Quantity Insensitive Stress

A Factorial Typology of Quantity-Insensitive Stress
Author(s): Matthew Gordon

Source: Natural Language & Linguistic Theory, Vol. 20, No. 3 (Aug., 2002), pp. 491-552
Published by: Springer
Stable URL: http://www.jstor.org/stable/4048055
Accessed: 27-09-2015 12:05 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Natural Language & Linguistic Theory.
http://www.jstor.org
This content downloaded from 193.54.174.3 on Sun, 27 Sep 2015 12:05:54 UTC
All use subject to JSTOR Terms and Conditions
GORDON
MATITHEW
A FACTORIAL TYPOLOGY OF QUANTITY-INSENSITIVE STRESS *
ABSTRACT. This paper presents an Optimality-theoretic(Prince and Smolensky 1993)

analysisof quantity-insensitivestress. A set of grid-basedconstraintsis shownby means of
a computer-generatedfactorialtypology to providea relativelytight fit to the full range of
stress systems attestedin an extensive surveyof quantity-insensitivestress patterns,many
of which have not been previouslydiscussed in the theoreticalliterature.
1. INTRODUCTION
One of the majorendeavorsin theoreticalphonology over the past quarter

century has been to develop a metrical stress theory which offers adequate empirical coverage of attested stress systems with a minimal
amountof overgenerationof unattestedpatterns.As our databaseon stress
has enlarged during this time, the challenge of constructing a restrictive yet sufficiently rich metrical theory has increased. Several theories
have emerged in response to this growing challenge, both in rule-based
frameworks(e.g., Libermanand Prince 1977; Prince 1983; Hayes 1980,
1995; Halle and Vergnaud1987, etc.) and, more recently,in a constraintbased paradigm(Crowhurstand Hewitt 1994; Kager 1999; Kenstowicz
1997; Walker1996; Alber 1997; Eisner 1997; Bakovic 1996, 1998; Elenbaas 1999; Elenbaas and Kager 1999). Optimality Theory (Prince and
Smolensky 1993), with its hierarchicallyranked constraints goveming
phonological well-formedness,has providedfertile groundfor evaluating
metricalstresstheories.By permutingthe constraintrankings,it is possible
to constructa factorialtypology of stress whose empiricalcoveragecan be
comparedagainstcross-linguisticstressdata.
This paper belongs to the researchprogramemploying factorialtypologies as a means of assessing the predictivepower of metricalconstraints
(cf. Eisner 1997; Elenbaas and Kager 1999). The proposed account tests
the stress patternsgeneratedby a set of constraintsagainst those found
* The authorgratefullyacknowledgesthe insightfuldiscussion and technicalassistance
of Bruce Hayes, as well as comments from Michael Kenstowicz and two anonymous
reviewers, which have greatly improved the paper.Thanks also to Donca Steriade and
audiences at UCSB and the 2001 SWOTmeeting for theirfeedback on this research.
A
Natural Language & LinguisticTheory 20: 491-552, 2002.

? 2002 KluwerAcademicPublishers. Printedin the Netherlands.
C)
492
GORDON
MATTHEW
in what is intendedto be a fairly exhaustivesurveyof quantity-insensitive

stresssystems, i.e., those in which syllable weight does not influencestress
placement.This surveyrevealsseveralstresspatternsthat,to the best of my
knowledge, have not been previouslyreportedin the theoreticalliterature
and turnout to be of interestfor metricalstresstypology.
An area in which the present work differs from other works on stress
within OptimalityTheory is in terms of its representationsof stress. A
theory of stress is developed that does not appeal to the metrical foot;
rather,the proposed constraintsrefer directly to the well-formedness of
different stress configurations,following earlier work by Prince (1983)
and Selkirk (1984) within a derivationalframework,and more recently
by Walker(1996) in a constraint-basedparadigm.While many of the adopted constraintsare familiar from the literature,some are novel either
in their method of evaluationor their formulation,including an expanded
set of anti-lapseconstraintsand a constraintwhich requiresthat stress be
aligned with both edges of a word. Althoughthe goal of the presentwork
is not to consider the entire body of evidence for metrical constituency
(cf. Kenstowicz 1994 and Hayes 1995 for discussion of such evidence,
e.g., segmentalrules, minimal word requirements,prosodic morphology),
it is shown that constraintsreferringdirectly to stress without the intermediary of foot structureare sufficient to account for the full range of
quantity-insensitivestress systems, while suffering from a modicum of
overgenerationof a typically non-pathologicnature.
1.1. Representationsof Stress
Althoughthe proposedtheoryshareswith much currentwork its adoption
of a constraint-basedparadigm,it differs from most contemporarywork
in its grid-basedratherthan foot-based representationsof stress, following Prince (1983) and Selkirk (1984). Following this work, the present
analysis assumes that the level of stress associated with a syllable is a
function of the number of levels of grid marks above a given syllable.
Thus, an unstressedsyllable has a single grid mark above it (abstracting
away from weight-sensitivestress), a secondarystressed syllable has two,
and a syllable with primarystress has three levels of grid marksabove it.
Adoptingthese representations,a binarystress system with primarystress
on the initial syllable and secondarystresson odd-numberedsyllables after
the firstwould display the following associationsbetween grid marksand
syllables for words with five and six syllables (1).
STRESS
A FACTORIALTYPOLOGYOF QUANTITY-INSENSITIVE
x
Primarystress
Secondarystress x x x
xxxxx
Syllable
ccrCYacaxu
493
x
x x x
xxxxxx
aa
auocxxa
Similar representationsare assumed throughoutthe rest of the paper

with unstressed,primary,and secondarystressed syllables differing from
each otherin the numberof dominatinggrid marks.However,to conserve
space, full representationsof metrical grids will be often be omitted;instead, following convention,secondarystress will be markedwith a grave
accent and primarystress with an acute accent.
1.2. A Typologyof Quantity-insensitiveStress
As a startingpoint in the investigation,an extensive survey of quantityinsensitive stress was conducted using a combinationof grammarsand
existing typologies of quantity-insensitivestress,includingthose in works
by Hyman(1977), Hayes (1980, 1995), and Halle and Vergnaud(1987). A
total of 262 languages,all languages with predictablequantity-insensitive
stress patterns(i.e., predominantlynon-phonemicwith a clearly dominant
pattern)for which I could find data,were includedin the presentsurvey.'
In analyzingthe results of the survey,it is useful to sort stress patterns
into three basic groups:languages with fixed stress (either a single stress
or two stresses), languages with binary alternatingstress, and those with
ternaryalternatingstress.The presentationin this paperwill centeraround
these threecore stress types. For each class of stress systems, resultsof the
cross-linguisticsurveywill be presented,followed by the constraintsrelevant for yielding the attestedpatterns,and the factorialtypology of patterns
arising throughpermutationof the rankings of the proposed constraints.
The coarse taxonomyof quantity-insensitivestress systems is summarized
in TableI, along with the section in which each system is covered.
2. FIXED STRESS
The most straightforwardsystem of phonologically predictablestress is

one in which there is a single stress per word falling a fixed distancefrom
the word edge. Interestingly,as pointed out by Hyman (1977), of the set
1 Many of the stress systems in the survey are sensitive to morphologyto some degree,
a factorthatis not treatedin the presentpaper.
494
MATTHEWGORDON
TABLE I
Coarsetaxonomy of stress systems

Stress system
Section
Fixed
Binary
Ternary
2
3
4
of logically possible docking sites for stress in quantity-insensitivesingle

stresssystems, only five are attested:initialstress,peninitialstress (second
from the left), final stress, penultimatestress (second from the right), and
antepenultimatestress (thirdfrom the right).
Both Hyman's extensive survey of stress systems and the present
one indicate asymmetries in the relative frequency with which each of
these stress sites is attested.2Languages with initial stress are abundant
in both surveys and include, among many others, Arabela (Rich 1963),
Chitimacha(Swadesh 1946), and Nenets (Decsy 1966). Similarly,many
languages have final stress, e.g., Atayal (Egerod 1966), Moghol (Weiers
1972), Mazatec (Pike and Pike 1947). Also very common are languages
with penultimate stress, e.g., Mohawk (Chafe 1977), Albanian (Hetzer
1978), and Jaqaru(Hardman-de-Bautista1966). Considerablyrarerare
the peninitialpattern,e.g., Lakota(Boas and Deloria 1933, 1941), Koryak
(Zhukova1972), and the antepenultimatepattern,e.g., Macedonian(Lunt
1952), Wappo (Radin 1929). The numberof languages in Hyman's survey and the current one displaying initial, peninitial, antepenultimate,
penultimate,and final stress are summarizedin Table II. A list of languages (and their genetic affiliation)instantiatingthese patternsappears
in Appendix 1. References for the languagesare availableat the author's
website:http://www.linguistics.ucsb.edu/faculty/gordon/pubs.
Therelative
rarityof peninitialand antepenultimatestress will be discussed furtherin
the context of the factorialtypology in section 2.2.
2 Hyman's survey differs from the present one in its inclusion of languages with
quantity-sensitivestress systems. The figuresfromHyman'ssurveyin TableII also include
languageswith binary stress.
495
A FACTORLAL
TYPOLOGYOF QUANTITY-INSENSITIVESTRESS
TABLE II
Numberof single stress languages

Hyman (1977)
Numberof Igs.
114
Initial
Penultimate
77
Final
97
6
Antepenultimate
Peninitial
12
Total
306
Presentsurvey
Numberof Igs.
%3
37.3
25.2
31.7
2.0
3.9
57
53.55
59.5
7
10
187
%4
30.2
28.8
32.0
3.7
5.3
TABLEIII
Dual stress systems
Initial
Peninitial
Initial
Peninitia-
Antpenultimate_
Penultimate
_
Final
______
Primary
Antepenultimnate Penultimate
1
67
Final
36
___,.
3 Note thatpercentagesdo not add up to 100, as figuresare rounded.

4 Percentagereflectspercentageof single stress languagesin the survey.
5 The .5 languages with penultimateand final stress are attributedto Gurage(Polotsky
1951; Leslau 1992), which displays penultimatestress in nouns andfinal stress in verbs.
6 Georgian(Zhgenti 1964; Aronson 1991).
7 Anyula (Kirton 1967), Awtuw (Feldman 1986), ChimalapaZoque (Knudson 1975),
Sibutu Sama (Allison 1979; Elenbaasand Kager 1999), Sanuma(Borgman 1989), Murut
(Prentice 1971).
8 CanadianFrench (Gendron 1966), Udihe (Kormushin1998), and most varieties of
Armenian,unless the initial syllable containsa high vowel, which does not carrysecondary
stress (Vaux 1998).
9 Walmatjari(Hudson 1978). An additionalcomplicationin Walmatjariis thatthe secondary stress falls on the penult in words with four syllables (see section 2.1.5), where
antepenultimatestress would entail a stress clash, and optionallyon the penult ratherthan
the antepenultin words with greaterthan four syllables:thus, dada but drad'aaor 6onaaa.
10 Lower Sorbian(Janas1984), Watjarri(Douglas 1981), Gugu Yalanji(OatesandOates
1964).
496
MATTHEW
GORDON
Another type of fixed stress system involves two fixed stresses per
word.The presentsurveyincludes 14 such 'dualstress' languages,a small
numberin comparisonto the 167 single stress languages.Table III summarizes the numberof languages (languagenames given in footnotes) in
the survey displaying each subtype of stress system involving two fixed
stresses (the location of primarystress falls on the x-axis and secondary
stress on the y-axis; unattestedpatternsare indicatedby a dash; logically
impossiblepatternsare shaded).
Interestingly,of the 20 possible combinations (factoring in patterns
which differ only in which of the stresses is primaryand which is secondary),only five are exploited. The numerousgaps in the set of attested
fixed stresssystems with two stressesareexploredfurtherin the discussion
of the factorialtypology analysis in section 2.2. All of the attested dual
stress patterns,with the exception of two of the three initial plus final
patterns,CanadianFrench and Armenian,do not allow stress clashes; in
clash contexts, for example, in trisyllabicwords in languages with initial
and penultimatestress, the syllable carryingprimarystress in non-clash
contexts takes the only stress.
2.1. A ConstraintSetfor Fixed Stress and ResultingPatterns
The set of fixed stress patternsis accountedfor in straightforward
fashion
by a relativelystandardset of Optimality-theoreticconstraints,certainof
which are formulatedin slightly novel ways. These constraintswill be introducedas we considerthe variousstresssystems whichprovideevidence
for them.
2.1.1. TheALiGNConstraintFamily
Followingpreviousworkin OT,the attractionof stressby edges is modeled
by constraintsrequiringstress to be aligned with word edges. Two of the
ALIGN constraintswhich play an importantrole in the theory developed
here serve the same functionas constraintsalreadyfamiliarfromthe theoretical literature,e.g., McCarthyand Prince (1993), Crowhurstand Hewitt
(1994), among others. One of these constraintsrequiresthat stressed syllables be aligned with the left edge of a prosodic word, while the other
demandsthat stressed syllables be aligned with the right edge. Given the
grid-basedrepresentationsassumed here, the ALIGN constraintscan be
formalizedin terms of alignmentof grid marksto the metricalgrid. The
relevantconstraintstake the form in (2) and requirethat every grid mark
be aligned with eitherthe left or rightedge of the immediatelylower level
of grid marks.
STRESS
OFQUANTITY-INSENSITIVE
A FACTORIAL
TYPOLOGY
(2)
497
ALIGN (Xjevein+j, tR/L}, level n, PrWd):Every grid mark of

level n + 1 is aligned with the Iright, left I edge of level n of
grid marksin a prosodic word.

The generalstress constraintsnot specific to primarystress emerge if n is
0, in which case the constraintis violated for every grid markon level 1
which does not align with the relevantedge of level 0 of gridmarks.ALIGN
(Xlevel1, L, level 0, PrWd)requiresthatlevel 1 gridmarksalign with the left
edge of level 0, while ALIGN (Xlevel1, R, level 0, PrWd) requires that level 1
grid marksalign with the rightedge of level 0. Following earlierwork,it is
assumedhere thatALIGNconstraintviolations are gradient,such thatthey
are calculatedfor each stress and summedtogether.For example, a word
with stress on both the second and the third syllables incurs three total
violations of ALIGN (xieveii, L, level 0, PrWd), two for the stress on the
thirdsyllable and one for the stress on the second syllable (see (5) below
for an exampletableau).
AnotherALIGNconstraintwhich is not familiarfrom the literatureis
a broad constraint ALIGN (EDGES, level 0, PrWd, xlevel1), which requires
that the edges of level 0 of a prosodic word, i.e., both the initial and
the final syllable, be aligned with a level 1 grid mark. Violations are
calculatedin a simple fashion:one violation is incurredif eitherthe initial
or the final syllable does not carrya level 1 grid mark,and two violations
are incurredif both the initial and the final syllables do not have a level 1
grid mark. ALIGN (EDGES, level 0, PrWd, xlevel1) is thus in some sense an
apodic conglomeration of the constraints ALIGN-LEFTand ALIGN-RIGHT
(Kager 1999; Elenbaasand Kager 1999), which requirethat words begin

and end, respectively, with a foot. ALIGN (EDGES, level 0, PrWd, Xlevel1)
will be shown to play an importantrole in languageswith both initial and

final stress; these include CanadianFrench (Gendron 1966), Armenian
(Vaux 1998), and Udihe (Kormushin1998), which are discussed below,
and Tauya (MacDonald 1990), which is considered in section 3.3. This
adoptionof ALIGN (EDGES, level 0, PRWD, X1evel1) is adoptedratherthan
an atomisticalternativeinvolving separateconstraintsfor each edge, e.g.,
grid-basedversions of ALIGN-LEFTand ALIGN-RIGHT,since the atomistic approachresultsin increasedovergenerationin the factorialtypology
discussed in sections 2.2, 3.3, 4.1. The atomistic approachresults in 16
additionalunattestedpatterns(multipliedby two if one factorsin variation
in which of the stresses is primary)not generatedgiven the formulation
adopted here: four additionaldual stress patterns,three binary patterns,
and nine ternarypatterns(see sections 2.2 and 3.3 for furtherdiscussion).
GORDON
MATTHEW
498
The threeALIGNconstraintsare formalizedin (3) 1

Constraintsmodeling the attractionof stress to edges
(3)
ALIGN (Xievel1, L, level 0, PrWd): Every grid mark of level 1 is
a.
aligned with the left edge of level 0 of grid marksin a prosodic

word.
b.
ALIGN (xlevel
1, R, level 0, PrWd):Every grid markof level 1 is
alignedwith the rightedge of level 0 of gridmarksin a prosodic

word.
ALIGN (EDGES, level 0, PrWd, xlevel1): The edges of level 0
c.
of grid marksin a prosodic word are aligned with level 1 grid

marks.
There are also ALIGNconstraintswhich requirethat level 2 grid marks
align with the edge of level 1 of grid marks. These constraints,ALIGN
(Xlevel2, L, level 1, PrWd) and ALIGN (Xlevel2, R, level 1, PrWd), thus refer
specifically to primarystress. They are formulatedin (4).

(4)
a.
Constraintsmodeling the attractionof primarystress to edges

ALIGN(xlevel2,L, level 1, PrWd):Every grid markof level 2 is
aligned with the left edge of level 1 of grid marksin a prosodic
word.
b.
ALIGN(x1,ve12,
R, level 1, PrWd):Every grid markof level 2 is
alignedwith the rightedge of level 1 of gridmarksin a prosodic
word.
Violationsof ALIGN (Xievel2, L, level 1, PrWd)andALIGN (Xlevel2, R, level

