PHONOLOGY OF ARnCUtATlON
PHONOLOGY OF ARnCULAnON
by
by
ALICJA GORECKA
AllCJA GOREC~A
Submitted to the Oepartmsnt of Linguistics and Philo.;ophy
on October 22. 1989 in partial fulfillment of the
requirements for the degree of Doctor
Philosophy in
Linguistics
Magisterium. Uniwersytet Warszawski
.,r
( 1979)
A9STRACT:
Submitted to the Department of
Linguistics and Philosophy
in partial fulfillment of the requirements for the degree of
In this thesis, I focus on the phonological representation of the
articulatory component of speech producti"o. I argue for a phonological constituent called the Constriction Node, which represents
the constriction gesture. Within the segmental tre9 (cf. Clements
(1985), Sagay (1986». the Constriction Node is a daughter of the
Root Node (in the sense of the Clemenis/Sagey model). It itself
dominates the Site. which correspof)~c; to the constriction location,
and the Articulator. which corresponds to the organ executing the
constriction gesture. The model of s~gment representation which
includes the Constriction Node is represented as follows:
DOCTOR OF PHlLOOOPHY
at the
MASSACHUSETIS INSTITUTE OF TECt-f\JOlQGY
October 1989
e
Alicja Gorecka, 1989
The author hereby grants to M.I.T. permission to reproduce and to
distribute copies of this thesis document in whole or in part.
Root Node
Signature of Author
Department of Linguistics 2ndPhiiosophy
Octaber.19B9
LBI"\IlO!f)'
I1enn er
Feet,,~
r eeture!
(ConstrIction)
Certified by
Site
Danca Sterrade
ProfesS"'·. linguistics
Thesis Supervisor
L ~n,lrh:tion
Arll tuleter
In this model. the Sits ranges over the features Labial, Anterior,
Palatal. Velar, and Pharyngeal. The Articulator can have either of the
values: Lower lip, Tongue Blade, Tongue Body, Tongue Root.
Accepted by
The constriction model reconciles two traditions of segment
characterization: the IPA tradition, which stresses the importance
Wayne O·Neil. Chairman
Departmental ComnliUee
ARCHIVES
MASSACHUSETTS INSTITUTE
OF TE~HNO'-OGY
AUG 18 1992
IIIClOa.gl~~
2
of the constrictiof location features. and the SPE (Chomsky and
Halls (1968» lradi!ion of giving priority to the features based on
the acl;ve articulator. It adopts the hierarchical view of segment
representation introduced by Clements (1985) and Sagey (1986). In
particular. it adopts the concepC of an Articulator as it is developed
in Sagey (1986).
The constncti~n model offers a view of segment represl=llntalson closely modeled on the physical reality which it represents:
the constrictioil t: stiJre is a necessary element of speech
production.
t'. ..
tion
On the phonologie:tl side. the rr.aio advantage of ihe constricmodal is that it makes ail articulation features equally
available to consonants and vowels; this is what allows it to
account for voweUconsonant interaction phenomena (e.g., mutual
vowel/consonant palatalization. vela!'ization. and pharyngealization.
among others). AlSO. by offering more stru=tured representations,
this model g09S even further than the previous models in the
direction of iimiting the number of phonological P:-OCSSS9S which do
not occur.
This th3Sis is organ;zed as follows: chaptqr 1 introduces the
constriction model. Chapters 2 and J argue for the phonological
constituent· Articulator. and lh9 phonOlogical const!tuent SHe
(respect;Yely) on the basis !1f the examples which can b9 anaiyzed
only in terms of these featur9S. Chapter 4 deals with the structure
within the Constriction Node. In chapter 5. the articulatory basis of
Site ans Articulator features. and the phonological motivation for
treating vowels in terms of such features are discussed. Chapter 6
discusses some residual questions in th'3 rep~ lsentation of consonants. Finally. chapter 7 considers the conssqt..ences that follow
from applying the constriction model in the analysis of various
phonological proC85ses.
Thesis Supervisor:
Title:
Danca Steriade
of linguistics
Pr~fessor
Acknowledgements
It has been a great privilege to have Donca Steriade. Ken Hale.
Jim Harris, Michael Kenstowicz. and Ken Stevens as members of my
committee. I won't even try to imagintJ what this thesis would have
been like without their help and guidance.
I am grateful to Ken Stevens for giving me so much of his time,
and for being such a kind and patient teacher. Ken's teaching has
inspired just about every idea in thas thesis.
It has been my goJd fOrtUr:9 to be working on my thesis while
Michael Kenstowicz was visiting at MIT. I am grateful to Michael for
the interest he h&s taken in my work, and for his art of painless
criticizing. Also. son.a of the nicest data in this thesis have been
pointed Olit to me by Michael.
I a:n grateful to Jim Harris for support and encouragement. as
well as the tough questions which have made me rethink a number of
points I have been taking for granted. I also want to thank Jim for
helping me wiih the mattars of presentation.
01 all the people I have met at MIT, Ken Hale and Dunca
Steriade havo affected my life in the way that no one ever has. I
have always been lao shy to thank them properly for everything they
have done tor me.
I want to thank Ken for being my friend in the best and in the
worst of times. for hours of conversation •• about lingUistics and
everything else. for letting me in on his way of thinking about
things. for pizza and Chinese food. for employment when I was
broke, but above all, for believing in me even when I myself had the
doubts.
Danca Steriaoe has been the best teacher I have ever had, both
in the c~assroom and in tile o~fice. Always expecting most from her
students and always fair. Working with her was the happiast cart of
my education.
Danca stole my heart fur phonology some time in the middle of
my second semester at ~AIT. and I've never r~vared. To her lowe
my fascination with feature phonology, which eventuaHy has led to
the propos;al in this thesis.
4
3
Speaking of the thesis, there is hardly an idea here which did
not impro.~ ~US8 of Danca's comments. I am grateful to her for
this contribution, and for guidance, and for the enormous amount of
time she has giv~ me.
During the time I l.!V8S working on this thesis. I have received a
lot of help from various people. For this help, whether in the form of
advice. or criticism. or suggestions, or data, I want to thank Susa~
Boyce, Lisa Cheng. Noam Chomsky. Morris Halle. Richard Heck.
Michael Heggarty. Michel Jackson. Itziar Laka. Li Ya'ei, Ian
Maddieson, Sharon Manuel, John McCarthy, Joe PerkeU, Alan Prince,
and Brian Sietsema.
For the care anti effort they have put into my education, I also
want to thank my other teachers, both at MIT. and at Indiana
University: Noam Chomsky. John Goldsmith, Fred Householder,
Richard Kayne, Richard Larson, WGyne O'Neil, and Luigi Rizzi.
One doesn't survive a graduate program at MIT without a circle
of friends. For tons of support. for dinners. parties, chats and
serious conversations, I want to thank my classmattts -- Andrea
Calabrese. Ewa Czaykowska-Higgins, David Feldman. Betsy KI;ppie.
Sandra Levenson, Katie McCreight-Young), Janis Melvold, Mark Ryser,
and Betsy Ritter. and outside my class ... Jose Boneau, Viviana
Deprez, Isabelle Haik, Michael Haggarty, Kyle Johnson. John Lumsc.1en,
Tova Rapoport, Peggy Speas. and last. but not least -- Li Yare; -- a
great friend on every occasion.
TItIEl Pil(38
1
Abst:rc:Iil::t.
2
Acknowledgements
'
TCIt»Ie of Cootents•••••..•..........••...••.••••..••••..••.......••.•......•.....•...........•......•...•.••............6
1.
Introduction
9
1. 1 The Phonetic Reality of the Constriction Model.
13
1.2
Naura-physiological Basis of Artj~!aiory Features
18
21
1.3 Phonological Motivation for the Constrictia" ~.. od81.
1.4 The Structure of the Constriction Model.....................•....•..................26
1.4.1
An Overview of Nodes and Features
26
1.4.2
The Structure of the Repres8ntati~n
Simple VS. Complex Segments
30
2. Articulators
32
2. 1. Lower lip
__
32
2.1.1 St.x::iarl9S8 ArcIIJic...............•...•••.......•......•.....•..•.••.•....•....•.•..........•••...••..•..••.. 3 4
< .0 • • • • • • • • • • • • • • •
22 TOflgllQ ~••...•••••.•••.•••.•••.••.•••••.•••••••••..•••••••.•.••••••...••••.•.•.••.•.••••••••••.••••••••••.35
2.2.1 StJdarlese ArcItJic:..•..•......•••••.•.•••...•.•...•.36
2.2.2
Classical Arabic
2.3
Tongue Body
_
36
37
2.3.1 St.dartese ArcIIlic:
40
3. Sil9S
3.1 l.a.biaI Sits
41
:
41
3. 1.1 Velarization in Veflda..............••...............................................................4 1
3.1.2
Palatalization of B~labials in Zulu
_
43
3. 1.2. 1
3.1.2.2
3.1.2.3
3.1.2.4
3.1.2.5
3.1.2.6
_
3.2 Antaric>r Site...............
_.._.._
44
46
50
52
_
53
55
.
56
3.2.1 Ve!arization of Anterior Consonants in Standard Thai
(Siamase}
57
3.2.2
Cases in Which Labiodentals are excluded from the
Phenomena affecting Dental/Alveolar ConsQnants
60
3.3
Palatal Sito.............................................................................•..•....................65
3.3.1
Palatalization
Unstressed Vowels in Russian................... . 66
3.3.2
ale Alternation in 8ulgarian
_
74
3.3.3.
The ale and ole Alternation in (Old; Polish
77
3.3.4 ale Allemation in Arumanian
79
3.3.5
Palatalization of lal in Margi
80
3.4
Velar Sile
83
3.4.1 Velar Palatalization in Polish
83
0'
5
6
3.4.2
3.5
Vowel Hardening in Kinyarwanda
88
fJtaaI~ ~••••••_._
95
3.5.1 Vowel LowerirIg in Tiberian J-tebrew.•.•..••••..•..•............•.•.....•.•••.........98
3.5.2 Ablautirlg Vowreis ir1 Arabic.......•........................................................... 102
3.5.3
Vowel lowering Before Uvulars in Greenlandic Eskimo
109
4.
Site and Articulator Within the Phonological Tree
The Constriction NDCte•.__
._ _
11 2
4.1 The Constriction Model vs. The Model with 8 Place Node
Dominating Sites and Articulators
__ __
11 4
4.1. 1
Palatalization Spread in Polish
11 5
4.2 TIle SI8tIJS of tt. PIcM:::e ~
11 9
4.3
The Structure Within the Constriction Node
120
5. Segment Representation
Vowels
.._
__
_..__
124
5.1
The Articulatory Basis of Sile Features in Vowels
126
5.2
Phonological Argumer:ts for Site Features in Vowels
130
5.2.1
labiat Site in Vowels
_,
_ _ _
130
5.2.2 Palatal Site ill VCJ\l!r8ls
132
5.2.3
Velar Site in Vowels ..
_
134
5.2.4
Pharyngeal Site in Vowels
136
5.3
The Articulatory Basis of Site F9atures in Vowels
137
5.3.1 The A.~latory Basis of the Labial S:te
1 37
5.3.2
The Articulatory Basis of Palatal Site
_
_ 138
5.3.2.1 Velarizatic)n irI S~ Tllai
1 39
5.3.2.2 Anterior Vowels in Other Languages
14 1
5.3.2.3 Second Velar Palatalization in Potish
143
5.3.3 Articulatory Basil of the Velar Site in Vowels
145
5.3.4
Articul:2lory Basis of the Pharyngeal Site in Vowels
153
5~4
Articulator Features in the Representation of Vowels
157
5.4.1 T11e Tc:xlgtJ8 Bcxty
157
5.4.1.1 The Feature (high]
158
5.4.1.2 The Feature (bac.k] ..- ..__
183
5.4.2 The ~ ~
_
168
5.4.3 TOI'lIglJ8 Ba:Ie
169
5.4.4. The Tongue Root
_
172
5.5
Evidence for the Constriction Node in Vowels
1 76
5.5.1
V,Jwel COale~nce
- .._
176
5.5.2
Vowel Raduction _
_
178
5.6
Vowel Representation
Residual Problems .__.._
__
_
180
5.6.1 [)o,rsaI Pharyflg8aJ Vowels
1 80
5.6.2 Front Rounded Vowels _
_
_
1 82
6. Rep:essntation of Speech Sounds
n
7
•••••••••••••••••••••••••••
Consonants
1 94
6.1 l"t't& R~entaOOn of /If..'J••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••• 184
6.2
The Representation of lei _.._
18 S
6.3 lJvulars as Complex Segm8flts,.••...•............•.•....... ~
189
7. Phonological Processes
192
7. 1 ~me Predictions about the Processes Which Affed Nonadjacent Segments
_
192
7.2
A Problem for the Constriction Model and Possible
Solutions
-_ _ _
198
Bi)tic)g~
207
8
1.
Introduction
In order to communicate by means of speech. humans must be
able to aller the shape of ,heir vocal tract so as to produce sounds
which the human auditory system is able to distinguish. There are a
number of things wa can do to create the contrasts we can hear -we can vibrate (or nol) our vocal cords. lower (or raise) the basic
frequency of vibrations. lower (or raise) our velum. blow the air out
(or draw it in). but most likely. Ihsse gestures alone would not
suffice to develop a good phonemic inventory. In fact, no known
human language relies solely on these gestures.
W~thout doubt, the richest source of audible contrasts are the
gestures executed with the tongue and the lips, which form the
articulatory component of speech production. The exact role of these
gestures has hact various interpretations in the field of phonetics. In
one school of thought (e.g., Ladefoged (1962). (1975). CaUord
(1977). Malmbarg (1963), sic.). the activity of the tongue and lips is
interpreted differ9ntly for consonants and for vowels. In the
production of consonants. the Tole of the tongue and lips is to
obstruct the air flow -- complotely for stops. and partially for
continuants. In the production of vowels. neither the lips nor the
tongue are thought of as obstructing the airflow; rather. they
assume various shapos and ultimately modify the shape of the vocal
tract.
This view of speech production is reflected in the
characterization of speech sounds by the International Phonetic.;
Association. The IPA system characterizes conson~nts in terms of
the point of maximum constriction. and vowels i~ terms of the
position of the highest point of the tongue.
More recently. a different view of speech production, based on
the notion that all speech sounds involve a constriction gesture. has
dominated phonetic res9arch~ It is proper to trace the origins of this
view to the research don~ in acoustic phonetics on the relationship
9
between the acoustic properties of speech sounds and the vocal
tract con!iguralions and particularly to the work by Stevens and
House (1955), who have exlendod the concept of constriction to the
treatment of vowels. Th9~r work has influenced the tone of
discussion for both acoustic and articulatory phonetics (ct., Steyens
(1972), (1988). Fant (1960, 1973). Jakobson and Waugh (1979), Wnod
(1975). (1979), (1982), Perkell (1979). Perkell and Cohen (1987),
(~989). Perkell and Nelson (1982), (1985), Mrayati. Carrs. and Guerin
(1988). Carre 3nd Mrayatl! (1989). etc.). but has had very little effect
on the phonological treatment of speech sounds.
The model of feature representation proposed in this thesis is
designed around the concept of constriction. Constriction is
represe'tied as a function of two variables: the Site (- point of
articulation) and the Articulator. where the Site may range through
Labial. Anterior (cover for the dental and alveolar sounds). Palatal,
Velar and Pharyngear specifications. The Articulator may take on
the following value!:: Lower Lip. Tongue Blade, Tongue Body, Tongue
Root and Glottis. Let us represent this model as follows (the tree
below contains only the articulatory features (without the stricture
features. though); it ignores all other speech production processes
(phonation. nasality, and airstream mechanism»:
( 1)
Root Node
A
Constrtctlon Node
A
51 te
(Constrictfon Node)n
Art1culetor
~h9 model in (1) formally expresses the idea that out of all possible
gestures that can be executed with the tongue and lips. only those
• Follow'~g McCorlhy (1969>.
10
that reBuit in a ~onstriction can create vocal tract configurations
suitable for the production of SDeech sounds.
The model in (1) claims also that all articulatory
configurations that can result in a constriction gesture (in other
wards. all configUl'stions that are physically possible) will result in
a possible speech sound. A comparison between the articulatory
spectrum of an average hUm9'1 and the inventory of the attested
speech sounds (e.g., Maddieso:l (1984»
reveals that this
generalization
is
larg9!y
correct.
Consider
all
logical
SiielArticulator combinations, with the phys· :ally jmpossible onss
marked with an asterisk: Labial/Lower lip, labial/Tongue Blade.
labial/Tongue Body, ·labial/Tongue Root. Anterior/Lower Lip,
Anterior/Tongue Blade, Anterior/Tongue Body, • Anterior/Tongue
Root. ·Palatal/Lower Lip, Palatal/Tongue Body, • Palatal/Tongue
Root. ·Velar/lower Lip, ·Velar/Tongue Blade, Velar/Tongue Body,
·Velar/Tongue Root. ·Pharyngeal/Lower Lip, ·Pharyngeal/Tongue
Bla~8, Pharyngeal/Tongue Body, Pharyngeal/Tongue Root. After
weeding out the starred ones, we are left with ten combinations
which represent physically possible gestures. The only ones in this
Group that may raise doubts as the candidates for possible speech
sounds are: labial/Tongue Blade. Labial/Tongue Body. and
Anterior/Tongue Body. The remaining combinations pattern onto
attested speech sounds as follows:
(2)
Labial/Lower Lip:
p, b. m,
Anterior/lower Lip:
Anterior/Tongue Blade:
t. d. s. z,
Palatal/Tongue BI:tde:
c, di, s, i
Palatal/Tongue Body:
Velar/Tongue Body:
Phsr ?ngeaVTongue BGdy:
Phar/ngeallTongue Root:
c, J, ~, i. e
k. g, X. );, Q, X,
f. Y,
Q,
fB,
U, 0
rn
9, ~,
~, U. 0
Q
h, <I, e
As for the three questionable ones -- the Labial/Tongue Blade
combination is utilized by soma languages, as demonstrated by
Maddieson (19B7). Also, there are arguments for positing the
II
existence of vowels which involve an Anterior/Tongue Body
constriction (see 5.3.2 for the discussion). Only the Labial/Tongue
Body combination does not appear to be utilized by any language,
despite the fact thai it is possible for a human to uttAr a
labial/Tongue Body sound. Note however. lilat such a sound would
involve the longue blade either protruding from the mouth, or curled
up underneath the tongue b~dy. 80th of these gestures are
articuJatorily awkward, and do not occur in any speech sc,und. It is
conceivable that the conditions on possible speech sounds prohibit
such gestures.
The approach I am taking here is not unreasonable: there are
numerous restrictio. ',s on possibls speech soundp, which have to do
with the physical limitations, and which must be treated as such
under any model of phonological representation. A speaker of a
language does not need to be instructed not to uUer a pharyngeal
nasal, an oral labio·dental stop, or a nasal fricative, given that such
sounds cannot be produced. In other words, the model of linguist~c
competence should not be concerned with physical impossibilities.
Most likely, impossible feafure combinations such as Tongue
Root/Anterior are ruled out by the same mechanism which prevents
a normal human being from voluntarily bending the knee the wrong
way.
The model in (1) conceptualizes the articulatory component of
speech. In that respect, there is nothing new about it. as mo~t
'eature classifications stress the importance of the articulation. Its
main novell:' lies in the fact that it departs from the tradition of
expressing the articulatory features eilhtir primarily in terms of the
constriction location (as in the IPA system of treating consonan~s),
or primarily in terms of the features describing the beha~ior of the
articulators involved in sp9&Ch production (as in Chomsky and Halle
(196B). and the; proposais that have recently evolved from theirs, ct..
Halle (1983). Clements (1985). Sagey (1986». Instead, I argue th31
both aspects of speech sound production aro phonologically relevant,
12
and I propose
representation.
to
incorporate both
in
the
model
of
feature
There is yet another difference between this proposal and the
previous ones: it introduces the idea that the mental representation
of a speech sound includes a constriction node which mediates the
relationship between the Situ and the Articuistor leatures. Since
this model has no means of representing sounds in terms of an
articulator or a site which do nol panicipate in a constriction, it
makes a very strong claim about the nGture of phonological
representation of speech sounds. namely. it ilredicls that only the
features of the site and the articulator involved in a const~iction
may be phonologically relevant.
The main (and perhaps. the ultimate) strength of a feature
representation lies in its explanatory power as applied to
phonological processes. conditions on representations. observed
phonemic contrasts, etc. The prosent thesis is almost el~tirely
devoted to the goal of defending the constriction model on these
"rounds. However.. in order to justify attributing certain properties
to this model or its various nodes, I shall appeal at times to various
physicai aspects of speech production. To make these moves licit. I
shall begin by establishing the links between the components of the
constriction model and the phonetic reality which it (indirectly)
represents.
1.1
The Phonetic Reality of the CrJnstriction Model
II is an interesting question ~also one that I will not pursue
too deeply), how closeiy phonological representation should
resemble phonetic representation. While there are obvious
advantages to trealing phonological segments in terms of labialily,
coronality. or anteriority. that is. in terms of the configuration of
the vocal Iract necessary for the production of the sounds which the
phonological s9gments represent. it is fairly obvious that certain
aspects of phonetic representation do not have phonological
counterparts. Some well known examples of this include thase:
13
speech is a continuum that does not lend itself to ~ny simple
segmentallon. but phonological segments are discrete 'Jnits; speech
sounds occur in time, and are by no means the same at all stages, but
phonological segments are icons which represent various target
positions for various articulatory structures, not necessarily
simultaneously achieved (in other words. for some sounds there is
no -time slice- which would correspond to the phonological
representations associated with these sounds). These facts show
that bringing phonology closer to phoneticg is not an a priori
desirable objectjve.
We are not yet sufficiently knowledgeable about the processes
which relate phonological to phonetic representations, to decide
apriori where the boundaries of phonology-pnonetics correspondence
lie. The position that I take in tttis thesis with regard to this
problem is pragmatic: to the extent that certain properties of
phonetic representation can be useful in accounting for aspects of
phonology, they should be incorporated into phonological
represen fa tion.
I appea~ to the phonetic reality of the constriction gesture in
order 10 be able to attribute to the Constriction Node the properties
with which the constriction gestufe is associated. namely. having
on. location and one articulator which executes it Of course, If the
Constriction Node has a ·physical reality-. so must the nodes under
it. This means that the site and the articulator faatures should map
uniquely onto real articulatory properties of sounds.
It is rather easy to demonstrate that about the articulator
:eatures: X-rays and oiher phonetic records show that the prirnary
articulator a.,d its shape ar9 largely invariant for all tokens of
(what W9 perceive as) one sound. A;so. we can call on the
considerable success of phonolt..gical theories which have based
phonological features on tho behavior of the articulators, for the
support for this aspect of .he constriction model.
14
As for the ~ite features, we are in a somewhat different
situation. While the importance of the constriction location in the
production of consonants is well recognized (ef. the IPA treatment
01 consonants), there has been a considerable tradition of
characterizing vowels without any reference to constriction.
Instead. the model of vowel articulation that has dominated
research in phonology, and to some ex:ent. in phonetics, over the
last century, is the tongue arch model introduced by Alexander 8011
in 1867. In essence, this model expresses the articulatory
properties of vowels in terms of the position of the highest point of
the tongue arch in the front-back and high-low space.
The tongue arch model has been widely accepted in phonology,
as ~t seems to be fs!rly well suited for expressing the p~"onological
processes aHecting the vowels (such as the backness and height
harmonies). In phonetics, on the other hand, the success ~f this
model has been dimmed br X-ray findings (Meyer (1910) . Russel
(1928), etc.}, which !lave revealed a discrepancy between the
predictions that this model makes with respect to the "eight of the
vowels, and the height of thEi tongue arch actually observed. The xray tracings of vowels from languages that display the lense·lax
opposition show persistently higher ~ongue position for the tense
vowels characterized as mid by the model, than for the lax voweln
which are labeled high.
Dissatisfaction with the tongue arch model has led to a number
of proposals (Jakobson, Fant. and Halle (1952), Jakobson and Waugh
(1979). ladefoged. DeeJerk, lindau, and Pap~un (1972), 9tC.~ which
viewed vowel production in terms of reaching acoustic targets, by
possibly variable articulatory means.
Renewed interest in the articu:atory aspects of vowel
p.oduction has sprung up in the context of research whict. studies
the relationship between the acol-stic effects of speech sounds and
the vocal tract configurations {Stevens and House (1955}). S!6vens
and House view the production of vowels as !nvolv;ng a constriction
vowels in terms of the location and the degree of the constriction,
and the degrft8 of lip opening. These parameters, when translated
into numbers (con!!triction location -- in terms of its distance from
tho glottis, degree of constriction -- in terms of the radius of the
opening between the articulator and the upper surface of the vocal
tract at the constriction), suffice to deter,njne the dimensions of
the rest of the vocal tract. Given these, the acoustic properties of
speech sounds can be computed.
Tho -quantitativa- model of vowel articulation develo9ed by
Stavens and House (1955) has influenced acoustic research in vowel
prOduction, e.g., Fant (1960), Fant (1973), Lindblom and Sundberg
(1971), Stevens (1972), (1988), Mrayati, Carre, and Guerin (1988),
Carre and Mrayati (1989), etc. It has also inspired the articula!ory
studies of vowels (e.g.,Wood (1975), (1979), (1982), Perkell (1979),
Perkell and Cohen (1987), (1999), Perkell 2nd Nelson (1982), (1985),
etc.). ~4ost importantly, ~t least from the point of view taken in this
theSIS, this model has opened a possibility that the constriction
location might represent an invariant aspect of vowel articulation,
and therefore may satisfy the requirements of a phonological
featurs.
This view was taken up in the work of Sidliey Wood (mentioned
above). Wood (1979) rejects tne tongue arch model of Alexander Bell
(1867), arguing that the disparity between the predicted and
actually obsorved tor.gue heights disqualifies this model as a tool
for describing the articulatory properties of vowels. Instead, he
advocates the return to the tradHion
characteriziilg vowels in
terms of the place oi maximum constriction. (The articulatory model
based on the constri\:tion location has been known and handed down
to us by the Sanskrit grammarians, cr. Varma (1929), Allen (1953)).
0'
On the basis of X-ray tracings of 38 vowel sets, Wood argues
for a model of vowel pr~duction which distinguishes four regions of
maximum narrowing in the vocal tract: palatal (for i-like and ,ft-Jike
in the vocal tract. :and they express the articulatory properties of
15
16
vowels}, palato-velar (for u.-liks vowels). upper pharyngesl (for
like vowels) and lower pharyngeal (for a-like vowelsf.
a.-
Wood's additional arguments for expressing articulatory
proplli1ies of vowels in terms Qf the place of maximum constriction
come from physioiogica' considerations. He analyzes the position
and the movemenf of the extrinsic tongue muscles: genioglossus,
styloglossus. hyoglossus. and palatoglossus, and concludes that each
of these muscle groups is ideally situated for executing tongue body
cOflstrictions at the four target locations.
In a similar vein, Stevens and Perkell (19n) point out the
inadequacy of characterizinv features (low) and [high] solely in
terms of the high-low position of the midline of the tongue. Instead,
they propose to specify these featuras in terms of -tongue-surface
contact with other surfaces· (Stevens and Perkell (1977). p. 330).
Virtually all of Woed's strongsr conclusions (e.g.. about the
phonological adequacy of his model. about the sufficiency of U'!@
constriction location and jaw opening parameters for artfculatory
characterization of vO~9lst and even about invariability of the
constriction location for some vowels. have been challenged: by
Halle (1983). on the strength of phonological arGuments. Catiord
(1981). and Jackson (1988) on the phonetic grounds, and by FischerJorgensen (1985), on both grounds.
In the chapter on vowel representation (chapter 5). I shall
consider all of these criticisms. and try to develop a view of vowel
representation that is both articulatoriiy and phonologically
acceptable. The result of Wood's work that will be dQfended and kept
as the basis of phonological representation of vowels proposed in
this lh ~sis is that the region of constriction constitutes a property
of vowel articulation s~able enough to be used for vowel
id8~tification. The claim that production features which are based
on the behaY~or of the dorsal art:cu:ator are not relevant in the
phonological characterization of vowels win be rejected.
1.2
Neuro-physiological Basis of Articulatory Features
A necessary criterion for a phonological featura i9 that its
execution be verifiable by the brain. Neither the articulators (lower
lip. tongue blade. tongue dorsum) nor the sites (labial. anterior.
palatal, velar. etc.} can communicate with the brain directly via
molor control: ~i1e articulators are rather passive bodies of flesh
which are mClnipulat9d by the various muscles of the tongue (or. in
case of the lower lip, the mUl~~les of the lip), and the sites
con9titute a fixed structure. Whet then allows the brain to operate
an features like [anterior)' or [coronal)?
It has been suggested in the literature that the implementation
of phonological features is aided by a tac.tile experience: duo to high
sensitivity of both the articulator surfaces and the walls of th~
Yocal tract (evideficed. for example. by the results of the
experiments testing stereognostic· capabilities of humans, ef.,
Grossman, Hotlis, and Ringel (1965», th9 brain is able to evaluate
both the regions and the extent of contact between the articulators
and the vocal tract surfaces.
Studies t.~ve t'gen dona which reveal a direct relationship
between the articulatory skills and £tereognostic ;J.ciiities in
humans. For example, in an experiment reported by Ringel. HOllse.
Burk, Dolinsky, and Scott {1970~, stereognostic abilitias of children
with no speech disorders and children with mild-te-severe
articulatory disorders were comrared. Th9 results of the experiment
have shown that children with articulatory disorders m=-ke more
errors than children with no disorders. and that the number of errors
increases with the degree of the impairment in a child.
• Obviously.. therl or. dlfferlnCIS betwlen Wood·, proDosal end m!n•. os far es tho
Inventory of Sltl. (Pflmary constrtctlo., locatlDn!) Is concerned. Thesl differences
..",111 bl discussed In more detail In 5.3.
·On thl bBSls of Ilectromyograohic Ivldence csn:J e.netomtcel d••crtot Ions.
I7
• Oral ster.ognosl' : Cblilty to recognlzo sheDes 01
18
obJ.ct~
Dlacod In mouth.
In a paper w:-'ich 8xp!ores the concept of orosensory·
correlates of phonological features, Stevens and Perkell (1977)
show that commonly postulated articulatory correlates of a number
of features (9.0... [high]. [low). [anterior]. etc.. ), which orient the
articulators in space. do not represent the actual gestures wit~
which these features are implemented.. They propose to rAplace the
traditional articulatory definitions of features with the orosensory
correlates which are -distinctive tactile ~nd proprioceptive
responses wi'lhin the vocal tract- (Stevens and Perkell (1977). p.
325) .. For example. the feature [+anterior] is characterized in this
framework as involving a contact between the tongue tip and the
edges of the tongue blade with the p31ate at its periphery. which
gives rise to the tactile stimulation in both tho tongue and the
palate.
Ideally, all phonological features which refer to articulatory
structures should be definable in terms of orosensory responses that
they trigger. In the phonological representation proposed in this
thesis. the Constriction Node and most nodes be~ow it are easily
characterized in those terms. This is trivially true of the site and
articulator features.. As for the features oorninated by the
articulators (e.g.. (distributed), [high), [back), etc.), most of them too
yield to orosensory interpretations: for example. the feature
(distributed] in its negative value can be characterized by
orosensory responses in the tongue tip ~rea {contact in the case of
stops. and turbulence in the case of fricatives} and no such
responses in the tongue blade area.. In its positive value. this feature
is associated with an orosensory experience along the tongue blade..
The feature [lateral] is characterized by a break in the contact
between the sides of the tongue blad't and the palate. and the
turbulence in those areas.
• SttYlns end Porklll deUne this term os rol1ows~ .hl term orosensory I..) forer,
to tha sensory systems In the yarlous Yocel-trQct !tructures from thi Je.rynx to
the IIpl. It Include, s.nlatlons dUI to leetH. $Umuletlor... pressurls. end turbulent
elrflow. ISS Will ~! sonsory Informetlon from musclQ sp5ndl.s. tendon org~ns ond
JC1lnt flcIPlors- (Sleyen, end P,rkel1 (1977>'
I9
p. 325).
There are, however. features which are not essily interpreted
in terms of the orosensory response.. For example. the feature [back).
which traditionally has been characterized in tern1S of the tongue
body position in the front-back space, resists such a treatment. al
least in terms of the tactile response.. •
A possible orosensory
correlate of this 'eature could be the motor activity required for
shifting the tongue body forward and backward.
EV'l" more troublesome in that respect are the stricture
features. Here, however, the difficulties lie (probably) in the fact
that we are far from understanding the phonology of these features .
For example, observations of various sonority hierarchy phenomena
suggost that there is one stricture feature which ;s multi-valued,
and which ranges over stop· fricative - nasal - liquid - glide .. high
- mid - low vah~9s. However, with the excepti9n of height harmonies
among vowels, there are very few phonological phanomena which
manipulate these valu·:ls. Again, the fact that stricture fealunts can
be nlanipulated among vowels. but not among consonants is puzzling..
As for positing an articulatory correlate of a single stricture
feature, one that would trigger distinct orosensory responses for all
its values, we are probably very much at the guessing stage. As I
have already observed. the degree of constriction opening is not the
most felicitous choice for characterizing vowels, as it does not
diff9fentiate between vowels as distinct as i and a.. A possible way
of looking at this 'eature would be in terms of the degree of the
airflow obstruction, where the unobstructed airflow would be
characterized by the vocal tract configuration for breathing. What
gives this approach a certain attractiveness. is that it affords us a
5 Stevens and Perkell (1977> sugglst thet (- b8lcklinvolv8S th. conleet
or thl
edg •• of the tongue tip wUh the lowor- Incisors. wh.reas (+beckJ rlQulrl1 that
thlre be no such eontect. I do not consider this approach. tor two reasons: first It
apD.crs D.rflctly neturel ror me to meJce cont"ct between the tongue tip end the
lower Incl!or!l for any vowel EVln" this contect Is not reQulreciln the ces. of
b8CIc vow lis, tho fecl thet II Is optional somehow Implies thot b6Ck vow.ls C$n bo
ootlonaJly 1-~GCk' (11). Second. the D~chltectuiCt of thl moall thet I em develoPing
hertl dlsallowg characterizing e sound In terms of the ccliliity of lhe ertlculolor
which does not DerllclDstl In the constriction.
20
unified view of the physical reality behind the sonority hierarchy
phenomena. We can associate various steps in tha sonority scale
.ith varying degrees of air pressure in the cavity behind the
constriction, prior to the release. Highly sonorous sagmenls would
be associated with small air pressure, and the segments at the
opposite end of th:- scale would be characterized by high air
pressure behind the constriction. We could h~poth8Sizf' thai sonority
constraints may really be constraints on the air pressure
main:enance in the vocal tract: steady decrease or increase may be
preferred over the choppy pattern.
It is clear that tho understand~ng of vowel/consonant
interaction phenomena is limited by the extent to which "OWt!ls and
consonants are allowed to share phonological features. Since none of
the currently available feature classifications provides -equal
feature access" to vowels and consonants. none is capable of
accounting for these ph8no~ena fully.
Consider, for example the feature systom which underlies the
characterization of sounds in the International Phonetic Alphabet:
this system characterizes consonants strictly in terms of the
constriction location features: labial, labiodental, dental alveolar,
palatoalveolar, palatal. velar, uvular, pharyngeal, and laryngeal: and
vowels -- strictly in terms of the tongue body features:back. high,
low. Obviously, no vowel/consonant int~raction processes can be
exp'ained in this system.
l
1.3
Phonological Motivation for the Constriction Model
On the phonological sid9. the constriction model draws
support from the fact that it is capable of explaining a variety of
phonological phenomena which have been left unaccounted for by the
earlior models of segment representation.
One of the outstanding problems in phonology are vowel
consonant interaction phenomena. These include palatalization of
consonants in the environmeni of iii and leI-like vowels,
pharyngoalization of consonants in the environment of non·high
vowels. pala:alization of vowels before cc.nsonants with the palatal
constriction location, vowel lowering conditioned by uvular.
pharyngeal and laryngeal consonants (cf. McCarthy (1989),
labialization of consonants by vowels. diss:milatory processes
involving consonants and vowels, vowel hardening and consonant
softening, etc.
Soma of these processes have already ,-eceivod satisfactory
accounts: this is true. for instance. about round vowel/labial
consonant interaction phenomena. which have been insightfully
analyzed by Sagey (1986), Yip (1988). others. Not surprisingly. the
language of the phonological representation introduced by Sagey
(1986) aliows labial consonants and round vowels to share a
phonological feature to the exclusion of all other segments.
21
Next. consider the feature system of Chomsky and H::!!e
(1968): it is based on the assumption that primarily, the moving
parts of the vocal tract. not the regions. form the basis of the
articulatory features (feature [anterior] is an exception in this
system. S98 the comment below). This view also forms the basis of
the propos~ls in ~fall9 (1983) and Sagey (1 ga6) (I will refer to these
proposals a.; the -articulator model-). in which the concept of the
articulator, corresponding to the anatomical organ that executes thE'
constriction. is made sxplicit and fully developed. While these
proposals afford a more uniform view of vowel and consonant
phonology, as they represent all segments in terms of articulators.
surprisingly, they are not capable of addressing the issues of
vowel/consonant interaction, except for the cases in which the two
are allowed to share the articulators. However, aven in these cases,
the explanation is not always available. For example, the velar Ik/.
owing 10 iis characterization by Sagey as a clain dorsal, has as much
in common with 1'19 vowels articulated in the velar region as it does
with all other vowels (owing to the fact that !n Sagey's system all
vowels are dorsal).
22
Interestingly, the SPE model (as well as the ClementsiSagey
model which it inspired) does recognize (even if not explicitly) the
limited need for the constriction location features. For example. it
posits the feature (anterior) in order to account for the phonemic
contrasts under the coronal node. There is no correlate of [anterior)
other than the constriction location. An unfortunate consequence of
making (anterior] a dependant of the Coronal Node (as in Sagey
(1986)) is that only the coronal segments can be characterized in
terms of this feature. This automatically rules out a possibility of
an assimilation uccount of the palatalization phenomena within the
articulator theory: while palatalization is triggered by segments
which are dorsal (!ront vowels), its output are the coronal sounds·.
From a purely phonetic point of view. it is obvious that many
examples of vowel/consonant interaction involve processes that are
sensitiv6 to the location of the constriction, not to the articulator
which executes it (s8e chapter 3 for such examples). Since the
informgtion about the constriction location can be manipulated by
phonological rules. it must be stored in the form accessible to such
rules. in other words. it must have a mental representation. The
constriction modeJ gives mental representation to the constriction
location, in the form of Site features, argued for in chapter 3.
Support for the constriction model comes also from the fact
that a significant number of phonolo~ical contrasts cannot be
cAptured without the Site ieatures. Consider the contrast between
the bilabials and labiodentals (/+ 1 vs. IfI; I PI 'Is. Ivl), utilizod in
languages like Ewe (Warburton at al. (1968)). Since both classes of
sounds are represented with the Lat'ial Articulator in Sagey (1986)),
and since this articulator does not dominate any constriction
location feature on her model, both classes of segments are assumed
• Although. elamlnS hey. De.n med. to Gccount for pelalallz!'tlon w Ilhln thl limits
Imposed by thl -~rtleulotor- theory. ror ,xampll. Clements (1976) treals frorat
!!,
vowIl1
coronol. ond analyze, psJetaJlzation ~s -coronallzotlon-. This account.
Phrased within lhl SPE terminology. hel blDII translated Into the C1Dmints/Sogey
modal by Ito and Mesler (1989) Thera ore Dnumbor of unfortunetl cons.aulnces of
such on 8nDlySlt~ which Dr. discussed tn detail tn 5.4.
23
to be non-distinct articulalorily-wis8. The SPE-suggesled treatment
of '+I-/f! contrast appeals to two features: (strident). and
[distributed]. However, phonological evidonce suggests that this is
not a possible solution: IfI. which according to this treatment is
supposed to be [-di::ltributed, +strident] never patterns in the
processes in which these features are activated. This is shown in
some detail in chapter 6.
On the constriction model, the I+I-/fl contrast is treated in
terms of di"erer,t Sites: L3bial in 1+1, and Anterior in Ifl.
Maddieson (1988) reports enoeher contrast which the
articulator model cannot capture: in a number of Austronesian
languages spoken on the islands of Espiritu Santo: Tangoa. Araki.
Mafaa. Aore.Vaa, Mpotovoro, etc.• phnnemic contrasts obtain between
bilabials, dentals (or alv90Ia,-s). and the sounds articulated with
either the tip or the blade of the tongue a~ainst lh9 upper lip. It is
clear that the feature [anterior], which the articulator model
employs in order to account for the ph"nemic contrasts under the
Coronal Node. cannot account account for the phonemic catrasts
among the coronal sounds in these langu'ages. Instead. what appears
to be needed to characterize these contrasts. is the reference to the
location of the constriction.
The Dorsal Articulator can execute a constriction at three
(possibly four) different locations: against the hard palate, soft
palata. and in the pharynx. Among vowels, the resulting contrasts
can b~ handled successfully in terms of [high], [!ow], and {baCk] -features which are independently mo~ivated for the vowels .
However. there is at least one contrast among the dorsal consonants
which these features definitely cannot handle: it is the contrast
between the palatal stop a!1d fronted velar. While few languages
have underlying fronted velars in their inventory, most have a
process which fronts velars adjacent 10 front vowels. Often
languages that have also palatal stops (Macedonian t Basque).
contrast the two sounds before front vowels.
24
Given the limitations of the articulator model, it can, at best
treat both sounds as [-back]. On the constriction model, the palatal
stop is a dorsa' palatal, and the fronted velar is a [-back] dorsal
velar (for a number of arguments to thai effect see chapter 6).
Finally, the support for the constriction model comes from tho
fact that it rules out a number of phonological processes which
never occur. Consider this: the articulator model has no intrinsic
constraints on how many out of the three articulators can be hooked
onto the Place Node at any given point (aside, maybe, from the
language-spocific constraints on complex segments). Since it
specifically predicts the spread of an articulator past any sound
that is not specified for that articulator, it allows and predicts, for
example, a process which spreads the Coronal Articulator across a
vowel (the Dorsal Articulator). onto a Place Nods that dominates a
bilabial, followed by the delinking of the labial articulator.
Basically, it allows an Iml to Inl assimilation across a vowel.
While the processes of ImJ to Inl assiinilation do exist (and are
commonly known under the lable of ·place assimilations·), they are
almo!t invariably local. Short of stipulation, there seems to be no
good way of ruling out a long distance, feature changing spread of
coronality, labialily, etc., on the Sagey's model. and. by the same
token, no way of 9xpiaining why such processos must be local.
On the constriction model. such processes are ruled cut by the
fact that a well-formed constriction is a function of two variables.
the Site and the Articulator. While unspecified representations. with
the value for one (or both) variables not entered are allowed (in
agreement wih most thoories of underspecification). a constriction
with more than one value for either the Site or the Articulator is
ill-formed, and is ruled out under a general assumption th~lt
ungrammatical representations are universally blocked (underivabla)
at all intermediate levels (for the discussion of this assumt:'tion see
chapter 7).
One consequence of this is that all spreading of Articulators or
Sites to a fully specified constriction node is imposs!bI9. both long
25
distance and locally. Spreading of either tho Site or the Articulator
onto a constriction will be allowed just in case the representation
is not complete for the value of the Site or the Articulator (or
both)-. In those cases, the focus and the target need not be adjacent
on the skeletal tier, unless the process is subject to proaodic
cOilstraints. Finaily, processes in which all articulatory features are
assimilat9d, are treated as the spread of the Constriction Node.
Nothing prevents the spread of this node -- after all, this is the
primary source of deriving comple~ segments.
1.4
The Structui9 of the Constriction Model
In this section, I discuss the format for ttae representation of
phonological segments within the constriction model. I begin by
reviewing the content of the nodes and the features in the segmental
tree.
1.4.1
An Overview of Nodes and Features
Following Clements (1986) and Sagey (1986), I assume that aU
segments are represanted with the Root Node. which dominates aU
other features in the segment. The Root Node is necessary in
explaining assimilations which spread the totality of features. as
observed, for example, in the compensatory lengthening processes. It
directly dominates the degree of closure-. air stream. and phonation
features. Following Piggott (1987). and McCalihy (1988), I assume
• This could b. the esse with thl so celled consonente' helrmanl.s obt,rved In
child speeCh. ADPertfntly. there has betJ" e dlgs.rtalton out or thl University of
illinois. by C~thy Echols. which explores thl Idea that at early stsg.s. the ,,:ord,
ere not funy repr,!St:'nted In th8 sDlech of children. lheat on~y the most sallont
saunas (8 g. In stressed syllables. at thl beginning of lhe \fiord. etc.> er. funy
represented. and tha~ conso.,enlal sortedt"; might bit the m.ons of filling out
tncomple\e reprS!8ntetlons (I em grateful to MlchDel Kenslow tcz for brl"gtng this
to my et tentlon.>
• This Is Pi'ettty much e hypothesil. An oJternetive view would be to ossoelet& the
stricture f,elur8' with the constriction node. Willie ~om8 complex segments haYI
IClenUcaJ stricture features on ell constrictions. there DrB segments which hov.
conso~8ntol eonstrlc ttons comblnod w Hh vOCDllc ones, as. for exompll. Is thl C!S.
W tth poiotallz9d ~8gm9nlS. For the time botng. I wsnt to troat the QUlsllon 0' th.
10ceUon of thl stricture feelurel as unresolved; pIecing them under the Root Node
Is Just tI tDmDorery solution.
26
that the nasality iestures too are dominatsd directly by the Root
Node. Finally, the Root Node dominates the Constriction Ncde(s). the
seat of all articulatory features.
Pharyngeal is the Site involved in the production of all gutturals:
uvulars, pharyngeals and laryngesls. I al~o assume, that the low
vowel a !nvolves a pharyngeal constriction-.
The Constriction Node dominates the Site and the Articulator
Nodes. I assume that this is the canonical form for the Constriction
Node, and that at no point in the derivation are there more than one
Articulator or more than one Site under this node. Derivations which
violate this condition are universally blockQd.·
Before leaving the topic of the Sites, let us recall that in
·Nood's ~1979. 1982) studies of vowel articulation. which have
inspired the treatment of vowpls proposed in this thesis. four rather
than three basic constriction locations for vowels are proposed: the
palatal. palato·~8Iar, uppsr·pharyngeal and lower-pharyngeal. The
difference between Wood's proposal and mine boils down to thtt
presence Y5. absence of subdivisions within the pharyngeal region.
Again, my treatment draws on the results of McCarthy (1989), which
reveal that segments with the pharyngeal constriction pattern
together in various phonDlogical processes, and therefore form a
natural class. Since, as McCarthy points out, as many as three
different articulators may ba involved in the production of
gutturals, the point of articulation (pharyngeal) is the only
candidate for a feature which wauld capture this property of the
gutturals. If so, the source of distinctions among the pharyngeal
sounds must not be derived from a different constriction location
but from other factors. Later on (in chapter 5) I will argue that
different pharyngeal sounds may aptJrJar to have constrictions in
different areas of the pharyn)( as a consequence of the fact that they
are execuled with different articulators.
The concept of the Site corresponds to an earlier concept of
the ·passive articulator·. Sites correspond to the regions in the
vocal tract with which the articulators come in contact during
speech production. They are ·fixed" structures that cannot be
otherwise modified, therefore, Sites do not dominate any other
features.
The Labial Site maps onto the upper lip: the constriction region
in bilabials and linguo-Iabials (the latter: sounds observed in several
Austronesian languages (Maddieson (1988)) which are articulated
with the tip or the blade of the tongue).
The Anterior S:19
corresponds to the region which extends from the upper incisors to
the alveolar ri~ge (inclusive): labiodantals as well as all denials and
alveolars invoive this Site. The Palatal Silo maps onto the hard
palate -- the constriction location of the coronal and dorsa!
palatals, as wsll as front vowels. The Velar Site maps onto the soft
palate which is involved in the production of velar consonants and
high back vuwels. The Pharyngeal Site maps onto the back wall of
the pharynx: following the revealing study of guttural phonology by
McCarthy (1989) (discussed in detail in 3.5). I assume that
• While Idlos of this sort Df' highly spoculativi. ~t llcst at thl Dresent stage ~f
our know ledge of genlUteUy conditioned behetyior, It Is '.at unreesoneble to
alsumo thet clrtlSln g.,turl' thst humans perform er. -prewlred", lhet Is.
genltlcslly encoded. ThlSI Idlas hove been 8)(o!oreCSln thl context of 7eseerch on
othar orally performed gesturl9. such Dt chew lng, suckl&lQ. crying, etc. My wey of
lhhlJ:lng aboul Un) rtpresantatlon of the constriction glsture hes been InSDlred by
such Ideal, end DsrtlculGrly by lhl suggeltlon mad. by P.r~.n (1979). w ~Ich
extends thl concept of "preexisting motor clrcuUry- CP. 368) to $Plech gostures
27
ArticulatC'rs map onto tho organs capable of performing
vertical or horizontal movements. With the excaption of the Tongue
Root, which I adopt after Ladefoged and Maddieson (1985), and the
Glottis. they are exactly the articulators of the Sagey (1986) model,
with the names modified so as to avoid confusion with the Siles (the
Site names are adjectival, and the Articulator names are nominal).
The following is the inventory of the Articulators: Lower Lip
(Sagay's labial Articulator), active in the production of bilabials and
haYI pointed out th~ Sitos for the ·exlrftme- vowels L.I.I. and A. ond left
of tho discussion the mid. as weU 6S the -Intlrlor- yowtls. laler on, I will
suggest" treatment of vowels wtllch Dr.tty much dar1v,s thl romolnlng QUsUtl81
from thea. three tleslc onl!.
• So fer. I
CU~
28
labiodentals; Tongue Blade (Sagey's Coronal Articulator). the
articulatar active in dentals. alveolars. coronal palatals. as well as
the linguo-Iabials: the Tongue Body (Sagay·s Dorsal Articulator). the
articulator active in dorsal palatals. v8Iar~. uvuJars. and all non-low
vowels; the Tongue Root -- active in the production of the low vowel
II. and the so called -pharyngeai:i-: the Glottis, the articulator of the
glottal stop and the laryngeal fricative h.
Articulators capable of assuming distinct. phonologicaUy
relevant shapes. are associated with the features which represent
these sh~s. For example. the lip do~inat9s the feature [round]
which refers to the shape of the lips.
The Tongue Blade dominates the [distributed]. the feature
which distinguishes laminCi. «(+distributed]) and apical ([d:stributed» sounds. I assume that [-distributed] is sufficient for
characterizing an apical sounds. whether with a slight raising of
the tongue tip. ?r extrsme fetroflel:ion: to my knowledge. no
Janguage
differentiales between the two varieties for any given
Site.
Following th6 arguments in Levin (1987). I assume that the
Blade (the coronal articulator) dominates the feature (lateral). in
addition to {distributed].
The Tongue Body dominates the features [high] and [back].
Traditionally, the feature (low) has also been included among the
dorsal features. However. since the constrictl.Jn of the IOYJ vowel a.
is produced with tho Tongue Root. th9 feature [low] is no lor:ger
needed for the purpose of distinguishing this vowel from the mid
vowels·. The status of this feature in the constriction model is
discussed in 5.4. along with other tongue body features.
