← Blog
4 min read

A Glossary for Better Machine Translation: How to Build One in 10 Minutes

A glossary doesn't need to be exhaustive to be useful — ten to twenty terms that are specific to your company or industry, listed with their intended translation, is enough to fix the errors that actually recur in practice.

What belongs in it

Start with proper nouns: your company's name, your product names, and the names of any clients, partners, or team members likely to come up repeatedly. These are exactly the words a general-purpose translation model has no way to get right on its own, because there's no "correct" translation to infer — only the one your organization has already chosen.

After that, add acronyms and internal shorthand specific to your industry or team — the kind of term that has an obvious meaning to everyone in the room but no standard translation a model would already know. A handful of these, listed once, prevents the same term from being mistranslated differently every time it comes up.

What doesn't need to be in it

Ordinary vocabulary doesn't belong in a glossary — general-purpose translation already handles common words and phrases well, and a bloated glossary is harder to keep accurate than a short, focused one. The test for whether a term belongs is simple: would a translation model, given no other context, plausibly get this wrong or render it inconsistently? If not, it doesn't need an entry.

Keeping it current

A glossary is only useful if it's current — a five-minute review before a recurring meeting (a new hire's name, a renamed product) is worth far more than a large glossary built once and never revisited. Treat it the same way you'd treat a shared meeting doc: a living list, not a one-time setup step.

Be first to try real-time interpretation yourself.

In active testing — join the waitlist and we'll email you when it's ready.