Generalization gradients and stimulus control in delayed matching-to-sample.

Pigeons learned a delayed matching-to-sample task. They saw a color sample, waited 0–8 s, then picked the matching key from two choices.

Researchers tracked how stimulus control changed as the wait grew longer.

With short delays the birds chose the correct color. With longer waits the gradient flattened.

Control shifted from the sample to random parts of the display, like the edge of a key.

THOMAS et al. (1963) showed you must master zero-delay matching first. Sanders (1969) used that same teaching order, then revealed how control later erodes.

Donahoe et al. (2000) extended the finding. Accuracy fell to chance after 3–9 s, yet birds kept pecking. The 1969 flattening is now seen as recall fading while motivation stays high.

Neuringer (1973) mapped similar flat gradients after weak discrimination training. Both papers warn: flat gradients can mean either long delays or poor initial teaching.

When you add a delay, probe often and early. If the learner starts guessing, shorten the wait or tighten the discrimination before re-introducing the pause. Watch where eyes or fingers drift; those incidental cues may steal control from the target stimulus.