Delattre, Pierre. Studies in French and Comparative Phonetics – T18

Some Factors of Vowel Duration and Their
Cross-Linguistic Validity *1

Perhaps no subject has been studied more thoroughly than vowel length before single
consonants in American English. Outstanding among old and new articles are one by
H. Rositzke in Language 12 in the late thirties, a series of six by R-M. S. Heffner in
American Speech 23 ending in the early forties, and two recent articles in this Journal,
one by Peterson and Lehiste, 34 the other by Arthur House. 45 Each study is more revealing
than the preceding one. The last author, perceiving that some factors had not
yet emerged so clearly as they could because of unsatisfactory experimental conditions
in earlier works, devised an experiment in which he used abstract nonsense syllables
instead of actual words, thus eliminating a number of noisy effects and gaining in
uniformity. The attempt was rewarded. Factors of vowel duration that had been
shady were now brought out in full light.

The data of those four major studies show considerable agreement. Apart from
stress and tempo, eight factors of vowel duration in American English emerge, three
“internal” factors that are in the vowel itself, and five “external” factors all to be found
in the single consonant that follows the vowel. In the correlating order shorter vowel/
longer vowel these eight factors are: (V1) Vowel abridging/vowel expanding, 56 (V2) less
open vowel/more open vowel, (V3) monophthong/diphthong, (C1) surd consonant/
sonant consonant, (C2) stop consonant/fricative consonant, (C3) liquid consonant/
solid consonant (all except r and l), (C4) oral stop consonant/nasal stop consonant, (C5)
more front consonant/more back consonant (within each of the six categories: surd
stops, surd fricatives, sonant oral stops, sonant nasal stops, sonant fricatives, liquids).133

When it comes to explaining these factors, however, agreement has little chance to
be complete since speculation plays a considerable role.

We would like to bring here additional facts to complement the discussion on
“explanations,” started by House in “vowel duration in English”.

The questions are whether, under one or more of those eight influences, variations
in vowel length rest upon learned habits of the phonemic structure of American
English, or are conditioned by inherent articulatory features; and to what extent
conditioning factors operate in other languages as they do in English.

(A) We agree with House that the “conditioned” explanation “… is untenable as
an over-all explanation”, and that the type /i/ɪ/ difference of length, where longer/
shorter correlates with closer/opener is “not explicable in the same articulatory terms”
as, say, the type /ɛ/ɪ/ difference of length, where longer/shorter correlates, on the
contrary, with opener/closer. Historical facts help confirm House's contention and
indicate that the /i/ɪ/ difference of length is learned, whereas the /e/ɪ/ is conditioned: /i/
is longer than /ɪ/ today not because it is less open — due to an articulatory conditioning
— but because of the survival of a former (Middle-English) distinctive feature long/
short /iː/ɪ/ which gradually changed to a rather less central/more central articulatory
distinction /i/ɪ/ with attenuation (but not extinction) of the old long/short distinction.
Thus it seems to be distinctive shortness of the vowel that caused centering (and
indirectly opening) and not the opposite. We have good reasons to accept this since we
know that the present difference in quality came after the difference in length. Besides,
in the evolution of English, vowels shortened by unstress tend toward the center of the
phonetic triangle, namely the schwa. Example: in nationalist, a former [o] and a
former [a] tend toward [ə] or even complete disappearance. The /ɛ/ɪ/ difference of
length, on the other hand, has no historical source — it seems to be due to articulatory
conditioning within the category of shorter vowels (more open vowel is longer), just
as the /e/i/ difference of length within the category of longer vowels. A differentiation
of timbre similar to that of English /i/ɪ/ occurred in Latin to Vulgar Latin. The long
[iː] remained close [i]; the short [i] centered to [ɪ], which the scribes started to write e
for lack of a better grapheme; and eventually the distinction of length disappeared
entirely. In English, although the /i/ɪ/ length distinction has not yet disappeared, it is
no longer distinctive per se. Duration still does play a part in the /ɛ/ɪ/ phonemic
distinction, however — it is one of its acoustic correlates; whereas in the /e/ɪ/ phonemic
distinction duration is negligible as an acoustic correlate. And, let us add, the /ɛ/ɪ/
duration difference is not negligible just because it is small. In French, for instance,
the distinctive difference of length /mɛtr/mɛːtr/ (20-30) is much smaller than the nondistinctive
difference /sɛk/sɛr/ (15-40), but the first one is phonemically learned, the
last one is phonetically conditioned. 67 Examples could also be drawn from English.
Some speakers will make a distinctive difference of length between bomb and balm
but they will make a larger difference of length — though nondistinctive — between
134leap and leave. And the naive subject will easily be made conscious of the first
difference of length but not of the second.

