(PDF) On Feature Spreading and the Representation of Place of Articulation

1 Introduction
1.1 Background
Since the introduction of feature geometry (Clements 1985, McCarthy 1988), four major innova-
tions in the theory have been proposed:
1. Unified Feature Theory, which employs a single set of Place features for both consonants
and vowels (Clements 1989);
2. Vowel-Place Theory, which maintains that vocalic and consonantal Place features are
segregated (Clements 1989, 1991, 1993, Nõ´ Chiosáin and Padgett 1993, Clements and
Hume 1995);
3. Partial Spreading, which allows spreading of two or more features that are not exhaus-
tively dominated by a common node (Sagey 1987, Halle 1995, Padgett 1995);
4. Strict Locality, which prohibits spreading from skipping segments (Nõ´ Chiosáin and Pad-
gett 1993, 1997, Gafos 1998, Flemming 1995).

Many thanks to Ben Bruening, Andrea Calabrese, Nick Clements, FranCois Dell, Máire Nõ´ Chiosáin, Michael Kens-
towicz, Jaye Padgett, and Cheryl Zoll for helpful comments on earlier drafts of this article, which was initially circulated
in 1993.

Linguistic Inquiry, Volume 31, Number 3, Summer 2000
387–444
q 2000 by the Massachusetts Institute of Technology 387

388 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

However, no consensus has been reached concerning which of these innovations should be incor-
porated into the theory of feature geometry. Nõ´ Chiosáin and Padgett (1993) employ 2–4 but not
1; Clements and Hume (1995) employ 1–2 but not 3–4; Halle (1995) employs 3 but not 1–2, 4;
and so on.
The purpose of this article is to set out the empirical and theoretical problems each of the
innovations encounters and to propose a new model of feature geometry and feature spreading
that avoids these problems. The new theory of phonological representations and operations that
we propose, Revised Articulator Theory, keeps Partial Spreading, but rejects Unified Feature
Theory, Vowel-Place Theory, and Strict Locality.

1.2 Outline of Revised Articulator Theory
Like other theories of phonetics/phonology, the Revised Articulator Theory (RAT) views utter-
ances as acoustic signals actualizing sequences of words, each of which is itself made up of one
or more morphemes. The morphemes are stored in speakers’ memories as sequences of discrete
phonemes. It is widely assumed that phonemes are not the ultimate constituents of speech; rather,
phonemes themselves are complexes of features. The features have a dual function. On the one
hand, they serve as mnemonic devices that distinguish one phoneme from another and hence one
morpheme from another in speakers’ memories. At the same time, each feature also serves as an
instruction for a specific action of one of the six movable portions of the vocal tract: Lips, Tongue
Blade, Tongue Body, Tongue Root, Soft Palate, and Larynx. Each articulator is capable of a
restricted set of actions of its own, and each one of these actions is associated with a particular
feature. Following Halle (1992, 1995), we distinguish two kinds of feature: articulator-bound and
articulator-free. Articulator-bound features such as [round] and [back] are necessarily executed by
one and the same articulator; articulator-free features such as [continuant] and [consonantal] are
executed by different articulators in different phonemes. The articulator executing the articulator-
free feature(s) of a phoneme is called its designated articulator. Given that the articulator-free
feature [consonantal] must be specified for every phoneme—it is in this way that consonants are
distinguished from vowels and glides—every phoneme must have a designated articulator.
Among the major postulates of the Revised Articulator Theory are these:
1. Designated articulators are indicated by features, rather than nodes in the geometry (Chom-
sky and Halle 1968, Halle 1989, 1992, 1995, Sagey 1986, Clements 1989, 1991, 1993,
Clements and Hume 1995).
2. Spreading is implemented as operations on terminal nodes in the feature tree (Halle 1995,
and in a somewhat different sense Padgett 1995).
3. Features are fully specified in underlying representations, and rules and constraints can
be indexed for marked, contrastive, or all feature specifications (Calabrese 1995).
We sketch each of these postulates in turn after first outlining the basic feature geometry that we
employ.

ON F EA TU R E SP R EAD ING 389

1.2.1 Basic Geometry We follow Mascaró 1983, Mohanan 1983, Clements 1985, and much
subsequent work in assuming that the universal set of phonological features is organized into a
hierarchy that is appropriately represented by a tree diagram. The particular structure that we
propose is given in (1).

(1) [suction]
[continuant]
[strident]
[lateral]
[round]
Lips
[labial]
[anterior]
[consonantal]
[distributed] Tongue Blade Place
[sonorant]
[coronal]
[high]
[low]
Tongue Body
[back]
[dorsal]
[nasal]
Soft Palate
[rhinal]1
[ATR]
[RTR] Tongue Root
[radical]
[spread glottis] Guttural
[constricted glottis]
[stiff vocal folds] Larynx
[slack vocal folds]
[glottal]
Theories of phonetics/phonology differ regarding the naturalness of the constituents that
make up the feature tree. RAT and some of its predecessors such as the theories of Sagey (1986)
and Halle (1988, 1989, 1992) base the organization of the tree on anatomical properties of the
articulators. Such theories draw a sharp line between tree nodes, which are always nonterminal,
and features, which, with one important exception, always occupy terminal positions in the feature
tree. We refer to such theories as Articulator Theories (AT). (The tree (1) is typical of this type
of theory.) In Articulator Theories the groupings of features in the tree reflect aspects of the
anatomy of the vocal tract. Thus, in (1) the lowest constituents (nodes) are made up of features

The articulator feature executed by the soft palate, corresponding to [labial] and the lips, [dorsal] and the tongue
body, and so on.

390 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

executed by each of the six articulators, and the next highest constituents (nodes)—Place and
Guttural—refer to articulator groups that are anatomically adjacent. The exception noted above
is the features [consonantal] and [sonorant], which (following McCarthy 1988) are assigned to
the Root node of the tree.
Articulator Theories stand in opposition to Vowel-Place Theories (VPT), which admit trees
where features dominate other features and where the nonterminal nodes of the tree are not directly
related to the anatomy of the vocal tract. For example, in most VPT accounts the feature [coronal]
dominates the features [anterior] and [distributed] (see Clements and Hume 1995:270), and the
tree geometry is based partly on the anatomy of articulators and partly on the degree of constriction
in the vocal tract. Moreover, in VPT the Place features of vowels and of consonants are grouped
under separate tree nodes, even though both groups involve the same articulators. In particular,
the V(owel)-Place features are grouped under a node that is subordinate to the C(onsonantal)-
Place node. It is the absence of a direct connection between anatomy and feature organization
that makes VPT less natural and hence more arbitrary than Articulator Theories.
The differences between the feature geometries assumed by AT and VPT are well illustrated
by their respective representations of secondary and complex articulations. In order to distinguish
phonemes with two or more primary articulations, such as kp, from phonemes involving the same
articulators but having only one primary articulation, such as kw , we need some formal way of
distinguishing primary articulations from secondary articulations. The Halle-Sagey AT model
(Sagey 1986, Halle 1988, 1989, 1992) assumed that this distinction was implemented via a pointer,
which extended from the Root node to the articulator node that executed the primary articulation.
This is illustrated in (2a). The problem with this device is the formal nature of the pointer itself,
which stands outside the normal inventory of primitives in feature geometry—features, nodes,
and association lines. Proponents of VPT avoid the problem of the pointer by encoding the
distinctions it makes within the feature tree itself: the primary articulator is dominated by the C-
Place node, whereas secondary articulations are dominated by the V-Place node, as shown in
(2b).

(2) a. Primary and secondary articulations according to Sagey (1986:91, 207)
kp kw

Root Root

Place Place

dorsal labial dorsal labial

[+round]

ON F EA TU R E SP R EAD ING 391

b. Primary and secondary articulations according to VPT (irrelevant nodes omitted)
kp kw

C-Place C-Place

V-Place

[dors] [lab] [dors] [lab]2
1.2.2 Articulator Features The model of feature geometry that we develop here also discards
the pointer, but its method of encoding distinctions between primary and secondary articulations
does not require the radical step of separating consonantal and vocalic Place features. Instead,
we accept the proposal in Halle 1992, 1995 that the underlying feature specification of every
phoneme includes an indication of the designated articulator(s) of that phoneme. Since the desig-
nated articulator executes the articulator-free features of the phoneme such as [consonantal] and
[continuant], a phoneme for which no designated articulator is indicated is incompletely specified;
crucial information for the realization of its articulator-free features is missing.
Padgett (1995) recognizes that stricture features must be supplied with information about
the articulator that is to execute them, but chooses to implement this formally by supplying a set
of stricture features to each of the Place articulators. Thus, for Padgett the Place nodes in the
feature tree are labeled with a set of features ([consonantal], [approximant], and [continuant])
that he calls the ‘‘articulator group.’’ We reproduce Padgett’s scheme in (3) (1995:17, figure
(15)).

(3) Place

Labial Coronal Dorsal

cons cons cons

approx approx approx

[cont] [cont] [cont]
This is not a viable proposal, for it implies among other things that there may be two kinds of
labiodorsal phonemes, one like the widely attested kp and gb, where both articulators are assigned

This [labial] specification would be [round] in the model of Nõ´ Chiosáin and Padgett (1993).

392 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

[`consonantal], and an undocumented one in which (e.g.) the Lips execute a [`consonantal]
articulation while the Tongue Body executes a [1consonantal] articulation.
We propose instead that the articulator that executes the stricture features of a given segment
is encoded in the form of a feature that behaves much like the other more familiar features. What
differentiates articulator features from the other features in the list is that articulator features are
unary (cf. Clements and Hume 1995:252), whereas the rest are binary. There is one articulator
feature for each of the six articulators; hence, the inventory of articulator features consists of
[labial], [coronal], [dorsal], [rhinal], [radical], and [glottal].
The use of articulator features to distinguish primary and secondary articulations is illustrated
in (4).

(4) Primary and secondary articulations in RAT
kp kw

Root Root

Place Place

Lips Tongue Body Lips Tongue Body

[–round] [lab] [dors] [+round] [dors]

articulator features
The idea of articulator features is not new. To the best of our knowledge it was first proposed
in Chomsky and Halle 1968 and was subsequently developed in Sagey 1986, Clements 1989,
1991, 1993, and Clements and Hume 1995. What has not been previously recognized is the special
character of articulator features and the vastly different roles played by articulator features and
by articulator nodes in the feature tree. It is the formal recognition of this distinction that differen-
tiates our framework from those of Sagey (1986) and Clements and Hume (1995). We believe
that the model in (4) is preferable to the V-Place model in (2b) because (along with other reasons to
be discussed in sections 2 and 3) it makes a formal distinction between entities that are empirically
different—features and nodes—and it does not postulate the abstract nodes C-Place and V-Place
(not to mention Vocalic and Aperture, which are not considered here).
Proponents of VPT justify the greater abstractness of their theory by referring to a body of
empirical phenomena, the most important of which are scalar assimilation of vowel height, long-
distance spreading, and consonant-vowel interactions. The scalar assimilation of vowel height
has been addressed in detail by Zetterstrand (1998a), who demonstrates that AT provides a more
satisfactory analysis of the facts than VPT. We summarize some of her more compelling arguments

ON F EA TU R E SP R EAD ING 393

in section 2.1.2. Discussion of the remaining two topics, long-distance spreading and consonant-
vowel interactions, forms the core of section 2. We conclude that attempts to deal with the
problems that these three phenomena pose by splitting the Place node into a V-Place node and
a C-Place node give rise to insurmountable problems.
As we demonstrate in section 3, RAT not only accounts for the problems that VPT attempts
to solve and avoids the problems that VPT creates; it also preserves structural cohesiveness, which
VPT is forced to abandon. Specifically, we show that Clements’s (1989, 1991, 1993) version of
VPT (section 2.1) critically redefines the notion of tier, whereas Nõ´ Chiosáin and Padgett’s (1993)
version of VPT (section 2.2) redefines the notion of feature. As we demonstrate in section 2,
each of these redefinitions significantly weakens the predictive power of these aspects of feature
geometry.
1.2.3 Terminal Spreading Perhaps the most important innovation of RAT is the idea that only
terminal nodes in the feature tree are allowed to spread. Many versions of terminal spreading
have been proposed in the literature (see Steriade 1987, Selkirk 1991, Goldsmith 1990, Padgett
1995); we adopt the interpretation proposed in Halle 1995, on which we elaborate below.
All versions of terminal spreading are proposed in response to the apparent existence of
partial spreading, that is, spreading of a group of features that do not constitute a natural class
in the feature tree in (1). Partial spreading of this sort is problematic for theories that allow only
one node or feature to spread at a time and that assume that blocking of a spreading node entails
blocking of all features contained under that node. One such case of partial spreading that has
received much attention is Barra Gaelic Vowel Copy, which spreads the contents of the Place
node even if the feature [back] is blocked by an intervening consonant. (We consider the Barra
Gaelic facts in detail in sections 2 and 3.)
In order to account for the Barra Gaelic facts as well as for other problems, Halle (1995)
proposes that all assimilation processes are formally characterized by spreading terminal nodes
in the tree hierarchy. According to this proposal, assimilation processes involving more than one
feature are limited to feature sets dominated exhaustively by a nonterminal node in the tree. A
set of features $ f1 f2 f3 } is exhaustively dominated by a node N if N dominates all of and only
those features. For example, node N1 exhaustively dominates the set $ f1 f2 f3 } in (5), but does
not exhaustively dominate the set $ f2 f3 } or the set $ f1 f2 f3 f5 }. In RAT the only sets of features
in (5) that would be allowed to spread together are $ f1 f2 f3 }, which is exhaustively dominated
by N1 , $ f1 f2 f3 f4 }, which is exhaustivelydominated by N3 , and $ f1 f2 f3 f4 f5 }, which is exhaustively
dominated by N4 . As shown in (1) and (5), the tree representing the features of a phoneme is a
two-dimensional object. A sequence of phonemes is therefore a three-dimensional object where
the third dimension stands for sequential order or time. To take explicit account of this fact, each
phoneme is supplied with a timing slot, represented in (6) by a capital X, which is linked to the
tree comprising the features that make up the phoneme. In a sequence of such trees each feature
(terminal node) is assumed to fall on a single straight line, and the same is true of each nonterminal
node and each timing slot. As a result, a sequence of trees constitutes a set of planes of the kind
illustrated in (6). In phonological discussions it is common to refer to planes whose bottom line

394 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

(5) N4

N3

N1 N2

f1 f2 f3 f4 f5

(6) X1 X2 X3

R1 R2 R3

N1 N2 N3

A1 A2 A3

B1 B2 B3
connects instances of a particular feature as the tier of that feature. Thus, the two bottom planes
in (6) are tiers of the features A and B, respectively. When a feature is spread from a given
phoneme to an adjacent one—for example, when feature A is spread from the third phoneme in
(6) to the second—a line is drawn on the plane A1 A3 N3 N1 connecting A3 to N2 , and the line
connecting A2 to N2 is erased. The situation becomes somewhat more complex when instead of
a single feature a group of features is spread from one phoneme to another. The theory underlying
feature geometry (see, e.g., Clements 1985) dictates that what is specified as spreading in assimila-
tion rules is a single element, be it a single (terminal) feature or a single nonterminal node. As
noted in Halle 1995, in the case of multiple feature spreading identical effects can be notated in
two distinct ways: either we can spread the subtree dominated by the spreading node so that
adjacent phonemes share this part of the tree, or we can spread the terminal features that are
dominated by the nonterminal node in question and leave the rest of the tree structure intact.
These two methods of feature assimilation are illustrated in (7). A number of compelling examples
in favor of the second method of feature spreading were discussed in Halle 1995. A particular
advantage of this formalization is that it allows for partial spreading of the features grouped under
a nonterminal node. Since only terminal features are spread, all spreading takes place on the tiers
(or planes) at the bottom of the different trees. When a single feature is assimilated, spreading
takes place only on one of these bottom tiers; when a set of features is assimilated, spreading

ON F EA TU R E SP R EAD ING 395

takes place on several of the bottom tiers. In cases of spreading on multiple tiers, it is natural to
posit that when spreading is blocked on one of the tiers, the blocking does not affect the spreading
of the features on the other tiers. This is illustrated in (8), where the blocking of spreading of
feature b does not prevent features a and c from being spread.
(7) Two versions of spreading
a. Traditional constituent spreading
Target: the Place node
Implemented as: spreading of the Place node
X X

Place

a b c
b. Terminal spreading in RAT
Target: the Place node
Implemented as: spreading of the terminal features dominated by the Place node
X X

Place

a b c

(8) X X X

Place Place Place

a b c b
In section 3 we show how terminal spreading accounts for Vowel Copy in Barra Gaelic and
Complete Vowel Copy in Tarahumara.
1.2.4 Full Specification RAT differs from other theories of feature geometry in its representation
of underspecification. Clements (1985) assumes that only contrastive features are specified in
underlying phonological representations. As a consequence, not all features in the tree in (1) are
specified in underlying representations of phonemes of every language; only contrastive features
are. However, Mohanan (1991), McCarthy and Taub (1992), Steriade (1995), and others have
pointed out a number of serious theoretical and empirical problems with existing theories of

396 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

underspecification. In this section we illustrate one of the problems for traditional underspecifica-
tion-based treatments of vowel harmony, drawn from the Turkic language Uyghur.
Treatments of vowel harmony (Clements 1976 and elsewhere) commonly assume that vowels
showing harmonic alternations are underlyingly unspecified for the harmonic feature, and that
disharmonic vowels are underlyingly prespecified for the harmonic feature (Clements and Sezer
1982 and elsewhere). To take a concrete example, the forms in (9ai) show that the Turkish
high-vowel suffix -kI alternates for the harmonic feature [round]; therefore, it is assumed to be
underlyingly underspecified for [round] (9aii). On the other hand, the suffix invariably surfaces
with a [1back] vowel (e.g., jar n-ki, not *jar n-k ) and therefore is assumed to be underlyingly
specified as [1back]. This differs from the behavior of all other high-vowel suffixes—for exam-
ple, the first-person possessive -Im, which alternates for backness (9bi) and is therefore underly-
ingly unspecified for the feature [back] (9bii). (Data and analysis are taken from Hahn 1991.)
(9) a. i. Simdi ‘now’ Simdi-ki ‘current’
bugyn ‘today’ bugyn-ky ‘today’s’
jarén ‘tomorrow’ jarén-ki ‘tomorrow’s’
ii. [round] tier
k I
[back] tier [1back]
b. i. eS ‘partner’ eS-im ‘my partner’
aS ‘food’ aS-ém ‘my food’
ii. [round] tier
I m
[back] tier
Vowel harmony in the Turkic language Uyghur works like vowel harmony in Turkish, with
a few minor differences. Like Turkish, Uyghur possesses a small number of disharmonic suffixes;
one of these is the modal suffix -Tae, which invariably surfaces as [1back] regardless of the
[back] specification of the root to which it attaches (10) (Hahn 1991:93–94). Since the vowel
in -Tae does not alternate for backness, it should be underlyingly specified as [1back] according
to the assumptions above, as shown in (11).
(10) Surface form Gloss
a. [1back] roots tyrk-T{ ‘(in the) Turkish (manner/language)’
b. [`back] roots ujGur-T{ ‘(in the) Uyghur (manner/language)’
taG -T{ ‘(one) as big as a mountain’
kitap-T{ ‘booklet’
on-T{ ‘about ten’
(11) T {
[1back]

