Hierarchical cluster analysis of the birds= performance with the object displays.

Neither during training nor during transfer, however, was there any evidence that within-category object displays was any harder to discriminate than between-category object displays. This result suggest that these pictorial stimuli were not being viewed as representatives of object-based categories. What then was the basis for the discrimination of these object displays?

To answer this question we conducted a hierarchical cluster analysis of the birds= performance with the object displays. Cluster analysis is one of several widely used multivariate techniques for detecting and modeling the psychological similarities which exist within a data set. Such techniques have been highly useful for illuminating the stimulus structure controlling an animal=s behavior in a variety of discrimination tasks (Blough, 1985, 1988, 1990; Matsuzawa, 1990, Tomonaga & Matsuzawa, 1992; Sands et al., 1982). In fact, one of the advantages of the present Same-Different task is that it readily provides measure of perceived similarity among sets of images. Our analysis was performed on 100 sessions of baseline performance collected soon after the completion of Experiment 2. During these additional sessions we had continued to expand the pool of object images presented to the birds. By the end of this time, the birds were being tested regularly with 32 object pictures (12 birds, 12 flowers, 4 humans, & 4 fish). Because the human and fish pictures had only been added towards the end, however, the number of pairwise observations involving these images was fairly limited and they were not included in the analysis described next. From these 100 sessions, we computed the average performance across the five birds for each pairwise combination of object pictures and then submitted these values to a cluster analysis using the average method (SPSS software package). The figure below shows the final clustering solution for the 12 bird and 12 flower images..

Its structure is easily interpreted. Three major clusters of perceived similarity can be identified. One cluster forms around those pictures best characterized as having a bluish coloration, a second centers around images having a reddish-brown cast, and a third forms around images involving combinations of greens and yellows. Thus, like the above analyses, this one provides no evidence of any object-based categorical encoding of these images. That is, the natural divisions we see and semantically label as birds and flowers were not being employed by the pigeons. Rather these complex compound stimuli seem to be primarily discriminated along the globally dominant color or pair of colors present in each picture.

Why were these birds not sensitive to the object-based categorical structure of these stimuli? There is nothing inherent about these particular stimuli that should have prevented such an outcome, since both categories have been shown to support this type of psychological organization in other settings (Bhatt et al., 1988; Cook et al., 1990; Ryan & Lea, 1994). One very likely possibility comes in the form of the dimension-analytic Ainstructions@ provided to the birds through the other display types. In the present setting all of their concurrent experience with the feature, texture, and geometric display types provided explicit reinforcement for attending to exactly this type of dimensional information, perhaps resulting in the overshadowing of the object-based categorical information present in these images.