User preference in conditional and unconditional grounding in turn-based spoken dialogue systems

Spoken dialogue systems are used more and more in everyday life. Unfortunately, there are often misunderstandings between the dialogue system and the user. To give the user the opportunity to repair these misunderstandings, grounding is implemented in spoken dialogue systems. Grounding is the process of reaching mutual belief that every person partaking in the conversation understands what has been said. This paper attempts to find which implementation of grounding users prefer. To do so, two types of grounding have been defined, conditional and unconditional grounding. Conditional grounding implements speech reparation by means of explicit verification through confidence levels. Unconditional grounding implements speech reparation by means of implicit verification, repetition of the answer of the user in the next question. In this paper, users (n=20) have been asked to interact with both a conditional and an unconditional prototype, to determine which type of grounding in spoken dialogue systems is preferred by users.