From Neural Machine Translation to Generative AI: A Systematic Review of English Translation Teaching in Higher Education

Abstract

This systematic review aims to synthesise empirical research on GenAI- and NMT-assisted English translation teaching in higher education from 2015 to 2025 and to evaluate cutting-edge teaching outcomes and emerging pedagogical implications. Drawing on 31 studies identified through PRISMA procedures, the review maps the rapid evolution from NMT-based post-editing and output comparison to GenAI-supported feedback, prompting, and collaborative revision. Organised through a task-, course-, and model-level framework and informed by Cognitive Load Theory and Sociocultural Theory, the synthesis shows that NMT and GenAI-supported feedback can improve efficiency, error awareness, and terminological consistency in informational and general academic texts. However, evidence remains conditional and sometimes contradictory: effects on cognitive load, learner motivation, and translation quality vary according to task type, tool quality, learner proficiency, self-regulation, and instructional scaffolding. The review also identifies risks of overreliance, superficial revision, weak prompt or post-editing competence, assessment ambiguity, and unequal access. It therefore argues for sustained, process-oriented pedagogy in which teachers scaffold comparison, revision justification, feedback evaluation, ethical disclosure, and human oversight. Ultimately, the study argues that translation education must strengthen human agency within AI-rich learning, prioritising contextual understanding, value judgement, and intercultural decision-making as core competencies.