Intelligent translation systems: how to reproduce Ukrainian feminine forms in English

Abstract The paper deals with feminine units in Ukrainian and the problem of their machine translation to and from English. The purpose of the paper is to study how feminine units are translated from Ukrainian into English and vice versa by machine systems. To solve this problem, a qualitative comprehensive analysis was applied using a corpus-based approach and quantitative data verification. To study this issue, a corpus bitext was compiled consisting of an array of parallel texts in Ukrainian and English. Texts from mass media discourse have also been studied. Three intelligent translation systems were tested: Google Translate, DeepL, and GPT. The relevance of the study is also reinforced by the fact that it is based on the rules of the new Ukrainian orthography of 2019, which regulates the formation of feminine nouns in the modern Ukrainian language. The paper analyses and summarizes the types of errors (lexical-semantic and stylistic) in machine translation of gender-specific constructions. The research also contributes to improving the quality of machine translation of feminine units offering the practical recommendations for rendering Ukrainian feminitives into English and English constructions that may have masculine and feminine equivalents into Ukrainian.