When learners make errors when constructing a planar geometrical figure according to
a teacher's formal specification, provide relevant explanation is a difficult
task for an ITS (Intelligent Tutoring System) diagnosing their behaviour.
I based my approach to solving this problem on the assumption that learners' productions
are consistent with their reasoning. Then by using induction on the domain theory and
learners' examples, it was possible to construct a theory that could explain these
productions, i.e. building a model of learners' misconceptions.
Because of the knowledge representation formalism used in the ITS (i.e. First-Order Logic),
Inductive Logic Programming offered interesting possibilities for this work.
The method I followed during this short project was to reuse existing ILP systems,
following a black-box paradigm. Although this method gave some encouraging results,
my research illustrated that the lack of counter-examples of students' conceptions
limits the induction process supplied by the systems.
More interactions between the system and learners will be required in order to generate
and classify the examples that are missing to prune the induction process.