Dimension-Wide vs. Exemplar-Specific Attention in Category Learning and Recognition.

摘要: Items that violate a category rule are remembered better than items follow the rule. This finding cannot be predicted by exemplar models when all exemplars share same attention along dimension. With dimension-wide attention, violating and rule-following treated equally. When each selects which dimensions to attend to, can predict memory advantage for items. exemplar-specific is distributed uniformly encoding but allocated dimension of rulefollowing differential makes distinctive in memory. In addition need ability distinguish important errors from negligible ones stronger weaker Humans confronted with more information they process. Consequently, selectively salient fundamental our cognitive behavior. Many learning utilize at locations representational space (e.g., Kruschke, 1992; Love, Medin, & Gureckis, 2004; Nosofsky, 1986). The well suited many artificial studies, categories symmetric members differentiated values on dimensions. For example, A may large size dimension, whereas B small. natural categories, however, there inconsistent do not structure. penguins fly birds, bats belong mammals. Dimension-wide fare as contain members. Some laboratory work does suggest specific region Aha Goldstone, Barsalou 1986; Lewandowsky, Kalish, Ngang, 2002). different an item depending context in. humans color shopping clothing much computer. Exemplar have long history explaining key psychological phenomena research (Kruschke, Medin Schaffer, 1978; However, one problematic people remember structure rule) structure-consistent rule-following) (Palmeri 1995; Sakamoto press). By storing every studied separate trace using current treat consistent fashion this paper, we show model (see 2001 related “specificity”) differentiate While result non-rule We further exemplarspecific mechanism accentuates larger minimizes impact smaller remainder review previous examines recognition items, introduce models, present fits findings, discuss modeling results. Memory Advantage Exceptions Palmeri Nosofsky (1995) found exceptions their study, subjects learned classify geometric stimuli into two contrasting categories. An imperfect successfully classified majority study most small were A, B), violated was member A). Following learning, completed test consisting novel served foils. basic recognized special status also suggested schema Rojahn Pettigrew, 1992) studies. such ALCOVE (Medin 1978), exceptions. store trace. likelihood recognizing determined item’s absolute similarity both B. enhanced because relations other do. distinguished exceptions’ assignment runs counter According reversal germane recognition. contrast, rule-based model, RULEX (Nosofsky, Palmeri, McKinley, 1994), constructs rules stores rules. Rule-following individually stored rather captured Information about explicitly stored. exception store. storage allows entire pattern Nosofsky’s exceptions, over predict, encode very little Thus, combined Erikson 1998 similar approach involving knowledge gating). able data. inability suggests just represent support idea, Allen Brooks (1991) even applied item, still somewhat sensitive between previously encountered examples. line argues systems alternative clustering SUSTAIN (Love, 2004). has aspects violation introducing will allow those violation. Exemplar-Specific Attention 1992), treats manner (ES-ALCOVE) should ESALCOVE shifts To correct category, distribute so opposing category. receive ES-ALCOVE could “differentiated” simulate results intuition. next section, formalize ES-ALCOVE.