The Shape of Formant Space

2680 and 1673 vowels were measured for the two speakers,^6.13 according to the methods described in Chapter 5; outlying measurements were examined individually and either corrected or confirmed. As argued in the Acoustics chapter, measurements of the first two formant frequencies reflect the articulatory dimensions of mouth-opening and of tongue-body backness and lip rounding. All the F1, F2 measurements for the two speakers are displayed in Figures  and  .

Figure: 2680 F1/F2 measurements.

38#38

Figure: 1673 F1/F2 measurements.

39#39

Comparing the overall shape of the vowel space (that is, the distribution of measurements of F1, F2 tokens) across dialects shows some real differences in the envelope of acoustic variation (Labov 1991).

These vowel spaces differ from those of other dialects in that the location of maximum density of tokens -- that is, the mode of the distribution -- is closer to the bottom corner of the triangle. The Chicano vowel space has two modes in the high front and high back; while the Chicagoans' vowel spaces are more evenly distributed. None of the other speakers, including the Alabama speaker, has a mode in the low corner. Since both Juba and Roasta have modes in the low corner, this feature is not an individual idiosyncrasy.

Juba's vowel space, which was analysed most extensively, is quite different from those found for the other English dialects (pages ff, , ). The shape of this vowel space is roughly triangular, or rather, V-shaped, with tokens relatively sparsely distributed at the top, and in the middle of the V. The paucity of tokens in the upper-middle region of the distribution is consistent with the claim that the high-central region of vowel space is typically unoccupied (cf. Liljencrantz & Lindblom, 1972). The distribution of tokens is densest on the front and back edges of the triangle, and more dense towards the bottom. Given that the overall shape is a triangle, two rules summarize the picture: the lower, the denser; the closer to the front or back edge, the denser. This makes the distribution look like a bottom-heavy V shape.

As with all speakers studied here, there is some front-back asymmetry in the overall envelope of variation. The distributions are rather more dense in the high-front than in the high-back, and the back edge of the triangle is possibly more vertical than the front edge (a difference which may be eliminated by transforming the scale to barks, mels, or a logarithmic scale.)