Metrical Phonology - Annurev - An.24.100195

Annual Reviews
www.annualreviews.org/aronline
Annu. Rev. AnthropoL 1995. 24:313-42

Copyright ©1995 by AnnualReviewsInc. All rights reserved
METRICAL PHONOLOGY
Annu. Rev. Anthropol. 1995.24:313-342. Downloaded from arjournals.annualreviews.org
by Instituto de Eclogia, A.C, Consorcio CONACyT on 02/19/07. For personal use only.
Michael Hammond
Department
of Linguistics, Universityof Arizona,Tucson,Arizona85721
KEY
WORDS:
stress, accent,foot, meter,optimality
Metrical theory is a branchof phonologythat posits a hierarchical structure to

represent stress patterns in the minds of speakers. This review examinesthe
basic argumentsfor this theory and surveys the central issues of the field over
the past 18 years. These issues include questions about whether the foot
typology is symmetric,whetherthere is a strict binarity requirement, and how
to treat ternary iteration. The review concludeswith a brief overviewof the
impact of constraint-based phonologyon metrical theory.
INTRODUCTION
Metrical phonologyis the branch of linguistic theory concerned with stress

phenomenain natural language. It is distinguished from previous approaches
in that it posits a hierarchical structure reminiscentof the structures used in
traditional discussions of poetic meter,1 hencethe namemetrical theory. It is
also muchbroader in coveragein that it relates stress to several other domains.
Example1 illustrates nominalstress in Lenakel(24, 25, 32, 68). Mainstress
falls on the penult. Secondary
stresses fall on everyother syllable to the left. In
orthodox linear generative phonology(8) such a pattern wouldbe described
terms of an n-nary stress feature. For example,wordstress is formalizedas in
Example2 and might be governed by the rules in Example3.
1 There
arealsoanalyses
of poeticmeterin termsof this theoryof stress(seee.g. 33,41,45,57).
0084-6570/95/1015-0313505.00 313
Annual Reviews
314 HAMMOND
n-~im N-fish ’fish’

k61ey ’sweet potato’
kav~vaw ’hat’
l-~gabn~-bon in-morning ’in the morning’
kay~law~law ’kind of. dance’
hl~-du~blug~ilk in-lungs ’in the lungs’
kay~law~law
02010
V --, [1stress] / _ CO VCO# 3b.

The numbersindicate degrees of stress: "1" indicates primary stress and "0"
indicates no stress. The rules apply to place primary stress. The first rule
elevates all even-numbered syllables to primaries. The secondrule elevates the
final stress to a primary and--via the Stress Subordination Convention~de-
motes all other stresses by one. The Stress Subordination Conventionrequires
that all stresses in a domain be demoted by one when a primary stress is
assigned.
In metrical theory, such a pattern is described in terms of binary trochaic
stress "feet." These feet are aligned with Lenakel forms in a right-to-left
fashion and position stress on even-numbered syllables counting from the right
edge of the word. Primarystress is placed with a superordinate structure. Both
kinds of structures indicate stress in terms of "headedness."Theseconstitu-
ency and headedness relationships can be diagrammedas follows:
X at
x x
x (x x) (x x)
k a y ~ la w ~ la w
Howcan these theories be distinguished and whyshould an anthropologist

care? Briefly, the theories are distinguished in terms of the elaboration of the
structure that indicates stress. Followingstandard generative thinking, both of
these theories makeclaims about the wayindividuals represent stress patterns
in the mind. The linear view would have it that each stressed syllable is
independent from its neighbors and numerologically related to them. The
metrical view holds that the syllables are groupedtogether in structures like
Annual Reviews
METRICAL PHONOLOGY315
the feet above. If these latter structures have any phonetic or phonological
reality, then that wouldconstitute evidencefor the metrical theory.
Whyshould a nonlinguist--an anthropologist specifically--care about this
theory? There are several reasons. First, in general, linguistic theories make
claims aboutthe limits of languagelearning, i.e. whatis or is not learnable. To
the extent that anthropologists are interested in whatis cultural, as opposedto
what is psychological, they should understand what is unlearnable, hence what
is not subject to cultural control or variation. Second,this theory merits spe-
cific attention becauseit ties together domainsof investigation that previously

have been treated separately: stress or accent, poetic meter, reduplication,
minimal word phenomena, and prosodic morphology. The theory makes
claims about the covariation that can be exhibited in these different domains,
and anyoneinterested in linguistic differences should understandthe range of
possible variation. Third, metrical theory is the domainof phonologyin which
optimality theory has developed the most. This newdirection in linguistic
theory has already achieved someextremely important results and has wide
applicability both within and beyondtraditional linguistic domains. A clear
understanding of metrical phenomenaprovides an essential prerequisite to
understandingoptimality theory.
OVERVIEWOF LINGUISTICS AND PHONOLOGICAL

THEORY
Let us briefly review someof the central assumptions relevant to metrical
theory: generative linguistics, phonologicaltheory, and phonetic theory.
WhatIs GenerativeLinguistics ?
The basic assumptions of generative linguistics are perhaps familiar to most
readers, but manyof the central ideas of phonological theory becomefar less
murkyif these assumptions are laid out clearly. Manyof these ideas are
examinedin someof Chomsky’smore classic works (e.g. 7, 8).
First, the object of description is not languageper se, but the unconscious
knowledgeresponsible for an individual’s intuitions of grammaticality (i.e.
competence).The force of this assumptionis that aspects of language maynot
be reflective of grammarand maybe excluded from phonological description.
Standard examplesinclude speech errors, spoonerisms, and false starts, but
this domainmight also include phonetic implementation, processing limita-
tions, etc.
Second,the focus or grail for generative linguists is Universal Grammar,or
the innate endowmentthat, through acquisition or language development,
makes the adult grammarpossible. It has been most forcefully argued by
Annual Reviews
~16 HAMMOND
Chomsky that adult grammarsare startlingly uniformgiven the great variabil-

ity in linguistic experience. Onthe generative view, this uniformity follows
from the commongenetic endowmentthat enables people to learn language
quickly and to comeup with effectively the same grammarsdespite varying
experience.
Third, grammaticaldescriptions should be explicit. This concept tradition-
ally has been cast in terms of a computationaldescription of generativity and a
notional focus on rules. However,other perspectives are consistent with a
general focus on explicitness (see below).

The standard evidence that we do need a grammarto account for linguistic
behavior is linguistic creativity. Just as speakers of a language can produce
sentences they have never heard before; they can judge the well-formednessof
phonological patterns exhibited by novel or nonceforms.
What Is Phonology?
Generative phonology focuses on sound systems. The sound system of a
languagetypically exhibits Certain patterns. For example,the distribution of
stresses in Lenakelis restricted, as shownabove. The claim behind generative
phonologyis that although languages do not exhibit the samerestrictions on
their soundsystems, the restrictions that do occur are bounded.For example,
not all languagesexhibit stress on every other syllable from the right. Lan-
guagesexhibit iteration fromdifferent directions and at different intervals, e.g.
left-to-right, right-to-left, edge-in, every other syllable, every third syllable
(see below). However,somepatterns are completely unattested, e.g. middle-
out iteration or stress on every fourth syllable.
Following the generative orthodoxy, such limits are ascribed to Universal
Grammar (8). The basic idea is that the child comesto the language task with
certain predispositions about iterating stresses. Theseexpectations reduce the
learnable patterns to the observed(and predicted) cases.
In early generative theory, these expectations were encodedin a rule for-
malism. The idea was that the child was endowed with a template for
phonological generalizations and used this template to comeup with rules. As
noted above, this was done by supposing a rule schemaand a set of terms that
could appear in that schema(8). The schemais given in Example5 below, and
the terms werethe distinctive features.
A--~ B/C D 5.
In morerecent generative phonology,the focus has shifted awayfrom rules
and toward representations. For example, in metrical theory, rather than re-
strict the rules that assign stresses, the set of possible feet is restricted. For
example, one might propose that feet can only be binary and ternary, thus
excluding as a possible language any system that exhibited stress on every
Annual Reviews
METRICALPHONOLOGY 317
fourth syllable. Recently, there has been a shift back away from repre-
sentations. 2 Interestingly, the mostrecent shift awayfrom representations has
not been accompaniedby a shift toward rules (see below).
Phonetics
Metrical phonologyis largely a theory of prominence, but the phonetics of
prominenceare far from clear. Moreover, metrical phonologyhas been ex-
tended to account for a number of phenomenain which prominence per se
does not appear to be at issue.