ON F EA TU R E SP R EAD ING 397

Uyghur also possesses a rule, formalized in (12), that changes low vowels into high vowels in
medial open syllables (Hahn 1991:84); illustrative alternations are provided in (13).
(12) Uyghur Raising

Rime

(Onset) Nucleus

[–cons, +son]

[+high] [+low]

(13) Underlying form Surface form Gloss
a. /a/ bala bala ‘child’
bala-lAr balilar ‘children’
bala-lAr-I baliliri ‘his/her/its children’
b. /{/ iS{G iS{k ‘donkey’
iS{G-lAr iS{xl{r ‘donkeys’
iS{G-I iSig i ‘his/her/its donkey’
iS{G-I-GA iSig ig { ‘to his/her/its donkey’
As expected, Raising applies to the suffix -Tae- when followed by a suffix that places it in a
medial open syllable (14a); when a following suffix places -Tae- in a closed syllable, it does not
undergo Raising (14b).
(14) Underlying form Surface form Gloss
a. n{j-T{-DA n{jTid{ ‘child’
kita:b-T{-DA kitapTida ‘in the booklet’
b. n{j-T{-m-DA n{jT{md{ ‘in my little flute’
kita:b-T{-m-DA kitapT{md{ ‘in my booklet’
The forms in (14b) show that -Tae- is disharmonic and spreads its own [1back] specification to
following vowels. Given prespecification, we predict that when -Tae- undergoes Raising it should
maintain its disharmonic status, since its underlying [1back] specification (see (11)) is unaffected
by Raising. The form kitapTida in (14a) shows that this prediction is incorrect: when the ae raises
to i, it becomes transparent to [back] harmony, allowing the [`back] specification of the preceding
a to spread through it to the harmonic vowel in the following -DA suffix.
In the prespecification analysis of disharmony sketched above, the transparency of the raised
output of -Tae- requires postulation of an ad hoc rule that deletes the [1back] specification of

398 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

the i. The problem here is that the prespecification analysis misses the connection between i that
results from Raising and i that comes from underlying i, which is transparent to [back] harmony
in Uyghur (see iSig ig ae in (13b)). What is called for is a theory of harmony that takes into
consideration the role of the i in the vowel system as a whole, regardless of its origins. The system
of sensitivity and full specification that we adopt in our theory does exactly this, as we demonstrate
below.
We follow Calabrese (1995) and Halle (1995) in assuming that segments are fully specified
in all representations. It is not the case, though, that all features are visible to all rules. These
differences in feature visibility,which account for phenomena previously explained by underspeci-
fication, derive from the different status assigned to features in Calabrese’s theory, which we
sketch briefly here.
To express formally the fact that each language has its own inventory of phonemes and to
capture the crosslinguistic infrequency of certain feature combinations, Calabrese proposes that
part of the innate knowledge that humans bring to the learning of a language consists of a set of
marking statements or filters prohibiting particular feature pairs from occurring in phonemes. The
marking statement limiting the occurrence of front rounded vowels is given in (15).
(15) *[`round, 1back] in the environment [ , 1cons]
Particular languages may deactivate a given marking statement based on exposure of language
learners to positive evidence of their violation.
In each marking statement there is a marked feature, which is underlined in (15). It goes
almost without saying that marked features can play a role only in languages in which the filter
prohibiting the marked feature complex has been deactivated. Thus, only in a language in which
(15) has been deactivated will there be both [`round, 1back] and [1round, 1back] vowels.
Following Calabrese (1995), we will say that in such languages the two values of the feature
[round] are contrastive. These distinctions among features play a fundamental role in the operation
of the phonological rules.
Unless specifically noted, only contrastive features are visible to a phonological rule. In
addition to such ordinary rules there are two kinds of special rules: rules for which only marked
features are visible, and rules for which all features are visible. An example of a rule for which
only marked features are visible is Rendaku in Japanese (see Calabrese 1995:413–418); a rule
for which all features are visible is coronal assimilation in English (as illustrated by the coronal
stops in dream, hundredth, and hardship; see Clements 1985:235–236). A typical unmarked rule,
for which only contrastive features are visible, is the Barra Gaelic rule discussed in section 2.
Let us now consider how this theory of full specification and sensitivity allows us to make
sense of the behavior of the Uyghur -Tae- suffix. As noted, the problem is that the vowel of the
suffix becomes transparent to harmony after Raising changes it to i, when the prespecification
analysis predicts that it should remain disharmonic. This problem arises because the prespecifica-
tion analysis misses the relation between the i produced by Raising and the i derived from underly-
ing i, which is transparent. The theory of full specification and sensitivity that we adopt captures
this relation directly, because it derives the behavior of vowels in harmony systems from their

ON F EA TU R E SP R EAD ING 399

role in the phonemic inventory of the language, rather than from their derivational history.
Uyghur [back] harmony, for example, is analyzed as rightward spreading of contrastive
[back] specifications (16). (See Vaux 1993b and Calabrese 1995 for further elaboration of the
treatment of vowel harmony within a theory of sensitivity and full specification.)

(16) X Y

[back]
where X = a segment contrastively specified for [back]
Y = a segment that can bear a [back] specification
Since the rule is sensitive only to contrastive [back] specifications, it ignores segments that are
not contrastive for [back], such as the neutral vowel i. Crucially, this holds for i whether it results
from underlying i or from underlying a or ae that have undergone Raising. Our theory therefore
predicts (correctly) that the -Ti- allomorph of -Tae- should be transparent to [back] harmony,
because its i is not contrastive for the feature [back]. Furthermore, it predicts that there is no
language that is exactly like Uyghur, save that the output of raising a disharmonic suffix remains
disharmonic. This option is simply not available in our model, because it ignores the derivational
history of segments. We take the fact that the known data conform to our prediction, and fail to
conform to the prediction of the prespecification analysis, to support the full specification approach
adopted here.
Each of the theoretical postulates set out in the preceding sections—articulator features,
terminal spreading, and full specification—finds ample independent support in the phonological
literature. In section 3 we show how a theory incorporating these postulates enables us to account
for the difficulties encountered by theories that segregate vocalic and consonantal Place features.
It is to these theories that we now turn.

2 Vowel-Place Theory
We begin our evaluation of Vowel-Place Theory (VPT) by examining the weaknesses that have
been observed in the Halle-Sagey model of feature geometry and by arguing against the remedies
that have been proposed for these weaknesses by adherents of VPT. We confine our attention to
the two varieties of VPT that we consider to be the most promising: Unified Feature Theory
(Clements 1989, 1991, 1993, Clements and Hume 1995) and Redundant Vowel-Place Theory (Nõ´
Chiosáin and Padgett 1993). In sections 2.1 and 2.2 we present the two theories individually,
considering problems with each as they arise; in section 2.3 we examine how the two versions
of VPT deal with long-distance assimilation. In section 2.4 we summarize our arguments concern-
ing the viability of VPT.

2.1 Unified Feature Theory
The primary motivation for Unified Feature Theory (UFT) lies in the observation that the Halle-
Sagey model has difficulty accounting for (a) interactions between consonants and vowels in

400 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

terms of place of articulation and (b) scalar assimilation of vowel height. UFT deals with these
problems by reorganizing the geometry of Place features in two ways. First, it proposes a unified set
of features encoding place of articulation for consonants and vowels, in which [dorsal] subsumes
[`back], [coronal] subsumes [1back], [labial] subsumes [`round], and [pharyngeal] subsumes
[`low]. Second, it replaces the features [high] and [ATR] with a set of [open] features. We
consider these two proposals in sections 2.1.1 and 2.1.2, respectively. In section 2.1.3 we examine
a case study from Arabic in order to determine how UFT deals with more complicated consonant-
vowel (C-V) interactions, and we demonstrate that in its modeling of these interactions, UFT
creates severe problems of tier inconsistency.
2.1.1 Consonant-Vowel Interactions Many linguists have observed that interactions between
labial consonants and round vowels, coronal consonants and front vowels, dorsal consonants and
back vowels, and pharyngeal consonants and low vowels are fairly common in human languages.
A typical example involves the fronting of vowels by adjacent coronal consonants, as in the Agn
dialect of Armenian (17a), where the back vowels o and u become ø and y, respectively, after
all coronal consonants (Maxudianz 1911:28–30, Vaux 1993a).3 Noncoronals do not cause fronting
(17b).
(17) Classical Armenian Agn Gloss
a. do€ dh øÒ ‘tremor’
ga€t-uk gh aÒdyg ‘secret’
ath orÇ ath ørÇ ‘chair’
morÇ atsh o€ morÇ tsh øÒ ‘forgetting’
Th ors Th ørs ‘four’
Tux a Th yx a ‘cloth’
Dur Dh yr ‘water’
nor nør ‘new’
x oSor x oSør ‘large’
sox søx ‘onion’
galo€ gh aløÒ ‘coming’
heru hery ‘last year’
b. botsh bh otsh ‘flame’
port bord ‘navel’
ph o€kh ph ox g ‘throat’
MuSe€ MuSex a personal name
kotsh gh otsh ‘closed’
kh orÇ kh orÇ ‘unit of grain’
gud gud ‘grain’
x utsh x urtsh ‘room’
The C-V interaction documented in (17) poses a problem for the Halle-Sagey model, insofar as
it is not clear from their formalism why coronal consonants should cause vowels to become

€ 4 [`back] l, rÇ 4 trilled r.

ON F EA TU R E SP R EAD ING 401

[1back]. Coronality in consonants and backness in vowels are represented with distinct features,
and therefore interaction between the two falls outside the notion of assimilation as spreading of
a feature.
UFT proposes to solve this problem by replacing the Halle-Sagey geometry of Place features
with a single set of features characterizing place of articulation in both consonants and vowels
(see Clements and Hume 1995:275ff.). The relevant feature geometry that UFT posits is shown
in (18) (modified slightly from Clements and Hume 1995:292; irrelevant nodes omitted).
(18) [anterior] [labial]
[coronal]
[distributed] [dorsal]
[pharyngeal]
C-Place
[labial]
[coronal]
V-Place
[dorsal]
[pharyngeal]
Vocalic
[open1]
[open2]
Aperture
[open3]
Armed with this feature geometry, UFT analyzes coronal fronting of the Agn type as in (19)
(intermediate nodes omitted).

(19) C V

C-Place C-Place

[cor] V-Place
As noted above, the Halle-Sagey model does not adequately handle the interactions between
coronal consonants and front vowels encountered in such palatalization processes. To deal with
this problem, RAT has recourse to Calabrese’s (1993) ‘‘equivalency relations.’’ This idea is
implemented formally by positing that Universal Grammar includes a special set of rules whose
adoption, though not obligatory, is favored over the adoption of other rules. One such rule (modi-
fied slightly from Calabrese’s original formulation) states that there is an equivalency between
[dorsal, 1back] in vocalic contexts and [coronal] in consonantal contexts. When the [coronal]
specification of a consonant spreads to a vowel, creating a vocalic [coronal] segment, this new
configuration is subjected to the equivalency relation, yielding a [dorsal, 1back] segment. The
same interaction holds in the opposite direction for palatalization processes that convert dorsal

402 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

consonants into coronals before front vowels. Equivalency relations avoid the host of new prob-
lems created by UFT’s attempt to deal with the above phenomena by assuming that front vowels
are [coronal] and back vowels are [dorsal].4
2.1.1.1 Coronal Dependents For example, UFT assumes that in consonants the feature [coronal]
dominates the features [anterior] and [distributed] (see (18)). This entails that in some cases these
features must spread to the vowel as well. However, it is not clear within this system how these
features are implemented in vowels, or even if they exist for vowels (diagram (40) on page 276
of Clements and Hume 1995 implies that they do not; diagram (62) on page 292 implies that
they do). Since in Clements’s framework [coronal] in vowels is directly equivalent to traditional
[1back], we must suppose that he treats [coronal] as a terminal feature in vowels, but as a
nonterminal node in consonants.
2.1.1.2 The Lingual Node UFT (in the form represented by the feature tree in Clements and
Hume 1995:292) is also forced to assume that what appear to be unitary processes of [back]
harmony found in Turkish, Hungarian, and other languages actually involve two distinct harmonic
phenomena, of which one spreads [coronal] and the other spreads [dorsal]. Similarly, it is not
clear in UFT why segments with lexical [dorsal] or [coronal] specifications block harmony in
such languages. For example, in Turkish vowel epenthesis (Lees 1961) certain consonant clusters
are broken up with a high vowel except when followed by a vowel-initial morpheme. Harmony
then applies to the epenthetic vowel; thus, g j ønl-de ‘heart-dat’ becomes g j ønIl-de, which harmo-
nizes to g j ønylde. However, the three pairs of consonants that are contrastive for backness (or
coronality/dorsality)—k j -k, g j -g, and l j -l—create their own harmonic domain that blocks the
spread of [coronal] and [dorsal]. Thus, in vak j t ‘time’ the UFT geometry in (18) predicts that the
[dorsal] V-Place specification of the a will spread across the [coronal] V-Place of the k j onto the
epenthetic vowel (20), yielding the incorrect surface form *vak j t.

(20) v a kj I t

V-Place V-Place

[dors] [cor]
In reality, the [`high] vowel I surfaces as [1back]/[coronal] i, rather than [`back]/[dorsal] .
In order to avoid an arbitrary rule that would block [dorsal] spread across [coronal] V-Places,
UFT would be forced to posit three rules: one spreading the [dorsal] specification of the a onto
the I, a second spreading the [coronal] V-Place of the k j onto the I, and a third to repair the

Calabrese actually assumes that the equivalency between dorsality and coronality holds only for coronals that are
laminal (i.e., [`distributed]). This theory predicts that coronal fronting can only be caused by [`distributed] coronal
consonants; languages like English whose coronal stops are [1distributed] should not trigger vowel fronting. The Agn
coronal consonants are therefore expected to be [`distributed], but this can no longer be tested since the dialect is now
dead.

ON F EA TU R E SP R EAD ING 403

resultant ill-formed corono-dorsal feature complex. Although UFT could add an ad hoc statement
prohibiting corono-dorsal segments, it would still have no way of accurately predicting which of
the two corono-dorsal specifications is to delink in this third rule. Such a prediction would depend
on the specification of the preceding consonant: after palatal consonants the [dorsal] specification
would delink, producing a front vowel, whereas after nonpalatal consonants the [coronal] specifi-
cation would delink, producing a back vowel.
To account for these coronal-dorsal interactions, Clements and Hume (1995:290ff.) propose
that [coronal] and [dorsal] are dominated by an intermediate node, which they call Lingual (after
Browman and Goldstein 1989); they then postulate that it is the Lingual node that spreads in
backness harmony. Support for the Lingual node in the literature is quite sparse, however, and
all of the main examples adduced to support it are easily dismissed or reanalyzed, as we now
show.
(i) In the first proposal of the Lingual node in the literature, Browman and Goldstein (1989)
simply offer the idea and leave the burden of proof to later researchers. They base their proposal
on the simple anatomical observation that the tongue tip and the tongue body both belong to the
tongue organ; but by this logic the tongue root should also be included in the Lingual node.
Instead, as shown in (1), we have grouped the tongue root with the larynx under the node Guttural.
This grouping is supported in part by the fact that the movement of the tongue root is controlled
by the inferior pharyngeal constrictor muscles, which are closely related to the muscles of the
larynx rather than to muscles that control the movement of the tongue body. More importantly,
the grouping of the tongue root and the larynx is supported by the fact that they pattern together
in phonological processes, as documented extensively by Vaux (1999a) and McCarthy (1994).
(ii) Christdas (1988) claims that there is need to refer to a [1labial] class of phonemes, later
recast as a [`lingual] class, in order to deal with the facts of palatalization and secondary articula-
tion in Tamil.5 She considers two suffixes that manifest alternations between ‘‘palatal’’ and
‘‘nonpalatal’’ allophones, in which underlying k becomes k j , t becomes T, and the labials surface
without a secondary [1back] articulation (Christdas 1988:38–40, 333–339; unfortunately, she
supplies no example of a labial-initial suffix).
These facts are readily handled in RAT by means of a single prohibition against multiply-
articulated segments (i.e., segments with more than one articulator). Consonants assimilating
[1back] from an adjacent vowel come to violate this prohibition, and trigger different strategies
in order to repair the violation. No repair occurs in the case of the dorsal consonant k j (21c),
since this consonant has only one articulator. The coronal t j and the labials p j , b j , . . . involve
two articulators and therefore violate the prohibition against multiply-articulated segments. The
coronal consonant t j is subject to repair, which turns it into the [1anterior] T (13a), by Calabrese’s
(1993) equivalency relation between [dorsal, 1back] and [coronal, 1anterior]. In this case the
[dorsal, 1back] specification of the palatal secondary articulation in the t j triggers the equivalency

Clements (1989:37ff.) also invokes a [1labial] specification, in order to account for Igbo Labial Assimilation (cf.
section 2.2). However, his reasons for doing so are flawed: he claims that the assimilation rule must refer to the root
vowel, but in fact it only refers to the vowel of the reduplicant prefix.