Finally. the Root and the Glottis are not modified by additional
features.
1.4.2
The Structure of the Representation: Simple
Segments
CompJex
In the prasent modeJ, simple seGments are represented with
one Constriction Node. complex segments with two or more. I
a~sume that there are no strllctural constraints on the number of
Constriction Nodes in a complex segment. and that the existing
limitations are most likely acoustically or perc8ptuall~ conditloned.
For example. Ladefoged dnd Maddieson (1988), show that the number
of primary constrictions in a surface complex stop must be no
greater than two. as the acoustic cues lor an additional constriction
would necessarily be lost between the acoustic cues for the onset of
one constriction and the release of the second.
The only structural constraint on complex segments is that no
Articulator or Site may be involved in more than one constriction in
any given segment This constraint is somewhat reminiscent of
SagGy's (1986) prohibition against complex ~egments in which one
articulator executes two cOlistrictions. However. J. Harris (p.c.) has
poi!lted out to me that deriving such ~ constraint from the shear
impossibility of one object being at two locations (whjr' appears to
be the reasoning behind Sagey:s proposal) is not necessarily
justified. For example. given the size and the strategic location of
the dorsum (Tongue Body). it is not oevious why it could not
participate in two constrictions (say. palatal and velar)
simulataneousiy. Or. given the fact that both the lower lip and the
coronal articu!ator. can form a constriction in the region of upper
incisors (Anterior Site). there should be no technical difficulty for
the A(1terior Site to be involved in two constrictions. !abial and
coronal. simultaneously.
• later ~n. 1 will consider 0 posSibility thel leng~o9'1 mey hOyt e non-l'!lgh. nonround. boclc vowel mesas wtth lh& Tongu. Body constrIction. I wtll melee 8 point of
:;hOw In; thst such IS vcwltl e~t! bolt PhcnologlCtHy end Dhonetlcelly ~s i-hlghl.
29
YS.
30
Given the fact ihat no language contrasts sounds solely in
terms of such possibilities·. it appears necessary to rule out the·
phonological combinations of features not on physical. but on some
other grounds. I would like to suggest. tentatively, the following
constraint:
(3)
• Root
A
ConItr
Canetr
I
Ir
.r
ft
The constraint in (3) is pretty much a descriptive statement.
Hopefully, with a better understanding of phonological features it
can be replaced by an explanation.
2.
Articulators
Given the results of feature rspr9sentation research c:>nducted
within the articulator model (Steriade (1985), Sagey (1986), Yip
(1988), McCarthy (1988), elc.). it may appear somewhat superfluous
to argue for the phonological constituency of articulators. To some
extent this is true, therefore I shall devote a limited amount of time
to this issue, mostly summarizing the arguments that have already
besn proposed. However. the status of articulators is somewhat
different in the present framework, since they are not the sale
source of articulatory features (cf. tha Site features already present
in the system). In order to establish an independent need for the
articulator features, it is necessary to prove that there exist
phenomena which can be dealt with only in terms of these features.
Therefore, in what follows, I shall concentrate on such
phenomena: I shall consider processes which affect either the entire
natural classes formed by the sounds sharing the same articulator.
or the subsets of these classes defined only in non-articulatory
terms. For example, if a language treats in some special way all
coronal continuants. but not the stops, we still consider this as
£tvidence for the constituency of the Tongue Biade articulator. On the
other hand. processes which involve. say. a palatal subset of the
coronal sounds. cannot be considered to support the constituency of
the Tongue Blade articulator.
• B,'ore this can b. assertod. I wOtlld Ilk. to edress the claim that ha! beln mDO.
on .Iveral occasions about the peJoloJ sound. In Russian; sccordlng to this cletm.
thtr. II a polete!/Dalotellzld dllU-.ctlon In Rusllen. This distinction Is SUDDOSld
to account for tM cantrall betw.,n the so celled -soft- Dolotel Dnd (hi ~hord~
poiat al. cf Ounatcv ( 196Jt Schmal.tteo (1964). dl ArmonCS ( ! 966). It cHow !V8r •
It Is not nle.lsort; to SDI81 to secondary polate) ertlcuJetion. in order to 8J(Dre"
the contro!t blll., ••n tM two tYPG. of oelelols; the -soU- palatel Is .xecuted ..,., Ith
thl tongue blsae. whll, tho -herd- polatol Involves: thl tongue lt~; the contresl Is
thin balw ••" the • encl- valu.:. of the r.eture (dl!trlbutldl. In foct. this Is not
lhl only ce-ntrest. es the apical frlcatlv. Is olso y.lerlz.d. el ShOwn on the X-rDYs.
ef. 8rVzgunovB (1963).
3i
The phonological phenomena that will be used as evidence are
the coocurrence restrictions on segments \~hich have identical
features (best treated in terms of the Obiigatory Contour Principle.
which prohibits adjacency of identical segments (L3ben (1973),
McCarthy (1979) McCarthy (1981). McCarthy (1986). Yip (1987),
(1988;. etc.). and phonological processes which affect. or are
triggered by. the natural classes whose existence we are ,iying to
11
esta~lish.
2. 1. Lower Lip
32
The natural cia.. characterized by this feature comprises
segments Ip. +~ I, m. b, u, 'X, 0, iii, etc., and other dimensions of
the.e articulations.
Many languages lack labio-dentals, thereforo, tile class of
sounds articulated with the Lower Up in such languages matches
exactly the class of sounds with the labial constriction site. Even if
there exist phenomena in such languages which single out the labial
sounds (el. the ~occurr8nce restrictions on labials and round
vowels in Taiwanese, lin !1989), labialization of the epenthetic
vowel in the environment of Iabials in Chukchee (Sagey (1986), etc.),
they cannot be analyzed unambiguously in terms of a Site or an
Articulator feature. Therefor., they do not provide unambiguous
support for the Anculalor feature Lowar Up, or for the Site feature
Labial in a system that contains both.
2. 1. 1
In Sudanese Colloquial Arabic· (Hamid (1984), after
Kenstowicz (1989», the following praeesses take place: Voicing
Assimilation and Articulatory Assimilation. The first rule
assimilates the voicing of word-final obstruents to the voicing of
the obstruent of the following word. The second rule assimilates
word-final obstruents to all articulatory features of a following
obstruent, if both s6gments share the same articulator. In addition,
stops assimilate in continuancy to fricatives (but not vice versa, M.
Kenstowicz p.c.). I begin by presenting the entire paradigm now, and
later on, I repeat the portions of the data relevant to each
articulator:
(4)(a)
I have been unable to find an example of a 'language which
would contrast the labial and labiodental sounds. and, in addition,
would display phenomena in which the two types of sounds would act
like a natural class. However. I know of at least one example of a
language in which the interaction between !abial sounds can be
analyZed only in terms of the Lower lip articulator. This ianguage is
Sudanese Arabic.
k1 teab 'book'
kt t8e(t I Fetht
klt8elp) Seomye
ktt8elp) Xelltd
Even those languages which do have labiodentals do not always
provide an ideal ground for testing either of these features, as the
presence of labiodentals in the surface inventory of a language does
not necessarily guarantee that they are present in the underlying
inventory. Only 13nguagBs which display surface bilabial and
labiodental fricativas cannot be doubted to have underlying
labi()("'A'ntals. In languages which do not display such a contrast, it
may often be the case that labiodentals serve as the surface
manifestation of bilabials: bilab.al fricatives have much less noise
than labiodentals; therefore, for perceptual reasons. labiodentals
might be preferred over bilabials in surface representation. This
iSSU8 is discussed in more detail in 3.2.
Sudanese Arabic
(b)
bit
'girr
bUtl fert Id
bUtl hesen
?elbt(z) z8ekefel
?e1bt(s) S88fet
?81blldJ ~trf8t
bUs) se8mytt
bH5J 3t1iaJ
behtd
btl!) !em1l1e
btldJrarlibe
'the girl studied'
'the girl saw'
'the girl knew'
'beautiful girl'
'strange girl'
'country'
bele(t) f8rttd
bnle(d) Gcesim
bole(s) semllr
?elbe18Id) geddem 'the country offered'
?elbeles seerek
'the country shared'
?elbele(z) zere~
'the country sowed'
baleij) ! nlce1
(c)
some(k) fetht
YS.
semelxl xoeltd
• I am grateful to Michel) Kenslowtc% for m~lctng thiS' dBt~ svctJoble to mi.
33
34
sema(k) samllr
seme(k) tartlf
seww~8g
same(¥) 1(Dr1tb
'driver'
sewwe(k) rerltd
YS.
s8wwelkl hasen
sewweehcl xeltd
sewwBe(g) Ssyytd 'good driver'
sew w esl¥' J ¥erltb
'strange driver'
SDWWG81k) keslaen 'lazy driver'
Other forms of evidence for the coronal articulator come from
the phonological phenomena in which the sounds articulated with the
tongue blade/tip form a natural class. I now turn to thrJ discussion
of such phenomena.
2.2. 1
Sudanese Arabic
Since the Sudanese Arabic has both anterior (t.. d.. S., Z, t, cJ~ ~,
and palatal (I, !.) coronals, the data to be pr&sented provide a
particularly convincing argument for '~h8 constituency of the Tongue
Blade articulator. All coronals pattern alike in the Articulatory
Assimilation:
~)
The data relevant for the diSCUSSIon of labial5, shown in (48)
demonstrate that while IbJ assimilates in voicing to any obstruent,
it assimilates in the remaining features only to If/..
The paradigm in (4) may be interpreted as follows: there is an
OCP triggered deletion 01 the Constriction Node in the first of the
consonants, followed by the spread of the Constriction Node of the
consonant on the right
2.2
(5)
?elbllz) zeekaret
?elbl(§) seefet
?elbt{dJ ~trfet
btl!)
~emltl8
bUd) ~ertlba
let us now tum to phonological phenomena which support the
constituency of the Tongue Blade (coronal) Articulator. This class of
sounds includes, among others, the follOWing: l .. t. d, ~, S, $ .. z.. ~, e,
~, n.. 1, r. $" ~ .. C, ¢, Z. Z, 31 etc., as well as the complex segments
involving these articulations (e.g.,
t J" t~. etc.). I am assuming here
that no vowels involve the Tongue Blade articulator. I discuss this
issue in more detail in 5.4.3.
beJad
bale(tJ
bele(d)
bele(s)
bala(a)
c.....
35
'girl'
bllt) ferttd
bHt)hesen
btls) stUlmye
blij) 3elel
Tongue Blade
One form of evidence for the existence of the articulator node
Tongue Blade is the existence of features that are universally
dependent on this articulator. Sieriade (1985), argues for the
constituency of the corona! articulator (as oposed to the feature
(coronal]) on the basis of the need for hierarchy within that node: she
points out that fealures such as (lateral] and [distributed], which
may only be used distinctively to characterize the coronal sounds.
and which are active phonologically only within the coronal class,
are best treated as the dependants. not the sisters of the coronal
node.
bl t
·the girl studied'
'the girt saw'
'the girl knew'
'beautiful girr
'strange girl'
'country'
fertld
Geaslm
semI tr
aalaal
?elbele(d) geddem 'the country cffered'
?albales seerek
'the country sharecf
?elbal elz) zare
'the country sowe<f
In (5). the anterior consonant assimilates in point of articulation
and continuancy to / S/ /3/, /
/zl - r.9 coronal consonants. but
not to If / I'll or /f:('- the labial, velar and pharyng9al consonant
respectively.
s/,
II
II
2.2.2
Classical Arabic
• Thl Delete' ! hes bien argued to be Gn underlying 9 In Areblc dialects. cr.
McCcrthy (1989). Howlvlr. at the level of repr.sentatlon at which Articulatory
Alslrr.lletlol'l applies In the SudanlSI dlellct.! cPPlars to pattern wllh the
coronClls.
36
Kenslowicz and Kisseberth (1979) report the following
phenomenon as evidence for the SPE 'eature [coronal): -in Classical
Arabic. the nJ of the definite article I? a II completely assimilates
to a following dental or palatal consonant. but remains unchanged
before labials. velars, uvulars. pharyngeals and laryngeals· (p. 249).
They present the following data:
(6)
?sl beeb
·the door"
?al faras
?el kelb
?al )(fultam
?el Qelb
"lhe horse"
"the dog"
781 barb
?el ?eb
·the ring"
"the heart·
·the war"
"the f ether"
Ys.
?It tex
?ed dear
?es sanduuq
?ez zeyt
?Dr reiul
?en nees
rai sems
"the bed"
"the hous."
·the bo)f
·the olr
"the man"
"the people"
"the sun"
Given that phonological processes affect natural dasses of sounds.
the phenomenon illustrated in (6) constitutes evidence for the
feature Tongue Blade Articulatar. since it takes place only when IV
is a:ljacenl to a coronal segment.
2.3
segments (following the line o~ argument for the constituency of the
coronal articulator proposed by Steriade (1988)).
I t would mDke very little sense to chllrectertz8 sounds such as
Ip/, It I , or le'- in terms of features [high) or (back]. On the other
hand. there is massive evidence for the existence of such features in
the phonology of sounds in whose production the tongue body is
(distinctively) active. Just to give a few examples -- Calabrose
(1985) analyzes the process of metaphony in SaJentina. a southorn
Italian dialect. which raises a stressed vowel followed by a high
vowel in an adjacent syllable. This harmony is blocked by a segment
which Calabrese tra,lscribes as /kJ/ and calls a ·palatal occlusive(p. 17). Below are a handful of examples in which the harmony has
applied, and a few cases in which the intervening -k!.- ir;hibits the
process:
t
(7)(a)
-->
9
i I _ _ + high
+ slress
sing:
plur: -i (word-final mid vowels ant
-9
raised)
TO!lQue Body
I now tum to tl.e arguments for the feature Tongue Body
articulator.. Other than in Yocoids. this articulator ;5 active in the
production of~. J• .c, j. (voiceless and voiced dorsal-palatal
fricatives and stops respectively). k, g, )(# 't I') (velars) and Q. G, X,
ts, N, R (uvulars)- . As for the vowels and glides. I assume that
Tong ll8 Body is an active articuiator in all of them. except for the
-low· vowel JaJ. which I assume to be articulated with the Tongue
Root (S99 section 5.4.4).
I
One way to argue for the phonological constituency of the
Tongue Body. is to appeal 10 ti1e fact there exist features which are
universally associated with what we consider as the dorsal
• Fc..Howlng McCerthy (19B9). , aSlume thet uvulert art campl,x segments
InvolYlng bOth Viler end pharyngeal constrIction 1ocetlonl. The only difference
bltw ••n McCarthy's treelmlnl of uvulars ond mini ronows from different
assumptions about the nature of Phono1ogleel repr.sentatlons.
37
paritt
reell
mast
(b)
0
-->
"weiHs)"
"nQt(sr
"monlh(s)"
u , __ + high
+
(c)
par(tl
r(tt
m(sl
stress
croci
nocl
cruct
nuel
stasonl
slesunt
the blocking effect:
vskkJu
'old'
supirkJu
·oulr Og9'
• See lhe discussion of I~/ In 3.3
38
'cross(es)"
·nul(s)"
·seeson(s)"
The onty way to understand the blocking effect in (7a). is to assume
that the segment -kl-, intervening between the trigger (the high
vowel) and the target (the stressed vowel) shares with the trigger
the spreading feature (high]. As to the nature of the blocking
segment - based on Calabrese's description which characterizes it
as a complex segment composed of a velar and a front glide -- one
can hypothesize that it is. most likely a front v81ar redundantly
specified as [+high] on the dorsal node. Coming to the point of this
presentation. the Salentina facts show that the feature l+high)
associates ~ith a non-low vowel and a vetar consonant -- sagments
articulated with the Tongue Body. It does not associate with any
segment in the language which is not articulated with Tongue Body.
The only way of capturing this fact in a simp~8 way is to have a
theory with a phonological constituent Tongue Body. on which the
feature (high] depends..
f
A different 9xsmple in which it is convenient to assum9 that
velars. which are inherently dorsal consonants, may be redundantly
specified as [+high]. is found in Cantone58, where a velar consonant
is not g,1'ow&rd in the rhyme of a syllable whose ~ucleus is a high
vowel.
!n
Jar v~in. something as common as the assimilation of
a velar to the lv'-.~u8 body position of an adjacent vowel (discussed
in detail in chapter 4) provides evidence for an intimate relationship
between the height and the backness features on thtt one hand. and
the Tongue Body Articulator on the other: after all. non-dorsal
segments are not typically affected by the frontness or backness of
an adjacent vowel; mutual influences in terms of these features are
possible only among the dorsal segments.
L
Lat us
consonants
dorsals are
the Tongue
'
Site less than easy on the consonantal material. Phenomena which
single out vowels and velar consonants are probably even less
common. in fact I don't know of any. Here, I continue the discussion
with the presentation of Articulatory Assimilation of the wordfinal dorsals in Sudanese Arabic, on the assumption that this
process is across the board articulator-sensitive.
2..3.. 1
SudMese Arabic
The only dorsal consonants in Sudanese Arabic are velars. The
Articulatory Assimilation affects thttm as follows:
(8)
samek
seme(k) f eth1
semD(k) samtlr
sema(kJ sarti f
seWW8tJQ
sewwel~J
Ys.
·drtver·
fer1td
YS.
·good drlyer'
'stronge driyer'
sewwce(k) !<esloen "lezy dr1ver"
sowwee(gJ 30yytd
sewwtJ(~) ~er11b
sDwwa(k) hasan
sewwee(KJ xa1id
I believe that interpreting the above data together with the rest of
the paradigm presanted in (4) (namely as a reflex of the Articulatory
Assimilation) is well justified. The generality of the process
suggests that this is the correct approach (otherwise the ianguage
would have to have two, very similar, but structurally unrelated
processes). Also. the absence of this phenomenon UndE'f the
constriction location (Site) identity (pharyngeals and laryngeals are
not affected by Articulatory Assimilation) points to the same
conclusion.
now tum to phonological phenomena which group dorsc:1
into natural classes. Among consonants. non-velar
fairty uncommon, which makes the task of arguing for
Body as a phonofogical entity distinct from the Velar
39
semehcl xeaHd
sama(r) ¥ertlb
40
3. Sites
In this chapter I consider evidence for the phonological
constituency of the Site features: Labial, Anterior. Palatal, Velar.
Pharyngeal. I follow the format of the argument adopted in the
preceding chapter, and argue lor the Site 'eabJres on the basis of the
phenomena which can be explained only in terms of these features.
3.1
Labial Site
The sounds which are characterized by this feature include the
following: , .... fR (lingua-Iabials), P. b. m,
IS. k", t W• pW. u. O. u. 0,
etc. At present. I do not know of any phonological phenomena in the
languages which have lingua-Iabials in their inventories. which
would group bitabials and linguo-Iabials into a natural class~ .
Therefore. the only possible way of arguing for the feature Labial
Site, is on the basis of !he phenomena which single out bilabial
sounds in languages which have both bilabials and labiodentals in
their inventories: Since in the constriction medel labiodentals are
represented with the Lower Lip constriction at the Anterior Site.
they contrast min!mally with the bilabials. which are represented
with the Lower Up constriction at the Labial Site. I argue that the
process of labial dissimilaiion in Venda. Zulu and other Bantu
languages (Pedi. Tswana, etc.) is an example of a process which
distinguishes among segments on the basis of their Site
specification.
+.
3.1.1
Velarization in Venda
Venda (segment inventory according to Oake (1954): I. L. U, O.
m,., ~, r, Y, pf. pfh. b\\ -by. t', t h, d. "d. n. 1.
9, E, o,~, 0; P'. ph, b, Ab.
\'. \h,
i,~.
q,
"~# lS'.. t$h .. d~# "d~, f).
e', c
h
•
d~. "di,
J.
r. s.. z, 1. s·,
'lv,
ts.... ts hW •
w
v
dZ • "dz .. P.
k', k3't, Q. "9. t). w. h) has a dissimilatory process
• Th". Is II practlcD! explanation for this: thl IXlstence or lInguo-lebtals hes be.n
conflrm.d only rlcently (ct. Maetdl.son (1987»; naturolly, languagls wh!ch or. said
to hlV' tha•• sounds. er, poorlly dDscrlb.d In thl linguistic Ilteratur.: WI simply
do not know whet the.r phonology Is
41
affecting the stem-final bilabials (+. P. m. ph. P. b) before the
passive suffix l-waJ. which results in a velarlz8tion of either the
first of the second segment. The following alternetions are
observed: bilabial fricatives turn into valars: +w ._> )(W; ~w ->
}{ w: when bilabial stops are combined with IW/. the Iwl turns into
a velar fricative: ph W --> ph X ; pw --> P~ . bw ._> b¥'; nasal ImJ
may pattern with either fricatives or stops: mw --,. I)w (ml);
(Doke (1954». The interaction of l-waJ with stem-final conson,ants
is illustrated below:
(9)(a)
1'0+
•
wa
--)
diP
+
we
--)
nap
+
we
--)
khoph
(b)
•
we
poxwa/po+lwe
dtl(we/dlplwe
napxtJ/neptwe
--) khoph)(D/khopht we
beb +
tum +
we - - ) beb¥'tJ/beblwa
we --> IUf')we/luml)l!
fun
rend
Wit
+
+
--)
funwa/funtw8
we --) rendw e/re"ui W D
"be tied·
"be known"
"be
'be
"be
·be
switched"
broken off cob'
begot ten'
bitten'
·be loved'
'be pret sed·
According to Doke (1954). p. 157, only bilabials are affected.
This is important. in view of the fact that there are labiodental
fricatives in Venda.
I would like to suggest the following analysis of the
velarization facts: Vanda has a constraint on sequences of segments
that have a labial point of articulation. This might be interpreted as
an OCP effect ap~lying at the labial Sita tier. When a bilabial stop
becomes adjacent to a round glide. the labial component of the glide
is delinked. The resulting segment is a velar glide: since such a
segment does not exist in the underlying inventory of Venda (see
above). it is reanalyzed as a velar fricative, which assimilates in
voicing to the preceding segment. It appears that in the case of
• Velarlzatlon ello occurs w"ln thl glottal -rrlceUve· his fOIJOW9C1 by 0 round
gild•• I.g, Buleh • WD --) pulerwe "to bl kill.,,". I essum. that b Is rtfilly en
BPproxlmont. Just like thl glldO. and what epPlers at velerlzetlon, reeny servI.
Uti parpose of Slr.ngth,n'n9 tho first segmenl. VI Uh thl volar compon.nt bltng e
result of IUh,r
8
de'Dull rule or spreading from thl gild'.
42
bilabial fricatives followed by a round glide. the labial component is
deleted in the fricative, and the glide remains intad. Ther. are two
way. of interpreting the velar component in the frir~tiv8: either it
is inserted by a default rule, or it spreads from the following glide.
Agait!, it is outside the scope of the present discussion to decide
what the actual mechanism might be.
3.1.2.1
In Zulu (segment inventorf (Doke (1926)): consonants: ph ~ b,
f, Y, m. 6, p1, t h , d, ts", dz, s, Z, n, 1, r, ~, 4,
di, At fl. k, kh , g, H, 1'),
h, fi, 1, lh, 19 , 1, ~, ~h, ~', ~, (, (h,
C; yowels: t, u, e, :I, plus the
same, seriss with length), bilabials are palatalized in a number of
morphological environments. The following examples illustrata the
behavior of stem-final bilabials before the passive suffix /-waJ:
The difference in the behavior of fricative. and stops before a
round glide can be interpreted as due to a constraint against
sequences of obstruent fricatives in Vend.; if the derivation
proceeded in the same fashion for fricative. as it does for the stops.
the output 01 the process would have been a sequence of 1+1 (/IS/)
and IxI (/11). Apparently, such sequences are not found tn V"nds.
(10)(a)
The above analysis crucially depends on the existence of an
articulatory feature by which the bilabial and labiodental fricatives
may differ. I am suggesting that such a feature is the Labial Site
which corresponds to the labial constriction location. Without this
featufe. phonological interpretation of the velarization facts in
Venda is not available.
3. 1.2
e',
-tU:ph
(d)
we
-4u:swe-
--)
-6d:ph
+
we
--)
-6d':iwa
-bu:b
+
WI)
--)
(kwe:}(?)
(b)
(e)
+
c, c",
-be te8sed'be tied'
-bu:5W8
derivattve of
·dle'
-be trapped-
-be:b
+
we -->
-be:!w8
-4d:6
-48:6
+
+
we
we
--)
-4d':c'wa
-ia:c'w8
-t h m
+
wa
--)
-t h U:PW5
'be senr
-hJ:m
+ we
--)
-lu:J1wa
'be bitten·
u:
--)
'be written·
'be s tebbed·
Exampls:& belCiw show that no other segments are palatalized before
/-wa/:
Palatalization of Bilabials in Zulu
(11)
Palatalization ot bilabials is a wide spread phenomenon in
Southern Bantu languages. The fact that it occurs most commonly
when a bilabial consonant becomes adjacent to a round glide (either
as a result of affixation or a phonological process). suggests an
analysis in terms of dissimilation. Of interest here is the fact that
labiodentals are never aiieeled by the process. This suggests that
the dissimilatory effect is not triggered by the same articulator,
but instead. by the feature which is based on the same constriction
location -. the feature Labial Site.
I discuss the palatalization of bilabials on the material from
(Doke (1926. 1927). Meinhof (1932}). and, whenever necessary,
contrast the facts of Zulu with those of Tswana (Stahlke (1976),
Herbert (1977) and Venda (Ziervogel (19B1), Meinhof (1932)).
Zul~
-60n + we --) -lionw8
-thend + we --) -thnndwe
-thendts + we --> -thendlswtJ
-lhendel + W8 --> -t h endelw8
-60') + we --) 6of)w tJ
"be loved·
'mode to love·
'be I Dyed for·
·be prelsed'
Oaks (1926). p. 139.. 140 emphasizes that only bilabials are
palatalized in the context of j-waJ. While Doke himself gives no
examples of passivized verbs ending in a labiodental, the fact that
the labiodenta;s are indeed unaffected ,n the presence of the passive
• Nllthlr Dok. nor MelnhOr consider the postconsonanta' glide 01 loblollzDtton on
the consunant. I assume that conlonent-gt!~,onsets ere ellow.d by th. syl!ebll
structure of Zulu. l~l.r on. ergumlnts will btl provided to show that at I.ast ot thl
point when D!letelllDtion 8DP1l85, 1( Is D s'gmlnl SIDDrat, from thl .tlm-flnel
consonant. Only this much Is naldld for the analysis of the Zulu facti to bl
pr.sented.
43
·be ssen·
44
their suffixed counterparts are not. I would like to suggest the
fo~lowing explanation: the rule responsible for the delelelian of the
labial node of the consonant applies in derived environments only. In
It 9 tto' :60/, the sequence of bilabials exists in the underlying
representation. In les1tOf'e:nt/. I assume that since vowel sequences
are not allowed in Zulu. the final vowel of the verb stem is delinked
when followed by the vowel of the suffix. This creat911J an input to a
syllabification process which resyllabifles a round VOW8~ as an
oncet. together with the preceding consonanl The daia involving the
suffix 1-W8I show that such inputs are syllabified only if thay do
not violate tJ1e prohibition against adjacent bilabials. Otherwise, the
first bilabial dissimilates to a palatal-coronal.
Let us now consider the data in (11) and (14-16). These
examples, show that an analysis which attributes palatalization of
the labials to the presence of an underlying palatal segment in lwei, or the oyert vowel Iii in ths locative, is not possible.
However, an analysis which attributes the palatalization of
labials before the passive suffix I-w~ to an underlying palatal
segment has, in fact. been proposed by Khumalo (1987). On the basis
of the forms in which the passive suffix is preceded by Iii (e.g.•
when following moncsyllabic verb siems: uku: + phS + W G --)
uku:p h 1wa '08 given'), Khumalo analyzes I-wal as having an
underlying form l-iwaJ. and derives the forms in (10) via a process
that converts Iii to Ij/, which in turn palatalizes the labials.
Khumalo treats palatalization of labials in the locative as a result
of a dissimilatory process; first, the rounded vowsl following the
labial turns into a round glide; tlien Labial Dissimilation applies to
tum this glide :nto Ij/: this is followed by the palataUzation of the
labial by the palatal glide.
This analysis has one obvious advantage over the treatment I
am proposing. namely. it can account in simple lerms for the fact
that in the locative, but not in the passive, the labial component of
the trigger vowel completely disappears in surface representation.
lis disadvantage, however, is that it wrongly predicts the
47
contracted variant of the passive suffix to trigger the palatalization
of non-Iabials as well.
This point has been raised by Herbert (1977), in his discussion
of the analysis of Tswana, another Southern Bantu language. by
Stahlke (1976). Stahlke 100 capitalizes on thE' fact that the passive
suffix surfHces in some forms as l·iw9J; he suggest! the process of
segmental fusion of IiI and the bilabial in forms which do not
display IiI on surface. Herbert (1977) points out that in the cases in
which palatalization can be attributed to a palatal element (namely.
before the diminutive suffix I-anal), labials are not the only
segment! that are affected. It is difficult not to agree with
~erb8rt's point. which also applies straightforwardly to Zulu (s8e
the comment about palatalization in the diminutive below).
Conceivably, Khumalo's analysis of the locative could be
adopted without his treatment of the passive: particularly if his
suggestion that the phenomena observed in the locative are
morphologically circumscribed is followed. Note that this analysis
must also be expressed in terms of the OCP triggered by the
identical Sites, not Articulators.
As for the presence of Iii in some forms with the passive
suffix - one can hypothesize that it is an epenthetic vowel, used in
Tswana as an optional means of pre\"Jnting the creation of consonant
clusters, and in Zulu, as a means of moeting the metrical
requirements of the suffix·.
The last environment in which palatalization of labials takes
place involves the diminutive suffix I-anal (I-anal). In this case.
however. the quality of the vowel that follows a bilabial is not
relevant; also, bilabials are not the only palatalized segments:
optionaUy, this process affects the alveolar plosiv8S. These two
facts separate the phenomenon of palatalization in the diminutive
from the cases presented above. We shall return to this phenomenon
• Thct suffixes do hav8 SUch requirements. hl!S beln shown
McCDrlhy and Pr1nce (1 ga7).
48
conVincingly by
when we discuss palatalization processes triggered by palatal
vowe's.
The analY!'8s of (10) and (13) proposed above capture what
these data have in common, and explains the contrast between the
behavior of bilabials vs. other consonants before round Yowels. Most
importantly, it explains the absence of the dissimilatory effect with
the labiodentals: since in the model I am proposing the labiodentals
are produced with a constriction at the Anterior Site, they are not
sensitive to the conditions against adjacent bilabials.
Such analyses are not available within a framework which
dOf)s not recognize the phonological constituency of point of
articulation features, and, in this case, the feature Labial Site (or an
equivalent). It this analysis should turn out to be the only way of
treating the facts of Zulu in phonological terms, it would constitute
an argument for including the point of articulation feature Labial
Site among the phonological features. So far however, all I have
established is that this is a plausible interpretation of the Zulu
lacts: in order to promote it to an argument for the feature Labial
Site, it is necsssary to show that no other treatment of these facts
in phonological terms is available.
To this end, I will present arguments agains. an interpretation
of the Zulu facts in terms of a festure resolution in a disallowed
complex segment, as well as an interpretation which would
attribute the contrast in the behavior of bilabials and labiodcntals
to some arbitrary feature, e.g., [continuant), by comparing the facts
of Zulu with those of Tswana, in which 'a~lial fricatives are bilabial,
and subject to dissimilation.
Finally, I will show that ~he proposed treatment of Zulu is
compatible with the asst:mptions I make about the underlying
representation 01 labiodentals: in 3.2.3, I propose that if phonology
of a languago fails to treat labiodantals as anterior. and the
language lacks a surface contrast among the labials sounds, the
surface labiodentals in such a languago should be treated as
49
underlyingly
otherwise.
Labial,
unless
the
phonological
evidence
points
Zulu belongs among the majority of world languages which lack
the bilabial-labiodental contrast; given the assumption stated
above, Zulu labiodentals are the prime candidates for being analyzed
as underlyingly represenlsd with the labial SUe. In order to be able
to treat them as underlyingly Anterior, I have to find evidence that
the phonology of the language treats them as such.
I will present two arguments to this effect: the first argument
will be based on the strength of the comparative reconstruction
analysis of the rise of labiodentals in Souther Bantu. I will show
that labiodentals in the ancestor language of Zulu arose while the
Ur-Bantu bilabials were still in its inventory. and when the
dissimilation of bilabials has already been a part of the Southern
Bantu phonology. Given the assumption that in a language that also
contains bilabials, labiodentals are represented with the Lower Lip
constriction at the Anterior Site. I will be able to conclude that the
labiodentals were originally conceived as Anterior-Lower Lip. and
treated as such by the dissimilation process. The second argument
for the -anteriority" of labiodenta!~ will be batec on their
patterning with the anterior co!onals in the process of
palatalization in the diminutive.
3. 1.2.3
A conceivable way of interpreting the data in (10) and (13).
is to consider the palatalization eUect as a repair strategy (cf.
Calabrese (1988)). applied to a segment that has been created during
the derivation -- say, a round9d labial. Presumably. in such a case,
any feature by which g. and If! differ could be used to designate the
class of bilabials as seg"isnts that are incompatible with rounding.
For example. a grammar of a language could plausibly contain a
statement such as -(-continuant. labial. +round]. I argue against such
an analysis on the basis of the data which involve verbs combined
with applied or causative. in addition to the passive suffix. which
50
reveal that dissimilation of the first labial takes place before the
merger of segments occurs (if at aUr. This is shown in (18) below:
(18)
-6oo h
-bub
-Ium
EI + we--) -6osE':lwD
Is • we --) -bu!f:sW8
Isis + we --) -luJ1ts(:swe
•
+
+
·to
°be t ted foro
°be killed for·
be bt t ten herd·
Next, compare the forms in (18) with the ones below, which do not
involv8 the passive suffix:
(19)
-60ph
•
-bUb
+
-lum
+
81 -->
-60pll~:IG
Is --)
-bubf:lse
Isis --) -lumis(:s8
·tle ror·
·klll for·
·btte herd·
The data in (18) show that the palatalization effect persists in the
forms suffixed with /..wal even when /-waJ is no longer adjacent
to the verb stem. The comparison between th~ forms in (18) and (19)
reveals that this effect is in no way related to the front vowel of
either J-isJ, I-ell, or I-isis!.
t
I want to suggest an analysis of the data in (18) and (19)
whose chief component is the assumption that morphemes may
attach directly to a root, ev&n if other affixBs are alreBdy attached
to the same edge. [Notice that an analysis based on the idea that
dissimilation can affect segments -long distance-, skipping the
intervening morphemes. because they lack bilabials, is not possible:
examples such as /6onwel (not ·conwe) 'be seen' show that only the
affixal material is invisible to this process. Also, the dissimilation
affect does not operate long distance across the epenlhetic vowel
lif. cf. uku: + p"e • we --) Ukuphl:we , which proves the same.
• The rlalon for this uncertainty about the phonetic slotus of consonant-glide
s_qulnc•• Is thl follow Ing: on onl hend. n.lth.r Dokl nor M.lnhof lr.li th.,.
s'Qulnce, It: anything but c:lult.f.; on the othlr hand. tllir. II It con,Id.rablg
linguistic tradition 01 Dlsumlng that at J.ft'l on surface, Bantu languag•• heYI a
Ilmpl.~ CV syllabi. structur •. Since from the Dotnt of vii., of U,. analysll adopted
hlr. U dot. not matter what stelul thll. I.qu.nces heyo. I do not alt.mpt to
r••olvl this Issul.
• Tt5. order In which the morp".mes appear In (1 B) dOls not Imoly thl ordlf of
derivation; s •• thl discussion billow.
• , OWl thIs
Dolnt to Kon Hoi•.
5I
The idea that affixes might be prespecified in terms of targets
to which they are allowed to attach, has originally been proposed by
McCarthy and Prince (198?') in their work on reduplication. McCarthy
and Prince have observed that the so called -infixes· appear to
attach to a stem after ·skipping- a predictable amount of
phonological material (usually definable in prosodic terms). or els8,
attach to a target that represents a prosodic unit. An idea similar in
spirit. has been developed in Aronoff (1987), who allows affixation
processes called -head operations·. whereby affixes may attach 10
morphologically specified targetso In ge~8ral, such operations can
pick out morphemes which head the words for the purposes of
morphological derivationo
Applied to the data in (18). the concept of targeted derivations
allows us to understand why the passive suffix I..wa! triggers
dissimilation in a stem final labial despite the appearance of being
separated from it by other morphemes: suffixation of I·wal
precedes any other suffixation; l-isaJ, I-sial, atc., attach to the
head of the construction -- the root.
One ccnsequence of this analysis is that it rules out a
possibility of interpreting the palatalization of labials before l-waJ
as a result of feature rearrangement in a complex segment: since
there are good reasons for assuming that in a complex segment.
articulations are phonologically simultaneous. and not linearly
ordered (el. Sagey (1986»). we cannot allow morphological processes
capable of separating such articulations.
3.1.2.4
Let us now turn to the facts of Tswana (segment inventory
i,
fl, dZ, k,
(Cole (1955»: p, ph.. m, b, t, t, t h~ n, r, J, ~, ts, ts h,
1') .. l . h; t, u, e, E.. 0 .. J, e) for evidence that the palatalization process,
which Tswana shares with Zulu. is in no way related to the
[continuancy) contrast between bilabials and labiodentalso An
analysis that would appeal to this feature is not very plausible, and
I doubt if anyone would actually consider it; in this sense, bringing
in the data from Tswana may seem unnecessary. However, there is an
additional advantage in comparing Tswana and Zulu: their parent
s, e, e"..
52
languages developod labial fricatives after thay acquired the
palatall2alion process: but when the parent language of Zulu was
..developing labiodentals (se8 paragraph 3.1.2.4 below). the parent
language of Tswana acquired bilabials.
The analysis of the Zulu facts I am proposing predicts that
since the process 01 palatalization is triggered by the OCP effect
applying al the labial Site tier. it should affect the bilabial
fricatives 01 Tswana. This indeed is the case, as shown by the
following examples:
(20)(a)
-bop
-gap
+
•
-llheba
-rOb
-bOt
-Ie'
(b)
-bOn
•
•
+
·
-ret
•
-rot
+
we
WI!
•
--)
--)
we
-bofwe/-boplwa
'be moulded'
'be plunderad'
-llhe!w e/-tlhablw e 'be slabbed'
-g~cwe/-geptwtt
--)
\*le
--)
-rO!we/-rObtw8
'be broken'
we
we
--)
-bOswa/-bO+twD
'be bound'
"be paid'
we
we
we
--)
--)
--)
--)
-leswe/-Ie+twB
-bOnws/-bOnlwe
-ratwa/-retlwe
-rokw e/-roklw a
'be seen"
'be loved'
'be sewn"
Examples above show that just as in Zulu, only labials are
palatalized before the passive suffix l-waJ in Tswana (el. the forms
in (20b). The stems in the last two examples in (20a) contain a
bialbial
in the passivized forms of these verbs. 1+/ converts to
lsi. Thus the data from Tswana reveal that first, continuancy has
noihing 10 do wito the palatalization phenomenon: second,
palatalization. even though dating back to the times which preceded
the rise of bUabial fricatives in Tswana, has all the earmarks 01 a
phonological rule which affects all sounds that meet its structural
description.
'+';
3.1.2.5
Even though Zulu does not have a contrast bilabiailabiodental, a roasonable case can be made for the -anteriority - of
If! on the basis of the following: According to Meinhof (1932) p.27.
the majority of Southern Bantu languages have experienced a process
53
whereby tha Ur-Bantu obstruents have turned into labia~ fricatives
before the tense vowel lui (among the exceptions are" e.g., Pedi.
Tswana. etc.). These fricatives must have arison as labiodentals,
because the process which created them had been tak~ng place while
the Ur-Bantu fricative Jl was still a part c~ ~he inventory: it is
possible to assume that the phonetic change affocting I ~/ was a
later development for the simple feason ttiat it has taken dinersnt
routes in different languages, giving rise to either Iw/. (Swahili
Meinhof p. 32), Ibl (Kongo. Meinhof p. 1561. or /6/ (Duals, Zulu.
Konde. Meinhof p. 32). Also. in some languages. the newly created
labiodentals did not suppress the bilabial fricatives: for example in
Venda, both series of fricatives have been retained (Venda has 1+ I ,
11'1 and IfI. Ivl Doke (1954): soe the discussion of Venda above).
p
The identity of the newly arisen labiodentals and the ProtoBantu bilabials must have been kept apart until that change, because
there was never a merger of tho bilabials and the labiadentals
(except, of course. for the cases in which obstruents turned into
labiodental! before tense lui), and no labiodentals ever shared tho
fale of the bilabials. (Also. the fact that anteriority was an
important component of the ne\vly created fricatives, can be seen
from the fact that in some languages (e.g., Herero). IfI has
progressed all the way to lsi (Meinhof p. 27).)
Since the process of bilabial dissimilation in Southern Bantu
is older than the phonetic change which created labiodentals (it is
found in Tswana in the form virtually identical to that of Zulu. and
somewhat modified in Pedi), it follows that there must have been a
period in the history of Zulu, when lhis process had to distingLiish
between bilabials and labiodentals solely in terms of the presence
vs. absence of the Labial Sile. Since the mode of application of the
process has not changed (ct. the comparative evidence from Tswana
above). jUdging from the fact that it does not affect the
labiodentals. one can conclude that underlyingly. labiodentals are
not represented in Zulu with the Labial Site.
54
3. 1.2.8
The point that Zulu labiodentals are not represented
underlyingl, as Labial (. with Labinl Site) receives additional
support from the facl that
patterns together with the alveolar
fricatives of Zulu (I. z. t. ~) in that it fails to undergo palatalization
in the diminutive. The diminutive paradigm is illustrated below:
'f'
(21 )(a)
U:ph 8:p hS
+
ana --)
'tether·
u:phe§e:na
ene --) u:[u!e:ne
'meal-wetar'
+ ene --) 1')k'ec'ci:nB/n I)k'a68:neJ
'ox'
Ints'ti':mp'a + ana --) tnt=lupc'ti:ne
·W8rt·
u:[u:bu
Il)k'ti:61
(b)
(c)
(d)
+
u:tht + ana --) u:t"a:ne/lu:ie:na)
umgo:dl + enD --) lumg:JdG:na)/umgj3 G:ne
uku:~e
Imbu:zt
+
+
ens --) ukUlje:na
tina --) Imbuz8:ne
·sttck·
'm'ne'
·food'
'goat·
IS(:f~ + sne --) tstf8:nD/lstr~a:ne 'disease'
(:fu + ~nQ --> t:fs:na/t:fwa:na
·cloud·
In the model of feature representation I am proposing, S, Z, J~
~. involve Tongue Binda constriction at an Anterior Site (recall that
Anterior Site covers dental and alveolar point!! of articulation): a
true labiodental. on the other hand, is characterized as involving a
Lower lip constriction at the Anterior Site. Therefore. the feature
which groups these sounds into a natural class in this system. is the
Anterior Sit9. The fact that these two groups of sounds pattern
together in Zulu. provides additional ovidencethat Zulu
kabiodentals are represented underlyil'lgly as anterior sounds.
With the discussion of Zulu. I conclUde the section devoted to
the feature labial Sile. ! have considered the phenomena which
ap.,ear to have all the characteristics of dissimilatory processes.
and which affect only sequences of bilabial consonants and round
vocoids, when such sequences arise as a result of derivation.
• Of cour,_. I concede that this IYPI of ,yldencl, since It oxpr8sSIs lht- fellure of
Uti procell to totl plec•• Is not es slrong os posnty, Ivld"nco.
5S
Because labiodental cOnsonants have at all times remained
unaffected, patterning with non-Iabials, I have proposed that the
conditions on representations responsibls for thase processes be
stated in terms of an articulatory property which makes bilabial
consonants and round Yocoids different from labiodentals. The only
such property, short of creating ghost features, is the constriction
location: labial in bilabials and round vocoids, and dontal (sAnterior)
in labiodentals.
3.2
Anterior Sita
I now turn to evidence for the existence of a feature Anterior.
In the model I am proposing, this feature is defined differently from
that of SPE or ~.lg9Y (1986). Recall that in SPE, this feature divided
all consonants into (+/-ant8rior] classes. such that the segments
articulated forward of tho palato-alveolar fegion were [+anterior).
and all others were [-antorior]; in Sage) {1986}. this feature is
distinctive only for the coronal sounds. and dominated by the coronal
node; [+anterior] are the sounds with a constriction forward of the
palato-alv8Dlar region: [-anterior] are the remaining coronals. I use
the label -Anterior- for a privative feature which Ch~;"Ci{;i8rIZes the
sounds with a constriction lo~!!an ill the dental or alveolar :ogion.
The class of sounds characterized by this feature includes: r, Y, t. d,
\,4. s, Z, $. ~, a, &.. ts, dz, n, r:t, etc., and complex segments involving
those articulations. I assume that in addition to the above
consonants, the class of Anterior segments may include vowels Ii/
and lei. in languages in which thesa vowels are articulated with a
constriction in the dental/alveolar ragion. The arguments :or the
ante.iority of such vowels are presented in chapter 5.
A controversial aspect of the Jist of sounds which J propose 10
characterize as Anterior, is tha! it includes the labiodentals with
denIals and alveolars in a natural class. As far as I know, no system
of feature classification has ever aiiowad thesa segments to form a
natural class to the exclusion of aU other segments. Yet /f/. Is/,
/t/. Inl, atc., share an articulatory property -- they are all produced
with a constriction irl the dental/alveolar region of the mouth.
56
Current phonological theories do not que.tion the fact that sounds
articulated in that r8lIion form a natural class. as long as they share
the coronal articulator: these theories do recognize the need for the
INtur. ANTERIOR which maps onto the dental/alveolar region in the
mouth. What thoy do not recognize is a possibility thai non-coronal
sounds may also be articulated in that region, and that this aspect of
their pronuncialiGn can be ~honologee.lly relevant. and can manifest
itself in the same way that it doh in the comnaJ souns.
In this section. I present an argument from Standard Thai for
the feature Anterior which groups aU segments with the
constrk:tioo in the dental/alveolar region into a natura' ciass.
3.2.1
Velarization of Atltsrior
(Siamese)
Consonants
#
c c.
til:
·to htt-
51:
·four·e boUt
r't:
(b)
ttt
·ey€f
sual
·pretty·
·s sky·
f8:
Other consonants in the language are not valariz9d before front
vowels:
(23)
ke:
dl:
·to b! twisted·
·good·
Velarization process described by Harris is a IO~ (eval
phenomanon: first it seems to be excep!ionless. second. it is not a
57
This process can be captured in terms of a natural class
phenomenon only in a mode' of feature r9prsssntation wh;ch allows
classifying together sounds on the basis of their constriction
location: sounds like If, S, tJ form a natural class only in terms 01
the (salure Anterior Site (I disregard the 'act that these sounds
share specifications for voicing and sonorancy, as these are clearly
not the only features which pick them out of the segment inventory
of Siamese).
in Standard Thai
An ~,'t8resting phanomenon of consonant·Yowel interaction is
feportad by Harris (1972) in his phonetic description of Siamese
consonants. According to that description, certain consonaras of
Siamese become velarized before ·close front vowels·. The
consonants alfected by this process are If, 5. tJ. out of the
followinc segr;.ant inventory (Ruhle" (1975)): consonants: ph.. P. b. r
m~ l",l t. d. s. n. J. r. tl •
k". Ie. g, 1'), ? h; vowels: t. t. u. 8, 8. 0, £. ~,
a. plus the same serios with le"gth). The dala that illustrate this
phenomenon (from Harris (1972)) are shown in (22) below:
(22)(a)
neutralizing process: velarized segments are po~itjonal variants of
the basic phonemes. Theretore. it cannot be disposed of as if it were
an obscure morphophonemic alternation that has lost its
phonological conditioning-.
At this point. it is not obvious what mechanism is responsible
for the veiarization phenomenon. why Anterior segments should
acquire a secondary velar articulation when followed by a front
vowel. Ordinarily, ~alata·lilation phenomena are exp&eted in such an
environment. An analysis of this phenomenon depends on certain
claims about vowel representation that I have not motivated yet. In
chapter 5, I proposo a treatment of VOWf?I, which relies on the claim
that similar vowels of different languages may have different
underlying representations. and that this factor is responsible for
the differences in their phonological p2tterning. For examplo, I
suggest that the vowel transcribed as IiI may have a constriction
• H.nderson (198'a>' whllo not suggetUng. thot the v91arlzollon Dhenomtnon It
morphologically clrcumserlbtd. do.s wavi away thl feet that If I pott.rns
toglther with ant.rlor consonants.. by undermining the phonemic st,tul of /f/ 81
If I: Sh. tuggests that v.lorlZld If I I' r.elly an und.rlylng lebla1tz.d Dsplret8"
viler that may surreel with v.lerlzetlon 81 a reflox of Itl underlying ShOP'. But
this could not bl correct Horr', dOl. roport tho velartzed IfI .. el will as ordlnGfY
If I 01 Y8l'Ienti 0' Ikhw I In thl SPIOCh of tom. Slame'l.. but h. emDl\eISzl' thel
both luch pronunciation, or. considered ·Iow elals- among lhl .dUealed Slom•••
so.et,rs. who do not marge Ikhwl with IfI In any pOlltlon; v.lerllJtd /f/ before
etos. frent YOWII'. on the other hand. II con,tde;-.d e standard (Herr's (1972). D.
11 end 17). \Jhot this shows It thot trut If / must ". vllertzld In the stondard
spatch. snd tilet mergIng of Ikhw / w Ith If/Is v',wed to b. In poor testo. finally.
sugge.tlng that Si~m.s. (fOIS not htwe /f' Is co~pl.t.'ly unJu,tlfled on hlstorlCQI
groundl. Comparative 'vldlne. (ll (1977» suggests that Proto-Thai did have/fl.
end Storn•• t IM,rll.d It.
58
kK:ation at either the Palatal or Anterior Site, and I show that such
two differ&nt Iii's map onto sounds with the predictable acoustic
diff.rences.
In chapter 5, I present arguments, based on the acoustic
measurements, that Iii in thai should be classified as an Anterior
segm.i't. Assuming this treatment of Iii now, it is possible to
analyzG the velarization phenomenon in Thai 8S a dissimilation
proeess triggered by the oep. Given the characterization of the
vetarized consonants as pronouncftd with a velar, IW/-like on-glide
to the vowel (Henderson (1983). I assume that the velarization
proce$s corfsists in the insertion of a glide which separates an
Anterior consonant from an Anterior vower. and I propose to
represent it as follows:
3.2.2 Casel in Which labiodentals are excluded from the Phenomena
affecting DentaVAlveolar Consonants
The phenomena which single out tha so called anterior coronals
phsoomena
in which the d9ntallalv8olflr and the labiodental consonants pattern
together. Some examples include a condition on repre,entation in
Basque (Huald. (1988)), which on~y allows anterior coronals in the
coda (with palatal coronals and labiodentals in the invontory of
segments), ccioccurrenC8 restrictions in Chinese (referoncel) which
dis~low anterior curonals before front vowels in the same syllabl9,
(with labiodentals in the segment inventory), apparent transparency
the anterior coronals only to the spread of leatures to the
epenthetic vowel in Fula (Paradis and Prunel (1988»
(with
labiodental IfI in the inventory). etc. etc.
ft. s. Is, etc) are generally much more Ct:Immon than the
0'
(24)
R
I
C
I
Anterior
R
R
R
I'"
I
~
C
-tOnI
Arltenor
C
I
Antenor
C
R
-CanI
I
Velar
"
C -cons
I
Anter10r
0'
This concludes the discussion
velarization in Thai. This
phenomenon is the only example I know of in which III patterns
together with the Anterior lsi. 11J. etc. It is not inconceivable that
phonologies of languages contain more examples or such patterning.
and that such examples may go unnoticed or be dismissed as
-oddities-. However. given the amount of descriptive knowledge that
has accumulated in phonology in the recent years. it is not likely
that such phenomena are extremely common. But if so. there should
be systematic explanation as to why this should be the case. In the
next section. consider such an explanation.
