Manuscript for the Thematic Issue of Phonetica Entitled Progress in Experimental Phonology: from Communicative Function to Phonetic Substance and vice Versa Title of This Manuscript: Exploring Prosody in Interaction Control

This paper investigates prosodic aspects of turn-taking in conversation with a view to improving the efficiency of identifying relevant places at which a machine can legitimately begin to talk to a human interlocutor. It examines the relationship between interaction control, the communicative function of which is to regulate the flow of information between interlocutors, and its phonetic manifestation. Specifically, the listener's perception of such interaction control phenomena is modelled. Algorithms for automatic online extraction of prosodic phenomena liable to be relevant for interaction control, such as silent pauses and intonation patterns, are presented and evaluated in experiments using Swedish Map Task data. We show that the automatically extracted prosodic features can be used to avoid many of the places where current dialogue systems run the risk of interrupting their users, and also to identify suitable places to take the turn. 1 Introduction Conversation is a primary means for human communication. An important function of conversation is to exchange propositional content, and during this exchange the interlocutors must somehow regulate the flow of information to make it proceed smoothly and efficiently. This interaction control is a collaborative effort where interlocutors continuously monitor various aspects of each other's behaviour, including semantics, gestures and prosody, in order to for example make decisions about turn-taking and feedback. The term turn-taking includes how to take the floor without interrupting, how to keep the floor, and how to pass the initiative on to others. Feedback is used to indicate to the speaker that the listener is attentive, understanding, agreeing, etc. The aim of this work is to improve the interaction control in spoken human-computer dialogue. The primary motivation for this is that if a spoken dialogue system is to be 3 perceived as a good conversational partner, it has to be able to understand the speaker's organisation of the content, and to know how to time its own contributions to the dialogue appropriately, in addition to recognising and responding to verbal input and generating verbal output. Current spoken dialogue systems commonly detect where the user ceases speaking in order to find out where they should take their turn. The method is based on the assumption that speakers have finished what they intended to say when they become silent, and that these points in time are also suitable places for the system to speak. Such endpoint detection triggers on a certain set amount of silence, or non-speech. …