Reducing Gender Bias in Machine Translation through Counterfactual Data Generation
This work addresses gender bias in machine translation, which is a critical fairness issue for users of translation systems, but it is incremental as it builds on existing methods.
The paper tackles gender bias in neural machine translation by supplementing a handcrafted balanced dataset with random base corpus samples to reduce catastrophic forgetting, and proposes a domain-adaptation technique using counterfactual data generation to improve accuracy on the WinoMT test set for English to French, Spanish, and Italian without significant quality loss.
Recent advances in neural methods have led to substantial improvement in the quality of Neural Machine Translation (NMT) systems. However, these systems frequently produce translations with inaccurate gender (Stanovsky et al., 2019), which can be traced to bias in training data. Saunders and Byrne (2020) tackle this problem with a handcrafted dataset containing balanced gendered profession words. By using this data to fine-tune an existing NMT model, they show that gender bias can be significantly mitigated, albeit at the expense of translation quality due to catastrophic forgetting. They recover some of the lost quality with modified training objectives or additional models at inference. We find, however, that simply supplementing the handcrafted dataset with a random sample from the base model training corpus is enough to significantly reduce the catastrophic forgetting. We also propose a novel domain-adaptation technique that leverages in-domain data created with the counterfactual data generation techniques proposed by Zmigrod et al. (2019) to further improve accuracy on the WinoMT challenge test set without significant loss in translation quality. We show its effectiveness in NMT systems from English into three morphologically rich languages French, Spanish, and Italian. The relevant dataset and code will be available at Github.