Heteroclisis and Paradigm LinkageStump Gregory T Gregory Thomas 1 - PDF document

Heteroclisis and Paradigm LinkageStump, Gregory T. (Gregory Thomas), 1954-Language, Volume 82, Number 2, June 2006, pp. 279-322 (Article)Published by Linguistic Society of AmericaDOI: 10.1353/lan.2006.0110For additional information about this article Access Provided by University of Kentucky at 10/18/12 9:30PM GMT http://muse.jhu.edu/journals/lan/summary/v082/82.2stump.html HETEROCLISISANDPARADIGMLINKAGET.SUniversityofKentuckyHeteroclisisisthepropertyofalexemewhoseinflectionalparadigminvolvestwoormoredistinctinflectionclasses.Althoughheteroclisisiswidelyobservable,itsimplicationsforgrammat-icaltheoryremainunderexplored,perhapsbecauseitscanonicalinstanceshavetheappearanceofsporadiclexicalexceptions.Butheteroclisiscannotbeassumedtolackanyroleinthedefinitionofalanguagesmorphology,since(i)itissometimeshighlysystematic,involvingwholeclassesoflexemes,and(ii)itobeysauniversalconstraint.Thesetwofactsshowthatheteroclisisisrule-governed.Ontheassumptionthatinflectionalmorphologyinvolvesalinkageofcontent-paradigmswithform-paradigms(Stump2002),heteroclisiscanbeseenasakindofmismatchregulatedbyrulesofparadigmlinkage.Suchrulesaccountfortherangeofempiricalphenomenasubsumedbyobservations(i)and(ii).*.Thetermreferstothepropertyofalexemewhoseinflectionalparadigmcontainsformsbuiltuponstemsbelongingtotwoormoredistinctinflectionclasses.Consider,forexample,thedeclensionoftheCzechnominallexemespring,sourceinTable1.Inthesingularportionofitsparadigm,inflectsasamemberofthesoft-masculinedeclensionexemplifiedbyinthepluralportionofitsparadigm,itinflectsasamemberofthehard-masculinedeclensionexemplifiedbyHeteroclisisisawidelyobservablephenome-noninnaturallanguage:itisnotrestrictedtolexemesbelongingtoanyparticularsyntacticcategory,norareIndo-Europeanlanguagestheonlysourceofexamples.Indeed,languagesthathaveinflection-classdistinctionstendtoexhibitheteroclisis;Table2listsasamplingofnon-Indo-Europeanlanguagesexemplifyingthistendency.(See§5.2forspecificdiscussionofHausaandNgiyambaa,§5.4forChukchi,§§3.1,5.2,5.4forFula,and§§5.2,5.5forTakelma.)Despiteitswidespreadincidence,theimplicationsofthephenomenonofheteroclisisforgrammaticaltheoryremainrelativelyunexplored.Perhapsthisisbecausethecanoni-calinstancesofheteroclisishavetheappearanceofsporadiclexicalexceptionsandhavethereforegenerallybeendismissedashavingnosystematicroleinthedefinitionofalanguagesmorphology.Thatis,perhapsthepremisesoftheLEXICALAPPROACHtoheteroclisisin1havebeenassumedtoaffordanadequateaccou

ntoftheobservedpropertiesofheterocliteparadigms.(1)Thelexicalapproachtoheteroclisis:Heteroclisisisalexicalphenome-non„theeffectofalexicalstipulationassociatingaparticularstemandinflectionclasswithparticularsetsofmorphosyntacticproperties.*IwishtothankDunstanBrown,GrevCorbett,LauraJanda,BrianJoseph,andTomStewartfornumeroushelpfulcommentsonanearlierdraftofthisarticle;DavidForis,GeoffKimball,andConstanceKutschLojengaforhelpfulinformationontheinflectionalsystemsofSochiapanChinantec,Koasati,andNgiti,respectively;TanyaIvanova-Sullivan,HelenaRiha,andAnastasiaSmirnovafortheirhelpwiththeBulgariandata;myresearchassistantVirginiaHenriksenforhelpinidentifyinginstancesofheteroclisisandinconstruct-ingadatabaseofexamples;mycolleagueRaphaelFinkelforputtingthedatabaseonline;andmycolleaguesintheEnglishDepartmentattheUniversityofKentuckyforgrantingmeaWilliamJ.andNinaB.TuggleResearchProfessorshipfor2002,whichfacilitatedthepursuitofthisresearch.ThedistinctionbetweensoftandhardstemsisonlyvestigiallyphoneticinCzech.Nominalstemsendinginapalatalconsonantorthedentalaffricatealwaysfollowthesoftdeclension,whilestemsendinginotherconsonantsusuallyfollowtheharddeclension,thoughwithexceptions(Heim1982:20).Initsoriginaluseinthestudyoftheclassicallanguages,heteroclisisreferredspecificallytoirregularitiesofGreekandLatindeclension;here,Iassumethemodern,moregeneralsenseoftheterm. LANGUAGE,VOLUME82,NUMBER2(2006) room spring bridge DECLENSION:SOFT-MASCULINEHARD-MASCULINE GEN moste GENVOCLOC pokojupokojupokojepramenymostymostumostu 1.HeterocliteinflectionofCzechspring,source(Heim1982:22,41f.,176).SAMPLEHETEROCLITEFAMILYLANGUAGELEXEMESOURCEAfro-AsiaticHausatotearoffNewman2000:708AltaicMongolianwoodGrønbech&Krueger1955:24AustralianNgiyambaatowalkDonaldson1980:158Chukotko-KamchatkanChukchifriendSpencer1999Niger-CongoFulawindArnott1970:120f.Nilo-SaharanFurtoaskBeaton1968:127NorthCaucasianTsakhurtorememberKibrik1999:62,881Oto-MangueanZapotectoaskButlerH[aworth]1980:78PenutianTakelmatoarriveSapir1922:166SiouanLakotatogoBuechel1939:75UralicMari(negativeauxiliary)Kangasmaa-Minn1998:2322.Somenon-Indo-Europeanlanguagesexhibitingheteroclisis.Accordingtothisapproach,theheterocliteinflectionofisaneffectofthelexicalstipulationin2,wheretheinflectionofasamemberofthesoft-masculinedeclensionisassociatedwithpropertysetscontainingsingularandtheinflectionofasamemberofthehard-masculinedeclensionisassociatedwithpropertysetscontainingplural.(2)StemInflectionclassSetsofmorphosyntacticpropertiessoft

-masculinedeclensionallwell-formedsetscontaininghard-masculinedeclensionallwell-formedsetscontainingWhilethelexicalapproachmightseemtobeadequatetoaccountforthespecificexampleof,theconclusionthatIasserthereisthatthisapproachisinadequateforanygeneralaccountofheteroclisis.Ontheempiricalplane,Idiscusstwofactsaboutheteroclisisthatarenotaccountedforbythelexicalapproach:(i)thefactthatheteroclisisissometimeshighlysystematic,involvingwholeclassesoflexemes(§4),and(ii)thefactthatheteroclisisuniversallyobeysanobservableconstraint(§5).Iarguethatanadequateaccountofthesetwofactscannotbeachievedbymeansofpiecemeal HETEROCLISISANDPARADIGMLINKAGE281lexicalstatementssuchas2,butmustinsteadinvolvemorphologicalrulesofamoregeneralsort.Onthetheoreticalplane,IdeveloptheapproachtoheteroclisisproposedinStump2002,accordingtowhichheteroclisisisonekindofmismatchbetweenalanguagescontent-paradigmsanditsform-paradigms(§3).Inthisapproach,hetero-clisisisregulatedbyrulesofparadigmlinkage;theseaffordasimpleandexplanatoryaccountoftherangeofempiricalphenomenasubsumedby(i)and(ii).Beforeproceed-ing,itisusefultoconsiderthedefiningpropertiesofheteroclisisatsomewhatgreaterlength(§2),withparticularattentiontotherelationbetweenheteroclisisandother,superficiallysimilarphenomena.OMEBASICCHARACTERISTICSOFHETEROCLISIS.IntheCzechparadigmsinTable1,thelexemesspringandroominflectalikeinsomecellsbutdiffer-entlyinothers;thispartialsimilaritybetweenparadigmsisanessentialeffectofhetero-clisis.Itisimportanttorecognize,however,thatsuchpartialsimilarityisnotalwaysaneffectofheteroclisis:itmayinsteadbeaneffectofdefaultinflectionalpatterns.Consider,forexample,theVedicSanskritparadigmsinTable3:theparadigmofthefemininenoungoddessandthefeminineparadigmoftheadjectiveTheseparadigmsexhibitdifferentinflectionalcharacteristicsinsomecells,butareinflectionallyalikeintheshadedcells. SINGULARDUALPLURAL devy-ái ´sdevy- ´sdevy- ¯-sÚCI´úci-s´úce´úci-m´úcy-a¯ ~ s´úcš¯ ~ s´úci´úcay-e´úce-s´úce-s´úca´úcau´úcš´úcš´úcš´úci-bhya´úci-bhya´úci-bhya´úcy-os´úcy-os´úcay-as´úcay-as´úcš¯-s´úci-bhis´úci-bhyas´úci-bhyas´úcš´úci-s 3.DeclensionoftwoVedicSanskritnominals(Macdonell1916:80ff.,87ff.).Wordsexpressingthesamemorphosyntacticpropertysetareshadediftheyinflectasmembersofthesameinflectionclass.Onecouldportraythisoverlapasaninstanceofheteroclisisbyarguingthatthe-stemdeclensionoffollowsthefeminine-stemdeclensionofcertaincells(orviceversa).Thiswouldbeaquestionablestance,however.ThereisnocompellingreasontoregardthesharedinflectionalpatternsinTable3asbasictoth

efeminine-stemdeclensionandintrusiveinthederivative-stemdeclension(noristhereanygoodreasontoregardthemasbasictothelatterdeclensionandintrusiveintheformer).AmoreplausibleassumptionisthattheshadedcellsinTable3involvedefaultrulesofinflectiontowhichmembersofboththefeminine-stemdeclensionTheevidencepresentedhereisdrawnfromadatabasecomprisingmorethanonehundredheterocliteparadigmsinthirtylanguagesfromtwelvedifferentlanguagefamilies;thisdatabasecanbesearchedonlineathttp://www.cs.uky.edu/cgi-bin/cgiwrap/gstump/heteroclisis.cgi.(Iinvitereaderstosendmeadditionalexamplesforinclusioninthedatabase,whichremainsanongoingproject.) LANGUAGE,VOLUME82,NUMBER2(2006)andthederivative-stemdeclensionareequallysubject.Somesuchrules(e.g.therulesuffixing-inthegenitive/locativedual)actasdefaultsforeverynominalinthelanguage;others(e.g.therulesuffixing-intheaccusativeplural)actasdefaultsforfemininenominalswhosestemsendinvowels,includingmembersofboththefeminine-stemandderivative-stemdeclensions.Factssuchasthesemotivatetheviewthatalanguagesinflectionclassesfunctionasnodesinadefaultinheritancehierarchy.example,thehierarchyofVedicSanskritdeclensionclassesmightbe(partially)repre-sentedasinFigure1;inthishierarchy,stemssituatedatthederivative-stemnominalsnodehavecertaininflectionalcharacteristicsincommonthatdistinguishthemfromstemssituatedatthefeminine-stemnominalsnode,butstemsatbothnodesinheritvariousdefaultinflectionalcharacteristicsfromhighernodes,forexamplethefeminineaccusativepluralsuffix-(fromthevowel-stemnominalsnode)andthegenitive/locativedualsuffix-(fromthetopmost,nominalsnode).Genuineinstancesofheteroclisis,however,cannotsimplybeattributedtotheexistenceofdefaultinflectionalpatterns,becausetheyinvolveajuxtapositionofinflectionclassesthatordinarilycon-trast:theinflectionof,forexample,juxtaposesthesoft-masculineandhard-masculinedeclensions,whichcontrastintheinflectionofroomand vowel-stem nominalsconsonant-stem nominals stem nominals 1.PartialrepresentationoftheVedicSanskritdeclensionalhierarchy.Heteroclisisiscloselyconnectedtothephenomenonofstemsuppletion.Ininflec-tionalinstancesofstemsuppletion,alexemesparadigmisbuiltupontwoormorestemswhosedifferencesdonotfollowfromregularrulesofinflectionalexponence.Consider,forexample,theinflectionoftheSanskritnoun)heart(Table4):thedirect(i.e.nominative,vocative,andaccusative)caseformsof)arebuiltonthestem,whileitsremaining,obliquecaseformsarebuiltonthestemThisstemalternationissuppletive:itisnottheeffectofanyregularruleofinflectionalexponence,butisinstead

simplystipulatedin)slexicalentry.Itisbecauseofthisstemsuppletionthattheparadigmof)isheteroclite:becauseisaneuterstemendingin,itfollowstheneuter-stemdeclension(likethesteminTable4),butbecauseisaneuterstemendinginaconsonant,itinsteadfollowstheneuterconsonant-stemdeclension(liketheneuterformsofinTable4).PresumablytheheterocliteinflectionofCzechalsoonceinvolvedanalterna-tioninstemform;inmodernCzech,however,thecontrastinghardandsoftdeclensionalpatternsarenolongerstrictlyassociatedwithanyregulardifferenceinstemphonology(Heim1982:20).Nevertheless,evenifthestemremainsphonologicallyconstantintheparadigmof,onecanmaintainthatthisparadigmexhibitsakindofstemSeeCorbett&Fraser1993,Fraser&Corbett1995,andotherworkinNETWORKMORPHOLOGYforasystematicandformallyexplicitdevelopmentofthisidea. HETEROCLISISANDPARADIGMLINKAGE283 ¯syam¯sya¯syam¯syena¯sya¯ya¯sya¯syasya¯sye¯sye¯sya¯bhya¯syayos¯sya¯ni¯syais¯syebhyas¯sya¯na¯syes¯niSYAdbhya ‘threefold’(neuter forms) ‘heart’ 4.HeterocliteinflectionofSanskrit)‘heart’(Whitney1889:§397).suppletion:hereastembelongingtooneinflectionclassalternateswithastemthat,thoughphonologicallyidentical,belongstoadistinctinflectionclass.Onthisview,heteroclisissimplyentailssuppletion.Thereverse,however,isnottrue.Consider,forexample,theparadigmoftheRussian‘mother’inTable5.Twostemsparticipateinthedefinitionofthisparadigm:aradicalsteminthesingulardirect-caseformsandanextendedstemelsewhere.Therelationbetweenthesetwostemsissuppletive:thereisnoregularinflectionalrulethatgivesrisetothisalternation,whichmustinsteadbeseenastheeffectofastipulationinthelexicalentryof.Nevertheless,theentireparadigmoffollowsthethirddeclension;noteitsparallelismtothesingle-stemthird-declensionparadigmof‘mother-in-law’inTable5.Thus,suppletioninitselfdoesnotnecessitateheteroclisis.Ananonymousrefereehasquestionedthisassumption,sincethepremisethatallsuppletionisphonologi-callyovertsuppletionisessentialtomaintainingtheNOBLURPRINCIPLE(i)NOBLURPRINCIPLE(NBP):amongtherivalaffixesforanyinflectionalcell,atmostoneaffixmayfailtobeaclass-identifier(inwhichcase...thatoneaffixistheclass-defaultforthecell).(Cameron-Faulkner&Carstairs-McCarthy2000:816)Theproblemallegedhereisthis.TheNBPisseeminglydisconfirmedbyPolishcasemorphology,neitherofwhoselocative-casesuffixes-and-appearstobeaclass-identifier;Cameron-FaulknerandCarstairs-McCarthy(2000),however,arguethattheNBPcanbereconciledwiththePolishevidenceiftheselectionalrequirement‘minoritystemalternant’isassumedtobepartofthemeaningof-(whichjoinswithaspecialstemalternant)butnot

of-(whichdoesnot);onthatassumption,-and-arenotrivalsinthesenserequiredby(i),sinceinanycontextinwhich-and-compete,-willbefavoredover-byPaprinciple.(Accordingtothisprinciple,competitionbetweentwoormoremorphologicalmarkingsisresolvedinfavorofthemarkinghavingthenarrowest‘meaning’.)Thereferee’sobjectionisthatifphonologicallycovertsuppletionwereallowed,thentheNBPwouldbereducedtovacuity,sinceanypairofaffixesapparentlycounterexemplifyingitcouldbereconciledwithitbytreatingtheselectionalrequirement‘suppletivestemalternant’aspartofoneaffix’smeaning.Thisisnot,however,aconvincingbasisforrejectingtheassumptionthatsuppletivealternantsmaydifferininflection-classmembershipwithoutdifferinginform,sincetheNBPisinanyeventwidelydisconfirmed. LANGUAGE,VOLUME82,NUMBER2(2006) SINGULARPLURALSINGULARPLURAL 5.DeclensionoftheRussian3rd-declensionnounsmotherandHEPARADIGMLINKAGEAPPROACHTOHETEROCLISIS.Inthissection,IdevelopanapproachtoheteroclisisbasedonthenotionofparadigmlinkageproposedinStump2002.Ipresentthefundamentalprinciplesofthisapproach,thenshowhowitaccountsforsomeoftheobservablepropertiesofheterocliteparadigms„inparticular,thedis-tinctionbetweenmorphosyntacticallyconditionedheteroclisisandmorphologicallyconditionedheteroclisis;thecooccurrenceofheteroclisiswithdeponency;andthefactthatheteroclisissometimesoffsetstheeffectsofmorphosyntacticpropertyneutraliza-tionanddefectiveness.Theneedfortheapproachtoheteroclisisproposedhereisdemonstratedbyarangeofphenomenathatexcludethelexicalapproachin1.ULESOFPARADIGMLINKAGE.FollowingStump2002,Iassumethateverylex-emeLinalanguagehasanassociated,asetofcellseachofwhichconsistsofthepairingofLwithamaximalconsistentsetofmorphosyntacticproperties.Thecellsinalexemescontent-paradigm(thatis,its)correspondtothedifferenttypesofnodesintowhichformsofthatlexememaybeinsertedinsyntacticstructure;accordingly,eachcontent-cellhasarealizationthatisaccessibleforuseinsyntax.Similarly,everystemsinalanguagehasanassociated,asetofcellseachofwhichconsistsofthepairingofswithasetofmorphosyntacticproper-ties.Thecellsinalanguagesform-paradigms(its)areaccessibletorulesofrealizationalmorphologyandarethebasisfordeducingthemorphologicalrealizationofthecellsinthatlanguagescontent-paradigms.Consider,forexample,theinflectedformsoftheCzechnominallexemebridgeinTable1.Theseformsrealizethecontent-paradigmin3andtheform-paradigmin4. InSanskrit,forexample,thedualdirect-casesuffixes-and-disconfirmtheNBP.Thedistributionofthesesuffixesisnotphonologicallyconditioned;forinstance,theyappearinthesamephonolog

