Significance of Chain of Thought in Gender Bias Mitigation for English-Dravidian Machine Translation

TL;DR

The paper addresses gender bias in English-to-Telugu/Kannada MT arising from gender inflection and data bias. It evaluates bias in Google Translate and ChatGPT and applies Chain-of-Thought prompting with language-specific strategies to generate more gender-neutral translations, complemented by API-based tests. Results reveal that bias is highly language- and domain-dependent: Kannada often achieves near-neutral translations with CoT, while Telugu shows variable improvements and residual bias in some domains. The work highlights the need for language-specific prompting and data preparation to advance fairness in multilingual MT systems and points to directions for future research and practical deployment.

Abstract

Gender bias in machine translation (MT) sys- tems poses a significant challenge to achieving accurate and inclusive translations. This paper examines gender bias in machine translation systems for languages such as Telugu and Kan- nada from the Dravidian family, analyzing how gender inflections affect translation accuracy and neutrality using Google Translate and Chat- GPT. It finds that while plural forms can reduce bias, individual-centric sentences often main- tain the bias due to historical stereotypes. The study evaluates the Chain of Thought process- ing, noting significant bias mitigation from 80% to 4% in Telugu and from 40% to 0% in Kan- nada. It also compares Telugu and Kannada translations, emphasizing the need for language specific strategies to address these challenges and suggesting directions for future research to enhance fairness in both data preparation and prompts during inference.