1, PrWd) are calculated by -countingthe number of secondary stressed
syllables separatingthe primarystressed syllable from the relevantedge.
Thus, for example, the seven syllable form uacra'aoaincurs two violations of ALIGN (Xievel2, L, level 1, PrWd), since two secondary stressed
syllables intervene between the left edge of the word and the primary
stressed syllable, and one violation of ALIGN (Xlevel
2, R, level 1, PrWd),
since one secondarystressedsyllable separatesthe primarystressfromthe

rightedge. A tableauillustratingthe evaluationof all the proposedALIGN
constraintsfor a schematic six-syllable word appearsas (5). Where there
11 The ALIGN constraintsare often abbreviatedthroughoutthe paper by eliminating
the final two arguments,i.e., 'level x' and 'PrWd',and the subscripted'level' in the first
argument.ALIGN (EDGES, level 0, PrWd,xievel1) is abbreviatedas ALIGN EDGES.
499
A FACTORIAL
TYPOLOGY
STRESS
are several violations of a constraintcommittedby a form, the numberof

violations appearsin parentheses.
Adopting the ALIGN(x2, X, 1, PrWd)constraintsas opposed to constraintswhich count absolutedistance of the primarystress from an edge
(cf. McCarthyand Prince 1993) has the empiricaladvantageof creating
a more constrainedfactorialtypology of stress systems12as well as the
formal advantage(in a grid-basedtheory of representation)of being more
principled, assuming that all grid marks above level 0 must dominate a
lower level grid mark(Prince's(1983) ContinuousColumnConstraint).
(5)
Evaluationof the ALIGNconstraints
CaCaCaaa
AUGN
(xl, L)
AUGN
EDGEs
**
ALIGN
(x1, R)
*****(5)
o xx xx xx
Cy a6CF___Cy
a
X
2
x
1X
ALIGN
(X2, L)
****(5)
****** (6)
AUGN
(X2, R)
*********
(9)
**
o xxx x xx
2
****** (6)
**
****** (6)
X
'x
**
**** (4)
adC6 aCy C
X
x x
Ilx
O xx x xx x
****** (6)
*********
(9)
******(6)
*********
(9)
****** (6)
********* (9)
Ilx
*********
(9)
I x
x
x
Oxx xx x x
o xx x x xx
X
lx
x x
ox x x xx x
2x
x x
lx
Oxx xx x x
d
a da
da
.__
**
.
_ _-__
Before proceeding with analyses employing the ALIGNconstraints, it

should be noted that, although the ALIGNconstraintsdiscussed in this
paper will make referenceto the word as the stress domain,it is assumed
that other membersof the ALIGNconstraintfamily sensitive to different
stress domains, such as the root and different phrasallevels, also exist.
These ALIGNconstraintsplay an importantrole in characterizingmorphologically sensitive stress (see Alderete 1999 for morphologicalstress
12 The ALIGN(X2, {R/L), 1, PrWd)constraintsadoptedhere generateonly 79 distinct
stress systems as opposed to 93 generatedby their hypotheticalcounterpartswhich count
absolute distance of the main stress from an edge. The extra patterns,none of which are
attested,fall underthe class of fixed stress systems displayingtwo stressesper domain(see
section 2.2 for the factorialtypology of fixed stress).
500
MATTHEWGORDON
within OptimalityTheory) as well as phrase-level stress (see Gordon in

preparationfor discussion of phrase-levelALIGN constraints).
2.1.2. Initial Stress and Final Stress
Given the ALIGN constraints,we are in a position to generatethreeof the
fixed stress patterns.The generalrankingschema which generatessingle
fixed stress systems involves one general ALIGN constraintdominating
other ALIGN constraintsand *LAPSE. Following standardprocedurein
OptimalityTheory,variationin the edge referredto by the highly ranked
ALIGN constraintyields differences between languages in the edge or
edges towardwhich stress is attracted.Stress systems with a single stress
on the initial syllable reflect an undominatedALIGN (x1, L) rankedabove
ALIGN (x1, R), and ALIGN EDGES. In stress systems displaying a single
stress on the final syllable, ALIGN (x1, R) is inviolable and rankedabove
ALIGN (x1, L) and ALIGN EDGES. Finally,ALIGN EDGES plays a crucial
role in the CanadianFrench patternreportedby Gendron(1966), which
involvesprimarystresson the finalsyllable andsecondarystresson the initial syllable, even in clash contexts.13Vaux(1998) reportsa similarpattern
involving primarystress on the ultima and secondarystress on the initial
syllable in most varieties of Armenian;the secondarystress on an initial
syllable containinga high vowel is suppressed.Certainvarietiesof Udihe
also place primarystresson the ultimaand secondaryon the initial syllable
(Kormushin1998), with Nikolaeva and Tolskaya (2001) suggesting that
the secondarystress on the initial syllable is either completely suppressed
or very weak in disyllables.14
To capturethese initialplus finalstresspatterns,ALIGN EDGES ensures
that syllables at both edges of the word are stressed.The promotionof the
final stress to primarystatus in CanadianFrench, Armenian,and Udihe
reflects the rankingof ALIGN (X2,R) over ALIGN (X2,L). The schematic
example in (6) provides a closer look at the rankings yielding primary
stress on the final syllable and secondarystress on the initial syllable.
13 In his studyof CanadianFrench,Gendron(1966, p. 150) contrastsvarietiesof French

spoken in Canada,which place secondarystress on the initial syllable, even in disyllables,
with ParisianFrench,which lacks this secondarystress. He observes that vowels in initial
and final syllables are longer and articulatedmore strongly than vowels in word-medial
syllables, which, in the case of the high vowels lil and lul, may devoice or delete wordmedially. Other accounts of CanadianFrench I have located only mention the primary
stress on the final syllable.
14 NikolaevaandTolskaya(2001) also discuss a varietyof Udihe with quantity-sensitive
stress.
501
A FACTORIALTYPOLOGYOF QUANTITY-INSENSITIVESTRESS
(6)
Initialand final stress

UUT
tALIGN
EDxFs
A,GN
(x2R)
ALIGN
ALIGN
ALIGN
(xl)
(x1, R)
I(X,L
~oa
The third and fourth candidatesboth fail because each lacks a stress
on one of the edge syllables in violation of ALIGN EDGES. The second
candidate fatally violates ALIGN (X2,R) because the rightmoststressed
syllable does not bearprimarystress.If one assumesthatthe initialsyllable
in Udihe is not stressedin the clash context arisingin disyllables,*CLASH
must outrankALIGN EDGESwhich in turnis rankedabove ALIGN (x1, R).
ALIGN (xl, R) is rankedhigherthanALIGN (x1, L) therebyaccountingfor
the survivalof the stress on the final syllable in disyllables.
and PenultimateStress
2.1.3. NONFINALITY
Repulsion of stress from the right edge is capturedin OptimalityTheory by NONFINALITY(e.g., Prince and Smolensky 1993; Walker1996).
For purposes of the present grid-based analysis, NONFINALITYsimply
prohibitsa level 1 grid markon the final syllable (7).
(7)
NONFINALITY:Stressdoes not fall on the finalsyllable. (A final
syllable does not have a level 1 grid mark.)15

NONFINALITYplays a role in capturinglanguageswith a single fixed stress
on the penult. In such languages,NONFINALITYis rankedabove ALIGN
(x1, R).16This rankingproducesthe 'rightmostwithoutbeing final' stress

patterncharacterizinglanguageswith penultimatestress.17
2.2. The *L4PSEConstraintsand Peninitialand AntepenultimateStress
The *LAPSEconstraints(Prince 1983; Selkirk 1984; Greenand Kenstowicz 1996; Elenbaas 1999; Elenbaas and Kager 1999) ban sequences of
unstressed syllables. Two non-position-specific*LAPSE constraintsand
15 It is assumed that the ContinuousColumn Constraint(Prince 1983) rules out a final
syllable carryinga level 2 grid markwithout a level 1 grid markalso.
16 In languages with penultimate stress in disyllabic and longer words but which
have stressed monosyllabic words, NONFINALITY is violated in order to ensure that
monosyllables are stressed.
17 Thanksto a reviewerfor pointingout thatNONFINALITYallows for an alternativeset
of rankingsfor generatinginitial stress:NONFINALITY>> ALIGN EDGES.
MATTHEW
GORDON
502
three sensitive to lapses at word edges are assumed here. The two general LAPSE constraintswill play a critical role in the binary and ternary
stress pattems to be discussed in sections 3 and 4, respectively.The first
of these constraints,*LAPSE,bans adjacentstressless syllables and is important in binary stress systems (section 3). Violations of *LAPSE are
assessed gradiently;thus, a string of three adjacent stressless syllables
incurs two violations of *LAPSE, while a string of four adjacent stressless syllables incurs three violations of *LAPSE. Similarly, a word with
two non-contiguous sequences of two unstressed syllables suffers two
violations of *LAPSE.
The second constraint,*EXTENDEDLAPSE, is decisive in ternarystress
systems (section 4); it bansa sequenceof morethantwo adjacentstressless
syllables.Violationsof *EXTENDEDLAPSE arealso gradient;thus,a string
of four adjacentstressless syllables incurs two violations of *EXTENDED
LAPSE (andalso threeviolationsof *LAPSE),as do two non-contiguoussequences of threeunstressedsyllables. The two general*LAPSEconstraints
are formalizedin (8).
(8)
a.
b.
*LAPSEconstraints
*LAPSE:A stringof more than one consecutive stressless syllable may not occur.(A sequenceof more thanone consecutive
syllable lacking a level 1 grid markis banned.)
*EXTENDEDLAPSE: A string of more than two consecutive
stressless syllables may not occur. (A sequence of more than
two consecutivesyllables lackinga level 1 gridmarkis banned.)
It should be noted that the formulationof *LAPSE is differentfrom other

anti-lapseconstraintsin the literature,includingversionsreferringto metrical feet, e.g., Kager's (1994) PARSE-2and Green and Kenstowicz's
(1996) *LAPSE, as well as versions referringto the metrical grid, e.g.,
Elenbaas(1999) and Elenbaasand Kager(1999).18
The members of the *LAPSE constraint family crucial for capturing
the remaining fixed stress patterns are specific to stress lapses at domain edges, following work by Alderete (1999). Avoidance of stress
lapses at word edges is perhaps most obvious in quantity-sensitive languages with stress windows, e.g., Piraha (Everett and Everett 1984; Everett
1986, 1988), Kobon (Davies 1980 in Kenstowicz 1997), Chukchi(Skorik

18 Elenbaas and Kager's *LAPSEallows sequences of two consecutive stressless syllables at a word edge, unlike the *LAPSEconstraintadoptedhere (see Elenbaasand Kager
1999 for comparisonof theirgrid-basedanti-lapseconstraintwith foot-based ones).
STRESS
TYPOLOGY
A FACTORIAL
503
1961), thoughlapse avoidanceat edges also figuresprominentlyin certain

quantity-insensitivepatterns,as we will see.
The constraintsresponsiblefor creatingstress windows are membersof
the *LAPSEEDGE sub-familyof constraints.A two-syllable stress window
at the right edge, e.g., Kobon, is the result of a highly ranked *LAPSE
RIGHTconstraintwhich bans more than one stressless syllable separating
the rightmoststress from the right edge of a word. A two-syllable stress
window at the left edge, e.g., Chukchi, is attributedto a highly ranked
*LAPSE LEFT banning more than one stressless syllable separatingthe
leftmost stress from the left edge. A three-syllablewindow at the right
edge, e.g., Piraha, is attributedto a highly ranked *EXTENDEDLAPSE
RIGHT constraintbanning more than two stressless syllables separating
the rightmoststress from the right edge.
The *LAPSE EDGE constraints are assumed here to be categorical;
thus both incur a single violation of *LAPSE
the forms acrou and aocscyac
RIGHT. The latter form also violates *EXTENDEDLAPSE RIGHT. The
*LAPSEEDGE constraintsare definedin (9).
*LAPSEEDGE constraints
(9)
a.
*LAPSE RIGHT:A maximum of one unstressed syllable separates the rightmost stress from the right edge of a stress domain.
(No more than one syllable separates the rightmost syllable

with a level 1 grid markfrom the right edge.)
b.
*LAPSELEFT:A maximumof one unstressedsyllable separates
the leftmost stress from the left edge of a stress domain. (No
more than one syllable separatesthe leftmost syllable with a
level 1 grid markfrom the left edge.)
C.
*EXTENDEDLAPSE RIGHT: A maximum of two unstressed

syllables separates the rightmost stress from the right edge
of a stress domain. (No more than two syllables separate the
rightmostsyllable with a level 1 grid markfrom the rightedge.)

We are now in a position to analyze two additionalfixed stress patterns:
peninitial stress and antepenultimatestress. Both of these stress patterns
are the result of an ALIGN constraintpulling stress towardone edge and
an overriding*LAPSE EDGE constraintpulling stress towardthe opposite
edge. The result of this struggle is demonstratedin (10) which illustrates
the antepenultimatestress patternresultingfrom the ranking*EXTENDED
LAPSE RIGHT>> ALIGN (X1, L) >> ALIGN (X1, R). This ranking ensures
MATTHEWGORDON
504
thatstressfalls as far to the left as possible withoutfalling to the left of the

antepenult.19
(10)
Antepenultimatestress
*ExT LAPSE
caaCa3
ALIGN(x I R)
ALIGN(X1, L)
RIGHT
edaaa
aaaad
It.
..-.
Fixed peninitial stress is attributed to an analogous ranking except for

switchingthe referenceedge of the ALIGN and *LAPSEEDGE constraints.
In peninitialsystems, *LAPSELEFTis ranked above ALIGN (X1, R) which
in turntakes precedenceover ALIGN(X1, L). The effect of this rankingis

to force stress as far to the right as possible withoutfalling to the right of
the peninitialsyllable, as shown in (11).
(I 1)
Peninitialstress
Goaaa
*L"SE LLF
ALIGN(x1, R)
ALIGN(Xi, L)
Interestingly, I am not aware of any stress system with a three syllable win-
dow at the left edge which would necessitateadoptionof the hypothetical

left edge counterpartto *EXTENDEDLAPSE RIGHT.20
19 Note that penultimate stress also follows from essentially identical rankings to
those driving antepenultimatestress except for replacing *EXTENDEDLAPSERIGHT
with *LAPSE RIGHT.Thus, fixed penultimate stress reflects two possible rankings:
NONFINALITY
?>ALIGN(X1, R) >> ALIGN(X1, L) or *LAPSERIGHT?>ALIGN(X1,
L) >> ALIGN(X1, R). Despite this functional overlap,both NONFINALITY
and *LAPSE
RIGHTmust be maintainedas separateconstraints,since they serve independentfunctions
in many cases. For example, NONFINALITY
plays a crucial role in binary stress patterns
with final stress avoidance(section 3.2.1), whereas*LAPSERIGHTcapturesstresswindow
effects and plays a decisive role in dual stresssystems (section 2.1.5), and in certainbinary
plus clash (section 3.3) and binaryplus lapse stress systems (section 3.2.2).
20 Thanksto an anonymousreviewerfor pointingout the stresspatternof Terena,which
potentially provides evidence for a *LAPSEEDGEconstraintprohibitingmore than two
consecutive stressless syllables at the left edge of a word. According to Harden(1946),
stressin Terenais phonemic andmay fall on any one of the firstthree syllables of the word,
providedthe stressedsyllable is non-final.In his detaileddescriptionof Terenastress,however, Bendor-Samuel(1963) shows that stress is phonemic within a two-syllable window
at the left edge, with third syllable stress arisingundercertainphrase-levelmorphological
505
2.2.1. Dual Stress Systems

Thus far, we have accountedfor all of the single fixed stress patterns:initial stress, peninitialstress, antepenultimatestress, penultimatestress, and
final stress. In addition,we have analyzed the initial plus final dual stress
pattern.This leaves only two types of dual fixed stress systems to account
for: ones with both initialandpenultimatestress(e.g., SibutuSama,Lower
Sorbian)and those with initial and antepenultimatestress (e.g., Georgian).
Since both of these stress patternsresult from very similar rankings,we
will only analyze one subtype of the more common patternin detail, the
initialplus penultimatepattern.Differencesbetween the rankingsunderlying this patternand those behind closely relatedones will be pointed out
later.
Our exemplar of the initial plus penultimatepatternis Sibutu Sama
(Allison 1979; Elenbaasand Kager 1999), in which primarystress falls on
the penultimatesyllable and secondarystress on the initial syllable,2' with
the addedprovisionthatonly the penultis stressedin trisyllabicwords;this
patternis exemplifiedin (12) (examples from Elenbaasand Kager 1999).
(12)
SibutuSama stress
a. bissala
'talk'
b. bissalahan
'persuading'
c. bissalahainna
'he is persuading'
d. bissalahankaimi
'we are persuading'
Before developing an analysis of Sibutu Sama stress, one additionalconstraintmust be introducedto account for the fact that adjacent stresses
are prohibitedin trisyllabicwords in SibutuSama. This ban on clashes is
indicativeof the high rankingof a constraintbanningstress clashes. This
constraint,which was introducedby Prince (1983) as a rhythmicprinciple
prohibitingadjacentstresses (cf. Kager 1994; Alber 1997; Elenbaas 1999;
ElenbaasandKager 1999 for *CLASH in OptimalityTheory),is formulated
in (13).
and syntactic conditions which would fall under the rubric of intonationalphenomena
in other languages, e.g., marking contrastivefocus, negative imperatives,nominal nonspecificity.Thus, althoughthere may be a window effect in Terena,it seems unlikely that
third syllable prominencein Terenaresults from the same class of *LAPSE EDGE stress
constraintsrelevantfor cases discussed in the text.
21 This penultimateplus initial stresspatternis reportedfor unprefixedwords (see Kager
1997 for discussion and analysis of prefixedforms which display additionalcomplexities).
506
(13)
GORDON
MATTHEW
*CLASH: A stress domain does not contain adjacent stressed

syllables. (Adjacent syllables carryinga level 1 grid mark are
banned.)
A string of two adjacentstresses incurs one violation of *CLASH.An additional violation is incurredfor each additionalstressed syllable in the
sequence. A sequence of three consecutive stresses thus violates *CLASH
twice, as does a word containing two stress clashes separatedby one or
more unstressedsyllables.
We now consider one set of rankingswhich generatethe Sibutu Sama
stress facts. As the forms in (12) indicate, in order for both stresses to
be realized, a word must have at least four syllables; this indicates that
*CLASHis undominated.It is crucially rankedabove ALIGNEDGESas
trisyllabicwords have a single stress on the penult in violation of ALIGN
EDGES: bissala, not *bissala. *LAPSERIGHTis also undominated and is
instrumentalin accountingfor the stress on the penult;it is rankedabove

ALIGNEDGES,as evidenced by the stress on the penult ratherthan the
initial syllable in trisyllabic words: bissala, not *bissdla. *LAPSERIGHTis
also ranked above both ALIGN (X1, R) and ALIGN (X1, L), since words of
at least four syllables stress the penult, even thoughthis entails additional
violations of ALIGN constraints: bissalahanna, not *bssalahanna. ALIGN
EDGESis ranked above ALIGN(X1, R), as all words with at least four
syllables have a second stress on the initial syllable in addition to the
one on the penult: bissalihan, not *bissldhan. NONFINALITY
is ranked
above ALIGNEDGES,as the final syllable is never stressed: bissaldhan,
not *bissalahdn. Finally, ALIGN (X2, R) is ranked above ALIGN (X2, L)
therebyensuringthat the penult takes primarystress in forms containing

two stress:bissalhhan,not *bissalcihan.The rankingsfor SibutuSama are
summarizedin (14).22 A tableau demonstratingthe rankings for Sibutu
Sama appearsin (15).
22 Note thatthe rankingsin (14) are not the only ones which generatethe Sibutu Sama
stress system. If ALIGN (X1, L) is ranked above ALIGN (X1, R), but below *LAPSE RIGHT,
NONFINALITY
ceases to play a decisive role in yielding the stresson the penult.This alternate set of rankingsis nearly identical to the rankingswhich generatethe antepenultimate
plus initial stresspatternfound in Georgian(see below).
507
(14)
Constraintrankingsfor SibutuSama
*CLASH, NONFINALITY, *LAPSE RIGHT, ALIGN (X2, R)
ALIGN EDGES
4
ALIGN (X1, R), ALIGN (X2, L), ALIGN (X1, L)
(15)
bissala
ON- *LAPsE
ALIGN
Eixim
ALIGN
(x1, R)
(x2
*CLSH NON- *LAPSE ALICN

FINAL RIGHT (x2. R, 1)
ALIGN
EDGES
ALIGN
(XI, R)
ALIGN
(X2. L?
FINAL
gE bissla
bissala
l.
bissaldi
b'{ssalE
_m
bwissalahan
'b'sssal
ihan
blssalahanl
btssslahan
bi-ssalahin
(XL
ALIGN
I)
jE_E__
,.
bissalahan
bissalahani
Ritr'r
l5ssasla
bissalahan
ALIGN
ALIGN
(x,. R, 1)
*CLSH
ALIGN
(X1L)
;:
_*
__
I_
**
"
'
--------E
_____
The Lower Sorbianpattern(Janas 1984), in which the initial stress rather

than the penult is preservedin trisyllabicwords, e.g. w5sitsojska 'fatherland', psiijasiel 'friend', and the primarystress is on the initial syllable
ratherthanthe penultin words in which both the penult and the initial syllable arestressed(see section 2.2.2 for discussionof the link betweenclash
resolution and primary stress placement), results from some minor rerankingsrelativeto those observedin SibutuSama. First, FLAPSERIGHT
is rankedbelow ALIGN EDGES, therebyaccountingfor the stress on the
initial syllable ratherthe penult in trisyllabicwords. *LAPSERIGHTmust
nevertheless still be ranked above ALIGN (X1, L) as the penult carries
stress in all words over three syllables. Finally,ALIGN (X2, L) is ranked
above ALIGN (X2, R), reflectingthe fact that the stress on the initial syllable ratherthan the stress on the penult is promotedto primarystress in
GORDON
MATTHEW
508
words with two stresses. The rankings for Lower Sorbian are summarized
in (16).
(16)
Constraint rankings for Lower Sorbian