icalcontextintheparadigmofpowerful(masc.dualdirect-case,neut.dualdirect-case).Moreover,bothsuffixescutacrossarangeofdeclensionclasses:-regularlyappearsinmasculineandfeminineformsinalldeclensionsexceptthe-and-stemdeclensions(whosemasculineandfemininedualdirect-caseformsaresuffixlessformswithalengthenedstemvowel)andthe-stemdeclension;-regularlyappearsinfeminineformsinthe-stemdeclensionaswellasinallneuterforms.Neithersuffixcoincidesregularlywiththeuseofaspecialstem,andbecauseneitherisspecificastogender,bothmustbeassumedtorealizethepropertyset.Sinceneitherisaclassidentifier,theseaffixesdisconfirmtheNBP.Evidenceofthiskindisnotrare;seeStump2005cfordiscussionofadditionalSanskritcounterexamplestotheNBP.Invirtueofsuchempiricalevidence,theNBPcannotbeseenascompellingarejectionofthepossibilitythatsuppletivestem-alternantsmightdifferintheirinflection-classmembershipbutnotintheirphonologicalform.SeealsoAckerman&Stump2004,Ackermanetal.2006,andStewart&Stump2006.NotethatStump2002usesdifferentterminologyforthesenotions,referringtocontent-paradigmsandform-paradigmsassyntacticparadigmsandmorphologicalparadigms,respectively. HETEROCLISISANDPARADIGMLINKAGE285(3)Content-paradigmoftheCzechlexememasculinenominativesingularmasculinegenitivesingularmasculinedativesingularmasculineaccusativesingularmasculinevocativesingularmasculinelocativesingularmasculineinstrumentalsingularmasculinenominativepluralmasculinegenitivepluralmasculinedativepluralmasculineaccusativepluralmasculinevocativepluralmasculinelocativepluralmasculineinstrumentalplural(4)Form-paradigmoftheCzechstemmasculinenominativesingularmasculinegenitivesingularmasculinedativesingularmasculineaccusativesingularmasculinevocativesingularmasculinelocativesingularmasculineinstrumentalsingularmasculinenominativepluralmasculinegenitivepluralmasculinedativepluralmasculineaccusativepluralmasculinevocativepluralmasculinelocativepluralmasculineinstrumentalpluralForeachcontent-cellinthecontent-paradigmofalexemeL,therealizationofisdeterminedbyacorrespondingform-cellinsomestemsform-paradigm;thisform-cellisthe.Thus,theform-celldativesingularin4cistheform-correspondentofthecontent-celldativesingularin3c,andsoon.Theofaform-cellistheformresultingfromthesystematicapplicationtosofallapplicablemorphologicalrulesrealizingthepropertyset;therealizationofacontent-cellis,inturn,therealizationofthatcellsform-correspondent.Thus,therealizationofboththeform-masculinedativesingularin4candthecontent-celldativesingularin3cisthewordform(cf.Table1).Therelationbetw

eenaform-cellinastemsform-paradigmanditsrealizationisdefinedbymorphologicalrealizationrules(rulesofexponenceandrulesofreferral;Zwicky1985,Stump2001);rulesofthissortaresensitivetoastemsinflection-classmembership.Bycontrast,therelationbetweenacontent-cellinalexemescontent-paradigmanditsrealizationisdefinedbyRULESOFPARADIGMLINKAGE,byreferencetotherealizationofthatcontent-cellsform-correspondent.Thecanonicalrelationbe-tweenacontent-cellanditsrealizationisdefinedbytheuniversalruleofparadigmlinkagein5. LANGUAGE,VOLUME82,NUMBER2(2006)(5)Universaldefaultruleofparadigmlinkage:Ifisacontent-cellandstemrisstipulatedastherootoflexemeL,thenasitsform-correspondent(i.e.therealizationofthecontent-cellisthatoftheForinstance,inviewofthestipulationthatthestemistherootoftheCzech,rule5entailsthattherealizationofthecontent-celldativesingularisthatoftheform-cellmasculinedativesingulartherealization),andsoon.Inaccordancewithrule5,content-paradigmsarecanonicallywiththeform-paradigmsbywhichtheyarerealized;thatis,foragivencontent-paradigmPthereis,inthecanonicalcase,exactlyoneform-paradigmPsuchthateachcellinPhasanidenticallyspecifiedform-correspondentinP,whererisLsroot.Sometimes,however,thisstateofaffairsdoesnothold;thatis,onesometimesfindscontent-paradigmsrealizedbyform-paradigmswithwhichtheyarenotfullycongruent.Indeed,thisstateofaffairscanariseinmorethanoneway;amongthepossiblesourcesofincongruencearethephenomenaofmorphosyntacticpropertyneu-tralization,deponency,andheteroclisis.Iconsidereachoftheseinturn.Onesourceofincongruencebetweencontent-paradigmsandform-paradigmsisthephenomenonofMORPHOSYNTACTICPROPERTYNEUTRALIZATION;ininstancesofthisphe-nomenon,themorphosyntacticdistinctionsmadeamongthecellsofacontent-paradigmareonlypartiallypreservedamongthosecellsform-correspondents.Forexample,intheinflectionofneuternominalsinSanskrit,noindividualdeclensioneverdistinguishestherealizationofanominativeformfromthatofitsaccusativecounterpart;inordertoportraythisfactasarealgeneralizationratherthanasacoincidenceinthedefinitionofthevariousrulesrealizingSanskritcasemorphology,itisreasonabletoassumethatalthoughnominativeandaccusativecellsaredistinguishedinaneuternominalscon-tent-paradigm,eachnominativecontent-cellinsuchaparadigmsharestheform-corres-pondentofitsaccusativecounterpart.Thus,supposethattheoperatorisdefinedasjoiningtwomorphosyntacticpropertiesp,ptoproduceanewproperty[p]suchthatanyrulerealizingporpalsorealizes[p];onecanthenassumethatthecontent-cells6a…cintheparadigmoftheneuterlexememouthhavetherespect

iveform-cellsin7a…castheirform-correspondents.(Thisanalysismakesexactlytherightpredictionsaboutaneuternominalsdirect-caseforms,entailing,forexample,thatthedefaultrulerealizingtheaccusativesingularas-appliesintherealizationofbothofthecontent-cellsin6a,yieldingthesyncreticform(6)a.neuternominativesingularneuteraccusativesingularneuternominativedualneuteraccusativedualneuternominativepluralneuteraccusativeplural(7)a.neuter[nominativeaccusative]singular(realization:aneuter[nominativeaccusative]dual(realization:aneuter[nominativeaccusative]plural(realization:aHereandthroughout,Iusethetermtomeanalexemesdefaultstem.SeeBaerman2004andBaermanetal.2005forargumentsfavoringanapproachtosyncretismsimilartotheoneassumedhere. HETEROCLISISANDPARADIGMLINKAGE287Inthisexample,twocontent-cellscarryingdistinctvaluesforthefeatureaform-correspondentwhosevalueisneutralwithrespecttothatdistinction.Butinstancesofmorphosyntacticpropertyneutralizationmightsometimesinvolvecontent-cellsthatcarrydistinctvaluesforafeatureforwhichtheirsharedform-correspondentissimplyunspecified.Thus,inlateSanskrit,forinstance,nodeclensionrealizesgenderdistinctionsoutsideofthedirect(nominative,vocative,andaccusative)cases;forin-stance,intheinflectionoftheadjectivebrave,masculineandneuterformsaredistinguishedinthenominativesingular(masc.vs.neut.)butnotinthegenitivesingular(masc./neut.Onecanaccordinglyassumethatgenderissimplyunspecifiedintheoblique-casecellsoftheform-paradigmsoflateSanskritnominals;thecontent-cellsin8a,b,forinstance,canbeassumedtosharetheform-correspondentin8c.Inviewoftheincidenceofmorphosyntacticpropertyneutraliza-tionssuchasthoseexemplifiedin7and8c,therelationbetweenalexemescontent-cellsandtheirform-correspondentsisoftenmany-to-one.(8)a.neutergenitivesingularmasculinegenitivesingulargenitivesingular(realization:všAnothersourceofincongruencebetweencontent-paradigmsandform-paradigmsisthepossibilitythatacontent-cellmayhaveaform-correspondentsuchthathavecontrastingvaluesforoneormoreinflectionalcategories;theeffectofthissortofcontrastisthephenomenonof.Consider,forexample,theLatinconfess.Itscontent-paradigmcontainsactivecellssuchas9a.Butbecauseisdeponent,theactivecellsinitscontent-paradigmhavepassiveform-cellsastheirform-correspondents;forinstance,thecontent-cellin9ahastheform-cellin9basitsform-correspondent.Forthisreason,therealizationsofarepassiveinformbuthavethemorphosyntacticcontenttypicalofactiveforms.(9)a.1stsingularpresentnonperfectactive1stsingularpresentnonperfectpassive(realization:fateor)Heterocli

sisisathirdsourceofincongruencebetweencontent-paradigmsandform-paradigms.Ininstancesofheteroclisis,thecellsinasinglecontent-paradigmdrawtheirform-correspondentsfromtwoormoredistinctform-paradigms.Thus,consideragaintheCzechheteroclitenounspringinTable1.Theform-correspondentsofscontent-cellsaredrawnfromtwodistinctform-paradigms:thatofthestem(amemberofthesoft-masculinedeclension)andthatofthestem(amemberofthehard-masculinedeclension).Inparticular,eachsingularcellscontent-paradigmhastheform-cellasitsform-correspondent;bycontrast,eachpluralscontent-paradigmhastheform-asitsform-correspondent.Here,too,thecanonicalcongruenceofcontent-paradigmswiththeform-paradigmsbywhichtheyarerealizedissuspended.Astheforegoingexamplesreveal,therearetwodimensionsofdeviationfromthecanonicalcongruenceofcontent-paradigmsandform-paradigms.First,acontent-cellsmorphosyntacticpropertysetmaydifferfromthatofitsform-correspondent;thismayThefeminineformsareirrelevanthere,sincetheyinflectasmembersofthedistinct-stemdeclension. LANGUAGE,VOLUME82,NUMBER2(2006)bebecausetheform-correspondentspropertysetincludesfewermorphosyntacticcon-trasts(asin7or8c)orbecauseitsfeaturespecificationsareactuallydistinctfromthoseofthecontent-cell(asin9).Second,theform-correspondentsofcellsbelongingtothesamecontent-paradigmmaybelongtodistinctform-paradigms,asinthecaseof.Anadequatetheoryofinflectionmustprovideforbothofthesedimensionsofdeviationfromthecanonicalcongruenceentailedby5.Myproposalisthatdeviationsfrom5are,inbothofthesedimensions,aneffectoflanguage-specificrulesofparadigmlinkage.Aruleofthissortstipulatesthattheform-correspondentofaparticularcontent-issomeform-cellotherthanthedefaultform-correspondentthat5wouldotherwisesupplyfor;inthisway,itoverrides5,suspendingthecanonicalcongru-encethatitotherwiseenforces.Thus,consideragainthetwodimensionsofdeviationfromthiscanonicalcongruence.Theevidenceofmorphosyntacticpropertyneutralizationanddeponencyshowsthattherelationbetweenacontent-cellspropertysetandthatofitsform-correspondentisnotalwaysoneofsimpleidentity;todevelopthisidea,itisusefultodrawaterminologi-caldistinctionbetweenalanguages(anypropertybelongingtoina)andits(anypropertybelongingtoinaform-cellGiventhisdistinction,therelationbetweenacontent-cellspropertysetandthatofitsform-correspondentcanberepresentedasafunctionfromsetsofs-propertiestosetsofm-properties;IhererefertofunctionsofthiskindasPROPERTYMAPPINGS.Inthesimplestcases„thosedeterminedbythedefaultruleofparadigmlinkagein5„therelevantpropertymappingisanidentityfunction

.IntherealizationoftheSanskritcontent-cellsin6and8a,b,bycontrast,therelevantpropertymappingisthefunctiondefinedin10,andintherealizationoftheLatincontent-cellin9a,therelevantpropertymappingisthefunctiondefinedin11.(10)DefinitionoftheSanskritpropertymapping:Whereisanygenderandisanyobliquecase:(i)ifneutnomXneutaccX,thenneut[nom(ii)if,then(iii)otherwise(11)DefinitionoftheLatinpropertymapping:IfactiveX,thenpassiveX;otherwisePropertymappingssuchasvaryintheirscope:forinstance,whilerelevanttotheinflectionofeverynominalinSanskrit,isrelevanttotheinflectionofonlyatinyminorityofLatinverbs.Thescopeofagivenpropertymappingmustthereforebestipulated;Iassumethatthisiseffectedbylanguage-specificrulesofparadigmlinkage.Thus,givenanycellinthecontent-paradigmofaSanskritnominalLhavingrootr,theruleofparadigmlinkagein12entailsthattheform-correspondentof;itisthisrulethatassuressrelevancetothedefaultinflectionofeverynominalinthelanguage.Butwhereisacellinthecontent-paradigmofaLatinverbL,theruleofparadigmlinkagein13suppliesasaform-correspondentonlyinthoseinstancesinwhichLisdeponent;itisthisrulethatassuresthatsrelevanceislimitedtodeponentverbs.Thisdistinctionderivesfromandcanbeequatedwiththedistinctionbetweens-featuresandm-featuresoriginallydrawnbySadlerandSpencer(2001:72).In10and11andthroughout,IuseXasavariableoversequencesofmorphosyntacticproperties;thus,neutnomXrepresentsanysetofmorphosyntacticpropertiescontainingneutandnom. HETEROCLISISANDPARADIGMLINKAGE289(12)Sanskritruleofparadigmlinkage:WhereLisanominallexemehavingrasitsroot,thecontent-cellasitsform-correspondent.(13)Latinruleofparadigmlinkage:WhereLisadeponentverballexemehavingrasitsroot,thecontent-cellasitsform-correspondent.Byrule12,thecontent-cellsin6a…cand8a,bhavethecorrespondingform-cellsin7a…cand8castheirform-correspondents;byrule13,thecontent-cellin9ahastheform-cellin9basitsform-correspondent.Becausetheserulesaremorespecificthanthedefaultruleofparadigmlinkagein5,theyoverrideit,blockingtheform-correspond-entsthatitwouldotherwisesupplyandsuspendingthecanonicalcongruencethatitwouldotherwiseentail.Theevidenceofheteroclisisshowsthattheform-correspondentsofthecellsinalexemescontent-paradigmneednotalwaysbelongtothesameform-paradigm.Forinstance,someofthecellsinthecontent-paradigmofCzechspringhaveform-correspondentsbelongingtotheform-paradigmof(amemberofthesoft-masculinedeclension),whileothershaveform-correspondentsbelongingtotheform-paradigmof(amemberofthehard-masculinedeclension).Iassumethatthechoiceamongcompetingstems(i.e.amongcompetingf

orm-paradigms)intheinflectionofaheteroclitelexemeis„likethescopeofapropertymapping„determinedbyrulesofparadigmlinkage.Inthetworulesofparadigmlinkagein12and13,theform-correspondentsstemis(asin5)simplytherootofitslexeme.Butininstancesofheteroclisis,theform-correspondentsstemgenerallydiffersfromtherootofitslexemeinsomeway,eitherinitsformorinitsinflection-classmembership(orboth);ifitdiffersinanyoftheseways,IrefertoitasaCORADICALSTEM(orsimplyasaThus,supposethatinCzech,isidentifiedassrootandhencethatshard-masculinecoradical.Onthatassumption,form-correspond-entsforthepluralcellsinscontent-paradigmaredrawnfromthecoradicalsform-paradigm,inaccordancewiththeruleofparadigmlinkagein14.(14)Czechruleofparadigmlinkage:WhereLisanominallexemethatbelongstotheclassandhassasitshard-masculinecoradical,if,thenthecontent-cellasitsform-correspondent.Thedefaultruleofparadigmlinkagein5sufficestoguaranteethatanycontent-cellwillhavetheform-cellasitsform-correspondent.Butinthepluralcellsofspara-digm,5isoverriddenbythemorespecificruleofparadigmlinkagein14,whichcausesanycontent-cellscontent-paradigmtohavetheform-cellasitsform-correspondent.Whenaruleofparadigmlinkagespecifiesform-correspondentscontainingthecora-dicalsofsomelexemeL,smay(asinthecaseof)belongtoaninflectionclassdistinctfromthatofLsroot;suchisbydefinitionthecaseininstancesofhetero-clisis.ButthecoradicalspecifiedbyaruleofparadigmlinkagemaybelongtothesameinflectionclassasLsroot(asinthecaseofRussianmother,Table5);instancesofthissortarecasesofsimple(nonheteroclitic)stemalternation.Intheinflectionof,thedistributionofthecoradicalisdeterminedbyasingleruleofparadigmlinkage;butthedefinitionofacoradicalsdistributionOrdinarily,theappearanceofaheteroclitelexemesrootisspecifiedbythedefaultruleofparadigmlinkagein5,whichisthenoverriddenbyoneormorerulesspecifyingtheappearanceofitscoradicalsinanarrowersetofcontexts;butinprinciple,thelatterrulescouldinturnbeoverriddenbyanevennarrowerrulespecifyingtheappearanceoftheroot.Comparetherulesinn.22. LANGUAGE,VOLUME82,NUMBER2(2006)sometimesinvolvesmorethanoneruleofparadigmlinkage.Thus,considertheCzechpresident(Table6),whoserootinflectsaccordingtothehard-femininedeclensionandwhosecoradicalinflectsaccordingtothehard-masculinedeclension.Asamemberoftheissubjecttotheruleofparadigmlinkagein14,whichlicensesintheplural;butbelongstoasubclassofmasculineanimatemembersoftheclassthatfollowthehard-masculinedeclensioninthedativeandlocativesingular.Toaccountforthelatterfact,thelexeme(andtheothermembersofitssubclass,forex

ample,colleague;Heim1982:116)mustbesubjecttotheadditionalruleofparadigmlinkagein15. edsedovéedsedu edsedu edsedyedsedovéedsedechedsedyedsedoviedsedoviedsedaedsedyedseduedsedoedsedou DECLENSION:HARD-FEMININEHARD-MASCULINE GEN PREDSEDA (m.) enaenyeneenuenoeneenouenyenenámenyenyenáchenamifilosoffilosofafilosofemfilosofovéfilosofu filosofy 6.HeterocliteinflectionofCzechpresident(Kopeckyetal.1976:761,770,773).(15)Czechruleofparadigmlinkage:WhereLisanominallexemethatbelongstothesubclassandhassasitshard-masculinecoradical,ifsingularX,thenthecontent-cellasitsform-correspondent.Inaheterocliteparadigmsuchasthatof,oneinflectionclass(thatoftheroot)servesastheparadigmsdefault;theparadigmsotherinflectionclass(es)canthusbeseenasintrusive,overridingthedefaultclassinparticularcells.Iassumethatuniversalcriteriadeterminewhichofaheterocliteparadigmsinflectionclassesisitsdefaultandwhichmustbeseenasintrusive.Forpresentpurposes,Iassumethecriteriain16;ultimately,thesecriteriamaywellneedtoberefinedandextended.(16)WhereSaredisjointsetsofcellsinaheteroclitecontent-paradigmPandtheformsrealizingthesesetsofcellsareinflectedasmembersofinflec-tionclassesCandD(respectively):scoradicalsteminflectsasamemberofthehard-masculinedeclensioninthedativeandlocativesingular,itisunlikeordinarymembersofthisdeclensioninthatitdoesnotexhibitasanalternativeto-inthesecases.Iassumethatthisalternativesuffixisrestrictedtorootsbelongingtothehard-masculinedeclensionandthereforedoesnotappearintheinflectionof(which,thoughamemberoftherelevantdeclension,isacoradicalbutnotaroot).Thealternationof-with-raisesanumberofinterestingtheoreticalissuesquiteapartfromitsfailuretoappearintheinflectionoftypeheteroclites;fordetails,seeSmilauer1972:144ff. HETEROCLISISANDPARADIGMLINKAGE291a.If(i)thereissomenonemptymorphosyntacticpropertysetsuchthatallandonlycellsinPcontaininganextensionofbelongtoS(ii)thereisnononemptymorphosyntacticpropertysetsuchthatallandonlycellsinPcontaininganextensionofbelongtoSthenCisanintrusiveclassinP;otherwise,b.CisanintrusiveclassinPifthecellcontainingtheleastmarkedmorpho-syntacticpropertysetbelongstoSBy16b,thehard-masculinedeclensionistheintrusiveinflectionclassintheparadigmsofCzech.Criterion16bmay,however,beoverriddenby16aininstancesinwhichthelatterapplies.Consider,forexample,theparadigmofFulawindinTable7:inthisparadigm,thecellcontainingtheleastmarkedmorphosyntacticpropertysetistheneutralsingularcellcontainingtheD-gradeform;16bmightthereforebetakentoentailthattheAgradeistheintrus

iveinflectionclassinthisparadigm.ButthesetofD-gradecellsinthisparadigmincludesallandonlycellscontaininganextensionofneutralsingular,andthereisnononemptypropertysetofwhicheachoftheA-gradecellscontainsanextension.Criterion16athereforeoverrides16b,entailingthattheDgradeisinfacttheintrusiveinflectionclassinthisparadigm. Diminutivesingularsingular(noun class 5)-umken-umbaa-AugmentativesingularNeutralsingular-iken-ihen-nduwaa-ndu 7.HeterocliteinflectionofFulawind(Arnott1970:89,99,120f.).TheCzechlexemeseachhaveonlytwostems„theirrootandonecoradical.Butheteroclitelexemessometimeshavemultiplestems;toaccommo-datesuchinstances,onemustnaturallyassumethatdifferentrulesofparadigmlinkagemaydeterminethedistributionofalexemesdifferentcoradicals.Thus,inCzech,nounsbelongingtothehard-masculinedeclensionfollowthesoft-masculinedeclensioninthelocativepluraliftheirstemendsinabackobstruent;forinstance,thenounphilologistinTable8generallyfollowsthehard-masculinedeclension,butexhibitsthelocativepluralform,whosestem-finalpalatalizationandsuffix-marksofthesoft-masculinedeclension.Moreover,heteroclitenounsofthetype(whichfollowthehard-masculinedeclensioninthepluralbutnotalwaysinthesingular)exhibitthissameidiosyncrasy;thus,thenounservantinTable8exhibitsboththepatternofheteroclisisdisplayedby(cf.Table6)andthatdisplayedbyinthelocativeplural.Thus,thelexemehasthreestems:isanisa(properorimproper)subsetof;seeStump2001:41foramoreexactdefinitionofextension. LANGUAGE,VOLUME82,NUMBER2(2006)arootbelongingtothehard-femininedeclension,acoradicalbelongingtothehard-masculinedeclension,andasecondcoradicalbelongingtothesoft-masculinedeclen-sion.Thedistributionofthesethreestemscanbeattributedtofourdifferentrulesofparadigmlinkage:thehard-femininerootisusedbydefault,inaccordancewith5;thisdefault,however,isoverriddenby14and15,inaccordancewithwhichthehard-masculinecoradicalappearsinthepluralandinthedativeandlocativesingular;rule14isinturnoverriddenbyrule17,bywhichthesoft-masculinecoradicalappearsinthelocativeplural. DECLENSION:HARD-FEMININEHARD-MASCULINESOFT-MASCULINE (m.)(m.)philosopherFILOLOG (m.)MU GENDATACC enaenyenuenoenoueneenyenenyenyenamisluhasluhysluhusluhosluhousluhovisluhovifilosofovi, filosofufilosofovi, filosofufilosoffilosofafilosofafilosofefilosofemfilosofu filosofechfilosofovéfilosofu filosofovéfilosofysluhu sluhu sluhovésluhyfilologovi, filologufilologovi, filologufilologfilologafilologafilologufilologemfilologu filologu filologovéfilologymuovi, muimuovi, muimumuemuemuimuemmuu mui,

muovémuu emui, muovémui 8.HeterocliteinflectionofCzechservant(Kopeckyetal.1976:761,766,770,773).Shadedlocativepluralformsfollowthesoft-masculinedeclension.(17)Czechruleofparadigmlinkage:WhereLisanominallexemehavingsasitssoft-masculinecoradical,iflocativepluralX,thenthecontent-cellasitsform-correspondent.Nounshavinghard-masculinestemsthatendinabackobstruentarequiteregularinshiftingtothesoft-masculinedeclensioninthelocativeplural;toaccountforthisregularity,Iassumethegeneralruleofsteminferencein18,whoseeffectistoguaranteethattheruleofparadigmlinkagein17willapplyintheinflectionbothofnounssuch(whosehard-masculinestemisacoradical)andofnounssuchas(whosehard-masculinestemisitsroot).(18)Czechruleofsteminference:IflexemeLhasastemsbelongingtothehard-masculinedeclensionandsendsinabackobstruent,thenLhassasitssoft-masculinecoradical,wheresislikesexceptthatitbelongstothesoft-masculinedeclension.Tosummarize:theproposedapproachtoheteroclisispresupposesadistinctionbe-tweenalexemescontent-paradigmandtheform-paradigm(s)ofitsstem(s);thoughthetwotypesofparadigmsaredistinct,theyarecruciallyrelated,sinceacontent-cellsrealizationisequatedwiththatofitsform-correspondent.Ininstancesofheteroclisis,distinctstemsbelongingtodistinctinflectionclasses(andthereforepossessingdistinctform-paradigms)participateinthedefinitionofasinglelexemesrealization.This HETEROCLISISANDPARADIGMLINKAGE293participationismediatedbyrulesofparadigmlinkage,whichdeterminetheform-correspondentforanygivencontent-cell;theseincludetheuniversalruleofparadigmlinkagein5andoverriding,language-specificrulesofparadigmlinkagesuchas14,15,and17.Thepostulationofoverridingrulesofparadigmlinkageisindependentlyrequiredtoaccountforinstancesofmorphosyntacticpropertyneutralizationanddepo-nency(asinthecasesof12and13,respectively).Partoftheplausibilityofthisapproachtoheteroclisis(theparadigm-linkageap-proach)stemsfromitseffectivenessinaccountingforanumberofobservableproper-tiesofheterocliteparadigms.Here,Idiscussthreesuchproperties:(i)thedistinctionbetweenmorphosyntacticallyconditionedandmorphologicallyconditionedhetero-clisis,(ii)thecooccurrenceofheteroclisisanddeponency,and(iii)thefactthathetero-clisismayineffectcounteractthephenomenaofmorphosyntacticpropertyneutraliza-tionanddefectiveness.ORPHOSYNTACTICALLYCONDITIONEDVSMORPHOLOGICALLYCONDITIONEDHET-.IninstancesofMORPHOSYNTACTICALLYCONDITIONEDHETEROCLISIS,thechoiceofinflectionclassintherealizationofaparadigmsindividualcellsisdirectlydeter-minedbythemorphosyntacticpropertysetsexpressedb

ythosecells;ininstancesMORPHOLOGICALLYCONDITIONEDHETEROCLISIS,thechoiceofinflectionclassintherealizationofaparadigmsindividualcellsisinsteaddeterminedbyanindependentlyobservablepatternofstemalternation.AsIshowhere,theparadigm-linkageapproachtoheteroclisisaffordsasimpleaccountofthedistinctionbetweenthesetwosortsofTheheterocliteinflectionofismorphosyntacticallyconditioned:whetheragivenwordforminsparadigmfollowsthesoft-masculineorthehard-mascu-linedeclensiondependspurelyandsimplyonthemorphosyntacticpropertiesthatthatwordformexpresses.Thereare,however,instancesinwhichthealternationbetweenaheterocliteparadigmscompetinginflectionclassesfollowsanindependentlymotivatedpatternofstemalternationobservableinnonheterocliteparadigms;ininstancesofthissort,therulesofparadigmlinkageneededtoaccountforthisindependentpatternofstemalternationalsoaccountfortheincidenceofheteroclisis.Sanskritfurnishesaclearexampleofsuchmorphologicallyconditionedheteroclisis.AnimportantcharacteristicofSanskritdeclensionalmorphologyisapatternofstemalternationwidelyobservableamongstemsendinginconsonants.Amasculineorfemi-ninenominalparticipatinginthispatternofalternationexhibitsitsStrongsteminthedirectcasesofthesingularanddualandinthenominativeandvocativeoftheplural;elsewhereitassumesaWeakform.AneuternominalexhibitsitsStrongstemonlyinthedirectcasesoftheplural,assumingaWeakformelsewhere.Somealternatingnominalspossessonlytwostems:aStrongstemandasingleWeakstem.OthershaveaStrongstemandtwoWeakstems:aWeakeststemappearingbeforevowel-initialcaseendings,andaMiddlestemappearingelsewhere.ThemasculineandneuterparadigmsofthealternatingadjectivewesterlyaregiveninTable9;intheseparadigms,theadjectivesStrongstem-(sandhiform:)appearsintheunshadedcells,itsWeakeststem-appearsinthelight-shadedcells,andCorbett(2006)arguesforacomparableconclusionwithregardtosuppletivestemalternationsingeneral.Strongstemsasaclassvaryintheirformation;thatis,itisnottheirformationthattiesthemtogether,buttheirparticipationinacommondistributionalpattern.ThesameistrueofMiddleandWeakeststems.SeeStump2001:Ch.6fordiscussion. LANGUAGE,VOLUME82,NUMBER2(2006)itsMiddlestem-(sandhiforms:-)appearsinthedark-shaded MASCULINE FORMSNEUTER FORMSSINGULARDUALPLURALSINGULARDUALPLURAL¯ca¯ce¯cas¯cas¯ci¯cos¯cas¯ca¯ca¯ce¯cas¯cas¯ci¯cš¯cš¯cš¯cos¯cos¯ca 9.MasculineandneuterformsofSanskritwesterly(Whitney1889:§408).UnshadedformsarebasedontheStrongstem;light-shadedformsarebasedontheWeakeststem;dark-shadedformsarebasedontheMiddlestem.Althoughtheyexhibitextensivestemalterna

tion,themasculineandneuterparadigmsarenotheteroclite:alloftheirformsfollowthegeneralconsonant-stemEvenso,rulesofparadigmlinkagemightbeinvokedtoaccountfortheobservedpatternofstemalternation.OntheassumptionthatsrootisitsStrongstem,theruleofparadigmlinkagein5entailsthatformsbuiltuponthisStrongstemwillappearbydefaultinsparadigm.Suppose,now,thatthecontrastingnotionsofstrongandweakpropertysetsaredefinedasin19;inthatcase,thedistributionofformsbuiltonsMiddleandWeakestcoradicalsiscorrectlyaccountedforbytheoverridingrulesofparadigmlinkagein20a,b(wherethepropertymappingisasin10).(19)Definitionofstrongandweakpropertysets:Whereismasculineorisanydirectcase(nominative,vocative,oraccusative),andisanynumber(singular,dual,orplural),a.instancesofbydefault;b.butinstancesofaccusativeplural;inaddition,c.anygender/case/numbercombinationthatisnotstrongaccordingto(a)(20)Sanskritrulesofparadigmlinkage:a.WhereLisanominallexemehavingsasitsWeakestcoradicalandisaweakpropertyset,iftherealizationoftheform-cell[vowel]X,thenthecontent-cellasitsform-corres-b.IfLisanominallexemehavingsasitsMiddlecoradicalandisaweakpropertyset,thenthecontent-cellasitsform-Thefeminineparadigmofis,however,basedonthestem,whichdoesnotparticipateinthepatternofstemalternationexemplifiedinTable9butinsteadfollowsthederivative-stemdeclension.ThisdeclensionisfrequentlyrecruitedfortheexpressionoffemininegenderinSanskrit;see§4.2fordiscussionofthisnotionofrecruitment.Clauses(a)and(b)indefinition19appeartobecontradictory;butbecause19bisthemorenarrowlyapplicableofthetwoclauses,itoverrides19a,inaccordancewithPainisprinciple. HETEROCLISISANDPARADIGMLINKAGE295Together,20aand20bapparentlycauseanycellinscontent-paradigmthathasaform-correspondentcontainingtheWeakestcoradicaltohaveasecond,competingform-correspondentcontainingtheMiddlecoradical.ButontheassumptionthatPainisprincipleregulatestheinteractionofrulesofparadigmlinkage,20awill,asthenarrowerofthetworules,override20binanyinstanceinwhichtheformerruleisapplicable(i.e.intheweakestcases,whoserealizationinvolvesavowel-initialcaseTherulesofparadigmlinkagein20areindependentlymotivatedbytheneedtoaccountforthepatternofstemalternationinnonheterocliteparadigmssuchasthemasculineandneuterparadigmsof.Astheyare,however,theyalsoaccountforthepatternsofdeclension-classalternationexhibitedbyheteroclitenounssuchasday(Table10).Theparadigmofisbuiltonthreestems:anditszero-gradecounterpartinflectaccordingtotheneuter-stemdeclension,whileinflectsaccordingtotheneuter-stemdeclension.Itwouldbemessyto

statetheparadigmaticdistributionofthesestemspurelyintermsofthemorphosyntacticpropertieswithwhichtheyareassociated;buttheirdistributionfollowsdirectlyfromtherulesofparadigmlinkagein20ifisidentifiedassroot(itsStrongasitsMiddlecoradical,andasitsWeakestcoradical.Thisexamplerevealsthefundamentalsimilaritiesanddifferencesbetweenmorpho-syntacticallyconditionedandmorphologicallyconditionedheteroclisis.Bothsortsofheteroclisisinvolvealexemehavingarootandatleastonecoradicalbelongingtodistinctinflectionclasses;bothinvolvearuleofparadigmlinkagespecifyingthecon-textsinwhichthelexemescontent-paradigmdrawsitsrealizationsfromtheform-paradigmofthecoradicalratherthanfromthatoftheroot.Intheparadigm-linkageapproach,however,thereisafundamentaldifferencebetweenmorphosyntacticallyconditionedandmorphologicallyconditionedheteroclisis:instancesofthelatterinvolverulesofparadigmlinkagethat(likethosein20)defineanindependentpatternofstemalternationwhoseincidencedoesnotalwaysinvolveheteroclisis.ETEROCLISISANDDEPONENCY.Theparadigm-linkageapproachalsoaccountsstraightforwardlyforthecooccurrenceofheteroclisisanddeponency.Ininstancesofsfeminineinflection,therulesin20areoverriddenbyaruleofparadigmlinkagebywhichanyfemininecontent-cellasitsform-correspondent;cf.n.18.TheidentificationofsWeakestcoradicalneednotbestipulatedlexically,sinceitfollowsfromadefaultruleofstemindexingthatisindependentlyneededinSanskrit;seeStump2001:186ff.forInthe-stemdeclensions,theStrongstemoptionallyappearsinthelocativesingular;thus,thelocativesingularformsinTable10haveasoptionalalternants.Inaddition,membersoftheneuter-stemdeclensionoptionallyhavetheStrongstemintheirdualdirect-caseform;thus,thedualdirect-caseformsasalternants.Thisvariationimpliesthefollowingoptionalrulesofparadigmlinkage.(i)WhereLisanominallexemewhoseStrongstemrbelongstoan-stemdeclension,iflocativesingularX,thenthecontent-celloptionallyhasasitsform-correspondent.(ii)WhereLisanominallexemewhoseStrongstemrbelongstotheneuter-stemdeclension,ifdualXforsomedirectcase,thenthecontent-celloptionallyhasasitsNotethatevenwithrespecttotheseinstancesofoptionality,therulesofparadigmlinkagerelevanttothedefinitionofsheterocliteparadigmareindependentlyneededforthedefinitionofnonheterocliteparadigms(suchasthatofSanskritpresentsnumerouscomparableinstancesofmorphologicallyconditionedheteroclisis,forexam-plethoseofroad,andsoon. LANGUAGE,VOLUME82,NUMBER2(2006) MAN (n.)as-STEM (n.)(n.) ¯ni¯mnš¯mnas¯mna¯ma¯mne¯mni¯masu¯mna¯ma¯ni¯mabhis¯mabhyas¯mabhya¯mnos 10.HeterocliteinflectionofSanskri

tday(Whitney1889:§§414,424,430).deponency,morphologythatisordinarilyusedintheinflectionofformspossessingonemorphosyntacticpropertyisinsteadusedintheinflectionofformspossessingsomecontrastingproperty;thoughdeponencyandheteroclisisarelogicallyindependentphenomena,itisnotunusualtofindparadigmsthatarebothdeponentandheteroclite.Thus,considerthecaseofOldEnglishpreterite-presentverbs„forexample,theverbknow(Table11).Inthepresenttense,generallyexhibitsthepast-tensemorphologytypicalofastrongverbsuchasInthepasttense,exhibitsthepast-tensemorphologytypicalofweakverbs(e.g.theverbTheinflectionofisdeponent,sinceitexploitspast-tensemorphologyintheformationofitspresent-tenseparadigm;itis,atthesametime,heteroclite,becauseitexploitsthemorphologyofstrongverbsinonepartofitsparadigmandthatofweakverbsintheotherpart.Intheparadigm-linkageapproach,thereisanimportantsimilaritybetweenhetero-clisisanddeponency:bothinvolveanoverrideofthedefaultrulein5byalanguage-specificruleofparadigmlinkage.(Recall,forexample,thatintheinflectionoftheLatinverbconfess,theruleofparadigmlinkagein13causestheactivecellsscontent-paradigmtohavepassiveform-cellsastheirform-correspondents.)Moreover,thereisnologicalreasonwhyasingle,overridingruleofparadigmlinkagecouldnotsimultaneouslyengenderbothheteroclisisanddeponency.Ontheassumptionsoftheparadigm-linkageapproach,aruleofparadigmlinkageassociatesacontent-cellwithaform-correspondent:ifsisacoradicalbelongingtoaninflectionclassdistinctfromthatofLsroot,heteroclisisresults;ifisapropertysetthatcontrasts,deponencyresults;butnothingexcludesthepossibilitythatmightcoinci-dentallysatisfybothoftheseconditions.Thus,consideragainthecaseof.Ontheassumptionthatsweakstemisitsroot,thedefaultruleofparadigmlinkagein5correctlyassociatesthepast-tensecellsinscontent-paradigmwiththepast-tensecellsinsform-paradigm;toaccountforspresent-tenseparadigm,however,theoverridingruleofparadigmTheformistheonlydivergencefromthisregularity:thedefaultsecond-personsingularindicativedesinence-isordinarilyoverriddenby-inthepast-tenseparadigmsofstrongverbs. HETEROCLISISANDPARADIGMLINKAGE297 ¯ h ¯ h ¯ h ¯ h ¯ h ¯ t ¯ ð Indicative1sg singularsingularplural FANLAN dršdršdršdrš witonwitewitenwast dra drifondrifewistewistewistonwistestwistewistenh ¯ h ¯ ðh ¯ h ¯ h ¯ dršdrš 11.HeterocliteanddeponentinflectionoftheOldEnglishpreterite-presentverbknow.Shadedcellsinsparadigminflectlikeshadedcellsinsparadigm.linkagein21aisnecessary.Thisruleengendersheteroclisisbysubstitutingstrongcoradicalforitsroot;atthesametime,iten

gendersdeponencybycausingspresent-tensecontent-cellstohavespast-tenseform-cellsastheirform-(21)a.OldEnglishruleofparadigmlinkage:WhereLisapreterite-presentverballexemehavingsasitsstrongcoradical,ifpresentX,thenthecontent-cellasitsform-correspondent.b.DefinitionoftheOldEnglishpropertymapping:IfpresentXpastX;otherwiseMORPHOSYNTACTICPROPERTYNEUTRALIZATIONANDDEFECTIVE-.Theparadigm-linkageapproachalsoaffordsastraightforwardaccountofthefactthatheteroclisissometimesoffsetstheeffectsofmorphosyntacticpropertyneutraliza-tion.Thekeyobservationhereisthatamorphosyntacticpropertydistinctionthatisneutralizedintheinflectionofordinarylexemesmaybepreservedintheinflectionofaheteroclitelexemepreciselybecauseitsinflectioninvolvesmorethanonestem;inparticular,theheteroclisisinducedbyrulesofparadigmlinkagecomparableto14mayhavetheeffectofcounteractingmorphosyntacticpropertyneutralizationssuchasthoseinducedbyrulesofparadigmlinkagecomparableto12.SochiapanChinantec(Oto-Manguean)furnishesastrikingexampleofthissort.InSochiapanChinantec,dynamicverbsinflectforperson/numberagreementprincipallythroughthemodulationoftheirstemstoneandstress;Table12showsthestemmodula-tionofthreeverbs.Astheseexamplesshow,numberdistinctionsarepartiallyneutral-izedintheinflectionofSochiapanChinantecverbs(Foris2000:56):whileaverbscontent-paradigmcanbeassumedtodistinguishnumberinallthreepersons,theassoci-Rule21aisitselfoverriddenbyanarrowerrule,accordingtowhichthecontent-cellasitsform-correspondentif2ndsingularpresentindicativeSochiapanChinantechasthreesimpletones(//)andfourcomplextones(/MHLMHLML/);seeForis2000:18ff.fordetails. LANGUAGE,VOLUME82,NUMBER2(2006)atedform-paradigmdistinguishesnumberonlyinthefirstperson.Informalterms,thismeansthatacontent-cellhavingasitspropertysethasaform-correspondentwhosepropertysetis),wherethepropertymappingisdefinedasin22a;accord-ingly,averbsrealizationsare,inthedefaultcase,determinedbytheruleofparadigmlinkagein22b. kuõukuõu 12.SteminflectionofthreeSochiapanChinantecverbs(Foris2000:56,57).Becausehortativemorphologyisobligatoryinnonsecond-personprohibitiveforms,suchformsexhibitthestemiflectionofhortatives(Foris2000:151).HereandinTable13,XXXsignifiesapotentialcellwhoseactualexistenceisexcludedbyindependentgrammaticalfactors,forexamplethelanguagespropertycooccurrencerestrictions(Stump2001:41).(22)a.DefinitionoftheSochiapanChinantecpropertymapping:Where2ndor3rdandsingularorplural,;otherwise,b.SochiapanChinantecruleofparadigmlinkage:IfLisaverballexemehavingrasitsroot,thenthecontent-cellasitsform

-Althoughthepropertymappingin22aeffectsamorphosyntacticpropertyneutraliza-tion,thiseffectiscounteractedintheheterocliteinflectionofmotionverbs.forexample,theparadigmoftheverbwalkinTable13:inthisparadigm,thesingularandfirst-personpluralformsarebasedonthestem;theremainingpluralThismismatchbetweencontentandforminSochiapanChinantecrecallsacomparablemismatchinEnglish:althoughtheparadigmofrevealsthatEnglishverbinflectionmakesacontentivedistinctionbetweenthreepersons,nootherverbinthelanguageformallydistinguishesthreepersonsanywhereinitsForis(p.c.,2002)hasidentifiedsixteenverbsthatevincethispattern;theirglossesaresit,walk,standon,stayat,standup,liedown,enter,arrivehome,arriveat/stayatnonhome,arriveatnonhome,comenonhome,comehome,gohome,gononhome,leave,falldown,anddie. HETEROCLISISANDPARADIGMLINKAGE299forms,bycontrast,arebasedonthedistinctstemNeitherofthesestemshasaform-paradigminwhichnumberisdistinguishedinthesecondandthirdpersons;buttogether,theyalloweveryperson/numbercombinationinthecontent-paradigmoftohaveadistinctrealization.Inparticular,therulesofparadigmlinkagein22band23causecontent-cellssuchasthosein24tohavedistinctform-correspondentssuchasthosein25. CLASS A PASTAMBULATIVEHORTATIVEHODIERNAL PASTANDATIVE …… 13.Steminflectionoftheheterocliteverballexemewalk(intransitiveanimate)inSochiapanChinantec(Foris2000:75).DisyllabicverbsarerestrictedtoclassesBandC(p.77).(23)SochiapanChinantecruleofparadigmlinkage:WhereLisaverballexemeofthetypehavingsasitspluralcoradical,if2ndor3rd,thenthecontent-cellasitsform-correspondent.(24)a.2ndsingularfuture3rdsingularfuture2ndpluralfuture3rdsingularfuture(25)a.2ndfuture2ndfuture3rdfuture3rdfutureJustasaruleofparadigmlinkagethatengendersheteroclisismaycounteracttheeffectsofamorphosyntacticpropertyneutralization,itmaylikewisecounteractthephenomenonofdefectiveness.TheparadigmPofalexemeLisdefectiveifoneormoreofthemorphosyntacticpropertysetsthatarerealizedintheparadigmsofotherlexemesbelongingtothesamesyntacticcategoryasLremainunrealizedinP.Thereisacloseconnectionbetweendefectivenessandheteroclisis,becausespeakersmaycompensateforaparadigmsdefectivenessbyfillingitsgapswithformsdrawnfromadistinctparadigm„formswhoseinflectionclassmaycontrastwiththatofthedefectiveparadigmsownforms.Consider,forexample,theparadigmofSanskritinTable14.ThestemsinTable13belongtotheconjugationclassesAandBexemplifiedinTable12.Bothclassespossessanumberofsubtypes;forthisreason,thepatternsofstemmodulationinTable13arenotexactlylikethoseoftheClassAan

dBstemsinTable12.Nevertheless,thevarioussubtypesofClassAverbsarealikeinexhibitingapatternoftone/stressmarkingbywhichfourperson-numbercategoriesareformallydistinguished:firstpersonsingular,firstpersonplural,secondperson,andthirdperson;subtypesofClassBarealikeinshowingalessdifferentiatedpatternoftone/stressmarkinginwhichonlythirdpersonandnonthirdpersonareformallydistinguished;andsubtypesofClassCarealikeinfailingtodistinguishpersonandnumberbyanydifferenceoftone/stressmarking.DisyllabicverbsareingeneralrestrictedtoclassesBandC.SeeForis2000:56ff.fordiscussion. LANGUAGE,VOLUME82,NUMBER2(2006) yakne DECLENSION:GENERAL (n.)(n.) ¯ma¯mne¯mnas¯mna¯mni¯ma¯ni¯mabhyas¯mna¯mabhis¯masu¯mnos¯mabhya¯mnš¯, na¯manš 14.SupplementaryheteroclisisintheinflectionofSanskritliver(Whitney1889:§§398,432).Theoblique-caseformsofarebasedontheroot,whichinflectsasamemberoftheneuter-stemdeclension(fullyexemplifiedinTable14bytheparadigmname).Butlacksdirect-caseforms;intheirplace,direct-caseformsoftherootareused.Therootfollowsthegeneralconsonant-stemdeclension,andunlike,itgivesrisetoafullparadigmofrealizations;seeagainTable14.Thus,Sanskritpossessestwosynonymouslexemesforliver,one()seeminglydefective,theother()not.Onemightreasonablyask,however,whethertheparadigmofisgenuinelydefective:onecouldinsteadmaintainthatisheteroclite,buildingitsoblique-caseparadigmonthestemanditsdirect-caseparadigmonthestem.Clearly,alargerissueisatstakehere:canasingleform-cell(e.g.neuter[nominativeaccusative]singular)serveastheform-correspondentofcellsbelongingtotwodifferentcontent-paradigms(e.g.thecontent-cellsneuternominativesingu-neuternominativesingular)?Ifso,thenthestemmaybeassumedtofunctionbothastherootofthelexemeandasthecoradicalofthe);inthelatterinstance,theruleofparadigmlinkagein26drawsuponthecellsinsform-paradigmintherealizationofscontent-paradigm.(Rule26alsoentersintotherealizationofcertainotherheterocliteswhosebehaviorparallelsthatof,forexample,forearm,andbroth;Whitney1889:§432.)(26)Sanskritruleofparadigmlinkage:WhereLisaverballexemeLthatbelongstotheclassandhassasitscoradical,ifforsomedirect,thenthecontent-cellasitsform-correspondent.Thisanalysisimpliesageneraldistinctionbetweentwosortsofheteroclisis.Inin-stancesofNONOVERLAPPINGHETEROCLISIS,twoormoredistinctform-paradigmspartici-pateindefiningtherealizationofasinglecontent-paradigm.AnexampleistheheteroclisisofSanskritday(Table10):theform-paradigmsofthestemsparticipateinthedefinitionofscontent-paradigmandofnoother.IninstancesofOVERLAPPINGHETEROCLISIS,bycontras

t,aform-paradigmservesdoubleduty,participatingbothintherealizationofanonheteroclitecontent-paradigmand HETEROCLISISANDPARADIGMLINKAGE301(togetherwithoneormoreotherform-paradigms)intherealizationofasecond,hetero-clitecontent-paradigm;theheteroclisisofisofthislattertype.Summarizing,theparadigm-linkageapproachtoheteroclisisproposedherecorrectlyrepresentsinflection-classmembershipasapropertyofstems(rootsandcoradicals)ratherthanoflexemes,andbydrawingadistinctionbetweenalexemescontent-paradigmandastemsform-paradigm,itcorrectlyallowsstemsbelongingtomorethanoneinflectionclasstoparticipateinthedefinitionofasinglelexemesrealizations.Thisapproachreadilyaccountsforthedistinctionbetweenmorphosyntacticallycondi-tionedandmorphologicallyconditionedheteroclisis,forthefactthatdeponencyandheteroclisisfrequentlycoincide,andforthefactthatheteroclisissometimescounteractstheeffectsofmorphosyntacticpropertyneutralizationanddefectiveness.AsInowshow,theparadigm-linkageapproachisstronglymotivatedbytwokindsofevidence:theincidenceofsystematicheteroclisisandtheexistenceofauniversalconstraintonYSTEMATICHETEROCLISIS.Intheclearest,canonicalinstancesofheteroclisis,theinflectionclassesparticipatinginthedefinitionofaheteroclitelexemesparadigmParecompetitors,inthesensethatasinglestembelongingtoeitherclasscouldinflectforthefullrangeofmorphosyntacticpropertysetsinP;forinstance,thesoft-masculineandhard-masculinedeclensionclassesthatparticipateindefiningtheparadigmofCzech(Table1)arecompetitors,sincethenounroombuildsitsentireparadigmonasoft-masculinestem,andthenounbridgebuildsitsentireparadigmonahard-masculinestem.Thereare,however,lesscanonicalinstancesofheteroclisisinwhichthejuxtaposedinflectionclassesarenotcompetitorsatall,orcompeteonlyincertaincon-Idiscussseveralexamplesofsuchnoncanonicalpatternsofheteroclisisinthissection.AsIshow,thesedifferfromcanonicalinstancesofheteroclisisinasecondwayaswell.Canonicallyheteroclitepatternsofinflectiontendtobeunsystematicorexcep-tionalinthesensethattheyareembodiedbyonlyahandfulofparadigms;thus,ifallheteroclisiswerecanonical,thelexicalapproachtoheteroclisisin1mightseemlikeaviablemodeofanalysis.Butlesscanonically,heteroclisismayplayacentralroleinaninflectionalsystemsdefinition,inthateachmemberofanentireclassoflexemesmayinvolvemorethanoneinflectionclassinitsrealization.Instancesofthissortcannotbeplausiblyaccountedforbymeansofpiecemeallexicalstipulationssuchasthosein2,butinsteadnecessitatethepostulationofrulesofmoregeneralscope;suchinstancesthereforefavortheparadigm-linka

geapproachtoheteroclisisadvocatedhere.Inthissection,Idiscusstwomaintypesofnoncanonicalheteroclisis:(i)heteroclisisinvolvingnoncompetinginflectionclasses(§4.1)and(ii)heteroclisisinvolvinginflec-tionclassesthatareinpartialcompetition(§4.2).Heteroclisisoftype(i)maybeoftwosubtypes,accordingtowhetherthejuxtaposedinflectionclassesareassociatedwithdistinctmorphosyntacticpropertiesorwithdistinctsyntacticcontexts.Heteroclisisoftype(ii)mayalsobeoftwosubtypes,accordingtowhetherthecompetitionofthejuxtaposedinflectionclassesisconditionedlexicallyormorphosyntactically.Iemphasizethatinthefollowingdiscussion,thetermheteroclisisistobeunder-stoodasreferringtothepropertyofanylexemewhoseinflectionalparadigmcontainsformsbuiltuponstemsbelongingtotwoormoredistinctinflectionalclasses,whetherTheterminologicaldistinctionproposedhereismeanttorecalltheparalleldistinctionbetweennonover-lappingandoverlappingsuppletionmadebyJuge(1999:186).Lesscanonicaldoesnot,ofcourse,meanlessusual.SeeCorbett2003fordiscussionofthecanonicalapproachinlinguistictypology. LANGUAGE,VOLUME82,NUMBER2(2006)ornottheseclassesarefullcompetitorsandwhetherornotthejuxtapositionoftheseclassesissystematic.Thisis,inasense,adeparturefromordinaryusage,inwhichheteroclisistendstobeequatedwithwhatIamcallingcanonicalheteroclisis(thejuxtapositionofcompetinginflectionalclassesincertainsporadicparadigms).YSTEMATICHETEROCLISISINVOLVINGNONCOMPETINGINFLECTIONCLASSES.Whenstemsbelongingtononcompetinginflectionclassesparticipateinthedefinitionofalexemesparadigm,thechoiceamongthesestemsmaybedeterminedbythemorpho-syntacticpropertysettoberealizedorbythesyntacticcontextinwhichtherealizationistobeused.Inthissection,Idiscussinstancesofheteroclisisofbothofthesesorts.AsIshow,botharecharacteristicallysystematicandthereforefavortheparadigm-linkageapproachtoheteroclisisoverthelexicalapproach.HEPRINCIPALPARTSPHENOMENON.Inthesimplestsystemsofinflectionclasses,theentireparadigmofalexemeLmaybedeterminedbythemembershipofLsrootinaparticularinflectionclass;inEnglish,forexample,theentireparadigmofthelexemeisdeterminedbythemembershipoftherootintheclassofweakverbs.Butalexemesparadigmisnotalwaysfullydeduciblefromthemembershipofitsrootinaparticularinflectionclass.InSanskrit,forinstance,thefactthattherootthelexemeshinebelongstothefirstconjugationaccountsforthemorphologicalrealizationofspresent-systemforms(i.e.itspresentindicative,presentopta-tive,imperfect,andimperativeforms),butnot,forexample,forthatofitsaoristforms(whichfollowtheroot-aoristconjugation).Indeed,

thepatternofaoristinflectionexhib-itedbyissharedbyverbswhosepresent-systeminflectionfollowsaconjugationotherthanthefirst(e.g.beashamed[3rdconjugation],beequalto[5thgreet[6thconjugation],andsplit,bore[7thconjugation]);bythesametoken,severalfirst-conjugationverbsformtheiraoristdifferentlyfromscatter[-aoristconjugation],beadapted[reduplicatedaoristprotect[-aoristconjugation],doubt[-aoristconjugation],cryout[-aoristconjugation]).Thus,ingeneral,theinflection-classmembershipthatdeterminesaSanskritverbspresent-systeminflectionneitherpre-dictsnorispredictedbytheinflection-classmembershipthatdeterminesitsaoristToaccountforsuchinstances,traditionalgrammarianshavecustomarilyassociatedalexemewithasetofPRINCIPALPARTS:asetofformsrealizingthatlexemewhichonlyjointlydeterminetheentiretyofitsparadigm.Traditionalgrammarianshavegenerallyassumedthateachofalexemesprincipalpartsisafullyinflectedword,butonecouldjustaseasilyassumethatitisastemfromwhichsomepartofthelexemesparadigmisdeducible.TheprincipalpartsofSanskrit,forexample,mightbeassumedtoincludethefirst-conjugationpresent-systemstemandtheroot-aoriststemThisprincipal-partsphenomenonisonekindofsystematicheteroclisis.GivenanylexemeL,LisbydefinitionheterocliteifstemsbelongingtotwoormoredistinctinflectionclassesparticipateinthedefinitionofLsparadigm.Incanonicalheteroclisis,theparticipatinginflectionclassesarecompetitorsinthesensethatforeachoneofThereare,tobesure,occasionalcorrelationsbetweenpresent-systemandaorist-systeminflectionclassesinSanskrit;forinstance,verballexemesfollowingthetenthconjugationinthepresentsystemgenerallyalsofollowthereduplicatedaoristconjugation(althoughthereverseisnotinvariablytrue;Stump2005a).Theexistenceofsuchcorrelationsdoesnot,however,alterthefactthatingeneral,averbspresent-systeminflectionanditsaorist-systeminflectionaremutuallyunpredictable. HETEROCLISISANDPARADIGMLINKAGE303theseclasses,therearenonheteroclitelexemesthatbuildtheirentireparadigmonastembelongingtothatclass.Butininstancesoftheprincipal-partsphenomenon,theparticipatingclassesarecomplementaryratherthancompeting:foreachcellinLsparadigm,therulesavailabletorealizenevermakeessentialreferencetomorethanoneoftheparticipatingclasses.InSanskrit,forexample,rulesrealizinginstancesofthepropertysetaoristXneverrefertotheinflectionclassesin27,onlytothosein28;similarly,rulesrealizinginstancesofthepropertysetspresentXneverrefertotheinflectionclassesin28,onlytothosein27.Thus,intheprincipal-partsphenomenon,wehaveheteroclisisinvolvingnoncompetinginflec

tionclasses.(27)Present-systeminflectionclasses(28)Aoristinflectionclasses1stconjugation6thconjugationroot-aorist2ndconjugation7thconjugation3rdconjugation8thconjugationreduplicatedaorist4thconjugation9thconjugation5thconjugation10thconjugationTheparadigm-linkageapproachtoheteroclisis„unlikethelexicalapproach„affordsageneralaccountoftheprincipal-partsphenomenon.SupposethatLisalexemebelongingtoasyntacticcategorywhosemembersexhibittheprincipal-partsphenom-enon.Inanysuchinstance,Lissubjecttoadifferentruleofparadigmlinkageforeachofitsprincipalparts.TheSanskritlexeme,forexample,issubjecttorulessuchas29a,b.(29)Sanskritrulesofparadigmlinkage:a.WhereLisaverballexemehavingsasitspresent-systemcoradical,ifpresentXimperfectX,thenasitsform-b.WhereLisaverballexemehavingsasitsaorist-systemcoradical,ifaoristX,thenasitsform-correspondent.isaninstanceofpresentX,29aentailsthattherealizationofthecontent-isthatoftheform-cell,whereisamemberofthefirstconjugation;whereisaninstanceofaoristX,29bentailsthattherealizationisthatof,whereisamemberoftheroot-aoristconjugation;andsoon.Merelexicalstipulationssuchas2wouldfailtocapturethesystematicgeneralizationsexpressedby29.YNTACTICALLYCONDITIONEDINFLECTIONCLASSES.Ininstancesoftheprincipal-partsphenomenon,thejuxtaposedinflectionclassesareassociatedwithcontrastingmorpho-syntacticpropertysets;intheinflectionoftheSanskritverbshine,forinstance,thefirstconjugation(towhichthestembelongs)andtheroot-aoristconjugation(towhichbelongs)areassociatedwiththerespectivepropertysetsspecifiedinOnthisview,itismisleadingtocharacterizeaheterocliteparadigmasamixedparadigm:ifaparadigmisheteroclitebecauseitembodiestheprincipal-partsphenomenon,thenallparadigmsbelongingtothesamecategorywillbesimilarlyheteroclite;thatis,theresimplywontbeanyunmixedparadigmsofthatcategory.Nevertheless,evenaparadigmembodyingtheprincipal-partsphenomenoninvolvesacontrastbetweendefaultandintrusiveinflectionclasses(inthesenseofthecriteriagivenin16).Thesestatementsare,Iwouldargue,trueeveninthecaseofverbswhosepresent-systeminflectionfollowsthetenthconjugationandwhoseaorist-systeminflectionfollowsthereduplicated-aoristconjugation(cf.n.32),sincethecorrelationbetweenthesetwoconjugationsisnotbothanecessaryandasufficientone. LANGUAGE,VOLUME82,NUMBER2(2006)29a,b.Butheteroclisismayalsoinvolvethejuxtapositionofnoncompetinginflectionclassesdistinguishedbytheirassociationwithcontrastingsyntacticcontexts.AfamiliarexampleofthistypeofheteroclisisisthatofGermanadjectivaldeclension.InGerman,adjectivesinflectaccordingt

othreedifferentdeclensions:thestrong,themixed,andtheweak.Eachdeclensionsuppliesaformforeachgender/casecombinationinthesingularandagender-neutralformforeachcaseintheplural;theparadigmsofgoodinTable15illustrate.Asitsnameimplies,themixeddeclensionisheteroclite:itfollowsthestrongdeclensioninthesingularofthedirectcases,andotherwisetheweakdeclension.Eventhoughallthreedeclensionssupplyrealizationsforthesamemorphosyntacticpropertysets,theyarenotincompetition,sincetheirrealizationsareassociatedwithcomplementarysyntacticcontexts.Inattributiveposition,strong-declensionformsareusedintheabsenceofadeterminer;weak-declensionformsareusedinthepresenceofasubclassofdeterminersincludingthis;andmixed-declensionformsareusedinthepresenceofadistinctsubclassofdeterminersincludingtheindefinitearticleSTRONGMIXEDWEAKSINGULARPLURALSINGULARPLURALSINGULARPLURALMASCNEUTFEMMASCNEUTFEMMASCNEUTFEMgutergutesgutegutegutergutesgutegutengutegutegutegutengutengutesgutegutegutengutesgutegutengutengutegutegutengutengutengutergutergutengutengutengutengutengutengutengutengutemgutemgutergutengutengutengutengutengutengutengutenguten15.Strong,mixed,andweakdeclensionsofgoodinGerman.Becausetheheteroclisisofthemixeddeclensioniscompletelysystematic,involvingeveryadjectiveinthelanguage,thelexicalapproachtoheteroclisisdoesnotaffordaplausibleframeworkforitsgrammaticalrepresentation.Theparadigm-linkageapproachdoes.Onthislatterapproach,Germanadjectivesareuniformlysubjecttotheruleofsteminferencein30.GivenanyadjectivallexemeLhavingrasits(weak-declension)rootandsasits(strong-declension)coradical,thecontent-cellhasbothasform-correspondents:thefirstoftheseisthedefaultform-correspondentsuppliedbyrule5,whilethelatterissuppliedbyalanguage-specificruleofparadigmlinkage.Inordertoaccountforthefactthatanadjectivallexemestwostemsyieldcomplementaryratherthancompetingwordforms,Iassumethatthelatterruleitselfimposesacontextualrestrictionontherealizationsofanadjectivallexemescoradical,asin31.Inaccordancewith31a,realizationsofstrongform-cellsareusedindeterminer-freecontexts;inaccordancewith31b,realizationsofstrong,singular,direct-caseform-cellsarealsousedinthecontextof-typedeterminers;andbydefault,realizationsofweakform-cellsareusedeverywhereelse.(30)Germanruleofsteminference:IfanadjectivallexemeLhasarootrbelong-ingtotheweakdeclension,Lalsohasanotherwiseidenticalcoradicalbelong-ingtothestrongdeclension.(31)Germanruleofparadigmlinkage:IfLisanadjectivallexemehavingsasitsstrong-declensioncoradical,thenthecontent-cellasaform-correspondent,whos

erealizationisusedattributively:a.indeterminer-freecontexts,andAdjectivesinpredicativepositionarenotdeclinedinGerman. HETEROCLISISANDPARADIGMLINKAGE305b.inthecontextof-typedeterminers,providedthatsingularXforsomedirectcaseInsummary,noncompetinginflectionclassesmaybejuxtaposedwithinasingleparadigmineitheroftwoways:noncompetingclassesmaybeassociatedwithcontrast-ingmorphosyntacticproperties(theprincipal-partsphenomenon)orwithcomplemen-tarysyntacticcontexts.Heteroclisisinvolvingeithersortofjuxtapositionisnotexceptionalatall,buthighlysystematicinthoselanguagesinwhichitappears.Anadequateaccountofeithersortofjuxtapositioninvolvesgeneralizationsoverentireclassesoflexemes,necessitatingthesortsofrulesaffordedbytheparadigm-linkageapproachtoheteroclisis.YSTEMATICHETEROCLISISINVOLVINGPARTIALLYCOMPETINGINFLECTIONCLASSESCanonicalinstancesofheteroclisis(inwhichalexemesinflectiondependsonstemsbelongingtotwoormorefullycompetinginflectionclasses)andheteroclisisinvolvingnoncompetinginflectionclasses(§4.1)areatoppositeextremesofacontinuum:thatis,thereareintermediateinstancesofheteroclisisinvolvinginflectionclassesthatareinpartialbutnotfullcompetition.AsIshowhere,instancesofthissortareoftwokinds,accordingtowhetherthecompetitionofthejuxtaposedinflectionclassesisrestrictedmorphosyntacticallyorlexically.These,too,motivatetheparadigm-linkageapproachtoheteroclisis.EFECTIVEINFLECTIONCLASSES.Someparadigmsareheteroclitebecausemembersofoneinflectionclasssystematicallyinflectasmembersofadistinctclassintherealizationofparticularmorphosyntacticpropertysets.Consider,forexample,thein-flectionofSanskritaorists.InSanskrit,therearesevenaoristconjugationclasses(thoselistedin28).Eachexhibitsadistinctpatternofinflectionintheaoristactive.Inthemiddlevoice,however,averbbelongingtotheroot-aoristconjugation(whoseformslackanystem-formingsuffix)insteadfollowseitherthe-aoristconjugation(inwhichthestemhasan-suffix)orthe-aoristconjugation(inwhichthestemhasan-suffix)„theformeriftheverbbelongstoPaclass,thelatterifitbelongstoclass.Theactiveinflectionoftheverbgive,forexample,followstheroot-aoristconjugation,butitsmiddleinflectioninsteadfollowsthe-aoristconjugation(Table16).Similarly,verbsbelongingtothe-aoristconjugationfollowtheconjugationinthemiddlevoice,andmostverbsbelongingtothe-aoristconjugationinflectasmembersofthe-or-aoristconjugationinthemiddlevoice. ¯va¯tam¯ta¯tha¯ta¯ma¯ta ¯ ¯ ¯tha¯taataAORIST CONJUGATION:ROOT-AORIST give16.HeterocliteinflectionofSanskritgive(Whitney1889:§§829,882).Shadedwordsinflectasmembersoft

