Searching for the Model of Common Ground in Human-Computer Dialogue

Natural language dialogue is a desirable method for human-robot interaction and human-computer interaction. Critical to the success of dialogue is the underlying model for common ground and the grounding process that establishes, adds to, and repairs shared understanding. The model of grounding for human-computer interaction should be informed by human-human dialogue. However, the processes involved in human-human grounding are under dispute within the research community. Three models have been proposed: alignment, a simple model that has been influential on dialogue system development, interpersonal synergy, an automatic coordination emerging from interaction, and perspective taking, a strategic interaction based on intentional coordination. Few studies have simultaneously evaluated these models. We tested the models’ ability to account for human-human performance in a complex collaborative task that stressed the grounding process. The results supported the perspective taking model over the synergy model and the alignment model, indicating the need to reassess the alignment model as a foundation for human-computer interaction.