For Knowledge Information Processing Systems in Expert-Systems or CAD Systems, it is important to handle massive knowledge efficiently. For this purpose, it is feasible to use Relational Database to construct Knowledge Information Processing Systems. But a mere connection of a Prolog Machine with a Relational Database is insufficient because of its communication overhead between the two elements. We take integration-approach ; enhancement of Relational Databases with a facility of unification. To implement resolution for deductive databases based on integration-approach, it is important to extend relations so as to represent terms, and to enhance relational algebra with unification power. We develop EXTRA ; extended relational algebra with unification power, and a resolution algorithm using EXTRA. Compared with conventional algorithms, our algorithm is more efficient in some cases. We verified this by counting processing steps by number of tuples processed. To make our resolution algorithm more efficient, some facilities must be added. One is to evaluate goals according to the informations about the goals. We found it is effective to evaluate goals according to the types of definitions and numbers of definitions for the goals. Another facility is to use EXTRA sequences instead of resolution. It is effective when the no recursion is involved. We will integrate these facilities into one algorithm. To do this, some preprocessing must be needed. We also study programming environment of deductive databases for expert systems, and architecture to support extended relational algebra. We will integrate the results of these studies to make up a deductive database based on relational algebra.

PAL is an extended Prolog to deal with inheritance hierarchy (IH), which is a tree structure formed by the inclusion relation among classes and instances. It adopts class bound variables (CBVs), which are variables with class constraints. CBVs improve the inferential efficiency greatly by suppressing unnecessary backtracking and exhaustive searches <(2)>. In this paper we propose for IH Prolog two new techniques : IH compiler and IH clause indexing. IH compiler associates classes with intervals in order to reduce the ancestor-descendent relation of classes to the inclusion relation of intervals. This makes the unification of CBVs very efficient. IH clause indexing adopts class keys for hashing as well as instance keys which are commonly used in ordinary clause indexing, and enables PAL to avoid unnecessary tests of matching for clauses that do not satisfy class constraints of CBVs. These techniques are indispensable especially when we deal with large scale data bases, where much irrelevant knowledge interferes with fast inference. They can suppress the unnecessary information processing caused by such irrelevant knowledge, and can make the inference in the IH Prolog far more efficient.

Discontinuity is a useful notion in understanding the behavior of physical systems without warrying about unmotivated details. In this paper, we present a method of analyzing discontinuous changes at a commonsense level. We take simple pulse circuits as subject. The center of the method is an algorithm for predicting the behavior of given circuits in causal terms. Potential obstacle in developing this algorithm was that ordinary circuit equations lack sufficient information for causal analysis, resulting in ambiguity. To overcome this problem we assume discontinuous changes are very rapid continuous changes to choose a standard explanation from candidates. We also introduce the notion of mythical state to qualify intermediate stages of discontinuous change which does not satisfy circuit equation. The algorithm is fully implemented using Utilisp. This paper consists of three parts : we first overview the qualitative reasoning method which we base our algorithm on, we then study properties of discontinuous changes, and we present the algorithm. The solution to all examples shown in this paper is outlined in appendix.

In this paper we propose a natural and general model of updates of knowledge bases, where we represent a knowledge as a closed formula of first-order logic and consider a knowledge bases a theory. For this purpose, we extend the model of updates of theories proposed by Fagin et al., and define the join of theories. Furthermore we extend the concept of join to treat logical databases of Fagin et al., and show that we can formulate the concept of insertion as a special case of the join of logical databases. We also give an example of updates of a knowledge base to illustrate our ideas. This model gives the semantics of the update of knowledge bases and will work as a basic model for investigating the behavior of the knowledge base management systems.

The clause form is very popular knowledge representation method for logic based reasoning systems. Recently, most working systems adapt Horn clause form. In fact, Horn clause set covers a great part of our knowledge. It will be true, however, that there exist problems which cannot be formalized by only Horn clausis problem can be solved, ordinary mechanical provers cannot reach the solution because of physical restrictions. Namely, slmple resolution ends in combinatorial explosion of search space. Some solution methods have been reported for this problem. At present, many-sorted resolution is the best solution method among them. In this paper, we propose an extended clause form and its resolution method. They are called guarded clause form (GCF) and guarded linear resolution (GL-resolution) respectively. GCF reflects meta-level knowledge about roles of predicatees. Schubert's steamroller is one of such problems. Although it is clear that ths in a clause. GL-resolution makes use of this feature of GCF to gain high search efficiency. We demonstrate the advantage of guarded resolution by proving Schubert's steamroller in comparison with many-sorted resolution.