Turn-Taking

Who speaks when? How to know when to speak? How to get to speak? How to know when to stop speaking? Turn-taking in human-human interaction involves an interplay of clues through syntax, intonation, pausing, eye-contact and gestures. A dialogue system, on the other hand, may only listen when it decides it is the user's turn and otherwise ignore any input, or it may use push-to-talk. How do any such constraints on turn-taking influence the dialogue? What challenges arise when we start crossing those barriers (e.g., when we want to handle barge-in)?

• [Cassell
  2#2
  1998] discuss the function of gaze in human-human conversation as signalling both turn-taking and information structure;
• [Duncan1972,Duncan1974] is cited in [Cassell
  2#2
  1998] as suggesting several cues that the speaker employs to indicate the end of a turn or invite the hearer to take a turn. These cues include falling pitch at the end of a sentence, the drawl of a syllable at the end of sentence, the termination of a gesture, specific phrases at the end of syntactic units, and changes in gaze direction, such as the speaker s looking away from the hearer as an utterance begins and toward the hearer as the utterance ends.
• Several links to more recent investigations into prosodic cues of turn-management can be found by submitting ``prosody turn taking'' to the google search engine.
• [Sidner
  2#2
  2003] address the modelling of engagement, i.e., the establishment, maintenance and abortion of contact in dialogue. They focus in particular on gestures, such as looks at/away from the conversational partner, pointing behaviors, (bodily) addressing the conversational participant and other persons/ objects in the environment, all in appropriate synchronization with the conversational, collaborative behavior.

• intelligent barge-in, e.g. [Ström and Seneff2000]

Reading: [Sidner
2#2
2003]