Acoustic and Prosodic Correlates of Social Behavior

Abstract

We describe acoustic/prosodic and lexical correlates of social variables annotated on a large corpus of task-oriented sponta- neous speech. We employ Amazon Mechanical Turk to label the corpus with a large number of social behaviors, examining results of three of these here. We find significant differences between male and female speakers for perceptions of attempts to be liked, likeability, speech planning, that also differ depending upon the gender of their conversational partners. There has been much work in the speech community on the acoustic-prosodic and lexical indicators of classic emotions. Similar approaches have also been used to identify other related types of speaker state, including uncertainty, confi- dence, and deception, as well as less clearly 'emotional' states as charisma, sarcasm, personality, and medical conditions such as depression. More recently researchers have begun to explore the acoustic and prosodic cues that may be correlated with the production and perception of social behavior in conversation, including flirtation, agreeableness and awk- wardness. In this paper we examine the perception of three types of social behavior in conversation: likeability, the attempt to be liked, and conversational planning. These behaviors represent part of a larger ongoing study of social behavior in task-oriented conversation in the Columbia Games Corpus. Section 2 describes previous research in this area. In Section 3 we describe the corpus. Section 4 discusses the annotation of social behavior we elicited using Amazon Mechanical Turk. Our current experiments are described in Section 5 and we discuss our conclusions and future research in Section 6.