Round licensing, harmony, and bisyllabic triggers in Altaic

52 Pages·0.491 MB·English

by  Walker Rachel.

RACHELWALKER
ROUNDLICENSING,HARMONY,ANDBISYLLABICTRIGGERS
INALTAIC(cid:1)
ABSTRACT. ManyAltaiclanguagesrestrictroundvoweldistribution.Thispaperexam-
inesroundharmonyinClassicalManchuandOroqen,whereroundspreadingoccursonly
when the first two syllables of a word are round, that is, it requires a bisyllabic trigger
(Zhang1996).Itisarguedthatthebinarythresholdemergesfromconflictbetweenwell-
establishedphonologicaldemands–numericreferenceisneithernecessarynordesirable.
The study isolates two distinct restrictions on rounding in bisyllabic trigger languages:
initial round licensing and round spreading – requirements occurring independently in
Classical Mongolian and Ulcha. Separating these restrictions is key: each is active in
languages withbisyllabic triggers, but they are ranked asymmetrically with respect to a
conflictingconstraintthatrestrictsfeaturestoatautosyllabicdomain.Rankingthetautosyl-
labicconstraintbetweenroundlicensingandspreadingpreventscross-syllablespreading
except when violations of tautosyllabicity are independently necessitated by round li-
censing. As a result, spreading is initiated only when the first two syllables are round.
Implicationsareidentifiedforthecharacterizationoffaithfulness.Positionalfaithfulness
constraintsplayakeyroleinrealizingtheprivilegedstatusoftheroot-initialsyllablein
roundlicensingandharmony.Inaddition,theanalysissupportstheseparationofIDENT(F)
intoIOandOIconstraints,whichdistinguishbetweenthelossandgainofprivativefea-
turespecifications,respectively. Thedistinctionprovesessentialinthecaseofbisyllabic
triggers. Theconstraint interactionthat produces thetwo-syllabletrigger threshold isan
instanceofageneralphenomenonexploredhere,termedParasiticConstraintSatisfaction.
Thiskindofinteractionariseswhentherearetwoconstraintsorconstraintsets,αandβ,
whosesatisfactioneachnecessitatesviolatingaconstraint,γ,andtheyarerankedα(cid:1)γ
(cid:1) β. When satisfaction of α compels violations of γ that also permit satisfaction of β,
thenβisdescribed asparasiticonα.TwooutcomesforParasiticConstraintSatisfaction
arediscussed.Thefirstisanemergenceoftheunmarked,occurringwhenβisamarked-
nessconstraintwhoseactivityemergesincontextswhereitisparasiticonα.Thesecond
outcome,whereβisafaithfulnessconstraint,isanemergenceofthefaithful.
(cid:1) IwouldliketothankthreeanonymousreviewersandEllenBroselowfortheiruseful
comments.Earlierportionsofthisworkwerepresentedatthe1997AnnualMeetingofthe
LinguisticSocietyofAmericaandthe1997AnnualMeetingoftheCanadianLinguistics
Association,andIamgratefultothoseaudiencesforhelpfuldiscussion.Earlierversionsof
thispaperhavealsobenefitedfromcommentsandsuggestionsfromJillBeckman,Diaman-
disGafos,JunkoItô,JohnMcCarthy,ArminMester,JayePadgett,GeoffPullum,Cathie
Ringen,BarrySchein,andmembersofphonologygroupsattheUniversityofCalifornia,
SantaCruzandtheUniversityofMassachusetts,Amherst.
NaturalLanguage&LinguisticTheory 19: 827–878,2001.
©2001KluwerAcademicPublishers. PrintedintheNetherlands.
828 RACHELWALKER
1. INTRODUCTION
Within the Altaic family, restrictions on the distribution of round vowels
are pervasive. In this paper, I explore the relation between three nonhigh
roundvowelpatterns inAltaic.Atthecoreisastudyofroundharmonyin
Classical Manchu (CMA) and Oroqen (Tungus branch of Altaic), which
presents an interesting complication to the usual pattern of Tungusic har-
mony.Theselanguages requireroundvowelsinthefirsttwosyllables ofa
wordinordertoinitiateroundspreading,thatis,thestructurethatinitiates
roundharmony–thetrigger–mustbeminimallybisyllabicinsize(Zhang
1996; Zhang and Dresher 1996). This contrasts with the more familiar
conditionunderwhicharoundvowelinthefirstsyllableissufficientalone
totriggerspreading. Thedistinction isrepresented schematically in(1).In
the familiar or canonical Tungusic round harmony, [Round] linked to the
firstsyllable spreads tosucceeding syllables (1a). Inthebisyllabic trigger
case, [Round]doesnotspread ifitislinked onlytothefirstsyllable (1bi),
but spreading occurs ifithaspre-existing affiliations withbothofthefirst
twosyllables(1bii).
