Adopted from pdflib image sample (C) - DSpace@MIT

PARAMErERS AND EE'E'~ OF WClID ORIER VARTATlOO

by

Lisa de~ Travis

B.A. Yale University(1974)

SUBMI'ITED IN PARI'IAL EULFILIMENrOF THE REOOlREMmrS OF THE

DEX3REE OF

at the

M1\SS.l\CHUSEITS INSTITtJrE OF TECHl\l)IOOY

Jtme 1984

@) Lisa deM:ma Travis 1984

'!he autoor hereby grants to M. I.T. pennission to reprcrluce arrl todistribute copies C'f this thesis dccunent in whole or in part.

Signature ofAuthor

) I

Depar~t of Lirguistics am Philosophy19 June 1984

5l>

certifiedby---------....,.",.------1'-....,.------.......--------

Noanl Chansky'1'hesis SUpervisor

..Ii? r :::::::Y:Accepted by

-----.---------¥--~-~--....---+lt----------

samuel Jay Ke}serMA5sAcH~~r~sT~innan Den:ar+-\:lental CoTmitteeOF TECHNOLOGY , ~ \.4

JUN 271984L\BAARtES

2

It is difficult to know \tJhere to start and stop thanking people since somany have had an effect on me leading up to this thesis.

'Ib my camri.ttee who al~ys stood by: N:>am O1ansky helped me reachfurther than I tb:>ugh PJssible and gave me supp:>rt \tJhen I rrost needed it.I.lligi Rizzi treated me to his infectious love of a good puzzle. Wayneo I Neill provided steady c.nuncil and gCXJd hurror. M::>rris Halle supplied awisdan, ling1.rl.stic and otherwise, that I ooj?e never to lese.

Also, Ken Hale, even in absentia, guided me in my linguistic interests.His love of and resj?ect for the less studied languages is as i.mp:>rtant tome as any theoretical insight. Arrl Paul Kiparsky had faith that arrong thecrazy ideas was one that might \\Ork.

Generations of linguists have passed through MIT during my four yearsand every one of them had sanething to offer ItlCe. 'll1e list 'below iscertainly not canplete and I ap:>logize to anyone inadvertantlyanitted.

In my generation, I thank Nigel Fabb for chocolate cake and energy tospare, Jeri Kisala for a type of supp:>rt that is all her C'IWIl, MaIroru Saitofor a colleague and an example I admire witb:>ut limit, and Mariofvbntal'betti for p:>etry and wit.

In past generations: Ha.git Borer, Denis Bouchard, Jill ])mcan, JimHuaIlCJ, Judy Kegl, Beth Levin, Al~ Marantz, M)hanan, IB.T,;?id Pesetsky, mugPulleyblank, Malka Rapt1ap:>rt, Anne RJchette, Ken safir, Barry SChein, TimSt£Jwell, Craig 'Ihiersch.

In future generations: Diana Archangeli, Mark Baker, Andy Barss, MaggieBrowning, Isabelle Balk, Kyle Jolmson, Juliette Levin, Diane Massaro, TovaRap:>p:>rt, Peggy Speas, Richard Sproat, and Betsy Walli.

The visitors and passers-by at MIT are as i.mp:lrtant as any other facet:Mriana Belletti, Peter Q:lopnans, Joe Ehonds, SUzanne Flynn, Bob Freidin,Yosef Groozinsky, Mike Hantoond, Celia Jacubowicz, Dany Jaspers, HildaKcx:>pnan, Jalclin Kornfilt, Dave Le'beaux, Pino longobardi, canne Picallo,Patti Price, Carles OXicoli, Eric Feuland, Ian Ro'berts, Esther 'Ibrrego,Eric wehrli, and John Whitrnan.

I thank in particular Bob Freidin, Tova Iep:>p:>rt, Dany Jasj?ers, BarrySchein, and Geoff Pullum for careful readings of pJrtions of this thesis.I will probably regret not follCMing their advice rrore closely.

A special thanks goes to the carmtmity of \\OTten here at MIT. '!hey, rrorethan any other group, have made the past year 'bearable.

. Thanks to Tova, Anne, Janet, Dany, Jaklin, Beth, and Kyle for lastminute help and supp:>rt.

'lb Maggie Carracino al'rl Nancy Peters for the much rrore than necessarythat they do, and Ibn Wilson for infinite patience. I thank my studentsat Next Ibuse for 'being a breatlL of fresh air.

'Ib my Malagasy infonnants: Charles R3ndriamasimanana, Rina Fajoharison,and .N:>ro Ramahatafandry i and my Yiddish informants, ca.vid Fishman andSylvia Freid.

'lb Glyne Piggott whose enthusiasm started me off, and Filwin Williamswhose enthusiasm kept me going.

To friends: Anne, who taught me the Irost imp:>rtant thing I know aboutlirguistics i Gideon, who got me to laugh ·Nhen I didn't feel like smiling iJanet, who gave generously of her time and insights i and Beth for Anneniantreats and endless enC'Ouragenent. Also, for standing by - Debbie, Cirrly,Bob, Yossie and Steve.

With much love to my parents and the rest, fran engine to caboose:Peter, June, Sean, Jared, Matthew, Faith, Ward, Julia, SUzarme, Claire,Mark, Dianne, Jerany, sara, and Lew. '!hey knew I could do it even \\lhen Ihad doubts.

Finally, to Irving, \\lho helped me make sense of it allde I' aube claire jusqu' a la fin du jour ••

3

4

PARAMETERS AND EFFECTS OF WORD ORDER VARIATIONby

Lisa deMena Travis

Suhnitted to the ~partment of Linguistics and PhilosophyJune 1984 in partial fulfillment of the requirements

for the D3gree of Ibctor of Philosophy

ABSTRACT

'!his t11esis investigates different issues raised by the problans oflanguage specific tM:>rd orders. The aim of Chapter 2 is to replace PhraseStructure Rules which stipUlate the order of constituents. '!he D:rnainAdjacency Condition (DAe) predicts tl1e order of non-head constituents inrelation to one another, while the parameters of the direction of caseassignment and Q-role assignment, together with thehead-initial/head-final parameter, establish the order of non-headconstituents in relation to the head. The parameter of direction ofpredication accounts for the p:>sition of a subject in relation to itspredicate. Word order typology is row rrore detailed because the S,-V-Otype of description is expanded to include the INFL nC>de, subcategorizedPPs (PPls) and non-subcategorized PPs (PP2s), and it is rrore restrictivebecause languages must fit into such a parametric descr;~tion.

Chapter 3 presents an accx:>unt of Gennanic word order which reliesheavily on the INFL nC>de in toth the synchronic and diachroniccharacterization. '!he Head t-bvanent Constraint .is intrcrluced to accountfor the rrovenent of V into INFL, and INFL into o:MP. In sane historicalspeculating, it is s"l1cMn that contradictory demands on INFL create tensionin granmatical systems and pranote reanalysis of d-structures.

Chapter 4 extends the notion of proper goverrnnent to include the propergovermLlent of a subject NP by an adjacent VP. '!his account relies on theDAe and a functional defini,tion of proper goverrnnent. It is s"l1cMn thatthis type of proper government, canplanent goverrnnent, explains not onlyes-insertion in Gennanic languages, but also extraction facts in Italianand O1am::>rro, and pro-drop facts in Italian and Irish. It is furthernoted that while the EJ:p requires the notion of canplanent goverrrnent, tl1eCEO still requires the rrore restricted notion of proper goverrrnent,lexical goverrnnent.

O1apter 5 presents several problems related to pleonastic.:s including theinteraction of pleonastics with verb agreanent, case assignment, and chainfonnation. It is claimed that there are t\\O types of pleonastics, theI-type and the T-type, and that these represent a l1ierarchy of featureswhich divide pleonastic constructions in predictable ways.

Thesis Supervisor: Nc>am ChanskyTitle: Institute Professor

TABLE OF CONI'ENrS

Chapter 1 INTRO~IOO

1.1 Introouction to Goverrnnent Binding 'Iheory1. 1. 1 Granmatical MXlel

1.1.1.1 ~structure

1.1.1.2 S-structure1.1.1.3 Predication1.1.1.4 PF1.1.1.5 IF

1.1.2 Projection Principle1.1.3 Goverrment

1.1.3.1 C-oammand1. 1. 3. 2 Proper Goverrment and the EX:P

1.1.4 Governnent and PRO1.1.5 Case Theory

1.1.5.1 case Filter

1. 2 Introduction to the Thesis1.2.1 Chapter 21.2.2 Chapter 31.2.3 Chapter 41.2.4 Chapter 51.2.5 Chapter 6

Chapter 2 WORD ORDERp~ AND TYroLOOY

2. 1 Constituents of the Verb Phrase2.1.1 Facts

2.1.1.1 Archaic Chinese2.1.1.2 MOdern Mandarin

2. 1. 2 The Contrcwersy2.1.3 Parameters in Chinese

2.1.3.1 Archaic Chinese2. 1. 3. 2 MOdern Mandarin

2.1.4 L&Tis Speculations2.1.4.1 Preposed Objects2.1.4.2 Case Barameter

2.1.5 Conclusion2.1.6 Postscript

2.2 Huang's Analysis2.2.1 Problens for the case Analysis

2.2.1.1 Adjectives2.2.1.2 PF Adjacency

2.2.2 Problens for Huang's Analysis

9

99

101213151516181820212222

232324242627

28

39404041434444455656565961

6263636568

5

2. 3 Branching

2.3.1 Ordering2.3.2 Constituency

2.3.2.1 M:>ve-a2.3.2.2 Pronominalization2.3.2.3 Cb-ordination

2.3.3 Sumnary

2. 4 Constituents of !NFL I

2.4.1 NP am. !NFL2.4. 2 !NFL and VP2.4.3 NP and VP

2. 5 l'hrd Order and TyJ;x:>logy2.5.1 Steele2.5.2 Possible counterexamples

2.5.2.1 Diola-Fogny2.5.2.2 Chinese2.5.2.3 Chorti

2.6 Discussion2.6.1 Sumnary2.6.2 Questions

Chapter 3 WORD ORDER IN GERMl\NIC IANGUAGES

3.1 Gennan and D.ltch: the accepted analysis

3.2 Yiddish3. 2. 1 Yiddish is not !NFL-final3.2.2 Yiddish is rot INFL-initial

3.2.2.1 Pronouns3.2.2.2 Extraction

3.2.3 Yiddish is S-I-VP

3.3 Gennan Revisited

3.4 'lbeoretical M:>tivations3. 4. 1 r-bvanent of heads3. 4. 2 !NFL ncvanent

3.4.2.1 !NFL ncvanent: Left3. 4. 2. 2 :rNFL rrovement: Right

3.4.3 Sunmary

3.5 Historical Speculations3.5.1 Ass~ons

3.5.1.1 ~ and V adjacency3~5.1.2 Case adjace~

3.5.1.3 unity of case assignment direction3.5.1.4 Inoonsiste~

3.5.2 Gennanic languages

71

73777780B485

87888990

9293959699

101

103103104

108

110

114114115117118119

120

129130135135139145

146146147147147148148

6

3.6 Conclusions am Typ:>logical Speculations

Chapter 4 ~) GJ~

4. 1 Fs Inser1:ion4.1:1 Facts

4.1.1.1 Gennan4.1.1.2 Yiddish

4.1.2 safir's analysis4.1.3 VP gov,~ent

4.2 Government4.2.1 Identification

4.2.1.1 Gap identification4.2.1. 2 Recoverability of features

4. 2. 2 VP Goverrment4.2.2.1 Fbrmal definition4.2. 2.2 Identification va. Features

4.2. 3 Structural government4.2.4 ~irical consequences

4.2.4.1 Italian4.2.4.2 INFL+V peripheral languages: Chanorro4.2.4.3 Irish and null sUbject languages

4. 3 Revision of the CEO4.3.1 Spanish4.3.2 Chinese4.3.3 Welsh4.3.4 CEO - revised

4.4 Conclusion

Chapter 5 PLIDNASTICS

5. 1 'Ihe hierarchy5.1.1 Justification for the hierarchy

5. 2 Pleonastics: language specific5.2.1 ~y sUbjects

5.2.1.1 INFL features5.2.2 Qptional pleonastics5.2.3 I-type vs. T-type5.2.4 Irish

5.3 T-type pleonastics5.3.1 Revie\t1S5.3.2 T~ype pleonastics are case5.3.3 Analysis

5.4 I-type pleonastics5.4.1 ES va. ES/O in Gennan

155

159

162162163164165166

169170173174175178184187197197200206

209210211212213

214

216

217218

226227229231232234

237237241247

252253

7

5.4.2 BET vs. ER in tutch5.4.3 Analysis

5.4.3.1 Passive'5.4.3.2 Argument SiS

Chapter 6 CO~USION

6.1 Ibnains

6. 2 Imrx>rtance of !NFL

6. 3 Word order effects

255257258261

268

268

270

272

8

Cllapter 1

ntis chapter is intended to intrcrluce the reader to the content of

the thesis and to review several concepts alld definitions of the

Government Bindirg nteory (GB) of Chansky (1981). It is, l1aYlever, not a

sufficient intrcrluction tc) the theory. For further details I refer the

reader to Lectures on Governrnent am Binding (1981) (LGB) and references

cited therein.

1.1 Introduction to GB '!heory

1. 1. 1 '!he Granmatical M:xiel

The graIllMtical m:de1 presented in IGB is giVeIl in (1). The granmar

is divided into several eatlIX'nents in the following tnnfiguration. In the

discussion below, we will see how'these canponents fit into the research

of this thesis.

9

(1) D-icture

S-structure

PF

10

1.1.1.1 D-structure

D-structure is generated by beise rules (see Jackendoff 1977). The

fonna.t of these beise rules is restricted by X I -theory am the

sUbca'tegorization frames of the lexical items.

X I -theory will insure that every category has a head whi.ch carries

the same categorial features as the maximal projection. The beisic rule

template is:

(2) >fl -> ... >fl-1 ••••

This template captures the generalization that an NP has an N as its head,

am a VP has a V as its head, etc.

SuperiItt.l;nsed en this template are the sUbcategorization frames of the

heads. For instance, a base rule for a verb phrase m.ight 'be represented

as shC1#Jn 'belO'#l, rneanil1CJ that a verb phrase consists of a verb and an

optional NP.

(3) VP -> V (NP)

Ibwever, since the verb smile does not subcategorize for an internal

argmtent,. if the V in the rule alx>ve is sm.Ue, the VP internal NP may not

'be realized. Similarly, the verb donate l-equires an internal argunent,

therefore the expansion of the rule al:x:Jve \'ihich does not include an NP is

rx>t possible.

'!he theta-eriterioll will insure that donate has the argunent it

requires, and the Principle of Full Interpretation will insure that sm.i.le

will not be generated with a VP internal NP.

Verbs assign theta-roles (a-roles), or thematic roles (see Gruber

1965) to their argunents. Since subcategorization is the syntactic

realization of these argunents, by requiring that all the 6-roles of a

verb be realized, we are requiring that all subcategorized elanents be

realized. '!he '!heta-criterion is (:Jiven below.

(4) '!heta-criterion (1GB, p.36)

Each argunent bears one and only one 6-role, and each 6-roleis assigned to one and only one argunent.

(5) a. *Irving donated.b. Irving donated his time.

'!he l-eason that (Sa) is ungranmatical is that donate has an internal

o-role to assign, but no argument to assign it to. In (Sb), this o-role

may be assigned to the VP-internal NP his time.

'!he 'lheta-criterion requires that all 6-IOles be assigned, and that

all arguments be assigned o-roles. '!he latter part of this requirenent

also appears in the Principle of Full Interpretation \\hich basically

insures that every constituent has a function in the sentence.

(6 ) Principle of Full Interpretation:

At PF a.n:1 LF, every' element must be licensed by saneappropriate interpretation \'Jhere interpretation can beachieved through

(i) subcategorization(ii) predication

If an elanent is genera'ted at D-structure without being

subcategoriZed, it must then be predicated of sane other elanent.

11

12

(7) a. *Sara smiled I.ew.b. Sara smiled.

In (7a) above, the st..ructure is ruled out because lew j.s not licensed

by subcategorization or predication (we will see ITOre on predication

below) •

D-structure is assumed to be a Itdirect representation of GF-oll

(IGB,p.43) meaning that e-roles are assigned to graI11llC:ltical ftmctions

«GFs), such as Subject, obj~t), and that these GFs (GF-8s) must be

filled at D-structure. GFs which are not assigned e-roles (such as

pleonastic subjects, see O1apter 5) are not filled at D-structure since

they are not required to be so by the e-criterion or the Projection

Principle (see below).

Since this thesis is concerned with word order and typology, much of

the discussion will center on the d-structure level of the granmar. In

Chapter 2, I examine the restrictions on the base rules in llOre detail and

will address not only the issue of what elanents are included., but also in

which order they appear. In C1apter 3, I argue for ad-structure 1NOrd

order for Gennanic languaget:3 that differs fran other accounts ani. I

speculate on the historical changes that produce this 'NOrd order. My view

of historical change assunes a large role for d-structure arrl the

cx:>nstraints put on IX>ssible d-structures.

1.1.1.2 S-structure

S-structure is created fran D-structure by a general ~e-a (or

13

Affect-a. (see Iasnik & ~;8aito 1984» rule \tJhich says tnQ\1e any category

anywhere. M:lvement wil"1 be restricted by other systens of the gramnar

such as sUbjacency, bincling, etc (see LGB for discussion of these

restrictions) .•

The level of s-structure is irnI;x:>rtant to this thesis in the way that

it obscures d-structure. In the discussion of typolCXJY (Chapter 2), we

will see that sane apparent 'NOrd order variation.s !X'sited by Steele (1978)

are due to llOVanent rules. '!his places the level at which these

variations appear at s-structure, then, not at d-structure.

The llOVanent rules of S-structure are also imp:>rtant to the vie'l/ of

'NOrd order presented in this thesis since I claL-n that the restrictions on

llOVanent (such as the Ehpty Category Principle discusse:i below) are

affected ~.l \\Ord order variations. '!his is the central concern of Chapter

4.

1.1.1.3 Predication

Williams (1980) argues for a level of predication structure (PS) oo.to

Which s-structure is mapped. Since pre:iication is :i.mp:Jrtant for the

Principle of Full Interpretation (see above) as a licensing mechanism, and

since elanents IlUlst be licensed at l:x:>th PF am LF, I will assure that PS

fits into the granmar as shown below.

(8) D-structure

I ~S-structure

fPS

....---.PF LF

Williams' Rule of Predication and his C-cx:mnand Condition on

14

Predication are given belCM:

(9) Rule of Predication (Williams 1980, p. 206 )

Coindex NP and X.

(10) '!he C-a:mnand Condition on Predication (Williams 1980, p.206)

If NP and X are coinc1exed, NP must c-cannand X or a variable of X.

FDthstein t.i.ghtens the c-cx:mnand condition adding that the predicate'\

must also c-ccmnand the sUbject, Where Williams requires the l<X>ser nc>tion'

of c-subjacency. (In footnote 1, l'xJwever, Williams also mentions the

IX'ssibility of needing mutual c-cannand for predication.) Her rule of

PredicaLe Linking is given below:

(11) Rule of Predicate Linking (for ED]lish)(Rothstein 1983, p.27)

a. Every non-theta-marked XI? rm.tst be linked at s-structureto an argunent Which it intnediately c-a:mnands and whi.chimnediately c-cannands it.

b. Linking is fran right to left (i.e. a subject precedes its predicate l .)

Examples of predication are found in the following examples.

(12) a. [Jonathan]i [did the puzzle]i

b. Janet ate [the carrots]i [raw]i

1. 'Ihis will be subsumed in the parameter of predication in Olapter 2.

c.

d.

Miriam hamnered [the nail]i [flat]i

Bruce considers [Karen]i [brilliant]i

15

Both Williams and PDthstein show that any category may be a predicate

(examples are taken fran Williams 1980, p. 206).

AP: John merle Bill sick.NP: John made Bill a doctor.PP: John kept it near him.VP: John died.

Predication is :i.mIx:>rtant in the thesis, roth as a directional

parameter for \\lOrd order (Chapter 2), am. as it contrasts with

exmplanentation (Olapter 4).

1.1.1.4 PF

PF is the phonological canp:ment of the gramnar. PF interacts with

\\Ord order throtgh stylistic rules (see Rxharont 1978). Because of the

position of these rules in the PF canp:ment of the granmar, they must

£011011 all syntactic rules and nust not affect 12 interpretation. These

rules are :i.mIx:>rtant to my view of historical \\Ord order change as

discussed in Chapter 3.

1.1.1.5 12

12 is the level of the gramnar which represents a type of lCXJical

fonn. For instance, questions \\Ould have the a-structure am. IF

· · bel 2representat10n g1ven CM.

16

(13) a. What telePhone. did Gideon bu¥ t.1 1

b. for Which x: x a telephone, did Gideon buy x?

RLtles that map s-structure onto the level of LF are Quantifier

Fai.sing (May 1978), Focus Interpretation, and, in sane languages,

wh-rtDVement (Huang 1982). 'lhese rules do not affect the \\Ord order of a

la.ngtBge since, lY.I the nOOel given here, they occur independently of the

poonetic representeation. H:1wever, LF is i.ntpJrtant to this thesis since,

like the nnvernent rules of s-structure, the ITOVement rules of IF are

subject t.D restrictions that may be affected by \\Ord order (see Chapter

4) •

1.1.2 ~jection Principle

(14) Projection Principle (WB, p.29)

Representations at each syntactic level ( i .e ., IF, and D-am S-structure) are projected fran the lexicon, in that theyobserve the subcategurization properties of lexical itans.

2. I leave out a detailed bracketing of structures until later chapters.

(WB, p.38)

Given[9 a b ]

[9 b a ]

17

(i) if b is an imned.iate constituent of 9 at L.,1

and 9 = a', then a 6-marks bing

(ii) if a selects b in 9 as a lexical property, then

a selacts bing at L.~

(iii) if a selects b in 9 at L., then a selects1

bing at L .•J

(\filere L. and L. range over syntactic levels)1. J

I use the Projection Principle and the '!heta-eriterion as a fonn of

proper goverrment in that they seIVe to identify gaps. If a constituent

has a 6-role to assign, by the '!heta-eriterion there must be an argunent

to receive this 6-role. At d-structure, the argument will bear the GF

which is assigned the 9-lU!e. ~ver, s-structure, through rrovanent, may

obscure this relationship. '!he Projection Principle insures that the

Q-role is assigned at every level, meaning that the IX>sition must be

there, even if its contents have rroved. In this way, the gap is required,

and tllerefore identified.

'!he Extended Projection Principle simply adds to the Projection

Principle the stipulation that all sentences require subjects. We have

seen above that I<>thstein subsumes this requirement in her Rule of

Predicate LinkiB3.

1.1.3 Goverrrnent

Government is a structural notion. Chcmsky' s definition is given

below.

