The main goal of the work presented in this thesis, is to provide a system for automatically classifying utterances into different types known as moves. In order to do this, one can take advantage of certain constraints found in natural dialogues. Utterances of the same type follow each other with a degree of regularity. This study joins together these three aspects to perform automatic move detection, which is used in an automatic speech recognition system to constrain the possible number of words. This system is successful in identifying utterances types and subsequently reduces the word error rate of the recogniser. It also provides an in depth study into how people express the discourse function of an utterance through intonation and provides a unique model of dialogue structure.