Towards the detection of social dominance in dialogue

When developing human–computer conversational systems the complex co-acting processes of human–human dialogue present a significant challenge. Para-linguistic features establish rapport between individuals and direct the conversation in directions that cannot be captured by semantic analysis alone. This paper attempts to address part of this challenge by considering the role of para-linguistic features in establishing and manipulating social dominance. We propose that social dominance can be understood as an interaction affordance, revealing action potentials for each signalling participant, and can be detected as a feature of rapport not of the individual. An analysis of F0 and long-term averaged spectra (LTAS) correlation values for conversational pairs reveals a high degree of accommodation. The nature of this accommodation demonstrates that others will adjust their speech to match the current dominant individual. We conclude by exploring the implications of these results on the role of rapport and outline potential advances for the detection of emotion in speech by encompassing the entirety of pleasure-arousal-dominance emotional space.