Replace crowdin by Weblate

[comradekingu]

(30 min later we return to what I was actually meant to be doing.)

In the spirit of

Lichess is free/libre open source software. You can download, read, use and modify every bit of source code.

Crowdin is anything but, and moreover not even functionally conducive to facilitate quality translations. It is all Weblate these days, save for translatewiki, but I think projects need some connection to the wikimedia foundation to use the official platform. Without a way to find multi-lingual suggestions, or span large swaths of strings at once, arriving at any level of consistency is difficult. Made harder still by the default settings of the voting system, and dragged down in buggy defaults.

Get the languages translated fully.

Is still open in #6524

6 years on from starting Crowdin, basic coverage is a fairly low ball to clear with 2 906 contributors, were it not for drive-by translations, as a core problem of the chosen platform. Moving everything but websites currently seems like a no-brainer to me. :)

(Full disclosure I have used both, contributed to Weblate, and got a T-shirt from Weblate at FOSDEM.)

[superuser-does]

Weblate is a much better platform now than it was when Lichess moved from its own custom solution over to Crowdin. These are the key functionality gaps between them:

• Weblate has a gap in the screenshot management feature, which is much more powerful in Crowdin. I checked a few months ago out of interest, and I don't believe there is a ready-made importer.
• Crowdin doesn't support fuzzy strings to my awareness (i.e. where management marks a strings as outdated due a change in upstream, without deleting existing translations). Weblate does, which could support future improvements to text on the site and app.

We might also lose other valuable information in the thousands of comments left on individual strings over the years. It's also worth remembering the workflows that have been built around Crowdin and the inevitable monitoring and security infrastructure that system administrators would have to set up. I say 'set up' because Weblate's free plan may not meet our needs. We have 140 languages turned on in Crowdin, mainly to stop the delay of a translator coming in and requesting that their language is enabled. You would note though that almost all of them have had at least a few words localised. It's a starting point, and we don't want to discard it. Weblate Hosted for libre projects is limited to 90 languages. I would also remind that Transifex too was once FOSS through-and-through. There is no guarantee that Weblate won't go down a closed path in due time. I don't believe strongly in this counter-argument to adopting it even though I'm the one making it, but it is again a note that could encourage us to reconsider.

I think it is fair to consider Weblate for the long term but we would lose a lot if we made this as a snap decision. At the moment, Crowdin meets our most important needs.

It is all Weblate these days, save for translatewiki, but I think projects need some connection to the wikimedia foundation to use the official platform.

were it not for drive-by translations, as a core problem of the chosen platform.

I want to comment on this part at length. I have to disagree based on my experience in Lichess's translation project along with those of many others. Having a complicated sign-up process discourages contribution. We rarely offer proofreader rights within Crowdin's contribution model and when we do, discourage proofreaders from approving strings unless they are absolutely certain. Along the same lines, I disagree with criticism of the voting system and 'buggy defaults'.

If by 'drive-by contributions' you refer to the wider Crowdin community, we have by choice not listed Lichess in the Crowdin 'Explore' menu, preferring to recruit from our own playerbase and this GitHub project (and let's not forget GitHub is also proprietary!).

I believe Crowdin could be better at exposing the glossary for the uninitiated, while the TM functionality is good but poorly presented. However, these are not much better on Weblate, which I'd argue gives an overall lesser experience for translators. I don't think it's far behind but it's important to look past ideology for a moment and be fair in our assessment.

[comradekingu]

From my point of view, it seems moving and then fixing the source strings early on makes sense. The WL libre plan allowances are not hard limits, and it would be far from irrational to pay for the service. I can help with both of those to some extent.

If the platform doesn't lend itself to continual improvements and feedback that reaches and improves the source stringbase, it all becomes a sunken-cost fallacy.

As a measure of potential quality, the comments are valuable, and in the same vein should not pile up. Keeping them where they are with the translation locked prevents any loss. Fixing comments gets a bit easier if the platform cooperates and bears promise of having something that makes more sense.

Transifex was never good, and still isn't. Their culture is the same, and Weblate can't do what TX did as it has a lot more outside contributors to its copyleft-only codebase. GitHub is at least functional, and I am saying that as a GitLab stockholder.

There is a possibility of just pointing to a directory of screenshots in WL now. Not so sure how that works.

I don't believe in the model of quality by assumed authority, and it also is more costly to keep most people out. My main gripe is that nobody has an easy time gaining any overview once the UI is in the way. The buggy defaults extend to having the wrong locales, for example.

I think the Crowdin glossary is the best in terms of putting in different info and being less clunky, but having stuff show up next to the string and the functionality of preventing illegal translations is good enough.