Temporal Processing for Syntax Acquisition: A simulation study

Early perceptual processing capabilities are likely to contribute to the categorization of lexical vs. grammatical words by newborns. This lexical categorization could be performed by detecting differences in the prosodic structure of these word categories. Here we demonstrate that a Temporal Recurrent Network (TRN) that allows realistic treatment of the dynamic temporal aspect of prosody performs this lexical categorization task on French and English. We then examine the functional relation between this capability, and non-linguistic temporal discrimination. We reduce sensitivity to temporal structure in the TRN by increasing the network time constants. This yields (1) a reduction in performance in the lexical categorization task, and (2) a deficit in the processing of brief auditory events similar to that observed for children with SLI. While our principle result is that the TRN can perform lexical categorization based on prosodic structure, it is of interest that the impaired TRN suggests a functional link between impaired temporal processing, and impaired lexical categorization in SLI.