The use of matrix training to promote generative language with children with autism

Five preschoolers with autism learned noun-verb phrases like "dog jump" and "cat eat."

The team used matrix training. They taught only diagonal pairs on a grid. They never drilled every combo.

Kids got 5-10 trials a day in play-based sessions. The study ran until each child could say new combos without teaching them.

Four kids started saying untrained phrases after the diagonal training.

One child needed a second round with more matrix cells.

All kids kept the new skill two weeks later.

Kemmerer et al. (2021) looked at 40 matrix studies and found big differences in how grids are built. Frampton’s small grid and diagonal-only teaching match the leaner end of that range.

Polo-López et al. (2014) used conditional-discrimination drills and saw mixed generalization. Frampton’s matrix gave clearer gains, likely because noun-verb combos are easier to recombine than abstract auditory-visual pairs.

Tullis et al. (2021) also got emergent language, but they needed tact and match-to-sample plus extra instructive feedback. Frampton shows a simpler path: teach the diagonal and let the matrix do the work.

You can grow a child’s phrase list without drilling every pair. Pick 6-8 nouns and 6-8 verbs. Build a grid. Teach the diagonal. Probe the rest. If the child stalls, add more cells instead of switching methods. This saves session time and keeps therapy fun.