Labeling references to laws and court rulings

[danfejzo-spec]

Examples taken from Duranteye (2020):
https://pdf-tei-editor.panya.de/#pdf=e7z9b6&xml=nf62nu

Currently, there is no adequate label for references to laws like this:

Labels explicitly identifiying the titel ('UrhG') as a law or norm and a differentiation between section ('§') and subsection ('Abs. 2') should suffice for capturing the information.

Court rulings, on the other hand, are often embedded in a complex reference structure leading to automatic mislablings like this:

Generally, the segmentation could be done in terms of currently available labels, e.g. like this:

However, at least three new issues arise with this segmentation:

1. 'Tz.' (= marginal number) is not actually equal to page numbers and can be especially misleading when actual page numbers either not exist at all or are derived from the version of the ruling published in a journal ( in this case 'GRUR'). Should there be a differentiation between page and marginal numbers?
2. 'GS Media/Sanoma' is a header used as shorthand for finding the case in question. It is not essential for the core information of the reference, but it inherently belongs to it (as opposed to elements usually labeled as notes). In any case, it has to stop being identified by the model as starting a new .
3. The court is not really an author and the case number not really a title. Is Orgname more adequate? Should the actual case id have its own label?

@lfoppiano @cboulanger

[cboulanger]

Thank you @danfejzo-spec ! I have asked Claude to suggest some solutions for us, with examples. I asked it to research if there aren't already existing practices but it didn't come up with anything. We can devise our own and then the TEI guys can correct us - it will be easy to update the schema and annotations if our own annotation scheme is fine-grained enough.

https://github.com/mpilhlt/grobid-footnote-flavour/blob/main/docs/spec-legal-references.md

[cboulanger]

I think the suggestions make sense. @danfejzo-spec do they convince you? @lfoppiano , what do you think from the perspective of model training?

[danfejzo-spec]

Thank you @danfejzo-spec ! I have asked Claude to suggest some solutions for us, with examples. I asked it to research if there aren't already existing practices but it didn't come up with anything. We can devise our own and then the TEI guys can correct us - it will be easy to update the schema and annotations if our own annotation scheme is fine-grained enough.

https://github.com/mpilhlt/grobid-footnote-flavour/blob/main/docs/spec-legal-references.md

The examples annotated according to the recommended profil look sensibly well, i.e. I can't see obvious counter examples for now. Re decisions, I agree with the prioritarisation with A and B carrying the bulk in terms of preserving information. Re Rollout, I would add to 4. a reannotation of an english text from the batch. the issue as described is not confined to german citations [see 1. Motivation], so ideally our solutions should work just as fine there, too.

[lfoppiano]

Looks good.

Regarding decision B, we could leave it open, because legal_case seems more correct but decision is more precise...

Minor: I see mixed camelCase and snake_case in the attribute values.

[lfoppiano]

The examples, only the information not already provided by the tag should be inside. I know that § means section, but I'm not sure if Abs. is uniquely identifying sub-section:

<bibl type="legislation">
  § <citedRange unit="section">19a</citedRange>
  Abs. <citedRange unit="sub-section">2</citedRange>
  <title level="m" type="legislation" key="UrhG">UrhG</title>
</bibl>

[cboulanger]

I wonder, though, if we shouldn't include the country-specific labels as meaningful data rather than as training feature. "§" is the official way of designating a section in a German legal code, a number won't do it. And "Abs." literatlly means paragraph, which is a numbered subsection, even though the "§" means "Paragraph" as well. The most finegrained pointer is "Satz" which is sentence. I am not sure what's the best way here - another question for the TEI guys.

[lfoppiano]

We could, for our benefit, but at this point everything will be placed in the same model.