I often use the automated suggestions in Crowdin. In many cases, I directly use the save button behind them, which saves that as the translation.
In other cases, I use them as a base to improve them as needed. Often, one of the words used does not match how the word has been translated before.
Translating a certain feature with a consistent name is quite important.
But using the DeepL translation is often enough, but it should be done by people who can judge if the translation makes sense in that context. Then Discourse could do that automatically and hope people would fix what isn’t translated that well. From my experience, most people don’t; reports about translations that need improvement here in the forum and directly on Crowdin are rare. You often need to know the original English text to notice that the meaning is actually not the same, and it’s rare that people look at both and notice the difference.
I’d say use the suggestions as often as possible and skip the texts where those aren’t helpful or you need more information to decide if they work in that context, so you get many translations as quickly as possible, but don’t add translations that you know need to be edited again. It will be very difficult to find those later.
But I am only someone who translates Discourse, maybe the team has a different opinion on this. I think they have added AI translations for some texts at the beginning of this year - which had a similar quality your approach would have ![]()
