Intraspeaker Comparisons of Acoustic and Articulatory Variability in American English /r/ Productions

The purpose of this report is to test the hypothesis that speakers utilize an acoustic, rather than articulatory, planning space for speech production. It has been well-documented that many speakers of American English use different tongue configurations to produce /r/ in different phonetic contexts. The acoustic planning hypothesis suggests that although the lr/ configuration varies widely in different contexts, the primary acoustic cue for /r/, a dip in the F3 trajectory, will be less variable due to tradeoffs in articulatory variability, or trading relations, that help maintain a relatively constant F3 trajectory across phonetic contexts. Acoustic data and EMMA articulatory data from seven speakers producing /r/ in different phonetic contexts were analyzed. Visual inspection of the EMMA data at the point of F3 minimum revealed that each speaker appeared to use at least two of three trading relation strategies that would be expected to reduce F3 variability. Articulatory covariance measures confirmed that all seven speakers utilized a trading relation between tongue back height and tongue back horizontal position, six speakers utilized a trading relation between tongue tip height and tongue back height, and the speaker who did not use this latter strategy instead utilized a trading relation between tongue tip height and tongue back horizontal position. Estimates of F3 variability with and without the articulatory covariances indicated that F3 would be much higher for all speakers if the articulatory covariances were not utilized. These conclusions were further supported by a comparison of measured F3 variability to F3 variabilities estimated from the pellet data with and without articulatory covariances. In all subjects, the actual F3 variance was significantly lower than the F3 variance estimated without articulatory covariances, further supporting the conclusion that the articulatory trading relations were being used to reduce F3 variability. Together, these results strongly suggest that the neural control mechanisms underlying speech production make elegant use of trading relations between articulators to maintain a relatively invariant acoustic trace for /r/ across phonetic contexts.