*CLASH, NONFINALITY, ALIGN (X2, L)
4
ALIGN EDGES
4
*LAPSE RIGHT
4
ALIGN (X1, L), ALIGN (X2, R), ALIGN (X1, R)
Georgian (Zhgenti 1964; Aronson 1991) displays a slight variation on the
Sibutu Sama pattern. In Georgian, stress falls on both the initial and the
antepenultimate syllables in words over four syllables long and only on
the antepenult in tetrasyllabic words.23 The stress on the antepenult reflects an undominated *EXTENDEDLAPSE RIGHT (rather than *LAPSE
RIGHT) crucially ranked above ALIGN (X1, L). The secondary stress on
the initial syllable in words of at least five syllables reflects the ranking
of *LAPSE LEFT >> ALIGN (X1, R). The fact that the initial syllable
rather than the peninitial syllable is selected for stress within the window
defined by *LAPSELEFT is ascribed to the ranking of ALIGN (X2, L) over
ALIGN (X2, R). Finally, ALIGN (X2, R) is ranked above ALIGN (X2, L),
thereby accounting for the promotion of the stress on the antepenult to
primary stress in words with two stresses. The rankings for Georgian are
summarized in (17).
(17)
Constraint rankings for Georgian

*CLASH, NONFINALITY, *LAPSE LEFT, *EXT LAPSE RIGHT,
ALIGN (X2, R)
ALIGN (X1, L)
4
ALIGN (Xl, R), ALIGN EDGES, ALIGN (X2, L)
23 The two sourcesconsultedon Georgianstressdo not describethe same facts, although
the data in each are mutually compatible with the other source. According to Zhgenti's
(1964) phonetic study of Georgianprominence,Georgianwords are stressed on the antepenultimatesyllable. Aronson's(1991) grammarmentionsan additionalstresson the initial
syllable in words of at least five syllables.
A FACTORIAL
TYPOLOGY
STRESS
509
Walmatjari(Hudson 1978) displays a variantof the Georgianpattern,

but with primary stress on the initial syllable and secondary stress on
the antepenult,except in tetrasyllabicwords which have secondarystress
on the penult ratherthe antepenult.This patternfollows from essentially
the same rankingsas Georgian,save for the promotionof ALIGNEDGES
above the otherALIGN constraintsand above *EXT LAPSE RIGHT, ensuring stress on the initial syllable ratherthan the antepenultin words with
four syllables. The option of stressingthe penultratherthanthe antepenult
in words longer than four syllables follows from the optionalrankingof
*LAPSERIGHTabove ALIGN(X1, L) (but below *CLASHand ALIGN
EDGES,as trisyllabicwords lack a stressfalling withinthe windowdefined
by *LAPSERIGHT.)The promotionof the initial stress to main stress in
words with two stressesreflectsthe rankingALIGN(X2, L) >>ALIGN(X2,
R).
2.3. A Factorial Typologyof Fixed Stress
We now consider the factorialtypology of fixed stress systems generated
by the constraints.The factorialtypology consisted of determiningall the
sets of stresspatternsgeneratedby permutingthe constraintrankingsin all
logically possible orders.All logically possible stress patternsfor words
containing between one and eight syllables were considered as candidates. The only potentialcandidatestresspatternsnot consideredwere ones
which violated Prince's (1983) CULMINATIVITY
condition,i.e., it was assumed that all words have one syllable that is more prominentthan all
others.24
The candidate forms consisting of two and three syllables in (18)

provide a relatively small example of the set of potential output forms
being evaluated.(1 is a primarystressedsyllable, 2 is a secondarystressed
syllable, and 0 is a stressless syllable.)
(18)
Candidateforms containingtwo and threesyllables

2 syllables
[01]
[10]
[12]
[100]
[010]
[001]
3 syllables
[102]
[120]
[012]
[210]
[021]
[201]
[122]
[212]
[221]
[21]
24 I leave the issue open as to whetherCulminativityis a (likely
inviolable)constraintor
a universalpropertyof Gen which limits the pool of candidatesavailablefor evaluationby
the constraintset.
510
MATTHEWGORDON
The number of possible stress patternsincreases commensuratelywith

word length according to the formula n * 2(n-l), where n
number of
syllables: 4 in disyllabic words, 12 in trisyllables, 32 in words with 4

syllables, 80 in words with 5 syllables, 192 in words with 6 syllables, 448
in words with 7 syllables, 1024 in words with 8 syllables.
The twelve constraintsin the factorialtypology are repeatedas (19).
(19)
Constraintsin the factorialtypology

1. ALIGN (X1, L)
5. ALIGN (X2, R)
9. *LAPSE RIGHT
2. ALIGN (X1, R)
6. *CLASH
10. *EXTENDED LAPSE RIGHT
3. ALIGN EDGES
7. *LAPSE
11. *LAPSE LEFT
4. ALIGN (X2, L)
8. *EXTENDEDLAPSE
12. NONFINALITY
The only fixed ranking assumed involved constraints referring to

primary stress. It was assumed that either ALIGN (X2, L) or ALIGN (X2,
R) was rankedbelow all other stress constraints.This rankingreflectsthe
cross-linguisticparameterizationof primarystressplacementin languages
with multiplestressesin a single word:languagesare internallyconsistent
in eitherpromotingthe rightmostor the leftmost stressto primarystress, a
consistency reflectedin the parameterizedapplicationof either End Rule
Right or End Rule Left in derivationalframeworks(Prince 1983).25In
terms of constraints,it is thus hypothesizedthatthere are no languagesin
which both ALIGN (X2, L) and ALIGN (X2, R) are rankedhighly enough
to be active, an observationwhich is capturedby demoting one of them
below all otherconstraints.26
25 As a reviewerpoints out, the fact that only one of the ALIGNconstraintsreferringto
main stress is highly rankedenough to influence the grammarin any given languagecould
be takenas evidence thatthereis a single ALIGN(X2, X) constraintwhich is parameterized
on a language-specificbasis accordingto which edge is relevant.
26 As an exercise, the factorialtypology was also computedwithout fixing the ranking
of either ALIGN(X2, L) or ALIGN(X2, R) below other stress constraints.This resulted
in a proliferationof generated stress systems (including binary and ternarysystems, as
well as fixed stress systems): 302, as opposed to 152 with the a priori rankingassumed,
includingboth those with an undominatedALIGN(X2, L) and those with an undominated
ALIGN(X2, R). The extrageneratedpatternsarise when both ALIGNconstraintsreferring
to primarystressare rankedabove one or both of the general*LAPSEconstraints,*LAPSE
and *EXTENDED
LAPSE,which themselves are rankedabove the ALIGNconstraintsreferringto secondarystress.This set of rankingshas two effects. First, it limits the number
of stresses, since increasing the numberof stresses necessarily increases violations of at
least one of the ALIGNconstraintsfor main stress. Second, it positions those stresses
in such a way as to minimize violations of the highly ranked*LAPSEconstraint(s).The
overalleffect is to shift the location of the stress(es) as a functionof numberof syllablesin
the word, a mechanismwhich appearsto be absent in attestedlanguages.Although space
511
The task of calculating a factorial typology for a set of twelve constraintsis appreciable.Permutingthe rankingof twelve constraintsin all
possible ways yields an a prioritotal of 12! or 479,001,600 logically possible rankings. Given the tremendousanalytic complexity of calculating
the output sets generated for all these rankings, the task was delegated to Hayes et al. (2000) OTSoft software package, which computes
factorialtypologies for a set of constraintsand candidatesusing Tesarand
Smolensky's(1993) ConstraintDemotion algorithm.The inputfiles for the
OT softwareconsist of a set of constraintsand a set of candidateforms and
theirconstraintviolations.To expeditethe process, the candidatesandtheir
violations were also generatedby computer.27
Each outputset generatedby the typology consists of a series of eight
forms, each one correspondingto a word with a differentnumberof syllables. The set of forms in (20) is one sample output set, corresponding
to the stress patternfound in Sibutu Sama, generatedby the typology:
primary stress on the penultimatesyllable and secondary stress on the
initial syllable in words of at least four syllables.
(20)
Sample outputset (SibutuSama)

[1], [10], [010], [2010], [20010], [200010], [2000010],
[20000010]
Overall results of the factorial typology indicated a strikingly good fit

between the survey of attested stress patterns and the proposed constraint set. Of the 479,001,600 logically possible rankings and the
10,823,318,000,000 logically possible output sets given a set of candidates consisting of between one and eight syllables,28a mere 152 of these
limitationsprohibitdetailedexaminationof these patterns,we may considerone such case,
involving final stress in di- and trisyllabic words but penultimatestress in longer words.
This system results from the ranking ALIGN (X2, L), ALIGN (X2, R), *LAPSE RIGHT >>
*EXT LAPSE >> ALIGN (X1, R) ?> ALIGN (X1, L), ALIGN EDGES. The inviolable status
of both ALIGNconstraintsreferringto main stressensures only one stressper word, which

must fall on one of the final two syllables due to highly ranked*LAPSERIGHT.Because
*EXTLAPSEcannot be violated in disyllables and trisyllables,the stress is free to fall on
the final syllable, therebysatisfyingthe highest rankedALIGNconstraint,ALIGN(X1, R).
However,in longer words, where *EXTLAPSEviolations become an issue, the stress shifts
to the penult in orderto minimize violations of *EXTLAPSE.
27 Thanksto Bruce Hayes for providinginvaluableassistancein this endeavor.
28 This figure of 10,823,318,000,000 reflects the cross-classificationof the numberof
stresspatternsfor wordsrangingfrom one to eight syllables. It is arrivedat by multiplying
the numberof possible stress patternsfor a word with n syllables by the numberof stress
patternsfor a word with n+l syllables by the numberof stresspatternsfor a word with n+2
syllables, etc., where the numberof syllables rangesfrom one to eight.
TABLEIV
Single stress languagesgeneratedby factorialtypolog

1.
Initial:Chitimacha
2.
Peninitial:Lakota
3.
4.
Peninitialplus nonfinality:No q.i., Hopi (q.s.)

Antepenultimate:Macedonian
5.
6.
Penultimate:Nahuatl
Final: Atayal
AL(X1,L) >> AL(XI ,R), A

*LAPSELEFT>> AL(X1,
*LAPSE LEFT, NONFINA
*EXTLAPSERIGHT>> A
NONFINALITY
>> AL(X1
AL(X1,R) >> AL(X1,L), A
513
possible outputsets were generatedby the proposedconstraints.29Of the

152 generatedstresssystems, only 79 were distinctstresspatternswith differing locations of stress;the other 73 of the 152 had counterpartsamong
the 79 differingonly in which of the stresses is the primarystress. In fact,
all stress patternsinvolving more than a single stress per word have a
counterpartdifferingonly in which of the stresses is the primaryone; this
symmetry is attributedto differences in which of the ALIGN constraints
referringto main stress is highly ranked.
All of the stress patternsin the survey of stress systems were generated, demonstratingthat empiricalcoverage was sufficient.While certain
unattestedpatternswere generated,these unattestedpatternsappearto reflect for the most part non-pathologicdeviations from attested patterns.
This point will become clearer as we consider the results in greaterdetail, considering stress patterns in turn by their taxonomic description.
While the focus in examining the coverage of proposed constraintsis on
quantity-insensitivestress systems, stress patternsin words consisting of
only light syllables in quantity-sensitivelanguages were also considered.
The justificationfor inclusion of such patternsis the assumptionthat the
same constraints are operative in both quantity-sensitiveand quantityinsensitive systems with quantity-sensitivityresulting from additional
weight-sensitiveconstraintswhich are highly rankedin quantity-sensitive
but not in quantity-insensitivesystems.30
2.3.1. Factorial Typologyof Single StressSystems
The single fixed stress systems generatedby the typology are considered
in TableIV, which containsa verbaldescriptionof each stresspattern,one
set of rankings which generates each pattern,and a quantity-insensitive
language(q.i.) instantiatingthe pattern.In case a patternis not found in a
quantity-insensitivesystem but does occur in words containingonly light
syllables in a quantity-sensitivestress system (q.s.), this is also indicated.
Six single stress patternswere generatedby the factorialtypology, all of
which are attested, including all five of the attested quantity-insensitive
29 A complete list of stress pattems and stratifiedconstraintrankingsbehind these pattems are availableat the author'swebsite:
http://www.linguistics.ucsb.edu/faculty/gordon/pubs.
30 It should be noted thatcertainhighly rankedconstraintsin quantity-sensitivesystems
have the effect of banningstress on light syllables. For this reason, some quantity-sensitive
systems have stresspattemsin wordsconsisting of only light syllables thatarenot captured
by the constraintsdiscussed in this paper,but are capturedby other constraints,such as
ones which repel stress from peripheralsyllables not sonorous enough to supportboth
stress and otherprosodicpropertiesassociatedwith peripheralsyllables (see Gordon2001,
in preparationfor discussion).
514
MATTHEWGORDON
single stress patterns(initial,peninitial,antepenultimate,penultimate,and

final stress),plus a sixth pattern,in which the peninitialsyllable is stressed
unless it is final. This patternis not attested in any quantity-insensitive
systemsI have found,butis foundin Hopi (Jeanne1982) in wordsin which
the firsttwo syllables are light.
2.3.2. Dual Stress SystemsGeneratedby Factorial Typology
Dual stressresultsfromthe low rankingof the general*LAPSEconstraints,
coupled with the relativelyhigh rankingof one or more constraintswhich
requirestresses at or near edges: ALIGN EDGES and/or a *LAPSE EDGE
constraint(s).I briefly summarizethe generalrankingschemas generating
the core types of dual stress systems in terms of their stress locations (see
Table V for the rankingswhich generatethe subpatterns).As we saw in
section 2.1.2, a highly rankedALIGN EDGESis responsiblefor dual stress
patternswith initial and final stress.If a *LAPSEEDGE constraintoriented
toward the right edge is rankedhighly in conjunction with the ranking
NONFINALITY>> ALIGN EDGES, dual stress systems involving stress on
the initial syllable plus eitherthe penult (high ranked*LAPSE RIGHT) or
the antepenult(high ranked*EXTENDEDLAPSE RIGHT)result.Peninitial
plus final stress reflects a highly ranked*LAPSELEFT and either a highly
ranked *LAPSERIGHTor *EXTENDEDLAPSE RIGHTin conjunction with
the rankingof ALIGN (X1, R) above both NONFINALITYand ALIGN (X1,
L). In such systems, the stress on the final syllable is in response to a

*LAPSE EDGE constraintpulling stress toward the right edge. Because
ALIGN (X1, R) is rankedabove NONFINALITYand ALIGN (X1, L), the
rightmoststress falls on the finalsyllable withinthe window definedby the
LAPSE EDGE constraint.Peninitialplus penultimatestress follows from
the ranking *LAPSE LEFT, NONFINALITY?>ALIGN (X1, R) >> ALIGN
(X1, L). The stress on the peninitial syllable follows from the rankingof
LAPSE LEFT above ALIGN (X1, R), while the penultimatestress results
from the rankingof NONFINALITYabove ALIGN (X1, R). Because all the
peninitialplus penultimatepatternsinvolve an inviolable NONFINALITY,
all have penultimatestress ratherthanpeninitialstress in disyllables.
Hypothetical systems involving peninitial plus antepenultimatestress
are not generated, as both peninitial and antepenultimatestress are the
result of highly ranked*LAPSE EDGE constraintsacting antagonistically
toward an ALIGN constraintpulling toward the opposite edge. Because
one of the constraintsALIGN (X1, L) or ALIGN (X1, R) must be ranked
above its sister constraint pulling in the opposite direction, rankings
will never give rise to the bi-directionalpull requiredfor peninitialplus
antepenultimatestress.
515
All of the dual stresssystems come in variantsdifferingin which of the

stresses is the main one, accordingto whetherALIGN(X2, L) or ALIGN
(X2, R) is highly ranked,and differing in whether stress clashes are allowed or not, dependingon the rankingof *CLASH.In case clashes arenot
allowed, there are different strategiesfor resolving the clash, depending
on the relative rankingof the constraintsproducingthe stresses oriented
towardeach edge of the word.
Table V summarizesthe dual fixed stress systems generatedby the
typology. The shaded stress patterns are discussed later. '1' represents
both primary stressed and secondary stressed syllables in Table V and
subsequenttables, wheredifferencesin which of the stressesis the primary
one are considered a separate dimension from the distinction between
stressedand unstressedsyllables. Stresspatternswhich are not predictable
from the verbal descriptionare indicated in bracketswhere '1' indicates
a stressed syllable and '0' indicates a stressless syllable. For example,
[10] following a stresspatterndescribedas initial and final means thatthe
stress on the initial syllable survivesin the potentialclash context arising
in disyllabic words.
Seventeen distinct patternswere generated, each of which occurs in
two variantsdifferingin which.of the stresses is the primaryone, depending on the relative ranking of ALIGN (X2, L) and ALIGN (X2, R). All of
the dual stress patternsfound in the survey were generated.Other novel

generatedpatternsinvolvedcombinationsof elements attestedin isolation
but not in all combinations.Thus, unattestedpatternsdeviate from others in their treatmentof stress clashes, their combinationof stress sites,
and their location of primarystress. For example, pattern2 differs from
pattern 3 only in resolving stress clashes in disyllables in favor of the
initial ratherthan the final syllable. Pattern6 is analogous to the attested
initialplus penultimatepatterns4 (LowerSorbian)and 5 (Sanuma)except
for its tolerance of stress clashes in trisyllabic words. Two of the more
interestingpatternsare 7 and 11. Pattern7 displays initialplus penultimate
stress whereverthis will not trigger a stress clash, but surprisinglylacks
the stress on the initial syllable in four syllable words where there is no
potentialfor a stress clash with the penultimatesyllable. This system has
the interesting property of having *EXTENDEDLAPSE be highly ranked
enough to influence the stress system, but yet not highly enough to be
inviolable, as it is in binaryand ternarysystems. The asymmetrybetween
words with four syllables which lack stress on the initial syllable and
longer words which stress the initial syllable is attributedto differences
in the possibility of satisfying *EXTENDEDLAPSE. For every word over
four syllables, positioning a stress on the initial syllable results in fewer
GORDON
MATTHEW
516
TABLEV
Dual stress systems generatedby the factorialtypology
] AL EDGr s >> *CLASI.
.1.
&
clash OK
Initia- final,
Prinmanon final. Caadian Fiench
Pmaryon initial:Nonie
2. Initial& final, no clash, [10]
*CL,Asii>> AL,EDGLs`> AL 1x L) or NONFINALITY

t
Rt
1>>XAl,
Primar;on initial:No.-ne
3 Ilntial& final,no clash, [0 ])
~~~~~ ~
'ta~
>> ALIEI
*CLASH
NOR
R)1R > A.t,(x_L)
finalUie
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
i-nmron~
Y >> AL EDwNS >> *LAISE RIGIIr
*C'LASt NONtINALIT
AL(X1,L) AL (XI, R)
N
i
S >>
RIiGITAL EDGE
NONFI\ALrry ->> *LA E1.
*CLASH1,
4. Initial & penult, n1oclash, [100]
>>
Primair on inittal Lower Sorbiai

5. IntAal & pentult, no clash, [0101
AI
k(x1,L), Ai ttx, R)
on penuilt.Sanuma
~~~~~~~~~Pnnmary
s
6. Initial& peniul.t.claVshOK.
.................................
No\taS,i\T-Y,
Primarvon initial: None
7. Initial & penult, no clash, [010], [0001
Primarvon iniitial:Nonie
9. Initial & antepenutlt no clash, [0100]
____________________________________
12. Peninitial & penult,clsh OK, [I01

i;
_ _
13. Peninitial& penult,no clash, [10], [1010]
_s~isiW :.
:
14. PeninitiaI& penult,no clatsh,[10], [0100]
..............
..................................
..........................
...........................
Pn
rmary on finmal'No-ne
on peniinitial:N'one
Primiarv
._
RiGITr A
*Exs LAPSE
*CiLASH,
AI(x1,L) >>
.LvsE >>
ALh?. R? > Ai EDGE S
P2mro nepnult: None
, *LAPSE LEFT AL (XI,
NaNTFINALITYGirTAPSERGI
RI> i>A.,(x1,L>A>At.Eixiws
aryonpenultimate:None
-~~~~~~~~~~Pni
*L.AisERioIT, *LAIp,sLEFT>>
NONFINALITY,
*CLASH,
S
AL (X1,R)>>AL(X1,L), ALEDGE
Primaryon penultiniate:None
. *CIA.SE *LAPSELEFT NONINA.SX.
ITY,*EXTLAPSFE
R
s
Riau >> AL(X1,R >>
AL'(x1,L),ALEDGE
. ....
... ......... . ........ ....... ........ - -..................
.............
Primr on penultitnate: None
FRGHT -> A, (XI, Ri).>*LA.5iL LEFT, *LAPSE
AlIX ,L) NFeiiINALiiY
16. Peninitial & final no cilash, [0o1] [0O10]

__
rim on peniniitial:None
17. Penii tial & final, nioclsh [010]
aryonAt
Prmayonpniiia:NoneN
xon antepenlt: Gieorgian

Pnimar
Ai,
*CLAs1,. NoNNzAIITY, *EXT L.PSE RI(GiT Al;>
>> AL(x R)
EDGE S>> .AiL)?
on initial. Wqlmahjari
Primary
([10010] posiblel
Ii. Initial& antepenult,niocIlash,[1010,] [00100]
...........
>> *EXT LTYS.I
*EXTLASE
*CLASH,No,NFINA.LD.LnY,
RIGH(1T,*ULAPs
LUFT> > A .(x 1,L.)>> Ai. EDGES Al x 1 R)
10. Initial& antepenult,no clash,[1 010]
Pnmary on veninitial: None

I1S.Peninitial & fin al, cash OK
Riti'r>> Ai.tx,L,(
*CLASH, *LAXPSiE
-->AL
R)>- AL EDGES
Prim r, on pnul None

*EXTL.APSRIIIT >> ALEDGES,
NONFINALITY,
A(X1,L>) AL(X1, R)
.Primal, on antepenult:None
3_
8. Initia & antepentit,clashOK
..........
*LA
APSE.RiI:T >> Ai.,E.Gol S >>
,L) AL,(Xi, R)..........................................................

