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.     

Mechanisms underlying speech sound discrimination and categorization in humans and zebra finches

Speech sound categorization in birds seems in many ways comparable to that by humans, but it is unclear what mechanisms underlie such categorization. To examine this, we trained zebra finches and humans to discriminate two pairs of edited speech sounds that varied either along one dimension (vowel or speaker sex) or along two dimensions (vowel and speaker sex). Sounds could be memorized individually or categorized based on one dimension or by integrating or combining both dimensions. Once training was completed, we tested generalization to new speech sounds that were either more extreme, more ambiguous (i.e., close to the category boundary), or within-category intermediate between the trained sounds. Both humans and zebra finches learned the one-dimensional stimulus–response mappings faster than the two-dimensional mappings. Humans performed higher on the trained, extreme and within-category intermediate test-sounds than on the ambiguous ones. Some individual birds also did so, but most performed higher on the trained exemplars than on the extreme, within-category intermediate and ambiguous test-sounds. These results suggest that humans rely on rule learning to form categories and show poor performance when they cannot apply a rule. Birds rely mostly on exemplar-based memory with weak evidence for rule learning.     

The effects of intradimensional and extradimensional shifts on visual discrimination learning in humans and non-human primates

Human subjects and non-human primates (the common marmoset) were trained on a series of reversals of both a simple (stimuli varying along one dimension) and compound (stimuli varying along two different dimensions) visual discrimination, using computer-generated stimuli. They were then shifted to a third series of reversals using completely novel compound stimuli. Those humans and marmosets for which the previously relevant dimension remained relevant, following the shift (shapes to shapes or lines to lines; intradimensional shift) made fewer errors than those for which the previously irrelevant dimension became relevant (shapes to lines or lines to shapes; extradimensional shift). These findings suggest that both humans and marmosets can learn to attend to the specific attributes or dimensions of a stimulus and use this information in visual discrimination learning.     

Asymmetry of attentional set in rhesus monkeys learning colour and shape discriminations

We trained rhesus monkeys on six visual discrimination problems using stimuli that varied in both shape and colour. For one group of animals shape was always relevant in these six problems, and colour always irrelevant, and for the other animals vice versa. During these “intradimensional shifts” (ID) the problems were learned at equal rates by the two groups, shape-relevant and colour-relevant. We then trained three further problems in which the other dimension was now relevant (“extradimensional shifts”, ED). The animals showed slower learning when shifting from colour-relevant to shape-relevant, but not when shifting from shape-relevant to colour-relevant. These results show that monkeys' ability to selectively attend to a relevant stimulus dimension and to ignore an irrelevant dimension depends on the experimenter's choice of relevant and irrelevant dimensions.     

Asymmetry of attentional set in rhesus monkeys learning colour and shape discriminations

We trained rhesus monkeys on six visual discrimination problems using stimuli that varied in both shape and colour. For one group of animals shape was always relevant in these six problems, and colour always irrelevant, and for the other animals vice versa. During these "intradimensional shifts" (ID) the problems were learned at equal rates by the two groups, shape-relevant and colour-relevant. We then trained three further problems in which the other dimension was now relevant ("extradimensional shifts", ED). The animals showed slower learning when shifting from colour-relevant to shape-relevant, but not when shifting from shape-relevant to colour-relevant. These results show that monkeys' ability to selectively attend to a relevant stimulus dimension and to ignore an irrelevant dimension depends on the experimenter's choice of relevant and irrelevant dimensions.     

Discrimination of multidimensional geometrical figures by chickens: categorization and pattern-learning

Japanese bantam hens were trained to discriminate between geometrical figures varying along four integral dimensions. Only one dimension predicted food: selections of sharp-cornered figures were reinforced, while selections of rounded figures were not. In experiment 1, hens were subsequently trained to discriminate between nine figure pairs in a simultaneous discrimination task. Because single pairs contained multiple redundant cues, whereas the relevant dimension was obvious only across stimulus pairs, the results revealed effects of both generalization and reversal learning. Accordingly, learning speed was enhanced for later discriminations. Experiment 2 tested the hens’ transfer performance to unknown pairs, following experience of 9 or 18 figure pairs. Four of seven hens showed reliable transfer after experience with 9 figures, but only three showed transfer after experience with 18 figures, indicating lower transfer with higher number of stimulus pairs learned. In experiment 3, hens were trained to discriminate 27 figure pairs. Discrimination ratios further decreased and the groups of pairs differed significantly in their ratios of discrimination. Individual hens’ pecking behaviour was analysed in relation to each dimension of single figures and in relation to relative differences in the levels of dimensions between paired figures. Hens were shown to be oriented towards irrelevant information and more towards relational and configurational than elemental and dimensional aspects. The results are discussed in the biological context of individual recognition in chickens’ dominance hierarchies, in which we suppose that chickens identify individual flock mates by representation of their visual pattern rather than by single characteristics. Received: 18 September 1998 / Accepted after revision: 23 December 1998     

Why children learn color and size words so differently: evidence from adults' learning of artificial terms.

An adult simulation study examined why children's learning of color and size terms follow different developmental patterns, one in which word comprehension precedes success in nonlinguistic matching tasks versus one in which matching precedes word comprehension. In 4 experiments, adults learned artificial labels for values on novel dimensions. Training mimicked that characteristic for children learning either color words or size words. The results suggest that the learning trajectories arise from the different frames in which different dimensions are trained: Using a comparison (size-like) training regimen helps learners pick out the relevant dimension, and using a categorization (color-like) training regimen helps the learner correctly comprehend and produce dimension terms. The results indicate that the training regimen, not the meanings of the terms or the specific dimensions, determines the pattern of learning.     

Another fundamental social category? Spontaneous categorization of people who uphold or violate moral norms

Studies of social categorization have shown that humans robustly categorize others along dimensions of sex, age, kinship, and coalition. Drawing on a functional perspective, we predicted that humans should also categorize others along the dimension of morality (i.e., violating vs. upholding moral norms). In a study employing the memory confusion paradigm, university undergraduate participants (N = 39) were presented with same-sex targets randomly matched with sentences indicating violation or upholding of moral norms, and with sentences indicating competence or incompetence. Results showed that participants spontaneously categorized along morality but not along competence, suggesting that morality (beyond general valence) may be an important dimension of social categorization.     

Form follows function: Learning about function helps children learn about shape

Object functions help young children to organize new artifact categories. However, the scope of their influence is unknown. We explore whether functions highlight property dimensions that are relevant to artifact categories in general. Specifically, using a longitudinal training procedure, we assessed whether experience with functions highlights shape as important for categorization. Seventeen-month-olds were provided experience with novel categories of similarly shaped objects. In Study 1, the function group learned about the objects’ shape-based functions; a control group did not. In Study 2, 17-month-olds were trained on the same categories, but the objects’ shapes and functions were no longer causally related. Only the function group in Study 1 subsequently used shape reliably when categorizing novel objects. These results suggest that function is instrumental in establishing a ‘shape bias’ in early categorization and that it does so via conceptually based processes.     

Encoding Shape and Spatial Relations: The Role of Receptive Field Size in Coordinating Complementary Representations

An effective functional architecture facilitates interactions among subsystems that are often used together. Computer simulations showed that differences in receptive field sizes can promote such organization. When input was filtered through relatively small nonoverlapping receptive fields, artificial neural net-works learned to categorize shapes relatively quickly; in contrast, when input was filtered through relatively large overlapping receptive fields, networks learned to encode specific shape exemplars or metric spatial relations relatively quickly. Moreover, when the receptive field sizes were allowed to adapt during learning, networks developed smaller receptive fields when they were trained to categorize shapes or spatial relations, and developed larger receptive fields when they were trained to encode specific exemplars or metric distances. In addition, when pairs of networks were constrained to use input from the same type of receptive fields, networks learned a task faster when they were paired with networks that were trained to perform a compatible type of task. Finally, using a novel modular architecture, networks were not preassigned a task, but rather competed to perform the different tasks. Networks with small nonover-lapping receptive fields tended to win the competition for categorical tasks whereas networks with large overlapping receptive fields tended to win the competition for exemplar/metric tasks.     

Goal relevance and artificial grammar learning

This investigation used a newly developed artificial grammar learning (AGL) paradigm in which participants were exposed to sequences of stimuli that varied in two dimensions (colours and letters) that were superimposed on each other. Variation within each dimension was determined by a different grammar. The results of two studies strongly suggest that implicit learning in AGL depends on the goal relevance of the to-be-learned dimension. Specifically, when only one of the two stimulus dimensions was relevant for their task (Experiment 1) participants learned the structure underlying the relevant, but not that of the irrelevant dimension. However, when both dimensions were relevant, both structures were learned (Experiment 2). These findings suggest that implicit learning occurs only in dimensions to which we are attuned. Based on the present results and on those of Eitam, Hassin, and Schul (2008) we suggest that focusing on goal relevance may provide new insights into the mechanisms underlying implicit learning.     

