ner.teach with a model made using span.manual

Does ner.teach recipe work with a model made with span categorizer (spans.manual)? If not is there any way, to teach my model in order to enhance the prediction accuracy?


Hi @hmousa961 !

It won't work out-of-the-box because the ner.teach recipe is optimized for the entity recognizer which predicts non-overlapping spans in the doc.ents.

If you want to improve prediction accuracy, I suggest using spans.correct. You can also pass the --update parameter if you want to annotate with a model in the loop.

Thank you so much for your reply. I will try the spans.correct with passing the --update.