Multilingual LLM Prompting Strategies for Medical English-Vietnamese Machine Translation

2025-09-19

AI Summary

This study addresses the challenges of terminology inconsistency and poor domain adaptation in English–Vietnamese medical machine translation (En-Vi MT), where Vietnamese is a low-resource language. We propose a terminology-aware prompting framework tailored for multilingual large language models (LLMs). Our method integrates Meddict—a specialized medical dictionary—to construct terminology-enhanced prompts and incorporates embedding-based dynamic example retrieval, supporting zero-shot, few-shot, and dictionary-augmented prompting paradigms. Experiments demonstrate substantial improvements in translation accuracy and terminology consistency, outperforming baseline models by +4.2–7.8 BLEU across multiple medical test sets, with particularly strong gains in zero-shot settings. To our knowledge, this is the first systematic validation of the synergistic effectiveness of terminology guidance and embedding-based retrieval for optimizing low-resource medical translation. The framework provides a reusable technical pathway for professional translation in resource-scarce languages.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Medical English-Vietnamese machine translation (En-Vi MT) is essential for healthcare access and communication in Vietnam, yet Vietnamese remains a low-resource and under-studied language. We systematically evaluate prompting strategies for six multilingual LLMs (0.5B-9B parameters) on the MedEV dataset, comparing zero-shot, few-shot, and dictionary-augmented prompting with Meddict, an English-Vietnamese medical lexicon. Results show that model scale is the primary driver of performance: larger LLMs achieve strong zero-shot results, while few-shot prompting yields only marginal improvements. In contrast, terminology-aware cues and embedding-based example retrieval consistently improve domain-specific translation. These findings underscore both the promise and the current limitations of multilingual LLMs for medical En-Vi MT.

Problem

• Evaluating multilingual LLM prompting for medical English-Vietnamese translation
• Addressing low-resource language challenges in healthcare communication
• Improving domain-specific translation through terminology-aware strategies

Innovation

• Multilingual LLM prompting strategies
• Dictionary-augmented terminology-aware cues
• Embedding-based example retrieval method