| |
Abstract:
Phonology rules that 'garden' may be pronounced 'gardem,' if
it occurs before 'bench,' but not before 'chair.' Gaskell and
Marslen-Wilson (1998) argued that listeners filter out such
variation at an abstract phonological level. 'Garden-bench' and
'gardem-bench' should thus sound more alike than 'garden-chair' and
'gardem-chair' due to phonological inference. To test this idea, we
used a passive oddball task while measuring mismatch negativity
(MMN). If lawful variation is filtered out, a minor or no MMN is
expected in response to a lawful variation (standard: garden bench,
deviant: gardem bench), but a significant MMN is expected if the
same variation is not lawful (standard: garden-chair, deviant:
gardem-chair). The results revealed no MMN in any condition,
despite consistent evoked potentials (EP) patterns over subjects;
e.g., the EPs in the 'chair' and the 'bench' conditions differed,
possibly reflecting the different amplitude envelopes of the two
words. The results of a follow-up word-identification experiment
revealed a positive bias for a perception of a final /m/ ('gardem')
in the 'chair' context, while favoring /n/ in the 'bench' context,
thus possibly abolishing an MMN effect. These data support Gaskell
and Marslen-Wilson's assumption that phonological variation is
filtered out at a perceptual level. However, in contrast to Gaskell
& Marslen-Wilson's assumption this filtering may be based on
general acoustic contrast rather than speech-specific phonological
inference.
|
