List license for language models. about argos-translate HOT 1 CLOSED

Very good question. Any work done by the Argos Translate project towards training models should be considered under the same license as the code (either MIT or CC0). When training models for the package index we avoid some of the data on OPUS that is under non-commercial restrictions, some of the data used is under Creative Commons licenses but a lot of it is created by academics with citation information but no explicit license.

I think training a ML model on publicly available data should be fair use, but I'm not a lawyer. The strategy has been to just list all of the data used in the README with any available citiations/licenses for transparency and then release models with no restrictions.

from argos-translate.