data/1000_novels/1000_novels.md

1000_novels

1000 Novels Corpus (CLARIN-PL), public-domain subset

Dataset description

• Source (upstream): https://clarin-pl.eu/dspace/handle/11321/312
• Domain: literature
• Language: Polish (pl)
• License: public-domain (public domain in Poland; per-row basis in 1000_novels.attribution.jsonl)
• Created (range): unknown
• Added: 2026-06-26; rebuilt 2026-10-05 as the public-domain subset; rescreened 2026-10-06; texts already in wolne_lektury or eltec_pol dropped 2026-10-06

Licensing — traceable basis

The CLARIN-PL record states CC BY 4.0, but CLARIN-PL confirmed on 2026-10-05 that the handle holds metadata only: it does not distribute the novels or translations and holds no documentation of rights in them. No licence covers the texts.

The release therefore keeps only the texts that are in the public domain in Poland, from a per-text screen (polish-dynaword-v2/evidence/1000_novels_rights_screen.csv, script polish-dynaword-v2/scripts/novels_rights_screen.py):

status	texts	released
public domain by term: the author and every credited translator died in 1955 or earlier (Art. 36)	887	yes
in copyright: 17 published by Wolne Lektury under CC BY-SA 3.0, 1 with a translator who died in 1960	18	no
author or translator not verified, or the text does not match its file	95	no

Of the 887 public-domain texts, 666 are not released because wolne_lektury or eltec_pol already carries them (see Filters applied); 221 are released.

Translator death years added on 2026-10-06 cite Biblioteka Narodowa and VIAF authority records, listed in the screen script. They identify the cryptonym "A. W." on 29 Chekhov texts as Antoni Lange (d. 1929). They also show that Wola mocy (1000_novels_0462) was co-translated by Stefan Frycz (d. 1960), so it is dropped although its Wolne Lektury footer says public domain. A Wolne Lektury public-domain footer alone does not clear a text: 1000_novels_0734 (translator "A.L.", not identified) is dropped as unresolved. Edition and critical-edition rights (Art. 99¹–99²) are not assessed.

Attribution

The author column credits the author and, for translations, the translator: Hans Christian Andersen (tłum. Cecylia Niewiadomska). 1000_novels.attribution.jsonl has, per row: author, author's death year, translator, title, how the title was read (file: header text matching the CLARIN file name, 213; position: the line beside the author line, 1; manual: read by hand, 7), CLARIN file name, screen status and basis.

Provenance

Pulled from SpeakLeash's public redistribution (speakleash-ds-pub, key 1000_novels_corpus_CLARIN-PL) of the CLARIN-PL corpus; row i matches the i-th file of the CLARIN listing. Rebuilt by src/clean_1000_novels.py from the v0.2.5 Parquet, the screen CSV and the wolne_lektury and eltec_pol Parquets it deduplicates against, all pinned by SHA-256 in the script. Text, added, created and token_count are unchanged.

Statistics

documents	characters	tokens (tiktoken proxy)
221	26,625,861	10,759,728

Per-document license metadata

license	documents
public-domain	221

Author metadata present for 221 documents.

Filters applied

• rights screen: read 1,000 → 887 public domain (dropped 18 in copyright, 95 unresolved)
• cross-source dedup (polish-dynaword-v2#45): 887 → kept 221. A text is dropped when at least half of its 5-word shingles (near_dedup tokenisation, Wolne Lektury footer cut first) are in wolne_lektury documents (629 texts) or in eltec_pol documents (37; none in both). No text covers between 0.0727 and 0.5178 against wolne_lektury or between 0.0462 and 0.7771 against eltec_pol. Each screened text's covers and matched document ids are in 1000_novels.dedup.tsv. One dropped text has a part the other sources lack: the last section (IV, about 1,100 words) of Orkan's Listy ze wsi (1000_novels_0493, cover 0.5178 against Wolne Lektury's Płanety).
• no length, language, OCR or within-source deduplication filters in this rebuild

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