Consider: maybe you know the keyword that should be in the document you want to find... but the author of the document you want to find, didn't use that keyword. Instead, they used a synonym for it, because they were unaware of the idiomatic keyword to use.
Example: medical journals. Eventually, everyone agrees on what the term for something is. But the very first few papers on that thing, might not call it that, because they were inventing it.
Do you want to fail to find those first (and most important) few papers on the subject? Or do you want Google to do the text-parsing equivalent of "snap to grid", ensuring that documents that say the right thing the "wrong" way still show up?
As far as I can tell, as far as the output goes, correcting your language and correcting the document's language are basically indistinguishable. Google could just be thinking of what it's doing as the latter, and yet it ends up having the same effect as the former.
I see it a lot Google searching for electronics terms, where combining common and unusual term sometimes causes unusual term to be ignored. The quotes used to help, but they no longer do.
As for your last point, metadata is all about addressing these issues. The context has to remain canonical, otherwise God knows what sort of legal issues would arise.