Straight talk about linear separability

It is intuitive that prototypes and additive similarity calculations might underlie human categorization, promoting a special appreciation of linearly separable categories. The failure to document empirically this appreciation has helped focus interest instead on exemplar strategies, multiplicative similarity calculations, and theory-based categorization. However, existing studies have mainly sampled poorly differentiated categories with small exemplar sets. Therefore, the present research repeated existing studies on linear separability, using better differentiated categories better stocked with exemplars. Both the data patterns and modeling suggest that prototypes and a linear separability constraint may have a stronger influence on categorization for these alternative category structures. The information-processing basis for this result is discussed.