Grounding

To ground an utterance means for the hearer to be able to integrate its meaning with her representation of the context (as captured in the information state). Different levels of understanding: hearing and recognizing words, understanding literal meaning, understanding non-literal meaning. Moreover, the hearer may understand something (perhaps with some degree of (un)certainty), but not want to accept it, e.g., when it conflicts with some of her information. Can a speaker ever be sure the hearer understood as intended? Can a hearer ever be sure she understood as intended? How does grounding work in a dialogue system? How to signal/check understanding? How to handle grounding problems? How to recognize and handle misunderstandings? Given the state-of-art in speech recognition, spoken dialogue systems have much greater need for feedback, clarifications and corrections than appear in human-human interaction. On the other hand, including explicit feedback and verification after every user's turn can make dialogue longwinded and inefficient. Optimizing the amount of explicit feedback is therefore an important challenge for a dialogue system.

• Modelling grounding in ISU-based systems: EDIS [Matheson
  2#2
  2000]; GoDIS [Larsson2003]
• [Martinovski and Traum2003] study causes of irritation in human-machine interaction, such as over-fragmentation, over-clarity, overcoordination, over-directedness, and repetitiveness of verbal action, syntax, and intonation. These may cause human reactions such as tiredness, tolerance, anger, confusion, irony, humor, exhaustion, uncertainty, lack of desire to communicate. Breakdowns are especially acute in cases where the user refuses to go on with what she perceives to be a farcical or impossible situation, and they may also lead to increased reluctance to interact with dialogue systems in the future.
• Determining problem-awareness places [Hirschberg
  2#2
  2001]
• Determining recignition-confidence and type of grounding feedback in spoken dialogue systems: Malte Gabsdil's dissertation;
• Producing helpful feedback in cases of speech recognition failures [Hockey
  2#2
  2003]
• Grounding in face-to-face communication [Nakano
  2#2
  2003]

Reading: [Matheson
2#2
2000]; [Larsson2003]; [Hockey
2#2
2003]