Mapping Speech Sound to Mental Representation : Neurophonological Evidence from Event-Related Brain Potentials

Abstract

To date it is unclear how the brain deals with the enormous variation in speech and how words are recognized with such varying input. It also remains unknown what exactly is stored in the mental lexicon and how detailed phonological information must be represented to ensure fast and adequate speech comprehension and processing. Different kinds of psycholinguistic models have been proposed in dealing with such variation and how speech sounds are stored in the mental lexicon. This dissertation aims to resolve this question by using electrophysiological methods in the investigation of the neuronal representation of phonological information.Chapter 1 gives an introduction to four different models of speech perception and representation. The account, which is tested in this research the Featurally Underspecified Lexicon Model (FUL; Lahiri & Reetz, 2002, 2010) assumes abstract and underspecified representations of phonological features. When dealing with variation, all models except for the FUL model assume full specification of lexical form that relies on experience or context. The Mismatch Negativity, an automatic change detection response in the brain, has been shown to be a powerful tool in this line of research as it enables the investigation of the representation of specific phonological features by contrasting speech sounds. Five MMN experiments were set up to test predictions of the FUL model in contrast to other models that differ in the degree to which phonological feature information (i.e. place and manner of articulation) is specified or underspecified. Only models assuming underspecified phonological features such as FUL predict an asymmetry of MMNs within the reversal of sound contrasts, presented as standard and deviant. For instance, sounds that are underspecified regarding a certain feature, cannot build up a representation when presented as standard. Consequently they do not conflict with the incoming deviant sound, which is reflected in reduced MMN amplitudes. In the reversed case, when a sound is assumed to be specified for a specific feature, it will be pre-activated by the standard. Therefore a conflict occurs with other incoming deviant sounds, which will elicit larger MMN responses. Models proposing full storage of all phonological information would predict equal MMNs within the reversal of sound contrasts.Chapter 2 presents the general methods used in all of the altogether six experiments. Adult subjects were presented with speech sounds embedded into words, nonwords, and syllables that were contrasted, each serving as standard and deviant in an MMN design.In Chapter 3 the studies with their results are presented and discussed. More specifically, the MMN Experiments 1-5 tested the FUL model s assumptions of the different place and manner features. The last three MMN experiments (3-5) were aimed at generalizing the underspecification account to other feature dimensions such as manner of articulation. Experiment 6 was a behavioral discrimination study to control for acoustic and attentional effects of all the consonantal sound contrasts used in Experiments 3-5. Breaking new ground by looking at different feature dimensions in this dissertation such as place and manner of articulation with different speech sounds, provides further support for the notion that mental representations of phonological features are not a one-to-one relation between the acoustic speech signal and their mental representations. Rather phonological information may be more abstract compared to theories suggesting full storage of all phonological details. These reported MMN asymmetries between conditions for the same acoustic contrasts in Chapter 3 clearly suggest that the brain refers to underspecified phonological representations as a basic principle in the functional organization of the mental lexicon during speech perception.Chapter 4 gives a general discussion of all results, discussing them in the light of different psycholinguistic models and further influencing factors such as acoustics and frequency effects.