Unsupervised category learning with integral-dimension stimuli

Despite the recent surge in research on unsupervised category learning, the majority of studies have focused on unconstrained tasks in which no instructions are provided about the underlying category structure. Relatively little research has focused on constrained tasks in which the goal is to learn predefined stimulus clusters in the absence of feedback. The few studies that have addressed this issue have focused almost exclusively on stimuli for which it is relatively easy to attend selectively to the component dimensions (i.e., separable dimensions). In the present study, we investigated the ability of participants to learn categories constructed from stimuli for which it is difficult, if not impossible, to attend selectively to the component dimensions (i.e., integral dimensions). The experiments demonstrate that individuals are capable of learning categories constructed from the integral dimensions of brightness and saturation, but this ability is generally limited to category structures requiring selective attention to brightness. As might be expected with integral dimensions, participants were often able to integrate brightness and saturation information in the absence of feedback—an ability not observed in previous studies with separable dimensions. Even so, there was a bias to weight brightness more heavily than saturation in the categorization process, suggesting a weak form of selective attention to brightness. These data present an important challenge for the development of models of unsupervised category learning.     

Unsupervised category learning with integral-dimension stimuli

Despite the recent surge in research on unsupervised category learning, the majority of studies have focused on unconstrained tasks in which no instructions are provided about the underlying category structure. Relatively little research has focused on constrained tasks in which the goal is to learn predefined stimulus clusters in the absence of feedback. The few studies that have addressed this issue have focused almost exclusively on stimuli for which it is relatively easy to attend selectively to the component dimensions (i.e., separable dimensions). In the present study, we investigated the ability of participants to learn categories constructed from stimuli for which it is difficult, if not impossible, to attend selectively to the component dimensions (i.e., integral dimensions). The experiments demonstrate that individuals are capable of learning categories constructed from the integral dimensions of brightness and saturation, but this ability is generally limited to category structures requiring selective attention to brightness. As might be expected with integral dimensions, participants were often able to integrate brightness and saturation information in the absence of feedback--an ability not observed in previous studies with separable dimensions. Even so, there was a bias to weight brightness more heavily than saturation in the categorization process, suggesting a weak form of selective attention to brightness. These data present an important challenge for the development of models of unsupervised category learning.     

非监控类别学习的分类策略与表征

类别学习是人类对不同类别加以归类的过程。类别信息的表征、分类策略运用的特点一直是类别学习研究的重点。非监控类别学习可分为直接的非监控类别学习和间接的非监控类别学习。直接的非监控类别学习(非限制任务, 限制任务)中被试的分类策略具有分类“单维度倾向”策略特点,类别变异程度会影响类别表征; 间接的非监控类别学习更倾向形成相似性表征, 直接的非监控类别学习则为基于规则表征。现有的非监控类别学习的理论对分类策略和表征的解释仍显薄弱, 不同学习任务下类别迁移和知识效应的研究还存在不足, 未来研究还需要进一步验证知识效应对非监控类别学习的认知加工过程的影响、探索影响类别表征形成的因素等问题。     

Sequence-sensitive exemplar and decision-bound accounts of speeded-classification performance in a modified Garner-tasks paradigm

Among the most fundamental results in the area of perceptual classification are the “correlated facilitation” and “filtering interference” effects observed in Garner’s (1974) speeded categorization tasks: In the case of integral-dimension stimuli, relative to a control task, single-dimension classification is faster when there is correlated variation along a second dimension, but slower when there is orthogonal variation that cannot be filtered out (e.g., by attention). These fundamental effects may result from participants’ use of a trial-by-trial bypass strategy in the control and correlated tasks: The observer changes the previous category response whenever the stimulus changes, and maintains responses if the stimulus repeats. Here we conduct modified versions of the Garner tasks that eliminate the availability of a pure bypass strategy. The fundamental facilitation and interference effects remain, but are still largely explainable in terms of pronounced sequential effects in all tasks. We develop sequence-sensitive versions of exemplar-retrieval and decision-bound models aimed at capturing the detailed, trial-by-trial response-time distribution data. The models combine assumptions involving: (i) strengthened perceptual/memory representations of stimuli that repeat across consecutive trials, and (ii) a bias to change category responses on trials in which the stimulus changes. These models can predict our observed effects and provide a more complete account of the underlying bases of performance in our modified Garner tasks.     

