Only those three languages? Or were those the only one that happened to be in the sample that was tested? I'm asking because if those three languages – which are pretty huge languages – fail, there are tons of smaller languages I would also expect to fail.

Can it pronounce names in these languages when they appear in the parentheses after the article title? For example, can it read the article Kaifeng from the English Wikipedia, and usefully pronounce the strings "开封" and "開封" in the opening paragraph?

An answer to this may help to investigate the technical reasons for why it fails.

However, I also recognize that a language selection control is not the same as article content. The most relevant questions are:

1. Does ULS provide something useful to a blind English speaker who is on an English Wikipedia article and wants to read the corresponding article in Chinese, Korean, Russian, or some other language? (It's possible that the user doesn't actually know this other language, but still wants to open the article, and it should be provided.)
2. Does ULS provide something useful to a blind Chinese speaker who is on an English Wikipedia article and wants to read the corresponding article in Chinese? It's quite possible that this person doesn't know English, but landed on that article because that's what was given by a search engine, and they want to find an article in the language they know best.

(In both cases, English is just an example. It could be any other language.)

ULS in general and Compact Links in particular did get some accessibility testing in 2017 or so, before the wide deployment of Compact Language Links, and as far as I can recall, it was found to be mostly accessible. Nevertheless, it won't hurt to do another round of testing, or perhaps even develop something new that fully takes blind users into account. (Actual prioritization of this is up to the relevant product managers.)