Conceptual graphs (a kind of semantic networks) [Sowa 84] are finite, bipartite graphs. The nodes are either concepts or conceptual relations. The two kind of nodes are connected by directed arcs. Each concept consists of a type label and a referent field pointing the individual in question (see Figure 3). All types form a taxonomic type hierarchy. Concept types and relations can also have type/relation definitions.
There are four canonical formation rules which are used for derivation: copy, restrict, simplify, and join. Another operation is projection which extracts a subgraph pi v of u which expresses how the "pattern" graph v participates in graph u. Projection is the basic operation used in DB-MAT to extract knowledge relevant to the user request. For instance, from the graph shown on Figure 3, i.e., 'Oil particles emerge on the water surface if they are lighter than water" and from a similar graph that 'Oil particles precipitate if they are heavier than water", DB-MAT can produce the explanation that 'Oil particles emerge on the surface or precipitate under certain conditions'.