GENDER BIAS IN MACHINE TRANSLATION DATABASE: CASE STUDY OF GOOGLE TRANSLATE FROM ENGLISH TO SERBIAN
GENDER BIAS IN MACHINE TRANSLATION DATABASE: CASE STUDY OF GOOGLE TRANSLATE FROM ENGLISH TO SERBIAN

Summary/Abstract: This study aims at analyzing gender bias in machine translation (Google Translate) from English into Serbian. We used the descriptive method of analyzing the machine translation (MT) from English into Serbian through gathered samples of English-to-Serbian translations of sentences: She is (name of a profession) and He is… (name of a profession). The results acquired prove the hypothesis that Google Translate offers biased translation from English to Serbian relying on the biased database it draws on. The second hypothesis sheds light on the influence of gender-specific adjectives such as beautiful, nice (feminine prevalence), and robust, aggressive (masculine prevalence) that qualify the names of professions dominantly by feminine or masculine gender, respectively. The third hypothesis of the research undertaken argues that certain professions are prevailingly translated by masculine gender even if they are preceded by a feminine personal pronoun (coach, economist, doctor, engineer, chemist, and lawyer) whereas certain names of professions such as nurse and maid are translated by a feminine gender even if they are preceded by a masculine personal pronoun. In addition, we hypothesize that the main reason for the gender bias in the machine translation from English to Serbian lies in the biased database used for the retrieval of data and certain algorithm improvements might mitigate the lack of gender sensitivity in the MT.

• Details
• Contents