(1)a. Canonicalroundharmony
b. Bisyllabictriggerroundharmony
Central to this investigation is understanding what underlies the two-
syllable requirement –stating this as aminimal size condition on triggers
simply expresses adescriptive generalization. Here,Iargue thatthebisyl-
labicthresholdisproducedthroughtheinteractionofwell-established and
conflicting phonological demands, formalized in Optimality Theory (OT)
as ranked, violable constraints (Prince and Smolensky 1993). Insight is
drawn from comparing simpler but related round vowel restrictions. This
studyidentifiestwodistinctrequirementsonroundinginbisyllabictrigger
languages: (i) initial round licensing, where [Round] must be linked to
theinitial syllable, and(ii)roundspreading. Theseconditions occurinde-
pendently inotherAltaiclanguages. Licensing aloneisactive inClassical
Mongolian (CMO)(Mongolian branch ofAltaic), and spreading from the
ROUNDLICENSING,HARMONY,ANDBISYLLABICTRIGGERSINALTAIC 829
first syllable occurs in Ulcha, which displays canonical Tungusic round
harmony. The rounding distributions sanctioned by initial licensing are
illustrated in (2). The structures in (2a-b) represent well-formed config-
urations since [Round] is associated to the first syllable, but (2c), where
[Round]hasonlyanon-initial link,isill-formed.
(2)a.
The separation of the requirements of initial round licensing and round
spreadingiscrucial.Inlanguageswithbisyllabictriggerseachrequirement
is visibly active, but they are ranked asymmetrically in relation to a con-
flicting constraint that restricts features to a tautosyllabic domain, that is,
aconstraint limitingalllinksofafeaturetoasinglesyllable. Thisranking
structure is key to understanding the two-syllable condition. I argue that
threshold effects are an instantiation of a kind of constraint interaction
exploredhere,termed PARASITIC CONSTRAINT SATISFACTION (PCS).In
bisyllabic trigger languages, this interaction arises as a consequence of
interleaving the tautosyllabicity constraint between licensing and spread-
ing: thelower-ranked spreading constraint issatisfied only whenitcanbe
parasitic on tautosyllabicity violations produced by round licensing. The
PCSconfigurationachievesthebinarythresholdstraightforwardlythrough
constraint conflictandranking,withoutnumericreference. Thisisadesir-
ableresult,sinceothertriggersizesthatcouldbecharacterizednumerically
(e.g., three syllables, four syllables, and so on) are unattested. A parallel
approach isshowntocaptureabisegmental triggerphenomenon.
A connected matter concerns the nature of the constraints. Pivotal to
the analysis is a family of constraints enforcing tautosyllabicity for fea-
tures,extendingItôandMester’s(1999)CrispEdgeconstraintonprosodic
constituency. Such constraints are shown to be independently supported
by various syllable-bound feature spreading in Altaic and elsewhere. An
important development proposed here is that tautosyllabic feature con-
straintsareassessedbottom-up,withaviolationaccruedforeachoffending
feature.Thisassessmentprovesnecessarytounderstandingbisyllabictrig-
gers.Asimilarevaluation isadopted forviolations offeaturalmarkedness
constraints–astepthatiscentralintheaccountofroundlicensing.Extend-
ingresearch byBeckman(1997, 1998), positional faithfulness constraints
are assigned a key role, realizing the prioritized status of the root-initial
syllable. These constraints not only capture the trigger role of the initial
syllable in round harmony, but also explain its status as a licensor of
[Round] via association. The analysis that is proposed for initial round
830 RACHELWALKER
licensing involves an asymmetric ranking of positional and nonpositional
faithfulnesswithrespecttofeaturalmarkedness,anotherinstanceofaPCS
configuration – in this case with parasitic satisfaction of faith. An altern-
ative substituting positional markedness constraints (Zoll 1996, 1997) for
positional faithfulness proves unsuitable, since it cannot prevent feature
specifications deriving from anon-initial syllable from overriding ones in
the root-initial syllable. Toachieve round spreading, constraints areadop-
ted along the lines of those proposed by Kaun (1995), motivated by her
extensive cross-linguistic study of round harmony. The relation between
thethreeroundingpatternsoflicensing,canonicalharmony,andbisyllabic
trigger harmony is accomplished in the account via minimally distinct
rankings.
