A General Architecture for a Real-Time Discourse Agent
and a Case Study in Oral Reading Tutoring
Graduate student, Computational Linguistics
Carnegie Mellon University
Committee members
Dr. Jack Mostow, advisor
Robotics Institute/Language Technologies Institute/Human Computer Interaction
Institute
Dr. Nancy Green
Robotics Institute
Dr. Alex Rudnicky
School of Computer Science
Abstract
Humans engage in a wide range of turn-taking behavior, including
backchanneling, multiple turn-taking, interrupting (barge-in), and pausing.
Spoken dialogue systems generally engage in only a small subset of these
behaviors: single turn-taking and indefinite pauses.
Spoken discourse relies on appropriate turn-taking.
In tutoring discourse in particular, instructor timing impacts student
learning.
Human tutors use rich turn-taking; computer tutors that listen should too.
In this project, we present a general architecture for a real-time discourse agent,
with particular emphasis on the ability to use a rich set of turn-taking behaviors
when participating in spoken multimodal dialogue.
As a case study, we have instantiated this architecture
in a computerized oral reading tutor being developed by Project LISTEN.
The tutor listens to children read aloud
wherever possible and offers assistance wherever necessary.
Table of Contents
Assignments
Assignments are no longer available over the Web.
Contact Information
Any technology, sufficiently advanced, is indistinguishable
from magic. -- Arthur C. Clarke
Corollary: Any technology distinguishable from magic is insufficiently
advanced.