Language assets
Glossary
Store approved term pairs for a language pair and apply them during text translation, document translation, and review.
A glossary stores source terms and their translations for a language pair. A source term can have several alternatives. Relevant, unambiguous choices from your selected glossary can be supplied with a translation request; review the resulting wording in context. Available from the Starter plan up.
Glossaries in your workspace
Step by step with screenshots: Glossaries.
- You can keep several glossaries: per client, per product, per language pair.
- Terms are added by hand or from CSV or XLSX files.
- A glossary is switched on in the dashboard translator or in the editors; the switch shows how many terms are loaded.
Approved, alternative and forbidden wording
Each term has a usage label, a definition, an example and notes. Approved is the wording you want, and the AI receives it. Alternative keeps another valid translation: the AI never receives it, but both editors show it in the segment resources panel, where a reviewer can preview it or save it as the choice for that segment; a saved choice is used when the segment is retranslated. Forbidden marks wording to avoid: the AI is told not to use it. If a source term has two different approved translations, automatic glossary selection leaves it unresolved, so make one of them an Alternative. When you approve an extracted term whose source already has an approved translation in the glossary, the new translation is added as an Alternative. The panel also lists similar glossary entries: the same words in another form or order, or most of the same words. The selected segment resource panel flags forbidden wording found in the saved translation and missing preferred wording. These are literal matching checks, so review inflection and meaning yourself.
Terminology review is available in both editors once a Word or editable PDF translation finishes; pending edits are saved first. It takes a source term, the current target wording and a replacement, and previews the change. Only source segments containing that term are considered. You pick which eligible segments change. Approved, confirmed and protected segments are excluded, and matches crossing formatting runs are left for manual editing. Applying saves drafts without consuming AI words. If the document changed since your preview, preview again. Project history can restore the old wording; Export rebuilds the download.
CSV import and export preserve status, definition and usage_example columns alongside source, target and notes. Importing the same bilingual pair updates it; a different target remains a separate alternative. Definitions and examples are reference information for reviewers.
How glossary terms are applied
The active term pairs are included with the translation request. DeepReference does not rewrite the translation afterwards: QA checks report every segment where a required target term is missing, with optional case sensitivity in CAT settings.
Use case sensitivity and morphology matching in CAT settings to control which source terms match. Review target wording in context and use glossary QA to identify missing terms. Separate Strict and Preferred modes are not offered.
Glossary and translation memory together. The priority order in CAT settings decides which wins. With the glossary first (the default), an exact memory match that leaves out a glossary term found in the source is not reused: in text and AI document translation that segment is translated again with the term, and in a document you translate in the editor it stays empty with the memory match shown as a suggestion. A memory match that already uses the approved term is kept.
For names and codes that must stay unchanged (brand names, product codes, part numbers), add them to your glossary with the same wording on both sides, for example ACME-X1 → ACME-X1, and check them in review. Automatic protection of URLs, email addresses and codes is not active at present.
Review repeated glossary entries
In the glossary list, choose Review duplicates beside a glossary. Review each group with the same source and language pair. Keeping an entry removes only matching copies of its wording, status and explanatory metadata. Different translations, notes or definitions remain alternatives. If an entry changed after loading, reload the review before removing copies. Groups larger than 100 entries need review in smaller groups in the main library.
What belongs in a glossary
Use a glossary for terms where your organization has already chosen the target wording: product names, defined legal terms, interface labels, or recurring technical phrases. For example, both *pannello di controllo* and *cruscotto* can translate "dashboard" in Italian; the glossary records which one this project requires.
Don't have a glossary yet? Terminology extraction can build one from documents you have already translated.
Keep exploring
Terminology extraction and termbase
Find bilingual term candidates, review them, and link approved terms to reusable glossary entries.
Quality assurance checks
Check target segments for missing placeholders, formatting-tag differences, unusual length, inconsistent translations, and glossary violations.
Smart suggestions
Complete words and phrases from the active glossary, approved termbase entries, and Translation Memory while editing a target segment.
CAT settings reference
Translation defaults, translation memory, terminology, quality checks, and presets.