Local contrast and maintained generalization.

Researchers trained pigeons to peck when they saw a vertical line. They used a variable-interval schedule—food came after unpredictable time gaps.

Next they added extinction. Birds saw lines tilted 5° to 30° left or right. No food appeared for these angles. Pecks were counted to build generalization gradients.

Early sessions showed big bumps—extra pecks—next to the reinforced line. These “local contrast shoulders” looked like tiny peak shifts.

After more training the bumps melted away. The gradient settled into a smooth curve. Stimulus similarity and training stage set the size of the early errors.

Brinker et al. (1975) said peak shift can appear after just three minutes of change—no long contrast history needed. Locurto et al. (1980) agree shift can pop up fast, but show it vanishes if you keep training.

Schmidt et al. (1969) got sharper peaks by making food duration obvious. The 1980 data add that once cues are clear, extra contrast edges fade.

Harrison et al. (1975) found more avoidance training steepens both excitatory and inhibitory slopes. The 1980 study mirrors this under positive reinforcement: longer VI plus extinction smooths the final gradient.

Expect brief “shoulders” of heightened responding next to the S+ during early discrimination. Don’t panic—they usually disappear with continued teaching. If you run stimulus generalization probes, collect enough sessions to see the stable curve, not the first bumpy one. This saves you from over-diagnosing peak shift or changing procedures too soon.