Consider first the question of prominence.Metrical theory can naively be
understood as a theory of where the stresses can go. The basic idea-~in
generative terms--is that the child comes to the task of determining what
generalizations governthe stress systemof his or her languagewith expecta-
tions about whatcan be a stress system. Theseexpectations limit the range of
possible stress systems.
It is not clear whata child mightlisten for in trying to figure out whatis or
is not stressed. Theprincipal cues for stress are assumedto be loudness, pitch,
and length. Stressed syllables are typically louder, higher pitched, and longer
(62). However,as Lehiste (62) argues, several systems defy this simple char-
acterization. The most well-knowncase is English. Hayes(38) argues that the
secondsyllables of wordslike veto and mottocontrast in stress. This difference
in stress is manifestedonly in the presenceor absenceof flapping or aspiration
on the medialconsonant,e.g. vitl~6 vs m~iro.
Thetask for the languagelearner is not to find stresses and then fit them
to the limits imposedby metrical theory. Rather, the learner comesto the
acquisition task with metrical theory and looks for virtually anything upon
whichto imposemetrical structure. Therewouldseemto be a clear bias for the
usual cues of stress to be governedby metrical structure, but as the example
from English shows, other properties can be governedby metrical structure as
well.
WHAT IS METRICAL PHONOLOGY?
Although there are a number of competing versions of metrical theory, I

provide here a fairly neutral version. As outlined above, the theory provides
for a set of constituents that can be built over a string of syllables. Theset of
constituents and the algorithms available for assigning those constituents to
strings comprisethe guts of the theory.
2
This fascinating ebb and flow between rules and representations over the history of phonology
is described in Anderson (2).
Annual Reviews
318 HAMMOND
The central notion is the metrical foot. Most versions of metrical theory
adopt at least two kinds of feet: iambsand trochees. Aniambic foot dominates
two syllables and places the stress on the secondsyllable. Atrochaic foot also
dominatestwo syllables and places the stress on the first one. Lenakel is an
exampleof a trochaic language.
Aklan (Example6) involves iambic feet (37). Stress falls on alternating
syllables from the right edgeof the word. If a heavysyllable (one ending in
consonant)is encounteredin the scansion, it receives stress and the count is
restarted. Whichstress is primary is complexand not relevant here (see 37 for

details).
n~g-k-in-~-lisfid ’worry-actor-past’ 6.
k-in-~-put6s ’wrap-instr focus-past-post’
aslrt~r ’lucky’
Mmb~g ’speak’
Hayes argues that iambic systems almost exclusively depend on syllable
weight. Trochaic systems are not limited in the same way. Systems like
3Lenakel do not depend on syllable weight, but other trochaic systems do.
English is a familiar yet complex example (see 8, 22, 23, 31, 37,
50). Basically, the final syllable is skippedand trochaic feet are built right to
left.
hilt tfible 7.
~nimal Am6rica
informfitional munlcipfility
hhmam~lidfinthemum
That syllable weight plays a role in this systemis shownby the "heavy penult
rule": The rightmost stress of an English wordcannot occur to the left of the
penult if the penult is heavy(i.e. contains a long vowelor is closed by a coda
consonant).
light penult long penult closed penult

Am6rica ar6ma ver~inda
cinema balalaika ag6nda
asp~iragus hi~itus cons6nsus
metr6polis horizon syn6psis
j~ivelin thromb6sis am~ilgam
v6nison cor6na ut6nsil
3
Onecan argue that Lenakeldoes have a covert syllable weight system(see 26, 32).
Annual Reviews
Feet can be assigned to a string in at least twodirections: left-to-right or

right-to-left. Lenakelexhibits right-to-left footing. Maranungku exhibits left-
to-right trochees (92).
tiralk ’saliva’ m6rep~t ’beard’

yfingarmhta ’the Pleiades’ lfingkar~tefi ’prawn’
w61ep~nem~nta ’kind of duck’
Iambic systems can also exhibit different directionalities. Aklanaboveis
right-to-left; Munseeis left-to-right (16). Somelanguages exhibit several

different directions (see 25, 64). Such "bidirectional" systems typically

build a foot at one edge of the domainand then iterate towardthat edge from
the other.
Assigningfeet to wordsor phrases will capture a systemthat distinguishes
between stressed and stressless syllables. However,some systems exhibit
several degrees of stress. For example,Lenakelhas two degrees of stress: main
and secondary. To capture such systems, higher-order structures are built over
feet. FollowingPrince (80), this can be described in terms of the "EndRule."
This rule selects out a peripheral foot for main stress; all other feet have
secondary stress. In Lenakel, the rightmost foot is promotedby the "EndRule
Right."
Wecan represent the structures discussed so far using the typographically
convenient constituentized grid, an exampleof which was given in Example4
above and is repeated below as Example10.4
x line 2 10.
x x line 1
x (x x) (x x) line
k a y ~ 1 a w ~ l a w
Syllabic constituency is indicated on line 0. Syllables with any degree of
stress are markedon line 1, and the result of the EndRuleis markedon line 2.
Additional evidence for unstructured nonbinary stress is that the EndRule
also appears to apply to unfooted spans directly. In somelanguagesthe first,
last, first heavy, or last heavy syllable is promotedto prominencewith no
evidencefor binarity. Czech(with initial stress) or French(with final stress)
provide familiar and simple examples of such systems. Khalkha Mongolian
(Example11) provides an exampleof a systemin which quantity matters (90).
In Khalkha,the leftmost long vowelis stressed. If there is no long vowel, the
first syllable is stressed.
4
Accordingto Hammond (23, 27, 33), the constituentized grid is really a notational variant of the
(albeit less typographicallyconvenient)arboreal grid.
Annual Reviews
320 HAMMOND
bosg6ul ’fugitive’ 11.

bari~iad ’after holding’
fili ’which’
xoyordugfiar ’second’
x6tolb~r~ ’leadership’
garfiasaa ’from one’s ownhand’
Theprecise analysis of such systemsis controversial, but clearly there is no
evidence for metrical constituency in such systems, binary or otherwise (see

22, 26, 37 for different approaches).

Finally, there is evidence for an intermediate level of prominenceand
constituency betweenfeet and main stress: the superfoot or colon. Hungarian
provides the clearest example(28). In Hungarian,mainstress falls on the first
syllable of the word. Subsidiarystresses fall on alternating syllables counting
rightward. Thesesubsidiary stresses are ranked in an alternating pattern: For
example, the third, seventh, and eleventh syllables have less stress than the
fifth, ninth, and thirteenth. Representative data are given in Example12.
Primary stress is markedwith an acute accent, secondary stress is markedwith
a circumflex, and tertiary is markedwith a grave.
vi:z ’water’ 12.

kSp~ ’hoe’
kSpa:vSl ’with hoe’
t~ri:tb:vel ’with tablecloth’
f6:lem~let~n ’on mezzanine’
f6:lem~let~id ’your mezzanines’
k~ku:nf~:le~h~:za:bSn ’in Kiskfnf61egyh~iza’
m~gvestegeth~tetl~nek ’unbribable (ones)’
m6gvest~geth~tetl~nekn~k ’to those unbribable’
~lka:pbsta:~i:tottSlonli:tott ’ decabbagized (!)’
l~gmegv~steg~thet~tlen~bbekn~k ’to those least bribable’
~lka:pbsta:~:tottSlon~:tott~:tok ’you’ve decabbagizedit’
This pattern can be analyzedas follows: Trocheesare built fromleft to right
on syllables. Trocheesare then built again on the roots of the original trochees.
End Rule Left promotes the first foot of the second round of trochees. A
sample derivation is given in Example13.
X
13.
X X
x x x (x x) (x)
x ,, x x ,~ (x,,) (x x) (x) (x x) (x (x)
f¢: lem~ le t~n-, f¢:le m~le t ~n--, f~:le m~le t~n
Annual Reviews
Passamaquoddy has also been cited as an exampleof cola or superfeet (89).

