Combating Gender Biases from Source Language in Machine Translation

Effective Chinese language translations that communicate clear and unbiased messages form the core fabric of peaceful relationships among people in society. Existing machine translations models like Google Translate, DeepL, Youdao Translate or even LLM sometimes generate Chinese-to-English translations filled with prejudice against females. The prejudice-prone Chinese-to-English translations are inherited from some terms in the source texts which are merely norms inherited from Chinese history and may not indicate female discrimination in terms of context. Such a gap must be eradicated to minimize gender biases and gender disparity in Chinese-to-English translation. Addressing the aforementioned gap, this study employs deconstructive translation theory to demonstrate the translational phenomenon of the remaining unnecessary female discrimination expression from the source Chinese to the target English language through the lens of feminism. The study also proposes an improved model of machine translation to hi-jack the gender-biased expression by identifying and modifying the gender-biased expressions in Chinese-to-English texts from a context perspective. The findings and application of the research will improve the well-being of females by preventing the cultural flow of gender biases, and contribute to creating a more amicable language environment for females.