Other, more general improvement might come from elaboration of the the mapping between domain knowledge and text semantics which underlies the explanations. At present DB-MAT extracts relevant (sub-)facts from the domain knowledge with the projection operation and forms the explanation semantics with the help of text schemas and graph join (see Section 5 and [Angelova & Bontcheva 96b]). Further elaboration requires a detailed study of the way domain knowledge underlies the semantics of the future explanations.