The effect of correlated linguistic dimensions on speeded classification of visually presented trigrams

The influence of two types of linguistic dimensions, word-nonword and consonant pronunciation, on classification speed of trigrams in card-sorting tasks of two levels of complexity was examined. In complex grouping tasks, which required the evaluation of more than one letter to classify each stimulus, sorting times were faster when the linguistic dimension was correlated with, rather than orthogonal to, the response categories. For tasks in which each stimulus could be classified on the basis of a single letter, no effect of the correlated vs orthogonal linguistic dimension was observed, even when performance was degraded by visual noise. These results provide further evidence that, while linguistic properties of visual stimuli may influence classification time in complex tasks, they are of little importance in the performance of tasks only requiring the discriminalion of a single visual feature.     

Range effects of an irrelevant dimension on classification

In univariate classification tasks, subjects sort stimuli on the basis of the only attribute that varies. In orthogonal classification tasks, often called filtering tasks, there additionally are trial-to-trial variations in irrelevant attributes that the subjects are instructed to ignore. Performance is generally slower in filtering tasks than in univariate control tasks. We investigated this slowing in experiments of how the range of irrelevant trial-to-trial variation affects responses in pitch/loudness classification tasks. Using two levels of pitch and of loudness as stimuli, Experiment 1 replicated prior work showing that responses are slowed more when the range of the irrelevant dimension is made larger. Also in Experiment 1, sequential analyses showed that response time depends both on sequence and on the stimulus set independent of sequence. Experiments 2 and 3 used several levels on the irrelevant dimension and showed that responses to categorize loudness are slowed more by larger trial-to-trial pitch differences, but only on trials when the response repeats. When the response changes, performance is essentially unaffected by trial-to-trial irrelevant variation. This interaction supports the conclusion that slowed average performance in orthogonal classification tasks, which is known as Garner interference, is not due to difficulties that subjects have in filtering stimulus attributes. It is due to how subjects process successive stimulus differences. We call for more frequent reports of sequential analyses, because these can reveal information that is not available from data averages.     

Consistent contrast and correlation in free sorting

Two experiments investigated free sorting, a type of unsupervised learning, with multiattribute drawings of alien animals. In previous research on concept formation, with simpler stimulus structure than ours, participants were insensitive to correlational structure in the stimuli, producing primarily "ID sorts," based on the values ofjust 1 dimension or attribute. Our experiments showed that participants used many strategies in categorizing but preferred to generate groupings that reflected the correlations in input when this did not violate consistent contrast. The second experiment used hierarchically structured stimuli to show that participants' sort strategies favor consistency within a set of contrasting categories, distinct from any preference for 1D sorting. Finally, both experiments show that correlational sorts are much more likely when the correlation-based sort contrasts consistently. Our data show complexity at work in free sort tasks: People are sensitive to multiple and sometimes conflicting biases for consistency and correlational structure in the category systems they create.     

Limitations in advance task preparation: Switching the relevant stimulus dimension in speeded same—different comparisons

When participants switch between relevant stimulus dimensions in speeded classification tasks, taskswitching cost is reduced by advance preparation. Previous studies in which speeded classification tasks were used have suggested that this effect results from attending to the relevant stimulus dimension. Because selective attention to the relevant stimulus dimension in same—different judgments is relatively poor (e.g., Santee &; Egeth, 1980), it was predicted that advance task preparation for a shift in the relevant stimulus dimension would be compromised. This prediction was borne out in two experiments comparing dimension shifts (shape vs.fill) with task rule shifts (same? vs.different?) and shifts in the mapping of right—left keys toyes and_to responses (yes—no vs.no—yes). The results indicate that advance attentional selection of the relevant dimension is an optional preparatory strategy in task switching, employed only in conditions enabling flexible refocusing of attention.     

On the generality of the affective Simon effect

In affective Simon studies, participants are to select between a positive and negative response on the basis of a nonaffective stimulus feature (i.e., relevant stimulus feature) while ignoring the valence of the presented stimuli (i.e., irrelevant stimulus feature). De Houwer and Eelen (1998) showed that the time to select the correct response is influenced by the match between the valence of the response and the (irrelevant) valence of the stimulus. In the affective Simon studies that have been reported until now, only words were used as stimuli and the relevant stimulus feature was always the grammatical category of the words. We report four experiments in which we examined the generality of the affective Simon effect. Significant affective Simon effects were found when the semantic category, grammatical category, and letter-case of words was relevant, when the semantic category of photographed objects was relevant, and when participants were asked to give nonverbal approach or avoidance responses on the basis of the grammatical category of words. Results also showed that the magnitude of the affective Simon effect depended on the nature of the relevant feature.     

