As conceptual models are both the result of and subject to communicati
on between domain experts and information analysts, it is of the highe
st importance to explicitly relate conceptual models to the language i
n which they are communicated. To achieve this, we identify a set of s
emantic primitives in a suitable conceptual modelling framework and pr
esent the prototype of a knowledge extraction tool which helps analyst
s to find useful primitives in free natural language texts. In additio
n we suggest possible extensions of the presented approach to other ph
ases of conceptual modelling.