Datasets:
id stringlengths 4 7 | text stringlengths 18.9k 6.29M | source stringclasses 1
value | added stringdate 2026-11-11 00:00:00 2026-11-11 00:00:00 | created stringdate 1508-01-01 00:00:00 1930-01-01 00:00:00 | token_count int64 4.21k 1.53M | license stringclasses 1
value | author stringlengths 13 323 | language stringclasses 1
value | text_origin stringclasses 1
value | generator stringclasses 0
values | derived_from stringclasses 0
values |
|---|---|---|---|---|---|---|---|---|---|---|---|
pg15 | "The pale Usher—threadbare in coat, heart, body, and brain; I see him now. He was ever dusting his(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1851 | 286,951 | public-domain | Melville, Herman (1819–1891) | en | human | null | null |
pg16 | "A Millennium Fulcrum Edition produced in 1991 by Duncan Research. Note that while a copyright was i(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1911 | 60,106 | public-domain | Barrie, J. M. (James Matthew) (1860–1937) | en | human | null | null |
pg23 | "In the month of August, 1841, I attended an anti-slavery convention in Nantucket, at which it was m(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1845 | 50,088 | public-domain | Douglass, Frederick (1818–1895) | en | human | null | null |
pg44 | "I\n\nDr. Howard Archie had just come up from a game of pool with the Jewish clothier and two travel(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1915 | 192,467 | public-domain | Cather, Willa (1873–1947) | en | human | null | null |
pg107 | "In reprinting this story for a new edition I am reminded that it was in the chapters of “Far from(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1895 | 178,648 | public-domain | Hardy, Thomas (1840–1928) | en | human | null | null |
pg135 | "So long as there shall exist, by virtue of law and custom, decrees of damnation pronounced by socie(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1887 | 748,695 | public-domain | Hugo, Victor (1802–1885) | en | human | null | null |
pg139 | "CHAPTER I\n\n\"There Are Heroisms All Round Us\"\n\nMr. Hungerton, her father, really was the most (...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1912 | 93,873 | public-domain | Doyle, Arthur Conan (1859–1930) | en | human | null | null |
pg147 | "Perhaps the sentiments contained in the following pages, are not yet sufficiently fashionable to pr(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1776 | 26,614 | public-domain | Paine, Thomas (1737–1809) | en | human | null | null |
pg148 | "Benjamin Franklin was born in Milk Street, Boston, on January 6, 1706. His father, Josiah Franklin,(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1909 | 83,007 | public-domain | Franklin, Benjamin (1706–1790) | en | human | null | null |
pg166 | "I\n\nA girl came out of lawyer Royall's house, at the end of the one street of North Dormer, and st(...TRUNCATED) | gutenberg_en_narrative | 2026-11-11 | 1917 | 71,617 | public-domain | Wharton, Edith (1862–1937) | en | human | null | null |
Slayer Narrative EN
Long-form English narrative text (novels, short stories, adventure, historical fiction, narrative non-fiction) for Slayer continued pretraining. Every document is in the public domain in Poland and the EU (every contributor with a right in the text died in or before 1955; life + 70) and in the United States, and follows an identified source edition (edition years 1508–1930). Whole books, chapter structure kept. The texts were transcribed by Project Gutenberg volunteers; this dataset is not affiliated with or endorsed by Project Gutenberg, see NOTICE.md.
The layout follows the SlayerLab dataset-team-submissions standard: one source per directory under data/, the
12-column row schema, a stats.json and a datasheet per source, per-row license, takedown procedure.
Status: private. Version 0.2.0, built 2026-11-11. Not yet reviewed by a second team member.
| Documents | 17,054 |
| Characters | 7,272,886,258 |
Tokens (cl100k_base, token_count) |
1,672,951,432 |
| Licence per row | public-domain (all rows) |
| Text origin | human (all rows); no synthetic or machine-translated text |
| Genre | Documents | cl100k tokens |
|---|---|---|
| narrative_nonfiction | 7,265 | 792,088,910 |
| novel | 5,247 | 481,294,376 |
| adventure | 2,434 | 212,177,850 |
| historical_fiction | 966 | 109,811,993 |
| short_stories | 1,142 | 77,617,596 |
(Genre token counts are measured before PII masking.)
Row schema
id, text, source, added, created, token_count, license, author, language, text_origin, generator, derived_from — the SlayerLab shard.SCHEMA. source is gutenberg_en_narrative, added is 2026-11-11,
created is the source edition year, id is pg<ebook number>. Title, genre, contributors with dates and
rights evidence per document are in attribution/ and keyed by id.
Held-out splits for evaluation are given as id lists: splits/train.txt (16,723), splits/validation.txt
(171), splits/test.txt (160). The parquet stores rows in that order.
