Recombinative generalization of subword units using matching to sample.

The team built a computer matching game. Adults saw a nonsense syllable on the screen. They picked the matching onset-rime chunk from four choices.

After each correct pick, the screen showed the full new word. The program mixed in trials that asked adults to pick the whole word they just saw.

All 25 adults could already read. The study wanted to know if the game would teach them to build brand-new words from the trained chunks.

Adults learned to match the chunks. More important, they could read new words they had never seen. The chunks recombined on their own.

No one taught the final words directly. The simple act of matching parts, then seeing the whole, sparked recombinative generalization.

Hansen et al. (1989) showed that adults with intellectual disability need both sample and comparison training. Catherine et al. used the same two-part idea, but with literate adults and reading chunks.

Marya et al. (2021) got the same recombinative magic with matrix training and speech devices. Different method, same outcome: teach a few pieces, get many new responses.

Tenneij et al. (2009) tweaked MTS by showing choices first. Catherine et al. tweaked it by adding whole-word trials. Both fixes cut errors and boosted learning.

You can steal the whole-word trial trick tomorrow. When you run MTS for sounds, letters, or signs, slip in trials that ask for the finished product. It costs zero extra time and may give you free generalization. Try it with kids who know some phonics but freeze on new words. Five extra trials could unlock dozens of untaught readings.