404 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

relation, producing the [coronal, 1anterior] segment T. The labials p j , b j , . . . also violate the
prohibition against multiply-articulated segments; they are repaired by delinking the [1back]
feature ((21b); see Calabrese 1993 for more details).
(21) Palatalization and secondary articulation constraints in Tamil
a. tj ® T (via the equivalency relation between [–back] and [–anterior])
X ® X

Place Place

Blade Body Blade

[+ant] [–back] [–ant]
b. pj ® p
X ® X

Place Place

Lips Body Lips

[–round] [–back] [–round]
c. kj ® kj
X ® X

Place Place

Body Body

[–back] [–back]
(iii) Clements and Hume (1995:291) (following Clements 1976) cite a constraint of Mandarin
that prohibits nonlabial obstruents before high front vowels i, y unless the obstruents are laminal
and palatoalveolar. This excludes the dorsal, retroflex, and dental consonants k, kh , x , ts, tsh , s,
, h , from positions before high front vowels. One could argue that this set constitutes the
natural class of linguals.
The problem with this analysis lies in its inability to account for the fact that the dental stops
t, t occur freely in this position. Without these two phonemes, the remaining consonants do not

ON F EA TU R E SP R EAD ING 405

constitute a natural class. We offer no reanalysis other than the possibility that the apparent
constraint on linguals before high front vowels is actually an accidental gap, resulting from
historical changes that have subsequently been obscured. Clearly a fuller understanding of this
phenomenon would require a more extensive study of Mandarin phonology.
(iv) That no [lateral] sounds are executed by the lips is not evidence for the existence of a
Lingual node. According to Ladefoged and Maddieson (1996), Northern Waghi (a language of
New Guinea) has, in addition to laminal and apical laterals, laterals executed by the Dorsal
articulator. This establishes that [lateral] is articulator-free, since the feature can be executed by
several articulators. Although an articulator-free feature can be executed by different articulators,
it is not the case that each of the six articulators may execute each articulator-free feature. Halle
(1995:7, 12) proposes that for [`consonantal] phonemes the choice of designated articulator is
limited to the Lips, Tongue Blade, and Tongue Body, and that articulator-free features other than
[consonantal] are not available to [1consonantal] phonemes. Since, as we have just shown,
[lateral] is articulator-free, it is available only to [`consonantal] phonemes, and the choice of
designated articulator for these phonemes is limited to the three articulators dominated by the
Place node. The feature [lateral] is furthermore subject to the prohibition *[`lateral, labial],
which is included in the set of universal constraints on feature cooccurrence. (See Clements and
Hume 1995:291 for arguments for the position of [lateral] under the Root node.)
To recapitulate the discussion in this section: Clements and Hume (1995) are required to
postulate a Lingual node dominating [coronal] and [dorsal] in order to account for the many
instances of what is traditionally considered to be [back] harmony. However, the independent
arguments adduced in support of the Lingual node break down under closer examination, leaving
UFT with no viable explanation for the problems of [back] vowel harmony. (This may be the
reason why Clements and Hume abandon the Lingual node in the feature tree that they finally
adopt on page 292.)
2.1.2 The Replacement of [high] and [ATR] by [openn ] In addition to the difficulties UFT
encounters with coronals and dorsals, it runs into problems with the traditional features [high]
and [ATR]. Clements (1991) replaces vocalic [high] and [ATR] with the family of aperture
features [open1 ], [open2 ], [open3 ], and so on, which encode height differences between high, mid,
and low vowels, respectively. This proposal makes several incorrect phonological predictions.
By separating the aperture features from consonantal and vocalic Place features, and by asserting
that these aperture features characterize only vowels and not consonants, Clements predicts that
they will not show the same interactions between consonants and vowels that are found with
Place-dependent features such as [coronal] and [dorsal]. The extensive evidence for interactions
between consonant voicing and vocalic [ATR] values documented by Trigo (1991) and Vaux
(1996) is thus problematic for UFT, whereas it receives a straightforward account within AT (see
Vaux 1996).6
Clements (1991) and Odden (1991) have argued that the AT model (1) has difficulty in

In many languages vowels become [`ATR] adjacent to voiced obstruents, a phenomenon that Trigo and Vaux
attribute to a [`ATR] specification in voiced obstruents. This sort of interaction is not readily captured in Clements’s
[ATR]-less model.

406 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

accounting for cases where [ATR] and [high] values apparently spread together in Bantu languages
such as Kimatuumbi, Kinande, and Esimbi (data and arguments summarized in Kenstowicz 1994:
476ff.). Halle (1995:62ff.) has shown that these data can be interpreted without recourse to the
feature [ATR] and therefore do not pose a problem for AT.
Several compelling arguments against the multivalued [open] feature family have been pre-
sented by Zetterstrand (1996a,b, 1998a,b). We briefly rehearse two of her arguments here. The
first argument involves the problem of constraining vowel height features: it is necessary to have
enough features to distinguish all of the height contrasts attested in natural languages, yet one
must avoid postulating so many height features that unattested height contrasts are predicted. This
is not a problem in AT, where Calabrese’s (1995) theory of marking statements makes explicit,
articulatorily grounded claims about the sets of possible and impossible height contrasts. No such
theory of constraints limits the set of possible height contrasts in Clements’s [open] model, how-
ever; even a straightforward constraint like *[`high, `low] would have to be stated in terms
of an arbitrary statement *[`open1 , 1open2 ], which has no articulatory or acoustic basis (Zetter-
strand 1998b).
Zetterstrand’s second argument concerns the interaction between height in vowels and uvu-
larity in consonants. It is well known that nonhigh vowels often trigger uvularization of consonants
(e.g., in Yakut (Vaux 1999a) and Sibe (Li 1996:202)), and conversely that high vowels often
impede or undo uvularization of adjacent consonants (e.g., in Turkana (Zetterstrand 1996b)). In
AT these sorts of interactions are easily explained in terms of the feature [high]. Following
Chomsky and Halle’s suggestion (1968:305) that dorsal consonants are [`high] and uvular conso-
nants are [1high], we can analyze uvularization of consonants by nonhigh vowels as spreading
of [1high] from the vowel to the consonant. Similarly, de-uvularization by high vowels can be
analyzed as spreading of [`high] from vowel to consonant.
In Clements’s model, on the other hand, these interactions have no natural sense, because
the feature that characterizes nonhigh vowels, [`open2 ] in the case of the Turkana vowel system
(Zetterstrand 1996b:485), does not and cannot characterize uvular consonants, whose distinctive
property is either a [dorsal] secondary articulation (Zetterstrand 1996b) or a combined [dorsal]
and [pharyngeal] secondary articulation (Herzallah 1990). The reason that uvulars cannot be
[`open2 ] lies at the very heart of the logic of Clements’s model: the idea that the organizing
principle of feature geometry is constriction. In this view, in consonants the degree of constriction
is expressed in terms of the feature [continuant], whereas in vowels it is expressed in terms of
the multivalued feature [open]. For Clements, this distinction between consonantal and vocalic
constriction is essential; consonants can never bear vocalic [open] strictures, and vowels can
never bear consonantal [continuant] strictures. It is therefore impossible for uvulars, which are
consonantal, to contain specifications for the vocalic feature [open].
The result of our discussion in this section is that the features [high] and [ATR] are essential
to any viable theory of feature geometry. The UFT attempt to replace [high] and [ATR] with a
family of [open] features is unconstrained, fails to capture well-known articulatory constraints in
a principled way, and sacrifices the ability to explain a host of relatively straightforward C-V
interactions.
2.1.3 Problems of Tier Interaction In the preceding two sections we demonstrated that UFT

ON F EA TU R E SP R EAD ING 407

creates problems in its attempts to deal with C-V interactions and scalar assimilation of vowel
height. In this section we argue that UFT also runs into theory-internal difficulties involving
interactions between the C-Place and V-Place tiers. We first exemplify some of these problems
by considering a rule of dorsalization in Palestinian Arabic, and we then discuss UFT’s problems
with tier interactions in more general terms.
2.1.3.1 Palestinian Arabic Dorsalization UFT assumes that the C-Place and V-Place nodes of
(18) are identical in every respect except for their position in the feature tree and the fact that
the C-Place node dominates the V-Place node. In defense of this assumption Clements (1993),
drawing on an analysis by Herzallah (1990), cites a process in Palestinian Arabic illustrated in
(22). Closer examination of Herzallah’s analysis reveals problems that severely undermine her
claim (and Clements’s).
According to Herzallah (1990), emphatic and uvular consonants in Palestinian Arabic cause
the root vowel in the imperfective (normally a or i—(22a)) to become u (22b).
(22) Perfective Imperfective Gloss
a. Regular verbs nidim ji-ndam ‘regret’
kibir ji-kbar ‘grow up’
katab ji-ktib ‘write’
b. u-verbs qatal ji-qtul ‘kill’
sax an ji-sx un ‘get hot’
nabaÒ ji-nbuÒ ‘excel’
t.Alab ji-t.lub ‘ask for’
According to Herzallah, emphatics and uvulars are characterized by a [dorsal, pharyngeal] articula-
tion. For emphatics, these features are dominated by the V-Place node, with [coronal] as their
primary (C-Place) articulation; for uvulars, [dorsal] and [pharyngeal] are the primary specifications
and thus attach to the (C-)Place node. Consequently, Herzallah attributes the -u- in the forms in
(22b) to a [dorsal] feature that spreads from emphatics and uvulars within the root, as in (23).

(23) Herzallah’s (1990:181) representation of ‘‘[—open] Dorsalization’’7
C V]

a -Place Place

[dors] [phar] Vocalic

V-Place Aperture

[–open]
The [1open] means that the rule applies to [`high] vowels. Herzallah follows Clements (1991) in her use of

408 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

Because of complications in another rule, however, she is forced to assign [1dorsal] and [1pha-
ryngeal] V-Place articulations to the uvulars, as illustrated in (24).

(24) Herzallah’s (1990:125) representation of emphatics and uvulars (‘‘back velars’’)8
d. s. r äK

Root Root

Place Place

[cor] V-Place [dors] [phar] V-Place

[dors] [phar] [–dors] [–phar]
Besides weakening the UFT position that vocalic features are privative, this makes it unclear
which set of specifications is spreading to the vowel in the dorsalization rule (23). Since uvulars
now have V-Place, and not simply redundant V-Place but a V-Place that contrasts with the
segment’s C-Place specifications, the formalization proffered in the alpha notation of (23) becomes
questionable.
Although UFT allows rules of the sort in (23), we note that this is the only spreading rule
cited by Clements that cannot be adequately represented using Nõ´ Chiosáin and Padgett’s (1993)
more robust account of C-V interactions (see section 2.2), in which all C-V interactions occur
on the V-Place tier. It cannot be V-Place that spreads in the Palestinian Arabic case, since the
negative specifications of the uvulars interfere. In the version of Clements’s UFT that she adopts,
Herzallah might have had recourse to underspecification, had she not ordered the specification
of the uvulars’ [1dorsal, 1pharyngeal] V-Place before the dorsalization rule.
In AT, on the other hand, the Arabic facts are explained as the result of spreading [`back]
(25c).9 In (25) we illustrate how this works for the imperfective form in (22b), ji-sx un (irrelevant
structure is omitted in what follows). First, we assume that the underlying form is ji-sx in ((25a);
cf. ji-ktib in (22a)) and that uvulars and emphatics in Palestinian Arabic contain the features
[`back, 1high] ((25b); see Chomsky and Halle 1968). We can then state that the rule responsible

the feature [open] to encode height differences. The term a -Place means that the node in question can be either C-Place
or V-Place.
Herzallah uses K to transcribe the Palestinian Arabic ‘‘back velar’’ analogue of Classical Arabic’s uvular q.
Other considerations in Herzallah 1990 also find simple explanations within an AT framework. For instance,
Herzallah considers a rule of pharyngealization triggered by coronal emphatics wherein the sound ae (which she transcribes
with the letter a for typing ease; Herzallah 1990:29) becomes a. AT treats this as spreading of consonantal [RTR], which
through a prohibition *[`RTR, `ATR] changes the [`ATR] of the ae to [1ATR], producing a. For evidence that ae
is specified as [`ATR], consider the three-way unrounded low vowel distinction ae, a, a found in various dialects of
English (e.g., Boston): baet ‘bat’, ba ‘bar’, bau ‘bath’. Not only is [ATR] the only vowel feature left to account for this
distinction; in addition, ae shows the acoustic properties of [`ATR], such as a lowered F1 and raised F2 relative to its
[1ATR] counterpart a.

ON F EA TU R E SP R EAD ING 409

for the forms in (22b) spreads [`back] from a consonant to a following vowel (25c); the applica-
tion of this rule to ji-sx in is shown in (25d). Spreading of [`back] to the underlying i produces
a [`back, 1low] vowel, which triggers a redundancy rule that ensures that all back vowels are
[`round] (25e).

(25) a. Underlying form
ji-sc in
b. Underlying form of emphatics and uvulars
d. s. r q (= Herzallah’s < ä K >)

Place Place

Blade Body Body

[cor] [+back] [–high] [–low] [dors] [+back] [–high] [–low]
c. [+back] spread
C V

[+back]
d. [+back] spread in ji-sc un
j i s i n

Place Place

Body Body

[dors] [–high] [–low] [+back] [–back] [+high] [–low]
e. Redundancy rule supplies rounding in back vowels
j i s u n

Place Place

Body Body Lips

[dors] [–high] [–low] [+back] [+high] [–low] [+round]

410 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

By hypothesis, plain dorsal consonants contrast with uvulars by being [1back]; for this reason,
they do not trigger the [`back] spreading rule in (25c).
2.1.3.2 Demotion UFT formalizes C-V spreading by means of demotion rules of the form in
(26), which delinks the vowel’s original V-Place and converts its new consonantal features into
vocalic features. The rule is exemplified in (26) for the Agn form nor ‘new’ . nør in (9a)
(irrelevant nodes omitted).
(26) Palatalization (coronalization) according to UFT
a. Underlying representation
n o

C-Place C-Place

V-Place

[cor] [dors] [lab]
b. [coronal] spreads from the consonant to the vowel
n o

C-Place C-Place

V-Place

[cor] [dors] [lab]
c. Intermediate representation
n o

C-Place C-Place

V-Place

[cor] [dors] [lab]

ON F EA TU R E SP R EAD ING 411

d. Demotion (ignoring the n)

C-Place

V-Place

[cor] [dors] [lab]
e. Intermediate representation

C-Place

V-Place

[cor] [dors] [lab]
f. Delinking of incompatible features

C-Place

V-Place

[cor] [dors] [lab]
g. Surface form

C-Place

V-Place

[cor] [lab]

412 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

To avoid the theory-internal intricacies of (26b–d), UFT might simply allow the consonantal
feature to ‘‘dock’’ onto the V-Place node (as implied, but not stated directly, by figure (44)
in Clements and Hume 1995). Both demotion and docking, however, present the same problem:
if C-Place and V-Place are truly separate, then these procedures violate the notion of tier separation,
in that two tiers are interacting without referring to a common superior node. To introduce a
second means of tier interaction besides the one allowed by the feature tree itself is to undermine
the very purpose of the hierarchical model of feature geometry. UFT leaves C-Place and V-Place
in an indeterminate state where they remain on separate tiers but interact as if on the same tier
as needed. In order to decrease the theoretical power of such mechanisms, UFT must propose an
ad hoc restriction on the use of docking and demotion to C-Place/V-Place interactions.
Supporters of UFT may counter that no such restriction is needed, since no two superior
nodes other than C-Place and V-Place access the same set of features. This still fails to explain
how UFT retains the ability both to overlook specifications for any given feature owing to their
differing superior Place nodes (e.g., spreading of V-Place [labial] over the intervening C-Place
[labial] specifications of labial consonants in rounding harmony) and to let these specifications
affect each other even though they are not dominated by the same node (as in Agn coronal
fronting, where C-Place [coronal] interacts with V-Place [coronal]). The notion of independent
tiers as originally formulated remains breached.
In contrast, in the AT account the interaction or noninteraction of consonant and vowel
places is determined solely by the contrastiveness or markedness of features. AT has no difficulties
with sporadic tier independence, since it admits only one Place tier.
2.1.4 Interim Summary To summarize the discussion thus far: UFT differs substantially from
AT in its treatment of the features [1back], which it interprets as [coronal], and [ATR] and
[high], which it interprets as a family of [open] aperture features. We have shown in this section
that replacing [ATR] and [high] with [open] makes incorrect predictions about the interaction of
consonantal and vocalic [ATR] and [high] values. In the domain of Place features, we demonstrated
that the [coronal] representation of front vowels creates more problems than it solves, that the
proposed Lingual node remedy draws little support, and that the problems that [coronal] representa-
tions of frontness intend to solve are easily managed within an AT framework. Furthermore, to
the extent that the same anatomical articulators are involved in the production of both consonants
and vowels, the introduction of two Place nodes—V-Place and C-Place—treats phonological
aspects of speech sounds (in this case, the feature-geometric nodes dominating the place of articula-
tion) as distinct from their articulatory aspects (in this case, the place of articulation). In the
absence of compelling evidence to the contrary, we prefer a theory of feature geometry to maintain
as close a mapping as possible between the phonetic and the phonological properties of speech,
and UFT fails to provide this compelling evidence. Finally, and perhaps most importantly, UFT’s
appeal to docking and demotion rules renders the notion of tier inoperative.

2.2 Redundant Vowel-Place Theory
In their version of VPT, Redundant Vowel-Place Theory (RVPT), Nõ´ Chiosáin and Padgett (1993)
address UFT’s problems with tiers by postulating that all C-V interactions occur on the V-Place

ON F EA TU R E SP R EAD ING 413

tier. In this section we present the basic tenets of RVPT and demonstrate that this model, though
it solves some problems, creates additional problems in feature consistency and is incapable of
accounting for some of the data sets that it invokes to support its postulates. Moreover, despite
attempting to avoid the issue of unified features (features specifying both consonants and vowels),
RVPT’s working assumption, that there are no unified features, encounters problems within a
VPT framework.
2.2.1 Redundant V-Place As mentioned above, some advocates of VPT, and Clements (1993)
in particular, use docking or demotion rules to bridge the gap between C-Place and V-Place in
C-V interactions, as illustrated in (26). Nõ´ Chiosáin and Padgett (1993) take a more conservative
position. Within their theory C-V and V-V interactions alike are interpreted as operations on the
V-Place node, so that there are no relations between consonantal Place features and vocalic Place
features and thus no need for docking or demotion. Furthermore, in view of problems they see
in UFT (notably the bifurcation of the feature [back]), they assert only that there are equivalence
relations between [labial] and [round], [coronal] and [1back]/[`high], [dorsal] and [`back]/
[`high], and [pharyngeal] and [`low]/[`back], and they propose the Place geometry in (27)
(Nõ´ Chiosáin and Padgett 1993:4).