Finally, the mechanism of extrametricality allows a syllable at the edge of
the footed span to be skipped. In the earliest work on metrical theory, this
skipping option was thought to be limited to edges. English (see above) is
exampleof this pattern; WesternArandais another particularly striking exam-
ple (11, 91). Stress in Arandafalls on odd-numbered syllables counting from
the left, with two exceptions. Word-finalsyllables never receive stress and a
vowel-initial word receives stress on even-numberedsyllables. Davis (11)
cites the following data (pp. 399-400):

tt~kura ’ulcer’ 14.

kfitungflla ’ceremonialassistant’
w6ratf~ra (a place name)
ergdma ’to seize’
artj~nama ’to run’
utn~daw?~ra (a place name)
k~ima ’to cut’
flba ’ear’
wfima ’to hear’
Thesedata receive the following analysis here:

(a) Makethe final syllable extrametrical; 15.
(b) makea word-initial vowelextrametrical;
(c) build trocheesfromleft to right.
This analysis assumesthat the entire span cannot be madeextrametrical. This
principle is most well knownas exhaustivity (36, 37, 39) and is needed
account for the difference betweenforms like ilba and ergftma. The former
bears initial stress becauseinitial extrametricality is blockedbecauseit would
violate the exhaustivity principle. Thelatter bears secondsyllable stress be-
cause initial extrametricality succeeds because exhaustivity is not invoked.
Samplederivations are given in Example16. Extrametricality is markedwith
angled brackets.
x 16a.
x x x <x> (x) <x>
i Ib a --, i Iba ~n /a --, i Ib a
x 16b.
x x x x x <x> <x> x <x> <x> (x) <x>
e rgum a ~ e rgum a --, e rgurna --* e rgu m a
Annual Reviews
322 HAMMOND
In summary,metrical theory provides for a set of constituent types (feet,

and cola or superfeet) and a set of algorithms for assigning those constituents
to strings. That set of constituents is supplementedwith the End Rule and
extrametricality. The test for this theory is whetherthe constituency claimed
has any empirical consequences.If so, this wouldconstitute a dramatic argu-
mentin favor of metrical theory over any linear alternative (e.g. 8).
EVOLUTIONOF THE FIELD

This section traces the history of metrical theory since 1968, starting with the
earliest challenges to linear phonologyand concludingwith an introduction to
optimality theory.
Linear to Nonlinear
Chomsky&Halle (8; see above) conceived of stress as no different in kind
from other features, e.g. [nasal], [coronal], [strident]. This theory had the
desirable property of uniformity; the substantive differences betweenspeech
soundswere formalized in terms of the samesorts of formal objects.
This theorized uniformity, however,leads to someproblems(see 66, 85).
For example, stress is not like other phonetic variables. Properties such as
nasality are absolute in that a loweredvelumis equated with [+nasal] regard-
less of context. Stress, on the other hand, is relative in that the stress of a
syllable is determinedin comparisonwith other syllables. This relativity was
expressed in two waysin Chomsky &Halle’s theory. First, there are different
degrees of stress and these needed to be encodednumerically, e.g. [lstress],
[2stress], etc. This is in stark contrast to all other features, whichweretreated
as binary in the phonology. 5 Second,this also led to the Stress Subordination
Convention(see Example3), which allowed local changes to a single segment
to affect all other segmentsin the string undergoingthe relevant rule. Noother
features required a similar convention.
There were also empirical problems with Chomsky& Halle’s analysis.
Liberman & Prince (66) show that the interaction of Chomsky& Halle’s
CompoundStress Rule (CSR) and Nuclear Stress Rule (NSR) produce incor-
rect forms. Basically, the NSRstresses the last wordin a phrase, whereasthe
CSRstresses the penultimate word in a compound.
compounds: b6ttle br~sh 12 17.

phrases: J6hn rfns 21
5
Chomsky &Halle did maintain that all binary phonologicalfeatures were to be translated into
integer values in the phonetics, but accordingto their theory, the only integer values to whichthe
phonologyneededto refer were those associated with the feature [stress].
Annual Reviews
In morecomplexphrases, both rules apply cyclically and invoke the Stress

Subordination Convention. Cyclic application meansthat the rule applies to
innermostconstituents and then reapplies at each successively larger domain,
invoking the Subordination Convention each time. Thus a compoundlike
bottle brushhandleundergoesa three-step derivation as follows. First, stress is
assigned to each word independently. Then compound stress is assigned to the
embeddedcompound bottle brush, assigning primary stress to the penultimate
primary, invoking the Stress SubordinationConventionand reducing the stress
on brush from primary to secondary. Finally, compound stress is reapplied to

the whole compound,assigning, pri.mary stress to the penultimate primary
bottle skipping brush--and demoting brush and handle again.
bottle brush handle 18.

1 1 1 step 1
1 2 step 2
1 3 2 step 3
Notice howcyclic application produces a rising profile amongthe demoted
stresses to the right of the main stress. Notice too howcyclic application
demotesbrush on the inner cycle before the compound stress rule has a chance
to apply to the whole compound,preventing the ungrammatical*bbttle brfish
h~ndle.
Phrasal stress is assigned in a similar fashion. Example19 contains a
three-part cyclic derivation. First, each wordis assigned primarystress. Next,
phrasal stress is assigned to the embeddedphrase sees Mary, promoting the
final primary and demotingsees. Finally, phrasal stress is reassigned to the
entire phrase, demotingJohn and sees.
John sees Mary 19.

1 1 1
2 1
2 3 1
Liberman & Prince (66) point out that although the Chomsky& Halle
algorithm works fine with compoundsembeddedin phrases, it does not work
whenapplied to phrases embeddedin compounds. Example20 shows how the
algorithm succeeds with a compoundembeddedin a phrase.
John sees bottle brushes 20.

1 1 1 1
1 2 CSR
2 1 3 NSR
2 3 1 4 NSR
Annual Reviews
324 HAMMOND
Consider how the rules would apply to a compoundmotor unit neural

control, containing the phrase neural control. First, each word is assigned
primary stress. Then the CSRapplies to motor unit and the NSRapplies to
neural control. Finally, the CSRapplies to the whole thing. Notice that be-
cause the CSRseeks out the penultimate primary stress, it incorrectly pro-
motesmotor, rather than control. (The attested pronunciation has the strongest
stress on control.)
21.
mo to r unit neural con t ro 1

1 1 1 1
1 2 2 1
*1 3 3 2
The problem is that the Chomsky& Halle rules are sensitive only to the
numerical values for the feature [stress]. Liberman&Prince (66) propose
an algorithm based on the overt branching structure. In their algorithm,
all constituency is indicated .with binary branching. All branches are la-
beled either strong or weak,.wherestrength is the formalization of stress. In a
phrase, the right branch is alwaysstrong. In a compound,the right branch is
strong if and only if it branches. The examplesabove are thus realized as
follows:
22.
bottle brush handle
23.
W W S
Annual Reviews
Motor unit neural control 24.
S W W S
Liberman & Prince’s algorithm accounts for the problem with phrases
embeddedin compounds,but it provides for several other desiderata as well.
First, it providesan expressionof the insight that stress is relative rather than
absolute. Second,it extends naturally to account for word-internal stress pat-
terns (66). Third, it has no need for the cycle.6 Fourth, it accountsreadily for
other kinds of stress systems(21, 37, 70). This last result is extremelyimpor-
tant. Chomsky &Halle’s analysis (8) was of English. Hyman (47) did the first
real typological generative workon stress. He argued that a numberof gener-
alizations restrict the stress systems of the world. These patterns are not,
however, easily captured within Chomsky& Halle’s paradigm. This kind of
typological work was a central impetus for the developmentof a parametric
metrical theory.
Several of the argumentsabovehave since been shownto be too strong. For
example, the elimination of the cycle was not possible. This was first shown
by Kiparsky (58; see also 20, 31). Also, conditions on branching have been
shownto be unnecessary(44, 80).
Grids and Trees

Liberman&Prince’s theory (66) has three components:a binary stress feature
[+_stress], the metrical tree, and the metrical grid. Whetherall three repre-
sentations are necessaryand, if not, whatkind of hybrid representations should
be adoptedin their stead was a hotly debatedtopic.
Wehave already seen empirical and theoretical motivation for the metrical
tree in the treatment of compoundstress in English. The need for a binary
stress feature can be seen readily by comparingEnglish words such as h4lix
and ndrthOx. Under the metrical theory outlined above, these words would
have the samebinary metrical tree labeled’s w.’ However,the latter differs
from the former by having a final secondary stress. Someadditional structure
was needed to encode this difference and Liberman& Prince (66) posited
binary stress feature. The two representations are given belowin Example25.
6
Rischel(84) madea similarargument
withrespectto Danish
compounding.
Annual Reviews
326 HAMMOND
25.
h@lix
+ - +
SW S W
V
There are other ways to capture this distinction as well. Hayes(37) pro-
poses to do awaywith the binary stress feature by adopting levels in the
metrical tree, i.e. the foot. In Hayes’snotation, stress mustbe strong in a foot.
This is exemplified in Example26 below, where the horizontal line separates
the foot level.
26.
Liberman&Prince also use a secondrepresentation: the metrical grid. They

motivate this in their discussion of the rhythmrule in English.7 The rhythm
rule is responsible for the leftward shift of stress in a modifier whenit is
followed too closely by a stress in the next word (Example27) (see also
42, 65 for discussions of rhythmin English).
thlrt~en thlrt6en m6n---, thirt~en m6n 27.

