From yesterday in #wikimedia-tech:

16:52:13 <legoktm> related to the discussion about automatic translations in Discourse vs MediaWiki and privacy implications of using Google Translate (or other external providers) - https://blog.mozilla.org/en/mozilla/local-translation-add-on-project-bergamot/ came out today, fully local translations
16:52:39 <AntiComposite> that's cool
16:57:16 <legoktm> it's mostly webassembly, would be interesting if it could be wired into a gadget that lets you translate individual talk page comments or sections

I have been testing the SectionTranslation tool (is only available for some languages) and is doing a great job translating sections and sentences, and even providing a way to store them, treat them and "officialy translate" the correcting the MT. So this seems a good path towards this feature.

Google Translate supports 109 languages. Project Bergamot apparently supports 14 (4 of which are still in beta), almost all first-world languages. (The quality seems pretty impressive though - worse than Google, but not by much.) So it doesn't seem very useful right now.

In any case privacy is a trivial matter of proxying through Wikimedia (or Wikimedia-affiliated) servers, it's the least challenging aspect of translating MediaWiki comments.

(Here's a matrix of translation API language support I just put together, FWIW.)

I can't edit the doc, but we are using Elia at Wikimedia for Basque, Catalan, Galician, Spanish, French and English languages.

Assuming that comments can be in any language, we will also need to identify the language of the comment first.

While this seems like a good thing to integrate into MediaWiki, it is also something that is supported on the browser side. I regularly use Google Chrome to translate Spanish documents to English through Google Translate. If we want to do this, how could we do it *better* than Google Translate does?

In T309920#8179397, @Mike_Peel wrote:

While this seems like a good thing to integrate into MediaWiki, it is also something that is supported on the browser side.

That's an extremely fair point. For any new feature which ostensibly contributes to already substantial tech debt, I think it is wise to ask if there already exists an easy-to-use, free-ish solution (this would qualify for consideration IMO) instead of attempting to reinvent the wheel with the Foundation's and Community's limited resources.