hi @Tiziri!
Thanks for your question.
Actually, you can simply reload past annotations using dataset:[name of annotated dataset]
as your source.
For example, let's say you've run:
python -m prodigy ner.manual ner_news_headlines blank:ar ./Tiziri.jsonl --label TYP,SYM,ORG,INGR,OBJ,TRANS
And you've saved X number of annotations into ner_news_headlines
.
You can run ner.manual
again with the annotated data by running:
python -m prodigy ner.manual ner_news_headlines_updated blank:ar dataset:ner_news_headlines --label TYP,SYM,ORG,INGR,OBJ,TRANS
Even cooler, you can also add :accept
or :reject
to annotate only those that are accepted, rejected, ignored, etc.
See these docs for more details.
Hm... can you provide a few examples in a .jsonl
file? I wonder if there's any issue with your data. I know of several users who have been successful using Arabic in Prodigy (FYI, we recently updated our right-to-left for many interfaces like rel.manual
so it should work.)
Ideally, if you can give me a fully reproducible example (prodigy command, prodigy version (run prodigy stats
), and input file), that would help us as it's really hard to diagnose problems without a reproducible example.