Generalization and discrimination of shape orientation in the pigeon.
Expect bumpy, two-humped generalization when teaching orientation, and use slow angle steps plus explicit “wrong” examples to sharpen control.
01Research in Context
What this study did
Researchers taught pigeons to peck when a parallelogram sat at one angle.
They then tilted the shape in tiny steps and watched how the birds responded.
The goal was to map how orientation alone controlled the birds’ choices.
What they found
The birds did not give a single smooth peak of responding.
Instead they showed two bumps: strong pecking near the trained angle and again at the mirror opposite.
Extra discrimination sessions sharpened the main peak and flattened the extra bump.
How this fits with other research
Bickel et al. (1991) later saw the same twin-peak pattern when pigeons learned with Pavlovian cues instead of peck cues.
This shows the bimodal curve is not tied to one training style; it appears whenever the task is orientation.
Nevin (1967) worked with monkeys on line-tilt and found that gradual shaping gave cleaner control than sudden jumps, matching the step-wise method H et al. used to tighten the pigeons’ performance.
Together the three papers say: expect wide, lumpy generalization at first, then use slow steps and extra contrast trials to narrow control.
Why it matters
When you teach a child to sort diagonal lines or clock hands, first responses may spread to similar but wrong angles.
Build programs that start with wide differences, then shave the angle down a few degrees at a time and add clear “not this” examples.
This old pigeon data reminds you to keep probes mixed and to stay patient; tight visual discrimination needs many small steps.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Take the current angle-matching task, widen the difference by 15 degrees, then fade it back 3 degrees every two sessions while mixing clear non-matches.
02At a glance
03Original abstract
Pigeons learned to peck a green key on which parallelogram-shapes were projected; they then received generalization tests in which the orientation of the parallelogram was varied. Nondifferential training produced very little eventual stimulus control along the orientation dimension, but when training included S- trials (absence of the parallelogram) subjects responded consistently more to certain orientations than to others. Unlike typical results for visual generalization (e.g., line-tilt), the tilt gradients obtained for this complex stimulus were bimodal, supporting predictions on the basis of human perceptual data. However, unimodal gradients could be produced by specific discrimination training along the orientation dimension. Other forms of intradimensional training also produced relatively steep gradients, often characterized by unexpected but consistent secondary peaks. An attempt to obtain inhibitory gradients (S+: green key; S-: parallelogram on a green background) resulted in virtually zero responding all along the shape-orientation dimension; therefore, specific inhibitory control could not be evaluated. All these experiments suggest that definition of this complex stimulus dimension in terms of mere "angular orientation" is inappropriate, and alternative interpretations are discussed.
Journal of the experimental analysis of behavior, 1968 · doi:10.1901/jeab.1968.11-753