This Title All WIREs
How to cite this WIREs title:
WIREs Cogn Sci
Impact Factor: 3.175

Revisiting domain‐general accounts of category specificity in mind and brain

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

Theories about the neural basis of semantic knowledge have been strongly influenced by reports that particular semantic categories can be differentially impaired by neuropathology and can differentially activate particular regions of cortex in brain imaging studies. One well‐known interpretation of these data is that the brain has evolved distinct functional and anatomical modules for storing and retrieving knowledge about different kinds of things. We review the evidence supporting an alternative view: that category specificity arises from many heterogeneous factors and so tells us little directly about the cognitive and neural architecture of semantic memory. We consider four general hypotheses about domain‐general causes of category‐specific patterns, their roots in early work, and their reemergence in contemporary research. We argue that there is compelling evidence supporting each hypothesis, and that the different hypotheses together can explain most of the interesting data. We further suggest that such a multifactor domain‐general approach to category specificity is appealing partly because it explains the important findings with reference to theoretical claims that are already widely accepted, and partly because it resolves several puzzles that arise under the alternative view. This article is categorized under: Psychology > Language Psychology > Memory Neuroscience > Cognition
Examples of category‐specific double‐dissociations in picture naming following brain injury. The bars show proportion correct for three individual patients naming animals versus man‐made objects (left and middle) or man‐made objects versus fruits and vegetables (right). In each case the pattern of naming impairment was said to exemplify a category‐specific knowledge impairment selectively affecting knowledge of animals, man‐made objects, or fruits and vegetables, respectively.
[ Normal View | Magnified View ]
Semantic relevance explains category‐specific impairment of living things in some patients with Alzheimer's disease and category‐specific patterns of activation in visual ventral cortex of the healthy brain. (a) When the patients were tested in picture‐naming and naming‐to‐definition tasks with commonly used items they demonstrated a more severe impairment of living things. (b) When the testing items were matched on semantic relevance in name‐to‐definition task, the performance discrepancy between semantic categories disappeared in the same group of patients for both high‐relevant and low‐relevant items, but overall performance was better for items with high rather than low semantic relevance. (c) In a picture‐naming task, tools often elicit greater activation than animals in medial posterior fusiform cortex. This pattern appears to be driven by the fact that tools are often higher in semantic relevance. When relevance is controlled, the category‐specific pattern is eliminated. (Reprinted with permission from Ref . Copyright 2006 Elsevier)
[ Normal View | Magnified View ]
Crowding effects explain category‐specific activation for animals in the lateral posterior fusiform cortex of the healthy brain. (a) Ventral temporal regions showing more activation for animals than for vehicles in a picture categorization task, when specificity of the category label is not taken into account. The observed lateral posterior fusiform peak is consistent with similar effects reported in many other studies. (b) Mean‐adjusted effect size in lateral posterior fusiform on each side. The category effect is only observed in the basic‐level condition. When the same items were classified at a more specific level, the same region showed significant and equivalent degrees of activation for both animals and vehicle items. (c) Mean‐centered effect sizes for peaks of the interaction term in posterior lateral fusiform gyri in both hemispheres. In all three sites, animal concepts elicited greater activation than vehicle concepts only in the intermediate condition. (Reprinted with permission from Ref . Copyright 2005 MIT Press)
[ Normal View | Magnified View ]
Schematic illustrations of a model architecture similar to that proposed by Farah and McClelland (hereafter FM) and one variant of an embodied framework for semantic knowledge similar to that proposed by Ref 25. In both panels, black arrows indicate bidirectional weighted connections between pools of neuron‐like processing units, and blue arrows indicate unit pools that receive direct input from the environment. (a) In the FM model semantic representations of both living and nonliving things were distributed patterns of activity over semantic features, with different features capturing knowledge about either sensory or functional properties of familiar items. Animals were proposed to have a higher ratio of perceptual to functional semantic properties than artifacts. Semantic representations received inputs from and directed outputs toward visual and lexical surface representations, and were reciprocally connected. When trained to compute mappings among visual, semantic, and lexical representations describing various animals and man‐made objects, the model could simulate performance on a range of semantic tasks including picture naming, word comprehension, and property verification. Semantic representations of animals and artifacts were attractor states that relied to different degrees on perceptual versus functional features for their stability. Consequently simulated damage to perceptual versus functional properties produced a category‐specific double‐dissociation for animals and man‐made objects. (b) Contemporary embodied views of semantic representations differ from the FM model in two respects. First, the dichotomy of perceptual versus functional properties is extended to a multitude of modality‐specific representations, including shape, color, motion, sound, praxis, function, and so on. Second, the semantic system is hypothesized to encompass these modality‐specific representations as well as cross‐modal representations, though different theorists vary in the details of their proposals. Embodied views raise the possibility that the same mechanisms captured by the FM model, but operating over a broader range of surface representations, may explain a broader range of category‐specific patterns of impairment.
[ Normal View | Magnified View ]
Schematic illustration of the semantic crowding effect and evidence from a computational simulation and from patients with herpes simplex viral encephalitis (HSVE). (a) Basic‐level categories (e.g., dog, cat, and horse) in the animal domain are represented in a highly crowded space because they share many properties, but basic‐level categories in the vehicle domain (e.g., car, boat, and plane) are more well‐separated because they have fewer common properties. Small distortions in inputs to the semantic system may lead the system to confuse animals from different categories (e.g., cat instead of dog) because these have quite similar representations. The same distortion may not affect basic‐level categorization of vehicles because the neighboring category is quite remote from the target category. If pressed to generate more specific category names, however, crowding should disrupt performance in both domains, because different items within basic‐level clusters are highly crowded in both domains. (b) When naming animals and artifacts at the basic level (e.g., dog and car), a model based on the framework shown in Figure 2(b) exhibits a category‐specific pattern of impairment when its weights are perturbed with noise. Patients with HSVE show a similar pattern. (c) When naming animals and artifacts at a more specific level (e.g., poodle and Ford), the same model predicts that the category‐specific pattern should be eliminated, with patients showing equally poor performance for both domains. Patients with HSVE show the predicted effect. (Panels b and c: Reprinted with permission from Ref . Copyright 2007 Oxford University Press)
[ Normal View | Magnified View ]
Discriminability differences in low spatial frequency content across different semantic categories. (a) Black and white images of three animals (wolf, fox, and dog) and three vehicles (car, airplane, and boat) from different basic‐level categories. (b) The same images subjected to Gaussian low‐pass filtering which removes high spatial‐frequency information. As a consequence the animals become more difficult to recognize at the basic level than do the vehicles. The example illustrates how different categories may rely to differing extents on visual information encoded across different spatial frequencies.
[ Normal View | Magnified View ]
The effect of expertise on the deterioration of knowledge in semantic dementia, a domain‐general semantic disorder. (a) For six patients with semantic dementia (SD), mean proportion correct on a picture‐naming task for animals and man‐made items matched for frequency, familiarity, and visual complexity. As is commonly reported, the patients show equal degrees of impairment to both categories. (b) Performance of two individuals with SD on tests of car and plant knowledge. Patient 1 had expert knowledge of cars prior to diagnosis, whereas patient 2 had expert knowledge of plants. As a consequence of these premorbid differences, the patients show a strong double‐dissociation of knowledge about cars versus plants in the context of a semantic disorder that typically affects all categories equally. Yet their knowledge is not completely preserved compared to healthy experts in the same category, serious impairments are observed in both cases. (Reprinted with permission from Ref . Copyright 2011 Elsevier)
[ Normal View | Magnified View ]

Browse by Topic

Neuroscience > Cognition
Psychology > Language
Psychology > Memory

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts