Distributional expectations and the induction of category structure.

Abstract

Previous research on how categories are learned from observation of exemplars has largely ignored the possible role of prior expectations concerning how exemplars will be distributed. The experiments reported here explored this issue by presenting subjects with category-learning tasks in which the distributions of exemplars defining the categories were varied. In Experiments 1 and 2 the distributional form of a category was found to affect speed of learning. Learning was faster when a category's distribution was normal than when it was multimodal. Also, subjects in the early stages of learning a multimodal category responded as if it were unimodal. These results suggested that subjects enter category-learning tasks with expectations of unimodal, possibly normal, distributions of exemplars. Experiments 3 and 4 attempted to manipulate subjects' prior expectations by varying the distribution of exemplars in the first of two consecutive category-learning tasks. Learning a multimodal category was influenced by the shape of a previously learned distribution and was facilitated when the earlier distribution was either multimodal or skewed, rather than normal. These results are interpreted as support for a dual-process model of category learning that incorporates the effects of prior expectations concerning exemplar distributions.

Primacy or recency effects in forming inductive categories

“…This finding is consistent with the extensive literature on expectation biases in cognitive processing. Prior expectations bias people to detect correlations and learn categories that are consistent with those expectations (e.g., Crawford, Huttenlocher, & Hedges, 2006;Flannagan et al, 1986;Trolier & Hamilton, 1986). For instance, such a primacy effect may facilitate the maintenance of stereotypes even in the face of disconfirming evidence (Johnston, 1996).…”

“…A variety of experimental studies have addressed how people form inductive categories from experience with stimuli that vary along one or more feature dimensions (Duffy & Kitayama, 2007;Flannagan, Fried, & Holyoak, 1986;Fried & Holyoak, 1984;Huttenlocher et al, 2000;Posner & Keele, 1968). In many of the tasks employed in these studies, participants have learned a new category f by observing and making judgments about a series of category instances consisting of stimuli that varied along d some dimension, such as square grid patterns presented as works of abstract art (Fried & Holyoak, 1984), dot patterns of geometric figures that varied in degree of distortion (Posner & Keele, 1968), blood cells that varied in size (Duffy & Kitayama, 2007), or lines that varied in length .…”

Effects of category diversity on learning, memory, and generalization

“…Are there any effects of diversity above and beyond item similarities, and if so, are these effects the same across the stimulus space? Some studies have controlled or factored out item similarities (Cohen et al, 2001;Posner & Keele, 1968;Rips, 1989;Stewart & Chater, 2002; but see the discussion in E. E. Sloman, 1994, andJohansen, 2000), and some have examined generalization right across the stimulus space (Flannagan et al, 1986;Fried & Holyoak,1984;Homa & Vosburgh, 1976;Peterson et al, 1973). However, no study has done both.…”

“…This has been shown both for items near the boundary between categories (Cohen, Nosofsky, & Zaki, 2001;Fried & Holyoak, 1984;Rips, 1989; for examples of boundary items, see Figure 2, items A, I, and G) and for peripheral items on the outer fringes of previously seen exemplars (Flannagan, Fried & Holyoak, 1986;Fried & Holyoak, 1984;Homa & Vosburgh, 1976;Posner & Keele, 1968; for examples of peripheral items, see Figure 2, items D, E, and F). In contrast, category diversity has a negative effect on generalization in the vicinity of the prototype (Flannagan et al, 1986;Fried & Holyoak, 1984;Homa & Vosburgh, 1976). 1 A noteworthy exception to the general pattern was reported by Peterson et al (1973), who observed negative effects of category diversity on generalization across a wide range of distances from category prototypes.…”

Bayesian generic priors for causal learning.

“…The first psychological model of categorization based on the Bayesian framework, proposed by Fried and Holyoak (1984), simply assumed that perceptual categories are learned by updating the mean and variance of a multidimensional normal distribution. Flannagan, Fried, and Holyoak (1986) extended this notion by proposing that people have priors for the abstract form of the distributions of quantitative dimensions-priors that favor learning of categories with unimodal and symmetrical distributions. Flanagan et al demonstrated that learning a category that violated this distributional form was relatively difficult but facilitated subsequent learning of a second category (based on different perceptual dimensions) that also violated the unimodal and symmetrical prior.…”