(15) Cb:msky (1981)[a governs gin]

[b ••• g... a... g... ], where

(i) a = >f

(ii) Wlere b is a maximai projection, if b daninates g,then b daninates a

(iii) a c-cx:mnands 9 (p. 165)

Since this definition depends en the definition of c-a:mnand, below

\aJe lex>k at three different definitions of c-eatma1n.

1.1.3.1 C-oammand

The follOiling three definitions of c-oammand mcike very different

predictions concerning c-cannand relations.

(16) Reinhart (1976)Ncrle A c-cx:mnands node B if neither A nor B daninates theother and the first branching rnle daninating A daninates B.

(17 ) h:>Ln1 am Sfx:>rtiche (1981)A c-cx:mnands B if A and B are X", A-+B, and every In.J.Ximal

proj~tion daninating A daninates B.

18

19

(18) Chomsky (1981) (p~166)

a c--a:mnands b if and only if

(i) a does not contain b

(ii) suppose that 91 , ••• , 9n is the maximal sequence such that(a) 9n =a

(b) 9. =aj

1-

(e) 9i inmaiiately dcminate 9i+1

'!hen if d dan.i.nates a, then either

(I) d daninates b, or

(II) d = g. and 91

dan.i.nates b~

In the structure given below, 'Ne can see the difference in the

predictions. Reinhart I S definition deperrls crucially on branchi.ng, A&S' 5

definition depends on maxi.maJ. projections, and 01ansky' s definition

deperrls on the projection of like categories.

(19) V"

V~XIIWith} ""UII

~W' y"

~W Z..

For Reinhart, W c--a:mnands only Zll. This is because W' is the first

branching node that daninates W arrl the only other constituent that W'

daninates is Z". Z" will also c--a:mnand W.

A&S differ fran Reinhart in that Wwill c-caunand not only Z" but

also Y··. Since W· is riot a maximal projection, the c--a:mnand dcmain is

defined by the node above it, W.. (1), \lihich is a maximal projection. Z"

am y'l will also roth c-carmand. W.

Chomsky's definition of c-cx:mnand allows W a greater danain of

c-carmand. than either of t.he other definitions, but allows Zll a narrower

danain than in 1\&8 and the same danain as in Reinhart. '1lle difference

'between W am Zll is that W is a head. In Chansky's definition, heads

c-a::mnand every constit\lent in the danain of the highest projection in a

chain of projections that share categorial features. W, WI, W"(l), and

WI (2) are a chain of projections widch are all of the category W.

Reinhart IS danain of c-cannand for 1:x:>th W am Zit is that of WI since

WI is the first branching node for roth constituents. For 1\&8, the

c-carmand danain of 1:x:>th W am ZII is the danain of W" (1) since W.. (1) is

the first maximal projection. For Olansky, the c-ccmnand of W is the

danain of W" (2) while the c-cannand danain of z.. is WI.

The distinction between these definitions is discussed in Saito

(1983) • 'Ibis will also 'be reviewed :in O1a.pter 2 where we see that, by

changing the notion of branching, the definitions cane very close to

collapsing into one definition.

1.1. 3. 2 Proper Goverrnnent and the ECP

Propel: goverrnnent in 01ansky (1981) is a subset of government which

depeoos on the gO\Ternor.

(20) Proper government: (fran 1'Dtm & Sportiche 1981)A governs B if and only if A governs B and(i) A is lexical, or(ii) A if:f c.niooexed with B

20

21

Proper gavernnent is used in the Elnpty Category Principle (~) which

states:

( 21 ) Empty category Principle:[ e ] must be properly governed

Proper govermnent and the EX:P account for the following contrast.

(22) *Wl-1O did Cirrly say that t had a cat?(23) What did Cindy say that Ken had t?(24) Win t had a cat?

Since the empty category in (22) is governed only by !NFL which is

not lexical, it is not properly governed and is thus ruled out by the

EX:P. In (23), the enpty category is governed by the V, a lexical

category. Finally, in (24), the empty category is governed by an element

in CXMP whi.ch is coindexed with it. '!he defini.tion of proper government

will be discussed, and revised, in Chapter 4.

1.1.4 G:>vernment am PRO

In 1GB, government is used to account for the distribution of PRO.

'!he argunent is that, since PRO is both an anapoor and a pronoun, and

since an anaphor must be b:nmd in its governing category and a pronolID

must be free in its governing category, that PRO may not have a governing

category. In this thesis, ~ver, I will be assuming the frarnev.ork of

EDuchard (1982), in that I will be assuning that control PRO is an anapoor

which must be bound in its governing category, and arbitrary PRO is a

prornun Which IlU.1st be free in its governing category. Control verbs, like

raising verbs and EIM verbs, trigger S· deletion, accounting for the

similarity of birding in the structures given belorN.

(25) a.b.c.

Tim., seaned [ t, to be tired ]Tim~ expected rPftO~elf, to wi.n ]T' 1 led [ .1 ]:un, trl. S · to W1n

1 1

22

Rather than using govenunent to account for the distribution of PRO,

aJuchard uses case marki~. Since the sUbject NP in the enbedded sentence

of (25c) is not assigned case, it is not lexicalized and it surfaces as

PRO (see Bouchard for details). 'lhe i.rnIx:>rtant consequence of this is that

PRO may be governed. In fact, this is what I assume in languages where

subject NPs appear adjacent to the VP (see Chapter 4).

1.1. 5 case 'lheory

case is assigned to an NP by a lexical category which governs it. I

will be assuni~ that INFL[+tense], V, and P are case assigners.

1.1.5.1 case Filter

The case Filter states that every NP lTUst be assigned case. 'lhe

definition fran Chansky (1980, p.49) is given belO\t/.

(26) case Filter:*NP if NP has phonetic CXJntent and has no Case.

