Abstract EN:

Participants had to generalize repetition-based structures or ordinal structures without repetitions. General symbolic mechanisms should generalize both structures equally well; associationist mechanisms should process the ordinal structures better. Participants readily generalized repetition-based but not ordinal structures, structure changes elicting rapid ERPs only for repetition-based structures. When they found that participants could generalize repetition-based structures only when the repetitions were in sallient positions (in sequence edges but not middles). We then studied the limitations of more language-related computations. Participants learned phonotactic contraints in word-edges but not middles. In the next experiments, participants extracted category-based generalizations from speech streams, but only when the critical syllables were in word edges but not middles. In conclusion, the mind may use a computational toolbox of specialized and constrained perceptual primitives.