MIT Working Papers in Linguistics

# I Papers on Syllable Structure, Metrical Structure and Harmony Processes (1979) #2 Issues in Japanese Syntax (1980) #3 Papers on the Gramm ar of Semitic Languages ( 1981) #4 Papers in Formal Syntactic Theory (1982) #5 Papers in Grammatical Theory (1983) #6 Papers in Theoretical and Appli ed Lingui stics ( 1985) #7 Papers in Morphology ( 1984) #8 Papers in Theoretical Linguistics ( 1986) #9 The 25th Anni versary of MIT Linguistics ( 1987) # I 0 Functional Heads and Clause Structure ( 1989) # II Papers from the First Student Conference in Linguistics, MIT (1989) # 12 Papers from the Second Student Confe rence in Linguistics, MIT ( 1990) # 13 Papers on wh-Movement (1990) # 14 Papers from the Third Studen t Conference in Linguistics, MIT ( 1991) # 15 More Papers on wh-Movement (!991) # 16 Papers from the Fourth Student Co nference in Linguistics, Ohio State University

( 1992) #17 Proceedings of the Kwa Comparati ve Syntax Workshop, MIT (1992) # 18 Papers on Case and Agreement I ( 1993). # 19 Papers on Case and Agreement 11 (1993) #20 Papers fro m the 5th Student Con fe rence in Lingui stics, Uni versity of Washington

( 1993) #2 1 Papers on Phonology and Morphology ( 1994) #22 The Morphology Syntax Connection ( 1994) #23 Papers from the 6th Student Conference in Linguistics, Uni versity of Rochester

( 1994) #24 Formal Approaches to Japanese Linguistics: Papers from the May 1994 MIT

Conference ( 1994) #25 The Interpreti ve Tract (May 1998) #26 Papers on Language Processing and Acquisition ( 1995) #2 7 Papers on Minimalist Syntax ( 1995) #28 Papers on Language Endangerment and the Maintenance of Linguis tic Di versity

( 1996) #29 Formal Approaches to Japanese Lingui stics: Proceedings of FAJL 2 ( 1996) lBO PF: Papers at the Interface ( 1997) #3 1 Proceedings of the Eighth Student Conference in Linguistics ( 1997) #32 Papers from the UPenn/M1T Roundtable on Argument Structure and Aspect (May

1998) #33 Papers on Morphology and Syntax, Cycle One (!999) #34 Papers on Morphology and Syntax, Cycle Two (1999) #35 Papers from the UPenn/MlT Roundtable on the Lexicon (1999) #36 Proceedings of the 9th Student Conference in Linguistics (2000) #3 7 Proceedi ngs of the 1Oth Student Conference in Linguistics (200 J) #38 Proceedi ngs of the 12th Student Conference in Linguistics (2001) #39 A Few from Building E39 (2001 ) #40 Proceedings of the I st HUM IT student conference on Language Research (200 I) #4 1 Proceedings of FAJL 3 (2002) #42 Phonolog ical Answers (and Their Corresponding Questions) (2002) #43 Proceedings of HUM IT 200 l (2002)

Proceedings of AFLA 8: the Eighth Meeting of the Austronesian

Formal Linguistics Association

MIT Working Papers in Linguistics

Volume 44

Edited by

Andrea Rackowski and Norvin Richards

November 2002

MITWPL MITWPL is an organization of the

graduate students in linguistics at MIT. Department of Linguistics, E39-245 MIT

77 Massachusetts A venue Cambridge, MA 02139

mitwpl @mit.edu http://web.mit.edu/mitwpl/

, f

I

I \

1

I

Backward Control: Evidence from Malagasy*

Maria Polinsky and Eric Potsdam University of Californ ia, San Diego and the University of Florida

1. Introduction

Control is an asymmetric interpretati onal dependency between two sy ntac ti positions in which the referential properties o f an overt one, the CO NTROLLER de termine the referenti al pro perti es of a non-overt one , the CONTROLLEI (Bresnan 1982) . In the canonical case o f FORWARD CONTROL , the contro ller i. s tructura ll y superior to the contro llee (represented atheoreti ca ll y as .0.), (la; Howeve r, a second pattern, which we will call BAC KWARD CONTROL, is a ls1 possibl e, in whi ch the controllee is structurall y superi or to the controller, (l b) .

(l ) a. The di ver; tried .0.; to hold hi s breath.

b. .0.; tried the di ver; to ho ld his breath. (hypothetical)

Backward Control has been reported in the literature for Japanese causati ves anc tokoro -clauses (K uroda 1965 , 1978, Harada 1973), Braz ili an Portuguese causati ves (Farrell 1995), and Nakh-Daghestani an aspectual verbs (Kibrik 1999 Po linsky and Potsdam 2000, 2001 a). This paper contributes to the growing bod) of empiri cal evidence supporting the ex istence o f Backward Contro l (BC) b) documenting and analyzing the construc tion in Malagasy (see Polin sky anc Po tsdam 2001 a,b for a fuller treatment).

The paper is organized as fo llows: Sec ti on 2 presents the M alagas) constructi on and lays o ut our structural proposal. Sec tions 3 and 4 argue agains1 two analytical alternati ves . Section 5 summarizes our empirical findings and set ~ the stage fo r poss ible theoretica l analyses . To th at end , secti on 6 attempts tc ana lyze BC within a PRO-based Princ iples and Parameters (P&P ) framework and conc ludes that P&P quite genera ll y rules out BC. Section 7 turns to a more recent Minimali st-based movement analys is of Control in whi ch the controller­controllee relationship is a res ult of movement. Our fundam ental c laim is th ai movement in Forward Control is overt while movement in the case of Bac kward Contro l is covert. BC is thus expected given widely recogni zed overt versus covert parametric variation in movement.

