Challenges of Phraseological Translation in the Era of AI: Georgian-Russian MT Perspectives and the Role of Intermediary Languages

Abstract

This article examines the core challenges of translating phraseological units in Machine Translation (MT) and Artificial Intelligence (AI), focusing on the Georgian-Russian language pair. Phraseologisms, characterised by cultural specificity and semantic indivisibility, pose a significant hurdle for automated systems that often rely on morphological and syntactic analysis rather than deep semantic synthesis. This research adopts a complex comparative-experimental methodology, analysing translations of idiomatic expressions from Mikhail Javakhishvili’s novel Arsena Marabdeli within the digital framework of the Electronic Text Analysis Group at Tbilisi State University (ETAG-TSU). By comparing professional human translations with machine-generated outputs, this study identifies critical failures in AI systems, including literalism, an inability to distinguish homonyms, and misinterpretation of cultural and biblical realia. Furthermore, the paper investigates the computational stylistics of phraseological transformations in literary prose, highlighting how algorithmic models struggle with authorial style and figurative nuances. A key finding is that English serves as a “pivot” or intermediary language in the Georgian-Russian MT process, introducing additional phonetic and lexical errors. The paper concludes that, for AI to achieve functional proficiency in translating idioms, it must be equipped with extensive electronic corpora, Parallel Idiomatic Corpora, and Specialised Knowledge Graphs in which phraseologisms are treated as independent, indivisible semantic components.