MORPHOLOGICAL ENCODING OF GENDER BIAS IN AI TRANSLATION: A STUDY OF URDU VERB AGREEMENT IN THE TRANSLATION OF ENGLISH PROFESSIONAL NOUNS

Abstract

The study aims to examine the morphological encoding of gender bias in Urdu verb agreements by AI when translating English professional nouns. The study adopted the mixed method approach for analyzing the data. The quantitative phase leads to grouping of three categories: male-stereotyped professional nouns, female-stereotyped professional nouns and gender-neutral professional nouns, in which gender neutral/mixed professional nouns outnumber the other two categories. The findings of the qualitative phase showed that AI tools (ChatGPT and Gemini) reproduce the male-stereotyped professional nouns by consistently using the masculine form during translation. On the other hand, slight variations were seen in the encoding of gender markers in Urdu agreements when translating the nouns of the female-stereotyped category. ChatGPT inconsistently included both gender while Gemini consistently used the masculine verb form, except with ‘nurse’ and ‘teacher’. In contrast, despite the absence of an explicit gender biased prompt, both tools consistently used masculine verb forms with professional nouns under the gender neutral/mixed professional nouns category.