06 Advanced Translation & Glossaries

The default output is a plain, conversational translation. To make translations follow your brand's wording and your industry's terminology, use the two tools for the job: Enhanced Translation (style / domain / custom instructions) and glossaries (which pin down how proper names are translated).

1. What you can set once it is on

Tick Enhanced Translation in the new-project form and these groups appear:

• Translation Style (7 options): Standard, Formal, Casual, Colloquial, Literary, Technical, Child Friendly.
• Industry Domain (10 options): General, Medical, Legal, Finance, Technology, Gaming, Marketing, Academic, Culinary, Travel.
• Custom Instructions: a free-text box capped at 500 characters. Whatever you write here goes to the translation model verbatim — for example "always use the formal 'you'", "no exclamation marks at the end of sentences", "leave company names untranslated".
• Use Glossary: click Select Glossaries to attach several glossaries at once; the chosen ones are listed as tags, click × to detach one, and the "Glossary Preview" beside them lets you look over the entries first.

If "Enhanced Translation" is left unticked, everything above collapses and not a single glossary applies (even ones you had selected before).

2. Instruction presets: keep the requests you reuse

The Advanced Translation page in the sidebar has two tabs: Glossary Management and Custom Instructions.

In "Custom Instructions" you can save an instruction as a preset: fill in Preset Name + Instruction Text, and it saves automatically when you finish typing. Afterwards, pick it directly in the new-project form, or edit it inside a project and click Save as new preset. When creating a preset there is also a Create from choice: Blank to write from scratch, or Based on existing to take a preset you already have as the starting point and revise it. Presets are not split by language — any project can pick any of them; the most recently used ones sort to the top, so the list order changes over time. That is by design, not a glitch.

3. Glossaries: how to build one

In the "Glossary Management" tab, the left side is the list and the right side is the content of the selected glossary. Click New Glossary (the + button) and fill in three things: Glossary Name, Source Language, Target Language (default English → Chinese; change it to your own direction — 20 languages are available).

Once created, the right side is a table with only two columns: Source Term / Target Term. The last row has a +; click it to add a row. You never press a save button — edits are written to disk about 0.8 seconds after you stop. The × at the left of a row has to be clicked twice (the first click turns it into "Confirm") before the row is really deleted.

When the list grows and editing row by row is too slow, Export CSV, edit the file, then Import CSV. The importer reads column 1 = Source Term and column 2 = Target Term; a first row that is a header is skipped automatically (headers in English or Chinese are both recognized).

Two things you must know, otherwise it will feel like "the import lost my data":

• Case sensitivity and whole-word matching are not offered in the interface. The program always treats terms as case-insensitive, whole-word matches; even if your CSV carries those two columns, they are ignored on import.
• Rows missing either column are dropped silently, with no error. After an import, count the entries in the preview to be sure.

4. How a glossary actually works (and why it sometimes seems not to)

A glossary does not just stuff a word list into the prompt and let the model freestyle. It works in two steps: the term list is sent along while translating, and after the translation is done the program checks every line once more — if the target term you specified does not appear in the translation, that term is attached to the line as an annotation.

That gives three real boundaries:

• Only short terms count. When the specified translation is longer than 3 words (phrase-level), that entry is not attached. Fixed phrase-level wording belongs in "Custom Instructions" instead.
• The language pair must match. A glossary is built for one "Source Language → Target Language" pair; when it does not match the project, the whole table is skipped — translation continues normally and nothing is reported. If terms seem ignored, check this first.
• It attaches as an annotation. It adds a "this should be translated as X" constraint to the translated line; it does not force-replace every matching word in the original sentence.

A glossary has no limit on the number of entries and no limit on how many glossaries you keep, but deleting one asks for confirmation, and once it is deleted any project that referenced it can no longer load it on its next translation.