When using the built-in translation functionality I get horribly bad results for my use case (translate to German). So bad in fact that some texts become nearly unreadable gibberish. It feels like how machine translations were 20 years ago.

I know that Firefox’s translation system works with offline data only and locally on my machine, and thus is very, very, very limited compared to specialized systems like DeepL – and even the common LLMs.

But I wonder: Is there a way to improve the system? Can I feed it more data to make it better? Can I do more than just download the translations in the settings?

I know I can always use extensions for sending pages or text to the common online translation services, but since Firefox has that built-in I’d love to use it – but it needs to return better results.

Any ideas are welcome!

Thank you and vielen Dank :)

  • solrize@lemmy.ml
    2·
    2 天前

    I expect there’s a project you can join, but otherwise just hope that the models improve over time. It sounds like the English to German one is worse than the German to English. The latter isn’t great but it helps. So it’s a matter of more data and more training.

    • 𝘋𝘪𝘳𝘬@lemmy.mlOP
      2·
      2 天前

      For English to German I don’t need machine translations, but yes: English to German translations are even more catastrophic. I sometimes translate from Polish. But I also tried some other languages … None of them produced results I’d consider good by modern standards.

      Translating something to English is slightly better, but if THIS is the best we can have right now, then the best we can have right now isn’t even close to what we COULD have right now (with the right training data and resources).

      I expect there’s a project you can join

      I did a quick research that took longer than I wanted it to take, and the Mozilla Corporation is really good in hiding anything even remotely related to how the translations work and what the community can do.