Clitics and ClauseStructure:The Late Medieval Greek System

Cleo CondoravdiPARC andStanfordUniversityPaul KiparskyStanfordUniversity

We rebutPappas’critique (this issue)of our treatmentof Late MedievalGreekclausalsyntaxand clitic placement(Condoravdi& Kiparsky 2001), point outsomeweaknessesof his counterproposal,andsuggestdirectionsfor further re-search.

Keywords: clitics, weakpronouns,functionalprojections,prosodicin-version,topic, focus,Greekdialects,MedievalGreek,historicalsyntax,syntacticchange,Cappadocian.

1 Clitic placementin Late Medieval Greek

In Condoravdi& Kiparsky2001wetracedtwo pathsof grammaticalizationin theevo-lution of Greekclausestructure:thereductionof Wackernagel(Xmax) clitic pronounsto word-levelaffixesviaanintermediatestageof word-level(X0) clitics, andtheemer-genceof auxiliariesandmoodmarkersasheadsof newfunctionalprojections.Thetwo trajectoriesareconnectedin thatthepositioningof Xmax clitic pronounsrespondsto changesin theclausalsyntax,therebyprovidinga diagnosticfor theemergenceofthearticulatedInfl projectionpreviouslypositedfor standardModernGreek(Tsimpli1995,Tsimpli & Roussou1996,Giannakidou1998,amongothers),andconfirmedbyourfindingsfor othermoderndialectsandfor LateMedievalGreek.1

Both theseGreekdevelopmentshavebroadparallelsin Indo-Europeanandbe-yond, which givesthemmorethanjust parochialinterest. In the contextof the re-centcontroversyover the actuationof syntacticchange,we arguedthatour findingssupportthe view that syntacticchangeproceedsin small but discreteincrementsincertainstructurally/functionallymotivateddirections,overtheopposingpositionthatit involves“catastrophic”reanalysiswith no intrinsicdirectionality.

Our analysisaddressed,in additionto datafrom contemporarydialects,theLateMedievalGreekcliticizationpatterndescribedin Mackridge1993.Weproposedthatitis substantiallythesameasthatof Cappadocianandtheotherdialectsthatwegroupedinto ourTypeA.

Pappas(2001) revisedMackridge’s descriptivegeneralizations,and proposedasynchronicandhistoricalinterpretationof them. In this issuehesummarizeshis re-sults,andarguesthathis analysisof LMG is superiorto ours. After consideringhisargumentswehavecometo theoppositeconclusion.Sincesomeof ourdisagreementsreflectdifferentviewsaboutthenatureof grammarandsyntacticchange,ourresponsewill haveto touchonsomemattersof principleaswell.

Thedescriptivegeneralizationsfor LateMedievalGreek(LMG) canbesumma-rizedasfollows.

(1) LMG clitic placement:thedescriptivegeneralizations

a. A clitic pronounfollowstheverbif theverbis

1. clause-initial,or immediatelyprecededby

2. a coordinatingconjunction,

3. a frontedobjectwith a resumptiveclitic pronoun,

4. theconjunctionsoti “that”, dioti “because”,i “if ”, or

5. thenegationu(k).

b. A clitic pronounprecedestheverbif theverbis immediatelyprecededby

1. a moodmarker,

2. a negationotherthanu(k) (mi, den),

3. a frontedobjectwithouta resumptiveclitic pronoun,

4. a frontedadverbial(otherthana temporaladverb),

5. a complementizer(pos,an), or

6. a Wh-expression(relativepronoun,clausalconnective,comparative).

c. A clitic eitherfollowsor precedestheverbif theverbis immediatelypre-cededby

1. a temporaladverb,or

2. a subject.

This schemeis simplerthanMackridge’s original onethanksto Pappas’findingthatMackridge’sfive-wayclassificationof contextscanbereducedto thethreeshownin (1). In particular,cases(c1) (temporaladverbs)and(c2) (subjects)do not haveasignificantlydifferenteffecton theplacementbeforeor after theverb. So amended,thegeneralizationsfit our theoryevenbetter. Hereis thephrasestructureweproposedfor bothLateMedievalandModernGreek(repeatedherebecauseit wasdisfiguredattheprintingstagein C&K 2001).

(2) CP

C′

ΣP

Σ′

TNSP

TNSP

TNS′

Spec C Spec Σ0 Cl Spec TNS0 VP{Wh-RelWh-Qu

} {FocXPEmphNeg

} {NegModPrt

}Vj . . . tj . . .

The first functionalprojectionabovethe VP is TNSP, which is headedby the finiteverb. This is whereclitics areadjoined.TNSP is thecomplementof ΣP, whoseheadcanbeoccupiedby a negationor a modalparticle(na, �a, as). Thespecifierof ΣPis thesiteof focusedconstituents,includingthosewith emphaticnegationsuchaska-nenas. We alsosuggestedthat in somedialects,probablyincludingLMG, [Spec,ΣP]canhostevenordinarynonfocusedsubjects. ΣP in turn is the complementof CP,which canbe headedby a complementizer(pos,pu etc.); its specifieris the site ofWh-phrasesandrelatives.Uncontroversially, topicalization(whichnormallyleavesaresumptiveclitic in caseof anobject)is adjunctionto theclauseat thehighestlevel(ΣPor CP).

On this understandingof the phrasestructure,the generalizationsaboutcliticplacementstatedin (1) canbestreamlined.In part(a)of (1), thethreecases(a1),(a2),and(a3)canbecombined.Theyall comedownto theclause-initialcase(a1),becausetopicalizedobjectsaswell asco-ordinatingconjunctionssit outsidethe clausesthatthey conjoin (this providesthe theoreticalmotivation for the “factor initial” whoseunity Pappasestablishesstatistically). Cases(a4)and(a5) requireanextraassump-tion, but not necessarilyabouttheclitic themselves;we suggestedthatuk, oti, dioti,and i areplacedin [Spec,CP]andthat the immediatelyfollowing finite verb is in C,whereit hoststheclitic.2 Giventhisadditionalassumption,all five casesof postverbalclitic placementin (a) fall underthesamegeneralization.

As for part(b) of (1), all casesfall underthegeneralizationthatclitics precedetheverbif thereis somethingbeforethemwithin theclause.A look at (2) showsthatthiscanbeoneof thefollowing elements(from right to left):

• a moodmarkeror negationin Σ0 (cases(b1)and(b2)),

• a focusedexpressionor emphaticnegativein [Spec,ΣP] (cases(b3)and(b4))

• a complementizerin C (case(b5)),or

• a Wh-phrasein [Spec,CP](case(b6)).

Finally, the apparentoption in (1c) is just a consequenceof the ambiguoussta-tus of preposedsubjectsandadverbialsin written texts. Unlike objects,they leaveno resumptivepronounsbehindwhen topicalized. So, in the absenceof sufficientintonationalor contextualcues,a preverbalsubjector adverbialcould be parsedaseithera topic or asa focusedelement.Hencethedescriptivegeneralizationthatafterapreverbalsubjector adverbial,a clitic maycomeeitheraftertheverbor beforeit.

