Glossary, dictionaries and memory

Three things make a second translation better than the first, and all three live on your device.

Your glossary

A glossary entry is a source term and the target term you want, for a language pair. It is enforced: when the source term appears, the translation uses your term, and when the engine's sentence cannot be fitted around it the note under the sentence says so rather than silently ignoring you. Add entries from Your words, from the term panel beside a result, or by importing a CSV file.

Entries are stored in your browser or app. With Pro, sync sends your glossary and history (and, from the browser, your translation memory) — encrypted on your device — to our server so another signed-in device can fetch them.

Topic dictionaries

Pick a topic (legal, medical, technical, and so on) and Traxlate loads a dictionary of terms for it from traxlate.com. A dictionary suggests: matching terms appear as chips beside the result, and one click promotes a suggestion into your glossary, where it becomes enforced. Dictionaries never rewrite a sentence on their own. We measured the alternative — forcing dictionary terms into every sentence — and it made translations worse, so suggestion is the rule.

Translation memory

The memory holds your corrections, not the machine's output. When you edit a sentence in the Desk or in a result, the source and your corrected translation are remembered for that language pair; the next time the same sentence comes back — the boilerplate paragraph in every contract, the safety line on every label — your correction is used instead of a fresh draft. Machine output is never cached, so an uncorrected sentence is always translated afresh with the current checks.

The memory is on your device. It travels with sync only if you turn sync on, and it is shared with other people only if you share it on purpose — see below.

Taking your memory with you

Your words has Translation memory with Export CSV and Import CSV, and beside them Export TMX and Import TMX — TMX is the format the professional translation tools read and write, so a memory built elsewhere can come in, and yours can go out. A TMX file names one language pair, so you pick the pair before exporting. An import never overwrites a correction you already made: it adds what you do not have, counts what already matched, and tells you how many of your own sentences it left alone. Anything in the file it could not read — an unknown language code, a segment with no text — is counted and reported rather than guessed at.

Analysis

Also under Translation memory: pick a document you have reviewed in the Desk and Analyse counts its words against your memory the way an agency quote is built — repetitions, exact matches, and the fuzzy bands between — with a CSV to download. For a language that does not put spaces between words, the report says the word count is not reliable instead of printing one.

Sharing with a team

A glossary or a translation memory can be shared by link from Your words. Whoever opens the link gets the entries into their own glossary or memory, on their own device, and can pull later updates. Everything shared is encrypted on your device first; the key is inside the link, so we store what we cannot read. Nothing is published until you press Share. Working with a team has the detail.

Does the glossary work in documents and PDFs too?

Yes. The same enforcement applies wherever a sentence is translated — Translate, Documents, PDF, Pictures, the Camera.

Can I export my glossary?

Yes. Export CSV in Your words writes a file with the term, the translation, the language pair and whether the match is case-sensitive; Import CSV reads the same format back.