Is common ground built during human-system dialogue device-specific? Examining knowledge transfer in multidevice systems.

When two people (or a person and a dialogue system) interact verbally, they build and then reuse mutual knowledge, or common ground. The goal of this work was to determine whether such common ground established with a system is device-specific or whether it can be transferred across devices. To answer this question, we used an adapted version of the matching task in which a participant had the opportunity to describe abstract pictures (Tangrams) to a Wizard of Oz system, thus creating common ground regarding how to refer to these pictures with the system. The participant then performed the same task again, but using a different device. The results suggested that the participants reused the common ground used with the first device when interacting with the second device, suggesting that this knowledge was perceived as shared with the system and as transferrable across devices. Theoretical implications as well as applications for the design of multidevice systems are discussed.