translation management: investigate crowdin

[avsm]

following #1268 and the tutorial updates by @johnwhitington and a question from @gasche, I asked around about what other systems do for configuration management.

I got two answers:

• @smorimoto suggests https://crowdin.com/features which helps coordinate translators and has a very nice open source support offer.
• @patricoferris suggests NetlifyCMS's i8n capabilities over at https://www.netlifycms.org/docs/beta-features/#i18n-support.

The metadata could also conditionally render something on the live site that (a) warns the content may be out of sync with English and; (b) a link for someone to either edit or confirm the information is good by comparing to English.
<...> It's open-source https://github.com/netlify/netlify-cms

Any more suggestions? We can start by evaluating both of these.

[smorimoto]

We've been using Crowdin for quite some time in the Mastodon project, so I can help you set it up if you need it. However, I've also used Netlify CMS (I was helping them a little around 2018) as well, but both of them are designed to maintain i18n files without problems even if we don't use them, so I think there is no problem using either of them.

Sounds promising. Here is a related old issue (ocaml/ocaml.org#376). I haven't done any homework on this yet, but I wanted to compare weblate with these other promising candidates whenever I do dive into this topic area. An ocaml based voting system (https://github.com/glondu/belenios) uses weblate, so maybe glondu can weigh in on its pros and cons?

Off topic: I really admire Mastodon's mission. I hope it keeps growing!

[smorimoto]

Crowdin UI is more convenient, permissions are easier (translator/proofreader), integration with GitHub is less buggy (in Weblate I would often have to reset the state because of merge conflicts), being able to vote/approve translations (don't remember if that was a thing in Weblate) and it saving history of changed translations so if there's vandalism it's easy to restore previous data.

I was talking with Garg on that, and we are not using Weblate for these reasons.

Off topic: I really admire Mastodon's mission. I hope it keeps growing!

I'm very very glad to hear that! We are now working on things related to anti-abuse quite seriously behind the scenes. Stay tuned.

[comradekingu]

I was sure I had contributed to Mastodon, but turns out it was only https://hosted.weblate.org/projects/mastodon-twitter-crossposter/, which isn't doing much worse in terms of coverage than its upstream project on Crowdin.
Had a look at the strings and it actually fares better than the usual Crowdin quality. That isn't to say it couldn't, or wouldn't have been better than it is. (The "no" locale should be nb_NO btw, or in the very least "nb")

@avsm
Crowdin doesn't have an open source offer. It has a non-commercial open source offer.
If you really read its terms and conditions you will find out how egregious they are. Get ready to untangle a lot of legalese that sounds nice till you look at it in detail.

The Crowdin UI is a further disaster for productivity. Any sort of overview or navigation is as slow as getting stuff through the voting system. You can of course do that on Weblate, but don't set it at default+2, that is just broken.

Restoring history is easy enough on Weblate, and pure vandalism is extremely rare in any system.

Curiously, I don't think I would have found Belenios without it being on Weblate.
There is no similar discovery or community on Crowdin, and there is no openness. No source code, no browsing strings without logging in, can't view all projects, etc.

Full disclosure, I am on the review team over at Weblate. Haven't used Netlify CMS at all.

I thought about this more. I suspect most content contributors to ocaml.org will be intimately familiar with github and even editing files in the github UI, so the main benefit of a headless cms is to do validation on data as it's being entered (https://www.netlifycms.org/docs/widgets/). (This reminds me of haskell-companies repository which gets frequent drive by edits)

It might be worth sacrificing the data entry validation, in order to get the most powerful translation management system? Patrick is the expert here in headless-cms landscape. I am just offering my thoughts for him to mull this over.