One of the translation options, LibreTranslate, can be configured to store the text stay local and doing no interaction to external server. My company does not want to send any data to external servers for auto translations, which might mean, for instance, sending IP to Google while using Google Translate. For forums hosted by Discourse, does anyone know if they are using a local instance of LibreTranslate?
To clarify, LibreTranslate is an application that may be installed by end users; the endpoints for translation may be exposed so you can use it via a network (such as on a self-hosted server), or it may be used as a local app where the endpoints are not available over a network, only as software running on the same machine (useful for translating a lot of documents or where network coverage is unreliable).
I’ve used LibreTranslate with Discourse Translator, with both the API hosted by the project and by self-hosting on a VPS. However, I did not try installing LibreTranslate on the same VPS as Discourse, nor in the same Docker container as Discourse.
The plugin does not forward the IP address to the services it uses; the translation services would be receiving data from the Discourse server.
The greater point is that you will need to read the terms of service for all the translation services, and decide if they are vendors your company can trust for your use case. And because it is open source, your company can check what the plugin is sending to the services: GitHub - discourse/discourse-translator
Also, LibreTranslate, as an API wrapper for Argos Translate, is fairly small. Language models are huge. We’re talking 10 gigs at a minimum. I would not want to store those on the same machine as Discourse (or even self-host, I just use the paid service ).