Prototype and Exemplar Accounts of Category Learning and Attentional Allocation: A Reassessment.

In a recent article. J. P. Minda and J. D. Smith (2002; see record 2002-00620-002) argued that an exemplar model provided worse quantitative fits than an alternative prototype model to individual subject data from the classic D. L. Medin and M. M. Schaffer (1978) 5/4 categorization paradigm. In addition, they argued that the exemplar model achieved its fits by making untenable assumptions regarding how observers distribute their attention. In this article, we demonstrate that when the models are equated in terms of their response-rule flexibility, the exemplar model provides a substantially better account of the categorization data than does a prototype or mixed model. In addition, we point to shortcomings in the attention-allocation analyses conducted by J. P. Minda and J. D. Smith (2002). When these shortcomings are corrected, we find no evidence that challenges the attention-allocation assumptions of the exemplar model. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Exemplar representation without generalization? Comment on Smith and Minda's (2000) "Thirty categorization results in search of a model."

J. D. Smith and J. P. Minda (see record 1999-15928-001) conducted a meta-analysis of 30 data sets reported in the classification literature that involved use of the "5–4" category structure introduced by D. L. Medin and M. M. Schaffer (1978). The meta-analysis was aimed at investigating exemplar and elaborated prototype models of categorization. In this commentary, the author argues that the meta-analysis is misleading because it includes many data sets from experimental designs that are inappropriate for distinguishing the models. Often, the designs involved manipulations in which the actual 5–4 structure was not, in reality, tested, voiding the predictions of the models. The commentary also clarifies various aspects of the workings of the exemplar-based context model. Finally, concerns are raised that the all-or-none exemplar processes that form part of Smith and Minda's (see record 1999-15928-001) elaborated prototype models are implausible and lacking in generality. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Exemplars, Prototypes, and the Flexibility of Classification Models.

J. P. Minda and J. D. Smith (2001) showed that a prototype model outperforms an exemplar model, especially in larger categories or categories that contained more complex stimuli. R. M. Nosofsky and S. R. Zaki (2002) showed that an exemplar model with a response-scaling mechanism outperforms a prototype model. The authors of the current study investigated whether excessive model flexibility could explain these results. Using cross-validation, the authors demonstrated that both the prototype model and the exemplar model with a response-scaling mechanism suffered from overfilling in the linearly separable category structure. The results illustrate the need to make sure that the best-fitting model is not chasing error variance instead of variance attributed to the cognitive process it is supposed to model. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Exemplar and prototype models revisited: Response strategies, selective attention, and stimulus generalization.

J. D. Smith and colleagues (J. P. Minda & J. D. Smith, 2001; J. D. Smith & J. P. Minda, 1998, 2000; J. D. Smith, M. J. Murray, & J. P. Minda, 1997) presented evidence that they claimed challenged the predictions of exemplar models and that supported prototype models. In the authors' view, this evidence confounded the issue of the nature of the category representation with the type of response rule (probabilistic vs deterministic) that was used. Also, their designs did not test whether the prototype models correctly predicted generalization performance. The present work demonstrates that an exemplar model that includes a response-scaling mechanism provides a natural account of all of Smith et al's experimental results. Furthermore, the exemplar model predicts classification performance better than the prototype models when novel transfer stimuli are included in the experimental designs. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Selective attention and the formation of linear decision boundaries: Comment on McKinley and Nosofsky (1996).

S. C. McKinley and R. M. Nosofsky (see record 83-21756) compared a linear decision-bound model with the generalized context model (GCM) in their ability to account for categorization data from experiments that used integral- or separable-dimension stimuli and required selective attention or attention to both dimensions. McKinley and Nosofsky found support for the GCM and concluded that decision-bound theory needs to incorporate assumptions about selective attention. In this commentary it is argued that (a) unlike the GCM, decision-bound theory provides a framework for independently investigating perceptual and decisional forms of selective attention; (b) the effect of stimulus integrality on the form of the optimal decision bound is misinterpreted; (c) averaged data is biased against decision-bound theory and toward the GCM; (d) many a priori predictions of the GCM are violated empirically; and (e) exemplar theory has lost much of its initial theoretical structure. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

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

Na?ve 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 article 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, the authors asked 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—the authors refer to this way of applying the GCM as “unsupervised GCM.” The authors systematically compared predictions of category intuitiveness from the unsupervised GCM and two models of unsupervised categorization: the simplicity model and the rational model. The unsupervised GCM compared favorably with the simplicity model and the rational model. This success of the unsupervised GCM illustrates that the distinction between supervised and unsupervised categorization may need to be reconsidered. However, no model emerged as clearly superior, indicating that there is more work to be done in understanding and modeling category intuitiveness. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

ALCOVE: An exemplar-based connectionist model of category learning.

ALCOVE (attention learning covering map) is a connectionist model of category learning that incorporates an exemplar-based representation (D. L. Medin and M. M. Schaffer, 1978; R. M. Nosofsky, 1986) with error-driven learning (M. A. Gluck and G. H. Bower, 1988; D. E. Rumelhart et al, 1986). ALCOVE selectively attends to relevant stimulus dimensions, is sensitive to correlated dimensions, can account for a form of base-rate neglect, does not suffer catastrophic forgetting, and can exhibit 3-stage (U-shaped) learning of high-frequency exceptions to rules, whereas such effects are not easily accounted for by models using other combinations of representation and learning method. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Combining exemplar-based category representations and connectionist learning rules.

