Glossaries

Each entry maps a source term to its target, with optional structure:

• Variants — alternate spellings or inflections to treat as the same term (“cookies” alongside “cookie”).
• Do-not-translate — lock the source string into the output verbatim, for names, product IDs, code identifiers.
• Forbidden targets — translations that must never be used (a common wrong rendering, a deprecated name).
• Context hint — one line on how the term is used or why it’s pinned. It sits with the term’s other optional details: part of speech, gender, number, domain, and an example sentence pair.

Most words don’t belong in a glossary. Pin a term only when the translation is a real decision: non-obvious jargon, a brand-voice choice, a proper noun to lock verbatim, a word with several spellings to unify, or a wrong rendering to forbid.

Skip 1:1 cognates, standard abbreviations (e.g., etc., Inc.), filenames and URLs, and ordinary vocabulary — pinning those just adds noise. Not sure? Run an auto-build on a document you’ve already translated; it extracts candidates and you accept, reject, or edit them in bulk. See Build from a whole document.

The Glossaries page groups your library into tabs: everything, personal, one tab per team, and items shared with you. The search box matches names and content, so typing a term finds the glossary that holds it even when the name says nothing about it. Where your glossaries carry languages, a language filter narrows the list to one direction.

Each card shows its first term pairs and its counts, and a preview opens on hover with the full picture. The card menu (⋮) covers the rest: rename, duplicate (a detached copy you own, useful for a per-client fork), auto-build from a document, move to a team, share, and delete. Deleting asks you to type the glossary’s name, and there is no undo.

The glossary’s own page carries the same actions. Export and Share sit in its header, and a ⋮ button beside them holds duplicate, merge, move to a team, and delete; renaming is the pencil on the title. A glossary still waiting on an AI sync review shows a Review pending badge and a Review button at the top of that page, so the review is reachable without going back to the library.

Every term row has a find-uses button. It opens the Search tab in the editor sidebar (also Cmd/Ctrl+Shift+F) with the term and all its variant forms filled in, and lists every block where any form appears, with a count per form. Blocks where the source term is there but the translation is missing the target term are flagged, which makes it a free consistency spot-check: exact text matching, instant, no AI and no cost. The deep version of that check is the AI glossary audit below.

The same panel works as a plain word search: type anything, pick source, translation, or both sides, and click a result to jump to it. From the Glossaries page it opens in a dialog with a document picker. See The editor.

Set the source language and a primary target on the glossary, then add more target languages as you need them. The glossary page lays them out as a matrix: one row per term, one column per language, with a coverage count on each language so gaps are visible.

Each language holds a complete profile for the term rather than a single translation. Alongside the rendering itself, a language carries its own source aliases, other acceptable renderings, renderings to avoid, gender and number, an example pair, a note, and its own do-not-translate flag — so a term can be left untranslated in Polish and translated in Czech. The primary target stays in the term’s main field. The full term editor opens from the glossary page and from the settings icon on a term row in the editor sidebar; a language selector at the top switches which language’s profile you are editing.

A language you have added but not filled stays empty everywhere. Transept never shows or uses one language’s value in place of another’s, and a document whose language the glossary has no profile for reads only the source terms and their source-level details.

Fill with AI sits on each added language — the sparkle button on the language chip on the glossary page, and the header of the glossary drill-in in the editor sidebar. It writes complete profiles for the terms that have none in that language and leaves every existing entry untouched, so a second run after a partial fill costs only what is left.

The run reads one of your documents, which is what makes the suggestions fit the way you actually use the words. Started from the editor it uses the open document; started from the glossary page it asks which matching document to read. It takes each term’s occurrences from that document, their current translations, and any open comments on those passages.

The dialog quotes a ceiling, “up to N words”, because only the profiles that actually land are billed, at their source-term words. The document snippets and comments it reads as context are free.

Everything it writes arrives marked for review: usable in translations immediately, and flagged until a person signs off. Approve and Reject sit on the matrix cell and in the sidebar, the language’s header menu approves or rejects everything pending at once, and editing a profile by hand approves it. Rejecting deletes the AI entry and leaves the cell blank. While anything is pending, the glossary’s library card shows an N AI to review badge. Only the languages you added can be filled this way; the primary target stays hand-curated.

Open the ⋮ menu on the library card, the same menu on the glossary’s own page, or the menu in the editor sidebar, and pick Merge…. Choose the glossary to bring in. It is only read, and nothing about it changes. Merging is free: no words are spent either way.

Picking one writes nothing. It opens a review of what the merge would do: how many entries it adds, how many blanks it fills where a language had no translation, how many it skips because both sides already agree, and how many hold a different translation on each side. Every one of those is a row you decide: drop an arriving entry, drop a fill, and answer each disagreement on its own row with the two values in front of you. Keep all and Take all settle the disagreements in one go when you would rather not read them individually. Long lists show the first 30 with the true total beside them, and the rest follow whichever of those two choices is in force. Nothing is written until you press Merge.

You do not have to assign languages first. If either glossary never had its source and target language set, Transept reads them off its own entries and saves them before the merge runs, so the preview and the write agree on where each value goes. A brand-new empty glossary takes the languages of whatever you merge into it, which is what makes “create one, merge several into it” work. Two glossaries that both stay unlabelled merge the way they always did, one target column into the other.

The merge stops in two cases. One glossary knows its languages and the other’s entries could not be read: set them by hand on that glossary’s page and merge again. Or the two read from different source languages, which nothing can reconcile, because their source terms are not in the same language to begin with.

While the merge screen is open, Literess can tell you what it currently holds and change those settings for you. Pressing Merge is still your click. See What Literess can do.