Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 236, in _generate_tables
                  pa_table = paj.read_json(
                             ^^^^^^^^^^^^^^
                File "pyarrow/_json.pyx", line 342, in pyarrow._json.read_json
                File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
              pyarrow.lib.ArrowInvalid: JSON parse error: Column() changed from object to string in row 0
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 93, in _split_generators
                  pa_table = next(iter(self._generate_tables(**splits[0].gen_kwargs, allow_full_read=False)))[1]
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 250, in _generate_tables
                  batch = json_encode_fields_in_json_lines(original_batch, json_field_paths)
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/utils/json.py", line 90, in json_encode_fields_in_json_lines
                  examples = [ujson_loads(line) for line in original_batch.splitlines()]
                              ^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/utils/json.py", line 20, in ujson_loads
                  return pd.io.json.ujson_loads(*args, **kwargs)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
              ValueError: Expected object or value
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 65, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                         ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Datasets in the GPT-NL Corpus

GPT-NL is made possible by the following parties in contributing their collections for LLM Pretraining:

GPT-NL Public Corpus

kb vng officiele-bekendmakingen woogle Tweede-Kamer Rijksoverheid Rechtspraak Nationaal Archief Utrechts Archief Noord-Hollands Archief Zeeuws Archief DANS Naturalis Wikiwijs EP CommonCorpus


GPT-NL Private Corpus

NDP Nieuwsmedia BNR NTvG ANP DNB ICTRecht ivdnt Movisie Centerdata Waarbenjij Iselinge Saxion Driestar

Note: all parties with collections in the Private Corpus have signed the Content Contributor Agreement (found on https://gpt-nl.nl/samenwerken/content-board). This agreement specifies uniform agreements and responsibilites of the GPT-NL team and the Data Contributor.

This page is meant to show the make-up of the full GPT-NL (Public AND Private) Corpus. More details about every collection can be found in the file GPT-NL-Corpus-metadata.json.




How to read GPT-NL-Corpus-metadata.json

The metadata for all collections from the Public Corpus like American-stories as well as for all collections from the Private Corpus like Instituut voor de Nederlandse Taal) can be viewed in GPT-NL-Corpus-metadata.json. For every collection, a couple of questions will be answered:

  • What is this dataset?

  • Where did it come from?

  • What are the usage rights?

  • What is the overall data quality?

  • What time period does it cover?
    Note: The filled in metadata is done in collaboration with the data contributors. The metadata collection for some sets is still ongoing and might be adjusted over time when we receive more information. If you have additional information about any of the datasets that could be of use, please write to us.

  • What does it look like “in numbers”?
    Note: The aggregated metadata is based on the output from the GPT-NL Curation Pipeline.

An example collection entry:

  "Instituut voor de Nederlandse Taal": {
    "description": "Het Instituut voor de Nederlandse Taal (INT) is a Dutch institute that focuses on the Dutch language and its use in society. It conducts research, provides language resources, and promotes the Dutch language.\"",
    "origin": "Multiple sources - each documented within this document; Lassy Groot-corpus Commercieel--- De dataset is samengesteld tijdens het STEVIN-project (“Essential Speech and Language Technology Resources for Dutch” 2004 - 2011). Dat was een Vlaams-Nederlands onderzoeksprogramma voor taal- en spraaktechnologie, gefinancierd door de Vlaamse en de Nederlandse overheid en gecoördineerd door de Taalunie. Binnen dat onderzoekprogramma was er het project Large Scale Syntactic Annotation of written Dutch (LASSY). Er werd materiaal van externe bronnen verzameld. DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- De dataset is samengesteld tijdens het STEVIN-project (“Essential Speech and Language Technology Resources for Dutch” 2004 - 2011). Dat was een Vlaams-Nederlands onderzoeksprogramma voor taal- en spraaktechnologie, gefinancierd door de Vlaamse en de Nederlandse overheid en gecoördineerd door de Taalunie. Binnen dat onderzoekprogramma was er het project DAESO (Detecting And Exploiting Semantic Overlap). Er werd materiaal van externe bronnen verzameld. JASMIN-spraakcorpus--- De dataset is samengesteld tijdens het STEVIN-project (“Essential Speech and Language Technology Resources for Dutch” 2004 - 2011). Dat was een Vlaams-Nederlands onderzoeksprogramma voor taal- en spraaktechnologie, gefinancierd door de Vlaamse en de Nederlandse overheid en gecoördineerd door de Taalunie. Binnen dat onderzoekprogramma was er het project JASMIN-CGN-project (Jongeren, Anderstaligen, Senioren en Machine Interactie voor het Nederlands). Het spraakmateriaal werd tijdens de loop van het project gecreëerd. De annotaties van het Corpus Gesproken Nederlands--- De dataset is samengesteld tijdens het project Corpus Gesproken Nederlands, gefinancierd door de Vlaamse en Nederlandse regering en door de Nederlandse Organisatie voor Wetenschappelijk Onderzoek. Het overgrote deel van het spraakmateriaal werd tijdens de loop van het project gecreëerd, de andere data werd verzameld bij externe bronnen, zoals radio- en televisiestations. Corpus Middelnederlands--- The main sources for this corpus are the rhyming texts and prose texts from the CD-ROM Middle Dutch, compiled bij the INL/INT and published in 1998 by the Flemish Standaard Uitgeverij and the Dutch Sdu. Het corpus is samengesteld op basis van overwegend kritische tekstedities; wetenschappelijk verantwoorde gedrukte uitgaven. Corpus Gysseling--- The Corpus Gysseling is the collection of all thirteenth-century texts that served as source material for the Vroegmiddelnederlands Woordenboek (VMNW; Dictionary of Early Middle Dutch). It is the digital edition, enriched with part of speech and lemma, of the thirteenth-century material from the Corpus van Middelnederlandse teksten (tot en met het jaar 1300) - Corpus of Middle Dutch texts (up to and including the year 1300) - published in the period 1977-1987 by the Ghent linguist Maurits Gysseling.",
    "modality": "text",
    "license": {
      "type": [
        "GPT-NL Proprietary"
      ]
    },
    "relevance": {
      "rating": "high",
      "rationale": "See the descriptions. Lassy Groot-corpus Commercieel--- Eindhoven Corpus, kranten, magazines, boeken, neergeschreven gesproken taal (groepsdiscussies, interviews, gesproken brieven, vrije conversaties in ongedwongen sfeer); EMEA corpus: documents from the European Medicines Agency; Europarl; proceedings of the European Parliament from 1996 to 2012; Wikipedia; Troonredes koningin Beatrix; Senseval; onduidelijk, maar vermoedelijk boekenmateriaal?; SONAR: zie volgende tabblad DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- Vertalingen van oudere boeken (Le Petit Prince, On the origin of species, Les Essais), autocue-ondertitelparen (NOS), nieuwskoppen (Google News), output van een QA-system (medical domain), persberichten (ANP en Novum) JASMIN-spraakcorpus--- Voorgelezen spraak (we selected texts of the nine different reading levels from books that belong to the reading programme Veilig Leren Lezen. For the non-native speakers we selected appropriate texts from a widely used method for learning Dutch as a second language. The texts were selected as to be suitable for learners with CEF levels A1 and A2. ) - Mens-machine-interacties (vaste onderwerpen, zoals het plannen van een reis of van een leuke dag) De annotaties van het Corpus Gesproken Nederlands--- Spontane conversaties ('face-to-face'); Interviews met leraren Nederlands; Telefoondialogen; Gesimuleerde zakelijke onderhandelingen; Interviews en discussie uitgezonden op radio en televisie; Discussie, debatten, vergaderingen (m.n. politieke); Lessen; Spontane commentaren (o.a. sport) uitgezonden op radio en televisie; Actualiteitenrubrieken en reportages uitgezonden op radio en televisie; Nieuwsbulletins uitgezonden op radio en televisie; Beschouwingen en commentaren uitgezonden op radio en televisie; Missen, lezingen, plechtige toespraken; Colleges, voordrachten, lezingen; Voorgelezen teksten Corpus Middelnederlands--- Diverse topics in een aantal verschillende soorten teksten, bv. bijbel, medische teksten, verhalen, liedjes… Corpus Gysseling--- Official documents, Literary texts and artes texts"
