On the learning of speechlike vocabularies

Summary A series of experiments dealing with the learning of ensembles of speechlike acoustic stimuli was performed. The stimulus ensembles differed with respect to the number of physical dimensions that were manipulated in generating the stimuli, and with respect to the extent to which the stimuli resembled speech. Results show that performance during learning is better when each stimulus is encoded into several physical dimensions than when the stimuli lie along a unidimensional continuum. Furthermore, as the stimuli become more like speech there is a deterioration of performance during learning, with the exception that performance is best when the stimuli are actually speech signals. Implications for theories of speech perception are discussed.