Approaching the Edge in Syntactic Derivation

Transcription

1 Approaching he edge 1 Approaching he Edge in Synacic Deriaion Phil Branigan Memorial Uniersiy July 2104 Absrac. Consideraion of he problemaic relaionship beween he No Tampering Condiion (NTC), and ucking in and feaure inheriance consrucions proides reasons o replace he narrow NTC wih wo general consrains on merge: Approach he Roo and Approach he Probe. Ineracion of hese consrains wih he labelling algorihm LA proides a basis for ucking in wh-moemen and for ECM, boh beliee and wager-classes. The model proides a principled mechanism for word order and heigh reersals in applicaie and causaie sruures, which are shown o make accurae predicions wih French episemic daie srucures, Albanian applicaie passies, Lubukusu causaies, and French causaies. 1. Ensuring he cyclic applicaion of Merge The idea ha ransformaion operaions mus obsere sric cycliy has been a mainsay of generaie grammar since Chomsky (1965). This general claim has been applied paricularly rigorously in recen minimalis analysis, and is embodied in he Exension Condiion (Chomsky, 1995) and, more recenly, he No Tampering Condiion 1 (NTC) (Chomsky, 2007, 2008): A naural requiremen for efficien compuaion is a no ampering condiion NTC: Merge of X and Y leaes he wo [synacic objecs] unchanged. If so, hen Merge of X and Y can be aken o yield he se X, Y, he simples possibiliy worh considering. (Chomsky, 2008) 1 Waanabe (1995) anicipaes he NTC by demonsraing ha he Exension Condiion can be deried if we require ha moemen operaions should no disrup he se of c-command relaions already buil ino a phrase marker. Collins and Sabler (2011) show ha he NTC and he Exension Condiion are no fully equialen, since he NTC would be saisfied by he formaion of n-ary (> 2) branching srucures a he roo, while he Exension Condiion would no.

2 2 Approaching he edge Gien he NTC, Chomsky coninues, he characerisaion of synacic moemen can be minimal, perhaps een perfec. The head of he roo may include an edge feaure, which permis he roo o undergo a Merge operaion. (Languages may ary, apparenly, in wheher an edge feaure remains acie pas he firs Merge operaion.) As such, when he semanics require ha moemen should occur, any phrase inernal or exernal may Merge wih he roo in he manner permied by he NTC. (Moemen which falls under he EPP is slighly differen, and no deried so direcly.) Thus, for example, wh-moemen occurs because C has an edge feaure, and herefore a wh-phrase may Merge as a specifier o he roo CP. In shor, he NTC plays wo roles: one empirical, and one meaheoreical. I precludes counercyclic operaions which are widely acceped as impossible, and simplifies he formulaion of he basic operaion of (narrow) synacic moemen. Boh are eidenly imperaie elemens of any ineresing model. Alhough he inuiie basis for he NTC is quie clear, he empirical grounds for i are less so. The NTC is immediaely undermined, in fac, by wo apparenly widespread phenomena: uckingin operaions 2 and he sor of couner-cyclic moemen which Chomsky (2008) and Richards (2007) rea in erms of feaure inheriance. The former phenomenon is well esablished in Richard s (1997) sudy of Bulgarian muliple wh-moemen (and oher similar consrucions). (1) Koj kogo ižda? who whom saw Who saw whom? (Bulgarian) As Richards shows, he deriaion of quesions like (1) inoles wo wh-moemen operaions. The firs one displaces koj o make i a specifier for CP. The second hen moes kogo and merges i as a lower specifier in he same CP. The laer operaion clearly iolaes he NTC, a leas if we inerpre he NTC wih is obious inuiie sense. 2 Collins and Sabler (2011) also noe ha ucking-in is precluded under heir formalisaion of Merge and he NTC.

3 Approaching he edge 3 The second class of problemaic deriaions emerges from Posal s (1974) raising o objec cases, such as (2). (2) We beliee John sincerely o be inelligen. As Posal obsered, he orienaion of he aderb sincerely in such examples is oward he erb in he marix clause, despie is posiion o he righ of he embedded subjec John. This word order herefore seems o show ha John has raised ou of is own clause ino he marix erb phrase. Lasnik (2001)) offers a minimalis accoun of his ype of daa in which he subjec of he infiniial complemen undergoes NP-moemen o a specifier posiion in he marix P. (More accuraely, Lasnik proposes he subjec raises o an AgrP projecion which ranslaes prey direcly ino P in more recenly refined models, bu for he purposes of his discussion, he difference does no seem o maer much.) For Lasnik, his necessiaes an EPP feaure in (or AgrP) which appears only when assigns accusaie Case. Saisfacion of he EPP feaure hen gies rise o he marix P srucure in (3). (3) P D we P beliee John Ad sincerely TP TP o be inelligen

4 4 Approaching he edge As Lasnik obseres, his analysis works only wih some connecion beween accusaie Case feaures and he EPP feaure in. Wha is more, gien Pesesky s (1992) Agen/ECM correlaion (4), he connecion beween he properies of and he subjec of an ECM complemen is sill sronger han Lasnik suggess. (4) Agen/ECM Correlaion If α assigns he θ-role Agen, α Case-marks β only if α θ-marks β. This correlaion isolaes he general rule ha erbs wih an agenie subjec canno normally paricipae in ECM. Thus, while (5a) is accepable, (5b) is no. (5) a. Mary discoered Bill o hae read he book. b. *John affirmed Mary o hae enered he room. In curren erms, his paern mus concern he relaionship beween he head which is responsible for he subjec θ-role and he ECM subjec. Eidenly, may bear unalued φ/case feaures in an ECM conex only when i has specific hemaic properies. The properies of he lower, lexical are essenially irrelean o undersanding he correlaion. Chomsky (2008) poins ou ha Lasnik s analysis can be improed if i is he feaures of which are responsible for he displacemen of he downsairs subjec. Then iself need no coincidenally bear an EPP feaure in exacly he circumsances in which mus assign accusaie Case. Bu implemenaion of his idea requires a mechanism o enable a couner-cyclic Merge operaion. Chomsky proposes ha may ransfer is EPP feaure (and ohers) o, which hen aracs he lower subjec o become is own specifier. Bu his means ha moemen o he specifier posiion of P canno ake place unil afer is presen in he srucure. Again, he deriaion clearly iolaes he NTC, unless some arful loophole can be deised o eade his conclusion. Chomsky (2013) sreamlines he analysis sill more, bu he he NTC conflic remains. In his analysis, moemen of he ECM subjec is drien by he need o break symmery in he downsairs

