Predicting category intuitiveness with the rational model, the simplicity model, and the generalized context model.

Abstract

This is the unspecified version of the paper.This version of the publication may differ from the final published version. Permanent repository link AbstractNaïve observers typically perceive some groupings for a set of stimuli as more intuitive than others. The problem of predicting category intuitiveness has been historically considered the remit of models of unsupervised categorization. In contrast, this paper develops a measure of category intuitiveness from one of the most widely supported models of supervised categorization, the Generalized Context Model (GCM). Considering different category assignments for a set of instances, we ask how well the GCM can predict the classification of each instance on the basis of all the other instances. The category assignment that results in the smallest prediction error is interpreted as the most intuitive for the GCM-we call this way of applying the GCM unsupervised GCM. The paper then systematically compares predictions of category intuitiveness from the unsupervised GCM and two models of unsupervised categorization, the simplicity model and the rational model. We found that the unsupervised GCM compares favorably to the simplicity model and rational model. This success of the unsupervised GCM illustrates that the distinction between supervised and unsupervised categorization may have to be reconsidered. However, no model emerges as clearly superior, indicating that there is more work to be done in understanding and modeling category intuitiveness.

On the adequacy of current empirical evaluations of formal models of categorization.

“…For example, Nosofsky and Palmeri (1996) (Nosofsky, Kruschke & McKinley, 1992;Nosofsky et al, 1994), prototype models (Smith & Minda, 2000), the Rational model (Nosofsky et al, 1994;Pothos & Bailey, 2009), and decision-bound models (McKinley & Nosofsky, 1996;Nosofsky & Palmeri, 1997b;Little, Nosofsky & Denton, 2011). And yet, limited progress appears to have been made in reducing the number of divergent, non-complementary models of categorization.…”

“…Wills, Lavric, Croft & Hodgson, 2007) is another example of a secondary DV, whilst measures of classification consistency (e.g. Haslam et al, 2007) or, relatedly, intuitiveness (Pothos & Bailey, 2009) are other examples of primary DVs on this definition. The question of whether reaction time measures (e.g.…”

“…Also, like SUSTAIN, the RMC is able to provide an account of both categorization and feature inference. One way in which attentional selection can be implemented in the RMC is in terms of prior biases for particular dimensions (Anderson, 1991; for an alternative approach see Pothos & Bailey, 2009).…”

“…Hence, the comparison was restricted not just to the same kind of evidence (classification probabilities) but effectively to variants of the same data set. Pothos and Bailey (2009) explored the ability of three different models (an unsupervised version of the GCM, the simplicity model, and the RMC) to account for five different data sets. While initially promising, as it turned out, none of the models were clearly superior across all five data sets, showing that a low ratio of data sets to models (5:3) was not adequate to discriminate between these models (equally, that the particular data sets were non-diagnostic in this comparison).…”

“…In categorization research (and in other areas of psychology) this is partly because informal theories make extensive use of verbal labels that denote intuitively obvious but computationally complex constructs. For example, in unsupervised classification (category formation in the absence of feedback), the category groupings selected most frequently by participants are often those that are most "intuitive" -but formally specifying what makes them so turns out to be quite complex (Pothos & Bailey, 2009). …”