Project Summary/Abstract
This project addresses a topic of critical significance in advancing research on speech communication in children
and adults: the adaptive mechanisms that enable listeners to navigate cross-talker variability in human speech.
Talkers’ pronunciations of segments, syllables, and words—and thus ultimately meaning—can differ
substantially depending on physiology, social identity, and language background. Scientific breakthroughs over
the last four decades have determined that listeners accommodate these differences by adapting their
interpretation of incoming speech for different talkers. However, what perceptual, linguistic, and cognitive
mechanisms achieve these adaptive changes has remained unclear. This status quo is recognized as a
major roadblock in advancing the field, limiting effective knowledge transfer to clinical and physiological research.
To address this question, we capitalize on two recent innovations from our team: a comprehensive
computational framework of adaptive speech perception (ASP) and a model-guided approach to high-
powered contrastive experiment design. ASP—for the first time—implements competing hypotheses about
how changes in perception, linguistic representations, and decision-making might jointly or separately contribute
to spoken language understanding. The model-guided design approach allows contrastive tests of these
hypotheses against human behavior. We use this approach to investigate one of the most complex adaptive
feats of human speech perception: adaptation to unfamiliar non-native accents. In 16 large-scale web-based
experiments (total N > 2,880 listeners), we investigate what combinations of mechanisms best explain such
adaptation, and how the engagement of different mechanisms depends on 1) specific acoustic-phonetic
properties of the accent, 2) the duration and amount of input, ranging from the first utterances to repeated
exposure over up to 12 months, and 3) cognitive demands and contextual affordances. Independent of these
primary goals, the proposed experiments make significant empirical, methodological, and theoretical
contributions to research on the perception of accents. This includes a new production database of Mandarin-
accented speech (the 2nd most commonly heard non-native accent in the US, incl. in healthcare contexts), new
insights on the perceptual and cognitive factors supporting accent adaptation; and new comparisons of
adaptation to different segmental and prosodic properties, some of which may require more exposure than
others. Our longitudinal testing extending over 12 months will be the critical first step towards understanding how
rapid adaptive changes in the first few minutes of exposure could lead to the well-known, long-lasting benefits of
talker and accent familiarity in speech perception. Finally, our data and computational models—both shared
without embargo in the form of open-access R libraries—will serve as a key resource for future experimental and
computational research, supporting effective and rigorous hypothesis testing.