Kurt VanLehn, Reva Freedman, Pamela Jordan, Charles Murray, Remus Osan,
        Michael Ringenberg, Carolyn Rose, Kay Schulze, Robert Shelby, Donald Treacy,
        Anders Weinstein, and Mary Wintersgill.
 

Model tracing tutors have been quite successful in teaching cognitive skills; however, they  still are not as competent as expert human tutors.  We propose two ways to improve model   tracing tutors and in particular the Andes physics tutor.  First, tutors should fade their caffolding.          Although most model tracing tutors have scaffolding that needs to be gradually removed (faded),  Andes' scaffolding is already "faded", and that causes student modeling difficulties that adversely   impact its tutoring.  A proposed solution to this problem is presented.  Second, tutors should 
integrate the knowledge they currently teach with other important knowledge in the task domain  in order to promote deeper learning.  Several types of deep learning are discussed, and it is  argued that natural language processing is necessary for encouraging such learning.  A new project,
Atlas, is developing natural language based enhancements to model tracing tutors that are intended  to encourage deeper learning.

            PDF 993KB