5 Approaching he edge 5 P, which is necessary for he label assignmen operaion (LA) o ake place. Moemen of he subjec ou of is original P enables he P o be labeled, bu he subjec mus end up below he phase head which riggers he operaion. Again, he counercyclic naure of he he moemen operaion is inescapable. Of course, i is always possible ha a specific and elegan way o inerpre he NTC migh be be found which would render he deriaion of (4) unproblemaic. Bu absen any such inerpreaion, he conclusion o be drawn a presen seems o be ha he NTC is oo resricie. Wha is more, he wo consrucions which undermine he NTC do so in differen direcions. Tucking in operaions inole an NTC iolaion in which he srucure higher han he probe is ampered wih; inernal objec shif requires ampering wih somehing inside he firs projecion of he probe. And since, under Chomsky s approach, he mechanism which leads o moemen ino a lower specifier posiion is accessed only for A-moemen, he prospecs for unifying hese wo problemaic cases are no encouraging. In fac, under he mos ransparen inerpreaions, he NTC is iolaed een in conexs where a phrase simply moes o become he unique specifier of a probe, such as English wh-moemen. If moemen someimes akes place o saisfy a requiremen of a probe as irually any minimalis analysis will assume hen moemen will neer leae he original phrase marker conaining he probe unchanged, because he moie for moemen is o aler he properies of he probe iself. To sum up hese general obseraions, he NTC, which proides a srong concepual basis for cyclic applicaion of moemen operaions, appears empirically indefensible in Chomsky s formulaion, because i blocks a range of aesed operaions in which phrases raise o posiions lower han he roo, and because i ensures ha moemen of any sor canno be drien direcly by he feaures of he head of he roo phrase. On he oher hand, i remains clear ha many arieies of counercyclic operaions mus be excluded, so he NTC canno be simply abandoned. Nor would we wish o rejec Chomsky s concepion of moemen as a perfec operaion, howeer perfecion is operaionalised. 3 The bes 3 The Moe-α operaion of Goernmen-and-Binding heory is arguable a perfec heory of moemen, as well, because nohing beyond he fac of moemen is required of is characerisaion. The proocaion operaion in Branigan

6 6 Approaching he edge approach would appear o be o idenify a new mechanism which permis he necessary empirical coerage and which sill adheres o he minimalis imperaie ha general deriaional consrains should be iable 3rd facor effecs. I will show ha he NTC can be eliminaed in faor of wo general consrains, he ineracion of which limis he range of landing sies which are permied for phrasal moemen operaions. This adjusmen o minimalis heory will proide subsanial empirical payoff wihou any increase in he basic complexiy of he heory, and wihou inroducing inolerably ad hoc consrains ino he model as a whole. The firs of he wo principles is a slighly weaker form of he Exension Condiion: (6). (6) Approach he Roo (ApR) Merge as close o he roo as possible. For many cases, ApR is sufficien o ensure ha cycliciy is obsered in synacic deriaions, jus as does he NTC. English wh-moemen, for example, will always saisfy ApR by merging direcly wih he CP roo of a clause. And gien Chomsky s (2007; 2001; 2008) phase heoreic premises, where only he heads of phases can rigger moemen, ApR will be saisfied auomaically by phrasal moemen which arges he phasal roo (P, CP,, ec.). The difference bewee ApR and he NTC/Exension Condiion is he as close as possible escape clause in (6), which implies ha here are condiions under which he roo migh no be only only permissable merge sie. To undersand wha hese condiions will be, i is necessary o consider he second moemen consrain: (7). (7) Approach he Probe (ApP) Merge as close o he probe (=phase head) as possible. The definiion (7) simply implemens Richards (2001) idea ha a phrase mus moe as close as possible o he probe which aracs i. Richards does no proide a label for his laer consrain. (2010) is comparable in his respec.

7 Approaching he edge 7 In Richards (2001), he suggess ha he effec follows from he Shores Moe principle, bu he chooses no o isolae any paricular precise implemenaion of Shores Moe. And since he erm Shores Moe has been used o mean a number of hings in he lieraure, i is imporan o hae access o erminology o isolae he specific idea ha phrases land close o he probe. For Richards, as here, he immediae consequence of ApP is ha languages like Bulgarian will employ ucking-in srucures in muliple wh-moemen, as discussed aboe. Crucially, ApP mus consrain moemen operaions raher han resuling represenaions, because he effec of uckingin is always o leae he firs specifier in a more disan posiion from he head han i was before. Wha is more, ApP does no iself rigger moemen. I merely consrains he choice of landing sie when a higher or lower specifier posiion wihin he same phrase are aailable opions. Consider how ApR and ApP are ealuaed in such a case. For an example like ((1)), where he srucure immediaely afer wh-moemen of he subjec will be (8). (8) CP D C koj C TP T ižda kogo C hen riggers moemen of kogo in (8). Now compare wo rees represening poenial oucomes for he second wh-moemen operaion: (9a,b).

8 8 Approaching he edge (9) a. CP b. CP D C D C kogo D C koj D C koj C TP kogo C TP T T ižda ižda The firs ree saisfies ApR maximally by merging kogo a he roo, bu iolaes ApP in doing so, because kogo is merged furher from C han possible. The second ree, which represens he correc word order, does he reerse. ApP is maximally saisfied, and ApR is saisfied only as much as possible, because he merge sie for kogo is no he roo bu one node down. Wha he exisence of ucking-in deriaions ells us hen (as Richards recognizes) is ha hese wo general consrains are enforced o differen exens, because ApP is ranked more highly han ApR in ealuaing he accepabiliy of poenial merge sies. Criically, hough, boh ApR and ApP require ha he landing sie for wh-moemen will be a specifier posiion wihin CP, so ha no moemen which arges a merge sie wihin TP will sill be impossible. (Exernal merge operaions will be subjec only o ApR and no ApP, because hey do no inole a probe-goal relaionship a all.) Chomsky s (2013) labeling heory inroduces a new grammaical force ino he picure. The labeling/projecion algorihm (LA), which applies a he phase leel, deermines which of wo daugher caegories are o proide he label for he moher. While he merge operaion produces hierarchical srucure, i does no proide caegorial labels for he resuls. Bu caegorial labels are essenial for a leas wo grammaical phenomena: selecion (in he synacic deriaion), and inerface inerpreaion. According o Chomsky, LA assigns labels on he following bases: (10) LA a. When X 0 merges wih a phrase, X 0 is idenified as he label. b. When a caegory moes o a higher posiion, is siser is he label.

