Edit saved annotations

,
1

Hi,

I’ve been using the ner.manual to annotate some entities and I would like to fix some of the annotations, not only the previous 10 showed in the History section. Is there a way to do that using Prodigy interface instead of modifying the exported .jsonl file manually? I tried to launch ner.manual with the annotated file but it loads the sentences as they were not annotated. I would like to see the selected entities, correct some of them and keep others.

Thanks a lot,

2

That's strange – are you using the latest version of Prodigy? The manual NER interface should definitely respect spans that are already present in the data. That's also how the ner.make-gold workflow does it.

Do you have an examples of the tasks in your dataset? And could you extract a few samples that definitely include annotated spans and try loading that with ner.manual? You can also always edit the source of the recipe and print the stream, to verify that the examples are going out correctly. (And finally, you could also check the console in your browser's developer tools to see if there's an error here – that could indicate a bug on the front-end. Maybe there's some edge cases that's not handled correctly!)

3

Hi Ines,

Thanks for your answer.

I'm using Prodigy v1.2.0, that could be the cause, thanks for pointing that out. I'll update and see if it still happens.

I also don't see any error in the browser. One annotated example from the file looks like this:

{"text":"long white dresses","_input_hash":584053789,"_task_hash":-386224016,"tokens":[{"text":"long","start":0,"end":4,"id":0},{"text":"white","start":5,"end":10,"id":1},{"text":"dresses","start":11,"end":18,"id":2}],"spans":[{"start":0,"end":4,"startIdx":0,"endIdx":0,"label":"ATTRIBUTE"},{"start":5,"end":10,"startIdx":1,"endIdx":1,"label":"COLOUR"},{"start":11,"end":18,"startIdx":2,"endIdx":2,"label":"SHAPE"}],"answer":"accept"}

Thanks,

4

Ah yes, that would explain a lot! Allowing pre-defined spans in ner_manual and the new ner.make-gold workflow are both features that were introduced in Prodigy v1.3.0. So if you upgrade to the latest version, it should all work as expected!

Btw, in case you haven't seen it yet, you can also set --answer accept when you export your annotations with db-out. This will filter out all examples you've rejected or ignored, so you won't have to click through those again when you re-annotate your data.

5

Hi Ines,

I just updated Prodigy to v1.3.0. Everything worked perfectly and I can edit the annotations now. I’m sorry for the misunderstanding and thanks a lot.

1 Like