I Primaryon penilt: Nonie
AL(X
........
............................... ...... .....................................................
y.
*CLASII *LAPSLLEFT,*EXTLAPSERicHT >> AL (XI,

RR>?>Ai.,(x L)
AI Ec;TE;s
LEFT>> Ai (X1.R) >>

*C AsH *LAPSERGHT, *LAPSF
4, NONFINALITY
violations of *EXTENDEDLAPSE than another candidate without stress

on the first syllable; thus, the candidate five syllable form [00010] fails
because it violates *EXTENDED LAPSE while the winner [10010] does not.
31 Udihe instantiatespattern1, however,if the initial syllable actuallycarriessecondary
stress in disyllables; Nikolaeva and Tolskaya(2001) express uncertainlyabout whetherit

is completely stressless or carriesweak stress.
A FACTORIAL
TYPOLOGY
STRESS
517
Crucially,the winner incurs no more violations of higher rankedALIGN

(X1, L) thanthe failed candidatewhich outranks*EXTENDED LAPSE. The
winning candidateincursmore violationsof ALIGN (X1, R) thanthe winner, but this is irrelevant,as ALIGN (X1, R) is rankedbelow *EXTENDED
LAPSE. Similarly, candidates for words longer than five syllables incur
fewer violations of *EXTENDED LAPSE if they stress the initial syllable
thanif they do not; hence, the stresson the initial syllable in words longer
than five syllables. In four- syllable (and shorter)words, in contrast,there
is no need to position a stress on the initial syllable, since doing so does
not result in any improvementin the satisfactionof *EXTENDED LAPSE;
thus the winner [0010] satisfies *EXTENDED LAPSE as well as the failed
candidate[1010] but has the advantageof betterhonoringALIGN (X1, R).
Pattern11 is nearly identicalto pattern7 except that *EXTENDED LAPSE
RIGHT is rankedhighly ratherthan *LAPSE RIGHT. Patterns7 and 11 are
unusualin lacking an expected stressin non-clashcontexts and appearnot
to have analogs in real languages.32
A relevantobservationis that the proposed constraintsonly generate
dual stress systems involving combinationsof stresses which are attested
in isolation. This observation is not trivial, since the set of logically
possible two stress systems is quite substantial;these include systems
with combinations of stresses not found in isolation, e.g., initial plus
preantepenultimate,postpeninitialplus antepenultimate,etc., as well as
combinationsdiffering as a function of numberof syllables in non-clash
contexts, e.g., initial plus final stress in three syllable words and initial
plus penultimatestress in four-syllablewords. It is also a desirableproperty of the proposed constraints,though the possibility of accidentalgaps
precludesestablishingthis with confidence, that they fail to generateunattestedtwo stress systems in which both stresses fall nearthe same edge,
e.g., initial plus peninitial stress, antepenultimateplus penultimatestress,
penultimateplus final stress.
Turningto the matterof overgeneration,the small numberof dual stress
languages (14 in the survey) makes it difficult to assess whetherthe unattesteddual stress systems generatedby the constraintsreflect accidental
gaps in the set of attesteddual stresssystems or whetherthey aregenuinely
pathologicaland thus destinedneverto occur regardlessof the size of the
sample. The situation with two stress systems thus differs from the case
of single stress systems, where the relativelylarge numberof single stress
systems makes it easier to distinguish between accidental gaps and illformedpatterns.Thereare a few points,however,in favorof the accidental
gap interpretationfor the absence in the real world of many logically pos32 Pattern7 is generatedby
foot-basedconstraintsas well (see Gordonin preparation).
518
MATTHEWGORDON
sible dual stress patterns.The first of these is that dual stress languages
as a whole are relatively rare relative to single stress languages. While
we cannot be sure of the reasons for the rarityof dual stress in general,
we may speculate that their rarityis relatedto the fact that a stress at or
near a single edge is sufficientto serve the demarcativefunction of stress
suggested by Hyman (1977). Pursuingthis line of reasoning, adding an
additionalstressat the opposite end of the wordwould providelittle if any
benefit from a parsing standpoint,perhapseven complicating the online
parseby introducinga second stress for the listenerto track.
The second argumentin favor of the accidental gap interpretationis
that most of the unattesteddual stress patterns,with the exception of the
unattestedinitial plus final system (pattern2), involve elements which are
independentlyrare. The first of these rare propertiesare stress clashes,
which are avoided by most quantity-insensitivelanguages regardless of
their type of stress system. Thus, all but two of the dual stress languages
(12 of 14) foundin the survey,the exceptionalones being CanadianFrench
and Armenian,do not tolerate stress clashes. This observationis particularly obvious in the case of initialplus penultimatestress, the most widely
attestedtype of dual stresssystem.All nine languageswith initialplus penultimatestresseitherposition stresson the penultor the initial syllable, but
not both, in trisyllabicwords. Avoidanceof stress clash is also apparentin
binarysystems (section 3) and ternarystress systems (section 4).33 Given
the rarity of stress clashes in general, it is plausible that the unattested
but generatedpatternsinvolving stress clashes fail to emerge because they
violate an anti-clash constraint.The fact that clash avoidance is more
widely attestedthanlapse avoidancesuggests thatit is morethanjust sheer
rhythmicprincipleswhich militateagainstclashes. One may speculatethat
clash avoidance effects are particularlystrong because adjacentstressed
syllables detractfromthe relativeprominenceenjoyedby syllables in nonclash contexts. Nevertheless, although clash avoidance appearsto be an
importantfactor in creatingmany of the gaps in the factorialtypology of
dual stress, there are two unattestedbut generatedpatternswhich stand
out from others in lacking an expected stress even in non-clash contexts:
patterns7 and 11. These would appearto be the most likely candidatesfor
unnaturalstatusamong the dual stresssystems generated,althoughthereis
a possibility thattheirabsence is an accidentattributedto the overallrarity
of dual stress systems.
33 It is interesting to note that clash avoidance is another area in which quantityinsensitive and quantity-sensitivestress systems appearto differ from each other,as stress
clashes involving at least one heavy syllable are common in quantity-sensitivelanguages.
TYPOLOGY
A FACTORIAL
STRESS
519
Another rarely attested property shared by many of the unattested

systems is their positioning of stress on syllables which tend to not be
stress-attractingin single stress systems. Several of the generatedbut unattestedpatternsinvolve peninitialor antepenultimatestress in additionto
anotherstress at or nearthe opposite edge. Peninitialand antepenultimate
stress are independentlyrare even in single stress languages (see Hyman
1977 for possible explanationsfor the rarityof these stresssystems). Given
the independentrarity of peninitial and antepenultimatestress in single
stress systems, it is not surprisingthat systems displaying one of these
stress locations in combinationwith anotherstress do not show up in any
language.
In summary,there are three factors which plausibly conspire to create
many if not all of the gaps in the match between generatedand attested
patterns:the rarityof dual stress systems in general,the avoidanceof stress
clash, and the rarityof peninitial and antepenultimatestress. Because all
of the unattestedpatternsgeneratedby the factorialtypology shareat least
one of these properties,and most display morethanone, it is reasonableto
assume that their likelihood of surfacingis substantiallyreducedrelative
to other stress systems displayingmore favorablecharacteristics.
In discussing overgeneration,it may be notedthatan atomisticalternative to ALIGN EDGES involving separateconstraints,each requiringstress
at a differentedge of the word, has the undesirableeffect of generating
an additionalfour unattesteddual stress patternsnot generatedby ALIGN
EDGES. One of these patternsinvolves stress on only the final syllable
in words of two and three syllables, but both initial and final stress in
longer words,i.e., [01], [001], [1001], [10001], [100001], etc. This pattern
reflectsan undominatedconstraintrequiringa stress at the rightedge. The
stress on the initial syllable in words of at least four syllables is due to the
rankingALIGN (X1, L) >> *EXTENDEDLAPSE >> ALIGN (X1, R). The
initial stress in tetrasyllabicand longer words resultsin better satisfaction
of *EXTENDED
LAPSEwithout any additionalviolations of ALIGN(X1,
L). An additionalunattestedpattern generatedby an undominatedconstraintrequiringstress at the right edge involves final stress in disyllables,
initialand finalstressin trisyllables,andpeninitialandfinalstressin longer
words, e.g., [01], [101], [0101], [01001], [010001], etc. The unexpected
stress on the initial ratherthan the peninitial syllable in trisyllables reflects the workof an undominated*CLASHcombinedwith an undominated
*LAPSELEFT and *LAPSERIGHT.The thirdunattesteddual stress pattern
involves initial stress in disyllabic and trisyllabic words, and initial and
final stress in longer words: [10], [100], [1001], [10001], [100001], etc.
In this case, a constraintrequiringstress on the leftmost syllable is un-
520
MATTHEW
GORDON
dominated as is *EXTENDEDLAPSE RIGHT,which comes into play only
in words of at least four syllables. The rankingALIGN(X1, L) >> ALIGN

(X1, R) ensures that the final syllable within the three syllable window
at the right edge is stressed, while ALIGN(X1, L) is ranked above the
constraintrequiringstress on the rightmostsyllable. The final unattested
dual stress patterngeneratedby the atomistic versions of ALIGNEDGES
involves initial and penultimatestress in words of at least four syllables,
andinitialand finalstressin trisyllabicwords:[10], [101], [1010], [10010],
[100010], etc. The constraintrequiringthat stress fall on the leftmost syllable is undominated, as are *LAPSERIGHTand *CLASH, which together
yield stress on the final ratherthanthe penultimatesyllable in trisyllables.

Withinthe two syllable windows at the left and rightedges the location of
stress is determinedby the rankingALIGN(X1, L) ?>ALIGN(X1, R).
2.3.3. Uniformityof PrimaryStressPlacement
Turningbriefly to the shaded sub-patternsin Table V, all of which are
unattested,there is a propertythat differentiatesthem from those which
areunshaded.In the shadedpatterns,primarystressfails to fall a consistent
distancefrom a word edge across wordsof differentlengths. For example,
in the unattestedsubtypeof pattern5 (primarystresson the initial syllable
and secondary stress on the penult in words with two stresses), primary
stress falls on the penult in words with two or three syllables, whereas
primarystressfalls on the initialsyllable in wordsof at least foursyllables.
Thus, the location of the primarystress relative to the word edge in the
unattestedvariantof pattern5 differs as a functionof numberof syllables
in the word. This contrastsminimally with the attested variantof 5, in
which primarystress consistently falls on the penult in words of varying
lengths. Only in words that are not long enough to provide a docking site
for primarystresswhich is consistentwith thatfoundin longer words,e.g.,
in monosyllabic words, is the preferencefor uniformityof primarystress
placementrelativeto an edge violated.The absence of stress systems such
as the unattestedvariantof 5 reflectsa strongcross-linguistictendencyfor
stress systems to be consistent in their location of primarystress across
differentwords lengths, as observedby van der Hulst (1984) and Hammond (1985). As Hammondpoints out, this has consequences for stress
clash resolutionin shorterwords. The survivingstress in clash contexts is
characteristicallythe one carryingprimarystress in longer words where
multiplestresses are present.Thus, in the attestedvariantof pattern5, the
stresson the penultratherthanthe stresson the initial syllable is preserved
in trisyllabic words where stressing both the initial and the penultimate
syllables would entail a stress clash.
521
While we cannot be sure of the motivationsbehind the preferencefor

uniformityof primarystressplacement,one might speculatethatpositioning primarystress the same distance from a word edge across words of
differentlengths is functionallymotivatedin that it provides a consistent
and reliable means for determiningword boundaries.34Given a string of
speech consisting of several words, it is thus possible to use the primary
stress,the perceptuallymost prominentstress, to parsethe stringinto individual words, therebyaiding in lexical access. Hyman (1977) has made a
similarsuggestionregardingthe demarcativefunctionof positioningstress
near word boundaries;hence the popularityof languageswith stress at or
near word edges.
Whateverthe ultimatereason(s) for the cross-linguisticpreferencefor
uniformityof primarystress placement, we can attributethe absence of
the shaded stress patterns,which comprise a largepercentageof the unattested fixed stress patternsgeneratedby the typology, to theirinconsistent
positioningof primarystressacross words of differentlengths.Ultimately,
the principlethat excludes the shaded candidatesshould be codified as a
formalconstraintblocking theirgeneration.However,the formulationand
implementationof such a constraintgoes beyond the scope of the present
paper, since it would appearto requirecomplex formal mechanismsnot
currentlyavailablein OptimalityTheory,in particular,output-outputcorrespondence constraintsevaluated across forms differing in number of
syllables ratherthan forms which are paradigmaticallyrelatedto the same
root.35
3.
BINARY
STRESS
Anotherwidely attestedtype of stresspatternplaces stresson every second

syllable. These 'binary' stress patterns are rhythmic in the sense that
stressesare separatedby a single unstressedsyllable. Binarystresspatterns
34 The preferencefor positioning primarystress a uniform distance from a word edge
is strongerin quantity-insensitivestress systems thanin quantity-sensitivesystems, where
it is routinely violated (see the discussion of binary stress in section 3). The issue of why
quantity-insensitiveand quantity-sensitivestress systems should behave differently with
respect to primarystress placementis an interestingone thatmust awaitfurtherresearch.
35 See Myers (1999) for discussion of output-outputcorrespondenceconstraintsrequiring identitybetween forms which arenot relatedto the same rootbut which are similarwith
respect to one or more phonologicalproperties.Myers's constraintsaccountfor analogical
extension of phonological altemations from one paradigm to other paradigmscontaining forms bearing similar properties,e.g., extension of velar softening altemations and
irregularverb paradigmsin English.
522
MATTHEWGORDON
TABLE VI
Binary stresspatterns
Schematic forms
Pattern
1. Odd-numberedsyllables 6a6a&, 6acr&a6a
from L to R
2. Even-numberedsyllables oo&oo, o6&cyo&
from L to R
3. Even-numberedsyllables oraor6,&aor6r
from R to L
4. Odd-numberedsyllables
Ka&a6,a&u&a6
from R to L
Example Lgs.
Czech (Kucera1961),
Maranungku(Tryon 1970)
Araucanian(Echeverraand
Contreras1965), Sirenikski
(Menovshchikov1975)
Cavinefia(Key 1968),
Warao(Osbom 1966)
# of igs.
1237
Chulupi(Stell 1972),
UrubuiKaapor(Kakumasu1986)
differ as a function of whether the alternatingpatternorigins at the right

or the left edge of the word and whetherthe patternbegins with a stressed
or an unstressedsyllable (Hayes 1980; Prince 1983). There are thus four
logically possible purebinarystresspatterns.The simple binarystresspattems are summarizedin Table VI, along with the numberof languages in
the surveyinstantiatingeach pattern.A list of languageswith binarystress
is found in Appendix 1.
One of the moreinterestingobservationsaboutTableVI is the existence
of three languageswith stress on even-numberedsyllables counting from
the left (pattern2) and five with stresson odd-numberedsyllables counting
fromthe right(pattern4). These patternsareinteresting,since they involve,
in foot-basedterms,iambs; thus, (ad)(ad)(ad). The existence of such patterns,thoughthey arerelativelyrarecomparedto patterns1 and 3, is rather
unexpectedgiven the hypothesizedconfinementof iambic-typestresspatterns to quantity-sensitivesystems (Hayes 1985, 1995). This hypothesis
has spurreda body of literatureattemptingto explain the apparentasymmetricaloccurrenceof patterns2 and 4 in TableVI in quantity-sensitive
but not quantity-insensitivesystems (e.g., Kager 1993; Eisner 1997). An
importantfeatureof such analyses is their exclusion of patterns2 and 4
in quantity-insensitivesystems. The attestationof all logically possible
binary stress patternsin the present survey, however, suggests a greater
degree of symmetryin binary stress systems than such theories allow for.
36 Three of these languages display optional final stress avoidancein odd paritywords:
Burum, Selepet and Northern Sami (see Appendix 1). This patternis observed in roots
in Waorani(counted among the 18 pattern 1 systems in Table VI); suffixes stress even
numberedsyllables countingfrom the right.
37 An additionallanguage,Waorani[Auca] (Pike 1964), stresseseven numberedsuffixal
syllables countingfrom the right.
TYPOLOGY
STRESS
A FACTORIAL
523
One favorablefeature of the proposed analysis, sharedwith stress theories positing symmetricalfoot inventories(e.g., Hayes 1980; Prince 1983;
Halle and Vergnaud1987; Elenbaas and Kager 1999), is its admittance
of all logically possible binary patterns,as will become apparentin the
discussion thatfollows.
3.1. Pure Binary Stress
A featureof all binarysystems is the undominatedstatus of both *LAPSE
and *CLASH, which togetherproduce the alternatingstressed-unstressed
pattern.The differentsubtypesof pure binarystress systems emerge from
differences in the rankingof ALIGN EDGES and the ALIGN (X1, IL/RI,
0), constraints.As a detailed analysis of a language with a simple binary
stress system, considera language which stressesodd-numberedsyllables
beginningfrom the left edge of a word, a patternwhich is instantiatedby
the forms in (21) from Maranungku(Tryon 1970).
(21)
a.
b.
c.
d.
e.
Maranungkustress
tiralk
mxraxpaet
mutl1iikin
ita
jaeijarm
'ltiritiri
Da
'saliva'
'beard'
'salt-waterturtle'
the Pleiades'
'tongue'
In Maranungku,the highest rankedALIGN constraintis ALIGN EDGES

which, nevertheless,is cruciallyrankedbelow both *LAPSEand *CLASH.
If ALIGN EDGES were rankedabove *LAPSE, we would get both initial
and final stress in even parity words: *jdyarrnatainstead of jdaVarmaita.
If
ALIGN EDGES were rankedabove *CLASH, we would get stress clashes
in even parity words: *jdJacrmatc!
*LAPSE is also
instead of jadrarnauta.
crucially rankedabove NONFINALITY,as the final syllable is stressed in
words with an odd numberof syllables: mdrwepdvt
ratherthan *merwpwt.
ALIGN EDGES is rankedabove ALIGN (X1, L), as evidencedby the stress
put. ALIGN (X1,
patternin odd paritywords:mcerwpd?t
ratherthan*mwr4w
in
R);
otherwise,
stresswould fall
L) is turnrankedhigherthanALIGN (X1,
on even numberedsyllables counting from the left in even parity words:
ratherthan the attestedjadVarmacta.
Finally, ALIGN (X2, L) is
*jajafrmatta
rankedabove ALIGN (X2, R), therebyaccountingfor the promotionof the
firststress to primarystatus:mWera?pdt
ratherthan*mircep(ft.
GORDON
MATTHEW
524
The rankings for Maranungkuare summarized in (22). A tableau