Minnes6ta Minnes6ta M~e--, M~nnesbtaMl’ke
T~nness6e T~nness6e~iir ---, T6nness6e
~iir
Mbnt~ina Mbnthna c6wbby ~ *M6nt?ana c6wbby
It is rather difficult to accountfor the possible shifts if only a tree repre-
sentation is used. Liberman& Prince propose to augmentthe metrical tree
with a metrical grid. The grid is constructed by aligning columnsof marks or
ticks with syllables, such that degree of stress is encodedin terms of the height
of the columns. Example28 showsthe tree representations for the phrases in
Example27, and Example29 showsthe grid representations.
7
Similar rhythmic processes apply in other languages, e.g. Italian (77), Tiberian Hebrew (79),
Tunica
(23).
Annual Reviews
METRICAL PHONOLOGY 327
Liberman &Princearguethat the rhythmrule applieswhenthere is a clash

in the metricalgrid. Aclash is definedas adjacentmarks
at twocontiguous
levels of the grid. (Clashesare marked
withhyphensbelow.)
~h.irte~ n~n Minneso~ Mike 28.

W S S S WSW S
~s~ ~r ~~ c~
S W S S W SW S W
X 29.
X X .... X
X .... X X X X
X X x x x X X x
t h i r t e e n m e n M i n n e s o t a Mi k e
X
X -- -- X X
X X -- -- X X x
X X X X X x x X X
T e n n e s s e e a i r Mort t a n a c owb o y s
Theobviousquestionis whytwoseparaterepresentations shouldbe neces-
sary. This question is explored in both directions. Prince (80) explores the
possibility that there is only the grid (see also 87). Heproposesthat there is
constituentfoot andthat stress in languages
like Lenakel
wouldbeassignedby
aligningwordswiththe perfectgrid.
x x x x 30.
¯ X X X X X X X X
Theperfectgrid is composedof peaksandtroughs.Thehighercolumns are

peaks;the lowercolumns
are troughs.In Lenakel,the perfectgrid is aligned
with strings fromthe right to left troughfirst.
Annual Reviews
328 HAMMOND
x 31.
X X
X X X X X
k ay ~ 1 aw~ law
Hammond (23) proposes a different solution, in which grids are constituen-
tized: the arboreal grid. Theresulting representation indicates clashes and foot
structure.
kay~law~law 32.
If constituency cannot be manipulated independently of metrical headship,

then this formalismis notational with a grid with constituency addedon: the
constituentized grid. This latter formalism is proposed by Halle &Vergnaud
(22) and is adopted here for typographical convenience.8
x 33.
X X
x (x x ) (x x
k a y ~ 1 a w ~ 1 aw
There is extensive evidence for metrical constituency (see below). This

evidenceconstitutes support for a constituentized representation (arboreal grid
or constituentized grid) and support for metrical theory in general.
Morphology
A rather dramatic developmenttook place after the debate over representations
above. McCarthy& Prince (72) argued that a broad class of morphological
operations are sensitive to metrical structures (see e.g. 63, 73 for refinements
of this theory).
Earlier work took a different perspective. Moravcsik(76) had provided
broad typological overview of reduplication, and Marantz (69) extended
McCarthy’s(70) CV-skeletontheory to account for these facts. The CV-skele-
ton theory maintained that phonological representations were organized
See Hammond(27, 30) for a discussion of the potential independence of constituency and
headship (see also 10, 17).
Annual Reviews
around timing units specified only for whether they were consonants (Cs)
vowels(Vs). Considerreduplication in Tagalog(4).
ipon i-ipon ’just saved’ 34.

bloaut bo-bloaut ’just gavea special treat’
trabaho ta-trabaho ’just finished working’
galit ga-galit ’just got mad’
Marantzwould characterize this pattern of reduplication as involving a

skeletal prefix of the form CV, which obtains its segmental content from a
copy of the base segments. Although this theory offers a concise formal
description of a widevariety of reduplicationtypes, it fails to limit themin a
satisfactory way. The theory predicts that any combinationof basic Cs and Vs
shouldconstitute a possible reduplicativeaffix, but this is not the case.
Rather, McCarthy and Prince argue, the set of possible reduplicative affixes
is defined by the set of prosodic categories. Thesecategories include various
sorts of feet and syllables. Moreover,they argue, these prosodic categories
define not just the set of reduplicative affixes, but the set of templatesand loci
in which morphologicaloperations can occur.9 YidinYprovides an exampleof
foot-based reduplication (12, 13, 38).
gindalba gindalgindalba ’k.o. lizard’ 35.

mulari mulamulari ’initiated man’
kalamparaa kalakalamparaa ’marchfly’
The existence of morphologicaloperations defined in terms of disyllabic

units obviously provides support for the theory of stress that groups syllables
together.
Other sorts of morphologicaloperations also convergeon this result. Con-
sider infixation in Chamorro(9): Mainstress falls on alternating syllables
counting fromthe right edge of the word.
mag~igu ’clothes’ m?~gagfi-fia ’his clothes’ 36.

kadtiku ’crazy’ m~n-kadtlku ’crazy’(pl)
Ignoring irrelevant complications, stress can be assigned by building tro-

chaic feet from right to left. The continuative is formedby reduplicating the
first syllable of the rightmostfoot.
9
Someof this work builds on an earlier paper by Broselow & McCarthy (5).
Annual Reviews
330 HAMMOND
[s~ga] sa[s~iga] ’stay’ 37.

[6gga] e[96gga] ’watch’
hu[g~ndo] huga[g~ndo] ’play’
Finally, there are normalaffixation operations that simplyfail if the base of

affixation does not meet someprosodic type. English comparativeformation is
limited in this way. The suffix -er can only be added to monosyllables or
trochees, e.g, bigger, htippier, but *aldrter, *c6rnicaler.
Treatmentof the two domainscontinues today (for other evidence of metri-

cal constituency,see e.g. 1, 18, 22, 29, 40, 71, 81).
Symmetry
Halle & Vergnaud (21), McCarthy(70), and Hayes (37) have proposed
metric and parametric metrical theories (see also 22, 49). Thesetheories were
parametricbecausethe set of occurringmetrical feet wasa function of setting a
finite set of parameters; they were symmetric because the parameters could
combine freely, producing the combinatorial maximum for the interacting
parameters.
Consider, for example, headednessand quantity sensitivity. Assumingonly
binary feet, headedness allows for either left-headed or right-headed feet.
Quantity sensitivity dictates whether nonheadsof feet can dominate heavy
syllables. (Quantity sensitivity allows heavysyllables to attract stress.) These
two binary parameters can cross to produce the combinatory maximum of four
foot types.
Headedness Quantity-sensitivit v foot 38.

left quantity-sensitive v
left quantity-insensitive
right quantity-sensitive
right quantity-insensitive
Hayes(43) argues that this perfect symmetrydoes not hold. The argument
is madeon the basis of a dramatic typological investigation and can be summa-
1°
rized in the followingchart.
10 Theseresults are anticipated in the overviewgiven at the beginningof this review(see also 44,
46, 72).
Annual Reviews
METRICALPHONOLOGY 331
Headedness Quantity-sensitivity Attested foot 39.

