Neural Correlates of Integral and Separable Processing During Category Learning Catherine Hanson Stephen José Hanson Tara Schweighardt Rutgers University, Newark, NJ Introduction Categorization is a fundamental cognitive function imposing order on stimuli; whether objects or actions, sights or sounds. Crucial to the ability to categorize is the ability to selectively attend to the relevant attributes of the tobecategorized input. Garner (1974) used the term perceptual structure to describe the way dimensions of a stimulus combine perceptually. Factors such as sensory constraints, development, or learning can affect whether attributes are integrally related, i.e., correlated, or separably related, i.e., seen as distinct dimensions. Attributes that are integrally related form a single percept. Separable attributes remain distinct and the percept is of a collection of attributes. Developmentally, integral processing has been shown to proceed the onset of separable processing (Smith, 1989) and adults can be biased smoothly towards either type of processing depending on category structure. In addition to the properties of the input, it is important to consider factors that may influence how the observer uses those properties during category learning. This study has two goals. First, to explore how strategy or attentional set interacts with stimulus dimensions of face stimuli during categorization. Second, to examine the neural correlates associated with processing categories that mediate the seamless switching between these two types of attentional processing.
Method Scanning was done on a Siemens Allegra 3T system. Stimuli were rear projected onto a mirror in the head cage. Responses were recorded with an MRI compatible button box. Subjects were normal, healthy adults.
that covaried (integral task). Subjects categorized the entire set of faces three times; i.e., there were three categorization blocks. 0
0.1
0.2
0.3
Eye Separation 0.4 0.5
0.6
0.7
0.8
0.9
1
0
0.1
0.2
0.3
Eye Separation 0.4 0.5
0.6
0.7
0.8
0.9
1
0.1 es.3,ml.1 es.4,ml.1 es.5,ml.1 es.6,ml.1 es.7,ml.1 es.8,ml.1 es.9,ml.1 es1,ml.1
0.1 es.1,ml.1 es.2,ml.1 es.3,ml.1 es.4,ml.1
es.7,ml.1 es.8,ml.1 es.9,ml.1 es1,ml.1
es.4,ml.2 es.5,ml.2 es.6,ml.2 es.7,ml.2 es.8,ml.2 es.9,ml.2 es1,ml.2
0.2 es.1,ml.2 es.2,ml.2 es.3,ml.2 es.4,ml.2
es.7,ml.2 es.8,ml.2 es.9,ml.2 es1,ml.2
es.5,ml.3 es.6,ml.3 es.7,ml.3 es.8,ml.3 es.9,ml.3 es1,ml.3
0.3 es.1,ml.3 es.2,ml.3 es.3,ml.3 es.4,ml.3
es.7,ml.3 es.8,ml.3 es.9,ml.3 es1,ml.3
0.4 es.1,ml.4 es.2,ml.4 es.3,ml.4 es.4,ml.4
es.7,ml.4 es.8,ml.4 es.9,ml.4 es1,ml.4
0.5 es.1,ml.5 es.2,ml.5 es.3,ml.5 es.4,ml.5
es.7,ml.5 es.8,ml.5 es.9,ml.5 es1,ml.5
0.6 es.1,ml.6 es.2,ml.6 es.3,ml.6 es.4,ml.6
es.7,ml.6 es.8,ml.6 es.9,ml.6 es1,ml.6
0.7 es.1,ml.7 es.2,ml.7 es.3,ml.7 es.4,ml.7
es.7,ml.7 es.8,ml.7 es.9,ml.7 es1,ml.7
0.8 es.1,ml.8 es.2,ml.8 es.3,ml.8 es.4,ml.8
es.7,ml.8 es.8,ml.8 es.9,ml.8 es1,ml.8
0.9 es.1,ml.9 es.2,ml.9 es.3,ml.9 es.4,ml.9
es.7,ml.9 es.8,ml.9 es.9,ml.9 es1,ml.9
1 es.1,ml1 es.2,ml1 es.3,ml1 es.4,ml1
es.7,ml1 es.8,ml1 es.9,ml1 es1,ml1
0.2
0.3 es.1,ml.3
M o u t h
0.4
M o u t h
es.1,ml.4 es.2,ml.4
es.6,ml.4 es.7,ml.4 es.8,ml.4 es.9,ml.4 es1,ml.4
0.5 es.1,ml.5 es.2,ml.5 es.3,ml.5
L o c a t i o n
L o c a t i o n
es.7,ml.5 es.8,ml.5 es.9,ml.5 es1,ml.5
0.6 es.1,ml.6 es.2,ml.6 es.3,ml.6 es.4,ml.6
es.8,ml.6 es.9,ml.6 es1,ml.6
0.7 es.1,ml.7 es.2,ml.7 es.3,ml.7 es.4,ml.7 es.5,ml.7
es.9,ml.7 es1,ml.7
0.8 es.1,ml.8 es.2,ml.8 es.3,ml.8 es.4,ml.8 es.5,ml.8 es.6,ml.8
es1,ml.8
0.9 es.1,ml.9 es.2,ml.9 es.3,ml.9 es.4,ml.9 es.5,ml.9 es.6,ml.9 es.7,ml.9 1 es.1,ml1 es.2,ml1 es.3,ml1 es.4,ml1 es.5,ml1 es.6,ml1 es.7,ml1 es.8,ml1
Filtration Task (separablel)
Condensation Task (integral)
Results Behavioral Data Forty subjects performed the judgement and categorization tasks outside of the magnet. An analysis of the correct response times during categorizataion yielded a significant main effect for Categorization Task, F(1, 68) = 7.43, p <.01, and a significant inter action between Judgement Task and Categorization Task, F(1,68) = 4.33, p <.05. An analysis on the mean accuracy of responses dur ing the categorization task yielded a similar pattern of results, i.e., a main effect for Categorization Task, F(1,68) = 46.24, p <.001. and an interaction between Judgement Task and Categorization Task, F (1,68) = 3.77, p <.06.
