JanEbbing
commented
1 year ago Hey, I transferred your feature request to the correct project and made the title a bit clearer. I'll discuss internally what we need to make this happen. (Also FYI, these repositories are for the DeepL API, which is a way to integrate DeepL translations into other software. Your request seemed more for the https://deepl.com product, but it also makes sense for the API).
Hi,
sorry for posting this here, but i found no better way to request a feature/give feedback without paying for deepl pro (which is not the way it should be imho), please forward this issue to the appropriate platform, thanks.
when i type some japanese text written with latin characters like
Ossan-tachi to tsuri
i get with deepl a result like this
but google is
if i use the transited text (from google or even a crude system like https://www.translitteration.com i get a result that is just fine:
lookes like it checks if it is japanese text and then translites it before it goes through the translation with some artificial inteligence behind..
that is the one thing that google translate dose better then your service, and i think it is not difficult to catch up to it, there is open source software like https://pypi.org/project/pykakasi/ that allows you to translit forth and back japanese text using the hepburn system. So you can just try that, try to translate it, score it and thereby determine (if you can use this term with neuronal processing systems) if it makes sense and then suggest it to the user like google dose. I guess a working prototype could be developed within a day assuming one has access to your source code and development systems...
So please look into it to make our live easier.
thanks a lot in advance and keep up the good work