There is no dataset for this. There's a standard.
Every item carries a permanent, stable ID. One English-anchored concept layer is reused across the whole family; each language attaches its own cartridge by ID. You license and query it like infrastructure — not a spreadsheet you have to clean.
A deep, layered, addressable corpus — not a word list.
the noun layer — names the body, per language, with plural + noun class
the language of a real consultation — symptom reports, instructions
concord made explicit — agreement that makes a sentence correct
the generative core — build sentences, don't match words
dictionary-grounded evidence behind every term
verbs applied in real clinical context
Watch it build the sentence.
A health verb plus a body concept generates the clinical phrase — by rule, not by lookup. That's the difference between a dictionary and a substrate a model can compute with.
Organized the way medicine is.
Concepts are grouped into clinical domains. The bars show prototype depth in the lead language — where coverage is partial, that's exactly what a partnership funds next.
Bars show depth in the lead language. The family-wide picture — breadth across all 21 languages — is below.
A living platform, growing on three axes.
All aligned to the same English anchor — every one is Anchored. New languages join comparable from day one.
143 concepts + 185 phrases. A living standard — it widens by version.
Languages with native audio underway — deepening Anchored → Filled → Voiced → Verified.
Not scrapable. Family-wide. Consent-gated. Compounding.
Evaluate it on your own infrastructure — data access only — then license it across the family.
Start your free evaluation →