Ok, as a new Wikidata editor I sense an opportunity to learn something here... I understand that you are managing bulk edits en masse (which is of course only sensible with the total number of items approaching 100 million), and I agree that the actual language of the description (or rather the label, which I believe is what we are talking about) may be unimportant for a disambiguation item that doesn't even have corresponding Wikipedia pages in those languages, but will any label, regardless of language, provide some kind of advantage over having no defined label at all in most languages? Or, is it perhaps an unavoidable side effect of the task your bot is supposed to perform? And why is the spelling important, if not the language?
I'm not objecting, just curious. I'm not primarily concerned about the linguistic accuracy of data that is hardly ever seen by anybody, but rather about the confusion it may cause among other editors (like myself) perhaps prompting us to spend time fixing unimportant "errors" instead of contributing useful new information. Anyway, this issue now makes me want to do some statistical data mining to try to get a better overall picture of what this database really contains!