Toward a Cognitive Architecture of Human–AI Collaboration in AI-Assisted Translation

Abstract

The growing adoption of generative artificial intelligence (AI) in higher education has transformed translation from an individual cognitive activity into a collaborative process in which learners continuously interact with intelligent technologies. Although AI-assisted translation has received increasing scholarly attention, existing research has primarily focused on translation quality, technology acceptance, and ethical issues, offering limited understanding of the cognitive processes underlying human–AI collaboration during translation. This study investigates the cognitive architecture of AI-assisted translation by examining how cognitive work is distributed between learners and AI, how learners regulate AI-generated outputs, and how they reconstruct translations while maintaining human agency. A qualitative multiple case study was conducted with five undergraduate students enrolled in a Translating General Text course in the English Literature Study Program at Universitas Negeri Makassar, Indonesia. Data were collected through AI-assisted translation tasks, classroom observations, and semi-structured interviews and analyzed using reflexive thematic analysis. The findings reveal three interconnected cognitive dimensions. First, distributed cognitive processing enables AI to support lexical retrieval, semantic interpretation, and syntactic reformulation during translation. Second, metacognitive regulation allows learners to evaluate, accept, modify, reject, and regenerate AI-generated outputs according to contextual and semantic requirements. Third, human reconstruction demonstrates that learners integrate AI-generated suggestions with their own linguistic knowledge to produce coherent and contextually appropriate translations. Cross-case analysis further indicates that AI-assisted translation operates as a recursive human–AI cognitive system in which cognitive resources are shared while executive control remains with the learner. The study proposes an empirical cognitive architecture integrating distributed cognition and metacognitive regulation, extending current understandings of human–AI collaboration in translation education and providing pedagogical implications for fostering critical judgment, learner autonomy, and responsible AI integration.