Effects of category learning on the stimulus selectivity of macaque inferior temporal neurons

Primates can learn to categorize complex shapes, but as yet it is unclear how this categorization learning affects the representation of shape in visual cortex. Previous studies that have examined the effect of categorization learning on shape representation in the macaque inferior temporal (IT) cortex have produced diverse and conflicting results that are difficult to interpret owing to inadequacies in design. The present study overcomes these issues by recording IT responses before and after categorization learning. We used parameterized shapes that varied along two shape dimensions. Monkeys were extensively trained to categorize the shapes along one of the two dimensions. Unlike previous studies, our paradigm counterbalanced the relevant categorization dimension across animals. We found that categorization learning increased selectivity specifically for the category-relevant stimulus dimension (i.e., an expanded representation of the trained dimension), and that the ratio of within-category response similarities to between-category response similarities increased for the relevant dimension (i.e., category tuning). These small effects were only evident when the learned category-related effects were disentangled from the prelearned stimulus selectivity. These results suggest that shape-categorization learning can induce minor category-related changes in the shape tuning of IT neurons in adults, suggesting that learned, category-related changes in neuronal response mainly occur downstream from IT.     

The effects of positive versus negative feedback on information-integration category learning

A number of studies have shown that in category learning, providing feedback about errors allows faster learning than providing feedback about correct responses. However, these previous studies used explicit, rule-based tasks in which the category structures could be separated by a simple rule that was easily verbalized. Here, the results of the first experiment known to compare the efficacy of positive versus negative feedback during information-integration category learning are reported. Information-integration tasks require participants to integrate perceptual information from incommensurable dimensions, and evidence suggests that optimal responding recruits procedural learning. The results show that although nearly all of the full-feedback control participants demonstrated information-integration learning, participants receiving either positive-only or negative-only feedback generally used explicit, rule-based strategies. It thus appears that, unlike rule-based learning, consistent information-integration learning requires full feedback. The theoretical implications of these findings for current models of information-integration learning are discussed.     

Selective attention in the speeded classification and comparison of multidimensional stimuli

Selective attention was studied when subjects were required to make either speeded classifications of single stimuli or comparisons of pairs of multidimensional stimuli. Experiment 1 established that subjects were able to attend selectively to form when the irrelevant dimension of size or shading varied in a speeded-classification (card-sorting) task. Experiment 2 confirmed this finding in a discrete-trials task. However, subjects werenot able to filter out irrelevant dimensional disparity in a comparably designed simultaneous-comparison (i.e., “same”-“different”) task. Mean “same” reaction time increased monotonically with increases in disparity between the two stimuli on the irrelevant dimension. Experiment 3 also revealed a monotonie increase in “same” RT as a function of irrelevant disparity in a successive-comparison task. These results were discussed in terms of a normalization model proposed by Dixon and Just (1978) in which it is assumed that a subject equates the two stimuli on the irrelevant dimension before deciding that they are the same along the relevant dimension. It was concluded that: (1) although subjects can efficiently filter out irrelevant disparity in a speeded-classification task, interference due to irrelevant disparity is obtained in the comparison tasks, (2) a common process such as normalization does not necessarily underlie performance in the speeded-classification and comparison tasks, (3) the ability to attend selectively to a stimulus dimension may be task determined as well as stimulus determined, and (4) contrary to the Dixon and Just proposal, normalization of irrelevant disparity occurs in a comparison task, even when the relevant dimension is represented as a separate encoding feature.     

Sequential analysis of the numerical Stroop effect reveals response suppression

Automatic processing of irrelevant stimulus dimensions has been demonstrated in a variety of tasks. Previous studies have shown that conflict between relevant and irrelevant dimensions can be reduced when a feature of the irrelevant dimension is repeated. The specific level at which the automatic process is suppressed (e.g., perceptual repetition, response repetition), however, is less understood. In the current experiment we used the numerical Stroop paradigm, in which the processing of irrelevant numerical values of 2 digits interferes with the processing of their physical size, to pinpoint the precise level of the suppression. Using a sequential analysis, we dissociated perceptual repetition from response repetition of the relevant and irrelevant dimension. Our analyses of reaction times, error rates, and diffusion modeling revealed that the congruity effect is significantly reduced or even absent when the response sequence of the irrelevant dimension, rather than the numerical value or the physical size, is repeated. These results suggest that automatic activation of the irrelevant dimension is suppressed at the response level. The current results shed light on the level of interaction between numerical magnitude and physical size as well as the effect of variability of responses and stimuli on automatic processing.     

Modeling graded response congruency effects in task switching

Compound cue retrieval is a computational model of a mediated route for response selection in task-switching situations. In previous studies, the model has been shown to account for response congruency effects when switching between two tasks, where response congruency reflects the degree of match between relevant and irrelevant task responses associated with a target stimulus. In the present study, the author derived a model prediction of graded response congruency effects in situations involving three tasks. The predicted pattern was observed for both response time and error rate in an experiment in which numerical categorization tasks were performed on single-digit targets. Implications for understanding response congruency effects and for developing models of task-switching performance are discussed.     

