Principal Investigator: Lorraine K. Tyler
Category Structure and Category-Specific Deficits: A Cognitive Science Approach
A fundamental human faculty is our ability to form categories of knowledge, to partition objects in the world into meaningful sets such as living things, man-made objects, animals and tools. During the process of acquiring a new concept, we learn not only its meaning but also the category to which it belongs. One of the most important category distinctions that people seem to represent is that between the domains of living and non-living things. This is a distinction which is observed in very young infants and it is seen in category-specific deficits when brain damage selectively damages one domain leaving the other intact.
The selective loss of specific categories of knowledge has traditionally been taken as compelling evidence that category structure is explicitly represented. The claim is that because categories of knowledge can be selectively impaired following brain damage they must be separately represented in the neural substrate. Recently, a different view has been developing in which category structure is seen as an emergent property of the structure and content of semantic representations. The present research builds on this earlier work in attempting to account for category-specific deficits within a distributed semantic system, in which category structure emerges from the complex interaction between the distinctiveness of semantic features, type of semantic feature (e.g. functional/perceptual) and correlations between features.
We propose to study this issue from a cognitive neuroscience perspective by developing a large-scale distributed computational model of semantics which incorporates several important theoretical assumptions; in particular, that categories differ in terms of the number and distinctiveness of features, and also in the number and strength of the correlations between features. The distribution of features in the model will be based on properties generated in a large-scale property norm study, thus providing psychologically valid properties for the model. The property norm study will involve asking large numbers of subjects to produce properties to living things and artefact concepts.
The model will be lesioned at increasing levels of severity to simulate the effects of brain damage, and the model's resulting behaviour will be compared to that of human subjects. We will develop experiments for unimpaired subjects and for patients with category-specific deficits to test predictions generated by the model. These experiments will use a variety of different types of tasks, in an attempt to probe the implicit, automatic activation of semantics as well as the more controlled aspects of cognitive processing. To the extent that the model's performance is similar to that of the patients with category-specific deficits, we can claim that category-specific deficits can arise from a damaged distributed system and thus semantic categories do not need to be explicitly represented.
To investigate the neural correlates of category knowledge we will obtain
MRI scans on our category-specific patients to try to relate the extent
of their brain damage with their performance on a variety of behavioural
tasks. In addition, we propose to carry out a small number of PET studies
with the patients, probing the issue of whether different brain areas
are differentially activated as a function of semantic category.