• We would like to thank our Malagasy consultant s Noro Brady and Solange Green for their he l ~ with the data . For numerous d iscussions we thank Stan Dubins ky, Norbert Hornste in, Ed Keenan, Y uki Kuroda , Ileana Paul , Matt Pearson, and the aud iences at the Austronesian Formal Lin gui sti c Assoc iat ion 8 a t MIT and the Subtropi ca l Summer Syntax Workshop at the Uni vers ity o f Georgia.

MIT Working Papers in Linguistirs 44, 257-272

Proreeding.1· '!lA FLA 8 © 2002 Maria Polinsky and Eric Potsdam

Mari a Polinsky and Eric Po tsdam

2. Control in Malagasy

Malagasy is an Austronesian language spo ken o n the is land of Madagascar. Its bas ic word order is VOS but VSO order is also typically possible. A canonical Forwa rd Contro l constructi on is given in (2) (Randriamasimanana 1986, Keenan 1976. 1995 , Law 1995, Paul and R anai voson 1998).1 Expectedly , VSO word order is also poss ible, (2b)2

(2) a. m- an-a ndrana [m-i-tondra ny fiara il;] Rabe, PRES-ACT- try PRES-ACT-dri ve the car R abe

b. m-an-andrana Rabe, [m-i-tondra ny fiara il;] PRES-ACT-try Rabe PR ES-ACT-dri ve the car 'Rabe is tryin g to dri ve the car.'

For such examples and VOS order in general, we assume a clause structure as in (3), fo ll owing Guilfoy le, Hung, and Travi s 1992. The predicate constitutes a VP and the c lause-final subject occupies a ri ghtward specifier of lP. 3

(3) lP

~ I' DP; ~ Rabe

I VP

~ V IP try ~

I' Ll;

~ I VP

~ dri ve the car

The verbs maHo111boka ' begin ', mahavita 'accompli sh', and mitsahatra 's top ' appear in superfi c iall y simil ar sentences, (4a)4 Upon closer investigation however, these verbs behave rather differently from Forward Contro l verbs. One difference is that they do not permit VSO order, *(4b).

1 The exact analys is of such structures is a topic of current debate . Law t995 argues that they do not instanti ate canoni cal control. Note for exa mple that the embedded clause is tensed; Malagasy has no morphological infiniti ves.

2 .We use th e fo ll owin g a bbrev iati ons in g loss ing: ACe-acc usativ e, A C T- acti ve, CIRC-circumstanti al, COM P-complcmenti zer, FUT- future, IMPER- imperati ve, NEG- negati on, OBL--<Jblique, PASS- pass ive, PRES- present, Q-questi on.

3 For alternati ve Malagasy clause structure, see MacLaughlin 1995, Pensalfini 1995 , Paul 1999, and Pea rson 200 1. We will call the c lause- fina l DP the subj ect, althou gh nothin g hinges on thi s terminology . 4 We illustrate only wi th IIIQI/0111 /Joka ' begin '. The other verbs behave simi larly, although some data on mitm lwtra (Keenan 1995 : 195- 196) suggest possible dialectal vari ati on.

258

(4) a.

b.

Backwards Control: Evidence from Malagasy

m-an-omboka m-i-tondra ny PRES-ACT-begin PRES-ACT-drive the

fiara car

*m-an-omboka Rabe m-i-tondra ny PRES-ACT-begin Rabe PRES-ACT-dri ve the 'Rabe is beginning to drive the car. '

Rabe Rabe

fiara car

We will argue that thi s and other contrasts between manandrana ' try ' and manomboka ' begin' arise from distinct clausal organizati ons . We claim that 'begin ', 'accomplish', and 'stop ' have a rather unique syntax schematized in (5) and we will call these verbs Backward Control (BC) verbs. The central claims of the BC analysis are that i) the overt controll er is structurally in an embedded clause, and ii) the main verb assigns an external theta rol e to a sy ntactic controllee. We defend these claims below by arguing against a lternati ves that do not embrace them . Sectio n 3 rejects a monoclausal analys is while section 4 argues against a sentential argument anal ysi s.

(5) m-an-omboka [m-i-tondra PRES-ACT-begin PRES-ACT-drive 'Rabe is beginning to drive the car. '

ny fiara Rabe ;] the car Rabe

3. Against Clause Union: A Biclausal Structure

Ll;

The verbs that participate in the unusual pattern above often form reduced cl ause constructions in other languages. Thus a reasonable hypothesis regarding the Malagasy BC verbs is that they are clause union predicates th at take a non­clausal VP complement, as in (6) . (7) presents our BC proposal for compari son ..

clause union hypothesis (6) IP

~ I' DP ~ Rabe

I VP

~ V VP

begin ~ drive the car

259

Mari a Polinsky and Eric Potsdam

BC hypothes is (7 ) IP

~ [' L\ ~

I VP

~ V IP

begin ~ I' DP; ~ Rabe

I VP

~ dri ve the car

A primary ad vantage of a clause union analysis is th at it is monoclausal and does not posit any empty categories. Below we give evidence from constituency, word o rder, and c lausa l domain identifiers against the monoclausal analysis .