9 Approaching he edge 9 c. When agreemen maches Force feaures on wo sisers, Force is he label. These hree (deried) rules coer many, bu no all srucure ypes. Typical head-complemen pairs fall under (10a). Phasal P falls under (10b), because he exernal argumen always raises ou. And wh-moemen srucures fall under (10c), wih he wh-phrase and C sharing an inerrogaie Force feaure. English TP srucures, howeer, do no fall under any of he hree descripions in (10). TP merges wih he senence subjec, which neer seems o sere as he label. Bu TP is no a head, he subjec may remain in Spec-T, and no Force feaure is presen in he subjec. Chomsky considers wo possible reamens of his siuaion. The firs one (which he ulimaely, enaiely, abandons, bu which I will adop) is ha no label is projeced a all in his case. In oher words, he srucure for a complemen clause like (11a) may be (11b), wih no projecion of TP pas he firs one. 4 (11) a. We said ha Robera had sho he sheriff. b. C ha Robera T TP P had sho he sheriff As Chomsky obseres, such a srucure migh be accepable gien how subjec moemen is characerised under phase heory. If moemen of he subjec occurs only a he CP phase leel, hen a he poin a which C selecs and merges wih is complemen, he subjec is sill conained wihin P. So LA plays no par in he selecion of TP in any case. As for he role projecion of he TP label migh play in inerpreaion, Chomsky obseres ha he surface subjec has a kind of secondary agency, as we see in such familiar examples as PRO o seem o be inelligen is no as easy as you migh hink.. This appears o be a sraighforward 4 The CP caegory label for he embedded clause in (11a) will iself no be esablished unil he following phase is complee.

10 10 Approaching he edge maer of CI-inerpreaion of he subjec predicae phrase α [he srucure formed by merging he subjec wih TP]. Surface subjec has oher special properies (wha Luigi Rizzi calls an abouness propery), illusraed in arious ways; in null subjec languages by he fac ha silen pro can refer o he surface subjec (bu no he objec) of he preceding senence. Such consideraions sugges ha α receies is own inerpreaion and should be labeled in accord wih he general principle ha all SOs ha reach he inerfaces mus be labeled. Bu as Chomsky himself obseres in Chomsky (2008), semanic properies which are associaed wih moemen of he subjec can irually always be inerpreed as properies of he TP in which T has agreed wih he subjec. Moemen iself may be irrelean. Wha does his enail for our undersanding of cycliciy? If we accep ha C riggers moemen of he subjec, which merges wih TP wihou LA proiding a label for he resuling srucure, hen he merge sie of he subjec mus also follow from he ineracion of ApR and ApP (and LA). I assume ha C bears he unalued φ feaures which allow phase heads o agree and assign Case, and ha C s EPP propery ensures ha aluaion of hese feaures compels he subjec o moe. (Wih agenie subjecs, such moemen soles he labelling problem for P, bu C mus bear he unalued φ feaures in oher clauses, oo.) In oher words, C is a probe wih respec o ApP. Suppose ha C has no wh-phrase. Then here are wo poenial merge sies o consider: (12). (12) a. subjec C TP b. C subjec TP T erb T erb If he subjec merges wih he roo, as in (12a), hen i saisfies ApR and ApP fully. Bu he resuling srucure will no acquire a label a he nex phase. While he C-TP srucure can be idenified as CP, here is no basis for labeling he srucure comprised of he subjec and CP. Selecion of he embedded CP by a marix predicae will hen fail. And inerpreaion of his srucure a he semanic inerface will presumably be problemaic, as well, since he lack of a label should make

11 Approaching he edge 11 i impossible o idenify he illucuionary force for he embedded clause. In conras, he merge sie for he subjec in (12b) saisfies ApP fully bu does no fully saisfy ApR. (The subjec is he daugher of he siser of C in (12b), and he siser of he moher of C in (12a). These consiue equidisan merge sies.) Bu he lack of a label aboe he subjec in (12b) does no lead o selecion/inerface problems. The only conclusion o draw is ha he as close as possible escape clause in he formulaion of ApR allows for a merge sie a his disance from he roo when a labelling problem would force he deriaion o crash oherwise. Noice ha his reamen of cycliciy makes i possible o dispense wih perhaps he leas desirable componen of he Chomsky/Richards phase heory. In Chomsky (2008), moemen of he subjec akes place because C ransfers is EPP feaure o T, along wih he arious alued feaures which are o be inerpreed a he PF inerface. Bu he EPP propery is ery differen from feaures like ense or number. While he laer are clearly linguisic maerial wih PF subsance, he EPP is jus he name we use for he problem of explaining he fac ha subjecs moe. Saying ha he EPP propery is ransferred o anoher head is informaie o approximaely he same degree as claiming ha all sasquaches hae six oes is, i.e., no a all. In he approach aken here, while he EPP propery of C sill raises foundaional quesions, a leas hese are no compounded by preending ha he erm EPP refers o more han he abiliy of C o rigger moemen of is nominal agreemen arge. To round hings ou, le us also briefly consider he siuaion when C does acquire a specifier hrough wh-moemen. In English, where only one wh-phrase raises, a pre-wh-moemen srucure may be (13). (α and β are no caegorial labels here, bu simply exposiory node capions.) (13) α C β /0 Robera TP has sho whom