Judgement Tasks 100
Condensation Filtration
Mean % Correct Categorization
1000
Mean Correct RT (msec)
Subjects first made similarity judgments about pairs of schematic faces. One group based their judgements on the similarity of features (eyes or mouth; separable priming) wheras the other group judged the similarity of the emotion expressed by the faces (integral priming).
800
600
400
200
Condensation Filtration
90
Emotion
Feature
Judgement Task
fMRI Analysis The fMRI scans were analyzed with the FSL software (Image Analysis Grouup, FMRIB, Oxford, UK). Each block of categorization trials was contrasted with a block of rest. The results of this analysis was used in a secondlevel, betweensubjects analysis. This analysis found a trend in the data similar to that found in the behavioral data. Specifically, a significant main effect was found for Categorization Task and for the interaction between Judgement Task and Categorization Task. Task Effect: The betweengroups analysis of the fMRI data mirrored that of the behavioral data in showing a significant task effect; i.e., the filtration task was easier than the condensation task. Areas demonstrating superthreshold activation were: anterior cingulate, caudate, cingulate gyrus, and medial frontal gyrus. Shown below are the tmaps for the third categorization block.
Anterior Cigulate (14, 24, 24) Caudate (12, 0, 20) Cingulate Gyrus ( 18, 2, 40) Medial Frontal Gyrus (18, 2, 40) Corrected cluster level, F = 4.12, p < .05
Interaction Effect: In addition to the task effect, the between groups analysis revealed a significant interaction between the judgement task (Emotion, Feature) and categorization task (Condensation, Filtration) reflecting those areas most active when the categorization task was incompatible with the judgement task.
70
60
Emotion
Feature
Following the judgement task, subjects were asked to categorize a set of new faces into two categories. All subjects categorized the same set of faces; however, for one group, the faces were categorized on the basis of a single feature (separable task), and for another group, the same faces were categorized on the basis of two features
Cingulate Gyrus
Parahippicampal Gyrus
Block 1
Block 2
Block 3
Our finding that anterior cingulate, caudate, cingulate gyrus, and medial frontal gyrus are involved in category processing replicates other work that has examined the neural correlates of category processing. What we believe to be unique is our finding that parahippocampal gyrus and posterior cingulate appear to moderate the way stimulus features are processed. These areas were most active when the attentional set (emotion or feature) conflicted with the the type of category structure being learned (integral or separable). In addition, the principal components analysis indicates that category learning is reflected in the location and orientation of patterns of activation specific to the level of knowledge attained.
References
Judgement Task
Feature Task
Categorization Tasks
PCA Analysis: Suprathreshold voxels were eroded using a connectivity measure to eliminate sparsely connected voxels. The remaining voxels were analyzed using the following principle components analysis: the voxel cooridinates of the of the remaining voxels were placed in an N X 3 matrix, M, and then an eigen decomposition was then performed on the covariance matrix Mt M.
Discussion
80
50
0
Emotion Task
judgements were better on the Filtration task than on the Conden sation task. Those subjects who had performed the emotion judgements were equally good on both types of categorization tasks.
Garner, W.R. (1974). The Processing of Information and Structure. Lawrence Erlbaum, Potomac, Maryland, 1974.
Performance on the categorization task was also analyzed for sub jects who were scanned. The pattern of results was similar to that obtained for subjects who had not been scanned. The data were analyzed separately for the three categorization blocks. All subjects found the Filtration (separable) task to be simpler than the Conden sation (integral) task no matter what judgement task they had origi nally performed. In addition, those subjects who had made feature
Smith, L.B. (1989). From global similarity to kinds of similarity: The construction of dimensions in development. In S. Vosniadou and A. Ortony (Eds.), Similarity and analogical reasoning (pp. 146 178). Cambridge: Cambridge University Press. Parahippocampal Gyrus (11 46 4) Posteriior Cingulate (12 44 8) corrected cluster level, F=3.98, p <.05
Acknowledgements We would like to thank the James S. McDonnell Foundation and NSF (EIA0205178) for their generous support. Thanks also to Donovan Rebbechi for his help with the principal components analysis.