• Consonants cluster ore allowsd In Thel: slnct thert erl no report, of vllarl%8t1on
off.etlng such clust.rs If they con:lst of Anl.rl~ consonenll. I aSlume that only I
-co"sonantall sigmenls m~y aDPlar In thl right-hand environment of thl rul,.
59
The phenomena which affect denials/alveolars, but do not
include labiodentals can be handled within tho constriction model,
given the fact that this model can designate a natura! class of
ant9rior coronals (as an intersection '), the class Anterior and the
clau Tongue Blade). HoweYer. it seems to me that such a treatment
would miss a broader generalization that appears to be at work.
A possible reason for infrequent patterning of labiodentals and
dQntallatveolars has been hinted at in the context of a discussion of
the articulator Lower Lip (section 2.1). There. I have suggested that
in a language with no bilabial fricatives. and no phonological
evidence for the -anteriority- of the labiodentals. the latier are to
be treated as underlyingly represented with the Lab~al Sits. I now
discuss various aspects of this proposai in greater delail. and
produce relevant arguments.
Given such a proposal. the task for a language learner is the
following: unless there is evidence to the contrary. labiodental
fricatives are mapped onto Labial underlying representations. This is
not a controversial point. as phonologies of languages are filled with
examples of sounds whi;:h do not pattern phonologically with the
60
class.. of segments with which they classify on the basis of their
surface acoustic and articulatory ~rop.rti•••. A well known example
of this phenomenon is the case of the velarized lateral in Polish.
Polish displays on surface a sound that is in no way different
from a round glide /wI. However, this sound, orthographically
represented in Polish u ,.,, pJ!Uems with liquids, not with Yocoids:
before a front vewel, it become. palatalized, and surfaces as IU,
e.g., kalWa/kole 'circle': in extrasyllabic positions, it does not
surface al lui as it ought to, given our current knowledge 01
voweUglide alternations (ef. levin (1985), Guerslel (1985», but
triggers epenthesis instead (Goracka (1988»: poswalposEw (E
indicates the epenthetic vowel here).
This sound is still pronounced as It J in the eastern dialects of
Polish, just like It I in Russian. Also, the stage pronunciation
requires It I. These facts suggest that the surface shift from It I to
Jwl has taken place recently.
The event which might havs led to It I becoming phonetically
/wI in Polish, has most likely been the shift of the underlying Iwl
to Ivl (the phenomenon which I am going to discuss in a moment on
the basis of the Russian material), which emptied the acoustic and
articulatory space of the round glide. The shift of It I 10 Iwl is easy
to understand - both sounds are velars.
Other
examples of camouflaged segments include the
pala~alized dental in Polish. which is represented as a distributed
palatal africate
Idi) in surface representation {a dental turns into
CIdi when followed by Iii or lei, e.g., kot/kot e 'cat', also. /di
depalataliz8 to Ud when a consonant.. initial suffix follows: k 0 S c'
(c
c
• It II neclslary to point out "era thot there Is only ;>erUel re,emblance between
tho CSSIS of -sound cemouf!og.- thet I am discussing below. snd thl CGSO I .", ant to
metl for thl treatmlnt of le~lodental frleeltvl'. In Ull formc.r. WI 'I' Cl.tlf
phonological pottlrnlng which JUltlfll' Doaltlng en unttlrlylng s.gm.nt dlffer.nt
from It. surlaet count.rpar t ; In thl leUlr. I urO»os. to consider If I to bt
und.rJylngl,-, rlCtrl'lntld as It/ wh.n no IY'd~nt;. It evelleDl1 Esslnl1ally. I
OIIU::n1 e unlvlrsal mechanism which s~.clfl•• thl sUt In the lablel frlcettve as
Labial. whln no Inform"Uon Is Qvctlebll_
61
'bone', but kose + k • a --) koslke 'little bone'): a voiced velar
stop in Standard Arabic, which is represented on surface as IJ /: I! I
fails te pattern with other coronals in the assimilation of the
dllfinite article IV (discussed in section 2.2.3 above), e.g., 111 +
saml -> Ilems 'the sun', but /1/ + gamest -> I!emeel 'the
camer (Brame (1970), cited in McCarthy (1989)), etc.
Ther. exist well motivated cases of camouflaged bilabials. The
best known example is that of a round glide in Contemporary
Standard RUlsian, which surfaces as Iv/. The phonemic status of
Russian Ivl has been discussed on numerous occasions (8.0.,
Andersen (1969), Coats and Harshenin (1971), Halle and Vergnaud
(1980), Hayel (1934), and others), because of its peculiar behavior
w.r.t. voicing assimilation: Ivl patterns with sonorants, in that it is
transparent to the spread of (de)voicing. while never acting as a
trigger, e.g., lot vdovwl (dvd] from the widow', !bez vpuskaJ (sIp)
without admission'. but Is vamil (svJ 'with you'. There is a fair
amount of agreement that undertyingly, lv/ must be lepr8S8nted as a
round glid~: the most important faclor is the phonological patterning
with sonorants. but other factors matter as well: first. there is no
surface round glide in the language, and so Ivl appears to fill the
gap: second, in the speech of many speakers. there is free variation
between Ivl and Iwl, which suggests that the shift from the
sonarant to the obstruent is not fully completed.
The relevance of the Russian data for the point I am pursuing
here, is obvious: the rule thai strengthens a bi~abial sonorant to an
obstruent produces a labiod&ntal, not a bialabial segment. This
suggests thai the status of bilabial fricatives in surface
representation is somehow more marked lhan the status of
labiodental fricatives. A point of departure for an explanation as to
why this should be the case, might be the fact that labiodentals have
greaier acouslic energy than 1+', I~ r. It has been observed
• Th. dlff.renc. In thl IntenSity hal onca served ot the besls for distinguiShing
b.tw.en lablodlntels end blleblal, In terms of 0 fiolur& lstrldentl. whit.
lablodlnlell have b.ln clasllfl.CI as (~'trld.ntt along with the soundllllee L 1.
62
(Strevens (1980), Nartey (1979)) that generally, languages prefer
high intensity over low intensity fricative.- .. Nartey (1979)
suggests another r••lOn for tho relative markedness of 1+1. namely,
tha' the structure 01 the mandible. might actually make the
&rticulation of Iff easier than the articulation of '+1. I would
hesiaate 10 adopt this idea, because it is not obvious to me how
bilabial stops are to be exempted from its broad consequences.
voiced velar fricative which merged with JgJ. Now compare the
obstruent inventory 01 Polish to that of Basque (Hl'a~ (1988»:
If only perceplual considerations are the reason for low
surface occurenee of 1+1 (In Nartey·. data base of 317 languages,
132 have JU, but only 20 have It" and only 31 have lI'f), then i~ is
not unreasonable to assume that the choice between bilabial and
labiodental articulation is a matter of a very late rule in the
derivation of phonetic representations.
The asymmetry of classifying IfI apart from bilabial consonants is
even mare conspicuous here: Basque has only one fricative
(voiceless) for each constriction site, ••cluding the bilabial and the
palatal fricatives.. The absence of the palatal fricative is just a gap
in the inveniory: it is not the case that Ir;/ groups asymmetrically
with some other class of sounds. However, the missing fricative
.under the Labial Site is Quite obviously IU. This observation squares
neatly with the fact reported at tho beginning of this section, that
labiodental is not allowed in the coda in Basque. even though the
other Anterior (and only Anterior) segments are.
There are (at leBst) two advantages c, assuming that most
labiodentals are underlyingly bilabial: :i;rsl, the inventories of
languages more often than not display a symmetry in the way the
segments are distributed within tho articulatory space. Consider.
for example. the obstruent invert'ory of Polish (d,sregarding the
palatalized articUlations):
labial
(25)
Anterior
Palatal
Velar
p
t
ts
C
k
b
d
dz
5
9
x
s
z
y
!
~
If labiodentals aligned with Ipl. fbi under Labial Site. the obstruent
inventory of Polish would be almost perfectly symmetrical. haying
two fricatives at each constriction site (minus the absence of
• te.
(Choms~
and Hall. (1968» (for II dllCUJlton of this .xt.nslon of (strldlnU.
(28)
Labial
Anterior
P
b
f
t is
d
tt
s
,
c
Velar
k
9
x
D. Steriade (p. c.) has pointed out to me another argument for
trealing /f/ in languages like Polish or Basque as underlyingly
represented with the Labial Site: a look at the inventories of these
languages shows very cl3arly that they are arranged in terms of
clalses de,.ned in terms of Site features. not the Articulator
features. This makes the asymmetrical behavior of Ifl in these
languages even more conspicuous.
The second advantage 01 representing IfI as underlyingly
bilabial in lanouages which show no evidence for its anteriority. is
that it explains infrequent patterning of labiodentals with other
Anterior segments: if If/ is underlyingly Labial/Lower Lip, then it
shouid come as no surprise that it fails te pattern phonologically
with the Anterior consonants.
s•• choDter 6).
• I am grateful to Kin Stavln. for dlscus.ing this topic wtth me.
• Inctd.ntoHy. the occurrence or other low Intlnslty frleetlYI. -- Intlrd.ntels -Is In thl laml region: 17 language. helvi .l.. end 20 lenguogl' h~Y. l·
63
Palatal
64
3.3
0'
Palatal Site
I assume that all sounds articulate:t with a constriction along
the hard palate are~ are characterized with the feature Palatal. This
include. sounds which have been v.rioully described in the
literatur. as pre·palalal. alveo-palatal. palata-alveolar, palatal,
dorso·palatal, etc. I assume that phonemic distinctions among the
Palatal sounds are carried Qut. first. by the articulators which
execute the constrictions (Tongue Blade and Tongue Body). and then,
by the features associated with the articulator. (in the case of
coronals -- [+/-distributedJ. (.,-Iateral): as far as the dorsal
palatals are concerned, I have not S88n a language that would
conlrast these sounds along more dimensions than [+/·Yoics),
therefore, I belieYe that the issue whether the featuras [back) and
(high) are ever used phonemically among them. is moor.
c.
s.
The following sounds are included among Palatals:
3.
Z
(palatal coronals), C. j. (palatal stops).~. J (palatal fricatives).
While the subject of the constriction location of front vowels will
be taken up properly in the chapter on representations. for the
purposes of this section. I am going to assume some of the claims
:hat are still to be argued, in particular. the claim that vowels Ii. a,
U" 01 belong among palatal sounds. Of course. to some extent. the
evidence presented below supports such a view. as it shows these
vowels interacting with Palatal consonants in ways that can only be
understood if the two classes of sounds share an articulatory
feature.
The status of the feature Palatal within the articulator model
(Sagey (1986» parana Is the status of the feature Anterior in that
model: since Palatal is (-anterior). and the feature [anterior) is
dominated by the Coronal Node. the only segments that can be
• Note lhet leI (thl Pllalal slop) ern! It'/ UronteCS Yolor). which ere both dorSfSls.
or. dlltlnglshed not In t,rms of the Tongul BOd,", fealurls. but In term. of Uti SUo:
lh. former Is representod with the Palatal Sit •• end thl 'otter -- with th. V.ler
Sltl. Thl.1 "Pr.s.ntellon, ore motlvetld In theDt.,
65
s.
characterized in terms of this 'eatur., are coronals. The piJint
this section is to show that Palatal constriction location can be a
phonological property of both coronal and dorsal sounds.
In order to argue for the feature Palatal as distinct from (anterior) of the Sagey model, i will focus on the phenomena in which
palatal coronals pattern with consonants whose palatal constriction
Is executed with Tongue Body: either dDrsaJ palatals (e.g., Margi,
Arumanian), or palatalized consonants (Russian (standard dialect).
Bulgarian).
Among the phenomena which justify positing the feature
Palatal Site. palatalization processes figure most prominently.
Since in these processes a dorsal front vowel quite commonly
induces a ·coronalization- effect in a consonant. it is here more than
anywhere els8. that the need for positing the feature Palatal Site
becomes strongly apparent. Because the treatment of palatalization
which I propose cuts across a number of theoretical points in this
thesis. I do not present it in this section. but discuss it separately
in chapler 7.
3.3.1
Palatalization of Unstressed VowelS in Russian-
In Russian (segment inventory (Jones and Ward (1969)): P. pJ, b.
bJ, f, ri , v. yi" m, m J" t. t J• d, cjJ" ts. s" sJ, Z. zj, n. nJ" t. Ii. f. r J• C, $. ill k,
g, x. I, e, UtI 0" e), the surface quality of unstressed vowels is
conditioned by a number of factors: the underlying quality of the
vowel. the position of the vowel w.r.t. the main stress. and the
quality of the preceding consonant.
It is this
will show in a
as Palatal on
which affects
regardless of
mi9ht be.
last factor thai is of theoretical interest to us: as I
moment. all consonants of Russian that are c1ass?'~ --J
the constriction model. pattern alike in the process
the quality of unstressed vowels in the language,
what their specification lor the Articulator feature
• ThankS to Mlcha,1 Kenttow lel for directing my etttntton to this p.... nomsnon.
66
In wnat follows, I outline the entire paradigm. and then focus
on the ca... in which the unstressed vowel assimilates to the
quality 01 the preceding consonant. I show that in order to have a
simple account of· such assimilati\)ns, it is necessary to appeal to
the concept of palatal constriction location (Palatal Site).
The following rules derive surface representations of
unstressed vowels In Russian (bu.t on Jon•• and Ward (1969»: III
and lui are laxed:
I, U ->
(27)
palatalized consonant (/t' " "" .... I). Afte, aD other consonants and
word-initially 181 and 101 surface as IA/. Thi'l is summarized in
(31). and illustrated in (32) below:
(31)
8,0 ->
a.o
(32)(a)
-stra.1
for example (stressed vowel is in bold type):
(28)
Jezd'ltJ
prJlfln~
IJIZdJtt'1
IprJtftnel
tJlsk't
It"skll
·to treyer
·cau••·
·ytc.·
orok
lurokl
·18SS0n"
Q~'
zntlJo
luie'
"alresdy·
IzneJul
·1 know·
(b)
lei is raised and taxed (presumably. by the same process that lues
high vowels):
(29)
8
->
(+high) I _ _
-sir,••
This is illustrated below:
(30)
tljlttrit CIS t VB
lelJettrJtcestvo
si1mi1nl
IsJemJenDI
'electrtct ty·
·seeds·
pJlri tmi ll Jt
Ipler'em181 Jt/
·grlnd", tmpef.
The quality of laJ and 101 depends on their position in the
word relative to where the main stress is. and on the quality of the
preceding consonant: in the pretonic position. these vowels surface
as III when following a front glide, a coronal palatal (If" $, !I). or a
67
__
..ar_
C
V
otherwise:
J __ •
1. Q
1/(f,t.!,tJ"J" .. "JJ
-:lit
c
A I
V
.11'."
·tongue-·
IJezttl
Ifa.t'
lp'atJtl
·welch·
",tv.·. gen.
IJoial
·hedgehog·
/2onal
·wlfe-
I~oltel
·was taking·, '1m.
SAme
Isemel
·s81'·, fem.
gtAZIi
znr\te
Igtazal
·,yes·
Iznatel
"know·" rem.
IxOct'lt'l
Idomfil
·houses·
Inogal
'leg·
·to go·
Before presenting the rest of the paradigm. I would like to comment
on the presence of '+1 (ItI, if unstressed) in some of the examples
above" The ill anernation in Russian is unrelated to the vocalic
phenomena in prelonic syllables; it represents a more general
process. whereby Iii is 'back8(f' when following tho so called -hard e
consonants. The class of -h&nr consonants includes all unpaJatalized
segments. as well as the (+continuant) coronal palatals 1$" tJ.
• The lIt eltornetlon II IXDltJlntd b.low
68
I adopt the standard anai)'sis of the i/+ alternation. which
treats the -hard- consonants of Russian as velarizecf. and considers
the presence of 1+1 after these consonants to be a resull of a rule
which spreads the b8cknsis of the vea.rized consonant onto the
vowel Iii (e.g., Jones and Ward (1989), Chomsky and Hane (1968».
This interpretation is wen supported by the evidence of the X-ray
tracings of the Russian sounds (e.g., Fant (1980), p. 170. 188), which
reveal secondary velar constrictions in p, s. I. 6. etc., rather similar
to the constriction present in the articulaoon of the vowel If1 (Fant
(1960), p. 170).
I now return to the characterizateon of vocalic alternations in
unstressed positions. and conclude the paradigm with the
presitntation of 181 and 101 in other thg:, prelonic positions. Wordinitially, 181 and loJ surface as IA/: otherwise, they surface as tel.
The statement of the rule, and the examples follow:
""-_:
-slr••s
a.
(33)
a,o
(34)(a)
"dYAket
",tAmal
levtometl
"dnAYO
lodnogol
"solicitor'
"automaton"
"of one-
ApAzdet
lopozdetl
'wes lete"
(b)
->
otherwise:
ledvokall
0
->
a
Ther. are several plausible analyses 01 the above phenomena.
For example, it is possible to tr.at the assimilation of a pretonic
vowel as a process separate from vowel reduction: in such a case.
the assimilation would involve spreading of the palatal constriction
of a consonant onto an adjacent (pratonic) vowel, followed by the
deletion of the original conelriction(s) of the vowel. Alternatively,
it could be assumetJ that Russian hal a general rule of vowel
reduction in unstressed peeitions. and that the assimilation of the
protonic vowel involves the SSKe8d of the Palatal Site of a consonant
onto an empty position of the vowel. Ultimately, I adopt the sacond
analysis (588 below), as it allows for the simplest account of the
phenomenon.
However, what needs 10 be stressed at this point. is thai
technical differencas sside, ail analyses at the assimilalion
phenomena must be able to characterize in terms of a natural class
the set of segments which condition the assimilation. As I have
shown, this set consists at all palatalized consonants of the
language. the coronal palatals. and the front glide.
taburiltke
IleburJelkel
'sloe"
Kenstowicz and Kisseberth (1979) propose an analysis of
vowel assimilation in Russian based on the assumption made in SPE,
thai leI and I~ I are (-back. +high). On this assumption, it is possible
to characterize the set
segments which condition vowel
assimilation in terms of a natural class. since both IV and the
palatalized consonants of Russian can also be treatset
(-back, +
parAxot
Iperexodl
·steamer·
high).
matAko
Imotokol
Igorodel
"mtlk"
There is a serious prob!em with the SPE treatment of If I, I~ I.
however: there are no documented cases of lei or I~ I acting like [back. +high) segments w.r.l. vowel harmony -- they Q) not block the
height or the back ness harmony. Consider, as an example, the
backnass harmony in Turkish: Turkish (segment inventory (Ruhlen
(1975)): I, ii, t, u, 9, O. 0.0; p, b, f, Y, m, t, d, s, z, n, I, C, c,!. 5, k, g.
?, h) has a harmony which assimilates in backness the suffixal
vowels to the the stem vowel. As shown below, this harmony is not
garAde
't'
·towns·
• Thl prlSlnco of
(as Will at It/) ~n thet grouD 'I not vlry .elY to ')(Dlaln· III
II e Dlletel consonant, and thll Ilonl shOuld luffici to block the valorization
procI.'" (•. g.,/f/ls not vilerlzld). On th. other hand" It II Dosslbl8 the I
veler'zallon of I t l l l D"parnle phenomlnon. unrel&ted to the rull which 8UICtS
non-peletel consonants. Unllk. If/./tl Is en apical con.onent: onl could vl,w
vllorlzatlon of It I es en .ntumclmcnt of Itl eDlcal Quellty. but SI. SloYlns end
0'
a,
K.ysor (1986» for a dUf.rent view.
69
70
blocked by either If I or 15 I, regardl••• of whAt value of [back]
sprudI (data are taken from Halle and Clementi (1984»:
(35)
sef
18ft
safe
satten
gu~
giiJii
gii5'
giiften
genf
genii
gin!.
genet.n
·helr·
secler"
·power'
giifl'ert
genflJ,r glnflltlrl 'young'
18~ler
guflJ.r
Also. If I and ,., do not spread the height feature individually (e.g.,
there are no cases of /01 and lei railing to lui and IV after
J•
III). The.e 'acts strongly sugg••t that coronal palatals are not
represented in terms of the tongue body features. Ther.tont, there Is
no fe.tur. or feature. within the articulator theory af phonologk:al
representation (Sagey (1988», that wouid characterize as 3 natural
class the segments which condition vowel assimilation in Russian:
under this theory, coronal palatals are [coronal], and IjI, as well as
the segments involving secondary palatalization, (e.g., It J" Ip'/.
etc.) are [dorsal].
Ie
Yel it is beyond doubt that in pur.'y phonetic terms. these
segments do share an articulatory property: they all Involve a
constriction in the palatal region of the mouth. I have pointed out
earlier that the model of feature representation of Sagey (1 986)
does recognize the need for a feature that represents the palatai
constriction location, but limits its use to the coronals by making it
:I dependant of the Coronal Node. Sinee it is clearly not true that only
coronal segments can have a palatal (. (-anterior)) constriction. by
making this feature independent of the articulator(s), we allow for a
system of representation more faithful to the phonetic reality, and
more capable of capturing relevant phonological generalizations. In
such a system. the set of segments which assimilate the reduced
pretonic vowel to Iii in Russian is characterized with one feature:
palatal point of articulation.
lei us now proceed with an analysis of the assimilation
phenomena, beginning with the processes affecting laJ and /0/. I
assume that vowels 181 and 101 in Russian undergo reduction in
7I
metrically weak, unstressed positions. The immediate output of thil
proce.. is a so called ·reduced- vowel - a segment with no Site
specification. (Thos treatment of the reduced vow... correl.tes with
the articulatory and acoustic facts: according to Fant (1980). p. 210.
an openly articulated sound, may have the formant pattern of a
neutral vowel (/8/) regardless of where the tongue constriction is
located; ther.for., the reduced vowel is a segment whose
constriction location is variabl., or -- in phonological terms -.
unspecified. For mor. discussion see chapter 5). Let us represent
this process as folloWl:
(38)
la/,/ol
R
/"'.c
I I
(C)
T. Body
Artlculetor
Pherynglel
Site
lei
R
I
c
I
TongUe Body
I
(+beck)
If the vowel /al remains in the metrically weak PGsition throughout
the derivation, it surfaces as Ie'. According to Jones and Ward
(1969) p. 54. -in the immediately pretonic position. which is the
second most heaVily stressed syllable in Rusian words. [... J the
Russian sound /8/ does not occur-. Instead. either vowel III (It/), or
occurs in this position.
'A'
Vowel II' surfaces in the pretonk: syUable. whenever preceded
by a palatal coronal. palatalized consonant or the fronl glide. I
propose to treat this process as spreading of the Palatal Site of a
consonant onto the Constriction Node of a stressed vowel. and I
represent it as follows:
72
(37)
+11'.8.
~
R
~
(C)
I
C
Pelatal
Tongue Body
Since the procels.. which manipulate the Articulator and the Site
node. may apply in 8 feature filling fashion only (sa8 1.3, and
chapter 7), the rule in (37) will not aftect stressed vowels which
are fully specified.
When fonowing a non·palatal consonant, the pretonic vowel
surfaces as 'AI. characterized by Jones and Ward (1969) as a back
vowel. clos. to l:ll, but with the lip rounding removed (p. 50). On the
basil of this description, it is posBible to hypothesize that I" I is a
vowel with a velar and pharyngeal constrictions, just like I~/, but
with no labial constriction. Since the same vowel is substituted for
lal and /01 in the word-initial position. I hesitate to derive its
quality from the influence of the velarized consonant. Instead, I
assume thai this quality is assigned to stressed reduced vowels by a
default rule. However. such an influence cannot be ruled out. as boSh
IAI and non-palatal consonants conta~n the velar component.
The precesses affe:ting Iii. lui, and lei in unstressed
syllables must necessarily be treated separately. Sinc~ they are of
little importance to the issues at hand, I will consider them very
briefly: the change of Ie/ to Iii is best treated as a reduction of a
complex segment -- more examples of similar phenomena will be
discussed in chapter 5. Laxing of Iii and lui. on the other hand. is a
manifestation of a more gener~j process: only vowels with main
stress may surface as [+!anse] in Russian. vowels in all other
positions are always lail. This can be expressed with the following
73
(38)
V
~
->
applied prior to the assignment of the pretonic
[-tense) I _ _
-stress
C
/~",""l
Arttculator
rule (which must
str&sl) :
This concludes the discussion of vocalic alternations in
ul1stressed syllables in Russian. The proposed analysis presents a
view according to which vowel reduction affects all cas.. of
unstressed lsi and 10/. This process creat.s representations which
are incompatible with the secondary stress assigned latar to
pretonic syllablel. The need for fully specified vowels in these
positions drives the assimilation proc... responsible for the
·palataliz~tlon· of the pretonic vowel.
The advantage of analyzing this assimilation as mediated by
the process that creates a reduced vowel (represented here as a
segment with no Sita specification) is that only those segments of
Russian whose constriction location is compatible with the Tongue
Body articulator (Palatal or Velar) will be allowed to contribute to
the quality of the pretonic Yowel. This analysis explains, for
example, why this vowel never acquires the labial component from
the preceding Ipl. lbi. Im/, etc.
Finally. by appealing to the phonological concept of Site, the
proposed analysis represents in a simplest ;lossible way the fact
that segments such as If I It JI and Ijl pattern together in a
phonological process. despite the fact thai they do nol form a
natural class w.rJ. the articulator activo in the constriction.
3.3.2
ale Alternation in Bulgarian
Scatlon (1984) reports the following alternation in Bulgarian
(segment inventory (Scatton (1984)): p, pi, h, bJ , f, ", Y, Vi, m, mi , i,
j
t J, d, d J, is, tsj, s, sj, z, Zi, n, nJ, I, ", r, r ,
dz, ~, i, k, g, x, I, ., e, U,
c,
74
o. e): stressed- 181 preceded by a palatalized consonant alternates
with leI, when it is followed directly, or acrass an intervening
consonant. by a coronal palatal, a palatalized consonant, front glide,
or a front vowel (in Bulgarian .- only Iii or 18/). Concomitant with
the change of 1&1 to lei is a depalatallzatton ot the consonant that
precede. the alternating vowel. This paradigm is illustrated below
(stressed vowels are shown in bold type):
(39)(a)
gr'ax
'stn"
"streight"
grelke
prect-
'error"
prJek
bJel
"whlt.'
"measure'
belJu
mlr J+
"I bleach"
"I measure"
"he sengi
'I bleeledo
peJ+
b!jax
bl_Je
"I sangO
"I blear
yill
°tl blew"
V.I
·It blows"
slJ ep
xlJab
"blind"
sJIPco
"bllndmen"
'br88d"
xllbfe
'Uttl. breed·
m'esto
'place"
mlstl +
"' moy,"
yJ ere
"felth"
"lefr
vert
"faiths'
"left, pI:
m'srke
p'el
", hInder"
for the alternation is a palatalization process in the sense defined
in Chapter 7: the process which spread. palatal constriction, When
the palatal constriction spreads onto the Root Node of a pha'Y"Geal
vowel laJ, the result is a complex segment with both pharyngeal and
palatal articulations: lei.
Most likely, the segment 10 which the vowel assimilates, is
the consonant on the teft. This hypothesis is baBed on the fact that
when a palatal coronal appears in that position, the assimilation
fails to take place (ct.. zyucel/zyucell 's;;unded'). The absence of
palatalization in this CBSe can be explained if W9 assume that only
the sagment on the left can spread the paJataJ constriction, and that
only the Palatal/Tongue Body constriction may spread onto a vowel.
Under these assumptions, the palatalization process may be
represented as follows:
(40)
(b)
IJIIY
'eYI
An analysis of the above facts can be offered only to ihe extent to
which we understand the phonological processes which require an
identica!. or at lesst a similar environment on both sides of the
target, a ·feature bridge- of sorts. Such phenomena are not
uncommon (e.g., Wolsaian (Howard (1973» has a rule which ra~s8s
lsi to lei between two high vowels: Kihunde {Goldsmith (1984)) has
a rule which raises a low tone flanked by two high tones. etc.}, but
so far. they are poorly understood,
For this reason, the interpretation of ale alternation proposed
here is rather tentative. I assume that the mechanism responsible
• Unstresl.d lsi surtett. &1 1.1 whontvlr Drlc.dld by e pelatollzild consonent;
there is no requlrlment that thlr. b. " psieteJ I.gmlnt on Ul' right.
.
7S
Raft
R.,'"
~
,~~-~1
A
A
(e)
c ..... ,.
Pelel T. Bod"
C
PII... T. Root
R ,..,
~
(C)
C
I
Peletsl
(Subscripts on the Root Nodes indicate adjacency conditions: the
right-hand palstal segment must appear in a syllable adjacent to the
syllable that contains the target.]
A dolail that remains to be explained is the alternation which
involves the consonant on the left: in the forms with laI, it is
palatalized: in the forms with lei, it appears to have lost
palatalization. Apparently, this phenomenon is not restricted to ale
alternation. It is true without exception in Bulgarian, that
palatalized consonants may not be followad by front vowels: when
such a sequonce arises as a result of affixation, the consonant is
depalatalized, e.g., koni • 0 -> konio 'horse, VDC..', but konJ + e
-->
kane 'horses'. The data in (39) indicate that depalalalization
also takes place when such a sequence arises as a result of a
76
phonological rule. For the purposes of this presentation let us
aasume that depalataUzation applies to any sequence of segments
which bear palatal specification. and that it deletes the palatal
constriction of the first segmenr. as shown below:
(41)
Root Node
Root Node
~
Constr. Constr.
A
Palatel
/"'.
/"'.
(ConstrJ
Tongue Body
Constr.
Peletel
Tongue Body
3.3.3.
The ale and ole Alternation in (OJd) Polish
• II Is pOlslble thet dlPGlelalizetlon proelss s.rYI. two different purposes. on on.
hsnd • It ,stlstl•• thl OCP r.Qulrlmlnt that there tNlIl not b. two odJDClnt paletel
con.trlctlonl • and on thl oth" hand. It saUsfies lhe conltralnt agalnat doubly
linked matrici. (lhe fact that glmlnat. consonants art not w.lI tol,retld In
eulgsrlDn (cf. Scetton (1984» supports thll point). I do not discus. thl.1
pOIA~bllltt.s Ir, detell. beCDl:SD thl probl.m of dlpelatcllzsUon Is only romotlly
related to 'hi lODlc at hand.
77
e,
j..
(42) (a)
0'
Let us conclude the discussion
ale alternation in BUlgarian,
by amphasizing the feleyBnC8
this phenomenon to the claim that
there exists a phonological feature Palalal Site, and that this
fealure maps onto palatal constriction location. Palatalization of a
stressed laJ takes pI~e only when the vowel follows a palatalized
consonant. and precad8s a cyllab:. containing either a palatal vowel
(IV or lei), a palatal coronal (/~I or IiI). a palatal glide (Ij/), or a
palatalized consonant (/pJ/, Il'l. 8tC.). Since the only articulatory
property that separatos Ii. e. ~. 9. J, pi, t J • etc.} from all other
segments of Bulgarian is the palatal constriction location, we can
conclude that this propar~ forms the basis of the phonological
feature which plays a role in the palatalization phenomenon, the
feature Palatal Site.
0'
The phenomenon that I am about to report has lost phonological
generality in the modern day language, but al some point. must have
been a part of Polish phonology. (Segment i~venlory 01 Polish
(Schenke, (1973): and my own corrections with regard to the apical
consonants): p, pi.. b, 1)1, f., fJ, V, vJ. m. m', i, d, ts, s. z. n, t, I, r,
dl.
t, f, s', c', Z', dZ', k, g, x. I, 8, U, 0, 8). The facts are. to some extent,
the mirror imagw of the situation found in Bulgarian: 101 and laJ
become lei when preceded by either a glide
a palatal coronal
/~ I or Ie I, or a palatalized consonant (/p', t
etc). and followed by
a pglatalized segment. The following are same of the example. that
have retained the alternation:
(b)
'tl.
Jedo
"they go"
Jedie (di =.;)
"gols·
kvJet
"flower·
kyJectst+
fern+
-black"
fern
·flowery"
"bleckn8SS·
·song, dim:
pJesn
·song"
·bBS·
pice I"
-bel. adj.·
pJosenke
ptfole
The analysis of the above data follows the same path as the analysis
proposed for the ale alternation in Bulgarian, oxcept that in Polish,
the palatal constriction must spread from the segment on the right:
when this segment is a coronal palatal. such as If I, the rule does
not apply. e.g. k§ok/ksec8k/kiecDsty 'bl!sMiUle bushibushV·. A1~.
in Polish. 101. in addition to Ial. participatos in the alternation. The
a --> e change can be explained as follows: when the palatal
constriction spreads onto 10/. represented here as a complex
segment with a velar and pharyngeal constrictions· (S88 chapter 5),
the velar constriction is deleted as incompatible with the palatal
constriction {vowel with both palatal and velar constriction would
necessarily have to have a single articulator (Tongue Body) involved
in two different constrictions: it has boen suggested earlier (see the
• In PoBSh. this segment surfacl! as
-c... a palatal attlcal.; $8' tho comment In
SecUorl 3.2.3.
• I consider rounding to bo r,dundently sPlclfled In
conltr Ictlon.
18
YOWII,
with
19
y,lar
Introduction, also Sagey (1986») that complex segments in which any
feature specification is repeated art' disallowed).
these similarities, I propose to analyze the Arumanian data much in
the same way that I have analyzed the Russian case.
The conclusion the: follows from examining the Polish data is
the same as the conclusion thai we have reached on the basis of the
Bulgarian 82tsmp18: in order to be able to capture the essence of a
phenomenon in which a front glide, a palatalized segment and a
palatal coronal pattern alike, a phonological theory must recognize
the feature Palatal Site (or its eqUivalent), independent of the
features which specify active articulators.
I assume that the assimilatory feature in Arumanian is the
Palatal Site. which spreads from a palataliZed consonant or a
coronal paiataJ onto the Constriction Node of a reduced Yowel. Here
again, what makes such a rule possibie. is the fact that the Site
specification is absent from the representation of the reduced vowel
(see 1.3, and chapter 7). Therefore, the assimilation is a featurefilling process.
3.3.4
ale Alternation in Arumanian
Arumanian (segment inventory (Golab (1984»: P. b. r, v, t, d, 5,
z, ts, dz, C. df.. §. ~, k. k'. g.. g', X, x'. 1(, (It'), m, n, raJ. 1, 1'= I, U, 9. O.
8e/m. a. 08/~. 8) is a Romanian dialect. now thoroughly mixed with
Macedonian (sometimes referred to as Macerlo-Romanian for that
reason), spoken in Macedonia. The phonology and morphology of this
language is .jescribeo in some detail by Golab (1984), on whose
description these comments are based.
lei is an underlying vowel of Arumanian, and it can appear
freely in almost any environment, except when preceded by a coronal
palatal or a palatalized segment. In the latter cases, morphological
/a/ surfaces as lei. This is shown in the examples below, where the
third person Aorist ending alternates between I-e / after nonpalatal endings and I-e/ after stems that end in a palatal:
(43)(a)
(b)
purt
+
a
--)
)ulcr
•
~
--)
b;s
+
8
--)
talJ
+
a
--)
'cerrledo
pL!rta
lukra ·worked·
bass
lalJe
3.3.5
Palatalization of /al in Margi
·klssed·
·cut·
The facts of Arumanian are qu~te similar to the ali alternation facts
in Russian, discussed in section 3.3.1. In both languagss the vowel
that assimilates io a palatal segment is /9/, and in both languages
the trigger can be either a dcrsal or a coronal palatal. Drawing on
79
Despite many similarities, there are also differences between
the Arumanian and the Russian case, the most important difference
being in the quality of the assimilated vowel ..- (II in Russian and
lei in Arumanian. A plausible explanation for this difference might
be that at the point when the rule applies in Russian. the height of
the reduced vowel (predictably [-high), 588 the discussion in chapter
5) is not specified. In such a case, follOWing the assimilation, the
feature
composition
of
the
newly
assimilated
vowel
(PalataVDorsal) will yield a redundant '[+high) specification on !he
Tongue Body Node. Supposo now that the reduced vowel in Arumanian
is specified as (-high] at the tnme when the assimilation takes place:
in such a case, after the spread of the Palatal Site, one of the
following two can happen: either the [+high) specification becomes
deleted, or else, the pharyngeal constriction is added to the Root
Node of the vowel (on the assumption that Ie/ is always a complex
segment). The latter appears to be the order of derivation in
Arumanian.
Hoffman (1963) reports the following alternation involving lal
W
in Margi (segment inventory (Hoffman (1963»: p, p'" ~ b~ b • t s'. dz". f ~
f"~ y~ y"l m. m"', ~ .., ·P. -b. ts, elZ. ts, J,\z, ·ts.. -elZ. lab"', 6. "ts. "elz. 6".
"ts", pt. b4 • tiff. c', dit.. t, d. C, dz. s. z, n, ). lW, ad, IIl C• D dz , Sill, ~, al,
"d. I'f, "'dZ, f, 4". "t, d. s, i, fl, "t". "1, f. ·s. -fl, 11I4, ell f~. vJ. J. ~. "e. k..
k"'. "k, 'k", ? g, gW. "g, "9". x, x·, l, ~". fl, 1')"'; 1, E, U, ~. a. e): following
80
palatal dorsals (c. j., ~, J), front glkle (i). and palatal coronals (c 5
i, f), word-final (but not phrase-final) ,., turns into IV. In the
absolute phrase-final position. ,., turns into a vowel which
Hoffman desribes as
centralized close back vowel without lip
rounding- (p. 21), and transcribes as ''II. This change affects only
lei. and it occurs regardless of the quality of the preceding
consonant. It is because of this additional alternation, that we can
distinguish an underlying Iii and IV derived from /81 after palatal
consonants: the underlying Iii is not affected in the phrase-final
position. The following examples illustrate the paradigm:
#
a.
(44)(8)
,., in word-final position:
lal in phrase-final position:
sa
sa)
mda)
mds
b6e
b6V
dlfe
def';
·thlng'
'persnn'
'chew'
'mush'
¥efl
¥ertj
·arrow·
fwi
ff
s(
il
ilj
Jl
JiJ
'like'
.;)
h;c)
If-J
'leer"
h,cV
'cafe for'
flJV
"burn"
~V
11*)
'speak'
'spln'
'clelm ftrst"
(b)
Iii in word-final position:
Iii in phrase-final position:
, ,
fmf
Iml
8I
"weter'
#
err(
cft!
!lrl
!lrl
The Margi facts greatly resemble eli alternation facts ~n
Russian. Therefore. we can assume that the analy~~~ whiQ"- we havo
proposed for Russian extends to Margi straightforwardly: the chang A
of /s I to Iii after a palatal segment involves tt;e spreading of the
Palatal Site of a consonant onto the Constriction Node 01 the reduced
vowel: again. what makes such a rule possible is the ract that the
reduced vowel is represented with no Site specification.
The fact that the assimilated vowel surfaces as IV suggests
that in Margi, just like in Russian, the rule spreading Palatal Site
applies prior to the rule which provides /el with the height
specification. The segment which results from the assimilation is
represented with a single Constriction Node: PalataUTongue Body;
such matrices are redundantly spsc!fied as l+high).
It is not clear why lal becomes palatalized in the word-final
position only. One possibility is that while not mentioned by
Hoffman, this position i5 metrically strong in Margi. On the other
hand, I have not found cases of iel following a palatal consonant in
the word-internal position. Apparently. this question cannot be
resolved until more is known about the language.
With the Margi case, I conclude the present section. Let us
briefly summarize its main points: (1) We have seen a number of
phenomena in which sounds produced with a constriction in the
palatal region of the mouth pattern alike. despite the fact that they
are not executed with the same articulator. These phenomena have
been shown to constitute evidence for the feature whose phonetic
base is the palatal region of constriction -- the feature Palatal Site.
(2) These phenomena have been analyzed in the manner which not
only captures the fact that segments with palatal constriction form
82
a natural class. but also explains the bahaviar of the.e segments in
phonological procul.l..
3.4
Velar Sit.
(b)
Ths region of the mouth which forms the articulatory basis for
tha feature Velar Site Is the soft palate. Since the only articulator
that can form a constriction in thl. region is the tongue body, all
velar sounds are necessarily dorsal_ This makes the task of arguing
for the phonological constituency of the V.lar Site somewhat
dlWlCult: any precess that anects a velar consonant (k, g. x, 1f, f)
can be ambiguously interpreted as targeted at either the dorsGI or
the velar specification; many procelses that affect "8lar vowel~
(lui and /01) can be analyzed in terms of backness or labiality,
ulually without greatly compromialng the phonological generalilation.
In order to argue for the Velar Site, it is necessary to contrast
the behavior of velar sounds with the behavior of other dorsals. In
W'hat follows. I present two such cases: the case in which velar
consonants undergo dissimilation before velar consonants and
vow.ls in Polish, and the case in which vetar vowels turn into velar
consonants in Kinyarwanda.
3.4.1
Ve~ar
Palatalization in Polish
Ther. are several suffhtes in Polish which trigger
palatalization of stenl-final consonants, but only if these
consonants are velars. Tne suffixes Bre I-kl (diminutive
lagentlferninine), I-usI (diminutive/agent), I-un I (diminutive). The
following 8it2mpZ8S illustrate their phonological behavior:
(45)(a)
pi J-ek
nog-&
pIJ-~C-k-EJ
U:t-o
nuz-k-a
ui-k-o
kot
mts-s
m1s-k-a
kot-k-s
83
'drukard·
"leg'
·eer'
feb-a
!eb-k-e
rur-a
rur-k-e
'pipe-
xtop
xtop-k-a
-pelsent'
nog-e
no!-ui-a
-leg'
Yllk
vllf-ul
"wolf'
tet-o
tet-us
,
mam-ul-e
4mom -
mem--,
reb-oY-a-c
stn
(c)
"rog-
-dedI
tsur-a
rab-us
stn-us
lsur-us
rik-e
nog-a
roc-un-o
"hancf
nof-uri-a
'leg-
mam-iJ
mem-un-a
·mom·
tet-o
beb-a
tat-un-o
-dad'
bab-un-a
tsur-o
tsur-un-a
-grendme'
'deughlar·
"thle"
'son-
4deughter'
In order to interpret the above data correctly, it is necessary
to compare the behavior of I-kl, I-usI, I-un I with the phonological
paUerning of other suffixes in the language. Polish suffixes can be
divided into two main categories, depending on whethsi they do CJr do
not trigger palatalization, I shall now characterize these categories
briefl1·
Palatalizing
suffixes
turn
stem-final
~abials
and
dental/alveolars into thoir palatalized counterparts. and va&ars cnto
palatal coronals, e.g.• J·iskol (augmentative): )( top I JC to pJ is k 0 'guy',
kot/koc I sko· 'car, pi JOk/p'jectsko 'drunkard'. The palatalization
process may be represented as the spread of the palatal constriction
of a vowel onto the Root Node or the precociing consonant: whon
applying to the fabials and denials/alveolars. this process derives
complex segments: palatalized labials, palatalized dentals. etc.
• Remomber from
Polich as
thl
preylous d!lcutlton ha.etton 3.2.3)
Ie. s. il --(+dI9irlbutldl p81el&ll.
84
th5t
It'• •'. zJI ,urfffe. In
Since these segments are present in the phonemic i;1veniory of
Polish, nothing can p:-ev8nt them from surfacing when they are
derived wia a phonological proce.s (I.. the disc~ssion of
palatalizatiol1 in chapter 7j. In the cue of Y'liars. it may be assumed
that following the spread of palatal constriction. the veiar
constridion is deleted. The resulting palatal segment is ronliztKt as
a coronar (because .here are no dorsal prdatala in the language).
Palatalizing suffix.. fall into two sub-categories: those with
an overt palatal component, e.g.. I-tskal (augmentative). '-ina!
(l1omlnaUzing). l-ikJ (diminutive). /-ell (nominallzing), and those
that on surface appear to begin with non"pQ~atal segments: l-arJr
(ave":), e.g.. mleko/mlefel 'milkman', nuda/nudz' e i 'bore'.
baba/bab"ai 'w:Jmanizer': !-skl (attributive). e.g., xud.'xudZ ak 'thin',
strax/stralak '188(. l-asl1 (adjectival). e.g.• kAsk/kiee8,tf 'bushy'.
df8Yo/d~.yjes\+ 'of trees', glinalgliri'estf- 'Q' clay'. etc. For the laner
category of suffixes, it is co~v8ni8nt to posit a floating segment
with a palatal constriction (cf. Gorecka (1986), Gussman (1987»)
which spreads onto the stem-final consonant, much in the same way
that the palatal constriction of an underlying palatal Yowsl does.
The class of non-palatalizing $uffixes consists of morphemes
that can begin with 101: I-os I (attribute), 8.g., yJ G I k 1/ yJ II k 0 SC
'greatness" jasnf/jasnos ~brighmess'. grub+/gruboi 'thickness': /oyl (3djectival), e.g., dOmJdomoy;' 'hom:', yJ.k/yJekov+ 'century',
bioza/bzozov+ 'birch': and with Ia/: I-acl (agent), 8.g., b'eg/bJegec
'runner', t!'pec/tapac ·cetcher·# I-avl (adjectival). e.g., rud+/rudev+
':eddish' sBrok'l/sefOkey+ 'wide" t~sJ/ttse'J+ 'baldish'. In addition,
c
c
c
there are a number of suffixes in Polish, which begin with palatal
vowels, but do not palatalize the stem-final consonants. Here are
some examples: l-iJ (Nom. plural), e.g., kotlkot+ 'ear. Z8 b BIz IS b ..
'frog'. nogaJnag', 'leg'; l .. emJ (Instr. sing., masc.) koUkotem 'caf,
pasJpasem 'belf. kroklkrok'em 'step', ete.
The absence of palatalization before these suffixes cwn be
explained under the assumption that pa~atalization is a cyclic rule
(cr. Rubach (1984). and thai affiles which begin with palatal
vowels, but do nol trigger palatalization, are non-cyclic.
Let us now go back to the dlsc.ussion of sufflxel l ..kI. I-us I,
and I-un/, with which we have opened this section. A brief look at
the data in (4Sa-c) ravsal. that as lar 81 the procesl of
palatalization is conc....rned. these suffixes do not pahern with any
other morpheme in the language: even if we tried to posit a floating
palatal segment in the underlying fctpresentation of these suffixes,
we could not explain the fact that the, do not palataJlze non-velar
consonants.
A simple analysis of Jhe process which affects ste,;,-Hnal
velars before I-k/, I-us I. and I-un I, is available under tile
assumption that constriction location features are phonalogically
relevant. It is a phonetic fact that 1kJ, IgI, lxi, and lui involve a
constriction in the velar region of the roof of the mouth. If we admit
a phonological feature Velar Site. which characterizes a:1 sounds
articulated in this region, then, capit~lizing on the fact that both
the trigger and the target in the pro~ess of valar palatalization
before I-kl. ' ..us I, and I-un I are velars, we ca", explain this
process as a dissimiiation effect. triggered by the DCP applying on
the velar tier ~ The dissimilation effect can be represented as
follows:
l
• It would G:>P•• tnsl th; TOrnJue Blmd. Is thl d.fault srtleulator In contonont,
with tha Petlot!al constriction;-thl. point Is taken up In mort dllell In tho" .lcUon
on DllatDlUzOUOft.
• Phon.UeeBy reallzld BI/-a!1 or I-eil (word-ftnolly).
85
86
(48)
Velar Dissimilation: application: cyclic
R
R
c
c
Velar
Veler
I
I
t
I
Let us complete this analysis with some additional details. First. I
assume that the Palatal Sits specification is supplied by a default
rliis. Second, I aSlume that it is not necessary to analyze
dissimilation as a process which deletes the 9ntire velar
con$triction, eYen though that is what it appears to be on surface. On
the assumption that predictab;e feature specifications may be
absent from the Uneertying Representation (er. Archangeli (1984),
may consider velar segments in languages that
Steriade (1987»,
do not contrast fronted and ordinary velars, 10 have an underlying
representation universally unspecified for the active articulator ••
after all, they may be executed with Tongue Body only. Given this,
.e do not have to assume that the dorsal specification of the velar
consonant imposes a limitation on the mechanism which supplies the
default Site specification for the dissimilated consonant
w.
Let us concluda the discussion of velar palatalization in Polish
by stressing the relevance of this phenomenon to the claim that
segments articulated in the velar region of the roof of the mouth do
pattern alike in a phono~ooicaJ process. In order for the theory of
phonology to be able to explain such phenomena in generaJ terms. it
must recognize tho feature which can group the velar sounds into a
natural class, the feature Velar Site.
3.4.2
Tha dissimilation analysis of the data in (45) makes some very
concrete predictions regarding the representation 01 affixes which
may cause velar palatalization: it predicts that only suffixes which
begin with a velar segment will ever trigger this process. This
prediction appears to be a correct one: apparently, no Polish suffix
that begins with a non·velar segment can selectively palatalize
velars.
There now remains one last fact to account for: while it is
true that all suffixes which trigger velar dissimi l 3tion begin with a
velar 580ment. it is not the case that aU veJar·inittal suffixes
condition this process. For example, the suffix /·uS! (nominalizing)
87
segmen~
8.g., nagilnagu5 'naked', yJere/yJerus
as I can tell, no suffix that starts with 10/
(r!'Ca1l that according to the proposal made
represented as a complex segment with a
".'~r, pharyngeal, and redundantly (ct. Stevens and Keyser (1989)) •
• lab'al constrictions). A plausible explanation for this gap might be
that tt~e dissimilation process applies in derived environments only,
and t"a~ suffixes whk:h do not trigger this process are no n--cycl ic. It
is pos.ib~. to C.efend these ctalm. for the following reasons: first,
v31ar con£onants and velBr vowell cooccur fr881y in underived
environment' (e.g., Ikubet J 'paU', lokulistal 'optometrisr, IguzJlkl
'button', /xud~1 'thin', 8tC.), second, the suffixti which do not trigger
velar dissimna~'on, despite the fact that they meet the structural
description of tb., rule, do not trigger any phonolooical rules, cyclic
rules included.
does not palatalize any
'soldier'. Also. aa far
~alat2lIz.s just velars
in this thesis, 101 is
Vowel Hardening in Kinya,wanda
A second argument for the faature Velar Site is based on a
phenomenon which Dccers in Kinyarwanda (segment inventory
f. Y, pf, t, d, n. 5, z. ts. r, S. i,
(Kimenyi (1979)): consonants: m.