On the relation between decision rules and perceptual representation in multidimensional perceptual categorization

This article examines the relation between changing categorization decision rules and the nature of the underlying perceptual representation. Observers completed a matching task that required them to adjust the length and orientation of a single line stimulus until they perceived it to "match" a second line stimulus (Alfonso-Reese, 1996, 1997). The same observers then completed four categorization tasks with the same stimuli. Data from the matching task were used to estimate a perceptual representation for each stimulus and observer. Three hypotheses regarding potential interactions between categorization decision rules and perceptual representation were examined. One assumed that there was no interaction between decision rules and perceptual representation. The second assumed that linear categorization rules affect the perceptual representation differently from nonlinear categorization rules. The third assumed that dimensional integration rules affected the perceptual representation differently from decision rules that require the observer to set a criterion along one stimulus dimension while ignoring the other; this is referred to as decisional selective attention. The results suggested that (1) the matching task perceptual representation provided a good account of the categorization data, (2) decisional selective attention affected the perceptual representation differently from decisional integration, and (3) decisional selective attention generally decreased the perceptual variability along the attended dimension.     

Prior experience with negative spectral correlations promotes information integration during auditory category learning

Complex sounds vary along a number of acoustic dimensions. These dimensions may exhibit correlations that are familiar to listeners due to their frequent occurrence in natural sounds—namely, speech. However, the precise mechanisms that enable the integration of these dimensions are not well understood. In this study, we examined the categorization of novel auditory stimuli that differed in the correlations of their acoustic dimensions, using decision bound theory. Decision bound theory assumes that stimuli are categorized on the basis of either a single dimension (rule based) or the combination of more than one dimension (information integration) and provides tools for assessing successful integration across multiple acoustic dimensions. In two experiments, we manipulated the stimulus distributions such that in Experiment 1, optimal categorization could be accomplished by either a rule-based or an information integration strategy, while in Experiment 2, optimal categorization was possible only by using an information integration strategy. In both experiments, the pattern of results demonstrated that unidimensional strategies were strongly preferred. Listeners focused on the acoustic dimension most closely related to pitch, suggesting that pitch-based categorization was given preference over timbre-based categorization. Importantly, in Experiment 2, listeners also relied on a two-dimensional information integration strategy, if there was immediate feedback. Furthermore, this strategy was used more often for distributions defined by a negative spectral correlation between stimulus dimensions, as compared with distributions with a positive correlation. These results suggest that prior experience with such correlations might shape short-term auditory category learning.     

Stages in concept learning. III

Abstract.— In a task-oriented concept learning situation with 30 children, 12–13 years of age, as subjects, the concept to be learned was made unfamiliar by using labels which had no bearing on the defining attributes. Under these conditions it was found that the categorization of the stimulus figures became more difficult and that identification of the defining attributes preceded an adequate categorization of the stimulus figures. But forcing the child into identification of the attributes before he started to categorize the stimulus figures did not result in better concept learning scores.     

Pigeons' categorization may be exclusively nonanalytic

Recent theoretical and empirical developments in human category learning have differentiated an analytic, rule-based system of category learning from a nonanalytic system that integrates information across stimulus dimensions. In the present study, the researchers applied this theoretical distinction to pigeons' category learning. Pigeons learned to categorize stimuli varying in the tilt and width of their internal striping. The matched category problems had either a unidimensional (rule-based) or multidimensional (information-integration) solution. Whereas humans and nonhuman primates strongly dimensionalize these stimuli and learn rule-based tasks far more quickly than information-integration tasks, pigeons learned the two tasks equally quickly to the same accuracy level. Pigeons may represent a cognitive system in which the commitment to dimensional analysis and category rules was not strongly made. Their performance could suggest the character of the ancestral vertebrate categorization system from which that of primates emerged.     

Naming and categorization in young children: IV: listener behavior training and transfer of function

Following pretraining with everyday objects, 14 children aged from 1 to 4 years were trained, for each of three pairs of different arbitrary wooden shapes (Set 1), to select one stimulus in response to the spoken word /zog/, and the other to /vek/. When given a test for the corresponding tacts ("zog" and "vek"), 10 children passed, showing that they had learned common names for the stimuli, and 4 failed. All children were trained to clap to one stimulus of Pair 1 and wave to the other. All those who named showed either transfer of the novel functions to the remaining two pairs of stimuli in Test 1, or novel function comprehension for all three pairs in Test 2, or both. Three of these children next participated in, and passed, category match-to-sample tests. In contrast, all 4 children who had learned only listener behavior failed both the category transfer and category match-to-sample tests. When 3 of them were next trained to name the stimuli, they passed the category transfer and (for the 2 subjects tested) category match-to-sample tests. Three children were next trained on the common listener relations with another set of arbitrary stimuli (Set 2); all succeeded on the tact and category tests with the Set 2 stimuli. Taken together with the findings from the other studies in the series, the present experiment shows that (a) common listener training also establishes the corresponding names in some but not all children, and (b) only children who learn common names categorize; all those who learn only listener behavior fail. This is good evidence in support of the naming account of categorization.     

