sucx3_ner

The original dataset does not contain an official train-dev-test split, which is introduced here; the tag distribution for the NER tags between the three splits is mostly the same. The dataset has three different types of tagsets: manually annotated POS, manually annotated NER, and automatically annotated NER.

For the automatically annotated NER tags, only sentences were chosen, where the automatic and manual annotations would match (with their respective categories). Additionally we provide remixes of the same data with some or all sentences being lowercased.