The Short Life CycleofExternal Arguments inGermanPassiveDerivations Gereon Müller,UniversitätLeipzig Abstract Inthisarticle,Ipursuetwomaingoals. First,Iargueforanewempiricalgeneral- ization: AnexternalargumentinGermanpassiveconstructions (DP )isaccessible ext from positions below it but inaccessible from positions above it. And second, I present anewtheoryofpassivization fromwhichthisgeneralization canbederived: I suggest that an elementary operation Remove should be postulated in phase-based minimalistsyntaxthatisthecompletemirrorimageofMergeinthatittriggersstruc- ture removal rather than structure building, and that obeys exactly the same restric- tions (with respect to triggers, strict cyclicity, etc.). Remove provides a principled approach to conflicting structure assignment and reanalysis in general, with short lifecycleeffects(derivable fromstrictcyclicity) atitscore. Keywords: Merge, StrictCycle Condition,passive,minimality,Agree(binding,control),reanalysis The Short Life CycleofExternal Arguments inGermanPassiveDerivations 1. Introduction In this article, I pursue two main goals. First, I argue for a new empirical generalization: An external argument in German passive constructions is accessible from positions be- low it but inaccessible from positions above it. And second, I present a new theoretical approach to passivization from which this generalization can be derived: I suggest that an elementary operation Remove should be postulated in phase-based minimalist syntax thatisthecompletemirrorimageofMergeinthatittriggersstructureremovalratherthan structure building, and that obeys exactly the same restrictions (with respect to triggers, strict cyclicity, etc.). I proceed as follows. In section 2, I show that there is conflicting evidenceconcerning thepresence ofan external argument DP in passiveconstructionsin German, and proposeto resolvethisconflict by postulatingtheAccessibilityGeneraliza- tion according to which the external argument is accessible from items that are lower in the structure and inaccessible from items that are higher in the structure. In section 3, I argue that from a conceptual point of view, there is every reason to postulate an opera- tion of structure-removal (Remove) as a counterpart to an operation of structure-building (Merge). Insection4,Iputthetwothingstogetherand showthatanapproach toGerman passive constructions in terms of structure removal accounts for the Accessibility Gen- eralization without further ado, by correctly predicting (via strict cyclicity) a short life cycle ofexternal arguments. Afterthis, section5 addresses thequestionof howvariation in the area of passivization can be accounted for in the new model. Section 6 draws a conclusionand presentsageneraloutlook. Inanappendix,Idiscussattemptstomaintain strict accessibilityorstrictinaccessibilityin thelightof theempiricalevidencepresented in section2. 2. Accessibility 2.1. Downward Accessibility Approaches to passivization differ with respect to the question of whether an external argument DP (DP ) is syntactically accessible or not. Over the last decades, some evi- ext dence has been presented that DP is indeed present in the syntax in passive construc- ext tions, and can be accessed by other operations; cf., e.g., Chomsky (1957), Perlmutter & Postal (1983), Roberts (1987), Baker, Johnson & Roberts (1989), Sternefeld (1995), Stechow (1998), Collins (2005), Harley (2013), Merchant (2013), and Georgi (2014b)).1 1 ForapproachesinwhichDP isnotsyntacticallyrepresented,seeHöhle(1978),Chomsky(1981),Bres- ext nan(1982),Kiss(1992),Wunderlich(1993),Müller,St. (2007)andKiparsky(2013),amongmanyothers; andforapproacheswhereDP issyntacticallyrepresentedbutnotaccessible, see Bach(1980),Keenan ext 1 A first, well-known argument for the syntactic accessibility of DP is that DP can ext ext control intoadverbialclauses;cf. thecaseofpurposeclauses in(1-ab). (1) a. Das SchiffwurdeDP versenkt[ um PRO dieVersicherungzu ext1 CP 1 the ship was sunk in order theinsurance to kassieren ]] collect “The shipwas sunkin ordertocollect theinsurance.” b. Der Reifen wurdeDP aufgepumpt[ um PRO dieFahrt ext1 CP 1 the tire was inflated in order thejourney fortzusetzen ] to continue “The tirewas inflated inorderto continuethejourney.” Second, DP can controlintosubject-orientedsecondary predicates,as in (2-abcd). ext (2) a. DieDaten wurden DP [ PRO nackt ]analysiert ext1 AP 1 the data were naked analyzed “The datawere analyzed bysomeonewho was naked.” b. Das HandoutwurdeDP [ PRO übermüdet]verfasst ext1 AP 1 the handout was tired written “The handoutwas writtenby someonewhowas tired.” c. Es wurde[ PRO absichtlich]einFehler gemacht AP 1 it was deliberately a mistakemade “Someonedeliberately madeamistake.” d. Dort wird[ PRO freiwillig] gearbeitet AP 1 there is voluntarilyworked “People worktherevoluntarily.” Third, control by DP into a regular complement infinitive is also possible; see (3-ab) ext (with impersonal passives based on a transitive control verb and a ditransitive subject control verb,respectively).2 (1980),Stechow(1987;1992),Bruening(2013;2014),Schäfer(2012b),Alexiadou&Doron(2013),Hole (2014),andAlexiadou,Anagnastopoulou&Schäfer(2015). 2 As argued by van Urk (2013), examples like (3-b) do not violate Visser’s generalization because this generalization should be taken to state that control by an implicit subject in the passive (DP ) into a ext complementinfinitiveisimpossibleifanovertDPagreeswithT;thisisnotthecaseintheexamplesin(3), whichbothexhibitimpersonalpassives(dativecaseisnotabsorbedbytheregularpassiveauxiliarywerden, andagreementisonlypossiblewithnominativeargumentsinGerman). Inlinewiththis, anexamplelike (i),wheredativecaseisabsorbedbythemarkedpassiveauxiliarykriegenandtheremainingovertargument (sie(‘she’))agreeswithT,iscorrectlypredictedtobeimpossibleunderVisser’sgeneralization. (i) *Gestern kriegteDP [ sie] versprochen[ PRO dasZimmerzureinigen] ext1 DP2 CP 1/2 yesterdaygot she promised the room to clean nom “Yesterdayshewaspromisedbysomeonethat(s)hewouldcleantheroom.” 2 (3) a. Es wurdeDP versucht[ PRO zu schlafen ] ext1 CP 1 it was tried to sleep “One triedto sleep.” b. Gestern wurdeDP [ ihr] versprochen[ PRO das Zimmerzu ext1 DP2 CP 1/∗2 yesterday was her promised the room to dat reinigen ] clean “Yesterday shewas promisedbysomeonethat (s)hewouldclean theroom.” Fourth, reflexive pronouns and reciprocal pronouns in German passiveconstructions can satisfy Principle A, which suggests the presence of an accessible co-indexed subject DP . ext (4) a. HierwurdeDP sich nichtgeprügelt ext1 1 here was REFL not hit “There wereno rowshere.” b. Es wurdeDP einander gedankt ext1 1 it was each otherthanked “People thankedeach other.” Fifth,DP cannoteasilybeinterpretedascoreferentialwithapropernameobject,which ext followsas aPrincipleC effect ifDP issyntacticallyaccessible;cf. (5-ab).3 ext (5) a. *Gestern wurdeDP Fritz eingeladen ext1 1 yesterday was Fritz invited “Yesterday, Fritzinvitedhimself.” b. ?*Gestern wurdeDP Fritz [ vonsich ] eingeladen ext1 1 PP 1 yesterday was Fritz by himselfinvited “Yesterday, Fritzinvitedhimself.” Thus, data such as those in (1)-(5) indicate that DP is present in the syntax. Note, ext however,thatalltheevidencepresentedsofarconcernsmateriallowerinthestructure,in domainsc-commandedbyDP . Therefore,weendupwiththefollowinggeneralization. ext (6) Downward AccessibilityGeneralization: The external argument in passive constructions (DP ) is accessible for items that ext itc-commands. 3 Baker, Johnson & Roberts (1989) analyze this as a strong crossover effect in English; however, since Germandoesnothaveobligatoryfrontingtosubjectposition(cf.Grewendorf(1989),Haider(2010),among manyothers),strongcrossovercannotsolelyberesponsiblefortheillformednessofthesentencesin(5). 3 2.2. UpwardInaccessibility In contrast, a question that does not seem to have been as widely pursued is whether the external argument in passive constructions is also accessible for items higher up in the structure. Closer inspection reveals that this is not the case. Note first that DP can ext never be bound by a quantified item in a higher clause (see Alexiadou et al. (2015)); this isshownforimpersonalpassives,wherenonominativeargumentremainsafterpassiviza- tion, in (7-a).4 Such bindingis unproblematicif theexternal argument is resumed as part ofa by-phrase, asin (7-b). (7) a. *Kein Student gibtzu [ dassDP schlechtgearbeitet wurde] 1 CP ext1 no student admits that badly worked was “No studentadmitsthathedidnotwork well.” b. Kein Student gibtzu [ dassDP schlecht[ von ihm ]gearbeitet 1 CP ext1 PP 1 no student admits that badly by him well wurde] worked was “No studentadmitsthathedidnotwork well.” In the same way, personal passives disallow binding of DP by a quantified item in a ext higherclause; cf. (8-a)vs. (8-b). (8) a. *Erhat den meistenLehrern gesagt [ dass DP derMaria Bücher 1 CP ext1 hehas the most teachers said that the Maria books dat dat nom geschenkt werden sollen] given are should “He toldmostteachers thattheyshouldgivebookstoMaria.” b. Erhat den meistenLehrern gesagt [ dass DP derMaria [ von 1 CP ext1 PP hehas the most teachers said that the Maria by dat dat ihnen ] Bücher geschenktwerden sollen] 1 themselvesbooks given are should nom “He toldmostteachers thattheyshouldgivebookstoMaria.” Third, as observed in Stechow & Sternefeld (1988, 447-451), Wunderlich (1989), Ste- chow (1989), and Haider (2010, 293), control infinitives must have an accessible subject argument. DP in passiveclauses can evidentlynot satisfy this condition; this is shown ext for controlintoimpersonalpassivesin (9). (9) a. *Erversucht[ DP gearbeitet zu werden ] CP ext hetries worked to be “He triesto ensurethatwork isbeingdone.” 4 Hereandhenceforth,DP inasyntacticrepresentationsignalsthatDP seemstobeinaccessible. ext ext 4 b. *weil [ bald DP geschlafenzu werden ]gewünschtwird CP ext because soon slept to be wished is “because someonewishesthatsleep comesoverpeople.” Fourth, non-overt material can in principle satisfy criterial movementconstraints in Ger- man. This is standardly assumed for extraction from embedded verb-second clauses as in (10-a) (where the intermediate movement step to SpecC can satisfy a verb-second C projection’s specifier requirement), and it also holds for topic drop constructions as in (10-b). (10) a. Wer glaubstdu [ –hat Recht ] ? 1 CP who think you is right nom “Who do youthinkisright?” b. – habeichschon gesehenheute (her) have I already seen today “I havealready seen hertoday.” However,DP canneversatisfyacriterialmovementconstraintinpassiveconstructions; ext cf. (11-a)vs. (11-bc)(with aby-phrasePP andan expletiveinSpecC, respectively). (11) a. *Ich denke[ DP istgut gearbeitet worden] CP ext1 I think is wellworked been “I thinkthat peopleworked well.” b. Ich denke[ [ vonihr ]ist gut gearbeitet worden ] CP PP 1 I think by her is well workedbeen “I thinkthat sheworked well.” c. Ich denke[ es istDP gut gearbeitet worden] CP ext1 I think it is wellworked been “I thinkthat peopleworked well.” Thefifthobservationconcernstheabsenceofminimalityeffectsinpassiveconstructions. As noted by Collins (2005), if DP is structurally represented in passive constructions, ext it is a priori unclear why movementof the internal argument to subject position can take place, given that movement obeys minimality: DP in Specv is invariably closer to ext SpecT than DP (i.e., an internal argument DP) in VP. The relevant derivation is given int in (12). (12) [TP [T′ T[vP DPext [v′ v [VP V DPint ]]]]] x A derivation as in (12) should be expected to be ill formed. However, DP moves to int SpecT in Englishpassiveconstructions, which suggests that DP is in fact not syntacti- ext cally accessibleand can thusnotintervene;see(13). 5 (13) [TP John2 was [vP DPext,1 [v′ v [VP killedt2 ]]]] DP canalsomovetoSpecTinGermanpassiveconstructions,whichthenalsoindicates int that that DP is not syntactically accessible. However, German differs from English in ext that movement to subject position in general is optional (see footnote 3), and that much TP-internal word order variation can be traced back to scrambling. A test for optional movement to SpecT in German is devised in Müller (2001): First, only a nominative subject argument DP can precede unstressed pronouns and at the same time follow C elements (cf. (14-a) vs. (14-b)); object DPs cannot do so (cf. (14-cd)). This strongly suggests a designated position in which only a nominative DP can show up: SpecT. ext On this view, unstressed pronouns move to a domain that precedes the landing sites for scrambling (viz., specifiers of vP or VP) but crucially follows SpecT. This rules out both (14-c) (where DP is scrambled to a position preceding the unstressed pronoun) and dat (14-d) (where DP is moved to a domain that cannot be reached by scrambling, viz. dat TP). (14) a. dass es [ derFritz demKarl t gegeben]hat 3 vP 1 3 2 that it the Fritz the Karl given has acc nom dat “that Fritzgaveitto Karl.” b. dass derFritz es [ t demKarl t gegeben]hat 1 3 vP 1 3 2 that the Fritz it the Karl given has nom acc dat “that Fritzgaveitto Karl.” c. *dass derFritz dem Karl es [ t t t gegeben]hat 1 3 2 vP 1 3 2 that the Fritz the Karl it given has nom dat acc “that Fritzgaveitto Karl.” d. *dass demKarl derFritz es [ t t t gegeben]hat 3 1 2 vP 1 3 2 that the Karl the Fritz it given has dat nom acc “that Fritzgaveitto Karl.” Given this state of affairs, it is clear that optional movement of a nominative DP in int German passive constructions has taken place in (15-a) (but not in (15-b)); and such movementisevidentlynotblockedby an interveningDP .5 ext 5 Notethatthisreasoningrequirestwofurtherassumptions.First,movementoperationslikescramblingand unstressedpronounfrontinghavesomewaytocircumventminimalityeffects,unlikeclassicalA-movement toSpecT;thishaslongbeenknownandholdstrueoftheseoperationsalmostbydefinition. Onepossible explanation for this is that the landing sites of scrambling and pronounfronting show evidence of being A-bar rather than A-positions(see Müller (1995)). Second, A-movementof DP in (14-b)or of DP ext int in(15-a)toSpecTacrossafrontedpronoun(orascrambledDP)alsocanavoidminimalityeffects,again incontrasttoA-movementthatwouldcrossanon-overtDP initsbaseposition. Again,itwouldseem ext reasonable to account for this by assuming that A-bar specifiers (scrambled DPs, fronted pronouns) can neverinduceminimalityeffectswithA-movement,incontrasttoitemsintheirbasepositions. 6 (15) a. dass derKarl2 ihr3 [vP DPext,1 [v′ [VP t3 t2 vorgestellt]v ]]wurde that the Karl her introduced was nom dat “that Karl was introducedtoher.” b. dass ihr3 [vP DPext,1 [v′ [VP t3 derKarl2 vorgestellt]v ]]wurde that her the Karl introduced was dat nom “that Karl was introducedtoher.” A sixthobservationisthatDP doesnotblock anaphoricbindingfrom aboveinpassive ext constructions,incontrasttootherexternalargumentsinGerman,whichactasinterveners (see Pitteroff (2014)). This is shown by the active/passive pair in an AcI construction with lassen (‘let’) in (16). In (16-a), the subject DP die Diener (‘the servants’) blocks a satisfaction of Principle A by sich with the matrix subject DP as antecedent; in contrast, in the lassen-passiveconstructionin (16-b), DP does not preclude sichfrom satisfying ext Principle A with the matrix subject DP as antecedent; in this derivation, the reflexive bears index 1.6 Alternatively, the reflexive can also take the embedded subject as an antecedent, signalled here by index 2. Thus, the two possible readings of (16-b) show both inaccessibility of DP (in the case of index 1) and accessibility of DP (in the ext ext case ofindex2,based onthesamereasoningas in(4)) inoneexample.7 (16) a. Der König lässt[ dieDiener sich rasieren ]} 1 act 2 ∗1/2 the king lets theservants REFL shave nom “Thekinglets theservantsshavethemselves.” b. Der König lässt[ DP sich rasieren ] 1 pass ext2 1/2 the king lets REFL shave nom “Thekingletspeopleshavethemselves.”/“Thekingletssomeoneshavehim.” Taken together, weend upwiththegeneralizationin(17). (17) Upward AccessibilityGeneralization: The external argument in passiveconstructions (DP ) is not accessible for items ext 6 NotethatGermanAcIconstructionssometimespermitlong-distancereflexivization,butthiseffectonly showsupwithPPs; cf. Reis(1976),Grewendorf(1983),Gunkel(2003),Barnickel(2014). Also,binding bythematrixsubjectin(16-b)cannotbeduetoraisingofthereflexivepronounsich tothematrixclause becausesichcanparticipateinVPtopicalization.Finally,lassen-passivesarespecialinsofaraspassivization isnotaccompaniedbyanymorphologicalreflex;however,thestatusoftheconstructionasaregularpassive isuncontroversial(amongotherthings,aby-phraseispossible,andlexicalrestrictionsonpassivizationare identicaltothoseactiveinstandardpassives). 7 Asamatteroffact,thereisathirdreadingofthestringin(16-b),irrelevantinthepresentcontext,where lassendoesnothaveacausativeorpermissiveinterpretation(aspresupposedsofar)andsichisnotanobject oftheembeddedverbrasieren;rather,sichlassenactsasamodalpassiveauxiliary.Onthisreading,(16-b) would mean“Itis possible to shave the king”. See Höhle(1978)fora comprehensivedescriptionof this construction. 7 thatit doesnotc-command. Combining the two generalizations in (6) and (17), the Accessibility Generalization in (18)emerges. (18) AccessibilityGeneralization(preliminaryversion): DP in passive constructions is accessible for items that it c-commands and in- ext accessibleforitemsthatitdoes notc-command. While I take (18) to be descriptively correct for the most part, the following subsection showsthatthereisafurtherempiricalphenomenonthatwillleadtoaslightmodification. 2.3. ARefinement: QuantificationalVariabilityEffects Consider quantificational variability effects in English, as they arise with indefinites (see Heim (1982), Diesing (1992)) and embedded wh-clauses (see Berman (1991)); cf. (19-ab), respectively. (19) a. A cat is usuallysmart ∼= Mostcatsare smart b. Johnpartly rememberswho cheated A standard assumption that I will adopt here (though see Hinterwimmer (2005) for a qualification) is that an indefinite DP (which can also be a wh-phrase) denotes an open sentence with a free individual variable that can be unselectively bound by an adverb of quantification; in line with this, existential binding in general comes about as a default operation. As observed in Alexiadou & Müller (2015), the external argument in German passive constructions is also subject to quantificational variability effects. This is shown bythedatain(20-ab),withtheadverbsofquantificationgrößtenteils(‘forthemostpart’) and zum Teil(‘partly’), and theinterpretationsas specified by thefree translations. (20) a. Es wurdegrößtenteils DP geschlafen beim Vortrag ext it was forthemostpart slept at thetalk “Mostpeoplesleptduringthetalk.” b. Dann wurdederSprecher zumTeil DP ausgebuht ext then was the speaker partly booed “Then apropersubsetofpeoplebooedthespeaker.” For binding by the adverb of quantification to be possible in (20-ab), DP needs to be ext c-commanded by it, and thus must be accessible for some item outside its c-command domain after all. This requires a modification of the Accessibility Generalization. As it turns out, there is evidence that the adverbs of quantification in question are properly included in the vP; thus, they can participate in vP topicalization, unlike, e.g., sentence 8 adverbialswhich aremerged outsideofvP; see(21-a)vs. (21-b). (21) a. [ Größtenteils DP geschlafen ]wurdebeim Vortrag ext for themostpart slept was at thetalk “Mostpeoplesleptduringthetalk.” b. *[ Wahrscheinlich DP geschlafen ]wurdebeim Vortrag ext probably slept was at thetalk “People probablysleptduringthetalk.” Note also that DP cannot be bound by an adverb of quanfication outside the minimal ext clause; see (22-ab) (where only a reading is available where the adverb quantifies over timespans).8 (22) a. Es war größtenteils so[ dassDP geschlafenwurdebeim Vortrag ] CP ext it was forthemostpart so that slept was at thetalk “*Most peopleslept during the talk.”/“Formost of thetimepeople slept dur- ingthetalk.” b. Es war zumTeil so[ dass derSprecher DP ausgebuhtwurde] CP ext it was partly so that the speaker booed was “*A propersubsetofpeoplebooedthespeaker.”/“Atsomepointsthespeaker was booed.” ThefactthatDP inGermanpassiveconstructionsissubjecttoquantificationalvariabil- ext ity effects can thenbe accomodatedby a minimalrefinement of(18), withtheconcept of c-command replaced by theslightlymoreliberalnotionofm-command. (23) AccessibilityGeneralization(final version): DP in passive constructions is accessible for items that it m-commands and ext inaccessibleforitemsthat itdoesnot m-command. If the Accessibility Generalization in (23) is correct, it implies that it is not possible to maintain either strict syntactic accessibility or strict syntactic inaccessibility of DP in ext German passive constructions. For the time being, I will abstract away from possible attempts to neverthelessmaintain existingapproaches where DP either is always, or is ext never, accessible in the syntax, in view of the empirical evidence presented so far; I will addressthisissueintheAppendix(andtheconclusionwillbenegative). Beyondthat,the Accessibility Generalization in (23) raises the question why it should hold. While there are several ways in which (23) might in principle be derived, the null hypothesis would clearly seem to be that accessibility is simply correlated with existence: Where DP ext 8 Thingsaredifferentwithovertindefinites,whichcanbenon-locallyboundbyanadverbofquantification; cf. Heim(1982). 9