Clause-BoundedMovement:StylisticFrontingandPhaseTheory AntonKarlIngason(UniversityofIceland) JimWood(YaleUniversity) 1 In this squib, we provide novel empirical support for treating the thematic domain—the “vP”—as a locality domain like CP (a “phase”), in agreement with a growing body of research (see Barbiers 2002, Fox 1999; Legate 2003; Rackowski & Richards 2005; Cozier 2006; Kahnemuyipour and Megerdoomian 2011; Buell 2012; and van Urk & Richards 2015; see Den Dikken 2006 for an opposing view). We show how vP phasehood explains a previously unsolved problem for defining the locality of Icelandic Stylistic Fronting (SF). We present novel data to show that SF of verbs and particles can only cross one phase boundary, a generalization that is empirically superior to “clause-boundedness.” Our study supports the view that v defines a phase edgewhethertheverbislinkedtoanexternalargumentornot(Legate2003). 1 StylisticFrontingandLocality Stylistic Fronting (SF) has provided an important data source for our theoretical understanding of EPP and verb-second effects (Maling 1990; Holmberg 2000; SigurDsson 2010; see Holmberg 2005 for a thorough overview). SF is movement of a non-finite verb, verbal particle, adverb, adjective, PP or DP to the left of a finite verb. Itispossible,andoptional,whenthereisasubjectgap;thatis,wheneverSpecTPisnot filledbyphonologicallyovertmaterial,suchasinsubject-extractedrelativeclauses.1 (1) [ Bókin [ semstolið hefurveriðstolið ]]erafar verðmæt. DP CP [ the.book[ that stolen.PTCP has been stolen]]is veryvaluable DP CP ‘Thebookthathasbeenstolenisveryvaluable.’ One important property of SF, which is the focus of this squib, is that it is clause-bounded:themovingelementmaynotcrossaclauseboundary,asillustratedfor afiniteclausein(2)andacontrolclausein(3)(examplesfromThráinsson1993:194).2 (2)*Þettaer[ stelpan[ sem stolið sagði[ að þú hefðirstolið bókinni ]]] DP CP CP this is [ the.girl[ whostolen.PTCPsaid [ thatyouhad stolenthe.book]]] DP CP CP (3)*Þettaer[ maðurinn[ sem lesa lofaði [ aðlesa allarbækurnar]]]. DP CP CP this is [ the.man [ whoread.INFpromised[ to readall the.books]]] DP CP CP WhileeveryoneagreesthatSFisclause-bounded,theanalysis,andeventheformal 2 description of this constraint, has proved recalcitrant. In Holmberg’s (2000) analysis, the closest frontable constituent moves; clause-boundedness must be stated separately. Note, for example, that lesa ‘read’ is the closest frontable constituent in (3) (since complementizers cannot undergo SF), and still it cannot move. Jónsson (1991) points outthatitisnotenoughtosaythatCPisabarrierforSF,sinceraisingverbs,standardly assumedtotakeaTPcomplement,donotallowit.3 (4) [ Þeir[ sem {*taka }virðast[ {taka}kartöflurupp]]]eruálfar. DP CP TP [ they[ who{*pick.INF}seem [ {pick}potatoes up ]]]areelves DP CP TP ‘Theoneswhoseemtoharvestpotatoesareelves.’ (5) [ Þeir[ sem {*brjóta }þykja[ {brjóta}niðurnemendur]]]eruálfar. DP CP TP [ they[ who{*break.INF}think [ {break }downstudents ]]]areelves DP CP TP ‘Theoneswhoarethoughttobreakstudents’spiritsareelves.’ Thráinsson (1993, 2007:374) proposes that the problem with (2) and (3) is not clause-boundedness per se, but the intervention of an overt complementizer head. He pointsoutthat,asoriginallyobservedbySigurjónsdóttir(1989)andSigurDsson(1989), SF is possible with some raising and control verbs when the complementizer að is omitted;thisfactwillbediscussedfurtherbelow. However, intervention of an overt complementizer head cannot explain the unacceptability of SF in (4). Moreover, this is not just a quirk of raising constructions: there are constructions that have not yet been mentioned in the literature which disallowSF.Forexample,thecausativeverbláta‘let/make’maytakeaninfinitiveverb complement with no thematic subject, and this construction does not allow SF. (We returntotheidentityofXPbelow.) (6) Þettaer[ maðurinn[ sem {*brenna} lét [ {brenna}allarbækurnar]]]. DP CP XP this is [ the.man [ who{*burn.INF}made[ {burn} all the.books]]] DP CP XP ‘Thisisthemanwhomade(someone)burnallthebooks.’ Similarly,fá‘get’maytakeaparticiplecomplementandmeansomethinglike‘manage’ (Taraldsen2010;Sigurðsson&Wood2012);thisconstructionalsodisallowsSF. 3 (7) Þettaer[ maðurinn[ sem{*brennt} fékk[ {brennt} allarbækurnar]]]. DP CP XP this is[ the.man [ who{*burned.PTCP}got [ {burned}all the.books]]] DP CP XP ‘Thisisthemanwhomanagedtoburnallthebooks.’ The same holds for SF of verb particles. (8) shows that the particle upp ‘up’ can undergo SF in the presence of the auxiliary skulu ‘shall’. (9)–(10) show that this is not possibleintheláta‘let/make’andfá‘get’constructions. (8) [ Þeir[ sem {upp}skulutakakartöflur{upp}]]eruálfar. DP CP [ they[ who{up }shall pickpotatoes {up} ]]areelves DP CP ‘Theoneswhowillharvestpotatoesareelves.’ (9) [ Þeir[ sem {*upp}létu takakartöflur{upp}]]eruálfar. DP CP [ they[ who{*up }madepickpotatoes {up} ]]areelves DP CP ‘Theoneswhomade(someone)harvestpotatoesareelves.’ (10) [ Þeir[ sem {*upp}fengutekið kartöflur{upp}]]eruálfar. DP CP [ they[ who{*up }got pickedpotatoes {up} ]]areelves DP CP ‘Theoneswhomanagedtoharvestpotatoesareelves.’ OnemighttrytosaythattheidentityofTP/XPin(4)–(7)issimplyCP,sothatthese examplessimplyillustratetheclause-boundednessofSF.However,Icelandicnon-finite CPs,suchasthecomplementsofcontrolverbs,haveverbraising,asshownbytheability oftheverbtomovetotheleftofclausalnegationekki‘not’.Raisingverbs,láta‘let/make’ andfá‘get’donotallowthis(Platzack1986;Thráinsson1986;SigurDsson1989). (11) Égreyndiaðbyggja ekkihúsið. I tried to build.INFnot the.house. ‘Itriedtonotbuildthehouse.’ (12) *Ég{virtist /lét }byggja ekkihúsið. I {seemed/made}build.INFnot the.house. INTENDED:‘I{seemedto/made(someone)}notbuildthehouse.’ (13) *Égfékkbyggt ekkihúsið. I got built.PTCPnot the.house. INTENDED:‘Imanagedtonotbuildthehouse.’ 4 Infact,thesekindsofclausesapparentlygenerallydisallowsentenceadverbialsinthefirst place(SigurDsson1989:85–87),reinforcingthepointthattheyaresmallerthanCP. 2 ANewGeneralization We propose that the novel data in (6) and (7), along with (4)–(5), reveal a previously unnoticed correlation: SF is disallowed past verbs that have their own argument structure—thatis,pastverbswhichhavetheirownvPlayerandextendedprojection.Láta ‘let/make’andfá‘get/manageto’canbeshowntohavetheirownargumentstructureon thebasisofthefactthattheycannotembedoblique-subjectverbs(seeThráinsson1986for discussionofthistest),illustratedwiththeverbtakast‘managetosucceed’. (14) Mér {mun/*lét }takast þetta. me.DAT{will /*made}succeed.INFthis ‘Iwillmanagetosucceedatthis.’ (15) Mér {hefur/*fékk}alltaf tekist þetta. me.DAT{has /*got }alwayssucceeded.PTCPthis ‘Ihavealwaysmanagedtosucceedatthis.’ Thesefactssupporttheviewthatláta‘make’andfá‘get/manageto’taketheirownthematic arguments,anddonotsimplyinherittheargumentstructureoftheircomplements.This,in turn,meansthattheseverbsprojecttheirownextendedvPlayer. Raising verbs like virðast ‘seem’ are different in that by definition, they involve raising of a lower argument; thus, they may embed oblique-subject verbs and preserve theembeddedcase,asshownin(16a).However,evidencethatraisingverbslikevirðast ‘seem’haveargumentstructureandprojecttheirownextendedvPcomesfromthefactthat theymaytakeexperiencerdativeargumentsoftheirown,asshownin(16b). (16) a. Mér virtist [ mér takast þetta]. TP me.DATseemed[ me.DATsucceedthis ] TP ‘Iseemedtomanagetosucceedatthis.’ b. Henni virtist [ mér takast þetta]. TP her.DATseemed[ me.DATsucceedthis ] TP 5 ‘Iseemedtohertomanagetosucceedatthis.’ Evenwhenthereisnoexperiencer,theimportantpointisthatraisingverbsoftherelevant kindarelexical,argument-takingverbs,andnotauxiliaries.Understandardassumptions, theywouldpatternwithunaccusativesinprojectinganunaccusativelittlev,whichdefines aphaseundertheviewthatallvPsarephases(Legate2003;Marantz2007).4 WhatdistinguishestheverbsthatdisallowSFfromtheonesthatallowSF,then,isthat theformerhaveargumentstructureoftheirown,whichincurrenttheoreticaltermsmeans thattheyprojectavPlayer.Inphase-theoreticterms,thismeansthattheydefine/project phasesoftheirown,andtheircomplementscontainseparatephases. 3 Analysis The correlation between having argument structure and disallowing SF is explained by phase theory, which treats vPs and CPs the same from a locality perspective. The descriptive generalization is this: verbs and particles undergoing SF can only cross one phase boundary, the one that the verb in question projects.5 Thus, they cannot cross a higherCPboundaryorahighervPboundary. ThisgeneralizationsupportstreatingvPs asphases,thatis,asboundariesforlocalityonparwithCPs(whichisnottosaythatvPs havethesamesetoffeaturesasCPs). The theoretical question is how to implement this generalization: why is it that a verb or particle can cross only the vP phase boundary that it projects, but no other? The answer comes from phase theory, specifically the class of theories subject to what Citko (2014:33) calls the “Weak Phase Impenetrability Condition” or PIC , which was 2 originally formulated by Chomsky (2001:13–14). Assuming that ZP is a phase that containsadistinctphaseheadedbyphaseheadH,PIC isdefinedasin(17). 2 (17) The domain of H is not accessible to operations at ZP; only H and its edge are accessibletosuchoperations. WhatthismeansisthatthedomainofHisaccessibletonon-phaseheadsthatintervene betweenZandH.ThisisbecauseTransfer/Spell-Outofaphaseisnottriggereduntilthe nexthigherphaseheadismerged.Inthepresentcase,thisrequiresonlythatweassume 6 thelandingsiteforSF(e.g.,SpecTP)islowerthanthenextphasehead.6 To take a concrete example, consider the distinction between (8) and (9), which is representedschematicallyin(18)and(19),respectively. (18) [ who[ up shall [ pickuppotatoes]]] CP TP vP (cid:120) (19) [ who[ up made[ made[ pickuppotatoes]]]]] CP TP vP vP (cid:120) × In (18), the domain of the phase vP is accessible to operations at TP, because TP is not itself a phase. Therefore, SF can apply and move either the verb or the particle to the preverbal position (assuming with Holmberg (2000) that verbs and particles form a head-complementrelation,makingthembothequidistanttothelandingsite).7However, in (19), there is an additional vP phase boundary between TP and the lower vP. Thus, as soon as that higher vP is constructed, the domain of the lower vP, including the verb andtheparticle,isnolongeraccessibletooperationsatTP.Theindependently-proposed PIC thus makes exactly the right cut: verbs and particles undergoing SF can cross one 2 andonlyonephaseboundary. 4 FurtherConsequences Allelsebeingequal,ouraccountpredictsthatcategoriesthatmovesuccessive-cyclically (e.g.,undergoA(cid:48)-movement)willnotbeclause-boundedinthesameway;forexample, DPs can move to phase edges, which should put them in a position to be Stylistically Fronted.Thispredictionisborneout(examplefromThráinsson(2007:374)). (20) ?Þettaermaðurinn[ sem allarbækurnar reyndi[ aðlesa t ]] CP i CP i this is the.man [ whoall the.books tried [ to readt ]] CP i CP i ‘Thisisthemanwhotriedtoreadallthebooks.’ Inthisexample,theDPappearstohaveStylisticallyFrontedpasttheCPclauseboundary, aswellaspasttwovPboundaries(forreyna‘try’andlesa‘read’).8 Thesameholdsfor PPs, which, like, DPs, can undergo long-distance topicalization as in (21). Adverbial proformspatternwithPPs;e.g.,þar‘there’canreplacethePPbelow.Theyarephrasalas shownbycomplexAdvPþaríbæ‘thereintown’,alsocompatiblewiththisposition. 7 (21) ...maðurinn[ sem [ í Osló]reyndi[ aðbúat ]] CP PP CP PP ...the.man [ who[ inOslo]tried [ to livet ]] CP PP CP PP ‘...themanwhotriedtoliveinOslo.’ It has been debated in the literature whether XPs should be considered to undergo SF, or whether apparent SF of XPs is really just topicalization (Rögnvaldsson & Thráinsson1990;Thráinsson2007:368;Hrafnbjargarson&Wiklund2009;Angantýsson 2011:218–221). The ‘just topicalization’ analysis suffers from the fact that in many environments(suchasrelativeclauses),the‘topicalization’isonlypossiblewhenthere is a subject gap—just like SF—and only in languages that allow SF (e.g. *...the man who the books, tried to read inEnglish)(Holmberg2000,2005).Nevertheless,Molnár (2010) has argued that in Icelandic, the information structural properties of SF of DPs andPPsdifferfromthosefoundinSFofverbs,particles,adverbsandadjectives(seealso Egerland2013). While we cannot go into details about the information structural properties of SF versustopicalization—notethatwhatiscalled‘topicalization’inGermanicisassociated with a variety information structural properties (Frey 2004, 2006)—in the present account, ‘long-distance’ SF shares syntactic properties with both topicalization and ‘clause-bounded’SF.Likewithtopicalization,long-distanceSFisfedbytheavailability ofsuccessivemovementtophaseedges;however,oncethatmovementputstheDPinthe samedomainastheSFlandingsite,itmovesbythesamemechanisms—whateverthose mightbe—thatderive‘ordinary’SF,explainingwhyasubjectgapisneeded. Considerhowsuccessive-cyclicmovementworksinaphase-basedtheory.Following Chomsky(2001),phaseheadscangenerallybeendowedwithoptional[EPP]featuresthat triggermovementtotheiredges(thoughseeSigurDsson2010foramorerefinedviewof EPPeffects).IfaDPistobetopicalized,itmovestothevPedgeinthisway.However, the DP cannot stay in place: it must find an appropriate landing site. In Scandinavian languages, movement of an object DP to the vP edge can feed Object Shift (OS) (Holmberg1999);inEnglish,thismovementcannotfeedOS,butmayfeedA(cid:48)-movement 8 to the CP domain (Chomsky 2001:26). Importantly, the [EPP]-feature responsible for successive-cyclic movement is not directly associated with the information-structural effectsassociatedwithOS,topicalization,wh-movement,orthelike.Thoseeffectsmust arise elsewhere in the derivation (but see López (2009), where information-structural effectsariseaspartofaphase-edgefeature-checkingoperation). Returning to SF, the generally held view is that any constituent close enough to the landingsiteisabletomovetothatposition(Holmberg2000).SinceDPshaveageneral wayofmovingsuccessive-cyclicallytoeachphaseedge,wenowexpectthatoncethey reachtheedgeofthephaseinsideaclausecontaininganavailableSFlandingsite,they shouldbeabletomovetothatsite.(20)and(21)showthatthisexpectationisborneout. TheDPmovestotheedgeoftheembeddedvP,fromtheretotheedgeoftheCPcontaining it,andthentotheedgeofthematrixvP(forreyna‘try’).Itcannotstayinthatposition,of course;itmustfindanappropriatelandingsite,whichinthiscaseistheSFposition.The presentaccount,then,predictsthatelementsthatcanmovesuccessive-cyclicallycan,in principle,undergoSFacrossclauseboundaries. Thepresentaccountalsoallowsforanexplanationofthefact,mentionedabove,that someraisingandcontrolverbsdoallowSF,aslongasthecomplementizeraðisomitted. Thisisillustratedin(22)–(23)(fromThráinsson1993:194–195).9 (22) Sásem reyndiaðlyftasteininumgafstupp. hewhotried to lift the.stone gave up (23) ?Sásem lyfta reyndi(*að)lyftasteininumgafstupp. hewholift.INFtried (*to) lift the.stone gave up ‘Theonewhotriedtoliftthestonegaveup.’ (23)seemspuzzlingatfirstglance,becauseitinvolvesSF,despitethefactthatitisacontrol construction,whichmeansthatthecontrolverbreyna‘try’hasavPofitsown.Assuming thatmuch,theoptionavailabletothepresenttheoryisthatin(23),itistheembeddedverb thatprojectsnophase.Thatis,itisarestructuringconstruction(Wurmbrand1998;Legate 2012). In the theory in Wurmbrand (1998), such cases involve no null PRO, but rather 9 involvecomplexpredicatesbuiltofstackedVPswithoneexternalargument.10 (24) [ EXT-ARGv[ try[ liftthestone]]] vP VP VP Assuming Icelandic að is higher than VP (standardly thought to be in C for control infinitives (SigurDsson 2008:407)), our analysis explains the absence of að: SF is only possiblewhenthecontrolverbtakesaVP-sizedcomplement,whichrulesoutað.11 Further, independent support for the restructuring analysis can be found when the embeddedverbselectsforanon-nominativesubject.Considerthecontrastin(25). (25) Context: A critic wants to write a negative review about a play, because the playwrightisherrival.Shegoestotheplaywithaverynegativeattitude. Gagnrýnandinnsem {?reyndiaðmislíka/*mislíkareyndi}leikritið the.critic who{ tried to dislike / dislike tried }the.play skemmtisér samtsemáðurvel. enjoyed REFLregardless well ‘Thecriticwhotriedtodisliketheplayenjoyedherselfregardless.’ The verb mislíka ‘dislike’ demands a dative subject, a requirement that can be met in controlconstructionsbyassigningdativecasetoPRO(Andrews1976;SigurDsson1991, 2008).SFisonlypossible,however,witharestructuringstructurelike(24),inwhichthere isnoPRO,makingitimpossibletomeettheargument-structuralrequirementsofmislíka ‘dislike’andreyna‘try’atthesametime;SFisthusimpossiblewithsuchverbs.12 In general, then, we conclude that in order for a control verb to allow SF of its verbal complement, it must involve restructuring in the sense of Wurmbrand (1998), with no PRO, and no phasal vP layer. The matrix and embedded verb behave as one predicate, and SF may cross the one vP layer projected by that predicate in a manner similartothatdiscussedforparticlesabove.If,however,thefrontingcategorymaymove successive-cyclicallyacrossphaseedges,asDPscan,thenclause-boundednessdoesnot hold in this form. The generalization that SF of verbs and particles is clause-bounded reducestotheclaimthatverbsandparticlesundergoingSFmayonlyleavethephasethat theyproject. 10