Having introduced models to formalise semi-structured data and DTDs, this section aims at formal models for the specification of ontologies. An ontology consists of domain concepts and further background knowledge. Feature graphs provide a means for the representation of domain concepts. Semantic information on the domain of individual type variables provides valuable knowledge for subsequent effective information inferencing. However, current approaches for ontological engineering as for example taxonomies or frame logic don't provide means to deal with arbitrary constraints. Feature graphs have an isomorphic declarative logic representation. While it is a standard practice to combine logical reasoning and constraint solving in ``constraint logics'', in the specific case of reasoning with graph-structured knowledge the straightforward translation approach will suffer from the generally inefficient handling of this kind of inferences in general-purpose logic. In order to combine the advantages of the more efficient term-based handling of feature graphs used in SEAMLESS with the expressiveness of constraint logic, a combined framework which allows for efficient graph operations without loosing the constraint solving capabilities is presented.
Exploiting the logical correspondences among graphs and logical specifications, it is proposed to model semantic constraints as constraint satisfaction problems that augment the declarative feature graph specifications. Domain concepts are represented as constrained feature graphs. Further background knowledge is represented by constrained feature clauses. These are clauses which embed graphterms that are interpreted as feature graphs.
Figure 
illustrates the representation of
a shared concept for a modem using constrained feature clause.
Figure: Formal Representation of the Domain Concept Modem
As each mediation starts with a user request formalised as query, first the procedure is sketched which retrieves those ontological concepts which are consistent with the query.