MELA introduces a multilingual benchmark for linguistic acceptability, highlighting the importance of in-language training data for cross-lingual transfer and syntax-related tasks.
