Abstract: Constraint-based modeling has been used in many application areas of
Intelligent Tutoring Systems as a powerful means to analyse erroneous student
solutions and generate helpful feedback. In contrast to domains where the
structure of the problem under consideration allows a constraint to (almost)
uniquely determine the possible cause of a particular student error, there are
other applications where a multitude of competing error explanations has to be
considered. In such cases constraint-based models alone hardly meet the
requirements for a student model. Instead a constraint-based model clearly
serves the purpose of error diagnosis and needs to be complemented by
additional components for diagnosis selection based on general or individually
tailored heuristics. By investigating the apparent and strong parallelism
between constraint-based modeling and model-based diagnosis, this paper
identifies four major sources of ambiguity that need to be considered when
using constraint-based modeling and describes options for dealing with
situations in which alternative error descriptions are available. Examples are
primarily drawn from the area of foreign language learning.