RETE and chart parsing from bottom-up call-graph caching

Abstract

A new mechanism is presented for graph instantiation. It is used to investigate various bottom-up data-driven AI algorithms. The author views instances as fairly autonomous objects that inherit graph information (e.g. links or relations) from their classes. Instances themselves can be instantiated. For example, a parse tree symbol is an instance of a grammar class-graph symbol, which itself is a symbol of instance. The model focuses on underlying computational processes, and makes little use of program text. Instead of transforming programs, CGC (call graph caching) is used to instantiate processes. Experimental results show that RETE matching, efficient context-free parsing, and truth maintenance are different manifestations of the same underlying computational process.<>