Abstract
Intelligent tutoring systems (ITS) can provide scalable and accessible education at scale. Generally, speech-based tutoring systems interpret the users' long pauses as a signal to take the floor. However, such behavior creates barriers for minority and low-literacy individuals who pause mid-sentence just to finish articulating their thoughts. For them, these systems inappropriately interrupt them, resulting in frustration. In this study, we present a bigram pause detection model that determines whether a pause is occurring in the middle or at the end of a sentence and implemented this model on an ITS (MODIFIED). We also compared the effect of this modification with a baseline ITS that interprets long pauses as the end of a turn (TRADITIONAL). Our results show that participants interacted longer with MODIFIED, and perceive it more human-like than the TRADITIONAL system. We argue that conversational systems need to be designed considering unique needs of their intended interlocutors.
Funding Information
  • National Science Foundation (2219586)

This publication has 15 references indexed in Scilit: