Teaching between-class generalization of toy play behavior to handicapped children.

Four preschoolers with developmental delays learned to play with toy sets. The trainer picked sets that looked different but worked the same way, like four kinds of cars or four kinds of blocks.

Kids first got praise for playing with one car. Then the trainer swapped in a new car without teaching again. They tracked if the child still played correctly with the new toy.

Training one toy in a set usually made the kids play the right way with the other toys in that set. It worked for 14 out of 16 sets.

When the trainer gave the child a toy from a different class, like moving from cars to a piano, the play skill did not carry over. Generalization stayed inside the toy family.

Watson et al. (2007) copied the idea using video instead of live training. Kids with autism watched clips of play, then generalized to similar toys. Both studies show the same fence: generalization hops to related toys only.

Garcia et al. (1971) saw the same fence earlier with imitation. Children learned to copy hand motions only if the new motion matched the trained class. The pattern repeats across decades: train within the class, get free responses inside the class.

Wilson et al. (2017) pushed the idea further. They used a matrix to teach play actions and got 86% new combinations plus peer requests. Matrix training keeps the within-class rule but adds social play to the mix.

You can save hours by teaching one toy from each class instead of every single item. Pick toys that share the same action, like stacking cups, ring stacks, or cars that roll. Do not expect the skill to jump to toys that work differently; plan separate lessons for those. This keeps programs lean and keeps you from guessing why generalization failed.