LINGUIST List 13.3330

Tue Dec 17 2002

Sum: High Vowels & Dorsal Consonants

Editor for this issue: Steve Moran <>


  1. Dave & Julie Eberhard, summary of responses re: query on hi vowels and dorsal consonants

Message 1: summary of responses re: query on hi vowels and dorsal consonants

Date: Sat, 14 Dec 2002 18:20:12 -0400
From: Dave & Julie Eberhard <>
Subject: summary of responses re: query on hi vowels and dorsal consonants

Replies to following query regarding link between [+hi] vowels & dorsal
consonants (Linguist 13.3174):

This question has to do with the spreading of place features from
vowels to consonants. The Mamainde language has a spreading process where
the high front vowel spreads [+hi] to the coda, creating a Dorsal, or velar,
or [+hi] place of articulation in the consonant. The output is not a palatal
consonant but a true velar. This is hard to explain via Clement's Unified
Feature Theory, or any other articulator theory for that matter since [hi]
is not available as a feature for consonants (they allow Open at the
Aperture node but this applies only to vowels).

Has anyone done or seen any research which shows high vowels spreading
the hi feature to consonants and creating dorsals (or velars)?


	From: "Nick Clements" <>
On your inquiry on the feature [+high]:

I'm not sure why you believe that [+high] or [-open] is not available
for consonants. [+high] is a consonant feature in the framework of
Chomsky and Halle, where it distinguishes velars from uvulars. In the
theory of Clements & Hume, the assignment of [+high] or its equivalent
to a consonant would create a vocalic secondary articulation involving
a high tongue body position. If the target consonant has no other
features, however, it might be realized as a high consonant such as a
velar by default. Joaquim Brandao de Carvalho has some data showing
relations between high vowels and velars that you might be interested
in - he has suggested that a feature (or element) of height might
define both high vowels and velar consonants,. You could get in touch
with him at

On your other question, Sharon Inkelas at Berkeley has suggested that
OT should allow underlying underspecification. You can get hold of
her papers at
- ---------
	From: "Ora Matushansky" <>

Dear Dave,

I'm extremely fascinated by the question you asked. A similar link
between dorsals and high vowels exists in Russian, where it seems to
be more pronounced because there are 2 high vowels, front and back
(the latter one is unrounded, though there's also [u]). These two high
vowels trigger a separate palatalization rule which affects only
velars: descriptively, when a velar appears before [y] (the high
unrounded vowel) (or [i], where it doesn't matter because Russian also
has the usual Indo-European "palatalize before front vowels" rule),
this velar is palatalized and the vowel is fronted.

E.g.: ruk- 'hand' + -y 'Nom-pl' -> ruk'i (final stress)

If what you're discussing in the quote below is an assimilation rule,
this makes the process described above a dissimilatory one.

The person you absolutely must ask about these matters is Morris Halle
(to whom I'm cc'ing this letter because I don't know whether he
subscribes to LL queries).
- --------------
	From: Joaquim Brandao de Carvalho

Dear colleague,

I have been developing a model of phonological primes (Carvalho 2002,
forth.) in which 'high' and 'velar' are but different manifestations
of the same abstract invariant. Two typical signs of this are that (a)
the high back vowel behaves as if it were underspecified wherever it
exists : since its backness plays no active role in phonology, only
[high] is to be viewed as distinctive ; (b) velar consonants are more
likely to undergo coarticulation and assimilation than those having
other place features, which is normal if it is assumed that there is
no velar place feature at all! The facts you mention remind me a
change occurring in the Low German dialect of K�ln (Cologne) :
coronals change into velars after high vowels (e.g., kink 'child, High
German Kind', hunk 'dog, HG Hund', luuk 'people, HG Leute', tsik
'time, HG Zeit'). McCawley's (1967) interpretation of this was that
velars are [+high] consonants. I think he was only wrong when he added
that labials and coronals are [-high] : I would say that they are all
'[+high]' since they are plosives ; labials and coronals (as /u/ and
/i/), however, have a place feature, unlike velars (and the high back
vowel), which are just [high]. In Carvalho (forth.), LG facts are
simply accounted for through OCP.

Carvalho, J. Brand�o de (2002). Formally-grounded phonology. From
constraint-based theories to theory-based constraints. Studia
linguistica 56, 227-263. [see first section]

Carvalho, J. Brand�o de (forthcoming). Repr�sentations versus
contraintes. Explications formelles et substantielles en
phonologie. A para�tre dans les Actes des 3�mes Journ�es d'Etudes
Linguistiques (Nantes, 21-23 mars 2002). Rennes : Presses
Universitaires de Rennes. [in French]

McCawley, J. ([1967] 1972). The role of phonological feature systems
in any theory of language. In V. B. Makkai (�d.), Phonological theory:
evolution and current practice. New York: Holt, Rinehart & Winston,
- ----------------
	From: "Dan Everett" <>

Thanks, Loren, for posting Dave's message. Mamainde is spoken about
600 miles or so south of Piraha and is unrelated to Piraha. However,
there *is* a process in Piraha similar to the one Dave mentions.