3.1. Constituency evidence

Two re levant strings can be tes ted to di stingui sh (6) and (7) : i) the embedded pred icate+overt subject (drive the car Rabe), which we will call E+S , and ii) the main verb+embedded predicate (begin dri ve tlze car), which we will call V +E. Under the clause uni on hypothes is, E+S is not a constituent but V+E is . The BC hypothesis in contras t c la ims th at E+S is a constituent but V+E is not. Data from ind irect pass ive and coordination show that the BC predictions are correct.

In Malagasy, verbs may appear in one o f three voices, active, pass ive, and c ircum stanti al (indirect pass ive), with a corresponding change in the DP that acts as the subject (see Keenan 1976, 1995 , Perlmutter and Postal1983) :

(8) a. n-i-vid y ny fiara hoan-dRasoa Rabe ACTIVE PAST-ACT-buy th e car for-Rasoa Rabe

b. no-v id-in-dRabe hoa n-dRasoa ny fiara PASSIVE PAST- buy- PASS-O BL.Rabe for-Rasoa the car

c. n-i-vidi -an-an-dRabe ny fi ara Rasoa CIRC PAST-ACT-buy-C IRC-OBLRabe the car Rasoa 'Rabe bought a car for Rasoa.'

The voice system is a useful constituency diagnostic because, if a string can advance to subject with one of the voices then it must be a constituent. (9) demo nstrates th at th e c ircum stantial form of ' begin ' is poss ible with E+S 111

subject pos ition. The bracketed string E+S must therefo re be a constituent.

260

(9)

Backwards Control: Evidence from Malagasy

anomboh-ana [mitondra ny fiara begin-ClRC dri ve the car 'Rabe begins to dri ve a car.'

Rabel Rabe

We now turn to coordinati on. First, ( 10) confirms that E+S is a constituent since it can be coordinated to the exclusion of the matrix verb. Second, ( 11) shows that V+E is not a constituent. It cannot be coordinated with another predicate, ( 11 a,b). Onl y the BC structure correctly predicts all of this straightfo rward consti tuency evidence5

( 10) a. nanomboka rnisotra toaka Rabe ary mi tady akangavavy izy began drink booze Rabe and5 seek pros titute he 'Rabe began to drink booze and chase women.'6

b. began [LP [IP drink booze Rabe] and5 [IP seek prostitute he]]

(11) a. *nanomboka namaky ny taratasy sy menahatra Rabe began read the letter andvr is .ernbarrassed Rabe ('Rabe began to read the letter and was embarrassed.' )

b. *[ vr [ vr began read the letter] andvP [ vr is.embarrassedl] Rabe

3.2. VP right edge identi fiers

Elements that mark the right edge of VP provide another tool for di stingui shing between the two hypotheses. Wi th reference to the structures in (6) and (7) , the subject linearly foll ows the ri ght edge of the matri x VP in the clause union structure, (6) , but precedes it in the BC structure , (7) . Keenan 1995 identifi es adverbs and the question particl e ve as VP-ri ght edge markers. In what fo ll ows we show that the grammatical placement o f these elements in the examples under investigation yields direct evidence for the BC structure.

The yes-no question particle ve is di scussed in Keenan 1976, 1995 , Paul 1999, and Pearson 2001. Keenan 1995:1 79 proposes th at it direc tl y foll ows the predicate phrase in a s imple clause:

(12) mitondra ny fi ara (ve) dri ve the car Q ' Is Rabe dri ving the car?'

Rabe (*ve)? Rabe Q

While we have encountered dia lectal variation, for some speakers, the question particle must fo llow the overt subject in the structure under inves ti gati on, ( 13), as predicted by the BC structure. For other speakers, however, the partic le must

5 Malagasy has a number of coordinati on parti c les whose exact functions have ye t to be in vesti gated. Two are used in the exa mples: sy coordi nates predi cates and we gloss it as ' andve' ; ary coordinates clauses and is glossed by 'ands' See Keenan 1976;274.

6 This exa mple a lso has a second meaning, which is irre levant fo r us, ' Rabe began to drin k booze and he chases women.'

261

Maria Polinsky and Eric Potsdam

precede the overt subj ect, making the data inconclusive.

(13) % manomboka mitondra ny fiara begin dri ve the car 'Has the teacher begun to drive?'

Rabe Rabe

ve Q

Fo ll owi ng observatio ns in Rackowski 1998 and Pearson 1998, one possible pos itio n fo r VP-adverbs in a simple M alagasy cl ause is immediate ly follow in g the predicate, with the clause-fina l, post-subject position not possible, (14a). Thi s sa me di str ibuti on holds in Forward Contro l struc tures, ( 14b), with the structure in ( 14c). VP-ad verbs are thus also VP right edge markers.

( 14) a.

b.

c.

niteny ity tonon-kira ity (indroa) Rabe (* indroa) knock this door this twice Rabe twice 'Rabe knocked twice on thi s door.'

nanandrana niteny ny tonon-kira (indroa) tried knock the door twice 'Rabe twice tried hard to knock on the door.'

[nanandrana [niteny ny tonon-kira ll.;]]yp

Rabe (*indroa) Rabe twice

indroa Rabe;

( !5) demonstra tes th at the ad verb fo llows the subject in the examples under inves ti gati on, in contrast to ( 14b,c). The pattern fo llows from the BC structure if the right edge of the VP is as shown in (15b) , with the overt subject internal to the matri x VP. The clause union hypothesis does not predict thi s possibility.

