Learning Perceptual Chunks for Problem Decomposition

How students learn to use diagrammatic representations is an important topic in the design of effective representations for problem solving or conceptual learning, but few good models of their learning exist. In this paper, we explore the learning process with an experiment using AVOW diagrams as a representation for solving problems in electric circuits. We find that the participants decompose each circuit into a similar set of groups when solving the problems. The natural question is whether these groups are an artifact of the visual form of the circuit, or indeed the result of prior learning. We argue that the decompositions are a result of perceptual chunking, and that they are (at least partly) a result of learning. In support of this, we describe a computational model of perceptual learning, CHREST+, and show that it predicts the decomposition of problems evident in the participants’ data.