Adaptive network and exemplar-similarity models were compared on their ability to predict category learning and transfer data. An exemplar-based network (J. K. Kruschke, 1990, 1992) that combines key aspects of both modeling approaches was also tested. The exemplar-based network incorporates an exemplar-based category representation in which exemplars become associated to categories through the same error-driven, interactive learning rules that are assumed in standard adaptive networks. Exp 1, which partially replicated and extended the probabilistic classification learning paradigm of M. A. Gluck and G. H. Bower (1988), demonstrated the importance of an error-driven learning rule. Exp 2, which extended the classification learning paradigm of D. L. Medin and M. M. Schaffer (1978) that discriminated between exemplar and prototype models, demonstrated the importance of an exemplar-based category representation. Only the exemplar-based network accounted for all the major qualitative phenomena; it also achieved good quantitative predictions of the learning and transfer data in both experiments. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Learning nonlinearly separable categories by inference and classification.

Previous research suggests that learning categories by classifying new instances highlights information that is useful for discriminating between categories. In contrast, learning categories by making predictive inferences focuses learners on an abstract summary of each category (e.g., the prototype). To test this characterization of classification and inference teaming further, the authors evaluated the two learning procedures with nonlinearly separable categories. In contrast to previous research involving cohesive, linearly separable categories, the authors found that it is more difficult to learn nonlinearly separable categories by making inferences about features than it is to learn them by classifying instances. This finding reflects that the prototype of a nonlinearly separable category does not provide a good summary of the category members. The results from this study suggest that having a cohesive category structure is more important for inference than it is for classification. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Attention, similarity, and the identification–categorization relationship.

A unified quantitative approach to modeling Ss' identification and categorization of multidimensional perceptual stimuli is proposed and tested. Two Ss identified and categorized the same set of perceptually confusable stimuli varying on separable dimensions. The identification data were modeled using R. N. Shepard's (see record 1959-05134-001) multidimensional scaling-choice framework, which was then extended to model the Ss' categorization performance. The categorization model, which generalizes the context theory of classification developed by D. L. Medin and M. M. Schaffer (see record 1979-12633-001), assumes that Ss store category exemplars in memory. Classification decisions are based on the similarity of stimuli to the stored exemplars. It is assumed that the same multidimensional perceptual representation underlies performance in both the identification and categorization paradigms. However, because of the influence of selective attention, similarity relationships change systematically across the 2 paradigms. Findings provide some support for the hypothesis that Ss distribute attention among component dimensions so as to optimize categorization performance and that Ss may have augmented their category representations with inferred exemplars. Results demonstrate that excellent predictions of categorization performance can be made given knowledge of performance in an identification paradigm. (51 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The changing composition of abstracted categories under manipulations of decisional change, choice difficulty, and category size.

The role of decisional factors in category abstraction was investigated. The major prediction was that a change in instructional set would primarily affect the more difficult choices of a category (boundary contraction hypothesis), with this outcome modulated by category size. Subjects classified patterns into three prototype categories until they reached an errorless criterion; then they immediately took a transfer test under a conservative or liberal set. Category size was varied independently of category frequency in Experiment 1; in Experiment 2, category size functioned as a between-subjects variable. The results showed that instructional set affected new, but not old, instances, with this effect additive across choice difficulty and category size. The boundary contraction hypothesis was rejected, and a two-stage model of classification was proposed to account for the results. A compositional analysis revealed that far greater levels of learning may be needed before selective decision making can occur. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

"Prototypes in the mist: The early epochs of category learning": Correction to Smith and Minda (1998).

Reports an error in "Prototypes in the mist: The early epochs of category learning" by J. David Smith and John Paul Minda (Journal of Experimental Psychology: Learning, Memory, and Cognition, 1998[Nov], Vol 24[6], 1411-1436). As a result of errors made in production, two equations in the article were printed incorrectly. The corrected equations are included in the erratum. (The following abstract of the original article appeared in record 1998-12790-005.) Recent ideas about category learning have favored exemplar processes over prototype processes. However, research has focused on small, poorly differentiated categories and on task-final performances--both may highlight exemplar strategies. Thus, we evaluated participants' categorization strategies and standard categorization models at successive stages in the learning of smaller, less differentiated categories and larger, more differentiated categories. In the former case, the exemplar model dominated even early in learning. In the latter case, the prototype model had a strong early advantage that gave way slowly. Alternative models, and even the behavior of individual parameters within models, suggest a psychological transition from prototype-based to exemplar-based processing during category learning and show that different category structures produce different trajectories of learning through the larger space of strategies. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Component and pattern information in adaptive networks.

