What are the effects of phonological vowel length on vowel quality? Insofar as formant-frequency measurements reflect perceptible vowel quality, we may explore these effects by plotting such measurements for pairs of long and short vowels. Consider the charts in Figure , which displays the five short and long vowels, /i(:), e(:), a(:), o(:), u(:)/, one pair per chart, for Juba, the first speaker studied. These charts show all tokens of the given phonemes, including all segmental environments, and all stress levels.
The expected result of examining a large number of tokens of different vowel phonemes, where short and unstressed tokens are included, is that the classes will overlap. There are many contextual effects that modify vowel quality: The vowel in well may sound closer to that in wool than that in head. As Lisker (1948) showed, even measurements of a vocalic minimal pair (pap, pep) in monitored laboratory speech may overlap in F1-F2 space, though this result was not confirmed in a replication conducted as support for the methods used here (page ). The problem in natural speech is much worse than Lisker's results would suggest: the distribution for a single vowel can cover almost the entire vowel space (as shown on page ). As Figure shows, overlap is no less pervasive in Jamaican than in Chicago. Very little of this overlap may be attributed to errors, as shown in the above discussion of the analysis of outliers (page ).
Thus we see that /i:/ and /i/ significantly overlap each other, despite having quite different means. Sometimes skim may sound like scheme, but usually not. Similarly for the other long/short pairs: the clouds overlap, but have different means. One exception may be /a:/ and /a/, where the mode of the short /a/ distribution is near the location of the mode of the long /a:/ distribution. The primary difference between /a/ and /a:/ appears to be that the short vowel is more widely distributed, perhaps because as a short vowel, it is more susceptible to context and stress effects.
The last chart on the page is of /ai/ and /ou/, which do not form a long/short pair, but conveniently filled the sixth position on the page so that I could repeat the point made above about the Great Vowel Shift. The nuclei of these vowels are identified as the same in some English dialects, such as my own, where they can be transcribed as /ay, aw/. In the Great English Vowel Shift, the nuclei of the high-front and high-back vowels, Middle English long //i:// and //u://, dropped down to the bottom center of vowel space, with the result that their nuclei could be identified as the same in some of the daughter languages. The Shift has not had the same result here, since the resulting classes /ai/ and /ou/ have nuclei that are phonetically quite different.