Marking sentences for classification

[spatiebalk]

I am currently using Prodigy to classify short texts. However, I would like to classify these short texts sentence by sentence. Instead of viewing and classifying each sentence separately, I was wondering whether it is possible to view text snippet by text snippet and then highlight the positive sentences in that current text snippet (in the same way that you can highlight tokens when you are training a NER model)?

[ines (Ines Montani)]

Hi! There are several ways you could solve this and I think it's definitely a good idea to focus on making each sentence a selectable unit (and not highlighting each sentence token-by-token, which would be pretty inefficient).

One simple approach would be to use the choice interface with "choice_style": "multiple" and create one option per sentence, maybe grouped by paragraph or some other sensible unit. Then you can go through them and select all positive sentences, either by clicking on them, or using keyboard shortcuts.

Alternatively, a similar use case was posted on the forum a while ago and they ended up using the manual span interface, but feeding in data with a "tokens" property, but with one sentence as a "token". This would let you view the sentences in their natural flow, and you could double-click on them to select them, and even assign them different labels. See here for details:

Annotation for document segmentation

Dear Explosion AI team, I am wondering if there is an option in Prodigy to annotate blocks of texts and highlight the annotation as blocks such as: [image] I think it would be useful for document segmentation tasks. In my case, I am trying to implement an annotation workflow for CV parsing with Prodigy, one step of which is to segment the work experience section into experience items. However, when we annotate items, it is hard to see in the interface whether two consecutive items are annotat…

[spatiebalk]

Hi Ines!

Thank you for your quick response. By formatting the data in such a way that each sentence was a token (so the second option) it gave the desirable behavior!

However, the lay-out is now quite odd of the marked sentences. I read the other post you referred to, but couldn't quite figure out how to adjust the shape of the highlighted areas. Could you possibly help me with this as well?

Thanks in advance!

[ines (Ines Montani)]

@spatiebalk

However, the lay-out is now quite odd of the marked sentences. I read the other post you referred to, but couldn't quite figure out how to adjust the shape of the highlighted areas. Could you possibly help me with this as well?

Glad it worked! And try setting display: block on the sentence "tokens", as described here to make the selection a block element:

Annotation for document segmentation

Yay, glad it worked You could try setting .prodigy-content span, .prodigy-content mark { display: block } in the global_css. This will make each "token" and selection (in your case, the lines) a block element. Just did a quick test locally and the selection and double-clicking still works as expected. If you also add something margin-bottom: 5px, it actually ends up looking very similar to your first screenshot.