Theupshotis that(2) allowsusto distill (1) to (3):

(3) LMG clitic placement:reviseddescriptivegeneralizations

a. A clitic followstheverbif theverbis initial in its coreclause.

b. Otherwiseit precedestheverb.

(3) is moreperspicuousthan(1), and,moreimportantly, it bringsLMG into line withthecross-linguistictypologyof cliticization workedout by Halpern1995andothers.Evenmore importantly, it furnishesthe basisfor a principledformulationof LMGclitic placement. Clitics are adjoinedto the functionalprojectionheadedby theirgoverningverb,which in LMG is TNSP. So thebasicclitic positionis preverbal,asshownin (2). Clitics areput after the verb asper (3a) only whenthereis nothingon their left for them to encliticize to. In minimal satisfactionof their needfor aprosodichoston their left, they cliticize to the verbby PROSODICINVERSION.3 Ifclitic adjunctionandprosodicinversionaregeneralmechanismsashypothesized,thentheLMG-specificresidueis thatclitics in this languageleanleftwards:

(4) LMG clitic placement:final formulation

• Clitics areenclitic.

The restwould be a matterof grammaticalprinciplesthat LMG shareswith all lan-guages.

In replacing(1) by (3) we went from a complexformulationframedin termsofcategoriesto a simpleoneframedin termsof structuralpositions. Both areequallytheory-laden;they just rely on different theoreticalconstructsandassumptions.Thereformulationpaidoff with thefurthersimplificationin (4). Thisfinal formulationhasrealexplanatorycontent,andmoreover,unlike thepreviousonesit coversa newsetof casesthathavebeenignoredin theliterature(section2.2below).

2 The analysisof clitics: syntactic issues

2.1 Pappas’alternative

Accordingto Pappas,theLMG distributionof clitics “cannotbecapturedby a singlegeneralizationor evena setof generalizations”.However,“the patterncanbeviewedastheproductof anon-goinganalogicalchange”.Beforegoinginto thedetails,let ustry to understandthesegeneralclaims,andthereasoningthatleadsPappasto them.

The statementthat LMG syntax“can be viewedas the productof an on-goinganalogicalchange”is in onesensetrivially true. It is uncontroversialthat languagesarealwayschanging,andthatsyntacticchangeis fundamentallyanalogical.What iscontroversialis whetherthoseanalogicalmechanismsarebestmodeledasreanalysisplusextension,asgrammaticalization,asoptimization,or assomethingelse.So,thesyntaxof every languagemustbe the productof on-goinganalogicalchange. ButPappasseemsto havea moredrasticclaim in mind: that LMG syntaxis somehowspecialin thata sequenceof analogicalchangeshasunderminedits clitic system,andleft it with a patternthat cannotbe capturedby any syntacticgeneralizations.Toocomplexto bestable,thispatternis soonregularizedagainby analogicalchange.Theparadoxis that analogicalchangeis hereportrayedboth as a sourceof chaosanddisorderandasa regularizingforce.4

Weconcurwith Pappasthatchaoticpartsof grammarareunlikely to bediachroni-cally stable.It is significant,therefore,thattheLMG clitic patternis thatof theTypeAdialects,wherehasremainedstablein manyseparatelocalities.FromMarioupoliandAzov in theUkraineto theancientGreekcommunitiesscatteredthroughCappadociaandBithynia, the Cycladic islands,the Dodekaneseislandsof Karpathos,Kos, andAstipalaia,andthe two communitiesof AjassosandPlomarion Lesbos,theTypeAsystemremainsin all essentialsthesame.5 Yet mostof thesecommunitieshavehadnosignificantcontactfor well overa thousandyears.In medievalGreek,accordingtoMackridge(1993),thepatternwasstablethroughnearly600years.6 Thisunderminesthenotionthat thereis no generalizationor setof generalizationsbehindtheTypeAclitic system.Logically, therearetwo possibilities.Either thedialectshadthesameidiosyncraciesfrom thestart,or theyarrivedat themindependently. If theyinheritedtheTypeA systemfrom atimebeforetheysplit,why hasit remainedsostablein themfor solong?Shouldn’tmostof theputativeanomalieshavebeenironedout in mostofthedialects?7 And if theTypeA featuresdevelopedindependentlyafter thedialectshad dispersed,why did all the dialectsconvergeon the samesupposedlyarbitrarydistributionof clitics? If, aswe propose,the clitic systemis firmly anchoredin thestructuralgivensof thelanguage,thenstability andconvergenceareto beexpected.

A secondmajor issueconcernsthe treatmentof variation. As far asclitic place-ment is concerned,LMG is basicallya Type A system,in the classificationof ourarticle,but it is notanunalloyedvarietysuchasweseein mostof themodernTypeAdialects.Many LMG textshaveanadmixtureof constructionsrepresentingthe later

systemof X0 clitics andlexicalprefixes.In particular,preverbalclitics in theenviron-mentslistedin (1a)(initial position,aftera coordinatingconjunction,etc.) cannotbederivedby theTypeA grammar. It is, however,theregularpatternof TypeB dialectswith proclitics,suchasthedialectof Kozani,andTypeC dialects,includingstandardModernGreek.Crucially, theLMG variationis not a randommixturebut anorderlycoexistenceof two independentlyattestedpatterns.Thevariationin clitic placementdocumentedby MackridgeandPappasfor LMG canbedescribedastheunionof en-clitic Type A syntaxandType B/C syntaxof the proclitic variety, with the formerstronglypredominating,of course.Everyoptionof clitic placementin LMG thatdoesnotfit theTypeA patternconformsto themodernproclitic patterns,andnearlyall thefeaturesof thelatteroccurat leastoccasionallyin theLMG texts.

Thereareseveralwaysto modelthis generalization,includinggrammarcompeti-tion (Kroch 2000),unrankedconstraints(Anttila 1997),andstochasticOT (BoersmaandHayes2001).Theyall unify variableandcategoricalpatternsin aprincipledway,allowing themto relatethevariationpatternin theLMG textsto thecategoricalpat-ternsof themoderndialects.8 Pappas,in spiteof his referenceto Kroch 1982/1989,insteadresortsto theolder“variablerule” technique,whichcandetectcovariationbutis too unconstrainedto modelthe orderly relationshipbetweencategoricalandvari-ablepatterns.His choiceof thisdescriptiveframeworkis relatedto hisview thateachenvironmentis in principlea law untoits own.

2.2 Preverbalclitics

Pappas’analogicalderivationof the LMG patternstartsfrom a stagein Early Me-dievalGreekwhenclitics werepostverbal.Whenthecategoryof theMoodPhrase(=our ΣP) arose,clitics beganto be placedafter its head,the new subjunctivemarkerna. Thisgaveriseto therulethatclitics arepreverbalafterna. Thatnewrulewasthenanalogicallygeneralizedfrom na to otherelementswhich typically appearimmedi-atelybeforetheverb,providedtheyhad“clause-typingor subordinatingproperties”.Suchelementswerenegativemarkers,complementizers,andWh-expressions,butnotcoordinatingconjunctionsandoti. Frontedconstituents(focusedNPsandPPs,andVPadverbials)becameassociatedwith preverbalclitic placementby a furtheranalogicalchange,basedon thegeneralizationthatpronounsarepreverbalafteranelementthatis “salient in the interpretationof theVP”. Pappassuggeststhat thereasonpreverbalplacementdoesnot extendto thecontextafter topicalizedobjects(preposedobjectswith resumptiveclitics) is “becausetheyarenotpartof theverb’sargumentstructure”.In thatcase,presumably, theclitics alonearetheverb’sarguments.

Pappasdoesnot saywhypreciselythis setof elements“typically appearimmedi-atelybeforetheverb”. Theyhappento be just thosewhich appearin theclausalpe-ripherybeforeTNSP (namely, in oneof thefourstructuralpositionsbetween[Spec,C]andΣ0 in (2)). Thus,our explanationsdiffer in how theyconceiveof therelationbe-tweenthesyntaxof theclause’s left peripheryandclitic placement.Forustherelationis direct, for Pappasit is mediatedby a sequenceof analogicalprocesses.For us it

reflectsan importantgeneralizationinherentin the grammaticalsystem,for Pappasthereis nogeneralizationto bemade.

Oneoverarchinggeneralizationis howeverbuilt into his system: that clitics inLMG alwayscomeeitherimmediatelybeforeor aftera verb. Surprisingly, this gen-eralizationturnsout to befalsebecausethereareverblesssentences,which canhaveethicaldative(“benefactive”)pronouns.

(5) a. tiswhat

suyou-Dat

ithe

doliawretched

tixifate

‘What a wretchedfatefor you!’ (Threnos105)

b. puwhere

mime-Dat

tothe

foslight

tothe

esjitonperceptible

‘Whereis theperceptiblelight for me(= my eyesight)?’(Belisarios487)

Theamendeddescriptivegeneralizationis thatclitics comenextto theverbif thereisone. Nothingneedbe addedto our analysisto coververblesssentences.Clitics areplacedat theleft edgeof theTNSP projectionthatis headedby theverb,andundergoprosodicinversionaroundtheverbif theyfind nohoston their left. This rule remainsvalid evenwhenthe TNSP lacksan overt head(or the headis a null copula). Ouranalysisalsocorrectlypredictsthat in modernStandardGreek,whereobjectcliticsareaffixedto verbsat theword level, theymustappearnextto a verb:

(6) tiWhat

jaFut

masus-Gen

tragudisissing

apopse,tonight,

keand

tiwhat

(*mas)(*us)

avriotomorrow

‘What will yousingfor ustonight,andwhattomorrow?’

Beforelookingfurtherintowhatthesynchronicevidencehasto say, letusnotethatthe two proposalsalsomakedifferentpredictionsaboutthehistory. Pappaspredictsthat preverbalclitic placementbeginswith the subjunctivemarkerna. We predictthatelementsin Σ0 (includingna if thatalreadywasa modalhead)andelementsin[Spec,Σ0] shouldbeginto inducepreverbalclitic placementat thesametime, sincethe specifierand the headareboth eligible hostsfor clitics. In general,we expectpreverbalclitic placementto haveevolvedhandin handwith the elaborationof theleft-peripheralfunctionalcategories.

2.3 Subjects

Turningnowto theapparentoptionbetweenpreverbalandpostverbalclitic placementrecordedin case(1c), recall that for us this option is not aboutclitic placementatall, but aboutalternativestructuralpositionsto which subjectsandadverbialscanbepreposed.A preverbalsubjector adverbialcouldbeeithera topic adjoinedto ΣP, ora focusin [Spec,ΣP]. If it is a topic, it occurswith postverbalclitics, if it is a focus,it occurswith preverbalclitics.9 (The sametwo positionsareavailablefor objects,but their placecanbedetectedevenin written discoursebecausetopicalizedobjectsnormallycorrespondto aresumptivepronoun.)Independentlyof cliticization,intona-tional criteriacandistinguishbetweenthesetwo structures.In written texts,we must

besatisfiedwith cluesfrom thediscoursecontext,whichconstrainthedistributionbutobviouslydonotalwaysuniquelydetermineit.

To counterour claim that thediscoursefunctionof a preposedelementis relatedto clitic placement,Pappascitesapairof exampleswith “the sameinterpretationwithrespectto thefocus/topicdistinction”butdiffering clitic placement.

(7) a. okapotsometime

apesosasinfinished-3pl

iljancame-3pl

isto

tothe

montorionMontorion

||||

othe

duksduke

tusthem

apodexjikenreceived-3sg

‘In time theyfinished[their journey],theycameto Montorion. Thedukereceivedthem.’ (Phlorios303-304)

b. kavalikevunride-3pl

arxondeslords

ipagungo-3pl

sto

tothe

palatipalace

||||

kjand

othe

vasilefsking

edexjinreceived-3sg

tusthem

‘Thelordsride,theygoto thepalace,andtheking receivedthem’(Phlorios938-939)

Pappas’argumentis basedon theclaim that(7a)and(7b) havethesamereading(hedoesn’tsaywhetherfocusor topic). Themerepossibility that theymight havea dif-ferentreadingwould invalidateit. It seemsto usthatit is morethanpossible.A focusreadingfor (7a)“the dukehimself receivedthem” actuallyenhancesthenarrativebyconveyingtheadditionalinformationthatthetravelerscouldwell havebeenreceivedby someother,lessimportantpersonage.At anyrate,thefocusreadingis consistentwith the context. On theotherhand,a plausiblereadingfor (7b) “the king receivedthem” hasfocuson theverbwith thesubjectconstruedasa topic. Thegeneralpointis that focusstructureis not automaticallyfixed by the textualcontext,or evenbythe extralinguisticcontext,becauseit dependson what the speakerhasin mind andwantsto express.A constituentmaybefocusedif thespeakeror writer thinksof it ascontributingaparticularlynoteworthyor surprisingpieceof information,or wishestorepresentit assuch,but in theabsenceof enoughsyntacticandintonationalcuesonewouldhaveto bea mind-readerto predictwhenthatis thecase.

A further complicationis that in certaintypesof presentationalsentences,evenmodernGreekliterarystyleallowstopicalizationof objectswithoutclitic doubling:

(8) tinthe

avliyard

tisof-the

filakisprison

eluzewashed

othe

iliossun

‘the prisonyardwasbathedin sunlight’ (K. Theotokis,O Katadikos)

HerecolloquialGreekwouldrequirea resumptiveclitic (Tin avli tis filakis tin eluzeoilios). Wasthis specialfeatureof presentationalsentencespresentin LMG already?Thisquestionmustbeleft for furtherwork.

Finally, as mentioned,evennon-focusedsubjectscan apparentlybe placedin[Spec,ΣP]. Thereseemsto besomevariationamongtextson this score.In Threnos,subjectsarepredominantlypreverbal,andconsistentlyinducepreverbalclitics.

2.4 Adverbs

Mackridgehadnotedthata classof one-wordadverbsof time allow preverbalcliticsmore readily than other preposedadverbials. We suggestedthat theseadverbsareoptionally nonprojecting(X0) categorieswhich appearwithin ΣP, just asa smallerset of time adverbslike panta ‘always’ and tote ‘then’ still doesin modernGreek(Alexiadou1994).10 As maximalprojections,thesameadverbsaretopicalizedoutsidethecoreclause.

Contestingthis,Pappasassertsthatall “temporalexpressions”areequallyassoci-atedwith preverbalandpostverbalcliticization. This is possible.However,we thinkthat he may havedismissedMackridge’s observationtoo hastily. Unfortunatelyhedoesnot tell us how he categorizedadverbs,but the one“temporalexpression”thathe doescite is the measurephrasexronuspeninda“for fifty years”. This suggeststhathemayhavemisclassifiedtherelevantadverbialcategoriesby conflatingtempo-ral adverbswith measurephraseswhichhappento referto time. It is well knownthatmeasurephrases(adverbialsof extent)suchas (for) fifty yearspatternsyntacticallylike spatialmeasurephrasessuchas(for) fifty miles, andunlike adverbsof time suchastodayandin fifty years. In manylanguagestheformerconstituteaclearcutseparatecategorywith characteristicsyntacticpropertiesof its own. For example,in Finnishtheyshift from accusativeto partitivecaseundernegation,like objectsbutunliketimeadverbials.In Sanskrittheyarethe only adverbialswhich passivize.In Greek,En-glish,Finnish,andotherlanguagestheyareusuallypreposedonly whenfocusedor ascontrastivetopics,with thespecialintonationcontoursappropriateto thosediscoursefunctions,whereasregularadverbsof timeareeasilypreposedunderneutraldiscourseconditions.

(9) a. simeratoday

perpatisaI-walked

dıotwo

ores.hours

‘TodayI walkedtwo hours.’

b. dıotwo

oreshours

perpatisaI-walked

simera.today

‘Two hoursI walkedtoday.’ (twohoursis focusor contrastivetopic)

Both in GreekandEnglish,(9a)canreceivenormalintonation,while (9b) requiresaspecialintonation.

In view of thiswewouldbesurprisedif amorefine-grainedinvestigationof LMGadverbs,basedonasyntacticallyadequateclassification,did not revealdifferentpref-erenceswith respectto topicalizationandfocusing,correlatingwith differencesin theclitic placementtheyinduce.

Pappas’own interpretationof the variableeffect of temporaladverbson cliticplacementis this: “Temporalexpressionsareusuallyclausaladjuncts,andsubjectsarerenderedpartlyredundantby pro,soI suggestthatspeakerswereambivalentaboutthesignificanceof suchelements,hencethe freevariationin theseenvironments.Itis quitepossiblethatthefocus/topicdistinctionwasalsoinstrumentalin creatingthis

distinctionsincefrontedconstituentsaremorelikely to befocusedelementsthansub-jects. It mustbeemphasizedthatsuchlinearrelationships,astheonesI propose,areby definition weakerthan structuralones,which would explainwhy the patternofvariationappearsto besofluid at first sight.”

Thenotionof speakers’“ambivalence”is a little vaguehere,but maybethe ideais that subjectsleavea null resumptivepronounwhich countsoptionallyasa verbalargument,andthatclitic placementwith preposedadverbsdependson whethertheyareadjunctsor subcategorized.Thesepredictionsarecertainlyworth testing;at thispointwehavenoevidenceeitherway. Wedonotseein whatsenseorderis a“weaker”relationthanconstituency. In point of fact,all of Pappas’proposedanalogicalgener-alizationsarebasedonlinearwordorderrelationsratherthanonconstituentstructure.Therefore,he canhardlyappealto the “weakness”of linear word orderrelationstoexplain why temporaladverbsand subjectsare specialin allowing seeminglyfreevariationin clitic placement.

2.5 Gerundsandimperatives

Althoughwe did not specificallydiscussgerundsandimperativesin our article,ouranalysisstraightforwardlypredictsthattheyrequirepostverbalclitics. Gerundsclearlyhavenoneof thediagnosticpropertiesof ΣPs.Theydonotallow focusedconstituentsor preverbalsubjectsor emphaticnegation,astheyshoulddo if theyhada [Spec,ΣP]position.And theydonotallow modalparticlesor den, astheyshoulddoif theyhadaΣ0 position.Theconclusionis thattheyarenot ΣPs(asPinon(1993)alreadyarguedonsimilar groundsfor Hungarian).On thatunderstanding,thereis nothingto theleftof the gerundthat can hosta clitic. Thereforeprosodicinversionmust takeplace,which is to saythatpostverbalclitic placementin gerundsis obligatory.

Pappas’claimthatwecannotderivethecliticizationpatternassociatedwith gerundsis simplyanerror,basedon thefalsepremisethatgerundsareΣPs.

As for the imperative:asa non-tensedbut modalverbalcategory, it oughtto bein Σ0 (just astensedverbsarein TNS0, andfor analogousreasons).This seemsto bebasicallycorrect.It predictsthat imperativescannotbenegated(imperativesmustbereplacedby subjunctivesundernegation),andprovidesthe[Spec,ΣP] positionneededfor preverbalfocus(to vivlio feremu“bring metheBOOK”). This sufficesto accountfor the main differencesbetweengerundsand imperatives.Gerundsdo not projecta ΣP or CP, hencedo not havepreverbalfocus, and havepostverbalcliticization.Imperativesdoprojecta ΣP andthereforeallow preverbalfocusin its specifier.

However,Pappashasdiscoveredan extraelementof variation in LMG impera-tives: whenanimperativeverbis precededby a focus,a following clitic is preverbalabouthalf thetime. Thevariationmayindicatethatanimperativeverbmay, insteadofraisingto Σ0, stayin T0, like aregularfinite verb.If it movesto Σ0, a following clitic ispostverbal,by thestandardencliticizationto theprecedingelement(unlikepostverbalcliticizationwith finite verbsor gerunds,which is theresultof prosodicinversion).If

it staysin T0, theclitic will bepreverbalif it hasa hoston its left, which is to sayinthepresenceof a preverbalfocus.11 Whenimperativeverbsstayin T0, we predictthepossibility that theymight benegated.And in fact, instancesof negatedimperativesoccurin LMG:

(10) tothe

thelimawish

moumy

plirosonfulfill

keand

apijisdisobedient

minot

ginubecome

‘Fulfill my wishanddonotbedisobedient.’ (Belisarios221)

2.6 Auxiliaries

Pappasfoundaninstanceof aclitic betweentheauxiliaryexoandtheinfinitive:

(11) naSubj

xeshave

toit

teliosifinish-Inf

‘If only youhadfinishedit’ (Rhodos454)

We hadpredictedtheexistencethis constructionaspartof an intermediatesyntacticsystem(with procliticV0) andfoundit in themodernKozanidialect.However,Pappasdismisses(11)asan“aberration”on thegroundsthatit is theonly examplehefound,in accordwith themaximunustestis,nullustestis.