12 12 Approaching he edge whom in (13) mus raise o ensure a wh-quesion inerpreaion for he clause as a whole. If i merges wih β, hen he choice of merge sie saisfies ApP, bu no ApR. Wha is more, he lack of a wh-phrase as he daugher of he roo means ha he inerrogaie force feaure will no be seleced as he roo label, which will lead o inerpreie problems a he semanic inerface. If α is seleced as he merge sie, he appropriae label will be proided for he roo, and boh ApP and ApR are fully saisfied. Therefore α wins ou oer β, producing he srucure (14). (14) α whom C β /0 Robera TP has sho Reurning o he analysis of Posal s ECM examples like (15a), le us consider how he posiion of he lower subjec migh be ensured gien his reised model of cycliciy. The srucure of he marix P prior o labelling and moemen of John can be assumed o be (15b). (15) a. We beliee John sincerely o be inelligen. b. α D β we P beliee Ad sincerely P TP o John be inelligen

13 Approaching he edge 13 Wih Chomsky, we may idenify he marix as he probe riggering moemen. The se of poenial merge sies for he lower subjec is hen hree nodes in he marix clause: α, β, and P. Bu boh α and β can be ignored, because merge of he lower subjec wih α or β will ensure a srucure in which LA canno idenify as he source of a P label. The P merge sie opion saisfies ApP, and he degree o which ApR is no fully saisfied is olerable if here is no beer opion aailable, which here is no in his case. The lower subjec herefor merges wih P, producing (16). (16) D we γ beliee John Ad P P sincerely TP o be inelligen Since selecion of P by has already aken place a an earlier poin in he deriaion, he lack of a label for γ causes no problem, and he deriaion can now ulimaely conerge. In shor, nohing needs o be sipulaed o ensure he subjec in English ECM complemens will always Merge as a specifier of he marix P. And once again, he use of a feaure inheriance operaion o ransfer an EPP feaure is unnecessary for his paricular consrucion. 2. Wager-class ECM Posal (1974) obseres ha wager-class erbs are consisen wih ECM-like srucures only when he subjec of he downsairs infiniial clause undergoes wh-moemen: (17). (17) a. I consider Jenna o be a srong swimmer. b. *I wager Jenna o be a srong swimmer.

14 14 Approaching he edge c. Jenna, who I wager o be a srong swimmer. Pesesky s Agen/ECM correlaion does no apply when he subjec is exraced; wager cerainly has an agenie subjec, bu he subjec of he infiniie sill succeeds in acquiring Case somehow in (17c). Kayne (1981) and Rizzi (1982) show ha his same paern is expressed more generally in French and Ialian, respeciely. In oher words, while ECM is generally barred in hese languages, wh-moemen of he subjec of an infiniie resuls in ECM-like behaiour. (18) a. *J imagine Maurice saoir bien nager. I consider Maurice o know well swim I consider Maurice o be able o swim well. b. Maurice, qui j imagine saoir bien nager. Maurice who I consider o know well swim Maurice, who I consider o be able o swim well. One way o characerise he difference beween English and hese Romance languages is ha Pesesky s Agen/ECM correlaion idenifies an English-specific mechanism which allows Case o be assigned by o a non-argumen, bu his special mechanism is inacie in French and Ialian. In all hree languages, hough, wh-moemen of he subjec of an infiniial complemen gies i access o anoher Case source. In all hree languages, here is eidence from raising ha he infiniial clauses in quesion are a leas opionally defecie clauses, which lack he C phase head. (19) a. Jenna was wagered o be a srong swimmer. b. Maurice a éé imaginé saoir bien nager. Maurice has been imagined o know well swim Maurice was imagined o be able o swim well. In English, he unaccepabiliy of a conrol clause as he complemen o wager furher suggess ha a defecie TP is required in his conex, raher han opional.

15 Approaching he edge 15 (20) *I wager PRO o be able o finish his whole pie. The conras beween (17b) and (17c) hen mus reflec a difference in how he deriaion copes wih wo ery similar iniial srucures (21a) and (21b). (21) a. α β I P wager- b. α TP o Jenna be a srong swimmer β I P wager- TP o who be a srong swimmer While agenie canno assign Case feaures o a non-argumen, i mus sill be able o rigger moemen, presumably wih a defecie se of φ feaures. In boh (21a) and (21b), hen, he subjec of he infiniie will hae o merge wih a node ouside of TP: α, β, or P. In (21a), Jenna can only merge wih P, for all he reasons which apply when ECM is successful. If Jenna merges wih α or β, hen LA will be unable o assign he P label o α or β, wih unaccepable consequences for selecion and inerface inerpreaion. Bu who is differen. Een if who merges wih α or β, i will undergo a subsequen wh-moemen operaion when he nex phasal caegory is inroduced. Tha means ha who will no be considered when LA looks for a label for he nodes

16 16 Approaching he edge aboe, and P can be correcly assigned. Therefore LA does no preen α and β from sering as merge sies for who. Only ApR and ApP are relean, wih ApP haing he sronger influence on he choice of merge sies, as always. While ApR would prefer α, hen β, ApP will be saisfied by eiher β or P. The bes soluion hen is β, and (21b) hen gies rise o (22). (22) α I who β P wager- TP o be a srong swimmer Boh I and who mus be exraced in (22) o higher posiions in order o aoid inerfering wih LA, bu his is guaraneed in any case by he feaures of C when i is merged wih he res of his clause. How hen does he higher posiion of who in (22) allow Case o be assigned o he subjec of he infiniie. Clearly, can play no role, since who in (22) has already exied he c-command domain of he probe. Insead, here mus be a source for Case ouside P, which can affec a nominal a he edge of he P phase. The only conclusion which can be drawn is ha he higher C mus be able o assign Case o boh I and who. In oher words, English/French/Ialian hae he abiliy o assign muliple (nominaie) Cases, jus like Japanese does. In mos srucures, of course, non-subjecs will remain inside he P phase, where C canno access hem. Bu in his paricular ype of ECM-srucure, he subjec of a lower infiniie mus reach he accessible periphery of P o saisfy general principles, and hen i may collec Case feaures from C afer he rue subjec does so. Heay-NP shif produces much he same resuls as wh-moemen wih wager-class ECM (Posal, 1974; Boškoić, 1997).

