Are terms from the glossary automatically taken into account by the LLM?

[marcoverheiden]

Hi,

I was wondering whether terms in the glossary are automatically used by the selected LLM during translation (both segment-by-segment and batch translation)? I did notice the AI Assistant where it can automatically generate new terms to use during translation, but it would be great if it would also add the terms from the active glossary by default (comparable to the Studio AI Assistant plugin 'terminology aware' LLM translation').

[michaelbeijer]

Hi Marco,

I'm now sure about this. I will investigate!

I actually created the AI Assistant a while back, but got side-tracked and haven't worked on it in a bit. My workflow currently is as follows:

• I import my file(s) into memoQ
• export a memoQ bilingual file
• import this into Supervertaler
• ask chatgpt.com to draft a prompt (based on a previous one that worked well)
• translate the Supervertaler project with this prompt
• export a translated memoQ bilingual file
• reimport into memoQ

However, I find myself increasingly doing the translation straight in the Supervertaler grid. And whenever I have time, I also import the source documents straight into Supervertaler via Import monolingual, translating them and exporting them. However, Supervertaler still needs quite a bit of work when working on monolingual documents. However, the memoQ bilingual file route is pretty bomb-proof, especially when you combine it with prompts generated elsewhere. I am also working on improving the Supervertaler AI Assistant so that eventually it will produce prompts as good as chatgpt.com.

[marcoverheiden]

Hi Michael,

Thanks for looking into this.

Your workflow makes a lot of sense, and it explains the current focus on prompt quality rather than deep terminology integration. That said, from a CAT-user perspective, automatic injection of the active glossary into the LLM context would be a big win, especially if that can be done in batch translation mode. This would auto-feed the right set of terms into the LLM during the translation process, which eliminates 1) having to add glossaries manually to the LLM prompt and/or 2) to post-edit terminology.

Looking forward to what you find. Happy to test or give feedback once you experiment with this.

I also saw the new release regarding the API key fix, will give it a spin later today.

Best,
Marco

[michaelbeijer]

Incidentally, try clicking the button called "Preview Combined". This will show you exactly what (prompt) is being sent to the AI.

Having the program automatically inject a specific user glossary into the prompt that is sent to the AI (both during batch mode and single segment translation mode ("Ctrl+T")) should be pretty straightforward. I'll try to implement this as soon as I have a moment.

Michael

[marcoverheiden]

Hi Michael - coming back to this thread, I was wondering if there have been any updates or news regarding the terminology injection into the LLM in Supervertaler? I just pipped my way up to release 1.9.174 and trying to get up to speed with all the latest updates!
Which Claude subscription and model are you using, Opus 4.5 on Max 20x? Or mostly Sonnet 4.5?

[michaelbeijer]

Hi Marco, I am implementing glossary injection as we speak and will get back to you here when it's done!

[michaelbeijer]

Done!

available in: v1.9.175 - January 28, 2026

[marcoverheiden]

Amazing, thanks! It works, that's a tremendous help during post-editing.
Strangely, I can't see any glossary matches being displayed in the Termview pane next to the grid, while I can clearly see that the AI is applying the right term from the connected glossary. Any thoughts?
Also, is the glossary in Supervertaler sensitive to the language direction (Studio/MemoQ isn't, so just want to make sure). i.e. a will a TB nl to en also work for en to nl projects?

[michaelbeijer]

That's great to hear! Let me know if you experience any issues with it, as it is a brand new feature!

I will look into this. I have a test glossary activated for AI, and I am seeing matches in the Termview pane to the right of the grid.

Also in the other Termview pane, which can be shown/hidden via: View > Termview under Grid.

An nl>en glossary should give matches in an en>nl project, and vice versa. I will test it ASAP.

Incidentally, have you tried clicking the "Preview Prompts" button below the grid? This will also show you any activated Glossary being sent to the AI.

[michaelbeijer]

It seems it has not yet been implemented. I am now adding bidirectional termbase matching. Stay tuned!

[michaelbeijer]

Done! Bidirectional termbase matching is now available in v1.9.177-beta.