Generalization and discrimination as a function of the S-D-S delta intensity difference.

BERRYMAELLIOTT et al. (1962) trained rats to press a lever when they heard a quiet tone and to stop pressing when the tone got louder. The team made the loud-quiet gap bigger or smaller across groups.

They then tested the rats with brand-new tones that sat between the trained sounds. The goal was to see how the size of the original difference changed later generalization.

Bigger gaps taught the discrimination faster. Rats reached error-free performance in fewer sessions when the S-D and S-delta tones were far apart.

Generalization gradients also slid in a predictable way. A rat trained with a 40 dB difference gave very sharp stop-go boundaries; a 5 dB difference produced flat, sloppy responding.

Cicchetti et al. (2014) extends the same rule to vision. Pigeons learned a four-color discrimination quicker when the colors were farther apart, echoing the intensity effect in rats.

Thompson et al. (1974) conceptually replicates the finding in pigeons working under an avoidance schedule. Orderly excitatory and inhibitory gradients summed algebraically, supporting Spence’s gradient-interaction theory first mapped in the 1962 rats.

Sanders et al. (1989) looks like a contradiction: monkeys failed to transfer a same-different rule from static to moving shapes. The difference is task complexity, not a flaw in the intensity principle. Simple sensory continua (tone loudness, color distance) generalize smoothly; abstract rules with shifting cues do not.

When you set up a new discrimination program, start with stimuli that are obviously different. A bigger S-D/S-delta gap cuts errors and gives you cleaner stimulus control. Once the learner is accurate, you can fade toward the natural range. This saves therapy time and reduces frustration for both you and the client.