Word recognition in auditory cortex
Although spoken word recognition is more fundamental to human communication than text recognition, knowledge of word-processing in auditory cortex is comparatively impoverished. This dissertation synthesizes current models of auditory cortex, models of cortical pattern recognition, models of single-word reading, results in phonetics and results in language acquisition to form a novel model of auditory word-form recognition. The model is expounded and assessed empirically in two studies and discussed with respect to results from clinical neuroscience.Study 1 uses a series of meta-analyses to quantitatively synthesize functional brain imaging results. The analytical method assesses the anatomical concordance of prior results for a given experimental paradigm. Through consideration of multiple paradigms and stimulus types, it assesses concordance for several experimental phenomena pertinent to aspects of the model presented. Two novel findings are reported. First, a processing cascade for speech sounds is observed in left auditory cortex, extending from mid superior temporal gyrus (STG) to anterior STG to anterior superior temporal sulcus (STS). This corresponds with increasing scales of speech sound processing, from phonemes to words to phrases. Second, correlates of selectivity and invariance for speech sounds are observed within the same left mid-to-anterior STG region, the former being with respect to both artificial control stimuli and non-speech natural sounds, the latter being with respect to category-specific neural adaptation.Study 2 tests predictions about the types of speech sounds encoded by auditory cortex. Data from linguistic corpora are used to model the natural statistics of the heard phonetic environment. These statistics are then used to characterize the regularity of phoneme orderings within non-word stimuli. Two novel results are reported. First, using functional magnetic resonance imaging (fMRI), sensitivity to the natural statistics of phoneme ordering is observed in left anterior STG, as predicted by distributional learning and hierarchical feature coding. Second, correlation is observed between activity in left anterior-lateral planum temporale (PT) and phoneme-level prediction error for an ideal observer, consistent with predictive coding. According to the model presented, PT is the site of sub-phonemic representation. Error signal in PT is therefore suggestive of inaccurate top-down contextual predictions for acoustic-phonetic features.
Showing items related by title, author, creator and subject.
Chevillet, Mark Allen (Georgetown University, 2011)Current theories of speech perception in the human brain propose two separate, hierarchically organized cortical processing streams, a ventral and a dorsal stream. The ventral stream is generally believed to mediate acoustic ...