17 Approaching he edge 17 (23) I would wager o be a srong swimmer eery child who has grown up in he Laurenians. Like wh-moemen, HNPS remoes a phrase from is posiion wihin P. As such, he analysis of he wh-moemen cases will exend direcly o accomodae daa like (23), een if he analysis of HNPS remains conenious. As long as he lower subjec can be remoed from he edge of he marix P, LA will no inerfere wih i aking up emporary residence here, and C can hen assign nominaie Case o i. In French, cliic pronouns may originae as he subjec of an infiniie, oo. Pollock (1984) proides he examples in (24). 5 (24) a.??marie, Pierre l a longemps crue aoir résolu ce problème. Marie, Pierre her has long belieed o.hae soled his problem Marie, Pierre has long belieed her o hae soled he problem. b.??jean, on l a longemps di aoir beaucoup aimé Marie. Jean one him has long said o.hae much loed Marie Jean, people hae long said ha he loed Marie a grea deal. As wih wh-moemen, examples like hese can be deried direcly from he fac ha cliic pronouns canno remain wihin P. As such, he inermediae srucure (25) will be permissable for (24a). 5 Pollock also proides persuasie argumens ha here is anoher source for Case for posposed indefinie subjecs of unaccusaie infiniies in French, which is acie in examples like (i). (i)?je croyais n êre enrés que deux linguises. I belieed no o.be enered bu wo linguiss I belieed only wo linguiss o hae enered. As he properies of his second ariey of Case assignmen are quie differen from hose which appear when he subjec is exraced ino he marix clause, his argumen does no affec he analysis deeloped here.

18 18 Approaching he edge (25) α on D la β P croire- TP aoir résolu ce problème Adjuncion of cliic la o T during he formaion of he nex phase will remoe i from he posiion i occupies in (25), allowing LA o idenify as he label for he projeced P. C will coninue o c-command la and can herefore assign Case o i, afer firs Case-marking he subjec on. 6 This approach o he special case of wager-class ECM srucures acually suggess a soluion o he ousanding quesion of he phasal saus of passie and unaccusaie Ps. The problem wih inegraing passies ino phase heory is ha he objec nominal in a passie erb phrase mus be accessible o he C probe, which agrees, assigns Case, and displaces he objec. On ha basis, i seems ha passie Ps mus no be phasal. Bu Legae (2003) finds eidence from inerface inerpreaions which indicaes ha all P ypes mus be phasal. This resul implies ha here mus be a mechanism which allows C o access objecs inside a phasal P. And his mechanism whaeer i is mus do so wihou undermining he foundaions of phase heory. The obious soluion is ha unaccusaie/passie erbs mus be able o displace heir objecs o he phase edge. The equally obious hich is ha objecs in he corresponding acie erb phrases mus no allow such objec displacemen, which would generae he wrong word order. Forunaely, he analysis of wager-class ECM srucures now offers an explanaion for he lack of moemen in acie erb phrases generally. If an acie, ransiie displaces any objec nominal, he merge sie mus remain inside he complemen of unless ha objec will undergo subsequen moemen a 6 Casual reference is ofen made o he le/la/les cliic series as accusaie Case forms, which would hen raise quesions abou how C migh assign accusaie Case. Bu i is clear, and amply documened in he lieraure, ha hese cliic pronouns are simply weak arians for he sronger lui/elle/eux Case-neural full pronoun se. As such, here is no reason o regard he cliics as exclusiely accusaie.

19 Approaching he edge 19 he nex phase leel. Consider he minimal conras in (26), where moemen of he objec o he edge of P is accepable in he passie case bu no in he corresponding acie. (26) a. The sheriff was [ P sho [ P ]] b. *I [ P he sheriff sho [ P ]] If P is always phasal, hen (26a) will be generaed if passie has unalued φ feaures o rigger moemen bu no assign Case. In he acie (26b), he Case/φ complex in mus also hen be permied o rigger moemen, bu he moemen will block LA from applying, since he sheriff remains in place as a siser o P. The general resul, hen, is ha may be permied o rigger moemen of nominal arges freely, because oher facors will filer he resuls o ensure ha his process does no oergenerae. 3. Argumen Inersion and he APP 3.1 Deriing inersion The proposiion ha moemen can arge a lower specifier posiion when necessary, and subjec o LA, ApR and ApP, makes a clear predicion in cerain complex siuaions. While he ucking-in operaion preseres he iniial relaie ordering of he displaced specifiers, moemen of muliple specifiers o a posiion below a probe should hae he opposie effec, and produce a mirror image relaie ordering. Consider a srucure (27), where F is a moemen-riggering (phasal) wih he specific marked propery ha F may be alued muliple imes, riggering moemen each ime. F is like Bulgarian C in his respec, bu unlike C in is abiliy o permanenly accomodae specifiers. (27) F GP G HP... α... β...

20 20 Approaching he edge Le α and β boh be phrasal caegories which mach he unalued feaures of he F probe. In his siuaion, F will displace α, he closer goal, and hen β in urn. When he deriaion mus inegrae α, i canno be merged as a specifier for F, so o saisfy ApP, i mus ake he nex bes opion and merge wih GP, generaing (28). (28) F α GP G HP... α... β... When β mus be merged, he same logic applies, bu as GP already has α as is siser, β mus eiher be ucked-in or merged ouside α. The Merge sie is now deermined as follows. LA preens β from merging wih he moher of F, so he only possibiliies are GP and he unlabeled moher of GP. Merge wih he moher of GP will esablish β as he daugher of he siser of F; merge wih GP will leae β as he daugher of he daugher of he siser of F. As he former is closer o F, ApP forces he higher merge sie for β. Thus, in his siuaion, (28) will gie rise o (29), in which he base ordering of α and β has been deriaionally inered. (29) F β α GP G HP... α... β... The feaure inheriance model, where an EPP propery is ransferred from a phase head o he head of is complemen, makes he opposie predicion. If we sar from he (27) ree, and if F hen ransfers is EPP feaures o G, so ha G may arac boh α and β, hen α should merge wih GP,

21 Approaching he edge 21 and hen β should uck in by merging direcly wih GP as well. The resuling srucure will hen be expeced o look like (30). (30) F α β GP G HP... α... β... As I now show, a range of consrucions in a number of languages confirm he predicions of he ApR/ApP/LA model, and falsify he EPP-inheriance approach. 3.2 Episemic daies The firs consrucion o consider is he French episemic daie consrucion documened by Ruwe (1982). Some examples (from Ruwe, ranslaions mine) appear in (31). (31) a. Je lui croyais une maîresse dans chaque por. I him/dat belieed a misress in eery por I belieed him o hae a misress in eery por. b. Je lui roue beaucoup de charme. I him/dat find los of charm. I find him o hae a lo of charm. c. Le médecin préoi une issue faale à cee maladie. he docor predics an oucome faal o his illness The docor predics ha his illness will hae a faal oucome. This consrucion has a number of inriguing characerisics. Mos immediaely, he consrucion proides a meaning of possession, in some sense, which unies he indirec objec and he direc objec, bu in a conex in which he erb iself is separaed from he possession relaionship. This can happen, as in he (31) examples, because he erb iself lacks any possessie sense inherenly, bu i can also occur wih erbs which normally impar a sense of a possessie relaion-

22 22 Approaching he edge ship, bu which become diorced from ha relaionship by he nuance of conex, as in he (32) examples. (32) a. On lui aribue beaucoup de bonnes forunes. one him/dat aribues los of good luck People suppose ha he has a lo of luck. b. Les Alliés prêaien à Napoléon l inenion de bare en reraie. he Allies loaned o Napoleon he inenion of bea in rerea The Allies supposed ha Napoleon had he inenion o bea a rerea. The same effec is acually marginally possible wih some English double objec erbs (Branigan, 1992), like (33). (33) a. We gran Brad a cerain naie charm. b. I gie him 10 years, no more. In (33a), alhough he erb gran can be used as a erb of giing, in his conex i akes on a purely episemic meaning ( I gran ha P ), and he sense ha Brad and his charm are in a possession relaionship comes from he synacic conex more han from he erb. Similarly, in (33b), gie means somehing more like predic and wha is prediced is he proposiion ha he person in quesion will hae 10 years of somehing. Wha is paricularly significan abou he French episemic daies is ha he possessor behaes as a subjec wih respec o binding condiions, as can be seen in (34). (34) a. Je leur crois du respec l un pour l aure. I hem/dat beliee of respec he one for he oher I beliee hem o hae respec for each oher. b. *Ils me croien du respec l un pour l aure. hey me/dat beliee of respec he one for he oher c. Je leur prêe de mauaises inenions l un eners l aure. I hem/dat lend of bad inenions he one owards he oher I suppose hem o hae bad inenions owards each oher.

23 Approaching he edge 23 d. *Ils me prêen de mauaises inenions l un eners l aure. hem me/dat lend of bad inenions he one owards he oher Under sandard assumpions, he fac ha he daie possessor can bind he reciprocal l un l aure shows ha he daie c-commands he accusaie phrase. The fac ha he possessor is he only accepable anecedan shows ha i is he only c-commanding aneceden wihin a cerain clause-like domain. In ha respec, hese daies behae much like he subjec of a small clause complemen: (35). (35) *Pierre e Paul croien Marie amoureuse l un de l aure. Pierre and Paul beliee Marie amourous he one of he oher Pierre and Paul beliee Marie in loe wih each oher. The binding properies of episemic daies lead inexorably o he conclusion ha he srucure of he erb phrase in hese examples mus be somehing like (36), where E migh be eiher an Applicaie head, or some oher less familiar caegory which can be seleced by he erbs in quesion. (36) P subjec P EP possessor E possessum EP eidenly consiues a binding domain. In he srucure (36), we should no expec A-moemen of he possessum o a posiion ouside P eer o be possible, because here are wo higher nominals which should inerene. Een if he subjec were remoed, as in a passie ersion of (31), he presence of he possessor should

24 24 Approaching he edge be sufficien o block A-moemen of he possessum. And his expecaion iniially seems o be confirmed in examples like (37). (37) a. *Une maîresse dans chaque por lui éai cru. a misress in eery por him/dat was belieed He was belieed o hae a misress in eery por. b. *Beaucoup de charme lui es roué (par ou le monde). los of charm him/dat is found by all he world He is found o hae los of charm (by eeryone). Bu Ruwe aribues o Paul Posal he obseraion ha he passies of some episemic daies are bad een when no moemen occurs. So (37a) is no improed if an expleie is used in he subjec posiion. Compare he saus of (38a) wih he grammaical resul of using an expleie in a passie formed from a regular double objec erb. (38) a. *Il lui es cru une maîresse dans chaque por. i him/dat is belieed a misress in eery por b. Il lui a éé prêé beaucoup de lires. i him/dat has been loaned los of books There hae been loaned los of books o him. Gien he saus of (38a), he ungrammaicaliy of he (37) examples need no be aribued o he A-moemen of he possessum. When passie srucures are formed from oher erbs (especially hose which hae as heir core meaning some noion of ransfer of possession), i urns ou ha he episemic daie possessums can undergo A-moemen. (39) a. Une issue faale es préue à cee maladie (par le médecin). an oucome faal is prediced o his illness by he docor This illness is prediced o hae a faal oucome. b. Beaucoup los de bonnes forunes lui son aribuées à or. of good luck him/dat are aribued by misake

25 Approaching he edge 25 He is hough o hae a lo of luck wrongfully. c.?une maîresse dans chaque por lui es souen imaginée. a misress in eery por him/dat is ofen imagined He is ofen imagined o hae a misress in eery por. We face a seemingly paradoxical siuaion. Gien he binding facs, he possessor mus be higher han he possessum. Gien he A-moemen facs, he possessum mus be he mos accessible goal for a T probe. The only way o resole he paradox is by allowing he deriaion o reerse he order of possessor and possessum midway. The possessum mus be able o raise pas he possessor before T eners he deriaion. Gien he srucure (36), he necessary inersion resuls if has he properies of F in (27). canno olerae he inroducion of a as siser o is moher in French, as i canno in English. Once again, we may aribue his paern o applicaion of LA o such srucures, where [ P ] has no unambiguous source for he label of he whole. We may accoun for he Case-marking of boh daie and accusaie objecs by supposing ha is able o Agree and assign srucural Case wice, as argued by Bobaljik and Branigan (2006). Assuming ha Case model, mus assign he marked (daie) Case firs, and he unmarked (accusaie) Case second. Person-Case Consrain effecs hen reflec feaure conflics locaed wihin a single acie probe :diransiies. 7 And jus as in English ECM (where assigns srucural Case only once), he φ/case feaures of rigger moemen of heir arge. Gien hese properies of in French episemic daie srucures, he srucure in ((36)) mus be ransformed ino (40) in order o saisfy boh ApP and he feaural requiremens of is componen pars. The binding relaions are eidenly esablished wihin EP before he probe assigns Case. 7 See also Adger and Harbour (2007) for deelopmen of his approach o he Person-Case Consrain.

26 26 Approaching he edge (40) subjec possessum j possessor i P EP i E j Afer (40) has been generaed, howeer, he possessum is acually closer o T han he possessor, so in a passie P, T can arac he possessum wihou worrying abou any daie inerener. 3.3 Leapfrogging The inersion mechanism which he ApR/ApP consrains make possible resembles wha McGinnis (1998) calls leapfrogging. Some of McGinnis s own cases appear o reinforce he conclusions already drawn from he examinaion of episemic daies, in fac. In Albanian, for example, sandard quanifier-pronoun binding ess show ha indirec objecs assymerically c-command direc objecs. McGinnis illusraes his wih he (41) daa, from Massey (1992). (41) a. Agimi ia dha secili djalë pagën e ij. Agim-NOM CL gie each boy-dat pay-acc his Agim gae o each boy his pay. b. *Agimi ia kheu secilin liber auori ë ij. Agim-NOM CL reurn each book-acc auhor-dat is Agim reurned o is auhor each book. In passie senences, howeer, where he direc objec is he arge of agreemen, he reerse is rue. (42) a. Secili libër iu khye auori ë ij. each book-nom CL khye-nact auhor-dat is Each book was reurned o is auhor. b. *Secili djalë iu dha paga i ij. each boy-dat CL gie-nact pay-nom his Each boy was gien his pay.

27 Approaching he edge 27 The reersal of binding possibiliies is o be expeced if Albanian double objec passies require o displace boh he daie Goal and he lower Theme nominals in urn, and if his moemen inoles merge o he complemen of, i.e. firs P, hen is moher.. In ha case, once again, he hierachical ranking of accusaie and daie argumens will be reersed from he base posiions. Noice ha acie senences in Albanian mus no undergo he same process, despie he presence of boh daie and accusaie Case. Wha his implies is simply ha he accusaie Case assignmen does no rigger moemen in Albanian, bu agreemen of wih he lower nominal in a passie, when Case is no assigned, does require moemen (presumably o render he lower objec accessible o T). This hen mus be an opion which he Case parameers esablished by uniersal grammar make aailable. A similar accoun can be gien o anoher of McGinnis cases: Briish English dialecs in which he lower of wo objecs is accessible for A-moemen in passies. Thus, in (43b), a book, which presumably originaes as he complemen in an applicaie phrase, ends up as he subjec, while he higher objec Colin remains inside he erb phrase. Again, he reersal of word order can be aken o reflec a double displacemen of he wo objecs by, wih merger of each wih he complemen of. (43) a. Colin was gien a book for his birhday. b. A book was gien Colin for his birhday. McGinnis obains hese resuls in Albanian, Briish English, and in oher similar cases, by imposing her own condiion on he how muliple specifier ordering is resoled. She proposes ha a uckingin ordering, which preseres he underlying order, is enforced when he aracing probe alues he same ype of feaure in Agree. This is he case for Slaic muliple wh-moemen. Bu if he probe alues wo disinc ypes of feaures, hen he second specifier o be moed mus become a higher specifier. Thus, for example, in McGinnis s analysis of Japanese scrambling, a single head may alue boh φ/case feaures and a scrambling feaure, and he scrambling feaure will be saisfied by a phrase which raises o a higher specifier posiion.

28 28 Approaching he edge This ype of explanaion will only be compelling o he exen ha a ypology of feaures is aailable o ground i. Unless an independan meric of similariy is in place which allows us o judge when wo feaures are o coun as similar enough o produce a ucking-in order, he only way o idenify wha makes feaures similar is by looking o see wheher hey rigger a uckingin resul, or an inersion resul. Bu his amouns o a circular logic, and mus be abandoned. In conras, he claim ha he he ApR/ApP conrols when ucking-in is preferred o inersion appears no o be circular, and is herefore preferable. In boh he French episemic daies and he Albanian and Briish English leapfrogging srucures, inersion of he Goal and Theme argumens makes he laer accessible o C, for raising and nominaie Case assignmen. One remaining, paricularly inricae insance of wager-class ECM is worh considering, in which he same sor of inersion feeds muliple nominaie Case-assigmen. This is he case where he marix erb is assure (Kayne, 1984): (44). (44) a. *They assured us PRO o be srong swimmers. b. *They assured us Jenna o be a srong swimmer. c. They assured us ha Jenna was a srong swimmer. d. We were assured ha Jenna was a srong swimmer. e. They assured us sincerely ha Jenna was a srong swimmer. f. Jenna, who hey assured us o be a srong swimmer. g. They assure us o be a srong swimmer eery kid who has compleed heir swim lessons. Example (44a) shows ha he infinial clause seleced by assure mus be defecie, because i canno sere as a conrol clause. So like oher episemic erbs, assure selecs a bare infiniial TP, and no an infiniial CP. Example (44b) shows ha assure canno paricipae in full ECM. This is o be expeced for wo reasons. Firs, since assure has an agenie subjec, i should fall under Pesesky s Agen/ECM Correlaion, which prohibis accusaie Case for non-argumens. Second, he srucural Case associaed wih is acually assigned o he higher Goal argumen anyways.

29 Approaching he edge 29 This is shown clearly by he raising of he Goal o subjec posiion in he passie example (44d). And he posiion of he aderb sincerely o he lef of he Goal shows ha us has been araced up from is iniial posiion as a specifier in he Applicaie phrase o become a siser of he P complemen. Despie he presence of he Goal objec in (44b), exracion of he subjec of he infiniie sill produces a grammaical resul, where exracion can be ia wh-moemen or Heay-NP Shif: (44e,f). We can undersand he grammaical logic being exploied here by examining he srucure in which aracion by mus operae. For (44d), his will be (45). (45) hey assure- P ApplP Appl us Appl TP T o P who be a srong swimmer The probe assigns accusaie Case o us and in doing so, riggers moemen, which mus be accomplished by merge wih P. This preens he displaced nominal from hindering LA from projecing, and saisfies ApP. assigns srucural Case only once, bu i may sill rigger moemen of a second nominal arge, and does so wih who. The merge sie for who mus sill saisfy all he usual consrains, bu his is accomplished by ucking in who o make i he siser of he moher of. Since who will be moing on shorly, is presence will no affec LA, and ApR prefers his posiion o he alernaie posiion wihin he complemen of.

30 30 Approaching he edge Afer boh of he nominal arges of hae been displaced, he srucure is (46). (46) hey who assure- us P ApplP Appl Appl TP T o P be a srong swimmer As wih he simpler cases of wager-class ECM discussed aboe, he wo nominals a he edge of P can (and mus) now be Case-marked by C, which displaces hem in urn: firs o merge wih TP, and hen o merge wih CP. I doesn maer a his poin ha can assign Case o only one nominal arge, because i is C ha supplies he exra Case anyways. To be horough, le us consider some of he remaining assure srucures which migh be generaed along hese lines. If he subjec of he infiniie in (46) were no a wh-phrase, bu a simple, hen i could no merge ouside, bu LA would be saisfied if i merged wih he complemen of. The resul would be (47). (47) *They assured Jenna us o be a srong swimmer. The ungrammaicaliy of (47) can be aribued o he lack of Case-assignmen for Jenna. in (47) has already exhaused is Case-assigning poenial before Jenna is displaced, and C canno access Jenna inside he P phase.

31 Approaching he edge 31 If assure were passie, hen one migh imagine ha Jenna could merge ouside and hen be displaced higher a he CP phase leel. The resul hen migh be (48). (48) *Jenna was assured us o be a srong swimmer. Bu his ime, alhough Jenna will be assigned nominaie Case successfully, and LA can idenify as he label for P, he passie erb will be unable o assign Case o us. Again, he deriaion mus fail for a Case-heoreic reason. 4. Causaies In many languages, causaie consrucions show sriking resemblances o applicaie consrucions, wih paired daie-accusaie Case-marking paerns found commonly in boh (Baker, 1988). In fac, because he srucure of he upper par of P is probably beer undersood han he srucure of ApplP is, a careful examinaion of how he word order is generaed in causaies may be more insrucie han are he applicaie deriaions already discussed. Of course, he lieraure on causaies is far oo deep and exensie o do more han ouch on a few paricular aspecs in his paper, bu i is neerheless reealing. I briefly examine how ransiie causaies are formed in wo languages: Lubukusu (Banu) and French. In hese arious applicaie consrucions jus examined, he reersal of underlying word order resuls because he probe always riggers moemen of boh is daie goal and is accusaie goal. If only he daie were displaced, no inered order would be generaed. Bu here appear as well o be languages in which he reersal opion is exploied as one of wo choices. As described by Baker e al. (2012), Lubukusu causaies do exacly his. I is safe o assume, wih Baker e al. ha Lubukusu causaies are formed by incorporaion of he - head of a downsairs P ino a higher affixal causaie suffix. Incorporaion of he lower erb leaes behind all he argumens wihin P, including agenie subjecs (if he erb inroduces hem) and inernal argumens (heme, goal, ec.). In his language, he sranded argumens of he lower erb display ariable word order of a

32 32 Approaching he edge specific ype. Alongside he heme-agen order in (49a), he agen-heme order in (49b) is possible (excep where blocked by he Phase Edge Prominence Consrain 8 ). (49) a. Wafula a-nyw-esy-a Wekesa ka-ma-lwa. Wafula Agr-Tns-drink-Caus-f Wekesa Agr-Agr-beer b. Wafula a-nyw-esy-a ka-ma-lwa Wekesa. Wafula Agr-Tns-drink-Caus-f Agr-Agr-beer Wekesa Baker e al. (2012) show ha each order in (49) can feed A-moemen operaions, so ha a passie causaie can be formed wih eiher he causee or he downsairs objec as he final subjec of he senence. This means ha he deriaion mus include some mechanism for opionally reersing he underlying order wihin he causaie complemen. Baker e al simply sipulae an opional A-moemen inernal o P which iolaes Closes Moe/Superioriy in order o accoun for he reersed word order in (49b), bu he ApR/ApP model offers a beer approach. We need only suppose ha he srucure of boh of hese clauses includes he subsrucure (50) afer he causaie suffix is inroduced, bu before LA applies wihin he marix phase. (50) Wafula esy Wekesa nyw- P kamalwa The usual incorporaion operaions will raise he lower erb up o he causaie suffix (Baker, 1988); our concern is wih he objecs. The agenie subjec of he lower erb mus be raised ou 8 The Phase Edge Prominence Consrain preens he objec in he downsairs P from being raised ou if i is a 1s or 2nd person. The characerisaion of his conrain presened in Baker e al. (2012) remains alid wih he slighly differen srucures defended here.

33 Approaching he edge 33 o allow LA o operae on he downsairs P. If he marix now simply displaces Wekesa in (50), hen LA and ApP will ensure ha he causee mus merge in he complemen of he, as siser o he upsairs P. If he marix has he opion of displacing kamalwa as well, hen he objec will also be require o merge below, bu ApP (and ApR) will force he merge sie o be he unlabelled moher of he upsairs P, and he order in (49b) will resul, wih he srucure in (51). (51) Wafula esy- kamalwa Wekesa P P P nyw- When he marix P is passie (wih kamalwa raising ino he marix subjec posiion, hen he merge sie for he second displaced nominal will be higher han indicaed in (51), because kamalwa mus be accessible o roo C. In ha case, kamalwa will merge wih he moher of marix, which LA oleraes precisely because kamalwa will moe sill higher a he nex phase leel. If a passie marix aracs only Wekesa, hen kamalwa hen remains in is base posiion, and Wekesa will be he nominal which merges ouside, and hen moes on. In shor, Lubukusu causaie word order requires no fundamenal aleraions o he model deeloped here. And he elegan Baker e al. (2012) analysis becomes sill idier wih he eliminaion of heir special moemen operaion which reerses he word order of agen and heme wihin he erb phrase.