I think it is important to make a policy: all comments reviews must be available on-site at least in English, although (perhaps?) it is acceptable to place a foreign language original side by side with an accurate and complete translation of this text into at least readable English.
The reason is that allowing untranslated discussions makes it possible to run into a situation where a group of people post and vote in cliques which are closed to review by outsiders.
But I don't think it is ok to similarly restrict the language of submissions, because there are excellent papers that have never been translated for one reason or another. But we can require a locally hosted complete and correct translation for any non-English submission to reviews, in the description or perhaps a linked pdf, better onsite than off (so it doesn't disappear later). A cheap trick to get attention is to host competent English translations of technical work that has not been translated. This should usually happily be provided by the author as part of the submission process.
But there is another consideration, which is forcible translation of important work against the author's will. Perversely enough, this is necessary in at least one unusual case. Grothendieck shut down the Grothendieck circle a few years ago, one of their projects was collecting his work and providing an English translation. Because of his explicit cease-and-desist, this work is still only available in annoyingly high-level French. This is completely unacceptable to me (mostly because I still have a hard time reading it).
In this case, we can translate the text in complete technical paraphrase, meaning work through the definitions and theorems and give completely original different text containing the same technical content, as judged by a competent technical reader of both languages. This is the analog of a clean-room free-software port of existing closed software, there is nothing crazy old Grothendieck can do to stop it.
I know for sure that hosting an English translation of Grothendieck will get us a bunch of new eyeballs. Simlarly true for missing translations of Soviet era Russian language literature.