Digital translation as a factor of variability in single-text translations: A case study of Siberian mining poetry

Abstract

This paper identifies the factors determining variability in the creation of secondary (translated) texts. The study employs a comparative analysis of machine translation (MT) results produced by three neural networks, ChatGPT, DeepSeek, and Gemini, to evaluate variants of secondary texts based on their degree of equivalence to the source. The focus is on three poems by Siberian authors, each concerning the labor of miners. The analysis establishes that digital translators significantly influence text variability, primarily due to discursive elements such as metaphors, idioms, and culturally specific professional vocabulary, which act as “points of tension.” The study reveals distinct stylistic profiles for each engine. ChatGPT frequently employs poetic modulation and compensation, achieving a balance between accuracy and artistic expression. DeepSeek prioritizes technically correct but literal solutions, often compromising the poetic nature of the work. Gemini tends toward calquing and transliteration, which reduces the emotional and cultural resonance of the target text. The findings indicate that the highest levels of expressiveness and accuracy are achieved through modulation, functional equivalence, and adaptation. Ultimately, the choice of a specific neural system confirms the role of the digital translator as a key determinant in translation variation and the overall quality of the poetic secondary text.