Hi B. Rieger , I'm just trying to reproduce this issue, but not successful; might be that I'm missing something.

Can you confirm that the steps I interpreted from your description are correct here?

1. You merge two individual segments in the editor.
2. Move focus to the merged segment.
3. The provider returns a translation for the individual segments in the translation results view as opposed to a single merged segment.

Can you provide a project package so that I can reproduce the issue you are reporting here?  if yes, please send to phartnett@rws.com with a link where I can download the project package.

Hi B. Rieger , I'm just trying to reproduce this issue, but not successful; might be that I'm missing something.

Can you confirm that the steps I interpreted from your description are correct here?

1. You merge two individual segments in the editor.
2. Move focus to the merged segment.
3. The provider returns a translation for the individual segments in the translation results view as opposed to a single merged segment.

Can you provide a project package so that I can reproduce the issue you are reporting here?  if yes, please send to phartnett@rws.com with a link where I can download the project package.

Hi Patrick Andrew Hartnett ,

Thank you for your response!

Actually, we didn't merge the segments. But in the initial translation suggested by the provider, only the first part is shown in the target segment. The rest can be found in other segments in the "Translation results" window.

I have sent a project package to your e-mail address. Thank you in advance for your attention!

Hi B. Rieger , Thank you for forwarding the package; very helpful.  I confirm this is a bug in the plugin.

I'll add the bug to our backlog this morning + circle back once we have a fix in place.

Hi Patrick Andrew Hartnett , Thank you! We are looking forward to the solution!

Hi B. Rieger , After reviewing the code, I've confirmed that multiple translations are appearing primarily because the source content isn’t fully segmented. As a result, the AI model interprets each portion as a separate transUnit, which leads to multiple responses. It’s interesting to note that the AI handles this segmentation on its own without any additional prompts.

We’re currently refining our prompt engineering approach for all AI-driven features in production to handle scenarios like this more gracefully. However, implementing these changes requires thorough testing to ensure stability across various content types and use cases. As such, I can’t offer a quick fix right now, but please be assured that we’re actively working on a robust solution for the future.

Hi Patrick Andrew Hartnett ,

Thank you anyway! Do you perhaps have any suggestions for workarounds to optimise the segmentations as a temporary solution?

Best,