ris-at: build Austrian law from RIS

#18
by grundtvig-bot - opened
Danish Foundation Models org

What I need from you: nothing.

Austrian legal information system (RIS): official works that are not protected by copyright (§ 5 UrhG), bulk-downloadable.

A parser plus a bulk download rather than a Hub download. It adds Austrian law to a legal domain that is currently German federal plus EU.

Order: after europarl-de.

Danish Foundation Models org

Correction to the opening post, from measuring the API today: RIS is not bulk-downloadable. The OGD terms say so explicitly (no ZIP/snapshot/dump of any kind) and impose: max 0.5 requests/second, no parallel connections, queries "only outside office hours (20:00-05:00) or at weekends/Austrian holidays", and a mandatory announcement to ris.it@bka.gv.at before any use beyond a single lookup, with user agent, IP and time window.

What that costs, measured against the live API today:

  • BrKons (consolidated federal law) 442,057 documents; BgblAuth 19,003; BgblAlt 28,547 - 489,607 for federal law alone, and the Judikatur applications are larger on top.
  • Enumeration is cheap (100 hits per page), but the text is one request per document, so at 0.5 req/s single-connection that is about 11 days of continuous fetching, or about 30 days inside the mandated 9-hour window.

Three ways forward, all fine by me:

  1. send the announcement to ris.it@bka.gv.at and crawl nights/weekends (long, but complete);
  2. ship only the gazette (BgblAuth + BgblAlt, 47,550 documents, ~3 days) and leave consolidated law out;
  3. drop Austrian law.

Licence is not the question: the texts are amtliche Werke and OGD use is CC BY 4.0 with attribution, which the subset card would carry.

Tell me which and I will start it; nothing has been crawled so far.

Danish Foundation Models org
•
This comment has been hidden
KennethEnevoldsen changed discussion status to closed
Danish Foundation Models org

Decision (5 Nov): crawl it. Austrian law is in, and we crawl rather than wait for a dump - RIS offers none (the OGD terms refuse ZIP/snapshot/dump outright), so the announced crawl is the only route.

  • Scope: legislation only - BrKons 442,057 + BgblAuth 19,003 + BgblAlt 28,547 = 489,607 documents. The Judikatur is left out for now; it would add weeks.
  • The mass-download announcement went to ris.it@bka.gv.at today, so the terms are binding: 2 s between requests, one connection, 20:00-05:00 Vienna time or weekends/Austrian holidays.
  • The crawler starts tonight at 20:00 CET and runs inside that window: about 281 request-hours, so documents land over the next few weeks and the subset is built from the cached XML as it accumulates.

Licence unchanged: amtliche Werke, CC BY 4.0 with attribution (RIS - Rechtsinformationssystem des Bundes), on the subset card and in every row.

Danish Foundation Models org

The first slice of this source is in the corpus: ris-at, v0.0.20, 519 acts / 3.55M tokens.

One grain question was settled in the build and is worth recording here. RIS BrKons serves the consolidated law provision by provision - a document per §, per Anlage, per Artikel, about 13 per act (median 989 characters), each repeating the act's title and Kundmachungsorgan. Shipping that grain would have produced 442,086 documents with no act boundary left. The build instead merges the provisions of one act back into the act, grouped by the Gesetzesnummer, ordered by the act's ELI: one document per act, whole (median 1,111 tokens, longest the Abfallwirtschaftsgesetz 2002 at 391K). If you would rather have the provisions as rows, it is one filter in create.py plus a rebuild.

Coverage is partial by construction: the enumeration is complete at 442,086 documents and the crawler adds about 3,500 per night inside the agreed window, so this release holds the first 6,971. BgblAuth (the gazette from 2004) and BgblAlt (1848-1940) are enumerated after BrKons finishes, so expect months rather than nights. The card says all of this under Limitations.

Sign up or log in to comment