Two methods of characterizing phonetic vowel qualities are used in this book. Impressionistic transcription of vowels according to a standard set of vowel qualities such as those of Daniel Jones is very useful, but only when readers are thoroughly trained to recognize the precise auditory qualities of the symbols used. Unfortunately, this means that phonetic descriptions are unreadable except to a relatively small, properly trained audience. Efforts to expand this audience by making it easy to learn standard vowels include Jones' original publication of a 78rpm record of the Cardinal Vowels, and my Macintosh program, ``Jones' Phones'', which trains the user in identification of these vowels. A limited amount of IPA impressionistic transcription is presented, since it is useful for calling to mind the phonetic qualities without referring to charts (though it is only interpretable when the reader knows the values of the symbols).
Another time-honored method of characterizing vowel quality is through plots of measurements of the first and second formant frequencies: the F1-F2 chart. Formants are interesting for several reasons. They are more objective than impressionistic categories, especially when the sounds measured are relatively short in duration. They are as good as any other spectral representation (under the assumptions of an all-pole model).4.4 They distinguish audible vowel qualities extremely well, as shown in the next section. And they can be used to infer articulatory configurations, as justified in detail in the Acoustics chapter. Thus they are good representations for static vowel qualities, analytically, perceptually, and articulatorily. Therefore charts of F1 vs. F2 are the core of the data presented in this work.
In 1949, Leigh Lisker wrote perhaps the first phonetics dissertation to make use of the recently declassified spectrograph machine. He made the disconcerting discovery that the main cues to vowel quality, namely the frequencies of the first two formants, were unable to distinguish reliably between clearly distinct pronunciations of two vowel phonemes. The phonemes involved are adjacent in formant space -- the vowels in the words pap and pep. Since he made sure each token was clearly identifiable as the intended form, throwing out tokens which were ambiguous to the ear, the similarities between /æ/ and // were not so great as to make any token of one sound like a token of the other. This result is one of the fundamental facts about F1-F2 charts: vowels that sound different may overlap in F1-F2 space (``formant space''). How then can we trust formant measurements as reliable representations of vowel quality?
Lisker (1949) prevents us from making a certain class of inference, that vowels which overlap in formant space are in fact phonetically identical, since /æ/ and //, for example, may overlap but sound quite different. Similarly, /ey/ and // may overlap in F1-F2 space: /ey/ is a long up-gliding vowel, while // is a short, frequently ingliding vowel. They may pass through the same location in formant space, but they are going in opposite directions. Thus the fact that measurements at single time-slices show overlap between the two categories is not evidence that the sounds are phonetically identical.
The purpose of the next two sections is to argue for the validity of the converse inference: vowels occupying significantly different parts of formant space cannot be phonetically identical.