First-impression bias in auditory processing as a window to perceptual inference and learning

The auditory system has a sophisticated ability to detect when prior regularities in sound change even without conscious attention. This automatic change detection is demonstrated by the mismatch negativity (MMN) component of the event-related potential that is elicited following any irregular change (deviation) from an established pattern in sound. Variability in MMN amplitude is believed to reflect a relevance-filtering function, weighted by the relative probabilities of sounds at any given time in order to prioritise neural resources to the most unexpected deviations. In this thesis, we use modulations in MMN amplitude to investigate a recently revealed “first-impression bias” in how predictions underlying sensory filtering are formed. This bias represents a profound higher-order modulation of neural responses to sensory information which is in contrast to a traditional understanding of perception as a passive, bottom-up process and speaks to sophisticated top-down hierarchical learning processes. Four diverse studies were conducted to test the ability of this bias to inform stimulus-driven/afferent mechanisms, top-down modulations, and clinical impairments in perceptual inference respectively. In each case, first-impression bias is successfully exploited to elucidate higher-order aspects of perceptual inference and learning that cannot be revealed by MMN alone. It is argued that to adequately account for these results requires the implementation of traditional predictive coding accounts of MMN generation and general brain function within a more sophisticated hierarchical learning scheme informed by more recent computational and Bayesian learning models. In this way the results support the versatility of first-impression bias as a window to the various higher-order mechanisms which contribute to perception and inferential learning.