Avian Expert Articles

New Pepperberg Parrot Research: Recognizing 2d Objects

Grey Parrot Anatomy ProjectOne marker of advanced intelligence, in humans and nonhumans, is the ability to transfer an acquired concept to a novel situation. African grey parrot Alex, for example, went from stating, “None” if nothing (color shape, material) was either same or different between two objects, to stating, “None” when asked about the complete absence of a set of objects in a number study — that is, showing a zero-like concept. He wasn’t as advanced with respect to labeling shapes: Trained to identify solid wooden objects of roughly the same size but varying form as “one-, two- three-, four-, five-, six-, eight-corner”, he transferred those labels to other shaped items of different materials, sizes and colors; however, he was limited to three-dimensional (3D) objects. Could our younger parrot, Griffin, similarly trained, now go further, transferring, for example, from 3D objects to 2D drawings? Furthermore, could he possibly identify the types of occluded objects and optical illusions shown in Figure 1 below? The formal terms for such abilities are, respectively, “amodal” and “modal” completion. (Note: objects involved in modal completion are also called “subjective contours” and Kanizsa figures, after the scientist who designed them.)

Shapes-Fig-1-Cat-hiding-in-tall-grass-Tony-Alter-FCC-width-600
Figure 1. Photo of cat hiding in tall grass, by Tony Alter via Flickr Creative Common

This study, co-authored with a Harvard colleague, was particularly interesting because demonstrating such abilities in nonhumans has been difficult. Although neither a human brain nor visual system seem needed for such capacities — many studies, on insects to nonhuman primates, find appropriate responses to 2D objects that are partially occluded or partly represented in outline form by subjective contours — the results are actually ambiguous. The tasks are designed so that subjects might be focusing on simple aspects of the object without understanding what they are really seeing (e.g., matching the angle in the ‘pacman’ of Figure 1 to a test angle without necessarily recognizing the imaginary triangle). However, given that Griffin understood symbolic representation — i.e., he could vocally identify various shapes as did Alex — we could test him just like humans, who are usually given a few stimuli and simply asked to label what they see.

We asked Griffin to label figures drawn on paper, something he had never before done. In the amodal task, we used variously colored figures for each shape he could label, of different sizes, occluded by black circles (e.g., asking ‘‘what shape blue?”). Griffin had not been trained to label either circles or the color black making it unlikely that he would attempt to label them. We also used other black shapes as occluders, to see if being able to label occluders would distract him from the task. To ensure that Griffin was responding appropriately to occluded figures, we also asked him to identify irregular shapes that were not occluded (“detached probes”) — e.g., figures that looked as though a bite had been removed, with an appropriately sized adjacent circle (Figure 2a). For the modal task, figures were constructed using black ‘pac-men’ to form imaginary figures, again for each shape he could label, of different sizes, on colored paper. Controls (‘‘probes”, one or two for each of the #-cornered shapes) involved placing additional circles or ‘pac-men’ near the figure so Griffin could not simply quantify black objects (e.g., Figure 2b). Each trial was unique with respect to color, size of polygon, or size of occluder/pacmen. He was given only 38 trials for each type of task.

Fig2_Pepperberg
Figure 2

Unlike most experiments that test nonhumans by repeatedly presenting identical probes during many (again) repeated training trials, and either reward the subject for all probes (potentially encouraging guessing) or, after decreasing rewards to a set percentage similar to the proportion of probe trials, for none of the probe/test trials (potentially discouraging possible correct attempts), we decided that the strongest test would be to observe Griffin’s response to single presentations of each probe—thus avoiding issues of familiarity, training, or encouraging either guessing or discouraging correct attempts.

Griffin identified all non-probe figures at statistically significant levels, and was 100% on all probes. Importantly, he was correct on his very first trials, showing that no training was occurring and that transfer from 3D to 2D figures was immediate. Interestingly, for the detached probes (e.g., Fig. 2a), where no occlusion occurred, he responded to figures never before seen (irregular shapes) with the number of visible corners. For modal test stimuli, in contrast to amodally completed stimuli, where there is nothing in common between his 3D training objects and test stimuli, his accuracy was identical to that for the amodal stimuli. He also was not counting pac-men or numbers of circles, because no errors correspond to their quantity.

Against Griffin’s remarkable success, much data exist showing that, as noted above, animals generally either do not show these completion phenomena at all or show some degree of success only after having undergone considerable training with very closely related stimuli. In other instances, actual tasks differed considerably among laboratories (with respect to, e.g., motion, 2D vs 3D stimuli, CRT vs LCD monitors [i.e., flicker fusion effects] and pre-exposure to stimuli) with the consequence that results also varied considerably.

What could account for Griffin’s success? Other creatures must solve problems involving at least some form of amodal completion in their daily lives. For example, processing information about a partially hidden predator and reacting is safer than not, even if some false alarms exist. Modal completion may rely on similar processing; imagine three black circles on a colored background that are occluded by a triangle of the same color. Griffin’s results may be a consequence of two capacities, although others may also be involved. First, he already understood symbolic representation: that a sound could stand for a physical object. Thus understanding that a 2D picture could represent 3D reality, including depth perception, would not be surprising. Interestingly, baboons failed when tested for amodal completion in a task involving only 2D figures but did succeed (although only after several hundred training trials) when stimuli provided cues indicating that the occluder was indeed in front of the targeted object (e.g., seeing the occluder actually move over the object that was being occluded). Second, Griffin was raised in an extremely rich environment for a laboratory subject, providing him with the same kind of experience that seems to enable young children to succeed on such tasks. Thus he saw and manipulated real-world 3D objects, of all forms, materials, and colors, both in full view and occluded, during a period of over 16 years before the study began. Such experience may be a prerequisite for carrying out these tasks. Whatever the reasons for his success, what is striking is that he did succeed without the training needed by other nonhumans, thus exhibiting the ability to transfer: an ability that shows advanced intelligence.

Subscribe to our newsletter