Extracting Social Meaning: Identifying Interactional Style in Spoken Conversation

Automatically extracting social meaning and intention from spoken dialogue is an important task for dialogue and information extraction applications. We describe a system for solving the new task of detecting elements of interactional style: whether a speaker is awkward, friendly, or flirtatious. We create and use a new spoken corpus of approximately 1000 4-minute speed-dates. Participants rated their interlocutors for these elements of style. Despite the difficulty of automatically determining interactional style from noisy natural data, we are able to detect flirtatious, awkward, and friendly styles in conversations with between 55% and 65% accuracy, compared to a 50% baseline, based on prosodic, lexical, and discourse features. We describe simple ways to extract relatively rich dialogue features, and analyze which features performed similarly for men and women and which were gender-specific.