Would it be possible for the program to identify unique language in each paper and scan your document for that language as a sort of "citation check"?

Probably :) I would do that like this (http://en.wikipedia.org/wiki/Vector_space_model), and probably like this (http://en.wikipedia.org/wiki/Latent_semantic_analysis) too, which put together is just like "compare things using vectors and linear algebra and the blood of newts".

... would you want that if you had to do it as a separate step, though? Anyway, I think I might hold off on this feature, but it's a really interesting idea!