A unified account of categorical effects in phonetic perception

Abstract

Categorical effects are found across speech sound categories, with the degree of these effects ranging from extremely strong categorical perception in consonants to nearly continuous perception in vowels. We show that both strong and weak categorical effects can be captured by a unified model. We treat speech perception as a statistical inference problem, assuming that listeners use their knowledge of categories as well as the acoustics of the signal to infer the intended productions of the speaker. Simulations show that the model provides close fits to empirical data, unifying past findings of categorical effects in consonants and vowels and capturing differences in the degree of categorical effects through a single parameter.

Perception in a Variable but Structured World: The Case of Speech Perception

“…For extreme cases, like the case of selective adaptation studied here, where there is no (or minimal) variance in the physical signal, this assumption leads to implausible asymptotic behavior: with more and more exposure, the category should shrink to nothing, leading to an asymptotic aftereffect of −1. Of course, between 128 and 256 exposures, the negative aftereffect induced by selective adaptation appears to continue to grow stronger (Figure 2.8 Clayards et al, 2008;Feldman, Griffiths, & Morgan, 2009;Kronrod, Coppess, & Feldman, 2012), this is expected to reduce or eliminate predicted effects of variation in physical variance.…”

“…This suggests that listeners not only have beliefs about the cue distributions for a particular single talker (as suggested by previous work, Clayards et al, 2008;Feldman, Griffiths, & Morgan, 2009;Kronrod et al, 2012), but also have implicit beliefs about the ways in which talkers tend to differ from each other, and hence what to expect from an unfamiliar talker.…”

“…Figure 4.1: Each subject heard one of these five synthetic accents, which differ only in the distribution of VOTs of the word-initial stops. Black dashed lines show VOT distributions from a hypothetical typical talker (as estimated by Kronrod et al, 2012). Note that the 0 and 10ms shifted accents are reasonably close to this typical talker, while the -10, 20, and 30ms shifted accents deviate substantially.…”

“…Adaptation was assessed based on listeners' classification function, or how they labeled each VOT as /b/ or /p/, which was continuously assessed throughout the experiment. For a typical talker, we use the distributions reported by Kronrod et al (2012), who determined the underlying means and variances that best explained a combination of listeners' VOT classification and discrimination behavior. These distributions are qualitatively very similar to those reported by Lisker and Abramson (1964), based on productions from four talkers.…”

“…Kronrod et al (2012) also use a Bayesian model to estimate the underlying distributions of VOT for /b/ and /p/ based on comprehension behavior.…”