Transfer language selection for zero-shot cross-lingual abusive language detection

We study the selection of transfer languages for automatic abusive language detection. Instead preparing a dataset every language, we demonstrate effectiveness cross-lingual learning zero-shot This way can use existing data from higher-resource to build better detection systems low-resource languages. Our datasets are seven different three families. measure distance between using several similarity measures, especially by quantifying World Atlas Language Structures. show that there is correlation linguistic and classifier performance. discovery allows us choose an optimal zero shot