Cross-lingual pretraining sets new state of the art for natural language


#1