illustratingthe rankingsappearsin (23).
(22)
Constraintrankingsfor Maranungku
*CLASH, *LAPSE, ALIGN (X2, L)
ALIGN EDGES
ALIGN (X1, L)
4,1
NONFINALITY,ALIGN (X1, R), ALIGN (X2, R)
(23)
mLerspet
*CLASH
iw nMrrpt
m : epa-L
ja '(rn1ata
*LAPSr
ALITN
(X;2, L)
ALIGN
.;XfS
ALIGN
(XI, >
AL IGN
NoNFINAL
w
(X2,w
ALIGN
I (XI, R)
___Emm
E...
,(
L )22 E
*CLks:A LIGNt}tT*LAPSE ALIGnN A I(.N
NUN.L}INATALIGN 3ALIG,N
w?j4 arma~ta___
iL'1armbLtA
~j
1___
jaiy rmatAXI
j4qlannatil
____
____
Othersimple binarystress patternsdiffer minimally from Maranungkuin

their rankingof the ALIGN constraints,following the spirit of foot-based
analyses of binarity(e.g., Crowhurstand Hewitt 1994). The UrubuKaapor
(Kakumasu1986) pattern(pattern4 in Table VI), involving stress falling
on odd-numberedsyllables counting from the right ratherthan the left,
resultsfromALIGN (X1, R) being rankedaboveALIGN (X1,L). The effects
of this rerankingare seen in even paritywords:ocra in UrubuKaapor,but
6ucuain Maranungku.The location of primarystress on the last stressed
syllable in Urubu Kaaporis the result of ALIGN (X2, R) being ranked
above ALIGN (X2, L).
The Sirenikski (Menovshchikov 1975) stress pattern (pattern2), in
which stress falls on even-numberedsyllables countingfrom the left, results from the rankingof ALIGN (X1, R) above both ALIGN (X1, L) and
ALIGN EDGES. Ranking ALIGN (X1, R) above ALIGN (X1, L) ensures
that stress skips over the initial syllable in even paritywords:u6oo rather
than*6a&a.The rankingof ALIGN (X1, R) over ALIGN EDGES is decisive
in avoiding both initial and final stress in odd parity words: a6o and not
*,
* ciu
CF
A FACTORIAL
TYPOLOGY
STRESS
525
The Cavinenia(Key 1968) pattern(pattern3), involving stress on evennumberedsyllables counting from the right, differs minimally(beside the
rerankingof the ALIGNconstraintsreferringto primarystress) from the
Sirenikskipatterndue to ALIGN(X1, L) being rankedabove ALIGN(X1,
R). Owing to this ranking, stress skips over the ultima and falls on the
penult insteadin even paritywords:ououcand not *ucsu6.
3.2. BinaryPlus Lapse Systems
3.2.1. BinaryPlus ExternalLapse
Many languages display a basic binarypatternwith stress lapses arising
undercertainconditions.One subtypeof these 'binaryplus lapse' systems
is found in languagesin which stress falls on odd-numberedsyllables from
left to right except for final syllables, e.g., 6a'aaa, 6aocxyxcxc.
Final stress
avoidance results in a stress lapse at the right edge of a word containing
an odd numberof syllables. Languagesdisplayingthis type of stresslapse
at the right peripheryinclude Pintupi (Hansen and Hansen 1969, 1978)
and Karelian(Leskinen 1984). I have located 14 languages(listed in Appendix 1) with this pattern.This patternis the result of nearly the same
constraintrankingswhich generatethe strictbinarypatterninvolvingstress
on odd-numberedsyllables from left to right, the importantdifference
being that NONFHNALITY
is undominated(or nearly so, subordinateonly
to CULMINATIVITYin languages with monosyllabic words) in the binary
plus lapse system. This rankingproduces an externallapse in odd parity

words, as the lapse occurs at a word edge. Crucially,because there is no
counterpartto NONFINALITY
at the left edge, there is no mirrorimage to
the Pintupipatterndisplayinga lapse at the left edge of a word.In fact, such
patternsappearto be unattested,a point to which we returnin section 3.4.
We now look more closely at the constraint-basedanalysis of the binary
plus externallapse system found in Pintupi:stress on odd-numberednonfinal syllables counting from the left (Hansen and Hansen 1969, 1978).
Examples of stressin Pintupiappearin (24).
(24)
a.
b.
c.
d.
e.
Pintupistress
tJutaja
puiijkalatJu
tkamulimpatiujku
tftii uulampatiu
kiranJiulullmpatJuqa
'many'
'we (sat) on the hill'
'ourrelation'
'the fire for our benefitflaredup'
'the firstone (who is) our relation'
The crucial constraint reranking which differentiates Pintupi from

Maranungkuis NONFINALITY above *LAPSE. This rankingyields final
526
MATTHEW
GORDON
stressavoidancein odd paritywords,even thoughthis createsa stresslapse

in final position: in Pintupi, t1utaja instead of *tjtap. NONFINALITY
is also ranked above ALIGN EDGES, as indicated by the lack of stress
on final syllables. ALIGN EDGES takes precedence over *LAPSE, as odd
parity words stress the initial syllable even though this entails a stress
lapse later in the word: tJu'tajanot *tJutaja. *LAPSE is still rankedabove
ALIGN (xl, L), since stresslapses are avoidedin non-finalcontexts even if
this costs additionalviolations of ALIGN (xl, L): t1admuli'mpat'
irjku not
*tJamulimpjaitJ'uku.
Finally, the ranking of ALIGN (X2, L) over ALIGN
(x2, R) correctlypromotesthe first stress to primarystatus in words with
not *ti iamulimpatJ
multiplestresses:tdamulimpatJui.yku
u'4ku.The rankings
of the stress constraintsfor Pintupi are summarizedin (25). A tableau
demonstratingthese rankingsappearsin (26).
Constraintrankingsfor Pintupi
(25)
NONFHNALITY, *CLASH, ALIGN (X2, L)

ALIGN EDGES
*LAPSE
ALIGN (xl, L)
ALIGN (xl, R), ALIGN (X2, R)
(26)
puti3kalatiu
pii~ik~iau
P'iLjkaiilatiio
a1&_ti1
putli3k
p4tiIklatju ..
L
No'QFINAIA AL(N
(X2}L)
ALIGN
EDGES
*LAPSE
ALIGN
(X1, L)
ALIGN
(XI, R)
ALIGN
(X2. R)
*CLASH
ALIGN
EDGES
*LAPSE
AUGN
(X1, L)
ALIGN
(X1, R)
ALIGN
(X2, R)
____:__~
_____
________
*CLASIH
_______
------_-----_----_--------
'.***!
tiamuulimpatiuqkuNONFINAL
tiJmuflmipatjuqku_
________
__l
q*k
----X--
ALIGN
(x2L)
A FACTORIAL
TYPOLOGY
STRESS
527
3.2.2. Binary Plus InternalLapse

Anothertype of binaryplus lapse stress system involves an internalrather
than an external lapse. A general feature of these systems (termed "bidirectional"by Elenbaas and Kager 1999) is their positioning of a fixed
stress at one edge and a binarypatterninitiating at the other edge. The
fixed stress at one edge is the resultof a highly rankedconstraintrequiring
stress to fall at or near an edge; constraintswhich have this effect are the
*LAPSEEDGE constraints and ALIGN EDGES. The origination point of the
basic binarypatternfound in binaryplus lapse systems is a functionof the

ranking of the ALIGN (x1, {L/R}, 0) constraints and ALIGN EDGES.
An example of a binary plus internal lapse system is provided by

Garawa(Furby 1974), in which stress falls on even-numberedsyllables
counting from right to left but skips over an even-numberedpeninitial
syllable in favorof stressingthe initial syllable. The resultis a stresslapse
following the initial syllable (which carriesmain stress) in words with an
odd numberof syllables.38Examplesof Garawastress appearin (27).
(27)
a.
b.
c.
d.
Garawastress
jami
ala
puinJ
'eye'
'white'
'armpit'
watJimpaIJu
'at your own many'
naTneinmufkunJinamira
The proposedanalysis of the differencebetween Garawaand Pintupifollows the basic insight of McCarthyand Prince's (1993) work within a
foot-basedframeworkandinvolves a differencebetween the two languages
in the relativerankingof ALIGN(X1, L) and ALIGN(X1, R). In Garawa,
unlike in Pintupi,ALIGN(X1, R) is rankedabove ALIGN(X1, L), thereby
accounting for the fact that the stress lapse in odd parity words occurs
immediatelyafter the initial stress ratherthan immediatelybefore the final stress; thus, nadriginmukun
inamira, not narivlnmukin'inamira. The
Garawa rankings are summarizedin (28), followed by a representative
tableauin (29).
38 Stresseson syllables otherthanthe initial one and the penultarereportedto be weaker

than those on the initial and penultimate syllable in Garawa,a fact not capturedin the
presentanalysis, which assumes two levels of stress.
528
MATTHEWGORDON
Constraintrankingsfor Garawa
(28)
*CLASH, NONFINALITY, ALIGN (X2, L)
ALIGN EDGES
*LAPSE
ALIGN (X1, R)
ALIGN (X1, L), ALIGN (X2, R)
(29)
punJala
NOFIN
pdiinala
ALIGN
' (X2, L)
*CLSH
ALIGN
EDGES
*LPSE
ALICGN
(X1, R)
ALIGN
(X22R)
ALIGN
(X1,L)
___
puniala
nariqiinmukunhnamira
NONFIN
ndri.nmaW=manimira
nArixjinmuiikun1inkimira
I3rlin niniamira
ndriginirukniinamira
ALIGN
*CLS11 ALIGN
(x2L)
*LPsE
ALIGN
ALIGN
x
,Lx
I~
(X,
EDGES
ALIGN
,
R)
(x2
_
__"_
___
'__8_
_1__*"*!
***
___
--
19!
Garawais not the only binaryplus lapse stress system with internalstress
lapses, though such systems appearto be fairly rare.A virtuallyidentical
system exists as an option to strict binarityin Polish (Hayes and Puppel
1985), as well as in Spanish (Harris 1983) and Indonesian(Cohn 1989),
with Spanish and Indonesiandiffering from Polish and Garawain being
sensitive to weight in locating primarystress. Piro (Matteson 1965) also
displays a binary plus lapse stress system, in which stress falls on the
penult,on the initialsyllable, andon odd-numberedsyllables following the
initial syllable;Piro thus differsfrom Garawain commencingthe alternating stress patternat the left edge ratherthan the right edge. The Spanish,
Indonesian, Polish, and Piro patternsare all generatedby the proposed
constraints,as we will see in section 3.4.3.
3.3. Binary Plus Clash Systems
Anothervariantof the binarystresspatternis found in languagesin which
stress follows a basic binarypatternbut falls on adjacentsyllables under
A FACTORIAL
TYPOLOGY
STRESS
529
certaincircumstances.These 'binaryplus clash' systems are the result of

nearlyidenticalrankingsto those relevantin binaryplus lapse systems, except for the low rankingof *CLASHand the undominatedstatusof *LAPSE
in binaryplus clash systems. Like binaryplus lapse systems, binaryplus
clash systems involve a fixed stress at or near one edge, due to the high
rankingof one of the *LAPSE EDGE constraintsor ALIGN EDGES, and a
binarycount originatingat the oppositeedge of the word.Binaryplus clash
patternsappearto be rarein quantity-insensitivesystems, but are found in
Tauya (MacDonald 1990), Biangai (Dubert and Dubert 1973), Southern
Paiute(Sapir 1930, Harms 1966), and Gosiute Shoshone (Miller 1996).39
To illustratethe rankingswhich generatebinaryplus clash systems, let
us considerin detail the analysis of the binaryplus clash system in Tauya,
a TransNew Guineanlanguagespokenin PapuaNew Guinea(MacDonald
1990). In Tauya,primarystress falls on the final syllable, and secondary
stressfalls on the initialsyllable and on alternatingsyllables countingback
from the final one. The requirementthat the initial syllable be stressedin
Tauyacreates a stress clash between the first two syllables in words with
an even numberof syllables; forms illustratingstress in Tauyaappearin
(30).
(30)
a.
b.
c.
d.
Tauyastress
nono
Tuneta'
momunepa
japatijWf6
'child'
'mat'
'X sat and X...'
'my hand'40
The constraintrankingsat work in Tauyaclosely resemble those operative in Maranungku.One importantdifferencebetween the two languages,
however,lies in the relativerankingof ALIGN EDGES and *CLASH. The
presence of stress on both the initial and the final syllables is indicative
of ALIGN EDGES being rankedabove *CLASH in Tauya,as a stress clash
regularly arises in words with an even number of syllables. The Tauya
39 It shouldbe notedthatthe SouthernPaiuteandGosiute Shoshone stresspatternscould
be treatedas displaying a binary patternat the level of the mora, and not necessarily at
the level of the syllable (see Sapir 1930; Harms 1966; Hayes 1995 for descriptionand
discussion).
40 Note thatunstressedvowels optionallyreduce to schwa.
530
MATTHEWGORDON
constraintrankingsare summarizedin (31), followed by a tableauin (32)

illustratingthe relevantrankings.
(31)
Constraintrankingsfor Tauya
ALIGN EDGES,
*LAPSE, ALIGN (X2,
R)
4
ALIGN (X1,
L)
4
NONFINALITY,
*CLASH,
ALIGN (X1,
R), ALIGN (X2, L)
(32)
?uneta
ALIGN
EDGES
momunepa
ALIGN ;ALIGN
EDGES
momnunepa
mfOnulp
ALIGN *tLPSE
A-IGN
*LPSE
(X24R) tX(X2,)
ALIGN
*LPsE
ALIGN NONFINIAL'*CLASH ALIGN1 ALIGN

(x14L)(XL)X1R
ALIGN
NONFINAL,4CLAS1H ALIGNIALIGN
(XI R)
t(X-,
ALIGN
NONFINAL
__
,______
'6
4!''
momrunepa
in6rnZ~p,
__
*CLSH ' A LIG
ALIGN
4*E
!~;
4
*****!
_4*
4*
3.4. A Factorial Typologyof BinaryStress

The binary stress patternsgeneratedby the factorial typology appearin
TableVII, along with the rankingsgeneratingeach patternand an example
of a language, if any exist, displayingeach pattern.Each system includes
a schematic patternfor an even and odd parity word (1 = stressed syllable, 0 = stressless syllable), as well as one for shorterwords whose stress
patternsare not recoverablefrom the description,e.g., in potential stress
clash contexts in systems not toleratingclashes. Shaded systems fail to
position primarystress a uniformdistancefrom an edge, following earlier
discussion in the context of dual stress systems in section 2.2.3.
3.4.1. Pure Binary SystemsGeneratedby the Factorial Typology
A total of 23 distinctbinarystress systems were generatedby the factorial
typology, including simple binary,binaryplus lapse, and binaryplus clash
systems. All 23 patternswere generatedin pairs differingin which of the
stresses is primaryand which is secondary.Looking at the subclasses of
A FACTORIALTYPOLOGYOF QUANTITY-INSENSITIVE
STRESS
531
TABLE VII
Dual stress systems
Simple BinarT
1.Even-numberedfrom L: 11010 [01010101J
(issl
-.
most isprima:Arau
bri oati
ian
Odd-nLiumbered
from R: [CLOASIIA [0E1)0101
(IAH Ls>>
ArI\ RI' *i,li

o
r
:
ALEr rs's AL(x1,Ry>,-
1 s prima: Utrubu
KaapoLI
_
*CIAMS *LAPSi' A
-A
_ ___R_
_ _ _h_
_ _ _ _ _ _ _ _ _ _ _ _ ___C_vi_ef_
___________
is primain sv
Rig------h_mot
"(I Ii *LAPSi EwAIE
.
"rs>>AL.(XL) >I>
"Resm
0.4 0,1 hspim'No1
[9J,[01010101]
3. .ven-numb'dfromR
Leflmo
prima . lalalo 1
o iPso ( a,)
4 Odd-numb5dfromL:[101010], [1
1, I010
~ ~
Lefmostis p)rimary:
Maramm'ku
5. Even-numb'dfromL & final, clash O1K:
.........
t
.(....l0-.....O-IO--01
LefimostiS primary:Noite
O
noi
s
r
i
.
r(;
Binary plus clash
>>' AL.(.xR) >AL EXL$Gs"- ALSX
*LAPSE
*CLAsii
1,L),
.................
l1 Yupik
RightmostiSprimaiy:No qi1. CentralA ian
___tterance-nSeliaal
6. Eve-numb'dfroinL & peitult,clash OK:

I9LPJIQIO19)1g A rS.1^L,
Ltmost is primary:Sotitlem Paiute
7, Odd-numb'd
from L & final, clashOK
hI.L)
1 R<)>>
I RwAcr lR)N1issli3A
(
*L.Ant
All$ 1L1 Cssi
1- El
.
RigInmostis prnmaryNon
s
Extu's Alix O >>
Aii,4
A
t.:xix1L),
I.NASs
Lefimostis prismary:
GosiuteShoshone
8. Odd-nui b'd froit I & penolt,clashOK:
[10101 101,[IOIOIOIOL
Left ost is .rimarv: None
9. Eves-numb'dfont R & iniitial,clashOK:
t,
,,,,,,,,,,,,,,,,
[1)10ll19 [[l].[l01.1AU.X
Lefimosiis primaar:None
I0. Odd-numb'd from R & initial, clash OK:
*CLsiil
is prim
None
Rig1100
LAS
t,nt'N\sINA.ITY *LAeSt RiGHT>>, Eww s>
g,x Ii *CiASi
AtLx RI
RiLsltmostis prinmaryNone.
NT.INH,
MUS
Al 1>: LI, ALEDGE.ss"
,,v,,,,,,,,,,,,,,,,,,,,,,,,,CLA,
RI 5AtSIlSA
..........................
.....,,..........
...f+?,
Rightmosris primar at: 'ii
Al 0
Lsis
Lefimst is primary:None
AI
is.
L) >Ax1iRI,
GAISI
is primary Tanya
Rigitmnsot
BinAri ohis lapse

I 1.Odd-niwhb'dfromL & penult no claslit [010]
.CLs,,tsH. oNos
INALITY1
*LAPE RiGHirn>> LEis""S'
>> .i..x
R
,
L PSEis
[10100101i, [10101019 j,
1
.....................-.
I.......
12. Odd-nunib'dfrost L & penult,ntoclaslir[1001
>>
*CLArS
1NeNrisso\tNIzT
A5,'
ElYi-5s>>
*L5 LPSI
vLsRiltrir
0(R 1> Ar..............
) >""AIA1(X]
[19]09!M101010101.I*Oplj'
Lfimostis primar. None_____________________

13. Odd-numb'd from L. minusfinal:
i0031]
..
Lentmst inprima':
15. Odd-nuntb'd front L &
_[Tt
[P)199
penult,
-~~~~~~~~~~~~~~t
*CLAsH, *LAPSERI'HTs>>"AL EDS">"

Ai
>>AS(N LOh ....
. .'m
i
*Ls'.s...
clash in
""s (
............ .
mo,st Lisprimary: None
Rig'tmtost
16. Odrd-numb'd fromL & final, clash is disyllales:

[101010)1,
.,,
..o-,o
......
,,,,e.i
........
,,[,,,1. [10101001].~~~AL(X1j,R3
Lt4mast is primar:None
>'
Ai,EDGES
20. Evren-noins'd from Rplus peitinitial: [1010]
[Q(919!PL[9.(99]9[9] .
Left a Lprimary:No.
==00IgjiolilA:
..
Xo~~~~~~~~~~~~~~~~~~s
Leftnmost
is prmary None
from,R & initial:[10]
23. Odd-numbered
.[!t).19191LL9 ..........tO.'..RI:.........L.
Leftmoatis, prima'Nose
None
> LAPSE >>Al,(XL):
........... .................
NON-FIN
ALiYi
>i>
AL EDGEs>> *LAP'S'
sa~~~~~~~~~~~~~~~~~~~~~~~~~~.........
Oh...."... ..........L...
'
Rightmo is primary.Indonesian
f >
>>Al
NNCFs1INAtirs
AiT E0s>
Cl,ASH>>*LAPS.>>
Ri hntmo i prinmaryNoie
"
L
CL
ASIH
No.'?NALITY
A,iEiIor >>
4W xg RI. [.(A1,L)
a
imryN
Nos'TirNA LTYs
*L.
*C.LA'Hi
~isA
22. O id-n lunsb

fl'om Oh& initial,e4shitni_ dilyllables'
ti primar
>
AIt-
..... ........
'CrLASi
21 . Odd-nuntb"d from R & initial no clash: [01 ]
*CLASiI
L)"
Rightmostisprimary: 'one
17, Even-numb'd from R & initial, no clash: [010]
Letimostisprintary:None
19. Even-numib'd
froisO
R & initial,iIo tasb:[1
[101010], [10101010]1
Lftmiosis primas':Garawa
>> AL IXSts
NONIFtNAlIStY*LAPsts RGTOiT
trisyllables:
[It) ]0i0].[i 101010]. ,

.
.
'
,
=Lees =isprim 'iN n
18, Eves-oiumh'dfiont R & iiiitial,clash iii trisyllable
>
AP,
""
A (t~L" (A>jtk RI
' ois'
a''N
Lftntnstis primary':
Pintopi
14. Odd-nunb'd from L & [fina1:[I0
..[l..)91)1.
41010
At..c,rs
. CLAS;ii,No.NF'1
11>
NA1X5
92.1.9]9.91,.3
[1010tO
LEs'E
L
AL1,OhRI"";-
LI 5L.elsFrAl DOt'.8
is simar_tNone
g~~~~hnmost
CiJAs>H"Ai,
.At.
s'L> 5*Ltsrs>>
'
pfs
EIF '
,s arI
LAsi'S
"-" AL
.L....
.
........
1,Ri >>
............
None
Rightintsitsspritnarsv:
*C.LASl>>"AlE..i.ws >>, *LAPSgNoNFiNALirry">>
OsIms
532
MATTHEW
GORDON
binarystress systems in detail, all four of the simple binarypatterns(patterns 1-4) are attested:even-numberedfrom left to right, even-numbered
fromrightto left, odd-numberedfromleft to right,andodd-numberedfrom
rightto left. These aredifferentiatedfromeach otherby the relativeranking
of the ALIGNconstraints,following discussion in section 3.1. .
3.4.2. BinaryPlus Clash SystemsGeneratedby the Factorial Typology
Of the six binaryplus clash systems, five are attested:patterns5, 6, 7, 9,
10, with one of the patterns(5), however,foundonly in all-light wordsin a
quantity-sensitivesystem. The unattestedpattern(8) differsfrom Southern
Paiute (pattern6) in stressing odd-numberedratherthan even-numbered
syllables; the directionalityof the stress count is the same in both pattern
6 and pattern 8. The difference between patterns 6 and 8 is attributed
to differences in the ranking of the ALIGNconstraintsrelative to each
other.In pattern6, ALIGN(X1, R), is rankedabove both ALIGN(X1, L),
and ALIGN EDGES, whereas in pattern 8, ALIGN (X1, R), is sandwiched
between higher ranked ALIGN EDGES, and lower ranked ALIGN (X1, L).
An importantfeatureof the binaryplus clash systems is that the fixed

stressoppositethe originationpoint of the binarypatternis always situated
at or nearthe periphery:on the initial syllable, the penultimatesyllable, or
the final syllable. A fixed stress on either the initial or final syllable is
attributedto a highly ranked(thoughnot necessarilyundominated)ALIGN
EDGESranked above the ALIGN(X1, {L, RI) constraintpulling stress
toward the opposite edge, as in patterns 5, 7, 9, and 10. The result is
an additionalstress at the opposite peripheryfrom the originationpoint
of the binary pattern.A fixed stress on the penult (patterns6 and 8) is
attributedto an undominated*LAPSERIGHTin conjunctionwith undominatedNONFINALITY.
This rankingforces stress on the penult, even if the
antepenultis stressed.
A key feature of the proposed constraintsis that they only generate
binary plus clash systems with a fixed stress on the initial syllable, the
final syllable, or the penult. This confinementof clashes to the periphery
capturesthe confinement,in foot-based terms, of degeneratefeet to peripheralpositions (see Kager 1989; Hayes 1995 for discussionof degenerate
feet). The only threeclash contexts generatedin the presenttheoryinvolve
the initial and peninitial syllables, the antepenultimateand penultimate
syllables, and the penultimateand final syllables;in fact, these arethe only
three clash environmentsattestedin actualquantity-insensitivesystems.
At firstglance, it might appearthatthe proposedconstraintswould predict additionalclash contexts, such as peninitialplus postpeninitialstress
or preantepenultimateplus antepenultimatestress, given the existence of
A FACTORIAL
TYPOLOGY
STRESS
533
*LAPSE LEFT and *EXTENDED LAPSE RIGHT constraints. The rankings,

however,fail to generatethese hypotheticalclash types, as, for every constraint,there is at least one othercandidatelacking these types of clashes
thatcommits fewer violations. Candidateswith peninitialstress and rightwarditerationof stressstartingwith the postpenitialsyllable, i.e. '6du6cr6,
requirethe rankingof ALIGN (X1, L) above ALIGN (X1, R). Under this
ranking,however,thereis anothercandidate,6ac6uFa6,which bettersatisfies ALIGN (X1,L) andfaresno worse withrespectto constraintsotherthan
lowly rankedALIGN (X1, R). Candidateswith antepenultimatestress and
leftwarditerationof stress startingwith the preantepenult,i.e., oduuauu,
requirethe rankingof ALIGN (X1, R) above ALIGN (X1, L). Under this
ranking,though, there is anothercandidate,a6csc6cca6,
which better satisfies ALIGN (X1, R) and fares no worse with respect to constraintsother
thanlowly rankedALIGN (X1, L).
3.4.3. BinaryPlus Lapse SystemsGeneratedby the Factorial Yypology

Turningto the binary plus lapse systems, of the 13 generatedpatterns,
four are attested.Pattern11, odd-numberedfrom left to rightplus penult,
is found in Piro (Matteson 1965). Pattern13, odd-numberedfrom left to
right except final syllables, is widely attested,as, for example, in Pintupi
(Hansen and Hansen 1969, 1978). It is worth noting that the proposed
constraintsfail to generatethe mirrorimage counterpartof Pintupi,oddnumberedfrom right to left except the initial syllable. This pattern is
not generatedbecause there is no left-edge counterpartto NONFINALITY
which bans stress on the initial syllable. The failureto generatethe mirror
image counterpartto the Pintupipatternis a virtueof the constraintset, as
it is unattestedto the best of my knowledge.
Pattern 17, even-numberedfrom right to left plus initial, occurs in
languages with the 'initial-dactyl' effect (see Prince 1983; Hayes 1995
for discussion), e.g., Indonesian,Spanish. This patterndiffers minimally
from pattern 19, which is also attested (e.g., in Garawa),in preserving
the stress on the penult ratherthan the initial syllable in clash contexts
in three-syllablewords. Most of the unattestedpatternsclosely resemble
attestedpatterns,differingminimallyalong a small set of dimensions:the
directionalityof stress assignment,the anchoringsite for the stress fixed at
the oppositeedge from the originationpoint of the alternatingstresscount,
and the resolutionof stress clashes.
It is interesting to note that some binary plus lapse systems display
stressclashes in relativelyshortwords.For example, stresspatterns16 and
22 have stress clashes in disyllabic words and patterns15 and 18 display
clashes in trisyllabic words. The clash in disyllabic words in patterns16
534
MATTHEW
GORDON
and 22 arise when ALIGN EDGES is ranked above *CLASH which in turn
is rankedabove *LAPSE.This rankinggives rise to stress clashes but only
in disyllabic words where ALIGN EDGES can otherwise not be satisfied.
A similar situationcan arise in trisyllabic words in a language (patterns
15 and 18) in which *LAPSE RIGHT and NONFINALITY are undominated
and cruciallyrankedabove ALIGN EDGES which in turnis rankedabove
*CLASH. Under this scenario, the initial syllable and the penult will always be stressed even if this entails a stress clash, as arises in trisyllabic
words. This subset of binary plus lapse systems involving stress clashes
in either disyllabic or trisyllabic words might be regardedas pathologic,
since they representan unattestedtype of hybridsystem with eitherclashes
and lapses dependingon the length of the word. Similarternarysystems
are also generatedby the proposedconstraints(see section 4.1). There is,
however, at least a possibility that this gap is an accidental one related
to the independentrarityof quantity-insensitivestress systems which are
tolerantof clashes, following discussion in section 2.2.2. I leave this an
open matterfor futureresearch.
In all of the binaryplus lapse systems featuringan internallapse, the
fixed stress situatedat the opposite edge of originationpoint of the binary count falls on one of four syllables: the initial syllable, the peninitial
syllable, the final syllable, and the penult. Of these four docking sites generated in binary plus lapse systems, two are attested:the initial syllable
(patterns17, 19) and the penultimatesyllable (1 1). Whetherthe absence
of binaryplus lapse systems with a fixed stress on the final or peninitial
syllable reflects an accidentalgap or not is unclear,thoughthe accidental
gap interpretationseems reasonablegiven the small numberof binaryplus
lapse systems observedcross-linguistically.
If the atomistic alternativeto ALIGN EDGES were adopted, an additional threeunattestedbinaryplus lapse patternswould be generated.One
of these patternsis nearly identical to pattern 14, differing only in the
preservationof the stress on the final syllable ratherthanthe initialsyllable
in disyllables: [01], [101], [1001], [10101], [101001], [1010101]. This
patternresults from a combinationof an undominatedconstraintrequiring stress at the right edge in conjunctionwith an undominated*CLASH.
The rankingALIGN (X1, L) >> ALIGN (X1, R) produces a stress lapse
immediatelypreceding the final stress in even parity words. This pattern
is not generatedby the unitaryALIGN EDGES constraint,which necessarily conflicts with *CLASH in disyllables. If ALIGN EDGES outranks
*CLASH, both syllables are stressed in disyllabic words, yielding pattern
16 in Table VII. If *CLASH outranks ALIGN EDGES (assuming ALIGN
(X1, L) ?>ALIGN (X1, R), pattern 14 results. Two additionalunattested
A FACTORIAL
TYPOLOGY
STRESS
535
binaryplus lapse systems generatedby separatingSTRESSEDGESinto individualconstraintsreferringto differentedges display stress on the initial
and penultimatesyllable with a binarypatternoriginatingfrom either the
initial or penultimatesyllable, dependingon the relativerankingof ALIGN
(XI, R) and ALIGN(Xl, L). These patternsdiffer from patterns 12 and
19 (attested in Garawa), generated by STRESS EDGES, in having initial
and final stress in trisyllabicwords ratherthanonly initial stress, i.e. [101]

ratherthan [100]. The two novel systems involve an undominated*CLASH
and *LAPSE RIGHT, in addition to an undominated decomposed version
of ALIGN EDGES requiring stress on the leftmost syllable. NONFINALITY
is also highly rankedbut is violated in trisyllabicwords in orderto satisfy

both *CLASHand *LAPSERIGHT.
3.4.4. PrimaryStress Placementin Binary Systems

Before concluding discussion of binary stress, there are a couple of interesting points about primary stress which are apparentin Table VII.
First, parallel to dual stress systems, in virtually all binary stress languages, primarystressfalls a consistentdistancefromthe wordedge across
words differing in numberof syllables.41A second observationgermane
to primary stress is that there exist several patternsin which the location of primarystress in words with heavy syllables is inconsistent with
the location of primarystress in words containingonly light syllables. It
may be noted that many subtypes of purely quantity-sensitivestress are
inherentlyincapableof positioningprimarystressa uniformdistancefrom
a word edge while simultaneously stressing heavy syllables regardless
of their position. Thus, the quantity-sensitiveanalog to the Araucanian
patternattested in Creek (Haas 1977; Tyhurst 1987), but with the final
stress as primary,positions its leftmost stress on the initial syllable if it is
heavy, otherwise on the peninitial syllable. Thus, promotingthe leftmost
stress to primarystress would not ensure consistency in the location of
primarystress across different words, since words with an initial heavy
syllable would have primarystress on the initial syllable while those with
an initial light would place primarystress on the second syllable: io...
41 Malakmalak(Birk 1976) appearsto
be an atypical quantity-insensitivelanguage in
which primarystress does not fall a consistent distance from a word edge across words
differingin length. Stress is reportedto fall on even numberedsyllables countingfrom the
right, with the leftmost stress being the main one, e.g., [102020], [0102020]. Given the
preferencefor uniformityin stress placement, one would expect the rightmoststress, the
one falling closest to the edge at which the binary patternoriginates(i.e., the one on the
penult) to be the primarystress. Malakmalakappearsto be the only quantity-insensitive
stress system which violates the preference for positioning primary stress in the same
position across the entire vocabularyof words with phonologicallypredictablestress.
536
MATTHEWGORDON
but .6... A similar situationobtains in quantity-sensitivesystems which

stress heavy syllables and even-numberedsyllables going from right-toleft. It is only in quantity-sensitivesystems which stress odd-numbered
syllables from either left-to-rightor right-to-left,i.e., 'peak first' systems
in Prince's(1983) grid-based framework,or those with an additionalfixed
stress at a word edge, that it is possible to position primarystress a uniform distancefrom a wordedge acrossthe entirevocabulary.Interestingly,
peak-firstquantity-sensitivesystems which do not position primarystress
a uniform distance from a word edge appearto be rare; possible cases
include EgyptianRadio Arabic (Harrell1960) and PalestinianArabic(see
Hayes 1995 andreferencescited therein).Giventhe paucityof such cases it
is difficultto stateat the presentwhetherquantity-insensitiveandquantitysensitive systems truly differ in their strengthof preferencefor placing
primarystress a consistent distance from the word edge across words of
varying lengths. Hopefully, furtherresearchwill shed more light on this
issue.
4. TERNARY STRESS
There are a very small numberof languages with ternarystress patterns,

in which stress falls on every thirdsyllable. Probablythe clearestexample
of a quantity-insensitiveternarysystem comes from Cayuvava(Key 1961,
1967), in which primarystressfalls on the antepenultand secondarystress
falls on every third syllable before the antepenult.Examplesof Cayuvava
stress appearin (33). (Note that sequences of vowels, includingidentical
vowels, are heterosyllabic.)
(33)
a.
b.
c.
d.
e.
f.
g.
h.
i.
Cayuvavastress
'tail'
'stomach'
'I ran'
'he came already'
a.ri.ui.u.tfa
d3L.hi.ra.ri.a.ma
'I must do'
ma.ra.ha.ha.e.i.ki
'theirblankets'
i.ki.ta.pa.re.re.pe.ha
'the wateris clean'
tiA.a.di.r6.bo.3u.ru'.tfe 'ninety-nine(firstdigit)'
'fifteeneach (second digit)'
me.d'a.ru.tfe.tfe.i.ro.hi.i.jie
e.pje
fi.ka.he
ki.hi.be.re
The analysis of the Cayuvavapattern,and more generally, the analysis

of ternaritypursuedhere follows the basic insight of Elenbaasand Kager
A FACTORIAL
TYPOLOGY
STRESS
537
(1999), who attributeternary intervals to the ranking of an anti-lapse

constraint requiring that every stressless syllable be adjacent to either
a stressed syllable or a word edge above a foot-based alignment constraint,which in turnis rankedabove a constraintrequiringthat syllables
be parsed into feet. In the present work, ternary stress results from a
lowly ranked*LAPSE but an undominated*EXTENDEDLAPSE. The low
ranking of *LAPSE ensures that stress lapses entailing two consecutive
stressless syllables are freely tolerated, yet the undominatedstatus of
*EXTENDEDLAPSE guaranteesthat stress lapses do not exceed two syllables. *EXTENDED LAPSE is ranked above all the ALIGN constraints
referringto level 1 grid marks,as any pressureto push stresses towardan
edge is supersededby the prohibitionagainst stress lapses of greaterthan
two syllables. Variationin the rankingof the ALIGN constraintsaccounts
for different sub-types of ternarystress. In Cayuvava,in which the ternary patterninitiates with the antepenult,we have the rankingALIGN (X1,
L) >> ALIGN (X1, R) >> ALIGN EDGES. The rankingof ALIGN (X1, R)
above ALIGN EDGES is decisive in wordsin which the initial syllable goes
unstressed;thus, ariuutfa, ratherthan *ariuiutfa.The rankingof ALIGN
(X1, L) above ALIGN (X1, R) plays a crucialrole in determiningthe location of the stress in candidatesincurringequal violations of *LAPSE;thus,
kihi'bere,ratherthan *kihibere.The promotionof the rightmoststress to
primarystress is the result of ALIGN (X2, R) being rankedabove ALIGN
(X2, L). Summaryrankingsfor Cayuvavaappearin (34), followed by a
representativetableauin (35).
(34)
Constraintrankingsfor Cayuvava
*EXTENDED LAPSE, *CLASH, NONFINALITY, ALIGN (X2, R)
4
ALIGN (X1, L)
4
ALIGN (Xi, R)
4
ALIGN EDGES, *LAPSE, ALIGN (X2, L)
538
GORDON
MATTHE-W
(35)
kihibere
ET NON-: ALIGN
Lpsr-'FINAL: (X2, R)
sw
kihibere
kihibe're
____
0WIhbere
__
*CLSH
ALIGN
(x1,L)
*!
__
re
keihi-b
ariuutfa
ow marAhaha.61iki
mnakahaha.6.i.ki
mnarAhaha.,.i.ki
**
.
.....N....**.**.
*CLSH~ ALIGN
*ETM-ALiGN
LPSE FINAL: (X2
a.ri.d.u.tfa__
marahahaeiki
**
kihibere
kThiber6
ALIGN ALIGN *LpsE: ALIGN

(x1I, ) EDGEs:
(X2, L)
*
*
(X1,44
__
___
*EXT: NON-: ALIGN:*CLSH ALIGN

LFSE :FINAL:(X2,R):
(X1,L)
___
ALIGN ALIGN *LPSEI ALIGN

(x,4
(xl, R) EIxEs
__
ALIGN. ALIGN *LPSE ALIGN

EDGES
(X2, L4
(x1,R)
7__
4!_________
4M,
The descriptionof Joway-Otoin Whitman (1947) also suggests a tern

ary stress system following the primary stress, though the author does
not provide examples illustratingthe pattern.There are a few quantitysensitive stress systems which also display ternarityundercertainweight
and/or morphological conditions, e.g., Pacific Yupik (Leer 1985; Rice
1992), Sentani(Elenbaas1999; Elenbaasand Kager 1999),42 Finnish

