Rethinking Geminates, Long-Distance Geminates, and the OCP Sharon Rose

Buildingon the analysis of long-distancegeminates as reduplication, thisarticle argues that the OCP mayapply to identical consonants acrossan intervening vowel. This is adduced from the behavior of gutturalconsonants in Semitic. It isfurtherargued that antigemination isnot resistance to an OCP violationbut avoidance of gemination; syncopebetween identical consonants avoids an OCP violationby creatinga geminate.This entails that there is no surface representa- tionaldistinction between true and fake geminates. Finally, cases of reduplicationare examined in which the standard reduplicant is changedto avoid either gemination or an OCP violation. Keywords: OCP, geminates,gutturals, reduplication, Semitic

1Introduction Perhapsno ‘‘principle’’ ofgenerative phonology has received as much attention as the Obligatory ContourPrinciple (OCP), whichhas been used to explaina myriadof phonologicaland morpholog- icalphenomena. The OCP prohibitsadjacent identical elements, where ‘ ‘elements’’ maybe de- finedas tone, segments, features, syllables, or evenmorphemes (Yip 1998). A commonformula- tionis takenfrom McCarthy1986b. (1) ObligatoryContour Principle At themelodic level, adjacent identical elements are prohibited. Althoughthe OCP was originallyformulated to apply to tone(Leben 1973, Goldsmith 1976), its applicationoutside the domain of tone to Semitic root consonants (McCarthy 1979, 1981) is considereda landmarkin autosegmental phonology. The combination of theOCP withautoseg- mentalspreading provided an elegant analysis of theskewed distribution of repeatedconsonants inSemitictriconsonantal verb stems, that is, of why verbs like samam areattested and not those like *sasam. If theOCP appliesto Semitic roots, verbs such as [samam] cannotcontain two

Manythanks to my consultants Mussie Bakit (Tigre), Hewat Asmelash, SennaiMussie, and Beraki Woldeabzghi (Tigrinya),all fromEritrea. I am gratefulto audiences at theUniversity of Southern California, the University of California, SanDiego, and the University of California, Irvine, for useful feedback, especially Rachel Walker,Barry Schein, Hagit Borer,Jean-Roger Vergnaud, Bernard Tranel, Kathleen Hubbard, and David Perlmutter. Thanks also to Matthew Chen andKeiichiro Suzuki for discussion of related issues.Finally, many thanks to theanonymous reviewers fortheir suggestions andcomments.

Linguistic Inquiry, Volume 31, Number 1,Winter 2000 85–122 q 2000 bythe Massachusetts Institute of Technology 85 86 S H A R O N R O S E instancesof /m/,butmust be analyzed as biconsonantal /sm/ andmapped to a triconsonantal templatefrom leftto right.The second consonant is spread to thefinal consonant position of the template,creating what is referred toas a ‘‘long-distance’’ geminate,a singleconsonant spread overtwo positions with a vowelintervening between the consonants, as in(2a). 1 Thevowels are associatedon a separatetier. Such a structureavoids violating the OCP, whereasa structurein whichthe two identical consonants are represented as separate root segments /smm/ wouldviolate theOCP (2b).

(2) a. s m b. s m m OCP violation

CVCVC CVCVC

a a Amechanismreferred toas Tier Conflation (attributed to Younes 1983 in McCarthy 1986b) alignsthe vocalic and consonantal tiers into a singlelinearized tier at a specifiedpoint in the derivation.Tier Conflation also aligns affixal morphemes lexically arrayed on separatetiers. The interveningvowel splits the long-distance geminate into two separate consonants during Tier Conflation.This is representedin (3).

(3) s m s a m a m No OCP violation

CVCVC C V C V C

a Inthe post –TierConflation structure, intervening vowels prevent violation of the OCP. As phonologicaltheory has progressed and, with the advent of Optimality Theory (OT) (Princeand Smolensky 1993, McCarthy and Prince 1993, 1995), has taken some revolutionary twists,previous analyses and assumptions have begun to be questioned. For example,Gafos (1996,1998) and I (Rose1997) reject the long-distance geminate represented in (2a).We propose thatSemitic roots characterized by doublingof thefinal consonant (i.e., samam)shouldinstead beanalyzed in terms of reduplication and that vowels and consonants are not segregated onto separatetiers. The biliteral analysis of the root as /sm/ remainsthe same. 2

(4) s a mi a mi

1 Therepresentation is alsoassumed forvowels, but the focus of thisarticle isonconsonants. 2 Associationof segments totemplates andthe characterization oftemplates are notaddressed here. See Buckley 1997a,Rose 1997,and Gafos 1998on derivations of (primary) Semitic verbstems inOT. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 87

Thespreading analysis in (2a) was adoptedpartly to avoid OCP violationsthat would ensue if theconsonants were separateand adjacent. A questionthat arises from theanalysis in (4) is whetherreduplication of this type violates the OCP, particularlysince the vowel intervening betweenthe two identical root consonants belongs to adifferentmorpheme. In thisarticle I will answerin the affirmative: the OCP is violatedby two identical consonants separated only by a vowel,a situationthat often results from reduplication—notjust in Semitic,but in other languages aswell. Although the OCP haspreviously been demonstrated to ignore intervening material (McCarthy1988, Yip 1988, 1989, Suzuki 1998), this analysis applies only whenindividual features ofthe two relevant consonants are being considered, not when the consonants are completely identical.However, I willargue that an intervening vowel is ignored when computing the OCP betweenconsonants, in both cases. The violability of constraints in OT allowsthe OCP tobe violatedin surface forms underpressure from higher-rankedconstraints. Thearticle is organized as follows. In section 2 Iassessthe behavior of gutturalconsonants inTigrinya and Tigre, two North Ethiopian Semitic languages. Gutturals may neither geminate noroccur as long-distancegeminates, but they may be reduplicated.At firstglance this appears tosuggestthat geminates and long-distance geminates pattern alike, a factthat could be attributed toa similarlinked representation. Nevertheless, I showthat the pattern of gutturals is most fruitfullyanalyzed as a widespreadOCP-driven restriction on gutturalrepetition across an inter- veningvowel. This is a welcomeresult since, if long-distance geminates are really a form of reduplication,any similarity between them and true geminates cannot be attributed to a linked representation.Furthermore, normal /overapplicationeffects of palatalizationand labialization in someEthiopian Semitic languages demonstrate that verbs with reduplication (i.e., f tf t) and verbs withfinal doubling or long-distance geminates (i.e., fet t)dopattern alike and should therefore berepresentedalike (Gafos 1998,Rose 1997). 3 Insection 3 Iextendthe proposal that the OCP appliesbetween consonants across an interven- ingvowel to languagespreviously hailed as the prime examples of antigemination,or resistance to syncopebetween identical consonants (McCarthy 1986b). I arguethat the standardantigemination

CiVCi sequenceitself actually violates the OCP, justas it doesin Tigrinyaand Tigre with respect tothe guttural subclass (see Berkley1994 for asimilarproposal about homorganic consonants ina CVCconfiguration). Resistance to syncope is reanalyzed as resistance to gemination, not avoidanceof an OCP violation.Previous explanations were forcedto stipulate identity restrictions onsyncope,but this new account provides a moreinsightful explanation for whysome syncope cases apply only when theconsonants are identical (to avoid an OCP violation),whereas others apply except when theconsonants are identical (to avoid geminates). The typology of syncope andepenthesis patterns that are sensitive to identical consonants documented in Odden 1988 is modeledas a differencein constraint ranking within correspondence theory (McCarthy and Prince

3 Variousauthors have pointed out that long-distance and local geminates donot always behavesimilarly with respect togeminate integrityor inalterability(Inkelas and Cho 1993, Hayes 1986,Schein and Steriade 1986). It is usually assumed thatthe long-distance geminate behavesdifferently because ithas beensplit by Tier Conflation(McCarthy 1986b)and so no longerfunctions like a geminate.Tier Conflationis unnecessary in an analysis that adopts (4). 88 S H A R O N R O S E

1995).In section 4 Iexaminethe surface distinction between fake and true geminates and show thatthe behavior of fake geminates can be capturedby referenceto domainsof OCP application. Insection 5 Ipresentexamples from languagesin which the shape of the reduplicant is altered toavoidviolations of eitherthe OCP oraconstraintagainst gemination. I concludewith a summary insection 6.

2Gutturals inEthiopian Semitic 2.1Long-Distance Guttural Geminates Itis wellknown that guttural consonants (pharyngeals, laryngeals, uvulars) resist gemination in someSemitic languages (McCarthy 1994). This is true of thoseEthiopian Semitic languages with arangeof guttural consonants, notably Tigre and Tigrinya, which have [ ? h ¿ É].TheTigrinya imperfectiveforms in(5) demonstratethat when the paradigm requires gemination, guttural conso- nantsdo not geminate. Type A andType B regularverb roots are shown for comparison(lexically determinedconjugation patterns of differentroots in Ethiopian Semitic are referred toas‘ ‘types’’). Insome cases, such as the passive imperfective, another consonant geminates instead of the guttural(data from Berhane1991). 4 (5) Type A Type B ‘whip’‘hurt’ ‘pull’ Imperfective yé- @rréf yé-béddél yé-séÉéb (*yé-séÉÉéb) Passiveimperfective yé- érr@f yé-bédd@l yé-ssaÉab (*yé-saÉÉab) Causativeimperfective y @- érréf y@-b@ddél y@-sÉéb (*y@-saÉÉéb) Aconstraintbanning double linking of guttural consonants would account for thefailure of gutturalsto geminate (G 4 guttural).

(6) * G

X X Ina theorythat allowed long-distance geminates, such a constrainton double linking would naturallyextend to long-distance geminates and would predict that long-distance guttural gemi- natesare also ruled out. This is exactlythe case in Tigrinya and Tigre. There are no verbs with finaldouble gutturals of theshape *C aGaG or *CaCGaG,thatis, no verbs like sa a or bas a .5 Onthe other hand, reduplication of gutturals is permitted,as the Tigre total-copy verb stems in (7) illustrate(the a isa transitionalepenthetic vowel that prevents gutturals from appearingin codas).

4 Gutturalconsonants lower adjacent /@/ to[a]. See Berhane 1991for contexts. 5 Buckley(1990) analyzes triliteralroots with final doubling such as k’ rd d-‘dice’as involvingreduplication but thoseformed from biliteral roots such as d d-‘worsen’as cases ofspreading. Angoujard (1988) makes asimilar assumptionfor the triliteral type, but does not pronounce on thoseformed from biliterals. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 89

(7) CGCG GCGC a faÉ faÉ-‘rubtogether’ ÉanÉan-‘ carefor someone’ a 6 na? na?-‘tobe unsteady, waver’ ¿ab¿ab-‘ incubate,cover to let ripen’ a sa¿ sa¿- ‘dance’ ?af?af- ‘lecture,criticize incessantly’ hathat-‘ rushsomeone’ Theseverbs are formed from biliteralroots of theshape CG orGCbycopying the root entirely. Theexistence of total-copy verbs with final gutturals as in (7) demonstratesthat roots of the shapeCG arepossible; they simply cannot expand by doublingthe final consonant: *CVGVG. 7 Underthe long-distance geminate analysis, the lack of finaldoubling verbs and the existence of theverbs in (7) followdirectly from theban on guttural linking. Nevertheless, I showin thenext sectionthat the reduplication analysis of final doubling verbs is better equipped to handle addi- tionaldata from Tigreand Tigrinya.

2.2Different Gutturals Despitethe apparent similarity between guttural gemination and long-distance guttural gemination, abanon linkingcannot extend to otherfacts from Tigrinyaand Tigre. First, it isnotonly identical gutturalsthat are prohibited in the verb root, but also two different gutturals.Thus, there are no verbforms suchas C a a - or Ca ah-inTigre and Tigrinya. This gap follows from theban on adjacenthomorganic consonants applied to Semiticroots (Greenberg 1950), attributed to the OCP (McCarthy1988, 1994, Yip 1989, Padgett 1992, Pierrehumbert 1993, Elmedlaoui 1995, Buckley 1997b,Frisch, Broe, and Pierrehumbert 1997). Second,the juxtaposition of gutturals from separatemorphemes is also avoided in Tigre. Thisoccurs with the causative morpheme /?a-/,the1st person singular nonperfective subject marker /?é-/,andthe broken plural prefix /?a-/ (Raz1983). Broken plurals are formed by internal adjustmentof the consonants and vowels rather than by affixationalone. When the singular form ofanounis of the shape CVCVC in Tigre, the broken plural frequently has an /?a-/ prefix. (8) SingularPlural a. k@béd ?a-kbud ‘belly’ b. météd ?a-mtud ‘stake’ c. w@rék’ ?a-w@rrék’ ‘silver’ d. béÉar ?a-bÉur ‘sea’ e. déh@b ?a-dhub ‘gold’

Whenthe initial consonant is aguttural,the /?a-/ prefixcannot be used;instead, a differentbroken pluralpattern is adopted.

6 Thisverb is used only in a fixedform with the conjugated auxiliary bela ‘to say’: na ana bela. 7 There isactuallya distinctionbetween thepharyngeals and laryngeals in Tigre. I foundno verb roots of theform Ch, andonly one of the form C .Verbs withpharyngeals in the second position are more common.In Tigrinya there are verbslike lahl h-‘pantfrom fatigue’ (Buckley 1990). 90 S H A R O N R O S E

(9) SingularPlural a. ?ékél ?akal *?a-?kul ‘corn,crop’ b. Éabél Éabéllét *?a-Ébul ‘rope’ c. Ééwar Éawr@t *?a-Éwur ‘foal,small donkey, camel’ d. Éaréb Éaréb *?a-Érub/*?a-Éarréb‘water-skin’ e. Éakél Éakéllét *?a-Éakul/*?a-Éakkél ‘hoe’ Anexplanation based on a banon gutturals occurring in coda position (McCarthy 1994, Rose1996) could not account for twoaspects of the data. First, this ban normally induces the epenthesisof avowelfollowing the guttural (e.g., * s’a n s’a a n ‘load’), sowe mightexpect ! a form like a- a kul in(9a).Second, such an analysis would explain only the plurals of theshape a-GCVC, wherethe guttural would be incodaposition, and not those of the shape a-GVCCVC (byanalogy with (8c) a-w rr k’),wherethe guttural would be in onset position. Furthermore,it isclear that the [ ?]oftheprefix is not doubly linked at the root to theinitial guttural,since in many cases the guttural is adifferentconsonant. 8 Finally,this is nota banon twogutturals cooccurring in the same word or in adjacent syllables, as the examples in (8d –e) demonstrate: a-dhub. Therelevant environment is two gutturals separated by a vowel. Thesame phenomenon is foundin verbswith the causative prefix /?a-/ inTigre, shown with regularverbs in (10a–b).If theverb begins with a guttural,either the initial vowel is lengthened asshown in (10c)or, more commonly, the prefix is supplemented with the consonant [t], making itidenticalto the causative reflexive /passiveprefix: /?at-/ (10d–e).The verb in (10d–e) alsohas the extra [@]betweenthe second and third root consonants characteristic of reflexive forms. All verbsare given in the 3rd masculine singular perfective, marked with the suffix - a. (10) Type A Causative a. k’@tla ‘kill’ ?a-k’t@la‘ causeto kill’ b. s@bra ‘break’ ?a-sb@ra ‘causeto break’ c. Éarsa‘ plough’ Éa:r@sa ‘cultivate’ d. ¿ak’ba ‘ guard’ ?at-¿ak’@ba‘ causeto guard’ e. Éad a ‘leave’ ?at-Éad@ a‘makeleave’

Asimilaravoidance strategy is foundwith the 1st person singular subject prefix, which is /?é-/ inboth the imperfective and the jussive, as shown in (11a– d). This prefix does not occur if the rootbegins with a gutturalconsonant (11e –h),but it canbe realized if anotherconsonant intervenes betweenthe two guttural consonants, either a rootconsonant as in (11c) or the reflexive /t-/ as in(11g). Finally, with no prefix, the 1st person singular jussive would be identical to the 2nd personmasculine singular imperative form [ Ééd@ ]‘leave!’; todistinguish them, the first stem vowelof the 1st person singular jussive is realizedas [i], asin(11h). 9

8 Prunet(1990) analyzes theplural prefix as /@/ withan epenthetic glottal stop to avoid an onset violation. Either way,the point is the same, thatsequences ofgutturals are avoided. 9 Thesame phenomenonis foundwith the 1st person plural marker /?én-/,whichis realized as [na-](the shape C a- instead of C -isdue to lowering triggered by theinitial guttural): n-s bb r ‘we break’but na- ass’ b ‘we wash’. This couldbe analyzedas beingformed by analogy with the 1st person singular. The plural differs from the singular in the imperfective onlyby the presence of[n]: - versus n-.Inthe jussive the 1st person plural has no[ ?]: n -sb r ‘let us break’ and n - s’ b ‘let uswash’ . GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 91

(11) a. ?é-m@zzén ‘I weigh’ b. ?é-s@bbér ‘I break’ c. ?é-s’¿én ‘I load’ d. ?é-sb@r ‘letme break’ e. Éas’s’éb ‘I wash’ f. ¿addém ‘I join’ g. ?é-t-Éas’s’ ab ‘ Iwashmyself’ h. Éid@ ‘letme leave’ Thedouble-linkingaccount is incapableof explainingthe casesof gutturalavoidance outlined above.Furthermore, the examples in (10d–e) and(11c,g) demonstrate that a gutturalinitial prefix isacceptable if another consonant appears between the guttural consonants. The only possible explanationis thatthere is abanon sequences ofgutturalconsonants when separated by only a vowel,but not when separated by aconsonant.Moreover, no explanationfor theTigre examples ispossiblein terms of tiersegregation. If morphemesare lexically represented on theirown tiers (McCarthy1986b), the two gutturals, belonging to differentmorphemes, would not be subjectto theOCP. Theywould only reside on thesame tier once Tier Conflation had applied and the affix andverb stem were aligned.But once this occurred, the vowel of the prefix would intervene betweenconsonant and vowel. Therefore, the OCP effectmust apply between the guttural conso- nantsacross the intervening vowel.

2.3Some Apparent Exceptions Threekinds of affixes with gutturals appear despite the guttural OCP restriction.These are the frequentativeinfix, the negative marker /?i-/ (e.g., i- -m zz n ‘Idon’t weigh’), and h-initial3rd personpossessive and object clitics such as /-hu/ ‘his’, whichoccur with vowel-final nouns and verbs:for example, de-ha ‘her hand’, u-hu ‘hisbrother’ , s m a-hu ‘listento him’ (the [h] is epenthetic;it does not occur with consonant-final nouns: ra as-u ‘hishead’ ). Iclaimthat the undominatedconstraint O NSET (‘‘Syllablesmust have onsets’ ’), unviolatedin Tigre,forces inser- tionof the epenthetic [h] betweenvowels and forces maintenance of [ ?]ininitialposition of the negativemarker, despite the OCP violation.See Myers 1997, Yip 1998, and others on the OCP asa violableconstraint in OT. Withthe other guttural affixes, alternative means of expressing themorphosyntactic features are available and are used instead of allowing the OCP violation. Notethat in each case the alternative means does not involve simple deletion of the offending 1 0 affixalguttural, which would leave an O NSET violation. Frequentativereduplication is another case in which a sequenceof gutturals is allowed in Tigreand Tigrinya. The frequentative verb form isformedby copyingthe penultimate consonant

10 Analternative analysis would be tostate thatthese affixes reside ina domainin whichthe OCP does not apply oris rankedlow. (See Buckley1996 on constraint domains in OT.) The negative marker andthe clitics are theoutermost ofall affixes thatmay attach totheverb or noun.If anyaffixes are imperviousto the OCP restriction, it should be those thatare theleast integratedinto the verbal or nominalcomplex. I claim insection 2.4 that the guttural OCP in Tigrinya isrestricted tothe basic stem. Lowenstamm andPrunet (1986) also provide evidence that object clitics inTigrinya form aseparate phonologicalword from the basic stem andtherefore escape OCPrestrictions (see section4.2). 92 S H A R O N R O S E asthe onset of aninfix [a], or[a:] inTigre.The usual meaning is ‘ intensive’or ‘ distributive’in mostEthiopian Semitic languages, but ‘ diminutive’in Tigre. The following examples are from Tigre:

(12) a. d@ ma‘ tell,relate’ d @ a: @ma‘ tellstories occasionally’ b. @rfa ‘whip’ @ra:r@fa ‘whipa little’ c. ba?asa ‘fight’ ba?a:?asa‘ fighta little’ d. sa¿ana ‘load’ sa¿a:¿ana‘ loada little’ If thepenultimate consonant is a guttural,reduplication does take place (12c –d),violating the OCP restrictionon gutturals.A long-distancegeminate analysis would claim that these examples aresimply cases of reduplication, and since reduplication is allowed (i.e., forms like ab ab ‘incubate,cover to let ripen’ ), thisis entirely as predicted. I proposethat these cases can be handledby constraintsrequiring both reduplication and identity between base and reduplicant to 1 1 expressthe frequentative: M AXB R and IDENTB R (Rose,to appear, Buckley 1998).

(13) a. MAXB R Everyelement of thebase corresponds to an element in the reduplicant.

b. IDENTB R Correspondentelements in the base and the reduplicant have identical values for feature F.

Theconstraint M AXB R requiresreduplication as part of the formation of internal reduplication. Noother kind of consonant may substitute, so identity of thebase and reduplicant consonants is requiredand ensured by I DENTB R .Thenormal epenthetic consonant in Tigre is [h] (Raz1983) andin Tigrinya is [ ?],bothgutturals. Failure to reduplicatewould result in vowel hiatus, exclu- sivelyruled out: * ba.a: as. Tigreand Tigrinya flout other constraints in order to realize the frequentative,such as one on multiple reduplication ( n d d- n da:d d ‘burnagain’ ) and ! anotheron the size of the template, allowing frequentatives with five consonants ( m sk r ! m s ka:k r ‘testifyagain’ ) (Rose,to appear). I concludethat the frequentative base-reduplicant faithfulnessconstraints outrank not only restrictions on multiplereduplication and templatic size, butalso the OCP appliedto gutturals.

2.4Tigrinya and Domains Affixationin Tigrinyacontrasts with affixation in Tigre in allowingsequences of gutturalssepa- ratedby anintervening vowel across the basic verbal or nominalstem boundary. I claimin this sectionthat the OCP inTigrinya is in effect within a moredelimited morphological domain. Tigrinyapatterns similarly to Tigre with respect to the distribution of gutturals in basic verb

11 Itshould be clarified thatin anOT analysisof reduplication, complete identityis notnecessary between areduplicant segment andthe base segment itcopies, but the two segments are stillrelated toeach othervia correspondence. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 93 stems,but it allows the affix /?a-/ tooccuradjacent to root gutturals, as shownin the following causativeforms: (14) Regular Causative a. ¿ayy@n@ ‘spoil’ ?a-¿ayy@n@ ‘causeto spoil’ a b. ?as@r@ ‘arrest’ ?a-? s@r@ ‘causeto arrest’ c. ¿add@ @ ‘buy’ ?a-¿add@ @ ‘causeto buy’ Thefollowing plural forms alsoclearly show that the OCP restrictiondoes not operate across themorphological boundary of theplural prefix /?a-/ inTigrinya.Where available, cognate Tigre wordsare shown for comparison. (15) Tigrinya Tigre Singular Plural SingularPlural a a. É@wi ‘brother’ ?aÉ wat Éu Éaw a b. ¿of ‘smallbird’ ?a¿ waf a c. ?om ‘tree’ ?a? wam a d. ?é ri ‘foot’ ?a? ar ?é ér ?é ar a e. ¿arki‘ friend’ ?a¿ réxti TheOCP asapplied to thegutturals operates at the level of the basic stem in Tigrinya, but at the levelof the word in Tigre. That the OCP maybe restricted to morphological domains in this manneris not new. Past interpretations of the OCP (Yip1988), as well as recent reanalyses (Suzuki1998, Alderete 1997, Ito ˆ andMester 1998), also refer togiven domains of application, suchas thestem or prosodicword. (See alsoBuckley 1996 on constraint domains in OT.)

2.5Gradient OCP andthe Intervening Consonant AnOCP restrictionon the guttural subclass is independentlyrequired to accountfor thelack of rootscontaining two different adjacent gutturals. Furthermore, it has been demonstrated that interveningconsonants detrimentally affect the strength of theOCP inSemitic roots. Pierrehumb- ert(1993) and Frisch, Broe, and Pierrehumbert (1997) argue that cooccurrence constraints on Arabicroot consonants should be assessedgradiently using a similaritymetric (see Berkley1994 ongradientOCP inEnglish).They show that OCP constraintson homorganicconsonants apply overintervening consonants in Arabicroots, but with a far weakereffect than between adjacent consonants.The existence of total-copy verbs with gutturals in Tigreand Tigrinya shows that an interveningconsonant reduces or blocksthe impactof theOCP oversurface sequences of guttural consonants,but an intervening vowel does not. It follows that roots of the shape GCG should alsobe attested,although they are predicted to befar from numerous.Tigre has several roots of theshape GCG, witheither identical or different guttural consonants.

(16) a. ¿ar¿a‘causesomeone to pasture cattle’ b. ?ar?a ‘shove’ c. Éas¿a‘lackbutter /milkin food; be dry due to lack of oil’ d. Éan?a‘twistankle, leg’ e. had?a‘calmdown’ 94 S H A R O N R O S E

Buckley(1997b) reports 6 occurrencesof such roots in hiscorpus of 2,744Tigrinya roots (based onda Bassano’ s (1918)dictionary), all 6 withnonidentical gutturals. It should be pointed out thatwithout considering the OCP theexpected number of nonadjacent guttural triliteral roots totals 49,1 2 indicatingthat the OCP maystill function across intervening consonants, but only weaklyto reduce the likelihood that such a rootwill occur. The data in (16) and Buckley’ s results furthersupport the claim that the guttural OCP appliesabsolutely across intervening vowels, but onlyweakly across consonants.

2.6Locality and the Intervening Vowel Theconclusion that a vowelintervening between two consonants does not impede the application oftheOCP mustbe assessedwith respect to theissue of localityor adjacency.It is well attested thatwhen the OCP appliesto features and not whole segments, it may apply across intervening segments(Steriade 1987, Padgett 1992, Odden 1994). In a feature-geometricframework thisis oftenattributed to the hypothesisthat features occupy different tiers and that intervening segments unspecifiedfor thefeature in questionare irrelevant. For example,Akkadian labial dissimilation (vonSoden 1969, Hume 1992, Odden 1994, Suzuki 1998) operates across intervening nonlabial consonantswithin the basic stem (excluding suffixes). The nominalizing prefix /ma-/ dissimilates to[na-] ifthe stem contains a labialconsonant (17c –d). (17)a. ma-zuukt ‘ mortar’ b. ma-Skanu-m‘ place’ c. na-pÉar ‘totality’ d. na-rkabt‘ chariot’

Atier-dependentanalysis of thiseffect is givenby Hume (1992), who attributes it toaconstraint againsttwo Labial C-Place articulators. Since the articulators of the other consonants are on separatetiers, they do not intervene, and the two Labial nodes are considered adjacent on the Labialtier, as illustrated in (18).

(18) m a r k a b t Place Place PlacePlace Place

Cor Dor Cor Lab Lab

TheTigredata clearly exhibit a restrictionon the guttural subclass and hence the restriction should operateat the level of a featuralnode such as Pharyngeal (McCarthy 1994), just as the labials doinAkkadian.However, this predicts that intervening consonants, unspecified for Pharyngeal, shouldnot block the restriction in Tigre,and yet they do. An intervening consonant is important

12 See Pierrehumbert1993, Buckley 1997a and Frisch, Broe, and Pierrehumbert 1997 for more detailson how this figureis calculated. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 95 inTigre,but immaterial in Akkadian. This suggests that Akkadian has no adjacency restriction, butTigre does. Odden(1994) identifies three important parameters of adjacencyin accounting for arange ofassimilatoryand dissimilatory phenomena: root adjacency, syllable adjacency, and unbounded adjacency.1 3 TheOCP canbe restricted to apply between adjacent segments, between adjacent syllables,or across an entire word. As theystand, none of theseparameters can account for the factsin Tigre.The unbounded adjacency of Akkadianclearly does not hold in Tigre,as discussed above,where intervening nonguttural consonants block. If theguttural restriction applied to adja- centroot nodes, the vowels would be expectedto blocksince they too have root nodes, yet they donot. If therestriction applied between adjacent syllables, or evenadjacent onsets, forms like ar a ‘shove’ and -t- as’s’ab ‘Iwashmyself’ would not be possible. Again, the blocker is an interveningconsonant. The CVC configuration does not fit neatly within Odden’ s adjacency typology. As theTigre guttural OCP appliesbetween adjacent consonantsevenwhen these are separated bya vowel,I proposethat this configuration be incorporatedinto the adjacency typology. 1 4 (19) Consonantadjacency Twoconsonants in sequence are adjacent irrespective of interveningvowels. Consonantadjacency may also apply in theconfiguration CC, providing the consonants are non- identical.This allows for dissimilationof root-adjacent and vowel-separated consonants. Both arefound in Yimas(Foley 1991), where two liquids are disallowed in root-adjacentposition and acrossa vowel.The perfective suffix /-r/ isrealized as [c] following /l/: ta`mpu`tpul`r`rm tamputpulcrm ‘theydidn’ t hitthose two’ (p. 255). Elsewhere the /-r/ isdeletedfollowing /r/: ! wu`r`ra wura ‘takein hand(perf. sg.)’(p. 245). Across avowelthe second of twoliquids ! isrealizedas [t]: w`ar`ark warat k ‘makerepeatedly’ (p. 54), apr`ara aprata ‘open, ! ! spread(inchoative)’ (p. 290). I claimin section 3 thatidentical consonants in a CCconfiguration aregeminates and do notviolate the OCP. Theadjacency typology thus includes root adjacency

13 TransplanarLocality is another parameter, usedto assess interactionbetween consonantsand vowels. Odden(1994) cites manycases where theOCP applies under syllable adjacency. Since all ofthem involvelanguages withopen syllables, they could be reanalyzed as cases where theOCP applies under consonant adjacency. The only exceptionis Dahl’ s Law (velar voicingdissimilation) in Southern Gikuyu (Davy and Nurse 1982),which appears tobe sensitiveto syllable adjacency: witness contrastingpairs such as [a.ke.o.ki. ®a] ‘andhe trod on it’ with no voicing dissimila- tion and [a.åeo.ke.ra]‘ andhe got up’ with voicing dissimilation. Davy and Nurse (1982)state thatthe assignment of the twovowels to separate syllablesin the example [a.ke.o.ki. ®a] is representativeof careful speech andpossible if the second vowelconstitutes a vocalicprefix: /o-/ ‘Class 3prefix’or /e-/ ‘reflexive’. Infast speech suchexamples are pronounced withdiphthongization and voicing dissimilation. Dahl’ s Law can besuspended in careful speech inother Bantu languages, forexample, Gusii: [i.nko.a.ke.aa. åe.te] ‘youhit it’ or [a.ro.ke.te] ‘ hewas circumcised’, [ka.ko.ro.ma]‘ he(Class 12)is biting’. Despite these examples, thesyllable generalization does appear tobe robustfor Southern Gikuyu, so I willleave aside theissue ofreplacingsyllable adjacency inthis article. 14 Consonant refers toall consonants,including consonantal glides. Hume andOdden (1996) argue against the feature [consonantal],but only as traditionallydefined to excludelaryngeals and glides. The restrictioncould also be formulated as ‘‘Noadjacent C-Place,’’ butthe model of Clements andHume (1995)assumes thatvowels also have C-Place. There may alsobe avoweladjacency parameter, butI donot explore this here. See Suzuki1998 for possible examples, such as Kera, Marshallese, Woleaian,Ainu, and Tzeltal. 96 S H A R O N R O S E

(allowingfor consonant-voweldissimilation or consonant-consonant dissimilation), consonant adjacency,syllable adjacency, and unbounded adjacency. Following Suzuki (1998), the OCP may applyunder any of these adjacency conditions and can therefore be split into a familyof con- straints.However, since unbounded adjacency subsumes the other cases, high ranking of a con- straintsubject to unbounded adjacency entails OCP violationslocally and at a distance. 1 5 Theinterpretationof consonant adjacency runs afoul of recentassumptions about the internal structureof segmentsand locality. Gafos (1996)argues that in aCVCconfiguration the consonant gesturesare not local; the intervening vowel prevents interpreting them as such. However, he arguesthis primarily to prevent assimilations betweenconsonants. Clements and Hume (1995) stipulatethat both vowels and consonants have a C-Placenode, a conclusionthat also appears tobepartlydriven by theneed to prevent consonant-consonant assimilation across an intervening vowel.However, in assessing the OCP, whichregulates phonotactic constraints, locality is rou- tinelydefined over longer strings of segmentsthat are never subject to assimilation. There is no attestedassimilation between labial consonants across other consonants, but labials may dissimi- lateover other consonants at a distance,as seen in the Akkadian example (17). Dissimilatory effectsare not subject to thesame adjacency requirements as assimilatory ones, a distinctionthat couldfollow from constraintson local spreading (N ´õ Chiosa´inandPadgett 1997, Walker 1998). Inconclusion, interaction between consonants separated by a vowelis permissible in cases of OCP-triggereddissimilation effects, whereas it may not be permitted for assimilation. Itmight appear that consonant adjacency could be analyzed as the OCP applyingbetween onsetsin adjacent syllables. Forms like ar a ‘shove’ and -t- as’s’ab ‘Iwashmyself’ in Tigre precludethis analysis, but there are also no examples showing dissimilation between an onset anda codasince gutturals are not allowed in codaposition. However, cases of ‘‘syllable-bound’’ OCP effectsdo bearon the issue, since they involve dissimilation between onset and coda across a vowel. Onecase is thatof Seri glottal deletion, cited by Yip (1988), whose discussion in turn isbasedon Marlett 1981 and Marlett and Stemberger 1983. In Seri the glottal stop of the passive prefix /a:?-/ isdeletedif itoccursin codaposition following another glottal stop across a vowel: /?a-a:?-sanx/ -a:-sanx ‘whowas carried’or /?i-?-a:?-kasÏ ni/ i- -a:kasÏni ‘mybeing bitten’ . ! ! Thisdoes not occur if thepassive prefix precedes vowels: t-a: -a: -itax ‘was itmadeto burn?’ (Marlett1981:77). Although Yip interprets this to mean that the ban on glottals is restricted to thedomain of the syllable, another interpretation is possible: the OCP isstill violated in t-a: - a: -itax , butthe glottal stop may not be deletedbecause of a vowel-hiatusrestriction, which is

15 Suzuki(1998) proposes a proximityhierarchy, designed to capture a rangeof intervening material inassessing adjacency.Intervening material may benothing,as ingeminate structures(which Suzuki assumes are asequence oftwo identicalsegments), a singleconsonant, a mora,two moras, twosyllables, and so on, ranked according to the following universalformat, in which X standsfor elements affected bythe OCP: (i) Proximityhierarchy X . . . X $ XX . . *X-C0-X . . *X-m -X . . *X-m m -X . . X-s s -X . . . . . X-` -X} Althoughthis hierarchy is intriguing, I donot adopt it since itallowstoo many unattested possibilities (i.e., two moras, twosyllables). GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 97 consistentlyrespected in Seri, with the exception of abstract consonant-initial stems (Marlett and Stemberger1983). Thus, Seri can be analyzedas a caseof consonantadjacency operating between onsetand coda. 1 6

2.7Guttural Geminates and the Root /Stem Ihaveestablished that guttural restrictions in Tigreand Tigrinya are regulated not by the linked representationof long-distance gemination, but by a generalOCP restrictionon guttural conso- nantsapplying across a vowel.Assuming that guttural geminates in Semitic are doubly linked, theymust be ruled out (a) bya separateconstraint against guttural geminates, (b) bya surface constraintagainst a gutturalsequence regardless of linking, or (c) bya banon gutturalcodas or moraicguttural codas (McCarthy 1994). In both Tigre and Tigrinya an epenthetic [a] isinserted followinggutturals to prevent gutturals from appearingin codas, except when they occur word- finally.Since word-final codas are usually analyzed as nonmoraic in Semitic (McCarthy and Prince1990, Rose 1997, but cf. Buckley 1997a), an analysis banning moraic gutturals does double dutyand captures both the lack of gutturalcodas and the lackof guttural geminates. The constraints onguttural geminates (a) ora sequenceof gutturals (b) wouldalso work, but in each case a separateconstraint against guttural codas would also be required. If theguttural sequences in verbs like sa a -areruled out by theOCP, whyare other verbs like f f-‘collect,amass’ allowed, since they would presumably violate the OCP appliedto the articulatorLabial? I claimthat all forms ofreduplication are subject to the OCP; verbs such as f f existbecause formation of triconsonantal verbs via reduplication is more important than obeyingthe OCP restriction. 1 7 Thisanalysis holds for allconsonants except gutturals, which also showstronger restrictions in nonadjacent positions in the root (Buckley 1997b). The ranking of constraintswould be along the following lines, where T EMPLATE isused informally to refer to theexpansion of the root to conform to the triconsonantal templatic shape, ranked within a hierarchyof placerestrictions (McCarthy and Prince 1993):

(20) OCP/PHARYNGEAL . . TEMPLATE . . OCP/VELAR, OCP/LABIAL . . OCP/CORONAL Thatthe gutturals would have a strongerOCP constraintthan other places of articulation could followfrom markednesstheory. 1 8 Theanalysis in (20)also avoids duplication between a separate

16 TheSeri case appliesonly with the passive prefix. Cantonese labial consonant dissimilation is another case often citedas arisingfrom a syllable-boundconstraint operating between onsetand coda, but Yip (1988) shows that it actually reflects amorphemestructure constraint, as italso applies across syllablesin language games. 17 Therightward nature of Semitic reduplicationcan beattributed either to affixation, as Gafos (1998)claims, orto aconstraintthat reduplication copy the rightmost base consonant(Rose 1997).The affixation analysis would encounter problemsin explaining the lack ofTigrinya CGG verbstems, since theguttural OCP in Tigrinya does not apply across affixboundaries. 18 Lombardi(1995) proposes that Pharyngeal is theunmarked place ofarticulationbecause oflaryngealconsonants. Themarked natureof pharyngeal consonants [ É ¿]isdue to combinationswith other features. Lombardi’s accountdoes nottake intoconsideration the behavior of laryngeal consonants depending on contrasts within a language(Rose 1996). Thismay notbe acrucial issue,however, since some languagesplace OCPrestrictions on subclasses ofaparticularplace ofarticulation(i.e., coronal; also labial— see Elmedlaoui1995) but not on other places ofarticulation. 98 S H A R O N R O S E lexical,morpheme structure constraint OCP andthe same kind of OCP constraintapplying to the output.1 9 Toconclude this section: I haveshown that an apparent similarity between long-distance gutturalgeminates and true guttural geminates in Tigrinya and Tigre is acoincidence.Sequences ofgutturalgeminates are prohibited because the OCP appliesstrictly across an intervening vowel, butonly weakly across a consonant,whereas guttural gemination is prohibited because moraic gutturalcodas are banned. This is awelcomeresult, since it wipesout a potentialcounterexample toproposalsto doawaywith long-distance geminates, and it does so by makinguse of anindepen- dentlynecessary mechanism to restrict root formation in the two languages. The proposal that theOCP appliesacross vowels but not consonants has significant implications, in thatit suggests thatthe OCP maybe violated in variousareas of thegrammar ofalanguagewhere it was previously thoughtnot to apply. I nowturn to one of thosecases, that of antigemination.

3Antigemination 3.1The Tier Conflation Approach Antigeminationdescribes the effect whereby a phonologicalrule, such as syncope, is resistedif theresultingstructure would violate the OCP bycreatinga sequenceof adjacent identical segments (McCarthy1986b). Typical examples are found in Afar, aCushiticlanguage of Eritrea.A syncope ruledeletes unstressed vowels in a peninitialtwo-sided open syllable (21e –f). Thisrule is blocked iftheconsonants are identical (21g –h)(data from Bliese1981). (21) a. di ib-t-e´ ‘shemarried’ e. di b-e´ ‘hemarried’ b. wa er-t-e´ ‘shereconciled’ f. wa r-e´ ‘hereconciled’ c.xarar-t-e ´ ‘sheburned’ g. xarar-e´ *xarre´ ‘he burned’ d.danan-t-e ´ ‘shewas hurt’ h. danan-e´ *danne´ ‘hewas hurt’ If thevowel were deletedin (21g–h),the two identical consonants would be adjacent and constitute anOCP violation,as shown for (21h)in (22).

(22) d a n a n *d a n n

C V C V C] C V C C] Inthis manner, the OCP actsas a filteron derivations. A crucialassumption under this approachis thatif syncopehad applied, the OCP violationwould not be repairedby automatically fusingthe segments to create a geminate.McCarthy (1986b) allows this to occur only when Tier

19 Theanalysis in (20) ties inwell withrecent reinterpretationsof theOCP in terms oflocal conjunction of markedness constraints(Alderete 1997,Ito ˆ andMester 1998).In this theory the OCP is reinterpretedin terms ofmarked structure occurringmore thanonce in a givendomain. A constraintlike *P LACE/PHAR isself-conjoined, so that two occurrences are worse thanone. This constraint is rankedseparately; thus,it is functionallyequivalent to a separate OCP /PHAR constraint.I willnot adopt the local markedness approachhere, as itdoesnot explicitly capture adjacency, a fundamental propertyof theOCP; see Suzuki1998 for comparison of localconjunction of markedness andthe OCP. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 99

Conflation(the mechanism that aligns separate morphemic tiers) is applied, although Yip (1988) questionsthis assumption. Cases where antigemination is ignored and syncope applies regardless areattributed to either (a) phoneticimplementation rules outside the purview of phonological constraintsor (b) separationof vowels and consonants onto distinct tiers, allowing for long- distancegeminates. Those languages that resist syncope, like Afar, eithermust apply syncope followingTier Conflation or mustnot have vocalic and consonantal tier segregation in the first place.Those languages that allow syncope between identical consonants would apply the rule beforeTier Conflation when the identical consonants form ageminate.This is theanalysis pro- posedfor ClassicalArabic. ClassicalArabic has a process(referred toasIdentical Consonant Metathesis in Brame1970) thateither syncopates or metathesizesa vowelin anopen syllable between two identical consonants intheverb stem. Although Brame analyzes syncope as metathesis followed by vowel shortening (/madada/ [maadda] [madda]‘ hestretched’), thesurface result is loss of thevowel between ! ! twoidentical consonants. A regularForm I(see McCarthy1979, 1981 on forms or binyanim) perfectivestem has the shape CVCVC throughout the paradigm: for example, katab-tu ‘I wrote’ and katab-a ‘hewrote’(23a,d). A verbwith final doubling has the shape CVCC when followed byvowel-initial affixes as in(23e– f) (examplesfrom McCarthy1986b and Brame 1970). (23)a. katab-tu ‘ Iwrote’ d.katab-a ‘ hewrote’ b.samam-tu ‘ Ipoisoned’ e.samm-a ‘hepoisoned’ c.madad-tu ‘ Istretched’ f. madd-a‘ hestretched’ Inthe imperfective verb form, andin some plural forms, metathesisoccurs instead of syncope, asseen in (24d), where the verb stem shape is CVCC instead of the standard CCVC of regular rootsas in (24c). (24)a. ta-ktub-na ‘ you(f.pl.) write’ c.ya-ktub-u ‘ hewrites’ b.ta-mdud-na ‘ you(f.pl.) stretch’ d.ya-mudd-u ‘ hestretches’ Theverbs with final identical consonants are represented as cases of long-distance gemination. Deletingor displacing the vowel leaves an acceptable geminate structure, as shownin (25). 2 0 (25) s m s m

C V C V C C V C C

a a Sincethe rule applies only between identical consonants, and only between tautomorphemic consonants,it is formulated so as to include the long-distance geminate in the representation (McCarthy1986b), as shown in (26).

20 Itis unclearwhether the vowel position or just the vowel is deleted. Either way, the point is the same. 100 S H A R O N R O S E

(26) Arabic Identical Consonant Metathesis a

i C V C V 1 <3>j 2 4 5 1 2 3 4 5 Condition: i É ´j McCarthy(1986b) further points out that consonants originating in othermorphemes, such as the infix /t/,whichappears in the Form VIII verbmeasure, are not subject to the metathesis rule: k-t-atab-a ‘hecopied’, * kattab-a.As theinfix /t/ isaseparatemorpheme from theroot /t/, they arenot represented as a long-distancegeminate. Therefore, deleting the vowel would create a sequenceof identicalconsonants, as in (27), violating the OCP. (27) k t t b *k t t b

C C V C V C C C C V C

a a a As seenin (25), the long-distance geminate respects the OCP andantigemination is not violated whensyncope applies. Antigemination will only be violated if the consonants are separate and adjacent,which occurs when Tier Conflation splits the segments (22). The syncope of Classical Arabicdoubled verbs is therefore a caseof syncope applying before Tier Conflation. Thisanalysis relies crucially on TierConflation. Even apart from thefact that the application ofTier Conflation at an intermediate stage in the derivation is incompatible with current OT approachesto phonology and morphology, Tier Conflation has had a checkeredand problematic history.Odden (1988) criticizes the separation of consonantsand vowels onto distinct tiers, which entailsTier Conflation, and also questions the universality of antigemination. Bat-El (1988) shows howreference to morphemic distinctions is requiredeven following Tier Conflation. Rules that apparentlyhad to apply before Tier Conflation have been reanalyzed: McCarthy (1983) relied on thelong-distance geminate representation to capture Chaha floating palatalization, but I have shown(Rose 1994) that Chaha floating palatalization must follow Tier Conflation, and Gafos (1998)and I (Rose1997) provide reanalyses in termsof reduplicative identity within OT. Banksira (1997)shows that Tier Conflation is irrelevant to Chaha devoicing, which operates on surface representations,rather than operating before and after Tier Conflation as argued by McCarthy (1986a).Gafos (1998)argues that V /Cplanarsegregation and Tier Conflation are overly powerful toolsthat are not required if long-distance geminates are reanalyzed as reduplication as in (4). Inthe following section I willshow that the alternative interpretation of theOCP thatI proposed insection2 for Tigregutturals, combined with the rejection of long-distance geminates and Tier Conflation,does not create a problemfor analyzingantigemination, but actually explains more effectivelywhy syncope in a languagelike Classical Arabic applies only between identical conso- nants. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 101

3.2Antigemination as OCP Violation Toexplainthe difference between Classical Arabic and Afar withoutappealing to Tier Conflation andlong-distance geminates, I applythe proposals made in section 2 for EthiopianSemitic guttur- alsand make the following assumptions:

(28)a. A surfacesequence C iVCi violatesthe OCP underconsonant adjacency.

b.Any surface C iCi sequencein a givendomain is a geminateand does not violate the OCP. Iassumethat any output sequenceof twoidentical consonants within the same domain (i.e., prosodicword, basic stem) constitutes a geminate,a singleconsonant with a longduration. This isinline with phonetic evidence, which has found no distinction between surface true and fake geminates,as documentedby Miller(1987) for LevantineArabic. See also Boersma’ s (1998:427) statementthat ‘ ‘adjacentidentical gestures may be heardas a singlegesture.’ ’ Therefore,when syncopeapplies between two identical consonants, a geminateautomatically results, an assumption alsomade by Kisseberth(1970) and Yip (1988). I followKeer (1998)in proposing that coalescence oftwo segments violates only I DENT constraintspertaining to features and not the constraint UNIFORMITY (McCarthyand Prince 1995). Given this assumption, it followsthat coalescence of identicalsegments violates no faithfulnessconstraints. Keer arguesthat fusing adjacent identical segmentsis alwayspreferred to leaving them separate, as two separate segments would violate markednessconstraints (i.e., *L AB, *COR)twicerather than once. They would also violate the OCP. Thishypothesis entails that underlying geminates will always be asingleinput consonant byLexicon Optimization (Prince and Smolensky 1993) and that any geminates arising from syncopewill automatically fuse and be representedas a singlegesture with long duration. 2 1 This alsoensures that a geminateresulting from syncopedoes not violatethe OCP. Thedistinction betweena ‘‘true’’ geminateand a ‘‘fake’’ geminate,as in Tigrinya (Kenstowicz 1982), arises from morphemeconcatenation. I maintainthat the different phonological behavior of fake and truegeminates can be attributedto theirmorphological affiliation, a pointI willelaborate on in section4. Odden (1988) provides evidence that identical adjacent consonants even in separate wordsmay resist regular epenthesis in languagesas diverseas Yir Yoront,Tondano, and Icelandic, suggestingthat (28b) may even be applicableacross word boundaries in some languages. Turningnow to the individual analyses, I proposethat syncope applies in Classical Arabic finaldoubled verbs precisely to avoid violating the OCP operatingunder consonant adjacency. Thisdirectly contradicts the antigemination analysis in McCarthy 1986b, which argues that a pre–Tier Conflation C iVCi configurationdoes not violate the OCP. Butthe antigeminationanalysis offers noexplanationfor whysyncope would apply only when the flanking consonants are identi- cal.This is built into the description of the rule as a stipulation.The analysis proffered here providesan explanation: to avoidan OCP violation.Nevertheless, the two analyses do coincide

21 Myers(1997) allows two adjacent Htonesto surface inawinningcandidate, but the second is realized phonetically as downstep.Different phoneticbehavior of fake andtrue geminates, however, is unattested. 102 S H A R O N R O S E inthatthe structure resulting from syncopeis analyzedin both cases as a geminate.The analysis Iproposeis inspired not only by the Ethiopian Semitic data, but also by data discussed by Hoberman(1988). He citesseveral examples from differentSemitic languages, where gemination (inhis terms, local spreading) is preferred to long-distance spreading. In Syriac, for example, biliteralroots are positioned to avoid vowels occurring between identical consonants (29c –d), evenif this violates the normal left-to-right distribution of root consonants in Semitic (29a– b), andresults in discrepancies between the members of the paradigm. (29) Root pk ‘ smash’ Root ktb ‘write’ a. pekk-et ketb-et ClassI perfect1sg. b. paakk-aa kaatb-aa ClassI activeparticiple 2sg. fem. c.te-ppok *te-pkok te-ktob ClassI imperfect2sg. masc. d.appek *apkek akteb ClassIV perfect3sg. masc. Inmy terms, the OCP wouldbe violatedin a sequencesuch as -pkok, butnot in a sequencesuch as -ppok because[pp] is ageminate.Syriac prefers geminationto violating the OCP. Toaccountfor Form Iverbs(23), a constraintpreventing geminates (N O-GEM)isnecessary inadditionto theOCP, aswell as familiar faithfulness constraints within correspondence theory such as MAXIO ,whichprevents deletion. N O-GEM isexpressed as a banon long consonants.

(30) a. NO-GEM Longconsonants are disallowed. b. OCP-[C-Adj] Asequenceof adjacent identical segments is disallowed(consonant adjacency).

c. MAXIO Everyelement of theinput has a correspondentin the output.

InClassical Arabic the OCP mustbe ranked higher than N O-GEM toaccount for caseswhere syncopeoccurs only when the consonants are identical, but not elsewhere. It must also be ranked higherthan M AXIO .Theoutput of syncope [madda] in (31a) violates N O-GEM and MAXIO , but unlikethe losing candidate in (31b),it doesnot violate the OCP. 2 2 Iabstractaway from thenature oftheinput and represent the root-aspect vowel complex as a fullyformed stem, which may be concatenatedat a differentlevel. See Orgun 1996 on morphological levels in OT.

(31) madda ‘he stretched’

/madad-a/ OCP MAXIO NO-GEM

+ a. madda * *

b. madada *!

22 Thecandidates mada and maada inwhich the consonant is deletedwould violate a high-rankingconstraint against consonantdeletion. They would also violate constraints on templatic shapeand M AXBR,as noreduplication is apparent intheoutput form. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 103

Aregularverb would not undergo syncope since there are no violations of the OCP, asshown in (32).

(32) kataba ‘he wrote’

/katab-a/ OCP MAXIO NO-GEM

a. katba *!

+ b. kataba

Thisanalysis provides an explanation for whysyncope occurs only between identical conso- nants.Although it iscertainlyplausible that there is aconstraintrequiring syncope of unstressed vowelsin opensyllables, it wouldnecessarily be rankedbelow M AXIO .Sucha constraintwill be necessaryfor otherlanguages such as Afar andother dialects of Arabic,which have more wide- spreadsyncope, but the driving force behind the deletion of thevowel in ClassicalArabic is the OCP violation.The same ranking, OCP . . NO-GEM,alsodrives the metathesis effects illustrated in(24).No reference is made to long-distancegeminates or TierConflation in thisanalysis. The rankingsin (31), combined with the new interpretation of the OCP, providea moreinsightful accountof thesyncope effects in ClassicalArabic than previous analyses, which were forcedto stipulatethe identical consonants in the structural description of the rule. Furthermore, languages thatdisplay similar deletion effects between identical consonants (i.e., Maliseet-Passamaquoddy, Koya;see Odden 1988) may be analyzedthe same way without relying on anotherwiseunwar- ranted(i.e., not driven by the morphology) separation of vowels and consonants onto distinct tiers. Beforediscussing antigemination cases, I wouldlike to address the problematic Form VIII ktataba ‘hecopied’in Classical Arabic. Recall that the Tier Conflation explanation for theverb’ s failureto undergo metathesis to *[kattab-] relied on the stipulation that the rule apply only to long-distancegeminates. In addition, the two instances of [t] haveseparate representations under theTier Conflation hypothesis, and bringing them to adjacent positions via metathesis would violatethe OCP. Thereare two problems with assuming that the consonants would violate the OCP ifthevowel were deleted.First, according to themorphemictier hypothesis, the twoinstances of[t] wouldbe onseparatemorphological tiers at the point at which syncope applies, so no OCP violationswould occur. An OCP violationcould occur after Tier Conflation applied, since the twoinstances of [t] wouldthen be adjacent. But McCarthy (1986b:257) states that Tier Conflation automaticallyfuses separate adjacent identical consonants into true geminates, so the structure shouldbe legitimate.Second, if theapplication of theOCP totwo adjacent consonants is really responsiblefor blockingmetathesis, we mightexpect metathesis or another avoidance strategy toapply when the root begins with a [t], whichshould also place two separate consonants in adjacentposition ([t-t-aba ¿-]‘follow,pursue, investigate’ ), yetthis does not occur (*[tatba ?]). Of course,if metathesisis stipulatedto applyonly to long-distance geminates, such a responsewould beruled out, and the metathesis rule would be inapplicable to [ktatab-]. I havealready shown theweakness of an approach that stipulates the linked representation in the rule formulation. I 104 S H A R O N R O S E proposeinstead that the failure of metathesis to apply in [ktatab-] is due to a constrainton morphosyntacticfeature realization requiring distinctness between forms (Gnanadesikan1997, Rose1997, Urbanczyk 1998, Walker 1998). If metathesiswere toapply, the output would be identicalto Form II—kattab—andwould violate constraints on morphological distinctness. Form II‘causative’is indicatedmorphologically by geminationof the medial consonant, whereas Form VIII ‘reflexive’is conveyed by the position of the affix [t] followingthe first consonant. This doesnot entail that a givenverb must have a Form IIfor themetathesis to fail to apply, only thatif metathesis did apply, the verb would no longerresemble Form VIII andwould be identical toa Form IIcausative.Tolerating the OCP violationinduced by the two instances of [t] is preferableto violating distinctness of morphological expression. Letus turnnow to Afar. Bliese(1981) notes that the identity condition on syncope is unex- pectedsince Afar hasgeminate consonants. This would seem to suggestthat N O-GEM should not outrankthe OCP. However,given the violable nature of constraints in OT, N O-GEM does not needto beasurface-trueexceptionless condition in a languagein orderto manifest its presence. Itis sufficientfor itto berankedhigher than the OCP. InAfar, syncopeoccurs in regularverbs, soa constraintis necessary to capture it. I informallylabel the constraint DELETE2 3 and it must outrank MAXI O ,asshown in (33).

(33) digbe ‘he married’

/digib-e/ DELETE MAXIO

a. digibe *!

+ b. digbe *

For averbwith consonant repetition, we mustconsider the other constraints N O-GEM and the 2 4 OCP. NO-GEM mustbe ranked over the OCP, andover the syncope constraint D ELETE, since nodeletion occurs. The resulting form violatesthe OCP, butthis is preferableto violating N O- GEM,as(34) illustrates.

(34) danane ‘he was hurt’

/danan-e/ NO-GEM DELETE MAXIO OCP

+ a. danane * *

b. danne *! *

Thus,Afar isaparadigmcase of antigemination:it avoidscreating geminates. McCarthy (1986b) showshow syncope does apply between two identical consonants that belong to different mor-

23 Thiscould be formulatedas abanon adjacent lightopen syllables. See Zawaydeh1997. 24 Ihavefound little extra evidenceto support this ranking over the opposite one. Violations of both constraints abound.One potentialcase isthat vowel-final (feminine) nouns form the plural by reduplicatingthe last consonantand adding [a]: amo ‘head’, amooma ‘heads’(Bliese 1981:177).Consonant-final (masculine) nounsform the plural by adding a suffix -wa: alib ‘tendon’, alib-wa ‘tendons’. If reduplicationwere appliedto consonant-finalstems, ageminate would result: *alib-ba. Derivedgender is also formed by applyingthe same reduplicationstrategy to bothfeminine and masculine vowel-finalnouns, but derived gender is not marked onconsonant-final nouns. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 105 phemesin Afar. Thisis to be expectedif the OCP islimitedto certain morphological domains. Ireturnto this in section 4. Thoughspace limitations preclude my reanalyzingthe numerousother examples of antigemi- nationand post– and pre– Tier Conflation rules given in McCarthy (1986b) and the other relevant examplesin Yip1988 and Odden 1988, I notethat all are amenable to reanalysisin termsof the OCP applyingunder consonant adjacency. A furtherconsequence of theanalysis presented here isthatthe lack of syncope between identical consonants can be directly tied to the overall lack ofgeminatesin some languages, for whicha N O-GEM constraintis clearlynecessary. Under the antigeminationaccount, only the OCP isarguedto berelevant in syncope.For example,Kisseberth (1970)claims that Tonkawa syncope does not apply between reduplicated consonants precisely becausethe language lacks geminates: for example, /picena-?/ [picno?]‘hecuts it’ but ! /hewawa?/ [hewawo?]‘heis dead’ . McCarthy(1986b) argues that Kisseberth’ s gemination ! explanationis inferior to one relying on antigemination. However, his objections are based on (a) theassumption that the output of syncopewould not be a truegeminate and (b) thefact that geminatesarise in the language via morpheme concatenation. Both of these objections can be addressedwithin the presenttheory. The outputof syncope does violate N O-GEM;myinterpretation coincideswith Kisseberth’ s viewof surfacing geminates. Second, in many languages without underlyinggeminates (e.g., English, Modern Hebrew), geminatesmay arise at morpheme bounda- riesor viaassimilation. This only suggests that N O-GEM isveryhigh ranking but not completely undominated,as permitted within OT. Constraintson faithfulnesswould prevent deletion of input segmentsdespite creation of geminates.Given the hypothesis that input geminates always consist ofasinglesegment (Keer 1998),the only faithfulness constraint violated by simplification of a tautomorphemicgeminate would be M AX-m .Aheteromorphemicgeminate, on the other hand, resultsfrom theconcatenation of two separate input consonants. Deleting one of these will violate MAX-Cinaddition to M AX-m .Theranking of N O-GEM withthese two faithfulness constraints onsegments and moras gives a three-waytypology of languages with geminates. (35) a. Languagewith no geminates NO-GEM . . MAX-C, MAX-m b. Languagewith only heteromorphemic geminates MAX-C . . NO-GEM . . MAX-m c. Languagewith heteromorphemic and tautomorphemic geminates MAX-C, MAX-m . . NO-GEM MAX-m . . NO-GEM . . MAX-C Theinteraction of theseconstraints does not produce a systemthat allows tautomorphemic gemi- natesbut disallows heteromorphemic geminates by deletion. As far asI know,this is a correct prediction.

3.3 NO-GEM andEpenthesis

Inlanguages that have no geminates on the surface, N O-GEM willnecessarily be highlyranked, inmanycases above the OCP. Itfollows that epenthesis or deletionmay occur to preventgeminate structuresfrom arisingon the surface. Deletion occurs in Burum (Gasaway 1997),a Papuan 106 S H A R O N R O S E languageof the Finisterre-Huon group: /ek-ket/ [eket]‘ theysaw’ or /nen-nini/ [nenini]‘ our ! ! sister’. Epenthesisis reported in Yimas (Foley 1991), another Papuan language. Underlying /mml/ isrealizedas [m émél]‘Javanesefile snake’ and identical consonants in differentmorphemes are separatedby epenthesis: /nakatémayk- kiak/ nakat maykikiak ‘Icallhim’ . Unlikein English ! orModernHebrew, inthese languages the geminate ban is absolute,even eliminating geminates thatmight arise through morpheme concatenation. More challenging is the behavior of ‘ ‘long- distance’’ geminatesin a Semiticlanguage that places restrictions on geminates, the subject of thissection. Chahaand Muher are two closely related Ethiopian Semitic Gurage languages. Both lan- guageshave biliteral verbs with final doubling (i.e., long-distance geminates). However, Muher hasgeminate consonants, whereas Chaha has geminates only in certainrestricted environments: word-finallyin participlesused as part of compoundverbs, word-initially preceded by an epen- theticvowel in asmallgroup of nominals,and across morpheme boundaries with certain sonorant- initialaffixes (see Banksira1997:70, 116, 154, 158 for details).The location of thegeminates at wordedges suggests that they have arisen via morpheme concatenation, but I haveno explicit evidenceto supportthis hypothesis for allcases. In all other areas in whichMuher has geminates, cognateChaha words have a singlesurface consonant. The‘ ‘long-distance’’ geminates(or biliteralroots with reduplication) behave differently in thetwo languages. In theimperfective form oftheverb, there is normallyno epenthesisbetween thetwo final consonants of a triliteralroot if a suffixis added, no matter the nature of the consonants. (36) RootImperfective Muher kft yé-k@ft-émw -t‘ theyopen (m.pl.)’ sb r yé-s@b r-émw -t‘ theybreak (m.pl.)’ Chaha kft yé-k@ft-o ‘theyopen (m.pl.)’ sb r yé-s@b r-o ‘theybreak (m.pl.)’ Theonly exception to thisis found in finaldoubled verbs, formed from biliteralroots. In Muher, whichallows geminates, there is noepenthesisbetween the consonants. In Chaha, which generally doesnot allow geminates, 2 5 epenthesisseparates the two identical consonants. (37) RootImperfective Muher sd yé-s@dd-émw -t‘ they(m.) chase’ mz yé-m@zz-émw -t‘ they(m.) extractfrom abundle’ df y-a-d@ff-émw -t‘ they(m.) liein wait’ Chaha sd yé-s@déd-o ‘they(m.) chase’ mz yé-m@zéz-o‘ they(m.) extractfrom abundle’ df y-a-d@féf-o ‘they(m.) liein wait’

25 Butsee Banksira1997, where itis arguedthat Chaha has underlyinggeminates thatare simplifiedon thesurface inall contextsexcept those listed above. Note that words with double z havea tendencyto form geminates inChaha: forexample, y f zz ‘itis better’ rather than y -f z z. Thus, y m z zo can alsobe pronounced as y m zzo. I have no explanationfor why [zz] is favored. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 107

As shownin (38), the Chaha data can be accounted for byrankingN O-GEM higherthan the OCP andthe faithfulness constraint preventing epenthesis, D EPIO ,‘‘Everyelement of the output has

acorrespondentin the input.’ ’ (38) y s de do ‘they (m.) chase’

/y -s didi-o/ NO-GEM DEPIO OCP + a. y s d d-o * *

b. y s dd-o *!

BecauseMuher freely allows geminates, N O-GEM isranked lower than D EPIO andno epenthesis occursto prevent gemination. Unlikein the verb forms, inother cognate words that have geminates in Muher we finda singleconsonant in Chaha,not epenthesis: for example,Chaha k m r versusMuher k mm r ‘pile’ (Leslau1979). We canreasonably assume that Chaha had surface geminates at one time, since we findtraces of them. Devoiced obstruents occur where geminates are found in Muher: for example,Muher s bb r-versusChaha s p r-‘break’. Geminateintegrity (the resistance of gemi- natesto epenthesis; Hayes 1986) can be captured by the ranking N O-GEM . . OCP . . MAXIO (i.e.,it’ s betterto simplify the geminate by deleting a morathan to insert a voweland violate theOCP). Thefinal-doubling verbs bypass the N O-GEM . . OCP . . MAXI O rankingbecause of ahigher-rankedconstraint on templates. I arguedin section 2 thatthe templatic size constraint forcesreduplication with biliteral roots in Semitic. The templatic size constraint on the verb is satisfiedif epenthesis applies, but would be violatedby lackof reduplication(i.e., s d-o). Thus, epenthesisapplies regardless of the OCP violationit entails.No size constraint is in forcefor the nouns,and simplification of the geminate is chosen instead of epenthesis, in order to avoid an OCP violation.In the next section I willshow how the Chaha case fits into the typology of interactionamong the OCP, N O-GEM,andthe faithfulness constraints M AX and DEP. Undera TierConflation /long-distancegeminate analysis, the Chaha /Muherdata prove prob- lematic,partly because the mechanics of Tier Conflation are not clearly defined. Since Chaha allowslong-distance geminates but disallows geminates, there must be somemeans to distinguish them,namely, an (empty) vowel position between the two final consonant slots: s dVd (Rose 1992,Prunet 1996, Prunet and Petros (Banksira) 1996, Banksira 1997). If therewere novowel position,the (disallowed) geminate would not be split by epenthesis in Chaha, either before or afterTier Conflation, as thiswould violate geminate integrity. But even adopting the emptyvowel slotraises problems. It is unclearwhether (a) themere presenceof thevowel slot causes the long- distancegeminate to split, or (b) alexicalvowel occupying a templaticslot is necessaryto separate along-distancegeminate. Under scenario (a) thelong-distance geminate should break apart in bothChaha and Muher, resulting in a sequenceof twoidentical consonants in eachcase, a violation ofthe traditional OCP (unlessfusion of the offending consonants is an option in Muher —but McCarthy(1986b) states that fusion is aconsequenceof Tier Conflation alone). Under scenario (b) thelong-distance geminate should remain after Tier Conflation. No epenthesis would be 108 S H A R O N R O S E necessaryeven in Chahasince the structure is not atruegeminate, and the surface form y s ddo ‘they(m.) chase’is incorrectly predicted for bothChaha and Muher. This kindof factis unproblem- aticwithin the present analysis. N O-GEM mustbe a high-rankedconstraint in Chaha to account for thelack of geminatesin thelanguage in general.By ranking it overD EP,we allowepenthesis toapplyto prevent creation of anygeminates on thesurface,even when epenthesis is notindepen- dentlynecessary for syllabification.

3.4Range of Identity Effects in Syncope and Epenthesis Comparisonof Afar andClassical Arabic responses to syncope illustrates just two of thepossibili- tiesavailable to indirectly reference identical consonants in syncope or epenthesis processes by usingtwo well-motivated constraints, the OCP andN O-GEM.Odden(1988) summarizes the kinds ofsyncopeand epenthesis rules that occur in languagesand the relevance of identical consonants. Thecases he discusses can be accounted for byranking N O-GEM andthe OCP withrespect to faithfulnessconstraints (D EP and MAX)andwhatever constraints drive epenthesis (I NSERT) and syncope (DELETE).Thedeletion cases and relevant languages are given in (39). (39) a. Deletea vowelunless flanking Cs are identical (Afar) NO-GEM . . DELETE . . MAX, OCP NO-GEM . . DELETE, OCP . . MAX b. Deletea vowelblindly (Hindi,Klamath, Maltese Arabic, Akkadian) DELETE . . MAX, NO-GEM, OCP c. Deletea vowelonly if flanking Cs areidentical (ClassicalArabic, Koya, Telugu) OCP . . MAX, NO-GEM . . DELETE OCP . . MAX . . DELETE, NO-GEM Case(39a) is Afar resistanceto syncope (antigemination), captured with either ranking. Blind deletion(39b) is foundwhen D ELETE isrankedabove all other constraints, which are not crucially rankedwith respect to each other. Odden cites Hindi, Klamath, Maltese Arabic, and Akkadian aslanguages of thistype, in whichvowels are deleted even if the flankingconsonants are identical. If MAX isranked over D ELETE asin (39c),then no deletionoccurs, unless there is ahigh-ranked constraintsuch as the OCP forcingdeletion. Thus, deletion occurs only when the OCP-violating configurationC iVCi wouldotherwise result. This is the Classical Arabic syncope case; Odden alsocites non-Semitic examples, among them Koya and Telugu. Theinsertion cases are similar. Epenthesis applies everywhere except when a violationof theOCP wouldoccur (40a). Geminate integrity could be recast under this ranking as a special caseof OCP avoidance.Geminate integrity is foundin a languagelike Palestinian Arabic, where epenthesisoccurs between two word-final consonants, but not if they form ageminate.I will returnto this in section 4, as it touches on the issue of true versus fake geminates. Epenthesis applieseverywhere in the case described in (40b); Odden cites Yimas, Chukchi, and Hua as languagesof this type. In (40c), where D EP isrankedabove I NSERT,epenthesisapplies only when aviolationof N O-GEM wouldotherwise occur, as inChaha or Lenakel. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 109

(40) a. Inserta vowelunless flanking Cs are identical (PalestinianArabic) OCP . . INSERT . . DEP, NO-GEM OCP . . INSERT, NO-GEM . . DEP b. Inserta vowelblindly (Yimas, Chukchi,Hua) INSERT . . DEP, NO-GEM, OCP c. Inserta vowelonly if flanking Cs areidentical (Chaha,Lenakel) NO-GEM . . DEP . . INSERT, OCP NO-GEM . . DEP, OCP . . INSERT Inconclusion, ridding phonological theory of TierConflation and long-distance geminates doesnot deprive us ofan accountof antigemination.On the contrary, by appealing to twogeneral constraintson identical consonants, N O-GEM andthe OCP, Ihaveshown how the ranking of theseconstraints with respect to faithfulness constraints captures the syncope and epenthesis processesthat refer toidenticalconsonants, without incorporating that reference directly into the syncopeor epenthesis rules themselves. Furthermore, this account explains why epenthesis and syncopeoccur only whenthe flanking consonants are identical, an insight that was missingin previousanalyses. Finally, it obviates the need for V /Csegregationin languages that otherwise showno evidencefor it.In the following section I returnto thedistinction between fake and true geminates.

4Fakeversus True Geminates Oneof the intriguing results to emerge from autosegmentaltheory was thedifference between fakeand true geminates. Fake geminates arise in the course of thederivation as morphemes are concatenatedor if syncope applies. True geminates are underlying or maybe produced through totalassimilation of oneconsonant to another. A truegeminate is doubly linked (41a), whereas afakegeminate is a sequenceof two identical consonants (41b). (41) a. d b. d d

C C C C Inthis article I havetaken the position that the output of syncopebetween identical consonants producesa geminate,which violates the constraint N O-GEM.As statedearlier, this is inline with phoneticevidence. In this section I willexamine two classic cases of fake and true geminates andshow how the notion of morphological domains helps capture their behavior.

4.1Palestinian Arabic and OCP Domains Atypicalexample of how fake and true geminates behave differently is found in Palestinian Arabic.In this dialect epenthesis splits word-final clusters, but does not affect true geminates (42c)(Abu-Salim 1980). Fake geminates, which arise from theconcatenation of a suffixwith a consonant-finalstem, pattern like other consonants in inducingepenthesis. In (42e) the 1st person singularsuffix /-t/ isattachedto a t-finalstem and epenthesis ensues. 110 S H A R O N R O S E

(42) a. ?akl ?akil ‘food’ ! b. ?ibn ?ibin ‘son’ ! c. sitt *sitit‘ grandmother’ d. sitt-na sittna*sititna ‘ ourgrandmother’ ! e. fut-t futit*futt ‘ Ientered’ ! Theranking OCP . . NO-GEM,onemeans of capturing geminate integrity, can account for (42c), butnot (42e), for which*[futt] would be expected. The representational approach handles these casesbecause only true geminates can be subject to geminate integrity. Therepresentational approach makes no reference to morphological affiliations, but it is clearthat the two instances of [t] in[futit]belong to differentmorphemes, whereas the geminate [tt]in [sitt] is wholly contained within a singlemorpheme. As discussedin section 2, the OCP mayapply within certain domains, such as aninternal stem or aphonologicalword. If theOCP appliesonly within the basic stem in Palestinian Arabic, we canaccount for theepenthesis that appearsto disregardthe OCP. 2 6 In(42c)the OCP wouldbe violatedif epenthesisoccurred within themorpheme. The other constraints, such as N O-GEM and NO-COMPLEX-CODA,areunranked withrespect to each other.

(43) sitt ‘grandmother’

/sitt/ OCP NO-GEM NO-COMPLEX-CODA

a. sitit *!

+ b. sitt * *

As for theform futit, ifthehigh-ranked OCP appliesonly within the basic stem, it isnot violated, andepenthesis may occur. (For analternative approach to domainsof applicationusing different constraintrankings, see Buckley 1996.)

(44) futit ‘I entered’

/fut-t/ OCP NO-GEM NO-COMPLEX-CODA

+ a. futit

b. futt *! *

Thereare other cases of geminate integrity in Palestinian Arabic. The definite article /l-/ assimilatestotally to a followingcoronal. As (45d–f) illustrate,total assimilation results in a geminate.

26 Aconstraintsuch as I NTEGRITY (McCarthyand Prince 1995; ‘ ‘Noelement oftheinput has multiplecorrespondents intheoutput’ ’) couldalso be usedto capture the resistance ofgeminates toepenthesis. However, if theeffects can be derivedfrom the interaction of the OCP with other constraints, then I NTEGRITY willnot be required. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 111

(45) a. l-?amar ‘themoon’ b.l-walad ‘ theboy’ c.l-furun ‘ theoven’ d. S-Sams‘ thesun’ e.z-zalame ‘ theman’ f. t-talaata‘ Tuesday’ Assimilationconfirms phonological integration of the /l-/ prefixwith the noun stem. Guerssel (1978)argues that assimilation across a morphemeboundary serves to obliterate the boundary. Epenthesisin the middle of such a geminatewould violate the OCP: * zizalame. Thisprediction canbe tested on words beginning with two consonants. Palestinian Arabic syllable structure disallowsCCC sequences (i.e., three timing positions filled by consonants), even across words, andepenthesis normally occurs between the first twoconsonants of thethree-consonant sequence (46a–b). If thereare four consonants, it occurs between the second and third: CC iCC (46c). However,with a geminateresulting from totalassimilation (46d), epenthesis occurs after the first consonant—C iCCC—andnot internal to the geminate. (46)a. l-walad i kbiir ‘theboy is big’ b. l-walad i zåiir ‘theboy is small’ c. l-walad likbiir ‘thebig boy’ d. l-walad iz-zåiir *lizåiir/*zizåiir‘ thesmall boy’

Theexample in (46d)violates both the ban on CCC sequencesand the constraint on geminates. Assimilationis more important than obeying either of these restrictions; hence, the constraint requiringassimilation must be highlyranked. A SSIMILATION canbe loosely formatted as follows:

‘‘Inasequencecoronal 1 -coronal2 ,alignleft edge of coronal 2 gestureswith left edge of coronal 1 .’’ In (47a–b) ASSIMILATION hasfailed to apply.In (47c –d)A SSIMILATION applies,but (47c) incurs aviolationof the OCP, whichis ranked above N O-GEM.

(47) l-walad izz iir ‘the small boy’

/. . . l-ziir/ ASSIMILATION OCP NO-GEM NO-CCC

a. liz iir *!

b. lz iir *! * *

c. ziz iir *!

+ d. zz iir * *

Unlikein Palestinian (and Moroccan) Arabic, in Algerian Arabic assimilation is suppressed if theoutput would create a sequenceof threeconsonants (Guerssel 1978): /t-dr@b/ [t@dr@b] ‘you ! hit’— not*[ddr @b],which is the cognate verb form inMoroccan Arabic. This shows that the constraintagainst a triconsonantalcluster outranks A SSIMILATION inAlgerian (Guerssel (1978) 112 S H A R O N R O S E capturesit via rule ordering). In summary, it ispossible to capture the different behavior of fake andtrue geminates in PalestinianArabic without explicit reference to theirlinked representation.

4.2Tigrinya Spirantization Tigrinyaspirantization is another classic case of arepresentationaldistinction between fake and truegeminates (Schein 1981, Kenstowicz 1982, Lowenstamm and Prunet 1986). Voiceless velar stops[k k’ ] andtheir labialized counterparts [k w kw ’]arespirantized in postvocalicposition. The spirantizedversion of the velar ejective has various pronunciations ranging from [ x ] to [¿] to [?].2 7 Itranscribeit in all the following examples as [x’ ]. (48) a. k@lbi ‘dog’ b. ?a-xlab ‘dogs’ c. k’@til-u‘ hekilled’ d. té-x’@tl-i‘ you(f.sg.) kill’ Theverb forms in(49)illustrate the alternations between stops and fricatives. In theimperfective form themedial consonant is geminated as part of the paradigm. If itis a voicelessvelar, it is notspirantized. (49) PerfectiveImperfective Jussive k@f@t@ yé-x@ffét yé-xf@t ‘open’ m@x@r@ yé-m@kkér yé-mk@r ‘advise’ m@x’@z@ yé-m@kk’éz yé-mk’@z‘cut(hair)’ b@t@x@ yé-b@ttéx yé-bt@x ‘sever,snip’ Othercases of geminationare found with object clitics (Kenstowicz 1982). Vocalic clitics cause optionalgemination of the preceding consonant of the stem (50a –c).Velars geminatedby this processare not spirantized (50b –c).Consonant-initial object clitics trigger gemination if the precedingsegment is a vowel(50d –e).Again, velar geminates do notspirantize (50e). When the geminateis derivedvia complete assimilation, no spirantizationoccurs either. This is foundwith thepassive prefix /t-/,whichundergoes total assimilation to the following root consonant, as in (50f–g).

(50) a. yé-sb@r-o [yésb@rro]/[yésb@ro] ‘lethim break it’ b. yé-bt@k-o [yébt@kko]/[yébt@xo]‘ lethim sever, snip it’ c. yé-barék-o [yébarékko]/[yébaréxo]‘ lethim bless him’ d. k’@t@l-u-ni [k’@t@lunni] ‘theykilled me’ e. m@x@r-u-ka [m@x@rukka] ‘theyadvised you (m.sg.)’ f. yé-t-k@f@t [yékk@f@t] ‘letit be opened’ g. yé-t-k’@t’k’@t’ [yékk’@t’k’@t’] ‘lethim be beaten up’ cf. [t@x’@t’k’@t’e] ‘hebeat up (someone)’

27 Theejective stop[k’ ] is sometimes transcribedas [q].In southern Tigrinya /k k’/ are alsospirantized following gutturalsand glides (Denais 1990,Berhane 1991). GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 113

However,when two identical voiceless velars abut across a morphemeboundary, the first onespirantizes.

(51) a. mérax-ka‘ your(m.sg.) calf’ b. ?amlax-kum‘ your(m.pl.) god’ c. yé-baréx-ka‘ heblessesyou (m.sg.)’ d. barix-ki ‘you(f.sg.) blessed’ Spirantizationapplies to heteromorphemic or fakegeminates, but not to true geminates. First,let us considerwhy geminate consonants are immune to spirantization.There must be abanon geminate velar fricatives: *[xx]. Kirchner (1998) shows that geminate fricatives are disfavoredcrosslinguistically because greater effort isneededto maintain the constriction over alongerduration. Further support within the Ethiopian Semitic family comes from theGurage languageMuher, in whichtautomorphemic geminate /x/ isrealizedas [kk]: m kk r- ‘headvised’ versus y -mx r ‘lethim advise’ (cf. b tt x-x m ‘you(m.sg.) extracted’ vs. b tt x- m ‘he ex- tracted’). Second,there must be a constraintbanning spirantization of thefirst part of ageminate sequence:*[xk]. That constraint is availablein theform ofthe OCP. If spirantizationtakes place, theresult is a sequenceof two velars, a violationof the OCP appliedto the Dorsal place of articulation.Recall that the Semitic Place OCP applieswithin the basicstem, thus affecting strictly adjacentconsonants or thoseseparated by avowel.A geminateconsonant [kk] avoids violating theOCP. Thequestion then arises why the forms in(51) escape the OCP. Onthe basis of independent evidence regarding nasalization and stress, Lowenstamm and Prunet(1986) argue that the possessive and agreement clitics in (51) are separate phonological wordsfrom thestems and that the (traditional) OCP operatesonly within the phonological word. Thetwo instances of /k/ thatare adjacent across a phonologicalword boundary do not violate theOCP andhence are not forced to fuse into a truegeminate to escape an OCP violation. Spirantizationfreely applies to thefirst /k/ asit has no connectionto the second. This is another instanceof the OCP beingrestricted to a givendomain. Furthermore, it was shownin section 2 thatthe guttural OCP alsoonly applies within the basic stem in Tigrinya. Tableau (52) for the jussiveform ‘lethim bless you’ shows the inapplicability of the OCP tothetwo adjacent velars. Theconstraint *[Vk] banspostvocalic velar stops.

(52) y bar xka ‘let him bless you (m.sg.)’ /y -bar k-ka/ *[xx] OCP *[Vk]

a. y bar x-xa *!

b. y bar k-ka *!

+ c. y bar x-ka

Letus now turn to the other cases of gemination derived via assimilation or prolongation (spreading)of the single consonant over two timing positions. As inPalestinian, in Tigrinya assimilationsignifies phonological incorporation into the basic stem. Spirantization would then causean OCP violation.It is less clear whether the geminate [kk] in forms suchas barix-u-kka 114 S H A R O N R O S E

‘heblessedyou (m.sg.)’ is within a singleOCP domain.Regardless of thisquestion, the gemination triggeredby the object clitics reflects a morphologicalgemination requirement on a parwith stem-internalgemination. This is shownby thefact that, at least for somespeakers, gemination ofconsonant-initial clitics occurs with the gerundive (used to express the past tense) (53a) but notwith perfective or imperfectiveforms (53b–c).

(53)a. barix-u- kka‘ heblessed you(m.sg.)’ b. yé-baréx-u-xa‘ theybless you(m.sg.)’ c. bar@x-u-xa‘ theyblessed you(m.sg.) (punctual)’

We cancall this gemination-requiring constraint M ORPHOLOGICAL GEMINATION (MORPH-GEM). As (54)shows, any half-spirantization will cause this constraint to be violated.

(54) barixukka ‘he blessed you (m.sg.)’

/barix-u-ka/ *[xx] MORPH-GEM *[Vk]

a. barixuxxa *!

b. barixuxa *! *

+ c. barixukka *

d. barixuxka *!

Thisanalysis whereby the OCP appliesonly within the domainof thebasicstem also predicts thata biliteralverb with a finaldoubled /k/ shouldnot allowhalf-spirantization if the two instances of /k/ occurin string-adjacentposition within the stem. This is confirmedby the verb ‘ toscratch off’, formedfrom theroot /Ék/. (55) PerfectiveImperfective a. Éax@x-@ yé-Éakkéx ‘hescratches (off)’ b. Éax@x-u yé-Éakk-u ‘theyscratch (off)’ Itcan be inferred that the geminate [kk] in y - akk-u isnot the same as that in the 3rd person masculinesingular form, butis formed from thebase and reduplicant consonants occurring in thesecond and third positions of the base. This is by analogy to regular verbs that take the imperfectiveverb stem C CCwhenfollowed by a vowel-initialsubject suffix (e.g., y -s br-u ‘theybreak’ but y -s bb r ‘hebreaks’; seeBuckley 1997a for ananalysis of thealternation). In allthe forms exceptthe imperfective 3rd person masculine plural, the /k/ isspirantizedaccording tothe requirement of postvocalic spirantization. The spirantization requirement even outweighs base-reduplicantidentity between reduplicated velars. However, with y - akk-u thereis anoppor- tunityto avoid the OCP violationby realizing both velars as [k] andforming a geminate.This motivatesthe ranking OCP . . *[Vk],shown in (56). GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 115

(56) y hakku ‘they scratch off’

/y -hakiki-u/ *[xx] OCP *[Vk] a. y haxx-u *!

b. y haxk-u *!

+ c. y hakk-u *

Withthe other verb forms inthe paradigm in (55), all of the verb forms willviolate the OCP whetherthey spirantize or not.This is becausethe OCP appliedto consonantsalso operates across interveningvowels. Therefore, spirantization applies to satisfy*[Vk], even outweighing I DENTB R . Insummary, the notion of domainin theapplication of theOCP playsa rolein determining spirantizationin Tigrinya, supporting the claims made by Lowenstamm and Prunet (1986).

4.3Conclusion Thereare several different options for dealingwith fake geminates that arise in the course of a derivation.One, the two consonants may reside in separate domains and have no impacton each other(Tigrinya spirantization). Two, they are separated by epenthesis, which may either be a generaloperation within the language (Palestinian Arabic) or occur specifically to avoid gemi- nates.Three, one of theconsonants may be deleted. This occurs if thelanguage strictly enforces thegeminate ban, as does Burum. If noneof theseprocesses occur, and adjacent identical conso- nantsare allowed to surface, they are treated as true geminates and will violate N O-GEM.

5Non-SemiticExamples and Reduplication Withthe exception of Afar, mostof the languages discussed in this article have been Semitic, andeven Afar belongsto theAfro-Asiatic language family. However, the generality of constraints suchas the OCP andN O-GEM issuch that their effects should be foundin otherlanguages. One predictionthe present analysis makes is thatmany cases of reduplication might violate the OCP, particularlythose that reduplicate a singleopen syllable. Yip (1998) assumes that the OCP can beoutrankedby constraintsrequiring reduplication (e.g., R EPEAT:‘‘Theoutput must contain two identicalelements’ ’), ora constraintpertaining to morphologicalrealization, forcing reduplication tooccurto expressmorphological features, even if in sodoing the OCP isviolated. In this section Ishowhow some languages may restrict the size of the reduplicant to avoid violations of both the OCP and NO-GEM.Ifirstexamine languages that avoid geminates and then turn to ones that avoidviolating the OCP.

5.1Reduplicants Avoid Geminates Urbanczyk(1996:283) shows that reduplication in Lushootseed resists adjacent sequences of identicalconsonants— inmy terms,geminates. The distributivereduplicant is usuallyof the shape CVC(57a– b), but when the first two consonants of the stem are identical, the shape is CV (57c–e). 116 S H A R O N R O S E

(57) Base Distributive a. ?@pu´s ‘aunt’ ?@p-?@pu´s ‘aunts’ b . l@ w @b‘youth,young man’ l @ w -l@ w @b‘youths,young men’ c. l@lwa´?s@d‘sleepingplatform’ l @ -l@lwa´?s@d‘sleepingplatforms’ d. w´õ w’su ‘ children;little’ w´õ -wiw’su ‘ little(pl.)’ e. c’´õ c’al ‘ longfeathers with c’ ´õ -c’´õ c’al-b ‘ sproutedwings’ thickstems’ Lushootseedgenerally bans geminates, allowing only w w , qq, and ll, butmost geminates arereduced to a singleconsonant. Urbanczyk argues that an OCP constraint(here applied in the traditionalway to identical consonants resulting from reduplication)and a N O-LINK (4 NO-GEM) constraintare both necessary to rule out an adjacent sequence of identical consonants. As shown in(58),these are both ranked above a M AX constraintapplied to thebase-reduplicant relationship.

(58) c’í-c’ íc’ al-b ‘sprouted wings’

/RED + c’ic’ al/ NO-LINK OCP MAXBR(Root)

a. c’ic’ c’ ic’ al-b *! **

b. c’ic’ c’ ic’ al-b *! ** + c. c’ic’ ic’ al-b ***

Underthe analysispresented here, no distinctionis necessarybetween the fake and true geminates foundin candidates (58a) and (58b); N O-GEM issimply ranked above the OCP, butthe OCP is interpreteddifferently, as applying across vowels. Both candidates violate the OCP inthat they containtwo instances of identical consonants separated by a vowel,but this constraint proves irrelevantwhen faced with high-ranking N O-GEM,asshown in (59).

(59) c’í-c’ íc’ al-b ‘sprouted wings’

/RED + c’ic’ al/ NO-GEM OCP MAXBR

a. c’ic’c’ ic’ al-b *! ***! **

+ b. c’ic’ ic’ al-b ** *****

Thus,Urbanczyk’ s analysis(reinterpreted in (59)) providesanother example of a languagethat restrictsgeminates, but tolerates OCP violations. Aparallelexample is found in Burum(Gasaway 1997).As statedin section3, no geminates arefound in this language, and any arising via morpheme concatenation are simplified. This is trueof reduplicants as well: the normal bimoraic reduplicant shape seen in (60a –b)is reduced in(60c– d) toprevent gemination. (60)a. toko ‘ tomeet’ tokotoko ‘meeting’ b.mal ‘ tolive’ malmal ‘life’ c.korak ‘ earwax’ korakorak *korakkorak ‘ earwax’ d.tat ‘ tosit’ tatat *tattat ‘sitting’ GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 117

Spaelti(1997) provides other cases of geminateresistance in thereduplication of WestTarangan dialectsof Indonesia.

5.2Reduplication Avoids OCP Violations Thereverse of the Lushootseed and Burum situation is found in Nukuoro (Carroll and Soulik 1973)cited by Odden (1988). In this language intensive reduplication takes the prefixal shape CVCV unlessthe first twoconsonants are identical, in which case the shape is CVC.Maximal reduplicationis not achieved, not because of a banon geminates, but because of the OCP. In (61a–c) normalreduplication occurs. In (61d –g)the CVC shape is found. (61) Base Intensive a.balavini bala-balavini ‘ awkward’ b. badai bada-badai ‘meddlein others’ affairs’ c. gili gili-gili ‘coralrubble; skin in an unhealthy condition’ d. bobo bob-bobo ‘rotten’ e. lili lil-lili ‘easilyangered’ f. sasa sas-sasa ‘stammerfrom nervousness’ g. nanu nan-nanu ‘complainrepeatedly’ Theform in(61c) shows that the reduction of the CVCV shapeis notthe result of aconstraint againstcomplete identity between base and reduplicant, such as *R EPEAT (Yip1998). Instead, theCVC shape appears only when the first twoconsonants of the base are identical. Even the vowelsmay be nonidentical,as in (61g). Tableau(62) illustrates how creating a geminatereduces the number of OCP violations. 2 8

Havingonly one syllable reduplication violates M AXB R inthattwo segments are not copied (62c), oriftemplatic constraints are imposed on thesize of thereduplicant,it doesnot achievea bimoraic minimum.

(62) bobbobo ‘rotten’

/RED + bobo/ OCP MAXBR NO-GEM

a. bobo-bobo ***!

+ b. bob-bobo ** * *

c. bo-bobo ** **!

Asimilarexample is found in Nakanai (Johnston 1980:257, Spaelti 1997). When a CV syllableis reduplicatedeither once or twice, syncope occurs to createa geminate.The syncopated vowelis the one that occurs between the two identical consonants. If thereis more than one

28 Analternative explanation would invoke the constraint I NTEGRITY,as inBuckley’ s (1998)account for a similar effect inManam. 118 S H A R O N R O S E reduplicant,the syncopated vowel is eitherpretonic or, if stress occurs elsewhere, the vowel of thesecond reduplicant. (63) a. /lololo/ [llo´lo] ‘listening’ b. /lololoau/ [lolloa´u] ‘listeningto me’ ´ c. /mememe/ [mmEmE] ‘shiny’ ´ d. /mememeti/ [mEmmEtI]‘alreadyshiny’ e. /matatutuluti / [matattulu´tI]‘alreadysleepy-eyed’ ´ f. /uru pepeho/ [uru ppEho]‘ verybig’

Inconclusion, these examples show the varied ways that the OCP andN O-GEM may be rankedin different languages, and how they may interact to affect the size of reduplicants.

6Conclusion Ibeganthis article by examininga puzzlingdistribution problem in two North Ethiopian Semitic languageswith respect to guttural consonants. At firstglance it appeared to provide support for adistinctionbetween repetition of consonants via long-distance gemination and repetition via reduplication.However, I showedthat the long-distance geminate analysis cannot be upheld, becausethe repetition involves different kinds of gutturals.To handle the resistance of gutturals toappearing within roots or stems when a vowelintervenes, but not when a consonantdoes, I proposedthat the OCP shouldbe ableto applyacross intervening vowels within certain domains (theprosodic word in Tigre, the basic stem in Tigrinya). This proposal has immediate conse- quencesfor theanalysis of antigeminationeffects in Semiticlanguages and others that crucially relieson an interpretation of the OCP appliedto strictly adjacent consonants and not those sepa- ratedby a vowel.Yet this new interpretation of the OCP actuallyhandles the antigemination casesin a moreexplanatory fashion, by justifying the role of identical consonants in syncope applicationand resistance. I reanalyzeddata from ClassicalArabic and Afar, relyingon this alternativeinterpretation of theOCP anda generalconstraint against geminates within OT. Ialso extendedthe analysis to cases of reduplication that restrict the size of the reduplicant to avoid geminationor OCP violations.

References Abu-Salim,Issam. 1980. Epenthesis and geminate consonants in PalestinianArabic. Studiesin theLinguistic Sciences 10(2):1–11. Alderete,John. 1997. Dissimilation as localconjunction. In NELS 27, 17–32.GLSA, Universityof Massachu- setts,Amherst. Angoujard,Jean-Pierre. 1988. Ge ´minationet redoublement (Gemination and doubling). LanguesOrientales AnciennesPhilologie et Linguistique 1:1–15. Banksira,Degif Petros. 1997. The sound system of Chaha. Doctoral dissertation, Universite ´ du Que´bec a` Montre´al.[To appear as Themorphophonology of Chaha. TheHague: Holland Academic Graphics.] Bassano,Francesco da. 1918. Vocabolariotigray-italiano e repertorioitaliano-tigray (Tigray-Italianvocabu- laryand Italian-Tigray index). Rome: Casa Editrice Italiana. Bat-El,Outi. 1988. Remarks on TierConflation. LinguisticInquiry 19:477–485. GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 119

Berhane,Girmay. 1991. Issues in the phonology and morphology of Tigrinya. Doctoral dissertation, Uni- versite´ du Que´bec a` Montre´al. Berkley,Deborah. 1994. The OCP andgradient data. Proceedingsof FLSM V.Studiesin the Linguistic Sciences 24(1/2):59–72. Bliese,Loren. 1981. Agenerativegrammar of Afar. Arlington,Tex.: Summer Institute of Linguistics. Boersma,Paul. 1998. Functionalphonology: Formalizing the interactionsbetween articulatory and percep- tualdrives. (LandelijkeOnderzoekschool Taalwetenschap 11.) The Hague: Holland Academic Graphics. Brame,Michael. 1970. Arabic phonology: Implications for phonological theory and historical Semitic. Doctoraldissertation, MIT, Cambridge, Mass. Buckley,Eugene. 1990. Edge-in association and OCP ‘‘violations’’ inTigrinya. In Proceedingsof the NinthWest Coast Conference on Formal Linguistics, 75–90.Stanford, Calif.: CSLI Publications. [Distributedby Cambridge University Press.] Buckley,Eugene. 1996. Constraint domains in Kashaya. In Proceedingsof the Fourteenth Annual West CoastConference on Formal Linguistics, 47–61.Stanford, Calif.: CSLI Publications.[Distributed byCambridgeUniversity Press.] Buckley,Eugene. 1997a. Alignment and weight in the Tigrinya verb stem. Paper presented at the 28th AnnualConference on AfricanLinguistics. [To appear in the proceedings.] Buckley,Eugene. 1997b. Tigrinya root consonants and the OCP. In Pennworking papers in linguistics 4.3, 19–51. Penn Linguistics Club, University of Pennsylvania, Philadelphia. Buckley,Eugene. 1998. Integrity and correspondence in Manam double reduplication. In NELS 28, 59–67. GLSA, Universityof Massachusetts, Amherst. Carroll,Vern, and Topias Soulik. 1973. Nukuorolexicon. (PALI LanguageTexts.) Honolulu: University of HawaiiPress. Clements,G. N.,and Elizabeth Hume. 1995. The internal organization of speechsounds. In Thehandbook ofphonologicaltheory, ed.John Goldsmith, 245 –306.Cambridge, Mass.: MIT Press. Davy,J. I.M.,and Derek Nurse. 1982. Synchronic versions of Dahl’s Law:The multiple applications of a phonologicaldissimilation rule. Journalof AfricanLanguages and Linguistics 4:157–195. Denais,Michel. 1990. E ´ le´mentsde phonologieet de morphologietigrigna (Elements of Tigrinyaphonology andmorphology). The `sedu doctorat (nouveau re ´gime),Universite ´ deNiceSophia Antipolis. Elmedlaoui,Mohamed. 1995. Ge ´ometriedes restrictions de cooccurrencede traitsen se ´mitiqueet en berbe`re: Synchronieet diachronie (Geometry of feature cooccurrence restrictions in Semitic and Berber: Synchronyand diachrony). Revuecanadienne de linguistique 40:1–126. Foley,William. 1991. TheYimas language of New Guinea. Stanford,Calif.: Stanford University Press. Frisch,Stefan, Michael Broe, and Janet Pierrehumbert. 1997. Similarity and phonotactics in Arabic.Rutgers OptimalityArchive ROA-223, http: //ruccs.rutgers.edu /roa.html. Gafos,Adamantios. 1996. The articulatory basis of locality in phonology.Doctoral dissertation, The Johns HopkinsUniversity, Baltimore, Md. Gafos,Diamandis [Adamantios]. 1998. Eliminating long-distance consonantal spreading. NaturalLan- guage& LinguisticTheory 16:223–278. Gasaway,Eileen. 1997. Burum morphophonemics. In Papersin Papuan linguistics 3, ed.Andrew Pawley, 59–99. (Pacific Linguistics Series A-87.) Canberra: The Australian National University. Gnanadesikan,Amalia. 1997. Phonology with ternary scales. Doctoral dissertation, University of Massachu- setts,Amherst. Goldsmith,John. 1976. Autosegmental phonology. Doctoral dissertation, MIT, Cambridge, Mass. [Published New York:Garland, 1979.] Greenberg,Joseph. 1950. The patterning of root morphemes in Semitic. Word 6:162–181. Guerssel,Mohamed. 1978. A conditionon assimilation rules. LinguisticAnalysis 4:225–254. 120 S H A R O N R O S E

Hayes,Bruce. 1986. Inalterability in CVphonology. Language 62:321–351. Hoberman,Robert. 1988. Local and long-distance spreading in Semiticmorphology. NaturalLanguage & LinguisticTheory 6:541–577. Hume,Elizabeth. 1992. Front vowels, coronal consonants and their interaction in nonlinear phonology. Doctoraldissertation, Cornell University, Ithaca, N.Y. [PublishedNew York:Garland, 1994.] Hume,Elizabeth, and David Odden. 1996. Reconsidering [consonantal]. Phonology 13:345–376. Inkelas,Sharon, and Young-Mee Cho. 1993. Inalterability as prespecification. Language 69:529–574. Itoˆ,Junko,and Armin Mester. 1998. Markedness and word structure: OCP effectsin Japanese. Rutgers OptimalityArchive ROA-255, http: //ruccs.rutgers.edu /roa.html. Johnston,Raymond. 1980. Nakanaiof New Britain:The grammar of anOceaniclanguage. (PacificLinguis- ticsSeries B.70.) Canberra: The Australian National University. Keer,Edward. 1998. Spirantization and geminate inalterability. In Workingpapers from Rutgers University 1, 147–167.Department of Linguistics, Rutgers University, New Brunswick,N.J. Kenstowicz,Michael. 1982. Gemination and spirantization in Tigrinya. Studiesin the Linguistic Sciences 12(1):103–122. Kirchner,Robert. 1998. An effort-based approach to consonantlenition. Doctoral dissertation, UCLA, Los Angeles,Calif. Rutgers Optimality Archive ROA-276, http: //ruccs.rutgers.edu /roa.html. Kisseberth,Charles. 1970. Vowel elision in Tonkawa and derivational constraints. In Studiespresented to RobertB. Leesby his students, ed.Jerrold Saddock et al., 109 –138.Champaign, Ill.: Linguistic Research. Leben,William. 1973. Suprasegmental phonology. Doctoral dissertation, MIT, Cambridge, Mass. Leslau,Wolf. 1979. Etymologicaldictionary of Gurage(Ethiopic). Wiesbaden:Otto Harrassowitz. Lombardi,Linda. 1995. Why Place and Voice are different: Constraint interactions and feature faithfulness inOptimalityTheory. Rutgers Optimality Archive ROA-105, http: //ruccs.rutgers.edu /roa.html. Lowenstamm,Jean, and Jean-Franc ¸oisPrunet. 1986. Le tigrinya et le principe du contour obligatoire (Tigrinyaand the Obligatory Contour Principle). Revue que´be´coisede linguistique 16:181–208. Marlett,Stephen. 1981. The structure of Seri. Doctoral dissertation, University of California, San Diego. Marlett,Stephen, and Joseph Stemberger. 1983. Empty consonants in Seri. LinguisticInquiry 14:617–639. McCarthy,John. 1979. Formal problems in Semitic phonology and morphology. Doctoral dissertation, MIT, Cambridge,Mass. McCarthy,John. 1981. A prosodictheory of nonconcatenativemorphology. LinguisticInquiry 12:373–418. McCarthy,John. 1983. Consonantal morphology in the Chaha verb. In Proceedingsof theSecondWest Coast Conferenceon Formal Linguistics, 176–188.Stanford Linguistics Association, Stanford University, Stanford,Calif. McCarthy,John. 1986a. Lexical phonology and nonconcatenative morphology in thehistoryof Chaha. Revue que´be´coisede linguistique 16:209–228. McCarthy,John. 1986b. OCP effects:Gemination and antigemination. LinguisticInquiry 17:207–263. McCarthy,John. 1988. Feature geometry and dependency. Phonetica 43:84–108. McCarthy,John. 1994. The phonology and phonetics of Semitic pharyngeals. In Phonologicalstructure andphonetic form: Papers in laboratoryphonology III, ed.Patricia Keating, 191 –234.Cambridge: CambridgeUniversity Press. McCarthy,John, and Alan Prince. 1990. Prosodic Morphology and templatic morphology. In Perspectives onArabiclinguistics I, ed.Mushira Eid and John McCarthy, 1– 54. Amsterdam: John Benjamins. McCarthy,John, and Alan Prince. 1993. Prosodic Morphology I: Constraintinteraction and satisfaction. Ms.,University of Massachusetts,Amherst, and Rutgers University, New Brunswick,N.J. McCarthy,John, and Alan Prince. 1995. Faithfulness and reduplicative identity. In Universityof Massachu- settsoccasional papers in linguistics 18, 225–247.GLSA, Universityof Massachusetts,Amherst. Miller,Ann. 1987. Phonetic characteristics of Levantine Arabic geminates with differing morpheme and GEMINATES,LONG-DISTANCEGEMINATES,ANDTHEOCP 121

syllablestructures. In TheOhio State University working papers in linguistics36: Papers from the LinguisticsLaboratory 1985 –1987, 120–140. Department of Linguistics,The Ohio State University, Columbus. Myers,Scott. 1997. OCP effectsin OptimalityTheory. NaturalLanguage & LinguisticTheory 15:847–892. N´õ Chiosa´in, Ma´ire,and Jaye Padgett. 1997. Markedness, segment realisation, and locality in spreading. RutgersOptimality Archive ROA-188, http: //ruccs.rutgers.edu /roa.html. Odden,David. 1988. Anti antigemination and the OCP. LinguisticInquiry 19:451–475. Odden,David. 1994. Adjacency parameters in phonology. Language 70:289–330. Orgun,Orhan. 1996. Sign-based morphology and phonology with special attention to Optimality Theory. Doctoraldissertation, University of California, Berkeley. Padgett,Jaye. 1992. OCP subsidiaryfeatures. In NELS 22, 335–346.GLSA, Universityof Massachusetts, Amherst. Pierrehumbert,Janet. 1993. Dissimilarity in theArabic verb roots. In NELS 23, 367–381.GLSA, University ofMassachusetts,Amherst. Prince,Alan, and Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Ms.,Rutgers University, New Brunswick,N.J., and University of Colorado, Boulder. Prunet,Jean-Franc ¸ois.1990. The origin and interpretation of Frenchloans in Carrier. InternationalJournal ofAmericanLinguistics 56:484–502. Prunet,Jean-Franc ¸ois.1996. Some core properties of Semitic morphology: Evidence from the Far South. In Currenttrends in phonology, 617–652.European Studies Research Institute, University of Salford. Prunet,Jean-Franc ¸ois,and Degif Petros (Banksira). 1996. L’ interactionentre sche `meset racines en chaha (Theinteraction between templates and roots in Chaha). In Studiesin Afroasiatic grammar, ed. JacquelineLecarme, Jean Lowenstamm, and Ur Shlonsky, 302 –336.The Hague: Holland Academic Graphics. Raz,Shlomo. 1983. Tigregrammar and texts. Malibu,Calif.: Undena Publications. Rose,Sharon. 1992. De la palatalisationen chaha(On palatalization in Chaha). Master’ s thesis,Universite ´ du Que´bec a` Montre´al. Rose,Sharon. 1994. Palatalization, underspecification and plane conflation in Chaha. In Proceedingsof the TwelfthWest Coast Conference on FormalLinguistics, 101–116.Stanford, Calif.: CSLI Publications. [Distributedby Cambridge University Press.] Rose,Sharon. 1996. Variable laryngeals and vowel lowering. Phonology 13:73–117. Rose,Sharon. 1997. Theoretical issues in comparative Ethio-Semitic phonology and morphology. Doctoral dissertation,McGill University, Montreal, Quebec. Rose,Sharon. To appear. Multiple correspondence in reduplication. In Proceedingsof the Twenty-third AnnualMeeting of the Berkeley Linguistic Society. Schein,Barry. 1981. Spirantization in Tigrinya.In MITworkingpapers in linguistics 3: Theoreticalissues inthe grammar of Semitic languages, 32–41.MITWPL, Departmentof Linguisticsand Philosophy, MIT,Cambridge, Mass. Schein,Barry, and Donca Steriade. 1986. On geminates. LinguisticInquiry 17:691–744. Soden,Wolfram von. 1969. Erga¨nzungsheftzum Grundriss der Akkadischen Grammatik. (Analectaorientalia 33.2.)Rome: Pontificium Institutum Biblicum. Spaelti,Philip. 1997. Dimensions of variationin multi-patternreduplication. Doctoral dissertation, University ofCalifornia,Santa Cruz. Steriade,Donca. 1987. Locality conditions and feature geometry. In Proceedingsof NELS 17, 595–618. GLSA, Universityof Massachusetts, Amherst. Suzuki,Keiichiro. 1998. A typologicalinvestigation of dissimilation. Doctoral dissertation, University of Arizona,Tucson. Urbanczyk,Suzanne. 1996. Patterns of reduplication in Lushootseed. Doctoral dissertation, University of Massachusetts,Amherst. 122 S H A R O N R O S E

Urbanczyk,Suzanne. 1998. Avoidance of the marked. Rutgers Optimality Archive ROA-268, http: //ruccs.rutgers.edu /roa.html. Walker,Rachel. 1998. Nasalization, neutral segments and opacity effects. Doctoral dissertation, University ofCalifornia, Santa Cruz. Yip,Moira. 1988. The Obligatory Contour Principle and phonological rules: A lossof identity. Linguistic Inquiry 19:65–100. Yip,Moira. 1989. Feature geometry and co-occurrence restrictions. Phonology 6:349–374. Yip,Moira. 1998. Identity avoidance in phonology and morphology. In Morphologyand its relation to phonologyand syntax, ed.Steven Lapointe, Diane Brentari, and Patrick Farrell, 216– 246. Stanford, Calif.:CSLI Publications.[Distributed by Cambridge University Press.] Younes,R. 1983.The representation of geminateconsonants. Ms., University of Texas, Austin. Zawaydeh,Bushra Adnan. 1997. On an optimality-theoretic account of epenthesis and syncope in Arabic dialects.In Perspectiveson Arabiclinguistics X, ed.Mushira Eid and Robert R. Ratcliffe,191– 213. Amsterdam:John Benjamins.

Departmentof Linguistics Universityof California,San Diego 9500Gilman Drive LaJolla, California 92093-0108 [email protected]