Comparing zero-shot self-explanations with human rationales in multilingual text classification

Open in new window