(27) [labial]
[coronal]
[dorsal]
[pharyngeal] Place
[round]
[back]
V-Place
[high]
[low]
In order to account for cases where a consonant’s C-Place specification spreads to vowels,
Nõ´ Chiosáin and Padgett postulate that plain consonants have inherent, redundant secondary V-
Place specifications that parallel, according to the above equivalency relations, their primary C-
Place features. In this theory the plain labial m would have the representation in (28), provided
that the language does not also contain a mw phoneme.

(28) Inherent/redundant [round] in m (Ní Chios in and Padgett 1993:2)

Place

Labial V-Place

[round]

414 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

In a language that contrasts m and mw phonemes, on the other hand, m would not have a
redundant [round] specification (Nõ´ Chiosáin and Padgett 1993:12); if it did, of course, it would
be indistinguishable from mw . This is illustrated in (29).

(29) [round] in an m : mw system (Ní Chios in and Padgett 1993:2)
mw m

Place Place

Labial V-Place Labial

[round]
Nõ´ Chiosáin and Padgett employ the representations in (28) and (29) to distinguish between
two types of C-V interactions, which they call type I and type II (1993:8–9). Type I interactions
involve the activity of distinctive V-Place features, as in Abaza (and many other Caucasian
languages), where vowels become round after labialized consonants, as in xw c ‘small’ ! xw uc
(Nõ´ Chiosáin and Padgett 1993:9).1 0 Nõ´ Chiosáin and Padgett analyze this process as spreading
of the consonant’s distinctive [`round] V-Place secondary articulation to the following vowel,
as in (30).

(30) Type I interaction in Abaza
xw @

Dorsal V-Place V-Place

[+round]
Type II interactions involve a change in the place of articulation of a vowel under the influence
of a neighboring plain consonant, as in Turkish Labial Assimilation, in which, according to Nõ´
Chiosáin and Padgett, ‘‘labial consonants cause the rounding of a following high back vowel if
preceded by a’’ (1993:10), as in armId ‘pear’ ! armut.1 1 Nõ´ Chiosáin and Padgett assert that
this phenomenon involves spreading of the redundant/inherent [round] V-Place specification of
the m to its neighbor I, as depicted in (31).

10
We assume on the basis of the Abkhaz cognate x w T j that the c employed by Nõ´ Chiosáin and Padgett is actually
a palatalized affricate T j.
11
The capital I here represents a high vowel unspecified for rounding and backness.
Zimmer (1969), van der Hulst and van de Weijer (1991), and several others argue against Labial Assimilation being
an active generalization in the synchronic phonology of Turkish; we employ it here only because it is one of Nõ´ Chiosáin
and Padgett’s main examples and serves to illustrate their point.

ON F EA TU R E SP R EAD ING 415

(31) Type II interaction in Turkish
m I

Labial V-Place V-Place

[+round]
Note that the Turkish plain m is allowed to have a redundant [round] specification, unlike Abaza
x, because Turkish contains no phoneme mw with which m contrasts for rounding.
The assumption that plain consonants can contain redundant V-Place specifications enables
RVPT to postulate that all C-V interactions occur on the V-Place tier, thereby avoiding the
problems of tier interaction faced by UFT. This assumption takes on an interesting character in
Igbo, where both plain labials and labialized consonants trigger a rule of Labial Assimilation,
contrary to what we might expect.
In Igbo the gerund is formed by prefixing a VCV sequence to a CV root. The initial V- of
the prefix is o or , depending on the [ATR] specification of the root vowel. The consonant of
the prefix is a copy of the root consonant, and the second prefixal vowel is [`high] and again
shares the [ATR] value of the root vowel. What is of interest here is a process of assimilation
that targets the second prefixal vowel when the root vowel is [1high]; in this case the vowel of
the reduplicated syllable surfaces as round before labial (32a), labiovelar (32b), and labialized
(32c) consonants, and as nonround elsewhere (32d) (Hyman 1975:53, Clements 1989:35).1 2
(32) Verb stem Gerund Gloss
a. Plain labial bè ò-bù-bè ‘cut’
bà Ò-bV̀-bà ‘enter’
fè ò-fù-fè ‘cross’
b. Labiovelar gbé ò-gbú-gbé ‘crawl’
c. Labialized kw é ò-kw ú-kw é ‘agree’
d. Elsewhere lé ò-lõ´-lé ‘look’
lá Ò-lÌ -lá ‘return’
mõ` Ò-m Ì-mÌ ‘dry’
In Nõ´ Chiosáin and Padgett’s system, Igbo Labial Assimilation would be interpreted as spreading
of the redundant [round] specification of the root consonant to the prefixal vowel, as in (33).
This treatment of plain labials as having an inherent [round] specification does obviate the need
for UFT’s docking and demotion rules. Nevertheless, as we mentioned above, some allowance
must be made to distinguish plain from rounded labials, such as p and pw , which are contrastive

12
Though Clements (1989:35) states that ‘‘whether the prefix vowel is labial (rounded) or not depends on the nature
of both the following consonant and vowel’’ (our italics), the data do not actually make clear whether the assimilation
process is triggered by the preceding or the following labial consonant. Since this is not relevant for the point we are
making here, we have simply followed Clements in assuming that the spreading is leftward.

416 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

in the Melanesian language Nambakaengo (Maddieson 1984), Nupe (Smith 1967), Ponapean, and
Mokilese (Mester 1986), or m and mw , which are contrastive in Washkuk (Maddieson 1984),
Nupe, Ponapean, and Mokilese. In such cases Nõ´ Chiosáin and Padgett posit that plain labials do
not receive the redundant specification (1993:17); that is, they are specified for (consonantal)
Labial alone and not for V-Place [round] (or possibly they are specified for [1round]).
(33) V C

Place Place

V-Place V-Place Labial

[round]
The interesting fact about Igbo is that it has both plain and labialized consonants: k contrasts
with kw , g with gw , and so on. However, the Igbo labials happen not to be contrastive for [round];
there is an m, for example, but there is no mw with which it contrasts. In Nõ´ Chiosáin and Padgett’s
theory m can therefore be redundantly specified for [round], since it does not contrast with a mw
phoneme. This allows Nõ´ Chiosáin and Padgett to account for the fact that both rounded consonants
and plain labials spread rounding in (32).
Although the predictions made by RVPT nearly match those made by AT, RVPT invokes
considerably more abstractness in its understanding of features than does AT. Features no longer
have a unique articulatory characterization, since a [`round] labial surfaces as rounded in some
languages but as unrounded in others. In fact, this mismatch between the phonology and the
phonetics can occur even within a single language, as in Igbo, where the phonologically [`round]
labialized consonants are phonetically [`round], but the phonologically [`round] plain labial
consonants are phonetically [1round]. (Nõ´ Chiosáin and Padgett (1993:16–17) discuss a similar
case in Berber.) RVPT thus provides the universal feature tree with some amount of language-
specificity, increasing its descriptive power.
RAT accounts for the Igbo facts in a manner that avoids the phonetics-phonology mismatches
and language-specificity required by RVPT. In RAT, Igbo Labial Assimilation involves leftward
spreading of the contents of the Lips node from the consonant to the preceding vowel. Given the
theory of terminal spreading that we set out in section 3, propagation of contents of the Lips node
is implemented as spreading of the terminal features dominated by that node: the articulator
feature [labial] in the case of labials and labiovelars, and the feature [round] in the case of labialized
consonants. (See figure (1) for the relevant details of the feature geometry we assume.) The results
of spreading [`round] to a preceding vowel are straightforward. Spreading of the articulator
feature [labial] to a vowel produces a vocalic [labial] segment; by an equivalency relation of the
sort discussed in section 2.1.1, this becomes [`round]. (For further discussion of the theoretical
machinery involved in this sort of process, see section 3.)

ON F EA TU R E SP R EAD ING 417

Let us now return to RVPT’s prediction about type II languages (see (31)), namely, that
plain consonants that are contrastive for a given feature will never act as if they have a positive
specification for that feature. Nõ´ Chiosáin and Padgett (1993:17) provide the following in-
stantiation of this prediction: ‘‘in a language with distinctive palatalization (represented say by
[1back] . . .) we do not expect to find fronting . . . of vowels around plain coronals, segments
which normally could be redundantly specified for these features.’’ This prediction works nicely
for Abaza and several other languages considered by Nõ´ Chiosáin and Padgett, but does not hold
true for all languages. Ironically, one such language is Turkish, which is one of the keystones of
Nõ´ Chiosáin and Padgett’s argument.
Turkish manifests a contrast for the feature [back] in three consonants: k, g, and l; we
represent the [1back] counterparts of these consonants as k j , g j , and l j , respectively. The fact
that these consonants are contrastive for [back] entails in RVPT that k j , g j , and l j have [1back]
specifications under the V-Place node (34a), and k, g, and l do not (34b).

(34) Turkish [back] contrasts in RVPT
a. kj, gj lj

Place Place

Dorsal V-Place Coronal V-Place

[–back] [–back]
b. k, g l

Place Place

Dorsal Coronal
Given these representations, we adapt Nõ´ Chiosáin and Padgett’s proposal to the Turkish case as
follows: ‘‘in a language with distinctive palatalization (represented say by [1back] . . .) we do
not expect to find fronting . . . of vowels around plain coronals, or backing of vowels around
plain dorsals, segments which normally could be redundantly specified for these features.’’ How-
ever, Turkish displays just such a process, wherein vowels are backed by plain dorsal consonants.
The data, which have already been discussed by Clements and Sezer (1982) and Clements
and Hume (1995), involve ‘‘exceptional’’ [`back] k. This exceptional k sometimes appears after
[1back] vowels, where we expect its [1back] counterpart k j because members of the set k, g,
l normally agree in backness with neighboring vowels. A typical example of exceptional k occurs

418 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

in the word tasdik ‘confirmation’ (Clements and Hume 1995:290), where we would expect *tas-
dik j .1 3 Now, in RVPT the last two segments in this word should take the form in (35).

| |
(35) i k

| |
Place Place

V-Place Dorsal

[1back]
The tenets of RVPT presented thus far require that the k be unspecified for [back], because it
contrasts with the [1back] k j . Crucially, we cannot stipulate that the k is specified as [`back],
for in doing so we would invalidate the prediction quoted above by allowing this [`back] feature
to interact with neighboring vowels.
Given our assumption that this k is unspecified for [back], we expect that it should neither
block propagation of [1back] in Turkish vowel harmony nor be able to spread [`back] (or
[1back], for that matter) to neighboring vowels. When we add to tasdik the accusative singular
suffix -I, for example, we expect that the [1back] specification of the i of the root should spread
through the k to the harmonic vowel of the suffix, yielding tasdiki (or tasdik j i, if the [1back]
feature of the following i is allowed to spread to the k). This process is formalized in (36).

(36) i k I

Place Place Place

V-Place Dorsal V-Place

[–back] [+high]
Both of the above predictions are wrong; the attested surface form tasdik demonstrates not only
that the [1back] specification of the i fails to spread to the suffix, but also that the k itself spreads
[`back] to the suffix.
Both of these facts are easily explained by assuming that the k is in fact specified as [`back];
it is then expected to block [1back] spread from the preceding i, and to spread its own [`back]
specification to the suffixal I, as depicted in (37). As illustrated in (37), this is thoroughly unprob-
lematic in RAT. RVPT could easily employ the same solution by abandoning its analysis of

13
Exceptional ks appear in loanwords involving Persian or Arabic q, or French dorsals in consonant clusters (e.g.,
French crédit ! Turkish k redi, not *k jiredi).

ON F EA TU R E SP R EAD ING 419

contrastive phonemic oppositions, but this would entail abandoning one of the central advantages
that Nõ´ Chiosáin and Padgett attribute to their theory. Furthermore, it would invalidate the VPT
analysis of Vowel Copy, which we discuss in section 2.3.2.

(37) i k I

Place Place Place

V-Place Dorsal V-Place

[–back] [+back] [+high]
RVPT encounters similar problems with rounding harmony in the Kwa language Nawuri,
as already noted by Casali (1995). According to Casali (1995:651), the vowel of the Nawuri
singular noun-class prefix gI- becomes round before a round vowel in a following syllable (38a);
this process is blocked in careful speech by intervening plain labial consonants (38b).
(38) Underlying form Surface form Gloss
a. gI-sU g U sU ‘ear’
gI-lO gUlO ‘illness’
gI-jo gujo ‘yam’
gI-ku: guku: ‘digging’
b. gI-mu gému ‘heat’
gI-fufuli géfufuli ‘white’
gI-pula gépula ‘burial’
gI-bo:to: gébo:to: ‘leprosy’
gI-kpo: gé-kpo: a type of dance
Crucially, Nawuri contrasts plain and rounded labial consonants in its phonemic inventory: p
contrasts with pw , b with bw , f with f w , and m with mw (Casali 1995:650). Given these contrasts,
RVPT requires that the rounded labials have a [round] secondary articulation under the V-Place
node (39a) and that the plain labials have no secondary [round] articulation (39b).

(39) a. Rounded labials b. Plain labials
Pw P

Place Place

Labial V-Place Labial

[round]

420 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

Since RVPT’s analysis of contrastive secondary articulations prevents the plain labial consonants
in Nawuri from having a redundant V-Place [round] specification, RVPT predicts incorrectly that
rounding should spread through plain labials, as shown in (40).
(40) g I m u

Place Place Place

V-Place Labial V-Place

[round]
Predicted output: *gumu
The blocking of rounding propagation by plain labial consonants is easily explained in RAT.
Let us assume that the rule responsible for the changes in (38a) takes the form in (41) (ignoring
the morphophonemic restrictions on the process, which are not relevant here).
(41) Spread contrastive [round] right to left from a [1consonantal, `sonorant] segment.
Given our theory of full specification (section 1.2.4), rounded labials are contrastively specified
as [`round] (42a), and plain labials are contrastively specified as [1round] (42b). (For the sake
of clarity, we omit irrelevant nodes.)

| |
(42) a. Pw b. P

| |
Place Place

| |
Lips Lips

[`round] [1round]
Since the rule in (41) is sensitive to contrastive [round] specifications, it is blocked by the con-
trastive [1round] value of plain labials, as exemplified in (43).
(43) g I m u

Place Place Place

Lips Lips Lips
[–round] [+round]

ON F EA TU R E SP R EAD ING 421

The result of the discussion in this section is that the proposal of redundant V-Place features
creates a major problem for RVPT—loss of parallelism between phonology and phonetics—and
fails to solve one of the central problems that the theory claims to account for.
2.2.2 C-V Affinities Returning to our main point, we note that Nõ´ Chiosáin and Padgett also
question the existence of unified features, commenting that they ‘‘believe the issue requires more
thought’’ (1993:3). We point out that if, as they suggest, there are no unified features in V-Place
theory, then the concept mentioned above that inherent V-Place specifications ‘‘parallel’’ their
C-Place articulatory counterparts becomes meaningless. In UFT, V-Place and C-Place features
act in parallel because they are identical. In Nõ´ Chiosáin and Padgett’s model, however, no
internal structure motivates their particular ‘‘C-V affinities’’ except the statements themselves. AT
encounters similar problems regarding coronalization and lowering processes, which it resolves by
recourse to Calabrese’s (1993) equivalency relations, as discussed in section 2.1.1. However,
RAT gives a straightforward account for labialization and dorsalization processes, by linking
them through the articulators Lips and Tongue Body, as we show in section 3. Nõ´ Chiosáin and
Padgett’s model not only has the same problems with coronalization and lowering processes that
AT does, but also has no structure-internal way to connect Labial with [round] or Dorsal with
[back].
2.2.3 Changes in C-Place Another major piece of evidence Nõ´ Chiosáin and Padgett (1993)
use to support their theory of V-Place-only interactions is the observation that, crosslinguistically,
C-Place does not change under the influence of a following vowel; thus, *su ! fu, *za ! a,
and *f ! x . In the few cases where such changes have actually been observed, Nõ´ Chiosáin and
Padgett posit a series of less dramatic historical steps, which remain in a telescoped form in
morpholexical rules. Note, however, that at some point in the chain of small steps a change must
occur for which Nõ´ Chiosáin and Padgett’s model has no account. Take, for instance, the Bantu
chain they cite from Hyman 1976 (Nõ´ Chiosáin and Padgett 1993:28): pi ! ph i ! ps i ! ts i !
si. Between the aspirated ph and the affricate ps something of the vowel’s articulation must become
a coronal articulation in the consonant, unless the coronality is accounted for by a default rule
calling on the underspecification of Coronal. Such an account, however, could not easily explain
why the rule applies only before i and not before other vowels. Nõ´ Chiosáin and Padgett might
attempt to account for the weak link of this historical chain with a restructuring rule specific to
palatalizations, but this would not explain the other Bantu data they cite, such as ku ! fu, or
Japanese hu ! fu.
2.2.4 Place Assimilation One final item of evidence invoked by Nõ´ Chiosáin and Padgett to
support a V-Place node is the behavior of nasals in modern Irish (Nõ´ Chiosáin and Padgett 1993:
7). As illustrated in (44), word-final coronal nasals assimilate the primary place of articulation
of a following stop, but crucially do not assimilate the secondary articulation of the stop. Thus,
in (44a) the palatalized n j assimilates the Dorsal primary articulation of the following g, but does
not assimilate its [`back] secondary articulation. In (44b) the [`back] n assimilates the Dorsal
primary articulation of the following g, but does not assimilate its [1back] secondary articulation.
(In the discussion below, it is important to bear in mind that Irish contrasts plain and palatalized
versions of most consonants in its inventory. This contrast is formally expressed by assigning
these consonants the feature [5back].)

422 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

(44) Form Gloss
a. dj ekj hj inj ‘I would see’
dj ekj hj iÎj gan e: ‘I would see without it’
b. dj i:l@n ‘a diary’
dj i:l@Îgj i:vj rj i ‘a winter’s diary’
According to Nõ´ Chiosáin and Padgett, this is an instance of Place assimilation. Hence, if we
assume that palatalization in dorsals is represented as a [1back] specification under the Dorsal
node, we expect that Place assimilation in (44b) should produce *d j i:l j g j i:v j r j i, as depicted in
(45).