'lhe assigrment of case and the case fi1ter are crucial for the view

of 'NOrd order presented in this thesis. In Olapter 2, I claim that a verb

in a head--final VP may assign case to the right, causi~ the object to

nove to the right of the verb at s-structure. The assignnent of case is

also imJ;x>rtant to my vieN of pleonastics presented in Chapter 5. It will

be used to accolIDt for the distribution of empty pleonastics and to

23

accotmt for the agreanent of the verb with VP :lnternal arguments.

1. 2 Introduction to th~ thesis

1112.1 Olapter 2

Olapter 2 deals with the problem of Phrase Structure Rules. M:>re

specifically, it raises questions as to wether PS Rules are needed to

detennine the order of constituents within a maximal projection. 'Ib

account for the order of non-head const.i.tuents as they relate to one

another, I propose the D:main Mjacency Condition which insures that a

danain, such as a case danain or a 6-<kmai.n, will not be interrupted. 'Ib

accolIDt for the order of non-head constituents in relation to the head, I

argue, using data fran Chinese, that the parameters of direction of 0-role

assigrment, case-assigrment, and predication must be added to the

head-initial/head-final parameter.

. .Since an articulated version of branching has been used to account

for ordering of danains (Jackendoff 1977), and for the \\Ord order of

Chinese (Huang 1982), I investigate the need for intennediate projections

between >f and xnax.· The CX)nclusion I reach is that rn not.i.on of

constituency is needed other than maximal projections am lexical itans,

and that a rrore restrictive theory of gramnar results fran a restriction

of branching to these t\\O levels.

Finally I vie'N 'NOrd order typ:>logy as parametric variation, using

Steele (1978) as a starting IX>int. Certain laIXJuages fran Steele I s

language sample appear problematic, but I show that these problems

disappear through reanalysis of the data.

1.2.2 Chapter 3

Chapter 3 presents a view of Gennanic word order Wh.i.ch is different

fran current analyses • Evidence fran Yiddish suggests that Yiddish is

S-I-VP (Subject-INFL-VP). Since similar data are found in Gennan, I

suggest that Gennan is also S-I-VP rather than S-\1P-I. By having INFL

sentence internally, not only are these data accounted for, but

differences between Gennan anc:i Ureal" verb final laIXJuages can be

explained. 'lbe INFL node, the novanent of V into INFL, and the novanent

of INFL into cx::MP play crucial roles in nw account of Gennanic w:>rd

order. For the novanent, I proIX>se the Head M)vanent Canstraint which

states that an >f nay only nove into the ..p which properly governs it.

In the final part of Chapter 3, I offer sane specUlations on

historical change in Gennanic languages. Again, the !NFL node plays an

i.mIX>rtant role. I claim that the danarrls on INFL are mcmy and no

structure meets all of them. Because of this: JTOVanent of lNFL, or

novanent of another c..unstituent dependent on INFL, often results in a

reanalysis of the d-structure.

1.2.3 Chapter 4

Chapter 4 presents an investigation of sane consequences of 'tJ()rd

24

25

order differences. My claim is that, given the lXmain Adjacency

Condition, if a subject is adjacent to its 9-role assigner, the VP (i.e.,

a pl10netically realized INFL does not inteIVene), then it is in the

canplement danain of the VP. Further, I argue for a funct.ional definition

of proper governnent similar to Jaeggli (1980) or S1:c1Ire1l (1981). Given

these tw:> results, I argue that subject NPs that are adjacent to the VP

are properly governed.

This extended notion of proper goverPJnent, canplanent government,

will accuunt for LF extraction facts in Chinese (whi.ch is S-VP-I), and

s-structure extraction facts in Charrorro (which is I-S-VP). '!he proper

gOV'ernnent of J.X)st-verbal subjects in ltal.ian am 8,panish is also

accolIDted for, even without assuming ItOVement of the SUbject into the VP.

In the inverted structures, the order \\Ould 1:Je I-VP-S, a structure which,

in this acculmt, has a properly governed subject position. I also claim

that pro-drop facts are aCCDunted for. If tNe assume that pro, like other

anpty categories, needs to 1:Je properly governed, the fact that SUbject

inversion and pro-drop seem to co-occur in S-I-VP langua.ges is not

surpris~n:J• In I -VP-S structures, but not in S-I -VP structures, the

subject J.X)sition is properly governed. '!his will also explain why Irish,

an I-S-VP larguage, -has pro-drop but no subject inversion.

In the las~ part of Chapter 4, I discuss the need to distinguish

canplement gaverrlnent fran lexical governnent. Huang's (1982) Condition

on Extraction lXmains (CEO) is sensitive to lexical government rather than

canplanent governnent, since while SUbjects thanselves ma.y 1:Je extracted,

they are not J.X)ssible danains of extraction. '!he conclusion is that the

26

~p requires canplanent government of an anpty category while the CEO

requires lexical government of an extraction danai.n.

1.. 2.4 Chapter 5

O1apter 5 is a discussion of pleonastics, pranpted by the

investigation of Gennan pleonastics in Olapter 4. I assume that languages

have t~ types of pleonasticsi the I-type (like it in English), aI'..d the

T-type (like there in English). rrhe I-type of pleonastic in Gennan is es,

an.1. the T-type is 0, or the anpty pleonastic pro. I discuss a hierarchy

of constructions that take pleonastics, what types of pleonastics a

variety of languages use, and \\hich pleonastics they use for which

constructions.

I claim that the distribution of empty subjects is due to the

features of INFL which allCJ\tlS a four-way distinction in thf.~ "richness" of

!NFL. Italian !NFL, the richest, allows all pronaninal subjects (\'thether

referential or pleonastic) to be anpty, Yiddish INFL allows all

pleonastics (roth I-type and T-type) to be empty, Germc-m !NFL allows only

T-type pleonastics to be anpty, an.1. Erglish has no anpty subjects. Since

. rx:>t only must empty categories be properly governed, but their content

llUlst be identified, I assune that the ability of INFL to identify a

properly governed subject depends on the features that it carries.

