Artable

Artwork records digitally published by museums and cultural institutions across the UK, Europe, the Americas and Asia, harvested from each institution's own API or open-data endpoint and normalised into one schema.

Scope of this release

Every row in this release has a picture. The working catalogue holds 23,010,062 records; the 5,604,792 with no image reference are held back for a later release rather than shipped as rows a reader cannot look at. Nothing was deleted — they return when they have pictures.

Configs

Two configs, because the two tables have different schemas. Without this file Hugging Face concatenates both Parquet files into a single split, and the schema conflict breaks the search API.

config	rows	one row is
artwork (default)	17,405,270	an artwork, with an image reference
artist	56,989	an artist, with a biography

Join on artwork.artist = artist.name.

artwork

id, title, title_lang, title_zh, title_alt, artist, artist_zh, year_start, year_end, date_display, medium, category, institution, institution_zh, city, country, region, accession, image_url, thumb_url, iiif, image_commons, image_commons_credit, license, source_url, description, description_zh, history, story, story_zh.

artist

name, name_zh, story, story_zh, story_lang, story_url, born, died, works.

story_lang matters: 11,952 biographies exist only in French, German, Italian or Japanese. The column says which language the text is actually in, so a French article is never presented as though it were written or translated for this dataset.

Honest limits

Stated plainly, because a catalogue that overstates itself is worse than a smaller one.

• Unknown is null. No accession number, licence, language tag or date is ever inferred.
• Chinese titles are marked by provenance. A title in title_zh is one the museum itself published. Machine-translated titles are held in a separate column and labelled in the app; they are never merged into title_zh.
• 40.2% of works carry a Traditional Chinese title, almost all machine-translated through a judged pipeline (each batch is graded by an independent reader and written only on a pass). The remainder are untranslated, not silently left in their source language under a Chinese label.
• "Story" is three different things. 7,812,526 works carry the museum's own description or curator's comment, 2,398,011 more reach a biography of their artist — but only 4,256 have a narrative about that specific object. Do not read the aggregate as the last number.
• 1,301,017 rows had an artist value that named nobody — "Okänd", "Ingen uppgift", "Onbekend / Unknown (Fotograaf)", "vermeld op object" — in 22 languages. These are now null so that "unknown" is not presented as a person's name. Values carrying a real attribution ("Anonymous British", "Unbekannt (Stecher)" where a nationality or role adds information) were kept.
• 192,623 rows carry a raw URI in the artist column instead of a name (http://www.wikidata.org/entity/Q…). These are unresolved identifiers, a known defect, not names. Do not treat them as artist labels.
• Coverage is uneven by source, not by harvest. Several institutions cap paging while reporting success; where a page could not be retrieved it is recorded as a gap rather than silently dropped.

Images

Not included, and must not be mirrored. Rows carry image_url / iiif pointing at the holding institution, which is both the licensing position and why this dataset is ~1.1 GB rather than terabytes.

Attribution

Attribution obligations travel with the data — Europeana, Joconde and Palissy require it. Every row carries institution and license; keep those columns.