What I found is that translated texts generally look rather fine, but I immediately found certain often-used single word items at a very wrong place.
Which means that the method to fish words from a big pond in order to throw them at UI texts to be translated is very wrong because it lacks context awareness.
Also the official recommendation to wholesalely translate UI items by free public translation services, without the necessary context machinery, sometimes seems to produce very unusable translations.
There is a translation centralization mechanism and effort via some public frappe site, but behind said “one-size-to-fit-all-translation-needs” recommendation there seems to be no “one-size-fits-all committee” (and how could it exist) “in charge” (and how could this easily come into being in FOSS) to clean up behind said public free naive-translation machine, and you can provide translations into said DB, but getting them into the actual product seems to be a lottery, so people complain about wasted efforts, as far as I understand from forum posts.
From experience I know that it’s not always easy to agree on translations between people with different backgrounds-and-little-domainspecific-knowledge.
So the method seems to be: produce sometranslation with little effort taken as an MVP and hope that someone takes over to do the next step of producing a real and coherent translation. Which is a valid method for distribution of develeopment+translation efforts.
But as long as contextual info is not very tightly integrated in the fishing-from-the-pond step, coherently producing a clean and at least correct translation is impossible even for a functioning committee, because the incomplete machinery might always reshuffle items which must stay at the intended place because it doesn’t use context-awareness (e.g. in which menu a translatable items sits).
This method is a situation where translation-competent people will quickly be frustrated, and indeed so much so that they might not even start to try participation. And I don’t say this lightly, having decades of professional translation experience in very different domains. Frankly, I wouldn’t touch the translations as long as this contextlessness persists, because I know the result beforehand, from experience. Many a time some project manager without knowledge of translation, via several intermediate sub project managers and even several intermediate enterprises who all want a share for their “work”, just sent a list of items to translate, often even alphabetically sorted (which takes away even the last remnants of contextuality), which is a dead horse order to begin with, which you only accept out of misery and if you had no income for some time and no bread any more, knowing full well that it will come back later because “the translator did a bad job”, in fact it’s all the PMs before who didn’t, and the translator COULDN’T do a good job, because in this situation it is IMPOSSIBLE. So this is how all these jokes about strange translations come into being. But the translator is the least responsible for it, most of the time.
The translator should have the web page in front of his eyes when translating and the machinery behind it MUST ensure that an item doesn’t pop up just anywhere else just because the word or the expression is the same. This is the part which is naive and even uncultured, because misunderstandings because of lack of context happen in any language in daily life all the time, and programming a misunderstanding-mechanism is a pre-alpha-“effort” at best (if not outright sabotage: maybe someone in the early development of gettext and .po files was sent by some industry spy/operative to lure inexperienced OSS fans into a dead end for later products, and sow discord as a byproduct? or were these just inexperienced youth? whatever, there is a matureness of software products which needs to be acquired). It might, but ONLY AFTER HUMAN CONFIRMATION. Without this, the translation endeavor will rest defective and frustrating and produce a situation where the product, more or less at first glance, will give an impression of incompetent work. If software-evaluating people see this year after year, it’s certainly bad publicity, and only adventurous people will look further.
Other than that, there also might be a need for a style guide and a terms dictionary (which certainly might evolve, too, adding to the communication needs) if any group (in any language) decides to produce and/or to try to help to produce a quality translation.
Please also note that the splitting of the monolith into modules (some time ago) might pose some additional communication and mutual education requirements if the whole thing is to have a uniform, coherent look.