Voicing-conditioned vowel alternations
|Submitter Email:||click here to access email|
Thanks to everyone who replied to my query about vowel quality
alternations conditioned by the voicing specification of adjacent
consonants: Dave Odden, Paul Johnston, Thorsten Schroter, Michael
Johnstone, Paul Boersma, Mary Paster, Remy Viredaz, Dieter Wunderlich,
Geoffrey S. Nathan, Wolfgang M. Schultze, Price Caldwell, Tobias Scheer,
and Viktor Tron. To the examples I already had from English (Canadian
Raising/Southern Monophthongization) and Polish (morphophonemic u/o
alternation), they added some from Nilotic, Madurese, Scots, Shetlandic
Scots, Buchan Scots, Yakut, Maastricht Limburgian, and Czech. They also
made short work of my putative diachronic German example. A summary is
I'm interested in this because of an article by Erik Thomas in the Journal
of Phonetics in 2000, in which he raises the possibility that there is
hyperarticulation before voiceless obstruents. I think this is right.
Voiceless obstruents are themselves hyperarticulated (review: Maddieson &
Ladefoged _Patterns of Sounds_ 1996:95-99). In English, mid and low
vowels become lower before voiceless (Summers 1987 JASA, Wolf 1979 JPhon,
Fujimura & Miller 1979 Phonetica), with more lowering closer to the
consonant. Meanwhile, the high offglide of Eng. /ai/ becomes higher
(Thomas 2000 JPhon), as do the offglides of /au oi ei/ (Moreton 2001 LSA
and in prep, ask if you'd like a draft).
Since the hyperarticulation of [-voice] obstruents is widespread, I'd
expected phonologized alternations on the pattern of Canadian
Raising/Southern Monophthongization, and Polish, to turn up in more
places. The present harvest is encouraging, but most of the
well-documented examples are from near relatives of English and Polish.
The Madurese alternation goes the "wrong" way; however, there the
consonant precedes rather than follows the vowel (and does not affect its
length), so it is probably a separate phenomenon. Thanks again,
David Odden writes:
Not unrelated is something that Keith Denning did in his Stanford
dissertation, regarding the relation between breathy vs. modal voice and
vowel quality in Nilotic. Essentially, what he pointed out is that
perceived vowel height is related to vocal tract geometry, that the "back
tube" can be lengthened either by raising the tongue or lowering the
larynx, and that breathy vowels often also involve larynx lowering; hence
they also seem to be a bit higher.
The interesting thing about these cases, in comparison to the ones you
mentioned, is that they don't involve vowel length (i.e. voice and vowel
length are related, and length is related to height). <end David Odden
This dissertation is a treasure trove of examples. Denning proposes a
"In lgs in which ther eis a regular correlation between perceived vowel
height and phonation (i.e. pitch, phonation type or voicing in
consonants), greater height is associated with relative laryngeal laxness
and/or voicing." (p. 59)
POLISH and CZECH (with an excursus on GERMAN)
Raising in Polish -- a morphophonemic alternation in
which some /o/s surface as [u] before an underlying voiced
coda consonant (Gussmann 1980 _Studies in Abstract Phonology_; Kenstowicz
1994 _Phon. in Gen. Gr._ 74-78):
nom. sg. m[u]d 'fashion'
gen. pl. m[o]da
Tobias Scheer <Tobias.Scheer@unice.fr> writes:
this alternation is indeed very common in Polish, but it is heavily
restricted by morphological and lexical parameters (the latter related to
frequency of the word). There is no synchronic activity for sure since
present-day [u] > alternates only with [o] if it comes from a former [o].
[u]s that have always been [u] do never alternate. Polish spelling notes
this difference: alternating [u] is written "o with an acute accent",
identifying its diachronic source, while regular [u] is spelt "u".
The same alternation concerns the two nasal vowels of Polish (written E
for the front, O for the back one hereafter):
dOb "oak NOMsg" - dEb-u "oak GENsg
etc. both the O-E and the o-u alternation are instances of the same
original process, which, alas for you, does not modify vowel quality, but
o > oo / __C+voice #
and there are two conditions, not just one: the input-o must occur before
a voiced consonant which on top of being voiced needs to be word-final.
