Illustration of the shift of view of vowels from points (top) to vectors originating at a vowel space center point (bottom). Vowel formant measurements were obtained from Peterson and Barney (Ref. 23).
Vowel angles obtained from the data set of Peterson and Barney (Ref. 23). Within each vowel, the group of box plots show angles obtained from children (left), males (middle), and women (right).
Box plots of VFD values categorized by vowel space corners. The corner categories are defined bythe angle component of the vowelvector: , , , and .
The relative frequency (in percent) for which statistical testing of a homogenous data set divided at random into two sets would indicate a significant difference between the sets at the 0.05 level. The metric on which the testing was conducted is indicated on the horizontal axis. The 5% level expected in an unbiased testing is indicated by a dashed horizontal line. A dotted white line indicates a 1% level for the oppose of visual reference.
Obtained probabilities of no difference being detected in statistical testing (one tailed t-test) between the full vowel space (Ref. 24) and a reduced version of the same vowel space, categorized by metric (right side) and the range of number of productions of each vowel included in the calculations (top). The scaling factor (percent of original vowel distance from the vowel space centroid) applied to compute the reduced vowel space is shown on the x-axis, and the probability of the two sets being indicated to be the same (the p-value) is shown on the y-axis. The 5% alpha level where a statistically significant difference would predominantly be reported is indicated by a dashed horizontal line.
Article metrics loading...
Full text loading...