Français
crise convulsive
back-translation — “convulsive crisis”
Turn specialist English terms into a multilingual glossary — with every risky term routed to a human for sign-off. Machine speed, human authority.
The skill never presents a machine translation as final. It does the legwork, then hands you a short list of real decisions — and waits for your sign-off.
Give Claude your terms — or a document to pull them from — plus the domain, 3–5 target languages with their region, and who the translation is for.
Claude translates each term, adds a literal back-translation as an accuracy check, rates its own confidence, and flags anything risky.
A focused worklist surfaces only the cells that need a human — false friends, no-equivalents, regional splits — each with one specific question.
You approve, revise, or reject. Nothing is marked final until a named reviewer signs the workbook for accuracy, appropriateness, and acceptability.
You don’t get a wall of text — you get a termbase your team can filter, review, and keep as a living record.
| English term | French | Spanish | Confidence | Flags | Status |
|---|---|---|---|---|---|
| nature-based solutions | solutions fondées sur la nature | soluciones basadas en la naturaleza | High | — | approved |
| seizure | crise convulsive | convulsión | Medium | register-risk | needs review |
| intoxication | intoxication / ébriété | intoxicación / ebriedad | Low | false-friend | needs review |
| tort | responsabilité délictuelle | responsabilidad delictual | Medium | competing-equivalents | needs review |
| tolerance stack-up | accumulation de tolérances | acumulación de tolerancias | Low | no-equivalent | needs review |
Catches false friends and taboo terms before they reach a patient leaflet or consent form.
Flags concepts with no clean equivalent instead of forcing a misleading near-match.
Pins regional variety and competing standards so suppliers read the same thing you wrote.
Keeps a consistent, signed-off termbase across markets and reporting frameworks.
Download the file below, open it in Claude.ai, and click Save skill on the file card. (You’ll see that button if skill creation is enabled for your account.) After that, Claude reaches for it automatically.
Describe the job in plain language. Name your languages and their region — it’s the single best thing you can do for quality.
Your terms stay in your own Claude.ai workspace. The skill adds structure and review — it doesn’t send your glossary anywhere.
No. It’s decision-support that gets a qualified reviewer to a confident answer faster. Claude drafts and flags; a human — ideally a native speaker with domain knowledge — makes the call and signs off. For safety-critical work, that human step is essential, not optional.
Inside your own Claude.ai account. The skill is a set of instructions Claude follows; it doesn’t transmit your glossary to any third party or store it elsewhere.
Any language Claude works in — 3–5 at a time is the sweet spot. Crucially, it rates its confidence per language and marks it lower where Claude is less reliable, so weak spots are visible rather than hidden.
An Excel (.xlsx) workbook with three sheets: a Cover with sources and a sign-off block, a colour-coded Glossary with a drop-down for each review status, and a Review Queue listing only the cells that need a human decision.
Never. Every cell starts unreviewed, and the workbook stays “draft” until a named reviewer checks the sign-off box. That’s the whole point of the skill.