There is a competition on building categorization on the fly which is called TRECK? Natural Language Processing.
Some good approaches to my particular field are the Cataloging provided by Ontologies, but this is not enough because ontologies do a very precise thing and they do it very precisely, on the other side we want to have a very specific solution which accomplish a very definite task.
The other theoretical field to look at is the Spatial Reasoning, particularly in the field of AI. On the the examples may be provided by the Discourse Rhetorical Theory (DRT) [Ludwig, 2000]. Also, Brendan Bennet, University of Leeds.
Also the text categorization as per Moore and Paris.
Another extra piece of work to look at is the Question Answering, particularly the work of Maarten de Reijke.
Concerning the Ontologies, probably the best approach is to look at Formal Ontologies which mix psychological, logical and philosophical approaches. Two researchers to look at may be Berry Smith and Cristiano Castelfranchi.
A possible procedure that I should try to put in place for the thesis’ work is to define a psichological model, then translate this model into a logical description and finally translate this into a software implementation. After this it is possible to run the “Prova di Completezza”, which verifies if the algorithm is actually responding to the logical method. Logic is based on semantics therefore a model can be validated from the right semantic implications (meaning).