The temporal segmentation of dialogue as a basis for a multivariate analysis of speech convergence

[Truncated abstract] While much of the communication between humans is carried in the linguistic content of our utterances, a great deal of information is transmitted in more subtle ways. Speech convergence; seen as the increasing similarity of a range of linguistic and prosodic variables in the speech of interacting partners represents one such form of communication. While the communicative outcomes of convergence are increasingly well understood, exactly how it is produced by speakers and interpreted by listeners is as yet unclear and represents a distinct gap in our knowledge of human communication. A number of conflicting theories seek to account for the available evidence of this phenomena via either social or automatic accounts. The present thesis seeks to contrast the socially motivated theory of Communication Accommodation (Giles 1973)with the Interactive Alignment Model (Pickering and Garrod 2004); an automatic theory in which convergence is seen as a behaviour that allows interacting partners to parsimoniously share the burden of language production in order to gain a fluency benefit. Based on the assumptions underlying each theory, a number of predictions can be derived for experimental testing. Central to the present thesis is the notion that the study of individual variables is insufficient to test the differing assumptions of both Communication Accommodation Theory and the Model of Interactive Alignment. Instead, we suggest that such questions can only be addressed in a multivariate fashion by observing convergence over multiple variables simultaneously; something that few studies have attempted previously. Furthermore, as relationships between variables represent an aspect of interest, the analysis of one variable needs to be orthogonal and uninfluenced by the presence of other variables.