Editing rel.manual to allow for multiple passes to account for nested entities

[rgandhasri87]

I am trying to make a custom recipe based off rel.manual to annotate relations between entities. I have some entities nested within other entities that I want to access, but only the outermost/longest entity appears in the annotation interface when using rel.manual, hence the need for the custom recipe.

I found this thread which suggests writing a stream generator to repeatedly send out the same example until I accept it, so I would be making multiple passes over the same data. I think this is what I want to do, since I can "delete" the longer outer entity and then the inner entity will be accessible on the annotation interface during the next pass.

My question is, how would I do this? I looked into pipeline components but that doesn't seem applicable at all.

Thanks for your time.

[ryanwesslen]

hi @rgandhasri87!

Thanks for your post and welcome to the Prodigy community

@rgandhasri87

My question is, how would I do this?

Did you also see this post?

Nested Labels while using rel.manual

Hi! Prodigy's interfaces for NER (and the span annotation features within the relations UI) are set up to work best with the way named entity recognition models generally work: contiguous, non-overlapping spans, with clear beginnings and endings. If the relation annotation has to represent any number of nested spans with each span and token being potentially connected to each other, this easily gets very messy and inefficient, and there's usually a better way to structure the task that's both e…

It's a more recent post that provides some background on approaching this problem.

Alternatively, are you aware of spancat? It was designed (in part) to handle nested/overlapping entities as NER cannot.

This post provides some background on ways to combine spancat with relations:

Can Dependencies & Relations work after Span Categorization?

Hi! Interesting use-case I think your suggested approach of using a spancat to extract the "rights" makes sense, as this is indeed not a typical NER task. What isn't clear to me is how you're defining the actions like "contact us" or "access your account settings". Are these extracted with a spancat model as well? With respect to relation extraction: There is builtin support in Prodigy to annotate both entities and relations at the same time with rel.manual. This does assume th…

Hope this helps provide more direction and let us know if you have further questions!