This paper describes 'Mr. Collins', a student model for intelligent co
mputer assisted language learning (ICALL), which is based on the resul
ts of empirical studies conducted in the target domain. 'Mr. Collins'
is constructed and updated through student/system collaboration. This
collaborative approach to learner modelling aims both to promote stude
nt reflection through discussion of the student model, and to provide
more detailed and accurate information for the model. Due to his own c
ontributions the student is less likely to regard the model with suspi
cion. Because the student model is the focus of the system, it must co
ntain more information than is typically found in learner models. Ther
efore, in addition to domain knowledge and misconceptions, 'Mr. Collin
s' also models learner confidence, background languages and learning s
trategies. The typical acquisition sequence for the target rules is ex
plicitly modelled to provide an additional perspective from which the
system may assess learner input.