Yann LeCun & Xiang Zhang produce first study of 473 encoding models for multilingual text classification