Exactly the same alternation o-u is found in Czech, with the same
conditioning. Furthermore, Czech has the advantage of not having
eliminated vowel-length as Polish did, to the effect that the original
contrast in length is still visible (the forms hereafter are not spelling,
kuuN "horse NOMsg" - koNe "id. GENsg" - koN-sky "id., adjective"
etc. much more data and illustration is available in a handout of a class
of mine which you can download at
then go to "classes taught at Warsaw", and click on the course-handout.
Polish-Czech o-u is in section 12.13 on p.44ss and a parallel German
alternation in strong verbs is discussed on p.45: beissen "bite" where
<ss> is [s] derives a preterite biss where the [i] is short whereas
preisen "worship" where <s> is [z] derives a preterite pries where the
<ie> is long.
<End quote from Tobias Scheer>
The course handout mentioned above is very detailed and informative, and
could provide a whole semester's worth of problem sets.
ENGLISH: Canadian Raising/Southern Monophthongization
Canadian Raising / Southern Monophthongization (English)
- a very widespread alternation in which /ai/, and sometimes
also /au/, is higher before voiceless codas than voiced ones
(Chambers 1973 CanJLing 18:113-135):
tight t^It taIt
tide taId ta:d
Paul Boersma <email@example.com> and Geoffrey S. Nathan
<firstname.lastname@example.org> suggest that the alternation is not conditioned directly
by the voicing of the following consonant, but by the difference in vowel
length (which in turn depends on voicing).
SCOTS: Scottish Vowel Length Rule
Michael Johnstone <email@example.com> and Viktor Tron
<tron@firstname.lastname@example.org> pointed out that the famous SVLR induces quality
alternations in /ai/. The SVLR applies to Modern Scots reflexes of
certain Middle Scots vowels:
Middle Scots (16th cent.) /ei i: e: u: * ui ou iu E a o/
Modern Scots /% e u u ! @i ^u ju E a o/
(/%/ see below; /*/ = slashed o, Cardinal 10, upper mid front rounded; /@/
= schwa; /!/ = several diphthongs; see Aitken 1981 pp 132-133). These
vowels are long
before a morpheme boundary, or
before a voiced fricative and a morpheme boundary, or
before [r] and a morpheme boundary
and short elsewhere (Aitken 1981 p. 135). The present-day reflex of MS
/ei/ undergoes a quality change like that of Canadian Raising/Southern
Monophthongization: _price_ [pr@is], _prize_ = [pra:ez]. A slew of
references are provided by Viktor Tron:
A. J. Aitken (1981) The Scottish Vowel Length Rule In. M. Benskin and M.L.
Samuels (eds) So Meny People, Longuages and Tonges, pp 131-157. Edinburgh:
Middle English Dialect Project.
Scott Allan (1985) A note on AYE distribution. Journal of Linguistics 21.
J. Derrick McClure (1977) Vowel duration in a Scottish accent. Journal of
the International Phonetic Association 7. pp 10-16
April M. S. McMahon (1991) Lexical Phonology and Sound Change: the case of
the Scottish Vowel Length Rule. Journal of Linguistics 27. pp 29-53
James Myers (1999) Lexical Phonology and the Lexicon. Rutgers
Optimality Archive #330-0699
J. M. Scobbie, N. Hewlett and A. E. Turk (1999) Standard English in
Edinburgh and Glasgow: the Scottish vowel length rule revealed. University
of Edinburgh ms.
BUCHAN SCOTS: Height harmony blocked by [+voice] obstruents
Mary Paster <paster@socrates.Berkeley.EDU> writes:
According to Eugen Dieth's (1932) grammar of the dialect, the facts are
these: High, unaccented vowels in the second syllable undergo lowering to
a mid variant if the preceding vowel is non-high (i.e. mid or low -- so
it's not total height assimilation). The pattern holds for root-suffix
(with some suffixes being excepted for some reason), word-clitic, and,
apparently, within words. But the lowering harmony is blocked by voiced
obstruents (and a few select sequences of other consonants). This gives
rise to the (relatively) famously-cited pair [lase] 'lassie' vs. [ladi]
'laddie', where the lowering is blocked on the diminutive suffix in
'laddie' (I gather that Scots makes more use of the diminutive suffix than
we do, so Dieth's grammar has a ton of examples of it, and it's clearly a
robust pattern -- at least, it was in 1932). <End Mary Paster quote>
Paster also mentions an article by Colleen Fitzgerald in a recent or
near-future _English Language and Linguistics_ (and has one of her own in
SHETLANDIC SCOTS: Raising before voiced sounds
Paul Johnston <email@example.com> writes:
Voicing conditions vowel alternations in Shetlandic, and to some extent,
Orcadian Scots, though the "voiced" group may be more broken up.
