The possibilities of machine translation in the era of total digitalization seem limitless, however, for poetic text, electronic translation technology has limited application. For the analysis and comparison of machine and literary translation, the text of Coleridge’s 1798 poem “The Rime of the Ancient Mariner” was taken for it uses the theme of the sea, which is extremely important in the poet’s work. This poem was translated by N.S. Gumilev in 1919: the translation was made more than a hundred years after the date of the creation of the original, and the author of the translation had to take into account the context of the creation of the work. The translation generated by modern technologies involves the work of a self-learning neural network which is based on corrections made by users. Online translation technology allows users who believe that the translation is inaccurate to either choose from several proposed options or make their own. On the one hand, due to the constant work of users and their corrections, the translation has the property of self-learning and implies that each subsequent translation is better than the previous one. On the other hand, for texts created 200 years ago, this technology may turn out to be unproductive: most texts translated using an electronic translator are modern, they take into account current realities and contexts of use. To test this hypothesis, an experiment was conducted: translations of the same text by the translator-poet of the Silver Age and the modern electronic translator Google were compared. An electronic translation cannot explicitly take into account the rhythmic structure of the text, the assonances and alliterations used in it, etc., however, it can convey the general meaning of the text, its content. It seems interesting that in the text generated by electronic translation, new semantic connections arise which are absent in the original and in the translation of the poet.
Keywords: Translation, neural network, digitalization, Gumilev, Coleridge