Student Modeling for Language Tutors

Workshop at AIED 2005

12th International Conference on Artificial Intelligence in Education

18-22 July

Amsterdam, The Netherlands

Topics and goals:

Student modeling is of great importance in intelligent tutoring and intelligent educational diagnostic and assessment applications.  Modeling and dynamically tracking a student's knowledge state are fundamental to the performance of such applications. However, student modeling in CALL applications differs from more "classic" student modeling in other domains in three key ways:

1. It is difficult to determine the reasons for successes and errors in student responses.  In classic ITS domains (e.g., math and physics), the interaction with the tutor may require students to demonstrate intermediate steps, and there exist heuristics and approaches (e.g., model tracing) to determine where a student’s problem solving efforts goes awry.  For performance in language domains, much more learner behavior and knowledge is hidden, and having learners demonstrate intermediate steps is difficult or perhaps impossible, and at any rate may not be natural behavior.  (How) Can a language tutor reason about the cause of a student mistake? (How) Can a language tutor make attributions regarding a student's knowledge state based on overt behavior?

2. A priori cognitive modeling is harder in language domains.  A standard approach for building a cognitive task model is to use think-aloud protocols.  Asking novices to verbalize their problem solving processes while trying to read and comprehend text is not a fruitful endeavor.  How then can we construct problem solving models?  Can existing psychological models of reading be adapted and used by computer tutors?

3. It may be difficult to accurately score student responses.  For example, in tutors that use automated speech recognition (ASR), whether the student’s response is correct cannot be determined with certainty.  In contrast, in classic tutoring systems scoring the student’s response is relatively easy.  How can "scoring" inaccuracies be overcome to reason about the students’ proficiencies?

Given these differences, a focused workshop bringing together people working on student modeling in language tutors is appropriate as it provides a forum to discuss approaches to overcoming these problems.

This workshop will focus on student modeling for intelligent computer-assisted language learning (CALL) applications, addressing such domains as oral reading decoding, and reading and spoken language comprehension.  Domains of interest include both primary (L1) and second language (L2) learning. Hence, the workshop will address such questions as:

-         What should a student model for a reading tutor or other CALL tutors contain? What knowledge components and elements should be maintained?

-         How should information about users be represented? Using what representational formalisms?

-         With what (cognitive or other) design rationale?

-         How can information about the user’s knowledge be obtained (via interaction with the CALL application) and what sort of inferences can be made about a student’s knowledge based on empirical performance?

-         How, and for what tutor tasks, can the student model be utilized?

-         How can the student model help guide a tutor in terms of instructional or remedial interventions? In terms of assessment?

Target audience:

Researchers and developers of CALL applications that involve student modeling for intelligent diagnosis, adaptive intervention, and/or adaptive interaction.

Call for Papers and Proposals:

We welcome papers in thefollowing categories:

• Full papers (up to 8 pages) - Describes work (research, systems) that involves student modeling for language learning

• Position papers (up to 4 pages) - Describes your qualifications, background, and interest with regard to student modeling for language learning
  • We also welcome discussion or panel proposals

• Demonstrations (up to 4 pages) - Describes an application or other work to be demonstrated live at the workshop

Please contact the workshop chairs by email as soon as possible, briefly describing your intended submission.

Papers should be formatted according to IOS Press publication requirements.

When possible, please use these templates: MS-Word template, LaTeX template.

• Submissions must be in:
  • MS Word (.doc),
  • Portable Document Format (.pdf), or
  • Rich Text Format (.rtf)

• More details in the authors guidelines for the main conference (ignore the part labeled "Only electronic submission;" submit workshop papers to the chairs by email)

Important Dates:

Deadline for paper submission 25 April 2005

Notification of acceptance 11 May 2005

Camera ready version 20 May 2005

Workshop date 18 or 19 July 2005

Chairs (alphabetically): 

Sherman R. Alpert

salpert@us.ibm.com

IBM T.J. Watson Research Center

Phone: +1 914 945 1874

Joseph E. Beck

joseph.beck@cmu.edu

Center for Automated Learning and Discovery

Carnegie Mellon University

Phone:  +1 412 268 5726

Program Committee (alphabetically):

W. Lewis Johnson

Director, Center for Advanced Research in Technology for Education,
USC / Information Sciences Institute

johnson@ISI.EDU

Stephen A. LaRocca, Ph.D

Army Research Laboratory (ARTI)

slarocca@arl.army.mil

Lisa N. Michaud

Department of Mathematics and Computer Science,
Wheaton College

lmichaud@wheatoncollege.edu

Jack Mostow
Director, Project LISTEN,
Carnegie Mellon University
mostow@cs.cmu.edu

More about the chairs:

Sherman Alpert has been at IBM's T.J. Watson Research Center since 1987. He has worked and published in the domain of educational technology during much of that tenure, has built intelligent tutoring systems for Smalltalk programming and elementary algebra equation solving, and has done work in the area of user model based personalization. His latest work is on a Web-based oral reading tutor that uses speech recognition to assess student performance, provides scaffolded support for readers, and incorporates a student model that tracks readers’ knowledge in terms of underlying reading decoding heuristics. His homepage is at http://www.research.ibm.com/people/a/alpert/.

Joseph Beck has been working on student modeling in general for his entire career, and student modeling in a language tutor for the past three years (Project LISTEN at Carnegie Mellon University).  He has developed statistical models that use automated speech recognition to estimate student reading proficiency, and has worked on applying classic student modeling techniques such as knowledge tracing to this new domain.   See http://www.andrew.cmu.edu/~jb8n for his homepage, and http://www-2.cs.cmu.edu/~listen/pubs.html for his publications.