In my MA thesis (back in the buggy days) and in my grammar of Piraha
in the Handbook of Amazonian Languages, I mention that /h/ + /i/ or
/u/ is often (and especially in singing) realized as k. For example,
the negative word, hiaba 'no' is often pronounced as kaba. And the
word huaga (orthographically hoaga) 'nevertheless' can be pronounced
as kwaga or koga.

I analyzed this [k] as resulting from the conjunction of the features
[back], from the /h/, and [high], from the vowel. The labialization in
the case of /u/ is then free to connect to either the resultant k,
producing kwaga or the vowel, changing a to o.

Haven't thought about this for quite some time, but this still seems
right. So the Piraha process looks somewhat similar to Mamainde.

Perhaps someone on one of these lists will have seen something

Thanks again, Dave, for posting this.
Daniel L. Everett
Professor of Phonetics and Phonology
University of Manchester
- ---------

	From: "Mark Donohue" <>


This is brief, I don:t have access to my library )I:m lecturing in
Japan for a couple of weeks0.

The languages of the western Lakes Plain in Irian Jaya do what you:re
talking about, in both directions.

In some langauges a fianl consonant is allowed+ that consonant will be
r if the last vowel is e, a or o; it will be b if the last vowel is u;
and g if the last vowel is i. Clear relationship between high vowels
and high Cs.

In other langauges the final Cs have been lost, but there is
compensatory raising/friction on the vowels+ here a high C influences
the V.

I can send you references, or write to]

for more details.
- ------------

	From: "Kathy H." <>


I don't really have an answer to your question, but I wrote a paper
which united vowels and consonants into one system, for a graduate
course. Got a "C" on the paper. (Got an "A" in the course.)

Because the tongue is involved in the production of both vowels and
consonants, there HAS to be a way to unite the two. I found
information on Kabardian to be a bit useful, something by Choi and by
Wood, I think, but I'm doing this from memory now.

Anyway, I found your question interesting--and the language you're
studying potentially useful for supporting my proposed idea! I wonder
if you could tell me a bit more about the language, and if you want, I
can send you a copy of my (rotten) paper.

Thank you,
Kathryn Hansen, M.A.
Ph.D. student
Purdue University
- ----------------

	From: Lameen Souag <>

This may or may not be relevant, but in some Berber languages y and w
geminate to become /gg/ and /gg*/ (/ww/ > /bb*/ is also found.) Thus
Algerian Arabic ZRudeyya 'carrot' is borrowed as Kabyle ZRudegga, for
instance; ie, doubled (tense) high semivowels seem to become become

* indicates rounding.
- ----------------
	From: "Sudaporn Luksaneeyanawin" <>

Dear Dave Eberhard,

The term "spreading" has been used by us to refer to the distribution
of vowels in a vowel system completely covering the vowel area:
horizontally, front-central-back and, verically, high-mid-low.

We did not use this term in terms of process. However, there is an
evidence in the North-Eastern Thai dialect that high vowel /ii/ spread
the palatal quality to the preceding consonant (the onset) and change
this velar consonant to an alveolopalatal consonant (I represent it
with a [ch] because I can't insert IPA symbol here). Example: /phuu0
khiiaw4/ (green mountain) --> [phuu0 chiiaw4] /maaj3 khiit1/ (matches)
- > [maaj3 chiit1]

Sudaporn Luksaneeyanawin, Ph.D.
Director, Centre for Research in Speech and Language Processing-CRSLP
Department of Linguistics, Faculty of Arts
Chulalongkorn University
- ------------

	From: Jess Tauber

I remember in RMW Dixon's "The Languages of Australia" the description
of one language (whose name escapes me), where there is a palatized
velar phoneme listed in the inventory, this in addition to the usual
palatal position of articulation.

Dixon's theory at the time of the origin of unusual articulations
(such as retroflex and (inter)dental) was that the unmarked
antecedents interacted with adjacent vowels. Thus alveolars associated
with -u- and thus became retroflexed, or palatals associated with -i-
and thus became (inter)dental. He based his notions partially on the
idea that there were Jakobson-like zones of articulation. Thus gravity
he labels "peripherality" in articulation, given the obvious non-oral
centered nature of p/k, etc.

Still, to have a system with complete generality, one would like there
to be some greater symmetry to the entire series of changes. I
presented to him, a couple of years ago, my hypothesis of what this
might entail, using his own data from LOA. Turns out that if one looks
at manner of articulation (stops, laterals and rhotics, nasals), the
laterals plus rhotics set (which he didn't see as linked) tend to
cover more ground than the stops, which themselves tend to be more
symmetrically distributed in above sense than the nasals. Thus we have
an implicational hierarchy.

