That is a very broad question, I understand. I am working in a project where my aim is to detect linguistic features of deception as a person is speaking or writing. Some work has been done for written text (e.g. this paper by Myle Ott, et al), but I have not found anything that provides theoretical insight into the development of the constructs. For example, when an average person lies, does s/he tend to backtrack more often (i.e. say something like "I went there yesterday to ... umm ... no, I think it was the day before")?
I am from a computer science background, working in natural language processing. Any pointers to linguistic theories will be extremely helpful, and will allow me to connect statistical observations from my experiments to these theories.