ner.manual: highlight patterns

, , ,
1

Hi,

According to the documentation pre-defined patterns are highlighted in ner.manual.

{"label":"PERSON","pattern":[{"lower":"Obama"}]}
{"label":"PERSON","pattern":[{"lower":"Barack"}, {"lower":"Obama"}]}
{"label":"PERSON","pattern":[{"lower":"Putin"}]}

I run
prodigy ner.manual politicians blank:en --label PERSON --patterns politicians.jsonl

However, none of the above are highlighted. Rather, I have patterns highlighted that have nothing to do with PERSONs. Once I have highlighted an new PERSON (eg. Clinton) It is not identified as PERSON either

Where do I go wrong?

cheers,

Andreas

2

Hi! The problem here is that your patterns don't match: you're looking for lowercase tokens that match "Obama" or "Putin". This is never going to be true, since those strings are not lowercase. Prodigy uses the same patterns logic as spaCy and you can find more details here: https://spacy.io/usage/rule-based-matching You might also find the interactive demo helpful, which lets you test your patterns and make sure they match.

3

Hi Ines,

thanks; the modification did do the trick!
I thought the "lower" did not make a difference as I interpreted as "it will be compared in lowercase". So I will chance it to the verbatim "ORTH (TEXT)", as the names should occur verbatim

thanks again

Andreas