ner.teach annotations with incomplete n-grams

while using the ner.teach to tag ORGs, I tend to be confused when the model is tagging only part of the organisation (e.g. only Coca Cola is tagged as ORG in Coca Cola Company). Would you advise to accept, reject or ignore the annotation…?

I’ve found this answer really helpful when I’ve faced similar decisions: Ambiguous ner.teach decisions, and ner.make-gold workflow issue

1 Like

To add to the comment above: In general, you should always reject incorrect boundaries – even if they’re almost correct. If you click accept for “Coca Cola” in “Coca Cola Company”, you’re telling Prodigy that the correct entities for that phrase are ['B-ORG', 'L-ORG', 'O']. However, if you reject the suggestion, part of the entity may still be correct and the only feedback you’re giving Prodigy is "['?', '?', '?'] but not ['B-ORG', 'L-ORG', 'O']".

ty both

following your logic, when doing a manual annotation (ORG only), if I get a sentence w/o any ORG, i shall accept the annotation without further tag, right?

Yes, exactly! Including texts that don't contain an entity of the type you're annotating is almost as important as including annotations of the entity type.