False cognates are pairs of words in two languages that have a similar structure and sound, but a different meaning. Because of this, they are known to cause confusion among language learners and human translators. Hence, it can be expected for false cognates to pose challenges to machine translators as well. This master’s thesis focuses on English-Russian, English-Slovenian, and Russian-Slovenian false friends and their translations in context using three different neural machine translators: Google Translate, DeepL Translate, and ChatGPT. False cognates were extracted from Pahotin’s (2006) dictionary of English-Russian and Russian-English false cognates. Definitions in English were taken from the following dictionaries: Cambridge Learner’s Dictionary, Cambridge English Dictionary, Oxford Advanced Learner’s Dictionary, and Collins Dictionary. Definitions in Russian were taken from Большой толковый словарь русского языка (The Large Explanatory Dictionary of the Russian Language) and occasionally from Викисловарь (Wiktionary), while the Slovenian definitions were taken from the Dictionary of Standard Slovenian. A total of 50 false cognates were analysed across the English-Russian, English-Slovenian, and Russian-Slovenian pairs, resulting in 150 pairings. Between English and Russian, there were 28 absolute false cognates, 21 partial false cognates, and 1 cognate. Between English and Slovenian, there were 18 absolute false cognates, 31 partial false cognates, and 1 cognate. Between Russian and Slovenian, there were 13 absolute false cognates, 29 partial false cognates, and 8 cognates. The data set included a total of 178 translations in Russian and Slovenian. Among the 178 translations, there were 58 (33%). Among the 58 mistranslated sentences, 26 (45%) were translations into Russian, while 32 (55%) were translations into Slovenian. Among the 58 mistranslated sentences, 16 (28%) were produced by Google Translate, while both DeepL Translate and ChatGPT produced 21 (36%) each.
|
