1 USC Information Sciences Institute Yolanda GilFebruary 2001 Knowledge Acquisition as Tutorial...
-
Upload
aleesha-mills -
Category
Documents
-
view
212 -
download
0
Transcript of 1 USC Information Sciences Institute Yolanda GilFebruary 2001 Knowledge Acquisition as Tutorial...
1USC Information Sciences Institute Yolanda GilFebruary 2001
Knowledge Acquisition as Tutorial Dialogue:Some Ideas
Yolanda Gil
2USC Information Sciences Institute Yolanda GilFebruary 2001
Exploring Possible Synergies
Intelligent TutoringSystem
(ITS)
Intelligent StudiousSystem teaches
teaches
?
?
ITS
KA(RKF)
3USC Information Sciences Institute Yolanda GilFebruary 2001
Exploring Possible Synergies: Dialogue
Intelligent TutoringSystem(ITS)
Intelligent StudiousSystem(ISS) teaches
teaches
?
?
Good tutoring
strategies
Good tutoring
strategies
4USC Information Sciences Institute Yolanda GilFebruary 2001
What ITS community has
Mountains of example tutoring dialogues Can be analyzed for strategies, misconceptions, hints and
help E.g., http://www.pitt.edu/~circle/Archive.htm
Many and diverse tutoring system have been built Raised grades by 1.0 standard deviation units
Best humans raise grades by 2.0
5USC Information Sciences Institute Yolanda GilFebruary 2001
Main Approaches to ITS
Coached practice and review Socratic dialogue: questions discover student
misconceptions, avoid telling students what they need to know
Critiquing student solutions
6USC Information Sciences Institute Yolanda GilFebruary 2001
Model Tracing Tutors [Anderson et al. 85]
Contain a model of the cognition designers want students to engage
EXPERT MODEL
HIGH BANDWITH INTERFACE
PEDAGOGICAL MODULE
X X√
-----? -------? -----
7USC Information Sciences Institute Yolanda GilFebruary 2001
Model Tracing Tutors [Anderson et al. 85]
Expert Model: how student should reason simple, precise, complete problem solving strategies
HB Interface: where student displays reasoning goal trees, explicating
Pedagogical Module: feedback and hints immediate feedback, hint sequences with increasingly more help
EXPERT MODEL
HIGH BANDWITH INTERFACE
PEDAGOGICAL MODULE
X X√
-----? -------? -----
8USC Information Sciences Institute Yolanda GilFebruary 2001
Key Research Projects
CIRCLE Research Center @ CMU PACT Geometry tutor, Ken Koedinger Andes Physics tutor, Kurt VanLehn
– Model tracing approach
CST: CIRCSIM-Tutor, from Illinois Institute of Technology Socratic dialogue approach Domain: physiology Used in classrooms in a non-experimental basis
ACLS (& others) @ UMass teaches a new concept when relevant during a simulation of
ER Many, many others: NEOMYCIN, SIERRA, CASCADE,
SOPHIE,...
9USC Information Sciences Institute Yolanda GilFebruary 2001
Interactive Directive Lines of Reasoning [Rose et al. 2000]
Instead of mini-lessons, which require that students have prior knowledge and motivation
Tutor starts by presenting student with a scenario and lesson overview (“advanced organizer”) Useful to draw prior knowledge (e.g., stating an analogy) Useful to detect missing prior knowledge Useful to give context to the new knowledge
Tutor asks detailed questions Once student provides the desired answers, tutor
ends with a summary
10USC Information Sciences Institute Yolanda GilFebruary 2001
Interactive Directive Lines of Reasoning: An Example
Tutor: Let’s think about the difference between speed and velocity. A closely related distinction is that of the difference between distance traveled and displacement from the origin. Take as an example a bumblebee flying from point A to point B by means of a curvy path. If you draw a vector from point A to point B, you will have drawn the bee’s displacement vector. What does the displacement vector represent?
Student: The bee’s distance.[…]Tutor: So the equation for speed is the length of the
path traveled by the body divided by […], even if the path […]
11USC Information Sciences Institute Yolanda GilFebruary 2001
Fading and Deepening (I) [VanLehn et al. 2000]
Human tutors start with lots of scaffolding that later fades, while ITS tools are quite rigid: support one strategy
– st mix steps from different strategies– st wonders what to do next, tool’s advice seems random (but
he was!) force students to enter information they hold in memory provide too much scaffolding in detecting errors and
hinting solns– st looked for the last hint in the sequence that says what to
enter– hints are not bad, but may not make sense within student’s
context
12USC Information Sciences Institute Yolanda GilFebruary 2001
Fading and Deepening (II) [VanLehn et al. 2000]
Human tutors pursue deep learning
At most two nested strategies
e.g.: lesson on how acceleration opposes velocity when slowing down T: What is the definition of acceleration?
S: Velocity divided by time
T: Yes, it is the change of velocity divided by time
S: It’s the derivation of time
T: Well, forget about the definition of acceleration. Let’s try analogy. Suppose…
Tutor’s strategy: derive from definition
Almost right, tutor enters 2nd level strat.
Student is even more confused
Abandon top-level strategy for another one
13USC Information Sciences Institute Yolanda GilFebruary 2001
Fading and Deepening (III) [VanLehn et al. 2000]
Deep learning through knowledge construction dialogues Teach a domain principle
– Three main KC types: from definition, analogy, contradiction Teach to do right thing for right reasons (no guessing of
actions)– Tutor should ask to justify actions
Teach domain language – Tutor should ask to say “I applied <principle> to <objs> because
<goal>”
Emphasize basic approach instead of details– Tutor should ask student to state basic approach
Qualitative skills, not just quantitative – Tutor should ask qualitative questions during lesson
14USC Information Sciences Institute Yolanda GilFebruary 2001
Why do only some tutorial events cause learning? [VanLehn et al. 98]
Analysis of tutorial dialogues showed that depending on what is the rule being learned: Students that make an error (reach impasse) tend to gain Students that hear a generalization of a rule tend to gain Students that produce incorrect equation gained when
explained why it was wrong (though not when using calculus)
Suggested strategies for ITS: Tutors should let students make mistakes instead of
avoiding that by giving them strong hints Different rules may require different kinds of tutorial
explanations (e.g., stating generalization, showing why wrong, etc.)
15USC Information Sciences Institute Yolanda GilFebruary 2001
Discussion: Differences
ISS does not suffer lack of motivation ISS can be built with a lot more initiative and
participation than a human student ISS does not need “cognitive tricks”:
Eg, incremental hints, they can just be given the solution
16USC Information Sciences Institute Yolanda GilFebruary 2001
Discussion: Opportunities
Intelligent Student Systems Student guides dialogue using good teaching strategies
Training human tutors Tutor uses ISS to learn good teaching strategies
Simulated student colleagues “I think the tutor meant …”