From pattern-directed to adapter-driven computation via function-applying matching

Abstract

The generalization of pattern matching to adapter fitting. As implemented in the programming language FIT. is described semantically. Adapters are like patterns that contain functions which during fitting are applied to corresponding arguments contained in data instances. They are more concise, easier to read, and more efficiently implementable than equivalent LAMBDA expressions and pattern-action rules, because they can analyse data, like patterns, and manipulate Them, like functions, in one sweep. Variable settings created by pairing adapter elements with data elements are treated as expressions obeying a consistent-assignment rule, generalizing the usual single-assignment. While pattern-directed computation invokes transformation rules by matching their patterns to an expression and then performing the associated actions, adapter-driven computation only requires adapter/expression fittings. This permits a new representation of AI data bases, LISP functions. hypergraph operations, inference rules (incl. Wang's algorithm), Woods' RTNs, and Turing machines, showing that adapter-driven computation provides an AI-oriented general computational base. The efficiency of pattern-directed and adapter-driven computation is enhanced by introducing the SECURE operator as a functional alternative to PROLOG's cut.