Glossary vs Translation Memory: What Each Fixes

Quick answer — A translation memory reuses whole approved segments. A glossary enforces individual terms wherever they appear. Memory saves money on repetition; a glossary prevents inconsistency, and neither substitutes for the other.

Vitra.ai Universe keeps one memory and one quality gate across every format.

Different units, different jobs

A memory works on segments — usually sentences. It answers whether this exact thing has been translated before. A glossary works on terms — usually a word or a short phrase. It answers what this is called, always, everywhere.

The unit is the whole distinction, and most confusion between the two comes from ignoring it.

Where each one wins

The last row is the one that decides the argument. A sentence never seen before gets no help from memory — but the terms inside it are still governed, so the output is new prose using approved vocabulary.

That is the case glossaries exist for, and it is the majority of content in any growing organization.

Precedence matters

A pinned glossary term outranks a memory match.

Without that rule, a fuzzy match containing an outdated product name reintroduces the old name every time it is reused, and the glossary quietly loses to its own memory. Pinning means the term resolves the same way regardless of what the surrounding segment did.

Both are applied at translation time rather than checked afterwards, which is what makes the first output correct instead of the third.

Run both, size them differently

Memory should be large and grow without supervision — every approved translation and reviewer correction writes back, and more coverage is straightforwardly better.

A glossary should be small and curated, because every entry is a constraint and a wrong one propagates everywhere.

Large memory, tight glossary. The opposite arrangement — a thin memory and a sprawling term list nobody maintains — is the configuration that produces both high cost and poor consistency, and quality control findings are usually the first place it shows up. Industrial products make the strongest case for building the glossary first — manufacturing glossary.

FAQ

What is the difference between a glossary and a translation memory? The unit. A memory reuses whole approved segments and answers whether something has been translated before. A glossary enforces individual terms wherever they appear, including in sentences never seen before.

Which one saves money and which prevents errors? Memory saves money by reusing repeated segments. A glossary prevents inconsistency by governing terms in new content, where memory offers no match at all. Neither substitutes for the other.

Should a glossary term override a memory match? Yes. Without that precedence, a fuzzy match containing an outdated product name reintroduces it every time it is reused, and the glossary quietly loses to the memory it was meant to govern.

How large should each be? Memory should be large and grow automatically, since more coverage is simply better. A glossary should stay small and curated, because every entry is a constraint and a wrong one propagates everywhere.