My recent research has focused on the development of an intelligent
tutoring system (ITS) called CIRCSIM-Tutor
(CST).
CST tutors first year medical students studying cardiovascular physiology.
CST's domain consists of the physiology relevant to the human body's attempt
to maintain constant blood pressure.
Version 2 of CST is fully implemented and it runs gracefully. It accepts
free text (via the keyboard) from the student. The quality of the dialogue,
like other natural language systems, is poor. A design of CST v3 was initiated
with a thorough study of human tutoring transcripts. These transcripts
have been used to (1) identify the necessary knowledge for the knowledge
base, (2) study the role of student modelling, the selection of topics
and the selection tactics, (3) extract a lexicon and (4) study the protocol
of tutorial conversation. I have worked on the implementation of an enhanced
knowledge base (human circulatory system) and the student model. My most
significant contributions to this project have been:
-
An identification of how experienced tutors select topics to be tutored.
There are few research groups that study human tutoring for the purpose
of constructing a computer tutor. My findings are similar to the findings
of these other researchers; the differences need to be studied.
-
An analysis of hinting patterns by experienced tutors. There has been,
essentially, no other comprehensive study on hinting. Again, much work
remains in order to generalize my findings.
-
An analysis of the relationship between student responses and the selection
of tutorial tactics. Specifically, what type of responses will elicit a
hint from the tutor? An explanation?
back