Wells, A study of the formants of the pure vowels of British English

V. Duration

[SAMPA] [Unicode]

Measurement of duration

Although the question of vowel duration might be considered to fall outside the field defined by the title of this thesis, it has nevertheless been decided to present measurements of this feature in addition to those pertaining more strictly to vowel formants.

Measurements were made of the distances on the spectrograms between the point of onset of each vowel and the point of onset of the following /d/. Vowel onset was defined as the point where voicing began; the vowel formants were, of course, usually present considerably sooner, since /h/ regularly approximates in tamber to the vowel following it [21, §777-8]. The end of the vowel was taken as the point where the vertical striations representing the glottal pulses ceased, being interrupted by the contact of the tongue with the alveolar ridge for the articulation of /d/. This means, of course, that the vowel-consonant formant transitions were taken as part of the vowel.

The measurements taken (in hundredths of a centimetre) were treated statistically in various ways and then converted into seconds. Results are summarized in Table 6 and Table 7 and shown graphically in figure 9 and figure 10.

Relative versus absolute

Vowel durations can be measured straightforwardly in fractions of a second. Previous work has usually gone no further (e.g. [18]). In some languages, where there are no phonemic contrasts of length (or, in Jones's terminology, the language has only one chroneme [22, §§399,417]), there can be no objection to this. But it is not adequate for a language like English, where vowel length is contrastive -- whether the duration contrast is considered the chief difference between, say, /i/ and /ɩ/ (in which case they would usually be transcribed as i: and i [22, §402]), or whether it is held to be subsidiary to the main contrast which is one of tamber [22 §516].

If an important point of contrast between /i/ and /ɩ/, or between long and short vowels as a whole, is duration (length), then what is important is not so much the ABSOLUTE duration of a given vowel, but its RELATIVE duration compared with other vowels of a similar tamber, or with the speaker's overall average duration for all vowels. In any linguistic situation, it is contrasts that matter, and not absolute values.

The duration of each recorded vowel was measured from the spectrogram material. The durations of each individual speaker's 22 vowels were averaged to give the SPEAKER'S MEAN VOWEL DURATION. The resultant figures may be see in Table 6 (col. 4). Each vowel duration was then expressed as a ratio of the relevant speaker's mean vowel duration. It was very noticeable that the traditionally "short" vowels (i.e. /ɩ ɛ æ ɒ ɷ ʌ/) regularly had a relative duration of well under 1.00 (the only exception was an occasional /æ/, which reached in one case 1.06), while traditionally "long" vowels, /i ɑ ɔ u ɜ/, regularly scored well over unity (the minimum being one occurrence of /i/ with a ratio of exactly 1.00).

This result is of course not unexpected, since English vowels are well known to fall into two classes of markedly differing length or duration. It would however seem to show that opinions expressed here and there (e.g. [47 §§361(vi),362; 21 §§874-879]) that English may be losing the short/long vowel contrast are unfounded, at least as far as the kind of English spoken by the sample studied is concerned.

Relationship between duration and tongue height

More interesting is the finding recorded in Table 7 and shown graphically in figs. 9a and 9b: that the open vowels studied are regularly of greater duration than close vowels of the same linguistic "length". This applies equally whether the absolute (fig. 9b) or relative (fig. 9a) values are considered. Thus /æ/ is regularly longer than /ɛ/ and /ɒ/, which in turn are regularly longer than /ɩ/ and /ɷ/; although all these short vowels are of shorter average duration than even the closest long vowels /i/ and /u/. The mid long vowel /ɜ/ is in turn longer than /i/ and /u/; and the open long vowels /ɑ/ and /ɔ/ are longest of all. Points in figs. 9a and 9b are plotted at conventional distances along the abscissa (corresponding to the articulatory dimension front-back) in order to bring out the similarity between this feature of variation in duration on the one hand and the familiar vowel quadrilateral on the other. The only vowel noticeably out of place is /ʌ/, which in spite of its half-open to open articulatory position is nearest in average duration, whether relative or absolute, to the short close vowels /ɩ/ and /ɷ/. It is not clear why this should be so.

