The traditional model for language assumes a level of competence that
represents words composed of distinctive features that are invariant units
drawn from a small, universally specified set. The present work tests two
of these assumptions, representational invariance and the assumed small set
size of these features.
This study trained naive listeners on artificially lengthened voice-onset
times in specific target words and tested whether pretraining and
posttraining perceptual boundary locations differed. Results indicate
subjects were successfully trained on the target words, but this effect did
not generalize to other, structurally similar words. In addition, robust
frequency and lexical effects are reported, but do not apparently interact
with training. As predicted, frequency effects suggest subjects preference
for high-frequency words over low-frequency words, but, somewhat
surprisingly, lexical effects suggest a preference for non-words over words.
Results undermine foundational assumptions of the traditional model,
specifically the assumptions of representational invariance and a small set
size. To account for these results, an exemplar memory model is discussed
in which no principled restrictions are imposed on memory for language. The
exemplar models discussed are able to account for the facts reported here.
|
E-mail this message to a friend
