Infant categorization: memory for category-level and specific item information

In an attempt to determine whether it is necessary to postulate abstraction processes in infant categorization, three experiments assessed retention of category-level information and information specific to category members. Using a visual recognition memory procedure, 10- and 13-month-old infants were familiarized with category instances containing both shared dimensional information and idiosyncratic features. The addition of idiosyncratic features to members of the familiarization category enhanced specific item memory for 13-month-old infants. However, this was not the case for the younger infants. The results of within-category test comparisons indicated that 10-month-old infants regarded information common to all members of the category as more familiar than information specific to individual exemplars. This occurred despite evidence that specific item information was retained in memory and available for retrieval during recognition tests. The findings are discussed in relation to exemplar and abstraction models of categorization.     

Unsupervised categorization and category learning.

When people categorize a set of items in a certain way they often change their perceptions for these items so that they become more compatible with the learned categorization. In two experiments we examined whether such changes are extensive enough to change the unsupervised categorization for the items-that is, the categorization of the items that is considered more intuitive or natural without any learning. In Experiment 1 we directly employed an unsupervised categorization task; in Experiment 2 we collected similarity ratings for the items and inferred unsupervised categorizations using Pothos and Chater's (2002) model of unsupervised categorization. The unsupervised categorization for the items changed to resemble more the learned one when this was specified by the suppression of a stimulus dimension (both experiments), but less so when it was almost specified by the suppression of a stimulus dimension (Experiment 1, nonsignificant trend in Experiment 2). By contrast, no changes in the unsupervised categorization were observed when participants were taught a classification that was specified by a more fine tuning of the relative salience of the two dimensions.     

Generalization and similarity in exemplar models of categorization: Insights from machine learning

Exemplar theories of categorization depend on similarity for explaining subjects’ ability to generalize to new stimuli. A major criticism of exemplar theories concerns their lack of abstraction mechanisms and thus, seemingly, of generalization ability. Here, we use insights from machine learning to demonstrate that exemplar models can actually generalize very well. Kernel methods in machine learning are akin to exemplar models and are very successful in real-world applications. Their generalization performance depends crucially on the chosen similarity measure. Although similarity plays an important role in describing generalization behavior, it is not the only factor that controls generalization performance. In machine learning, kernel methods are often combined with regularization techniques in order to ensure good generalization. These same techniques are easily incorporated in exemplar models. We show that the generalized context model (Nosofsky, 1986) and ALCOVE (Kruschke, 1992) are closely related to a statistical model called kernel logistic regression. We argue that generalization is central to the enterprise of understanding categorization behavior, and we suggest some ways in which insights from machine learning can offer guidance.     

Rasch models for item bundles

This paper discusses the application of a class of Rasch models to situations where test items are grouped into subsets and the common attributes of items within these subsets brings into question the usual assumption of conditional independence. The models are all expressed as particular cases of the random coefficients multinomial logit model developed by Adams and Wilson. This formulation allows a very flexible approach to the specification of alternative models, and makes model testing particularly straightforward. The use of the models is illustrated using item bundles constructed in the framework of the SOLO taxonomy of Biggs and Collis.The work of both authors was supported by fellowships from the National Academy of Education Spencer Fellowship.     

Bayesian item fit analysis for unidimensional item response theory models

Assessing item fit for unidimensional item response theory models for dichotomous items has always been an issue of enormous interest, but there exists no unanimously agreed item fit diagnostic for these models, and hence there is room for further investigation of the area. This paper employs the posterior predictive model‐checking method, a popular Bayesian model‐checking tool, to examine item fit for the above‐mentioned models. An item fit plot, comparing the observed and predicted proportion‐correct scores of examinees with different raw scores, is suggested. This paper also suggests how to obtain posterior predictive p‐values (which are natural Bayesian p‐values) for the item fit statistics of Orlando and Thissen that summarize numerically the information in the above‐mentioned item fit plots. A number of simulation studies and a real data application demonstrate the effectiveness of the suggested item fit diagnostics. The suggested techniques seem to have adequate power and reasonable Type I error rate, and psychometricians will find them promising.     

Thirty-something categorization results explained: selective attention, eyetracking, and models of category learning

An eyetracking study testing D. L. Medin and M. M. Schaffer's (1978) 5-4 category structure was conducted. Over 30 studies have shown that the exemplar-based generalized context model (GCM) usually provides a better quantitative account of 5-4 learning data as compared with the prototype model. However, J. D. Smith and J. P. Minda (2000) argued that the GCM is a psychologically implausible account of 5-4 learning because it implies suboptimal attention weights. To test this claim, the authors recorded undergraduates' eye movements while the students learned the 5-4 category structure. Eye fixations matched the attention weights estimated by the GCM but not those of the prototype model. This result confirms that the GCM is a realistic model of the processes involved in learning the 5-4 structure and that learners do not always optimize attention, as commonly supposed. The conditions under which learners are likely to optimize attention during category learning are discussed.     

The effect of an extraneous added memory set on item recognition: A test of parallel-dependent vs. serial-comparison models

The linear increase in response time (RT) in item recognition as a function of memory set size may be attributed either to increased memory search, as in Sternberg’s serial-comparison model, or to increased interference and competition for a limited processing capacity, as in a parallel-dependent model. An additional memory load was imposed on the normal item recognition task. If the normal task involved letters, the additional task involved digits, or vice versa. The added set was presented before and tested after the normal core set was presented and tested. Across the four trial blocks, the subjects did come to dissociate the two memory sets, but this could be due as much to directed rehearsal as to directed search when other results, including the subjects’ postsession reports, are considered. Other findings, such as the higher error rate and shorter RT for positive than negative trials, seem to reflect differences in effective trace strength, and thus are consistent with the parallel-dependent model.     

Responding under time pressure: Testing two animal learning models and a model of visual categorization

Two experiments are reported, which employed a Pavlovian eyelid conditioning procedure with human participants. The experiments tested the predictions of three models of the time-course of processing under time pressure. These were the extended generalized context model (Lamberts, 1998), and two variants of the Rescorla-Wagner model (Rescorla & Wagner, 1972), which were activated in cascade mode. Reinforcement schedules in the experiments were equivalent either to an AND rule or to an XOR rule. The time available for processing the conditioned stimulus and initiating a conditioned response was manipulated by varying the interval from the onset of the conditioned stimulus to the onset of the unconditioned stimulus. The results were in accord with the predictions of one of the two variants of the Rescorla-Wagner model.     

Verbal learning: Incremental vs. all-or-none theory

The incremental and all-or-none theories of verbal learning are compared by means of a little-used but methodologically superior variation of the “drop-out” paradigm with paired associates. Earlier experiments purporting to be relevant to the controversy are rejected as failing to offer a conclusive distinction between the two theories. The results presented here are taken to support the incremental theory. It is suggested that irregularities in the results of this and other experiments are caused by several intrusive factors, and a “dual-factor” hypothesis which was put forward to account for these irregularities is questioned on logical and methodological grounds.     

Teaching by Examples: Implications for the Process of Category Acquisition

A new paradigm, the 'teaching-by-examples' paradigm, was used to shed new light on the process of category acquisition. In four experiments ( n = 90, 90, 115, 117), manipulating the variables of category structure, status of non-target category, learning mode, and teaching mode, participants first learned a category and then taught it to someone else. High agreement between participants on the teaching sequences was found across conditions, and a typical sequence was identified for each category structure. The typical participant-produced sequences startedwith several ideal positive cases, followed by anideal negative case and then borderline cases. The efficiency of such sequences for teaching was tested in another experiment ( n = 60), in which they were compared with sequences emphasizing category borders and sequences emphasizing each dimension separately. The typical participant-produced sequences induced the most efficient learning. It is proposed that the pattern of performance may provide a rich source of data for testing and fine-tuning models of category acquisition.     

Ontologies and Worlds in Category Theory: Implications for Neural Systems

We propose category theory, the mathematical theory of structure, as a vehicle for defining ontologies in an unambiguous language with analytical and constructive features. Specifically, we apply categorical logic and model theory, based upon viewing an ontology as a sub-category of a category of theories expressed in a formal logic. In addition to providing mathematical rigor, this approach has several advantages. It allows the incremental analysis of ontologies by basing them in an interconnected hierarchy of theories, with an operation on the hierarchy that expresses the formation of complex theories from simple theories that express first principles. Another operation forms abstractions expressing the shared concepts in an array of theories. The use of categorical model theory makes possible the incremental analysis of possible worlds, or instances, for the theories, and the mapping of instances of a theory to instances of its more abstract parts. We describe the theoretical approach by applying it to the semantics of neural networks.     

Category size effects revisited: frequency and masked priming effects in semantic categorization

Previous work indicates that semantic categorization decisions for nonexemplars (e.g., deciding that TURBAN is not an animal name) are faster for high-frequency words than low-frequency words. However, there is evidence that this result might depend on category size. When narrow categories are used (e.g., Months, Numbers), there is no frequency effect for nonexemplars. This result is confirmed, and is explained in terms of a category search model, which allows a "No" decision to be generated without access to the lexical entry for the target word. This explains the absence of a frequency effect, but not the presence of a strong masked repetition priming effect, which is assumed to have a lexical source. It is shown that this effect may not be lexical, since nonwords also show similar priming. Both of these priming effects disappear when a larger category is used. This pattern of results is explained on the assumption that category search is only possible with small categories, and that tentative category decisions are generated for the unconsciously perceived prime, which leads to a marked response congruence effect.     

Stochastic search item selection for factor analytic models

In this paper we implement a Markov chain Monte Carlo algorithm based on the stochastic search variable selection method of George and McCulloch (1993) for identifying promising subsets of manifest variables (items) for factor analysis models. The suggested algorithm is constructed by embedding in the usual factor analysis model a normal mixture prior for the model loadings with latent indicators used to identify not only which manifest variables should be included in the model but also how each manifest variable is associated with each factor. We further extend the suggested algorithm to allow for factor selection. We also develop a detailed procedure for the specification of the prior parameters values based on the practical significance of factor loadings using ideas from the original work of George and McCulloch (1993). A straightforward Gibbs sampler is used to simulate from the joint posterior distribution of all unknown parameters and the subset of variables with the highest posterior probability is selected. The proposed method is illustrated using real and simulated data sets.     

Minimax d-optimal designs for item response theory models

Various different item response theory (IRT) models can be used in educational and psychological measurement to analyze test data. One of the major drawbacks of these models is that efficient parameter estimation can only be achieved with very large data sets. Therefore, it is often worthwhile to search for designs of the test data that in some way will optimize the parameter estimates. The results from the statistical theory on optimal design can be applied for efficient estimation of the parameters.A major problem in finding an optimal design for IRT models is that the designs are only optimal for a given set of parameters, that is, they are locally optimal. Locally optimal designs can be constructed with a sequential design procedure. In this paper minimax designs are proposed for IRT models to overcome the problem of local optimality. Minimax designs are compared to sequentially constructed designs for the two parameter logistic model and the results show that minimax design can be nearly as efficient as sequentially constructed designs.     

When parameters collide: a warning about categorization models

Similarity-choice (S—C) models of categorization contain two principal mathematical transformations: an exponential-decay similarity function and a choice rule. However, there is a tension between the psychological processes that models emulate and the mathematics they use to do so. To illustrate this, I show that in these models an unappreciated interaction occurs between the mathematical transformations so that the stages of the model essentially cancel each other out. The result is that the model’s output reflects its input linearly. This cancellation phenomenon has potentially serious implications regarding the interpretation and use of S—C models. The phenomenon also raises questions about the simplification and psychological grounding of categorization models. Modelers broadly might benefit from an internal analysis of their models, such as that described here.     

Knowledge partitioning in categorization: constraints on exemplar models

The authors present 2 experiments that establish the presence of knowledge partitioning in perceptual categorization. Many participants learned to rely on a context cue, which did not predict category membership but identified partial boundaries, to gate independent partial categorization strategies. When participants partitioned their knowledge, a strategy used in 1 context was unaffected by knowledge demonstrably present in other contexts. An exemplar model, attentional learning covering map, was shown to be unable to accommodate knowledge partitioning. Instead, a mixture-of-experts model, attention to rules and instances in a unified model (ATRIUM), could handle the results. The success of ATRIUM resulted from its assumption that people memorize not only exemplars but also the way in which they are to be classified.     

Representational blending in human conditional learning: Implications for associative theory.

In two experiments, participants were presented with pictures of different foods (A, B, C, D, X,) and learned which combinations resulted in an allergic reaction in a fictitious patient, Mr X. In Problem 1, when A or B (but not C or D) was combined with food X an allergic reaction occurred, and when C or D (but not A or B) was combined with Y an allergic reaction occurred. In Experiment 1, participants also received Problem 2 in which A, B, C, and D interacted with foods V and W either in the same way as X and Y, respectively, or in a different way. Participants performed more proficiently in the former than in the latter condition. In Experiment 2, after training on Problem 1, participants judged whether or not novel combinations of foods (e.g., AB, CD, AD, CB) would cause an allergic reaction in Mr X. They were no more likely to indicate that AB or CD would cause an allergic reaction than AD or CB, but made their judgements more rapidly and with greater confidence on AB and CD trials than on AD and CB trials. These results (1) indicate that shared representations come to be addressed by the components of similar compounds (e.g., AX and BX) that have predicted the same outcome (an allergic reaction), and (2) are inconsistent with standard, associative theories of learning, but (3) are consistent with findings from nonhuman animals and with a connectionist interpretation of these findings.     

Fixed Effects vs. Random Effects Meta-Analysis Models: Implications for Cumulative Research Knowledge

Research conclusions in the social sciences are increasingly based on meta-analysis, making questions of the accuracy of meta-analysis critical to the integrity of the base of cumulative knowledge. Both fixed effects (FE) and random effects (RE) meta-analysis models have been used widely in published meta-analyses. This article shows that FE models typically manifest a substantial Type I bias in significance tests for mean effect sizes and for moderator variables (interactions), while RE models do not. Likewise, FE models, but not RE models, yield confidence intervals for mean effect sizes that are narrower than their nominal width, thereby overstating the degree of precision in meta-analysis findings. This article demonstrates analytically that these biases in FE procedures are large enough to create serious distortions in conclusions about cumulative knowledge in the research literature. We therefore recommend that RE methods routinely be employed in meta-analysis in preference to FE methods.