The Timecourse of Generalization in Phonotactic Learning

Abstract

There is considerable evidence that speakers show sensitivity to the phonotactic patterns of their language. These patterns can involve specific sound sequences (e.g. the consonant combination b-b) or more general classes of sequences (e.g. two identical consonants). In some models of phonotactic learning, generalizations can only be formed once some of their specific instantiations have been acquired (the specific-before-general assumption). To test this assumption, we designed an artificial language with both general and specific phonotactic patterns, and gave participants different amounts of exposure to the language. Contrary to the predictions of specific-before-general models, the general pattern required less exposure to be learned than did its specific instantiations. These results are most straightforwardly predicted by learning models that learn general and specific patterns simultaneously. We discuss the importance of modeling learners’ sensitivity to the amount of evidence supporting each phonotactic generalization, and show how specific-before-general models can be adapted to accommodate the results.