Our paper ‘What’s in a Name? Reducing Bias in Bios without Access to Protected Attributes’ won the Best Thematic Paper award at NAACL 2019!

In previous work, we studied the risks of compounding gender imbalances in occupation classification. We have also proposed algorithms for enumerating biases in word embeddings, and showed that widely used embeddings encode societal biases. Building on both these works, in our NAACL paper we tackle the question ‘how can we mitigate biases without requiring access to protected attributes?’ and explore ways of leveraging societal biases encoded in word embeddings towards this end.