The graves/peripherals can split just as well as the acutes- they just
do so later in the implicational series. One of these splits involves
the -ky- mentioned at the beginning of the letter. Whether this has
more to do with -k- or more with -p- is an open question (my own sense
is that both members of the pair are involved, and this will also
eventually be shown to be the case with the acutes as well). The last
split here produces labiodentals, also extremely rare in Australia.

I tend to have an extremist geometrical bias to structuring
phonological systems. The classical p,t,ch,k set, for instance, for me
can be mapped as a tetrahedron. Splits in the system eventually lead
to production of a set nicely rendered as a cube. All
symmetry. Interestingly one can see very similar implicational series
in other language families worldwide. The details vary, both in terms
of the primary feature strings making up simple sets and which
elements these then interact with to create splits (vowels in some,
rhotics or laterals in others, nasals in still others, and so on). But
the total attestable set of sets shows only a handful of patterns,
involving branching due to different prioritization of features in
strings (so obviously amenable to an OT rendering).

Anyway, that's my take on things. Never published any of this, or
presented it publicly (except on the internet). Shifted priorities in
my life as well! Hope this has some bearing on what you're doing.

Best of luck
Jess Tauber
- -----------------
	From: "Torre, E.J. van der"

Dear mr. Eberhard,

I know of at least two (dialects of) languages where this phenomenon
also occurs.

In reply to another query, Tobias Scheer gives some velarisation data
from Cologne German where after high vowels, coronal consonants become
velar (see and see
also S�g�ral & Scheer 2001).

Braun [braUn] [bruN] 'brown'
mein [maIn] [miN] 'mine'
fein [faIn] [fiN] 'fine'

Zeit [tsaIt] [tsik] 'time'
Braut [braUt] [bryk] 'bride'
seit [zaIt] [zik] 'since'

bunt [bunt] [buNk] 'colourful'
blind [blint] [bliNk] 'blind'
Mund [munt] [muNk] 'mouth'

A more restricted process can be found in the Antwerp dialect of Dutch
(Taeldeman 2001). In this dialect, coronal nasals become velar under
two conditions, i) they have to be preceded by a long high vowel and
ii) this long vowel has to be short in the output (a nice opacity

grune [Gry:n] 'green' gruun [GryN] 'green'
schoenen [sxu:n] 'shoes' schoen [sxuN] 'shoe'
zonen [z2:n] 'sons' zoon [z2N] 'son'

Interestingly enough this process does not occur when the vowel is
already short in the input:

kin [kin] *[kiN] 'chin'
zon [zon] *[zoN] 'sun'

Perhaps less related, and maybe besides the point; there are also a
few processes where the coronal nasal changes to a velar nasal after
other vowels. In some Afrikaans words, for instance, the originally
Dutch schwa-n endings have changed into [-IN]:

Dutch: koren [kOrn] Afrikaans: koring [kOrIN] 'wheat'
Dutch: doren [dOrn] Afrikaans: doring [dOrIN] 'thorn'

And then there are the curious fact of Wieringen Dutch (Daan 1950; van
Oostendorp 2000). In this, now extinct, dialect of Dutch, coronal n
becomes N iff it is preceded by a low back vowel and followed by a
coronal (the first realisation is Dutch, the second Wierings):

 dans [dAns] [dANs] 'dance'
 hand [hAnt] [hANt] 'hand'
 mond [mOnt] [mONt] 'mouth'
 hond [hOnt] [hONt] 'dog'


 vent [vent] [vent] 'guy'
 kind [kInt] [kInt] 'child'
 lamp [lAmp] [lAmp] 'lamp'
 man [mAn] [mAn] 'man'
 bon [bOn] [bOn] 'receipt'

It seems as if this is a process of both assimilation to the vowel and
dissimilation to the coronal obstruent (that is of course if you
assume that low back vowels have a similar place of articulation as
engma (as I do)). This phenomena is also reported to have occurred in
earlier stages of Dutch, and Hoeksema (1999) shows that in 17th & 18th
literature speech of farmers was, among other things, indicated by
spelling 'mond' as <mongt> (mouth), 'onder' as <ongder> (underneath),
and 'Frans' as <frangs> (French).

I am currently finishing my dissertation (Dutch Sonorants), and there
I discuss these data in quite some detail. As you can imagine I am
intrigued by the example you mentioned in the query. Have you or
somebody else published on these data?

best wishes,
Erik Jan van der Torre

Daan, J. (1950). Wieringer land en leven in de taal [Wiering land and
life in the language]. Doctoral dissertation, Universiteit van

Hoeksema, J. (1999). Velarisatie van /n/ in plat Hollands
[Velarisation of /n/ in coarse Dutch]. TABU 29:2, 94-96.

