A dimensional summation account of polymorphous category learning

Abstract

Polymorphous concepts are hard to learn, and this is perhaps surprising because they, like many natural concepts, have an overall similarity structure. However, the dimensional summation hypothesis (Milton & Wills, 2004) predicts this difficulty. It also makes a number of other predictions about polymorphous concept formation, which are tested here. In Experiment 1 we confirm the theory's prediction that polymorphous concept formation should be facilitated by deterministic pretraining on the constituent features of the stimulus. This facilitation is relative to an equivalent amount of training on the polymorphous concept itself. In Experiments 2--4, the dimensional summation account of this single feature pretraining effect is contrasted with some other accounts, including a more general strategic account (Experiment 2), seriality of training and stimulus decomposition accounts (Experiment 3), and the role of errors (Experiment 4). The dimensional summation hypothesis provides the best account of these data. In Experiment 5, a further prediction is confirmed --- the single feature pretraining effect is eliminated by a concurrent counting task. The current experiments suggest the hypothesis that natural concepts might be acquired by the deliberate serial summation of evidence. This idea has testable implications for classroom learning.

A Cognitive Category-Learning Model of Rule Abstraction, Attention Learning, and Contextual Modulation

“…In category learning, similar views are referred to as 'Combination Theory', which views configural instance memories as result of more complex and cognitively more demanding brain processes, relative to learning simple rules (see Lamberts, 1995;Wills, Ellett, Milton, Croft, & Beesley, 2019;Wills et al, 2015). Consequently, similar but not identical to the error-based formation of clusters in SUSTAIN, (Love et al, 2004) we assume that configural memory is psychologically distinct from rules, in that storing instances is most effective only if CAL's rules are erroneous, such that correct prediction would require feature combination beyond contextual modulation.…”

“…22 This constrasts with 'information integration' (II) explanation of Type IV performance (Ashby et al, 1998); for critiques of the II explanation, see Edmunds et al (2018); Lewandowsky et al (2012); Wills, Ellett, et al (2019). rule exceptions in configural memory (λ), and that attentional control (AC) 24 predicts…”