Hizlari gonbidatua: Shawn Cummings. Perceptual learning and the statistics of speech input

What: Perceptual learning and the statistics of speech input

Where: Online talk; BCBL Auditorium Zoom room (If you would like to attend to this meeting reserve at info@bcbl.eu)

Who: Shawn Cummings. PhD, University of California, Irvine, US

When:  Thursday,  Sep 25th at 03:00 PM.

Different talkers produce different acoustics in service of the same intended meaning. While at face value this ‘lack of invariance’ poses a massive challenge to comprehension, we often nevertheless understand one another with ease. Furthermore, when input misaligns with our expectations, we can rapidly adjust the way we perceive subsequent speech. This talk will discuss two seemingly contradictory facets of a particular paradigm designed to elicit such adjustments. First, learning is rapid and robust, being measurable after just a handful of words and remaining in the face of competing task demands and distractors. Second, the magnitude of learning is hugely variable between different input talkers, with some talkers eliciting no effect. Possible reconciliation of these factors lies in a theory of cumulative input sensitivity, where listeners continually update their beliefs but are exquisitely sensitive to the acoustic details of the input they receive. Through a series of studies using a novel large-scale dataset including a corpus of systematically manipulated speech from 16 talkers, behavioral outcomes are compared computational simulations reliant on Bayesian belief updating. A successful link herein promises to inform our theoretical understanding of the mechanisms that allow us to understand one another and the world around us.