M. A. Gluck and G. H. Bower (see record 1989-00340-001) presented an adaptive network model of human classification in which associative weights are modified according to R. A. Rescorla and A. R. Wagner (1972) conditioning theory, a special case of the Widrow-Hoff/LMS learning rule. Presenting empirical data from a series of artificial medical classification tasks, we argued that the network model predicts results that were unanticipated, given prevailing alternative theories of category learning. We consider here some alternative interpretations of this data suggested by D. R. Shanks (see record 1990-27514-001) and argue that they are not sufficiently compelling when compared to the network model's treatment of the data from all the experiments presented in our earlier study. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Attentional processes in stereotype formation: A common model for category accentuation and illusory correlation.

Stereotype formation may be based on the exaggeration of real group differences (category accentuation) or the misperception of group differences that do not exist (illusory correlation). This research sought to account for both phenomena with J. K. Kruschke's (1996, 2001, 2003) attention theory of category learning. According to the model, the features of majority groups are learned earlier than the features of minority groups. In turn, the features that become associated with a minority are those that most distinguish it from the majority. This second process is driven by an attention-shifting mechanism that directs attention toward group-attribute pairings that facilitate differentiation of the two groups and may lead to the formation of stronger minority stereotypes. Five experiments supported this model as a common account for category accentuation and distinctiveness-based illusory correlation. Implications for the natures of stereotype formation, illusory correlation, and impression formation are discussed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Base-rate effects in category learning: A comparison of parallel network and memory storage-retrieval models.

Exemplar-memory and adaptive network models were compared in application to category learning data, with special attention to base rate effects on learning and transfer performance. Subjects classified symptom charts of hypothetical patients into disease categories, with informative feedback on learning trials and with the feedback either given or withheld on test trials that followed each fourth of the learning series. The network model proved notably accurate and uniformly superior to the exemplar model in accounting for the detailed course of learning; both the parallel, interactive aspect of the network model and its particular learning algorithm contribute to this superiority. During learning, subjects' performance reflected both category base rates and feature (symptom) probabilities in a nearly optimal manner, a result predicted by both models, though more accurately by the network model. However, under some test conditions, the data showed substantial base-rate neglect, in agreement with M. A. Gluck and G. H. Bower (see record 1989-00340-001). (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Prototypes in category learning: The effects of category size, category structure, and stimulus complexity.

Although research in categorization has sometimes been motivated by prototype theory, recent studies have favored exemplar theory. However, some of these studies focused on small, poorly differentiated categories composed of simple, 4-dimensional stimuli. Some analyzed the aggregate data of entire groups. Some compared powerful multiplicative exemplar models to less powerful additive prototype models. Here, comparable prototype and exemplar models were fit to individual-participant data in 4 experiments that sampled category sets varying in size, level of category structure, and stimulus complexity (dimensionality). The prototype model always fit the observed data better than the exemplar model did. Prototype-based processes seemed especially relevant when participants learned categories that were larger or contained more complex stimuli. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The costs of supervised classification: The effect of learning task on conceptual flexibility.

Research has shown that learning a concept via standard supervised classification leads to a focus on diagnostic features, whereas learning by inferring missing features promotes the acquisition of within-category information. Accordingly, we predicted that classification learning would produce a deficit in people's ability to draw novel contrasts—distinctions that were not part of training—compared with feature inference learning. Two experiments confirmed that classification learners were at a disadvantage at making novel distinctions. Eye movement data indicated that this conceptual inflexibility was due to (a) a narrower attention profile that reduces the encoding of many category features and (b) learned inattention that inhibits the reallocation of attention to newly relevant information. Implications of these costs of supervised classification learning for views of conceptual structure are discussed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Muscarinic cholinergic neuromodulation reduces proactive interference between stored odor memories during associative learning in rats.

Previous electrophysiological studies and computational modeling suggest the hypothesis that cholinergic neuromodulation may reduce olfactory associative interference during learning (M. E. Hasselmo et al [see record 1992-38176-001]; M. E. Hasselmo and J. M. Bower [see record 1993-40370-001]). These results provide behavioral evidence supporting this hypothesis. A simultaneous discrimination task required learning a baseline odor pair (A+B–) and then, under the influence of scopolamine, a novel odor pair (A–C+) with an overlapping component (A) versus a novel odor pair (D+E–) with no overlapping component. As predicted by the model, rats that received scopolamine (0.50 and 0.25 mg/kg) were more impaired at acquiring overlapping than nonoverlapping odor pairs relative to their performance under normal saline or methylscopolamine. These results support the prediction that the physiological effects of acetylcholine can reduce interference between stored odor memories during associative learning. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The fan effect: A tale of two theories.

This article addresses J. R. Anderson and L. M. Reder's (see record 1999-05245-004) account of the differential fan effect reported by G. A. Radvansky, D. H. Spieler, and R. T. Zacks (see record 1993-16287-001). The differential fan effect is the finding of greater interference with an increased number of associations under some conditions, but not others, in a within-subjects mixed-list recognition test. Anderson and Reder concluded that the differential fan effects can be adequately explained by assuming differences in the weights given to concepts in long-term memory. When a broader range of data is considered, this account is less well supported. Instead, it is better to assume that the organization of information into referential representations, such as situation models, has a meaningful influence on long-term memory retrieval. (PsycINFO Database Record (c) 2010 APA, all rights reserved)