Abstract
Pronunciation time probability density and hazard functions from large speeded word naming data sets were assessed for empirical patterns consistent with multiplicative and reciprocal feedback dynamics – interaction dominant dynamics. Lognormal and inverse power law distributions are associated with multiplicative and interdependent dynamics in many natural systems. Mixtures of lognormal and inverse power law distributions offered better descriptions of the participant’s distributions than the ex-Gaussian or ex-Wald – alternatives corresponding to additive, superposed, component processes. The evidence for interaction dominant dynamics suggests fundamental links between the observed coordinative synergies that support speech production and the shapes of pronunciation time distributions.
Highlights
MethodsHolden et al (2009) Experiment 2 data set is composed of 30 participants’ pronunciation times to 1100 randomly selected single and multisyllabic English words that ranged from 4 to 15 letters in length with an average frequency of occurrence of 70.2 times per million (SD = 295.16) according to the Kucera and Francis (1967) norms
The cocktail description captured the vast majority of the individual English Lexicon Project (ELP) pronunciation time distributions
These analyses indicate that the refined cocktail model successfully describes the empirical pronunciation time distributions for a large majority of individual participants
Summary
Holden et al (2009) Experiment 2 data set is composed of 30 participants’ pronunciation times to 1100 randomly selected single and multisyllabic English words that ranged from 4 to 15 letters in length with an average frequency of occurrence of 70.2 times per million (SD = 295.16) according to the Kucera and Francis (1967) norms. Distribution fitting Standard maximum-likelihood estimation methods were applied to the empirical distributions to estimate a set of density function parameters for each participant’s pronunciation time distribution. We applied all the same statistical methods described in the previous experiment to the present data set. We again benchmarked our hazard discrimination routine by replacing, in turn, each of the 470 participant’s empirical pronunciation times with synthetic data derived from the best-fit parameters of the cocktail, the exGaussian, and ex-Wald models. Replacing the empirical data with data from known models allowed us to assess the ability of the hazard analysis to discriminate the power law from exponential models
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
