Hugging Face's logo

Datasets:
SlayerLab
/
polish-dynaword

Tasks:
Text Generation
Languages:
Polish
Size:
1M<n<10M
ArXiv:
Tags:
polish
pretraining
dynaword
License:
Dataset card Files
xet
 Community
117
polish-dynaword / data /govpl /govpl.md
ppuzio's picture
ppuzio
govpl: drop 188 PAP reprints (gov.pl third-party carve-out), add URL sidecar (#63)
c2099e6
preview code
|
Raw History Blame Contribute Delete
3.18 kB

govpl

gov.pl — Polish government press releases

Dataset description

  • Source (upstream): https://www.gov.pl/
  • Domain: government
  • Language: Polish (pl)
  • License: CC-BY-SA-4.0
  • Created (range): 2005-09-26, 2026-07-19
  • Added: 2026-07-19; PAP reprints removed 2026-10-06

Licensing — traceable basis

gov.pl publishes its text under CC BY-SA 4.0 (site footer and https://www.gov.pl/web/gov/prawa-autorskie: "Treści tekstowe publikowane w serwisie … są udostępniane na licencji CC BY-SA 4.0"). The same page excludes works whose copyright belongs to third parties; using those needs the owners' consent. Articles crediting PAP are therefore dropped (legal note).

Provenance

Ministry/agency press releases fetched directly from gov.pl (editor-content per article, HTML stripped) by src/fetch_govpl.py. Subsites + their non-uniform news-section paths are enumerated by src/discover_govpl.py into the committed manifest src/govpl_subsites.json (gov.pl has no sitemap/API). v0.2.6 removes PAP reprints from the v0.2.5 Parquet with src/clean_govpl.py, which also writes govpl.attribution.jsonl: each row's gov.pl URL, subsite, title and date.

Attribution

CC BY-SA 4.0 requires credit and a link to the source. govpl.attribution.jsonl has one line per row, in Parquet order: id, url (the gov.pl article), subsite (the publishing ministry or agency), title, date (as listed on gov.pl) and license. Rows were matched to the 2026-07-19 crawl on SHA-1 of the stripped text; for 1,189 rows the same text was fetched from more than one URL, and the first in fetch order is given.

Statistics

documents characters tokens (tiktoken proxy)
88,002 217,515,240 80,445,758

Per-document license metadata

license documents
CC-BY-SA-4.0 88,002

Author metadata present for 0 documents. Empty values mean the upstream record did not expose a machine-readable author field.

Legal scope note

gov.pl's copyright page carves out "utwory, co do których prawa autorskie przysługują osobom trzecim" from its CC BY-SA 4.0 grant. Articles crediting PAP (Polish Press Agency) as their source are such works, so 188 are dropped (rule fetch_govpl.pap_rule, applied by src/clean_govpl.py): a "Źródło:" line naming PAP (163), "Na podst. … PAP" (22), a "(PAP)" sign-off line (2) and one PAP interview transcript (1). PAP photo credits and articles that only quote or mention PAP stay. Other third-party text is not screened: 710 kept articles carry a "Źródło:" line naming some other source; they were not reviewed.

Filters applied

build_dynaword.py (v0.2.2), minimal per Dynaword guidelines:

  • drop documents < 200 chars: 0
  • drop non-Polish (diacritic ratio): 24
  • exact cross-source dedup (sha1): 1,926
  • read 90,140 → kept 88,190

src/clean_govpl.py (v0.2.6), third-party carve-out:

  • drop PAP reprints: 188 (Źródło: line 163, Na podst. 22, (PAP) sign-off 2, interview 1)
  • read 88,190 → kept 88,002

Token counts are a fast tiktoken (cl100k) proxy (~1% off Llama-3); the canonical Llama-3 count is computed at release.