argos-translate
argos-translate copied to clipboard
Consider dropping advertisment of experimental translations
Consider no longer advertisting translation models that aren't functional or useful.
For example, English to Chinese (vice versa) is completely broken, failing even to translate "goodnight" properly.
While it is nice to say a project supports X feature, if it doesn't actually work what is the point of showing it off?
This can be achieved in two ways.
- Separating broken models to "experimental" category with warnings discouraging production use, and encouraging development use.
- Only including broken models in development documents (inspired by @pierotofy) for future improvement.
While English <=> Chinese definitely scores very poorly on lots of translations, it's not "completely broken". The point is that we have a baseline from which to start improving.
Perhaps you could help train a better model.
While English <=> Chinese definitely scores very poorly on lots of translations, it's not "completely broken". The point is that we have a baseline from which to start improving.
Perhaps you could help train a better model.
Thank you for your response.
Perhaps I was a bit too rash with my original statement. With the information you provided in mind. I'll change the wording of this issue to "stop advertising" broken models.
New Chinese simplified and Chinese traditional models. https://libretranslate.com/?source=auto&target=zt&q=goodnight
Chinese simplified still struggles with "goodnight", but I think traditional is spot-on.
New Chinese simplified and Chinese traditional models. https://libretranslate.com/?source=auto&target=zt&q=goodnight
Chinese simplified still struggles with "goodnight", but I think traditional is spot-on.
This is great! I tried the en_zh package last week but didn't seem to work on the python interface.
How might one try to install en_zh locally? (as mentioned tried to install this with from_code en and to_code zh but "hello world" mapped to "hello world" still in english)
I think the argos-index was refreshed a few days ago, try to update the models now?