(Sadeniemi 1949; Hansonand Kiparsky1996), the varietyof Estoniandescribedby Hint (1973),43and Winnebago(see Hayes 1995 and references
cited thereinfor discussion).
4.1. A Factorial Typologyof TernaryStress
We now turnto the ternarysystems generatedby the factorialtypology.Because of the rarityof ternarystress,the systems generatedby the proposed
constraintswill be only briefly summarizedhere.
42 Sentani is sensitive to a three-way weight distinction with closed syllables and syllables containing diphthongsbeing heaviest, schwa in open syllables being lightest, and
non-schwa vowels in open syllables being intermediatein weight (see Elenbaas 1999
for discussion and analysis). The Sentani stress patternfound in words not containing
heavy syllables (i.e. words lacking diphthongs and closed syllables) is not generatedby
the quantity-insensitiveconstraintsdiscussed in this paper;ratherit requiresan additional
context-sensitive weight constraintbanning stress on light (i.e. insufficiently sonorous)
initial syllables due to tonal crowding avoidance (see Gordon 2001, in preparationfor
weight constraintsoperativeat the periphery).
43 It should also be noted thatthe Estonianstresspatterndescribedby Hint is the subject
of controversy;Eek (1982) reportsa differentpatternnot involving ternarity.
539
A total of 33 ternarysystems were generated,all of which aregenerated

in pairs differing in the location of primarystress.44 Given the extreme
rarity of ternarystress in general, it is not surprisingthat only a small
minorityof the generatedpatternsactuallyexists. The only clear quantityinsensitive ternary pattern, the Cayuvava system, was generated. The
Ioway-Otopatterninferredfrom descriptionswas also generated,as were
three quantity-sensitivestress patterns observed in words consisting of
only light syllables: Estonian,Pacific Yupik,and Winnebago.This leaves
a total of 28 generatedbut unattestedpatterns(31 if quantity-sensitive
patternsare excluded).
Four pure ternarysystems were generated:ternaryfrom left starting
with the third syllable, ternaryfrom left startingwith the peninitial syllable, ternaryfromrightstartingwith the antepenult,andternaryfromright
startingwith the penult.45A total of nine ternaryplus clash systems were
also generated.These systems are the ternaryanalogs of the binaryplus
clash patterns:ternarystress iteratesin one directionand there is a fixed
stress at or nearthe opposite edge of the word. In certainword shapes,the
combinationof ternarystress and the fixed stress trigger a stress clash.
While I am not aware of any ternaryplus clash systems in actual languages, this mismatchbetween generatedand attestedpatternsis plausibly
an accidentalgap attributedto the confluence of two independentlyrare
phenomena:stress systems with clashes and ternarystressin general.
The final type of ternarystress system generatedby the factorialtypology entails ternaritymixed with binarystress intervalsin certainword
shapes.All of these systems have the propertyof displayingmore thanone
ternaryintervalin at least one word type of sufficientlength to diagnose
iterativeternarity.These systems thus differ from the binary plus lapse
systems which have a maximumof one ternaryintervalin any one word.
The mixed ternarysystems are similar to the ternaryplus clash systems
in displaying ternarityin conjunctionwith a fixed stress at the opposite
44 The number of generatedpatternsincreases to 42 if ALIGN EDGES is broken into
separateconstraintseach referringto a differentedge. Of the nine additionalpatterns,two
involvepureternaritystartingfrom a peripheralsyllable (left-to-rightternaritystartingwith
the initial syllable and right-to-leftternaritystartingwith the final syllable) and 7 involve
ternaritymixed with binarystress intervalsin certaincontexts.
45 Pureternarypatternsdifferingfrom those generatedin initiatinga ternarycount from
an edge syllable, e.g., a hypotheticalpatterninvolvingternarystressiteratingfrom the right
edge andbeginningwith the final syllable, are not generated,since, in certainword shapes,
pure ternarycandidateswith stresses distancedfrom the peripheryhave fewer stressesand
thus bettersatisfy ALIGN (X1, L) and ALIGN (X1, R); for example, aaeacda' incursfewer
violations of both ALIGN (X1, L) and ALIGN (X1, R) than a'cravcr6. Crucially,because
ALIGN EDGES is lowly rankedin pure ternarysystems, it is unable to ensure that an edge
syllable is stressed.
540
MATTHEW
GORDON
edge of the word from the originationpoint of the ternarycount. Three

of these patternsare the ternaryanalogs to similar binaryplus lapse patternsinvolving clashes in shortwords (see section 3.4.3). In addition,three
left-to-rightternarypatternsoriginatetheir stress counts closer to the left
edge of the word in relativelyshort words in orderto avoid violations of
undominatedNONFINALITY. Such patternsare analogousto patternsseen
in single stress languages (e.g., Hopi = pattern3 in Table IV) and binary
stress languages (e.g., SouthernPaiute= pattern6 in TableVII).
A total of 20 mixed ternary systems were generated, including the
systems found in Pacific Yupik(ternaryfrom left startingwith peninitial
syllable plus a binaryintervalat the rightedge), Winnebago(ternaryfrom
left startingwith thirdsyllable plus a binaryintervalat the rightedge), and
Estonian(ternaryfrom left plus penultimatestress in non-clashcontexts).
5. FACTORIAL TYPOLOGY: A SUMMARY
In summary,the factorialtypology generatedby the proposed constraint

set offers sufficient empirical coverage of the quantity-insensitivesingle
stress, dual stress, binaryand ternarystress systems found in an extensive
survey.While the thoroughnessof the empiricalcoverage of the proposed
theorycomes at the price of a certaindegree of overgeneration,the amount
of overgenerationis relatively small, at least relative to the logically possible set of stress systems. Furthermore,the overgenerationcommittedby
the presentanalysis, with some possible exceptions (e.g., mixed clash and
lapse systems; see discussion in section 3.4.3), is plausibly attributedto
accidentalgaps resulting from the overall rarityof certaintypes of stress
systems. This is particularlyapparentin the case of ternarysystems which
are a vanishinglyrarebreedcross-linguistically.Thus, virtuallyany ternary
stress system generatedby the theory is certain not to occur just given
the paucity of ternarystress in general. Similar cases of overgeneration
which could plausiblybe attributedto accidentalgaps are observedin dual
stress and binary stress systems, where most of the unattestedbut generatedpatternsinvolvedelements which are attestedin isolation or in certain
combinationsbut happennot to exist in all of the combinationsgenerated
by the factorial typology. Although the present work can neither be regardedas a complete metricalstresstheory,as such a theorywould require
constraintsdesigned to also handle quantity-sensitivestress (see Gordon
to appear,in preparationfor analysis of quantity-sensitivestress), it nevertheless suggests that OptimalityTheory offers a promisingframeworkfor
expressionof grid-basedmetricaltheories.
A FACTORIAL
TYPOLOGY
STRESS
541
APPENDIX: LIST OF LANGUAGES WITH QUANTITY-INSENSITIVE

STRESS
Stress pattern
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial (of root)
Initial
Initial
Initial
Initial
Initial
Language
Family46
Afrikaans
Indo-European
Arabana-WangkanguruAustralian
Arabela
Zaparoan
Arawak
Arawakan
Sino-Tibetan
Chepang
Chitimacha
Gulf
Chutiya
Sino-Tibetan
Comox
Salish
Danish
Indo-European
Diola
Niger-Congo
Dizi
Afro-Asiatic
DjapuYolngu
Australian
Enets
Uralic
Even (Lamut)
Altaic
Gilyak
Isolate
Gondi
Dravidian
Gosiute Shoshone
Uto-Aztecan
Gurung
Sino-Tibetan
Hanty
Uralic
Hewa
Sepik-Ramu
Hualapai
Hokan
Huitoto
Witotoan
Irish
Indo-European
Jemez
Kiowa-Tanoan
Kalkatungu
Australian
Kambera
Austronesian
Kanauri
Sino-Tibetan
Ket
Yenesei Ostyak
Kolami
Dravidian
Korafe
Tran-NewGuinea
Kota
Dravidian
Latvian
Indo-European
Naga
Creole
Nenets
Uralic
Olo
Austronesian
Papago
Uto-Aztecan
ParintintinTenharim Tupi
Pomo, Eastern
Hokan
Sami, Eastern
Uralic
Sango
Creole
Santali
Austro-Asiatic
46 Genetic affiliationsare according to the 14th edition (2001) of the SIL Ethnologue
(CD version) edited by BarbaraF. Grimes.
542
MATTHEW
GORDON
Stress pattern
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial
Initial(mora)
Initial(nouns), final of root (verbs),
Initial(nouns), verbs-??
Initial (root)
Initial (root)
Initial (root)
Initial (root)
Peninitial
Peninitial
Peninitial
Peninitial
Peninitial
Peninitial
Peninitial
Peninitial
Peninitial(of root)
Peninitial
Antepenult
Antepenult
Antepenult
Antepenult
Antepenult
Antepenult
Antepenult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Language
Senoufo
Siona
Sumbanese
Swedish
Tigak
Tinrin
Tunica
Wembawemba
Yurok
Nama
Dumi
Pawnee
Coreguaje
Kung (Zu1'Haasi)
Mixe, Totontepec
Tewa
Assiniboine
Basque
Ignaciano
Koryak
Lakota
Lezgian
Paraujano
Zan
Siroi
Tolai
Cayubaba
Cora
Kela
Macedonian
Mae
Parnkalla
Wappo
Alawa
Albanian
Amara
Andamanese
Anem
Arapesh
Atchin
Bukiyip
Chamorro
Chickaranga
Chumash,
Barbareno
Cocama
Cofan
Family
Niger-Congo
Tucanoan
Austronesian
Indo-European
Austronesian
Austronesian
Gulf
Australian
Algic
Khoisan
Sino-Tibetan
Caddoan
Tucanoan
Khoisan
Mixe-Zoque
Kiowa-Tanoan
Siouan
Isolate
Arawakan
Chukotko-Kamchatkan
Siouan
North Caucasian
Arawakan
South Caucasian
Trans-NewGuinea
Austronesian
Isolate
Uto-Aztecan
Austronesian
Indo-European
Austronesian
Australian
Yuki
Australian
Indo-European
Austronesian
Andamanese
East Papuan
Torricelli
Austronesian
Torricelli
Austronesian
Niger-Congo
Hokan
Tupi
Chibchan
Stresspattern
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult (of phrase)
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult
Penult (nouns), final (verbs)
Final
Final
Final
Final
Final
Final
Language
Dayak
Djingili
Jaqaru
Kaliai-Kove
Kola
Kutenai
Kwaio
Labu
Lamba
Laz
Lingala
Lusi
Mohawk
Monumba
Movima
Mussau
Nahuatl
Onondaga
Paez
Pagu
Paiwan
Quicha
Quileute
Rapanui
Shona
Siriono
Solor
SouthernSotho
Tacana
Temate
Tetun
Tiwi
Tolo
Tolojolabal
Tonkawa
Tswana
Tuscarora
Usan
Wardaman
Wikchamni
Gurage
Abun
Alabama
Apinaye
Aramaic
Atayal
Azerbaijani
543
Family
Austronesian
Australian
Aymara
Austronesian
Austronesian
Isolate
Austronesian
Austronesian
Niger-Congo
South Caucasian
Niger-Congo
Austronesian
Iroquoian
Niger-Congo
Unclassified
Austronesian
Uto-Aztecan
Iroquoian
Paezan
West Papuan
Austronesian
Quechan
Chimakuan
Austronesian
Niger-Congo
Tupi
Austronesian
Niger-Congo
Tacanan
WestPapuan
Austronesian
Australian
Austronesian
Mayan
Coahuiltecan
Niger-Congo
Iroquoian
TransNew Guinea
Australian
Penutian
Afro-Asiatic
WestPapuan
Muskogean
Macro-Ge
Afro-Asiatic
Austronesian
Altaic
544
MATTHEW
GORDON
Stress pattern
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final (of root)
Final
Final
Final
Final
Final
Final
Final (of root)
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final
Final (of phrase)
Final (of phrase)
Final
Language
Bashkir
Cakchiquel
Canela-Krah6
Dani
Gagauz
GreenlandicInuktitut
Guarani
HaitianCreole
Hebrew
Than
Ivatan
Kabardian
Kavalan
Kayabi
Kayap6
Kazakh
Konkani
Kurdish
Lango
Maricopa
Mazatec
Moghol
Nanai
Pemon
Persian
Sa'ban
Semai
Stieng
Tajiki
TamazightBerber
Tatar
Temiar
Thai
Tolowa
Tsotsil
Turkmen
Tzutujil
Udmurt
Uighur
Uzbek
Waiwai
Yagua
Yakut
Yuchi
French,European
Wari'
Diegueno
Family
Altaic
Mayan
Macro-Ge
TransNew Guinea
Altaic
Eskimro-Aleut
Tupi
Creole
Afro-Asiatic
Austronesian
Austronesian
North Caucasian
Austronesian
Tupi
Macro-Ge
Altaic
Indo-European
Indo-European
Nilo-Saharan
Hokan
Oto-Manguean
Altaic
Altaic
Carib
Indo-European
Austronesian
Austro-Asiatic
Austro-Asiatic
Indo-European
Afro-Asiatic
Altaic
Austro-Asiatic
Daic
Na Dene
Mayan
Altaic
Penutian
Uralic
Altaic
Altaic
Carib
Peba-Yaguan
Altaic
Isolate
Indo-European
Chapacura-Wanham
Hokan
Stresspattern
Final (of root)
Final (of root)
Final (of root)
Final (of root)
Final (of root)
Final
Final (lary) & initial (secondary)
Antepenult(lary) & initial (secondary)
Initial (lary) & antepenult(secondary)
Initial (lary) & penult (secondary)
Penult (lary) & initial (secondary)
Even numberedfrom L
Even-numberedfrom L
Even numberedfrom L
Even numberedfrom R
Even numberedfrom R
Even numberedfrom R
Even numberedfrom R
Even-numberedfrom R (root)
Even numberedfrom R
Even numberedfrom R
Even numberedfrom R
Even numberedfrom R
Even numberedfrom R
Even-numberedfrom R
Even numberedfrom R
Odd numberedfrom R
Odd-numberedfrom R
Odd numberedfrom R
Odd-numberedfrom R
Odd-numberedfrom R
Odd-numberedfrom L
Odd numberedfrom L
(with optionalnonfinality)
Odd numberedfrom L
Odd numberedfrom L
Odd numberedfrom L
Odd numberedfrom L
545
Language
Paipai
Shilha
Waskia
Yavapai
Zapotec,Mitla
Zazaki
Armenian
French,Canadian
Udihe
Georgian
Walmatjari
Gugu Yalanji
Lower Sorbian
Watjarri
Anyula
Awtuw
ChimalapaZoque
Murut
Sanuma
SibutuSama
Araucanian
Hatam
Yupik,Sirenik
Anejom
Berbice
Cavinefna
Ese Ejja
Larike
Malakmalak
Nengone
Orokolo
To'aba'ita
TukangBesi
Ura
Warao
Asmat
Chulupi
Kamayur
UrubuKaapor
Weri
Bagandji
Burum
Family
Hokan
Afro-Asiatic
TransNew Guinea
Hokan
Oto-Manguean
Indo-European
Indo-European
Indo-European
Altaic
South Caucasian
Australian
Australian
Indo-European
Australian
Australian
Sepik-Ramu
Mixe-Zoque
Austronesian
Yanomam
Austronesian
Araucanian
West Papuan
Eskimo-Aleut
Austronesian
Creole
Tacanan
Tacanan
Austronesian
Australian
Austronesian
TransNew Guinea
Austronesian
Austronesian
Austronesian
Isolate
Trans-NewGuinea
Mataco-Guaicuru
Tupi
Tupi
Trans-NewGuinea
Australian
Trans-NewGuinea
Czech
Hungarian
Icelandic
Livonian
Indo-European
Uralic
Indo-European
Uralic
546
MATTHEW
GORDON
Stresspattem
Odd numberedfrom L
Odd numberedfrom L
Odd numberedfrom L
Odd numberedfrom L
Odd numberedfrom L
Odd-numberedfrom L
Odd-numberedfrom L
Odd-numberedfrom L
Odd numberedfrom L
Odd numberedfrom L
Odd numberedfrom L
Odd numberedfrom L (roots);even
numberedfrom R (suffixes)
Odd numberedfrom L minus final
Odd-numberedfrom L minus final
Odd numbered from L minus final (of
root)
Even numberedfrom L & penult
Evenloddnumbered(lexical) from
R & initial
Even-numberedfrom R & initial
Odd numberedfrom L & penult
Odd numberedfrom L & final
Odd-numberedfrom R & initial
Penult & odd-numberedfrom R/L
Temaryfrom R, Rmost on antepenult
Temaryfrom L startingwith
initial/peninitial
Language
Mansi
Maranungku
Murinbata
Ningil
Ono
Panamint
Sami, Northem
Family
Uralic
Australian
Australian
Torricelli
Trans-NewGuiena
Uto-Aztecan
Uralic
Selepet
Trans-NewGuinea
Sinaugoro
Timucua
Votic
Waorani
Austronesian
Arawakan
Uralic
Unclassified
Anguthimri
Badimaya
Bidyara/Gungabula
Dalabon
Dehu
Diyari
Karelian
Kate
Pintupi
PittaPitta
TenangoOtomi
Australian
Australian
Australian
Australian
Austronesian
Australian
Uralic
Trans-NewGuinea
Australian
Australian
Oto-Manguean
Wangkumara
Wirangu
Yingkarta
SouthemPaiute
Biangai
Australian
Australian
Australian
Uto-Aztecan
Trans-NewGuinea
Garawa
Piro
Gosiute Shoshone
Tauya
Polish
Cayuvava
loway-Oto
Australian
Arawakan
Uto-Aztecan
TransNew Guinea
Polish
Isolate
Siouan
A FACTORIAL
STRESS
TYPOLOGY
547
REFERENCES
Alber, Birgit. 1997. 'QuantitySensitivityas the Result of ConstraintInteraction',in Geert

Booij and Jeroen van de Weijer(eds.), Phonology in Progress-Progressin Phonology,
HILPhonology PapersIII, HollandAcademic Graphics,The Hague, pp. 1-45.
Alderete, John. 1999. Morphologically GovernedAccent in OptimalityTheory, unpublished Ph.D. dissertation,Universityof Massachusetts,Amherst.
Allison, E. J. 1979. 'The Phonology of Sibutu Sama: A Languageof the SouthernPhilippines', in C. Edrial-Luzaresand A. Hale (eds.), Studies in Philippine Linguistics, Vol.
3, No. 2, Linguistic Society of the Philippinesand SummerInstituteof Linguistics,pp.
63-104.
Aronson,Howard. 1991. Georgian:A Reading Grammar,Slavica, Columbus,OH.
Bakovic, Eric. 1996. 'Foot Harmonyand QuantitativeAdjustments',manuscript,Rutgers
University.Availableon RutgersOptimalityArchive,ROA-168,
http://ruccs.rutgers.edu/roa.html.
Bakovic, Eric. 1998. 'UnboundedStress and Factorial Typology', manuscript,Rutgers
University.Availableon RutgersOptimalityArchive, ROA-244,
Bendor-Samuel,J. T. 1963. 'Stress in Terena', Transactionsof the Philological Society
1962, 105-123.
Birk, D. B. W. 1976. The MalakmalakLanguage, Daly River (WesternArnhem Land),
AustralianNational University,Canberra.
Boas, Franz and Ella Deloria. 1933. 'Notes on the Dakota, Teton Dialect', International
Journal of AmericanLinguistics7, 97-121.
Boas, Franzand Ella Deloria. 1941. Dakota Grammar[Memoirsof the National Academy
of Sciences 33], United States GovernmentPrintingOffice, Washington,D.C.
Borgman,Donald. 1989. 'Sanuma',in Desmond Derbyshireand Geoffrey Pullum (eds.),
Handbookof AmazonianLanguages,Vol. 2, Mouton,New York,pp. 15-248.
Chafe, Wallace. 1977. 'Accentand Related Phenomenain the Five Nations IroquoisLanguages', in Larry Hyman (ed.), Studies in Stress and Accent, University of Southern
California,Departmentof Linguistics,Los Angeles, pp. 169-181.
Cohn, Abigail. 1989. 'Stressin Indonesianand BracketingParadoxes',Natural Language
and LinguisticTheory7, 167-216.
Crowhurst,Megan and MarkHewitt. 1994. 'DirectionalFooting, Degeneracy,and Alignment', manuscript,University of North Carolina and University of British Columbia.
Availableon RutgersOptimalityArchive, ROA-65, http://ruccs.rutgers.edu/roa.html.
Davies, H. J. 1980. KobonPhonology, AustralianNationalUniversity,Canberra.
Ddcsy, Gyula. 1966. YurakChrestomathy,IndianaUniversityPress, Bloomington.
Douglas, Wilfrid. 1981. 'Watjarri',in RobertDixon and BarryBlake (eds.), Handbookof
AustralianLanguages,Vol. 2, John Benjamins,Amsterdam,pp. 196-272.
Dubert,Raymondand Marorie Dubert. 1973. 'BiangaiPhonemes', Phonologies of Three
Languages of Papua New Guinea, SummerInstituteof Linguistics, Ukarumpa,Papua
New Guinea,pp. 5-36.
Echeverria, Max and Heles Contreras. 1965. 'AraucanianPhonemics', International
Journalof AmericanLinguistics31, 132-135.
Eek, Arvo. 1982. 'Stressand AssociatedPhenomena:A Surveywith Examplesfrom Estonian', Estonian Papers in Phonetics 1980-1981, Academy of Sciences of the Estonian
SSR, Instituteof LanguageandLiteratures,Tallinn,pp. 20-58.
548
MATTHEW
GORDON
Egerod, S0ren. 1966. 'A Statementon Atayal Phonology', ArtibusAsiae Supplementum

