Gender Bias in Multilingual Neural Machine Translation: The Architecture Matters
This research addresses the problem of gender bias in multilingual NMT for users of translation systems, showing that architectural choices can mitigate this issue.
This paper investigates how different multilingual neural machine translation (NMT) architectures influence gender bias accuracy when trained on the same data. They found that Language-Specific encoder-decoder architectures exhibit less gender bias than Shared encoder-decoder architectures across four language pairs.
Multilingual Neural Machine Translation architectures mainly differ in the amount of sharing modules and parameters among languages. In this paper, and from an algorithmic perspective, we explore if the chosen architecture, when trained with the same data, influences the gender bias accuracy. Experiments in four language pairs show that Language-Specific encoders-decoders exhibit less bias than the Shared encoder-decoder architecture. Further interpretability analysis of source embeddings and the attention shows that, in the Language-Specific case, the embeddings encode more gender information, and its attention is more diverted. Both behaviors help in mitigating gender bias.