I, then, discuss the ,problem of T-type pleonastics am bJw they

interact with case assigrment an.1. agreement on the verb. I reach the

conclusion that T-type pleonastics are simply a spell-out of case

27

assigrment independently of whether they foIm a chain with a VP internal

argunent or not. Agreement with an element other than the pleJnast.i.c

occurs only when a chain is fonn.~ bet~ this pleonastic and a

VP-internal argunent.

I-type pleonastics present a different sort of problen. '!he use of

this type of pleonastic differs in the laD;Juagos studi.Erl, as do the

extraction facts. D.1tch and Gennan ma.y use T-type pleonastics with

V(passive)-S' c:.Dnstructions 1Nhere Erglish Imlst use an I-type pleonastic.

Further, 1Nhenever an I-type pleonastic is used in D..1tch, extraction out of

the s' is not p:>ssible. 'lhis is not true in English I-type

constructions. I tie the difference in pleonastic distribution to the

p:>ssibility of havin:J impersonal passives in D..1tch and Geman. I, then,

accotmt for the extraction facts by claiming that word order differences

bet\tleen Geman and D..1tch on one hand and English on the other force

extrap:>sed S' s to join at different levels of the tree. A revised version

of Huang' s CEO will then aC(.l.)unt for the extraction asymnetries.

1.2.5 Chapter 6

Chapter 6, while reviewing p:>ints made througrout the thesis, also

highlights sane overall conclusions-. '!hese have to do with the i.ttlp:>rtance

of the !NFL node, the adjacency of danains, and the imp:>rtance of \tJOrd

order at all levels of the gramnar.

28

In this chapter I irwestigate problans of the base canponent and

specifically the problem of accounting for langua.ge specific word orders.

'!he goal is to avoid having to use stipu1ative means, such as the phrase

b.

a.

structure (or rewrite) rule below, to generate appropriate d-structures.

V" -> Vi - (have-en) - (be-ing) - (Advp) - Vi - «(:v~) - (S)

Vi -> V - (NP) - (Prt) - (f:\) -(t~p~) -(pp) - (pp) - (SI)

(1)

such a rule Ser'fleS many pl1r};X)ses. It sJ?eCifies (i) which elements

are obligatory, (i.e., the heads) (ii) which. elements are optional, (i .e. ,

the non-heads) (iii) What order tlle optional elements must have in

relation to one another, and (iv) What order they must have in relation to

the h,·,~. In the current GB literature, these tasks are factored out into

different sub-systans of the granmar.

'!he proFOsals in XI_theory (implicit in Harris (194:6, 1951) I explicit

in Cl'¥::msky (1970) am developed in Jackendoff (1977» acootmt for the fact

that NP expands into an obligatory N and other optional elements while VI?

exparrls into an Obligatory V and other optional elements. '!his is

captured within the basic phrase structure rule of (2) •

29

(2) ~ --> ••• xn-1 •••

AsSmt'ling that N, V, P, A, CCMP, and !NFL all enter into such a rule,

we have the resie template £01: all the phrase struc.:ture rules. I assume

that S I is equivalent to CXl-1P e and that S is equivalent to !NFL'.

What ma.y appear in the ellipses of (2) above falls under the

Principle of Full Interpretat.i.on (Olansky fall lectures, 1983) which, as

ha · Cha l' 1£ ~.' 'd ' 1we ve seen:tn .pter I 1tse u.1.V1 es :Lnto t'ttU parts.

(3 ) full Interpretat.i.on :

At PF and IF, every element must be licensed by saneappropriate interpretation, Where interpretation canbe achieved through

(i) subcategorization(ii) predicat.ion

'!he lexicon is responsible for indicating Which elenents are

subcategorized, and predication theory (Williams 1980) accoLUlts for those

elements licensed through. predication. Lcx:>king first at

s~ategoriza.tion, \tie can see that by having the lexicon as a sub-systen

in the base (Chansky 1965), we can derive further infonnation concerning

phrase structure rules. Ttnugh sane categories select canplements

independently of lnw they themselves are lexically filled (CXMP always

selects I NFL I , am !NFL always sel~s VI) I other categcries select a

differant canp1ement structure depending on the ~exical itemG Th.i.s is

1. In thi.s thesis I will have little to say about specifiers; that is,lexical non-hea:ls. Basically, I wi.11 be dealing with wtlat Ennrrls (1976)calls recursive c:.Dnstituents. I assume that the ideas presented. herecould be extended. in sanE! lla'tural way to include non-recursive categories.

30

ItDst obvious in the VP. .Persuade requires an NP and an S I, believe can

take just an S', put requires an NP and a PP, hit requires an NF, and

sn.i.le nonnally takes no canplanents at all. we, then, have the following

CXlnfiguration:3 •

(4) V NP 5' [p:!rsuade Joan that Mary would cane]

(5) V S' [believe that Mary \\QuId cane]

(6) V NP pp [plt the book on the table]

(7) VNP [hi.t the ball]

(8) V [sm.Ue]

We can see that by overlaying the subcategorization frames of the

lexicon on the XI rule tanplate, we can fill in the cQuplenent structure

of the phrase structure rule.

rrtlrough XI theory and subcategorization, t\tJO redundancies have been

avoided. X' theory captures the getleralization that every non-lexical

projection must have a head (i.e., an element with the same categorial

features) • Subcategorization places the canplanent structure of a lexical

item in the lexicon rather than in the phrase structure rules2• Since the

cx:mplanent structure differs with the lexical item, we tNOuld expect such

idiosyncratic infonnation to appear in the lexicon. The NP in the VP is

I'X)t optional in the case of every verb as the phrase structure VP->V(NP)

suggests. Since infonnation about canplanent structur~ must be carried by

the lexical item, there is no need to restate this in the phrase structure

2. We still have to account for the fact the CDIjJP alwals takes I I am 1°always takes V'.

31

rules.

