Meanwhile, the argument rages on about Wikipedia's use of a special "diaphonemic" transcription that is supposed to cover all accents of English in a single representation. The argument can be seen (until someone decides that it's all gone on too long and makes it disappear) on the Talk section of Help:IPA for English in a section headed simply "Bad", and is essentially between one or two people who don't like the transcription or can't see the point of it, and one or two people who support it. I have already mentioned that I have serious misgivings about the idea myself. I think it was James Sledd who pointed out that "diaphonemics" and "dire phonemics" are homophonous in many accents of English.  

I wrote some time ago about how the Wikipedia article on Prosody needed improvement. I have now put my reworked version in place of the old one, keeping in some of the original material written by someone else. It still needs work, but I think the basic structure is now easier to follow.

Most people who have been introduced to acoustic phonetics have seen formant plots, graphs arranged in such a way that the frequency values of the first two formants of vowels are plotted on a graph arranged to look roughly like a cardinal vowel quadrilateral.  Here’s a simple example – I superimposed a CV diagram. The scales are only approximate, and were chosen to suit my formants.

I have been discussing formant plots with a WP editor who wants to improve the Vowels section of the Norwegian phonology article by adding a plot of Norwegian formant values that he has found in a published paper. At present the vowels of Norwegian are shown on a traditional auditory vowel diagram. When the formant values are displayed, there are several vowels that seem to be a long way from where you would expect them to be. For example, the values for /æ/ put it at the right-hand side of the chart, where back vowels live.

The discussion has revived my doubts about how much value formant plots actually have. There are so many variables that can reduce their usefulness. Different size vocal tracts produce different formant frequencies for the “same” vowel; some voices are much harder to analyze and measure than others; the effect of lip-rounding produces values different from those of vowels with the same tongue position but no lip-rounding; some vowels have formants that are much harder to resolve than others; and there are various ways of organizing the plots (e.g. using different scales such as Bark or Mel), or using the value of F2 minus F1 for the horizontal axis. On top of those factors, I find that if I measure the formant frequencies of a particular vowel several times, I get quite widely varying results. There’s a useful summary of all this in Katrina Hayward’s book Experimental Phonetics (2000, Longman, Chapter 6), as well as in Peter Ladefoged’s Three Areas of Experimental Phonetics (1967, Oxford). In the end, the only use of a formant plot (or formant data in general) that I would feel confident about making would be for comparing a vowel of one speaker with another vowel produced by the same speaker in the same context.

Wikipedia has a short article on formants, and formant plots are briefly mentioned in it, but regrettably, no example is given. Nor do we find any such example in the dreary WP article on acoustic phonetics. Interestingly, WP itself has flagged up the fact that this article is of very poor quality and needs major overhaul. The bibliography looks as if it has simply been copied from someone’s reading list.

The other thing I am reminded of is the persistent male bias in formant data. The WP Formant article reproduces a single chart of values of F1 and F2 for cardinal vowels taken from Catford’s A Practical Introduction to Phonetics (1988, Oxford). As we find in most of the literature on formants, adult male values are taken as the norm, despite the fact that adult males are, of course, a minority of the population. In a way it’s surprising that some of the earliest work on formant plots did use female values as well as male: early American work used the Petersen and Barney formant frequency data (published in 1951) which included men, women and children, and Ladefoged’s cardinal vowel study included three female phoneticians among the speakers. However, until recently most descriptions of formants and formant plots have used adult male values as the norm. Alan Cruttenden’s Gimson’s Pronunciation of English is an honourable exception.