he-aoristconjugation.Regardingtheabsenceofstem-finalinthe2nd-and3rd-personsingularaoristmiddleformsofseeWhitney1889:§§881c,884. LANGUAGE,VOLUME82,NUMBER2(2006)Inheterocliteparadigmssuchasthatof,theparticipatinginflectionclassesareonlypartialcompetitors:theyarealternativesintheactivevoice,butnotinthemiddlevoice.Inparticular,theyarepartialcompetitorsbecausetheroot-aoristconjugationisDEFECTIVEINFLECTIONCLASS:itdoesnotfurnishrealizationsforallofthemorphosyn-tacticpropertysetsforwhichthe-aoristdoes.Notethatsayingthataparadigminvolvesadefectiveinflectionclassisnotthesamethingassayingthatitisadefectiveparadigm;theparadigmof,forinstance,isnotdefective.Note,too,thatinflection-classdefec-tivenessisdifferentfromthephenomenonofdefaultinflectionalpatternsdiscussedin§2.Itisnotsimplythattherootaoristandthe-aoristshareadefaultpatternofinflectionformiddleforms:inmiddleparadigmssuchasthatof,therootaoristisclearlygivingwaytothe-aorist,sincetheformsexhibitastem-final(oroneofitssandhialternants)„thedefiningmarkoftheIntheparadigm-linkageapproachtoheteroclisis,suchinflection-classdependenciescanbestraightforwardlyattributedtotheinteractionofrulesofsteminferencewithrulesofparadigmlinkage.Suppose,forinstance,thatifaSanskritverballexemehassasitsaoriststem,thenitisassignedanaoristmiddlecoradicals,whoseformandinflection-classmembershiparedeterminedfromsbytheruleofsteminferencein32;onthatassumption,thepropertiesofaoristmiddleformsfollowfromtheruleofpara-digmlinkagein33.(32)Sanskritruleofsteminference:WherelexemeLhassasitsaoriststem,itsaoristmiddlecoradicalsislikes,exceptthatifsbelongstotheroot-,-,or-aoristconjugation,thena.ifsbelongstothethensbelongstothe-aoristconjugation,andb.otherwisesbelongstothe-aoristconjugation.(33)Sanskritruleofparadigmlinkage:WhereLisaverballexemehavingsasitsaoristmiddlecoradical,ifaoristmiddleX,thenthecontent-cellasitsform-correspondent.ECRUITEDINFLECTIONCLASSES.Intheforegoingexampleofpartialcompetitionamonginflectionclasses,thecompetitionispartialbecauseoneoftheparticipatinginflectionclassesisdefective;thecontextsinwhichnocompetitiontakesplacecanthereforebecharacterizedinmorphosyntacticterms(asin33).Thereare,however,instancesofpartialcompetitionthatdonotinvolveinflection-classdefectivenessandinwhichthelackofcompetitionisassociatedwithlexicalcontextsratherthanwithamorphosyntacticcontext.TheseinvolveRECRUITEDINFLECTIONCLASSES,whichfunctionasordinaryinflectionclassesinsomeparadigmsbutwhichtakeonaspecificmorpho-syntacticfunctioninothers.Heteroclisisinvolvingrecruitedinflectioncl

assesoftenarisesinlanguageswithbothgenderdistinctionsanddeclension-classdistinctions.Insuchlanguages,membersofparticulardeclensionclassesmay,bydefault,beassignedtoparticulargenderclasses(Corbett1991:33ff.),andifanadjectivesstemsaresubjecttosuchdefaultgenderassignments,thenintheinflectionofthatadjective,declension-classdistinctionsareineffectrecruitedfortheexpressionofgenderdistinctions.Consider,forexample,theinflectionofadjectivesinAncientGreek.InAncientGreek,membersofthegeneral-stemdeclensionexemplifiedbytheparadigmofhorseinTable17aremascu-linebydefault(thoughasmallnumberoffemininenounsalsofollowthisdeclension);membersofthecontrasting-stemdeclensionexemplifiedbytheparadigmofgiftinTable17areinvariablyneuter;andmembersofthefeminine-stemdeclension HETEROCLISISANDPARADIGMLINKAGE307exemplifiedbytheparadigmoflandinTable17areinvariablyfeminine.Accordingly,thesethreedeclensionclassesarerecruitedforthesystematicexpressionofathree-waygendercontrastintheinflectionofAncientGreekadjectiveswithstem-(Jannaris1968:139f.);thus,theinflectionoftheadjectiveworthyinTable17isheteroclite,followingthe-typedeclensioninitsmasculinesubparadigm,the-typedeclensioninitsneutersubparadigm,andthe-typedeclensioninitsfemininesubparadigm.Thisistheregularpatternofinflectionforvastnumbersof-stemadjectives.Becausethe-typedeclensionincludessomefemininenounsroad),thegeneral-stemandfemininedeclensionsarecompetitorsintheinflectionoffemininenouns;butintheinflectionofanadjectivebelongingtotheclass,thesetwodeclensionsarenotincompetitionpreciselybecausetheyarerecruitedtoexpresscontrastinggenders. MASCNEUTFEMN. A. V.¯axiôaxiôchôrainaxioinN. V. 17.DeclensionofthreenounsandtheadjectiveworthyinAncientGreek.Theparadigm-linkageapproachtoheteroclisisaffordsageneralaccountofsuchgender-boundheteroclisisinthedeclensionofadjectives.Forinstance,ontheassump-tionthatsrootisitsmasculinestem(amemberofthegeneraldeclension),theformandinflection-classmembershipofitsneuterandfemininecoradi-calscanbepredictedbyaruleofsteminference.(34)Greekruleofsteminference:IfanadjectivallexemesrootXbelongstothegeneral-stemdeclension,thenitsotherwiseidenticalneutercoradicalbelongstotheneuter-stemdeclensionanditsfemininecoradicalXtothefeminine-stemdeclension.Giventhisrule,theadjectivallexemeisthensubjecttothetworulesofparadigmlinkagein35andto5aswell.Rule5definessmasculinesubparadigm:whereisaninstanceofmasculineX,5entailsthattherealizationofthecontent-cellCertainmasculinenouns(e.g.(m.)youngman)havestemsin,buttheyfollowthemasculine-stemdec

lension,whichdiffersfromthatofinthenominativeandgenitivesingular.Certainadjectiveshavingstemsinfollowthegeneral-stemdeclensioninboththemasculineandfemininegenders,forexampletame;indeed,thisisquitegenerallytrueof-stemadjectives,forexamplesonorous(Jannaris1968:142).Intheinflectionofsuchadjectivesoftwoendings,rule34isoverriddenbyamorenarrowlyapplicableruleofsteminferenceequatingthefemininestemwiththemasculinestem. LANGUAGE,VOLUME82,NUMBER2(2006)isthatoftheform-cell.Therulesofparadigmlinkagein35a,bdefineitsneuterandfemininesubparadigms,respectively:whereisaninstanceofneuterX,35aentailsthattherealizationofisthatof,wherebyisamemberoftheneuter-stemdeclension;andwhereisaninstanceoffeminineX,35bentailsthattherealizationofisthatof,whereisamemberofthefeminine-stemdeclension.(35)Greekrulesofparadigmlinkage:a.IfLisanadjectivallexemehavingsasitsneutercoradicalandneuterX,thenthecontent-cellasitsform-correspondent.b.IfLisanadjectivallexemehavingsasitsfemininecoradicalandfeminineX,thenthecontent-cellasitsform-corres-Inthissection,Ihaveexaminedvariousinstancesofnoncanonicalheteroclisisinwhichnoncompetingormerelypartiallycompetinginflectionclassesarejuxtaposedwithinaninflectionalparadigm.Eachoftheseisdistinguishedfromcanonicalinstancesofheteroclisisbyitssystematicityacrossanentirelexemeclass.Thereasonsforthissystematicityare,ineachcase,clear.Mostobviously,alexemewhoserootbelongstoadefectiveinflectionclasscannothaveacompleteparadigmwithoutbeingheteroclite;thus,heteroclisismustbesystematicinlanguageswithdefectiveinflectionclasses.Thesameisevenmoredramaticallytrueinlanguagesexhibitingtheprincipal-partsphenomenon,inwhichalloftheinflectionclassesrelevantfortheinflectionofstemsbelongingtoaparticularsyntacticcategoryareineffectdefective.Iftheinflectionofsomelexemeclassdependsonasystemofsyntacticallyconditionedinflectionclasses,itisnotpossibleforamemberofthatlexemeclasstobeusedintherangeofexpectedsyntacticcontextsunlessitsparadigmisheteroclite;heteroclisismustthereforebesystematicinlanguageswithsyntacticallyconditionedinflectionclasses.Andifcertaininflectionclassesarerecruitedfortheexpressionofsomemorphosyntacticdistinctionintheinflectionofsomelexemeclass,membersofthatlexemeclassmustbesystemati-callyheterocliteiftheirparadigmsaretoreflectthatdistinction.Forrepresentingsys-tematicpatternsofinflection-classjuxtapositionsuchasthoseconsideredinthissection,piecemeallexicalstipulationssuchas2arenotacrediblealternativetotheparadigm-linkageapproach,whoserulesregulatetherealizationofwholeclassesoflexeme

s.CONSTRAINTONHETEROCLISIS.Theincidenceofheteroclisisinaninflectionalsystemrevealsanimportantfactaboutinflectionalparadigms.Likeacrystallinemin-eral,eachexhibitsaparticularpatternofcleavage:justasamineralmaybemorelikelytosplitapartalongoneplanarsurfacethanalonganother,sothesplittingofaninflec-tionalparadigmintosubparadigmsbelongingtoseparateinflectionclassesmaybemorelikelyalongonemorphosyntacticboundarythanalonganother.Intheparadigmofspring(Table1),theinflection-classsplitisanespeciallycleanone,separatingthesingularandpluralpartsoftheparadigm.SplitsofthiskindarenotunusualinCzech:comparableexamplesaretheparadigmsofstrap,andsoon.Crystallinemineralsdontalwaysbreakapartalongaplaneofcleavage;lesspredict-ablebreaks(calledfractures)canalsooccur.Inthesameway,manyCzechnounshaveheterocliteparadigmsinwhichthesplitbetweendeclensionclassesisnotabsolutelyclean,fromamorphosyntacticperspective;examplesaretheparadigmsofthenouns HETEROCLISISANDPARADIGMLINKAGE309presidentandservant(Tables6and8).Butintuitively,eventheseparadigmsseemtocleaveprimarilyalongthenumberboundary.Inthissection,Iexaminethevariouswaysinwhichheterocliteparadigmsmaysplitintosubparadigmsbelongingtodistinctinflectionclasses.Iarguethatthereisauniver-salconstraintonthecoexistenceofcontrastingpatternsofinflection-classsplit;asIshow,thisconstraintamountstoawell-formednessconditiononrulesofparadigmlinkage.Beforeproceeding,itisusefultohavesomeprecisedefinitionsfordistinguish-ingdifferenttypesofsplit.OMEDEFINITIONS.ConsidertheheterocliteparadigmoftheCzechnounspring(Table1),whosemembersinflectasmembersoftwoinflectionclasses(thesoft-masculineandhard-masculinedeclensions)andexpresstwoproperties(singular,plural)oftheinflectionalcategoryofnumber.Inthisparadigm,thecorrelationbetweenpropertiesofnumberandinflectionclassesisperfect:thecellsinTable1thatfollowthesoft-masculinedeclensionareexactlythosethatarespecifiedassingular,andthosethatfollowthehard-masculinedeclensionareexactlythosethatarespecifiedasplural.Asawayoffacilitatingcomparisonsbetweensuchperfectparadigmsandothersthatarelessperfect,Iproposethefollowingempiricalmeasureofaheterocliteparadigmsdegreeofmorphosyntacticcorrelation:givenaheterocliteparadigmPandaninflec-tionalcategoryAhavingv,...,vasitspossiblevalues,theDEGREEOFTIONINPisaratiox/y,whereyisthenumberofcellsinPandxisthesumof,...,n,wherenisthelargestnumberofcellsinPthatcarrythespecificationvandinflectasmembersofthesameinflectionclass.Bythismeasure,thedegreeofnumbercorrelationintheparadigmofis14/14(i.e.1.0),sincea

llsevencellscarryingthespecificationsingularinflectasmembersofthesoft-masculinedeclensionandallsevencellscarryingthespecificationpluralinflectasmembersofthehard-masculinedeclension.Bythissamemodeofcalculation,thedegreeofnumbercorrela-tionis.86(12/14)intheparadigmofpresident(Table6),sinceallsevencellscarryingthespecificationpluralinflectasmembersofthehard-masculinedeclensionandfiveofthesevencellscarryingthespecificationsingularinflectasmembersofthehard-femininedeclension;similarly,thedegreeofnumbercorrelationis.79(11/14)intheparadigmofservant(Table8).Theintuitionthatthesenominalparadigmscleaveprimarilyalongthenumberbound-aryisconfirmedbycalculatingtheirdegreeofcasecorrelation:thisamountsto.50insparadigm,.64insparadigm,and.57insparadigm;ineachparadigm,thedegreeofnumbercorrelationexceedsthedegreeofcasecorrelation.Inwhatfollows,IrefertoaninflectionalcategoryAasanABSOLUTECORRELATEaparadigmsheteroclisisifandonlyifthedegreeofA-correlationinthatparadigmis1.0;thus,numberisanabsolutecorrelateofheteroclisisintheparadigmofrefertoaninflectionalcategoryAasaMAXIMALCORRELATEofaparadigmsheteroclisisifandonlyifthedegreeofA-correlationishigherthananyotherinflectionalcategorysdegreeofcorrelationinthatparadigm;thus,numberisamaximalcorrelateofhetero-clisisintheparadigmsofaswellasthatof.Finally,Irefertoheterocliteparadigmsthat(likethatof)haveanabsolutecorrelateasparadigms,andtothosethat(likethoseof)lackanyabsolutecorrelateasTheincidenceoffracturedparadigmsdisconfirmstheslabcodiciltoCarstairss(1987:81ff.)paradigmeconomyprinciple,theantecedentofthenoblurprinciple;cf.n.5. LANGUAGE,VOLUME82,NUMBER2(2006)CLOVENPARADIGMSABSOLUTECORRELATE.Itisstrikinghowfewinflectionalcategoriesappearasabsolutecorrelatesofheteroclisisinagivenlanguage.Thispaucityisareflectionofthewidelyobservabletendencyin36.(36)Clovenparadigmsoflexemesbelongingtothesamesyntacticcategorytendtohavethesameabsolutecorrelate.Thatis,justassomeminerals(e.g.covelliteorstibnite)alwayscleavealongasingleplane,soalsotheparadigmsofheteroclitelexemesbelongingtothesamesyntacticcategorytendalwaystocleaveinthesamedirection.Forinstance,Czechnounswithclovenparadigmshavenumberastheirabsolutecorrelate;heteroclitenounparadigmswhoseinflection-classsplitscorrelatecleanlywithanyotherinflectionalcategoryare,tomyknowledge,nonexistentinCzech(forinstance,thereisnonominalparadigmthatfollowsthehard-stemdeclensioninthenominative(singularandplural)butthesoft-stemdeclensionelsewhere).Numberappearsasanabsolutecorrelateofheteroclisisinclovennounparad

igmsinmanyotherlanguages,includingPolish(e.g.(propername)),Lithuanian(e.g.man),AncientGreek(e.g.ambassador),Latin(e.g.city),andsoon.Butnumberisnotalwaysthesharedabsolutecorrelateinclovennounparadigms.InSanskrit,forexample,clovennounparadigmshavecaseastheirabsolutecorrelateliver(Table14),water,andsoon).InFula,clovennounparadigmshaveevaluativetypeastheirabsolutecorrelate:forinstance,the)bellyfollowstheA-gradeinflectioninitsevaluative(diminutiveandaugmentative)formsandtheC-gradeinflectioninitsneutral(nonevaluative)forms(Table18);cf.also)hut,)donkey,)he-goat,andsoon(Arnott Diminutivesingularsingular(noun class 5)-umdeer-umunor-gumAugmentativesingularNeutralsingularree-duunor-du 18.HeterocliteinflectionofFula)belly(Arnott1970:89f.,120f.).Inlanguagesinwhichmembersofaparticulardeclensionclassareassignedbydefaulttoaparticulargenderclass,genderisnaturallyanabsolutecorrelateofheter-oclisisinclovenadjectivalparadigms;thisistrue,forexample,oftheAncientGreekadjectivesconsideredin§4.2.Inclovenverbalparadigms,theabsolutecorrelateofheteroclisisisoftentense(asinthecaseofArmeniancome(Table24below),Bulgariansleep,OldEnglishknow(Table11))oraspect(asinthecaseofHebrewapproach,Latinhit),thoughotherinflectionalcategoriesarealsoobservableasabsolutecorre-lates,forexample,mood(asinthecaseofNgiyambaawalk(Donaldson1980: HETEROCLISISANDPARADIGMLINKAGE311158)),finiteness(asinthecaseofHausatearoff(Newman2000:708)),number(asinthecaseofSochiapanChinantecenter),andsoon.Althoughitisthenormforclovenparadigmsbelongingtothesamesyntacticcategorytohavethesameabsolutecorrelate,theydontalways.Justassomeminerals(e.g.pyrolusite,enargite)maycleavealongeitheroftwononparallelplanes,clovenpara-digmsbelongingtothesamecategoryoccasionallyexhibitcontrastingabsolutecorre-lates.Takelma(Penutian)furnishesaclearcaseofthissort.InTakelma,theparadigmsofcertainheterocliteverbscleavecleanlyalongtheaorist/nonaoristboundary:forexam-ple,theverbshine(Table19)inflectsintheaoristasamemberofconjugationclassIII(theusualclassfortransitiveverbs)butinflectselsewhereasamemberofclassII(thatofmanyintransitiveverbs);similarly,theverb(Sapir1922:166)inflectsintheaoristasamemberofclassIIbutinflectselsewhereasamemberofclassI(towhichotherintransitivesbelong).Inparadigmsofthissort,theabsolutecorrelateofheteroclisisistense.Certainotherheterocliteverbs,however,cleavecleanlyalongthesecond-person/nonsecond-personboundary.Forinstance,thework(Table20)inflectsinthesecondperson(singularandplural)asamemberofclassII,inf

