| Abstract: | Two hundred consecutively seen aphasics, 142 of them with infarcts, were examined by tests of fluency, comprehension, repetition, naming, and information content. The language scores were subjected to a minimum variance clustering algorithm separately for the total and for the infarct groups. The latter generated 10 clusters on a dendrogram. Attribute analysis of each cluster provided a clinically meaningful profile of language performance for these groups. The degree of correlation of most computer generated clusters with clinically recognized groups was high, and the homogeneity of some of the clusters is striking. An exception appears to be “conduction aphasia,” which is bimodally distributed. One of these clusters, with high fluency and low comprehension scores, was renamed “afferent conduction” aphasia, and the other, with lower fluency and higher comprehension, was renamed “efferent conduction” aphasia. The Principal Components Analysis was used to evaluate the discriminatory value of language characteristics, and the Nearest Neighbor Network Analysis was used to evaluate the significance of clustering. The dendrogram for all aphasics showed a less specific and less homogenous six clusters. |