Layout
| Path | Content |
|---|---|
data/gutenberg_en_narrative/gutenberg_en_narrative.parquet |
the rows (12-column schema, zstd) |
data/gutenberg_en_narrative/gutenberg_en_narrative.stats.json |
docs, chars, tokens, drop reasons, PII record (Presidio version, entities, counts) |
data/gutenberg_en_narrative/gutenberg_en_narrative.md |
datasheet: Source, Rights review, Cleaning, Personal data, Sample |
splits/ |
train / validation / test id lists |
attribution/attribution.jsonl |
per document: title, authors, translators, contributors with dates, work URL, source edition, transcription credits, what was changed, text hash |
attribution/licences.tsv |
per document: licence label, PG rights field, latest contributor death year, end of EU protection, edition year |
attribution/license-evidence.json |
rights policy, documents held or excluded at the text screen with reasons, what was not checked |
provenance/records.jsonl |
per document: source file URL in the official archive, raw SHA-256, fingerprints, PII counts |
provenance/build_report.json, dedup_pairs.tsv |
counts, rejections by reason, near-duplicate pairs |
exclusions/ |
exclusion register: every rejected input and why, selected ids, config |
takedown/ |
procedure, blocklist, register |
SOURCE_CARD.md |
DatasetTeam source card |
files.sha256 |
SHA-256 of every file in this repository |
Rights review (summary)
Full review: the datasheet.
- Admitted: every contributor with a right in the text in the source catalogue (author, translator, editor, author of introduction, compiler, contributor, commentator, annotator, adapter) died in or before 1955; the title page names no translator, editor or introducer missing from the catalogue; the source edition is identified by an imprint or copyright year on the title page.
- Held or excluded (not in the data): contributor died after 1955 (public domain in the US only), unknown death year, anonymous or "Various", anthologies without authors in the catalogue, copyrighted items, and documents whose source edition is not identified (SlayerLab LESSONS #7: public domain without an identified edition is withdrawn).
- Death years come from the source's RDF catalogue, checked on a sample against Wikidata; the public-domain label on the platform is not taken as proof.
- We do not rely on the text-and-data-mining exception to publish anything.
Personal data
Contact identifiers (e-mail, URL, modern phone formats) were masked during the build; the dataset-team PII step
(Presidio 2.2.366, spaCy en_core_web_lg 3.8.0, names=False) then masked 6,736 [PHONE] and 268 [PII]
spans. A spot check found them to be false positives on numbers (date ranges, index page references, numeric
tables); they are kept as the rule requires. Names are kept: authors are credited, characters are fictional or
historical. A second PII pass over 200 seeded documents found nothing new.
Checks
A streaming equivalent of the dataset-team check.py (same functions) passed with 0 errors: schema, per-row
gates, recounted token_count, Dynaword text keys and upstream hosts (no overlap), takedown blocklists, English
language gate, duplicate ids and texts, stats, datasheet sections.
Disclaimer & legal
- Public domain is a status, not a licence. The label
public-domainstates our assessment that copyright in the text has expired in Poland/EU and the US. Status in other jurisdictions may differ; check your own law. - Provenance in good faith. We make no warranty about the status of any individual document. Provided "as is", without warranty of any kind. This is not legal advice.
- No ownership claim over the texts.
- Moral rights of authors (e.g. Polish copyright act art. 16) are not limited in time: do not present altered works as the author's original, and attribute when you quote.
- Notice-and-takedown. To ask for a record's removal, open a discussion in this repository's Community tab
with the record
id, a link or a phrase from the text, and the reason. Procedure: takedown/PROCEDURE.md; removals are listed in takedown/register.md.
License & attribution
The texts are in the public domain (above). The collection (selection, arrangement, metadata and sidecars) is
released under CC0-1.0, see LICENSE. Credit, though not required, is welcome: "Texts transcribed
by Project Gutenberg volunteers (https://www.gutenberg.org/)". Per-document credits are in
attribution/attribution.jsonl. "Project Gutenberg" is a registered trademark of the Project Gutenberg Literary
Archive Foundation; it is used here only to name the source.
Acquisition
Project Gutenberg's official weekly archive txt-files.tar.zip (SHA-256
6e3d284356e62c576e7861dde70f65ff087960ab187cc17a48828f8ad8010582, 11,333,943,015 bytes) and the official RDF
catalogue (rdf-files.tar.bz2, SHA-256 04ba71ed476d4aa986f68acc6ce7e09032e927d98042a00f5e0c5e920ad5f881),
acquired 2026-11-11 through the official offline channel; no scraping of www.gutenberg.org.
- Downloads last month
- 76