CONTEXTUAL RECOGNITION OF POLYSEMANTIC WORDS SEMANTICS BY MACHINE TRANSLATION SYSTEMS

Abstract

Тhis paper discusses the basic principles underlying the work of machine translation systems and presents a comparative analysis of texts of diff erent styles to see how machine translation systems cope with fi nding the contextual meaning of a word used in fundamentally diff erent contexts. Four machine translation systems, Google Translate, Яндекс.Переводчик, DeepL and PROMT were chosen for this study. The article presents main approaches to machine translation, a brief description of the operation principles of the four mentioned above machine translation systems, examples of using the same words in diff erent contexts, as well as variants of their translation into Russian by the above-mentioned machine translation systems. In case of incorrectly recognized meanings an extended context (paragraph instead of sentence) was added to the translator program, as it was assumed that this could improve context recognition and, as a consequence, the number of correct translations of diff erent meanings of the selected words. The hypothesis was that expanding the context in case of an unrecognized meaning should help the system to recognize it. Using a continuous sampling method we have selected forty sentences in which twenty of the same words but with diff erent meaning are used in diff erent contexts. Selecting the sample sentences for analysis we did not stick to specifi c topics. The main principle of selecting the words presented in the article was polysemy. At the end of the article we discuss the comparative analysis results and summarize main aspects of this study. We believe that the experiment described in this paper can serve as a good reference for further research in this area, in particular to understand how context is taken into account and how machine translation systems can be improved.