impresso-project/ner-hipe2020-hist-medium

Impresso HIPE-2022 NER model

Token-classification (NER) model fine-tuned from dbmdz/bert-medium-historic-multilingual-cased on the impresso-project/ner-augmentation dataset (HIPE-2022, NE-COARSE-LIT, IOB2). Part of the Impresso NER pipeline.

• Languages: fr, de, en
• Base model: dbmdz/bert-medium-historic-multilingual-cased
• Training data: impresso-project/ner-augmentation
• Label scheme: IOB2 over NE-COARSE-LIT (pers / org / loc / prod / time).

Test-set results (seqeval, entity-level)

metric	value
f1	0.7022
loc_f1	0.8028
org_f1	0.3944
pers_f1	0.6958
precision	0.6620
prod_f1	0.5292
recall	0.7476
time_f1	0.6947

License

Inherits the CC BY-NC-SA 4.0 license of the underlying HIPE-2022 data.