The average absolute durations of short vowels, long vowels, and all vowels pooled, together with the ratios of long vowel average to short vowel average, are shown for each speaker in Table 6. It will be seen that the overall average for short vowel duration is 0.16 secs., with a standard deviation of 0.03 secs., while the overall average for long vowel duration is 0.30 secs., with standard deviation of 0.08 secs. Speakers' mean vowel durations ranged between 0.18 secs. and 0.32 secs., with a grand average of all vowels for all speakers of 0.23 secs. The duration ratio of a given speaker's average for long vowels to his average for short vowels was found to range between 1.4 to one and 2.7 to one, with a mean of 1.9 to one.

Vowel duration statistics quoted by Jones

Measurements for the vowel durations in the words seed and lid in his own speech are quoted by Jones [22, §§403,405] as follows: seed, 0.252 secs.; lid, 0.135 secs. If the duration ratio between these two vowels (1.87 to one) is typical of all Jones's long vowels versus short vowels, then his ratio is very near to the average found in the present study.

Vowels in contexts other than /h-d/ will, of course, have different durations. Jones [22 §403] gives the following average lengths for his /i/ (which he writes i:) in various phonetic contexts:
see0.317 secs.
It will be seen from this data that the position before a voiced consonant, as exemplified in the words here studied, corresponds to the longest vowel duration of all preconsonantal positions, being exceeded only by prejunctural (final) position. Vowels occurring prejuncturally will be about one-quarter as long again as those before a voiced consonant; those occurring before nasals and voiceless consonants will be about four-fifths and half as long respectively, while vowels before intervocalic voiceless consonants will be only about one-third as long. All the foregoing applies exclusively to strssed vowels bearing the nuclear tone of the intonation pattern; vowels under other conditions of stress and intonation may be expected to be of considerably shorter duration.

Meyer [28] carried out a large number of kymographic measurements of duration of English speech sounds. His figures are based on 393 different monosyllabic and dissyllabic words spoken two or three times each by one speaker. He found the following absolute durations for vowels, including diphthongs, in the context /-d/:
-ɷd21.6 centisec.-id35.4 centisec.(-ɛəd38.3 centisec.)
(His transcription has been modified to accord with that used elsewhere in this dissertation. Diphthongs not covered here are placed in brackets above.)

It will be seen that these values for short vowels all fall outside the range recorded in the present sample: presumably the extra length is due to the fact that the words were not included in sentences but spoken i isolation. The long vowels, however, all fall within the range observed in the present sample. Expressed as a ratio of the speaker's mean vowel duration, the vowels have the following durations: /ɷ/ .66; /ɩ/ .76; /ʌ/ .81; /ɛ/ .94; /ɒ/ .95; /æ/ 1.09; /i/ 1.09; /u/ 1.09; /ɜ/ 1.16; /ɔ/ 1.23; /ɑ/ 1.23. It will be seen that Meyer's subject had a poorish short/long contrast, since all his short vowels last relatively longer, and all his long vowels relatively shorter, than the means observed in the present sample (Table 7). The discrepancy is usually less than two standard deviations: only in the case of /ɛ/ is it more than three standard deviations (.94 as against .64, s = .09). The ratio for long vowel average to short vowel average is 1.3 to one, smaller than any in the present sample. One reason may be that any aspiration after the release of the consonant preceding the vowel was taken by Meyer as part of the vowel even though it be voiceless.

Meyer summarizes his findings about vowel duration as follows (p. 107-8, my translation):

"Vowel duration is determined
  1. by the nature of the vowel:
    1. ... other things being equal, lax ('short') vowels are shorter than the corresponding tense ('long') vowels.
    2. ... the higher the raising of the tongue, the shorter the vowel.
  2. by the nature of the following consonant:
    1. ... before a tense final consonant a vowel is shorter than before a lax final consonant.
    2. ... before a plosive a vowel is in general shorter than before the corresponding fricative."
It will be seen that his conclusions under (i) are in accord with those of this investigation.

