Stimulus Generalization in Category Learning

摘要: Stimulus Generalization in Category Learning Matt Jones, W. Todd Maddox, and Bradley C. Love [mattj,maddox,love]@psy.utexas.edu University of Texas, Department Psychology, 1 Station A8000 Austin, TX 78712 USA representations the distinction between integral separable dimensions. We conclude by discussing broader applicability this new methodology as well its implications for nature perceptual category representations, attentional learning, roles short- long-term memory categorization. Abstract generalization is often regarded a fundamental component yet it has not been directly studied context. Here we develop technique measuring based on sequential effects subjects’ responses. find that patterns can adapt to global properties task, but only when structure defined perceptually primitive Implications are discussed learning both representations. Recency stimulus Perhaps most task facing brain use past experience determine useful behavior novel situations. For example, deciding whether particular snake poisonous, one might draw knowledge other, similar snakes whose toxicity was known. The details process be critical: Basing one’s response other color markings may effective, relying irrelevant such length could have disastrous consequences. In words, successful depends critically which variables relevant current prediction. One believed play an important role (Medin & Schaffer, 1978). However, contrast rich body data conditioning (see Shepard, 1987), investigated. Often assumed operates same these two tasks, functions empirically supported studies incorporated into similarity categorization models (Kruschke, 1992; Love, Medin, Gureckis, 2004; Nosofsky, 1986). richer involved (e.g., Maddox Ashby, 1993; Rosch et al., 1976; Sloman, Ahn, 1998) suggests domain far more complex than currently assumed. primary aim paper explore method assessing learning. technique, described detail below, close connection recency (Jones Sieck, 2003). present experiments designed validate approach relate previous findings Our results show good support illustrate how provide insight Proportion “A” responses Introduction robust phenomenon repeated judgment tasks. probability (repeated uncued forced-choice tasks), regularly found subjects biased select whichever reinforced trial Myers, 1970, review). Jones Sieck (2003) effect occurs cued categorization: Once identity controlled for, tend choose correct trial. This marginal from interpreted next, because reflects belief likely belong stimulus. Consistent with interpretation, magnitude stimuli, shown Figure 1. Stimuli were hypothetical medical patients varying presence or absence three symptoms. greatest successive stimuli identical decreased each cue mismatch, fully disappearing cases complete mismatch. approximately exponential decrease functional form commonly (Shepard, 1987). Previous A B Number M ismatching Cues 1: function number mismatching cues stimuli. (From 2003, Expt. 2, control condition.)