( 15) a. nanomboka niteny ity tonon-kira ity Rabe indroa began knock this door thi s Rabe twice 'Rabe twice began to knock on this door.'

b. [nanandrana [n iteny ity tonon-kira ity Rabe]]yp indroa

3.3. Biclausality

1\ thi rd diffe rence be tween structures (6) and (7) is in the number of clausal domains (IPs). In the c lause uni on structure , there is onl y one IP while in the BC struc ture there are two IPs. In what follows we show that the examples under investigation pass di agnos ti cs for biclausality.

Clauses are typi cally the domain o f independent tense and polarity under the assumpti on that IP is the locus of verbal morphosy ntactic categories. Under the BC st ructure. then, we expect to be able to independently specifiy tense and polarity on both verbs. U nder the clause uni on hypothesis, in contrast, there is onl y one infl ec ti onal domai n and we ex pec t to be able to specify only one va lue of tense and polarity for th e sentence. Since all verbs in Malagasy show tense infl ection, we expect obligatory tense concord with th e higher verb.

The examples below support the BC structure. (16) demonstrates that th e embedded verb may have a tense spec ification di stinc t from that of the matrix verb, although both may be identically specified. Similarly, both verbs

262

Backwards Control: Evidence from Malagasy

may be independently negated, (17a,b).

( 16)

( 17) a.

b.

m-an-omboka [m-itondra/h-itondra ny fiara Rabe] PRES-ACTIVE-begin PRES-drive/FUT-drive the car Rabe 'Rabe has begun to dri ve the car.'

tsy nanomboka nihomehy t-amin ' ny NEG began laugh PAST-PREP'the 'He didn ' t begin to laugh at the story .'

tantara izy story he

nanomboka tsy nihomehy t-amin 'ny tantara intsony izy NPI he began NEG laugh PAST-PREP' the story

'He began to no longer lau gh at the story.'

A second kind of evidence for biclausa lity comes from clause-bound operati ons in the grammar. One such phenomenon is VSO word o rder mentioned above. While we take no stand on how exactly VSO order is derived, the observation is that a subject can directly follow th e verb onl y if it is the subject of this verb. In other words , subject scrambling is clause bound. With referen ce to the structures in (6) and (7), VSX word order is predi cted to be possible with the c lause uni on structure but not the BC struc ture. It is ruled out in the latter case because the subject is in an embedded clause and cannot be scrambled into a hi gher clause. As we saw earli er, the latte r is the correct predicti on. Unlike with Forward Contro l verbs, (2b), VS word o rder is disallowed with 'begin' :

( 18) *manomboka ny mpianatra mitondra ny fiara begi n the student drive the car ('The student is beginning/has begun to drive the car.')

A s imilar state o f affairs arises with Subject-to-Object Raising (SOR) structures. A number o f Malagasy verbs govern SOR (Keenan 1976:283-285, Paul and Rabaovo lolona 1998) in which an embedded subj ect raises to the object pos ition of th e immedi ate ly hi gher verb , ( 19) . This movement targets onl y subjects and is clause-bound; it cannot target a multipl y embedded subject.

(19) mino an-dRabe; [ho n-itaraina t;] aho believe ACC-Rabe COMP PAST-ACT.complain I 'I believe Rabe to have complained .'

The clause union hypothesis predicts that a ' begin' example can be embedded under an SOR verb. The subject simpl y raises, paralle l to ( 19). Under the BC structure, SOR is correctly predicted to be impossible because the overt subject is too deeply embedded to raise. It would have to cross two c lause boundaries:

(20) *mino an-dRabe; [ho nanomboka [nitaraina t;]] Rasoa believe ACC-Rabe COMP began complai n Rasoa ('Rasoa believes Rabe to have begun to compl ain .' )

263

Maria Polinsky and Eric Potsdam

We conc lude from these various arguments that BC exampl es are bic lausa l. S ince the clause union structure does not represent this character istic of the construction , it should be rejected.

4. Against a Single Clausal Argument: Thematic Role Effects

Sec tio n 3 showed the examples under consideration contain a fu ll embedded c lause and the BC structure in (21) reflects this constituent. At the same time, our hypothesi s posits an additional non-overt argument 11 in the matrix clause that a prio ri seems unnecessary. In this sectio n, we support this additional comp lexity by considering and rej ectin g an alternative analys is in which the BC verb takes only a single c lausal argument, as in (22) . We argue below that i) the BC verb has an external argument which is not the c lause, and ii) th is argument is sy ntacticall y represented.

BC hypothes is (2 1) IP

~ l' 11;

....-~ I VP

~ V IP

beg i n ~ I' DP; ~ Rabe

I VP

~ dri ve the car

sin gle c lausal argument hypothes is (22) IP

~ I' IP

~ ~ I VP drive the car Rabe

I v begi n

4.1. Selectional restrictions

Selectiona l res trictions ind icate that the verbs und er considerati on ass ign a thematic role (theta ro le) to an extern al argument (in addition to ass igning a the matic ro le to the clausal argu ment) . These res tri ct ions seem to be borne by the o vert subject; however , because it is not a clause-mate of the hi gher verb , it also cann ot be an argtunent of that verb. Consequently, we pos it a co indexed , empty ca tegory in the same clause which bears the selectiona:l restrict ions.

264

Backwards Control: Evidence from Malagasy

(23) illustrates that a non-volitional, non-agenti ve DP cannot be used in the BC construction. We propose that (23b) is ungrammati cal because the selectional restrictions imposed by ' begin' are not satisfied.

(23) a. avy ny orana b. come the rain 'It 's raining/going to ra in .'

*nanomboka [avy ny orana) PAST.begin come the ra in (' It began to rai n. ' )

At the same time, these verbs can appear in the imperative, (24), a construction which requires an intentional agent (Perlmutter 1970, Farkas 1988).