While thismaximis asoundruleof thumb,it shouldnotbeappliedin amechanicalway. Evena singleoccurrenceof a constructionshouldbetakenseriously, aslongas(1) thereadinghasgoodtextualandeditorialsupport,(2) theconstructionitself is rareenoughthat the existenceof a singletokencouldbe dueto chance,and(3) thereisa principledexplanationfor theconstruction.Thesethreecriteria,which Pappashaselsewherearticulatedandinvokedhimself,12 arefully applicableto (11) andmilitateagainstdiscardingthereadingasascribalerroror othertextualcorruption.While notwishingto trumpetthis oneexampleasfurtherconfirmationfor thepredictedKozanipatternuntil we seeevidenceof theotherpartsof thatpattern,we find its peremptorydismissalasan“aberration”premature.

2.7 Thesyntaxof olos

Pappas2001:102notesthata topicalizedobjectunexpectedlyallowspreverbalcliticswhenaccompaniedby olos ‘all’. This possibilityariseswhenolos is focusedandtherestof thenominalappearsto its left in Topicposition:

(12) [ [ testhe

xorescountries

mumy

ologiraall-around

]DP [ [ olesall

]QP testhem

afanizidestroys

]ΣP ]ΣP

‘My countriesall around,hedestroysthemall.’ (Rimada1308,Pappas2001:79,Pappas,this issue)

It doesnot matterfor purposesof the presentdiscussionwhetherthe quantifierandthenominalphraseareseparatedby a movementprocess(theusualassumption),orbase-generatedin separatepositionsandcoindexedby aruleof construalor anaphora.

Whatis importantis thatsuchsentenceshavebotha topic anda focuscorrespondingto thesameargument,soto speak.Thegrossdescriptivegeneralizationsin (1) donotgiveaclearresultfor thiscase,sincetheydonotspelloutwhichphrasetheresumptiveclitic pronounis associatedwith. Themoreadequatedescriptivegeneralizationin (3)andthefinal formulationin (4) correctlypredictthat thefocusedelementwill hostapreverbalclitic whetherthereis a topicadjoinedbeforeit or not.

Pappascitesexample(13)asaproblemfor ouranalysis.

(13) tathat

idesyou-saw

keand

tathat

ikusasyou-heard

olaall

afigisunarrate

meme

tathat

‘what youhaveseenandheard,tell it all to me’ (Lybistros2784)

He overlooksthatthis sentencehasanimperativeverb. As we havejust seen,imper-ativeswork differently in that theycanmoveto Σ0, in which casea clitic will followthem.Pappas’objectionis thereforeunfounded.

In Pappas’own analysis,thepreverbalplacementof clitics afterolos is treatedasan idiosyncrasywhich hasonly a historicalexplanation.Pappasascribesit to con-taminationwith the“partitive” construction.His ideais that,at thestagewhencliticswerestill consistentlypostverbal,sentenceslike (14a)couldbemisparsedashaving,insteadof a genitiveclitic thatis partof theobject,anobjectclitic beforetheverb,asin (14b).

(14) a. [ olusall(Pl)

tusthem

]NP [ epatise]VP

he-conquered

‘all of them,heconquered’

b. [ olusall(Pl)

]NP [ tusthem

epatise]VP

he-conquered

‘All, heconqueredthem’

This misparsinggaverise to a new specialrule “object clitics canprecedethe verbafterolos”, which thengaveriseto sentenceswith unambiguouspreverbalclitics like(15b),replacingprevious(15a).

(15) a. olinall

gnorizisyou-know

tinit/her

‘you knowit all’

b. olinall

tinit/her

gnorizisyou-know

Theproposedanalogyis implausible,empiricallyunmotivated,andunexplanatory.

It is implausiblebecausehearersor learnerscouldhardlyhaveassignedthewrongstructureto sentenceslike (14) in thefaceof massiveevidencethatobjectclitics werepostverbalandpossessiveclitics postnominal.It is asif hearersor learnersof Englishwereto misparsesentenceslike (16a)as(16b),with themeaningof (16c),

(16) a. [ My friend in Athens]NP [ is happy]VP.

b. [ My friend ]NP [ in Athensis happy]VP.

c. [ My friend ]NP [ is happyin Athens]VP.

andcomeup with a rule thatreplacesJohnis happyin Athensby *John in Athensishappy.

Theanalogyis unmotivatedbecauseit doesnot relatethechangeto anythingelsein theTypeA system.Exactlythesameanalogycouldbeformulatedfor thepostverbalclitic systemof thePonticdialects,for example.Ouranalysisof Ponticclitics asword-levelencliticsexplainswhy thatchangeneverhappenedin anyPonticdialect.

Finally, theanalogyis unexplanatorybecauseit doesnot relatethe specialcliticbehaviorafterolos to thespecialsyntaxof olos. We might ask,for example,why ananalogousmisparsingof (17a)as(17b)did not give riseto therule thatobjectcliticscanprecedethe verb after pateras‘father’, as in (17c), which is ungrammaticalinthesedialects.

(17) a. [ tonthe

paterafather

masour

]NP [ esosan]VP

they-saved

‘our father,theysaved’

b. [ tonthe

paterafather

]NP [ masus

esosan]VP

they-saved

‘father,theysavedhim for us’

c. *[ tonthe

paterafather

]NP [ masus

tonhim

esosan]VP

they-saved

‘fathersavedus’

Our analysisoffers an accountfor why it is preciselythe quantifierolos that allowspreverbalcliticization,andnotsomeotherelementor classof elements.

2.8 uk andanuk

MackridgeandPappasnotedthatuk is associatedwith postverbalclitic placementandan uk is associatedwith preverbalclitic placement.We proposedthatuk is locatedinahigheroperatorposition([Spec,C], let usassume)andthatit attractstheverbto C0.But, sincean “if ” is acomplementizer(asits behavioron its ownshows,see(1b5)),itfills theC0 position,which precludestheverbfrom movingto C0. Thepositionof ukis thendeterminedby its scope.If uk hasscopeoveran in C0, it movesup to [Spec,C] (uk an “not if ”), if it comesunderthescopeof an, it remainsin thelower positionwherenegationnormallyappears(an uk “if not”). In eithercase,theclitic of courseremainspreverbal,sinceit hassomethingon its left to leanon.

Similar variationin theplacementof negationis foundin someof theotherolderIndo-Europeanlanguages.In Old English,thenegationne is assumedto bein [Spec,C] (or in somehigh operatorposition),whereit attractstheverbto the following C0

slot (see(18a)). But in the presenceof the complementizergif “if ” in C0, negationandtheverbstayin a lowerposition,asin (18b).13

(18) a. &and

nenot

bidis

dærthere

nænigany

ealoale

gebrowenbrewed

midamong

EstumEstonians

‘andnoaleis brewedamongtheEstonians’

b. GifIf

pesthis

bealdwyrdabold

biscopbishop

acwealdkilled

nenot

bidwill be

‘If thisboldbishopwill notbekilled’

Pappasinsiststhat an uk is a problemfor us. Somehowhe missedthe obvioussolution just presented,evenwhile proposinga similar oneof his own. Accordingto him, pronounsappearpostverballyafter uk becauseuk is proclitic andpre-emptsthe preverbalposition. When an anduk arecombinedthey form a phonologicallyindependentunit whichnolongerrequiresahost.As aresult,thepreverbalpositionisfree,and,sincetheverbis precededby a functionword, thepronounis placedto theleft of theverb.

Thoughbroadlysimilar in conception,our respectivesolutionsdiffer in the as-sumptionsthey rely on. Ours dependson the principle that the orderof operatorsreflectsscope,andon the constraintthat the complementizerslot cannotbe multi-ply filled. This so-called“doubly-filled Comp filter” holds for ModernGreekandis widely attestedelsewhere;it is likely to havebeenvalid for MedievalGreekaswell.14 Pappas’proposalassumesthat a clitic pre-emptsthe preverbalposition,sothattherecanbeonly oneproclitic perhost.15 But aputativeconstraintblockingmul-tiple clitics doesnotevenholdfor Greek.Pappashimself(2001:79)effectivelyrefutesit for LMG by his observationthat,whena verb is associatedwith two clitics, bothareput on thesamesideof theverb,eitherbeforeit or after it. So, if a procliticizeduk on theverbpreventsproclisisof a pronounto it, why doesn’ta procliticizedpro-nounon theverbpreventit? On Pappas’accountwe would expectthefirst pronounto pushthe secondinto postverbalposition,but Pappasstatesthat split cliticization(Pronoun+V+Pronoun)doesnotoccurin hisdatabase.

Theuk andan uk dataturn out to supportour analysis.Pappas’criticism is mis-placed. His own solutionrequiresan otherwiseunmotivatedconstrainton multipleprocliticswhich moreoveris inconsistentwith his own dataandwith his analysisofobjectclitics.

3 The analysisof clitics: prosodic issues

3.1 Preverbalclitics

On ouranalysisclitics arealwaysenclitic,while on Pappas’analysistheyareencliticonly in postverbalposition,andproclitic otherwise.Thepoint of disagreementis theprosodicaffiliation of preverbalclitics.

On historicalgrounds,it would not besurprisingif objectpronounswereencliticin LMG, as they were in the classicallanguageandcontinueto be in a numberof

moderndialects,includingPontic(TypeB), and,moreinterestingly, to at leastsomeCappadociandialects(Type A). In the latter, a clitic assignsa stressto a precedingmood marker,which is otherwiseunstressed(see(19c), Dawkins 1916:496,Janse1998).

(19) a. naMoodPart

seyou

pitaksoI-send

‘I wantto sendyou’

b. naMoodPart

seyou

tahim

dosoI-give

‘I’ll givehim to you’ (ibid.)

c. naMoodPart

galjepsoI-ride

‘for meto ride’ (ibid.)

Thiswhatwe calledactivesubcategorization,afterInkelas1989.

Preverbalproclisis in the LMG systemwould be unexpectedon our analysis,thoughnot impossible.Descriptively, it would meanthat thestructures(3a)and(3b)would beassociatedwith differentstresspatterns.A clitic in its basicpositionwouldattachprosodicallyto the precedingelement,whetherX0 or Xmax, without forminga prosodicword with it, andso would not assigna stressto the precedingsyllable.Prosodicinversion,on theotherhand,would resultin a moreintimatecombination,analogousto thatof anounwith apossessiveclitic, whichconstitutesaprosodicwordandsowould geta secondstressif thefirst oneis morethanthreesyllablesfrom theend.A rule to this effectcouldbeaddedto our analysis,thoughsucha complicationin anotherwiseutterlysimplesystemmight raisesuspicion.

What doesthe datatell us? Thereare two potentialsourcesof evidence: theaccentuationof the manuscripts,andthemetricalpracticeof poets. Let us considerthemin turn.

3.2 Thewrittenaccents

Beforea clitic, no rhythmicstressis markedon thefinal syllableof a precedingnounor otherprecedingconstituent,asin (20a),but it is markedon a precedingverb,asin(20b).

(20) a. sıselonwith-saddle

tonhim

epetaksathrew

‘I threwhim [off hishorse]togetherwith hissaddle’(Lybistros2047)

b. tremus´ınfear

tonhim

ithe

arxontesrulers

‘The rulersfearhim’ (Belisarios300)

If we takethesestressmarkingsat facevalue,asPappasdoes,we wouldconcludeonouranalysisthatprosodicinversionaroundtheverbcreatesa phonologicalword.

But somecautionis in order. Howdoweknowthattheaccentualdifferencein (20)actuallyrenderstheauthor’s speech?It mayreflecta spellingconventionintroducedin thecourseof theMS tradition,or evenby themoderneditorof thetext. Someof theLMG accentmarkingscertainlymustbejust that,andcannotbedirectrecordsof thespokenlanguageof thetime. Forexample,theancientrulethatclitics assignanaccentto thefinal syllableof properispomenabut not to thefinal syllableof paroxytona(see(21))is faithfully adheredto in thetext,eventhoughtheaccentualdistinctionbetweenacuteandcircumflex,andthe lengthdistinctionthat supportedit, hadbeenlost forwell overa thousandyears.

(21) a. poisè ton (Belisarios53)

b. poÐsw se (Belisarios197)

Similarly, preverbalclitics arealwayswritten with anaccent,eventhoughtheywerealmostcertainly unstressed,at least beforea following stressedsyllable, whereaspostverbalclitics werewritten without an accent. In suchcasesPappaswould pre-sumablyagreethatthespellingis artificial.

3.3 Theevidencefrom meter

To supporthis claim thatclitics do not assigna stressto a precedingsyllable,Pappasarguesonmetricalgroundsthatin alinesuchas(22),thesyllable-tescouldnotreceiveastressfrom thefollowing enclitic.

(22) ithe

erotescupids

tonhim

egenisangavebirth

‘the cupidsgavebirth to him’ (Achilleid 1113)

For if it did, “the hemistichwould havethreebeats,on thesecond,fourth, andsixthsyllables,whichis a rhythmicpatternthatI havenotencounteredanywhereelse” (ouritalics).

We arepuzzledby this argument,for two reasons.First, a three-beatrhythmicpatternwith stresson thesecond,fourth,andsixth syllablesoccurson virtually everypageof everytext in politikosdekapentasyllabosmeterthatwe haveseen.Openingour copy of Wagner1970literally at random,we find hemistichswith beatson thesecond,fourth,andsixthsyllablessuchas:

(23) a. kefali,head

xeria,hands

somatabodies

‘head,hands,bodies’(Threnos82)

b. dukades,dukes

kunti,counts

prınkipesprinces

‘dukes,counts,princes’(Threnos177)

Secondly, in (22)thelastbeatactuallyfalls ontheseventhsyllable,notonthesixthasPappassays,becausethefirst dipodyendsin anextraunstressedsyllable.But this

doesnot changetheconclusion,for hemistichswith this rhythm,havingbeatson thesecond,fourth,andseventhsyllables,arealsowell attested:

(24) a. mikronsmall

tothat

legeisays

othe

prologosprologue

‘what theprologuesaysis asmallthing’ (Threnos19)

b. amibut

sxolazo,I-stop

afinoI-leave

toit

‘But I stophere,I leaveit’ (Threnos42)

This is not in the leastsurprisingif we considerthe structureof the meter. It isbasedon four iambic dipodies,of which the last is catalectic. Thusit belongsto avastfamily of meters,which includesalsothe ancientGreekiambic tetrameter,theLatin septenarius,16 the Englishfourteener,and, in 20th centuryGreece,the balladmeterof Kavvadias.It is derivedsimplyby successivebinarybranching,sothata lineconsistsof two hemistichs(H), a hemistichof two dipodies,or cola(D), a dipodyoftwo iambic feet (F), andeachiambic foot of two syllables,exceptfor the last foot,which is unary(f).

(25)

H H

D D D D

F F F F F F F f

Thereis anobligatorycaesurabetweenthehemistichs.Everyeven-numberedsyllablefalls in a strongpositionandis a preferredlocationfor stress,althoughweakposi-tionsmaybestressed.Eachdipodymustcontainat leastonestress.Therefore,eachhemistichhasat leasttwo “clearly felt beats”. But it canalsohavemorethan two,aslines like (23) and(24) show.17 Hencewe think that theargumentfrom metricalusage,far from being“subtle” and“convincing”, is incorrect.

Pappas(this issue)arguesthataclitic is prosodicallyattachedto a following verb,on te groundsthat it canundergocontractionwith it. Again,we think this is a hastyconclusion,becausecontractionin LMG takesplaceevenbetweenfull words.Someof it is reflectedin theorthography, andthemetershowsevenmoreof it:

(26) Wordsin isolation: thelonaipo alla oligaTextaswritten: thelona’po all’ oliga

Metrical: thelso na’p

so ’ll’ ol

siga

But themeterrevealsthatevenpossessiveclitics, which certainlyencliticizeto theirleft, cancontractwith a following word.

(27) t’the

onomaname

m’my

unot

grafowrite

toit

‘I don’t write my name’(Rhodos1025)

On theotherhand,contractionis optionalevenwith preverbalclitics, asit is be-tweenwordsin general:

(28) mikrossmall

ior

megasbig

diafor

nato

pi,say

diafor

nato

seyou

onidisiblame

‘someonesmallor big to sayor to blameyou’ (Threnos166)

Herethe-eof theclitic secannotcontractwith thefollowing vowel,otherwisethelinewould lack a syllable. Thus,contractionis neitherrestrictedto clitic boundaries,norobligatoryat them.

In sum,theevidencefor preverbalproclisisin LMG is not compelling. The tes-timony of orthographyis suspectbecauseof its artificiality, andsofar at least,meteryieldsnofirm clueseitherway.

4 Conclusion

Wethinkwehaveshownthatouranalysisstandsupwell to Pappas’objections,thoughwecertainlydonotexpectit to remainthelastwordonthesubject.His ownview thatthe LMG patternis not subjectto any syntacticgeneralizationsseemsto us whollyunjustified.Thissaid,wewould like to emphasizethathiswork is themostimportantcontributionto LMG syntaxto haveappearedin a long time. The readerwill havenoticedhow muchof the aboveargumentationis basedon dataamassedby Pappashimself. With respectto carefuldocumentationhe setsa standardthat is only toorarelyattained.

Amongtheopenquestionsthatremainontheagendais therelationshipof LMG toearlierstagesof Greekandto themoderndialects,andpossiblefiner differentiationswithin LMG by chronology, style andgenre,andprovenance.Detailedanalysisofthepatternsof apparentoptionalitywill very likely revealadditionalfactorsat work.Heresyntacticanalysismustgohandin handwith statisticalmethodsof thesortusedby Pappas.Thesearea powerfulprobeinto grammar,but the answerstheyprovidecanbe only be asgoodasthe linguistic questionsthat go into them. Givena setofcandidatefactors,statisticscantell uswhichof themmakeadifference.Forexample,Pappaswasableto simplify Mackridge’sdescriptivegeneralizationsby showingthatsomeof the statisticaldifferencesbetweenenvironmentsareartifactsof meter. Butthesemethodscan’t detectfactorsthat are not codedinto the databasein the firstplace.Sinceit is impossibleto codeeverylogically possiblefactorinto thedatabase,the factorsthatdo getcodedhadbetterbewell chosen.This is wherethechoiceofsyntacticanalysisbecomescritical. Amongthedesideratafor futurework area morefine-grainedclassificationof adverbsbasedonthecategoriesknownto berelevantfor

word orderandconstituency, a closerinvestigationof subjectplacement,anda fullerexploitationof themetricalevidence.

Notes1We would like to thankBrian Josephandhis co-editorsfor giving us the opportunityto returnto the

topic, aswell asfor their helpful editorial comments.

2 In general,of course,ModernGreek,hasno V-to-C movement.That cases(a4) and(a5) really areresidualirregularitiesfrom anearlierstageis alsosuggestedby that fact that theyarelargelyeliminatedinthemodernTypeA dialects.In Cappadocian,u(k) getsregularpreverbalclitics after it everywhereexceptin thePharasadialect,anddi (thepresumedavatarof oti) seemsto behaveasaparticlewith theverb“say”;see(29), from Dawkins1916:492:

(29) Ipensaid

didi

jeand

jınothat(DEM)

othe

fugaraspoorman

kithat(COMP)

. . .

. . .‘and thatpoormansaidthat. . . ’ (with directdiscourse)

3In C&K 2001we suggestedthat prosodicinversionis not syntacticmovement,but a strategyfor re-solvingcompetingsyntacticandprosodicconstraints,in thespirit of OT. Severalotherwaysto think aboutprosodicinversionarecurrentlyunderdebate.For presentpurposes,thequestionis not important.

4In our view, analogybasicallyregularizesstructure.To besure,complexmixesof theold systemandthe newonecanoccuruntil the analogygoesto completion.Also, learnersmay internalizetheir own (oreachothers’)wrongoutputsat intermediatestagesof acquisition.SeeKiparsky1979,wherethesecasesarerespectivelyreferredto as PARTIAL ANALOGY andFALSE ANALOGY. Pappas’proposalseemsdifferent,though.Heseemsto besayingthatarobustantecedentsystemis disruptedby localanalogieswhicharenotrelatedto thenewsystem(ModernGreek)thatultimatelybecomesestablished.This seemsmoredubious.

5Settingasidemixed dialectslike that of Amisos,clitic-specificvariation in the dialectsis confinedtotwo things: the uniquebehaviorof the negationjo in the dialectof Pharasa,which preservedthe ancientsyntaxof its cognateuk (case(1a5)),andthemergerof moodmarkerswith afollowing clitic into aprosodicword in somedialects(thephenomenonof “active subcategorization”).Suchothervariationasthereexistshasto do not with the rules for clitic placementbut with the availability of the left-peripheralpositionsfor varioustypesof elements,notablywhethernon-focusedsubjectscanmoveto [Spec,ΣP], asmentionedabove.

6Pappas(this issue,section2.5)objectson thegroundsthatchangein thedirectiontowardsthestandardGreeksystemof preverbalcliticization is foundin 17thcenturyprosefrom Crete.But no-onehasclaimedthatTypeA systemsareabsolutelyimmutable.Thepoint is ratherthat theycanbestablytransmitted,andfor themostparthavebeen,which beliestheir supposedlyanomalousstatus.Like anyotherregularfeatureof language,they canchangethroughnormativepressureor borrowing,or evenby ordinaryendogenousprocesses.

7It is telling that the behaviorof the negationu(k), a featureof LMG which really is synchronicallyanomalous,wasregularizedeverywhereexceptin themoreconservativedialectof Pharasa(fn. 2).

8Remarkably,Anttila showsthat,in arangeof interestingcases,theobservedfrequenciesof thevariantsin agivenenvironmentarepredictedby theproportionof allowedrankingsin which it is theoptimaloutputin thatenvironment.

9As mentionedearlier,somedialectsalsoseemto allow non-focusedpreverbalsubjectsto beplacedin[SPEC,ΣP].Again this optionshouldcorrelatewith clitic positioningin TypeA dialects.This option,andits dialectaldistribution,remainsto beinvestigated.

10Thediagnosticis that theycancomeright aftera focusor a Wh-phrase:Pjon akomaskefteteo Yanis?‘Who is Johnstill thinking about?’

11This latter situationmay persistin the modernCretandialectsmentionedby Pappas.Kontosopoulos(1994:32)reportspreverbalclitics in imperativesin the presenceof preposedobjects(enagafemu kame“makemeacoffee”), suggestingthat in this dialectimperatives,like otherfinite verbs,remainin TNS0.

12Pappas2000defendstheauthenticityof ahapaxlegomenoncomparableto (11),arguingthatpreciselythesethreecriteriaoutweightheunustestis,nullustestisprinciple. It would havebeenconsistentto invokethesameconsiderationsheretoo.

13Theseexamplesarecitedfrom Traugott1992.

14The doubly-filled Compfilter is actuallyoneof the earliesttypologically generalconstraintsto havebeenformulatedin generativesyntax(ultimatelydatingbackto Ross1967).Wedonotneedto claim thatitis universal,of course,merelythat it holdsfor MedievalGreek,which is ratherlikely for thereasonsstatedin thetext.

15Of courseweclaim thatclitics areenclitic evenpreverbally,butwewill setasidethis point for thesakeof theargument.

16Blumenfeld2004makesa convincingcasethat theearliestform of theLatin septenarius,thatof Plau-tus,wasactuallytrochaic.

17Pappas’error may be due to a misunderstandingof Horrocks’s description(1997:257): “In eachhemistich,therearetwo clearly felt beats,which areprovidedby accentplacement.In thefirst hemistichthe accentmay occureither on the secondor fourth syllable andon the sixth or eight syllable, while inthesecondhemistichthe accentmayoccureitheron the tenthor twelfth syllableand,obligatorily, on thefourteenth.”

References

Alexiadou,ArtemisandElenaAnagnostopoulou.1998. “ParametrizingAgr: wordorder,verbmovement,andEPPchecking.” Natural LanguageandLinguisticTheory16:491–539.

Anttila, Arto. 1997.Variation in Finnishphonologyandmorphology. Ph.D.disserta-tion, Stanforduniversity.

Blumenfeld,Lev. 2004.“Foot structureandmeterin PlautineLatin.” Talk presentedat theLSA meeting.

Boersma,Paul,andBruceHayes.2001.“Empirical testsof theGradualLearningAl-gorithm.” LinguisticInquiry 32:45-86.

Dawkins,RichardM. 1916. ModernGreekin Asia Minor. Cambridge:CambridgeUniversityPress.

Giannakidou,Anastasia.1998.Polarity sensitivityas(non)veridicaldependency. Phil-adelphia:Benjamins.

Halpern,Aaron.1995.On theplacementandmorphologyof clitics. Stanford:CSLIPublications.

Horrocks,GeoffreyC. 1997.Greek:A historyof thelanguageandits speakers.Lon-donandNew York: Longman.

Inkelas,Sharon.1989. Prosodicconstituencyin the lexicon. Doctoraldissertation,StanfordUniversity.

Janse,Mark. 1998. “Cappadocianclitics andthe syntax-morphologyinterface.” InBrian D. Joseph,GeoffreyC. HorrocksandIrenePhilippaki-Warburton,eds.,Themesin GreekLinguisticsII , 257-281.Philadelphia:Benjamins.

Kiparsky, Paul.1979. “Analogicalchangeasa problemfor linguistic theory.” in B.Kachru(ed.),Linguisticsin theseventies:Directionsandprogress. StudiesintheLinguisticSciences8,2.

Kontosopoulos,Nikolaos.1994. Di�lektoi kaÐ >Idi¸mata thc Nèac <Ellhnikhc.[Dialectsandlocal varietiesof ModernGreek.] Athens.

Kroch,AnthonyS.1982/1989. “Function andGrammarin the History of English:PeriphrasticDo.” In Ralph Fasold,and DeborahSchiffrin (eds.) LanguageChangeandVariation, 133-172.Orlando:HarcourtBraceJovanovich.

Kroch,Anthony. 2000. ”Syntacticchange.” In Mark Baltin andChrisCollins, eds.,The Handbookof ContemporarySyntacticTheory, 629-739. Malden, MA:Blackwells.

Mackridge,Peter. 1993.“An editorialproblemin medievalGreek:thepositionof theobjectclitic pronounin theEscorialDigenesAkrites.” In NikolaosM. Panayio-takis, ed., >Arqèc thc neoellhnikhc logoteqnÐac / Origini della letteraturaneogreca,325–342.Venice.

Pappas,Panayiotis.2000. Unustestis,nullus testis? Working Papersin Linguistics54:171-176.OhioStateUniversityDepartmentof Linguistics.

Pappas,Panayiotis.2001. Weak object pronounplacementin Later MedievalandEarly ModernGreek.Doctoraldissertation,TheOhioStateUniversity.

Pinon,Christopher. 1993.“SigmaPandHungarian.” WCCFL11:388–404.

Ross,JohnRobert.1967.Constraintsonvariablesin syntax.Ph.D.thesis,MIT.

Traugott,Elizabeth.“Syntax”. In RichardM. Hogg(ed.),TheCambridgeHistory oftheEnglishLanguage,Vol. 1. Cambridge:CUP.

Tsimpli, IanthiMaria.1995. “Focusingin ModernGreek.” In Katalin E. Kiss, ed.,Discourseconfigurationallanguages, 176–204. Oxford: Oxford UniversityPress.

Tsimpli, IanthiMaria,andAnnaRoussou.1996.“Negationandpolarityitemsin Greek.”TheLinguisticReview13: 49–81.

Wagner,Wilhelm. 1970.MedievalGreektexts:beinga collectionof theearliestcom-positionsin Vulgar Greek,prior to the year1500. Amsterdam:Gruner. (Un-changedreprintof theLondon1870edition).

Me thn ergasÐa auth uposthrÐzoume thn prohgoumenh analus  mac (Condoravdi& Kiparsky2001)thc suntaktikhc domhc kai twn proswpikwn klitikwn antwnu-miwn sta metagenestera Mesaiwnika Ellhnika, antikrouontac thn kritikh touPappa pou dhmosieuetai s' auto to teuqoc. Sugqronwc epishmainoume merikaproblhmata sthn analush tou idiou tou Pappa kai upodeiknuoume jemata giaperaiterw ereuna.