BERT based Token Classification model based on tohoku-nlp/bert-base-japanse and trained to predict in a Japanese sentence without an explicit subject where the subject would be.

This model was trained as part of a bigger project to predict implicit subjects in Japanese text. You can find whole project here [https://github.com/Romi212/Japanese-Subject-Predictor-System]

Model was trained using dataset https://github.com/UniversalDependencies/UD_Japanese-GSDLUW

The dataset was reduced only to sentences with a subject, and the subject was removed from the sentence saving the position to train the model to predict where the subject should go.