Decision table language and its parallel execution architecture with applications in expert systems
John David Bezek · 1990
Decision tables have long been used for program documentation, design, and, to a limited extent, implementation. The inherent advantages of decision tables are: concise presentation of the conditions and actions of a program; ability to verify execution paths by direct use of the bounding conditions of the table and presentation of an algorithm by execution path; application to general purpose programming and to the programming of expert systems. The last point becomes the link between decision table implementation of an algorithm and the implementation of forward chained reasoning systems. Past decision table languages have been additions to conventional languages where the output of a table compiler or table translator is another high-level language. In this manner the decision table is used to augment the existing language. This study demonstrates that decision tables in and of themselves are viable for use as a programming language. Enhancements to the general principles make them an excellent tool for the implementation of parallel algorithms and expert systems. This study presents the background material on decision tables, content addressable memory, forward chained production systems and the LINDA tuple space primitives. The general concepts of decision tables are expanded into a full programming language. This language is then used to meld the above concepts into a powerful environment for general purpose programming and for the implementation of expert systems using forward chained reasoning. A case study using the map four coloring problem is given. Several implementations using the decision table language are given along with performance results and comparisons. It is shown that the use of content addressable memories along with a novel translation of forward chained rule sets using tuple spaces results in significant performance improvements. Several CAM-like chip designs intended for the acceleration of DTL condition section processing are given, at the functional level. Because of the demonstrated connection of decision tables and forward chained languages, these chip concepts can be applied for the enhancement of other logical inferencing systems.