks for low-resource languages, where maintaining logical correctness and cross-lingual consistency remains challenging. To this end, the authors introduce CUREMED-BENCH, the first high-quality multilingual medical reasoning benchmark covering 13 languages—including low-resource ones such as Amharic and Yoruba—and propose the CURE-MED framework. This framework integrates code-switching-aware supervised fine-tuning, grouped relative policy optimization, and curriculum-guided reinforcement learning to jointly enhance medical logical accuracy and multilingual stability. Experimental results demonstrate that the approach achieves language consistency and logical correctness rates of 85.21%/54.35% on 7B models and 94.96%/70.04% on 32B models, substantially outperforming existing baselines.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

While large language models (LLMs) have shown to perform well on monolingual mathematical and commonsense reasoning, they remain unreliable for multilingual medical reasoning applications, hindering their deployment in multilingual healthcare settings. We address this by first introducing CUREMED-BENCH, a high-quality multilingual medical reasoning dataset with open-ended reasoning queries with a single verifiable answer, spanning thirteen languages, including underrepresented languages such as Amharic, Yoruba, and Swahili. Building on this dataset, we propose CURE-MED, a curriculum-informed reinforcement learning framework that integrates code-switching-aware supervised fine-tuning and Group Relative Policy Optimization to jointly improve logical correctness and language stability. Across thirteen languages, our approach consistently outperforms strong baselines and scales effectively, achieving 85.21% language consistency and 54.35% logical correctness at 7B parameters, and 94.96% language consistency and 70.04% logical correctness at 32B parameters. These results support reliable and equitable multilingual medical reasoning in LLMs. The code and dataset are available at https://cure-med.github.io/

Problem

• multilingual medical reasoning
• large language models
• language consistency
• logical correctness
• healthcare applications

Innovation

• curriculum-informed reinforcement learning
• multilingual medical reasoning
• code-switching-aware fine-tuning
• Group Relative Policy Optimization
• CUREMED-BENCH