Debiasing Pre-trained Contextualised Embeddings
This addresses bias in contextualized embeddings for NLP applications, but it is incremental as it builds on existing debiasing methods for static embeddings.
The authors tackled the problem of discriminative biases in pre-trained contextualized embeddings by proposing a fine-tuning method applicable at token- or sentence-levels, finding that token-level debiasing across all layers yields the best performance and noting a trade-off between accuracy and debiasing across models.
In comparison to the numerous debiasing methods proposed for the static non-contextualised word embeddings, the discriminative biases in contextualised embeddings have received relatively little attention. We propose a fine-tuning method that can be applied at token- or sentence-levels to debias pre-trained contextualised embeddings. Our proposed method can be applied to any pre-trained contextualised embedding model, without requiring to retrain those models. Using gender bias as an illustrative example, we then conduct a systematic study using several state-of-the-art (SoTA) contextualised representations on multiple benchmark datasets to evaluate the level of biases encoded in different contextualised embeddings before and after debiasing using the proposed method. We find that applying token-level debiasing for all tokens and across all layers of a contextualised embedding model produces the best performance. Interestingly, we observe that there is a trade-off between creating an accurate vs. unbiased contextualised embedding model, and different contextualised embedding models respond differently to this trade-off.