stein, Loveland, & Cable, 1976; Lubow, 1974; ... slide tray, the red light came on again. The .... The maximum capacity of the slide tray was 40 pairs. Thus, to ...
Journal of Experimental Psychology: Animal Behavior Processes 1984, Vol. 10, No. 2, 256-271
Copyright 1984 by the American Psychological Association, Inc.
Pigeon Visual Memory Capacity William Vaughan, Jr., and Sharon L. Greene Harvard University This article reports on four experiments on pigeon visual memory capacity. In the first experiment, pigeons learned to discriminate between 80 pairs of random shapes. Memory for 40 of those pairs was only slightly poorer following 490 days without exposure. In the second experiment, 80 pairs of photographic slides were learned; 629 days without exposure did not significantly disrupt memory. In the third experiment, 160 pairs of slides were learned; 731 days without exposure did not significantly disrupt memory. In the fourth experiment, pigeons learned to respond appropriately to 40 pairs of slides in the normal orientation and to respond in the opposite way when the slides were left-right reversed. After an interval of 751; days, there was a transient disruption in discrimination. These experiments demonstrate that pigeons have a heretofore unsuspected capacity with regard to both breadth and stability of memory for abstract stimuli and pictures.
Little work has been done on long-term memory for visual stimuli in animals. It has been well documented, on the other hand, that the human capacity for picture recognition is large (Nickerson, 1965, 1968; Shepard, 1967; Standing, 1973; Standing, Conezio, & Haber, 1970). Standing et al. (1970) showed subjects 2,560 photographic slides, followed by test pairs for which subjects had to recognize which member of the pair had been previously seen. Accuracy was over 90%. Standing (1973) exposed subjects to up to 10,000 pictures. These subjects demonstrated recognition well above chance. Accuracy, however, decreased with the number of days between initial exposure and the recognition test in such tasks (Shepard, 1967; Nickerson, 1968). Human recognition of pictures is superior to recognition of nonpictorial'stimuli, such as words or sentences (Shepard, 1967) or of threedigit numbers (Shepard & Teghtsoonian,
This study was supported by NIMH Grant MH-15494 and NSF Grant IST-8100404 to Harvard University. We thank J. Cerella, It J. Herrnstein, and B. F. Skinner for suggestions and encouragement. Some of these results were presented at the Fourth Harvard Symposium on Quantitative Analyses of Behavior, June 20, 1981. The current address for Sharon L. Greene is Bell Communications Research, Inc., Room 2C-115,600 Mountain Ave., Murray Hill, New Jersey 07974. Requests for reprints should be sent to William Vaughan, Jr., Department of Psychology and Social Relations, Harvard University, Cambridge, Massachusetts 02138.
1961). Complexity or variability on many different dimensions, as is the case in most picture recognition studies, may facilitate recognition. However, variability within a class may make recognition more difficult, as shown by Goldstein and Chance (1970). Their results constituted an exception to the observed superiority of picture recognition over other forms of recognition memory. They argued that the variability in stimuli used in prior picture memory studies could lead to recognition of class membership of a picture without recognition of a particular picture. Goldstein and Chance had subjects view pictures of the same conceptual class. The subjects saw either faces, inkblots, or snowflakes. Accuracy was much lower than in prior studies. Faces were recognized best (71% correct), ink blots next (46%), and snow crystals were worst (33%). It has been found that when a specific detail is named as the basis for the recognition judgment in picture memory studies, performance is superior to recognition based pn simple general visual familiarity (Loftus & Bell, 1975; Loftus & Kallman, 1979). Due to the nature of the stimuli used in the Goldstein and Chance (1970) study, this labeling of detail was unlikely. The combined findings that (a) verbal labeling of details aids picture recognition, (b) picture recognition is not much better than other forms of recognition memory when the pictures are of essentially the same conceptual class but is superior when stimuli are concep-
256
PIGEON VISUAL MEMORY CAPACITY
tually varied and physically complex, and (c) picture recognition falls off appreciably with large retention intervals suggest the usefulness of examining picture recognition and longterm memory capacity in animals. Dual representation (i.e., verbal labeling and visual encoding) would not be an issue. Questions regarding capacity, the effects on recognition of graphic and conceptual similarity among stimuli, and the shape of decay functions are directly addressable. Animal memory research, to date, has mainly been concerned with short-term, or working, memory. Questions of representation have typically been investigated using delayed matching-to-sample procedures, spatial arrays, or sequences of visual stimuli (D' Amato, 1973; Hulse, Fowler, & Honig, 1978; Olton, 1978; Roberts & Grant, 1976; Roitblat, 1982; Straub & Terrace, 1982). Sands and Wright (1980) used pictures as stimuli in a probe recognition procedure, but this procedure did not test the animal's capacity for long-term retention of large numbers of pictures. The animal research that comes closest to addressing issues of long-term memory has not involved picture recognition. Hoffman, Selekman, and Fleshier (1966) demonstrated pigeons' retention of a conditioned suppression after several years. Research from another more naturalistic setting has shown that various birds, such as marsh tits and certain nutcrackers, can find a large number of sites in which they have cached seeds. The recovery of these caches by marsh tits occurs usually within 24 hours after caching (Cowie, Krebs, and Sherry, 1981; Sherry, 1982; Sherry; Krebs, and Cowie, 1981; Shettleworth and Krebs, 1982), but with Clark's nutcracker, recovery may not occur for several months (Vaaider , Wall, 1982). Experimental manipulations of a laboratory environment simulating a natural setting have implicated memory of specific sites in the recovery of caches. The large number of caches (up to several thousand) recovered by Clark's nutcracker indicates that these birds must have extremely large long-term memory capacities (Vander Wall, 1982). Recovery seems to be guided by visual cues in the environment. Concept formation research using animals implicitly entails memory for pictorial stimuli. The animals, usually pigeons, are shown large
257
numbers of slides that are divided into reinforced and nonreinforced categories based on some conceptual distinction. By reinforcing pecks to slides containing category members (e.g., trees) and not reinforcing responses to slides lacking category members, birds are easily trained to distinguish certain classes of pictures (Cerella, 1979; Greene, in press; Herrastein, 1979; Herrnstein & deVilliers, 1980; Herrnstein & Loveland, 1964; Herrnstein, Loveland, & Cable, 1976; Lubow, 1974; Malott & Siddall, 1972; Pietrewicz & Kamil, 1977; Poole & Lander, 1971; Siegel & Honig, 1970). Given that transfer of the discrimination is usually quite good, most of these studies claim to demonstrate conceptual abstraction. Under certain circumstances, however, the contribution of picture memorization to such discriminations is substantial (Greene, in press), further indicating the need to investigate the properties of long-term memory for pictures in pigeons. The following experiments directly address pigeons' long-term memory capacity for complex stimuli. The procedure used is similar to ones used in concept formation studies in pigeons and thus involves not only a recognition of a particular picture but a classification of the picture into the reinforced or nonreinforced category. The nature of the stimuli shown to different groups of birds allows a comparison between memory for natural pictures that vary visually along many dimensions and stimuli that are simple random shapes that appear to the human observer more difficult to distinguish. Several transformations of the stimuli (e.g., rotations, masking) were performed during the course of the experiments to investigate the visual cues that were important for recognizing and classifying the stimuli. General Procedure The following procedure, with slight variations, was employed in all four experiments. Pigeons were run in operant chambers equipped for the back-projection of slides onto a response key. Figure 1 illustrates the general sequence of events. At the beginning of a session, the response key was illuminated by a red light. A single response produced 3 s of reinforcement (access
258
WILLIAM VAUGHAN, JR., AND SHARON L. GREENE
]///\ Slide on i
iI 1 1
4 sec ITI
Slide off Relnf off Relnf on
Slide on Pos
K\N Rslnf on
i 1 I I 1 //////////////////A i j
10 sec Responses recorded
1 vi
J
1 J
1
i//AYWI
1
!
