Conversational structures affecting auditory likeability

Three-person telephone conferences of unacquainted people are conducted by means of pre-defined scenarios providing individual information and goals. Interlocutors rate the likeability of each other after a training session as well as after the actual conference. The recordings of the conferences are manually annotated concerning speaker’s verbal contribution, pauses, and back-channels. Regression analysis reveals likeability ratings after the conference to be dominated by the ratings before the conversation, but simple parameters like number of turns also contribute significantly to a descriptive model. The regression model for ratings averaged for both interlocutors provide a similar fit as the one for pair-wise ratings. Exchanging of the manually obtained parameters by automatically estimated values still results in significant regressions, indicating facilitation for future research.