lectingelsewhereasamemberofclassIII.Inparadigmsofthissort,theabsolutecorrelateofheteroclisisisperson.Inshort,tenseandpersoncoexistasabsolutecorrelatesofheteroclisisintheparadigmsofTakelmaverbs. CONJUGATION:CLASS 2sg1sg1pl [Sstop M kill (with 3sgobject agreement) ¯ ´K!ALshinet!omoma´[nt!omoma`tdouma´ndoumanaga´mt!omomana`kal-we´k!ala[nal-we´k!alatal-we´k!alt eeal-we´k!alp igamal-we´k!alana`khana´[sde[hana´[sdamhana´[sikha´n[sdeeha´n[sigam 19.HeterocliteinflectionofTakelmashine(Sapir1922:165f.,170f.,183).Inthedatabaseofheterocliteparadigmsassembledforthisstudy,nolanguageex-hibitsmorethantwoabsolutecorrelatesofheteroclisisintheparadigmsoflexemesbelongingtothesamesyntacticcategory(andtwoisitselfrare).Whilethislimitmaybeacoincidentalfeatureofthesampledlanguages,thefollowingfactisneverthelessaclearone.(37)ThenumberofabsolutecorrelatesofheteroclisisinagivenlanguageisalwaysmuchsmallerthanthatlanguagesfullinventoryofinflectionalThisfactisnot,ofcourse,alogicalnecessity.FRACTUREDPARADIGMSMAXIMALCORRELATE.Infracturedparadigms(thoseheterocliteparadigmslackinganyabsolutecorrelate),itissometimespossibletodistin-guishamaximalcorrelate(inthesensedefinedin§5.1).Maximalcorrelatesoftenparticipateinthefollowingtendency.TheTakelmaparadigmsinTables19and20arepartial,listingonlythoseformsspecificallycitedbySapir;notethatsingularandpluralsubjectagreementarenotdistinguishedinthird-personverbformsinTakelma(Sapir1922:159).This,therefore,iswheretheanalogyofheterocliteparadigmstocrystallinemineralsends,sincesomeminerals(e.g.sphalerite)haveasmanyassixplanesofcleavage. LANGUAGE,VOLUME82,NUMBER2(2006) wana k CONJUGATION: 2sg3sg2pl1sg2sg3sg1pl2pl1sg2sg1pl2pl Inferential SstopDOuM kill (with 3sg ¯ I-HEGWEHAGWwork ¯ ma´nd ¯ mada´[d ¯ ma nk‡d ¯ ma´t‡ba[d ¯ nt!omoma´tt!omomana´k‡t!omoma´´t‡p‡t!omõmhana´[sde[hana´[sdamhana´[shana´[sdap‡hana´[sik‡hegwehegwegwaaheegwaahegwegwenhegwehegwegwaa(= -k )heegwagwagwagwa ‡dõmk‡dõmk‡anak‡ dõmk‡[e tpgwa gwa ‡p‡ha´n[sdeeha´n[sda[ha´n[sdaba[ha´n[´n[sd ¯ sga ‡t‡k‡ha´n[sk!e 20.HeterocliteinflectionofTakelmawork(Sapir1922:165f.,170f.,182).(38)Theinflectionalcategoriesappearingasmaximalcorrelatesinfracturedpara-digmstendtobethesameasthoseappearingasabsolutecorrelatesinclovenparadigmsbelongingtothesamesyntacticcategory.Forinstance,justasnumberistheabsolutecorrelateofheteroclisisintheclovenparadigmofCzech,itislikewisethemaximalcorrelateofheteroclisisinthefracturedparadigmsof.Additionalexamplesoftendency38aregiveninTable21:ineachinstance,theinflectionalcategor

ylistedinthethirdcolumnappearsasanabsolutecorrelateofheteroclisisintheclovenparadigmlistedinthefourthcolumnandasamaximalcorrelateofheteroclisisinthefracturedparadigmlistedinthefifthcolumn;foreachfracturedparadigm,themaximalcorrelatesdegreeofcorrelationisgiveninparentheses.SYNTACTICINFLECTIONALLANGUAGECATEGORYCATEGORYCLOVENPARADIGMFRACTUREDPARADIGMArmenianVerbtensetake(.92)SochiapanVerbnumberliedown(.83)falldownwalk(.83)AncientNounnumberknee(.87)liver(.87)portent(.87)CzechNounnumberKULIcoolie(.861.0)museum(.79)PolishNounnumberjudge(.79)SanskritNouncaseday(.71)bone(.75)21.Somecorrespondencesbetweenabsolutecorrelatesofheteroclisisinclovenparadigmsandmaximalcorrelatesofheteroclisisinfracturedparadigms.FRACTUREDPARADIGMSINTERSECTIVECORRELATES.Althoughmanyheterocliteparadigmshavemaximal(andevenabsolute)correlates,manyothersdonot.Consider,forexample,theparadigmoftheCzechnounphilologist(Table8).Here, HETEROCLISISANDPARADIGMLINKAGE313thedegreeofnumbercorrelationisthesameasthedegreeofcasecorrelation:bothare.93(13/14);thus,eventhoughthecategoryofnumberisthemaximal(andabsolute)correlateofheteroclisisintheclovenparadigmofCzech,thereisnomaximalcorrelateofheteroclisisinthefracturedparadigmof.Comparableexamplesabound,forexample,inFula(cloven)bellyvs.fracturedTables18and7),Sanskrit()heartvs.water),Latin(cityvs.home),Lithuanian(manvs.wind),Polish(poetvs.pauper),andsoon.Clearly,aninflectionalcategorysstatusastheabsolutecorrelateinaclovenpara-digmdoesnotguaranteethatitwillserveasamaximalcorrelateinfracturedparadigmsbelongingtothesamesyntacticcategory.Evenso,theredoesseemtobeaweakercorrespondencebetweenclovenparadigmsandfracturedparadigmsofthesamesyntac-ticcategory.Thispointcanbemostclearlyelucidatedwithreferencetoanothersortofcorrelationamonginflectionalcategoriesandinflectionclassesinfracturedparadigms.WherePisafracturedparadigm,theinflectionalcategoriesA,...,ASECTIVECORRELATESofPsheteroclisisifandonlyifforeachwell-formedpropertyspecifiedforexactlythecategoriesA,...,AthereisasingleinflectionclassCsuchthateverycellinPrealizinginflectsasamemberofC.TheinflectionalcategoriesA,...,AMINIMALINTERSECTIVECORRELATESofPsheteroclisisifandonlyifthereisnopropersubsetof,...,AwhosemembersareintersectivecorrelatesofPsheteroclisis.Consider,forexample,thefracturedparadigmofAlbanianfall(Table22).Theinflectionalcategoriesoftenseandmoodareintersectivecorrelatesofthisparadigmsheteroclisis,sinceforanywell-formedpropertyset,thereisasingleconjugationclassCsuchthatea

chcellinsparadigmthatrealizesinflectsasamemberofC:eachcellrealizingpresentindicativeXasamemberofthesecondconjugation;eachcellrealizingaoristindicativeXasamemberoftheirregularconjugationtowhichsaybelongs;andsoon.Sinceneithertensenormoodisitselfanabsolutecorrelateofthisparadigmsheteroclisis,tenseandmoodarealsominimalintersectivecorrelates.Thedatabaseofheterocliteparadigmsassembledforthisstudysuggestsanimportantrelationbetweenclovenparadigmsandfracturedparadigmsbelongingtothesamesyntacticcategory.(39)SupposeSisanonemptysetofinflectionalcategoriesservingasabsolutecorrelatesinclovenparadigmsoflexemesbelongingtosyntacticcategoryCinlanguage.Inthatcase,everyfracturedparadigmofcategoryCinhasamemberofSasoneofitsminimalintersectivecorrelates.Thisclaimistriviallytrueoffracturedparadigmsinwhichonlytwoinflectionalcategor-iesaredistinguished;forinstance,giventhatthefracturedparadigmofCzech(Table8)distinguishesonlypropertiesofnumberandcase,theminimalintersectivecorrelatesofheteroclisisinthisparadigminevitablyincludenumber(theabsolutecorrelateofheteroclisisintheclovenparadigmofCzech).Nevertheless,frac-turedparadigmsthatdistinguishthreeormoreinflectionalcategoriesprovidenontrivialevidencefortheclaimin39;considersomeexamples.TenseistheabsolutecorrelateintheclovenparadigmoftheBulgarianverbsleep,whichfollowsthesecondconjugationinthepresentandthefirstconjugationelsewhere;correspondingly,tenseisoneoftheminimalintersectivecorrelatesinthefracturedparadigmoftheverb)eat,whichfollowsthethirdconjugationinthefirst-personsingularpresentandelsewherefollowsthefirstconjugation(Table23). LANGUAGE,VOLUME82,NUMBER2(2006) CONJUGATION:IRREGULARrashëhapa 22.HeterocliteinflectionoftheverbfallinStandardAlbanian(Barlow2002:51f.,70,88,93,Newmarketal.1982:44,46ff.,55f.).thats/heseizes(conjugation2).Similarly,tenseistheabsolutecorrelateofheteroclisisintheparadigmoftheArmeniancomeandoneoftheminimalintersectivecorrelatesofheteroclisisinthefracturedparadigmofArmeniancry(Table24).InHebrew,aspectistheabsolutecorrelateintheclovenparadigmoftheverbapproach,whichfollowstheconjugation(orbinyan)traditionallyreferredtoasthe alintheperfectandthatknownastheQalelsewhere;correspondingly,aspectisoneoftheminimalintersectivecorrelatesinthefracturedparadigmoftheHebrewverblead,whichfollowstheQalintheperfectandtheimperativeandelsewherefollowstheHiphšl(Kautzsch1910:220).Whatoneapparentlydoesnotfindarelanguagesinwhichclovenparadigmsandfracturedparadigmscoexistinsomesyntacticcategoryandinwhichsomefracturedparadigmsinthatca

tegoryhaveminimalintersectivecorrelatesnoneofwhichappearsasanabsolutecorrelateinanyclovenparadigmofthatcategory.Itshould,however,benotedthatagivensyntacticcategorymay,insomelanguage,havefracturedpara-digmsbutnoclovenones;claim39hasnoimplicationsforthefracturedparadigmsbelongingtosuchacategory.AnapparentexampleofthissortisthecategoryofnounsinChukchi(Chukotko-Kamchatkan),wheremembersoftheheteroclitethirddeclensionfollowthefirstdeclensioninthesingularandabsolutivepluralandmayfolloweitherthefirstortheseconddeclensionelsewhereintheplural(Spencer1999).IfthisistheonlypatternofnominalheteroclisisinChukchi,thenneitherofathird-declensionnounsminimalintersectivecorrelates(i.e.neithernumbernorcase)isanabsolutecorrelateofheteroclisisinthislanguage. HETEROCLISISANDPARADIGMLINKAGE315 1ST1ST3RD2ND 2sg2pl3pl1sg2sg2pl3pl1sg2sg1pl2pl Aorist ´P´kradékradéte´tdávadávadávatedávatgasjágasímgasígasítespímspjátku´pjaku´peku´pemku´peku´peteku´pjatku´pexku´peeku´pexmeku´peeku´pexteku´pexaku´paxku´paku´paxmeku´paku´paxaku´paxtejadéjadéte´tkradéeekradjáxtekradjáxaeejadjáxtejadjáxadávaxdávaeedávaxmedávaxtegasjáxegaséegasjáxmegasjáxtekrádekrádoxmekrádoxtekrádoxajádejádoxmejádoxtejádoxaeespjáxmespjáxtespjáxagasíxgasígasíxmegasíxtegasíxaspáxspáxtespáxadávadávaxte 23.Heterocliteinflectionoftheverbs)eatandsleepinBulgarian(Scatton1984:211f.,217,223,226f.,238,240). ST CONJUGATIONof secondary verbs3RD CONJUGATIONof primary verbs2ND CONJUGATIONof primary verbs 2sg3sg1pl2pl3pl1sg2sg3sg1pl2pl3pl1sg2sg3sg1pl2pl3pl Aorist ku gayiku ku ku gar gayik ku ayiku ku ku lac av  kardayik kardayirk kardark kardayin k tesnemk tesnesk tesnenkk tesne  tesne k tesnenk tesne k tesne ir  tesne r  tesne k tesne i k tesne i tesaytesartesavtesantesanktesakyekayyekaryekavyekanyekankyekakku gamku ku gay ku ku gak amku ku ku ku ku k xo k xo k xo k xo k xo k xo k kardayinkku ku k kardayikk kardac kardac k ik l c lac lac l ik xo  xo  xo  xo  xo ¯sec xo  k xo k xo k xo k xo k xo k xo 24.HeterocliteinflectionoftheArmenianverbscomeandcry(indicativeforms)(Kogian1949:82,84ff.,96ff.,118,121).FORMALINTERPRETATIONOFOBSERVATIONS36…39.Itisapparentfromtheobser-vationsin§§5.2…4thatthereisanasymmetryamongalanguagesinflectionalcategor-ies.Inparticular,36…39suggestthattheinflectionalcategoriesservingasabsolutecorrelatesofheteroclisisinagivenlanguageare,andthattheinflection-classboundariesinthatlanguagesheterocliteparadigmsareinsomewaysensitivetothepr

ivilegedstatusofthesecategories.Itisclearthatwhetheragiveninflectionalcategoryisprivilegedvariesfromlanguagetolanguage;thatis,theredoesnotseem LANGUAGE,VOLUME82,NUMBER2(2006)tobeanyinflectionalcategorythatisuniversallyprivileged.Moreover,mostlanguagesseemtorequirethatamongtheinflectionalcategoriesrelevantfortheinflectionofaparticularsyntacticcategory,therebeatmostonethatisprivileged(36);inCzech,forexample,numberistheonlyprivilegedinflectionalcategoryamongthecategoriesrelevantfornominalinflection.(InTakelma,however,tenseandpersonarebothprivi-legedamongthecategoriesrelevantforverbinflection.)Inanyevent,themajorityofalanguagesinflectionalcategoriesareunprivileged(37);indeed,theinflectionalcategoriesforwhichthememberlexemesofaparticularsyntacticcategoryinflectmayormaynotincludeanythatareprivileged.(IfthememberlexemesofasyntacticcategoryCdoinflectforaprivilegedinflectionalcategory,IhenceforthsaythatCtooisprivileged.)Ifaheteroclitelexemebelongstoaprivilegedsyntacticcategory,thenaprivilegedinflectionalcategoryparticipatesinconditioningthejuxtapositionofinflec-tionclassesinitsparadigm(38,39).Intheparadigm-linkageapproach,thenotionofaparadigmssensitivitytoaninflec-tionalcategorysprivilegedstatuscanbegivenformalsubstancebymeansofthefollowingwell-formednessconditiononalanguagesrulesofparadigmlinkage.(40)PRIVILEGEDCATEGORYRESTRICTION(PCR):Ifaruleofparadigmlinkageap-pliestolexemesbelongingtoaprivilegedsyntacticcategoryCandthisruleissensitivetothevalueofanyinflectionalcategory,thenitissensitivetothevalueofaprivilegedinflectionalcategoryformembersofC.ConsidertheimplicationsofthePCRforrulesofparadigmlinkageofvarioustypes.ThePCRhasnoimplicationsforthedefaultruleofparadigmlinkagein5,sinceitsapplicationisinsensitivetothevalueofanyinflectionalcategory.Thesameistrueoflanguage-specificrulesofparadigmlinkagesuchasthosein12and13.Aswehaveseen,somerulesofparadigmlinkagearesensitivetothevalueofasingleinflectionalcategory;examplesaretheCzechruleofparadigmlinkagein14(whichissensitivetonumber),theOldEnglishrulein21a(sensitivetotense),theSanskritrulesin26and29(sensitivetocaseandtense,respectively),andtheGreekrulesin35(sensitivetogender).Allsuchrulesengenderclovenparadigmsinwhichtheconditioninginflectionalcategoryisanabsolutecorrelateofheteroclisis(thatis,aprivilegedcategory);rulesofthissortthereforenecessarilyconformtothePCR.IfaruleofparadigmlinkageappliestolexemesbelongingtoaprivilegedsyntacticcategoryCandthatruleissensitivetothevalueofoneormoreunprivilegedinflectionalcategories,thenthePCRentailstha

titisalsosensitivetothevalueofaprivilegedinflectionalcategoryformembersofC.Consideraninstanceofthesatisfactionofthisrequirement.InCzech,theheterocliteinflectionof,anddependsontherulesofparadigmlinkagein14,15,and17;allthreerulesapplytoatoassignacoradicalform-correspondent.Theapplicationofrule14requiresthatpluralX;aswehaveseen,thisidentifiesnumberasaprivilegedinflectionalcategory(andnounsasaprivilegedsyntacticcategory)inCzech.Theapplicationofrule15requiresthatdativesingularXlocativesingularXandthatofrule17requiresthatlocativepluralX:althoughtheserulesaresensitivetothevalueofanunprivilegedinflectionalcategory(i.e.ofcase),theyarealsosensitivetothevalueoftheprivilegedinflectionalcategoryfornouns(i.e.ofnumber)andthereforesatisfythePCR.ThedatabaseofheterocliteparadigmsassembledforthisstudyexhibitsuniformcompliancewiththePCR.SomeoftherelevantevidenceispresentedinTable25.Foreachlanguageinthistable(column1),Ilistheteroclitelexemes(column4)belonging HETEROCLISISANDPARADIGMLINKAGE317tothesamesyntacticcategory(column2);foreachlanguage,atleastoneofthelistedlexemeshasaclovenparadigm(column3).ForeachlexemeIlisttheinflection-classmembershipofbothitsrootanditscoradical(column5).Column6specifiesthemorphosyntacticpropertysetthatcausesthelexemetobeassignedacoradicalform-correspondent.Theprivilegedinflectionalcategoryrelevantfortheheteroclitescitedfromeachlanguageisunderlinedincolumn6.Ineveryoneoftheselanguages,aninflectionalcategoryservingasanabsolutecorrelateofheteroclisisinclovenparadigmsalsoinvariablyconditionstheassignmentofacoradicalform-correspondentinfracturedparadigms.ThatthisshouldalwaysbesoistheessentialcontentofthePCR. SYNTACTICCATEGORYPARADIGMCLOVEN ORFRACTURED?HETEROCLITE LEXEMEROOT/CORADICALPROPERTY SETS WHOSE REALIZATION(privilegedinflectional category underlined)LANGUAGE Noun{ cloven HÊPAR liver SanskritNominalVerbclovenVerbclovenVerbclovenclovenclovenclovencloven3rd conj. of 1ary verbs/ 1st conj. of 2ary verbs3rd conj. of 1ary verbs/ 2nd conj. of 1ary verbsGrade A stem/Grade C stemGrade A stem/Grade D stem (i)u decl./  (i)a decl./(i)u decl.an-stem/general C-steman-stem/s-stemi-stem/in-stem{ { AGR { { AGR { EVAL { EVAL { EVAL { GAL come (Table 24) LAL cry (Table 24) MOGÙS JAS sleep (Table 23) JA(D) eat (Table 23) belly (Table 18) HEN wind (Table 7) YAKAN liver (Table 14) AHAN day (Table 10) RI water 25.Conformityofsomeheteroclitelexemestotheprivilegedcategoryrestriction(Jannaris1968,Kogian1949,Scatton1984,Arnott1970,Ambrazas1997,Whitney1889,Fori

