MELD¶

MELD is a multilingual and multi-domain dataset for Named Entity Recognition (NER) constructed from 60 existing datasets. Built with reproducibility and extensibility in mind, MELD currently provides gold-standard annotations for 60 languages across up to 14 domains with a total of 601 normalized entity labels. MELD was primarily designed for diverse mutlilingual and multi-domain evaluation but also includes all training and validation sets from its source datasets where available.

Learn how to get started with MELD here or browse the API documentation.