DETECTING AND MITIGATING OF BIASNESS IN MACHINE TRANSLATION

Abstract

Machine translation (MT) systems have become indispensable tools for cross-lingual communication, yet their ability to accurately convey gender-specific nuances remains a significant challenge. This study provides a comprehensive evaluation of popular MT systems like Google Translate and Bing Microsoft Translator, focusing on the Hindi-to English language pair, where grammatical gender plays a crucial role. We investigate the systems’ performance in translating gendered vocabulary, encom passing personality adjectives, professions, nouns, and pronouns. Analyzing the preva lence of female, male, and neutral forms in the translated output, we assess the systems’ capacity to preserve the original gender intent and identify potential biases. Our research emphasizes professional contexts, where gender bias can have profound implications for individuals and society. We scrutinize how MT systems handle gendered language nuances in these settings, examining whether they inadvertently memorialize stereotypes or introduce discriminatory language. Through rigorous statistical analysis of translation outputs, we aim to quantify bias in these systems. By identifying specific areas where biases occur, we can provide valuable insights to guide the development of more equitable and inclusive translation technologies. This study contributes to a deeper understanding of the capabilities and limitations of current MT systems in accurately representing gender-specific language constructs. Ultimately, our goal is to foster more nuanced and sensitive cross-lingual communication by highlighting areas for improvement and promoting the development of MT tools that respect and preserve the diversity of gendered language.