(24) manomboha mitondra begin.IMPER dri ve 'Begin to dri ve the car' '

ny fiara (ianao) the car you

T here is good evidence then that ' begin' pl aces thematic restn ct10ns on an external argument. Under the single clausal argument hypothesis however, there is no argument that bears these restrictions, so they are unexpected .

4.2. Empty category effects

Within the Standard Theory and its derivat ives, th e presence of selectio na l res trict ions entail s that there is a local syntac tic constituent that is ass igned the associated theta role. A theta ro le cannot be ass igned across a clause boundary and the argument cannot si mply be un rea li zed. Neverthe less, independ ent empirical evide nce fo r the syn tactic existe nce of this empty category is desi rable. Our ev idence comes from floating quantifiers. Whether fl oatin g quantifiers are adverbs or stranded DPs, it is widely agreed that they must be bound and cannot appear in a hi gher clause th an their antecedent.

Keenan 1995: 178 di scusses the di stribution o f the Ma lagasy uni versa l qua nti fier dalzolo. He indicates that daholo occurs at the right edge of VP and must be bound by the subject, (25), although it does not form a constituent with it. It can not be bound by the object in (25) nor can it be bound by a more deepl y embedded element, (26) .

(25) m-i-j ery ny mpi anatra daholo ny mpampianatra

(26)

watch the student all the teacher 'All the teachers watched the student(s).'

#'The teacher(s) watched all the students.'

*tsara daholo tonga ny mpi anatra good all arri ved the student (' It 's good all students arri ved. ' )

The distri bution of daholo supports the BC structure . First, (27a) demonstrates that th e quantifie r may ex pectedl y appear in the embedded clause where it is bound by the overt subject. Surprisingly, the quanti fie r can a lso appea r in the matrix clause, (27b), req uiring that there be a syntac tic argument there. With the BC structure, thi s non-overt argum ent is /1 . In a s in gle cl ausal argument

265

Mari a Polinsky and Eric Potsdam

structure , the re is no c lause- mate subj ect antecedent for th e quantifier and we expect (27b) to be ungramm atical, on a par with (26).

(27) a. nanomboka omaly [rnihomehy daholo ny anki zy] began yesterday laugh all the children 'Yesterday the children began to all laugh.'

b. %nanornboka daholo omaly [mihomehy ny ankizy] L'l; began all yes terday laugh the children 'Yesterday the children all began to laugh.'

A lthough more e vide nce supportin g th e empty category is des irable, we cons ider the floa ting quantifie r data quite stro ng. Together with the selectio nal res tri ctions it supports the c laim that BC verbs ass ign an external theta ro le to a c lausemate empty category. A single clausal argument is untenable.

5. Backward Control in Malagasy

For the exampl es under considerati on, sec tio ns 3 and 4 have supported the structure in (28) in whi ch the overt subj ect is structurally in an embedded clause but the main verb nevertheless also has an external argument.

(28) m-an-omboka [m-itondra ny fiara Rabe ;] L'l; PRES-ACT-begin PR ES-dri ve the car Rabe 'Rabe is beginning to dri ve the car.'

Si nce the non-overt argument is interpreted as coreferential with the overt embedded subj ect, we propose that a contro l relationship exists be tween the two. Two add itional observa tio ns support thi s characte rj za tion. First, the control interpreta ti o n is required . The matri x subj ec t cannot be interpreted as refe renti a ll y di stinct from the embedded subject, (29). Seco nd, the matrix argument does not alternate with an overt DP but must be unpronounced , (30).

(29) a.

b.

(30)

*manomboka mitondra ny fiara Rabe; ny mpampi anatrak begin drive the car Rabe the teacher ('The teacher has begun to have Rabe drive the car. ' )

*manomboka mitondra ny fiara Rabe; prok begi n dri ve the car Rabe ('Someone has begun to have Rabe dri ve the car.' )

*manomboka mitondra ny fi ara izy; begin dri ve the car he (' Rabe is beginnin g to dri ve the car.')

Rabe; Rabe

In summ ary, mano111 boka 'beg in ' , maha vita ' accomplish', and mitsahatra ' stop ' are Backward S ubj ect Co ntro l verbs in whi ch the overt co ntro ller is in the embedded clause and the no n-overt contro llee is in the hi gher cl ause. The

266

Backwards Control: Evidence from Malagasy

. remainder of the paper turns to poss ible analyses of this phenomeno n,

6. Principles and Parameters Approaches

Within a Princ iples and Parameters (P&P) analys is of Contro l (see fo r example Chomsky and Lasnik 1993) , the empty category co ntrollee in Control structures is PRO. The presence o f PRO is ma ndated by the Theta Cri ter io n, which requires that every 8-ro le be ass igned to exactly one DP and th at every contentful DP receive exactly one 8-role. Since the matrix subj ect receives a 8-ro le from the control verb, there must be a d istinct DP in the embedded clause to be the external argument of the embedded verb . In addition, because PR O is referentially dependent, it is widely assumed th at PRO must be sy ntactically bound to receive a referential interpretati on. When thi s condition does not obtain, the arbitrary interpretation o f PRO results: [PRO.rb to win] is always fun.

For the BC construc tion, the empty category subj ect of BC verbs identified above would be PRO:

(31 ) manomboka [mitondra ny fi ara begin drive the car 'Rabe is beginning to dri ve the car.'

Rabe;] PRO; Rabe

Despite the directness of extending the P&P analys is to BC, it faces a number of difficulties. F irst, PRO in (31) is no t bound . Neverth e less, th e a rbitrary interpretation of PRO in which the example would mea n 'Someone is beginning to ha ve R abe dri ve th e car ' is un available, see (29b) above. Second , the configuratio n in (3 1) vio lates Conditi on C o f the Bind ing T heory , which requires th at R-express ions like Rabe be free. The DP is not free but th e example is noneth e less grammati ca l. Since th e argume nts against the PRO an alys is are essentiall y independe nt of the de tail s of Malagasy sy ntax, we conclude that P&P quite generall y rules out Backward Control.

7. A Minimalist Movement Analysis of Control

Gi ven the failure of P&P to account fo r th e BC construction , we pursue an alternati ve under Minimali st assumptions. The leading idea that we will develop is given in (32). BC does in fact contain a control relationship but it does not exist in surface syntax. It is formed in the covert syntax, at Logica l Form (LF).

(32) there is an ordinary contro l rel ationship in BC but it does not exist on the surface, onl y in the covert syntax

Such a proposal requires a derivatio nal view of Control (see O ' Neil 1995, Manzini and Roussou 1999, and Hornstein 1999). In the remainder of the paper we develop (32) using the specifics of Hornste in 1999. We first present his movement analysis of Forward Control and then extend it to BC.

The leading idea in Hornstein ' s analys is is that the Control rel ation is created by movement from one thematic pos ition to another. The proposal is developed within the Minimali st framework and assumes that th is, and all other

267

Maria Polinsky and Eric Potsdam

movement, is driven by feature checking. The core assumptions are in (33), from Horns te in 1995:78 and Chomsky 1995. ·

(33) a. b.

c. d. e.

8-ro les are features a DP receives a 8-ro le by checking a 8-feature of a verb th at it merges with there is no upper bound on the nu mber of 8-roles a chain can have features may be strong or weak the EPP feature is strong, Case/8-role features are weak

Hornste in ass um es that 8-ro les are fea tu res and the ir ' assignment' is thus regu lated by the tenets of Checking Theory. A DP will ' receive' a 8 -role by chec ki ng the 8 -ro le fea ture of a verb. G iven th at Contro l is reduced to movement , mo vement fro m one 8 -posi tio n to another is necessarily permitted and there is no limit on the number of 8-role features that a DP can check.

T he needed assum ptions about features and featu re checking are largely standard within Chomsky 's ( 1995) M inimali st P rogram . Features of a head may be strong o r weak. S trong features must be checked prior to Spell Out while weak fe atures wi ll be checked as late as poss ible in the de ri vation due to Proc rasti nate (Chomsky 1993). We assume, at least for Malagasy , that Case and 8-ro le features are weak and the EPP feature is strong.

W ith these assumptions, the de ri vation o f an Engli sh Forward Contro l structure is (34), with the relevant features shown below the structure.

(34) a.

b.

Kim i hopes ri to win

[IP Kim lvp .!CASFJE:PP

hopes [IP

,( 8hope

to [yp t win]J]]

.!EPP .r 8win

The DP Kim first merges with the embedded V' win and checks the exte rnal 8-ro le feat ure. It then moves to the embedded spec ,IP to check the EPP feature o f the e mbedded r. From there it moves into the matri x clause and creates the control re lat ion. ft first stops in spec,VP to check the external 8 -role feature of the contro l verb hope and then moves to the matri x spec,IP where it checks the EPP feature and Case. The derivation converges with all features checked.

Our fund amental c laim about BC is that its analysis is identical to that of Forward Control except that the movement step that creates the Contro l re lation occ urs in the covert syntax. For the M alagasy example in (3 1), the Spell Out struc ture is (35) (Malagasy structure shown but English words substituted). T he embedded c lause is full y fo rm ed and the features there are checked. The overt deriva tion stops at thi s point yie ld in g the desired constituent structure.

(35) [IP l' [yp begin [1p [yp t dri ve carl y p Rabe]1p ]

8begi n .! 8drive .! EPP/CASE

lrp EPP/CA SE

As (35) shows , there are unchecked fea tures in the structure and we are led to ask whe ther they cause the deri vation to crash. For the Case and external

268

Backwards Control: Evidence from Malagasy

8-role feature of beg in, the answer is no because these are weak features and do no t need to be checked until LF. The EPP fea ture of the matrix clause however is strong and should cause the de riva tion to cras h if not checked overtly. There are a number of ways around this apparent EPP vio lation and the solution that we will pursue is that the EPP can be sati s fi ed by verb raising (Alex iadou and Anagnostopoulou 1998, Benm amoun 1999). Th at is, the EPP may be satisfi ed e ither by an XP in spec,IP or a verbal head ra ised to l '. In the latter case, no overt XP movement to spec,IP is necessary.

In the Malagasy structure then we hypothesize that the BC verb raises to I' overtly and checks the EPP feature. Thi s is shown in the LF representation be low. The proposa l doveta il s with Guilfoy le, Hung, and Travis' (1992) independent claim that verbs rai se overtly to I" in Malagasy. Benmamoun 1999 implements the movement by assumin g that the EPP feature is a strong D feature of I' and that it can be checked by a verbal head whi ch is spec ified with the feature [ +D] . Most verbs are [ -D] and thus are incapable of satisfying the EPP even if they raise to 1'. The [+D] BC verb however may excepti onally check the EPP upon raising. We be lieve that thi s approac h is on the r ight track in that it locates some of the exceptional sy ntax of the BC constructio n in lexical spec ifications of the BC verb itself.

