Bringing Context into MT Evaluation: Translator Training Insights from the Classroom

The role of technology in translator training has become increasingly significant as machine translation (MT) evolves at a rapid pace. Beyond practical tool usage, training must now prepare students to engage critically with MT outputs and understand the socio-technical dimensions of translation. Traditional sentence-level MT evaluation, often conducted on isolated segments, can overlook discourse-level errors or produce misleadingly high scores for sentences that appear correct in isolation but are inaccurate within the broader discourse. Document-level MT evaluation has emerged as an approach that offers a more accurate perspective by accounting for context. This article presents the integration of context-aware MT evaluation into an MA-level translation module, in which students conducted a structured exercise comparing sentence-level and document-level methodologies, supported by reflective reporting. The aim was to familiarise students with context-aware evaluation techniques, expose the limitations of single-sentence evaluation, and foster a more nuanced understanding of translation quality. This study provides methodological insights for incorporating MT evaluation training into translator education and highlights how such exercises can develop critical awareness of MT’s contextual limitations. It also offers a framework for supporting students in building the analytical skills needed to evaluate MT output in professional and research settings.