But the devil is in the details: Most english speakers don't know foreign languages, so they are unaware of lots of problems that occur from just a word list. E.g. there is a difference between "end" as a noun and "end" as a verb. In German, one would translate the former as "Ende", the latter as "beenden". So for the string table, the programmer would need two different annotated entries "end (n)" and "end (v)" for the translation to be correct. But usually, the translation team only gets "end" without context, annotations or anything, and of course all occurences of "end" will be conflated into one line in the translation table. There are other frequent problems, like sentences containing numbers that require logic to get right and distinguish between Singular/Plural, nothing and maybe even Dual cases. Or the problem that designers got their fingers into a GUI, leaving just enough room for English language strings, but the translations will be cut off or awkwardly skew the layout because they are too short/overflowing/unaligned.
This frequently leads to the situation where as a German, I have to translate to English and back to understand the meaning. Which is why I nowadays avoid translated software and do everything in English. I think translation without lots of effort is pointless. And nobody will invest any significant effort into "small overseas markets".
In FOSS it may be easier to get good translations because of the feedback from real users. They are often more proficient in the target language (than paid translators) and can recognize awkward phrasings and misleading situations. Translation is also an opportunity for small incremental improvements by tons of otherwise non-technical volunteers.
But for the aforementioned reasons, automatic translation makes things worse, not better. I have yet to find a useful translation of technical content, i.e. anything that goes beyond just understanding the general topic of a newspaper article.
I've tried to use only English language settings for decades. Otherwise, you just get an ugly (and often illogical or downright factually wrong) mish-mash.
Like when your locked-down work machine's OS is set to use your local language, but the software you use is only available in English. Lots of dialogs in English, but then the buttons at the bottom jarringly ask "Yes / No / Cancel?" in whatever language HR and IT thought was all your little mind could handle... Utterly infuriating.
Facebook's approach to this in their translation crowdsourcing interface is to provide extra context like an example sentence with the phrase, or a definition.
Git can handle this sort of thing with smudge/clean filters.
Say, you must pick identifier names from a subset dictionary of language concepts (pruned of synonyms and multiple senses), for which we have translations into X other languages. E.g. WidgetFactory is a valid name, since we can translate that, but FoznozzleBodega raises an error. So the resulting code can be trivially output in any language in the master dictionary.
There could also be multilingual literary programming by making the output syntax conform to the target language as well, in terms of conjugation, part of speech order, etc.
Code is a relatively low hanging fruit for debabelization.