grundtvig-bot's picture
doab: add the German open-access books (1,939 books, 124.7M tokens)
0b80fb5
|
Raw History Blame Contribute Delete
5.68 kB
metadata
pretty_name: Directory of Open Access Books
language:
  - de
license: cc-by-sa-4.0
license_name: CC-BY-SA 4.0
size_categories:
  - 1K<n<10K
task_categories:
  - text-generation
  - fill-mask
task_ids:
  - language-modeling
source_datasets:
  - coral-nlp/german-commons
domains:
  - Books

Dataset Card for Directory of Open Access Books

Full text of German-language open-access scholarly books from DOAB

The Directory of Open Access Books (DOAB) indexes peer-reviewed open-access books from academic publishers. This subset holds the German-language books among them, one document per book: 1,939 monographs, 124.72M Llama tokens.

The material is scholarly prose in German across the humanities and the social sciences — linguistics, history, sociology, literary studies — with the apparatus that comes with a book: footnotes inline, bibliographies, front matter. Unlike the newspaper and court subsets it is edited text rather than OCR, and unlike the treatises in the legal subsets it is not formulaic.

Dataset Description

  • Number of samples: 1.94K
  • Number of tokens (Llama 3): 124.72M
  • Average document length in tokens (min, max): 64.32K (10, 386.22K)

Dataset Structure

An example from the dataset looks as follows.

{
  "id": "doab_/shared/input/fd9218cd-37a5-4bba-98c8-d9a9c18215c4.pdf",
  "text": "Aus den theoretischen und praktisch-angewandten Aufgaben des Fremdsprachenunterrichts in den letzten[...]",
  "source": "doab",
  "added": "2026-11-01",
  "created": "1991-01-01, 2025-12-31",
  "token_count": 54669,
  "license": "CC-BY-SA-4.0"
}

Data Fields

An entry in the dataset consists of the following fields:

  • id (str): A unique identifier for each document.
  • text (str): The content of the document.
  • source (str): The source of the document (see Source Data).
  • added (str): The date when the document was added to this collection.
  • created (str): The date range when the document was originally created.
  • token_count (int): The number of tokens in the sample computed using the Llama 3 tokenizer.

Data Fields

An entry in the dataset consists of the following fields:

  • id (str): A unique identifier for each document.
  • text (str): The content of the document.
  • source (str): The source of the document.
  • license (str): the licence of the document, CC-BY-SA-4.0 for this subset.
  • added (str): The date when the document was added to this collection.
  • created (str): The date range when the document was originally created.
  • token_count (int): The number of tokens in the sample computed using the Llama 3 tokenizer.

Data Splits

The entire corpus is provided in the train split.

Dataset Statistics

Annotation Overview

This subset has no annotations: metadata.parquet carries nulls for every document. Annotation has to be computed from the text, which is what the small annotators in propella-bekko exist for.

Additional Information

Source Data

Taken from the Scientific subset of German Commons, source Directory of Open Access Books, which redistributes the full text of the books it indexes. The ids are the ingest paths German Commons kept (/shared/input/<uuid>.pdf), which carry no bibliographic information.

The books are whole: a document is one monograph, from the first page of the body to the last, and nothing is split or merged. The longest is 386K tokens, which is what a book is, and matches the length of the books already in the corpus.

Annotations

None. A uuid is not an id the annotation release knows, so the text has to be annotated from itself. metadata.parquet is an aligned table of nulls.

License Information

The books are CC-BY-SA-4.0, which the build asserts on every document rather than taking from the collection card.

Limitations

  • created is a bound, not a date. The ids carry no date, the text starts in the body of the book rather than at a title page, and German Commons keeps no date column for this source. Of the 64 books whose text names a DOI that Crossref resolves, the publication years run from 1991 to 2025, and every document carries that window. A book's real publication year is inside it and is not recorded here.
  • The corpus is German-language books only, so it is a small slice of DOAB: the directory is dominated by English titles, and this subset is 1,939 books.
  • Scholarly monographs, so the register is academic throughout, with footnotes inline in the running text and the citation apparatus included.
  • Some books are PDF renderings of a print original, so hyphenation is broken at line ends and tables arrive as columns of numbers.

Citation Information

@article{{gienapp2025germancommons,
  title   = {{{{The German Commons -- 154 Billion Tokens of Openly Licensed Text for German Language Models}}}},
  author  = {{Gienapp, Lukas and Schr{{\"o}}der, Christopher and Schweter, Stefan and Akiki, Christopher and Schlatt, Ferdinand and Zimmermann, Arden and Gen{{\^e}}t, Phillipe and Pottschast, Martin}},
  year    = {{2025}},
  journal = {{CoRR}},
  volume  = {{abs/2510.13996}},
  url     = {{https://arxiv.org/abs/2510.13996}}
}}