The influence of correlations between noncritical features and reinforcement on stimulus generalization

Song et al. (2021) asked adults without disabilities to learn a new sorting task on a computer.

Some shapes were linked to points; others were not. The team made the color of the shapes either match the points most of the time or match only by chance.

They then tested whether the adults kept choosing by the true rule or got stuck on the color cue.

When color rarely predicted the points, people learned the real rule faster and made fewer color-based mistakes.

When color often predicted the points, people picked by color even when it stopped paying off.

Low link between non-critical cues and reward gave cleaner stimulus control.

Preissler (2008) saw the opposite pattern in children with autism. Those kids kept picking the picture that had been paired with a word, not the real object. The two studies look opposite, but the gap comes from the learners: neurotypical adults versus autistic children with cognitive delays.

Reiss et al. (1982) also used stimulus-equivalence drills and showed that extra looks at printed words can teach spelling without direct spelling practice. Song et al. add a guardrail: keep non-critical features weak during those looks so the learner locks onto the real relation.

DeFulio et al. (2011) found that real words beat jumbled letters when teaching typing. Their advice and Song’s both say: strip out misleading cues to cut learning time in half.

When you set up a discrimination program, scan for accidental cues like color, position, or who delivers the token. Make those cues random so the learner contacts the true contingency. You will see faster mastery and fewer odd errors in probe tests. Try it with matching tasks, intraverbal drills, or conditional discriminations next week.