Incomplete Neutralization and Task Effects in Experimentally-elicited Speech: Evidence from the Production and Perception of Word-final Devoicing in Russian

Abstract

This dissertation investigates the role of grammatical versus methodological influences in the production and perception of final devoicing in experimentally-elicited speech from Russian. It addresses the question of how the partial preservation of the phonological voicing contrast in word-final obstruents is affected by (i) task-independent factors that reflect phonological and lexical properties of stimuli words (underlying voicing, word length, lexical competition) and (ii) task-dependent biases that arise due to the nature of the experimental task performed by the speaker (availability of orthographic inputs, presence of minimal pairs among the stimuli). Results of a series of acoustic production and perceptual identification tasks reveal that task-dependent factors account for the presence of robust and perceptually salient differences in the parameter of phonetic voicing. Several types of stimuli items also show limited but statistically significant differences in closure/frication duration and release duration that are independent of the presence of orthography or inclusion of full minimal pairs among test items. Taken together, these findings indicate that non-grammatical factors can play a prominent biasing role in both production and perception of the voicing contrast in experimentally-elicited speech, such that certain voicing-dependent cues are maintained only in the presence of task-dependent pressures. However, not all incompletely neutralized differences between phonologically voiced versus voiceless final obstruents can be attributed to the effects of orthography or inclusion of minimal pairs among the stimuli. In the theoretical domain, these results are argued to favour a less restrictive definition of neutralization and a model of phonology that views devoicing as a loss of the primary acoustic cue to the underlying voicing contrast rather than complete identity of the [voiced] feature.