(B) All authors, including House, agree that there is greater muscular effort during
the production of a surd consonant (bit) than during the production of the corresponding
sonant consonant (bid). But House assumes, in addition, that vowels preceding a
greater effort ought to be longer — that bit ought to be longer than bid — if the /ɪ/
duration is conditioned. Why such an assumption? We have always made the opposite
hypothesis, namely that the anticipation of a greater effort would make one
shorten the vowel more in bit than in bid. 78 Is it not natural that by anticipation of
more and longer closure in [t] than in [d] one should close earlier for the end of the
preceding vowel and thus shorten it? It is well known that, in the process of speaking,
one constantly anticipates. For instance, nondistinctive nasalization of vowels,
whenever it occurs, in any language, is often due to the following consonant, never to
the preceding one. Other observations might also help in realizing this effect of
anticipation. For an English speaker we have seen that [ae] is shorter in pack or pat
than in pad. But it is shorter yet in pact. Is this not indicative that the anticipation of
a greater effort for the articulation of the cluster [kt] shortens more than the anticipation
of a single consonant [k] or [t]? In the same order of things, Spanish [i] is
considerably shorter in pinta than in pina or pita.

Similarly, stops ought to shorten the preceding vowel more than do fricatives
because more closure (for the stops) requires more effort — anticipation of a greater
effort shortens the vowel more. In the matter of correlation between degrees of
aperture and degrees of effort, consonants and vowels do not have to behave similarly
as assumed by House, 89 but rather in opposite ways. We would think that in a vowel
a greater effort correlates with more opening; and in a consonant, with more closure.

(C) As evidence against the “conditioned” explanation, House relies on a study of
surd/sonant effects on the length of the preceding vowel which minimizes this effect on
Spanish vowels and implies that it does not obtain in Spanish as much as in English. 910

The conclusions of this study prove nothing against the cross-linguistic validity
of the “conditioned” explanation.

(1) The surd/sonant contrast practically does not exist, in Spanish, where it can
affect the length of a preceding vowel. In Spanish final or intervocalic position, there
is no true case of a consonant distinction based solely on the surd/sonant distinction.
The sonant counterparts of final or intervocalic surd stops are all fricativated:
135pito/pido, [pito], [piðo]. The closest thing to a surd/sonant affecting a preceding contrast
vowel is the type tez/sed, [teθ], [seð], but even this is not quite satisfactory for it
opposes a real fricative to a fricativated stop which is an allophone of a stop.

(2) The Zimmerman-Sapon study uses exclusively bisyllabic paroxytones in
Spanish (pito/pido) and monosyllabic oxytones in English (niece/knees). Is this a fair
comparison? No wonder the difference between surds and sonants is found to be
considerably greater in English (Spanish /p/β/, 93/130; English /p/b/, 126/200).
Obviously the shortening influence of a medial consonant in a bisyllable is much less
pronounced than that of a final consonant in a monosyllable. A fair comparison
would have been Spanish pito/pido with English bitten/bidden in order to have
bisyllabic paroxytones in both cases.

(3) In the Zimmerman-Sapon study, the average difference between vowel length
before surds and vowel length before sonants (18 msec) is calculated by adding all
surds on one side and all sonants on the other. This, frankly, is meaningless. Comparing
/k/ with /l/ proves nothing. But if instead of accepting the authors' conclusions
we look at the data itself, we find more attractive duration figures before the three
pairs of surd stops and their sonant (fricativated) counterparts: /p/ 93, /β/ 130;
/t/ 104, /ð/ 136; /k/ 108, /γ/ 137. It is not known to what extent these differences in
vowel length are due to voicing or to fricativation, but considering that they occur
before medial consonants and that they are nevertheless so consistent, these vowel
durations are certainly significant. It is not certain that English paroxytones of the
type bitten/bidden would yield more significant differences.

(4) The three pairs named above reveal a perfect front-to-back pattern: the vowels
being shorter before /pβ/ than before /tð/ than before /kγ/. This is in remarkable
agreement with the Peterson-Lehiste data on English!

(5) They also reveal a perfect agreement with the notion of “force of articulation”
suggested by Belasco. 1011 Force of articulation representing a combination of two
consonant factors: surd-stop vs sonant-fricative, the /p/β/, /t/ð/, /k/γ/ pairs represent
precisely the contrastive factors strong force vs weak force.

(6) In short, agreement with English appears in three factors: surd/sonant,
stop/fricative, fronter/backer. This is significantly in favor of the conditioned
explanation.

(D) This is not the place for a review of vowel-length behavior in other languages.
We shall simply state briefly that, on the basis of the studies we know, nearly all the
eight factors named above operate in French. This is very significant, for French and
English are at the two opposite extremities on the scale of world languages. They
differ in every possible way — tension/laxness, fronting/backing, vowel anticipation/
consonant anticipation, etc., yet they agree on the determinants of vowel length.

In other languages, Meyer for German, Metz for Italian, and Navarro Tomás for
Spanish had already found the three conditioning factors: longer before sonant than
136before surd, before fricative than before stop, and in more open vowels than in closer
ones, in the first two decades of our century. 1112

Having weighed all the additional facts we were able to contribute here, one will
perhaps agree, first, that variations in vowel length are phonemically learned under
only one of the eight factors listed above — the abridging/expanding factor (also called
lax/tense) — and that under the seven other factors, variations in vowel length are
physiologically conditioned; secondly, that the chance for conditioning factors to be
universal, to operate cross-linguistically, is far from negligible.137