A harbor seal can transfer the same/different concept to new stimulus dimensions

We investigated the formation of an abstract concept of same/different in a harbor seal by means of a two-item same/different task. Stimuli were presented on a TFT monitor. The subject was trained to respond according to whether two horizontally aligned white shapes presented on a black background were the same, or different from each other, by giving a no-go or go response. Training comprised of four stages. First, the same/different task was trained with two shapes forming two same problems (A–A and B–B) and two different problems (A–B and B–A). After the learning criterion was reached, training proceeded with new pairs of shapes. In the second experimental stage, every problem was presented just five times before new problems were introduced. We showed that training to criterion with just two shapes resulted in item-specific learning, whereas reducing the number of presentations to five per problem led to the formation of a same/different learning set as well as some restricted relational learning. Training with trial-unique problems in the third stage of this study resulted in the formation of an abstract concept of same/different which was indicated by a highly significant performance in transfer tests with 120 novel problems. Finally, extra-dimensional transfer of the concept was tested. The harbor seal showed a significantly correct performance on transfer tests with 30 unfamiliar pattern and 60 unfamiliar brightness same/different problems, thus demonstrating that the concept is not restricted to the shape dimension originally learned, but can be generalized across stimulus dimensions.     

Distinct mechanisms in visual category learning

The ways in which visual categories are learned, and in which well-established categories are represented and retrieved, are fundamental issues of cognitive neuroscience. Researchers have typically studied these issues separately, and the transition from the initial phase of category learning to expertise is poorly characterized. The acquisition of novel categories has been shown to depend on the striatum, hippocampus, and prefrontal cortex, whereas visual category expertise has been shown to involve changes in inferior temporal cortex. The goal of the present experiment is to understand the respective roles of these brain regions in the transition from initial learning to expertise when category judgments are being made. Subjects were explicitly trained, over 2 days, to classify realistic faces. Subjects then performed the categorization task during fMRI scanning, as well as a perceptual matching task, in order to characterize how brain regions respond to these faces when not explicitly categorizing them. We found that, during face categorization, face-selective inferotemporal cortex, lateral prefrontal cortex, and dorsal striatum are more responsive to faces near the category boundary, which are most difficult to categorize. In contrast, the hippocampus and left superior frontal sulcus responded most to faces farthest from the category boundary. These dissociable effects suggest that there are several distinct neural mechanisms involved in categorization, and provide a framework for understanding the contribution of each of these brain regions in categorization.     

特征的呈现方式对类别学习中规则和相似性知识获得的影响

采用原型变异任务,通过操纵规则和相似性特征的呈现通道,探讨特征的呈现方式对类别学习中规则和相似性知识获得的影响。结果发现,在听觉-视觉条件下习得规则的人数显著多于习得相似性的人数,而在视觉-听觉和视觉-视觉条件下则不存在显著差异;并且,三种条件下习得规则的正确率都高于习得相似性的正确率。这说明特征的呈现方式影响对规则和相似性特征的习得,在听觉通道呈现规则时被试更倾向基于规则分类。     

The nonindependence of stimulus properties in human category learning

Typically, models of category learning are verified through behavioral experiments with stimuli consisting of putatively independent dimensions such as shape, size, and color. The assumption of independence is critical in both the design of behavioral experiments and the development of models and theories of learning. Using the standard classification learning paradigm and a common stimulus set, the present work demonstrates that the assumption of independence is unwarranted. Systematic relations span stimulus dimensions and govern learning performance. For example, shape is not independent of size and color, because humans quantify size and color over shape when shape is relevant to the categorization. This quantification is reflected in natural language use (e.g., "blue triangle" as opposed to "triangle and blue"). In this example, color and size are predicates and shape is the argument. Across four experiments, the difficulty of mastering a classification rule can be predicted by the number of predicates that must be unbound in order to free rule-relevant stimulus dimensions.