Exploring the Impact of Training Data Distribution and Subword Tokenization on Gender Bias in Machine Translation
This work addresses gender bias in machine translation, an important fairness issue, but is incremental as it builds on existing tokenization and bias research.
The study investigated how training data distribution and subword tokenization affect gender bias in machine translation, finding that gender-form imbalance in the corpus is a major contributor to bias, and fine-tuning the token embedding layer reduced the gender prediction accuracy gap without harming translation quality.
We study the effect of tokenization on gender bias in machine translation, an aspect that has been largely overlooked in previous works. Specifically, we focus on the interactions between the frequency of gendered profession names in training data, their representation in the subword tokenizer's vocabulary, and gender bias. We observe that female and non-stereotypical gender inflections of profession names (e.g., Spanish "doctora" for "female doctor") tend to be split into multiple subword tokens. Our results indicate that the imbalance of gender forms in the model's training corpus is a major factor contributing to gender bias and has a greater impact than subword splitting. We show that analyzing subword splits provides good estimates of gender-form imbalance in the training data and can be used even when the corpus is not publicly available. We also demonstrate that fine-tuning just the token embedding layer can decrease the gap in gender prediction accuracy between female and male forms without impairing the translation quality.