Findings of the WMT 2024 Shared Task on Translation

This technical report presents the findings from the Ninth Conference on Machine Translation (WMT 2024) Shared Task, which focused on developing machine translation systems for low-resource languages in Spain, specifically Spanish–Aragonese, Spanish–Aranese, and Spanish–Asturian. A total of 17 teams participated, submitting 87 systems. The baseline system utilized was Apertium, a rule-based machine translation system. The report discusses the results of the submitted systems, noting the strengths of both neural and rule-based approaches. It outlines the objectives of the shared task, which included pushing the boundaries of machine translation development with limited resources and exploring transferability among low-resource Romance languages. The findings indicate that while some systems achieved statistically significant improvements over the baseline, the differences in evaluation scores were modest, suggesting that rule-based systems still play a crucial role in these translations. The report further details the submission categories and the resources provided to participants.