Vowel length in General American

House and Fairbanks [18], studying certain General ɑmerican vowels, namely /i e æ ɑ o u/ (bead, bayed, bad, bod, bode, booed), found the same relationship between greater duration and lower tongue position observed in this study to hold also in their material. "From vowel to vowel, duration ... varied in a manner systematically related to the usual conceptions of vowel physiology." [18, p.113].

They found too that in their material there was no distinction between "short" and "long" vowels such as has been observed in the present study. This is not surprising, since

  1. they were dealing with American English, not British, and it has frequently been observed that, in Jones's words, "the speech of many (or perhaps most) Americans does not exhibit consistent relationships between vowel length and quality such as are found in some types of British English. With these speakers all vowels may occur long." [21, App.D §2].
  2. in the controversial but widely accepted "overall pattern" analysis of American English stressed vowels (syllabic nuclei) [46], which provides for a contrast between "simple" and "complex" nuclei corresponding to what we have called "short" and "long" vowels, the six words concerned would for many American dialects all be transcribed with "complex" nuclei, as /bíjd/, /béjd/, /bǽhd/, /báhd/, /bówd/, /búwd/; it is not therefore surprising to find no "short" vowels among them, even if the corresponding words in British English have short vowels.
  3. their words "The results shed considerable doubt upon the classification of [i], [ɑ], and [u] as "long" vowels and of [e], [æ], and [o] as "short" vowels by Jones [Outline]" are based upon a simple misinterpretation of his transcription symbols: Jones uses the symbols [e] and [o] to represent the vowels of head and hod (or in some works the reduced allophone of the vowel whose strong form he writes [ou]), and not the vowels of hayed and hoed, which is what House and Fairbanks use them to mean. Further, when Jones classes [ɑ] as a long vowel, he means the vowel of r-less hard (since he is talking about British English) and not the vowel of hod (which he writes [o] or [ɔ]).

Three other reports have indicated that American English does have a short/long contrast, although not so distinctly as the type of British English here studied.

Forgie and Forgie [15] measured the duration of nine speakers' pronunciations of ten vowels in the context /b-t/. They print a graph from which it appears that the vowels concerned fall into the following ranking, from short to long duration: /ɩ ɷ ʌ i ɛ u ɚ ɑ æ ɔ/. These can be arranged by duration to correlate with tongue height only if divided into a "short" group /ɩ ɷ ʌ ɛ ɑ æ/ versus a "long" group /i u ɚ ɔ/. (/ɑ/ represents the vowel in "hod".) But the two groups overlap considerably.

Peterson and Lehiste [38], although primarily studying the influence upon vowel duration of surrounding consonants, remark that with respect to vowels "two groups may be established ... : instrinsically short syllabic nuclei, comprising the four vowels [ɩ ɛ ə ɷ], and intrinsically long syllabic nuclei, consisting of [i eɩ æ ɑ ɔ oɷ u r ɑɷ ɑɩ ɔɩ]."

House [17] comes to much the same conclusion about American English vowels: " ... some of the vowels measured in this study can be divided into contrastive long-short pairs. The high-front vowels [i] and [ɩ] contrast in duration, as do the high-back vowels [u] and [ɷ], the mid-front vowels [e] and [ɛ], and the low-back vowels [ɑ] and [ʌ]. The four vowels [ɩ], [ɛ], [ʌ] and [ɷ] have been described many times before as 'short' vowels, and they are in fact characterized by the shortest average duration of the vowels measured in this study."

British English vowels, then, differ from American English vowels in two major respects as far as duration is concerned:

  1. British English has a much sharper long/short contrast, since even the longest British short vowel averages shorter in duration than the shortest long vowel, while this is not the case in American English;
  2. The distribution of vowels to the two classes is not quite the same: in particular, /æ/ and /ɒ/ are short in British, though the corresponding American vowels /æ/ and /ɑ/ are long or indifferent.

  • Final note: Inverse filtering
  • Title page and contents