Fig. 5.

(A) Distributions of VOT and VL values for voiced and voiceless stops from the production data in Allen & Miller (1999). The locations of the lines indicate the mean of each category along each dimension, and the lengths of the lines are equal to the standard deviations of each distribution. Voiced and voiceless sounds are primarily distinguished along the VOT dimension, but there is also a small difference along the VL dimension. (B) Identification responses from listeners for sounds varying in both VOT and VL from McMurray et al. (2008). Listeners tend to identify sounds as voiced (/b/) for low VOTs and voiceless (/p/) for long VOTs. The shift in the identification function for the two different VL conditions indicate that they also use VL information when making voicing judgments (i.e. they are more likely to identify sounds as voiced for long VLs and voiceless for short VLs).