(45) n gj

Place Place

Coronal Dorsal

[–back]
Nõ´ Chiosáin and Padgett assert that in order to account for the nonassimilation of the palatal
secondary articulation in (44b), ‘‘we must adopt a structure in which the V-Place feature represent-
ing palatalization is independent of Dorsal’’ (1993:7).
Nõ´ Chiosáin and Padgett offer no support other than these Irish data for their assertion that
Dorsal and [back] are not connected in the feature tree, a position that would surely have repercus-
sions, if not crosslinguistically, at least elsewhere in Irish phonology. Moreover, Nõ´ Chiosáin and
Padgett’s analysis itself—that the phenomena in (44) are instances of Place assimilation—is not
the only, or necessarily the best, treatment of the data.
It is important to note in this connection that the data in (44) do not reflect the normal behavior
of Nasal Place Assimilation in Irish, which spreads both primary and secondary articulations
simultaneously. In (44) the Dorsal primary articulation of the second consonant spreads without
simultaneously spreading its [back] secondary articulation. We must therefore distinguish two
assimilation rules in Irish:
1. Nasal Place Assimilation, which spreads both primary and secondary articulations;
2. Dorsal Assimilation (44), which spreads only the Dorsal primary articulation.
It is true that Nõ´ Chiosáin and Padgett’s theory can account for both of these processes efficiently:
the former would involve spreading of the Place node, and the latter would involve spreading of
the Dorsal node.
A similarly efficient account is available in RAT. Like Nõ´ Chiosáin and Padgett, we assume
that Nasal Place Assimilation involves spreading of the features dominated by the Place node,

ON F EA TU R E SP R EAD ING 423

whereas Dorsal Assimilation involves spreading of the Dorsal articulation. However, our account
differs from theirs in that the Dorsal component that spreads is not the Dorsal node in the feature
tree, but the feature [dorsal]. Since, as we stated in section 1, articulator features can spread like
all other features, independently of the articulator nodes Lips, Tongue Body, and so on, spreading
of the terminal feature [dorsal] does not entail spreading of the other features dominated by the
Tongue Body node. In effect, what is spreading is the designated articulator [dorsal] for the
segment in question; the other features executed by the Tongue Body articulator remain unaffected.
(See section 3.3 for further elaboration of this analysis.)
Like most languages, Irish does not admit phonemes with more than one designated articula-
tor. As a consequence, when the [dorsal] articulator feature of the following consonant is assimi-
lated by the preceding nasal, the nasal loses its original [coronal] specification, resulting in the
feature composition [dorsal, `nasal]—that is, j or , depending on whether or not the nasal
was originally [1back]. When the designated articulator is assimilated, it is only this feature—and
no other feature dominated by the articulator—that is assimilated.
The Irish facts are therefore crucial to a proper understanding of feature geometry, as Nõ´
Chiosáin and Padgett rightly observe. However, these facts are crucial not because they point
exclusively at a VPT analysis—the same data are equally compatible with the RAT presented
here—but because they demonstrate the correctness of the traditional notion of articulator features.
We return to this point in more detail in section 3.

2.3 Long-Distance Assimilation
Recall from section 1 that the proponents of VPT have claimed that AT cannot account for long-
distance assimilation. In this section we examine how VPT treats the two logical types of long-
distance assimilation, involving consonants and vowels, respectively. We argue that both versions
of VPT fail to provide a satisfactory account of the former, and that RVPT is unable to reconcile
its account for the latter with the explanation it was forced to adopt for the Turkish facts discussed
earlier. The alternative RAT account of these facts encounters none of these difficulties.
2.3.1 Consonant Harmony, Vocalic Transparency, and Strict Locality Nõ´ Chiosáin and Padgett
(1993) note that V-Place theories in general have difficulty excluding such unattested interactions
as *nak ! ak, where the Onset consonant assimilates the designated articulator of the Coda
consonant. This sort of interaction is expected in VPT, since individual consonantal Place features
can spread across vowels, as long as they do not require the involvement of the Place node. In
order to exclude this class of unattested interactions, Nõ´ Chiosáin and Padgett introduce a constraint
stating that ‘‘spreading of C-Place articulator features is strictly local’’ (1993:47). This is an
undesirable move, since the factual evidence for the constraint is weak and, more importantly,
since the introduction of such locality constraints greatly weakens the strong hypothesis that all
tiers allow action at a distance.
That aside, such locality constraints incorrectly predict the nonexistence of consonant har-
mony systems. Cases of cross-vocalic Coronal harmony have been documented in Sanskrit (Steri-
ade 1986, Schein and Steriade 1986) and Tahltan (Shaw 1991, Halle 1995:39–42). In order to

424 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

account for the ability of the Coronal node to harmonize across vowels, and to disallow other
types of unattested consonant harmonies (e.g., Dorsal, Labial), Nõ´ Chiosáin and Padgett (1993:
47) postulate a Site node subordinate to Coronal, as shown in (46).
(46) Place

Coronal

Site

[cont] [ant] [dist]
Since no other evidence is cited to support this Site node, this is a purely ad hoc move. The
existence of consonant harmony and the uniqueness of Coronal harmony are not a problem for
AT, because vowels have contrastive Labial and Dorsal specifications to block the spreading of
these consonantal features, but they have no similar specification for Coronal (for details see
Halle 1995).
Nõ´ Chiosáin and Padgett’s version of VPT, which has no demotion rules, allows no further
possibilities for long-range consonant interactions. In UFT, however, demotion rules allow C-
Place specifications to become V-Place specifications. Thus, the geometry does not prohibit the
unattested scenario in (47). This type of process includes not only long-range labializa-
tion, but also long-range palatalization (*paS ! p j aS ) and long-range pharyngealization (*ti !
t i ). In Nõ´ Chiosáin and Padgett’s version of VPT, where demotion does not occur, and especially
in AT, where there is no C-/V-Place separation, these interactions are not allowed.

(47) [kap] ® [kwap]
a. Spreading
k a p

C-Place C-Place C-Place

[dors] [lab]

V-Place

[phar]

ON F EA TU R E SP R EAD ING 425

b. Demotion
kw a p

C-Place C-Place C-Place

[lab] [dors] [lab]

V-Place V-Place

[phar]

c. Output
kw a p

C-Place C-Place C-Place

[dors] [lab]

V-Place V-Place

[lab] [phar]
Limited long-range consonant interactions other than Coronal harmony do occur. Herzallah
(1990) notes a process in Palestinian Arabic whereby emphasis spreads leftward from a Coronal
emphatic onto other consonants throughout a word. Since Herzallah represents emphasis in Coro-
nals as a combined [dorsal, pharyngeal] V-Place specification, she must posit that Emphasis
Spread involves the entire V-Place node, which cannot spread through the V-Place of intervening
vowels. She formalizes Emphasis Spread by ordering it before tier conflation, so that the interven-
ing vowels’ V-Place nodes cannot interfere. It should be noted, however, that of the four rules
that she orders before tier conflation, this rule of Emphasis Spread is oddly the only nonmorpholog-
ical rule.
In the RAT framework developed here, appeal to tier conflation and extrinsic rule ordering
is not necessary. We instead account for Emphasis Spread in Palestinian Arabic by calling on
the noncontrastiveness of vowels for the feature [RTR], which we assume to be the feature
involved in the rule. Say that we have a hypothetical word tatatat., where the last consonant is a
Coronal emphatic that spreads emphasis onto the three preceding consonants, ignoring the inter-
vening vowels; the surface result is t.at.at.at.. If we postulate that Emphasis Spread spreads con-

426 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

trastive [RTR] specifications leftward, then we expect the [RTR] of vowels to be ignored, since
vowels are not contrastive for this feature. We illustrate this scheme in (48), omitting irrelevant
details; noncontrastive [RTR] specifications are shown in italics.

(48) t a t a t a t

[–RTR] [–RTR] [–RTR] [+RTR]
In the last few years a number of researchers have adopted a different strategy for dealing
with long-distance consonant interactions. Flemming (1995), Gafos (1996), Walker (1996, 1998),
Nõ´ Chiosáin and Padgett (1997), and others assert that nonlocal spreading does not exist at all,
and therefore consonants cannot interact at a distance unless intervening segments are also affected
by the process. In the words of Nõ´ Chiosáin and Padgett (1997:2), ‘‘[S]egments are either blockers
or participants in spreading; there is no transparency.’’ This theory of Strict Locality requires
that a word involving rounding harmony such as Turkish somun ‘loaf’ has the surface structure
in (49a), not (49b).
(49) a. s o m u n

[round]

b. s o m u n

[round]
Vaux (1999b) demonstrates that Strict Locality encounters fatal difficulties with processes that
spread a feature through segments that are contrastive for that feature. Space considerations prevent
us from repeating the arguments here, but we present one illustrative case from the Tungusic
language Sibe.
In Sibe the velar segments k and x (which are [`high]; see Vaux 1999a) become respectively
the uvulars q and x (which are [1high]) when preceded anywhere in the word by a [1high]
vowel (Li 1996:201–202). In (50a), for example, the suffixal dorsal consonant in the word for
‘long’ becomes uvular by virtue of the fact that it is preceded by the [1high] vowel ö; contrast
this with the forms in (50c), where the same suffix surfaces with a k, because it is not preceded
by a [1high] vowel. This type of uvularization, which is also found in the Turkic language Yakut,
can be analyzed as spreading of [1high] from a vowel to a dorsal consonant.
(50) Form Gloss
a. G ölmi(n)-qén ‘long’
Dalu-qun ‘full’
aDi(g)-qén ‘small’

ON F EA TU R E SP R EAD ING 427

b. bOdu-x u ‘to consider’
lavdu-x u ‘to become more’
ömi-x é ‘to drink’
c. éldé(n)-kén ‘bright’
ulu-kun ‘soft’
té-xé ‘to sit’
tyry-xu ‘to rent’
Much like Turkish, Sibe has two series of vowels that contrast for the feature [high].
(51) [`high] i y é u
[1high] E ö a O
Given this vowel system, the uvularization facts in (50) are counterexamples to Strict Locality:
in the word for ‘long’, for example, spreading of [1high] from the ö to the suffixal consonant
should cause the intervening i to become its [1high] counterpart E. In fact, Strict Locality leads
us to expect all vowels intervening between a [1high] vowel and a uvular to become [1high],
but the forms in (50a–b) show that this is not what happens. The problem for Strict Locality here
is that strictly local spreading would incorrectly neutralize phonemic contrasts in intervening
segments. As long as Strict Locality is required, this problem will not go away.
In RAT we analyze the Sibe uvularization process as rightward spreading of marked [high]
specifications to eligible consonants. Given the structure of the Sibe vowel inventory, we can
assume that [1high] is marked in vowels, and [`high] is unmarked. The uvularization rule
therefore spreads the height specification of nonhigh vowels, but not of high vowels. By the
same token, only marked feature specifications are visible to the rule, and intervening [`high]
specifications are thus ignored. The bare bones of this analysis are sketched in (52).

(52) a. Spread marked [high] rightward to dorsal consonants.
b. [high] is only marked in [–high] segments.
c. Unmarked [high] specifications are not visible to the rule.
d. l a v d u - X u

[–high]
2.3.2 Vowel Copy The final major argument advanced against AT by proponents of RVPT is
that vowel copy processes of the kind encountered in Barra Gaelic are incorrectly blocked in the
Halle-Sagey model. In this section we describe the Barra Gaelic facts and consider how they are
dealt with in the two versions of RVPT. We conclude that RVPT is unable to account for the
Barra Gaelic data, because of its assumptions about the nature of feature spreading.
Barra Gaelic (Borgstrøm 1937, 1940, Clements 1987, Sagey 1987) repairs certain sequences
with an epenthetic vowel that is an exact copy of the preceding vowel, except when the intervening
consonant has a contrastive [back] specification (in this dialect backness is contrastive for all conso-
nants except Labials and the Coronals n, R), in which case the epenthetic vowel agrees in backness
with this consonant. The basic facts are given in (53) (data from Clements 1987, Sagey 1987).

428 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

(53) Underlying form Surface form Gloss
a. After [`back] alp@ alap@ ‘Scotland’
sonorants sj {rv sj {rav ‘bitter’
sj {nx@s sj {nax@s ‘conversation’
dunx@g dunux@g ‘Duncan’
urpel urupel ‘tail’
Orm OrOm ‘on me’
marv marav ‘dead’
f{rk f{rak ‘anger’
b. After [1back] mörj v mörj ev ‘the dead’
sonorants bulj kj bulj ikj ‘bellows (gen.sg.)’
merj kj merj ekj ‘rust’
c. After noncontrastive tj imxj al tj imixj al ‘round about’
sonorants {msj irj {m{sj irj ‘time’
Proponents of RVPT have noted that the Halle-Sagey model fails to account for facts of the
sort in (53) (see Nõ´ Chiosáin and Padgett 1993:4–5). The reason for this failure is that Vowel
Copy in the Halle-Sagey model involves spreading of the vowel’s Place node, since the Labial
feature [round] spreads simultaneously with the Dorsal features [high], [low], and [back]. As
shown in (54), this would be incorrectly blocked by the Place node of an intervening consonant.
(We return to this topic in section 3.1.)

(54) Vowel Copy through a plain velar consonant: Halle-Sagey
[–cons] [+cons] [–cons]
Place Place Place

Labial Dorsal Dorsal

[round] [low] [high] [back]
RVPT proposes to resolve this dilemma by segregating the Place features of consonants and
vowels in the manner described above. This change enables the V-Place node of one vowel to
spread safely through an intervening consonant to the next vowel, as depicted in (55). However,
this scheme only works if we assume that plain consonants do not have redundant V-Place
specifications in such cases. If, for example, the Barra Gaelic pair r and r j were specified as
[`back] and [1back], respectively, the V-Place node of both consonants would block the spread
of the vowel’s V-Place node, as shown in (56a–b). In the RVPT account depicted in (56), Vowel
Copy will never apply through a contrastive consonant, because its V-Place node will always
block the propagation of the vowel’s V-Place node.

ON F EA TU R E SP R EAD ING 429

(55) Vowel Copy through a plain velar consonant: RVPT
[–cons] [+cons] [–cons]

(C-)Place (C-)Place (C-)Place

Dorsal

V-Place V-Place

(56) V-Place spreading from vowel to vowel blocked by intervening consonant
with contrastive V-Place specification: RVPT
a. V r V

C-Place C-Place C-Place
V-Place V-Place V-Place

[+back]
b. V rj V

C-Place C-Place C-Place
V-Place V-Place V-Place

[–back]
The requirement that the plain consonants be unspecified for [back] in Barra Gaelic does
not appear at first glance to pose any problems for RVPT; Nõ´ Chiosáin and Padgett (1993) state
explicitly that plain consonants do not bear redundant V-Place specifications in such situations.
The problem for this proposal is the Turkish facts discussed earlier: in order to account for the
behavior of exceptional k, RVPT must postulate that the plain members of contrastive pairs are
specified for the contrastive feature. In short, RVPT cannot have its cake and eat it too: by
invoking redundant V-Place, it accounts for Turkish but not for Barra Gaelic; and by rejecting
redundant V-Place, it accounts for Barra Gaelic but not for Turkish.
One might object at this point that it is possible to stipulate within RVPT that the instances
of exceptional k in Turkish are prespecified as [`back], but ‘‘unexceptional’’ occurrences of

430 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

[`back] k are underlyingly unspecified. This approach to exceptionality,advocated by Inkelas and
Cho (1993) and Inkelas, Orgun, and Zoll (1997), is reasonable within the phonologicalframeworks
espoused in those papers. However, this sort of variable prespecification deprives Nõ´ Chiosáin
and Padgett’s framework of any predictive power. Recall that Nõ´ Chiosáin and Padgett state that
their theory predicts that ‘‘in a language with distinctive palatalization (represented say by
[1back] . . .) we do not expect to find fronting . . . of vowels around plain coronals, segments
which normally could be redundantly specified for these features’’ (1993:17). By extension, we
should not expect to find backing of vowels around plain Dorsals. If we allow ourselves to account
for exceptions to this prediction by specifying just those exceptional segments with the necessary
feature specification, the prediction becomes meaningless.
An even more severe problem for the RVPT account is that no version of RVPT can account
for Vowel Copy through palatalized consonants. Since these consonants are assigned a [1back]
V-Place specification regardless of one’s theory of underspecification, they will invariably block
propagation of the preceding vowel’s V-Place node.
RAT, on the other hand, is able to account for both phenomena, as we show in section 3.

2.4 Interim Summary
We have shown in section 2 that the two versions of VPT fall short of their goals on every front.
They fail to demonstrate in many cases that AT is unable to account for the relevant data, and
the solutions they propose for the same problems sometimes fail, and almost always create more
problems than they solve. For these reasons, it is preferable to maintain AT, provided we can
demonstrate that it is able to account in a satisfactory manner for all of the objections raised by
proponents of VPT. Establishing that the modified version of AT developed here, RAT, is able
to account for these objections is the task of the next section.

3 Revised Articulator Theory
Our demonstration that RAT can account for all of the objections raised by advocates of VPT
consists of several parts. We organize our presentation according to the problematic facts that
remain to be accounted for: Partial Vowel Copy (section 3.1), Complete Vowel Copy (section
3.2), Irish Nasal Place Assimilation (section 3.3.1), and Irish Dorsal Assimilation (section 3.3.2).

3.1 Partial Vowel Copy
Recall from section 2.3.2 that in certain situations Barra Gaelic inserts an epenthetic vowel that
is an exact copy of the preceding vowel, except when the intervening consonant has a contrastive
[back] specification, in which case the epenthetic vowel agrees in backness with this consonant.
The relevant facts are repeated in (57). The problem that these data pose for the Halle-Sagey
model is that in (57a–b) the vocalic features [round], [back], [high], and [low], whose minimal
common mother node is the Place node, spread without [back], which is also dominated by the
Place node. According to the Halle-Sagey model, propagation of the Place node in this case
should either override the [back] specification of intervening consonants (58a) or be blocked
(58b). (Note that the solid arrows are pointers in Sagey’s (1986) sense, not spreading nodes.)