s2000).ThePCRisnot,ofcourse,alogicalnecessity.Onecouldimaginealanguagehavingjustthetworulesofparadigmlinkagein41.By41a,numberwouldbeanabsolutecorrelateofheteroclisisintheclovenparadigmsoftype-Iverballexemes,butby41b,tenseandperson(butnotnumber)wouldbeminimalintersectivecorrelatesofheter-oclisisinthefracturedparadigmsoftype-IIverballexemes;Table26illustrates.Thoughthissortofsystemislogicallyconceivable,thePCRentailsthatitshouldnotactuallyexist:eventhoughnumberwouldbeaprivilegedinflectionalcategoryforverbinflec-tion,theapplicationoftheruleassigningcoradicalform-correspondentstoatype-IIverballexemewouldbeinsensitivetonumber.(41)Rulesofparadigmlinkageinahypotheticallanguage:a.IfLisatype-IverballexemehavingsasitsclassBcoradicalandpluralX,thenasitsform-correspondent.b.IfLisatype-IIverballexemehavingsasitsclassBcoradicalandpast3rdX,thenasitsform-correspondent.ThePCRhasaclearfunctionalmotivationforthelanguagelearner.WheneveralexemebelongingtosyntacticcategoryCisfoundtohaveaclovenparadigm,the LANGUAGE,VOLUME82,NUMBER2(2006) TYPE-I LEXEMETYPE-II LEXEMEPASTPASTPRESENTPRESENT 26.HypotheticalcooccurrenceofclovenandfracturedparadigmsprohibitedbythePCR.absolutecorrelateAofitsheteroclisisis,bydefinition,aprivilegedinflectionalcategoryformembersofC;bythePCR,anyruleofparadigmlinkageassigningcoradicalform-correspondentstoamemberofCmustbesensitivetocategoryA.Inthisway,thePCRsubstantiallynarrowsthesetofrulesbywhichcoradicalform-correspondentsmightpotentiallybeassigned:acandidateruleisimmediatelyexcludedifitisnotconditionedbyaninflectionalcategorythatisprivileged.ThePCRisnot,ofcourse,theonlymanifestationofasymmetryamongalanguagesinflectionalcategories;accordingly,thePCRmightultimatelybederivableasatheoremofsomemoregeneralpatternofhierarchizationamonginflectionalcategoriesinnaturallanguage;thispossibilitymustawaitfurtherresearch.IMPLESTEMALTERNATIONANDPRIVILEGEDINFLECTIONALCATEGORIES.Hetero-clisisisonesortofstemalternation,butnotallparadigmsexhibitingstemalternationareheteroclite(§2).Evenso,thePCRshouldperhapsbeseenaspertainingbothtoheteroclisisandtononheterocliticstemalternations.Asitis,thePCRmakesasubstantivepredictionabouttheincidenceofsimple(nonheteroclitic)stemalternationininflectionalpara-digms.Ininstancesofsimplestemalternation,alexemesparadigmisbuiltonarootandoneormorecoradicalsdifferingfromtherootinformbutnotininflection-classmember-ship;examplesaretheparadigmoftheSanskritadjectivewesterly(inwhichthecoradicalsbelongtothesamedeclensionastherootTable9)andthatofRussianmother(inwhichth

estemstothesamedeclension;Table5).ThePCRentailsthatininstancesofsimplestemalterna-tion,asininstancesofheteroclisis,ifaruleofparadigmlinkageappliestoalexemebe-longingtoaprivilegedsyntacticcategoryandthatruleissensitivetothevalueofsomeinflectionalcategory,thenitissensitivetothevalueofaprivilegedinflectionalcategory.Thisentailmentappearstobeempiricallyconfirmed.Forinstance,intheinflectionof,therules20a,bassigningform-correspondentscontainingtheMiddleandWeakestcoradicalsaresensitivetocase,aprivilegedinflectionalcategoryforSanskritnouns;intheinflectionofRussian,theruleassigningform-correspondentscontainingthecoradicalmustbesensitivetonumber,aprivilegedcategoryforRussiannouns;andsoon.OtherapparentsimilaritiesbetweenheteroclisisandsimplestemalternationsuggestthatthePCRshouldbeinterpretedmorebroadly.In§5.5,acategoryisdefinedasoneservingasanabsolutecorrelateofheteroclisis.IfthenotionWitness,forexample,thefactthatkittenfollowsdeclensionIinthesingularanddeclensionIVintheplural.(Ihereassumethedeclension-classlabelsofCorbett&Fraser1993.) HETEROCLISISANDPARADIGMLINKAGE319ofprivilegeisdefinedinthisway,thennoinstanceofsimplestemalternationsufficestoestablishaninflectionalcategoryasaprivilegedone.Consider,however,thealterna-tiveofdefiningaprivilegedcategorymorebroadly,asoneservingasanabsolutecorrelateofeitherheteroclisisorsimplestemalternation(whereaninflectionalcategoryCisanabsolutecorrelateofsimplestemalternationinsomeparadigmPifandonlyifforeachvaluevofC,thereisasinglestemssuchthateveryrealizationofvinPisbasedons).Underthisreviseddefinition,aninstanceofsimplestemalternationmaysufficetoestablishaninflectionalcategoryasprivileged.Ifthedefinitionofprivilegeisbroadenedinthisway,thenthePCRmakesricherpredictions.Consider,forinstance,theincidenceofsimplestemalternationintheinflectionalparadigmoftheBretonprepositionof(Table27):thesteminthethirdperson,andthestemelsewhere.Becausepersonisanabsolutecorrelateofsimplestemalternationinthisparadigm,thispatternofstemalternationestablishespersonasaprivilegedinflectionalcategory(andprepositionasaprivilegedsyntacticcategory)inBreton.BythePCR,thisentailsthateveryruleofparadigmlinkageassigningacoradicalform-correspondenttoaprepositionallexememustbesensitivetoperson;theobservablepatternsofstemalternationintheinflectionofBretonprepositionssatisfythisprediction,asthepatternsinTable28show.Thisisnotapredictionmadeundertheearlier,narrowdefinitionofprivilege,sincetheinflectionofBretonprepositionspresentsnoinstancesofoutrightheteroclisis(eventhoughtheinflectingpre

positionsdofollowmorethanoneconjugation;Trepos1968:149).accordingtoSingular1sthervezonachanon2ndhervezoutachanout3rdmasculinehervezanfemininehervezianeziPlural1sthervezomachanom2ndhervezochachanoch3rdhervezoanezo27.SimplesuppletionintheinflectionofBretonPREPOSITIONNONTHIRD-PERSONSTEMTHIRD-PERSONSTEMtodi(sg),dide(1pl),de(2pl)dezaseveldeveltforevidevitwithgan(sg),gane(pl)gantwithouthebdheptexceptnemednemetagainstouzout28.StemsuppletioninBretonprepositionalparadigms.FurthercomparisonofheteroclisisandsimplestemalternationwillbenecessarytoconfirmthedesirabilityofbroadeningtheinterpretationofthePCRinthisway.UMMARYANDCONCLUSIONS.Inlightofthephenomenaexaminedinthisarticle,itisclearthatthelexicalapproachtoheteroclisisin1isuntenablefortworeasons.First,itisclearthatinstancesofheteroclisissometimesinvolveverysystematicassociationsbetweeninflectionclassesandsetsofmorphosyntacticproperties.InaForrelevantdiscussionofthegeneralphenomenonofsuppletion,seeCorbett2006,Hippisleyetal.2004,Maiden2004,Markey1985,andMelcuk1994. LANGUAGE,VOLUME82,NUMBER2(2006)theoryinwhichheteroclisiswereattributedpurelytopiecemealstipulationsinspecificlexicalentries,realgeneralizationsabouttheseassociationswouldnotbeexpressible:theseincludegeneralizationsaboutsteminferenceandinflection-classjuxtapositioninprincipal-partssystems(e.g.29a,b),insystemsassociatingparticularinflectionclasseswithparticularsyntacticcontexts(e.g.30,31),insystemsinwhichadefectiveinflectionclassisdependentonanondefectiveone(e.g.32,33),andinsystemsinwhichinflectionclassesarerecruitedfortheexpressionofspecificmorphosyntacticdistinctions(e.g.34,35).Second,heteroclisisuniversallyobeystheconstraintembodiedbytheprivilegedcategoryrestriction(40),awell-formednessconditiononrulesofparadigmlinkage.Inatheoryinwhichheteroclisiswereattributedpurelytopiecemealstipulationsintheentriesofindividuallexicalitems,conformitytothisconstraintwouldremainunac-countedfor.Thetheoryofparadigmlinkagedevelopedin§3satisfactorilyaccountsforthevariouskindsofphenomenathatexcludehypothesis1.Itallowsgeneralizationsaboutsystem-aticinstancesofheteroclisistobedirectlyexpressedbymeansofrulesofsteminferenceandparadigmlinkage,anditallowsauniversalconstraintonheterocliterealizationofmorphosyntacticpropertysetstoberepresentedasaconstraintonrulesofthelatterThefactspresentedherehavedecisiveimplicationsregardingthestatusofinflectionalparadigmsinmorphologicaltheory.Inrecentyears,anumberoftheoreticalframeworkshavebeenproposedthatportrayparadigmsasmerelyepiphenomenal

„ashavingnoessentialroleinthedefinitionofalanguagesinflectionalmorphology;thesearegener-allyframeworksinwhichinflectionalmarkingsareassumedtohavethetheoreticalstatusoflexicalentries.Thefactspresentedherecastseriousdoubtontheviabilityofsuchframeworks,extendingandstrengtheningthealreadysubstantialbodyofevidencethatparadigmsareindispensabletothedefinitionofinflectionalsystems(Stump2001,,andG.2004.Paradigmsandperiphrasticexpression:Astudyinrealization-basedlexicalism.Projectingmorphology,ed.byLouisaSadlerandAndrewSpencer,111…57.Stanford,CA:CSLIPublications.;andG.2006.Arealizationaltheoryofperiphrasticpredicates.Non-transformationaltheoriesofgrammar,ed.byBobBors-leyandKirstiBorjars.Oxford:Blackwell,toappear.(ed.)1997.Lithuaniangrammar.Vilnius:BaltosLankos.,D.W.1970.ThenominalandverbalsystemsofFula.Oxford:OxfordUniversity.2004.Directionalityand(un)naturalclassesinsyncretism.80.807…27.;andGG.C.2005.Thesyntax-morphologyinterface:Astudyofsyncretism.Cambridge:CambridgeUniversityPress.J.2002.Albaniangrammar:Albanianverbsexplained.Aldington,Kent:BayForeignLanguageBooks.,A.C.1968.AgrammaroftheFurlanguage.Khartoum:SudanResearchUnit,UniversityofKhartoum..1939.AgrammarofLakota:ThelanguageoftheTetonSiouxIndiansSt.Louis:JohnS.Swift.Swift.AWORTH],IM.1980.ticazapoteca:ZapotecodeYatzachielBajoxico,D.F.:InstitutoLingusticodeVerano.,andA.2000.Stemalternantsas HETEROCLISISANDPARADIGMLINKAGE321morphologicalsignata:EvidencefrombluravoidanceinPolishnouns.NaturalLan-guageandLinguisticTheory18.813…35..1987.Allomorphyininflexion.London:CroomHelm.G.1991..Cambridge:CambridgeUniversityPress.G.2003.Thecanonicalapproachintypology.Linguisticdiversityandlanguagetheories,ed.byZygmuntFrajzyngier,AdamHodges,andDavidRood,25…49.Amsterdam:JohnBenjamins.G.2006.Suppletion:Typology,markedness,complexity.Oninflection:InmemoryofWolfgangU.Wurzel,ed.byAndreasBittner,FransPlank,andPatrickger.Berlin:MoutondeGruyter,toappear.G.,andNM.F.1993.Networkmorphology:ADATRaccountofRussiannominalinflection.JournalofLinguistics29.113…42..1980.Ngiyambaa:ThelanguageoftheWangaaybuwan.Cambridge:CambridgeUniversityPress..2000.AgrammarofSochiapanChinantec.Arlington,TX:SILInterna-tionalandTheUniversityofTexasatArlington.M.,andGG.C.1995.Gender,animacy,anddeclensionalclassassignment:AunifiedaccountforRussian.YearbookofMorphology1994.123…50.,andJR.K.1955.Anintroductiontoclassical(literary).Wiesbaden:Harrassowitz..1982.ContemporaryCzech.Columbus,OH:Slavica.G.C;andD2004.Suppletion:Frequency,categoriesanddistributionofstems.StudiesinLanguage28.2.38

9…421.N.1968.AnhistoricalGreekgrammarchieflyoftheAtticdialectHildesheim:GeorgOlms.L.1999.Ontheriseofsuppletioninverbalparadigms.BerkeleyLinguistics25.183…94..1998.Mari.TheUraliclanguages,ed.byDanielAbondolo,219…48.London:Routledge.,E.(ed.)1910.GeseniusHebrewgrammar.2ndEnglishedn.Oxford:OxfordUniversityPress.,A.E.1999.lementytsaxurskogojazykavtipologiceskomocves.Moscow:,S.L.1949.Armeniangrammar(westdialect).Vienna:MechitharistPress.,L.V.;J.F;andO.L.1976..(2vols.)Prague:.1916.AVedicgrammarforstudents.Delhi:OxfordUniver-sityPress.[1977reprint.].2004.Whenlexemesbecomeallomorphs:Onthegenesisofsuppletion.Paperpresentedatthe11thInternationalMorphologyMeeting,Vienna.L.1985.Onsuppletion.2.1.51…66..1994.Suppletion:Towardalogicalanalysisoftheconcept.Studiesin18.2.339…410..2000.TheHausalanguage:Anencyclopedicreferencegrammar.NewHaven:YaleUniversityPress.;andP.1982.StandardAlbanian:Arefer-encegrammarforstudents.Stanford,CA:StanfordUniversityPress.,andA.2001.SyntaxasanexponentofmorphologicalYearbookofMorphology2000.71…96..1922.TheTakelmalanguageofsouthwesternOregon.HandbookofAmeri-canIndianlanguages,part2,ed.byFranzBoas,1…296.Washington:GovernmentPrintingOffice.A.1984.AreferencegrammarofmodernBulgarian.Columbus,OH:.1972.Naukaocmjazyku.Prague:Sta LANGUAGE,VOLUME82,NUMBER2(2006).1999.Chukchiwebsite.http://privatewww.essex.ac.uk/,andGT.S.2006.Paradigmfunctionmorphologyandthemorphology/syntaxinterface.TheOxfordhandbookoflinguisticinterfaces,ed.byGillianRamchandandCharlesReiss.Oxford:OxfordUniversityPress,toappear.T.2001.Inflectionalmorphology.Cambridge:CambridgeUniversityT.2002.Morphologicalandsyntacticparadigms:Argumentsforatheoryofparadigmlinkage.YearbookofMorphology2001.147…80.T.2005a.Delineatingtheboundarybetweeninflection-classmarkingandderivationalmarking:ThecaseofSanskrit-aya.Morphologyanditsdemarcationsed.byWolfgangU.Dressler,DieterKastovsky,OskarE.Pfeiffer,andFranzRainer,293…309.Amsterdam:JohnBenjamins.T.2005b.Rulesaboutparadigms.Morphologyandthewebofgrammar:EssaysinmemoryofStevenG.Lapointe,ed.byC.OrhanOrgunandPeterSells,49…82.Stanford,CA:CSLIPublications.T.2005c.SomecriticismsofCarstairs-McCarthysconclusions.ofMorphology2005.283…303..1968.Grammairebretonne.Rennes:OuestFrance.[1980reprint.].1889.Sanskritgrammar.2ndedn.Cambridge,MA:HarvardUniversityPress.M.1985.Howtodescribeinflection.BerkeleyLinguisticsSociety11.372…86.DepartmentofEnglish[Received26February2003;UniversityofKentuckyrevisioninvited6October2003;1215PattersonOfficeTowerrevisionreceived22February2005;Lexington,KY40506-