Expert analysis of the results of neural machine translation of popular science texts

Abstract

The main focus of this article is on the use of neural machine translation systems for translation of popular science texts from English into Russian and German, as well as their operating principles, advantages, and disadvantages. In the course of the research, the features of translating popular science texts were analyzed; the results of automated evaluation of NMT were considered; and the expert analysis of the NMT results was conducted. In addition, typical errors, current NMT systems make most often, were determined. As a part of the study, an analysis of the results of translation performed using Google Translate and DeepL Translate NMT systems into Russian and German was conducted. Thus, the features of translation of popular science texts were studied, which include: maintaining the simplicity of scientific information presentation, maintaining imagery in stylistic figures of speech, as well as maintaining the level of their expressiveness, proper translation of terminology, and specialized scientific vocabulary. The results of the study were presented in the form of diagrams, which showed that the use of NMT for popular science literature translation shows high effectiveness. Based on the analysis, it was revealed that the systems are generally effective at preserving the style and logic of the narrative. Reoccurring errors, such as grammatical, lexical, logical and spelling errors made by NMT systems during translation were identified. It should be noted that a full automatic assessment of the translation quality of NMT systems allows diagnosing both the errors of these systems and the shortcomings of machine translation programs, which in the future will allow solving these problems and improving these systems.