left quantity-sensitive
left quantity-insensitive
right quantity-sensitive
right quantity-insensitive
Accordingto the abovechart, there is no evidence for a right-headed quantity-
insensitive foot. All right-headedfeet appearto be quantity-sensitive. Further-
more, the left-headed quantity-sensitive foot is affirmed but not the one pre-
dicted by the symmetrictheory. The symmetrictheory maintained that a left-

headed quantity-sensitive foot was simply restricted in whether the nonhead
could dominatea heavy syllable. Hayesmaintains instead that the left-headed
quantity sensitive foot could dominateeither one heavysyllable or two light
11
ones. If such a span is unavailable,no foot is built.
Evidence for this new left-headed foot--the moraic trochee-~comesfrom
Palestinian Arabic (53, 54). Hayescites the followingexamples:
mak~iatib offices 40.

mool~dna our feast
,&irabato she hit him
b~kara cow
gajar~ituhu his tree
baarfikato she blessed him
b~iarako he blessed him
The analysis proceeds as follows: Moraictrochees are built from left to
right. The final foot is madeextrametrical, and End Rule Right applies. Some
samplederivations are given in Example41 below. Notice that if the rightmost
foot is not word-final, extrametricality is vacuous.
X 41.
X X
X X X X X (X X) X (X X)
m ak fi a t ib m ak fi a t ib m ak ~i a t ib
X
X X X X
x x x x x (x x) (x x) (x x) (x x)
moo 1 ~idn a moo 1 ad n a moo l~idna
11Hayes(44)claimsthata headless
footis built, butsubsequent
researchers
haveassumed
nofoot
is constructed
in suchcircumstances.
Annual Reviews
332 HAMMOND
X
X X X <X>
X X X X (X X) (X X) (X X) (x x)
,d ~i r ab a to ,d a r a ba to ,dfira ba to
X X
X X X (X X) (X X
b~ikara b~ka ra b~kara
X
X X X X
X X X X X (X X) (X X) (X X) (X X)
~a ja r~ tuhu ~a ja r~ tuhu ~a ja r~i tuhu
X
X X X X
X X X X X (X X) (X X) (X X) (X X)
baa r~ikayo baa r~ika to baa r~ika to
X
X X X < X >
X X X X (X X) (X X) (X X) (X X)
bfia rako b~a rako bfia rako
Left-headedquantity-sensitive
feet fail to account
for this pattern..Follow-
ing are the sameexamplesfootedwithtraditionalparametric left-headedquan-
tity-sensitive feet. Asterisksmarkformsthat wouldreceiveincorrectstress
underthis analysis.
X 42.
X X
X X X X X (X X X) X (X X X)
mak~a t ib mak~a t ib mak~a t ib
X
X X X < X >
X X X X X (X X) (X X) (X X) (X X)
moo 1 gdn a moo 1 ~idn a *moo 1 ~dn a
X
X X X < X>
X X X X (X X) (X X) (X X) (X
,d ~ r ab a t o ,d fir a b a t o ,d fir a b a t o
Annual Reviews
X
X X
x x x (x x) (x x)
bgkara b~ika ra b~ka ra
X
X X X X
x x x x x (x x) (x x) (x x) (x x)
~a ja rfi tuhu ga ja r~i tuhu ga ja r~ tuhu

X
X X X <X>
XXXXX (XX) (X X) (X X X) (X X)
b aa r~ika to b a a r~i ka to *b aa r~i ka to
X
X X X
XX X X (X X) X) (X X) X)
b~a rako bfia rako b~ia rako
The theory that emergesfrom this approach is no longer parametric in the

samesense. Underthe symmetrictheory, parameters combineto define a set of
basic feet. UnderHayes’s theory, the basic feet are primitive and are only
parametricin that they comefroma finite list.
Binarity and Catalexis
One of the issues that comes out of Hayes’s asymmetric typology is the
absence of a monosyllabicdefault foot. As exemplified in the above analysis
of Palestinian, if the spanis insufficient, nofoot is built. In the case of a moraic
trochee, if there is less than either a heavysyllable or two light syllables, no
foot is built. Previously, such feet were countenancedby the theory and were
termed degenerate feet. Whethersuch feet should be allowed has been hotly
debated.
Somelanguages appear to allow for superficial degenerate feet. In Ma-
ranungku(see Example9), syllabic trochees are built fromleft to right. Notice
though that a final degenerate foot is built in words with an odd numberof
syllables. This is an apparent counterexampleto the asymmetricfoot typology.
Hayessuggests the final stress in such cases is only apparent, that it is only
somekind of word-final lengthening.
An alternative explanation is proposed by Kiparsky (59), whosuggests the
mechanismof catalexis as a way of accounting for the following generaliza-
tion: Languageswith trochaic feet that permit final stress typically do not have
Annual Reviews
334 HAMMOND
a minimal wordconstraint. This generalization can be seen by comparingOno

and Diyari. Onohas final stress and no word minimum(78).
ktlm ’palm bark’ 43.

d6ne ’my eye’
~ril~ ~iril~
Diyari, on the other hand, has no final stresses and a bisyllabic word
minimum(3).
k~ina ’man’ 44.

pinadu ’old man’
wlqapina ’old woman’
The basic idea behind catalexis is that somelanguages allow for an invis-
ible (or catalectic) syllable at the edgeof a word.Coupledwith a strict binarity
requirement on feet, the option of catalexis will derive the two possibilities
above. Ono allows for a catalectic syllable (marked with square brackets
below); Diyari does not.
X X X X 45.
(x x) (x x) (x x) (x x)
k tim [o] d 6 n e ~ r i 1~ [o]
X X X X 46.
X (X X) (X X) (X X) (X
*o k ~n a p in adu w i 1 a p in a
The catalectic syllable allows for subminimalwords (monosyllables) and
final stress. The prediction is that there will be no (trochaic) languagethat has
final stress and a bisyllabic minimalword, and no languagethat allows mono-
syllabic wordsbut not final stress. Kager(51) tested these predictions exten-
sively.
There are other uses for catalexis as well. For example, Kiparsky argues
that the otherwise mysteriousstress systemof Tiabatulabal (see 93) receives
natural treatmentif there is final catalexis. Tiibatulabal has rigid final stress
accompaniedby stresses on every other mora to the left. The final stress
requirement suggests the feet are iambs, but the "every other mora" require-
mentsuggests the feet are moraictrochees.
t~iah~wil~ ’the summer’(obj.) 47.

haniilfi ’the house’ (obj.)
6ey6eti ’he became ashamed’
wit~iyhat~l ’the Tejonindians’
witayh~tal~iabacfi ’away from the Tejon indians’
Annual Reviews
Kiparskysolves this conundrum by proposingthat there is a final catalectic

syllable with moraictrochees built fromright to left.
x x x x x x x x 48.
(xx) (x x) (x x (xx) (x x) (xx) (xx) (x
taa haw i l~i [s] han ii la [o] 6e y6e
X X X X X X
x (x x)(x x) (x x) (x x) (xx) x (x
w i t ~tyh a tfi 1 [o] w f tayhfi ta l~tab a ctl [o]
Thoughmost linguists are naturally suspicious of such invisible elements, the

mechanism
of catalexis brings together a numberof generalizations.
Ternarity
A lot of recent research has focused on accountingfor apparent ternary itera-
tion. Considerstress in Cayuvava.Cayuvavawas first brought to the attention
of the metrical world by Levin (63) (see also 34, 46, 83, 88). The data below
are from Key(55, 56). Stress in Cayuvavafalls on every third vowelcounting
fromthe right end of the word(63, pp. 101-102),
(a) d~iru ’hand’ 49.

