Logo Lanfrica

Mizo Named Entity Recognition: silver-standard corpus, gold evaluation set, models and code

Domaine:

natural language processing

Type de record:

dataset
Créateur:
ThaHusDawVar
Éditeur:
Zenodo
Hôte:avatar
Silver-standard named entity corpus for Mizo (lus): 441,178 sentences, 590,655 annotations over 11 types, built by cross-lingual projection from English. Includes a 300-sentence gold evaluation set annotated by two Mizo speakers (Cohen's kappa 0.798), trained recognizers, and code reproducing every result in the accompanying paper. Labels are silver-standard: measured against the gold set the projection scores F1 0.6146.