Slide 16 of 25
Notes:
The standard version of cognitive modeling is committed to similarity as the measure by which one decides whether an item is or is not a member of a category, but category membership is neither sufficiently stable nor metric to be characterized by a single similarity array (see e.g., Barsalou, 1982, 1983).
People reliably violate the triangle inequality in judging similarity. China and Cuba are similar, Cuba and Puerto Rico are similar, but China and Puerto Rico are not similar.
Similarity ratings depend on the order in which items are presented. England is more like the US (12.84) than the US is like England (11.4). Prunes are more like apples (9.84) than apples are like prunes (8.77). These and other similarity judgments appear to be based on a dynamic system that includes contextual information, and which dynamically highlights and weights a set of dimensions for comparison. The highlighted dimensions are those that exhibit variability in the context. Similarity judgments are based on a mechanism that dynamically alters the importance of dimensions as a function of their variability. The process adjusts the weights attributed to dimensions to reflect the transitory diagnosticity of the dimension (Goldstone et al. 1997).
Barsalou (1983) found similar effects for ad hoc categorization. Categorization and similarity are not intrinsic to the stimuli, but rather are complex functions of the stimuli in context and are computed ad hoc when the judgments are being made.