6ne ’leaf’
(b) s~kahe ’stomach’
r~era ’leg’
(c) kihl’Bere ’I ran’
tak~asi ’old man’
(d) arik~jahi ’he has already fallen’
Bari6kimi ’seed of squash’
(e) p6pohec6Baka ’inside of cow’
B~idaca6ai ’my younger brother’
60 aB~reric~ikaA ’palate (Px)’
mar~haha6iki ’their blankets’
(g) ikit~iparer6peha ’the wateris clean’
tiBiBioafine ’she spanks meagain’
(h) c~adir6boBurfiruce(caad~iirob6irohline) ’ninety-nine’
(Bururuce) medfirucec6irohiine ’fifteen each’
The most straightforward account would be to admit a new foot type, the
dactyl, which would have three terminals with the head on the left. At least
four other approacheshave also been taken.
Annual Reviews
336 HAMMOND
Halle &Vergnaud(22) propose augmentingthe set of possible feet to allow

for anapests: ternary feet with the head in the middle. Coupledwith final
extrametricality, such feet can account for the patterns above. Hammond (34)
argues that these patterns can be captured with binary feet if the theory of
extrametricality is expandedto admit, as a markedcase, extrametricality in the
middleof words. Hayes(46) argues that there are at least two waysa span can
be parsed into feet. Understrong local parsing, binary feet are built next to
each other; under weaklocal parsing, binary feet are built one light syllable
awayfrom each other.

The binary approaches have two advantages over the approaches admitting
primitive ternary feet: First, binary approaches maintain a restrictive foot
typology. Second, they account for the emergenceof binarity even in ternary
systems. (For example, the minimumword in Cayuvavais two syllables, not
three.) Onthe other hand, both binary approachesexpandthe theory, in other
respects. The relativized extrametricality proposal must admit nonperipheral
extrametricality. The weaklocal parsing approach must allow for a newfoot-
12
ing algorithm. Howbest to account for ternarity is an openquestion.
Optimality Theory
A recent developmentin metrical theory (and in other areas of phonologyas
well) is optimality theory (OT)(e.g. 74, 82). This theory represents a
step forward in manyrespects, although in someways, it is a return to ideas
that were developeda numberof years ago.13 In somerespects, it is a rejection
of someof the central ideas that have dominatednonlinear phonologyfor the
past twentyyears.
The basic idea behind OTis that ordered phonologicalrules are rejected in
favor of ranked and violable constraints. For example, the stress patterns of
14
Lenakel above could be described by imposingthe following constraints.
PARSE:
Syllables must be footed. 50.
FTBIN:Feet must be binary. 51.
ALIGN:
Feet must be aligned with the left/right edge. 52.
TROCHEE:
Feet are trochaic. 53.
12Kager (52)andFitzgerald(14)bothtreat ternarityin anoptimality theoryframework.

13Constraints havebeenaroundin phonology for a longtime(seee.g. 61). SeeHammond(23)
an earlyconstraint-basedapproachto rhythm andSelkirk(86) for a constraint-based
approach
lexlcalstress.
14Allof theseconstraints are attestedin the OTliterature. I givethemherein a somewhat
more
informal formfor expository
ease.
Annual Reviews
The constraints FTBINand TROCHEE are unviolated and, in the terminology of

OT, undominated in the constraint hierarchy. The constraints ALIGNand
PARSE are both violable in order to minimizeviolations of the superordinate
FTBIN.Furthermore, ALIGN is violable to satisfy PARSE.These ranking rela-
tionships can be captured with the following hierarchy.
54.
~ FTBIN >> PARSE>> ALIGN
I. TROCHEE
Theseconstraints and ranking relationships can be exemplified in the fol-

lowingconstraint tableau. Rankingof constraints is indicated by left-to-right
ordering (and solid as opposedto dotted lines). UnderOT,a set of candidate
forms is generated by the function GEN.The constraint hierarchy then applies,
and violations of the various constraints are assessed. The guiding principle
here is "strict ranking":A violation of a higher constraint is damningif there is
an alternative candidate that does not induce the sameviolation. (Such damn-
ing violations are markedwith an exclamationpoint and rightward shading.)
/kayelawelaw/ FTBIN TROCHEE PARSE ALIGN 55.

kayelawelaw
(k~iye) (l~iwe)
(k~y)la(w~law)
~ ka(y~la) (w~law)
ka(yel~i) (wel~w)
(k~) (y~la) (w~law)
kayela(w~law)
Notice that none of the candidates escapes violating at least one constraint.
Multiple violations are reckoned in the relevant cases and indicated by the
numberof asterisks.
This approach has several advantages over a rule-based approach. One of
McCarthy& Prince’s strongest arguments comes from infixation in Tagalog
(15). In Tagalog,the progressiveinfix -urn- is positioned before the first vowel
of a word.
sulat ’write’ sumulat 56.

aral ’teach’ umaral
gradwet ’graduate’ grumadwet
This observation is an embarrassmentto the theory of prosodic morphology
developed by McCarthy& Prince (72) because, as noted above, the locus
infixation was thought to coincide with the edges of prosodic constituents, e.g.
feet and syllables. (A vowelis not a member of this set.)
Annual Reviews
338 HAMMOND
McCarthy&Prince (74) argue that the facts of Tagalog can be accounted

for by proposingthat the position of the infix is governedby prosodic factors
but not simply crude alignment of the affix with the edge of someprosodic
category. Specifically, they proposethat the affix goes as far to the left as it
can, subject to the constraint that coda consonantsare avoided.
Thusa form like aral undergoesinfixation as umaral, rather than as *aum-
ral because the latter contains a coda. The form *arumalis also not the output
because there is an alternative form still not containing an additional coda in
which the -um- occurs further to the left: umaral. Contrast this with grumad-
wet, in whichpositioning the infix further to the left alwaysresults in an extra
coda: *gumradwetor *umgradwet.
This relationship can be modeledeasily in OTterms, in whicha constraint
on position is subordinateto a constraint against codas. Theserelationships are
diagrammedin the following tableaux.
{um+aral} NoCODA LEFTMOST 57.

~ umaral *
aumral **! *
arumal * **!
~um+gradwet} NOCODA LEFTMOST 58.

umgradwet
gumradwet
grumadwet
graumdwet
The difference betweenstress in Tonganand English provides a nice metri-

cal example.Both languagesexhibit right-to-left footing with moraictrochees.
However,they differ in their treatment of a long vowelin penult position. In
English, such a vowel receives stress, presumably exhibiting the following
structure (if strict binarity is assumed).
x 59.
x (xx)
ar6:ma
In Tongan,such spans are reanalyzed so that the penult is realized as a
string of two light syllables. Compare
the following forms.
x x 60.
(xx) x(x x)
h fi : ’go in’ hu. fi f i ’open officially’
Annual Reviews
This splitting up of the penult allows the word-final foot to occur at the
edgeof the wordand still only dominatetwolight syllables.
Undera derivational view, the Tonganfacts wouldrequire two passes of
syllabification. The first pass wouldproducesyllables to whichthe stress rules
wouldapply. The secondpass wouldreadjust that syllable structure to accom-
modatethe metrical requirements. The problemis that there is no evidence for
twoseparate syllabification stages.
This difference betweenthe two languagesis readily accountedfor in terms
of constraint ranking. Prince & Smolenskyposit two constraints: RIGHTMOST
and ONSET.RIGHTMOST wants the foot to occur as far to the right as possible.
ONSET wants VVsequences to syllabify as long vowels. The difference be-
tween English and Tongan comes from differential ranking of these con-
straints. This is exemplifiedin the followingtableaux.
/huufi/ RIGHTMOST ONSET 61.

[hfiu]fi *!
~ hu[fifi] *
/arooma/ ONSET RIGHTMOST 62.