    },
    "quality": {
      "rating": "medium",
      "rationale": "Lassy Groot-corpus Commercieel--- Gevarieerde kwaliteit, maar de meeste data is gepubliceerd en is dus afkomstig van professionele schrijvers. DAESO-corpus --- parallelle Nederlandstalige monolinguale treebank Commercieel--- Gevarieerde kwaliteit, maar de meeste data is afkomstig van auteurs en journalisten. JASMIN-spraakcorpus--- Bevat spreektalige fenomenen, zoals aarzelingen, versprekingen, herhalingen,… en een deel van de data is afkomstig van personen die het Nederlands niet als moedertaal hebben. De annotaties van het Corpus Gesproken Nederlands--- Bevat spreektalige fenomenen, zoals aarzelingen, versprekingen, herhalingen,… Een deel van de data is afkomstig van professionele sprekers, zoals radio- en televisiemensen. Corpus Middelnederlands--- Onduidelijk, professionele auteurs? Corpus Gysseling--- Onduidelijk, professionele auteurs en ambtenaren?"
    },
    "temporal_coverage": {
      "distribution": {
        "pre_1950": 0.05,
        "y1950_2000": 0.75,
        "y2000_2020": 0.2,
        "post_2020": 0.0
      }
    },
    "acquisition": {
      "method": "Lassy Groot-corpus Commercieel--- Data werd aangeleverd, verzameld op het web, overgenomen uit andere corpora DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- De Google News-koppen werden van het net gehaald, de rest werd op een andere manier verzameld/overgenomen. JASMIN-spraakcorpus--- Opnames De annotaties van het Corpus Gesproken Nederlands--- Opnames van spraak + overnemen van spraakopnames uit andere projecten en van externe leveranciers Corpus Middelnederlands--- The Corpus Middelnederlands presented in this application contains classical works of Middle Dutch literature like Beatrijs, Van den vos Reynaerde, the abele spelen, the stories about King Arthur or about Charlemagne, all texts from the famous Gruuthuse manuscript (including the Egidius song), but also many of the lesser known or less researched texts, such as prose adaptations of the rhyming knight’s tales (the so-called ‘chapbooks’), collections of songs such as the Antwerp Songbook, several Bible translations, hagiographies, books of prayer, chronicles, and all kinds of religious, didactic and scientific treatises, medical manuals and recipes. Corpus Gysseling--- The Corpus Gysseling is the collection of all thirteenth-century texts that served as source material for the Vroegmiddelnederlands Woordenboek (VMNW; Dictionary of Early Middle Dutch). It is the digital edition, enriched with part of speech and lemma, of the thirteenth-century material from the Corpus van Middelnederlandse teksten (tot en met het jaar 1300) - Corpus of Middle Dutch texts (up to and including the year 1300) - published in the period 1977-1987 by the Ghent linguist Maurits Gysseling.",
      "collected_at": "Lassy Groot-corpus Commercieel--- Het LASSY-project liep van 2007 tot 2010, maar er werd ook data overgenomen van de DCOI- en het SoNaR-projecten, die respectievelijk liepen van 2005 tot 2006 en van 2008 tot 2008 tot 2011. Voor andere data is het onduidelijk (bv. Senseval) DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- Het DAESO-project liep van 2006-2009, maar er werd ook de Q&A-data werd overgenomen uit het ATRANOS-project (2000-2004). JASMIN-spraakcorpus--- De spraak werd opgenomen in de periode 2005-2007. De annotaties van het Corpus Gesproken Nederlands--- De spraakopnames werden verzameld/opgenomen in de periode 1999-2003. Corpus Middelnederlands--- Eind twintigste eeuw Corpus Gysseling--- jaren '70",
      "collector": "Lassy Groot-corpus Commercieel--- Journalisten, ambtenaren, auteurs, taalkundigen, particulieren,… DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- Voornamelijk journalisten, ook auteurs. JASMIN-spraakcorpus--- Een gevarieerde groep sprekers uit Vlaanderen en Nederland bestaande uit kinderen, senioren en anderstaligen. De annotaties van het Corpus Gesproken Nederlands--- Een gevarieerde groep volwassen sprekers uit Vlaanderen en Nederland (enkel moedertaalsprekers) Corpus Middelnederlands--- Middelnederlandse auteurs Corpus Gysseling--- Middelnederlandse auteurs"
    },
    "processing": {
      "modifications": "Lassy Groot-corpus Commercieel--- Bij het SoNaR-project werd er aan normalisatie en correctie gedaan - Where diacritics were missing and the word form without diacritics was not a valid word in its own right, fully automatic replacement was mostly possible and has been effected. - Text correction was performed by extracting all the word pairs from a corpus that display a particular difference in the bag of characters making up the words in the pairs. Onduidelijk voor andere subcorpora. Het INT heeft uit het Lassy Groot-corpus bestanden verwijderd die niet commercieel gebruikt mochten worden, wat resulteerde in een commerciële versie van het Lassy Groot-corpus. DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- \"Preprocessing in general involved converting all text material to XML format with UTF-8 character encoding. All book translations were (mostly) automatically converted from their original electronic format (raw text, MS Word, PDF) to XML adhering to the TEI Lite standard, the light version of the Text Encoding Initiative markup language(Burnard and Sperberg-McQueen 2006). The original document structure and formatting was as much as possible preserved in the markup; at a minimum, all books have markup indicating chapters and sections. In addition, manual markup was added to indicate parts of the texts which were not fit for our purposes, e.g., citations in a foreign language. - All text material was subsequently tokenized\". Het INT heeft de aangeleverde dataset niet gewijzigd. JASMIN-spraakcorpus--- Het INT heeft de aangeleverde dataset niet gewijzigd. De annotaties van het Corpus Gesproken Nederlands--- Het INT heeft de aangeleverde dataset niet gewijzigd. Corpus Middelnederlands--- Nee Corpus Gysseling--- Nee"
    },
    "notes": {
      "pii": "Lassy Groot-corpus Commercieel--- In de neerslag van de discussies, vrije conversaties uit het Eindhoven Corpus (maar data van 50 jaar geleden). DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- Geen persoonlijke data. JASMIN-spraakcorpus--- Vermoedelijk niet, want de gesprekken werden inhoudelijk zwaar gestuurd. De annotaties van het Corpus Gesproken Nederlands--- Mogelijk persoonlijke informatie in de face-to-face- en de telefoongesprekken, maar alle sprekers waren er zich volledig bewust van dat de data verspreid zou worden voor onderzoeks- en ontwikkelingsdoeleinden. Corpus Middelnederlands--- Onduidelijk of de data persoonlijke data bevat, maar sowieso niet relevant omwille van het historische karakter van de dataset. Corpus Gysseling--- Onduidelijk of de data persoonlijke data bevat, maar sowieso niet relevant omwille van het historische karakter van de dataset.\nLassy Groot-corpus Commercieel--- Als er al persoonlijke informatie in de data zit, dan werd die niet intentioneel opgenomen. DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- Niet van toepasssing JASMIN-spraakcorpus--- Als er al persoonlijke informatie in de data zit, dan werd die niet intentioneel opgenomen. De annotaties van het Corpus Gesproken Nederlands--- Als er al persoonlijke informatie in de data zit, dan werd die niet intentioneel opgenomen. Corpus Middelnederlands--- Niet van toepasssing Corpus Gysseling--- Niet van toepasssing\nLassy Groot-corpus Commercieel--- Nee DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- Niet van toepasssing JASMIN-spraakcorpus--- Het oorspronkelijke consortium heeft de sprekerinformatie geanonimiseerd. De annotaties van het Corpus Gesproken Nederlands--- Het oorspronkelijke consortium heeft de sprekerinformatie geanonimiseerd. Corpus Middelnederlands--- Niet van toepasssing, Corpus Gysseling--- Niet van toepasssing",
      "harmful_or_biased_language": "Lassy Groot-corpus Commercieel--- Er komen vloeken en scheldwoorden voor (bv. makak, klootzak, kut, hoer,…), voornamelijk in de ondertiteldata en de kranten/magazines. (Ook in de discussielijsten maar die data moet genegeerd worden). DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- Afgaande op de verschillende soorten tekst, vermoedelijk niet. JASMIN-spraakcorpus--- Vermoedelijk niet, want de gesprekken werden inhoudelijk zwaar gestuurd. De annotaties van het Corpus Gesproken Nederlands--- In de spontane conversaties en de voorgelezen boeken zit er in beperkte mate \"harmful content\", zoals offensive language. En de zakelijke onderhandelingen zijn gesimuleerd. Corpus Middelnederlands--- Het is mogelijk dat de data termen bevat die in onze tijd niet meer aanvaarbaar zijn. Corpus Gysseling--- Het is mogelijk dat de data termen bevat die in onze tijd niet meer aanvaarbaar zijn.\nLassy Groot-corpus Commercieel--- Onduidelijk DAESO-corpus; parallelle Nederlandstalige monolinguale treebank Commercieel--- Onduidelijk JASMIN-spraakcorpus--- Niet van toepassing, de inhoud van de spraakopnames stond vast of was sterk gestuurd. De annotaties van het Corpus Gesproken Nederlands--- Onduidelijk Corpus Middelnederlands--- Onduidelijk Corpus Gysseling--- Onduidelijk"
    },
    "aggregated_stats": {}