Supervised versus unsupervised categorization: two sides of the same coin?

Supervised and unsupervised categorization have been studied in separate research traditions. A handful of studies have attempted to explore a possible convergence between the two. The present research builds on these studies, by comparing the unsupervised categorization results of Pothos et al. ( 2011 ; Pothos et al., 2008 ) with the results from two procedures of supervised categorization. In two experiments, we tested 375 participants with nine different stimulus sets and examined the relation between ease of learning of a classification, memory for a classification, and spontaneous preference for a classification. After taking into account the role of the number of category labels (clusters) in supervised learning, we found the three variables to be closely associated with each other. Our results provide encouragement for researchers seeking unified theoretical explanations for supervised and unsupervised categorization, but raise a range of challenging theoretical questions.     

Classification of exemplars with single- and multiple-feature manifestations: the effects of relevant dimension variation and category structure

Most classification research focuses on cases in which each abstract feature has the same surface manifestation whenever it is presented. Previous research finds that people have difficulty learning to classify when each abstract feature has multiple surface manifestations. These studies created multiple manifestations by varying aspects of the stimuli irrelevant to the abstract feature dimension. In this article, multiple manifestations were created by varying aspects of the stimuli relevant to the abstract feature dimension. People given categories with the family resemblance category structure often used in psychology experiments had difficulty learning to classify when multiple manifestations were present, even though the variation was relevant. This effect was reversed when a family resemblance structure with nondiagnostic values was used.     

Response repetition benefits and costs

In a wide variety of tasks, choice reaction time (RT) is reduced for repetitions of the previous response. However, when the task itself or a relevant physical feature that successive trials have in common changes, costs for response repetitions can be observed. In a series of three experiments it was investigated whether the repetition of a response results in costs if the stimulus category changes. Furthermore, it was asked whether there need to be informative physical task features that successive trials have in common to produce response repetition costs. In alternating runs, participants had to respond to either one of four symbols or one of the letters with a binary choice reaction: Results suggest that a change of stimulus category is a sufficient condition to produce response repetition costs. It is hypothesized that any change of a task feature that is part of the task representation participants adopt leads to a disruption of repetition-based facilitation and tends to facilitate a response alternation.     

Switching of response modalities

When participants perform a sequence of different tasks, it is assumed that the engagement in one task leads to the inhibition of the previous task. This inhibition persists and impairs performance when participants switch back to this (still inhibited) task after only one intermediate trial. Previous task-switching studies on this issue have defined different tasks at the level of stimulus categorization. In our experiments we used different response modalities to define tasks. Participants always used the same stimulus categorization (e.g., categorize a digit as odd vs. even), but had to give a vocal, finger, or foot response (A, B, or C). Our results showed a higher reaction time and error rate in ABA sequences than in CBA sequences, indicating n - 2 repetition cost as a marker for persisting task inhibition. We assume that different response modalities can define a task and are inhibited in a “task switch” in the same way as stimulus categories are inhibited.     

Between-task transfer of learning from spatial compatibility to a color stroop task

Responses to a relevant stimulus dimension are faster and more accurate when the stimulus and response spatially correspond compared to when they do not, even though stimulus position is irrelevant (Simon effect). It has been demonstrated that practicing with an incompatible spatial stimulus-response (S-R) mapping before performing a Simon task can eliminate this effect. In the present study we assessed whether a learned spatially incompatible S-R mapping can be transferred to a nonspatial conflict task, hence supporting the view that transfer effects are due to acquisition of a general "respond to the opposite stimulus value" rule. To this aim, we ran two experiments in which participants performed a spatial compatibility task with either a compatible or an incompatible mapping and then transferred, after a 5 min delay, to a color Stroop task. In Experiment 1, responses were executed by pressing one of two keys on the keyboard in both practice and transfer tasks. In Experiment 2, responses were manual in the practice task and vocal in the transfer task. The spatially incompatible practice significantly reduced the color Stroop effect only when responses were manual in both tasks. These results suggest that during practice participants develop a response-selection strategy of emitting the alternative spatial response.     

Anterior prefrontal cortex contributions to attention control

In a series of event-related functional magnetic resonance studies, we consistently found activation in anterior prefrontal cortex related to visual dimension changes in singleton search tasks. I review these data and discuss possible contributions of anterior prefrontal cortex to attention control in visual search. It is proposed that anterior prefrontal cortex may detect task-relevant stimulus changes when the target is ambiguously defined. This process may occur in the absence of awareness and may support visual dimension weighting by inhibition of the old relevant dimension in favor of the new dimension.