TRANSLATION OF ETYMOLOGICAL CALQUES IN MACHINE TRANSLATION SERVICES GOOGLE TRANSLATE, DEEPL, CHATGPT

This article addresses the issue of translating etymological loan translations (calques) using machine translation services. The object of the study is the process of machine translation of etymological calques by contemporary automated translation systems. The subject of the research is the analysis of the methods used to render etymological calques in machine translation systems. The primary aim of the study is to examine the translation of 200 sentences using the machine translation services DeepL, ChatGPT, and Google Translate and to identify the errors made by these systems in translating etymological calques. The study clarifies the concepts of calque and literal translation, characterizes the operating principles of machine translation services, compiles a corpus of etymological calques for comparative analysis, conduct a comparative evaluation of translations produced by DeepL, Google Translate, and ChatGPT, and identifies the most common translation errors and difficulties with rendering etymological calques. The findings indicate that all three systems demonstrated a relatively high level of translation accuracy. ChatGPT achieved the highest performance, producing 164 correct translations out of 200, which suggests a superior ability to account for contextual and semantic nuances. Google Translate ranked second, while DeepL showed a slightly lower result, with 146 correct translations. None of the systems achieved perfect accuracy. The number of translation errors amounted to 36 for ChatGPT, 45 for Google, and 54 for DeepL. The analysis demonstrates that, despite the considerable progress made in modern machine translation technologies, the adequate rendering of etymological calques remains a challenging task. All of the selected services employ algorithms capable of recognizing context and reducing the likelihood of overly literal translation. Nevertheless, the overall quality of translation depends on the complexity of the expression, the context, and the availability of relevant training corpora. The results confirm that machine translation cannot fully replace human post-editing, particularly in the translation of phraseological units, idioms and culturally specific expressions as well as calques.