General Information

You’ll see:

  • description: a plain-language summary of what the dataset contains.
  • organization (optional): who produced or published it.
  • origin: where the content was sourced from (archive, institution, website, etc.).
  • modality: what kind of data it is --> always text.


Licensing Information

Look in license:

  • type: the license label(s) that apply (e.g., Creative Commons, proprietary).
  • details (optional): extra clarification, for example:
    • attribution requirements,
    • differences between the dataset license vs. license of underlying content,


Quality

Two short assessments:

  • relevance

    • rating: usually high / medium / low
    • rationale: why it’s considered relevant (or not) for the intended context of training GPT-NL
  • quality

    • rating: usually high / medium / low
    • rationale: why the data is considered clean/reliable (or potentially noisy—e.g., OCR issues)


Recency

Under temporal_coverage you’ll find:

  • distribution: a rough split of the dataset across time buckets:
    • pre_1950
    • y1950_2000
    • y2000_2020
    • post_2020

These values are proportions, giving a quick sense of how “old” or “recent” the content is.


Collections and Preparation

Some datasets include extra background:

  • acquisition

    • collected_at: when it was collected (often month/year)
    • collector: who collected/assembled it
  • processing

    • modifications: what was done to the data (OCR, filtering, cleanup, deduplication, layout extraction, etc.)


Notes, caveats, and risks (optional)

The notes section is based on observations from the original data collector or GPT-NL team.

  • pii: whether personal information may appear, and what the expected risk is.
  • harmful_or_biased_language: warnings about historical bias, offensive language, misinformation, or unfiltered editorial content.

This is especially common for historical corpora or sources that weren’t created with modern safety expectations.

Note: these observations are about the original content. It might be that a lot of the original concerns are alleviated by the GPT-NL Curation Process.


Links and references (optional)

  • links: URLs to relevant external sources. Numbered as referenced in the rest of the collection entry.


aggregated_stats (optional)

If present, aggregated_stats summarizes the dataset:

  • files

    • how many storage files there are (e.g., parquet files)
    • total size in bytes
  • rows

    • total number of records/documents
  • language_stats

    • total_counts: how many records per language (e.g., nl, en, de)
    • by_language: per-language summaries, including:
      • language_score (confidence score for that language -> often used as a quality indicator for text)
      • n_char (amount of characters in specific language)
      • n_non_symbol_words (amount of words -- excluding symbolic strings)
Downloads last month
6

Collection including GPT-NL/Collection-metadata