The control relation in the BC constructi on is created at LF:

'(36) [IP begin [y p t tbegin [!P [y p t dri ve car] y p t]1p] Rabe] 1p

.! EPP .! 8begin .! 8dri ve .! EPP/CASE .! CASE

In the covert syntax, th e contro ller DP moves into the matri x c lause and es tablishes a contro l re lation by checkin g the extern al 8-ro le fea ture of the BC verb. It then moves into the matri x spec,IP where it checks Case. The deri vation concludes with all fea tu res checked .

It would appear that the contro ller chec ks Case twice and th at we have a chain with multiple Case positions- a si tuation which is usually no t permitted . We have no solution to this problem but simply suggest two ways around it. One approach is that Case checking is opti onal (McCloskey and Sells 1988, Ura 1998). The hi gher I' does not have a Case feature and so the contro ll er does not check Case when it raises covertl y. A second approach is that multiple Case chains are in fact permitted. There is a fairl y large body of evidence supporting th is conclusion (Yoon 1996, Bejar and Massam 1999, and references therein) . We will leave it as an unresolved iss ue how a language would determine whether multiple Case chains are available.

To summari ze, there are three differences between the English Forward Control derivation in (34) and the BC derivati on in (35) and (36) . First, the timing of the A-movement into th e hi gher cl ause : In Forward Cont ro l the movement takes place overtly, in Backward Control, it takes place covertl y. Thi s di ffe rence determines whether the contro ller will surface in the higher clause or the lower clause. Second, the way in which the EPP is sati sfied at Spell O ut: In Forward Contro l the EPP is sati sfi ed by overt XP movement o f the contro lle r. U nder Backward Control , th is XP movement has not yet taken pl ace so another means of satisfying the EPP must be avail ab le if a language is to have BC. We proposed th at V ' -to-1' serves thi s purpose in Malagasy . Thi rd , th e Case

269

Mari a P o linsky and Eric P o tsdam

ass ign me nt characte ri stics of the cons truc tio n: In the E ngli sh Fo rw ard Control cons tructi o n , the contro ller DP checks Case in the hig he r subjec t positio n only, res u lting in a s ing ly Case-m arked c ha in . In the B C co ns truction, the controller DP m ust be able to c hec k Case at least in the lower subject positio n. It is unclear at thi s po int what the Case c haracteri stics o f the hig her subject positio n are. If C ase is a lso c hecked there, the n mul tiple Case cha ins must be permitted.

T he primary theore tical ques tio n that the construc tio n raises is w hether syntact ic theory sho uld perm it Backward Contro l as we have described it. If the answer is no; the n PRO-based ana lyses of Control pred ict the imposs ibili ty and existi ng cases, inc luding Malagasy, must be misana lyses. O n the o the r hand , if the a nswer is yes , the n we have suggested that a movement analysis of C ontrol pe rm its B C. Malagasy insta nti ates thi s possibility and thus provides interes ting ev idence fo r a moveme nt a na lysis o f Contro l.

References

Alexiadou, Artemis and Elena Anagnostopou lou. 1998. Parametrizing AGR: Word order, V-movement and EPP checking. Ni:aural Language and Ling uistic Theory 16: 49 1-539

Bejar. Susana and Di ane Massam. 1999. Multiple case checking. Syntax 2: 65-79. Ben ma moun , El abbas. 1999. Th e Featural Structure of Functional Categories: A

Comparative Study of Arabic Dialects. Oxford: Oxford University Press. Bresnan, Joan. 1982. Control and complementation. Linguistic Inquiry I3: 343-434. Chomsky, Noam. 1993. A minimalist program for linguistic theory. In Kenneth Hale and

Samuel Jay Keyser (eds.), The View f rom Building 20: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge, Mass.: MIT Press, pp. 1-52.

Chomsky. Noam. 1995 . The Minimalist Program. Cambridge, Ma. MIT Press. Chomsky , Noarn, and Howard Lasnik. 1993 . The theory of Principles and Parameters. In

J . Jacobs , A. von Stechow, W. Sternefeld, and T. Vennemann (eds .), Syntax: An lntem ational Handbook of Contemporary Research. Berlin: Mouton.

farkas, Donka. 1988. On obligatory control. Linguistics and Philosophy II : 27-58 . Farrell , Patrick . 1995. Backward control in Brazi lian Portuguese. In Janet M. Fuller, Ho

Han , and David Park inson (eds.), ESCOL I/ . Ithaca, N.Y.: Cornell Uni versity Department of Modern Languages and Linguistics, pp. 116- 127.

G ui lfo yle, Eithne, Henri etta Hung. and Li sa Trav is. 1992. Spec o f lP and Spec of VP: Two subjects in Austronesian languages. Na tural Language and Linguistic Theory 10: 375-414.

llarada, S.-l. 1973. Counter Equi-N P-Deletion. University of Tokyo Research Institute of Logopedics and Phoniatrics Annual Bulletin 7: 11 3-47.

I lornste in , Norbert. 1999. Movement and control. Linguistic Inquiry 30: 69-96. Keenan, Edward L. 1976. Remarkable subjects in Malagasy. In Charles Li (ed .), Subject

w zd Topic. New York: Academic Press, pp. 247-301. Kee nan, Edward L. 1995. Predicate-argument structure in Malagasy. In C. S. Burgess ,

K. Dziwirek , and D. Gerdts (eds.), Grammat ical Relations: Th eoretical Approaches to Empirical Questions . Stanfo rd: CSLI, pp. 17 1-216.

