Generalization across dimensions: A model for three‐alternative choice

Davison et al. (2025) worked with pigeons in a three-key chamber. One key paid food on a variable-interval schedule. A second key never paid. A third key paid at a new rate partway through the session.

The team asked: does raising the pay on the third key make the bird peck the unpaid key more or less? They tracked how dimensional distance between keys shaped the answer.

When the newly rich key sat close to the unpaid key on the stimulus line, pecks on the unpaid key jumped. When the rich key sat far away, pecks on the unpaid key dropped.

The results fit a generalization-across-dimensions model: reinforcer rate spreads like a gradient, boosting or suppressing extinction responding depending on distance.

Evenhuis (1996) showed that stimulus disparity already swayed choice in three-alternative setups. Davison extends that idea by adding extinction and mapping the gradient’s effect on resurgence.

Brown et al. (1972) found that pigeons would not generalize to untrained tones without direct teaching. Davison’s model agrees: generalization follows trained dimensional links, not smooth continua.

Fontes et al. (2018) brought back extinguished responding by punishing an alternative. Davison achieves the same resurgence but by enriching, not punishing, a dimensionally close alternative — a mirror-image outcome.

If you run concurrent schedules or multiple schedules in clinic, remember that changing pay on one alternative can leak into others. Boosting reinforcement for a similar task may revive an extinguished problem behavior. Dropping pay on a distant task could do the same. Plot your alternatives on their stimulus or response dimension and plan for these ripples.