Tag
This paper investigates the curse of multilinguality in lexical normalization, finding that training a single model on multiple languages leads to decreased per-language accuracy, with optimal performance when languages are trained in small groups.