Feature Request: Add Sentence-by-Sentence Translation Mode

[petertzy]

Description

Currently, the translation interface provides a full-text translation mode. I would like to introduce a translation mode selector for the full-text translation section, with the following two options:

1. Full-text Translation

   • Keep the existing translation behavior unchanged.
   • Translate the entire text as a complete passage.

2. Sentence-by-Sentence Translation

   • Add a new translation mode that displays each original sentence together with its corresponding translated sentence.
   • Ideally, the original sentence and its translation should be visually paired and easy to compare.

This sentence-by-sentence comparison mode would be particularly useful for language learning, as users could easily compare the original text with the translation sentence by sentence and better understand vocabulary, sentence structure, and expressions.

Expected Behavior

• Add a mode selector to the full-text translation section.
• Keep the existing full-text translation as one option.
• Add a new sentence-by-sentence comparison mode.
• Switching between the two modes should be simple and intuitive.
• The sentence-by-sentence mode should clearly associate each source sentence with its corresponding translation.

Please try to implement this feature while preserving the existing full-text translation functionality.

Thank you!

[Gambit-Checkmate]

Hi, I'd like to work on this. I checked the current AIPanel flow and the FastAPI /translate boundary. I think the cleanest approach is to keep the existing selection/document scope as-is and add a separate display mode for full versus sentence.

For sentence mode, I'd have the backend ask for and normalize structured source/translation pairs, instead of trying to split an already translated paragraph in the frontend. The existing full-mode response can stay backward compatible, while sentence mode adds optional pairs that the panel renders together under a small mode selector. I'd also cover the request/response normalization and make sure the existing full translation path remains unchanged. Does that response shape sound right before I start?

[pankajsrivatsankbs-jpg]

Hi @petertzy, I'd like to work on this — planning a new /translate/sentences endpoint that splits the source markdown into sentences and returns original↔translated pairs, keeping the existing /translate endpoint untouched.

[petertzy]

Thank you for your participation. I’ll assign the task to you now. Please feel free to give it a try, and be sure to consider the user experience of the software while working on it. Thank you!

[petertzy]

I just noticed that two different people have expressed interest in completing this task. Based on the order of the requests, I’ll assign the task to the person who asked first. Thank you both for your enthusiasm and interest!