| |
Abstract:
Native Japanese speakers often have persistent difficulty
discriminating English /r/ and /l/, even after years of immersion
in the US. A connectionist model (Thomas & McClelland, 1997)
suggests that rather than being the product of a biological
critical period, this phenomena can be explained by a Hebbian
principle that acts to reinforce the tendency to hear examples of
this non-native speech contrast as instances of a single perceptual
category. Two implications of this model are tested in a training
study: a) enhancing the discriminability of /r/ and /l/ stimuli
should induce rapid learning in Japanese adults, and b) this
perceptual learning should not depend on explicit feedback. A
resynthesis technique was applied to natural speech samples (e.g.
load, road) to produce intermediate and exaggerated versions of the
stimuli in a multi-step continuum. An "adaptive" training procedure
initially presented subjects with exaggerated stimuli, then
gradually reduced the exaggeration after every 8 correct responses.
A "set" training procedure presented non-exaggerated stimuli. In
all trials, subjects heard one of two stimuli and labeled them as
either starting with R or L without any feedback. Only the adaptive
procedure produced dramatic improvements across 3 twenty-minute
training sessions, and improvements partially generalized to
another resynthesized continuum (e.g. rock, lock).
|
