Phonetic and phonemic predictions in auditory memory

Hestvik, ArildIn natural language, speech sounds can be represented at several levels of mental representation – from acoustic to phonetic to phonemic. In order to learn new phonetic-phonemic mappings, the auditory system must be sensitive to sub-categorical phonetic differences, it must be able to extract phonetic generalizations from varying input, and it must be able to generate new representations from those generalizations. This dissertation has two primary goals: to determine the nature of auditory predictions generated in response to varying speech sounds, and to disentangle effects of prediction from effects of gradient auditory encoding in the neural processes indexed by the neurophysiological response, mismatch negativity (MMN). ☐ To achieve these goals, I conduct several experiments using the varying standards oddball paradigm and the random standards control condition introduced by Jacobsen and Schröger (2001). The central question of this dissertation is whether the predictions made by the auditory system in response to varying input are categorical (phonemic) or gradient (phonetic) in nature. The varying standards paradigm has been argued to constrain the formation of memory traces (and predictions). I test this claim by manipulating the phonetic distance between standards and deviants, operationalized as differences in Voice Onset Time (VOT). ☐ Two competing hypotheses are entertained. If the predictions made in response to varying input sounds are purely categorical, there should never be a modulation in MMN amplitude predicated on phonetic distance. At the phonemic level, all category members are equivalent. When evaluating a phonemic prediction, phonetic values should not affect how well the prediction has been satisfied – except insofar as they determine category membership. On the other hand, if the predictions made in response to varying input are gradient and phonetic, then the phonetic distance manipulation should result in a modulation in MMN amplitude predicated on phonetic distance. In this view, prediction evaluation is partially a function of phonetic distance. Comparing the phonetic value of the memory trace to the phonetic value of the deviant determines how well the prediction has been satisfied. A greater distance should result in a higher amplitude MMN. ☐ This dissertation will also disentangle the effects of prediction from the effects of auditory encoding. To achieve this, deviants are compared to an identical stimulus in a context where no prediction is made. By comparing deviants to themselves, auditory encoding is held constant. The only differences in brain response must be attributable to differences in prediction evaluation. ☐ In five experiments, while controlling for the effect of gradient auditory encoding, I observe no reliable phonetic distance effects. I conclude that while perception is gradient and phonetic as evidenced by gradience in the brain responses to stimuli with different VOT values, prediction is categorical and phonemic. The auditory system is sensitive to sub-categorical phonetic differences in varying input, but that sensitivity does not result in a phonetically detailed auditory representation of the input or a phonetic prediction. The underlying, pre-attentive prediction mechanisms in auditory sensory memory do not track phonetic contrasts that do not correspond to known categories.University of Delaware, Department of Linguistics and Cognitive SciencePh.D