Algorithms for constructing minimal deduction graphs

Two algorithms for constructing minimal deduction graphs (MDG) for inferring rules and facts in an extended version of the Horn clause logic are described. A deduction graph (DG) is minimal if the number of arcs in the graph is minimized. Horn clauses (HC) are extended to Horn formulas (HF), such that the head or the body of an HF can be a conjunction of positive literals or a disjunction of the bodies of some rule instances, respectively. Each algorithm constructs an MDG from its source to its sink, whose arcs infer the HF `if source then sink'. The construction of an MDG is based on a sound and complete set of inference rules of reflexivity, transitivity, and conjunction for HFs which proceeds by expanding a tree rooted at its sink until its source has a successful backtracking to the root. Then the MDG is extracted from the tree. The nodes being expanded in such a tree are classified into seven types, which are assigned by different priorities for their growing into subtrees or for their pruning to reduce the tree space