This comment contains a few logical fallacies.
> It states the cargo culted reasons, but not the actual truth
This dismisses existing explanations without engaging with the mentioned reasons. The following text then doesn't provide any arguments for this.
> Pronunciation is either solved by a) automatic language detection, or b) doesn't matter.
There are more possibilities than a and b. For example, it may matter for other things than pronunciation only. Also it may improve automatic detection or make automatic detection superfluous.
> If I am reading a book [...] I will pronounce it correctly, just like the screen reader will. If I see text in a language I don't recognize, I won't pronounce it correctly, and neither will the screen reader.
A generalization of your own experience to all users and systems. Screen readers aim to convey information accessibly, not mirror human ignorance.
> There's no reason that the screen reader will get it wrong, because <hungarian sentence> isn't ambiguous
This is circular reasoning. The statement is based on the assumption that automatic detection is always accurate - which is precisely what is under debate.
> If you can translate it, you already know what language it is in.
This a non sequitur. Even if someone can translate text, that doesn't mean software or search engines can automatically identify that language.
> The lang attribute adds nothing to the proces.
This absolute claim adds nothing to the logic.