Características de los participantes en el chaku

Since Warrington and her colleagues suggested that category-specific deficits may be explained in terms of disproportionate impairment to information from particular sensory/motor modalities, a number of authors have examined the performance of patients with category-specific impairments for living things on tasks designed to differentiate between various types of semantic knowledge. Although the fimctional/sensory hypothesis

has been extended to include specialized channels within each of the main sensory/motor modalities, these studies have concentrated on the broad distinction between sensory and functional (or visual and non-visual) information.

One of the first studies of this type was carried out by Silveri and Gainotti (1988). These authors found that, in a task of naming animals to definition, their patient (LA) performed much better when the definitions consisted of verbal metaphorical expressions or functional descriptions than when they stressed visual perceptual features. At first these results appear to support the view that the patient's deficit for the category of animals was associated with an impairment to visual knowledge. However, Stewart, Parkin and Hunkin (1992) have pointed out that the mean word frequency of the items described by functional attributes was higher than the mean frequency of the items that were described in terms of visual properties, suggesting that the superior performance with the former type of definition may have been an artifact of word frequency. Furthermore, Hillis, Rapp, Romani and Caramazza (1990) have noted that in all tasks, the patient demonstrated superior knowledge of domestic animals than wild animals. This distinction is important because 79% of the definitions containing verbal features referred to domestic animals, whereas only 18% of the definitions containing perceptual features did so. The possibility that this patient's greater success with functional rather than visual definitions was an artifact of other factors cannot, therefore, be ruled out.

A number of other studies support the view that category-specific deficits for living things are associated with an impairment to visual knowledge. Hart and Gordon (1992) and De Renzi and Lucchelli (1994) have both described patients who, when tested in the verbal modality, demonstrated impaired knowledge of the visual attributes of living things in the context of intact knowledge of their non-perceptual properties. Sartori and his colleagues (Sartori and Job 1988; Sartori, Coltheart, Miozzo and Job 1994) and Farah, Hammond, Mehta and Ratcliff (1989) have also reported patients whose knowledge of the visual characteristics of living things was much worse than their knowledge of their non-visual properties. Finally, Riddoch and Humphreys (1993) have described a patient who was good at naming living things from functional definitions, but poor at naming them from visual

definitions. These findings suggest that category-specific deficits for living things may indeed be related to impairments of visual knowledge, and hence lend some support to those theories of semantic processing which explain category-specific deficits in terms of damage to information from particular sensory/motor modalities.

However, another set of findings seems to contradict this view. Patients have been described whose deficit for living things affects both visual and non-visual information. For example, Laiacona, Barbarotto and Capitani (1993) tested the semantic knowledge of two patients (FM and GR) with a verbal questionnaire, and found a category effect that was unrelated to question type. They observed a general deficit for living things, which did not spare associative information, in the context of preserved knowledge of both the perceptual and fimctional attributes of non-living things. Sheridan and Humphreys (1993) have reported a similar case. Although their patient (SB) had intact stored perceptual knowledge when tested visually (by object decision), her ability to retrieve both visual and verbal information about concepts from their names was significantly poorer for living things. Finally, Mehta, Newcombe and de Haan (1992; see also Newcombe, Mehta and de Haan 1994) have also described a patient (MS) who had impaired knowledge of both the visual and functional attributes of living relative to non-living things.

The latter pattern of performance undermines the view that category-specific deficits for living things are the result of damage to visual knowledge, as these patients each demonstrated an impairment to both visual and functional knowledge about living things. This pattern of performance suggests that, for these patients, the impairment may be genuinely category-specific rather than attribute-specific. Hence cases such as these would appear to support models of semantic processing that posit that the semantic system is organized by taxonomic category.

However, Farah and McClelland (1991) have recently presented a model of semantic processing which, they claim, demonstrates that the observation of impaired functional knowledge of living things can be accounted for without rejecting the notion that category- specific deficits for living things are the result of damage to visual knowledge.

6.12 A computational model of semantic memory impairment

Farah and McClelland have recently developed a computational model of semantic memory which, they claim, demonstrates that category-specific impairments can be accounted for by a system that is organized by modality, without requiring further organization by category (Farah and McClelland, 1991; Farah 1994). The basic model consisted of three pools of units, corresponding to verbal inputs or outputs (name units), visual inputs or outputs (picture units) and semantic memory representations. There were bi-directional connections both within and between pools, with the exception that there were no direct connections between the name and picture units. The semantic representations were divided into visual and functional units. Any particular item was represented by a pattern of activation over the name and picture units, and a pattern of activation over a subset of the semantic units. The average ratios of visual to functional units in the semantic representations of living and non­ living things were determined by asking normal subjects to identify the number of visual and functional descriptors used in the dictionary definitions of the items used. The resultant ratios of visual to functional units were 7.7:1 for living things and 1.4:1 for non-living things. The overall ratio of visual to functional units was roughly 3:1.

The model was trained using the delta rule (Rumelhart, Hinton and McClelland 1986) to associate the correct semantic and name pattern when presented with each picture pattern, and the correct semantic and picture pattern when presented with each name pattern. The model was then damaged by eliminating different proportions of visual and functional semantic units, and the effects of such "lesions" on the model's performance on a simulated picture-name association task were explored. (Half of the trials simulated picture naming, while the other half simulated auditory matching-to-sample.) The normalized dot product of the observed pattern across the semantic units and the correct pattern, which provides a measure between 0 (completely dissimilar) and 1 (identical) served as the dependent measure. It was found that, with increased damage to visual semantics, the performance of the model on the picture-name association task dropped, and that this decline was more precipitous for living than for non-living things. This would appear to correspond to a

selective impairment of semantic memory for living things. By contrast, damage to functional semantic memory impaired knowledge of non-living things to a greater extent than living things.

The ability of the model to simulate category-specific deficits in this way is not particularly surprising, given the manner in which the two categories were represented by different ratios of visual to functional semantic units. However, it was also claimed that the model could simulate the observation that for some patients with impaired knowledge of living things, the deficit extends to the functional properties of these items. When the model's ability to produce the correct pattern in functional semantics on presentation of a name pattern or picture pattern was assessed, it was observed that, for all levels of damage to visual semantic units, the ability to retrieve functional semantic knowledge was more impaired for living than for non-living things.

Farah and McClelland therefore argued that the characteristics of a category-specific semantic memory impairment can be simulated by a parallel distributed model which only distinguishes between modalities of knowledge, as long as the model also incorporates the idea of active, distributed representations, in which the activation of a representation depends on mutual support amongst different parts of the representation. They argued that, although such representations are quite robust, large amounts of damage will deprive intact portions of the representation of the support necessary to attain their proper activation levels. Because the semantic representations of living things include more visual units than the representations of non-living things, the former category would be more impaired by damage to visual semantic units. In addition, because living things have a high ratio of visual to functional semantic units, damage to visual units would deprive the functional units of living things of the collateral activation necessary to attain their proper activation levels. Hence, the quality of functional semantic information about living things would be inferior to functional information about inanimate objects, despite the fact that functional units themselves would not be impaired.

demonstrated by some patients with category-specific impairments that initially posed a problem for those models of semantic processing which explain category-specific deficits for living things in terms of damage to visual information.

6.13 Which theory of semantic memory does the Farah and McClelland model