Pigeon Lab

As birds are highly mobile animals, vision is of particular importance to them. This fact is reflected in the huge relative size of their eyes and the large portion of their brains that are dedicated to visual processing. On an absolute scale, however bird brains are quite small. The brain of a pigeon is about 1/1000 the size of our own. But nevertheless, pigeons are able to solve many of the perceptual problems as we do with our much larger brains. Therefore, the study of pigeons is important to our understanding of the general mechanisms of visual cognition.

Our major interest is focussed on those phenomena that pave the way to understand the phylogenetic evolution of human reason. Applying a new method of laboratory experimentation (Huber 1994), we have demonstrated that pigeons, though devoid of language and presumably also of the associated higher cognitive capacities, can categorize photographs or drawings as complex as those encountered in ordinary human experience.

Pigeons classify stimuli either via learning by rote, via extracting local features, or by abstracting a category prototype. Exemplar models assume that intact stimuli are stored in memory, and that classification or recognition is determined by the degree of similarity between a stimulus and the stored exemplars. According to feature models, categorization is based on the abstraction of “common features”, i.e., a necessary set of defining features that characterize members of the same category. Prototype models, finally, postulate that categorization is accomplished by the acquisition of a prototypical representation of a category, which is assumed to be a summary representation that corresponds to the “central tendency”.

Currently, we are exploring the parameters that may determine which stimulus aspects will actually come to control categorization. So far, the available evidence suggests that stimulus perception, feature selection and processing mode strongly depend on the specifics of a particular task, including size and structure of the category, stimulus design, context, training history, attentional factors, individuum-specific preferences, etc. Therefore, our goal is to understand natural categorization as a flexible and adaptive mechanism of selecting from numerous strategies.

Learning chambers

Most of experiments are carried out in operant indoor chambers (“Skinner boxes”), which the birds enter from their respective outdoor compartment through a connecting channel (Huber, 1994). They are motivated solely by hunger and the knowledge that food is available inside. The person conducting the experiment intervenes only by pushing sliding windows up and down - one at the channel entrance and one at the rear of the box. After an acclimatisation phase of several days, they traverse the channel without fear. As each Skinner box is adjoined to one compartment, each bird is assigned to a single testing chamber. The interior size of the wooden chambers is 50 x 30 x 40 cm. In the center of the front panel there is a clear Perspex pecking key (5 cm diam.). Directly below the key there is an aperture for the food hopper of the grain feeder. A hopper light illuminates the receptacle area whenever grain is accessible. Each Skinner box is connected to a Pentium PC, equipped with a relay board (8 x input, 8 x output) by Keithley/Metrabyte and with a software package (PigeonLab, M. Steurer) that controls all events in the operant chamber during experimental sessions. Stimuli are presented on 15-inch LCD monitors, at a distance of 5 cm behind the pecking key.


We choose between one of the following two conditioning procedures: a go/no-go procedure invented in the Herrnstein laboratory and a multi-stimulus learning paradigm.


The pigeons are required to peck onto the key in the presence of positive stimuli in order to obtain food, and to refrain from pecking in the presence of negative stimuli. Discrimination performance is calculated from peck rates to positive as compared to peck rates to negative stimuli. Each bird runs one session per day, five days a week.


In some experiments a multi-stimulus, multiple-matching learning paradigm is employed, which, we believe, not only simulates the foraging behavior of pigeons in natural environments, but also more realistically tests their categorization. Actually, it incorporates a number of elements to which pigeons are well adapted. The key features are exploration of the search space and selection of targets among several distractors. A touch screen TFT panel, placed in a learning chamber, shows a stimulus configuration consisting of a 2-dimensional array of small bitmap images that changes from trial to trial. The pigeon's task during each trial is to identify the images that are defined as positive by the experimenter and to peck at these. While pecks at a negative figure have no consequences for the subject, pecking at a positive stimulus leads to its disappearing. When the last positive stimulus has disappeared, the trial ends with a small food reward. Using a probability tree, we derive a probability density function to statistically assess learning rates.