Some Weaknesses of Modern Machine Translation (by Example of Google Translate Web Service)

Abstract

Some weaknesses of machine translation carried out by means of neural networks (“neural translation”) are considered. The relevance of this topic is determined by the significant popularity of the relevant web services among translators, teachers, researchers, students and others who are interested in the introductory or urgent transcoding of the text from one language to another. The results of the experiment are presented: well-known web service “Google Translate” was offered to translate into English several dozens of Russian-language terms and concepts related to philology and pedagogy. Recommendations are given to correct some errors and inaccuracies made by the web service. The authors come to the conclusion that the use of neural translation has led to a significant improvement in the quality of services provided by the web service “Google Translate.” It is noted that this service still makes mistakes (semantic and stylistic) when translating the following categories of vocabulary: compound terms and concepts; lexical units and phrases that do not have clear equivalents in the English language; two words with the same English equivalent; abbreviations; author’s occasionalisms. It is reported that, according to the authors, such errors are made by other popular web services (“Yandex.Translator,” “Translate.ru” etc.)