Phonology impacts segmentation and generalization in speech processing
Luca Onnis, Padraic Monaghan, Nick CHater, & Korin Richmond
Abstract
A principal concern in language acquisition research is the nature of the computations involved in the child learning the structure of the language.
Theories differ over whether processing is dependent on statistical computations alone or whether algebraic computations are required.
Pena, Bonatti, Nespor, and Mehler (2002) recently argued that speech segmentation is a statistical process, whereas learning structural generalizations
is due to algebraic computations. We attempted to test the generality and robustness of Pena et al.'s results in a series of artificial language learning
experiments. We found that participants had strong preferences for phonemes in certain utterance positions, which overwhelmed any evidence for learning
the structure in the artificial language. Consequently, we found no evidence for the statistical/algebraic distinction: the results from Pena et al.
were a consequence of the impact of phonological preferences on language processing. We reassess the debate on algebraic versus statistical computation
in the light of our results.
Please email suggestions/errors to lo35@cornell.edu |