As before, stimuli for this experiment were generated randomly, given the constraints which defined each of the categories. As in Experiment 1, adjectives were organized along lexical dimensions, specified by the most relevant input dimension and the linguistic context input. In this case, there were four lexical dimensions, one each for the four input dimensions that specify the presented objects.
Unlike in Experiment 1, however, the adjective and noun categories were identical in every other way; in fact, the same set of 16 categories was used for the 16 nouns as well as the 16 adjectives. For all categories a single sensory dimension was most relevant; that is, the range of variation possible along that dimension was considerably narrower than on the other three dimensions. For example, one adjective category was defined in terms of ranges spanning 2/3, 2/3, and 1/3 of three of the input dimensions and 1/12 of the relevant dimension, and one of the noun categories was defined in exactly the same way. Whereas the noun and adjective categories overlapped completely (since they were identical categories), there was no overlap within the noun and adjective classes. This is necessary for the condition with no linguistic context specifying the lexical dimensions; without such linguistic input, it would be impossible to learn overlapping categories. Thus in this experiment, the only factor distinguishing the two classes of outputs is the presence of linguistic contexts associated with subsets of words and specific perceptual dimensions.