I think it's completely unreasonable to expect authors to have read and completely understood the details of every paper that they cite in their references. Most authors certainly do not do that. We'd never have time to write any of our own papers if we had to thoroughly digest every single other paper that is relevant to our work. As I see it, the rules should be: (1) don't cite anything you don't refer to in the paper. (2) only cite those papers which are relevant to your work, although it's okay to have some papers in your references which are only cited in the sense of "papers [x] and [y] and [z] also consider similar problems..." and (3) you should completely understand the statement of any result you use. If you haven't actually checked the details of the proof, most mathematicians with experience in their fields can skim over a proof and decide whether or not they believe it. If the reference has been published and later turns out to be incorrect (which sadly happens all too often) then you cannot be held responsible for that.

Of course, the language of the paper you cite is completely irrelevant. I've never before heard of a referee caring about this.