Reversal shift in phonotactic learning during language production: Evidence for incremental learning

Speakers implicitly learn novel phonotactic patterns while producing strings of syllables. The learning is revealed in their speech errors. For example, if /f/ is artificially restricted to the syllable onset position and /s/ is restricted to the coda position, speakers’ slips will respect these constraints, demonstrating learning. The mechanism behind this learning was investigated by reversal shift. In two experiments, speakers produced syllable sequences for 32 trials in which one consonant was constrained to be an onset and one, a coda. Then for the next 32 trials, the constraints were reversed; for example, /f/ is now a coda and /s/ an onset. Over a third block of 32 trials, the constraints were reversed again back to the original rule. The learning of the reversed rules was slow in comparison to rapid learning of the original rule, providing evidence that the underlying mechanism is incremental. To learn a reversed rule, first one must unlearn the original association before making progress on the new rule. This pattern of rapid initial learning and slow reversal learning was replicated in a nonspeech analogue of the syllable production task. Participants pushed sequences of buttons, in which finger presses corresponded to consonants and thumb presses to vowels. A control experiment found that non-reversal shifts between blocks are easily learned, showing that the difficulty is specifically with reversal shifts. A final experiment provided evidence that learning in these experiments is greater when productions are more errorful.