PSYCHOLOGICAL SCIENCE Research Report PIGEONS ARE SENSITIVE TO THE SPATIAL ORGANIZATION OF COMPLEX VISUAL STIMULI Edward A. Wasserman,' Kim Kirkpatrick-Steger,' Linda J. Van Hamme,' and Irving Biederman" '7/ie University of Iowa and ^University of Southern California Abstract—Two experiments investigated the role of spatial organization in the dis- crimination and generalization of com- plex visual stimuli hy pigeons. In Exper- iment I, after pigeons had been trained lo discriminate line drawings offour ob- jects, they were tested with novel pic- tures in which the same component parts ofthe objects were spatially rearranged. The -Spatially scrambled pictures led to a dramatic drop in recognition accuracy, but responding remained above chance. In Experiment 2, pigeons reached a high level of discriminative performance when required to choose among four dif- ferent spatial arrangements of the same object parts. These results conjirm Cer- ella's (1980) conclusion that pigeons discriminate the component parts of complex visual stimuli, but. unless it is assumed that the scrambling deleted or created emergent features, the results disconfirm his conclusion that spatial or- ganization plays no role in pigeons' pic- ture perception. In an important and widely cited study. Cerella (1980, Experiment III) trained two pigeons to discriminate car- toons of Charlie Brown (the positive stimulus class) from cartoons of other Peanuts characters (the negative stimu- lus class). Cerella then tested the birds with scrambled versions of Charlie Brown which involved vertical rear- rangements of Charlie's head, torso, and legs. Both birds responded at the same high rate of keypecking to the intact and the scrambled versions. The observation that test responding was unaffected by spatial rearrangement led Cerella to make several strong con- clusions about the pigeon's analysis of complex visual stimuli: (a) "Insensitivity to scrambling indicates that the absolute Address correspondence to E.A. Wasser- man. Department of Psychology. The Univer- sity of Iowa, Iowa City. lA 52242. or relative placement of features within the image is unimportant" (p. 5). (b) "The mechanisms of identification must operate . . . at the level ofthe local fea- tures which survive the fragmentation of the stimulus" (p. 5). (c) "The identifica- tion of [cartoons of Charlie Brown can- not] depend on the computation of global properties such as three dimensional structure" (p. 5). What are the implications of Cerella's results? It is implausible, for example, that scrambling a scene or an object would show no effect on humans' recog- nition of that stimulus. Indeed. Bieder- man and his associates (1972; Bieder- man, Glass, & Stacy, 1973; Biederman, RabinowitE, Glass, & Stacy, 1974) showed deficits in scene perception and in the recognition of objects in such scenes when the scenes were jumbled. Thus, if Cerella's conclusions were gen- erally true, we would be faced with the intriguing possibility that the pigeon's highly developed ability to recognize ob- jects by sight (e.g., Bhatt, Wasserman, Reynolds. & Knauss, 1988; Herrnstein, 1985) is both computationally and neuro- physiologically different from that of hu- man beings. Unlike humans, pigeons might "respond to complex line projec- tions as collections of local features rather than as representations of three dimensional objects" (Cerella, 1980, p. 1). Before accepting Cerella's thesis, we conducted two experiments designed to explore further the effect of component scrambling on the pigeon's discrimina- tion of visual portrayals of complex ob- jects. Before describing the experiments, we note that there is an asymmetry in what can be concluded if there is or is not an effect of scrambling on discrimi- native performance. If no effect of scrambling is obtained (i.e.. if pigeons respond as well to the scrambled ver- sions of a stimulus as to the original), then it can be concluded, as Cerella did, that these birds are indeed insensitive to the relations among the parts. Moreover, one can also conclude that whatever original features might have been lost or new features created as a consequence of the scrambling, those features are ir- relevant to the pigeons' performance. In contrast, if generalization pertbrmance suffers because of scrambling, then, without additional research or assump- tions, it is unclear whether the pigeons are sensitive to the component relations or to new or lost features produced by the scrambling. This uncertainty concerning the ef- fects of scrambling arises from the ambi- guity in the information that controls the pigeon's visual discrimination. There is evidence (Biederman & Cooper, 1991) that people represent a complex object as an arrangement of the convex regions that correspond to the object's parts. If it is assumed that pigeons do the same, then a decrement in generalization to scrambled stimuli that preserve the ob- ject's parts would be evidence that pi- geons were sensitive to the relations among those parts. Above-chance gener- alization to scrambled stimuli could be interpreted as evidence that pigeons also respond to the parts, independent ofthe relations among them. EXPERIMENT 1 In our first experiment, we adapted the four-alternative forced-choice method of Bhatt et al. (1988) to train pi- geons concurrently to discriminate line drawings of four common objects. Sub- jects were then tested with novel pic- tures in which the same component parts ofthe objects were portrayed, but rear- ranged both vertically and horizontally from their original positions. Would the pigeons continue to discriminate the scrambled drawings as accurately as they discriminated the original ones? Or would their accuracy drop as a result of spatial scrambling? 336 Copyright ® 1993 American Psychological Society VOL. 4. NO. 5. SEPTEMBER 1993