Social Power in Interactions

In this thesis, I investigate whether social power relations are manifested in the language and structure of social interactions, and if so, in what ways, and whether we can use the insights gained from this study to build computational systems that can automatically identify these power relations by analyzing social interactions. To further understand these manifestations, I extend this study in two ways. First, I investigate whether a person’s gender and the gender makeup of an interaction (e.g., are most participants female?) affect the manifestations of his/her power (or lack of it) and whether it can help improve the predictive performance of an automatic power prediction system. Second, I investigate whether different types of power manifest differently in interactions, and whether they exhibit different but predictable patterns. I perform this study on interactions from two different genres: organizational emails, which contain task oriented written interactions, and political debates, which contain discursive spoken interactions.