Relations on Contexts

As a result of our discussion with KSL's Ontolingua and ISI's Ontosaurus groups, we have agreed that the next big hurdle both systems must face is: identifying general patterns of translation/lifting which commonly arise in their applications. An understanding of relations between contexts in which information sources were developed may make the process of combining these inconsistent sources easier. Understanding these patterns would also provide the basis to identify useful defaults that the system could suggest to the user.

Our context research has already taken the first step in this direction. Prototypes include the lifting axioms which describe simple translations [BF95], and the lifting axioms which add a temporal argument to an ontology of spatial relations (i.e. on and above) as discussed in §B.1.2 and in [McC93]. We propose to do the basic research needed to identify the typical ways in which knowledge bases conflict and formalize the process of accommodating each type of conflict.

We see the process of combining knowledge bases as consisting of the following three parts:

1. Create an outer context which can be used to discuss and reason about the implicit assumptions of the KB's to be integrated.
2. Identify any conflicting assumptions.
3. Once identified, accommodate the conflicting assumptions by transforming the original KB's so that their implicit assumptions no longer conflict.

The formal theory of context already enables us to solve the first step. A good overview of our solution as well as various other solutions is given in [BM96]. We now need to focus our research on steps 2 and 3 above. Looking at a few simple examples we were able to identify, at least the following three sources of conflicts between knowledge bases:

different meanings of terms

different domains of objects

different propositions

implicitly assumed about the world.

And each type of conflict will need to be accommodated in a different way.