Continuing the discussion from Be inspired by features of other online translation services and discuss reverse engeneering them:
You wrote Wiktionary data is used to train the models but what about the corrections users can hand in? How are they used? What about copyright? The feature says nothing about licensing. What if the text in the original language was copied from a website (if it was not from somewhere else, why would someone want to translate it)?
These are good questions. The feature for collecting user data was just implemented and currently just saves the data to a database local to that LibreTranslate instance. We’ll need to figure out a system for collecting it, licensing, and hopefully publishing for others to use.
This is exactly the most interesting part. Is there any news there?
We made a Git repo to store submitted data (discussion) where we store data submitted to the API. We can also accept manual data submissions from pull requests to this repo.
Currently none of the models are trained on LibreTranslate data but after we’ve gathered enough data we can start using it.