Believe It or Not: Acoustic-Prosodic Cues to Trust and Mistrust in Spoken Dialogue

Trust is a fundamental component of human-human and human-computer interaction. In this work we examine the acoustic-prosodic features of trust in a corpus of interview dialogues. While previous studies have explored the characteristics of speech that is trusted or mistrusted by others, we study a complementary problem: what are the characteristics of trusting vs. mistrusting speech? That is, are there specific acoustic-prosodic cues in an interviewer’s speech that indicate whether the interviewer believes their interlocutor, or whether they are skeptical? We use a corpus of deceptive and truthful interview dialogues, where trust labels are explicitly provided by the interviewer for every question asked. We analyze acoustic-prosodic features extracted from interviewer turns and compare the features of trusting and mistrusting speech, finding several significant differences in features. Furthermore, we compare the features of trusting speech in our study of human-human dialogue, with previous findings from a study of trusting speech in human-computer dialogue. This work sheds light on the nature of trusting speech, and how it manifests itself when humans communicate with human vs. machine interlocutors.