mizo-ner-xlmr

micro F1 0.8739 on the silver test set; F1 0.6078 against human gold

Trained on the Mizo NER corpus, 441,178 sentences with silver-standard labels.

Usage

from transformers import pipeline
ner = pipeline("token-classification", model="haulai/mizo-ner-xlmr",
               aggregation_strategy="simple")
print(ner("Pu Lalthanhawla chuan Aizawlah thu a sawi."))

Important

Training labels are silver-standard. Against 300 sentences annotated by two Mizo speakers this model scores substantially lower than on the silver test set; both figures are given above. Treat the silver number as agreement with the projection, not accuracy.

Trained only on entity-bearing sentences, so behaviour on entity-sparse text is untested.

Citation

See the code repository.