Abstract Syntax Tree (AST) Infrastructure in Problem Solving Research
Joseph M. Scandura
Both top-down and bottom-up approaches to problem solving face major hurdles in converting theory to working technology. Top-down efforts must do a better job of converting high-level conceptualizations into working systems. Working systems often bear an indirect and non-exclusive relationship to the theories that motivated them. Bottom-up methods need the reverse – insuring that working systems accurately reflect coherent abstractions.
This article shows how Structural Learning Theory (SLT) generally and Abstract Syntax Trees (ASTs) specifically provide a rigorous, systematic way to bridge the gap. SLT puts the focus on higher order knowledge (SLT rules), a universal goal-switching control mechanism and a systematic method for structural (cognitive task) analysis for identifying higher (and lower) order SLT rules. ASTs make it possible to represent declarative (structural) and procedural knowledge simultaneously at ALL levels of expertise (i.e., knowledge).
The article concludes with implications of SLT and ASTs for research highlighted in this issue. A key question is the extent to which preanalysis of content domains would eliminate ambiguity and increase the precision with which one can predict behavior. Arbitrary irreducible relations, for example, can be refined systematically, and introducing higher order AST-based rules can dramatically reduce the number of nodes required in any given representation.