Mathematically rigorous NLP

Question

I'm looking for resources (books/articles/whatever) that provide mathematical formalization of NLP and statistical language theory. By that I mean clear exposition of the subject in terms of probability spaces (measure spaces) and so on. For example, many NLP books (like the Manning's one) use n-gram models which, as I see, may be modelled as Markov processes with word-states, but neither book states explicitly how the probability space for the process is constructed (I guess, there's something related to probabilities on formal languages?). I need such clear expositions. Thanks in advance.