~ a[r6o]ma *
aro[6ma]
Similar argumentsare presented in Prince (81) and Mester(75).
Most of the work in generative phonologythat has been presented at the
major conferencesover the past two years has been in terms of OT(but see 19,
49 for argumentsagainst a purely constraint-based phonology).15
SUMMARY
I have reviewedthe essential structure of the metrical theory of phonology,

focusing on the early argumentsfor it and reviewingsomeof the central issues
that haveanimatedthe field since 1968.
The central insight has been that rules are traded for constituency. Rather
than a derivational approachto stress, in whichsomen-nary stress feature is
assigned to domains,constituents are aligned with strings. The proper align-
mentof these constituents is governedby a restricted set of algorithms, which
can be conceivedof in terms of a ranked set of violable constraints. Mostof
the debates within the field have centered on the nature of these constituents
15
Startlingly, nothing has appearedto date in any of the major linguistics joumals. However,most
of the circulating papers are available via anonymousftp from the Rutgers Cognitive Science
Center (ruccs.rutgers.edu).
Annual Reviews
340 HAMMOND
and the disparate kinds of evidence that can be adduced in support of different
versions of metrical structure.
The past two years have seen a dramatic shift away from constituency to a
focus on how the constituents get placed. The leading idea being explored
today is that constituents are not placed by specific rules. Rather, a candidate
set of constituentized representations is generated and then winnowedthrough
by the language-specific constraint hierarchy.
It is impossible to predict the future, but I suspect that the constraint-based
view will continue to be explored and that there will be less reliance on
constituents as constraint-based theory is elaborated. However, it seems rather
unlikely that the constraint-based view will completely supplant the rule-based
view, and I hope a happy mediumwill develop.
ACKNOWLEDGMENTS
Thanks to Diana Archangeli, Jane Hill, and Diane Ohala for useful discussion.
AnyAnnualReviewchapter,as well as anyarticle cited in an AnnualReviewchapter,

maybe purchasedfromthe AnnualReviewsPreprintsand Reprintsservice.
1-800-347-8007;415-259-5017;emaiharpr@class.org
L itera ture Cited
1. Al-mozainy HQ, Bley-Vroman R, Australianlanguages.BerkeleyLinguist.

McCarthy J. 1985.Stress shift andmetrical Soc. 11:398~407
structure. Linguist.Inq. 16:135-44 12.Dixon RMW. 1977. Somephonological
2. AndersonSR. 1985. Phonologyin the rules of Yidiny. Linguist.Inq. 8:1-34
TwentiethCentury. Chicago:Univ. Chi- 13. DixonRMW. 1977. y.
A Grammar ofl’idin
cagoPress Cambridge:Cambridge Univ. Press
3. Austin P. 1981. A Grammar of Diyari. 14.FitzgeraldC. 1994.Ternarystress in opti-
Cambridge:CambridgeUniv. Press malitytheory.Univ.Ariz., Dep.Linguist.,
4. BowenJD. 1965. Beginning Tagalog. Tucson,Unpublishedms.
Berkeley:Univ.Calif. Press 15. FrenchKM.1988. Insights into Tagalog
5. BroselowE, McCarthy J. 1983.Atheoryof reduplication,infixation andstress from
internal reduplication.Linguist. Rev.3: non-linear phonology.Masters thesis.
25-88 Univ.Texas,Arlington
6. Chai N. 1971. A grammarofAklan. PhD 16. Goddard I. 1982.Thehistorical phonology
thesis. Univ.Penn.,Philadelphia of Munsee. Int. J. Am.Linguist. 48:16~8
7. Chomsky N. 1965.Aspectsof the Theoryof 17. HagbergL. 1993.Anautosegmentaltheory
Syntax. Cambridge,MA:MITPress of stress. PhDthesis. Univ.Ariz., Tucson
8. ChomskyN, Halle M. 1968. The Sound 18. HalleM.1987.Gridsandtrees in metrical
Pattern of English. NewYork:Harper& phonology.In Phonologica1984, ed. WU
Row Dressier, HCLuschiitzky,OEPfeiffer, JR
9. Chung S. 1983.Transderivational relation- Rennison, pp. 79-94. Cambridge:Cam-
ships in Chamorrophonology.Language bridgeUniv.Press
59:35~56 19. Halle M.1994. NewEnglanddialects pa-
10.CrowhurstM. 1993. Minimalityand foot per. MIT,Dep.Linguist., Unpublished ms.
structure in metrical phonology and pro- 20. Halle M, KenstowiczM. 1991. The free
sodicmorphology. PhDthesis. Univ.Ariz., elementconditionandcyclic versusnoncy-
dist. by IndianaUniv.Linguist.Club clic stress. Linguist.lnq. 22:457-501
11. Davis S. 1985. Syllable weightin some 21. Halle M, VergnaudJR. 1978. Metrical
Annual Reviews
METRICAL PHONOLOGY 341
structures in Phonology. MIT, Dep. Lin- in stress rules. ~erkeleyLinguist. Soc. 11:
guist., Unpublishedms. 429-46
22. Halle M, VergnaudJR. 1987. An Essay on 44. HayesB. 1987. A revised parametric met-
Stress. Cambridge, MA:MITPress rical theory. Northeast. Linguist. Soc. 17:
23. Hammond M. 1984. Constraining metrical 274-89
theory: a modular theory of rhythm and 45. HayesB. 1989. The prosodic hierarchy in
destressing. PhDthesis. UCLA;Revised meter. See Ref. 60, pp. 201-60
version dist. by Indiana Uuiv: Linguist. 46. Hayes B. 1994. Metrical Stress Theory.
Club. 1988. NewYork: Garland Chicago: Univ. Chicago Press
24. HammondM. 1985. Obligatory branching 47. HymanLM.1977. Onthe nature of linguis-
revisited. Northeast. Linguist. Soc. tic stress. See Ref. 48, pp. 37-82
15:145---67 48. HymanLM,ed. 1977. Studies in Stress and
25. Hammond M. 1985. Mainstress and paral- Accent. South. Calif. Occas.Pap.Linguist.,
lel metrical planes. BerkeleyLinguist. Soc. Vol. 4. LosAngeles: Dep. Linguist., Univ.
11:417-28 South. Calif.
26. HammondM. 1986. The obligatory- 49. ldsardi WJ. 1992. The computationofpros-
branching parameter in metrical theory. ody. PhDthesis. MIT
Nat. Lang. Linguist. Theory 4:185-228 50. Kager R. 1989.AMetrical Theory of Stress
27. Hammond M. 1987. Accent, constituency, and Destressing in Dutch and English.
and lollipops. ChicagoLinguist. Soc. 23 Dordrecht: Foris
(2):149-66 51. Kager R. 1993. Consequencesof catalexis.
28. HammondM. 1987. Hungarian cola. Pho- Proc. HILConf. Phonology, 1st
nology 4:267-70 52. Kager R. 1994. Deriving ternary rhythmin
29. Hammond M. 1987. Destressing rules in alignment theory. Utrecht Univ., Unpub-
metrical theory. MilwaukeeStud. Lang. 1: lished ms.
76-102 53. KenstowiczM. 1981. The metrical struc-
30. HammondM. 1989. Lexical stresses in ture of Arabic accent. Presented at UCLA/
Macedonian and Polish. Phonology 6: USCNonlinear Phonol. Conf.
19-38 54. KenstowiczM. 1983. Parametric variation
31. HammondM. 1989. Cyclic secondary and a~cent in the Arabic dialects. Chicago
stress and accent in English. West Coast Linguist. Sqc. 19:205-13
Conf. FormalLinguist. 8:139-53 55. Ke)~ H. 1961. Phonotactics of Cayuvava.
32. HammondM. 1991. Metrical Theory and lnt. J. Am. Linguist. 27:143-50
Learnability. Univ. Ariz., Dep. Linguist., 56. Key H. 1967. Morphology of Cayuvava.
Unpublished ms. Hague: Mouton
33. Hammond M. 1991. Poetic meter and the 57. Kip~rsky P. 1977. The rhythmic structure
arboreal grid. Language67:240-59 of English verse. Linguist. Inq. 8:189-247
34. HammondM. 1995. Deriving ternarity. 58. Kiparsky’P. 1979. Metrical structure as-
Coyote Pap. In press signment is cyclic. Linguist. lnq. 10:
35. Hammond M, NoonanM, eds. 1988. Theo- 421-41
retical Morphology. San Diego: Academic 59. Kiparsky P. 1991. Catalexis. Stanford
36. Harris J. 1983. Syllable Structure and Univ., Dep. Linguist., Unpublishedms.
Stress in Spanish. Cambridge, MA:MIT 60. Kiparsky P, YoumansG, eds. 1989. Rhythm
Press and Metkr. San Diego: Academic
37. HayesB. 1981. A metrical theory of stress 61. Kisseberth CK.1972. Onderivative prop-
rules. PhDthesis; Revisedversion dist. by crties of phonologicalrules. In Contribu-
Indiana Univ. Linguist. Club. 1984. New tions to Generative Phonology, ed. MK
York: Garland Brame, pp. 201-28. Austin: Univ. Texas
38. HayesB. 1982. Metrical structure as the Press
organizing principle of Yidiny phonol- 62. Lehiste I. 1970. Suprasegmentals. Cam-
ogy. In The Structure of Phonological bridge, MA:MITPress
Representations,Part I, ed. Hvan der Hulst, 63. Levin J. 1988. Generating ternary feet.
N Smith. Cinnaminson, Germany:Foris Texas Linguist. Forum29:97-114
39. HayesB. 1982. Extrametricality and Eng- 64. Levin J. 1988. Bidirectional foot construc-
lish stress. Linguist. lnq. 13:227-76 tion as a window on level ordering. See Ref.
40. HayesB. 1983. The role of metrical trees 35, pp. 339--52
in rhythmic adjustment. Northeast. Lin- 65. LibermanM. 1975. The intonational sys-
guist. Soc. 13:105-20 tem of English. PhDthesis. MIT
41. Hayes B. 1983. A grid-based theory of 66. LibermanM, Prince A. 1977. Onstress and
Enlish meter. Linguist. Inq. 14:357-93 linguistic rhythm.Linguist. Inq. 8:249-336
42. Hayes B. 1984. The phonology of rhythm 67. Lombardi L, McCarthyJ. 1991. Prosodic
in English. Linguist. Inq. 15:33-74 circumscription in Choctaw morphology.
43. HayesB. 1985. Iambic and trochaic rhythm Phonology 8:37-71
Annual Reviews
342 HAMMOND
68. Lynch J. 1974. Lenakel phonology. PhD 80. Prince A. 1983. Relating to the grid. Lin-
thesis. Univ. Hawaii. Univ. HawaiiWork. guist, lnq. 14:19-100
Pap. Linguist. 7 81. Prince A. 1991. Quantitative consequences
69. Marantz A. 1982. Re reduplication. Lin- of rhythmic organization. Chicago Lin-
guist, lnq. 13:435-82 guist. Soc. 26
70. McCarthyJ. 1979. Formalproblemsin Se- 82. Prince A, SmolenskyP. 1993. Optimality
mitic phonologyand morphology.PhDthe- theory. Rutgers Univ./Univ. Colo., Dep.
sis. MIT.dist. by Indiana Univ. Linguist. Linguist./Comp. Sci., Unpublishedms.
Club 83. Rice C. 1992. Binarity and ternarity in
71. McCarthyJ. 1982. Prosodic structure and metrical theory: parametric extensions.
expletive infixation. Language58:574-90 PhDthesis. Univ. Texas, Austin
72. McCarthyJ, Prince A. 1986. Prosodic Mor- 84. Rischel J. 1972. Compound stress in Dan-
phology. Univ. Mass./Brandeis Univ., Dep. ish without a cycle. Annu. Rep. Inst.
Linguist., Unpublishedms. Phonet. Univ. Copenhagen6:211-28

