There are always at least two knowledge graphs in use in the system, though it is usually more. In SemanTex documentation they are sometimes referred to as contexts, the terms are interchangeable. 

The long term context contains dictionary and thesaurus derived information and is the first source of the background knowledge needed for language processing. The dialog context contains the output of the language processing. Further knowledge-graph information can be accessed as required, perhaps the most frequently used are lists of given names, and the geo-names gazetteer for entity classification.

Further information on SemanTex language processing is here and information on the graphs is here 

​

The important point is this: The knowledge graphs both inform the language processor and are added to and refined by the language processor, knowledge is compiled and used to improve the knowledge compilation process.