Our work on bias evaluation in multilingual models titled “Comparing Biases and the Impact of Multilingual Training across Multiple Languages” is now on arXiv. Check it out here.