The FOSS app Argos Translate enables people to locally translate their documents without depending on an external service and then hoping their content is not snooped on (while simultaneously hoping to get translation service for free). Argos does okay with quite popular language pairs but it’s really not up to a good standard of quality overall.

The machine learning input into Argos known as “models” are trained on samples of (hopefully manual) translations. The models require huge amounts of data. Apparently the effort to gather large volumes of input leads to grabbing poor quality samples, which ultimately leads to bad translations. To worsen matters, you have a sparse scatter of different projects making their own models. So the effort is decentralised in a detrimental way. End users are then left with having to experiment with different models.

Shouldn’t Académie Française (the French language protection org) have some interest in the public having access to resources that give high-quality translations into French?

Consider that Académie Française members each spend €230k on clothes (yes, that “k” after the number is correct), surely they have money sloshing around to promote French. If playing dress-up is worth €9.2 million (€230k × 40 members), just imagine how much money they must have for their mission of supporting the French language.