ON F EA TU R E SP R EAD ING 431

(57) Underlying form Surface form Gloss
a. After [`back] alp@ alap@ ‘Scotland’
sonorants sj {rv sj {rav ‘bitter’
sj {nx@s sj {nax@s ‘conversation’
dunx@g dunux@g ‘Duncan’
urpel urupel ‘tail’
Orm OrOm ‘on me’
marv marav ‘dead’
f{rk f{rak ‘anger’
b. After [1back] mörj v mörj ev ‘the dead’
sonorants bulj kj bulj ikj ‘bellows (gen.sg.)’
merj kj merj ekj ‘rust’
c. After noncontrastive tj imxj al tj imixj al ‘round about’
sonorants {msj irj {m{sj irj ‘time’

(58) Vowel Copy in the Halle-Sagey model
a. Feature-changing Vowel Copy: intervening features are overridden
m rj V v

Root Root Root

Place Place Place

Labial Dorsal Dorsal Coronal

[–round] [–high] [–low] [+back] [–back]
b. Feature-filling Vowel Copy: intervening features block spreading
m rj V v

Root Root Root
Place Place Place

Labial Dorsal Dorsal Coronal

[–round] [–high] [–low] [+back] [–back]

432 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

However, once we assume that several terminal features can spread individually and simul-
taneously (recall section 1.2.3), a straightforward account for the Barra Gaelic facts becomes
available. This account states that the set of terminal features dominated by the Place node spreads
from a preceding vowel to the epenthetic vowel, except when a consonant intervenes for which
backness is contrastive; in this case the vocalic [back] feature is blocked by the Line-Crossing
Prohibition (Sagey 1986). This process is illustrated in (59).

(59) m rj V v

Root Root Root

Place Place Place

Labial Dorsal Dorsal Labial Dorsal

[–round] [–high] [–low] [+back] [–back]

Though terminal spreading represents a radical departure from previous theories of feature
interaction, we strongly believe that it is required to account for the Barra Gaelic facts. As it
turns out, terminal spreading also helps us to make sense of most of the objections raised by VPT
that have so far remained unanswered. We consider these points in the remainder of this section.
(For additional evidence illustrating the need for terminal feature spreading, see Halle 1995.)

3.2 Complete Vowel Copy: Tarahumara

In section 2 we mentioned that the phenomenon of Complete Vowel Copy over plain Dorsal
consonants is problematic for the Halle-Sagey model. In fact, it was this problem that provided
the initial motivation for VPT. With terminal spreading, though, a straightforward account for
Complete Vowel Copy is available.
A typical example of Vowel Copy is found in the Uto-Aztecan language Tarahumara, where
the vowel of the deverbal suffix -kV is an exact copy of the final vowel of the root (Nida 1949:
23).

(60) Verb Gloss Noun Gloss
miTiru ‘make shavings’ miTiru-ku ‘shavings’
reme ‘make tortillas’ reme-ke ‘tortillas’
paTi ‘grow corn’ paTi-ki ‘ear of corn’
opaTa ‘be dressed’ opaTa-ka ‘garment’

Tarahumara has the following phonemic inventory (Burgess 1984):

ON F EA TU R E SP R EAD ING 433

(61) Consonants
[labial] p b m w
[cor, `ant] t s n l r
[cor, 1ant] T j
[dorsal] k g
[glottal] ? h
Vowels
[1back] [`back]
[`high], [1low] i u
[1high], [1low] e o
[1high], [`low] a
In the Tarahumara consonantal system, the only contrastive features subordinate to the Place node
are the articulator features and the Tongue Blade feature [anterior], which distinguishes [`ante-
rior] t from [1anterior] T. Crucially, k and g do not contrast with any other Tongue Body
consonants and therefore have no contrastive features that are dependents of the Tongue Body
node other than [dorsal]. Given these facts, we can now represent the Vowel Copy process as
spreading of the Place node of the root-final vowel to the vowel of the suffix. In accordance with
(7b), the spreading of the Place node is implemented as spreading of the terminal features it
dominates, as shown in (61). Note that only contrastive features are visible to the harmony rule;
noncontrastive features—shown in italics—are transparent to this spreading process.

(62) m i T i r u k V

Place Place Place

Labial Dorsal Dorsal Labial Dorsal

[+round] [+high] [+back] [–low] [+high] [+back] [–low]
We conclude that once terminal spreading is assumed, AT is readily able to account for the
observed consonantal transparency in the case of Vowel Copy.1 4
Several more complicated instances of Vowel Copy cited by Odden (1991) in which only
the features [back] and [round] appear to spread have been shown by Halle (1995) to be consistent
with the analysis presented here. To the best of our knowledge, then, RAT is able to account for
all cases of complete and Partial Vowel Copy and is therefore to be preferred over both VPT,

14
The Klamath case discussed by Nõ´ Chiosáin and Padgett (1993:4–5) should be treated as an instance of partial
reduplication rather than of Vowel Copy; Gafos (1998) makes similar arguments for Temiar.

434 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

which cannot account for Partial Vowel Copy, and AT models employing nonterminal spreading,
which cannot account for either type of Vowel Copy.

3.3 Articulator Features
RAT must still account for one problem, which involves delinking of secondary articulations. In
this section we argue that this problem is best dealt with by reintroducing traditional privative
articulator features. We base our discussion on the two Irish assimilation phenomena discussed
in section 2: Nasal Place Assimilation (section 3.3.1) and Dorsal Assimilation (section 3.3.2).
3.3.1 Irish Nasal Place Assimilation Terminal feature spreading encounters problems in the
formulation of certain rules, such as Nasal Place Assimilation in Irish (see the discussion below
(44)). In nasal place assimilation rules, [coronal] nasals acquire the Place specifications of the
following consonant; for instance, in Irish coronal n typically surfaces as Dorsal when followed
by g. Within a terminal-feature-spreading framework, with full specification, this example must
be formalized as in (63).

(63) n g

Place Place

Blade Body Body

[+ant] [+back] [+back]
In this formulation there is no reason for the [`anterior] feature to delink, since no new [anterior]
specification is forcing it out. Furthermore, in Irish we cannot call on a repair rule to delink the
Tongue Blade specification because, as we showed in (44), double specifications for Tongue
Blade and Tongue Body are legal and even required by the phonology of the language, which
contrasts plain and palatalized consonants.
We propose to account for the problem raised in (63) by postulating the existence of articula-
tor features. The idea of articulator features dates back at least to Chomsky and Halle 1968 and
survives to this day in the articulator features employed by Clements and Hume (1995) (though
they assume that these features can dominate other features, whereas we do not).
The notion of articulator features is well grounded in the general architecture of speech,
besides being motivated by phonologicalevidence. We adopt Halle’s (1992) insight that all speech
is produced by actions of the six articulators manipulated in the human vocal tract: Lips, Tongue
Blade, Tongue Body, Soft Palate, Tongue Root, and Larynx. Each articulator is capable of a small
set of actions of its own. From this point of view the features that figure in discussions of phonetics
and phonology are instructions for specific actions of an articulator. This conception of the speak-
ing process is formally implemented in the RAT tree structure in (1), where features executed

ON F EA TU R E SP R EAD ING 435

by the same articulator are grouped under a common nonterminal node, whereas the higher nodes
Place and Guttural group anatomically contiguous articulators, and the articulator-free features
are directly dominated by the Root node of the tree.
The property of being a designated articulator shares all important characteristics with phono-
logical features. In particular, in assimilation processes the designated articulator can spread, just
like any other feature, leaving all other features intact (this is the case in Irish Dorsal Assimilation,
discussed in section 2.2 and below). This type of assimilation process cannot be readily expressed
in Sagey’s pointer notation. Like other features, the designated articulator must be specified in
the list of features characterizing a given phoneme in underlying representations of morphemes.
It is in this way that we capture the primary Place distinctions among consonants, such as that
between the labial nasal m and its coronal and dorsal counterparts n and .
To express this role of the articulators in formal terms, we supply each articulator with a
feature indicating that it functions as designated articulator. We have therefore included in (1),
under each articulator node, a terminal feature assigning designated articulator status to this
articulator. Although every phoneme has at least one designated articulator, a phoneme may
(rarely) have more than one such articulator. The labiovelar stop kp, for example, includes in its
lexical representation the articulator features [dorsal] and [labial] in addition to [`consonantal,
1sonorant, 1round, 1continuant, . . .]. By contrast, in most languages the labialized velar kw
has the feature complement [dorsal, `consonantal, 1sonorant, `round, 1continuant, . . .], with
no specification for the feature [labial] (see Halle 1995).
It is crucial not to confuse the terminal articulator features [labial], [dorsal], and so on, with
the articulator nodes Lips, Tongue Body, and so on, dominated by the Place node in the feature
tree in (1). Though they refer to the same component of the articulatory apparatus, the terminal
articulator features encode information about the role of an articulator in the production of a
particular phoneme. The nonterminal nodes, on the other hand, characterize groups of features
that can pattern together in phonological processes (spreading and delinking), and are generated
in an entirely predictable manner from the feature complement of a given phoneme. Furthermore,
as noted above, designated articulators, being terminal features, are able to spread from one
segment to another, without affecting other nodes and features. These distinctions are crucial in
accounting for Irish Nasal Place Assimilation and Dorsal Assimilation, as we now show.
Let us return first to the Irish Nasal Place Assimilation facts. Given our new theory of
articulator features, the Irish ng sequence would begin with the underlying representation in (64a);
note that the n contains the articulator feature [coronal], and the g contains the articulator feature
[dorsal]. Nasal Place Assimilation then spreads the contents of the Place node of the g—including
its [dorsal] articulator feature—to the preceding n (64b). Since Irish does not allow segments to
contain more than one articulator feature, the [coronal] articulator feature of the n delinks (64c),
which in turn entails delinking of the [`anterior] feature, which is not a valid secondary articula-
tion in Irish (64d). These two procedures yield the correct surface representation g (64e). By
postulating articulator features, then—which are motivated on independent grounds—we are able
to account for the recalcitrant subtleties of Nasal Place Assimilation in Irish.

436 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

(64) a. Underlying representation
n g

Place Place

Blade Body Body

[cor] [+ant] [+back] [dors] [+back]
b. Spreading of the contents of the Place node of the g
n g

Place Place

Blade Body Body

[cor] [+ant] [+back] [dors] [+back]
c. Delinking of the [coronal] articulator feature
n g

Place Place

Blade Body Body

[cor] [+ant] [dors] [+back]
d. Delinking of [+anterior]
n g

Place Place

Blade Body Body

[+ant] [dors] [+back]

ON F EA TU R E SP R EAD ING 437

e. Surface representation
n g

Place Place

Body Body

[dors] [+back]
3.3.2 Irish Dorsal Assimilation The striking facts of Irish Dorsal Assimilation that we discussed
in section 2.2 also fall into place once we equip AT with articulator features. Recall that without
articulator features, AT was unable to account for the fact that Dorsal Assimilation does not affect
Dorsal secondary articulations; we expect forms like *d j i:l j g j i:v j r j i rather than the attested
d j i:l g j i:v j r j i, from underlying d j i:l n ` g j i:v j r j i. With articulator features, though, we can
simply state that Dorsal Assimilation involves spreading of the articulator feature [dorsal], as
depicted in (65). The derivations of ‘I would see without it’ and ‘a winter’s diary’ then proceed
as in (66) and (67), respectively.

(65) Dorsal Assimilation
Spread [dorsal] to a word-final coronal nasal from a right-adjacent consonant.
[+cons, +son] # [+cons]

Place Place

Soft Palate Blade Body

[+nasal] [dors]

(66) djekjhjinj + gan e: ® djekjhji j
gan e:
a. Underlying form
nj g

Place Place

Blade Body Body

[cor] [+ant] [–dist] [–back] [dors] [+back]

438 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

b. Dorsal Assimilation
nj g

Place Place

Blade Body Body

[cor] [+ant] [–dist] [–back] [dors] [+back]
c. Delinking of disallowed second articulator feature, together with attendant features
nj g

Place Place

Blade Body Body

[cor] [+ant] [–dist] [–back] [dors] [+back]
d. Surface form
÷j g

Place Place

Body Body

[–back] [dors] [+back]

ON F EA TU R E SP R EAD ING 439

(67) dji:l n + gji:vjrji ® dji:l gji:vjrji
a. Underlying form
n gj

Place Place

Blade Body Body

[cor] [+ant] [–dist] [+back] [dors] [–back]
b. Dorsal Assimilation
n gj

Place Place

Blade Body Body

[cor] [+ant] [–dist] [+back] [dors] [–back]
c. Delinking of disallowed second articulator feature, together with attendant features
n gj

Place Place

Blade Body Body

[cor] [+ant] [–dist] [+back] [dors] [–back]
d. Surface form
gj

Place Place

Body Body

[+back] [dors] [–back]

440 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

4 Conclusions
In this article we have considered two models of feature geometry: Articulator Theory (AT),
which views phonetic features as instructions for actions of the six articulators, and Vowel-
Place Theory (VPT), which calls upon the notion of constriction degree to separate vocalic and
consonantal Place nodes in phonological processes. We have shown that RAT, our proposed
revision of AT, accounts in a principled manner for all of the objections that have been raised
by proponents of VPT. We have also shown that Unified Feature Theory (UFT; Clements 1989,
1991, 1993, Clements and Hume 1995) fails to capture the unitary behavior of [5back], which
plays a role in many phonological systems, and does not explain how its unified-feature Coronal
dependents behave in vowels. In addition, we have demonstrated that data traditionally taken to
support the proposed Lingual node revision of VPT can easily be explained without it. Further-
more, UFT incorrectly predicts that consonantal and vocalic [ATR] and [high] values will not
interact. Most importantly, in order to explain C-V interactions, UFT relies on processes of
docking or demotion, which significantly weaken the notion of tier and allow unattested consonant
harmonies such as *kap ! kw ap.
Nõ´ Chiosáin and Padgett’s (1993) version of VPT is just as problematic as UFT. In their
attempt to remedy UFT’s problem with tiers, Nõ´ Chiosáin and Padgett must invoke redundant
secondary articulations in plain consonants. This introduces excessive language-specificity in the
interpretation of Universal Grammar features and fails to account for certain Turkish and Barra
Gaelic facts. Moreover, Nõ´ Chiosáin and Padgett’s ban on C- and V-Place interactions is not
supported adequately either by the Irish data discussed here, which RAT accounts for by reintro-
ducing the traditional notion of articulator features, or by their reference to the absence of articula-
tor changes such as su ! fu, which we have shown to be untenable. Further, in rejecting unified
features, they have considerably weakened VPT’s structure.
In addition, neither variety of VPT considered here is able to account for the instances of
consonant harmony described in section 2.3, most notably Shaw’s (1991) Tahltan data. Attempts
to account for these data have resulted in ad hoc proposals for a locality constraint and a Site
node under the Coronal node.
The two versions of VPT considered above and the Halle-Sagey version of AT (Halle 1992,
Sagey 1986) are all unable to account for cases of Partial Vowel Copy as discussed in section
3.1, because their theory of spreading holds that all and only nodes in the feature tree can spread
at any one time. We have shown that once we limit spreading to terminal feature nodes in the
tree, we not only gain an explanation for the cases of Partial Vowel Copy, but also acquire the
machinery necessary to account for the instances of Complete Vowel Copy within AT. Since
these cases, which were the original motivation for postulating VPT, are now manageable within
RAT, there is no longer any reason to maintain VPT.
Finally, we have clarified the special role of articulator features and their behavior in various
assimilation processes. Of particular note is the result that all assimilation processes involve the
spreading of terminal features. This establishes a sharp distinction between terminal and nontermi-

ON F EA TU R E SP R EAD ING 441

nal nodes in the tree. Whereas the terminal feature nodes participate fully in the implementation
of phonological rules and in the determination of the conditions under which feature assimilation
is blocked, the nonterminal nodes are limited to defining the allowable sets of features that may
be spread by a single rule.
Since the structure of nonterminal nodes in a phoneme is entirely predictable from the set
of terminal features that make up the phoneme, the speaker need not remember this structure.
This suggests that the representations of individual morphemes in speakers’ memories are com-
posed solely of feature bundles, and that the nonterminal tree structure is constructed as needed
in each derivation or computation.This marks a return—albeit in a more abstract and sophisticated
form—to the older view of phonemes as unstructured feature bundles. RAT therefore represents
an amalgam of very solid notions that have stood the test of time with more novel concepts whose
validity has been established by rigorous confrontation with complex data from a variety of
languages. It is our hope that this article will lead others to explore our proposals further, so as
to correct them where necessary and confirm and extend them where possible.

References
Borgstrøm, C. Hj. 1937. The dialect of Barra in the Outer Hebrides. Norsk Tidsskrift for Sprogvidenskap
8:71–242.
Borgstrøm, C. Hj. 1940. A linguistic survey of the Gaelic dialects of Scotland. Vol. 1, The dialects of the
Outer Hebrides. Norsk Tidsskrift for Sprogvidenskap, suppl. Bind 1, Oslo.
Browman, Catherine, and Louis Goldstein. 1989. Articulatory gestures as phonological units. Phonology 6:
201–251.
Burgess, Don. 1984. Western Tarahumara. In Studies in Uto-Aztecan grammar, vol. 4, ed. Ronald Langacker,
3–149. Dallas, Tex.: Summer Institute of Linguistics.
Calabrese, Andrea. 1993. On palatalization processes: An inquiry about the nature of a sound change. Ms.,
Harvard University, Cambridge, Mass., and University of Connecticut, Storrs.
Calabrese, Andrea. 1995. A constraint-based theory of phonological markedness and simplification proce-
dures. Linguistic Inquiry 26:373–463.
Casali, Roderic. 1995. Labial opacity and roundness harmony in Nawuri. Natural Language & Linguistic
Theory 13:649–663.
Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. New York: Harper & Row.
Christdas, Prathima. 1988. The phonology and morphology of Tamil. Doctoral dissertation, Cornell Univer-
sity, Ithaca, N.Y.
Clements, G. N. 1976. Vowel harmony in nonlineargenerativephonology:An autosegmentalmodel. [Distrib-
uted by Indiana University Linguistics Club, Bloomington (1980).]
Clements, G. N. 1985. The geometry of phonological features. Phonology Yearbook 2:225–252.
Clements, G. N. 1987. Toward a substantive theory of feature specification. In Proceedings of NELS 18,
79–93. GLSA, University of Massachusetts, Amherst.
Clements, G. N. 1989. A unified set of features for consonants and vowels. Ms., Cornell University, Ithaca,
N.Y.
Clements, G. N. 1991. Vowel height assimilation in Bantu languages. In Proceedings of the Special Session
on African Language Structures 17S, 25–64. Berkeley Linguistics Society, University of California,
Berkeley.