C. I. fl, ~. k, g, x: vowels: i, a, U, o. a, pius the same series with
length). an Eastern Bantu language spoken in Rwanda and Burundi. Tho
discussion that follows is based on the description of Kinyarwanda
phonology by Kimenyi (1979).
a.
c,
Vowel .=Iusters are disallowed in Kinyarwanda, and are
eliminated whenever they arise as a result of the derivation. The
method of elimination depends on the qualities of the vowels
88
involved in the cluster, on the quality of the preceding consonant.
and often. on the morphological context. In what followa. I shall
dlaregard morphologically circumscribed proces•••, and concentrate
on puntly phonological phenomena. The p&ndIgm 01 these phenomena
is outlined I»low_
When • non--Iow vowel is preceded by a consonant and followed
by another vowttI, it merge. with the preceding consonant. forming a
camp". segment (el. Sagey (1988) for the tr••~ent of complex
••omantl in Klnya,.ancia) which retains the articulatory
characteristics
the vewel. Back rounded vowel adds two
constrictioM'to the consonant: VeiarlTongue Body and LablaULciwer
0'
(49)
ku
11'1
ku
+
•
+
• II --) kuaoa
o:ondo --) Ib'oondo
rl .. II --) kur'e
SI
Imt • u:ga --) Im"u:ge
tu
+
e:nga --) tlr'*a:l')ga
leu • dod • u
+
Pon • u
+
ku •
"we het.e --) kudod"'a "to bl lewld'
a --) ku~on"'D "to be s••n"
Kimenyi notal (p. 12) that after the labiodentall (pl. f. v), the
high back vowel is deleted. if it is followed by another vowol. This
is shown below:
(4&,
ku
ku
+
YU
+
pfu
e _.,
•
+
II
--)
(50)
be
+
ro
+
u:bek
n • za 8:mlr •
be
+
gupfD
A front vowel turns into a PalataUTongue Body constriction
when it merges with the preceding consonant:
• I hey, b.,n unobll to find ,xDmpl'l of 101 turning Inlo en obltru9nt. Sine. the',
a.ampl •• :»f U,. front mid yow.1 undergoing this proc•••• I Illume that lh.
generaUzetion Itsted by Klmenyl II corr.c·.
89
•
It
e --) beru:beke
--)
a: '!' I: k~r • t:r •
nz.:mlre
JI --) bOkor.JI
-they
worked for thlmsaIY••·
Kimenyi (1979) propose. the following analysis of the above
data: first. non-low vowell are glided when followed by another
vowel, then a consonanl is inserted between the glide and the
preceding consonant .- a palatal stop before the front glide. and a
Yelar stop before the back glide. Finally. the inserteet consonant
as,dmilat.. in voicing and nasality to the preceding consonant. and
the back glide f~S81 with the inserted consonant. while the front
glide is deleted.
kUYD
I have not been able to find a single example that would show
bilabials behaving differently from labiodentals. In fact. Kimenyi
mentions two cases with ImJ and IPI (p. 12), in which the first of
the sequence of high back vowels also fails to become an obstruent.
and is deleted instoad. ~r: the absence of more precise information. I
assume that round vowels are always deleted between 9. labial
consonant and a vowel.
11'"
·profl•• lons·
Finally, the illustration of the behavior of the low Yowel in a
cluate, with another vowel complet.. the paradigm. The lew vowel
181 is del.ted whenever followed by another vowel. aa shown below:
lip·:
(47)
'to grind"
'mud'
"to I .. t·
I adopt Klmenyi's position that there is a gliding process which
precedes any other rule that is at work here. Most likely (although
Kimenyi does not present the data that would help decida this issue).
a word-initial non-low vowel turns simply into a glide when
followed by another vowel. I analyze gliding as a process which
resyllabifies with an onset. a Yowel that has been demoted from the
nuclear position.
Rather than proposing a consonant epenthesis. I follow Sagey's
analysis of Kinyarwanda. and assume that the glide merges with the
preceding consonant. The support for this view comes from the fact
that. as pointed out by Sagey. and as illustrated by the data above.
none of these alleged consonant clusters involves any given Site
feature. or any given Articulator ~ture more than oncs.
90
The procesle. posited above are Illustrated in a sample
derivation in (51):
Sagey (1988», therefore, the offending vowels must be merged with
consonants.
·to be
The reduction of lei and 101 to J* I and f I respectively
requires an additional comment: in the preHnt model. th... vowels
are represented as complex segments: lei has a PalataUTongue Body
and Pharyngeal/Tongue Root constrictions: lof. constrictions are
Velar/Tongue Body and PharyngeallTongue Root. Since the,. cannot
be glid.. fully homorganic with the.. vowel., W8 may assume that
in the procesl of gliding. the PharyngeaVTongue Root constriction is
deleted from both of these segments.
(51) ku + dod + U + e
--)
kudodwlI
--)
kudodfWe
sewed-
(8)
(J
/NI
I
ku d
0
cr
fI
f1
N
fI
/1I
II I
I
I
d u
fJ
--)
I
ku d
II
0
A
du II
~
(b)
kudod us
kudod"e
--)
I
I
R
R
~
c
c
/1
c
/l
c
c
~
/1
lib Ltp VII Bod.
~
Ant 81 "el
~
80dII lib l. Ltp
The analysis of Kinyarwanda proposad here reUes on the model
of feature representation in which only the vowels IV and lui share
articulatory features with a palatal/dorsal, and a velar/dorsaS
respectively. By contrast, in the model of feature representation
proposed in Sagey (1986), only the velar (and labial) consonants
share articulatory leatures with Yowels, as pointed out in 1.3. On
that model, all cases of glide hardening and fusion in Kinyarwanda
should yieid a segment with a velarldorsal component (see an
elaboration of this point below).
On the constriction model, these phenomena are explained
under the assumption that representations which ~ntain clustered
nuclei must be repaired to conform to the conditions on syllablo
structure in the language. Since KIi1yarwanda allows only CV
~ syllables, only complex segments may be allowed in the onset (cf.
91
Finally, there remains the case of 1&1: the absence of the
merging effect with Ia! is perhaps due to the fact that glides
homorganic to laJ are either non·existent or very uncommon. The
deletion of 181 can be viewed as a default procedure whose purpose
is the same as thai of the gliding process: the removal of the hiatus.
The phenomenon of Kinyarwanda discussed above belongs
among the so called strengthening/weakening processes, which
affect strictur. features, but leave the articulatory component of a
segment intact. Such processes occur quite frequently, either in the
form of a sound change, or as a phonological rule. An exampltt of a
process of weakening which provides us with additional evidence for
a close articulatory match between the velars and the high back
vowels is the g ._,.. w change that occurred between Old and Modern
English, e.g., bugan .... > bow: lagu _..> law, etc. (Lass (1971».
0'
a strengthening process is found in Kpan (i,
Another example
e, U, 0, a, plus a nasalized series of same: p, b, f, v, m, W, t, d, S, Z, n,
r, ts, dz, k, Q, x, l .. I) .. j) {a Yukunoid la~guage discussed by Shimizu
(1971». In Kpan, front glide turns into a palatal coronal fricative
after the labials and the back rounded glide turns into a velar stop
(without rounding) aftor any consonant except a velar. After a velar,
the back rounded glide surfaces as a labial. The newly derived
consonants agree in voicing and nasality with the segment that
precedes. (Unfortunately, no data are quoted by the author.)
92
This phenomenon is very r.manlac."t of the process which
merg.. a vowel with • consonant in Klnyarwanda. with an
inter••ttng twist: her.. the front glide doe. not give a rise to a
palatal dorsal. but change. into a coronal legment instead. Of
Int.rest he,. is the fact that this proc... pres.rves the original
constriction location of a glkle. but not the active articulctor. If
this observation were to be generalized to tb. en"r~ p.rMt!!m, the
conciulion for Kpan would be that only the original constriction
location ia preserved in the strengthening process. and that the
aciive articulator is assigned by a default rul•.
Finally, a morphologically conditioned glide/consonant
alternation occurs in Fula (i, a, u. 0, a: P. b.
IS. m. 6. I. t. d, n, I, r.
f. !. ef. P. k. Q. rJ: every segment can be geminated: Ruhlen (1975)),
where depending on the gracie, a back glide surfaces as either a back
glide, a labial stop. or a vetar stop (Lieber (19M»).
+.
All of the abcv8 examples indicate an intimate relationship
between the consonants commonly ref.rred to as ·velars- and the
high (mid) back Ydwel. a relationship, which an adequate system of
phonologicat representation should be abl. 10 express.
One cOl~1d conceive of treating the strengthening processes
within Sagey'! articulator model by making use of the height and
backn... f••ture. ... the dependants of the Dorsal Articulator.
However. such an anatysis faces some serious difficulties. The first
difficulty arises in connection with the phonological representation
the articulator model assigns to the plain velar consonants: In order
to account for the fact that plain velars never block dorsal harmony.
regardless of the harmonic feature, Sagey (1986) proposes that IkI.
loJ, Ixl. If) I, etc.. ba represented wi!h the bare Dorsal Node. I agree
with Sagey that the behavior of velars in vowel harmonies is a
phonological fact that must be accounted for. and I adopt her
treatment of velars. 10 the extent that it can be incorporated into
the model I am proposing: I assume that velars are represented with
the Tongue Body constriction at the Velar Site. and that the Tongue
Body is not modified by any features.
93
Within the articulator model. representing velar. with an
unmodified Dorsal Node has the effect of making these consonants
-generic- dorsals, which have as much in common with the high back
vowel. as they do with any other dorsal segment, and that category
covers all vowels on the Sagey model. On that model. there Is no
reason why, in Klnyarwanda. the vowel laI, or for that matter, any
vowel, should not turn into a velar (dorsal) element. Structure
preservation cannot be of help her.. as it does not prevent /8/ and
10/ from gliding. Therefore. the unique relationship between lui. 10/
on the one hand, and the velar articulation an the other hand. cannot
be explained on this model.
Another difficulty in appealing to the dorsal features for the
purpose of explaining the voweUglide/consonant alternations. has to
do with the task of representing the palatal dorsals. and the
palatal/dorsal articulations in a complex segment. Presumably. the
articulator theory would have to treat these sounds as dorsal
segments with [-back] specification. However, evidence from
languag9~ which contrast palatal stops· and fronted ve!ars on
surface (e.g.• Basque, Macedonian. etc.), suggests that this is not a
possible solution. This topic is discussed in more detail in chapter 8.
Let us concl~de on the basis of the above, that the unique
simelarities between high back vowels and velar consonants cannot
be explained on the articulator model, but they are predicted by the
model of feature representation which recognizes the phonological
status of the velar point of articulation.
• Tho DhonologlcDe stotus of palatel dorsals Is dlscuII.d In morl detail In siction
6.2.
94
3.5
Pharyngeal Sile
The feature (pharyngNIJ. in the sen.. that will be used here.
has been introduced to the theory of phonology by McCarthy (1989).
in a comprehensive. revealing study of the so called -gutturals-•
consonants articulated with a constriction in the uvular. pharyngeal.
or laryngeal region of the vocal !ract.
McC..-!hy's discussion of the f••ture [pharyngeal) is based on
the phonological material of the Semitic .anguages. where ttte
guttur.11 occur commonl,. and In gr••t variety
phonemic
contrasts. In his study, McCarthy show. that the gutturals pattern as
a natural cia.. in a number of phonological processes: they have a
lowering effect on the adjacent vowels. they are SUbject to
restrictions on syUabification to which no other segments are
subject, and they appear to be transparent to some vowel
assimilation processes. We shall discuss some of these phenomena
lat.r on in this sectton.
0'
McCarthy argue. convincingly that the articulator-based
theory of feature representation has no means of explaining the fact
that gutturals pattern as a natural class, as the three classes of
segments: uvulars. pharyngeals and laryngeals do not share an active
articulator in the same sense that coronal or labial sounds do. He
points out thai -the gutturals are produced by three entir&ly distinct
gestures: a purely glottal one in the case of the laryngeals,
retraction of the tongue root and epiglottis and advancement of the
posterior wall of the laryngopharynx in the case of the pharyngeal,;
and a superior·posterior movement of the tongue dorsum in the case
of uvulars-.
He also dismisses a possibility that the gutturals might be
characterized with a unique acoustic feature: ·On the acoustic side.
the gutturals are also quite diverse. The laryngeals theoretically
have ths resonances of a uniform lube (though in practice, they
simply adopt the properties of the vocalic context). The pharynge81s
95
have a very high first formant, but the uvulars raise the first
formant only slightly above the value for a uniform tube. Both
uvula... and pharyngeals have a low second formanr.
McCarthy consider. earlier treatments of point of articulation
features in phonology: Jakobson, Fant and Halle (1951), Chomsky and
Halle (1988), Lad8fog~ (1975), Williamsen (1977). and concludes
that none of these frameworks are capable of characterizing the
guttural consonants as a natural clals. The feature (flat] of Jakobson
8t at, dcgfined acoustically ae equivalent to the low second formant,
can classify uvular. and pharyngeals together, but leav•• out the
laryngeals.
The SPE framework, while technically capable of
characterizing gutturals as a natural class (: ([-high, -anterior», can
do so only at the cost of drastically redefining the feature (high): in
SPE. the art;culatory basis of this feature has been v.rtica~
displacement of the tongue body from its neutral position. However.
ss McCarthy points out (and before him. Kenstowicz and Kisseberth
(1979), the tongue bcdy doe!» not participate in the production of
pharyngeal. or laryngeals. Also, McCarthy observes. the uvulars,
while grouped with the rest of the gutturals by (-high) In the SPE
framework. are articulated with a high tongue position. These
clashes between the phonological representation and the
articulatory reality lead McCarthy to reject the SPE approach to
representing the gutturals.
Finally, McCarthy shows that ladefoged's (1975) and
Williamson's (1977) theories of multi-valued point of articulation
features are also incapable of treating the gutturals as a natural
class, because in these theories. uvulars, pharyngeals and laryngeals
are characterized by throe different features, corresponding to three
different constriction locations.
In order to reconcile the phonological facts about the gutturals
with an assumption that place features must havs an articulatory
basis, McCarthy Introduces the feature [pharyngeal), whose
96
articulatory counterpart is the region between the uvula and the
larynx.. Thll ,. .tur. characteriz•• gutturals .1 having the same
constriction location.
McCarthy proposes to Incolp)rate (pharyngeal) into the model
of feature roprnent8tion of Clementi (1985)1Sagey (1986) in looh a
ny, that it reprosenll one more place f••ture, dominated by the
Place Nod., and hierarchically equivalent to the Coronal, Labial and
Dorsal Nodes.
111 the ' ...mework developed in this thesis. McCarthy's feature
0'
(pharyngeal] fits elegantly with the rut
the Site f••tur••: it
rep,...nt. a constriction location, along with the Labial, Anterior.
Palatal and Velar Sit....
that the non-high vowels and the pharyngeal consonants form a
natural ciasi.
In the remainder of this section, I review several casel that
McCarthy discusses in his paper, and I show that they can be given an
analysis consistent with the model that is being proposed in this
thesis.
3.5. 1
Vowel Lowering in Tiberian Hebrew-
McCarthy discusses the following phenomenon in Tlberian
Hebrew (segment invctntory (Prince (1975), who follows Lambdin
(1971), and Geseniul (1910»: i, u. e. o. (a). a, i:, U:. 8:. 0:, a:; P. (I). b,
(V), m, •• (8), tft, d, (&), S, $. i. s', Z, n, I, j, k, (x), g, (¥'). W, q. r, h. ~
? h: the diacritic •• • Indicates the so called ·emphatlc· or
pharyngealized consonants.): there is a group of noun stems in the
language (refer!ed to as the -segholate- nouns in Hebrew linguistic
tradition), whose surface forms alternate between the CVCVC shape
in word-final position and thA CVCC shape before a vowel-Initial
suffix, e.g., the underlying ImalkJ 'king' surfaces as Imelekr when
unsufflxed, and as Imallu (e.g... Imalk·iiJ 'my kino') when followed
by a vQwel·initial suffix_There are good reasons to believe that the
vowel which appears between the last two consonants of the stem is
epenlhetic: first, it fails to attract stress (ordinarily, if the last
syllable of a word ends in a consonant in Biblical Hebrew, it is
stressed (Rappaport (1981
and second. its quality is predictable:
it surfaces as laJ when adjacent to a guttural consonant, and as /9/
elsewhere. The follOWing are examples in which the epenthetic
vowel is adjacent to a guttural:
I
#
With the results presented in McCarthy (1989). !;~e task of
arguing for the Pharyngeal Site II virtually comp;etad. However.
since the staws of this feature in the constriction model is ratt1er
different than in the modified articulator model proposed by
McCarthy. an analylll of the -guttural- phenomena which accounts
for lhase differonces seems necessary.
The main difference betneen McCarthy's proposal and mine is
that on the constriction model, vowels articulated in the pharyngeal
region are included among the natural class characterized by the
Pharyngeal Sito. This is a consequence of the fact that in this model,
the constriction location must be represented in vowels as well as
in consonants. On this view, the class of pharyngeal sounds inctudes
? h. h. ~. ~, tt, R. G, atc., as well as a, 0, 8. ~, £-. Indeed, the
phenomena that McCarthy discusses point directly to the conclusion
#
»).
(52)(a)
Ibaht/
Ib8~1/
• RICIn from thl IntroctuctlDn thet lalll repretlntld with II Tongue Root
constrIction at the Pharyngeal Sltl. and 101 end IlJl er••nl.yzld et complox
I.gm.nt•• 1./: PelatollTongue eody. Phoryng.ellTongut Root: /0/: YelerlTongu.
3ody, Pheryng.ellTongue Root. Agsln, Uti, wey of rlDres.ntlng yow,lo will bl
In~.nd.nUy Motlvat.d In cMPter 5. howI"lr. sam. of t~1 'Dct, on thl blill of
which w~ 4hell arCJUI for thl PhonologIcal eonltltulncv of tM flatur. Pharyng.ol
Sit•• can elr .._'....... b. rtad QI IYldenc. for the.1 r'Dre'lntltionl.
97
·cost ly slone·master·
--)
-->
• I am grletful to John MCCerlhy. Alan Prince end Brien SI.tsema tor discussing
this phenom.non w I ttl me. Any .rror, In th. tnt.fpretet Ion of the dlta. and In the
onelYlis are min•.
• a/I elt.rnatlon I. dlsculled below.
98
(b)
Ibethl
--)
Ibal~1
--)
-nem. of city·
-swel1Dwtngo
Following Malone (1984), McCarthy analyzes the ale alternation in
the first syllable of the forms in (S2b) as a r.flex of a more general
assimilation rule which rail81 181 to lei when in an open syllable
and followed by lei_ He alia assum•• after Malone that there are
two processes of vowel iow.ring In the guttural environment, both
applying after the rule of epenthe.ls: one process affects vowels
alter the gutturals, and it take. place balor. the rule of vowel
raising jUl' rMntloned, the second ono foilowl the railing rule, and
affects the. vowels in front of the gutturals.
The analysis that I propose is allO based on the assumption
that there are two processec which derive low yowels In (52). First.
I analyze the procell 01 epenthesis .1 insertion of a schwa: a vowel
who.. constriction lOCAtion is unspecified. Let us represent this
process as follows:
(53)
•
R
->
le_c,
I
The -lowering- effect the guttural environment has on the
inserted vowel Is best analyzed 81 the spread of the Pharyngeal Site
of the guttural consonant onto the Constriction Node of a schwa, a
vowel which lacke the Site specification. This direction of analysis
is supported by. two facti: first, the underlying ,., does not undergo
the process of -towering-, 8.g., he • ~bl:r --) h.~.btr 'hit led across',
which suggestl that /8/ In the segholate noune is nol derived from
,., via a lowering Influence of the gutturals. Incidentally, the
repr.sentatlon which I propose for the vowel 181 rules out a
possibility of such a process:
is a complex segment with a
Tongue Body constriction' at the Palatal Site and a Tongue Roo~
constriction at the Pharyngeal Site. This makes it already a
pharyngeal vowel. The only maana for lei to become laI, is through
the d61ettQn of the palatal cf)"stricUon. Second, as discussed by
Prince (1975), p. 98, the lowering of schwa in the gutiural
environment is a more general procesl, not limited to the epenthetic
vowels. According to Prince, all schwas in the language turn to 181
in the gultural environmenr. Accordingly, I propose to represent the
process of pharyngeal spreading as follows:
I.'
(54)
Constr
I
Dorsal
I
R
R
I
I
Const~
-high
Since the schwa is [·highr, the raising of laJ to lei in the first
syllable of the segholate nouns can be attributed to the spread of the
height specification of the epenthesized vowel. (I shan postpone the
discussion of the exact mechanism which accomplishes the laJ -->
lei change until the tim. when I discuss the processes affecting the
Tongue Ebdy features (chapter 5».
• RICIn from 33.1, that on thl b0111 of Fonr. (1960) character'zatlon of lhe
schwa IJ II YOWl' with. fairly lerg. oPinIng. and on thl e•• umptlon thot the
high
tongue body palltlon II alsoclstld with I smell GP,rlur•• I proPoI.d to ropr.'lnt
It Phonologteolly II II-high) YOWl"
99
l"
Phar
Const~
'-~l
T.Body
Note that the phenomenon observed here in some ways
parallels the phenomenon of schwa palatalization in Russian and in
Margl, in section 3.3: in both cases. segments with different
Articulator specifications. but with identical Sites. trigger a
phonological change which can be treated only in terms of an
assimilation to the constriction location.
• Although, Prl~c. not•• only call. of I./lt'wlrlng aft.r e guttural. not blror•.
Glvln IImltld date, I hDY. not btan abll to d,t.rmln. whither aU Ichw lSI ar.
.fr,cled blfor. thl gutture'l.
100
n.8.
Following Malon.'. (1984) and McCarthy', (1989) insight, I
uaume thai there •• two proc..... which spread the Pharyngeal
Site: the rightward spread, responsible for the d8ta in (528). which
pI1tC9del the raising rule mentioned above. and the laftward spread,
r••panllb.. for the facti in (52b), which follows the raising rule.
This a.sumption accounts far the IIact that the epenth8sized vowel
triggers the raising of 181 in the preceding syllable only when it is
followed. not preceded by a guttural. Presumably, following the
spr••d of the Pharyngeal Site. the r.sultlng segment Is reanalyzed
.. 181: 8 -ament which lacks any height specification (s.. chapter
5). Such • ugment may not trigger tho ratslng rule.
The,. now remelna the matter of explaining why the
epenthesized schl!!a surfaces as
instead of "/: afte, all, there
have been schwas in the phonetic representation of Biblical Hebrew,
ct. Prince (1975). The answer to this question is given in Prince
(same). First, Prince obs.rvo. that there are other occurrences of
epenthesis in Ihe language: he nole. the forma like Ig. b u: II
'boundary', Iblritl 'covenanf, etc.• in which the presence of a schwa
cannot be attributed to vowel reduction. Prince concludes that the
occurrence of schwa in th... forms is duo to th. process of
epenthesis, which Indeed inserts a schwa. not lei, into a wordinitial or word-final consonant clustef. H. attributes the fact that
tho epenthetic vowel surfaces as /e' in the ·segholato· nouns to a
process which spell. oui I.' al lei in word·final closed syllables.
1.,
let us summarize the discussion of schwa pharyngealization in
Tiberia" Hebrew with a sample derivation, to illustrate the main
points of the analysis:
(55) Underlying fOim:
Gloss:
I conclude on the basis of the Tiberian Hebrew facts discussed
IlDove, that segments (both consonants and vow.I!) articulated with
• constriction in the pharyngealllar:/ngeal region of the vocal tract
p.tter~ as a natural class. These facts argue for the phonological
constituency of the '.sturs Pharyngeal Site.
3.5.2
Ablautlng Vowels in Arabic
The phenomenon that I consider next, the Arabic ablaut. has
been dlscUIAed earlier by Brame (1970) and McCarthy (1989). The
analysis proposed her. developa some of the ideal suggested in
the•• works.
In Standard Arabic (phon.lto inventory according to Maddleson
(1984): t:, U:, 8:,0:, ea:, l, 0,8; b, m, w, r, t, d, I, &, s, z, n, I, r, C, t~,
d~, ~~, 9~, z~, d!, I, J, k, g, q, X, ~, h, ~, ?, h; Isngth ignored) the
perfective and imperfective templates of the verbs in Form 1 are
CVCVC and ccve respectively. Brame (1970) proposes to derive the
CCVC template of the imperfective via a rule of vowel elisIon. He
noi.. that while the perfectlve stems (CVCVC) are always suffixed,
the imperfectives are always preceded by a prefix that ends in a
yowel. This process creates an environment for the application of
the elision rule. which he formulates as follows: V··> 0 / V +
C_CV.
The first vowel of the stem in the perfective is always laJ.
Tho root consonants remain the same between the templates. If we
adopt Brame's proposal, and derive the imperfective template via an
elision rule. we ara left with only one difference between the two
aspects: the difference in the quality of the last vowel of the stem.
On the basil of the vowel altitrnation observed in that position in the
two aspects, the verbs of Arabic are divided In~o five ablaut classes:
Imelk/
Ibahtl
IbDl~1
·k1ng·
'coSily
'swel1owtng'
mel8k
n.s.
mellk
bih&\
be'h8\
n.8.
n.D.
b81a~
ali
dlereb/d~rtb
n.8.
n.e.
b'la~
i/a
sertb/sreb
slons'
Epenthesis:
Rightward Pharyng.
a -> e
Leftward Pharyng.
101
b819~
n.8.
(56) a/u
kal8b/ktub
102
'write'
'beat'
'drink'
.,.
ulu
f.~.l/f~.1
'do'
'be stupid'
baludlblud
McCarthy \13S!) characterlz.. the ablaut cla.s•• U follows:
the occurrence 01 verbs In tho ulu ci..., and partially in the ila
cl.... II either .emantlcally or morphologically determined, the
occurrence In the all or aJu clan II entirely unpredictable, and
finally, the ala cl... conllatl 01 verbl which h8ve a guttural
conlOnant adllant to th. ablauting vowel.
The vie. of ablaut oftered by Brame (1970) I. somewhat
dlfferanl Brame too characterize. the uJu
al morphologically
circumscribed, but propos.. to treat the perfectives and
Imperfective. of the remaining clg.... al phonologically related. He
observes for example. that the height of the last Item vowel in the
imperfective il atways (outside the ulu and ala CI.Ssel, of coul.e)
opposite of the height of the lalt stem vowel in the perffICtiv8. Ha
propose. to derive the vowel of tho imperfective by a rule which
polariz.. the height of the underlying flowel .- tho Yowel which
appear. in the second position in the perfective. This proposal works
elegantly for the III class, but encounters a problem when applied to
the ali and alu verbs: if the vowel of the imperfectivG is indeed
derived from the vowel thai surfaces in the last stam position of
the perfectiv., how can liJ and luJ in the imperfective. of th~B'
verbl be predicted?
I suggest a way of looking at Arabic ablaut which captures
some of Brame's Inlightl, but derive. vowe·1 quality in alu, and ala
cia.... from the consonantal environment, and treats the 811 class
81 • default
It hal been long not!ced about Arabic, that the appearance of
181 In the Imperfective I. cont.xtually determined: 181 appear. if
one of the adjacent consonants Is • guttural. this is shown below:
(57)
eta..
Brame considers a possibility that there might be two low
vowels in the underlying inventory of Arabi~: a front vowel, which
surfaces, and the back vowel, which merges with IflJl on surface,
When polariZed, these vowels would yield Iii and lui respectively In
the imperiectiv8 forms. He rejects this possibility on the grounds
that nothing els8 in Arabic phonology requires positing I(J/ in the
underlying representation. Inst9ad, Brame propose!! to represent the
stems of the alu class with a diacritic +F. and he posits the rule oi
height polarization which associates [+round) with +F (Brame
qete\'-et
feleh-at
·shl cut·
tB-qte~-u
·she opined-
ba~e'-l!It
·shl,ent·
te-fteh-u ·shlopenste-b~88-u ·sh. s.nds·
Qere?-et
·shl read'she went·
ta-qre?-u 'she rlads·
ta-~heb-u 'she go••'
'eheb-et
-she cut.·
According to McCarthy (1989), out of 438 verbs which ablaut in laJ
(have 181 in the imperfectjve), 411 contain a guttural consonant
adjacent to the vowel. Of the remaining 25, 15 have a guUural in the
initial position.
Within the guttural class, the uvulars ' .... , and ItJ I behave with
some uncertainty: frequently, they do trigger the a-ablaut (e.g.,
l&a"X.are; a'- 'to preserve', IAerD'X.a; aJ 'to splinter', IbaHata; aJ
'come unexpectedly'). but aknost as often. they appear with the alu
class, e.g., Ifara~a: uJ 'be void', Ina~ aba: uI 'pick', /rasa)(, a: uJ 'be
stable', etc. This fact is noted by McCarthy (1989), p. 13. IqI is
virtually excluded from the guttural c~ass, in that it almost never
triggers the a-ablaut. Instead. it tends to associate with the aJu
verbl.
The phonological properties of the aJu class are not as
transparent as the properties of the ala class, yet on closer
examination, clear generalizations emerge. The results which I am
about to discuso are rather informal, based on a casual reading of
• Thll ,xampll II Quoted from Wehr'. (1971) dictionary. In which verbl 8fO entlrld
with the perfectivi form followed by the yowel of the Imp,rfectivI. Reth.f thon
tHing full Imperfectiv••• I follow the formal edogt.d In the dlctiDnory.
(1970), p. 157).
103
104
Wehr's (1971) dictionary. I have compiled a lilt of a little over 800
tri-litoral verba that belong to either aJu or all cl.... guided by the
main entries in the dictionary. Roughly. about five hundred of the.e
a,. aJu verba. The pre..nee of the yowel lui in the imperfective is
usually a••ociated with an adjacent uvular. velar, or labial
consonant. I found 44 cu.. (not counting the ca... with Ir/; .ee
below) in which lui appears without such an environment; of th••••
29 root. have • velar. uvular. or a latH.' as their first consonant.
The.. ara Nlm8 of the ••amp.. of alu verb., contrasted. with th.
••ample. of all verba:
(58)(8)
aJu:
netjeme; u/t
fattlre; e/u
·open-
na~1I18;
'sift·
u
ml'X.ete; e/u
'hum'
-blow nos,-
neql&e; u
-dIIlYlr'
,eqele; u
'try'
nekese: u
sekene; u
-turn eround-be stilt'
nad!eda; u
~ed~ema;
(b)
The pre••nce of lui seems to be required especially when any
of the•• consonant. pr.ced. the ablauting vowel: for example, out
of 124 alu and ali radical. with either N. IgJ (Which surfaces as
Idfl in Standard Arabic, cf. Breme (1970), McCarthy (1989), IqI,
rx./, or /HI in the s&CQnd position, 108 ablaut in lui_ And out of 125
aJu and all radicals with Ib/ or ImJ in the second position, 99 have
the ablaut vowel luI. In the third posatan. the velar and uvular
consonants stili have a strong Infiuence In determining the quality
of the ablaut vowel (out of 99 stems, 78 have lui in the
imper'ectlve); the influence of the third position lablals is
perceived In about SO% of the C8881.
u
~atemlJ; I
nedefa; I
hszame; t
hesame; I
he$tJr8; I/u
he\abe; I
heieme; I
,sIebe; I
-help-
-llsr
-hesitate'
-tees.-wrelf
'sev,r"
-surround-galher firewood"
-dsslroy-cross arms-
10.5
Whll. the numbers quoted above could hardly be a result of a
free and unrestricted distribution of segmants, it is also far from
obvioul that we are dealing here with a productive phonological
procell_ After aU, there exist a considerable n"-nber of exceptions
to every generalization stated above_ A plausible explanation for
this situation might be that the phenomenon which we observe in
Standard Arabic is a reflex of a phonological process possibly
productive in the past. but becoming lexicaUzed in the modern
language, or els8, an ongoing process. For one thing, a large number
of doublets (188 in my sample of ali and alu verbs) suggests that
Arabic ablaut is in a stato of transition_ A comparative hastorical
analysjl might reveal the direction of the change. However, for the
purpose of the present discussion t the exact history of the
phenomenon is not as important as is its mechanism, and this is
what I tUfn to nexl.
In order to account for the fact that the vowel faJ of the
perfective surfaces 81 either laI, luI, or Ii! in the imperfective, I
would like to suggest the following: the st~in-final vowel 181 is
pruned of ils articulatory features, the Articula~or and the Site, in
the imperfective. In fact, we may assume that this vowel is a
morphological marker of the imperfective. This aspect of the
analysis is crucial in that no other lars of the language are affected
in the way this vowel is_
106
The '.Iuliing segm.nt ...imU.t.. in Site to an adJecent
consonant: labial, velar, uvular, pharyngeal. or IaryngeaA. In the cue
of • labial, it
lmU.18. to the Labial Site, producing lui, in the
cu. of the v
- to the V.lar Site, and the output il lui. A uvu~ar
may Ipr••d .'ther the Vela, or the Pharyngeal Site (on the
...umption th.t uvul.... are complex I egmen ts. ct. McCarthy (1989),
.nd the dllcussion in chapt., 8), producing either lui or 181
,espectively. arid tM remaining gutturall may apr.1Id the Pharyngeal
Site, producing Jal.
When an •••imllatlon fslll to apply, the unlpeciflad vowel
IUrfacn with the default melody III. Tht. app..... 10 be the cetrrect
way of viewing the &15 cia.. (rath., than, tor example. attributing
IU to an ailimnaticn in the constriction location of the coronal
conlOnantl) for the following r881Ons: first. III I, the melody of the
• p....th.tic Yowol in Arabic: second tho largest number of exceptions
10 th. distributional generallzatlonl stated above il found among
the all V.rbl. Thl. is belt explained .1 an application of a default
rule applying when. for lome reason. th. phonological rule faile to
apply.
The above an.'YIII account. for the distributional
generaUzations sketched out above. and offers a plausible
.xplanatJon for the fact that uvulars can PBUem with either velars
or pharyngeallilaryngeali. Unfortunately, it give. us no clue as to
Why the fricative uvular. ('X, and IH I) do pattern both ways, but
IqI patterns with v.Jars only. I leava this question unanswered.
0'
During the discussion
the environments which condition the
ab!aut, I have left out of the consideration Ir/. the segment which in
Standard Arabic surfaces as an alveolar trill (Brame (1970». In
about two thirds of the verbs. /rl patterns with the consonants
which are associ.tod with the aJu class. Because of the frequent
occurrence of Irl in Arabic, thie makes for CD significant number of
cases (59) which appear to violate the generalization laid out above,
and makes Irl the only coronal consonant which regularly triggers
the lui ablaut. This. of course. is not the only problem caused by Ir/,
107
becau.. thor. il .'10 a large number of Irfs which do not trigger
the lui abalut. Ther.fore, •• are faced with the task of explaining
why some token. of a given sound trigger certain phonological
behavior (...umlng that the above generaUzatlona are to be tr•• ted
In phonological terms), and other. do not.
A Ilmllar situation arllel
W.f.t.
'",
also left out of the
conlldel1lUOn so fill: IVs divide evenly among the alu and ali C818••
Conceivably. this behavior could be attributed to the fact that Ipi
merged with IfI in the history of Arabic (Brame (1970». It could be
hypoth.slzed that the preunt situation r.flects that moment In the
history of the language when both segments coexisted. and when the
rule of ablaut distinguished between the segments with the labial
constriction location (/p/, Ibl. 1mJ), and other segments (If! among
them) .
Extendlnr; the same approach to the Irl cases, plausibly, the
modem day Irl could be derived from a historical ,., (8 uvular trill)
(such • possibility is not ruled out by the semiUcistl, cf. Moscatl
(1984), p. 32r. With this hypothesis. it Is reasonable to assume a
period in the history of the language when the ablaut rule had to
distinguish between the two repre.entations. The casel of modern
Irl appearing in the eJu vems could be attributed to the spreading of
the I.fs velar component, and the cases in which Irl appears in the
ali verbs. could be derived from an Underlying Ir/. In the laUer. the
the melody of the ablaut vowel would be derived via a default rule.
Lei us conclude the discussion of Arabic ablaut with this
comment: the primary importance of the ablaut phenomena, and the
reslon for considering it here, lies in the fact that it groups the
guttural consonants (modulo the behavior of IqI) into a natural
class, thus providing an argument for the feature (pharyngeal)
proposed by McCarthy (1989). I have extended the discussion to
o
AlthOugh John McCarthy (p.e.)
suggl,ts that more IIk.hj. the ,., of Hllbriw
dl",loDld from the common louree/r/. On thl InlorD".tation I am sugg.stlng, ,.,
would bl tho reconstruct.d •• gmlnt fOf thl Semitic longuag••• ond Irl would b. a
new dI'ltlopmlnl In ArabIc.
108
0'
Include the rut
the paradigm In order to show that McCarthy's
'eatur. (pharyng••q PAttern. iog_ther, or at '''It did patter" at
101M point in the history of Arabic. with other tuture. which
to the ~.triction location: Labial and Velar in this ca... Ther.for.,
the ablaut In ArabIc provides
for the statuI of (pharyngeal]
•• another site '.lure In the model of phonological r&prHentatlon
propoud In this thell•.
r.f.,
ltV.....
3.5.3
Vowel lowering Belore UvuIMl In GlMftlandic Eskimo
I now tum to vowel lowering pMnorMna of G....n.ndic eskimo
(I8Om.,.t l~entory aft., MaddlelOn (1984): I, u•••te). (0): p, It" m, f:,
t, t:, n, I, tl, ., i, jo, k, x:,
I), q, ~:, .) dlscuI.1et ••rller by
Kenltowlcz and KiII.blrth (1979). In W.st Greenlandic. only nonhigh VOwell .... found before uvulara IQ/. rX,I, and IH/:
r.
(59)
•••mlq
·glecter'
nlRdloq
·goo••·
ts••eq
'enkl.'
·bark'
IpIRDq
nenoQ
'harpoon strap'
'bier·
Q81eloq
No other consonant in the language hilS a lowering effecl on the
vewe'l:
(60)
nune
'lend'
ugSlk
Igdl0·
·cow'
l~neQ
'.Ibow'
'bluff'
'house'
Iga
'par
IkuSlk
An anAlysis of the above phenomenon within the articulator
model is tochnically possible, under the assumption originally
sugg~ted In SPE, that uvular consonants and vowels lei, 101, an dial
share the feature [-high]. In fact. Kenstowicz and Klsseberth's
analysis of the Gr.enlandlc Eskimo data follows this avenue.
However, as pointed out by McCarthy (1989). uvulars are definitely
not (-high] articulatorily·wis8.. In fact, if these segments were to be
characterized in terms of height, and if the height featuro were to
have any phonetic content, then, given the articulatory description
of thes. segmentl in McCarthy (1989), thsy could only be
characterized 8S (perhaps redundantly) (+hlgh).
The constriction model affords a slmp'e and elegant account of
vow.1 low.ring in Gr8enlandlc Eskimo: Flrlt, following McCarty's
Bugge.tlon, I a.lume that uvular. are complex legments.
Incorporated Into the constriction model, this sugge.tlon tranalatel
81 folioWI: uvular. have a Tongue Body constriction at the Velar
Site, and a Tongue Root constriction at the Pharynge8J Site. The
Tongue RooUPharynoeal constriction spreads onto thtft Root Node of
the procedlng vowel In the manno, luggnted below:
(81 )
R
R
-ca;;l- - - - _
c
A
- .. - _ C
(C)
A
Pheryn. T. Root
The
output of this rule is a complex vowel:
palataUpharyng••' leI, or a velaritJharyngsal 10/.
either
a
A procesl that is a mirror image of the one in Greenlandic,
occurs in Chemehuevi (inyentory according to Press (1980): I, +, u, 0,
e; the saml .Irlel with length; p, PlY, m m;, w, w', t, s, ts, r, n, n',
J, J', k, ¥, I), ,.', k", r", 1)7): before 101 or laI, a voiceless velar turns
into a uvular: tHea + IJU + eke --) ttkar)u8Q 'that you ear (the velar
nasal remains unaffected). Again, an analysis of this phenomonon ~~
straightforward within the constriction model: on this model, all
non-high vowels are represented with a pharyngeal constriction. In
Chemehuevi, the pharyngeal constriction spreads onto a voiceless
velar: the resulting segment is a consonant with a Velar/Tongue
Body and PharyngaaVTongue Root articulations, in other words. a
uvular.
The relevance of the above phenomena to the argument made in
this section is clear: since the non-high vowels are represented with
109
110
4.
• pharyng.al component In the constriction model, they are assumed
to form a natural ct••• with the guttural consonantl. Ther.fore,
proce.... such a. vowe. lowering before uvulars in Greenlandic, or
pharyngeallzatlon
IkJ before non-high vowell in ChemehuBvi
constitute support for the feature Pharyngeal Sita, because they
show that the non-high vowels and the gutturals pattern alike.
0'
Site and Articulator Within the
PhonologlC8~
Tree: The Constriction Node
Having
established
the
existence of bo~h Site and
the
Articulator featur.l, W8 are now faced with the task of
Incorporating these feature. Into the theory of phonological
representation. AI 'a, 81 the larger picture is concerned, I adopt the
theoretical framawork Introduced by Clemants (1985) and Sagey
(1988), whOle success in the recent years testifies to !ts besic
correctn.... I adopt such eram.nt. of the ClementslSagey structure
a. the Root Node and the Laryngeal Nod., and I assume along with
S8gey (1986) that the manner featur•• are directly dominated by the
Root Node. Finally, I follow Piggott (1987) and McCarthy (1988) in
including [nasa'J among the features dominated by the Root Node.
With the above. only the articulatory features remain left out of the
segmental tree.
Throughout this presentation, I have been assuming without
argument a structure in which the Site and the Articulator features
are dominated by a constituent which I have named the ·Constriction
Node-, and which I have represented
as repeated ~-e!ow:
as
a dependent of the Root Node,
.
(82)
Root Node
.,
llr1l1Ol11
Feetl.#es
"
(ConstrtctlanJ
~
tiM".,
FBobrH
COftstrttUon
~
Sit,
ArtitulDtor
Intuitively, the constituoncy of such a node is easy to accept. CiS it
represents the articulatory gQslure necessary in the production of a
speech sound. The physical parameters of this gesture are indeed the
11I
Ii2
lite af the maximum narrowing and the orgG" which execute. ir.
Howe¥Of', from the phonological point of view, the noed for such a
constituent ramain. to btt e.tAblished. The pr.s;tnt s...ctlon is
devoted to this goal.
Granted that we do need f••tur.. which represent the location
and the articulator 5Ctiv8 in the constriction, we ara 'eft with at
"ul two options other than (82) fer representing these features:
we ~n assume thai both Site and the Art~ulator are direct!)'
dominated by the Root Node: or
can pOsit 8 constituent equiva~,nt
to the Place Node of the ClementI/sago, mode', dominating a!1
Articulators and .11 Sites with no other node Inti'lYening. Til9 two
options are repre..ntecl in (63a) and (63b) r••pecfr...ly:
w.
(83)(8)
(b)
R
LrtII.
~r"l'.
Nt
Sill
Hlmr
F••l....
R
,~
t1Iftfter
P11CG
lIryn.
~l"
Art 1 Art 2
SU,! G1t82
assimilations-, which involve spreading of the Site and the
Articulator together (e.g.; nasal assimilation to a following
obstruent) .
The model in (63b) cannot be dismissed as eaeil}·, therefore. it
is discusa6d at gr••te, length in the next soctitJn.
4.1
Th8 Constriction Model VI. The Model with a Place Node
Dominating Sites and Articulators
When ~8gments with single articulations are considered, the
model In (83b) r.nd~r8 point of articulation asslmi;ations 8S natural
as
the constriction model: it tre8t5 thes. assimilations in
terml of the Piace N~e spreading. while the constriction model
tr.ats them as the spread of the Constriction Node.
do..
The difference, between the two represer.tations baeams
aparent in the case l" processes that involve complex segments.
Consider. for example a complex segment such as ItJ/: the model in
(G3b) assigns the following representation to it (the lary-ngeal and
manner features are omitted):
(64)
(The tree in (~b) is designed to fapresent a complex segment on
Pl;!!~i.: 21 we shtJI seG, on!)' when the complex sngments are
considered. will the difference between the model which assi'i"".~ a
separate node to every articulation (-constriction), and the model in
which all articu;ations are comb!ned under ono node, become
evident.)
Let us now CJnsider the above proposals in turn. The model in
(fl3a) is a transparent straw J"'1an: given that·, non-constituent
spreading is not allowed tct St9riade (1987», this .nodel makes
equally impossible procassa:tl which spread. say, an Articulator and
the laryngea' fealures~ and the so cal:ed ·point of articulation
• AI won ftl the eosrl16l. but WI de not cdr~s. lh& OhO"ologICO! Ititul 01 lno
strlelw-o f,.lurl'. nor ,~.lr DICitl In tha reprl,onlsll0n or ••• ;ment In lhll
thl1:I:.
1J3
~
t J:
I
Piece
MI.r~ T. !lcxt.I
T. Bl,d.
f'elel.J
Since thia mode! does not preserve the distinctness of the
Anterior/Tongue Blade and PalataVTongue Body articulations, it
predicts, given the assumption that only constiluenls may be spread
by phonoJogicai rules, 'hat the processes which spread t~elher the
nodes Palata! and Tongue Body of /t J, should be as unusual as, say.
!h, processes which sproad together th9 nodes Palatal and Tongue
tada 01 IlI/. ~"hile I know of no example of a process converting a
segment to
I before It J I. processas in which one of the
Ie
114
artiouIationa in a camplex segment spreads are not uncommon. Let
UI consider an example of such a process, involving palatalized
segments in Polish.
This rule must precede the spread of palatalization. because the
sibilant which precede. the Linderlylngly palatal segment also
surfacel unpalatalized:
4.1.1
(88)
Palatalization Spread In Polish
Polish hal • very "'8U molly.ted rul. which palatalizes
Anterior sibilants (lsi. IzI) before palatalized .egrnents:
moic.-
most
(85)
sne
Inu
IPSC
'pJt
organtzmJI
eorgentzm
-brtdg.-dream·to sle.p/h. 11•• p.·
-organism·
i~ followed by a front yowel: with an exception of
separated by the prefix juncturo (which ars Immune to
mOlt phonological procell.. In Polish), sequence. of lsi (Iv)
followed by a paJatalized segment are virtuldly l'xeluded from the
language_ Consider the following exampl••:
kOlt-ke
mltoltlce
but
bul
-bon,-lovl/fUrt·
It Is not the calO that all stem-final sequence. of pal..tallzed
segments are d.palaU:Uzed before a consonant·lnitial lufflx: In the
case 01 und.rl,lng sequences of
(IiI) followed by a palatalized
:iigment, only the lalt of the lequence II affected by
depalatallzation:
'I'
(89)
It is not the case that palatalization spreads onlo • sibilant only if
kOle
mltolc
vlan-e
tein-e
yts In-k-e-
but
but
teien-k-e
·sour cherry-belh-
the clustlrr
leq~8ncel
(86)
scene
SfiJeY8C
boJeiri
mlfosc
/'
-wen-to sing-
sm'lch
-feer-
kose
-'ov,-
kein
' ,
vesn
-laughter·
·tortur.-bon.·
-querrsr
At 'east for some Dr these cases. the fad that palatalization of /9 i
must be d.riv~ via a spreading rul9 can be proven: Polish has a
depalatalizaUon rule. which affects a stem-final palatalized
consonant before a consonant-initial suffix. for example:
(67)
hreb'-eI
Sin
hrep-sk-I
son-k-e
-count·onteroom-
In yiew of the aboYe, we must conclude that there is a rule in Polish
which sproads a palatal constriction
a paJatallzed consonant.
0'
The constriction model predicts that such processes should
occur naturally: on this model. the palatalization spread in Polish is
analyzed 81 the spread of the PaiatallTongu9 Body constriction. On
ths model in which all articulatory properties of a segment are
combined under on8 node, this analysis is not available_
Of course. it is possible to revise the model in (63b) so as to
make it capable
handling processes which spread one of multiple
articulations: this result can be achieved with indexing devices. and
with constraints which disaUow the spreading of nodes with
different indices_ HowevIJr. since the only roJa of such devices is 10
capture the generalizations that are delivered naturally by the
constriction model, their stipulal(Jry nature !s all too obvious.
0'
Another difforence between the constriction model and tho
representation in (83b) can be seen in the way the two handle
processes which spread the Articulator or the Site features
• Recln from 3.2.3 that Iii. Iii. lei_ ete.• art curfee. mon!f ••tatlonl of
'r/. Ir/. and IIJ! rl'PletlYI'Y.
uncIerlvlng
lIS
• ,,'II ,p,nlh.tlc. '" Gorlcke (19B8).
1 16
t_
0'
of each ot'1er. W. have already seen a number
casel
which ..... bttlt analyzed In terms of Site spreading: each time
howev.r, there were Good ,ealOna for tr••ting such procH.e. as
f••tur.-'ililnO r applying to .egmentl that lacked the Site
specification.
It would appear that only 'eature-rilling spreading of either
Site or ArtIculatar occurs. For In.tancet I have never s••n an
example of a rule converting a coronal paJatal into a dorsal palatal
before any vowel. Or, If the direction of this rule il marked .- Its
converse: a rule wh~ derivH ~ coronal palatal from a dorsal
palatal be':»r. any comnal sound (e.g., C -> C 1 _ t, S, etc.). I have
not seen a rule which would COI"iY8rt Ifl Into lsi before a coronal, or
• rule _thlch would change a coronal palatal into a dentaUalveolar
before il" lsi" It/, InI, Irl -- segments which are charaeterizctd BS
Anterior.
The constriction model predicts that feature-changing
a••imilations involving the Site or th, Articulator are impossible.
This prediction faUI out of the assumption that the Constriction
Node represents a real gesture which accompanies sound pr~!..~jon;
by all physical evidence avaUabkt. such a g.sture alwaya involves
one articulator and one area of obstruction for every articulation
known. To put this in phonok»gical terml - we kilOw of no phonemic
contrast which couki be attributed to the fact that in one segment.
the active articulator executes a constriction ~n one area of the
vocal tract. but in another. the same articulator participates in two
constrictions. On these grounds, I hays proposed to represent the
Constriction Node canonically as a function of two variables: the
Site and the Articulator.
Given that the Constriction Node may not dominate more than
one Site and one Articulator, and given the assumption that there are
no phonological processes which derive universally impossible
representations, the constriction model predicts the absence of
phonological processes which spread the Site or the Articulator to
replace an existing one.
I 17
The model in (83b) cannot dellvsr the same result: the Place
Node of thl. model may dominate multiple Sit.. and Articulators,
therefore. nothing should prevent the spread of an additional Sit. or
Articulator onto this node. Also. ;t does not appear to be possible to
correct this medel with a constraint that would block the spread of
a single Site or a single Articulator. beacause such a constraint
~IQuld rule out fealure-filiing applications of such processes 81
W'-':!.
The above considerations constitute suffk:lent grounds for not
regarding the tree in (83b) .s a Ilnoua candltate for a model of
phonologlca' representation. At the same time, the arguments
pr.sented in this lectlon provide strong support for the conclusion
that an adequate phonological theory must recognize the
constituency of a node which represents a single articulation: the
Constriction Node. This node must be defined In terms of the Site
and the Articulator - featuras which map onto the physical
parameter. of an articulation: the location of the narrowing and the
active artlcu'ator respectively.
E~8n with the status of the Constriction Node established, we
are stili left with at least two questions about the structure of the
segmental tree unanswered. The first question concerns the status
of the Place Node in the sense of Clements (1985) and Sagey {1986}:
in those models, the Place Node combines all articulations of a
segment; the question that I want to pose is whether such a
constituent, distinct from the Root Node, is at all motivated by the
phonological facts, or whether. with an independently motivated
Constriction Node, the Place Node can be disposed of entirely. Below.
I argue that the Place Node is not reqUired.
The second question that remains to be answered is that of the
structure within the Constriction Node. So far, I have been assuming
that the Site and the Articulator are free agents. independently
118
aUlpended 'rom the Conltrictlon Node. However, at ,1••lt two other
arrangarnentl ar. paaalb..: OM, In which the Articulator dominate.
the Site, and ia ItIetf domlncted by the Constriction Node, and one in
which the order of the Site and the Articulator is reversed. Har., I
win argue that the pr.f.rred .rrangement of the Site and the
Articulator under the Constriction N~. il on. in which the two
constituents .... dependent on the Constriction Node, but IOOepender,!
of NCh other.
Let
4.2
UI
now take up each of these quatian. in turn.
The Status of the Place Node
Sag., (1988) presents evidence for a phonological constituent
which combines all articulations of a segment: the Place Node. The
8vkSence consists of casel of nasal assimilations in which the nasal
acquire.
an
articulations of an adjacent segment. Sagey discusses
various examples of such phenomena In ~II., Yoruba, and Dan. Her.,
we look at the data from KpeU., where a syllabic nasal alslmilates
to a consonant on the right. according to the following paradigm:
(70)
N
•
pOlu
N + tie
N + k~~
N + k'tq
--)
--)
--)
--)
mbolu
"dte
,
F)g:)~
~·!ttl"j
-my beck·
·my taboo·
-my foor
·myself-
Sagey argues that since the syllabic nasal in Kpene and the
consonant that follows it must be represented on separate X-slots,
the only means of accounting for the fact that the nasal and the
consonant are arliculatorily identical, Js !o pOiit an assJml!atory
process in the language (thus ruling out. for example. a segmental
fusion analysis). She points out that to express such a process in
terms of a rule spreBding tho articulators: Labial" Coronal and
Dorsal, would be to miss tho generalization that all articulatory
features are involved in it. Sage)' concludes that the most insightful
way of analyzing the Kpelle facts, is in terms of a procesl which
3Pfeads a node that combines aU articulstionl of a segment.
I 19
In the Clements/Sage, model, an articulator (labial, Coronal,
Dorsal) rS:Jrsaents what in the constriction model is represented !b Y
the Constriction Node: a slngl. articulation. On that model, it is
impossible to fe'.r nan-specifically to a singl. articulation without
referring to aU articulations of a segment, i.8., without r.ferring I to
the Place Node. On the constriction model, on the other hand, .n
I
Instruction to spread the Constriction Node will result In the
•
I
spr••dlng of all qualifying nod.,. Thlrefor., on thlt model, II I. not
I
neceuary to .rect a node which combines all articulatory propertJrl
of a segment, the Place Nod., in order to account for the proceasrs
which spread mor. than one articulation In a segment. Since I know
I
of no other motivation r"r positing the Place Node, I conclude the
argument her•.
4.3
The Structure Within the Constriction Node
t'8
In this section, I present evidence that the Site and
Articulator are not hierarchcally dependant on one another, but
instead. are sisters under the Constriction Node.
I
We have already S8tjn arguments for not wanting to reprasent
the Site al emcedded
the segmental tree; recall the p,jiltalizati6n
I
of a schwa In Russian and in Margl (Sectk»ns 3.3.1 and 3.35, and t~e
ablaut facts of Arabic: in all these casel, the preferred analysis was
in terms of a rule spreading the Site independently of the
Articulator. I teke these analys8s to impl)' that Phonologlcb,
processes can access the Site features without involving the
I
Articulator features: th!s is possible only if the Site do.. "pt
CSominate the Articulator.
I
'rl
Let us now consider a hypothesis that Articulator dominates
I
the Sito, and is itself dominated by the Constriction Nods. Such an
arrangement would be consistent with the phonological facts whiJh
I
never required that the Articulator be manipulated independently of
I
the Site. Later on. we shall consider some analyses which w~1J
I
appeal to the possibility that the Articulator may spread
independently of the Site. However, the most persuasive evidenco rbr
120
not embttdding the Articulator Node above the Site within the
_-omental tr... com.. fram the phenomena In which It Is
advantaglJOu: to treat segments specified for the Site 'eatur••, but
lacking the Articulator. Apparently, such phenomena cannot be
tr••ted within 8 model In which tha Articulator II emtedded
between The Conatriction Node and the Site, because, on such •
model. it would not be pollible to specify the Site of a legment
without rep,..entlng the Articulator of thle segment.
Earlier. I have suggested that, taking advantage of the theory
underapecilicatlon
Sterl'" (1987), we may repr•••nt velar
segmenta, which are predictably dorsal (no other organ may form a
constriction in the velar region) with the Articulator node
underlying!y unspecified. Her., I show that there are considerable
benefits
such a treatment of velarl.
0'
••gmentl. I argue in that chapter that fronted velar. have the
following surface repr.lentatlon:
(71)
k':
I
c
A
Veler
Q'
0'
One of the puzzling aspect. of the phonological behavior of
velar consonants is the readlnell with which they .ssimllate to the
frontne.s of an adjacant vowel. Within the SPE phonology, the
fronting of the velars is trealed 81 a result of palatalization. What
makes this analysis problematic, is the fact that in moat languages,
only velarl are subject to sLJCh a treatment: if the fronting eftecl is
Indeed due to the ,S"ead of palatal constriction. there is no reason
why this should be the cale. t:~-~n more curious is the frequent
ab.ence of -true- palatalization offects in the environments in
which v.lars am fronted, in languages which do have palatalization.
For example. in Polish. lebials and donlals are palatalizeci before
mOlt suffixes which start with a front vowel, e.g., kat + e --> kat e
'eal', tap + 8 -> .apJ e 'paw, etc. But, as mentioned earlier, there are
front yowGI-initial suffixel which do not trigge~ palatalization:
nevertheless. the.e suffixes do have tha fronting effect on the
veIarI. e.g., kat + em -> kotem 'caf, xtop .,. em --~ xtolllm 'peasant',
but krok + em -> krok'em 'pace', rak + em -> rak'em 'crayfish', etc.
In Chapter 6, I bring u~ this and other phenomenm in which the
fronted vsJars (/k'l. Ig'I, etc.), do not pattern with the palatalizod
121
R
Tongue Body
I
-bDCk
Given the repre.entation in (71), thG fronting of velars before Iii,
'm/, etc., cannot be attributed to palatalization -- there is no
palatal component in that representation.
lei,
An adequate account of velar fronting should be able to 8xpla~n
the mechanism of the process and capture the fact that only velars
are fronted before front yowell. Below, I suggest that an analysis in
terml of the spreading of the Tongue Body articulator of an adjacent
vowel onto the Constriction Node of the velar which lacks the
Articulator specification does meet this requirement.
On the assumption that the value of an Articulator is always·
unspecified in velars, we can derive velar fronting via a
procesl that automatically spreads the articulator of an adjacent
segment, 8S the means of providing the missing part of a
representation: probably, not unlike the mechanism responsible for
the compensatory lengthening phenomena. Since only the Tongue Body
an form a constricticn at the velar region, only the dorsal segments
und8r~yingly
(either the non-low vowels or the dorsal palatals) may be the
• Vlttuaty every grammar that lists posinO"" allophone. of tho bale Hgments.
chlflderlzes vetarl .. front before front WId bactt belor. back VOW&IS. OC:casbnaUy,
Ihough, alanguaoe may have • v." IOOOd IdIoIyncratically specified as e~ + orback. ih!I miGht bathe caM in S~ PoIshr wher'l epp8aralO bfJ (+bIIc*) In IhI
URI BIll has the nma (+bac:kl aJIophont betor. thl franl and back vowelS.
122
apr.....
~
This
:s
consistent with the •••umption that phonological
5.
which create ungramrMllcal r......ntatioRl do nat e.ilt.
go..
This analYlls
through only If the articulator valu. is
unspecified in the v••r conlOnantl~
given the fact thai
phonologc.'Iy. QIarI beh.v. ••• doraJ segments with 110 tongue
bod)" f••turn. the •••Imitation of
to the balclcn... of an
.tjlant vowel would h8ve to be i'rMted . . • phonological proc•••.
In order to trigg., such •
unlv1Nully. •• would have to
posit • unlver.aI condition on .-.pr...ntalion which would make
vowetc and velar consonants IQI'H In bicknell. Thli il JUlt the sort
th. condition on repr•••nmtlon that do.. nol sound very
convincingly when applied universally. If. on the other hand, vel.ra
ar. assumed to have no artk:ulator specification. the only universal
condition needed to account for the fronting effect il that all
segments btt fully specified on surface.
Otherwt...
we....
proce..
0'
To conclude the above .rgument. the nsed for unspecified
articulator in a veiar sagment point. up the Advantage. of a
representation in which the Site and the Articulator Node. are
;ndttpendently dominated by the Conltrlction t-kKte. If the Articulator
domin.~ the Site. leaving the Articulator value unspecified would
entail leaving out the Site value al well. This would make velars
comple'ely unspecified segmentl. open to just about any
assimilation, not just .n assimilation to the tongue body features.
Segment Repr.sentation: Vow.11
Problems of vo.... representation are lome of th·. mOl'
camJ:le. in phonetics and phonology. First. becaUM of !he nature of
the org.n. Involved in vo~.1 production. there ar. dltflcuill.. In
determining the parameler. of vowel articulation. Given the fact
that vowell rotaln their auditory Identity throughout varying
environment., and from speaker to _peaker r at least 101M upects of
vowel articulation should remain constant. However, In the ongoing
d.bat~. there is little agreement al to which ••pects of vowel
pronunciation are Itable enough to qualify 8. artlculltory
parameters. According to some, vowell should be charscte,ized in
terml of the tongue body potltton: others appeal to the constriction
location al the articulatory parameter (Ie. Chapter 1 for
~;scus.lon).
For the theory of phonological representation which posits
features based on articulatory properties of sounds. the problem of
articulatory accuracy is also a matter of concern.
Another major problem for the theory of vowel repreaent&tion,
not present on the same scale in the phonology of consonants. comes
from the variation in the behavior of seemingly idflnticaI segments.
It is not uncommon that seemingly identical vO~8ls do not behave
alike w.r.!. seemingly identical pt1onological processes. For example.
in some languages, the vowel /81 is neutral w.r.t. a height harmony,
but In others, it actively participates in an identical procesi. Or, in
one language the vowel Iii patterns with palatals (f, I, C, ". etc.),
but in others, it patterns with the anterior consonants Ct. s. ts, etc.).
In the early years of generative phonology. when rulel were
expected to carry the ,,-..ain burden of crossllnguistic variation among
phonological phenomena. idiosyncrasies of the sort pointed out abovt
were of lesser concern. However, recent developments in phonology
have shown that many constraints on phonological rules are best
derived from the structure inherent in the repr8sentations to which
124
123
the ruee. apply. II il reasonable to hypothHlze that the sam. line of
rMMrCh can be profitable in IOIvIng the prabIom described above.
Finally, an outstanding problem for a theory of vawe'
r.prenntatlon ar. wowellconsonant InltwllCllon phenornen8. In an
Slttempl to IOIYo thil problem. Ct!omlky and Halle (1981) h.ve
propoled to ch8l8Cterize contOn_. which cb InterEt with vowell
In I8rmI of the tongue body fee..... For example. SPE nata If/·
like ~ndI as I+htgh. -bEkI. veIara U I+hlgh. +bllckJ, uvularl •• [high. +back). phary.... as 1+1ow1. etc. However. II hal been shown
(... ~ dllCUUion In chapter 3) that IhII proposal
descriptively
incorrect (a" leat In the cue of Hgmentl luch a. uvulars). and
phonologically Inadequate. For ex""', ..... .,. no known case. of
lei blocking or triggering tt:o height or the backn... harmony, 1kJ Is
a. •
tran.~rent to all do....1
and pharynge.l. are
not even arflculated with the tongue body. Given that the SPE or
SPE-like propo.a'i (such .1 Sag.y', (1988» cannot explain
Yowellconsonant interaction, the.. phenomena continue to be a
problem for the pr...nt day phonological theory.
'I
rut.
h.rmon....
addre..
In thl. chapte,. I
the problem* pointed out above in
some demit. S~nce I propose to characterize vowels in terms of both
the active articulator, and ~h. constriction location, I .r;aue that
both theories
vowel articulation mentionect above are .,s8ntially
correct. and I try to d••1 with the criticisms against them. For
obvioul r••sonl, , devote a greet de.~ more tim. defending the vlow
that vowel. can and should be represented In terms
tne
constriction location. Aft.r all. the vie. that articuiators are
imponant in the repre••nt8t1on of vowels il more or .8. slandard.
and needs littl. support.
0'
0'
I also suggest ways of solving some of the problems of
variation in the phonological behavio, of vowels by appealing to the
differences in the representation
seemingly identical segments.
It turns out that the problem of articulatory accuracy. and the
problem of variation are often surface manifestations of the same
phenomenon: often the ca.e. i~ which an IV·like vowel is observed
0'
125
to have • constriction In the antertor Instead of the pal.tal region
(cont,.ry to the predlctlonl of the theory which characterlz..
yowell In terml of the constriction location). t"rn out to be the
S8m. •• the ca... in which an IV·like vowel behlv.. phoneloglcall,
like an anterior segment. Thll il pouible to observe :n langu.g.. in
which the phonological propertl.. of an artlculatarlly -mllbehavtngvowel ar. reflected In the phonological prOCH....
Finally. the constriction modol offers a solution to the
problem of yowoUconlonant interaction phenomena: on this model.
an articulatory f.atur.. are equally acc•• slble ta vowels and
eanson.nla. This ha. nevo, been the ca.. In any of the ••,lle,
propo.als ... neither the IPA nor the SPE fe.tur. system. have
allawed to characterize vowels in terms of point of
.rtlculatlon/constrlctlo:l location f••tures.
This chapte, is organized al fellows: sections 5.1, 5.2, and 5.3
are devoted to Sit. f••ture. In Yowels •• the phonological
arguments In st!Ction 5.2 are a review of the arguments for the
constituency of Site feature., pr••ented in chapter 3. Sections 5.1
and 5.3 deal with the articulatory aspect of Site features: In 5.1, I
reviltw thl arguments from phonetics for the view that treatment of
vowell In terms of the constriction location is articulatorily sound.
In S.3 I present arguments that the Site features I posit colncid,
with the regions of articulatory stability in vowels. I also defend
the ~r.atm.nt of vowels in terml of the constriction location from
the criticism that It artlculatorily unreliable.
I then discuss the status of the Articulator in vowols, and
present evidence that the Articulator features are required in vowel
representation. in additfon to the rich system of Site features. Next.
I argue for the Constriction Node In yowels on the basis
phonological processes which can be understood only In terms of
Constriction assimilation. Flnlllly, various Issues concerning
individual segments are discussed.
0'
5.1
The Articulatory Basis of Site Features in Vowels: Part 1
126
(1928), etc., which
rch.,.
i..
Of particular interest to the present discussion is the proposal
by Sidney Wood (1979, 1982), whO canttrrna the relults of Meyer,
Ru•••n, etc.. in a larg. scale eros.llnauistlc study of vowel
articulation. Woad consid.... the discrepancy betw..., the tangue
arch model and the phon~ reality to be serious enOUGh to render
the mod" u••le.l. Instead, he pra90... a return ta the Sanskrit
grammarians' tradition of characterizing yowet articulation in
terms of the conl1J1ctton location. He recaoniZ81 four (rather than
three. ·"itlfe ctfnStriCtfoti tocaflaft -lor·vawell:--pItatal, vetar. upper
pharyngeal and lower pharyngeal.
The view that articulatory properties of yowe.e can b.
ftDrnHd in terml of the canstridtan location, originaety due ta the
Sanskrit grammarians (ct. Varm. (1929). Allen (1953). entoyed a
great d.. of popularitY in the tim.. precsding the introduction of
tit. tongue arch modet. Thrauotl cultUral contact. this view sprnd ta
China. J_n. Arabia. and to Euruo.. .tier. it is reft8Cted in the
workl of John Hart (1SS9) (dfscuss" in Oanietssan (1959), Jacob
Padsen at Aarhus (1589) (discuss" in Mel,. and Skautru;) (1930».
HeUW8CJ (1781), and Sed (1849)-,
amono
others.
In 1887. AI.zander Bell introdUC5d a new way of reoresanti"o
vowell: he ob,eMJd that the tongue position varied for dlHerent
vowets. and he proDQHd to account fer this fact in a rnodeJ in which
articulatary praDerties of vowels wer. expr.ssed in terms of the
position of the highest point -of ttt. tongue. Belrs model gave rise to
the hiOh-Sow. 'rant·back f.atur. system which forms the b•• is of
vinua!!7 all yowe' ctassifications currently available. Fallowing
Wood (1982). I will refw to this system as '1he tonaue arch mod..•.
For almost a csntury. the ta~. a:et1 model cmtrIhadowM aN
ather methods of characterizing vowet articulation, including. the
ancient mMhOd baud on the constriction location. Its success h••
not been diminished by the X-ray findings of Meyer (1910). Russell
eTltn, ref.onen ere QUOtH e't,r WOOCI (1982,.
-...
With the excaDdon of the feature [tensa). Wood rejects the
tongue body 'eaturet ((htoh). powl. [back)). In his model, vowat. are
characterized in terms of the fonowing binary f,.turel: (palatall,
where (+palatall are III, ILl. lei, III, lui, 10 I: [velar). where
[+velarJ are lui. /0/. 101. I~/: (pharyngeal). where [+pharyngesIJ are
101. fIJI, lal, /8/: (openl, where [+Open' are 18/,
/0/, 1'31. la/.
181. and (tensel. with the usual caveraoe. Notice that in Wood's
terms. the feature (openl takes ovar the function of the feature
[high). thus makino the model capable of capturing the height
,.1.
phenomenL
Wood defends this mod.. on the basi' of X-ray evidence which,
he aroues, reve.'s similar constriction locations fc;~~ similar voweJs.
both for tho tokens from the same language. and crosslinguisticatly.
He also comoare. the area functions of similar vowels 'ram var!ous
;anguag... and derives the same conctusion. The only crosslinguisttc
variation that he observes. is in the conntrietlon location of the
vowel fll. which tends to be -more anterior" (Wood (1979) p. 34) in
languages which have mora than two high Yowets. and in the vov-:ttl
lui, which may be near-psJatat velar or uyular (I will return to the
128
127
a_lidEC
:c:aaazrsz:
,ev_,ed
a discrepancy belWHn ttl. mode' and
the anlculatory reality. (The crittcilrns ra• • tty the res. .rchers
will be dlscus.ed in more detail in secttan 5.3.) N.v.rth.I••••
dllsattsfllctlon with the tongue arch model hu led same r....
to look for alternative means of chanleterizing vowel antculailon.
On. 0# tfte main claims af 1i'41. th.... is that the phonological
repr....1IIIIon af a speech sound indud.. felltUl"ft which carrupand
to the location of ttle canstrtetton(s) •••Dci.ted with sound
production. WhU. thll cteim is not new in phenology. it is
uncontrov....
anly .. far .. the trutment of canlonant. is
concarned: the constriction lacation pl.,. a rate in the IPA
cl•••jflcattan of conlonants (In terms of f••ture. [labial).
pabiodantaiJ. [dental). (alveolar). (palato·alveol.r). [palatall. (vel.rl.
(uvu....). [pharyngeal], and ParynoMID. and. to a I...., ~ in ttl.
SPE tramewark. which must appeal ta a 'eature like [anterior). in
order to aocaunt for the phonemic contrastll among the coronal••
2. . llJEiJiLiWilii&didlMi+
2
OJ
i_I
JFI£22iiUJ&iI
topic of cro.IUngulltlc v.rlalio:1 In the conltrictlon location of
vowell la.r on, in HCtion 5.3.)
Rn.Uy. Wood studle, tho EMG records of the activity of the
e.lnnlte 10"1:1. mulCltt., which are tf1ldlllon."y considered to be
r••ponllble for the pcsitlon of the tongue during yo.. ~ production.
On the basis of th... records, and anatomical descriptions, he
concludes that the faur placee
constrlctlon: palat.l. palatal".'ar, ~."r-pharyng••I, and pharyng••• ar. 1de.II, corr.latad with
1M movements those mulClel can ••ocuta.
0'
Whl.. the mocMl at legment rep,...ntlltlon d8velopsd In thll
th.sis differs In m.ny re.pects from
moGel, lome of his
'.Iuits are directly ,e-ov."t to the propos.' I am making. Wood
pres.1I • considerable body ot evidence for the claim that vowe'l
ean be reliably chs,.cterized In 'enne 0' the conltriction location.
Wooer.
The view that the place of constriction can be • parameter of
vowe' .rtlculation h.. atlo b•• n •• pre'led in the cont.xt of
r••earch on 8caustlc·artlculatory correspondence. (Stevens and
Houl. (1955). Fan. (1960), Itc.), AI point'" out by Fant (1980), the
articulatory par.met." of the tongue arch model are somewhat
.wkward to us. in computattonl relating the articulatory and the
acoustic data.
On the basil of the results in Stevens and Hous. (1955), Fant
(1980), and Wood (1979. 1982). it Is posibl. to conclUde thai since
similar quality Yowels are associated with similar constriction
locations, phonological f••lur.. based on the constriction location
of a vowel ar. sprlori possible. The que.tlon thai remains to be
snswered is whether the constriction regions for particular vowels
correspond to features which a phonologial might want to posit on
the basis of the phonological behavior of the vowels. Since this
question can be answered property only after tho inventory of Site
f•• ture. for vowelt is established. I return to it reliowing the
section which reviews the arguments for representing vowels in
terms of constriction location.
129
5.2
Phonological Argument' for Site Feature. in Vowell
I have presented the main body of evidence tor the phonological
constituency of the Site featur.. in vowell in chapter 4, when I
argued for the phonological statUI of th.s. f.ature. In gen.ral. In a
large number of ca.e.. the phenomena in which the Site f.Btur••
were activated, Involved conlonants al wen .s vowell, Therefore,
the present sect~on should be vleweel largely as a review of the
arguments pr.s.nted In chapter 4, with the discussion focused on
yowell to a greater extent.
Labial Site in Vowets
5.2.1
Vowels represented with the Labial Site include u. 0, O. ~, ii,
and
The evidence for the phonologically active llblal Site in
vowell camel from labial dla.inlUatlon phenomena in Venda and In
Zulu. In Venda, discussed earlier in 3.1.1, dislmUation effect is
observed with bilabial consonants (+, ~, m, ph, p, b), when they ar\l
folJowctd by the passive suffix /-waJ. The following alternatlonl are
reported by Oak. (1954): frJca~Y'1 become v~la,lzed before I-wel:
stops do net undergo any change, but Instead, the round glide that
foliowl them turns into a velsr 'rlcative. With the n8sal Iml both
pallerns of behavior occur. Before the /·iwaJ alternaNt, no
dissimilation il observed. The relevant data are rep.lted below:
o.
(72)(8)
lSo+
•
•
poxwa/Po+lwe
we
--)
we
--)
dtrwa/dl~lwe
nap • we
--)
nepMll/neplwe
dt~
khop"
+
we
--)
khDph)(S'~"~~hlwit
·be tied'be known'
'be switched'
·be broken of'
cob'
lum •
we
we
fun
we
beb
(b)
rind
+
+
•
we
IUrJwe/luml')8
·bft begot ten'
·bl bitten·
funwe/funlwG
rendw a/rendlVi 8
·bl loved'
·be pral sed·
blbre/blbtwe
--)
--)
--)
--)
Ooke (1954) reports that only bUlbi",. participate in the
dissimilation procesi. Since Venda does have labiodental fricatives
130
in its seern.m invalIDlY. ttl. dina in (72a) ,*,not be analyzed in
term. of dlutmilatlon at the Arttculatar tier. Th".fo~. th. rule
ntCPOnsible far ttt... data must be stahld in tenna of the labial
SUe.
While the
p....
(b)
•
•
•
•
1:'8110 • Int - ) 'I,.rywenl
I:zulu • tnt -) Izulwtnt
(c)
I
•
I:fu •
Inl
_e)
e • t:mbvu .. tnt
sutfin in Bantu i. conventionally writt.n
/.W8/ in the linouist!c nteratu.... it is more . .rapri.,. to analyZe
it .s a sequence of WI. . . . lui and 181. end. fallowinv Guers•••
(1981) and Levin (1985). who treat lui and Iwl •• twa
syllabification varianll of the same sooment. derive the surface
I-wel tram a pracft8 that gllclH the hlo"l YOWGt when it is failowed
by anotNr vawttt. SUIZPDIt for such an . . . . cames fram ttl. fllCt
that ttl.. ... no sequenc:ea of lui follcnNd by • vowel In VCIftdL
(d)
•
•
I
•
eft:nt/efwt:nt
_e)
•
·cloud·
."bvt:nt/eft1DYW t:nt ·S".IO'
tmDt • Int _e) Imptnt
Inte68 .. Int --) Inlaeent
The above data point to a dll.lmiiatory proc.s. which affects
bilabial consonants followed by round vowels (for detailed analysis
s.. 3.1.2). In order to ••etude the labiodental. 'rom undervoing the
prac... (el. (7~». it is nClC••••ry to limit the scop. of tile
dissimilation rul. to segments characterized by the Labial Site. This
is possible only if round vowels are represented with this f.ature.
Given thD abave analysis. it follOWl that the va'" lui mult
be repre••nted with tha Labia' Sit. in order to trioaer the
5.2.2
Palatal Site in VOWGta
dlnitnilatory ProcnI nlustrated in VMdL
Let us now consider briefly the dissimilatory process
affectlno bilabial, in Zulu (discussed mar. extensively in 3.1.2). In
Zulu. dissimilation of bilabials occurs nat only befar. the paslive
suffix. but also in twt! other enviranments when a bilabial is before
a starn·'ln.. round vawet. fallowed by the locative suffix l-iniJ.
Dissimilation affects bilabials only, and the relult is a pal.tal
segment. The rounded vowel tums into a glide. This is illustrated
betow (only a fractton of tne data from 3.1.2 is repeated here):
(73)(a)
-be:b
+
~U:D-
•
-t"u:m
(b)
we
•
-60n •
-8011
•
-t"snd •
(74)(a)
-4U:SVfe
-thu:pws
"be treDpecr
"be teeslcr
·be lint-
-6onwl!t
_e)
-80'lwe
we --) -tllsnClwe
118 sl.n"
°be pretsld°b. I C!'J5d"
we
we
we
wa
e •
1114,:6,
•
uml~:m~
•
-btt:dtwe
_e)
_e)
_e)
_e)
tnt -)
•
..
tnt
-)
e'14~c·8:nt
·frt IneS-
ernl~"I:n1
"mouth·
.~
The inventory of Vawetl whtch I propose to represent with the
phonological feature Palatal Site. consists of Iii. 11./. lei. and 11/.
Evidence far the Palata! Site in these vowels CGmes fram the facts
of schwa assimilation in Russian (3.3.1). Arumanian (3.3.4) and in
Maroi (3.3.S), fram vowel palatalization in Bulgarian (3.3.2) and in
Polish (3.3.3). and. malt importantly. from the palatalization
phenomena. to be discussed in chapter 8. , will briefly review here
some of these cases.
In Russian. we have seen a phenomenon in which a
phonomically unstressed 181 and 10/ turn into a schwa. which. if
followed by a stressed vow". assimilates to the pat.tal quality of
the preceding consonant. The recult of the assimUation is ILl. as
shown in (75a). Othent'ise. a schwa in the pretonic position surfaces
as /A/. This is shawn in (75b). (stressed vow" is in bald typ~):
(75}(a)
I ) slUe/
I~e.tl
• The 1/+ alternatton wtn be exolelnld Delow,
131
lJ2
-tongue'·
·wele'"
·h'dg."og·
rafa
",'Ite
IJalsl
l,.Joltsl
·wall taking'. feme
'arrow·
't8'"
SAm.
I.erne/ '
·Sllf·. fem.
gtAze
Igrezel
"'I".
fl
If
~"
ttl
xAcJitt'
Ixod'ttJ'
'to go'
I1ftl
J\!(b)
nAg_
_tat
Since in Russian. the ICftwcI can animilattt to any consonant with
the
con.uiction focatian (coronal or dersal), and only LO SUM
con.anantl. and stnc:e the rnult af tho ••Iimtl.tlon Is /t/, it
fol~ that /l' must be repreHl1ted with th. pata'" constriction
location (PaIataJ Site) . . ..I.
In Arurnanian. the I'ftult of a Procell whk:h assimilates sct1MI
to iI palatal conSOMllt (either caronal or dorsal) is lei. ·This is
shGWft beQcw:
f78l(a)
•
, '!lcr •
0•• •
PUrl
tltJ
•
•
•
•
--)
-)
•
-)
--)
purt•
lutra
bei.
t.~je
·carrted·
'workedo
'tle••d'
·cut'
H.,. 20. ~ the Articulator active in the constriction of a consonant
to \Which t:i. sctJwII assimilat. . mak. . nu diffetrenC8 in the quality
of th. au,milated VDWfIt. Thi, indicate. that ttle spreading feature
must be the Palat8t Sitae It fallows then that the phonological
reprnentatian ot lei mult inYGlYe a Palatal Site.
_tat
In MaIQi. a ward·final schwa tums to Iii only when preceded
bl' a
consonant (carenal or dorsai), or a palatal glide. This is
shown below:
(77)
III in word-final position:
si
bee
133
~V
J1~V
~l
'1.,'
Inogel
III in phra....final positian:
'Ipeetc'
'sotn'
'tie,'
'burn'
The argument made for the Russian and Arumanlan data holdl here as
well: slnee the aulmilatory 'nture may only be the Palatal SIte of
the trigger consonant. it fallcws that the resulting sflQment. IU.
must "Ia be repr..ented with the Palatat SIte.
I conclude he,. the arguments for the Palatal Sit. in the
representation of Iii. 'l/, lei, aftd II J. The only Yowet for which I
posited the Palatal Site. but did not present phonological evidence,
is III. However. s;nce this
is a lax counterpart of lei. it is
posllble to extend the conclusions about the constriction f.atur.. of
lei to the representatioft of III.
vow"
5.2.3
V..
Site in VOWfI
The inventory of yowet. charaderized with the '.atur. Velar
Site inctudes lui, 10/. loJ. and 1'3/. Fhonafooicai evidence presented
in support of the f••ture Velar Site in section 3.4 _. velar
dissimilation in Polish. and vawel hardening in Klnyarwanda. can
atso be quoted in support of the ctaim that this feature is present in
the representation of vowels lui. 10/. 101. and I'J/. I review bath
cases here !;)riefly.
In Polish. a stem-final velar consonant turns into a coronal
palatal (either If I, Id 1./. or 111). when follawed by a suffix that
starts with either a velar cansonant or with lui. The rehwant data
are mp••ted betow:
(78)(8)
pi J-slc
PtJ-aC-k-e
"09- 8
ux-o
nui-k-e
ui-k-o
134
'druterd·
'leg'
'ser'
ville
not-ui-ra
Ylle-ui
rik-e
nag-e
roe-un-a
not-uri-e
'hInd'
kot
'cer
fur-a
kot-t-e
rur-k-e
tet-a
reb-oY-e-c
tet-ui
feb-ui
'ded'
'thl.,'
mem-I
mem-un-e
tlur-e
tlur-un -II
'mom'
'cPfJughtlr'
nog-I
(b)
"ig'
'wolf'
"eg'
'pIp.'
In section 3.4.1, I have argued that the phenomenon illustrated above
ia best an8lyzed 8S a CB.e of dissimilation triggerGd by the OCP
applying ., the Vel.r Sit~ tl.r. Such an analysil Is possible only if
tho vowsl lui is represented with the Velar Site.
In Klnyarwsnd., • non-low vowel fuse. with the preceding
consonant whon fo~lowed by another vowel, according to the
followin, paradigm: a back roundltd vowel adds II labiovelar
W
component (t ) to the conlonant: a front vowel turns into a
palatal/dorsal componant. This is illustrated below:
(79)(a)
tu
•
ku •
ku ..
(b)
e:nge --) t,w e:rJua
'we het'·
dod • U • I --) kUdodfWa ·to bl sewed'
lSon + u • 8 --) kU Don""a ·to bl seen"
ku
~
91
•
8
--)
ku
•
rt
+
8
--)
kusoa
·to grind·
kur's
·t'O est·
Iml • u:ge --) Im"u:ga
The
flets of Klnyarwande support the claim that
·professlons·
luI and /01 mus! be
represented with the Velar Site, btaC8use only under this
:;;iirnptkiii can tha diff~iane: :~ th= b=h=~!ai vf fiGiii and back
yowell be explainCKI. In a theory in which vowels are characterized
sol@l, in terms of the active articulator. they are dJ&tinguished only
135
on the other hand, bear; no tangue
body feature. (••• the dlscu••lan In 3.4.2), lind so. It ought to
pattern with • vowel such !II lei In the same way it patt.,n. with
luI,
by the tongue body featur... 1kJ,
5,2.4
Pharyngeal Site in Vowell
The Inventory of vowell which 8r. Ch.'8ct.rlzed with the
feature Pharyngeal Site Includ•• 10/. I~/, lei. "'. and all '1II·lIke
vow.ll. Eyidence for the Pharyng••' Site In /01 and lei comel from
the vowel lowering facti ot Gr..nlandic eskimo. dllcu••8d In 3.5.4.
In Gr••nl.ndlc. high vowell III and lui are lowered to lei end 10/
respectively when followed by uvula,.. ,... , or
Data UJu.tr.tlng
1."
this phenomenon
(80)
I,.. repelted below:
s••meQ
"glacler'
neltoloq
·goos.~
tse!ltl!Q
"enklo'
IpIR8Q
Qasaloq
'bert'
nenoQ
'harpoon strap·
'bier'
Given that uvulars may not be characterized as [-high] (ct. McCarthy
(1989), and the dillCusslon in 3.5.4), vowel lowering in Greenlandic
cannot be mnalyzed In terms of height asslmllatlol'. Therefore, the
only way to treat this phanomenon, is te attribute it to the spread of
a leature which corresponds to the constriction location of the
trigger· . Uvular consonants, being complex segments (ct. McCarthy
(1989». have two constrictions: on" velar and one pharyngeal.
Cleaily, the velar constriction does not spread. a$ can be seen from
the fact that Iii turns into leI·· a non-velar segment. Therefore, it
must be the case that the pharyngesl constriction of a uvular
• Actuilly, It II DOI'lbl. to prODO•• on In'!;iii' of YOWl' low.rlng In Groonlondlc
conte'n~d enUrl'Y wlth'n thl artlcuiator t~.oru of phonologlc,I r'Dro•• ntIUon: on
thl .lsumPtion thlt UYulcrc er. fIDr ••• nled with both doreal end tDn~e root
artlculatorl, tho lowering Iff tel thl uvular. hlVI on the adJectnt vowIII can be
attributed to th. ,pr.ed of thl tongue root (of CO"",, YOWIII/I' and /0/ would
hi'!; to b. rtp'6•• nted ;1 coma.llx ••gm,ntllnvolvlng th, longul root el woll).
HoweY,r, thl Iuee••• of luch on InIIV.'1 II short liVid, .Inci. al Dolntld aut bV
McCarthv. ttte tongue root II not on leU". articulator In ell DherynglellOWldl. For
Ixomol., thl laryngeall may not b. Includ.d In I natural cll'o wlth,sey.
phar"ng••", on the b."1 of en Detty. articulator, but thlY do form I nlturol cl.la
with thl•• I.gm.nt. naY,rllle"'I.
136
spr••dB in Icelandic. Such an analysll entaUI that Ie' and !ol
contain a pharyngeal comporMnt in their repre.entatlon.
in their constriction location. Therefore, it is possible to conclude
that LabJal SUe Is artlculatorily well def!ned.
Evidence for the Pharynge.' Site In 181 comes from the
procesl.. In Tlborla" Hebrew, which turn the SChW8 to 1&1
befor.I.fter a guttural consonant (dllCUllld In more detail in 3.5.1).
The relevant data are repelted below:
5.3.2
ImeJkl --) meltle
l?fJr, I --) ";>8r.,
Iqadl' --) Qodl.
(81 )(a)
(b)
.
Ibahtl
--)
Ib8~11
--)
/bat""
--)
--)
Ib81~1
bah!\
be'i'el
'costly ston.·
bite"
'neme of City·
bi'lev
'sw .,llow Ing·
'mestlr·
In 3.5.1. I have argued for an analysis of the Tiberia" Hebrew facts
which attributes the 8 •• > a change to the sJ;read 0' the Pharyngeal
Site of a guttural onto the Constriction Node of the reduced vowel. It
follows from thhJ analysll that /81 must bear the phanological
specification for the feature Ph~ryng••1 Site.
This concludes the review of arguments for representing
vowels with a phonological f.ature: Slle. In the next section, I
address the question whether thl Site 'eature. which I p()sil (Labial,
Palatal. Velar and Pharyngeal) have solid basis in the articulatory
reality.
5.3
5.3.1
The Articulatory Basis of Site FeaturBs in Vowels: Part 2
The Articulatory Basis of the Labial Site
The feature labial Site is posited here for vowels lui, 10 I,
lu A/, /0 /, /0/, /~ I, /0 I, 151. and /a/. All rounded Yowels involve
labial gesture in which both lips participate. While there are reports
of certain amount of crossllnguistle variation in the lip shapes for
those vowels (cf. Linker (1982», I know of no example of variation
137
The Articulatory Basil of Palatal Site
Vowels which I propose to characterize with the fealure
Palatal Site, are III, ILl, JeI, and ",. According to Wood, th••• are
also the Yowels which are articulated with a constriction In the
palatal reglen (based on radiographic evidence). However, .a Wood
points QUI, not all cas.. of III, IL', etc., Involve palatal
constriction. After comparing the area functions of IV uttered by B
speaker of Southern British English, American English, and Egyptian
Arabic, he concludes that the constriction for the vowel uttered by
the Egyptian Arabic speaker is much more anlerior than the
constriction produced bV the Southern British English or American
English speakers (about 8 mm closer to central incisors thon the
constriction of the British English speaker's Iii, and about 13 mm
closer than tho vowel of the American English speaker). Wood
mentions that this shift in the constriction loeattan of /11 Is
particularly commo~ In languages which have more than two high
vowels (/+1 or
I, in addition to /il and lUI) In their Inventories
,although this is not t~e case in Egyptian Arabic). He quotes Fant
(1985) for similar observations about Iii in Scandinavian languages
and in Russian. compared to its American English counterpart. Fant
characterizes ,he RU5sian and Scandinavian Iii as -prepalatal-, and
the American English Iii as -mid-palatal-. Fisch&r-J~rgenson (1985)
also mentions vowels which have the maximum constriction at the
alveolar ridge, and quotes Straka (1978), who calls French III an
alveolcpafatal, or simply an alveolar,
lu
These facts soem to indicate that, contrary to the assumptions
on Which
the constriction model is based, namely, that the
constriction location of a vowel is stable, the constriction locateon
of a vowel may vary across more than on9 articulatory region. Given
Wood's (1979) and Fant's (1985) descriptions. some languages appear
to have the constriction for IV in the dental/alveolar region of the
138
yocal tract. Th!1 r8glon forma the bull of the f••tur. Anterior Site.
wall motivated by the phonology of the dentall.'vCtOI.r conge"a"tl.
po,..
Th. que.tlon ih.t naturally
ils.1I '. ~ thl. point. Is
whether the IiI of Scandinavian '.ngu",••, Egrpt'.n 'trablc. or
RUlsian I. stili to be trat., .. • p.....1 legment. If Ihe answer to
this question I.
then the concept of arttcu,-torUy defined Site
f••
should .'th., be Iblndoned. or modified so .1 to allow
dlf'.,en! articulatory definition, for conlOnantal lind vocalic Sites.
Ot cours., luch • propo..! weak.. the concept of the Site '.atur••
conBId.rably.
y...
tu,..
How.ver, this co~.pi. and th. constriction mode' which
depends on il. would not need to be compromised. if it could be
shown that the vowal Iii must be repreaented with the Anterior Site
in I."guage. in which it is articulated with • constriction in the
anterior raglon of the mouth.
Apparently, there exisl phonological phenomena whose
Interpretation require. that thlt Yowell III and I.' be treated as
Anterior. An example Of one is • y.lar~zatJon procesl aff"ctlng Ifi.
/11. and ItJ in Standard Thai. mentioned briefly In 3.2.1. I now
consider this phennmenon in sarno detail.
5.3.2.1
Velari:ation in Standard Thai
In his charactarizat~on of consonants of Standard Thai
(segmel"! inventory after Ruhbn (1975): i.•, u, e, 8. 0, I, ~, e, sam~
lertl. wtth '.ngth; P. ph. h, m. r, t, ttl, d, I, n, I, Irt f, f"# k. k h #g, '1,
? , h), Harris (1972) reports that If/, III. and III become veiarized
lyll81blo-iniUally. befors clo.. front vowell. Velarlzallon is
perceived al a bnck unroundsd off-gllde. The following are axamples
that illustrate this phenomenon (Efter Harrll (197~»:
(82)(a)
f'f:
sil:
'0 botr
·four·
tIl:
·to hit·
139
(b)
fa:
SUI'
ts:
·Sky·
'pret ty·
.'y,'
AI pointed aut In 3.2.1, !hit only way to charaeteri!. {f, I. tl ••
an artlculatorlly natural class, I. to appeal to the fact that th•••
segments are produced with a constriction in the d.nt.U.lvK~.r
region. Within the constriction model. thl. region forme the bU.I of
the f••tur. Ant.rlar Site.
Under the aSlumptlon that vow.l, /11 and /el In thai shar.
this feature with (f. I, tl. Yelarlzatlon of anterior consonantl can be
exptalned 81 • dl••'ml;atlon proceu. triggered by the OCP applyfng
Oil the Anterior Site tier.
While there are no radIographic data on the articulation of IV
and
In Thai, It is possible to speculate about the articulatory
characteristics c. t these vowels on the balis of the acoustic d.~r
which. fortunately. are available.
I.'
Wood (1979) reports that articulatory differences between the
lil of B.itlsh English and that of Egyptian Arabic are reflected in
the acoustic proptsrties
these vowels: -The consequence of the
Issl attterior mid-palatal constriction is a w'der pre-palatol part
and narrower post-palatal part. which will both yield a lower Fs •.
Conversely, the more ll~t'cioi vowel has a Gmaximally high Fs • (Wood
0'
(1979. p. 34».
At f:rst glanc~. the acoultlc data on Thai (Abramson (1962»
do not suggest that the vowel III has an anterior constriction. The
third formant of thiw Yowel ranges betweoil 2800 and 3200 cps. with
the mean average of 2910 cps. basad on 15 measurements of v:lwels
uttered by two (male) spsakers. By comparison. the third formant of
Iii uttered by the male speakers of American English averages
between 2775 cps (Lehiste (1984)) and 3010 cps (Petarscn and
Barney (1!:S2)). Given that the American EngUsh IV is articu:storily
characterized as mki..palatal (ct. the comments Clibove), ~hon. if the
third formant Is a tru~ indicator of the constriction location in high
VQwols. It would app~ar that III in Thai should have roughly ths
constrecton location of the American English li/.
140
H~we"er _ the F, at the 8111,.,... Iii is ;.Itarpr,ted differently.
when considered togethar ~th other formants. The F, of this vowel
,.no•• between 280 and 380 cps. averaging at about 322 cps.
Compcred 10 F.. 270 of the American English 11/ (Petereon and
Bame, (1952)). the FI of the Sa-meso Iii impllel a phoneticaily
somewh8t lower yowe" for tYhCt; the F. average of ~1 0 cps does
indicate • fairly anterior constrlction-. It il u..tul to compar. here
the 'orman! structure of the Siames. IiJ with the form.ntt' of the
Am.r~n Eng~.sh Ill, • front ;ax vowel. which. 8S • consecr.... nce of
laxness. h.1 3 lowar tongue body posftlGn than Iii. evan though it is
phonologically -hlgh-. lhls filld",paJatai vowel has F, of about 390_
and Fs
2550 cps (Pttterson and 3arney (1952»). What this shews is
that lowering of F. will result in lowering of F, without the shift in
the constrict~on ;ocatlon.
Beior. let."ing the pntS8nt topic. lat UI consiJer some of the
consequences of the proposal made above. If all vowels articulated
in the anterior region of the mouth 8ro to be tr•• ted ••
~honologlcally Anterior. then the III of Ruesilin. with. very forward
constriction (ct. ~. tracino... in F8nt (1980). p. 107) and the high F,
(3200 cps: Fant, !:a:ne. p. 109) to furtha: confirm the 'f~Udlty of the
X-ray reading, ought to behave phonologically like an Anterior
sfJgment. However. the faetl of Ru.s~a~ do not bear out this
prediction. If anything, the RUlsJan III acts Uk. a palatal ssgment.
~n that it trigger. a palatalization rule which converts v.lara Into
coronal paIShll!: muk-almuf-l-t J 'torture/to t"rtu r .'.
0'
c*
Whj~8 the ultimate ~vid8nce for the anleriorlty of Iii in
Siamese should come from the articula~ry data, either X-rays or
palatogram!. the acoustic properties of this vowel CRn be
interpreted to suggest the constriction location more antorior than
in an average Iii.
What can be concluded on the basis of the Siamese d3ta. is that
shift ;n the constriction location of a Yowe' is not
phenologically neutrt!1. Siame~ .. n:. which can be hypothesized 10
have an an!erior constriction location on the basis of it~ acoustic
properties. pattern~ in a phonological process like a segment
r9prese!1ted with a pho•. :1 logical feature Anterior Site.
t~e
5.3.2.2
Anterior Vowels in Other languages
• It I' nflt the cotl that tht higher F', of (hI- Slem••• III could' bl ot lrl~ut.O to 0
smeller ,Izi or t~; yocII trKt of tho SUbJlct.• !n the .xl7.rlm.nt If thll
~~.,tl)!.
~or"
tM tSI•• ell tormonts. far '11 Yow,ll should b' ~ sl ••d. tutl .1 lh.V ert
t".
In
cg:.'th of women rand avon mort 10. children. Far .KDmpl ••
~orment
Ilrueturt for 'hi Amer'tC!'I Enq!llh la/. ulll1-d 'Jli 1ft adUlt mI..... II f, = 730. ;, :
thl
I~~. ~'I s 2440 ePI Tho lem. YO",.I utt.,I:C! tty. women har i~ F•• es"-.,,. ::
1220. eM Fa ~ 28 10 COl: by " child -- F, & 1030.
i370. fa % :5 ; 70 epa
(p.l.rc~ end 8~ntt\j! I ~S2». 8lJ eornDlrls.on. tho Siem••• lal utterlt~ by 8 mall
'IDeoter he, Ft =670. F't z I~~. aM r, 1I 2330
f,.
141
One way out
the difficulty posed by tne paiataUz8Uon 'acts
to attrihUl8 the p.l~talizattf'n effect nat ta 111. but to a dlacr~~1c
with which suffixes that trigger th~ rule would be represonted. Such
a diacritic (or a flcaUng palatal segment -- a~ this point the
difference is not ;mportant) is ind".lendently needed In order to
account for palatalizing properties of SLHfix85 like I-n! (e.g .• knJt gi~
e/kn J I2-n-+J
'book'). which are not overtly represented with
segments that are likely to have pglamllzing propertlas. Also, or· t
all suffixes that begin wUh Iii trigger palatalization. e.g.. k "I I g- e J
kn't g-I 'book-, «hich shows that palatalization is not a true
phonological rule of Fiulsian. but a highly lexicalized phencmenoi
instead.
Another way of accounting for the behavior of the Russian IU
is to appeal to the fact that of the three high vo\.eis in the language,
only IU and lui are underlying s8(iments. 1+1 being derived from (II
by the process me~!inned earlier in 3.3.1. It is cone,,;vable th"t the
Anteriofity at Iii is a derived effect. one that follows the rule
deriving 1+1.
If you recall Wood's generalization mentioned earlier. the
-anterior- vowels are most likely to occur in languages with mere
than hYo high vowels. ThIs suggests that the -anteriority· of a vowel
is a product of those procesS8G which regulate the articulatory and
perceptual distances between sounds. in accordanca ..with the
142
-dfspersion- theory Initiated by Jakobson (19411),
and recen!ty further developed by LHjencranti and
Since if! Ru••lan, the -thlrer high vowel (/+1) is
unnJuonabie to aIIum. that t!le Anl.riorlty ~t
"t1enomenon a.
creatH
5.3.2.3
wen,
"'artinet (1955),
Undblom (1972).
derived, it is not
IV i. • derived
and that It II orcMred after the rule which
It/.
Second Velar P...aaUzation in PdJah
An Interesting fact of Polish Is that in some cases (mainly in
derivational suffix.s), iii aetI like a p."ta~ vowel, and palatalizes
the pr~edlno IkJ to If I, e.g., !hakl 'hook', Ihef -tk/· 'little hook',
but in ethers (mainly in inflectlona), It triggers -anlerlorization- of
the preceding velar (in the Slavic Iingui3tlc tradit~on, this
phenomenon is referred to as ·secoml velar patatallz8tton-), o.g.,
fjak/ 'how', IJats-I' 'which'. /vrog..a/vrodz-+' 'hostile', etc. The
anteriorizalion cannot be dismissed •• siiy as a morphological
idiosyncrasy. beacaus8 the sam. suffix.. that trigger this behavior
ir; velars. cause the labials and the dentmls to become ·palatalized(hero. in the sense of aCQuiring a vowel..like articulation), for
example Ikoloroy-a/koloroyJ·jJ 'colourful', If ervon-. If. r v 0 nJ·il
'red'. In this respect, they panem like sufflxe. which trigger k --> f
rule, for example: lok-ol 'eye'. ;z-of -+- I 'to notice,. a~d Iprav-oJ
'law', ipray J - t - I 'to preach', Iran·oJ ·wound'. IranJ - t 'to wound'.
c
c/
An assumption that a Yowel may change its Site specification
as a renex of the dispersion phenomena, could possibly explain the
• In PolIl" 'I' nevor IDPlorl eft.r If I. Id!/./I/. If I. Itl/. and Idzl It Is .IWB\lS
eonv.,tld to 1+1 oftOf' th"1 consonants.
143
'+'
(83)
Th5 above anatysis can ~ profitably extended to Polish, which,
a. far as the vowel s,ltem is cahC8l'l1ed. ronmblel tM situation
found n. Russizn in a number of wayl. Pol"~ too hal three high
vowell: Iii. lui, 8nd It/. Arttcula1OrHy, ~ .. $'oliSh III appears to be
very anterior, 8. shawn on the tracing In Wlerzchowskll (1985), p.
83. Unfortunately, the acoustic data are available only tor the first
and second formants of this vowet.
c
dual nature of the Polish Iii. AI argued by Gu•• mli~ (1998' and
Czaykowlkl·Hlgginl (1988),
In POUI" must allO be derived from
III. Su,po.e that prior to the .rrival of If-I (very likely, an
"nroundl<t velar), the vowel IU il reprnented with the Pahatal Site.
The procels which spreads the constriction of thlt vowel onto a
consonant can be roptnented 81 fallows:
Palatalization:
Root
'~od.
r Conltr.
Root Nodi
--.
- --- - Constr.
J
A
Pe1etel Tongul
Body
This rule rssuits in velars turning into coronal palatals: and nonvelar. acquiring the secondary IiI-like constriction. I analyze the
effect this rule has on velar consonants as a result of a festure
readjustment procell:
complex segments which mention an
Artlculato: or a Sit. more than once are not attested, lind In all
likelihood, are not passable. A palatalized velar woutd be atn eX8mp~
of such a segment. since it woukt have the Tongue Body ~::YOlved in
the velar constriction and in the palatal constriction as weir. In
Polish. the velar CO;'1strtction is deleted and the palatal constriction
surfaces with the default articulator ... the Tongue Blade.
Next suppose that the affixation of inflectional morphemes
place in a separate component of the grammar eel.
Czaykowska-Higg~ns (1987) for an argument to this effect), at the
point when 1+1 is already present In the inventory, and III Is
takes
e.
• Given the ,xDleniUon thet Dalelellz.a vllerl may not surface bieaul. thav ere
not DO•• lbll ••gmantl. 0 QUlstion may nlturol1y ertlQ'
to wh\llhOUld SUCh
••gmente b. derl"ld In the flr.t Dlac•• pertlcularl'-l. In vl.w of the 'l.umDUon
moCSI In thll th"'I. thot ImDO.,.'bl. I.;m.filt ef' not derlv.d bV pl'lonologicol
ruili. To Inlw.r thll QUlallon, I wolud like to DPDlel to thl 112'. lnttodUC.d
lor It or• naml'Y. thet the Artlculalor Yalu. II not ID.elfl'd In Yllere In tht
UndVlylnt RIDrI•• ntotion. Ther.for,. tt\t sprlad1ng or I constriction thlt Invol.....
the dorsil articulator Ihou;d not bl ~Iock.d tty ol such a stege.
144
a.
,..natrzed
a sogmen~ with an Anterior Sit. (for convenience. I
will tIM
a. a Graphic ~ for such • vowelj. The effect that
has on the pr8CCdr.g COftlOnanii no longer needt to be attributea
to • ..parate process (not
y to e.plaln under any
a.sumption). The Mmrionzation of *ar8 Is derived in the lame
YA1 that palatalization hal btten derived • moment ••rUer: the
p:ocen which spreads the entire canltrtctlon
/.1' 18k.. place. II
h•••••etty the same 'annat u the paJ8tallzat5t1n rua. In (12),
••cept that lhe vowel la n~ repreunted ~th the Ant.rior SUe.
firat. !he non-velar conlOnanls emerge from thl. proceSI with an
/.t,-Uke constriction (molt IIkel,. .n III-like con.lrlction. are
:eanaUzec:l a. ,., /-lik. at thll point). The /., I-like constriction
spreads onto vela... •• ..H. However, Iince the segment which
combines the Velar/Tongue Body and the Ant.riorlTongue BOdy
constrictions is not a po.sible sound, Itl velar constriction is
de..ted, and it surfaces as • segment with the Anterior constriction
only, and the Tongue Blate as the default artlcullltor -- either ItS/.
or Idzl, dopending on the inpuf.
I."
'.'1
v." •••
0'
0'
This concludn the discussion
articull:tQry variation among
Iii like vowell. let us ree.pltut.te the main points brlofly: while
considerable variation in the articulation
Iii has been observed,
thar. are good reasons for treating the -displaced- vowels precisely
in terms of the constriction location suggs.ted by the=r articulation.
This r••ult further supports Ihe position that pnonological features
based on the constriction location of speech sounds, in particulsr,
the 'eaturn Anterior Site and PaialaJ Site, ant phonst6ca11y sound.
0'
5.3.3
vnwell as [+ve.arJ. The main difference betwHn wooer. proposal and
mine la that I treat lui .tId 10 I al just ".'.r {with roundlna added
81 an enhancement teatur., ct. Stevena and Key• ., (1989). wher•••
Wood treate the.. vowels al both velar and palatal. While I do not
adopt Wood's proposal-, I rr-:ognlze 1M rationale behlrid It
Wood observe. a cert8in d~r" of ero••llnguistlc variation In
the conltrldlon location of the high back rounded vowoll: whU.
molt lurs ar. articulated In the region of the soft palate. lOme
have • conltrictlon at tna boUndary belW"" the soft and Ihtt hatd
palate, and some have a constriction right oppolite the uvula. Wood'.
observations are confirmed by a thorough Inveltlgatfnn of tho
articulatory propertl.. of lui and 10/. by Jacklon (1988), who
interprets the extent of articulatory variation in the•• vowel. 8S
evidence agalnsl positing parameters of Yowel articulatidn baNd on
the constriction location
a yowel. Since articulatorily. lui .IId
10/ pre.ent :ilff.rent problem., I shan discu.. the two vowel.
0'
sep.rately.
0'
The reports of vallatlon In the constriction location
lui
raise the same question that the -anterior· iV raised 68,1I8r.
namely. what is the phonological representation of lui articulated
outsd:te the boundaries of the soft palate, th~ artt:ulatory corrolate
of the feature Velar Site? If it is indeed the case that all lui-like
vowels are phonologically identical. regardless of the articulatory
diHerencel among them, then this is again a problem for the view
that there can be phonological features based on the constriction
location of a sound.
Articulatory Basis cf the VelAr Site in vowels
Vowell which I propose to represent with the Velar Site ar.
the following: lui, lei, 101 and I-:J/. To some extent. this is similar
to the proposal by Wood (1979, 1982), who. too. characterizes these
• ~n t"~. DfotHS./xl doel not Dlttlr" with Itl encJ 1;1. but Ollltlllzl' to III
Snltled. Thll ,. not tM onll Jl'onologlCQI DKUUw"g dlgp18V1d by this IOUnt! In
POUsb. P-'-O' It•• beh..,lor of lbl, IOWICI will" MUir \ftCI"..toccl once III
~109lcal Itllus
SI ••• o•••d correcl1;
145
• Pr,.umlbiy. thll Droco.el I. Int.nd.d to account fOt thol. tGkln. of lui which
are erUculatld with; constriction at thl DoundDr~ betw.en thl psletal und Yllar
rlglone. I do not ICSoDt thll DrOPOI.I, boelUS., glvln thl conlt~Cllnt Ig&Inlt compl,x
slgmentl which Involvi tht lime Articulator or the 10m. Site f.oture mort thin
one•. Ute ·conltrlctlon" mod.1 •• not CIOtti., af reprelontlng yewlle with
slmulteneous veler end Dllltll eonstrlctlon., II both such constriction. wouJd
hovt le Involv. the dorlal erUculetor (for 9rgumonts egolnlt reor •••ntlng vowIII
'II Ith U'I cafonetl ortlculator
5.4). Now. gly.n thet th.r. II no PhOnologlCI'
,vldenc, rrom ony langue;. thlt lui a:tl Uk•• pelltll/~.I.r I.gment, It I.
I"
ptrhepi on UMICCIILrV comproml •• to Dropos•• thlorv which 8nOwi such I
repr, ••nllt lo~.
146
Lot u. first IMtdr... the problem o! an lui-like vowel
ilJticulaled with • poelible palatal GOnetrlction. J8Cklan (1988)
1ftIIdIDI. _
OM such ....: it II the lui of French. ertJculated near
the boundlwy belwNII the 10ft . . . and the hard palate. Wood
(1171) reportI that the lui
btl EgyptIan ArabIc into,.,.". allo
appruched Ittls t-oundary. ItoWDwer. Iheu ca... do not .xactly
P8'deI 1M cae of an •.....,... N: here. ..... VOWII could sttn be
~ldlIred "..., (particularly. If . . adopt th. quantal vi.. of
'Ntu,.., ct. Stevena (1172. 1101»). without "'llng • nrioul
problem for ttt. conc.pl of the conatrlctlon location· baled
8ItIcuIatory palWMt8rs. And from . . phonological point
view, •
constrtcllon tocation of • vowel. MtbIguoua t•.-..n two po••lbIe
"'turn, would SNm
with the
f _ which
f.'wln
doedttde
compa....
pIIoI__
or
would shoe the vowel pattern In terms of eIthei of the f••turn.
A moro serious prob"m for the phonologicel theory which
posits the Site f••tur.. would be an lui·... vowel with the palatal
constriction location and the velar-like bohawiGr. So tar. nobody has
reported the exlltenee of luI-like YOwo-. with the plla'al
constriction location. and 10 the prediction. of the constriction
model cannot ba '"ted.
On the phonologal steIG, there is at least on. well known cas.
ot lui patterning with Iii and acting Uk. a palatal Yowel: it is lui
in Papego. In Papago (segment inv.,tt»'Y a~r Maddieson (19U): i, +,
u. ~. e. P. b_ m. t, ~. s. n. f. dl. ". d••• dJ. J. k, g, ?, h, w), dent1!ll JtJ
and JdJ are in complementary diltribution with If I and Id!/: iel and
Jd2/ appear only before high frenl vowel. iii, 1+1, and luJ: Itl and
IdJ appear in all other environments. This is shown below (dllta
quoted from Kenstowicz and Kilsaberth (1979)):
(84)(a)
clnlg
-to moy, the lip,'
clkpan
'work"
fuegte
·nlt
tutul
cukme
·chlekln°
f+posld
-to brand-
be~r
dftwlkon
-arrival'
'storm'
'to scrap.'
d!un'
dlu.ukll
-drlld cectul frlut'
-llzerd Ip.'
tlte.
tetel
tem'
'mother-. young brattier'
-gums'
dltwhted8U
dlt-gol
0'
(b)
-to rub'
-hlellng'
tohnto
todsld
-t.ndon'
-degen.ret.'
dekpon
-to frighten'
'to Illp'
do?eQ
'mounteln·
Assuming that the phenomenon illustrated above Is an exampfe of an
exceptlonl... allophonic distribution, it mUlt have • phonological
explanation. Kenltowlcz and Klss.berth (1979) suggest an analylll
which r.'ies on the characterization of If I and Id!/ •• redundantly
(+high). in the SPE tl1ldltion: the
did! alternation I. then treated
as an aallm!lation In height. However. there are good reasons to
be!ktYe that If I and Idtl do nol bear any tongue body specifications
tie.
(ct. the discussion in 3.3.1). therefor., a different explanatton for
the facti In (84) should be sought.
Danca Steflade (p.e.) sugpested to me thOat perhaps thlt
phenomenon observed in Papaao is of the same kind as the one which
occur. in Japane.e, where all high vowels trigger affricstJon of the
dentals. What mak.. this explanation plausible. Is the fact that
Papago lacks surflce dental affricates: it is not Inconc?ivable then,
thai the dent.' affricates, derived from denial stops b.forfJ high
vowell, surface In Pap_go as the p~~atal affricates. ~iYen this
analysis, it is not necessary make unusuai assumption. about the
repre.entation 01 high vowell In Papago-
-derk-
141
148
Let us now cona.ider tho.. cu.. In which the Yowel lui turns
up with • uvular or upper pharyngeal constriction on the X-ray.
Jacklon (1188) reports !hilt whJb th... CU8I ant rather rar•• th.,
neverttte.... CSo occur. For . . . . . . he reports, on the basis of X-ray
IrKingl In Svamy and Swam, (1855), that lUI In Chin••• hal .n
upp.r pharyngeal constriction. Wood (1979) reporta _ uvular
conltrlction in lui of the Southem BrItIsh English speake, who was
• subject in his .xpertment.
0'
I .m not . . . .
the phonological phenomena in Chin... or
SoLtthem British English, which would help eltablllh whether Indeed
the lui In the.. langueg.. must be tr••ted .. velar despite the
pharyng.al articul8tion. ., such ph.numen. do e.llt in th•••
language•• then tho, are d ••rty • problem for Wooer. theory. and
they a,. • problem for the vi... that the f•• tur0 Veta, Site il
artJculatorily sound_
The only example of an articulatorily shifted lui I know of is
the lui
Gwman. U illultrated on the tracings in M~.n. (1983):
However. the German lUI does not behave ,••• velar sound. In fact.
it patlaml with the other two pharynguJ vowels. IfJI and Ial. in 8t
least one phenological phenotMnOn. This phcNlomenon involves dorsal
fricativel in an allophonic aJ....tion known commonly as -Achlaur-:
when adjacent to lUI. 10/. or /al. the dorsal fricative surface. with
a pharyngeal constriction: :n all ether environment.. it surfaces
0'
with • palatal
(85){_)
constriction. This is stiown below:
e'Xtln
·to
",.d·
(b)
I~
or
YO~'
·w••ac
nlr;t
bu'"
au'"
-boolf
melt~.n
·no·
"glrr
·ello·
mll~
·mllt·
or lets Germ", lui In Del.ttr, (1971). "'ow thlt YOw~1 vr It" • v•• ar.
not wuJor eon,trlcUon. The only r,eton for thle thOt I could t"lnk of II thet
DW"&os lhll vow.1 ItWOl,," two cGMtrieUont whIch .... not o.eeuted
.tmuU
Ig. Sl~ X-rO'J 'rKI• • • orCln~lIy b4JAd on lUll PhOt09rIDftI.
tltty w nl r
rlftGlJr U... erUeulltlcn of e COfft;)Hnc Isgm"'t correetty.
• Ttl. trlClnp
lv
149
The above data can be analyzed 81 foliowl: the underlylna dorsal
fricative ill unspecified for the Sit. f••ture. An aUophony rule
spread. the Pharyl1G".1 Site of an adjacent vow., onto the
Constriction Node of the fricative. A default r!Jl. IUPPU., the
Palatal Site to the rem.I~lno dorsa' ,ncatIY...
Impli8i that the r••ultlng segment, which I have
transcribed .s a uvula" II repr•••nted with the Tongue Body
conltrlc\.on at the Pharyngeal Site. Even with the articulatory
d.tcripllons (cf. Delattre (1971
it Is not obvious lUI thll I. lhe
correct representation of thlt lOund. If German IXI should tum out
to be a true uvula,. and ther.far• • comple. segment. a different
anal,lll of the data in (85) should be conlkbred. A poc.lbl.
alternative is to tr••t the Achl.ut In torml of a f••tur. changing
process which spreads th. pharyngeal constriction of • vowel onto
the palatal frlcatlv., and whose output II late, r.analyzed a. •
this analys"
»).
uvular.
let UI now discuss the articulatory basil of ch.raeterlzlng 101
with the velar constriction location. According to J8ckson (1988),
101 il even more notorious for having variable conltrictlon locatlofll
than lui: here. it is appropriate to note that Jackson responds to th9
articulatory cha,acterlzation of 101 by Wood (1979. 1982). who
deslgn.i'. • special articul8tory regton tor JoJ: the upe:er pharynx.
While Jlckson admits that thl. il the preferred constriction rogion
for 101. he lists a number of cases in which /01 Is document~ with
a constriction In the velar and In the lower pharyngeal region.
At first sight. all of those observations appear problematic for
the treatment of 101 suggested here. Since I propose to repressnt
this vowel as a complex segment. consisting of a vel"r and
pharyng.al constrictions. I predict that both constrictions should be
refleded in its articulation. However. the reports by Wood and
Jackson do not bear out this prediction.
Yet it is not the case that the vowel 101 never appears with
two constrictions on the X-ray. For example. the two constrictions
ISO
are ...ety ObHrvod on th8 tl'laClngl 0' OhaLuo vowell shown In
J.cobIon (1978). Hbw II then the amount of variation in the observed
proptlrtin of 101 to be explained?
A number of thing- can consplr. to produce the ab••rved
_Hectl. Flr.l. the vowel 101 •• • comp'.. legment with
constriction. a' two anatomically MI!ac.nt reglans: It ." not
uncommon for such HOmenti to appeltl with the I\tIIO .d).cent
constrtcttons merged Into one. IOINMhere along 11. two regia",. For
••empIe, 1M segment such as It" (found In BuIgMan, Runlan, ele.),
which must be roprnented phonologICally whh the AnteriDrlTongue
Blade and hlataUTongue Body COftI!rictlone. ohn appears on the xray with one long conltriction. ct. scallOn (1984). Sam_tim.. the
two constrictions 'use .rlicula1Orltv: for example an underlying It"
in M8C8donian is ~..:~uncect .. II palatal ltop. It is not unlikely that
the -tusion- .ffKt il compounded by the 'act thai phYlically, the
two articulator. Involved in It' I ar. Iituated on the sam.
an.lamical organ. This il also tha ca•• with 10/, articul8ted with
the Tongue Body and Tongue Root, at the Vela, and Pharyngeal Sites
respectively.
Let UI now consider the cu•• reported by Jackson (1988), in
which he observed 101 with a velar constriction, but not lower
pharyngeal, or the low.r phl1rynge" constriction. but not the velar
on8. Tha.. effects are not very surprising, and can be eltplained
.a.ily by appealing to the possibility that In a complex segment,
articulations do not always nsad to be simultaneous. For 8x!lmple.
Maddieson and Ladefoged (1988) thaw that this is necessary in the
production of compler stops. in order for tho acoultic cues of two
articulations to be h.ard.
In a segment like 101. the sequential ordering betwee" the two
gesture. may t. caused by tho f.ct that the two articulator. active
:" the produdion of t;'11 vowel have lim. freedom from each other.
Since most trllcings available In the literature are based on stilt
photogr2phl. often only on~ crose-sectIon of the s_gment's
I SI
articulation may be captured. In the cue at 101 It may be either the
v••1r or the pharyngeal gntur•.
The above expl2natian i. made more plau.lbl. by the
deacriptlon at uvuIDr,·. consonant. homorganic to 101. by Delattre
(1971). According 10 oelanro (p. 135. 137, etc.). uvulatl, whk:h .r.
r.rel, shown on the tracing. with two simuitaneoul constriction.,
Involve two g_.ture., that .r. executed lequentlally. On. g.ltu,.
Involvol the movemenl
the tongue root horizontally toward. the
pharynK, and the second gHtu,. Invalva the tongue riling along the
pharyng_" wall towardl the uvula. Since the two gestur.. .,. not
.lmultaneouI, Delettre IIlu.~t.. them using two tracing••
0'
Thll concludes the discussion of articulatory properttes of /01
VI. the phonological representation assigned to this vowel within
the constriction model. While • straightforward support for the
velar constriction in this vowel is not available, I have tried to
offer an exp'.natJon al to why an X-ray of /01 artk:ulatton might not
.twaYI rev.a; a velar constriction.
The proposed explanation appeals. among others. to the
possibility that there can be some variation between spe.kers or
between languages, in the articulatory strategy for executing a
given speech sound, particularly, when it il a complex segmet. this
sort of variBtion Is found throughout phonetic deserfptlons. For
example. the strategies for tensing vary even among the speakers of
one language (English), .1 reported b~ Bell-Berti et al. (1979). And,
if the feature ATR is yet another manifestation of (tense). which is
sa;ogcsted by the fact that no language has eYer been r~orted to use
the two dimen~ions, then this variation is even greater (compare,
for example. the discussion of ATR vowels by Undau (1978) and
Jacobson (1978).
• RICIII, thlt follow Ing f'1cClIrthy ( 1989). I consld.r uvuler, to b. ccmol.x
'Igmlntl. rlOr•• lnted wltta • Yller and PMrvno•• J con.trlctlons, Ix.cut.d by th.
ton9u. body and tonCJUI root r,sD.ctlvtlv.
152
0'
AI far •• the articulation
101 II concemMt. the important
conclusion ill. that thl. wariatlon doe. not exCHd the bounda,ie.
Imposed on this vowel by the phonological charectertz8tton: no one
hal yet reported the .xistence 0' an 101·1.. vowel with a palatal
conltrlctlon.
5.3.4
Articulatory Basil 0' the PharyngHI Site in Vowels
Th. vowell which I propos. to repreSNt with the Pharyngeal
Site ar. lei. loJ. IfJI. I . I. la/. .Ic. I hey. dllcu•••d the
artlcul.tory properties of 101 In 101M delt.i!1 In the prcedlng section.
and conctuded that the apparent cvunhtntlampktl to the el.lm that
IQ/
both velar and pharyngeal hbe • natural ......tion.
'I
With reapect to lei -- YO'" which I propoM to repre••nt .1 _
complex segment with a palaltal and pharyngeal conltrietlonl the
.1
L
articulatory evidence Is sometimes .. unclear
it II In the case of
101. Although there seem to be no .Klmple. of the two articulations
merging halfway. (which Indicate. th.t there may not be lefl with
a vetar constriction), lei onen appears to be constricted more in
the pa..taJ than in the pharyng. .' reolon. (Thll might be ~. r.asan
why Wood (1982) tr••ts thele vawctll ~a p...
only.) For e.ample.
this Is how 181 in German end in French appears on tracings (M.rtens
(1983). and Bothe,., at al. (1918). respectively).
ta.
The pharyngeal constriction appears Ie; be more pronounced In
IiI, the lax counterpart of lei. This i. shown both in G.~n and In
French (same r.f.rencn). as weH as in DhoLuo {J acobson (1978)). A
possible r•••on for this con trait might be that muscl. tensing in
lei results in an upvrard may.".,,' of tongue body, which alIa pulls
the tongue root. Obviously, lax yoweis do not Involve tho tongue
bunching gesture, and .1 a rnult. they can be articulatorily true, to
thoir phonological representation.
In languages which do not employ the tense/lax contrast. such
for example Polish, the tongue body and the tongue root appear to
be equidistant from the hard palate and the pharyng8~1 wall
8.
IS3
r••pttetively (ct. Wlerzchowaka (1985). p. 88). and
constrlctlone appears to be ~ery tight.
that
neath.
of the
The ovitrall plctur. that emerg.. from th... ob.Natlon, II
Cfin be •••" with • pheryng••1 conl'rlctlon whenever
'el
tensn... doH not counter that geature.
Let us now turn to the 'amity Cif 1&1 voweta. On the model I 8m
the.. Yowell.'. char.ct.rlzed with • .'ngi.
propollng,
conltrlctlon, .xecuted In the pharynx. While the exact conltrlctlon
location in these vowell ,.y vary from the upp« pharyngeal to mid
no Iw·llke vowell with •
and Jow pharyngeal locations. thor.
nen-pharyngea' constriction. In 5.4, I shall argue that the varlatlan
In the conltrlction location in the laI-like vowe'. II • conlequence
of the fact that the.. vowell may be articulated with either the
Tongue Body or the Tongue Root. Given the Mammy of the tongue. the
.r.
Tongue Body constriction will molt Iik.l~ tt. opposite 0' the uppormid pharyngeal wall. and the Tongue Root constriction wUI mOlt
likely face the lowe!" pharyngeal wall. I will show In 5.4 thai thl,
difference is phonologk:ally relevant.
The concluding remark for thl. sub-section is that while !ome
inconsistencies betw•• n the posited phonological representations
and the articuletory r••llty may be observed. the overall phonetic
picture lends support to the phonological feature Pharyngeal Sits.
which maps onto the articulatory region pharyngeal .cll. The
observed inconsistencies have be.n explained 8S a fAsult of one
articulatory gesture oYfJrridlng the effects of another.
Let us summarize the present section briefly. and comment on
some of the proposal. that it Introducad. The main pUrpctSl of this
section ha. been to defend the anJculatory balll of !he Site
features in vowels: to show that the.e feature. are phonetically
sound. I have considered two kinds of apparent counterexample. to
tnis claim: cases In which a postulated complex segment was shown
with a single constriction. and cases In which • segment was
art!culated with a constriction different from the one that was
IS4
0'
po.ltod~ The fI,s' group
caunt.,.••mplM wer. Jhown to be •
,••u"
either imperfection.
s.1I1 photography. which cannot
rep,...nt sequ=ntlally ordered .rttculattonl, or . .,. cauMd by the
counteracting
of anoth.r articulatory a"lure..
0'
0'
."tid
The S8COnd group of coun~.amp". invotved articulations
shifted from the postulated point of artk:ulalloft. I~ I('me of such
casH. I have bMn able to Ihow that the phonology of the segments
in question reflected the re" point of articulation. not the one
posited. And so. an 8nt8rior N in ThaI h& t»en IhGwn 10 btthave like
an anterior segment. An anterior III In Polish hu been shown to
pattern with the palaml and the
or NgrMntl, but ~, .ffect
hal been attributed to th. fact that anteriDrtty is • d8rived PNpertr
of the Polish iii. A uvular lui in German ha. bMn .hofrlfl to pattem
with other pharyng.al vowels C••eluding /., .net III) by tr~erlno
Achlaut.
.n.. .
approach•• pre-palatal (anterior) regions. Wood (1979) polntl out
that such vowel. are charaderlzed by • considerably high.r Fa than
ttt. palatal lifl. Thi. is born out by hi, acoustic data far /U In
Egyptian Arabie. Also. the IV In Rusalan, reported by Fant (1185) to
have a IIpr.-p••ta'· constriction Is allo characterized by • hlgho,
than an average third formant (reported at 3200 cpI by Fan' (1910),
p. 109). Wood
points out tha; an lui-like vowel with • POlt·
Y.'.r constriction II likely 10 surface with an F. lower th.n an
ordinary, vel., lui. The acoultic data for C'1trman (Ferrari Diane,
(1983). p.. 38) show that the lui In thl. language II Indeed
characterized by a lower Ft than ttl Engll.h counterpart. What theM
facts show, i. that -dlsplacecr vo.... nHd not rety on phonological
evidence for their articulatory ,.f.renc., because they should be
porceived .1 acoustically distinct by the language learner.
.'10
Given that /., I (an anterior IV) Is acoustically dl.tinet rrom
IV (a palatal Iii), to the oxtent that the language leamer can ..sign
While such anal,••• deliver the d••ired results. ihey raise
im,~rtant qucastlons. whach need 10 be answered bt»fOf. the
anal,... .,. adopt". For example. even if It II p".11bkt to repr.sent
an anterior IU phonologa't" iM)w il the language 'nmn to know
which Iii, an anterior. or a paf8ta1 one. he or :he hears. One aMWef
to this question might be tNlt unlet.. Itt.... il phenological evidence
that establish.. U. representation of • vowel. !he Site of the vowel
default specification
is I." unspecified. and ev.ntally receive. 8
(•.G.. P8latal for a segfMnt .rtlcu~.ted with th. Tongue Body
dorr.inating featur•• (+high, -back). An unfortunate consequence of
Stich • solution is that in lenguage. which lack the necessary
phonological r.'erence, yowoli should alw8ys surface with their
supposed default v.luttl. As we have already ...n in Ru••ian,
Chin.... elC.. this is not true. Alt.rll8tive1,. on. could allow the
solution proposed above. with an addcDd fector
dlsp....lon: in
~U80" which lack !he _
~. but have more than
two high Yowels. the default Site for IV would alw.y. be Anlerior.
The only plausibl. explanatJo." I can think of is that while the
acoustic signals assodated with these sounds are distinct enough to
be correctly classified by a language learner (or even by an adult,
who easily detecls ·forei!Jn accent-), the acoustic distance between
them may not b1t sufficient to support a phonemic contrast.
It is not clear. however. thai eve:. this so!ution il necelsary.
!n ~is discussion of IiI-like Yowels whose constriction location
two
some
0'
• correct repr••ent.tion to these vowe'. on the basi. of the
acoustic signal. the next question that comes te mind Is why /., I
and Iii ,re neve, contrasted in the same Jangu8ge (at lo.st not in
any known language). Since such • gap Is not IIke5, to be eecidantal.
it d...",•• an expianatlon.
0' course. this is only a hypothesis. which can only be
confirmed by a broad scale study comparing acoustic distances
among the sounds which Iingulgss ordinarily contralt. with the
Beaustlc d!lt8nc. between , •• , and N, velar and uvular lUI, and so
on.
As a final remark, I would like to point cut that casel in which
phonologically distinct represantations fall to support a
156
ph~nemk:
contr••, are not limited to the on.. described above. For
e.ample, any grammrr should be capabl. of .s.igning d;stlnct
phonological repr...nteltonl to • p.l.tallz~ n...1 (/n'I), a coronal
do,..,
paletal na..l, and a palatal
naeal. As tar .s I know, no
languige employs this thr. .way contrast.
5.4
Articulator Fealur.. in the Representation of Vowel.
In this section I ciiitCUBI various lasue. ,elevent to the concept
of repr•••"ti"g vowels In terml 0' Articulator. -- phonological
'eawr.. Introduced bt S~ey (1988), who.. phonetic basis ate the
arg.n. active in the conctrtctOon guture. While the an.lomical
organs which eX9Cute the con8trlctlonl .r. tho lower Up and the
longue, Sagey adopts the id.. in Chomsky a.id Halle (1988),
ysveloped more explicitly in ....11. (1983). to further ,ubdivide the
longus into parts that can move r.lelly.,y independently of each
othe~. And so. she pOlitI the coron.' articulator, who•• anatomical
corre'.te is the tongue tip with the tongue blade, and the dor.al
articulator, whose anatomie81 correl.te il the tongue body.
Recently, this system of articulator. hal been enriched by the
additfon of the radical aFtlcu'ator• which corresponds to the root of
the tongue, by Ladefogod and M.tdlesnn (1988).
I adopt Sag.y's concept of the articulator, and with the
modification introduced by Lad.'oged and Maddie.on (1986), I
assume thai there are four diltinct articulator features: lower Lip,
Tonoue BIcxIe. Tongue Body, and Tongue Root. Solow, I disetms various
issues related 10 representing vowell in terms of Ar!lculator
'.atures: I consider the phonetic b.ail of the.. f8atur8. whenever Ii
controversy exists, and I dlsccis the (ole of the Articulators witt:in
the feature system which already cont~lns the Sitos.
5.4.1
The Tongoo Body
I assume thai the vowels which are represented in terms of
the "-onQU8 Body are: Iii, /1/, lUi, lui, 10/. and leI. /a/. ;0/, I:JI,
157
/1/·, I,. /. Th~1 'eaves 181, and possibly, its round counterpart. lDi,
as the only vowell nol repre.ented wit., this artlcul.tor.
I foUow the tradition of associat~ng feature! (h!gh) and iback)·
with the position of the tongue, and after Sagey (1988), I ••Iume
that the.. feature. are dominated by the dorsal articulator (the
Tongue Body). I leBve out of the pr818nt discussion the f••turs2
[ten••) and (ATR). since it is not c~.ar to me where th... f••tur••
(or f•• tu,., should (ATR] turn out to be an alter ego of [tense): se.
Halle and Clementi (1983) for a suggestion 10 this effect) bolong In
a segmental tr...•
In what folloWl, I discuss some of the articulatory aspects of
(high) and [back]. and, on the phonological side. I present Brgumenll
that these features are needed for the purpos.. of vowel
representation, in addition to the S!te features.
5.4. 1. 1
The Feature (high)
While participation of the tongue body in vowel production is
rarely if ever questioned, the articulatory basis of the featur••
• /a/ repr••lnt, I mid yow.1 higher than 1..1
• At thl. point. I rlfreln from Iny comltmlnt to thl ft,tur.nDwl. Llt,r on, I
conlld.r varioul pro'S and con's of 'ncludlng lIes.,.- I among thl Phonologlcl' f •• turl'.
• Nllther phonological nor phon,t Ie conlldorotlen. ert h,ID'U; In d.'.rmlnlno the
stltUI of Itlnlll and IATRI. For 1.lmpll, If tho.t era two dlf'eronl f •• turel. It II
not ell" w.uA t"IV ceMot eODCCUt In .. Ilngll languege. If. on the oth.r hind, they
or. 8 ilngll fl.tur •. thllr erUcul.to~u dlverllt~ Dtcoml. puzzling: whln f •• turt
Itlnl.1 WI' arlglnea" Introduced (Jakob.on end HIli. (1956». It WI' dlfln.d In
tlrms of thl tongue mUlcl. tln,lnG. Sinel thtn. I numtttr of dUflr,nt Drooo,," for
dl nerlnt languIO's hev. bien med.: Lldo'ogld ( 1964) hal pointed out that tl!~
mlchenllm b.h;nd en ODDMent tenl,-Iex dlaltncUon In IQbo II thl tongue root
ecl\,lnclm.nt. Llndlu ( 19715) hal dllcov,r'd thet thl m.chanlsm of tongue root
r.(raeUon mey bl secomplnlld b,. lervn. rol.lng In AkIn. Jlcoblon ( 1978>, who
studlel the corrllat,. of thl harmonic ,••turo In Dholuo r'Dor •• thai IP.llelr. ert
divided Into two grouDI. d.pendlng on whltt,.r thl" U.I tongue hllght or tongue
root ~dYeneum.nt for tho 11m. narmonlc ,f'lel. Int.r•• tlngly. h. doe. not obl,rYI
lory". mov,ment Illocleted with tho harmonic :.otur•. O'I1-B,rtl It II. (1979)
rlPort that the sD,ak.rs or Engllih are dlvldld Into dUrer,"t categorll', dlDendlng
on whlthtr th.V u•• tongul height or mUlcle tlnllng to product ten•• -Iax
contr.lt.
IS8
which .:8 dominated by th. Tongue Body articulatar are not alw8YI
obvious: the mngue arch model. despite ill luo=Hl in phonology, h••
bun criticized 'or itl 8rtlculatory inaccuracy, virtually linee Its
very beginning. The main obfectlon ralHd by tho researche,. (Mey.r
(1910). RU.I.I (1928). JOOI (1tta). ste.) hal been that the mode'
Incorrectly predicts the I•• vowel '&1 to have • higher tongue arch
position than the ten•• vowe' lei. On the basis of the articulatory
data from German, Dutch, Swedish. and EnoHsh. the, h8ve shown that
the vowel ILl. IUPPOled!y cha~rlzed by I+hlgh) tongu.. position.
consistently appe.,. with the tangue body lower than the ten••
vowel
In languages thlt contain both.
I.'
The relationship between tho real 800 the phonoltJgicai height
of Yowels h.. been Inve.tlgated mont recently by Sidney Wood
(1979. 1982), "!hOM deta constlt of multiple X-rey Ir.clllO_ of
vowels from '.5 langu.g••, som. of t~...m obtained with modern
ctneradiographlc methods to ••Iur. high accuracy. Wood's study
confirms the obleN.tlct" madlt by Me,er, Au...', Joos. etc., that the
tongue height predicted by the tongue arch model doe. not
correspond
to the phonetic data.
Wood considers this flaw sufficient to dispense
tangue arch model. Inst••d. he proposes a model
articulation which characteriz.. vow.!. in terms
constriction location, degree of jaw opening. and muscle
(see section 5.1). While I adopt Wooers proposal to treat
terms 01 the constriction location f•• tures. I take
sympathetic vi" of the tongue arch model.
with the
of vowel
of the
lensenes!I
vowels in
a morl
AI noted by Halle (1983). the tongue arch mod~ has persisted
the banis of vowel description in phonology "and phonstics
textbooks, despite ttle fact that the controve,ty InvolVing th.
feature (high) has baon known since the beginning of the century.
.una points out thai there i3 a very good reason for the model's
popularity r namely. grammars of iangu8get provide many examples
of phonolog;cal rules which &etivaiPt tho [height) feature, and which
treat vewels Iii, Il/ as high. and le/, II'
non-high.
81
8'
159
In similar vein, the,. are many e.ample. of rul.. (e.g.. height
harmonies) which group \fo.els IV. lUI on one hend, ~nd
101 on
the other hand. into natural cillsel. The model which charecrtertz••
vowel articulation solely in terms of the constriction location.
cannot repr•••nt vowels like III and lui u membera of • natural
cia.,.
I.'.
Obviously. Wood (1982) II not una••,. of this: while he
rejects the tongue height fe.tur. in phonetlcl and phonology. he
propos.. to account for the phonologal Gener.llzations which thl.
feature handlas, in t.rml of • f••tur. (open]. who.. articulatory
basil II the dear•• of jaw opening. He ch.raeterlz•• IV, ",. It/,
It'. Iii I. lui. 10 I. etc.. al (-open), and lei. 'II, 101. I'J I. 'II, 10,/.
etc.• U l+open).
However. the feature (open). whUe II may appear to solYe Iho
articulltory difficulties of the feature [high]. hal problems of Its
own: in the every-day speech, the jaw movements accompany vowel
production. generally, In conformity with Wood', predlc!lons.
However, experiments in which subjects have b••n I.ked 10
pronounce dlHerent quality Yowels while their jaws were locked in
a fixed position (ct. Llnblom (1989)), show that the variation in jaw
opening might facilitate the pronunciation of dUferent quality
Yowels, but that it Is by no means a necessary component of Yowel
production- .
The feature [open) faces even greater difficulty :n phonology.
particularly In the phonology which posIts highly structured
representations. Since it is not associated with any Articulator (nor
with a Sito. if the constriction model is considered). Its only
possible location can be the Root Node (or at IGBst a node which
dominates the Constriction Node). In such a ca,e, the festur. [open)
ought to be equally relavsn! in the representation of segments
articulated with eith" Lower Lip, Tongue Blads or the Tongue Root•
• I am grotoful to
ICI,~ St.':.nl for
Dolntlno thlt out to mi. t!nd for dlmDn,treting
thllKplrlmr.snt
160
•• It t. In the representation of sounds articulated with the Tongue
Body. Ho. ..,.,. • shown in 2. 3. the height phenorneM do not occur
with non-dorsa. segments: neither corona'i nor labial. ever trigger
height ch8"Oes in yo. .ls. And••• I will show 'ater on, legments
rapr•••ntttd with the Tongue Root .rtlculator ar. equally n.utral
w.r.~. height phenomena.
Lot us summarize Ih. dlecu••lon so far: while phonological
conlid8rations _rgu8 for inetudlng tho tongue height f••ture in the
repre.ent.tlon ot wowell, .rtlcul.torlly the tongufr arch model
app••
to be In~ccur.t., a' I.... •• far •• thle f••lur. Is
corarnfKI. Even though there IIPpe.... to be no substitute for (high)
(in view of what hal been observed about tho deg," ot Jaw opening
r.
'eatur. of Wood (1982)), the articulatory Ineccuracy nevertheless
pose, I! problem for a modei of feature r.pr••entatlon which
emploYil the tongue body '.atur.. (such .1, for e.ample. th~
constriction model).
A possibl. solution to th~. problem has been suggested by
Flscher-JIJrgenssn (1985), who ahlo recognizes the expl8natory
power of the feature [high). and defenda the statu. of this feature
ageinst Wood's critlciams. Sha pointe out that un••pecled
differences in tongue height are observed only when tens8 vowels
are compared with lax \;owell. naver when the vow.ls are compared
within the tense and Ial IQtI. She Sugg.lts treating ten.eness 8S a
leparatfk, dimension. independ&nt of ~ight: on thl. view. III and 10/
are not compared with ,., and 10/, but with 'al and ",, against
which thG)' are indeed [+hlgh).
In a way this explanation shifts the problem elsewher. instead
of solving it -- after all the f••tur. (tGnsel is not well undfirSlood
either. However, it is acewptable In SO far 81 there doel elist a
phonological feature (18"99), which artlcuJatorily interacts with the
'9alur. (high). S!nca this sxpfanatlon salvage. the concept of
representing vowefs in terms of the featurs [high). associated with
!hG positions the Tongue Body articulator assumes during the
production of different vowehl. I adopt it.
i 61
I now turn to the que.tion of
f•• lure (high). SInce this f••tur. il
all that remains to be shown is that
repr•• entlng vowell In .ddltlon to
phonolaoical motivation for the
wen ••tablished In phonolog,.
it II! needed for the purpo.. of
Site featur••.
Tradltlonaly, the feature (high) has divided vowell Into two
cl•••••: (+hlgh) I, I, U. 0, ii, w, ., etc., anG (-high) e. I, 0, " ii, At U
wen aa the so called -lowG yowell. I adopt thl, dl.tlnction with one
exception. nemely, I do not tr.at the vowel I~ in terml of the
f.atur. [high). .Inee I do nat conlkler it to be articulated with th.
Tongue Body.
Given the system of Site f.atures proposed In thll th.sll, the
need for the feature [high) I. not Immediately obvious, and mUI' be
reex.mined. While the,. i! little doubt thl!t phunologlee of languages
require 'hit feature (or itl equivalent) to explain such phenomena as
haight harmonies (er. the [+high)l[-h'ah) vow~1 harmony In P.llego
(McCarthy (1984», [+high) vowel harmony In Menomlnl (Bloomfield
(1939), Col. (1988), Ste,iade (1987))), or heiOht dlsharmcnl•• (ct.
height disharmony In Ngbaka and In Ainu (Ito (1984))), It is
reasonabls to ask whether the systam of site futuros dOH not mako
this feature superfluous.
An t'xaminatlon of the representations which the constriction
model assigns to vowels reveal. thai the value of the feature (high)
for any vow.~ can be predicted from Its constriction locatlon(s): all
Yowels with the pharyngeal component (s. 8, o. ~. 1\,
tit. 0, ate.) are
[-high), and aU vowels without the pharyngeel component (I, I. U. 0,
u, t. etc.) are [+high). This showl that til. value of (high] Is redundant
in Yowels. At the same timB. however, It is clear that onty those
segments which share the pharyngeal specSficstion can be
characterized as a natural class without a reference to height
Seoments which ar~ redundantl)' [+hlgh) do not share a phonological
featvre which would characterize them as a natural clmss, and
which could replace [high): Iii,
I. and III. are Palatal/Dorsal,
o.
lu
162
lu/-, Itl, and /01 •• Velar/Dorsal. etc. Since there •• ilt phenomena
Ce.D.. vowel harmon, In PuJego (McCarthy (198.»)) in which both
valuea of (high) ar. aetIwt, It falIoWI that th,~ f••tur. mUI' remain
In the ....,....,tI\tton of va.....
5.•.1.2
The
Featur. It*kl
Traditionally, the f••tur. (back) groups "ow8ls into the
following cia....: (-b8ck! I, t, I, I, Ii. 0, ., etc., and l+b8Ck) t, U, 0, Q,
.te. AI f.r a. I know, the,. Ie IIttta controver.y cance,ning the
artlcu~.tory .ccuracy of thll f••tur.-. Th.,.tore, this lectlon w:1I
bo concerned lal.ly with the qu••tlon wh.ther the,. 8.isls
Independent motivation for p,.ckJ In the f••tur. 'yltem which
contain. the Sitel.
A comparison tHltwoen tho segm8ntt t'.ditton.lly treated as (back) (I, I. I, I, ii, ii, ., etc.) and the 'Igments which I propose to
reproNnt with the Palatal Site (I. I, I, I, ii, 0, etc.) reveals that the
b!O 1811 overlap canlidorably. In ~. only ths vowe' ,., remains
•• • (-back] seament which il nol _tal. In a cue Uk. this it is
renonable to aile whether the 'oeture (-b8ck] could, in the interest
of economy. be replaced with the feature Palatal Site.
I want to suggest that despito • n.ar-pe~ect overlap between
tho
(-back) and palatal segments, ther. are good reasons for maintaining
the f&atur. [back) in addition to Palatat Site.
In order for the Palatal SilO to take over (ba:k), a number ot
conditions would have to be met: first It would h8V8 to be shown
that mil processes analyZed earU., en lerm8 of [-back) •• a feature
dominated by the To~u. Body Articulator - can M handlad In termo
of it f••ture which re'.rs to the conltrfction location. Stacond, it
Wbuld h8ve to be. shown that the... sr- no process.. which require
• Not COUfttt~ lui In IIftDUGpe Ut, Germlft.
C,"
While WOOd, 10112) rejects It
5.1). he doel to tvaly Decluel hi r'Ject. tha
tGftlU'lIrCII model, w Ilh wfttch thll '.Itere I. . .tOctltlJ.
(I
163
'.1,
the feature (+backJ. Finally,
since it palterns with palatal
vo.ela (see below), would have to be treated ., • palatal segment,
and represented in such a w.y as tD contrast with lei and ,.,. Below
I show that none of the.. conditions can be met.
In 7.1, I argue that because of their diflerent position, in the
segmental tr••, the two types of feature. .- an Artlculatordominated f••ture and a Site fe.tur" should reveal rather diffefent
behavior in phonological procelses. Since ihe SIt8 feature is
dominated directly by tha Constriction Node, and since all (fully
specified) segments contain this node. then, on the assumption that
all phonological proc81s81 are strictly local {ef. Steriade (1988».
there should be no case. of -segment skipping- spreading of this
feature. This restriction doe. not apply to the artitulator·dDmlnated
features, in this cas. the faature [back). Since it is dominated
directly by the Tongue Body, it may propagate across any segment
that is not specified for the Tongue Body articulator. Evidonce from
languages such as Chamorro (Topping (1968)). or Hungarian (Kiparsky
(1988), Vago (1976), Ringen (1980), etc.). in which two or more
vowell agree in backness (or palatality) without involving the
consonants. argues for the feature [back) Independent of Palatal Site.
CI8srly, the harmony processos in such languages cannot be
reanalized in terms of the feature Palatal Site.
Another argument against eliminating the feature (back) comes
from languages in which [+backJ is acti\t8 in phonological processes.
An 8~ample of such a language is Finnish, for which a process of
(+backl harmony has been motivated by Kiparsky (1981) (as well as
Steriade (1987), \vho adopts this aspect of Kiparsky's analysis).
Since vowels such as lui. /01. and 101 do not share th& constriction
location or the heig~t feature. they can be characterized as a natural
etass only with the feature [+back).
In a similar fashion. the vowel 1!lJ I. and its patterning with
palatal vowels in the harmony processes (e.g., Chamorro. as ana'yzed
by Kenctowicz and Kisseberth (1979)). constitutes a prodlem for a
feature systom without [bec!tJ. ThufB are both phonetic and
164
phonological ,•••onl for not tr••ting it al a palala. vowel:
phonetically -- It dao' nol Involv. a pal.'" conltrlctton (c.t. Wood
(1182»): phonologically
oyen though it pattern, with p.'atal
vowael In Ih. harmony pracel"', It falll to trlgge, pal.MUlatlon- .
Thereto,., It cannot be reprasamted with the f••lur. PI.tal Sit•.
The only ether feature that can pIBe It among vowell such a. lei,
N. lu I, lie., II ttl. ~tu... (-beck).
_0
On the baM of the abov. argurMnll It II poaibltt to conclude
that the f••tur. (bleli) II n8eded far the purpo... at vowel
,.,..ntatlon In addition to ths 81:0 "turn.
5.4.1.3
The Feature (!ow)
In this Iectlon I -roue again.' the f••tur. !Iow). I show that
1M natural cl... en.ctl amona leGmen..
(+Iow) can be
ClIP"'''' by tH "tuAt Pharyngeal Site. and I rOanlilyze an apparent
••ample of the phonologic.JOy ~. (-!ow],
,-fled ••
To my kn~ •• there are no atteltH case. of [+10_) vowel
hlmiOnlel. No languago hal e rule which converts aU vowell Ie
1.,
(+Iow) v"hen.v.r there il laI,
or 101 in a word. this is an
unexplained gap if Pow) is a 'cs8turw dom'naled by the TO"Gue Body
Artlculetor.
DI18imiiatory procell.. do occur among
t~8
so called -low-
vo..... For e••mple, In the Blzcayl!n dl818ct1 of 3uqu8 (ct. de Rijk
(1970», the first of two adjacent lar. tum. into leI. I.g.,
(18) siebe bot
'onl d5ughler'
elabl-a
"the daughter"
girl·
n'lkl-e
Ilere-I
'the gtrr
'thl shalt'
ne.ke bet
·r~1
' ••re bet
'one
.".or
A similar procell occurs en Wol••lan (Sohn (1975)): however. in this
f2M, • consonant may Interven. between tho two
-The rIDrI'Ift!IUon of 1,.1
e. IIlleu.ltd In mora delatlln 5.6.
165
yaferal
'my shoulder"
It il possible to analyze the dissimilation phenomena
illultr8ted above 81 an OCP effllCt applying et the Pharyngesl Site
Uer, without r.forence to tho feature [low). Of course. such an
analysis rail.1 an obvious quastion. namely why does the
dls.imilated vowol turn into leI, a segment which contillns the
offending pharyngeal component (given the representation of /el
propoled in this thesis)? A plausible explanation for tills fact might
be that the first step in the dissimilation procesl is the deletion of
the Pharyngeal Site of the first ~ow81. This rule produces a vowel
with en unlpecifitKt Site. a schwa (see 7.5). The u!timate conv3rslbn
of lal to lei can be attributed to the fact that in the five vowel
systems of Basque and Wolesian. /81 is articulatorily closest to
/sf: ;r is redundantly (-high). and it doe.· not involve lip rounding.
Stronger support for the feature [low) comes from the
process•• which raise 181 in the environment of non-low vowels. An
.xampl. of such a procesB is found in a great number of Basque
dlaleclll (ct. de Rljk (1970). Hualde (1988». In those dialects,
vowe's III and lui raise the fc~'ewing Ia! (either adjacent or
a'~ross a consonant) to leI. Nsither Ie/ nor 101 trigger this process:
(88) 8Qun-e
er'-kl
"the dey'
gllon-8
"throw'ng stones"
Dllole-seet 'throwing 6 ball'
'the men"
The raising phenomena of th:s sort seem to imply an asymmetry
beIWet'" the
and the -." vll1lue of the feature [low). On the one
hand, the absence of -lowering- harmonies suggests that there is no
festure (+Iow): on the other hand. the raising of laJ after high
vowel. seems to SU;g0St that there is a feature (·'ow). While the
possibility that there might be such a feature cannot be completely
-+-
vowell:
mfltlmamt 'our Iyl'·
(87) mite'
Jetere
'shoulder'
yaferlmamt
'our shoulder'
• NOlt IIkoly, the .eml m.ehsnlsm tl r.spon,lbl' for the fact that In meny
1ongulg••, la/curfaeo. as an ODlnthltlc YOWIL d••olta the fect thlllt It II not
IlmD'1 from the mrtlculotory point of vl.w.
166
r.".'rail
ruled out. •
of the dAta in (81) aMY. would be more
welcome on p(ti1dpIed groundl.
lax vowell in the sem. word). ThslG facts mske the an.l~s'8 I)f
vowel ,.ilin1 in Baequ8 sugge.ted abovo mar" plausible.
~ pasllble . .y to anetyze th. data In (88) without ref.rence
10 I-low) would be to •••urne a !JI'OCIII which tpl'Mdl ten• ."... of
111 Mel lui onto Ial. Slr1C4 Itt...
nc (+ten"l counte"." of /., In
In vie. of what has been discussed in this section, the
phonological behavior .If the featu!. (lOW] can be summarized 8S
JOUOWI: In its positive value, [low) '.UI to pattern Uk. an
artlculalor-dominated '••Cur.. •• mOlt notably, ~t faUs to trigger
vowel harmonle"o Indeed, [+lowJ actl more Uk. a feature dominated
directly by the Constriction Node •• it
r,ot skip 8egment. whon
It spr•• ds, yet It 'I vi,ible IcrOSI non-low segments in
dissimilatory proceSSt'I. in ;t. negative value, (low) does act like an
articulator-dominated '••tur., in that it spreads in the harmon;~
fashion. Howevar, it iaU. to pattern with (+IC'w) in that it re(Jver (at
least to my knowledge) participates in a diSiimiiatory process. In
other words_ ~8re are no examples of Iii, Jul. lei, or 10/ turning to
181 when followed by either Iii, lui, /8/, or 10/.
8DIque. ten..
/w
'I
II the n••t pla.1bIe alllmattve_
The above anl',sll, while not parflCUla,ly ••" molly.ted on
thai the mid front vowel
the ground. of Basque (excep! for th.
In this lenguaage I. indeed (+ten..) (d.
'act
"Diey.
(1170), and not (-
"""1 as would O.""'"Y be expect8d In a Dye vowel .ystem,
cf.
Cro"'.,. (1978»), ,**"1 mont IUpport from tn. comparison with
!gbkra ... another language In wh~ 181 il raIIMj to lei_
In Igbir,. (vowe' inventory: i, I, I, I, U. Q. 0, '. al, tho low
vowel II railed not Just in the .C'~'1aronment of high vowel. (in fact
the lax high yowsi! hav. no .ffect on It), but before any tgnse vowel
of the 18ngu8ge. Thil II shown below (I quote tho data after Bls.antz
eI eI (1187)):
(89) mlzt
mlZI
milo
metu
-, IMPlct'
-Ism weir
-, !rrllng.'
-. bOlt'
mell
-I em tn petn°
mazi
0; egr.,-
met,
0'
iJ~Ck.
mlto
OJ
send-
In (81), 1M 1st porson prefix is show to alternate be!Wsen leI and
/8/ In what appears to bit II de.r ~);.mpte of a lenilln... harmony.
Whet argua against the lax~ng ar.aiys!1 is the fact that /81 has e lax
couMerpart in thl Ilngu_g., !I/. Therefore, it would txt difficult ic
deiiv. JaJ from lei via 8 laxing ru'&. Deriving 16/ from laJ via
t.nllno :1 s,mp'.r, beeaus. 181 hili no Isnle cOllnterpart in ths
langultge_ Presumably, th& oa.'tput of tli. tensing of laJ is reanalyzed
80 81 to coSifoam to 8:. phonemic inventory of th" l&ngu8ge.
The Igbirr. facti constitute In example of a PrccGss which has
surface effect ~f r~lling /8/ to leI, in whh:h the feature [-low)
pillys no r~l. (aG chown by the 11:;\ .fBt lai co·.xists with non·low
•
161
do..
This behavior of (low) makes it a very peculiar feature, which
not resemble any other feature in the system. Of course, the
aspects of the behavior of (low) which appear pecuiiar in a single
artlculstor·dom 1 naQed fe.ture, are quite eonsistent with the
behavior of two t'eatures: Pharyng8~1 Site and [iensoJ. A reanalysis
of 1+low) in !erms of the Pharyngeal Site explains why th6r& are no
vowel harmony processes which spread !+Iowj .- aft(lr aU, no Site
fbslure acts en such away ° A reanalysis of (.lowJ in terms of
I+lens8) capture; the facts of vowel raising (as in Basque and
Igbirr$i), and explains why [·lowJ does not behave in any way like
{..low}. With these reanalyses, tho motivation for the fSilure [low)
disappears.
d~el
5.4.2
The Lower Up
While the role of the lower lip in vowel production is
uncontroversial, it is not an easy task to demonstrate that this
articulat~r is acti~e in vowel phtlnology. ObvIously, this is nDi
because labial vowels are somshow immune to phonological
processes. However, in order for the processes which manipulate
168
~:h ~. 10 be urcamtiguou~ Inte,1)re*, a. Involving the LQ:lfer
Up Art50uIidDt (U1d not the Labial Site). they wo·sld have t~ occur in
IanVIJ8G" in ~lch Ihtt cIaa 0' sounds artk:uleled with thg Lower
Up It waer than
clan of sound. invo~eng the labial Site. At
thIa potnt. I am net aware of the ex.~ of the l8t);al phllnort'lena
in bl~ ~at 111M! thlts condition. HoMIft'. molt lilte.y. this ~ap
Ie a ~ IM:ddont. slnca there IU.. eba, calel of proeesses
InvoMng ,.tfal consonantl ~ muat be analyzed in tarms of the
lo\..., Up \rticulato: (1M Z.. ~I}.
"'w
5.4.3
T~~·
Neither t~ IPA faature !Sy.~am nor
have mae;" allo•• nc.l for tr.atlng
t'"
~woJ.
orig;~1
SPE proposal
in tbrms of features
lh. proposal to treel front vowel. es coronal soems
problematic from tt''t phonological point of view. First it draws
support frem the definition of the feature (coronal) proposed In SPE.
a~'O:ding to which -coronal sounds are prodLh-~ with th. blade of
the tongue raised from its neutral positJort- (d. Clemtlntl (1978), Ito
and Uester (1989)). However, such i:l usa :If [ecronal) ignores what
appear. to be a univenlJal restriction on the articulatory t88tlJre~.
namely, that they may be phonologically rei6var:t ordy if they
characterize articulatory slates at the constriction. Since the
tongLt. b~a~ des! not participate in the constriction gestule of a
fron~ vo~ef (~S attelted by the X-ray evidence). the feature [coronal)
would have to c ,nstilule the 501& exe.ption to the above
g.nerali:atlon.
e
•• Ioct&.~ w!th the tangue ~1.d.Jllp srtlculatcr • The earliast
pmpclIIl ttl characterize Yowele in t.rml of coro,,;Uly is due to
ClorMttfD {1918;, who charaderized ~ni Yowel• •, [...coronal]. Ito
and tAe,;tor (1 981) int.rpr,,~ Cioments· 'j:!po.a~ within the
ille:arch!cal system or segment repr...ntation of SL~.Y (1~98).
The mein mutivation behind CI~mb~~' (as well al Ito and
Ynter's) p~saJ luI bee~ to provide In adequate sccount of the
palatalization phenomena. Clem~tl poinli out that the Invariability
of those phenomena with rerpeci to the trigger (front vowels) and
the output (coronal consonants). al well al the frequency of
occurrerril. suggest a possibility of an assimilation-based analysis.
Sinee such an .nalysi~ il not ayanabl. ;."'ithin U,. SPE feature
Iyltem (a. frenl yonis and paiat.i ~nlon ,nts de not fOriTI a
natliraJ c18u in this system), he ;m...~,:... .0 19ht8dy lhe situation by
tr•• ting front vowels as corona; (in addition to [-back)). and
palat.Uzatlon as caronallzation w~ich p1'8S0i"81 the consonant's
specification for [anterior).
------ai,ev••ton
• Thl
(1959).
In tht. I.cl II)!) 'I • It\ort~~ vertlon
or
the dltcUIDlcn In Gor~tte
Accordi,10 to Clements (1976) ar:i Ito and Mester (1989), front
vowels are both (+coronal) and [-back). In SagElY's (1988) terms this
trans!ttleS into two possibilities: either frent vnw"is are
both
cQronal and dorsal, with tne Coronal Node dom;naling the f'3st:Jre
(an~~:iorJ. and the Oorsa' NodfJ dominating (back) and [high). fjr eise.
the Coronal ;~oda is the seat of at: froh~ vOYrsl features. including
(high) and (back) (in fact Ita and f\1estar adopt ths latter 'J;ew).
Both vi9W~ have obVIOUS undesirable consequences: if front
vowels t:aSi be coronal in addition to being dorsal. why canr.ot back
vowels l"'e coronai ~s well (and also trigger palatalizarion)? If. on
the O~h8'- hand. front "owe's ~re just coronal. and back vJwe:~ are
just dorsal, then. given th!' treatmar! ot compl@x segments pro.,osed
by Ssgey (198S). based on the insight that any MO independent
articu~ators can CIt' combined in a complex segment, ~e should
expect to find complex vowlljls which are both frent and back. Olh~r
Gu6stions concern the representatjon of the paJatalized dentals.
found in Russian. Bulgaria". etc. (whose existence Ito and Mester
deny). the behavior 0: the coronal vs. palatalized segments in vowol
harmonies (simple Ct)ronaJ sounds ~': not block the spread of [·back)
hdrmo,n8S. but the palatalized sogmertis (includinO the palatalized
• WIl" • POts'bla c)CeeoUan or Uti 10 ~.lI.d ·thoteeIZtd- vowel$ which hay, b'ln
chr.cterlzoQ bV leam. Ce.g. ltdcfogRl.1' I PS2}) 'I r,iroflex.
169
170
coronda) do (ct. the analYll1
0' Turtdah
by CI.m~;4 and Sezer
(1182), Met 1M dllculcmn it1 3.3).
Itt..
0'
Evon
short list of problema ....rggwta that the tr••tment
front VOWOIa as coronal cannot bo eulJy Incorpa,.~ into modem
day phonology. n..ntfore. It fa .....nabl. to
whether there
au
.!TAItt sufflcf.nl pMnoJoglca1 mottvatlgn for luch e proposition.
90th Clementi and ftc and U••to, jUltlfy their propo.al by
demonltratlng Ita potentta' for ,olvlng the problem
_ . Aa far . . . know, tilla fa tNt only MotIY.~ion that hal
ftOI' bMn propoeed for the coronaJIty ot front vowell.
0'
10...
This naotivatlon
Ita strength whon considered In the
of the conltrictiofj model. ~Int» thle model recognize!
phonological f••tur.. ba,ed -)~ the constriction location of a
legment. It prcYide. the m,-_ns of analyzing the palatalization
phenomc.na (Ie. Gorecka (198~» in a ••y that captU~8S their
ailimll.-tory character. without alcumlnG that front vowels are
~ron.1. Briefly, the analysis ~raposed in this tt:alis treats
pal8taHzation u • process ytnieh spreade the palatal constric(jon of
• front y~.1 onto tha reo! node Df an adjacent consonant. The
immtldiate output of this prc..,_:s II a paJctalized ~ment. If such a
.~m.nt OCCUlt in the :Jndertying inventory of the language -noth1nv happens. The reduct:~n of • tXmplex segment to _ 8 simple
~"-tal sound takes pi8C8 if a laney.ge lacks the doUbly articulated
segment. Whether this sound il coronal or dorsal, depends again on
the Im.ntory of tile !a.'1gu8ge. The liiUllylis I propose predicts that if
• given language ha. doraai p.S.tals. then they win constitute the
final output of palatalization (this II what happens w~!h Margi
val." (Hoffman (~9S3». atnd with Basque dentals ~C;altar8I1i
(1988», ev.n though bcIth "ngu~.s hay. the coronal palatals too. If
the language lacks the dOllal palatals, the default art;culator fa,.
the Palatal site becomes tho Tongue Blade.
f~
The constric:ti9n model is also capable of explaining tha cases
of ·ant.rloriz.tionO~ quoted by Ciements (1978) as it primory
evidenc8 for the caronalily of front vowell. An examp'e 01 such a
i 71
~h&nomGnon.
involvinp a kltl alternation before frant vowell in
Polish. is discussed in 5.3.2.3. Thtt alternation is analyzed along the
lines propoud for pa.....llzstion. except thet the yowtlti triggering
anteriorizaUOn are Ir••tad 81 Anterior. not Palatal.
Finally. there .:cill phenomena involving front vowell and
antanor consonants. which can be explained only by appealing to ttl.
conBtriction location ,..
An .umple of such • phenomenon Is
the velarizatian process affecting If/. lsI. and ItJ before front
vowels in Thai (discussed in 5.3.2.1). This phenomenon cannot be
explained t:y appealing to the coronallty of fiOnt \fowels. because the
clal. of COrllOnanb which are velarized includes at Iablodltntal.
tu.....
In view of what has been said in this section. it is clear that
the constriction model dOdI noi need to appo81 to the coronality of
front Yowels in order to explain the palatalization phenomena.
5.4.4.
The TOl'lQue Root
The treatment of the Tonguo Root articl,;~atc,r I am suggesting
departs considerably from previous treatments. the main difference
being that I do not consider this articulator to be uniquely
associated with tho feature (ATR). Instead. I consider the Tongue
Root to be an Brti~lator in the same sense that the Tongue Body or
the Tongu Gt Blade is ... an Articulator capable of executing a
constriction ges!ure. Among the vowels represented with this
articulator are lei. 18/. 10/. l'.l/. and Ia!.
Since the tongue root can form a constriction against the
pnaryngeal wall (- Pharyngeal Site) or;!y, it is not possible to argue
ior the feature Tongue Root on the basis of the natural class
phenomena. However. there are other consitierations which support
positing this feature.
One of the most puzzling problerT,s for a theory of segment
representation is the existanc9 of the so called ·n~utral· segments • ~egm8nts whic.., fail tlJ undergo a harmony process. and which are
transparent to it. In vowel ~larmonies. the vowol faJ is most
172
commonly found !O -=t like. neutral -ament. Vet it is by no means
• rute tor IaI to behave in such • WIlY.
In 7..1 I propo.. to appro8Ch the probIsm of neutml segments in
.rm. of the milalng ...nding slt.-: a reatur. may be allowed to
prop8gate .croll • legment thet leek' the node to which this
toeture could att8Ch. WhAo thll COIIClIPt dDeI not ~zpI.in ail casel of
neutral HQmentI. II captu,.. rather elegantly th. general facti of
CDnIOnantai and vowel hannoniel (... 7.1).
It spptNUI to me that this canc»pt can b4I profitably exhtnded
to explaining the behavior o' 1M vowel laI, In cas.. In which i: acts
Uk. • neutral segment and in CUM In which It participates In the
harmony (disharmony) procet.... I want to sugges! that in those
ca... in which ttl8 vowel 181 .eta liko • neutr~1 segment w.r.t. to
the proeHl•• which spread the ,. .ture, dof'i1i~rded by the Tongue
Body, II I8cka the -landing site- onto which the h"Jrmonie feature
could spread. In such ca.... 1M vowel 181 is to be analyzed as
inwlYing the Tongue Root 80 an active Artlculater. Whenever III
r.'artleJpatn In processol which spread either (back) or [high]. it Is
to be
~ecJ
as involving the iongue Body as an active Articulator.
In
kwinte
-bIC8U•• of
In II kel(xu
In N;baka (se. S!8riade (1987), and r.f.rencrJ' there). (vowel
inventory (Steriade ('1987»: I. U, 8, I, 0, ~, a: plus the same series
with 13ngth)~ vowels within a single morpheme Ill. sUbject to the
following restrictions: vowels ot the same hAight c&n cooccur only
if they agree in al~ other 'eatures: If the vowe" af. nol identical,
they mUlt be of' a dlffe,,,"! height The vowel lei, though, can
caoccur with any other vowel.
Steriade (1987) explains the neutrality of 181 in •....so cases
as 'allows: since tha VC1W81 181 i~ [+Iow). ifs specification for (high)
is predictable (given thaI there are no (.high)[+low) seoments), and
does not need to be present in the Undertying Representation. The
absence of this specification explains the fact thol the vowel 131
cannot be subject to const~:nts which apply to [-high) s8gments,
the fact that it cannot spread [-high), and its tranlparency to the
spread of [-high)-.
This type of explanation is not available within the 'ramework
which dGes not employ the feature [low) (see 5.4.1 for the arguments
against this feature). If lal is not (+Iow), then its [-high)
to whk:h these concepts can be
specification ought to ~ determined by the same princjples by
applktd- In 1M Puiego dialect of Mcr:tan 8. Spanish (_"alyzed by
which lei and 101 are (-high). In fact. this is how I propose to treat
aU dorsal vowels with a Pharyngeal constriction.
Let us consider some exarnpln
McCarthy (1984). Steriad. (1987» (yowei inventory (McCarthy
(1984»: I, I, ' . U. Q. D. ~, a, and A•• ten.e counterpart of laI) a
height
harmony
spre.ds
either
[-high] or (+high) spedflcation of • s~sed vowel onto a non·low
YOlnt As shown below, the vowel JaJ (al well al Its tense
counterpart IAI) is tninsparent to the spread of thG harmonic
features, and fails to spread any vaJu. for height when it appsars in
• enand po Ii tion:
(iO)
blb-(.
-drink-
bIb-emus
bib-emus
II-me)
·the ,vir
173
.tnt- (s
·r88r
The neutrality of 181 w.r.t. height (dis)harmony in Pasiego and
Ngbaka can be explained wUhin the constriction model under the
assumption that laJ lacks the proper landing site for the feature
[tj~hJ to s~read on. This would be the case if the vowel faJ were
articulated with the Tongue Root rather than the Tongue Body. Such a
vowel cculd n(t\'er bear the height or the backness spttedlcation
(both features being the daughters of the Tongue Body Node), and it
would always 08 transparont to processes spreading these featuras.
·the log-
Giyen the aoovD proposal. one woukJ predict that the vowel Ia!
should never be affected by the processes spreading the Tongu8 Body
Slnt-emUI
stnt-eits
II mAcsiro
174
f••tures. Obviously. thil prediction i. fal... bee.use there are
IanguIIge8 In which Ja! aparenlty don participate in such proc......
Consldr. fer example Turkish (vowel Inventory (ClementI and Sezer
(1912»: '.,
u, W • •,
0, •• plus the sam. series with I.~gth),
u.
whera /81
(11)
o.
altematn with Iw In a bEk.... harmony:
ode-den
-room-
dire-din
odfl-l.r
dlrl-llir
ode-JI
d.r'-J.
·rtver·
In order for th. nat"","1 of Pui4tgo and Ngbaka lugges:.d
8bcMt ta go through. there would have to be two varietl.. of /81 ...
one artJculDld wtth 1M Tong.- Root (as In P o . and Ngbaka), and
one artIcuIeted with the Tongutt Body (at In Turkish).
At first light. such • move may appc-ai to be unwaranted .&fter all. It IntrodUCOil 8 phonological C:lltfnctlon which is neve' (at
••••t to my knowledge) uled to support • phonemic distinction.
However. the arth,:ulutory data make this tr.atment of laJ credible.
ComparatlYe articulatory evidence suggests that conltrictio~
location for Ial may vary betMten the mJd.to-upper phBrynx (as in
French (d. Bother.' at a' (1988» or Turkish (cf. SkalidisKonltantinidis (1981))}. end the lowe, pharynx (as in Polish (ct.
Wktrzchowska (1985»). In the former. a balJ.shaped tongue executes
the ccnltrtction: In the latte, - the bal. of the tongue is an active
artlculltor. Thl.e dlfferencel corre.pond to different acoustic
Ilgnall: for example. the Turkish Ia! has a fairly iaw F1
(approxlmetely 500 CPI, judging
irom the
spectrograms in Skalidis-
Konltlntlnidll (1981 », whUG th. F, of the Polish 181 is
appKJximatoly 800 cps (WlerzchoWika (1985».
The t....tm.nt of 181 suggolted here predicts that the variant
with a low pharyngeal constriction should alway. be neutral to the
proc~s'.1 sproadlng the 'aaturo= dominated by the Tongue Body
Artict..Jlator. and the variant with the upper pharyngeal constriction
should parUcipate in such prec.sl... Unfortunately. the
175
artlcuS8tory/acouslic data from '_nguag.. with the two type. of
harmoni.. are too ICIlIC8 to Yftritr ttlts pndictIon.
No articulatory or acoustic de:. ar. av.i&able for the 'utego
variety
Manta".1 Spanish. •,. however. the -.ow- \:owe' in this
dialect 01 Spanish should tum out. not implausibly. to b8 the same
a. the clow· vowel in another dia*=l
Spanish •• castilian. then.
given the acoustic measurements in Morimoto (19&8), according to
which the F,
laJ pronounced
spNker of Castitlan ,. 780
cps. the Manta"" 181 would be considor8d II Tongue Root vowel. Of
course, without the IIctuaI data •• this is pur. speculation.
0'
0'
by."""
0'
5.5
Evidence for the ConstriCtion Node in Vowe&t
I argue for the presence of the Constriction Node in the
rep'H8ntation of vowels on the basis of the phenomena which can be
analyzed only in lerm, of a node thai maps onto a sin9fe articulatory
ge.ture.
One of the most clear examples
single constriction node of a vowel
palatalization. In Gorecka (1989) I
consonants can be analyzed only as
Palatal constriction of a vowel onto the
of the phenomena ;n which a
is phonologically active. is
show that palatalization at
a process which spreads a
Root Nodtt at a consonanl.
Vowel coalescence and vO~81 simpHfication phenomena
provide pttrticularly strong support for the Constriction Node. and
fol' the Yowel representations proposed in this thesis. These
phenomena have long bee" a problem for the model of vowel
represe f -alion which can charaet~rize vowets solely in terms of the
tongue ,ody and the labial features (this applies to both the IPA and
the SPE treatment of vowels). Below I analylCi examples of these
phenomena. and show that they are best explained in terms of
operations 0 n phonological constituents which correspond to the
conslrl.::tlon gestures.
5.5.1
Vowel Coalescence
176
Vowel coaanc.ne8 il ~ con1mon phMomenon in languages of
the wortd. It is virtual., a rule in ttl. Bantu languagn. but it II by no
meant confined to !hi. family (ct. the dilCUSlion In SChane (1987»).
Most common I" vowel coalelcence conYflrtl the sequence. /81 + IV
and lei + lui to lei and 101 reepectlvel,.
On the constriction model. thil phonomenan hal a
Itrelghtforwerd ••plan.tlon: two Ilmpie legments merge .nd give
rIM to • doubly articulated legment via • P~II which co;3ap.es
IdMticII nodn. and =mputes non·1dentIcaI nodes.
While the coatucence phenomena can be handled 81 a single
Proc.Il within the IPA or SPE framework (with obvious diflicultles
ariling In predicting the f••ture output from the input that (,Ontains
• (+htgh] and [+Iow) elemsnr). a very similar phonomenon. Which
Involv.. • •• Imilation rather than the merger 0' the looments,
cannrJt IWCI!ve such • u.ctment within thol~ framework•.
AI an example. consider the fonow,ng procell in Tunica,
dllcullld by Kenstowicz and Klaseberth (1979), on the basis of the
data In HoI (1940). In Tunica. 1M sequence IlaJ is realized as leI,
and the sequfJnce /uaJ is realized u '-.,1. Such sequences arise when
• quotaUv. postfix I-anil attachea to words thai end in Iii or luJ:
(12) mill • &nl ---) milln'
nlku
+
enl ---)
ntk~nl
·It 11 rid·
·hIIOyS·
Kenstowicz and KiaHberth point Qut that this phenomenon cannol be
tr~ted 81 a simple ca•• of segmental fusion, i)ut rather, it must be
analyzed as a two-atep proce,.. involving an assimilstion and a
d·e~etJon of one of the Yowell. They argue fur thfs analysis on the
ba~1 of the fact that the aliematiun takes place even the vowels
N te) and 181 or lui (0) and Ial Dr. ssperated by a gloUsl stop.
Thil II shown below:
-n..nr.t.
'RIGn w"lIluCh dlfftcuIU •• diD not ft••" to erll; on ~h. con.trlctlon mod.1 la
tNt
ttll. mOde' doli not rtcognlD U,. 'estur. £loYlI (cr. ttee dllcu••lon In
5.-4.1>' IlCone!. btJU, Might yolu.1 or. prtdlCl8t'lt In thta m091 anC jo not n••d to
be DrlMftt In Uta UR wtten cHI.&cenCI tatOI ploc •.
177
(93)
po • ?8kt ---) PO?~kl
·Sh. looks·
pi • ?ekt ---> pf?£kt
·she .merg••·
On the constriction model, the above phenomenon involves the
of the Constriction Node of a yo. .I onto the following Ial.
T 8 only articulations that can spra.d. are naturally the Palatal and
the Vetar one. The fact that the glcitaJ stop does not inhibit this
process, but all other consonants do, can be e.pl8ined on tho
assumption that unlike other consonants, I?I lacks place f•• tures in
the Underlying Representation (el. Stlriada (1988)).
St~read
Neither the IPA nor the SPE- frameworks can treat the facts in
(92-93) in terms of a single procus. 8S the only feature that iiI,
Ju/, leI and lei sh&tft within thele framaworkm is [-lOW): obviously,
It.is feature cannot explain why 181 turns into l:al after lui and 101.
and into lei after IV and lef.
5.5.2
Vowel Reduction
Vowel reduction phenomena constitute a different type of
evidence for the constitullnl Constriction Node in the representation
of vowels. Reduction occurs generaUy iii unstressed Yowels, and
most commonly, converts a comple. vowel (either lei or 101) into a
simple vowel (Iii (I) and lui (0) respectively). Some examples
inc~ud8 vowel reduction en Syrian Arabic where a five vowel system
merges with IiI, lui and la! in unstressed syllables (Cowell
(1964». and in Bulgarian. where in a n:.lmber of dialects. ustressed
vowels form a three-vowel syzt,m (Sentton (1984)).
l
Thdse phenomena are difficult to explain within tt!e tongu9
arch model of ~owel representation: even if they C2n be analyzed
uni10rmly in terms of the reusing of mid vowels. what remains
unexpl,ined on this nlOdeI is why such phenomena should occur at all.
'h.
• That Inclu~••
ricin' mOMls dlYllooacJ 310m] thl SPE IIn'l. Clomentl (1 oes>.
Archang.1I end PullQyblenlc. fcrt!\comlng. Savoy (i 966>'
• Klnltowlcz arKJ KI ••• blrth (1979) tr,et III ena 1.1 8' t-rouncU. to achf.VD B onlrull enslVlls 01 the Tunics foets How""r. thtr. er, gOOd r•••Ofts to astuml thot
thl. rl.tw. d051 not eXIII. cf St.,IKe (1986).
178
An anawer to this quntion il polsiba. only within a frem.work in
which lsi and lei .re reprnentMI al complex IeGments (ct. Kaye 81
... (1815), Schane (1987».
respectively. and it deletes the entire snicuilitory component of a
low yOweI. Not unrHlOnably, thi! component il replaced by a lei,
insened by a default proceu.
The constriction model ••tllfle. this requirement. On this
model. vowel rMIuctIon phenomena are tNSted a. the reduction of a
comphlx --ament In a metncaUy weak poll~n {... Sage, (1988)
for moro disculclon of the procell.. whle'l rMuce compl••
legmentl !n metrically weak polttlonl).
To summarize this section -- since both Yowel coalescence
and Yowel reduction ph.nomen. h.V8 a naturae explanation when
analyzed in terms or the proceslttS affecting the Constriction Nodes
uf vowels. they support recognizing thes node ac a phonological
constituent.
Even though molt caHI 0' vowel reduction can be analyzed ~n
terml of • slngl., mkJ-to-hlgh railing ruPe, thle pos.lbUny is nat
=tway. avan.bJ.. In thll r.,pKt, consider the ca• ., of vowel
reduction roporfed by Topping !' ~AlJ In Chamt'rro, allo dlsculaed by
KenltOWk:z and Kisseberth (19)'9). In Chamorro (180ment inventory
(Topping (1988)): pc.. P:, f b, b:, m. m:, t, t:, d, d:, Iss, (lZ, S, S:, n, n:, I,
I:, 1'. r. ", k~ k:. iI. g:. 'I, ';». h, I. I, .., U, 0, a), unstressed vowell are
red~ according to the following pal1ldjgm: Ii! and '01 become lIlt
lui .rId 101 become 10/, and III and ,., ~ome ,.,. This is
iliullr.ted below:
5.8
#
(94) dfigo
pica
gwfh.n
p(gwI?
'yam'
'chest·
'flstf
'bltel nut'
t dlguhu
-my yam'
I p:cohu
t gw Iheani'lu
I Jllgwihu
-my chest'
"my'lsh'
ehe
constriction model, on the other hand. vowel reduction
In ChllmOtrO can be treatld .s • process which (in addition to l~xing)
deille. the Pharyngaa' constriction of a yowel: this pr!1cess
converts /01 and leI to lui (and later /0 I) and Iii (and later III)
179
In this section I consider some issue. concerning the
representation of !8Y8,aJ indiwidual Yowels which have been len out
of the discussion so far. I discuss the representation of the dorsal
pharyngeal vowels, in particUlar, the vowels Iml and 10/, and the
representation of the froilt rounded vowels.
5.6.1
Dorsal Pharyngeal Vowels
The vowels which belong in this category, are 1st I and 10/. I
propose to repreunt them as in (9Sa) and (9Sb) respectively:
(95)
(a)
(b)
lit:
Root Node
Q:
Constr
Constr.
A
Pheryng
Tongue Body
A
-high
Rool Node
I
I
·my betel nur
Nota that in Syrian Arabic and in Bulgarian the vowe' 181 remains
unaffected by the r.cluetfon procue: ~II i. what 8nO"~' the tongue
arch model to ~!ur. the rOiJuction proee'l in terms of a single
(vowel ratling) rule. In Charoorro, hOW2ver, aU vew.'s are a"ected:
unt.Il 'I; i., tht. language is trs.ted •• II nigh vowel, this process
eatlno! haVG 8 urJform interpretation in the tongue arch model.
On
Vowel Representation: R8IiduaJ Problems
-beck
/~Tongue Body
Pheryng.
A
-high
+beck
The treatment of these vowels as [·backJ and [+backl respect;v~ly is
weU supported by their phonology: they pan6m with other vowels
that are chsracterized in terms of these features. cf. the backness
harmony facts of Chamorro (Kenstowicz and Kisseberth (1979)).
Since they must be characterized in terms of a feature dominated ty
180
the Tongue Body, they have 10 be representlld with thl Tongue Body
u the active Articueatcr. Anlculatory dala for American English
I.' and lal (Perke" (1989)) confirm that indeed the•• vowels do
not Involve the tongUdt root constriction.
"nd
The ,.pre••ntation of's' and la! il of p.rtlcular Intere.t,
number of theoratlcal "Iuel
Olb it. Fe,' Instance,
ghen ttMt rttPreHntal," In (gsa). I.; appea... to be the only vowel
(10 fer) for which the f••ture [back) mUlt be malntllined In addition
~u• •
to the f. .lUr. P.latal Slt._ If ,., could
~
a••umed to have a
palatal conltrtctlcn (and 101 - • velar one). pam.pl the feature
(bIIc*] couki be abandoned (but IN tho dllCUlllon in 5.... 1).
y"t there .r. ,easonl for not considering
I.'
I'.
(8a) ~i~IP
Front Rounded Vowela
The constrit;tion model aliowl at least two representations
for the front rounded vowell: the.. are shown below (for
convenience, , shan limit the discusslen to lui:
(97)(8)
(b)
Root Nodi
A
Constr.
A
Peletel
T.
80dU
Root Nodi
'gam.-
lUlU
-mtstek.·
elhtrtlen'
Iisultl
·washroom·
Ion
-hand'
10m
·secl(
The YO weI /811 faUs to trigger :hll proceSI, al documented by the
forms I " Is. kI ·color'.
While _ single example of negattv. evidence cannot ciecide
or not • fo.lure il absent In the representation of a
wh.ttt~r
.egmant (fc;r mor. dlscuslion af 8 palatalizing capacity of Is I see
NHid (1 Si3r). the Korean facti mak. a rathe, good case for for the
• ....Id (1073) ...gun fM I hllfWChy of ,.eltlllzing ttren2Ut among frent Yowell.
hlerweiVJ./8' he. tht w"kl,l "letIIlZ!ng pro",\lo,. Dul he
elMl ~Ifd., It to It•• Dlillol legrnent Hvarthelc••. Howlver. Gn •• ,mgl•• of
;alatllEl'tlm tay 'el w~te!t Ntlld quotec (,.g., In 5wtdtl~. Fr.nch). Involve Viler ••
ffMI occur onlV II It hWalcrlCI2 CMlng:. not 18vnet'rentc rull. A pltaUclbJ.
"'1
111
A
Ccnstr.
A
lib'.'
Constr.
Lower
Constr.
AA
V,II'
T.
Dodu
Lta
Ltlbtel
Lower
LID
I
-back
(~887»:
11~'8m
A=orGlftg to
5.8.2
I.'
aa Pa!atal:
faUI to trigger palatalization. Consider 8S an
••am. the palataUzatlon procell In Korean (segment inventory
~~ (1984)): p.. pll. p', m" t, t~ .. t', I,
n. 1# f, e", t,1, J, k. k h ,
k', 'I, h, w; vow.ls: I, ii, u, t. u, 1,0,0, A, S, a: plus the same saris.,
with Ipgth): paiatali:ation Is an allophonic rul. which con~ert. lsi
to III beta•• palatal vowels, a. shown bolow (dllte quoted from
1,... rnattcaUy
BIIIantz et aI.
claim thai 1st I is not p.-.latal: the process is ailophonk: (note that
the reselting segment, Iii, is not an underlying segment of the
language): generally, aUophonic procels•• involve 21 trtgger. lhe
entire natural class of segments.
A questior; naturally arises as to whether both options are
utilized in natural languages. A relative rarity w,~ ~ which /u I
triggers palatalization (cf. Neeld (1973)) suggests thai the option in
(97b) should be more common. On the other hand, the articulatory
dat~ on these vowels in French (Bothore~ et at (1986)) and German
(Martens (1984») show luI articulated with the r>alatal conseriction.
Of course, neither of these languages has a palatalization process
which could help decide the underlying re:3resentation of lu I.
Be'ore the abovo question could be answered with a reasonable
of certainty a comparative study of the articulatory and
phonologfcal properties of these segments would have to be
a~:;~nt
J
I.planatlon for thl hlatorles) feell mSgftt D. thet YI'.r, fronte" bo,or. thl front
lal hlY' turnld Into ecoulucsn" Ilmltlr dOtlol ~&lltQ'1 (th. stagl lUll Obllrved
In SWldlsh). ond lU2)ieautntiv. Into 5.11 marked coronel Dllatoll. eat In Fr.nch.
182
perfonnecL At thll point. such • study II unr••Ultic. given the fact
dial th. artlcui.tory dar. .re un.v.iI.bl~ for the majority of
world'.
language••
8. Representation 01 Speech Sounds: Consonants
In this short chapter. I consider some re.idual issue. in the
phonology of consonants. In the cours. of thla dls•• rtatlon. I have
p'Jlited rapre.ant.tlons which In some ca... may appear
controversial and need st«1itional justification. and in others, if
succestuUy defended, provide additional suppan for the constriction
model. I now discuss th... ca• • i"dividuaUy.
6.1
The Representation ot /1:'/
fe'.,
I us. the symbol -k'- to
to what is known in the Uterature
as either a Gpalatalized velar· (otten transcribed as -k J , . or a
-ironted veiar·. and corresponds to a sound 8t1iculaled with P ~::'~r
constriction with the tongue body fronted. On X.rays. this sound
often appears with a closure somewhal forward of the region in
which ordinary velars are art!culated.
On the constrtcttoi" mod.I, it is not possible to represent Ik')
as ~ palatalized segment as a secondary articuiatiGn palatalization
Involves a front vowel constriction. This me8~1 t~al a palatalized
velar would necessarily have to be represented with the 70figue Body
Articulator simultaneously !txecctlng two constrictions: Palatal and
Velar. FaUowing Sagey (19&6) (but S88 the jjscussion in 1.4), I
acsumee that such
r8pre~ntatiarls
aro nat allowed.
On this medel, there are only two possibl., representations for
Ik'/; either Ik'/ involves a Tongue Body constriction at the Palatal
Site~ or a Tongue Body constrictiori at the Velar Site (the palatal
stop), with the Tongue Body fronted ([-back)). Choosing the first
representation would bo justified if no language e~8r contrasted lei
with /k'i. Only then could the two sounds be assigned it ~ingl8
phtlnological representation. perhaps associatod with slightly
varying acoustic signall in different languages.
However, 8vidar.cs from ~4nguages which do have palatal stops
(even U only ~n lheir surface inventories) suggests that this
183
184
trcatmMl of /k'/ la not po.lible. In a number of such langUages, !cJ
mull be represented dlfforentfy from /k'I. ~ocaule the two sounds
cDntral1 befor. front yow.I,·. This il true of Basque (as
demonetrmted to me by the n&ttvo spe.ker of the language). In
MSCIdonlan. tho pol.tal Imp .Ny mer;. with Ik'/ before front
vowelJ !n th. lpe8Ch of ",-me .-'Ion., but in the standard
pronunciation the twa .... k.-pt epart (Lunt (1152), p. 12).
In view of the above. the only reprnontatlan !lv.nabl. for Ik'i
within the c."'Onltrietion modee Is th.t of • VeCar sound articulated
eith the Tongue. Body modifittd by the f••ture I-back). al shown lil
(II):
segments (e.g., t • Si, d j , Zi (phonotically realized as distributed
'
coronal palatals: s.. 3.2.2),
:J, b', I', etc.), which exllt in the
undertying inventory of the language, but may alsc M creetfK! by the
phonological processes.
P'.
As is wen kno~n (cf. Rubach (1984»), suHI.e. which begin
with 2 front vowel in Polish (IV or ,.,) can be either cyclic. in
which case thf'Y trigger palatalization, or non-cychc. in which case
they do not. The e,tclic suffixes induce the following changes in the
slem·flnal consonants: lablals and dentals are pala:alized (p .... > pi,
b··> b', t ··>t J (followed by the c~ang. to
IU. while velars turn
into coronal palatals (k •• > ~. g .-> ~. X •• > i). Thi~ is illustrated
below:
Ie
(18)
Ik'/:
R
(99)(a)
I
(b)
/~
Below I argMe that this repre.entatlon il consiltent with
the
bohavior of /k'/ in phonological processe=.
Ik'j Iystematically fails to pattern with the palatal or
palamliz8d segments. In thii ralpact. consider tho following facts
from Polish. language which hal a wide fange of pa!atalb.~d
elncldtntDII\!. th'. feet conStitut.11 probl.m 16r Ut' .rlleuletor theory of S.tgty
line; It t:1n ropr.,e"t Ute contrelt. Imo", the dor.I' ,ou~dl In tsrn.s of
tM 'tetur•• (bICk', (highland flow I Oill".. It ,. not ClDel'l. r'Drl'tnt Jng lh.
CM1,nt "tween lei end It'l. The only r.cour•• for this ."oor" II to al,um. theat
"e'ltel 'ton ItO r.on-,trld.nt cDronale (oroyld.d thlt th. fOlllur.(coronaJlII
rltdlfl.led to Inclua tbl front Dart Of the dorsum). HOWIVlf. th'l trlltmlnt of
Hletl' ,tOt' 10 not JUIU"" b~ their Plt'.rntng In DftQnologtcl. procaa.ol. 01 J
0'
dMM;,alral. b.~ow.
18S
tuP'2C
xtop + tsk • 0 ---) xtOP'tsko
kat • Isle • 0 ---) ko,; tsko
nog
Tongu. Body
I
-beck
(! 986):
---)
red • I • t J ---) reditc
krok • t • t' ---) kro~tc
c
Viler
tup • I • t J
•
Isk
•
0
--- > no!tsko
·to rob'
·to advise'
·to merch·
'peasant, I'ugm:
·cril, Dugm:
·119. eugm:
I analyze th9 process illustrated above along the lines proposed in
Gorecka (1989): the Palatalization Rule spreads the Palatal!Tongue
Body constriction of Iii o"to the Root Node of a preceding consonant.
The immediate output of this processs is a palatalized segment,
which in the case of labia~s and dentals remains unaltered. as these
are als\» the underlying oegments in Polish. The effect the
palatalization processs has on velars is predicted by the constraints
on po~sible representations which I assume in this thesis: while a
PalataVTol1Que Body constriction may spread onto a Velar segment
at the stage when it lacks the Articulator specification (possible
only in languages which do not haV6 a k : It' contrast), such a
re~re88ntatlon
cannot surface as a fully specified. doubly
articulated segment and therefore, must be reanalyzed as a single
palatal (or the palatalization effect can be nullifiad. but obviously,
this; is not the course of events in Polish).
186
Wh.n followed by a non-cycUc suffix which begins with a
fror!t vow.l. stem-final conlon.nll are unaff8cted if they are
labial' or dentats: velaf., howeve, (with the ••coption
l'1.f) are
trontMt. Thll it shown below:
(101)(8)
0'
(b)
(100)(.)
(b)
nog •
trok
TM
---) !ebt
lab.
tot •
I
•
---)
kott
---)
I ---)
nag',
kroJ('t
_ned
front yowell in Polish heve on v.ier. would be. difficult
to axplaln If ", I were Ie be tr.ated al a
palatalized v.lar. While t~. data In (99), could plausibly be treated
I~ tttrmi of Structure P~es~ryatlo" (afte, all, Polish does not have
an underlyir-g /k'I), the data In (100) ar. more of a problem. Here
ve......... ·pa~t.llzeer and nothing el•• i~,
scene
·we'
"
"bon,·
sp'ett
·slnglng·
·bllnci"
"wrong·
slJlp+
spil
telk'.
"fe'-='OfS·
pstk'.m
zg'.tk
'not ••·
dfazg'l
kosi
il'.
·S"'P'·
·belt, Inltr."
'spllntlrl·
Anether example along th. same Une. comes from Bulgar,••l. where
palatalized segments are ••,,-aptianlesl'Y dopaiataU7ed before front
vowels ct. lkon'olkon81 'horsQ' (this phenomenon hal been discussed
8arUer in 3.3.2). In the s.m. environmont velara are fronted. cf.
Ibteg'i/ ·sw8of.
(It not Im!3o••lbl.)
This fact is not 8 problem for the treatment of 1k~1 adopted
Mre. In (100) vAl.,. are fronted (for the analysis of the mechanism
bahlnd thl. process .ee 4.3). TM r••son why no other segment Ir~ th~
18ng"'808 undergoes thtJ rule il si"aple: no other segment can be a
target of the procesl whit:h spreads the Tongue Body Articulator in a
f•• ture..filling ..a.hion.
Thrar. are many ether itxamples of phonological processes in
which AtI fsils to pattern with Ihe palatalized consonants. In
Polish. again. it faill to act like e pslatallzed segmen; in the
procesl of Palatalization Spread c:Hscuaeed earner in 4.1.1. This rule
palatallz". denial s!biiantl when they are followed by a palatalized
segment: .ith~i' Ip'!, ~/. or nfl, etc. This ia not a cyclic process. as
it appU~i aerOOI the board {except when the prefix boundary
In~fV.ne. b.Iw"~ the trigger and the targQt: for an 6xp!anation 8!t
10 why th!1 should be th9 case ... Czaykowlka-Higgins (198a)}. Yet
Ik'/ faill to trigger it. This is iliUltrated below:
• I bcWt no IGlenltion for thlt.
187
These facts 2rtl sasUy explained under the assumption that Ik~1
is a fr~nted velar: it cannat spread the palatal constriction in the
Polish case, and it cannot loose it in Bulgarian, becaUI9 U doss not
have it. While the above phenomena can also bo hsnd!ed on the
assumption thai /k'/ is a palatalized consonant. say, by invoking rule
ordering, the point that remains is that /k'/ never paUerns like a
palatalized segment This generalization is just an accident on the
theory that /k.'1 is a palatalized veJar.
6.2
The Representation of lei
In this section I argue for represeniing leI. the so called
·palatal stop· (as wen as O~lY sound from the family of palatals: I J/.
Ir;/, III) as a simple segment with e TOfigue Body constriction at the
Palatal Sito. Since • have already shown in 6.1 ~hat this sound cannot
be reduced to /k"1 ~a frortted velar). all that remains to b9 shown is
that ii cannot be represented as a coronal segmant with I-anterior)
specif;cation. distinguished frem If I in terms 01 stridency.
Consider again vowel hardening process in Kinyarwanda,
discussod in 3.4.2. In this process. a vowel fuses with a preceding
consonant when it is followed by another ~owel (the analysis ot this
phenomenon in terms of a complex segment formation follows Sagsy
(1986)). If it is a fr:Jnt vowel. :! gives
rise 10 a palatal stop
compon9nt in the resulting complex segment. Since the resulting
188
complex Ngment may contain an unqueltlonably coronal companant,
e.g., ku-••• -> ku.eg 'to griner. ku-ri·. -> kur'1 'to ~.r, then,
given Sag.y', (1988) conltrmlnl against camplsx segmentl
mentioning th. sam. articulator twa, it foliowl that the lei (/J/)
~t
In these ugments cannot be an.lyzed a. coiOnal.
Given the characterization t"
. . . one wouid
.I~et
lei. I flo If",
~te. al
the dorsal
tMt they can be determln8d •• dorsal by
the wame criterion by which the, have beten determined a. noncorama'. namely, they ar. predictee not to occur I I component
articulations in compl.. ..gmentl that have • velar/dor.11
component. The t••tlng ground II fa!rly limited In Klny.rwsnda .according to Klmenyi (1979) thcI only p...... segment that occur. in
the envimnment In which il could potentially merge with • demoted
HG\tI8Ver. for t9111 s»artlcular segment. the above
pr«llctlon appe.... to be corr~: the palatal nalal is alw.Y' deleted
velar vowel is
'ra/.
In such erMronmentl·:
(102)
iu • mil' •
tu • •••".
W
•
•
---)
kum''1''e (-tume"fJ-a) 'to b. known·
(-9UI8.fllI-e) 'to bl dlltroyed'
W• 8 ---) gU'I',,-"
leu .. 1'0 • II
---~
kU'1-S (-gUf'I')-a) ·let drink'
The concluliOn that lei cannot be tr••ted al a coronal. and
must be tr_ted •• a dorsal segment Instead, and the conclusion that
It mUlt be repr• •nted as distinct from /k'/ combine together into
an .rgalment against the .rttculator model of phonological
rwpra..ntatlon of Sag8, (1988), linea thll model cannot assign
distinct ropre.entallona to Ik'l and lei thai could explain the
~o~k:aI behavior Cif these lOunds.
313
McCarthy (1989) proposes to treat uvul.rs .1 comp~.x
segments which are both vel., and pharyng••I, and pre'."t. a
number of argument. for this view.
First. he observes that uvular. pattern with both vela,. and
pharyng88:S in d&termining the dlmlrlbutlon of segments In tfilit.ilIl
roots: as amply shown by McCarthy (1979. 1981, 1988. 1985).
Arabk: displays phonological processes whteh are belt understood
under the assumption thai in the underlying repr•• entation,
cc ·~:!nant. and vowels of Arabic occupy separate melodic tiers. This
assumption plays a key role in eEpleinlng the cooccurrence
restrictions on consonants In trillt"r.1 verbal roots. in particular.
the fact ihat these restrictions are more sever. among ~h.
consonants which are adjacent within a triliteial root (despite the
tact that they may be separated by vowols in surface 'aims) thsn
among the non-adjacant onea.
McCarthy (1989) shows that in trUiteral roots, uvular. faU to
cooccur w;th either velar or pharyngeal consonants. Given this fact
and the oenerallzatlon that cooccurrence restrictions In Arabic
prohibit the adjacency of segments which are artlculatorily similar,
it follows that uvulars must be perceived as articulatorify similar
10 both velars and pharyngeals. McCarthy's oonclusion is that uvulars
share articulatory features with both velars and pharyngeals. and
that In fact they are complex segments which involve the two points
of articulation.
McCarthy's insight, when incorporated into the framework of
the constriction model. translates Into an analysis of uvulars as
segments involving a Velar/Tongue Body and a Pharyngeal/Tongue
Root constriction.
UW..... 81 Complex Segments
I> Klmany' menUon. cmw cl... or I.CectiOM lo tflea pnerell:IUon: IDDar.ntly. the
'111tl•••HI II not ol,tld WhiR mlfgod With
v.ler VOWI' If It '1 derlv'd YIO
I mvtlU", rp.!•. Thll rul•• 'f.c~. t:cnlOMntl b.fGV. tlte C...IUv, markor /. J/. Th.
",IM,-'tton for thlt 'let t e.n ltd. or II thlt D"tteD' In 'MI' conDiruc!l~n,~
• "lDt.~/~.sl rte••1to r••• lyzef! ••• PO.atelicorDfiII.
U,.
-.n'"
il9
This characterization of uvulars has proven useful in the
analysis oi Arabic ablaut in 3.5.2: it made it possible to explain why
these segments pattern with either velars. causing the ablauting
vowel to surface as luI. or as pharyilQeals. changing the ablauling
vowe! to IBI. (A rather similar ablaut pettern has been discovered
190
(not lurprillngl,) in Pal.stinian Arabic. by Younes (1082). also
qwtld by McCarthy).
7. Phonological Processes
This chwpter is concerned with the constraints on phonological
processes which are not feature..particular. but appear to be
structura;ly determiliod instead. I attempt to derive some of tho9S
constraints from the general properties of the constriction model. In
cases where this mode' appears to OY8rgenerate. I introduce
constraints whJch limit its power.
7. 1
Some Predictions about the Processes Which Affect Nonadjacent Segments
In this section. I propose to derive the locality conditions on
the assimilatory processes from the position the spreading features
occupy in the segmenta! tr88: I show that non-Iccal spreading of
fsatures is possible only when intervening segments do not provide
the ·'anding site- (are not suitable hosts) for the 'aature that
spreads.
In terms of the locality conditions which ihoy observe,
phonological processes cen b@ superficiaUy divided into two
categories: those which may spread features while skipping
(specified) segments. and those which may never skip a specified
segment. The first category includes backness. height. tenseness
harmoilies among \lowels, and retroflexion and sibilant harmonies
among the consonants. The second category consists of all point of
articulation assimilations, palatallzation-. velarizat50n, voicing or
ttricture assimilai!ons, etc.
Putting aside the retroflexion and sibilant harmonies for the
momenr. it becomes immediately obvious what the processes in the
first group have in common: thay aU spread Articulator-dominated
features. Since this generalization does not appear to be a result of
• With one k"~~n Ixc,ptlon. neMlly. palatalization of III In korole.
• SI:tc. It 'I not obvious whet 'e"lur•• or. aprled In the•• prOCIII.S.
191
192
suffix is confirmed by the native speaker of the language (I am
graleful to Michael Kenstowicz for making these data availRble to
me):
(12) UkUXOY8
ukucofe
(16)
9
+
+
+
-to kneed- ukuxoYwa/ukuxovtw8
'Il It kneaded'
8
-to pre.s'
'It Is pressed'
9
ukudiokovi 'to sheke"
ukungwlva 'to ege·
kuJecofwa/kuJecofiwe
kuJedlokoywe/kuJadlotovtwa "tt Is
shaken'
kuJangw8YWe/kuJangw8Vlw& 'Il Is
aged'
Another environment in which bilabials are palatalized.
involves the locative morpheme Ie-... .-ini/: here. a bilabial must be
followed by a slem·final lui or 101 in order to undergo the rule:
(13)
Finally. to complete the locativltJ paradigm. no palatalization takes
place if labials are not followed by a round vowel:
8
+
I S16tf:p h O
S
•
u:(u:bu
Inl --) est60AB:nl
'grass rope'
Inl --) o:[usf:nt
'meDlte-meDI water'
e • 1snd:60" Int --) estfOe's:nt
'friend'
s • umlO:mO + tnt --) emIO.,e:nl
'moutn-
Impt +
Int86tJ
Inslmbt
+
tnt --) amptnt
'ermy'
tnt --) 8nt868nl
-hill'
+ tnt --) enstml'tnl 'metar
3. 1.2,2
I would like to suggest the following analysis of the
palatalization facts presented above: There are cooccurrence
restrictions in Zulu which prohibit sequences of segments with a
constriction at the Labial Site. This restriction may be interpreted
as an OCP effect applying at the labial Site tier. When a disallowed
sequence arises in the formation of the Passive. the labial
constriction of the first segment is deUnked, Let us represent this
process as follows:
+
+
(Note the disappearance of the labial element from both vowel and a
consonant in the surface forms.)
( 17)
R
/ \
R
I
c
/\
Lib
lip
If - )
~
Lib Lip
\ftl aodu
~
R
,f
/ \
c
/\ ;1 ~
Lei!