The paper is organized as follows. In section 2, I establish the de-
scription of canonical Tungusic round harmony, and then present data
from CMA and Oroqen illustrating the bisyllabic condition on triggers.
TheCMOdistribution ofroundlicensing withoutspreading isintroduced,
adding athird membertothesetofrelated patterns. Section 3turns tothe
constraint interactions that produce the spreading and licensing require-
ments.Insection4,Ifocusontheanalysisofbisyllabic triggers, outlining
the important function of the tautosyllabic feature constraint and determ-
ining its ranking in relation to the requirements of round licensing and
spreading. Analternativecondition-based accountoftwo-syllable triggers
isconsidered, and otherapplications ofPCSconfigurations arediscussed.
Section 5 considers typological implications, deriving differences in the
three rounding patterns through minimal reranking and examining exten-
sions to other rounding distributions. Section 6 contrasts an alternative
approach tolicensing, andsection 7presents theconclusion.
2. THREE ROUND HARMONY PATTERNS
The basic pattern of Tungusic round harmony is familiar from comparat-
ive Tungusic studies, such as Kaun (1995), Li (1996), and Zhang (1996),
alongwiththeprecursorsonwhichtheybuild.InthissectionIfirstreview
the core canonical pattern, which does not impose a size restriction on
the trigger for harmony, and then go on to describe the more complex
distributional restrictions on round vowels in languages requiring a two-
syllable trigger. I subsequently identify a connected pattern in Mongolian
thatdisplays initialroundlicensing withoutroundspreading.
ROUNDLICENSING,HARMONY,ANDBISYLLABICTRIGGERSINALTAIC 831
2.1. Canonical TungusicRoundHarmony
Anexampleofcanonical roundharmonyoccursinUlcha,aTungusiclan-
guageofRussia(Kaun1995drawingonSunik1985).ThevowelsofUlcha
aregiven in(3). Vowellength iscontrastive onlyinword-initial syllables;
and [(cid:1)] is also restricted to the first syllable. The vowels participating in
round harmony, [a((cid:2))]and[(cid:3)((cid:2))],arehighlighted inabox.LikemanyTun-
gusiclanguages,Ulchaalsoexhibitsatonguerootharmony.Thisharmony
will be apparent in much of the data in this paper but is not the subject of
analysis (onthisseetheTungusicstudiescitedabove).
(3) Ulchavowels
The main properties characterizing the canonical pattern of Tungusic
round harmony are as follows with illustration in (4). First, the trigger is
subject to apositional restriction that is widely apparent across the Altaic
family: the trigger for round harmony must be a vowel in the root-initial
syllable. In addition, it must be nonhigh. This is part of a more general
requirement thatroundharmonypropagate strictlyamongstnonhigh vow-
els; hence, targets – vowels that undergo round harmony – must also
be nonhigh. The Ulcha data in (4a) present examples of round harmony
from [(cid:3)((cid:2))]in theinitial syllable tofollowing nonhigh vowels. Inthis type
of sequence, rounding must spread, that is, forms matching the structure
∗[C(cid:3)Ca]generallydonotoccur.1Highvowelsinthissystemactasblockers
(vowels thatprevent propagation ofround harmony). Observe in(4b) that
high vowels block round spreading from a preceding vowel, and they are
nottriggersortargetsthemselves.Further,althoughroundnonhighvowels
never occur after unround orhighvowels, round high vowelsoccur freely
innon-initial syllables afterunroundvowels,asverifiedin(4c).
(4)a. b(cid:3)(cid:2)n(cid:3) ‘hail(weather)’ g(cid:3)r(cid:3) ‘far’
t(cid:3)(cid:4)d(cid:3) ‘straightahead’ t(cid:3)t(cid:3)(cid:4)g(cid:3) ‘multi-colored’
(cid:5) (cid:5)
k(cid:3)(cid:2)r(cid:3)t(cid:6)(cid:7)v(cid:7) ‘toregret’ d(cid:8)(cid:3)gb(cid:3)l(cid:3)v(cid:7) ‘toprick,stab’
1 AsmallnumberofexceptionsarenotedanddiscussedbyKaun(1995,p.76,n.19).
832 RACHELWALKER
b. (cid:3)j(cid:9)lav(cid:7) ‘leggings’ v(cid:3)lm(cid:9) ‘long’
k(cid:3)(cid:2)v(cid:7)lav(cid:7) ‘toraiseamast(naut.)’ b(cid:7)qta ‘fragment’
m(cid:7)r(cid:9) ‘horse’ bu(cid:2)li ‘lampwick’
(cid:5)
c. ba(cid:2)p(cid:7) ‘pack,bunch’ s(cid:9)lt(cid:6)(cid:7) ‘sackfortinder’
AsummaryoftherestricteddistributionofnonhighroundvowelsinUlcha
isgivenin(5)alongwithschematicforms.(“C”representsanyconsonant.)
(5) SummaryofUlcharoundharmony:
a. Triggersarenonhighroundvowelsintheinitialsyllable;targets
are also nonhigh, and round nonhigh vowels never occur after
anunroundedvowel.Well-formedstructuresinclude[C(cid:3)((cid:2))C(cid:3)],
[Ca((cid:2))Ca],butnot∗[C(cid:3)((cid:2))Ca],∗[Ca((cid:2))C(cid:3)].
b. High vowels block round harmony; after a high vowel, a non-
highvowelisunrounded, i.e.,[C(cid:3)((cid:2))C(cid:9)Ca]and[C(cid:3)((cid:2))C(cid:7)Ca]are
well-formed, butnot∗[C(cid:3)((cid:2))C(cid:9)C(cid:3)],∗[C(cid:3)((cid:2))C(cid:7)C(cid:3)].
2.2. RoundHarmonywithBisyllabicTriggers
An interesting complication in the round harmony of some Tungusic lan-
guages has been uncovered in research by Zhang (1996) and Zhang and
Dresher(1996).Theyobservethatsomelanguagesimposeasize-threshold
onthetriggerofroundharmony;inparticular, thefirsttwosyllablesofthe
word must be round in order to induce round spreading. Examples occur
inCMAandOroqen,asdescribed below.
2.2.1. ClassicalManchu
CMA(alsoknownasWrittenManchu) isthelanguage represented bythe
Manchu writing system. It was the language of the Manchu court from
about the seventeenth century to the early twentieth century and is con-
sideredtobebasedontheJianzhoudialectoftheseventeenthcentury.The
following description and data are mainly from Zhang (1996) and Zhang
and Dresher (1996) (drawing onNorman 1978; Seong 1989), supplemen-
ted by Li(1996). Thevowel inventory is presented in (6); the vowels that
alternate inroundharmonyare[a]and[o].
(6) ClassicalManchuvowels
ROUNDLICENSING,HARMONY,ANDBISYLLABICTRIGGERSINALTAIC 833
RoundharmonyinCMAcloselyresemblesthecanonical patternofUlcha
in most respects. Examples of round harmony in CMA are shown in
(7). Round spreading amongst nonhigh vowels from a root to suffix is
illustrated in (7a). The data in (7b) present instances of unrounded suf-
fix variants for comparison. In these forms we observe that high vowels
block round spreading and their rounding specification is independent of
precedingvowels,asinthecanonicalsystem.Theexamplesin(7c)present
cases of round harmony within atrisyllabic root. Note that if the first two
syllables contain nonhigh round vowels, a third nonhigh vowel must also
beround,thatis,∗[CoCoCa]isill-formed.
(7)a. dobo-no- ‘gotooffer’ dorolo-no- ‘gotosalute’
(cid:5)
bot(cid:6)o-(cid:4)go ‘colored’ osoxo-(cid:4)go ‘havingclaws’
mo(cid:4)go-ro- ‘speakMongolian’ obo-xo ‘towash’
b. baxa-na- ‘gotoget’ kofori-na- ‘tobecomehollow’
gosi-(cid:4)ga ‘loving,compassionate’ arbu-(cid:4)ga ‘image’
(cid:5)
mond(cid:8)i-ra- ‘wringthehands’ nomula-xa ‘topreach’
c. dorolon ‘rite’ foxolon ‘short’
osoxo ‘claw’
Theaboveshow examplesofthefamiliar roundharmony pattern informs
where the first two syllables of the root are surface-round. Thus far it
would be reasonable to infer that the round second vowel is determined
by [Round] spreading from the first vowel, as in Ulcha. However, further
data contradict this conclusion. The data in (8) show a rather unexpected
outcome for roots containing a nonhigh round vowel only in the initial
syllable: roundspreadingdoesnotoccur.[Round]intheinitialvowelfails
tospreadbothfromroottosuffix(8a)andwithintheroot(8b).
(8)a. to-(cid:4)ga ‘few,rare’ do-na- ‘alightinswarm’
jo-na- ‘formasore’ no-ta ‘youngersisters’
go-xa ‘breakapromise’(perf.)
(cid:5) (cid:5)
b. t(cid:6)oban ‘alever’ t(cid:6)ola- ‘tofry’
(cid:5)
doran ‘virginland’ pod(cid:8)an ‘firecracker’
(cid:5) (cid:5)
t(cid:6)ot(cid:6)ara- ‘toactcarelessly’
Based on these data, Zhang (1996) and Zhang and Dresher (1996) estab-
lish the descriptive generalization that the first two syllables must contain
nonhigh round vowels in order for [Round] to spread in CMA. I will call
this the bisyllabic trigger condition. The implication is that the first two
834 RACHELWALKER
syllables of the forms in (7a) and (7c) must underlyingly contain round
vowels. That isbecause round harmony actually occurs inthose forms, in
contrast totheonesin(8).
There is a further point concerning the distribution of round vowels in
CMA that must be taken into consideration. First, it must be absolutely
clear that vowels in the second syllable are not subject to the neutralizing
effect ofround spreading. Someminimalpairs contrasting solely interms
of the round specification of the second vowel are given in (9). These
unambiguously showthatroundinginasecondsyllableiscontrastiveafter
aninitialnonhigh roundvowel.
(9)a. dola ‘barrenland’ dolo ‘inside’
b. doxa ‘stick’ doxo ‘lime’
c. noran ‘apileofwood’ noron ‘longing’
d. oxa ‘obedient’ oxo ‘armpit’
Onthebasis ofthese pairs, itmaybeexpected thatround nonhigh vowels
occur freely in the second syllable. However, a round nonhigh vowel in
the second syllable is prohibited following an initial unrounded syllable;
in other words, ∗[CaCo] roots are ill-formed. Note that this rounding dis-
tribution is also excluded in Ulcha. In CMAit is clear that this restriction
cannot be attributed to the rounding agreement produced by harmony,
since the well-formedness of both [CoCo] and [CoCa] shows that round
harmony does not carry from the first to the second syllable – there must
beabisyllabic trigger.
Thecondition tobeexplained isthatroundnonhigh vowelsonlyoccur
in the second syllable when following a round nonhigh vowel. I sug-
gest that this distribution is the result of an initial licensing requirement,
wherebya[Round]featureonanonhighvowelmustbelinkedtoanonhigh
vowelinthefirstsyllable.BisyllabicstructuresofCMAthatsatisfylicens-
ing are shown in (10a–b). These may be contrasted with the structures in
(10c–d) that contain an ‘unlicensed’ [Round] feature. [CaCo] words are
thusill-formedbecause[Round]isnotassociated withthefirstsyllable.
(10)a.
c.
ROUNDLICENSING,HARMONY,ANDBISYLLABICTRIGGERSINALTAIC 835
Asummarydescription forCMAispresented in(11).
(11) SummaryofClassical Manchuroundharmonyandlicensing:
a. Licensing:Post-initialroundnonhighvowelsoccuronlyimme-
diately following a round nonhigh vowel, i.e., [CaCa], [CoCo]
and[CoCa]arewell-formed, butnot∗[CaCo].
b. Bisyllabic trigger: [Round] spreads to following nonhigh vow-
els when the first two syllables contain round nonhigh vowels.
Highvowelsblock round harmony, i.e.,well-formed structures
include[CoCo-Co],[CoCa-Ca],[CoCi-Ca],[CoCu-Ca],butnot
∗[CoCo-Ca],∗[CoCa-Co],∗[CoCi-Co],∗[CoCu-Co].
2.2.2. Oroqen
CMApresented anexample ofround harmony requiring abisyllabic trig-
ger.Oroqen,aminoritylanguageofnortheast China,isasecondTungusic
language that exhibits this kind of pattern. I focus here on the evidence
Oroqen offers concerning the behavior of long vowels in harmony with
a bisyllabic trigger condition (see Zhang 1996 for additional details of
Oroqen harmony). The language description and data are from Zhang et
al.(1989), Zhang(1996), andZhangandDresher(1996).
ThevowelsofOroqenarelistedin(12).Oroqenpresentsarichersetof
vowel contrasts than CMA; of particular interest is the contrast in vowel
length. Round harmony produces alternations between a((cid:2)) ∼(cid:3)((cid:2)) and (cid:10)((cid:2))
∼o((cid:2)).
(12) Oroqenvowels
The operation of round harmony in Oroqen is illustrated by the forms in
(13a).Hereweseeroundspreading fromtheroottoasuffixwhenthefirst
twovowelsoftherootareroundandnonhigh.Bycontrast,thedatain(13b)
show the occurrence of an unrounded suffix alternant after unrounded or
high vowels. Fromthese data itisapparent that Oroqen round harmony is
836 RACHELWALKER
subject to the usual height restriction: only nonhigh vowels participate in
roundharmony.
(cid:5)
(13)a. (cid:3)l(cid:3)-w(cid:3) ‘fish’(def.obj.) t(cid:6)o(cid:4)ko-wo ‘window’(def.obj.)
(cid:5)
m(cid:3)(cid:2)t(cid:6)(cid:3)n-m(cid:3) ‘difficulty’(def.obj.)2 (cid:3)lg(cid:3)(cid:2)-r(cid:3) ‘dry’(pres.)
olo(cid:2)-ro ‘boil’(pres.) mo(cid:2)ro-ro ‘moan’(pres.)
b. t(cid:3)r(cid:3)ki-wa ‘boar’(def.obj) min(cid:10)-w(cid:10) ‘me’(def.obj.)
(cid:7)r(cid:7)(cid:2)n-ma ‘hoof’(def.obj.) jab(cid:7)-ra ‘walk’(pres.)
s(cid:10)r(cid:10)-r(cid:10) ‘awake’(pres.) ku(cid:2)mn(cid:10)-r(cid:10) ‘hold’(pres.)
As in CMA, round spreading fails in Oroqen when just the first syllable
of the root contains a round vowel. Zhang and Dresher make the import-
ant observation that even a bimoraic (long) round vowel is insufficient to
triggerroundspreading onitsown,asseenin(14).
(14) m(cid:3)(cid:2)-wa ‘tree’(def.obj.) do(cid:2)-r(cid:10) ‘mince’(pres.)
n(cid:3)(cid:2)da(cid:2)- ‘throw’ ko(cid:2)rg(cid:10) ‘bridge’
These data make evident that the bisyllabic trigger condition in Oroqen
is truly a bisyllabic condition not just a bimoraic one. Since CMA lacks
a vowel length distinction, it is silent on this matter. A final point is that
Oroqen displays the same initial licensing requirement for [Round] that
was identified in CMA (and consistent with the distribution in Ulcha):
nonhigh roundvowelsoccurinthesecond syllable ofrootsonlywhenthe
initial syllable contains a nonhigh round vowel (that is, ∗[C(cid:10)Co], ∗[CaC(cid:3)]
rootsareill-formed).3
2.3. Classical Mongolian
TheaboveTungusicharmonieshavebeenobservedtooccuralongwithan
initial licensing distribution. I turn next to data from Mongolian that re-
vealanoccurrenceinAltaicofroundlicensingalone.CMOrepresentsthe
2 Zhang (1996, p. 189) glosses this form as in the objective case. I assume that it is
infactthedefiniteobjectcasemarker,inaccordancewithZhang’sglossesofotherforms
withthissuffix.
3 AsnotedbyZhang(1996)andZhangandDresher(1996),Oroqenpresentsafurther
restriction on rounding in nonhigh vowels: in order for [Round] to occur in a nonhigh
vowel,itmustbelinkedtothefirsttwomorasofthestem,i.e.,[Co(cid:2)]and[CoCo]arewell-
formed,but∗[CoC(cid:10)](withinitialshortvowel)isill-formed.Thisinterestingrequirement
plausiblyhasfoundationinperceptual considerations,sincerounding contrastsarerelat-
ivelydifficulttoperceiveinnonhighvowels(Kaun1995).Therestrictionwillnotbethe
focus of analysis here, since it isdistinct from the condition on trigger-size. As seen in
(13–14),spreadingmustbeinitiatedbyatwosyllabletrigger–notsimplyatwomoraone.