No, I am not google translating. I was just doing that as a "reverse test" to see what the translators had actually written. We are using translators who are proficient in the ___domain and speak 2 languages at least (English plus the target language). But they are still "external" to the team. They have some ___domain knowledge, but they don't have the true conceptual knowledge about what things are called in the program. Translators inside the development team (for the couple of languages covered by the developers) don't make the mistake of picking an incorrect synonym. The external translators might, despite good ___domain knowledge. The problem with it is: We don't notice. Because we don't speak the language!
E.g. if you make an architectural program, you might call things "floor", "level", or "storey". In an editor you might have "object", "entity" and so on) and all of these might exist and mean different things in the program. Language wise they may well be synonymous, but in the "___domain language" of the program they might be very different.
As a frustrated non-English user of recent widespread degradation in software translations, I can verify that you're completely correct in your observations. Recent advents of "I give you lists of short gibberish to make guesses on but I don't listen for excuses because they're short" platforms like Crowdin are starting to put toils on i18n users, just as you have been finding out from the other side.
Basically it has to be bunch of screenshots or has to be tweet-long sentences, and it always had been, until someone ignorant in power started forcing that naivety.
E.g. if you make an architectural program, you might call things "floor", "level", or "storey". In an editor you might have "object", "entity" and so on) and all of these might exist and mean different things in the program. Language wise they may well be synonymous, but in the "___domain language" of the program they might be very different.