Lip
lob Ltg
Vol Body
No other consonant becomes palatalized in this environment:
(14)
9
e
e
+
•
+
1:S8rJO + Inl --) eS8I'Jwent
u:khezo + Inl --> o:khezwenl
t:zulu + Int --) ezulwtnl
"Qelewey'
'spoon"
'sky'
Particularly. the labiodentals are not palatalized under these circumstances. However. they do not behave exactly like non-Iabials in
(11). as two surface forms are possible:
(15)
e
+
e
+
e •
f:lT)bvu
Isf:ro
f:ru
+
+
+
tnt
In1
Int
--) ernbv Lnl or embvw f-nt 'sheep'
--) eslfe:nt or eslfOe:nr 'dlseese'
--)
er!:nl or efw f:n!
·cloud'
(R • Root Node: C
Lower Lip Articulator:
Articu lator)
Co
Constriction Node; lab - labial Sito: Lip Val • 'lelar Site: Body • Tongue Body
I assume that palatal constriction is inserted either by default. or
by a ruls, but since the exact mechanism of that process is not
relevant to the present discussion. I do not pursue this issue.
The above analysis does not extend automatically to the data
involving the locative morpheme. First. observe that the nouns in
which the palatalization of bilabials takes place when thay are
suffixed with I-ini/, contain the allegedly disallowed sequence in
the underlying (and surface) roprasentation. The question arises as
to why forms like It s If t1 :601 'friend' are allowed to surface. but
• as Is a d1tahtong_
4S
46
an accktent, a SYltematlc explanation 'or the phonological behavior
01 the Articulator-dom~iiiiiw ie:iiures should be available.
Th. explanation which I propose derives the special status of
within the 'eature hi6rarchy.
Whll. not ruling out • pouibl!ity
a phanoBogical process may
make ,.'.renett to prosodic Information, and therefore require
-Jacency of the l8Gm.nt~ involv~, I want to suggest that all else
being equal, procelses that ·skip· segments, spread 'eatures for
which the -skipped- segments cannot provide a clanding siteD, in
other words .- a mother node. T;,is proposal predicts thai proc88ssa
which spr.ad the Conltrictlan Nod. or any reatur~ directly
dominated by the Root Node may never skip a segment, unless the
legment can be argued to be CQmpletely unlpecUied (a possible
example of thl. might be c:onlonant assimilation in K"ley-I,
discu••ed by Archangeli (~98a». Similarly, processes spreading
~rtlculatorlSit8 f••tures are predJcted to observe strict adjacency,
unteD Intorvening segments can be argued to lack the Constriction
th\,.. f•• tur.. from their position
'hi'
NodD.
Articulator-dominated 'eatuies (again, Ziside from a few narrowly
defined cases where the skipped se;gments ars severely
underspecified and Jack either the Constriction or the Root Node).
With this in mind. let us i10~ turn to tha harmony processes
among consonan!s. While such processes are far les9 comm'l" than
the harmonies affecting vowels, they do occur nevertheless. Several
examples commonly quoted in the literature include retroflexion
(e.g.. n-retroflexion in Sanskrit) and sibilant harmonies (e.g., in
Chumslh, NavQjo, Kinyarwanda, etc.). The analyses of th~se
phenomena in the recent literature do not support the -landing siteD
view of the locality conditions, &3 they attribute the harmonies
variously to the spreed of the Coronal Articulator (ct. Schein and
Stsriade (1986», or the featuro [anterior). (ef. Poser (1982)). Below.
I consider a possibility of reanalyzing th@ above cases of
consonantal harmonies in t'.lms of the Articula~or·dominated
features.
In Sanskrit (consonant inv&ntory (Whitney (1889)): P. m, t, n, S.
:t, f, c, 1', ~, k, 11). Ini became retroflex r:~I&r a retroflex
continuant (either Is6/ or Ir6/). with any amount of phonological
material intervening. as ~ong E.S none 0' !he segments were coronal.
Another condition on the rule, which I do not discuss beyond this
statement. was that In! ha~ ~o be followed by a liquid sonorant (see
Schein aJ1d Steriade (1986) for the d~scussion of this condition).
Here are several examples of the retroflexion harmony:
\,
Thll propos.' explain. why the Artlculator-dominated features
such a special status in phonology: since they can spread only
their mother Articulators, they can skip segmonts which iack
specifications. Thul, the features dominated by the Tongue
Articulator ([high). (bsck» are sllowed to skip all intervening
Itlgments which lack the Tongue Body specificati"n. This is indead
what W8 observe in the case of [bsdtness) or [height) harmonies: they
propagate across any consonant. Of course, they skip ordinary velar
consonants as well, despite the fact they are dorssl segments. but
Uils facl has a simple explanation, consistsnt with what has been
said about v@lars earlier: ,inca their Articulator specification is
universally predictable, it is assumed not to be piosant in their
underlying representation.
have
onto
such
Body
The -landing sitd view of lh~ lace-lily conditions on rules net
only exp~ains the special behavior of the [high] and ~ckJ f9aturas, it
also predicts that segment skipping ass~milalions may Involve only
WJ
193
r),
(103)(a)
pur -D:Qa-
vfk-r:u!-
·fIJl'
·cut up·
vs.
k~ ved-8:n8·hum·
mrd-ne:"be gracious·
The analysis of n-retroflexion proposed by Schein and Sieriade
(1986), also followed by Sagey (1986), attributes the process to a
rule which spreads the Coronal Node which dominates the features (.
anterior] and [-distrioutedJ_ Ths main reason for appaaUng :0 the
Articulator nodo rather than just one of the features below iI. i~
that such an analysis explain3 the blocking of the rule by an coronal
segments in the language.
194
Obviously, fram the point of view of the proposal I am
Introduclng In this section, such an account of Sanskrit nr.tmflexion is not possible. The only situati~n in which an
Articulator node could spread across a segment under this proposal,
would be if a segment lacked the Constriction tlode. This is clearly
not the cu. in Sanskrit, where any non-coronal segment may
intervene between the trigger and the target in n-retroflexion. The
only analysis avaUab" under this' proposal is in terms of the feature
(distributed): amang all the poSSible candidates. this feature is the
. 0I!1y
0n2
that can propagate across
no~ronal
segments.
Given the analysis of n-retroflexion in terms of [distributed).
how is the coronal blockage to be accountud for? After all. the
proposal I am Introducing makes no statements about the
transparency or opacity of segments. By this proposal, the feature
[diatribu~ should be aDowed to spread on all the coronals.
Yet the blocking effect can have
suppose that n-retroflexion applied
coronals (other than Inl) have
specificallons for [distnbuted]; in such
caused by the ioalure [distributed.
an independent explanation:
In Sanskrit after all the
received their redundant
a case. blocking would 08
Whil~ I 'mow of no independent evidence from Sanskrit for the
above analysis, there are good reasons for believing that it is
nevertheless correct:
the analysis in terms of the Articulator
spreading predicts that there should exist cases of long distance
spreading of the Articulators. not only among the sQunds that are
executed with the same articulator (note that under the Articulator
spreading analysis of n-retroflexion, the Coronal Articulator
spreaas to dislodge the Coronal Articulator). but also among the
sounda executed with different Articulaturs. I do not know of a
single example of such a phenomenon.
Next. let us consider briefly an example of a sibilant harmony,
such as the one which occurs in Chumash. Since I discuss this
)95
phenomenon in some detail in the next section, I refer the reader to
consult the data there.
In Chumash (segment inventory (Ruhlen (1975»: p, m. ph, m', l,
h,
ls, s, n, 1, t h , lSh, Sh, n', )1, t', S~J
A, Sh, k, k h , k', q, qh,
QUI,. X,. X', 7, h, I, s, t, u. 0, a), all sibilants in a word must belong to
either a /§ I, Ie I sei, or to a lsi. IIsI set. Alternations reveal that
the underlying quality of a sibilant is subordinated to the quality of
tho rightmost sibilant in the word. Since the sounds tn the I!/. I~I
series are laminal. and the sounds in the lsi, Itsl series are apical•
the harmony rule sHeets two changes simultaneously: it changes the
constriction location of a segment (Palatal <--> Anterior). and it
changes its specification for [distributed].
ts', c, c c',
Poser (1992) considers a possibility that the harmonic feature
in Chumash is (anterior). Clearly, such an account cannot be
accomodated under the proposal which derives the locality
conditions on assimilations from the position of the assimilatory
feature in the segmental tree: the only possible candidate for the
harmonic feature under this proposal is the feature [distributedl.
which is dominated by the Tongue Blade Articulator in the segmental
tree (recall !hat this feature may be relevant only in tha phonology
of the coronal segments).
Thaie are at least two arguments for positing [distributed]
(and not the Sites) ss the harmonic feature in Chumash. First, any
vowel can intervene between the trigger and the target. including
the palatal vowels (see the data in 7.2), without creating the
blocking effect Second (this arg.... 71ent is Gxternal to Chumash), of
all recorded cases of sibilant harmo"ie~ (cf. Bhat (1978». each and
every one involves the change of the value for (distributed), in
addition to a~y other change. No language in which the Anterior and
the Palatal c.oronals agree in the value of this feature displays a
sibilant harmony.
On tha basis of the preceding discussion, it is possible to
conclude that the facts of Sanskrit and Chumash. as weH as all
196
generalizations about consonantal harmonies. suppcut the proposal
which derives the locality conditions in assimilations from the
position of the assimilc\tory feature within the segmental tr88.
Before ieBvlng this section, G would like to consider the
troubl..some case of palatalization in Karok (segment inventory
(BrIght (1957): P. f. P. m, I. t, s. n, c.
(I), k, x, ? h, I, tt, u, I:, e:,
a:. I:. 0:. u:)_ As mentioned earlier, palatalization in this language,
triggered by faJ, lei and fJ/. can skip any consonant to reach lsi.
. Examples:
Tongue alade and the Tongue Body). On such an analysis. the
derivation of the surface form such as /I k z ( r t p § ipl 'to start to run'
would proceed as follows:
(105)
tkztr t
sip
R
R
R
c
A
c
A
c
A
c c
e.
l.-&'~~t -
Pel
T Lib
Body
(104)
?I:nflrup
-threshold of Indlen hous.-
?f:f'Ip
-to grow-to stert to run-laugh-
lkzfrlpilp
?tkiah
Since I propose to treat palatalization 8S a process which spreads
~. t'alatallTongue Body constriction of a vowel onto the Root Node
at a consonant (ct. Gorecka (1989), 1 predict that such a process
"iust be strictly local. Therefore. its ability to skip consonants in
Karok present!; a problem for the -landing site- theory of the locality
conditions.
Yet despite this appearance. the palatalization process in
Karok can be analyzed in a manner consistent with this theory. It is
possible to posit palatalization spread- in Karok, nol unlike the
process which I have motivated for Polish in chapter 4. The only
difterence between t:-te two cases wauld be that in Karok~ the
Palatal constriction would be delinked from any consonant that is
not lsi. The Iii would be assumed to surlace eventually with a copy
of the Palatal constriction
independent of the trigger vowel, to
satisfy the constraint which prohibits the association lines from
crossing (Goldsmith (1976». and also. to account for the fact that
the Palata; constrictions of the target and the trigger in this
palatalization process surfac9 with different articulators (the
-1
L Ml T
l to
81 Bde
sip
tkzlrtp
A
R
~
C C
p~ I AnilT~
BO~LJ
Bllde
I
l!bll
LI,
Pel
:
I
r
T
Body
Bllde
As a final comment to this section, I would like to point out
the following: while the -landing sito- theory of the locality
conditions on assimilations is conceptually independent of the
constriction model, and it can be applied in the phonological analysis
under any model of segment representation. of the presently
available proposals, only the constriction model provioes Ule
hierarchy of features which can produce the correct results under
this theory. Consider, for example, the articulator model of Sagtly
(1986): using the ~ame idea, it is possible to rule out on that model
long distance Articulator spreading: since Articulators are
dominated directly by the Place Node. no segment that already has a
Place Node may be skipped by such a rule. However. given that the
feature [anterior) is a dependent of the Coronal Articulator on this
model, and given the prediction that ArticlJlator-dominated fealures
can skip segments not specified for their mother Articulators. this
model predicts the existence of harmonies spreading the feature
[anterior] only. This prediction is not borna out by the available data
(see the discussion above).
7.2
A ProblslTl for the Constriction Model and Possible Solutions
-Tbl. anllya•• hes b.en ,ugg•• t.d to ml by MlchDII Kenstow'\:l.
197
~
p
R
198
In the preceding chapters. I have presented a number of
.argumentl for inducing the Site and the Articulator featuros in the
phonological ,........,tation. and I have proposed a hierarchy of these
f••tur.. within • phonological segment, capable of explaining
certain aapecta of their behavior In phonological processes. Still,
the constriction model. if left unqualified. allows for phonological
pIOC8IHI which. to my knowledge, do not occur.
Given the range of phonological processes observed In n=tural
languag.., an adequate model of phonological representation must
(106)
Root Node
Root Node
I
I
Constr.
Constr.
AAnterior
A
Lower Lip
PaJDtel
Tongue Blede
Or it should allow a conversion of a coronal palatal 10 a dorsal
palatal before vowels Iii, lui. le/. and JoJ, as demonstrated below:
aDow Independent spreading of Articulators, Sites. and the features
which are dominated by the Articulators «[back], [high],
(distributed]. etc.), as demonstrated extensively throughout this
th.,II. Obviously, the feature hierarchy proposed in chapter 4
delivers this result. At the same time however, there are reasons
lor wanting to constrain that freedom.
(107)
I
Constr.
A-
Pelatel
As it stands. the constriction model allows all features to
spread in cumulative fashion: a Constriction Node may spread onto
the Root Nod. of a segment. &Van if this node already dominates a
Constriction Node: an Articulator or a Site feature can spread onio
the Consbiction Node, even if this node is already fully specified in
terms of both the Articulator and the Site; finally. an Articulatordominated fealur. may spread onto an Articl.:ator that bears an
opposite specification of the same feature.
Howeyer. some of the processes which this model predicts t~
be possible, do not occur. Here are some examples: spreading of an
Articulator feature in cumulative fashion (with subsequent deletion
of the original Articulator sp~cificatlon) should allow an
.lsimUalion of If1 to 151 (or 18 I) before, say.
I; this is
Ie
illustrated below:
Root Node
Root Node
~
Constr. (ConstrJ
A
Tongue
Any
Tongue
Blade
Site
Body
If Sites were allowed to spread cumulatively, ihe following
processes should occur: It/ should be able to turn into lsi before
IfI. as shown below:
(108)
Root Node
I
Constr.
A--
PDI8tai
Tongue
B~ade
Root Node
I
Constr.
A
Anterior
Lower
Ltp
Or we should be able to find Iii commonly turning into luI before a
velar:
199
200
(109)
Root Nodi
I
APeletel Tongue
Constr..
Body
Root Node
I
Constr.
~
TCingue
Veler
Body
None of the procoSS8S illustrated In (106) through (109) occur
commonly or occur at all. This gap does not appear to be accidental,
which means that a systematic explanation is in order.
.
One way of blocking the derivations in (106-109) would be to
introduce a constraint which would prevsnt cumulative spreading of
features. Unfortunately, such a constraint is too powerful in that it
blocks some of the most common processes, such as palatalization,
velarization. labialization. in fact all ihe processes whi~h introduce
secondary articulations. Since all these processes spread the
Constriction Noda(s), it is reasonable to conclude that this node may
spread cumulatively.
Putting asids Articulator-dominated features for the moment.
It appears that only Sites and Articulators may not spread
cumulatively; the Constriction Node can. Before I attempt to account
for this fact, I want to point to a fundamontal difference between
the processes which spread Sites and Articulators on the ana hand,
and the Constriction Node on the other hand: while the latter
processes may result in possible speech sounds, no process which
spreads either Site Of an Articulator cumulatively can produce an
intermediate output that would map dir9cUy onto a grammatical
representation: there are no possible speech sounds which can be
characterized as in.oiving one Articulator executing a single (!)
constriction at two Sites, or two !~rtjculators executing a single
constriction at one Site.
201
Suppose now that phonological processes could not apply in
ways that would result in ungrammatical representations -- if this
indeed were the case, nothing would naed to be said about the
manner in which phonological features could spread; the derivations
in (108·109) would be automatically ruled out, because their first
stage would 8lway~ be an ungrammatical representation.
The above proposal, while somewhat radical at first sight, is
not really that far apart from the common sense intuitions that
most phonologists share. To my knowledge, no serious phonological
analysis has besn offered that posited impossible speech sounds in
the underlying inventory of a language. And ave" if sometimes
derivations are proposed that proceed through an ·ungrammatical
intermedia:e stage-, I am not aware of any analysis that would
actually take advantage of the impossible feature combinations and
derive phonological effects from them.
At this point, a question could be reasonably asked why
phonological processes could not apply in such a way so as to
accomplish the spreading of Site/Articulator with a simultaneous
delinking of an original Site/Articulator specification {:'&call that
this is precisely the kind of derivation I na;;t.t shown in each of the
cases in (106-109) in Q!dGi' io mai<e them look at all sensible). Given
a possibility of such derivations. the system would still be able to
overgenerate. Therefore, in order for the above proposal to work in
the manner intended, spread-and-delink derivations would have to be
disallowed.
Simultaneous spread-snd-delink derivations could be blocked
easily, if phonological processes were not alJC'wed to accomplish
more than one change at a time, in other words, if they were
assumed to be completely atomic. In such a case, delinking procedure
would have to be viawed as a phonological process separate from
spreading. Again, this is not an unreasonable assumption in the
present day phonology; after all, one of the main reasons for positing
segmental trees has been to allow the spreading of multiple
features by single (atomic) rules.
202
Stil on the same topic - In the discussion of palatalization
phenomena in Gorecka (1989) I conclude that at least among these
phenomena. there are no examples of spread-and-delink rules. Th!s
II based on ths observation that segments which are derived as a
relult of palatalization (complex segments with a front vowel
articulation) are never simplified. unles. there exist conditions in
the 1angU8ge which trigger such a proce....
If this generalization should turn out to be true of other
phonological proce.... as W8n, then the derivations in (106-109)
woukt no longer be an embarassment to the constriction model: they
would be rua.:l out on independent ground...
In this spirit. let us now consider processes spreading
features ([high]. [back), [distributed). etc.).
While early analyses of such processes have relied crucially on the
existence of spread~and-delink rules. under the assumptions of
underspeciflcation theory (Archangell (198-4» many harmony rules
have be.n reanalyzed 2S feature-filling (ef. the articles in
Phonology, vol. 5. and numerous r.'erences there). Even so, there
stili exist phenomena which, at least on the face of things, appear to
require a spread-and-delick analysis. Sibiiant harmony in Chumash.
analyzed by Pose, (1982), Steriade (1987;, may be considered a
classical example of such a phenomenon. Below, I propose an
analysis of the Chumash facts, which does not need to appeal to the
sp'ead-and-delink rules.
Artlculator-domlnat~
In Chumash. all sibilants in a word must agree with the
righ!most one in the value for [distributed]. The interpretation of
Chumash harmony in terms of [distributed] is my own (S99 the
discussion in 7.1); Poser (1982) and Stariade (1987) consider a
possibility lhGit the spreading feature is !anterior). My
reinterpretaiion is based on ihe information that palatals are
laminaJ and alveolars are apical in Chumash {cf. Bhat (1978».
Experience with other processes. similar to this one. shows that
sibilant harmonies occur Dilly when alveolars and palatals have
opposite specification for [distributed].. In 7.1. I have argued that the
203
only possible into!"pretation of such harmonies is in tarms of the
feature [distributed), despite the fact that other features are
affected along the way.
(110)(a)
(b)
k
+
sunon + us --> ksunonus ·1 obey him'
k + sunon + S --> ksunonts ·1 am obedient"
Poser shows that the harmony can change both values of
[distributed): causative suffix whose isolation form is Isul. appears
as Ii u I in words in which the last sibilant is It I, and the dual
subject marker, whose isolation form is IIA/. eppears as lis/ In
word9 w1th the IDst stbllent 1st. Here are the example9:
(111 )(a)
(b)
(112)(a)
(b)
su + weyen --) suweyen
'CDUSB
to hang"
k + su .. §oyln --> ksusoyln "I darken I r
p
s
+
+
Is
Is
+
+
81
+-
tlsl
nan' --) ~1!Clnan' ·don"l you two gO"
+
yep
+
us --) slstlplyepus ·they two show
him·
On the basis of these and similar examples Poser argues that the
sibilant harmony in Chumash is feature changing. Given the
appearance of the data in (111-112), one may be leat:i to such a
conclusion easily. However, despite this appearance. it is possible to
analyze the facts of Chumash without assuming that one value of
[distributed) spreads and delinks tile other value.
Evan though Chumash has both apical and laminal sounds in its
inventory, the apical/laminal distinction does not need to support a
phonemic contrast: laminals (/!/. Itt/. etc.) are always Palatal and
apicals (lsi, Ilsl, etc.j are always Anterior. This means that the
distribution of both values for [dis~ributedJ is predictable in
Chumash, and therefore. neither specification for [distributed] need
to be present underlyingly in this language.
204
On thta assumption, the harmony pmcass in Chumash may be
viewed .. spreading both values of [distributed) to segments that
lack any specification for this feature. While the spreading process
Itself
not feature changing- the effects of the harmony are:
palatal segments which receive [-dIstributed) specification must
ultimately be converted to anterior segments. and alveolars which
receiv8 (+distributed], must be converted to palatals. as the harmony
'I
I
II structure-prel8fYlng.
The support for the present analylls of Chumash harmony
comn from tho followl:1g facts (allO reported by Poser): there Js a
proce.. In Chumalh which converts lsi to IiI before Ifl. IV, Inl.
in dertvltd environments. The derived III does not undergo the
sibilant harmony. as shown in (113):
(113)
• + tl + Jep • us --) stl Jepus -hi iells him'
It is not the simple matter of ordering the rules·. as the derived IAI
does spread its value for (distributed]:
(114)
segment is no longer a possible host for the harmonic fe3ture, nor
can it be transparent to its spreading. much in agreement with the
current assumptions about the nature of opaquEt segments.
•
+
Is
+
Under the above analysis, the facts of Chumash are not only
consistent with the theory that spresd-and-delink rules are
unnecessary. but far more than that. they provide a strong argument
against positing such rules.
Whether or not spread-and-delink rules can be completely
abandoned In phonology. cannot be decided on the basis of one or even
several examples. The ultimate fats of such rules may depend on
what the final shape of underspecification theory will be. as the
questions about the way the rules apply are directly determined by
what phonological ropresentations they apply to. Richly specified
representations are more likely to require rules changing feature
specifications than the representations which are severely
underspecified.
t..? --) Itit..?+ ·he f1nds It'
The molt natural explanation for these facts is that following a rule
which spsciflel the value for (distributed] in a segment (made
evident by the fact that this value can be phonologically active). the
-In ordlr for tM bannony in Cmme.~ to apply In thl ,."l"'l-fl1l1ng fa'h!cn, It ,.
fttCnslty thot thl redundant IPDclflceUon fnr !dlltrlbUtedl not" , ••Ign.d to all
morphem•• Simultaneously. Whll. It I. not uncommon for thl harmony proc••••• to
lUborettn.'. all tergltl to thl mOlt prominent trlgpr In tM word: It em. b. tnhlr
• Itr.,.td yow.l. M the rlgfttmo't (liftmost} vowIl, Itt., It II not cllw howe
rut. CIn Diet out tt" rightmost coroneal. AI pointed out to me by Donco St.rled.,
PhonOloglca' prOCIIII. do not ept'lu to thl rightmost .Ibllent., obltru.nt.~ ItC. At
lbll point I do not MY•• proposel thet would .)CDloen thl. DSPlct cl my on6Iy'!I.
• Although lbll S. precl.lly thl position take:. by POI.r; Po••r .ugg•• t. the
pOIllbUlly that thl I\8rmo~ proc.I. oppllt. In two .teg.s: flrlt. both Yolu'l )f
tbI "'armonSc feotur. art dIUnk.d; ••cond th. hermonle fl.turl IPf.Dds from th'
rightmOil trIg;... S!ne, the,. orl two distinct proc•••••• onother phonological
pncnl Is allowed to InterY.",. Th. Change from J. to 0 pelota) b.forl1J..D t.
lbOu9l't to occur botwlen d.Untlng and IPr.edlng of the harmonic rlotur.. Whll.
thll lolutlon dOl. ecCCNnt for the dato, It leck. the .~plonotDrv p.r.uaslyen••• -.. 'l I. nitt clser why d'ltnlctng would DDPly In thl flrlt pioco.
20S
206
)
D. Slollenwerk. and Z. Zhang-Sheng (1987). Language Files.
Advccate PUblishing Gmup, Reynoklsburg, Ohio.
Bibliography
AbramIon. It. S. (1962). The Vowels and Ton.. of Standard Thai:
Acousticld
Measurementl and Experiments.
Intematiamtl
Joumal 01 American Ur.guisticl, Publication 20.
Ahn. W. S. (1953). Phonetics in Ancient India. London.
Andersen, H. (1989). The phonological statui of the Russian -labial
trIcatIvel. In Jou,-.,at of Ungu!lticl S. 121-127.
. Archangen. D. (1984). Underspeciflcatian In Vawelmanl Phonology
and Morphology. MIT Ph.D. dlu8rtaIIon, cambridge, Mass.
Bloomfield, L. (1939). Menomini Morphophonemics. Travaux du Cerels
llnguistique Prague 8, 105-115.
Bosch. A.• B. Need and E. SChiller (ads.) (1987). Papers from the
Parasession on Autosegmental and Metrical Phonology.
C:hlcago Unguistic Society, Chicago, III.
Bothorel. ~ .• P. Simon, F. Y./io land, and J. P. Zening (1986).
Clneredtogrephte des yoyelles at consonns! du Frenc;ais.
TravBux de rlnstitule de Phonotique de Strasbovrg.
Archangel. D. (1988). Kele!-j Consonant Assimilation. Coyote
Papers 8. 1-15. University of Arizona, Tuscon. Arizona.
Brame, M. (1970). Arabic Phonology: Implications for Phonological
Theory and Historical Semitic. MIT Ph.D. dissertation.
Cambridge, Mass.
Arch.ng_lI. D. and D. Pulleybtank (forthcoming). The Contant and
Structure 01 Phonological Representations. MIT Press.
Bright,
cambridg..
Mass.
Aronoff. M. (1987). Haad operations and strata in reduplication: A
linear treatment SUNV-8tony Brook Ms., Stony Brook New
YmL
'
Aronoff. M. and R. Oehrle, ads. (1984). Language Sound Structure.
Studies Presented to Marris Halle. MIT Press, Cambridg&,
Mass.
Armand. R. C. de (1966). -Palatal- and -pmataliz9f.j~ redefined. In
Canadian Journal of Unguistics 11:2 ,109-113.
Bell. A. M. ~1~). The PrirwCiplas of Speech and Vocal Physiology:
and Dictionary of Sounds. London and Edinburgh.
Bell. A. .... (1887). Visible Speech. London.
BeY-Berti, F.• L J. Raphael. D. B. Pisani. and J. R. Sawusch (1979).
Soma relationships between articulation and perception.
Phonetica 35. 373-383.
Btl_I. O. N. S. (1978). Palatalization. In Greenberg (1978), 47-92).
w.
(1957). The Karok Language. University of California
Press, Berkeley.
Bryzgunov8, E. A. (1963). Prakticheskaja fonetika i intonacija
russkogo jazika, Moscow.
Carre, R. and M. Mrayati (1989). New Ccncept in AcousticArticulatory-Phonetic Relations. Perspectives and
Applications. Present9d at ICASSP 1989. Glasgow.
Catford, J. C. (1977). Fundamental Problems in Phonetics. Indiana
University Press. Bloomington. Indiana.
Chomsky. N. and M. Halle (1968). The Sound Pattern of English.
Harper and Row. Now York.
Cl9m8nt~,
G. N. (1976). Palatalization: linking or assimilation?
Chicago Linguistic Society 12. 90-109.
Clements. G. N. (1985). The geonletry of phonological features.
Phonology Yearbook 2, 225-252.
Clemsnts. G. N. and E. Sezer (1982). Vowel and Consonant
Disharmony in Turkish. in van der Hulst and Smith (1982),
311-336.
Blssantz. A. S., K. Johnson. C. J. Godby. R. Wallace. C. Jolley, D. B.
Schaffer. J. W. Perkins. F. C. Latta. S. G. Geoghegan. C. McManis.
207
208
H., and A. Harshenin (1971). On the Phonological Properties 0:
Ruuian Iv/. In Slavic and East European Joumal 15, 486-478.
CoatI.
Cole, J. (1988). Vowel Harmony and Coalescence in Menomini. Ms.,
MIT.
CoweI. M. W. (1984). A Reference Grammar of Syrian Arabic.
G5nrgelDWn University P...... Washington, D. C.
Fant, G. (1973). Speech Sounds and Features.. t.4IT Press. Cambridge.
Mass.
Ferrari Disnor, S. (1983). Vowel Quality. 7he Relation between
Unwarsal and Langu3g.Specific Factors.. UCLA Working
Papers in Phonetics No. 58, University oi California, Los
Angeles.
Cro""'. J.
Fischer-J"rgensen. E. Some Basic Vowel Fez2ur@!, Their
Articulatory Correlates, and Their Explanatory Power In
PhonCilogy. In Fromkin (1985), p. 79-99..
Czaykowlka-Higglns, E. (1987). The Distinctness of Derivation and
IntIedIon in Polish, NELS 17, GLSA, University of
Fromkin, V.. A. &d. (1985). Phonetic Linguistics. Essays in Honor of
Peter Lafefoged. Academic Press Inc., Orlando. Florida.
(1978). Typology and Universal. Of Vowel Systems. In
Greenberg (1978). 93-152.
Uuaachunttl at Amherst, Massachusetts.
Czaykowsk.Higglns, E. (1988). Investigations into Polis!l
Morphoiogy and Phonology. MIT PhD Dissertation.
Danielslan. B. (1955).. John Harts Works. Part J. Stockholm
de Rljk. R. (1970). Vowel interaction in Bizcayan Basque. Fonles
lingua. Vasconum, 2:5, 149-167.
Dolattre. P. (1971). Pharyngeal Features in the Conoonants of
Arabic,
German Spanish. French. and American English.
Phonellea 23. 129-155.
Dolt•• C. M. (1926). The Phonetics of the Zulu Language. University of
Witwatersrand Press, Johannesburg.
Coke, C. M. (1927). Text-Book of Zulu Grammar. The University of
Witwatersrand Press, Johannesburg.
Coke, C. M. (1954). The Southern Bantu Languages.. Dawsons of Pall
Mali. london. (page numbers are quoted from the 1967
reprl;lt.)
Dunatov. R. (1963). Palatalized and Palatal - a Definition. The
SlRvic and East European Journal 7.4, 402-408.
Fant. G. (1960). Acouslic Theory of Speech Production. Mouton and
Co.'s-Gravonhage.
Gesenius-Kautzsch (1910). Gesenius' Grammar Hebrew Grammar as
Edited and Enlarg9d by the Late E.. Kautzsch. Oxford University
Press. Oxford.
Golab, Z. (1984). The Arumanian Dial&ct of Krusavo in SR Macedonia.
SFA YugoSlavia, Skopje.
GoJdsnlith, J. (1976). Aulosegmentai
Dissertation.
MIT Ph.D.
Goldsmith, J. (1964). Tone in Kihunde. Wiener Linguistische Gazette
Beiheft S. ad. J. Renncson. Institute lusr Sprachv~ss8nschaft
der Univesitaet Vie;,..
Goldsmith, J. (1985). Vowel Harmor.)' in Kt.alka Mongol:an. Yaka.
Finnish and Hungarian. PhonoloQY Yearbook 2, 253..275.
Gorecka, A. (1988). Epenthesis and the Coda Constraints in Polish.
Ms., MIT. Cambridge. Mass.
Go recka, A. (1989). Are Front Vowels Coronal? Ms. Purdue
llniversity.
Greenberg, J. H. (ad.) (1979). IJniversals of Human Lauguage. vol. 2.
Guerssel, M. (1986). Glides in Berber and Syllabicity. LJ 17. 1-12.
Grossman. H C.• B. F. HaUis. and R. L Ringel (1965). Oral Tactil~
Experience. Archives of Oral Biology 10, 691·705.
Fant. G. (1965). Formants and caviti8S. Procsedings of the Fifth
International Congress of Phonetic Sciences, 120-141. Basel.
209
~honology.
210
· ....... M. (1940). Tunica. Handbook of American Indian Languages, vol.
4. Smithsonian Inltitution, BUrMU of American Ethnography,
Washington, D. C.
HaIa.
B.,
Hai-,
M. (1983). On Distinctive F••ture. and Their Articulatory
Implementation. Natural Language snd Ungulltlc Theory 1, p.
M.. Romporit P. Janota (1988) Proceedings of the Sixth
Int8mational Congress of Phonetic SCiencei.
Hualde, J. I. (1988). A Lexical Phonology of Basque. Ph.D.
dissertation, University of Southern California, Los Angeles,
Cal.
Hulst, H. G. van dar. N. Smith (ads.) (1582). The structure of
phonological rep.'8sentations (Part II). Foris Publications,
Dordrecht - Holland/Cinnaminson • USA.
ItO, J. (1984). Mellodlc DIs81mtlelion In Ainu. Linguistic Inquiry 15,
505- 513.
91-105.
Halle, M. and G. N. Clementi (1984). Problem Book in Phonology. tAIT
Ito, J. and R. A. Mester (1989). F6sture predictability and
underspecificatJon. Language 85, 258-293.
HaU8..... and J.-R. Vargnaud (1980). Three Dimensional Phonology. In
Jackson, M. (1988). The articulation of lui: X-Ray Evidence.
Presented at the Linguistic Society of Am8ri~ Annual
Meeting.
p,.... Cambric:Ig4I.
Joum.
of Ungulstlc Research1, 83-105.
HamId. A. H. (1984). A Descriptive Analysis of Sudanese Colloquial
Arabic Phonology.. Ph.D. di.sertation, University of Illinois,
Urbana-Champaign, III.
Hania. J. G.
(19;'2). Phonetic Notes on Some -SIamese Consonants. In
J. G. Harris and R. B.. Nosa (eds.) TBi Phonetics and Phonology.
Uahk. JI University. Bangkok, p. 8-22.
Jacobson, L. C. (1978). DhoLuo Vowel Harmony. UCLA VJorking Papers
In Phonetics No. 43, University of California, los Angelss.
Jakobsen. R. (1941). Kindersprache. Aphasis, und allgemeine
lautg88etze. Reprinted in Selected Writings, p. J28-40 1.
Mouton, Hague.
Hart. J.
Jakobson, R" C. G. M. Fant, and M. Halle (1952). Preliminaries to
Speech Analysis. The Distinctive Features and Their
Correlates. The MIT Press, Cambridge, Massach....setts.
Havos, B. (1984). The Phonetics and Phonology of Russian Vo;cing
Assimi- IaOOn. In Aronoff and Oehrle (1984), 318-328.
Jakobson, R. anet M. Halle (1956). Fundamer~ls of Language. The
HeIIwag, C. F. (1781). Dissertatio inauguraJis physiologico-medico
de formatione Ioquelae.. TUblngsn.
Jones. D. and D. Ward (1969). The Phonetics of Russial1. Cambridge
University Press. Cambridge. England.
Herbert. R. K. (19n). Morphophonological Palatalization in Southern
Bantu: A Reply to Segmental Fusion. In Studies in African
Unguistics 8. 143 -171.
Joos, M. (1948). Acoustic Phonetics. Lan{,\lsge Monograph 23.
(1569). An orthographie, conteyning the due order and
reason howe to write or paint thimage of mannes voice, most
lic. to the life or nature. London.
c.
Hoffman,
(1983). A Grammar of the Margi Language. Oxford
University Pres., London.
Howaril. I. (1973). A Directional Study of Rule Application in
Phonology. Ph.D. Dissertation, MIT.
211
Hague: Mouton.
Kaye, J., J. Lowenstamm, and J.-R. Vergnaud (1985). The Internal
Structure of Phonological Elements: a Theory of Charm and
Government. Phonolog~ Yearbooi( 2. 305·3~B.
Kenstowicz, M. and C. Kisseberth (1979). Generative Phonology,
Description and Theory. Academic Press. New York.
212
KImenyI, A. (1979). Studies in Klnyarwanda and Bantu Phonology.
Current 1nqulry into Language and Unguistics 33. Unguistic
ReeeardI. Inc. Carbondale. USA, and Edmonton, Canada
KIparUy, P. (1988). How Abstract ia Phonology? Indiana University
Unguistlca Club mimeograph.
Kiparlky,
P. (1981). VowoJ
Harmony,
mi. Stanford University.
Lad8foged, P. (1982). The Nature of Vowol Quality. Reprinted In
lItdefaged (~987), 50-142lIdofogecl. P. (1984). A PhoMtic Study of West African Languages.
An Audltory·tnltrumentaJ Survey. Catnbridge University
Prell. ~ , England.
Ladetoged, P. (1987). Three Areas 01 Experimental Phonetics. Oxford
Univoralty Pres., London.
Levin. J. (1987). A Flaes for Lateral in Feature Geometry. Presented
at the Unguistic Society of America 1987 Winter Meeting.
Li. F. K. (1975). A Handbook of Comparative Tai. University P~ess of
Hawaii,Honolulu.
Uljencrants, J. and B. Lindblom {1912). Numerical Simulation of
Vowel Quality Systems: the Role of Perceptual Contrast.
Language 48(4), p. 839-862.
Un. Y...H. (1989). 'Consonant-Vowel Cooccurrence Restrictions in
Taiwanese. Presented at the StUdent Conference in
Linguistics, MIT, Cambridge.
Lindau, M. (1978). Vowel features. LanguHge 54. 541-562.
Undb!om. J. (1989). Lecture presented at MIT.
Laiefoged. P. (1975). A Course in Phonetics. Harcourt. Brace.
JovanovJch, New York.
Lindblom, J. and J. Sundberg (1971). Acoustic Consequences of Lip,
Tongue, Jaw and Larynx Movement. Joumal of the Acoustical
Society of America 50. 1168-1179.
Ladefoged. P., J. DeCI9fk. M. Lindau, and G. PapQUn ('1972). An
Auditory-MolDr Theory of Speech Production. UCLA Working
Papers L, Phonet~ 22. 48·75.
Linker, W. (1982). Articulatory and Acoustic Correlates of Labial
Activity in Vowels: A Crosslinguistic StUdy. UCLA Working
Papers in Phonetics 58, UCLA, Los Angeles. Cal.
Ladefoged. P. and I. Maddieson (1988). Sorno of ths Sounds of the
Wortcrs Language.. UCLA Working Papers in ?honetics 64. Los
Angeles. Cal.
Lunt. H. G. (1952). Grammar of the Macodonian Literary Language.
Skopje.
LasI. R. (1971). Boundaries as Obsuuents: Old English Voicing
Assimilation and Universal Stmgth Hierarchies. Journal
Unguistica 7. 15-3'~.
Leben.
w.
0'
(1973). Suprasegmental Phonology. Ph.D. dissertatio". MIT,
cambridg.. Mass.
Lehlsta. I. (1984). Acousticai Characteristics of Selected English
Maddieson, I. (1984). Patterns of Sounds. Cambridge University
Press, Cambridge. Cambridge.
Maddieson, I. (1987). Linguo-Iabials. UCLA Working Papers in
Phonetics 68. 21-45. UCLA, Los Angeles. Cal.
Maddleson, I. and P. ladefoged (1988). Multiply Articulated
Segments and Feature Hierarchy. Presented at the linguistic
Society of America
Winter 1988 Annual Meeting.
Consonants. Publication 34 of Indiana University Research
Anthropology. Folklore and Unguistics. Bloomington,
IneIlana.
Malmberg. B. (1963). Phonetics. Dover Publications. Inc.• New York.
Levin. J. (1985). A Metrical Theory of SyUabicity. Ph.D. dissertation,
Malone. J. (1984). Tiberian Hebrew Phonology. Ms., Barnard College.
Center in
MIT, Caillbridge. Mass.
213
Madsen. J. of Aarhus (1589). De Literis Libri Duo. Basel.
214
tMrtena. P.
Morlmolo, V. (1906). The Fiye Vowels of Cestillan and Jeponese. A
Spectrogrephtc Analysts. 18 t -191.
MmtInet, A. (1955). Economie del changementl phonetiques.
Francke, Berne.
Moscati, S.. (19M) .. An Introduction to the Comparative Grammar of
the Semitic languages. Phonology and Morphology. Otto
Harasowitz, Wiesbaden.
(1181). Oplilche Unterstii tzung Akult 1scher
Sprechltgnall. In Hele. Ramport1. end JanDte (1 9S?)JI p.
McCarthy, J. (1979).. Formal Problema in Semitic Morphology and
Phonology. Ph.D. dilsel1atton. MIT. cambridge. Mass.
McCarthy. J. (1981).. A ProsodJc Theory of Nal1CGncatenative
Morphology. In Unguiltlc Inquiry 12, 373-418.
McCarthY. J. (1984). Theoretical Consequences of Montane. Vowel
Harmony. Unguistlc Inquiry 15. 291-319.
McCarthy. J. (1988).. OCP Effects: Gemination and Antigemination. In
Nartey, J. N. A.. (1979). A Study rn Phonemic Universals •• Especially
Concerning Fricatives and Stops. UCLA Workino Papers in
Phonetics, 48. UCLA. Los Angeles, Cal.
Nash, D. (1979). Warlpiri Vowel Assimilations. In K.. Safir (ed.)
Papers on
Syllable Structure, Metrical Structure. and
Harmony Processes.. MITWorking Papers in Linguistics, 12-24.
tlOiaY8. G. (1970). Structure du dialects basque de Maya. The Hague:
Mouton.
Ungulatic Inquiry 17. 207-28.'.
McCarthy. J. (1988). Furore Geometry and Depenooocy. Ms...
University of Massachusetts. Amherst, Mass.
McCarthy, J. (1988). Feature Geometry and Dependency: A Review. To
appear in O. FujimuI"M, ed., Articulatory Organization - From
Phonology to Speech Signals, Phonetica Special Issue.
McCarthy, J . (1989). Guttural Phonology. MI. University of
Maasachusetts, Amherst. Mass.
McCarthy, J. and A. Prince (1987). Prosodic Morphology. Ms.. ,
University of
Massachusetts at Amherst and Brandeis
Ohnasorg, K.• and 0 . Svarny (1955). ~ tudes exparimentales des
articulations chinoi$9s. CeskosloYsnske Akademie Ved:
Rozpravy 65.5.
Perkell. J. (1969). Physiology of Speech Production: Results and
Impiications of a Quantitative Cineradiographic Study. MIT
Press. Cambridge. Mass..
Perkell. J . (1979). On the Nature of Distinctive Features:
imp!icatlons of a Preliminary Vowel Product~on Study.. In 8.
Lindblom and S Oehman, ads.. Frontiers of Speech
Communication Aesbarch. 365-380.. Academic Press, London.
University Ms., Mass.
S. and M.. H. Cohen (i 987) . Tok9!I-tCi-token Variation at
Tongue·body Vowel Targets: the Enact of Coarticulation.
Joumal of the Acoustic31 Society (.,f America. 82. S 17 fA).
PerkeU, J.
Meinhof, C. (1932). Intmdudion to the Phonology of the Ba~tu
languages. Dietrich Reimer/Ernst Vohl8n, Berlin.
Meyer, E. (1910). Untersuchungen ubef Lautblldung. Festschrift
Wilhelm Vietor, 168-248 (spedaJ
isI~8 of
Ole Neuren
Spradlen)
"Iler C. and P. Skautrup (eds.) (1930-31). Jacobi Matthis
ArhusiensiB, De literis libra duo, herausgegeben unci
erie utert yon Christen Meller und Peter Sl<autrup mit einer
Oanischlr U8b.fs8t~ung nebsl elnef Abhendlung uber Text unci
QUillen yon Franz Blatt. 2 Yots~ Aarhus.
21S
Pe.Kell, J. S. and M. H. Cohen (1989) . An Indirect Test of the Quantal
NlJture of Speech in the Production of the Vowels iiI, laJ and
lui. To appear Journal of Phonetics..
Perkell, J. S.. and W. L.. N61son (1982). Articulatory Targets and
Speech Motor Control: A Study of Vowel ProcJuction. In Speech
Motor Control (8. Grillner, A. Persson. B. Lindblom, and J..
Lubk&t', ads.), 187-204.
216
Perk..I, J. S. and W. L Negn ("i985). Variability in the Production
of Vowell IV and laJ. Journal of the Acoustical Society of
America, 77, 1889-1895.
Peterson, G. and H. L Barney (1952). Control mathods used in a study
of the vowels. Joumal of the Acoustical Sociel'; of Amaooa
24, 175-185.
pos.,
W. J. (1982). Phonological ~epreS8ntation8 and Action-at-aDistance. In van der Hulst and Smiitl (1S82), 121·~58.
Press, M. L (1979). Chernehuevi. A GramlMl' and Lexicon. University
of California Press, Loa Angeles.
Prince, A. (1975). Th& Phonology and Morphology of Tiberian Hebrew.
MIT Ph.D. Dissertation.
Rappaport, M. (1981). On the Phonology of Gutturals in Biblical
Hebrew. MIT Working Papers in Unguistics 3, 101·126.
. RIngel, L R., A. S. House, K. W. Burk, J. P. Dolinsky, and C. M. Scott
.
(1970).
Some Relations betwee" Orosensory
Aspects of Speech
Discrimination and Articulatory
Production. Journal of Speech and Hearing
Disorders 35, 3-
11.
Ringen, C. (1980). A Concrete Ana!ysis of Hungarian Vowel Harmony.
In Vega (1980).
Rubach, J. (1~84).Cyclic and Lexical Phonology. The Structure of
Polish. Foris Publications, Dordrecht. Holland.
Russell O. (1928). The Vowel. Columbus. Ohio.
Ruhlen. M. (1975). A Guide to the Languages of the WorkJ. Language
Universals Project, Stanford University, California.
Sage)', E. C. (1988). The Representation of Featurfj$ and Relations in
Non- Unear Phonology. MIT Ph.D. dissertation, Cambridge,
Mass.
Saltarelli, M. (1988). Basque. Croom Helm, London.
Scatton, E. A. (1B84). A Reference Grammar of Modern Bulgarian.
Siavica Publishers, ColumbUS, Ohio.
Schane. S. A. (19B7}. The R"solution of Hiatus. In Bosch at (1387),
279- 290.
Schein, B. and D. Steriade (1988). On Geminates. Linguistic Inquiry
17, 691- 744.
SchmBlstieg, W.. R. (1984). Palatalized and Palatal -- a Sharp
Acute Definition~ In The Slavic and East European
Compact
Journal. 8.2. 182-183.
Shimizu, K. (1971). C"mparative JLI~unoid: An introductory Survey.
Ph.D. Dissertation. University of Ibadan.
Skalidis-Konstantidinis, R. M. (1981). Las occlusives du lure parle a
Istanbul: Etude radiocinemalographique et acoustique.
Sohn, H. M. (1975). Woleaian Reference Grammar. The University
Press of Hawaii, Honolulu.
Stahlke, H. F. W. (1976). Segment Sequences and Segmental Fusion.
In Studies in Afr~can Lingui&tics 7 41-63.
4
Stariade, D. (1982). Greek Prosodies and the Nature of
Syllabification. MIT Ph.D. Dissertation, Cambridge. Mass.
Sleriade, D. (1985). A Note on Coronal. Ms., MIT, Cambridge. Mass.
Steriad9, D. (1986). Locality Conditions and Feature Geometry. NELS
17, GLSA, University of Massachusetts at Amherst.
Massachusetts.
Sleriade. D. (1987). Aedundant Values. In Bosch et al. (1987), 339-
382.
St9vens, K. N. (1972). The Quantal Nature of speech. In Human
communication: A unified view, ad. by E. E. David, Jr., and P. 8.
Denas, 51-66. New York. McGraw-Hili.
Stevens, K. N. (1989). On the Quanta) Nature of Speech. Journa; of
Phonetics, to appear.
Sawashima. M., F. S. Cooper (19m. Dynamic Aspects of Speoch
Production. University of Tokyo Press, Tokyo, Japan.
217
218
Stevens. K. N. and A. S. House (1955). Development of a QU-w1titative
Description of Vowel Articulation.. Journal ~f the Acoustical
SocIety of America 27, 484-495.
StevMa, K. N. and J. PerkeH (1977). Speech Physiology and Phonetic
Featur... In Sawuhlma and Cooper (1977). p. 323·341.
~"
K. N., S. J. Keyser, and H. Kawasaki (1988). Toward a
Phonetic and Phonological Theory of Redundant Features. In:
Invariance and Variablity in Speech PfOC8II8l, &d. by J. S.
Pttttcel and D. H. Klatt, 428-49. HIlIdaIe, New Jersey:
Lawrence Erlbaum Auoc1at•••
Stewn.. It N. and S. J. Keyser (1989). Primary Features and Their
Enhancement In Consonants. Language 85, 81-108.
Straka, G. (1978). Apropos du classement articulatoir. des vayelles.
In O. Y. Essen, C Gutknecht, J. P. Koester, and H. H. Wa_ogler,
eda., Hamburger phoneUsche Beitraege 25. 432-460. Helmut
Bulk., Hamburg.
Whitney. W. D. (1889). Sanskrit Grammar. Harvard University Press,
Cambridge. Massachusetts.
Wierzchowska. B. (1985). Wymowa polska. PZWN. Warszawa~
Williamson. K. (1977). Multivalued Features for Consonants.
Language 53, 843-871.
Wood. S. (1975). Tense and Lax Vowels -- Degree of Constriction or
Pharyngesl Volume? Working Pa~rs of Lund University
Phonetics Laboratory 11, 109-133.
Wood. S. (1979). A Radiographic Analysis of Constriction Location
for Vowels. Journal of Phonetics 7, 25-43.
Wood, S. (1982) X-ray and Model Studies of Vowel Articulation.
Working Papers of Lund University Phonetics Laboratory 23.
Yip, M. (1987). Feature Geometry and Co-ocurrence Restrictions. Ms.
Brandeis University, Waltham, Mass.
Stroven•• P. (1980). Spectra of Fricative Noise in Human Speech. In
L&nguage end Speech 3, 32-49.
Yip, M. (1988). The Obligatory Contour Principle and Loss of Identity.
Lklguistic Inquiry 19, 65-100.
Topping, D. (1988). Chamorro vowel harmony. Oceanic Linguistics. 7,
87-79.
Younes, M. (1982). Problems in the Segmental Phonology of
Palestinian Arabic. Ph.D. Dissertation, University of Texas.
Austin.
Vago. R. (1978). Theoretical Implications of Hungarian Vowel
Harmony, Unguistic Inquiry, 7, 243·263.
Vego, R., &d. (1980). Issues in Vowel Harmony. Studies in Language
Companion Series, 6. John Benjamlns B. V., Amsterdam.
van der Hulst, H. and N. Smith (&ds.) (1982). The Structure of
Phonological Representations Part 2. Foris.
Vanna. S. (1929). Critical Studies in the Phonetic Observations of
Indian Grammarians. London.
Warburton, I., P. Kpotufe, R. Glover (1968). Ewe Basic Course.
African Studies Program. Indiana University, Bloomington.
Indiana.
Wehr, H. (1911). A Dictionary of Modern Written Arabic. Spoken
Languag. Services, Ituaca, New York.
219
220