Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
What we do (not) know about the mechanisms underlying adaptive speech perception
There is a Poster PDF for this presentation, but you must be a current member or registered to attend SNL 2023 to view it. Please go to your Account Home page to register.
Poster A75 in Poster Session A, Tuesday, October 24, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
Xin Xie1, T. Florian Jaeger2, Chigusa Kurumada2; 1University of California Irvine, 2University of Rochester
One of the central questions in the neuroscience of language concerns the mechanisms by which listeners perceive and categorize speech amidst its inherent variability. Empirical research over the past two decades has documented that listeners make rapid, and potentially long-lasting, adaptive changes to accommodate a priori unfamiliar talkers or accents. However, the mechanisms underlying this remarkable adaptivity remain largely elusive. Studies have often attributed the adaptivity to mechanisms such as auditory signal transformations (Tang et al., 2017; Johnson & Sjerps, 2021) or the learning or “remapping” of cues to linguistic categories (Myers & Mesite, 2014; Blanco-Elorrieta et al., 2021). However, these hypotheses have not been contrasted with each other, limiting our knowledge of the unique contribution of each mechanism as well as the potential interplay between different mechanisms. To address this critical gap, we present a novel analytical framework aimed at formalizing the mechanisms underlying adaptive speech perception (“ASP” for brevity). ASP is the first comprehensive framework that integrates three mechanistic models of speech categorization: (a) low-level, prelinguistic auditory normalization; (b) mapping from prelinguistic percepts to linguistic representations; and (c) psychometric lapse-bias model of decision processes. Crucially, ASP also formalizes how each of these levels of mechanisms can dynamically change in response to recent experience ("change models"). We tested the predictions of the three change models using two standard paradigms: perceptual recalibration (Norris et al., 2003) and natural accent adaptation (Bent & Baese-Berk, 2021). Here we focus on the former. In this paradigm, listeners receive acoustically ambiguous tokens in a lexical context (e.g., crocodile), simulating talkers with shifted acoustic distributions. Listeners subsequently categorize phonetic input sampled along a continuum (e.g., /d/ vs. /t/) according to the exposure input, demonstrating a corresponding boundary shift. All three models predicted the characteristic boundary shift observed in human categorization responses. Simulations of a non-native accent adaptation experiment replicated this finding: even a computationally parsimonious mechanism (e.g., cue normalization or decision bias change) predicted data patterns often considered indicative of exposure-related changes in category representations. Our results thus demonstrated that common empirical results have limited diagnostic power when it comes to discerning the mechanisms of adaptive speech perception. In summary, our simulation studies underscore the pressing need to reevaluate current standards of empirical testing. In this presentation, we will explore how a formal modeling framework, such as ASP, can facilitate simulation-based experimental designs, enabling better differentiation of predictions derived from the three mechanisms. Crucially, this approach will advance the standards of neuroimaging research by generating more concrete—and thus more testable—predictions about the types of computations taking place in different brain regions or networks. Additionally, we will discuss how future behavioral and imaging studies can leverage ASP to investigate the interconnected auditory, perceptual, and cognitive processes that underlie complex human behavior in the face of input variability. By unraveling the puzzle of adaptive speech perception, we can gain a deeper understanding of the neural underpinnings of speech processing in naturalistic contexts.
Topic Areas: Speech Perception, Computational Approaches