On YouTube, why have the multiple languages audio streams disappeared?

[abolibibelot]

Not so long ago, on YouTube, when gettting the list of available formats with yt-dlp, for some videos there used to be multiple audio only streams for various languages, presumably corresponding to automatic translations, but those seem to have disappeared. Why?
I downloading a few videos with those alternate language streams and found the result technically impressive (if fully automated) but weird and uncanny to listen to...
Side question: normally, whenever there are multiple audio versions (currently for combined video + audio streams, formerly for audio only streams as well), the lower code number should correspond to the original audio in the native language (for instance “140-0” or “95-0”), but it's always the highest numbered version. Why? That makes it more difficult to automate the selection of streams to download (since the code number varies depending on how many streams are available for each video).

Example:
In January I downloaded that video (which is seriously hilarious by the way), and saved the yt-dlp output for future reference as I usually do; I selected streams “136+140-10” as there were 11 audio streams available with code “140-#” (from “140-0” to “140-10”) and the original English was the last one. It also listed multiple languages for formats “91”, “92”, “93”, “94”, “95”, “96” which contain both video and audio, and likewise, the one in original English language was the last one.
If I re-run the same command now, the list of available formats is much shorter. Since then, using a “JavaScript runtime” in combination with yt-dlp has become increasingly mandatory so I added a call to QuickJS (the only one still working on Windows 7 out of the four suggested), now more formats are listed, but there's only one “140” stream in original English, and the total number of video and/or audio streams went from 139 to 95. There are still multiple languages but only for video + audio streams.

Another:
In November I downloaded that video, selecting both “140-0” and “140-1” audio streams. Here “140-1” was the original audio in French and “140-0” was an automatic dub in English. Like I said, the result is jarring, both positively as it's technically impressive, and negatively as it's so fu##ing weird. It seems mostly accurate and except for a few seconds around 05:20 it correctly distinguishes between dialog and songs. At some points though the auto dubbing is abnormally fast. An extra oddity is that it mixes up some male and female voices – the host Thierry Ardisson (who died a few months ago at age 76 – that show was from the late 1990s or early 2000s) is dubbed as a “woman”, while Geneviève de Fontenay, the old woman with a hat (who used to organise the “Miss France” contest – incidentally, co-host Linda Hardy was a former Miss France, that's why they were joking about her “honesty”, since winners of the “Miss France” contest are supposed to be both beautiful bodies and spotless minds), is dubbed as a “man”. (See attached video if you'd like to check it out.)
Likewise, if I run the same command now to list the available formats, there is only one “140” audio stream available in original French.

Another:
In December I downloaded that video, selecting four “140-#” audio streams out of nine available (from “140-0” to “140-8”). Again, the original audio was the last one, “140-8”. The result is particularly goofy, with the automatic dubs rendering overblown alpha male grunts as a well mannered conversation between a posh advertising man and a posh advertising woman, but still impressive as the marvelous machine manages to distinguish between actual grunts and halfway intelligible utterances. (Video also attached.)
Likewise, if I run the same command now to list the available formats, there is only one “140” audio stream available in original English.
By the way, with yt-dlp, when choosing multiple streams in a MKV container, I found no way of properly naming or tagging each one; in most cases, the first stream is tagged as “English”, which turns out to be correct for that one, but is incorrect if the first stream is in French for instance, and all others don't have any name. For the example above there is no language tag at all, even though it was downloaded at the same period with the same method. If it's not possible to have correct tags, I would prefer no tags at all.