Contributing Greek legislation — question on URL fields for multiple sources

#2
by odychlapanis - opened
Hugging Face for Legal org

Hello! I'd like to contribute a subset of Greek legislation to this dataset. I think it would pair really well with GreekBarBench for a complete legal RAG benchmark.

Before I open the PR, I wanted to clarify how to handle URLs for the Greek case, since the source structure differs from legifrance.

Greek laws are published in Government Gazette (FEK) issues as PDFs, not per-article web pages (search.et.gr). There's no equivalent of legifrance's stable LEGIARTI... per-article deep links. A single article also often spans multiple FEK issues over time (original publication plus amendments).

About the dataset itself: it was assembled in 2020, but the source links weren't preserved properly. So the limitations are (a) the snapshot is from 2020, and (b) PDF URLs are missing for most articles. This isn't suitable for actual legal practice, but I think it's fine for RAG evaluation of LLMs.

Proposed approach:

  • url_sourcepage: null (no per-article URL exists)
  • url_sourcefile: FEK PDF URL where available, null otherwise (mostly null at first; we may be able to recover some post-hoc with a script)
  • Everything else fully populated, except for html: jurisdiction: "GR", language: "el", text, titles, IDs, dates, plus uuid and text_hash via the hf-for-legal package

Is null acceptable for these URL fields, or would you prefer a different convention for jurisdictions without per-article deep links? Happy to adjust based on what works best for the project.

Thanks!

Sign up or log in to comment