Kibrik , Alexander E. (ed .). 1999. Ele menty caxurskogo j azyka v tipologiCeskom os veSCe nii [Aspects of Tsaxur Gramma r From a Cross-Ling uistic Perspective]. Moscow: Nas ledie .

Kuroda, Sige-Yuki. 1965 . Generat ive Grammatical Studies in the Japanese Language . Doctoral dissertation, M IT, Cambridge, Mass .

270

Backwards Control: Evidence from Malagasy

Kuroda, Sige-Yuki . 1978 . Case marking, canonic~ ! sentence patterns, and Counter Equ in Japanese (a preliminary survey). In John Hinds and Irwin Howard (eds) , Problems in Japanese Syntax and Semantics, Tokyo: Kaitakusha, pp .. 30-5 1.

Law, Paul. 1995. On grammatical relations in Malagasy control structures. In C. S Burgess , K. Dziwirek, and D. Ge rdts (eds .) , Grammatical Relations . Theoretical Approaches to Empirical Questions, Stanfo rd: CSLI , pp. 27 1-290.

MacLaughlin, Dawn. 1995. Wh-movcment in Malagasy : An extrac tion asy mmetry. lr Akinbiyi Akinlabi (ed.), Theo retical Approaches to African Ling uistics Trenton, NJ : Africa World Press, pp. 11 7- 128.

Manzin i, Rita and Anna Roussou. 1999. A Minimalist theory of A-movement and control. University College of London Depa11ment of Phonetics and Linguistic~ Wo rking Papers in Linguistics 11 : 403-440.

McCloskey, James and Peter Sells . 1988. Control and A-chains in Modem Irish. Na tura. Language and Linguistic Theory 6: 143- 189.

O ' Neil , John. 1995. Out of control. In NELS 25. Amherst, Ma.: G LSA, Uni versi ty of Massachusetts, Amherst, pp. 36 1-171.

Paul , Il eana. 1999. Malagasy Clause Structure. Doctoral dissertation, McGi ll University. Paul , Ileana and Lucie Rabaovololona. 1998 . Rai sing to object in Malagasy. In Ileana

Paul (ed .) , The Structure of Malagasy fl . UCLA Occas ional Papers iz Lingui sti cs 20, Los Angeles: UCLA Department of Linguistics, pp. 50-64.

Paul, Il eana and Jeannot F. Ranaivoson. 1998. Co mpl ex verbal constructions in Malagasy . In Il eana Paul (ed .) , The Structure of Malagasy fl. U C L A Occasional Papers in Lingui stics 20, Los Ange les: UC LA Department o f Linguisti cs, pp. 111 - 125 .

Pearson, Matthew. 1998. Rightward object shi ft and the syntax of adverbs. In Ileana Paul (ed.). The structure of Malagasy fl. UCLA Occasional Papers in Linguisti cs 20. Los Angeles: UCLA Department of Linguistics, pp. 34-49.

Pearson, Matthew. 2001. The Clause Structure of Malagasy: A Min imalist Approach. Doctoral dissertation, UCLA , Los Angeles ..

Pensalfini , Robert. 1995. Malagasy phrase structure and the LCA. In Robert Pensalfini and Hiroyuki Ura (eds.) , Papers on Minimalist Syntax. MIT Working Papers in Linguistics 27, Cambridge, Ma.: MITWPL, pp. 209-222.

Perlmutter, David. 1970 . The two verbs beg ill. In Roderi ck Jacobs and Peter Rosenbaum (eds.), Readings in English Trw zsformational Grammar . Waltham, Mass.: Blaisdell , pp. I07- ll9.

Perlmutter, Dav id and Paul Postal. 1983. Toward a uni versa l charac terizati on of pass ivizati on. In David Perlmutter (ed.), Studies in Relat io1za l Grammar 1. Chicago: Uni versity of Chicago Press, pp. 3-29.

Poli nsky, Maria and Eric Potsdam. 2000. Backward control in Tsez. In R. Bill erey and B. Lillehaugen (eds.), WCCF / 9. Somerville, Ma.: Cascadilla Press , pp. 357-370.

Po linsky, Mari a and Eric Potsdam. 200la. Backward control. Ms. Uni versity o f California, San Diego and the Uni versity of Florida.

Polinsky, Maria and Eri c Potsdam. 200l b. Backward cont rol in Malagasy . Ms. Uni ve rsi ty of California, San Diego and the Uni versity of Florida.

Rackowski, Andrea. 1998 . Malagasy adve rbs. In Ileana Paul (ed.), The structure oj Malagasy Il. UCLA Occasional Papers in Lingui stics 20, Los Angeles: UCLA Department of Linguistics, pp. 11-33.

Randri amas imanana, Charles. 1986. Causatives of Ma lagasy . Honolulu : Uni versity of Hawai i Press.

Ura, Hi roy uki . 1998. Checking, economy, and Copy-raising in lgbo. Linguistic Analysis 28: 67-88 .

Yoon, James. 1996. Ambiguity of government and the Chain Cond iti on. Na tu ra l Language and Linguistic Theory 14: I 05-162.

27 1

Maria Polinsky and Eric Potsdam

Mari a Polinsky UCSD Lin guistics Department 9500 Gilman Drive, #0108 La Jo ll a, CA 92093-0108

polillsky@ ling. ucsd. edu http://lillg. ucsd. edul-pol i11 skyl

Eric Potsdam Linguistics Department P.O. Box 115454 Uni vers ity of Florida Gainesville, FL 32611

potsdam @ ufl. edu htt p:l/web. clas. ufl. ed~t!users/potsdam/

272