CatholicCorpus/catholiccorpus-text

CatholicCorpus — Extracted Text

Pre-extracted plain text from the CatholicCorpus — 2,000 years of the Catholic intellectual tradition, ready for NLP, RAG, and digital humanities.

This dataset contains 47,407 plain text files (5.7 GB, 2.64 billion GPT-2 tokens) extracted from the raw source corpus (PDF, EPUB, TEI XML, HTML). If you need the original source formats, see the raw corpus.

Quick Start

from datasets import load_dataset
ds = load_dataset("CatholicCorpus/catholiccorpus-text")

Or clone directly:

git clone https://huggingface.co/datasets/CatholicCorpus/catholiccorpus-text

What's Included

Collection	Files	Description
01 Git Repos	4,593	Aquinas Opera Omnia, LXX, Byzantine text, eBible
02 Corpus Corporum	9,945	Patrologia Latina, medieval Latin (TEI XML source)
03 Project Gutenberg	8	Catholic classics in English
04 Direct Downloads	1,232	SBL Greek NT, Vulgate, Canon Law, miscellaneous
05 CCEL	54	Ante-Nicene and Nicene Fathers
06 Catholic Encyclopedia	15	1913 Catholic Encyclopedia volumes
07 Latin & Franciscan	21	Peter Lombard, Bonaventure, Franciscan authors
08 Liturgical & Hymns	31,090	Divine Office, Gregorian chant texts
09 Archive & Misc	57	Hagiography, scholastics, devotional works
11 Catholic Bible	11	Haydock, Knox, Confraternity editions
12 English Catholic Thinkers	45	Newman, Chesterton, Belloc, Knox
13 Mystics	13	Teresa of Ávila, John of the Cross, Emmerich
14 Liturgical Expansion	13	Breviarium, Missale, Pontificale
15 Late Scholastics	42	Suárez, Molina, John of St. Thomas
16 Patrologia Graeca	214	159 of 161 Migne PG volumes
17 Modern Magisterium	54	Vatican II, CCC, modern encyclicals

Languages: Latin (60%), English (25%), Greek (~15%)

Known Gaps

This is the v1.1 text layer. Not every file in the raw corpus has a corresponding text extraction:

• 5 oversized PDFs (>200 MB each) were skipped: Butler's Lives complete edition, Haydock Douay-Rheims Bible, Breviarium Romanum (2 vols), and Missale Romanum 1962. These total ~1.85 GB of raw source not yet extracted.
• 115 files produced empty extractions (image-only scans with no embedded text layer, or non-prose XML schemas).
• 9,932 .pos.xml files were intentionally skipped (duplicates of the regular XML).

These gaps are documented in full in EXTRACTION_GAPS.md and targeted for the v1.2 OCR Enhancement milestone.

If you need the raw files for the gaps above, download them from the raw corpus.

Extraction Quality

Quality varies by source format:

• TEI XML and HTML sources (Tasks 01, 02, 04, 08, 17) produce clean, high-quality text.
• EPUB sources (Task 16 — Patrologia Graeca) produce good text with occasional formatting artifacts.
• PDF sources (Tasks 05–07, 09, 11–15) vary widely. Modern PDFs with embedded text layers extract well. Older archive.org scans may contain OCR artifacts (garbled characters, missing diacritics, broken line breaks).

Licensing

The same three-tier licensing model as the raw corpus applies:

• Tier 1: Public Domain — the vast majority (pre-1928 U.S. publications, authors deceased 70+ years)
• Tier 2: Creative Commons — SBL Greek NT (CC BY 4.0), Rahlfs LXX (CC BY-NC-SA 4.0)
• Tier 3: Copyrighted / Referenced Only — Task 17 (Modern Magisterium) content is under copyright

See the About page for full licensing details.

Links

• Raw source corpus: CatholicCorpus/catholiccorpus
• GitHub (scripts & docs): github.com/CatholicCorpus/catholiccorpus
• Website: catholiccorpus.org
• Contact: admin@catholiccorpus.org