“Cognition may imply the existence (and coexistence) of several different interpretation universes, and a specially important type of learning consists in acquiring new ways of interpreting the world or some aspects of it.
When a new interpretation domain is learnt (i.e. new segmenting of the world and new inference rules), the new concepts and new rules are not yet settled nor directly interpretable in the world. They have to be linked with known entities. Therefore, when a model is built in terms of target entities, it in fact refers to the world mostly through entities and relations belonging to the currently operational domain(s).
Aspects of this model might thus be filled in thanks to automatic (and unchecked) inferences within the source domain(s). This is the basis for the tunnel effect.“