Oostendorp, M. van (2000). Wieringse nasaalvelarisering [Wiering nasal
velarisation]. Taal en Tongval 52, 163-188.

S�g�ral, P. & T. Scheer (2001). Abstractness in phonology: The case of
virtual geminates. Constraints and preferences, edited by K.
Dziubalska-Ko&#322;aczyk, 311-337. Berlin and New York: Mouton de

Taeldeman, J. (2001). Vlaamse klankfeiten en fonologische theorie�n
[Flemmish sound facts and phonological theories]. Ms., Universiteit
- --------

	From: "Ron Artstein" <>

I have written a paper that you may find relevant:

 Artstein, Ron. 1998. The incompatibility of underspecification
 and markedness in Optimality Theory. In Ron Artstein and
 Madeline Holler (eds.), RuLing Papers 1: Working Papers from
 Rutgers University, pp. 7-13. New Brunswick, NJ: Rutgers
 Department of Linguistics.

You can download the paper from my web site:

The basic argument is that since an Optimality-Theory grammar cannot
give rise to an increase in markedness, an underspecified feature will
always end up being realized as the least marked value; this can only
lead to alternations if the least marked value is not the same in all
environments (e.g. Yoruba ATR harmony, where the least marked value is
[-ATR] before [-ATR], and [+ATR] before [+ATR]). This does not
necessarily lead to underspecification, but rather underdetermination:
for Yoruba prefixes that always harmonize we simply have no way of
telling whether the underlying representation is [-ATR], [+ATR] or
unspecified, and in a way this question doesn't matter since the
feature will be changed in the output anyway. Underspecification can
only have a desirable effect if you find alternations between
least-marked forms that coexist with forms that do not alternate: in
such a case the underspecified input will lead to an alternating
output, and the specified inputs will lead to the non-alternating
forms. This is not the case in Yoruba, where all forms harmonize.

My paper is a response to the following paper, which uses
underspecification in a way that allows an increase in markedness and
thus leads to undesirable typological predictions; in my paper I
explain why.

 Inkelas, Sharon. 1994. The Consequences of Optimization for
 Underspecification. Ms., University of California, Berkeley.
 Rutgers Optimality Archive ROA-40,

- --------------

	From: "Michael Johnstone" <>

Dear Dave,

There's a chapter in "Derivations and Constraints in Phonology" edited
by Iggy Roca (Oxford: Clarendon 1997):

Inkelas, Sharon, C. Orhan Orgun & Cheryl Zoll (1997) "The implications
of lexical exceptions for the nature of grammar." In Roca (ed.),

which argues on the basis of Turkish vowel harmony and final consonant
devoicing, and Spanish vowel alternations and other data, for
underspecification in the lexicon. The Spanish examples are old, going
back to early Generative Phonology approaches, but they've recouched
it all in OT.

They basically propose that non-alternating forms may be learnt fully
specified, but alternating forms will be underspecified. For example,
there is an alternation in Spanish between stressed [we] and
unstressed [o] as in ['bweno ~ bon'dad], but there are also instances
of [we] and of [o] which do not alternate when stressed/unstressed. So
the non-alternating forms are acquired as fully specified in the
lexicon, whereas the grammar settles on an underspecified form for the
alternating ones - in this case, [oX] with an underspecified
timing-slot - which then comes out as [o] or [we] according to stress,
due to some cunning constraints (which they do not give).

So it's not actually a matter of constraints on the input, but during
acquisition of the grammar and lexicon, underspecified forms are
allowed as the best input for these alternating surface forms.

As far as I remember, vowel harmony was also dealt with by
underspecification. I'm sure there's more on underspecification in the
Rutgers Optimality Archive
- ----------------

	From: "Ruth T Kramer" <>


I just read your post on Linguist List, and thought I could point you
to at least one article that is about the intersection between
underspecification and OT. I'm familiar with it because I recently
did a presentation on it for my phonology class. It is:

Ito, Junko, Armin Mester and Jaye Padgett. (1995) "Licensing and
Underspecification in Optimality Theory." Linguistic Inquiry 26,

The article discusses voice assimilation in Japanese, but maybe you
can extend the analysis to place assimilation too. It also
specifically addresses the concern of how underspecification should be
represented in OT (as feature of input/output)so I hope this helps!
- -------------

	From: "Irene Kr�mer" <>

Hi Dave,

sorry, I have no answer to your question. I am working withing a
group project on semantics, optimisation in interpretation, and I have
been asking myself how to deal with underspecification - in my view,
it would be a pity to lose underspecification as a concept in
semantics, so I've been wondering how it could fit in with OT. I hope
you will get some reactions, and that you will be able to post a
summary on linguist list.

Good luck with your work!

Irene Kraemer

Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue