Efficient English-Vietnamese Medical Machine Translation: Insights from the VLSP 2025 Shared Task
Main Article Content
Abstract
Abstract: In this paper, we present a comprehensive overview of the VLSP 2025 Medical Machine
Translation Shared Task, which focuses on English-Vietnamese translation in the medical domain
using small language models (SLMs). The shared task provides a large-scale, high quality parallel
corpus and encourages participants to develop efficient, domain-adapted translation systems under
resource constraints. We summarize the dataset construction, evaluation protocols, and model constraints, and review the diverse strategies adopted by participating teams-including parameter efficient fine-tuning, bidirectional training, retrieval augmented generation, and reinforcement learning
with reward optimization. Our analysis highlights the strengths and limitations of SLM-based approaches for medical translation, discusses key findings from the competition, and outlines future
research directions for building scalable, accurate, and practical machine translation systems in specialized domains.