Authors:

BOZEAT, S., Lambon Ralph, M.A., GRAHAM, K.S., PATTERSON, K., Wilkin, H., Rowland, J., ROGERS , T.T. & HODGES, J.R.

Reference:

Year of publication:

2003

CBU number:

5244

Abstract:

These two studies analysed pictures of concrete concepts drawn by patients with semantic dementia to provide additional insight into the structure and internal representation of normal and degraded conceptual knowledge. A feature-based scoring scheme was constructed from analysis of drawings produced by normal participants to obtain a relatively objective measure that focussed on the content of the drawings and minimised the influence of drawing skill. In Study 1, in which six patients with semantic dementia were asked to produce drawings of concrete concepts from dictation of their names, the patients' drawings were characterised by a loss of distinctive features. In the artefact domain, this feature loss resulted in representations that were increasingly boxlike. In the living domain, as well as distinctive features being lost, there was a tendency for patients to include incorrect features which resulted in more familiar and "prototypical" representations. We found significant correlations amongst performance on the drawing assessment, object naming and word-to-picture matching, suggesting that all of the patients' impairments may be attributable to degraded central conceptual knowledge.A second study included two further conditions in the drawing assessment: immediate and delayed copying of line drawings of concrete concepts. The scoring scheme was also modified to establish the distribution of shared and distinctive features across concepts and to enable quantification of incorrect "intruding" features. Analysis of the drawings produced by three patients with semantic dementia confirmed that overall performance was significantly influenced by the task condition (immediate>delayed) and severity of disease. The rate of intruding features, but not of omitted ones, was influenced by the domain of the item, with a greater proportion of intrusions in the living than in the non-living domain. There was also a significant effect of feature-distinctiveness on the proportions of these error types: intruded features were most likely to come from the pool of properties that are shared across domain. These results replicate the findings of feature-listing studies and support the notion that higher order structures emerge from a distributed collage of fine-grained attributes.