Hugging Face's logo

Datasets:
coral-nlp
/
german-commons

Tasks:
Text Generation
Modalities:
Tabular
Text
Sub-tasks:
language-modeling
Languages:
German
Size:
10M - 100M
ArXiv:
Tags:
german
commons
legal
scientific
cultural
political
License:
Dataset card Data Studio Files
xet
Community
2
german-commons / README.md
lgienapp's picture
lgienapp
Update README.md
f38a1d8 verified
preview code
|
raw
history blame contribute delete
18.1 kB
metadata
annotations_creators:
  - machine-generated
language_creators:
  - found
language:
  - de
license:
  - odc-by
multilinguality:
  - monolingual
size_categories:
  - 100B<n
source_datasets: []
task_categories:
  - text-generation
task_ids:
  - language-modeling
pretty_name: German Commons
tags:
  - german
  - commons
  - legal
  - scientific
  - cultural
  - political
  - web
  - news
  - economic
configs:
  - config_name: default
    data_files: subset=*/source=*/*.parquet
    default: true
    features:
      - name: id
        dtype: string
      - name: source
        dtype: string
      - name: subset
        dtype:
          class_label:
            names:
              - Cultural
              - Economic
              - Legal
              - News
              - Political
              - Scientific
              - Web
      - name: text
        dtype: string
      - name: license
        dtype:
          sequence: string
      - name: num_tokens
        dtype: int64
      - name: perplexity
        dtype: float64
      - name: ocr_score
        dtype: int64
    splits:
      - name: train
        num_examples: 36773579
  - config_name: cultural
    data_files:
      - split: blbooks
        path: subset=Cultural/source=BLBooks/*.parquet
      - split: dibilit
        path: subset=Cultural/source=DiBiLit/*.parquet
      - split: dibiphil
        path: subset=Cultural/source=DiBiPhil/*.parquet
      - split: germanpd
        path: subset=Cultural/source=GermanPD/*.parquet
      - split: mosel
        path: subset=Cultural/source=MOSEL/*.parquet
      - split: sbbfulltexts
        path: subset=Cultural/source=SBB Fulltexts/*.parquet
      - split: wikisource
        path: subset=Cultural/source=Wikisource/*.parquet
      - split: wikivoyage
        path: subset=Cultural/source=Wikivoyage/*.parquet
      - split: wikiquote
        path: subset=Cultural/source=Wikiquote/*.parquet
    features:
      - name: id
        dtype: string
      - name: source
        dtype: string
      - name: subset
        dtype:
          class_label:
            names:
              - Cultural
              - Economic
              - Legal
              - News
              - Political
              - Scientific
              - Web
      - name: text
        dtype: string
      - name: license
        dtype:
          sequence: string
      - name: num_tokens
        dtype: int64
      - name: perplexity
        dtype: float64
      - name: ocr_score
        dtype: int64
  - config_name: economic
    data_files:
      - split: tedeutenders
        path: subset=Economic/source=TEDEUTenders/*.parquet
    features:
      - name: id
        dtype: string
      - name: source
        dtype: string
      - name: subset
        dtype:
          class_label:
            names:
              - Cultural
              - Economic
              - Legal
              - News
              - Political
              - Scientific
              - Web
      - name: text
        dtype: string
      - name: license
        dtype:
          sequence: string
      - name: num_tokens
        dtype: int64
      - name: perplexity
        dtype: float64
      - name: ocr_score
        dtype: int64
  - config_name: legal
    data_files:
      - split: bverfgaes
        path: >-
          subset=Legal/source=Amtliche Entscheidungssammlung des
          Bundesverfassungsgerichts/*.parquet
      - split: bundesrecht
        path: subset=Legal/source=Deutsches Bundesrecht/*.parquet
      - split: bag
        path: subset=Legal/source=Entscheidungen des Bundesarbeitsgerichts/*.parquet
      - split: bfh
        path: subset=Legal/source=Entscheidungen des Bundesfinanzhofs/*.parquet
      - split: bgh
        path: subset=Legal/source=Entscheidungen des Bundesgerichtshofs/*.parquet
      - split: bgh20
        path: >-
          subset=Legal/source=Entscheidungen des Bundesgerichtshofs in
          Strafsachen aus dem 20. Jahrhundert/*.parquet
      - split: bpatg
        path: subset=Legal/source=Entscheidungen des Bundespatentgerichts/*.parquet
      - split: bverfg
        path: >-
          subset=Legal/source=Entscheidungen des
          Bundesverfassungsgerichts/*.parquet
      - split: bverwg
        path: >-
          subset=Legal/source=Entscheidungen des
          Bundesverwaltungsgerichts/*.parquet
      - split: eurlex
        path: subset=Legal/source=EurLEX/*.parquet
      - split: openlegaldata
        path: subset=Legal/source=Open Legal Data/*.parquet
    features:
      - name: id
        dtype: string
      - name: source
        dtype: string
      - name: subset
        dtype:
          class_label:
            names:
              - Cultural
              - Economic
              - Legal
              - News
              - Political
              - Scientific
              - Web
      - name: text
        dtype: string
      - name: license
        dtype:
          sequence: string
      - name: num_tokens
        dtype: int64
      - name: perplexity
        dtype: float64
      - name: ocr_score
        dtype: int64
  - config_name: news
    data_files:
      - split: anno
        path: news_anno_*.parquet
      - split: zeitungsportal
        path: news_deutsches_zeitungsportal_*.parquet
      - split: europeana
        path: news_europeana_newspapers_*.parquet
      - split: wikinews
        path: news_wikinews_*.parquet
    features:
      - name: id
        dtype: string
      - name: source
        dtype: string
      - name: subset
        dtype:
          class_label:
            names:
              - Cultural
              - Economic
              - Legal
              - News
              - Political
              - Scientific
              - Web
      - name: text
        dtype: string
      - name: license
        dtype:
          sequence: string
      - name: num_tokens
        dtype: int64
      - name: perplexity
        dtype: float64
      - name: ocr_score
        dtype: int64
  - config_name: political
    data_files:
      - split: btdrucksachen
        path: subset=Political/source=Drucksachen des Bundestages/*.parquet
      - split: eurovoc
        path: subset=Political/source=EuroVoc/*.parquet
      - split: germanpoliticalspeeches
        path: subset=Political/source=German Political Speeches/*.parquet
      - split: btplenarprotokolle
        path: subset=Political/source=Plenarprotokolle des Bundestages/*.parquet
      - split: reichtagsprotokolle
        path: subset=Political/source=Reichtagsprotokolle/*.parquet
    features:
      - name: id
        dtype: string
      - name: source
        dtype: string
      - name: subset
        dtype:
          class_label:
            names:
              - Cultural
              - Economic
              - Legal
              - News
              - Political
              - Scientific
              - Web
      - name: text
        dtype: string
      - name: license
        dtype:
          sequence: string
      - name: num_tokens
        dtype: int64
      - name: perplexity
        dtype: float64
      - name: ocr_score
        dtype: int64
  - config_name: scientific
    data_files:
      - split: arxiv
        path: subset=Scientific/source=arXiv/*.parquet
      - split: doab
        path: subset=Scientific/source=Directory of Open Access Books/*.parquet
      - split: polyjournal
        path: subset=Scientific/source=Polytechnisches Journal/*.parquet
      - split: wikibooks
        path: subset=Scientific/source=Wikibooks/*.parquet
      - split: openalex
        path: subset=Scientific/source=OpenAlex/*.parquet
      - split: wikiversity
        path: subset=Scientific/source=Wikiversity/*.parquet
    features:
      - name: id
        dtype: string
      - name: source
        dtype: string
      - name: subset
        dtype:
          class_label:
            names:
              - Cultural
              - Economic
              - Legal
              - News
              - Political
              - Scientific
              - Web
      - name: text
        dtype: string
      - name: license
        dtype:
          sequence: string
      - name: num_tokens
        dtype: int64
      - name: perplexity
        dtype: float64
      - name: ocr_score
        dtype: int64
  - config_name: web
    data_files:
      - split: onemillionposts
        path: subset=Web/source=One Million Posts/*.parquet
      - split: thestack
        path: subset=Web/source=The Stack/*.parquet
      - split: wikidiscussions
        path: subset=Web/source=Wiki Discussions/*.parquet
      - split: wikipedia
        path: subset=Web/source=Wikipedia/*.parquet
      - split: youtubecommons
        path: subset=Web/source=Youtube Commons/*.parquet
    features:
      - name: id
        dtype: string
      - name: source
        dtype: string
      - name: subset
        dtype:
          class_label:
            names:
              - Cultural
              - Economic
              - Legal
              - News
              - Political
              - Scientific
              - Web
      - name: text
        dtype: string
      - name: license
        dtype:
          sequence: string
      - name: num_tokens
        dtype: int64
      - name: perplexity
        dtype: float64
      - name: ocr_score
        dtype: int64

German Commons - 154 Billion Tokens of Openly Licensed Text for German Language Models

A comprehensive collection of German-language text data under open licenses for training German language models.

Dataset Description

This dataset is aggregated from 41 diverse sources and contains 154.56 billion tokens of German text data with 35.78 million documents spanning 7 thematic domains:

  • 🌐 Web Commons: 19.89B tokens source from Wiki projects, online discussions, code repositories, social media posts, YouTube transcripts
  • 💬 Political Commons: 3.57B tokens sourced from parliamentary documents, speeches, protocols, political vocabulary
  • ⚖️ Legal Commons: 2.99B tokens sourced from court decisions, federal law, legal databases, EU legal documents
  • 📰 News Commons: 72.67B tokens sourced from historical and current newspapers archives
  • 🏦 Economics Commons: 0.11B tokens sourced from EU public tenders
  • 📚 Cultural Commons: 54.49B tokens sourced from cultural heritage collections
  • 🔬 Scientific Commons: 0.84B tokens sourced from scholarly papers, books, and technical journals

Dataset Features

Each record contains the following fields:

  • id: Unique identifier string, as per each documents' source dataset
  • source: Source dataset name
  • subset: Thematic subset (Cultural, Legal, Political, Scientific, News, Web, Economic)
  • text: Main text content; deduplicated, quality filtered, with consistent formatting and encoding. Can be split at newlines to obtain paragraph text.
  • license: List of applicable licenses for each document, given as canonical SPDX license URL.
  • num_tokens: GPT-2 token count
  • perplexity: Text perplexity measured with a KenLM model trained on German Wikipedia text
  • ocr_score: OCR quality score measured using OCRoscope

Dataset Usage

  • Load the entire dataset

    from datasets import load_dataset

ds = load_dataset("coral-nlp/german-commons")

  • Load a thematic subset

    ds = load_dataset("coral-nlp/german-commons", "cultural")

  • Load individual source datasets

    wikipedia = load_dataset("coral-nlp/german-commons", "web", split="wikipedia")

Supported splits and constituent datasets are:

