Abstract
The presented article examines the features of coherence and cohesion means in Russian poetry and the related issue of preserving coherence and cohesion of these texts in their neural machine translation into English. The relevance of the work is determined by the specifics of the errors made by automatic translation systems in comparison with the errors occurring in traditional ‘human’ translation as well as by the novelty and insufficient study of machine translation of poetry. The theoretical section introduces the main categories of the text – cohesion and coherence – and discusses their specifics in poetry. The conclusions made are: a lower extent of coherence and cohesion in poetic texts, specific information structure, and the predominance of coordinating conjunctions within a sentence. Also demonstrated is the exceptional importance of such leitmotivs as lexical and synonymous repetitions and the cross-cutting use of one or more stylistic devices for preserving the coherence of poetic texts. In the second section, a neural network machine translation of six poetic works belonging to the Silver Age was done using the ‘Yandex Translate’ service. Next, the means of coherence and cohesion employed in the original texts and those that are absent in the translation texts were considered line by line. Thus, the validity of the main conclusions on the specifics of coherence and cohesion means in poetic texts is demonstrated. It was also found that in machine translation, the most frequent errors are those mainly responsible for reducing coherence and cohesion of the text: errors associated with the transmission of various leitmotivs (repetitive lexemes as well as individual semantic components, morphemes and stylistic devices). The specificity of these errors for automatic translation and their significance in the context of poetic translation based on automatically generated interlinear translation were discussed.