Jan 6, 2021
Did you try to use BERT multilingual? From a quick test, I see the number of tokens for racialized names is different from what you've found. Very interesting and important study!
Written by Riccardo Di Sipio
Senior Machine Learning developer at Dayforce. NLP, LLMs, graph neural networks. Formerly physicist at U Toronto, Bologna, CERN LHC/ATLAS.