Terminology database (termbase)
A ruled list of your product's terms: this one gets the approved wording, that one is never translated at all.
See it
What it is
A termbase is a small, ruled dictionary that sits beside your translations. Each entry names a source term, says what it means, marks its part of speech, and rules on how it should be rendered per locale. That ruling is richer than one word: a term can carry a preferred translation, admitted alternatives, prohibited renderings, and inflected forms, which is what heavily inflected languages need. Some entries are flagged do not translate instead, but DNT is a policy call that follows your brand and legal guidance, not something a term inherits automatically for being a product name or a plan tier. Recurring UI verbs (Save, Publish, Share, Workspace) go in with an approved target so they do not drift.
You want one the moment more than one person has ever translated your product, because five translators over three years will render 'Workspace' five different ways and your German UI will read like five different products. Every TMS ships a glossary panel that surfaces matches inline while translating and fails a QA check when a translation ignores one. The interchange format is TBX, and the old desktop anchor is SDL MultiTerm.
Gotcha: teams over-lock. Marking every noun DNT leaves your French copy stuffed with English and reading like a machine wrote it. Lock what is trademarked, what appears in contracts, and what users type into a support ticket. Let the rest be translated. Second gotcha: an entry without part of speech and a definition is barely useful, since 'share' the verb and 'share' the noun need different rulings.
Ask AI for it
Build a termbase for this product as a CSV with columns: term, part of speech, definition, do-not-translate flag, preferred translation per target locale, admitted alternatives, prohibited renderings, and a usage note. Scan the UI copy and pull out every product name, plan tier, feature name, and recurring interface verb as candidate terms. Fill in approved translations and DNT status only from sources that already exist: brand guidelines in this repo, shipped translation catalogs, or locale decisions I give you. Where no such source exists, leave the cell blank and mark the row 'needs-linguist-review' rather than inventing terminology. Keep entries singular and split homographs into separate rows by part of speech.