Identifying Stylometric Correlates of Social Power

This thesis takes a stylometric approach to the measurement of social power, particularly hierarchical power in an organisational setting. Following the social constructionist view of identity, we infer that construction of identity is an ongoing process incorporating the full scope of human behaviour, including linguistic behaviour. We test the primary hypothesis that stylistic choice in language is indicative of power relations, and that a stylometric signal can be extracted from natural language to enable prediction of relationship status. Additionally, we consider the effect of individual variation versus interpersonal variation, and the effects of aggregating predictions to boost the predictive strength of the model. Three different datasets are used to validate the proposed approach across three different genres: email, spoken conversation, and online chat. We also present a vector space approach to modelling linguistic style accommodation, and undertake a preliminary examination of the correlation between linguistic accommodation and social power.