XXIII(Felicitationvolumefor the 75th birthdayof ProfessorG. H. Luce), ArtibusAsiae,
Ascona, Switzerland,pp. 120-130.
Eisner, Jason. 1997. 'FootFormDecomposed: Using PrimitiveConstraintsin OT', MIT
WorkingPapers in Linguistics. Available on Rutgers Optimality Archive, ROA-205,
Elenbaas, Nine. 1999. A Unified Account of Binary and TernaryStress, Netherlands
GraduateSchool of Linguistics,Utrecht.
Elenbaas, Nine and Rene Kager. 1999. 'Temary Rhythm and the Lapse Constraint',
Phonology 16, 273-329.
Everett, Daniel. 1986. 'Piraha', in Desmond Derbyshire and Geoffrey Pullum (eds.),
Handbookof AmazonianLanguages, Vol. 1, Mouton,New York,pp. 200-325.
Everett, Daniel. 1988. 'On Metrical ConstituentStructurein Piraha',Natural Language
and LinguisticTheory6, 207-246.
Everett,Daniel and KerenEverett. 1984. 'Syllable Onsets and StressPlacementin Piraha',
Proceedingsof the WestCoast Conferenceon FormalLinguistics3, 105-116.
Feldman,Harry.1986. A Grammarof Awtuw,AustralianNationalUniversity,Canberra.
Furby, Christine. 1974. Garawa Phonology, Canberra,AustralianNational University,
Canberra.
Gendron, Jean Denis. 1966. Tendances Phonetiques Francais Parle au Canada, Les
Presses de L'universiteLaval, Qudbec.
Gordon,Matthew.2001. 'The Tonal Basis of Final Weight Criteria',Chicago Linguistics
Society 36, 141-156.
Gordon, Matthew. To appear. 'Syllable Weight', in Bruce Hayes, Robert Kirchner,
and Donca Steriade (eds.), Phonetic Bases for Phonological Markedness,Cambridge
UniversityPress.
Gordon, Matthew.In preparation.'A Grid-basedOptimality-TheoreticConstraintSet for
Stress', manuscript,Universityof California,SantaBarbara.
Green, Thomas and Michael Kenstowicz. 1996. 'The Lapse Constraint',Proceedings of
the 6th AnnualMeeting of the FormalLinguisticsSociety of the Midwest,pp. 1-15.
Haas, Mary. 1977. 'TonalAccent in Creek', in LarryHyman (ed.), Studies in Stress and
Accent[SouthernCaliforniaOccasional Papersin Linguistics4], Universityof Southern
California,Departmentof Linguistics, Los Angeles, pp. 195-208.
Halle, MorrisandJean-RogerVergnaud.1987. An Essay on Stress,MIT Press, Cambridge,
MA.
Hammond, Michael. 1985. 'Main Stress and Parallel Metrical Planes', Berkeley Linguistics Society 11: Parasessionon Poetics, Metrics, and Prosody,Berkeley Linguistics
Society, Berkeley,pp. 417-428.
Hansen, Kenneth and L. E. Hansen. 1969. 'Pintupi Phonology', Oceanic Linguistics 8,
153-170.
Hansen, Kenneth and L. E. Hansen. 1978. The Core of Pintupi Grammar,Institute for
AboriginalDevelopment,Alice Springs,Australia.
Hanson, Kristin and Paul Kiparsky. 1996. 'A ParametricTheory of Poetic Meter',
Language72, 287-335.
Harden,Margaret.1946. 'Syllable Structureof Terena',InternationalJournalof American
Linguistics12, 60-63.
Hardman-de-Bautista,
MarthaJames. 1966. Jaqaru:Outlineof Phonological and Morphological Structure,Mouton,The Hague.
A FACTORIAL
TYPOLOGY
STRESS
549
Harms,Robert.1966. 'Stress,Voice, andLengthin SouthernPaiute',InternationalJournal

of AmericanLinguistics32, 228-235.
Harrell,Richard.1960. 'A Linguistic Analysis of EgyptianRadio Arabic', in CharlesFerguson (ed.), Contributionsto Arabic Linguistics,HarvardUniversityPress, Cambridge,
MA, pp. 3-77.
Harris, James. 1983. Syllable Structureand Stress in Spanish: A Nonlinear Analysis
[LinguisticInquiryMonograph8], MIr Press, Cambridge,MA.
Hayes, Bruce. 1980. A Metrical Theoryof StressRules, Ph.D. dissertation,MIT.Published
in 1985 by GarlandPress, New York.
Hayes, Bruce. 1985. 'Iambicand TrochaicRhythmin Stress Rules', BerkeleyLinguistics
Society 11: Parasessionon Poetics, Metrics, and Prosody,BerkeleyLinguisticsSociety,
Berkeley,pp. 429-446.
Hayes, Bruce. 1995. Metrical Stress Theory:Principles and Case Studies, University of
Chicago Press, Chicago.
Hayes, Bruce and StanislawPuppel. 1985. 'On the RhythmRule in Polish', in Harryvan
der Hulst and Norval Smith (eds.), Advancesin NonlinearPhonology, Foris, Dordrecht,
pp. 59-8 1.
Hayes, Bruce, Bruce Tesar,and Kie Zuraw.2000. OTSoft,softwarepackage. Availableat
http://www.humnet.ucla.edu/humnet/linguistics/people/hayes/otsoft/index.htm.
Hetzer,Armin. 1978. Lehrbuchder vereinheitlichtenalbanischenSchriftsprachemit einem
deutsch-albanischenWorterbuch,Helmut Buske Verlag,Hamburg.
Hint, Mati. 1973. Eesti Keele SonafonoloogiaI, Eesti NSV TeadusteAkadeemia,Tallinn.
Hudson,Joyce. 1978. TheCore of WalmatjariGrammar,AustralianInstituteof Aboriginal
Studies, Canberra.
Hyman, Larry.1977. 'On the Natureof Linguistic Stress', in LarryHyman (ed.), Studies
in Stress and Accent, Universityof SouthernCalifornia,pp. 37-82.
Janas,Petr. 1984. NiedersorbischeGrammatikflrden Schul-gebrauch,Domowina-Verlag,
Bautzen.
Jeanne, Laverne. 1982. 'Some Phonological Rules of Hopi', InternationalJoumal of
AmericanLinguistics48, 245-270.
Kager, Rene. 1989. A Metrical Theoryof Stress and Destressing in English and Dutch,
Foris, Dordrecht.
Kager, Rend. 1993. 'Alternativesto the IAMB-TROCHEE
Law', Natural Language and
LinguisticTheory11, 381-432.
Kager, Rend. 1994. 'Ternary Rhythm in Alignment Theory', manuscript,
Utrecht University. Available on Rutgers Optimality Archive, ROA-35-1094;
Kager, Rend. 1997. 'Generalized Alignment and Morphological Parsing', Rivista di
Linguistica9, 245-282.
Kager,Rend. 1999. OptimalityTheory,CambridgeUniversityPress, Cambridge.
Kakumasu,James. 1986. 'Urubu-Kaapor',in Desmond Derbyshireand Geoffrey Pullum
(eds.), Handbookof AmazonianLanguages,Vol. 1, Mouton,New York,pp. 326-406.
Kenstowicz, Michael. 1994. 'Evidence for Metrical Constituency', in K. Hale and S. J.
Keyser (eds.), The Viewfrom Building 20: Essays in Linguistics in Honor of Sylvain
Bromberger,MIT Press, Cambridge,MA, pp. 257-273.
Kenstowicz,Michael. 1997. 'Quality-SensitiveStress', Rivistadi Linguistica9, 157-188.
Key, Harold. 1961. 'Phonotactics of Cayuvava', International Journal of American
Linguistics27, 143-150.
Key, Harold. 1967. Morphologyof Cayuvava,Mouton,The Hague.
550
MATTHEWGORDON
Key, Mary. 1968. ComparativeTacananPhonology with CavinefiaPhonology and Notes

on Pano-TacananRelationship,Mouton,The Hague.
Kirton,Jean. 1967. 'AnyulaPhonology', in D. Glasgow, K. Glasgow, J. Kirtonand W. J.
Oates (eds.), Papers in AustralianLinguistics,AustralianNationalUniversity,Canberra,
pp. 15-28.
Knudson,Lyle. 1975. 'A NaturalPhonology and Morphophonemicsof ChimalapaZoque',
Papers in Linguistics8, 283-346.
Kormushin,Igor Valentinovich.1998. Udykheiskii(Udegeiskii)Iazyk,Nauka,Moscow.
Kucera,Henry. 1961. ThePhonology of Czech,Mouton, 's-Gravenhage.
Leer,Jeff. 1985. 'Prosodyin Alutiiq', in Michael Krauss(ed.), YupikEskimoProsodic Systems:Descriptiveand ComparativeStudies,Alaska Native LanguageCenter,Fairbanks,
pp. 77-134.
Leskinen,Heikki. 1984. 'Uberdie Phonemsystemder KarelischenSprache',in PeterHajdu'
andLaszlo Honti (eds.), Studienzur PhonologischenBeschreibunguralischerSprachen,
AkademiaiKiad6, Budapest,pp. 247-257.
Leslau, Wolf. 1992. 'Outlineof GuragePhonology', in Wolf Leslau (ed.), GurageStudies:
CollectedArticles, Otto Harrassowitz,Wiesbaden,pp. 1-116.
Liberman,Mark and Alan Prince. 1977. 'On Stress and Linguistic Rhythm', Linguistic
Inquiry8, 249-336.
Lunt, Horace. 1952. A Grammarof the MacedonianLiteraryLanguage,n.p., Skopje.
Lynch, John. 1974. Lenakel Phonology, unpublished Ph.D. dissertation,University of
Hawaii.
Lynch,John. 1978. A Grammarof Lenakel,AustralianNationalUniversity,Canberra.
MacDonald,Loma. 1990. A Grammarof Tauya,Mouton de Gruyter,New York.
Matteson, Esther. 1965. The Piro (Arawakan)Language, University of CaliforniaPress,
Berkeley.
McCarthy,John and Alan Prince. 1993. 'GeneralizedAlignment', manuscript,University
of Massachusetts and Rutgers University. Available on Rutgers Optimality Archive,
ROA-7, http://ruccs.rutgers.edu/roa.html.
Menovshchikov,G. A. 1975. IazykNaukanskikhEskimosov,Nauka,Leningrad.
Miller,Wick. 1996. 'Sketchof Shoshone,a Uto-AztecanLanguage',in Ives Goddard(ed.),
Handbook of American Indian Languages, Vol. 17 Languages, SmithsonianInstitute,
Washington,D.C., pp. 693-720.
Myers, James. 1999. 'Lexical Phonology and the Lexicon', manuscript,GraduateInstitute of Linguistics, National Chung Cheng University, Taiwan. Available on Rutgers
OptimalityArchive,ROA-330,http://ruccs.rutgers.edu/roa.html.
Nikolaeva, Irna and MariaTolskaya.2001. A Grammarof Udihe, Mouton,New York.
LanguageStudies,
Oates, WilliamandLynetteOates. 1964. Gugu-Yalanjiand Wik-Munkan
AustralianInstituteof AboriginalStudies, Canberra.
Osborn,Henry. 1966. 'WaraoI: Phonology and Morphophonemics',InternationalJournal
of AmericanLinguistics32, 108-123.
Pike, Kenneth.1964. 'Stress Trainsin Auca', in David Abercrombie,Dennis Fry,P. A. D.
MacCarthy,N. C. Scott, andJ. L. M. Trim(eds.), In Honour of Daniel Jones, Longmans,
London,pp. 425-431.
Pike, Kennethand Eunice Pike. 1947. 'ImmediateConstituentsof Mazateco Syllables',
IntemationalJournalof AmericanLinguistics 13, 78-91.
Polotsky, Hans Jakob. 1951. Notes on Gurage Grammar,Israel OrientalSociety, Jerusalem.
A FACTORIAL
TYPOLOGY
STRESS
551
Prentice, D. J. 1971. The Murut Languages of Sabah, AustralianNational University

Canberra.
Prince,Alan. 1980. 'A MetricalTheoryfor EstonianQuantity',LinguisticInquiry11, 51 1562.
Prince,Alan. 1983. 'Relatingto the Grid', LinguisticInquiry14, 19-100.
Prince, Alan and Paul Smolensky. 1993. 'Optimality Theory: ConstraintInteractionin
GenerativeGrammar',manuscript,Rutgers University and University of Colorado at
Boulder.
Radin, Paul. 1929. A Grammarof the WappoLanguage, University of CaliforniaPress,
Berkeley.
Rice, Curtis. 1992. Binarity and Ternarityin Metrical Theory: ParametricExtensions,
unpublishedPh.D. dissertation,Universityof Texas, Austin.
Rich, Fume. 1963. 'ArabelaPhonemes and High-level Phonology', in Benjamin Elson
(ed.), Studies in Peruvian Indian Languages, Vol. I, Summer Instituteof Linguistics,
Norman,Oklahoma,pp. 193-206.
Sadeniemi,Matti. 1949. MetriikkammePerusteet,SKST, Helsinki.
Saelzer, Meinke. 1976. 'Fonologia Provis6riada Lingua Kamayura',Serie Lingidstica5,
131-170.
Sapir, Edward. 1930. SouthernPaiute: A Shoshonean Language, American Academy of
Arts and Sciences, Cambridge,MA.
Selkirk, Elisabeth. 1984. Phonology and Syntax: The Relation between Sound and
Structure,MIT Press, Cambridge,MA.
Skorik, Petr. 1961. GrammatikaChukotskogoIazyka, Izdatelstvo Akademii Nauk, Moscow.
Stell, Nelida Noemi. 1972. Fonologia de la LenguaAlulaj, Universidadde Buenos Aires,

Buenos Aires.
Swadesh, Morris. 1946. 'Chitimacha',in CorneliusOsgood (ed.), LinguisticStructuresof
Native America, Viking FundPublicationsin Anthropology,New York,pp. 312-336.
Tesar, Bruce and Paul Smolensky. 1993. 'The Learnabilityof Optimality Theory: An
Algorithm and Some Basic Complexity Results', Technical Report CU-CS-678-93,
University of Colorado, Boulder. Available on Rutgers Optimality Archive, ROA-2,
Tryon, Darrell. 1970. An Introductionto Maranungku,AustralianNational University,
Canberra.
Tyhurst,James. 1987. 'Accent Shift in Seminole Nouns', in P. Munro (ed.), Muskogean
Linguistics[UCLIAOccasional Papers in Linguistics6], UCLA LinguisticsDepartment,
Los Angeles, pp. 161-170.
van der Hulst, Harry.1984. Syllable Structureand Stress in Dutch, Foris, Dordrecht.
Vaux,Bert. 1998. ThePhonology ofArmenian, ClarendonPress, Oxford.
Walker, Rachel. 1996. 'Prominence-Driven Stress', manuscript, University of
California, Santa Cruz. Available on Rutgers Optimality Archive, ROA-172,
Weiers, Michael. 1972. Die Sprache der Moghol der Provinz Herat in Afghanistan,
WestdeutscherVerlag,Opladen.
Whitman,William. 1947. 'DescriptiveGrammarof loway-Oto', Intemational Journal of
AmericanLinguistics13(4), 233-248.
Zhgenti, Sergi. 1964. 'The Problem of Rhythmical Stress and Intonation Structureof
the GeorgianLanguage',Zeitschriftfur Phonetik,Sprachwissenschaftund Kommunikationsforschung17, 357-368.
552
MATTHEW
GORDON
Zhukova,Alevtina Nikodimovna. 1972. GrammatikaKoriakskogolazy/ka:Fonetika,Morfologiia, lzdatelstvo Nauka,Leningrad.

Received 23 August 2000
Revised 12 October2001
Departmentof Linguistics
Universityof California,
SantaBarbara,CA 93106
USA
<mgordon@linguistics.ucsb.edu>

Quantity Insensitive Stress

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Quantity Insensitive Stress

Hochgeladen von

Copyright:

Verfügbare Formate

A Factorial Typology of Quantity-Insensitive Stress

Author(s): Matthew Gordon

A FACTORIAL TYPOLOGY OF QUANTITY-INSENSITIVE STRESS *

ABSTRACT. This paper presents an Optimality-theoretic(Prince and Smolensky 1993)

One of the majorendeavorsin theoreticalphonology over the past quarter

Natural Language & LinguisticTheory 20: 491-552, 2002.

in what is intendedto be a fairly exhaustivesurveyof quantity-insensitive

Similar representationsare assumed throughoutthe rest of the paper

The most straightforwardsystem of phonologically predictablestress is

Coarsetaxonomy of stress systems

of logically possible docking sites for stress in quantity-insensitivesingle

Numberof single stress languages

3 Note thatpercentagesdo not add up to 100, as figuresare rounded.

ALIGN (Xjevein+j, tR/L}, level n, PrWd):Every grid mark of

grid marksin a prosodic word.

(Kager 1999; Elenbaasand Kager 1999), which requirethat words begin

will be shown to play an importantrole in languageswith both initial and

The threeALIGNconstraintsare formalizedin (3) 1

ALIGN (Xievel1, L, level 0, PrWd): Every grid mark of level 1 is

aligned with the left edge of level 0 of grid marksin a prosodic

alignedwith the rightedge of level 0 of gridmarksin a prosodic

of grid marksin a prosodic word are aligned with level 1 grid

specifically to primarystress. They are formulatedin (4).

Constraintsmodeling the attractionof primarystress to edges

Violationsof ALIGN (Xievel2, L, level 1, PrWd)andALIGN (Xlevel2, R, level

since one secondarystressedsyllable separatesthe primarystressfromthe

are several violations of a constraintcommittedby a form, the numberof

Evaluationof the ALIGNconstraints

Before proceeding with analyses employing the ALIGNconstraints, it

within OptimalityTheory) as well as phrase-level stress (see Gordon in

13 In his studyof CanadianFrench,Gendron(1966, p. 150) contrastsvarietiesof French

Initialand final stress

NONFINALITY:Stressdoes not fall on the finalsyllable. (A final

syllable does not have a level 1 grid mark.)15

(x1, R).16This rankingproducesthe 'rightmostwithoutbeing final' stress

It should be noted that the formulationof *LAPSE is differentfrom other

1986, 1988), Kobon (Davies 1980 in Kenstowicz 1997), Chukchi(Skorik

1961), thoughlapse avoidanceat edges also figuresprominentlyin certain

(No more than one syllable separates the rightmost syllable

*LAPSELEFT:A maximumof one unstressedsyllable separates

*EXTENDEDLAPSE RIGHT: A maximum of two unstressed

rightmostsyllable with a level 1 grid markfrom the rightedge.)

thatstressfalls as far to the left as possible withoutfalling to the left of the

Fixed peninitial stress is attributed to an analogous ranking except for

in turntakes precedenceover ALIGN(X1, L). The effect of this rankingis

dow at the left edge which would necessitateadoptionof the hypothetical

2.2.1. Dual Stress Systems

'we are persuading'

*CLASH: A stress domain does not contain adjacent stressed

instrumentalin accountingfor the stress on the penult;it is rankedabove

therebyensuringthat the penult takes primarystress in forms containing

*CLSH NON- *LAPSE ALICN

The Lower Sorbianpattern(Janas 1984), in which the initial stress rather

Constraint rankings for Lower Sorbian

Constraint rankings for Georgian

Walmatjari(Hudson 1978) displays a variantof the Georgianpattern,

The candidate forms consisting of two and three syllables in (18)

Candidateforms containingtwo and threesyllables

The number of possible stress patternsincreases commensuratelywith

syllables: 4 in disyllabic words, 12 in trisyllables, 32 in words with 4

Constraintsin the factorialtypology

10. *EXTENDED LAPSE RIGHT

11. *LAPSE LEFT

The only fixed ranking assumed involved constraints referring to

CLSH NON- LAPSE ALICN

CLASII LAPSLLEFT,*EXTLAPSERicHT >> AL (XI,

CLks:A LIGNt}tTLAPSE ALIGnN A I(.N

CLAsH, LAPSERI'HTs>>"AL EDS">"