442 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

Clements, G. N. 1993. Lieu d’articulation des consonnes et des voyelles. In L’architecture et la géométrie
des représentations phonologiques, ed. Bernard Laks and Annie Rialland, 101–145. Paris: Editions
du CNRS.
Clements, G. N., and Elizabeth Hume. 1995. The internal organization of speech sounds. In Handbook of
phonological theory, ed. John Goldsmith, 245–306. Oxford: Blackwell.
Clements, G. N., and Engin Sezer. 1982. Vowel and consonant disharmony in Turkish. In The structure of
phonologicalrepresentations,part II, ed. Harry van der Hulst and Norval Smith, 213–255. Dordrecht:
Foris.
Flemming, Edward. 1995. Vowels undergo consonant harmony. Paper presented at the Trilateral Phonology
Weekend, Stanford University.
Gafos, Adamantios. 1996. The articulatory basis of locality in phonology. Doctoral dissertation, The Johns
Hopkins University, Baltimore, Md.
Gafos, Diamandis [Adamantios]. 1998. Eliminating long-distance consonantal spreading. Natural Lan-
guage & Linguistic Theory 16:223–278.
Goldsmith, John. 1990. Autosegmental and metrical phonology. Oxford: Blackwell.
Hahn, Reinhard. 1991. Spoken Uyghur. Seattle: University of Washington Press.
Halle, Morris. 1988. The immanent form of phonemes. In The making of cognitive science, ed. William
Jeffrey Hurst, 167–183. Cambridge: Cambridge University Press.
Halle, Morris. 1989. The intrinsic structure of speech sounds. Ms., MIT, Cambridge, Mass.
Halle, Morris. 1992. Phonological features. In International encyclopedia of linguistics, vol. 3, ed. William
Bright, 207–212. Oxford: Oxford University Press.
Halle, Morris. 1995. Feature geometry and feature spreading. Linguistic Inquiry 26:1–46.
Herzallah, Rukaiyah. 1990. Aspects of Palestinian Arabic phonology. Doctoral dissertation, Cornell Univer-
sity, Ithaca, N.Y.
Hulst, Harry van der, and Jeroen van de Weijer. 1991. Topics in Turkish phonology. In Turkish linguistics
today, ed. Hendrick Boeschoten and Ludo Verhoeven, 11–59. Leiden: Brill.
Hyman, Larry. 1975. Phonology: Theory and analysis. New York: Holt, Rinehart and Winston.
Hyman, Larry. 1976. Phonologization. In Linguistic studies offered to Joseph Greenberg, ed. Alphonse
Juilland, 407–418. Saratoga, Calif.: Anma Libri.
Inkelas, Sharon, and Young-mee Yu Cho. 1993. Inalterability as prespecification. Language 69:529–574.
Inkelas, Sharon, Orhan Orgun, and Cheryl Zoll. 1997. Exceptions and static phonologicalpatterns: Cophonol-
ogies vs. prespecification. In Derivations and constraints in phonology, ed. Iggy Roca, 393–418.
Oxford: Oxford University Press.
Kenstowicz, Michael. 1994. Phonology in generative grammar. Oxford: Blackwell.
Ladefoged, Peter, and Ian Maddieson. 1996. The sounds of the world’s languages. Oxford: Blackwell.
Lees, Robert. 1961. The phonology of Modern Standard Turkish. Bloomington: Indiana University Press.
Li, Bing. 1996. Tungusic vowel harmony: Description and analysis.The Hague: Holland Academic Graphics.
Maddieson, Ian. 1984. Patterns of sounds. Cambridge: Cambridge University Press.
Mascaró, Joan. 1983. Phonological levels and assimilatory processes. Ms., Universitat Autònoma de Barce-
lona.
Maxudianz, Mesrop. 1911. Le parler arménien d’Akn (Quartier Bas). Paris: Librairie Paul Geuthner.
McCarthy, John. 1988. Feature geometry and dependency: A review. Phonetica 43:84–108.
McCarthy, John. 1994. The phonetics and phonology of Semitic pharyngeals. In Papers in laboratory
phonology 3, ed. Patricia Keating, 191–233. Cambridge: Cambridge University Press.
McCarthy, John, and Alison Taub. 1992. Review of Paradis and Prunet 1991. Phonology 9:363–370.
Mester, R. Armin. 1986. Studies in tier structure.Doctoral dissertation,University of Massachusetts,Amherst.

ON F EA TU R E SP R EAD ING 443

Mohanan, K. P. 1983. The structure of the melody. Ms., MIT, Cambridge, Mass., and National University
of Singapore.
Mohanan, K. P. 1991. On the bases of radical underspecification. Natural Language & Linguistic Theory
9:285–325.
Nõ´ Chiosáin, Máire, and Jaye Padgett. 1993. Inherent V-Place. (UC Santa Cruz Working Papers.) Linguistics
Research Center, University of California, Santa Cruz.
Nõ´ Chiosáin, Máire, and Jaye Padgett. 1997. Markedness, segment realisation, and locality in spreading.
Report LRC-97-01, Linguistics Research Center, University of California, Santa Cruz.
Nida, Eugene. 1949. Morphology. Ann Arbor: University of Michigan Press.
Odden, David. 1991. Vowel geometry. Phonology 8:261–290.
Padgett, Jaye. 1995. Feature classes. In University of Massachusetts occasional papers 18: Papers in Opti-
mality Theory, 385–420. GLSA, University of Massachusetts, Amherst.
Sagey, Elizabeth. 1986. The representationof features and relations in nonlinear phonology. Doctoral disser-
tation, MIT, Cambridge, Mass.
Sagey, Elizabeth. 1987. Non-constituent spreading in Barra Gaelic. Ms., University of California, Irvine.
Schein, Barry, and Donca Steriade. 1986. On geminates. Linguistic Inquiry 17:691–744.
Selkirk, Elisabeth. 1991. Major place in the vowel space: Vowel height features. Ms., University of Massachu-
setts, Amherst.
Shaw, Patricia. 1991. Consonant harmony systems: The special status of coronal harmony. In Phonetics
and phonology 2: The special status of coronals, ed. Carole Paradis and Jean-FranCois Prunet,
125–157. San Diego, Calif.: Academic Press.
Smith, Neilson. 1967. The phonology of Nupe. Journal of African Languages 6:153–169.
Steriade, Donca. 1986. A note on coronal. Ms., MIT, Cambridge, Mass.
Steriade, Donca. 1987. Locality conditions and feature geometry. In Proceedings of NELS 17, 339–362.
GLSA, University of Massachusetts, Amherst.
Steriade, Donca. 1995. Underspecification and markedness. In The handbook of phonological theory, ed.
John Goldsmith, 114–174. Oxford: Blackwell.
Trigo, Loren. 1991. On pharynx-larynx interactions. Phonology 8:113–136.
Vaux, Bert. 1993a. Coronal fronting in the Armenian dialect of Akn. Annual of Armenian Linguistics 14:
15–29.
Vaux, Bert. 1993b. The origins of Altaic labial attraction. In Harvard working papers in linguistics 2,
228–237. Department of Linguistics, Harvard University, Cambridge, Mass.
Vaux, Bert. 1996. The status of ATR in feature geometry. Linguistic Inquiry 27:175–182.
Vaux, Bert. 1999a. A note on pharyngeal features. In Harvard working papers in linguistics 7, 39–63.
Department of Linguistics, Harvard University, Cambridge, Mass. [An earlier version from 1993
entitled ‘‘ATR-Voice Interactions and the Geometry of the Lower Vocal Tract’’ was circulated in
manuscript form.]
Vaux, Bert. 1999b. On locality. Ms., Harvard University, Cambridge, Mass. [An oral version was presented
at the 1999 annual meeting of the Linguistic Society of America, Los Angeles.]
Walker, Rachel. 1996. Neutral segments and locality. Ms., University of California, Santa Cruz.
Walker, Rachel. 1998. Nasalization, neutral segments and opacity effects. Doctoral dissertation, University
of California, Santa Cruz.
Zetterstrand, Sylvia. 1996a. A formal comparison of two models of vowel height. Ms., Harvard University,
Cambridge, Mass.
Zetterstrand, Sylvia. 1996b. High vocoids in Turkana: Evidence for [high]. In NELS 26, 473–487. GLSA,
University of Massachusetts, Amherst.
Zetterstrand, Sylvia. 1998a. The phonological representation of vowel height. Doctoral dissertation, Harvard
University, Cambridge, Mass.

444 M OR R IS HAL LE, B ER T VAU X, AN D A NDR EW WO LFE

Zetterstrand, Sylvia. 1998b. Scalar effects in Italian metaphony. In Theoretical analyses on Romance lan-
guages (LSRL XXVI), ed. José Lema and Esthela Treviño, 351–366. Amsterdam: John Benjamins.
Zimmer, Karl. 1969. Psychological correlates of some Turkish morpheme structure conditions. Language
45:309–321.

(Halle)
Department of Linguistics and Philosophy
E39-218
MIT
Cambridge, Massachusetts 02139