Scots /a/ = Shetlandic [a] before vl. sounds, [ae] before vd. ones As
Scots /E/ = Shetlandic [E] before vl. sounds, [e] or [ei] before vd.
ones As in bet/bed
Scots /a:/ = Shetlandic [a:] before vl. sounds, [ae:] (North/West) or
[D:] (South--that's a low back rounded vowel) before vd. ones as in salt
Some of the other vowels show this alternation too, but only before
obstruents; voiced sonorants behave like voiceless obstruents, or
introduce a third allophone.
The exact equivalent of the first and third rules also operate in some
types of Northumbrian English, particularly of the Coalfield and East
Central Part of the county of Northumberland.
bat = [bat] bad = [bae@d]
salt = [sa:t] Maud = [mae@d]
The best [reference] for Orcadian/Shetlandic is probably mine: Johnston,
Paul. 1997. "Regional Variation" in Jones, Charles (ed.) . The Edinburgh
History of the Scottish Language, pp. 433-513. Relevant material is on
464, 485, 489. Also 464 (for /E:/). My data came from: Mather, James Y.
and Hans-Henning Speitel. 1986. The Linguistic Atlas of Scotlkand, Vol.
3: Phonology. London: Croom Helm. and observations.
The Nhb. stuff was personally observed. I mentioned it in my Ph. D.
thesis, but that's unpublished. However, I think Henry Warkentine's
(1964) hard-to-find thesis on Hexham, Nhb. dialect (I haven't got a full
ref.) may have it. If you want to plow, you can get the whole
distribution by looking at relevant words in Orton, Harold and Wilfrid H.
Halliday. 1962. The Survey of English Dialects, Vol. I. Basic Material:
The Six Northern Counties and the Isle of Man. Leeds: Arnold.
Look at Community Nb 4 particularly. <End Paul Johnston quote.>
MADURESE: Raising after voiced or voiced-aspirated C
David Odden <firstname.lastname@example.org> called my attention to this one,
which is described in detail in
Stevens, Alan M. (1968). Madurese Phonology and Morphology. American
Oriental Series, #52. New Haven: American Oriental Society.
Stevens describes a class of "higher determinant" consonants, which he
Lab Den Alv Pal Vel
voiced stops /b d d, z g /
aspirates /bh dh d,h zh gh/
plus /w/ and most instances of /j/ (exceptions for morpheme juncture).
The four principal vowel phonemes /a i u @/ occur in two versions, raised
and lowered. Following a higher-determinant consonant, vowels are raised.
This effect propagates through the word until stopped by a
lower-determinant consonant (anything that is neither higher-determinant,
/s/ or /j/ with a close juncture, [r], or [q]). Vowels not raised are
There is a relevant paper by Cohn and Lockwood (1994, not seen) in the
Working Papers of the Cornell Phonetics Lab.
YAKUT: Lengthening and breaking-to-diphthong before [+voice]
Remy Viredaz <email@example.com> pointed me to
Grammont, Maurice (1933). Traite de phonetique. Paris: Librairie
A paragraph on p. 187 reads:
We find in some languages a phenomenon which is in a certain sense the
opposite of the Latin phenomenon [devoicing of a stop with compensatory
lengthening of the preceding vowel, te:ctum < tegere]. In Osmanli, a
short vowel + voicless consonant remains short vowel + voiceless
/at/ 'horse' /at/
/ot/ 'plant' /ot/
/as-/ 'open' /atS-/
But a long vowel + voiceless consonant becomes a short vowel + voiced
/a:t/ 'name' /ad/
/uot/ 'fire' /od/
/a:s/ 'starved' /adZ/
/bu:t/ 'hip' /bud/ 'thigh'
/y:t/ 'milk' /yd/
/ky:s/ 'force' /gydZ/
G. doesn't discuss the breaking-to-diphthong, but it seems that Yakut
underwent a change /o/->/o:/->/uo/ before originally voiced consonants.
