Hugging Face's logo

Datasets:
CatholicCorpus
/
catholiccorpus

Tasks:
Text Generation
Text Classification
Feature Extraction
Languages:
Latin
Ancient Greek (to 1453)
English
Size:
10K<n<100K
Tags:
catholic
theology
patristics
scholasticism
latin
greek
License:
Dataset card Files
xet
 Community
catholiccorpus / BACKLOG.md
ddbaula's picture
ddbaula
Add files using upload-large-folder tool
943f1dc verified
preview code
|
raw
history blame contribute delete
3.35 kB

CatholicCorpus — Backlog

Deferred work, ordered by when it should be picked up. Add new items at the top of the relevant section.

Gaps to revisit

Task 16 — Patrologia Graeca volumes 16 and 86

  • What: Vols 16 and 86 of the Migne Patrologia Graeca are not on archive.org under any search pattern tried on 2026-04-15. The current script skips them via the MISSING_VOLUMES = {16, 86} set.
  • Coverage achieved without them: 159 of 161 volumes (98.8%) via three chained sources:
    • Primary: patrologia-volumes (155 vols)
    • Fallback: patrologia-graeca_202105 (fills 75, 106)
    • Single-vol items: patrologiae_cursus_completus_gr_vol_087_2 (87) and patrologiae_cursus_completus_gr_vol_007_irenaeus_tomus_unicus (7)
  • Content at stake: Both are early Patristics (1st–3rd c.). Vol 16 may contain material from Eusebius / Hippolytus; vol 86 from Sophronius / Maximus Confessor.
  • Where to look next:
    1. Google Books / HathiTrust (they may have scans archive.org doesn't).
    2. documentacatholicaomnia.eu — a sister corpus that mirrors some Migne material.
    3. Specific French university digital libraries (Sorbonne, BnF Gallica).
    4. Direct probing of unpadded and differently-split archive.org patterns we might have missed (e.g. patrologiae_cursus_completus_gr_vol_016_pars_*).
  • Action: Low priority — do once the main run completes and the rest of the corpus is stable.

Long-running downloads (save for the end)

Task 02 — Patrologia Latina full download

  • What: Run the full PL download from Corpus Corporum (corpus idno 38).
  • Command:
    cd 02_corpus_corporum
python3 download_script.py --collection 38 --resume
  • Scope: 5,277 texts, ~85.5M words. Produces TEI XML + POS-tagged XML where available.
  • Expected runtime: ~10–11 hours at default rate limiting (1.5s between nav requests, 2s between downloads). Discovery phase alone is ~2.8h because PL has 1,528 authors and 5,204 works.
  • Disk: ~2–5 GB estimated.
  • How to speed up (if needed): --delay 0.8 roughly halves total time to ~6h. Default is the polite setting for a university research server — only reduce if there's a real deadline.
  • Resume: State saved every 10 texts to .download_state.json. Safe to Ctrl-C and restart with --resume.
  • Run it in: nohup, screen, or tmux so laptop sleep doesn't kill it.
  • Validation before kicking off: run --collection 17098 (Libri sacri, 7 texts, ~1 min) first to confirm the download endpoint is healthy end-to-end.
  • Status: Script fixed and verified via dry-run on 2026-04-15. Ready to execute. Deferred — do last.

Task 02 — Other large Corpus Corporum corpora (optional)

Candidates if PL goes well and you want more Latin coverage. Sorted by size:

  • Acta Sanctorum (idno 21967) — ~52.7M words
  • Graeca miscellanea (idno 21155) — ~10.8M words (Greek, not Latin — skip unless wanted)
  • Thomas Aquinas (idno 11975) — ~8.1M words, but accessible_texts_count=0 on the root listing, so downloads likely blocked. Verify before scheduling.
  • Neolatinitas (idno 14001) — ~7.7M words
  • Auctores scientiarum varii (idno 12130) — ~7.3M words
  • MGH (idno 14010) — ~6.6M words
  • Scriptores Ecclesiastici (idno 14007) — ~5.8M words

Decide scope before starting — each is a multi-hour run.