(Vaux)
Department of Linguistics
Harvard University
Cambridge, Massachusetts 02138
[email protected]
(Wolfe)
Department of Linguistics
Harvard University
Cambridge, Massachusetts 02138
References (67)
Borgstrøm, C. Hj. 1937. The dialect of Barra in the Outer Hebrides. Norsk Tidsskrift for Sprogvidenskap 8:71-242.
Borgstrøm, C. Hj. 1940. A linguistic survey of the Gaelic dialects of Scotland. Vol. 1, The dialects of the Outer Hebrides. Norsk Tidsskrift for Sprogvidenskap, suppl. Bind 1, Oslo.
Browman, Catherine, and Louis Goldstein. 1989. Articulatory gestures as phonological units. Phonology 6: 201-251.
Burgess, Don. 1984. Western Tarahumara. In Studies in Uto-Aztecan grammar, vol. 4, ed. Ronald Langacker, 3-149. Dallas, Tex.: Summer Institute of Linguistics.
Calabrese, Andrea. 1993. On palatalization processes: An inquiry about the nature of a sound change. Ms., Harvard University, Cambridge, Mass., and University of Connecticut, Storrs.
Calabrese, Andrea. 1995. A constraint-based theory of phonological markedness and simplification proce- dures. Linguistic Inquiry 26:373-463.
Casali, Roderic. 1995. Labial opacity and roundness harmony in Nawuri. Natural Language & Linguistic Theory 13:649-663.
Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. New York: Harper & Row.
Christdas, Prathima. 1988. The phonology and morphology of Tamil. Doctoral dissertation, Cornell Univer- sity, Ithaca, N.Y.
Clements, G. N. 1976. Vowel harmony in nonlineargenerativephonology:An autosegmentalmodel. [Distrib- uted by Indiana University Linguistics Club, Bloomington (1980).]
Clements, G. N. 1985. The geometry of phonological features. Phonology Yearbook 2:225-252.
Clements, G. N. 1987. Toward a substantive theory of feature specification. In Proceedings of NELS 18, 79-93. GLSA, University of Massachusetts, Amherst.
Clements, G. N. 1989. A unified set of features for consonants and vowels. Ms., Cornell University, Ithaca, N.Y. Clements, G. N. 1991. Vowel height assimilation in Bantu languages. In Proceedings of the Special Session on African Language Structures 17S, 25-64. Berkeley Linguistics Society, University of California, Berkeley.
Clements, G. N. 1993. Lieu d'articulation des consonnes et des voyelles. In L'architecture et la ge ´ome ´trie des repre ´sentations phonologiques, ed. Bernard Laks and Annie Rialland, 101-145. Paris: Editions du CNRS.
Clements, G. N., and Elizabeth Hume. 1995. The internal organization of speech sounds. In Handbook of phonological theory, ed. John Goldsmith, 245-306. Oxford: Blackwell.
Clements, G. N., and Engin Sezer. 1982. Vowel and consonant disharmony in Turkish. In The structure of phonologicalrepresentations,part II, ed. Harry van der Hulst and Norval Smith, 213-255. Dordrecht: Foris.
Flemming, Edward. 1995. Vowels undergo consonant harmony. Paper presented at the Trilateral Phonology Weekend, Stanford University.
Gafos, Adamantios. 1996. The articulatory basis of locality in phonology. Doctoral dissertation, The Johns Hopkins University, Baltimore, Md.
Gafos, Diamandis [Adamantios]. 1998. Eliminating long-distance consonantal spreading. Natural Lan- guage & Linguistic Theory 16:223-278.
Goldsmith, John. 1990. Autosegmental and metrical phonology. Oxford: Blackwell.
Hahn, Reinhard. 1991. Spoken Uyghur. Seattle: University of Washington Press.
Halle, Morris. 1988. The immanent form of phonemes. In The making of cognitive science, ed. William Jeffrey Hurst, 167-183. Cambridge: Cambridge University Press.
Halle, Morris. 1989. The intrinsic structure of speech sounds. Ms., MIT, Cambridge, Mass.
Halle, Morris. 1992. Phonological features. In International encyclopedia of linguistics, vol. 3, ed. William Bright, 207-212. Oxford: Oxford University Press.
Halle, Morris. 1995. Feature geometry and feature spreading. Linguistic Inquiry 26:1-46.
Herzallah, Rukaiyah. 1990. Aspects of Palestinian Arabic phonology. Doctoral dissertation, Cornell Univer- sity, Ithaca, N.Y.
Hulst, Harry van der, and Jeroen van de Weijer. 1991. Topics in Turkish phonology. In Turkish linguistics today, ed. Hendrick Boeschoten and Ludo Verhoeven, 11-59. Leiden: Brill.
Hyman, Larry. 1975. Phonology: Theory and analysis. New York: Holt, Rinehart and Winston.
Hyman, Larry. 1976. Phonologization. In Linguistic studies offered to Joseph Greenberg, ed. Alphonse Juilland, 407-418. Saratoga, Calif.: Anma Libri.
Inkelas, Sharon, and Young-mee Yu Cho. 1993. Inalterability as prespecification. Language 69:529-574.
Inkelas, Sharon, Orhan Orgun, and Cheryl Zoll. 1997. Exceptions and static phonologicalpatterns: Cophonol- ogies vs. prespecification. In Derivations and constraints in phonology, ed. Iggy Roca, 393-418. Oxford: Oxford University Press.
Kenstowicz, Michael. 1994. Phonology in generative grammar. Oxford: Blackwell.
Ladefoged, Peter, and Ian Maddieson. 1996. The sounds of the world's languages. Oxford: Blackwell.
Lees, Robert. 1961. The phonology of Modern Standard Turkish. Bloomington: Indiana University Press.
Li, Bing. 1996. Tungusic vowel harmony: Description and analysis.The Hague: Holland Academic Graphics. Maddieson, Ian. 1984. Patterns of sounds. Cambridge: Cambridge University Press.
Mascaro ´, Joan. 1983. Phonological levels and assimilatory processes. Ms., Universitat Auto `noma de Barce- lona.
Maxudianz, Mesrop. 1911. Le parler arme ´nien d'Akn (Quartier Bas). Paris: Librairie Paul Geuthner. McCarthy, John. 1988. Feature geometry and dependency: A review. Phonetica 43:84-108.
McCarthy, John. 1994. The phonetics and phonology of Semitic pharyngeals. In Papers in laboratory phonology 3, ed. Patricia Keating, 191-233. Cambridge: Cambridge University Press.
McCarthy, John, and Alison Taub. 1992. Review of Paradis and Prunet 1991. Phonology 9:363-370.
Mester, R. Armin. 1986. Studies in tier structure.Doctoral dissertation,University of Massachusetts,Amherst.
Mohanan, K. P. 1983. The structure of the melody. Ms., MIT, Cambridge, Mass., and National University of Singapore.
Mohanan, K. P. 1991. On the bases of radical underspecification. Natural Language & Linguistic Theory 9:285-325.
Nõ ´Chiosa ´in, Ma ´ire, and Jaye Padgett. 1993. Inherent V-Place. (UC Santa Cruz Working Papers.) Linguistics Research Center, University of California, Santa Cruz.
Nõ ´Chiosa ´in, Ma ´ire, and Jaye Padgett. 1997. Markedness, segment realisation, and locality in spreading. Report LRC-97-01, Linguistics Research Center, University of California, Santa Cruz.
Nida, Eugene. 1949. Morphology. Ann Arbor: University of Michigan Press.
Odden, David. 1991. Vowel geometry. Phonology 8:261-290.
Padgett, Jaye. 1995. Feature classes. In University of Massachusetts occasional papers 18: Papers in Opti- mality Theory, 385-420. GLSA, University of Massachusetts, Amherst.
Sagey, Elizabeth. 1986. The representationof features and relations in nonlinear phonology. Doctoral disser- tation, MIT, Cambridge, Mass.
Sagey, Elizabeth. 1987. Non-constituent spreading in Barra Gaelic. Ms., University of California, Irvine.
Schein, Barry, and Donca Steriade. 1986. On geminates. Linguistic Inquiry 17:691-744.
Selkirk, Elisabeth. 1991. Major place in the vowel space: Vowel height features. Ms., University of Massachu- setts, Amherst.
Shaw, Patricia. 1991. Consonant harmony systems: The special status of coronal harmony. In Phonetics and phonology 2: The special status of coronals, ed. Carole Paradis and Jean-FranCois Prunet, 125-157. San Diego, Calif.: Academic Press.
Smith, Neilson. 1967. The phonology of Nupe. Journal of African Languages 6:153-169.
Steriade, Donca. 1986. A note on coronal. Ms., MIT, Cambridge, Mass.
Steriade, Donca. 1987. Locality conditions and feature geometry. In Proceedings of NELS 17, 339-362. GLSA, University of Massachusetts, Amherst.
Steriade, Donca. 1995. Underspecification and markedness. In The handbook of phonological theory, ed. John Goldsmith, 114-174. Oxford: Blackwell.
Trigo, Loren. 1991. On pharynx-larynx interactions. Phonology 8:113-136.
Vaux, Bert. 1993a. Coronal fronting in the Armenian dialect of Akn. Annual of Armenian Linguistics 14: 15-29.
Vaux, Bert. 1993b. The origins of Altaic labial attraction. In Harvard working papers in linguistics 2, 228-237. Department of Linguistics, Harvard University, Cambridge, Mass.
Vaux, Bert. 1996. The status of ATR in feature geometry. Linguistic Inquiry 27:175-182.
Vaux, Bert. 1999a. A note on pharyngeal features. In Harvard working papers in linguistics 7, 39-63. Department of Linguistics, Harvard University, Cambridge, Mass. [An earlier version from 1993 entitled ''ATR-Voice Interactions and the Geometry of the Lower Vocal Tract'' was circulated in manuscript form.]
Vaux, Bert. 1999b. On locality. Ms., Harvard University, Cambridge, Mass. [An oral version was presented at the 1999 annual meeting of the Linguistic Society of America, Los Angeles.]
Walker, Rachel. 1996. Neutral segments and locality. Ms., University of California, Santa Cruz.
Walker, Rachel. 1998. Nasalization, neutral segments and opacity effects. Doctoral dissertation, University of California, Santa Cruz.
Zetterstrand, Sylvia. 1996a. A formal comparison of two models of vowel height. Ms., Harvard University, Cambridge, Mass.
Zetterstrand, Sylvia. 1996b. High vocoids in Turkana: Evidence for [high].
Zetterstrand, Sylvia. 1998a. The phonological representation of vowel height. Doctoral dissertation, Harvard University, Cambridge, Mass.
FAQs
AI
What explains the differences between Revised Articulator Theory and Vowel-Place Theory?
add
The study finds that RAT effectively retains anatomical links in feature geometry, while VPT introduces contrasting nodes that produce inconsistencies, especially in vowel harmony.
How does terminal spreading function in Revised Articulator Theory?
add
Terminal spreading in RAT permits only terminal nodes in the feature tree to spread; for instance, features can spread independently from one another, facilitating partial assimilation.
What role do articulator features play in phonological representations?
add
Articulator features distinguish how phonemes are produced by various articulators and must be specified in underlying representations, influencing processes such as assimilation.
How does the Revised Articulator Theory address vowel harmony in Uyghur?
add
RAT provides a full specification approach, solving vowel harmony inconsistencies by accurately linking features to underlying representations, unlike traditional underspecification models.
What empirical support underlies the postulates of Revised Articulator Theory?
add
RAT's postulates, including terminal spreading and articulator features, are supported by numerous cross-language phonological phenomena, including detailed analyses of Irish and Turkish interactions.
Bert Vaux
University of Cambridge, Faculty Member
Papers
224
Followers
9,410
View all papers from
Bert Vaux
arrow_forward
Related papers
Feature Spreading and Vocalic Place
Bert Vaux
1994
Download free PDF
View PDF
chevron_right
Spreading In Functional Phonology
Paul Boersma
1970
The occurrence of and the restrictions on the temporal spreading of phonological feature values (assimilation, harmony) are the results of interactions between the functional principles of minimizing articulatory effort and minimizing perceptual confusion. This proposal is tested on the typology of opacity to nasal spreading. While the sonority approach of Gnanadesikan (1995) meets with insuperable problems with regard to the position of /h/ in the hierarchy, and the feature-geometric representational approach of Piggott (1992) needs to take recourse to ad-hoc conditions in UG in order to get the hierarchy right, the functional approach accurately predicts the attested typology. 1 The functional approach to spreading We can distinguish several fundamental functional principles, all of which can lead to the phenomenon of feature or gesture spreading. 1 This paper is chapter 19 of Boersma (1998), which is available from the author (http://fonsg3.hum.uva.nl /paul/) or from the publisher (http://www.hagpub.com). 2 I write underlying perceptual specifications between pipes, articulatory implementations between square brackets, and perceptual results between slashes. 3 I would like to use terminology that is unbiassed with respect to the oral/nasal distinction, i.e., I would regard [p] and [a] as oral and non-nasal, [m] as nasal and non-oral, and [a) ] as oral and nasal. The traditional term for this interpretation of 'oral' is 'continuant': an unfortunate leftover from the age of
Download free PDF
View PDF
chevron_right
Language-independent bases of distinctive features
Elizabeth (Beth) V Hume
Phonetic and Phonological Perspectives, 2011
In spite of the fundamental role that phonological features play in current linguistics, current research continues to raise many basic questions concerning the relation these representational categories of phonology have to measurable physical properties. This paper addresses the question of how [spread glottis] segments are phonetically implemented and proposes a language-independent definition of this feature. Language-independent feature definitions specify both articulatory and acoustic definitions. We show that the definition of [spread glottis] in terms of a single common glottal configuration or gesture would be insufficient to account for the full range of speech sounds characterized by this feature. For a feature to be present at the phonetic level, its defining acoustic attribute, or failing that its enhancing attributes, must also be present in the signal. Studies like this one lay the groundwork for future investigations that will make the study of features more concrete, by associating them with specific articulatory states and gestures and with equally specific acoustic cues.
Download free PDF
View PDF
chevron_right
On the atoms of phonological representation
Bert Vaux
Primitives of Phonological Structure, edited by Bert Botma, Andrew Nevins, and Marc van Oostendorp. Oxford: Oxford University Press.
Is the phonetic form of human language underlyingly organised in terms of words, distinctive features, elements, or articulatory gestures? After setting out the basic properties of the two leading responses to this question, couched within Distinctive Feature Theory and Articulatory Phonology, we argue that although the two are essentially interchangeable in many ways, they differ significantly with respect to a number of important empirical predictions, some resulting from the nature of features and gestures themselves and some from the theoretical frameworks within which they are generally embedded. In both regards Distinctive Feature Theory is to be preferred, insofar as (at least when situated within a conventional generative model such as Rule-Based Phonology: it correctly predicts the distribution of labial vs labiodental consonants; the existence of relativised locality, inversion, enhancement, and suprasegmental processes; autonomy of stricture and manner vs place of articulation; and phonological manipulations of individual segments such as insertion and non-assimilatory mutation.
Download free PDF
View PDF
chevron_right
A brief history of articulatory-acoustic vowel representation
Frederic Berthommier
HAL (Le Centre pour la Communication Scientifique Directe), 2015
This paper aims at following the concept of vowel space across history. It shows that even with very poor experimental means, researchers from the 17 th century started to organize the vowel systems along perceptual dimensions, either articulatory, by means of proprioceptive introspection, or auditory. With the development of experimental devices, and the increasing knowledge in acoustic and articulatory theories in the 19 th century, it is shown how the relationship between the two dimensions tended to tighten. At the mid 20 th century, the link between articulatory parameters such as jaw opening, position of the constriction of the tongue, or lip rounding, and the acoustical values of formants was clear. At this period, with the increasing amount of phonological descriptions of the languages of the world, and the power of the computer database analysis allowing extracting universal tendencies, the question of how the vowel systems are organized arose. The paper discusses this important question, focusing on two points: (1) how the auditory constraints shape the positioning of a specific set of vowel within the acoustic space, and (2) how the articulatory constraints shape the maximal extension of the vowel systems, the so-called maximal vowel space (MVS). In the past centuries, the vowel space representation evolved along two main dimensions: articulatory and auditory. One of the first articulatory representations of vowels was proposed by Robinson in 1617. By capturing the position of his tongue positions during vowel production, he proposed to categorize the vowels along the anterior-posterior position of the tongue (see for details). More than a hundred years later, Hellwag proposed in turn, one of the first triangle representations of the German vowel space with its degrees and its order relations ([2: §57], see fig. ). Now, the auditory perception seemed to be of importance in this description. As quoted by Nearey: « Hellwag is partly interested in anatomical descriptions of speech production. However, his introduction of the triangle seems to be motivated primarily by concern for some type of auditory, rather than articulatory relationship» ([3: 41]). Indeed, Hellwag was much influenced by Reyher's 1679 works on "tone heights", defined from whispered vowels and that relied on the timbre of the vowels independently of any fundamental frequency (see ). At the same period, the German physicist Chladni ([4]) proposed another representation of the 10 vowels, that he claimed to be the whole set of vowels, sorted into 3 series according to articulatory considerations 1 ([4: § 52], see fig. ). From the vowel /a/, he derived a 1 st branch with "open outside and slowly narrowing inside" (/a ɔ o u/), a 2 nd branch with an
Download free PDF
View PDF
chevron_right
The elements of phonological representation
Geoff Lindsey
Frontiers of phonology: atoms, structures, …, 1995
Download free PDF
View PDF
chevron_right
Model-Based Exploration of Linking Between Vowel Articulatory Space and Acoustic Space
Anqi Xu
Interspeech 2021, 2021
While the acoustic vowel space has been extensively studied in previous research, little is known about the high-dimensional articulatory space of vowels. The articulatory imaging techniques are limited to tracking only a few key articulators, leaving the rest of the articulators unmonitored. In the present study, we attempted to develop a detailed articulatory space obtained by training a 3D articulatory synthesizer to learn eleven British English vowels. An analysis-by-synthesis strategy was used to acoustically optimize vocal tract parameters that represent twenty articulatory dimensions. The results show that tongue height and retraction, larynx location and lip roundness are the most perceptually distinctive articulatory dimensions. Yet, even for these dimensions, there is a fair amount of articulatory overlap between vowels, unlike the fine-grained acoustic space. This method opens up the possibility of using modelling to investigate the link between speech production and perception.
Download free PDF
View PDF
chevron_right
Integration of featural information in speech perception
Dominic Massaro
Psychological Review, 1978
A model for the identification of speech sounds is proposed that assumes that (a) the acoustic cues are perceived independently, (b) feature evaluation provides information about the degree to which each quality is present in the speech sound, (c) each speech sound is denned by a propositional prototype in longterm memory that determines how the featural information is integrated, and (d) the speech sound is identified on the basis of the relative degree to which it matches the various alternative prototypes. The model was supported by the results of an experiment in which subjects identified stop-consonant-vowel syllables that were factorially generated by independently varying acoustic cues for voicing and for place of articulation. This experiment also replicated previous findings of changes in the identification boundary of one acoustic dimension as a function of the level of another dimension. These results have previously been interpreted as evidence for the interaction of the perceptions of the acoustic features themselves. In contrast, the present model provides a good description of the data, including these boundary changes, while still maintaining complete noninteraction at the feature evaluation stage of processing.
Download free PDF
View PDF
chevron_right
Phonological features and feature classes: The case of movements in sign language
Wendy Sandler
Lingua, 1996
The signs of sign language consist phonetically of hand configurations, locations on the body or in space, and movements. Some models claim that dynamic movements and static locations are the sequential segments of sign language, and even that movements are analogous to vowels. Others claim that movements are redundant, or in any ease should not be represented as fully-fledged sequential segments, The present study measures movements agains~ stringent phonological and morphological criteria for featurehood and classhood, in light of the current controversy over their status. Data from American Sign Language and from Israeli Sign Language support the claims made here, among them, that there is a set of phonologically contrastive features of movement which is phonetically coherent, and that these features constitute a class that is referred to in a blocking constraint on Multiple inflection and other processes. It is shown that the distinction between sequences of dynamic movements mid static elements in signs is exp]oited in templatic morphology in both sign languages. While this analysis supports the claim that movements are phonologically significant at the underlying level, it suggests that their linear position need not be lexically specified. i. The controversy From the earliest days of sign language linguistics, it has been accepted that there arc threc categories of phonological features: hand configuration, location, and ~' I am very grateful ~o Harry van der Hulsl and to anonymous reviewers for their helpful comments on this paper. Thanks as well to paflic:.pants for their questions and comments at the following conferences where earlier versLon:+ of this paper were presented: the Workshop on Sign Language Phonology and
Download free PDF
View PDF
chevron_right
Unified Parametrization of Phonetic Features and Numerical Calculation of Phonetic Distances between Speech Sounds (preprint version)
Maksym Vakulenko
Journal of Quantitative Linguistics, 2022
A metric method to numerically measure phonetic and phonemic distances, or contrasts, between speech sounds is put forward. The feature values of the compared phones taken from the standard IPA charts, are treated as independent parameters that give rise to corresponding Euclidean distances. As an illustration, the general phone set is mapped to Ukrainian phonemes. The proposed model agrees well with the historical linguistic facts, experimental phonetic data, and with general intuition. The described approach may find its due applications in various fields of linguistics and speech technologies, including historical and typological linguistics, language acquisition, phonetic studies, computational phonology, machine translation, information retrieval, text-to-speech.
Download free PDF
View PDF
chevron_right
Related topics
Cognitive Science
Mathematics
Phonology
Phonetics
Cited by
A feature geometric approach to Bondu-so vowel harmony
Abbie Hantgan-Sonko
Glossa: a journal of general linguistics
Christopher Green, Abbie Hantgan. A feature geometric approach to Bondu-so vowel harmony.
Download free PDF
View PDF
chevron_right
Sur les bases phonétiques des primitives phonologiques
Ivaylo Burov
Histoire Epistémologie Langage
Cet article s’interroge sur les bases phonétiques des théories traitant de la structure interne des segments. Il porte un regard sur la naissance et le développement de l’idée que le phonème n’est pas le constituant ultime de la chaîne phonique indissociable en unités plus petites, mais le produit de l’association d’une liste finie de paramètres universels appelés différemment par les divers modèles : traits, éléments, gestes. La rupture avec la conception « atomique » du phonème et l’élaboration des théories des primitives phonologiques vont cependant de pair avec une complication du formalisme et une hermétisation de la phonologie par son isolement de la phonétique.
Download free PDF
View PDF
chevron_right
The Jewish Neo-Aramaic Dialect of Dohok
Dorota Molin
2020
5 The Neo-Aramaic dialects are modern vernacular forms of Aramaic, which has a documented history in the Middle East of over 3,000 years. Due to upheavals in the Middle East over the last one hundred years, thousands of speakers of Neo-Aramaic dialects have been forced to migrate from their homes or have perished in massacres. As a result, the dialects are now highly endangered. The dialects exhibit a remarkable diversity of structures. Moreover, the considerable depth of attestation of Aramaic from earlier periods provides evidence for the pathways of change. For these reasons the research of Neo-Aramaic is of importance for more general fields of linguistics, in particular language typology and historical linguistics. The papers in this volume represent the full range of research that is currently being carried out on Neo-Aramaic dialects. They advance the field in numerous ways. In order to allow linguists who are not specialists in Neo-Aramaic to benefit from the papers, the exa...
Download free PDF
View PDF
chevron_right
Clinical Illustrations of Voice Quality
Allison Benner
2019
The objective of this work was to compare the extraction of phenolic compounds from Tilia argentea flowers and bracts by using conventional (solvent extraction) and novel (ultrasound assisted) extraction methods. Ethanol (70 %) extracts were analyzed for their antioxidant activities. Total phenolic content was determined using Folin-Ciocalteu method and the antioxidant potential was determined by DPPH radical scavenging and Ferric Reducing Antioxidant Power (FRAP) assays. To determine the effect of ultrasound treatment on the extraction, same extraction parameters were applied in both methods. The results showed that extracts obtained by ultrasound assisted extraction have higher total phenolic content and antioxidant activity.
Download free PDF
View PDF
chevron_right
Voice and Voice Quality
Allison Benner
Voice Quality, 2019
Worrying about the consequences of climate change. Climate behaviour and the meaning of caring for future generations The aim of the study was to see how care-worry about the consequences of climate change for future generations are experienced and handled by informants in a Swedish context. The term care-worry refer to an assembly of several approaches, such as empathy, care, responsibility, worry, and justice towards other people and coming generations. Care-worry is related to and initiates people's attitudes and behaviours towards future risks. Empirically, the study is based on thematic interviews conducted in the year 2016 and deals with informants who explicitly expressed worries about climate change for future generations. All 24 informants turned out to be parents. From the material, four ideal types were depicted, which in different ways can illustrate the parents' care-worry. Different climate behaviours also seem to be linked to these ideal types. The study as a whole shows that parents' care-worries affect them in various ways to reduce their own climate footprint in the future. Many of them choose a more sustainable lifestyle for the sake of future generations.
Download free PDF
View PDF
chevron_right
A featural paradox in Votic harmony
Ida Toivonen
Natural Language & Linguistic Theory, 2016
In this paper we present a novel argument against strict locality in vowel harmony: a vowel’s feature may have a double identity, active in one process and neutral in another. Such is the behavior of [back] in Votic [i]. It is invisible to harmony, while simultaneously triggering an assimilation process. We argue that no feature-sharing account of this phenomenon is plausible, including the relatively powerful extension of Span Theory that permits vowels in a harmonic span to remain unassociated (and unharmonized) with the span’s head. We offer instead an account based on the Agreement-By-Correspondence approach to long-distance assimilation.
Download free PDF
View PDF
chevron_right
A Corpus-Based Swadesh Word List for Literary Christian Urmi (New Alphabet Texts)
Alexey Lyavdansky
Semitic Languages and Cultures, 2020
Download free PDF
View PDF
chevron_right
Glide strengthening in Atayal: sonority dispersion and similarity avoidance
Hui-chuan J. Huang
Journal of East Asian Linguistics, 2020
Diachronic descriptions of lenition and fortition are more common than synchronic ones in the phonological literature despite that they are synchronically active in a number of languages (De Carvalho et al. 2008). This paper presents the data of Atayal glide strengthening, showing how Sonority Dispersion Principle (SDP, Clement 1990) and the constraint on homorganic glide-vowel sequences play a role in leading to the observed patterns in its
Download free PDF
View PDF
chevron_right
Arabic Loanwords in the Neo-Aramaic Dialect of Ankawa
Salam Hakeem
Semitic Languages and Cultures, 2020
Download free PDF
View PDF
chevron_right
Bengali nasal vowels: lexical representation and listener perception
Allison Wetterlin
Phonetica
This paper focuses on the question of the representation of nasality as well as speakers’ awareness and perceptual use of phonetic nasalisation by examining surface nasalisation in two types of vowels in Bengali: underlying nasal vowels (CṼC) and nasalised vowels before a nasal consonant (CVN). A series of three cross-modal forced-choice experiments was used to investigate the hypothesis that only unpredictable nasalisation is stored and that this sparse representation governs how listeners interpret vowel nasality. Visual full-word targets were preceded by auditory primes consisting of CV segments of CVC words with nasal vowels ([tʃɑ̃] for [tʃɑ̃d] ‘moon’), oral vowels ([tʃɑ] for [tʃɑl] ‘unboiled rice’) or nasalised oral vowels ([tʃɑ̃(n)] for [tʃɑ̃n] ‘bath’) and reaction times and errors were measured. Some targets fully matched the prime while some matched surface or underlying representation only. Faster reaction times and fewer errors were observed after CṼC primes compared to bo...
Download free PDF
View PDF
chevron_right
Explore
Papers
Topics
Features
Mentions
Analytics
PDF Packages
Advanced Search
Search Alerts
Journals
Academia.edu Journals
My submissions
Reviewer Hub
Why publish with us
Testimonials
Company
About
Careers
Press
Content Policy
580 California St., Suite 400
San Francisco, CA, 94104