Many more examples, though no published source, can be found in an
assignment from Bert Vaux's class at Harvard:
MAASTRICHT LIMBURGIAN: Lengthening and breaking-to-diphthong before
Paul Boersma <firstname.lastname@example.org> contributes a parallel breaking
[W]e can consider a similar case in Maastricht Limburgian, in which high
vowels became (or become, in an abstract analysis) diphthongs
if a final schwa dropped after a voiced consonant. Thus:
bli:2ve -> blEi1f 'stay-1SG'
bli:2ven -> bli:2ve(n) 'stay-INF'
So we have vowel alternations within verb paradigms, if the consonant is
voiced. In nouns, we have similar examples:
du:2ve -> dOu1f 'pigeon'
du:2ven -> du:2ve(n) 'pigeons'
The free ride in this case is provided by the change from second to first
tone, which occurs precisely when schwa is dropped after an originally
voiced consonant (or currently voiced; the sonorants do it as well, as in
Polish). So the real conditioning factor is probably the presence of the
first tone, though it is hard to distinguish this from conditioning by
voicing, since there are no underlying or original first tones on the
vowels /i:/, /u:/, or /a:/ (which becomes [O:] under tone change or voiced
schwa drop). The Maastricht case is further complicated by the fact that
unlike in Polish, there are original consonant-final forms, which don't
bli:2f -> bli:2f 'stay-IMP.SG'
So underlyingly, there may be a voiced consonant here as well, but
historically there isn't. This is some extra evidence for the causation of
vowel change by tone. Now one would like to know the details of what
happened in Polish... [the Maastricht change also occurred if there was
*no* consonant, as in ri:2e -> rEi1 'row' and ly:2(d)e -> loei1 'people']
<end Paul Boersma quote>
GERMAN: Not really
My original query asked about:
(3) A "well-known sound change" in German in which long
vowels are shortened and lowered before voiceless consonants
(mentioned briefly in Kohler 1984 Phonetica 41:150-174):
Several people wrote in to deny that this was anything of the sort:
Thorsten Schroter <email@example.com>, Paul Boersma
<firstname.lastname@example.org>, Remy Viredaz <email@example.com>, Dieter
Wunderlich <firstname.lastname@example.org>, Wolfgang M. Schultze
<W.Schultze@lrz.uni-muenchen.de>, and Tobias Scheer
That this is not synchronic is shown by Wunderlich's examples:
g[u:]ter 'good' strong masc.sg.nominative
Wolfgang Schultze connects it with the open/closed syllable distinction:
Though there seems to be a tendency to shorten long vowels before vl.
consonants [such as _V[a:]ter_ 'father' shifting towards _V[A]ter_ esp.
Rhenanian and some northern varieties], the standard formula is that there
is a correlation between
length/hight and V-final syllable ('open')
short/lowered and C-final syllable ('closed')
Hence we have
g[u:]-ter 'a good one' etc.
b[I]b-bern 'to tremble'
B[i:]bern 'for the beavers' etc.
And remember that the opposition M[U]tter / Br[u:]der is conditioned by
etymology (< *mat:r resp. *bhr:ter). <End Wolfgang Schultze quote>
Tobias Scheer adds more examples of short vowels before voiced stops:
wabbern, blubbern, daddeln.
Paul Boersma and Tobias Scheer point out that the only certain correlation
is between vowel height on the one hand and vowel length (of whatever
origin) on the other: [U] and [u:], never [U:] or [u].
Remy Viredaz comments:
The German example does not conform exactly to the formula you have used.
In many cases, a long vowel has remained long before a voiceless consonant
(Buch 'book', Hut 'hat', Mut 'courage', Haut 'skin', Euter 'udder'), and
even a short vowel has sometimes become long in that environment (Vater
'father'). I haven't looked through the 19th and 20th century grammars to
see if our "forefathers" have been able to see some order in all this.
However, the voiced vs. voiceless character of the vowel has certainly
played a role in many cases. MHG short i, u seem to have been lengthened
before a voiced obstruent (Glied 'member') or a sonant (viel 'much') and
not before a voiceless consonant (Schnitt 'a cut') or a consonant cluster.
<End Remy Viredaz quote>
Finally, Price Caldwell <email@example.com> mentions the tendency
of high vowels to disappear entirely (or at least become devoiced) between
Many thanks again to all contributors.
|Original Query:||Read original query|
Sums main page