Translation studies in the age of artificial intelligence are shaped by a fundamental tension between the formalized processing of linguistic knowledge and contextual understanding. The study examines seven translation versions of a fragment from Chekhov's The Black Monk, produced by human translators in different historical periods and by contemporary AI systems. The analysis focuses on the operational boundaries of large language models in handling culturally marked elements, particularly a biblical allusion and a diminutive form, which require three successive levels: semantic identification, pragmatic interpretation, and cultural decision-making. The methodology combines a diachronic analysis of translation theory and machine translation technologies with a comparative analysis of the versions, based on a three-level operational model of understanding and a method of gradient instructing of the AI. The novelty lies in the empirical substantiation of the functional boundary between algorithmic knowledge processing and human understanding. The results show that large language models achieve high efficiency in the search for and combination of linguistic means, whereas pragmatic interpretation remains an exclusively human competence. Each technological leap expands the boundaries of knowledge processing without eliminating the contradiction between knowledge and understanding. As technology approaches the threshold of human understanding, intrinsic differences become more pronounced. Attempts to replace human understanding with technological knowledge encounter epistemological limitations. A promising development lies in abandoning narratives of replacement and shaping a paradigm of symbiosis that combines machine efficiency with human advantages in contextual understanding and value creation, supported by the institutional entrenchment of ethical principles in translation education and professional codes of conduct.
Shi Qiu (2026) studied this question.