Subset Split Key Dataset Name Docs Tokens License Text Type Source
web wikipedia Wikipedia 2,930,224 2,948,751,608 CC-BY-SA-4.0 Various 🔗
web wikivoyage Wikivoyage 20,370 42,025,478 CC-BY-SA-4.0 Travel 🔗
web wikidiscussions Wikipedia Discussions 8,349,076 1,218,210,917 CC-BY-SA-4.0 Online Discussions 🔗
web youtubecommons YouTube Commons 2,809,714 14,478,850,964 Various Video Subtitles 🔗
web onemillionposts One Million Posts Corpus 946,082 94,872,633 CC-BY-4.0 Online Discussions 🔗
web thestack The Stack (Markdown and TXT Subsets) 421,466 1,105,173,228 Various Various 🔗
political reichtagsprotokolle Reichtagsprotokolle 522 703,495,637 CC-BY-SA-4.0 Parliamentary Protocols 🔗
political germanpoliticalspeeches German Political Speeches 6,678 29,409,655 CC-BY-4.0 Speech Transcripts 🔗
political btdrucksachen Corpus der Drucksachen des Deutschen Bundestages 3,017 528,769,669 CC0-1.0 Parliamentary Publications 🔗
political btplenarprotokolle Corpus der Plenarprotokolle des Deutschen Bundestages 1,833 316,034,708 CC0-1.0 Parliamentary Protocols 🔗
political eurovoc EuroVoc 245,838 1,988,111,462 EUPL Parliamentary Publications 🔗
legal bundesrecht Corpus des Deutschen Bundesrechts 3,217 1,004,294 CC0-1.0 German Federal Laws 🔗
legal openlegaldata OpenLegalData 249,909 1,915,956,613 CC0-1.0 Court Decisions 🔗
legal bfh Corpus der Entscheidungen des BFH 10,885 67,791,931 CC0-1.0 Court Decisions 🔗
legal bgh20 Entscheidungen des BGH in Strafsachen des 20. Jhd. 36,062 92,873,390 CC0-1.0 Court Decisions 🔗
legal bgh Corpus der Entscheidungen des BGH 77,258 292,832,709 CC0-1.0 Court Decisions 🔗
legal bverfg Corpus der Entscheidungen des BVerfG 8,028 39,503,223 CC0-1.0 Court Decisions 🔗
legal bpatg Corpus der Entscheidungen des BpatG 30,705 185,099,188 CC0-1.0 Court Decisions 🔗
legal bverwg Corpus der Entscheidungen des BVerwG 27,185 123,487,739 CC0-1.0 Court Decisions 🔗
legal bverfgaes Corpus der amtl. Entscheidungssammlung des BVerfG 919 24,427,294 CC0-1.0 Court Decisions 🔗
legal bag Corpus der Entscheidungen des BAG 5,624 48,248,111 CC0-1.0 Court Decisions 🔗
legal eurlex EurLEX 64,934 201,263,562 CC-BY-4.0 European Union Laws 🔗
news zeitungsportal Deutsches Zeitungsportal 8,076,164 43,871,094,547 CC0-1.0 News Articles 🔗
news europeana Europeana Newspapers 3,256,341 20,684,418,365 CC0-1.0 News Articles 🔗
news anno ANNO 1,910,281 8,103,825,248 CC0-1.0 News Articles 🔗
news wikinews WikiNews 23,266 14,222,520 CC-BY-4.0 News Articles 🔗
economic tedeutenders TEDEUTenders 57,214 110,611,112 CC0-1.0 Procurement Notices 🔗
cultural dibilit DiBiLit-Korpus 2,062 216,391,448 CC-BY-SA-4.0 Literature 🔗
cultural dibiphil DiBiPhil-Korpus 269 32,151,997 CC-BY-SA-4.0 Literature 🔗
cultural wikisource Wikisource 240,689 347,770,430 CC-BY-SA-4.0 Various 🔗
cultural germanpd German-PD 123,592 49,333,198,231 CC0-1.0 Literature 🔗
cultural blbooks BLBooks 3,714 1,012,047,216 CC0-1.0 Literature 🔗
cultural mosel MOSEL 3,127,203 3,181,917,752 CC-BY-4.0 Speech Transcripts 🔗
cultural sbbfulltexts SBB Fulltexts 2,605,569 358,514,283 CC-BY-4.0 Literature 🔗
cultural wikiquote Wikiquote 8,612 6,688,458 CC-BY-4.0 Quotes & Proverbs 🔗
scientific wikibooks Wikibooks 346 180,257,799 CC-BY-SA-4.0 Educational Books 🔗
scientific polyjournal Digitalisierung des Polytechnischen Journals 27,292 50,434,996 CC-BY-SA-4.0 Scholarly Papers 🔗
scientific doab Directory of Open Access Books 1,939 166,920,321 Various Scholarly Books 🔗
scientific arxiv arXiv 8 103,478 Various Scholarly Papers 🔗
scientific openalex OpenAlex 47,733 413,632,648 Various Educational Content 🔗
scientific wikiversity Wikiversity 16,371 27,802,099 CC-BY-SA-4.0 Scholarly Papers 🔗

Citation

If you use this dataset, please cite the correponding paper:

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

License

This dataset aggregation and metadata is released under ODC-BY license. Individual documents have their own specific licenses - please check the license field for each record.