Normalize NER Entities

Hello! I would like to tag a text with people names mentioned in it. So I pass the text to ner_ontonotes_bert_mult_torch model and get the needed entities extracted. That woks fine with English names as they never change. But Russian names could be in different cases, like: “Позвонить Ивану Иванову” so I get “Ивану Иванову” for my tags instead of “Иван Иванов". Same with cities etc. is there any way I could bring my tags to Nominative Case for further reference?

Just now I have found morpho_ru_syntagrus_pymorphy model which is probably what I need. But I can’t use TensorFlow models. Can I train it with pyTorch? Is there a config available for it?

Dear @vzhilov ,

We haven’t converted morpho_ru_syntagrus_pymorphy model into Pytorch yet. What I would recommend is to use either ner_ontonotes_bert_mult_torch or ner_rus_bert_torch.json for NER, then you can postprocess the named entities by using pymorphy2 separately.

Let me know if you need further assistance.

1 Like