The information presented here was collected from two large vowel datasets,
and Barney (1952) and
Hillenbrand et al. (1995). This data comprises
productions from 215 unique speakers, 76 adult females, 78 adult
males, and 61 children of either sex.
The figures below show average productions for different vowel
categories from two large datasets.
Individual tokens are plotted below. Obviously, there is a lot of
variation around vowel-category means. However, ellipses generally point
in the same direction and are closely aligned with the lines on the
figures. These lines indicate the direction of variation in which F1 and
F2 increase by a single scale-factor.
||If all of the tokens in the figures
above are centered around their category-means and superimposed,
we get a figure like the one on the right.
The ellipse enclosing tokens shows the orientation of the
distribution, the solid line is the line in which variation in F1 is
equal to variation in F2, and the dotted line indicates the
major axis of the ellipse (denoting the primary direction of
variation). The major axis of the ellipse differs from the solid
line by only
This suggests that within-category variation in formant-patterns
is primarily according to a single scale-factor. In fact, an
analysis presented in
Barreda & Nearey
(2013) indicates that around 80% of variation in
formant-patterns occurs along this dimension.