73. McCarthyJ, Prince A. 1990. Foot and word 85. Schane SA. 1979. Rhythm, accent, and
in prosodic theory: the Arabic brokenplu- stress in English. Linguist. Inq. 10:483-502
ral. Nat. Lang. Linguist. Theory8:209--82 86. Selkirk EO. 1980. The role of prosodic
74. McCarthyJ, Prince A. 1993. Prosodic Mor- categories in Englishwordstress. Linguist.
phologyI. Univ. Mass./RutgersUniv., Dep. lnq. 11:563-605
Linguist., Unpublishedms. 87. Selkirk EO. 1984. Phonology and Syntax.
75. Mester RA.1994. The quantitative trochee Cambridge, MA:MIT Press
in Latin. Nat. Lang. Linguist. Theory 12: 88. Spring C. 1989. Dactyls in Cayuvava.Univ.
1-61 Ariz., Dep. Linguist., Unpublishedms.
76. Moravcsik E. 1978. Reduplicative con- 89. Stowell T. 1979. Stress systems of the
structions. In Universals of HumanLan- world unite! MITWork. Pap. Linguist. 1:
guage, Vol. 3, Word Structure, ed. JH 51-76
Greenberg. Stanford, CA: Stanford Univ. 90. Street JC. 1963.KhalkhaStructure.Bloom-
Press ington: Ind. Univ.Publ.
77. NesporM, Vogel I. 1979. Clash avoidance 91. Strehlow TGH.1942. Arandic phonetics.
in Italian. Linguist. lnq. 10:467-82 Oceania 7:255-302
78. Phinnemore TR. 1985. Onophonology and 92. Tryon DT. 1970. An Introduction to Ma-
morphophonemics. Pap. NewGuinea Lin- ranungku.Pacific Linguist. Monogr.Ser. B,
guist. 22:173-214 No. 14. Canberra: Aust. Nat. Univ.
79. Prince A. 1975. The Phonology and mor- 93. Voegelin C. 1935. Tiibatulabal Grammar.
phology of tiberian hebrew. PhDthesis. Berkeley: Univ. Calif. Press
MIT, Cambridge, MA

Metrical Phonology - Annurev - An.24.100195

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Metrical Phonology - Annurev - An.24.100195

Uploaded by

Copyright:

Available Formats

Annual Reviews

Annu. Rev. AnthropoL 1995. 24:313-42

Metrical theory is a branchof phonologythat posits a hierarchical structure to

Metrical phonologyis the branch of linguistic theory concerned with stress

n-~im N-fish ’fish’

V --, [1stress] / _ CO VCO# 3b.

Howcan these theories be distinguished and whyshould an anthropologist

cific attention becauseit ties together domainsof investigation that previously

OVERVIEWOF LINGUISTICS AND PHONOLOGICAL

Chomsky that adult grammarsare startlingly uniformgiven the great variabil-

general focus on explicitness (see below).

does not appear to be at issue.

WHAT IS METRICAL PHONOLOGY?

Although there are a number of competing versions of metrical theory, I

restarted. Whichstress is primary is complexand not relevant here (see 37 for

light penult long penult closed penult

Feet can be assigned to a string in at least twodirections: left-to-right or

tiralk ’saliva’ m6rep~t ’beard’

right-to-left; Munseeis left-to-right (16). Somelanguages exhibit several

different directions (see 25, 64). Such "bidirectional" systems typically

bosg6ul ’fugitive’ 11.

evidence for metrical constituency in such systems, binary or otherwise (see

22, 26, 37 for different approaches).

vi:z ’water’ 12.

Passamaquoddy has also been cited as an exampleof cola or superfeet (89).

cites the following data (pp. 399-400):

tt~kura ’ulcer’ 14.

Thesedata receive the following analysis here:

In summary,metrical theory provides for a set of constituent types (feet,

EVOLUTIONOF THE FIELD

compounds: b6ttle br~sh 12 17.

In morecomplexphrases, both rules apply cyclically and invoke the Stress

on brush from primary to secondary. Finally, compound stress is reapplied to

bottle brush handle 18.

John sees Mary 19.

John sees bottle brushes 20.

Consider how the rules would apply to a compoundmotor unit neural

mo to r unit neural con t ro 1

Motor unit neural control 24.

Grids and Trees

Liberman&Prince also use a secondrepresentation: the metrical grid. They

thlrt~en thlrt6en m6n---, thirt~en m6n 27.

METRICAL PHONOLOGY 327

Liberman &Princearguethat the rhythmrule applieswhenthere is a clash

~h.irte~ n~n Minneso~ Mike 28.

Theperfectgrid is composedof peaksandtroughs.Thehighercolumns are

If constituency cannot be manipulated independently of metrical headship,

There is extensive evidence for metrical constituency (see below). This

ipon i-ipon ’just saved’ 34.

Marantzwould characterize this pattern of reduplication as involving a

gindalba gindalgindalba ’k.o. lizard’ 35.

The existence of morphologicaloperations defined in terms of disyllabic

mag~igu ’clothes’ m?~gagfi-fia ’his clothes’ 36.

Ignoring irrelevant complications, stress can be assigned by building tro-

[s~ga] sa[s~iga] ’stay’ 37.

Finally, there are normalaffixation operations that simplyfail if the base of

Treatmentof the two domainscontinues today (for other evidence of metri-

Headedness Quantity-sensitivit v foot 38.

Headedness Quantity-sensitivity Attested foot 39.

dicted by the symmetrictheory. The symmetrictheory maintained that a left-

mak~iatib offices 40.

~a ja rfi tuhu ga ja r~i tuhu ga ja r~ tuhu

The theory that emergesfrom this approach is no longer parametric in the