Iconic-Declarative Programming and Adaptation Rules

Abstract

Functional and logical languages permit declarative programming, i.e. executable high-level problem specifications. However, to obtain optimum intelligibility, programs should also be iconic, i.e. directly model their domain, as illustrated imperatively by object-oriented languages. After a discussion of iconic aspects in pure LISP (call nestings) and pure PROLOG (invocation patterns), an advanced iconic-declarative technique in the functional/logical AI language FIT is presented: adaptation rules join left-hand side invocation patterns and right-hand side call nestings of transformation rules into single pictorial contexts that process data by global tests and direct local transformations. Their utility is exemplified in three domains (list processing, set normalization, and graph searching). The conclusions criticize the predominance of transformation rules and contrast iconic-imperative with iconic-declarative developments.