Художественный текст, художественный перевод и искусственный интеллект

The article discusses the potential possibility of training neuronet to translate literary texts. Evaluating the quality of literary machine translation, researchers in most cases point to such shortcomings as incorrect choice of lexical correspondences, grammatical errors, inability to understand the meaning of phraseological units and figurative expressions and to find the corresponding units of the target language, and stylistic errors. All this characterizes the current state of machine translation as such, regardless of the type of text being translated, but it is especially detrimental to literary text. However, the inability of the neuronet to fully translate literary texts is not due to these shortcomings, which, judging by the pace of development of machine translation, can soon be eliminated. The specificity of literary translation is determined by the properties of the literary text itself, which a computer program cannot perceive and reproduce due to its inherent conceptual limitations. These properties include emotional insight and emotional impact on the reader, the presence of such non-formalized characteristics as atmosphere, tone, the use of allusions, appeal to personal associations, and much more, which, by definition, is inaccessible to a neural network that has neither personality and background knowledge nor the ability to be emotionally engaged