Those elements that are part of a rewrite rule, and Which are not

heads and not canplements, must, by the Principle of Full Interpretation,

be licensed through predication. We have seen examples in Olapter 1, sane,

of Which are repeated below.

(9) Janet ate the carrots raw.(10) 1-1i.riam harnnered. the nail flat.(11 ) Bruce considers Karen brilliant.

In any sentence, the SUbject p:lsition will be licensed through

predication. A subject is never a a:mplement as it is never in the

su1:x::ategorization frame of a lexical itan. Further examples of elenents

licensed by predication giVeIl above are E2'!!., flat, and I<"aren. In each

case, the elanent licensed by predication is in a predication relation

with an elanent which is licensed by a:mplenentation. In (9) and (10),

the elanent licensed by canplementation is the subject of the predication

relation (carrot, nail). In (11), ~ver, the a:mplanent is the

predicate (brilliant). In (9) and (10), then, the elanent licensed by

predication is the predicate (~' flat) and in (11) I the subject

(Karen) •

Theories of the lexicon, XI, and predication detennine what elements

ma.y appear withi.n the maximal projection, but there remains the question

of the order of these elements. Since much of the infonnation of Phrase

Structure rules foll()W'S fran other canponents of the grannar, we do not

want to resort to these rules to stipulate the order of the elements.

The problen of \\Ord order can be divided into t'tJO sub-problems. (1)

32

What order the maximal. projections (i.e., non-heads) must have in relation

to the head I and ( 2) What order they Im.1St have in relation to one

another.

The first problem is rrost easily solved in pure head-initial and

head-final laD3Uages. Greenberg I s stmy of \\Ord order (1963) sh:Ms that

V-o languages tend to be prep:>sitional (P-NP) I while o-V langua.ges tend to

be p:>stp:>sitional (NP-p). By categoriziBJ laD3uages as head-initial (V-o)

or head-final (a-v) I we capture these tendencies.

Usirg Malagasy as en example of a head-initial laD3uage and Japanese

as one of a head-final language, ~ can exenplify this \fJOrd order

distinction3•

(12) Malagasy: V-<l-SCG1P I -> CG1P!NFL' -> !NFLVI -> VN' -) N •••etc.

• •• fa manane ny ookin I ny mpianatra he an I ny ankizy Norothat PRFS+give the b::xJkIN the student to the children NJro

I that NJro gives the student I s l:x:ok to the children I

3. Ebr the time beiBJ, I am assuning that XI is ~. The reason.ingbehind this will be discussed in section 2. 3

(13) Japanese: S~VCCMP' -) CCMP!NFL' -> INFLV' --> VNt -> Netc.

CCMP'

~~N' V' I

~pi N' V

~N' P

33

Izum.i.-ga Hi.roko-kara tegami-o nnrat-te (keto)Izlllli.-NCM Hiroko-fran letter-ACC receive-PAST (the fa~ that)

'(the fact that) IZlIn:t received a letter fran Hiroko. I

[caMPI [I' N' [VI NI V] I] COMP]

'l.W:> further advances in the derivation of word order have been made

by Stowell (1981) am Pesetsky (1982).. Foth use the relation of case

assigrment to predict the position of elements. S~ll uses this

relation to predict the order of maximal projections in relation to one

another, while Pesetsky uses it to predict the order of the SUbject NF in

relation to the head, IO.

S~ll proposes the Parameter of case Adjacency and the Principle of

case Resistance. case Adjacency simply states that adjacency of sane sort

is a CX)ndition on case assignment. In other 'Nerds, if an NP is not

adjacent to a case assigner, then case ma.y not be assigned, the NP will

,receive no case, and the case Filter will be violated. The value of this

parameter ma.y be set differently for different laD3uages. In Warlpiri, no

4. koto, 'the fact that' I has been added to abstract away fran the effectsof discourse and topic.

adjacency is involved in the assignnent of case, as in (14) below 5. I

will be assuming that in languages with canpletely free word order, NPs

are inserted into a W* structure (see Hale 1980) with case already

assigned to them.

(14) warlpiri: a. Ngarrka-ngku ka wawirri panti-rniman-ERG PRES kangaroo spear-NPST

•The man is spearing the kangarco. •

b. wawirri ka panti-rni rgarrka-ngku.

c. Panti-rni ka ngarrka-ngku wawirri

(fran War1piri LexiCDn Project)

A rule of interpretation at IF will insure that every NP is licensed

either by being an argll'lleI1t of the verb, or by predication tllrough

coindexation to another element (see Hale 1982).

Lan:Juages that require adjacency for case assigrment nay requi.re

strict adjacency, as in Fnglish, or argunent adjacency, as in French.

(15) I like bcx:>ks very nnlch.*I like very much bcx:>ks.

(16) J' aime beaucoup les Iivres •I like very much the bcx:>ks

I I like bcx:>ks very much. I

*J I ai rnis sur la table Ie livre.I have put on the table the bcx:>k

•I put the bcx:>k on the table.'

In (15) we see that in Fnglish, nothing can .intervene between the

case assigner and the case assignee. In French, 'l"a.\rever, an adverb may

5. Adjacency may be involved if we consider the case endings on the nounto be the case assigners rather than the verb.

34

35

int:lrvene, but not another argunent. It is strict adjacency, then, that

is irnp:>rtant for Fnglish, and only argunent adjacency that is irnp:>rtant

for French.

When these adjacency requirements appear to be violated, it is

because of a rule of Heavy NP shift (or \tJhat S1:.cIY1ell argues is Fc>cus NP

shift) \\bich has noved the object NP to the end of the VP. The trace of

the novanent is still assigned case under the conditions of adjacency.

(17) I like t very much those 1:x:xJks that you told me to read.(18) J'ai mis t sur la table bouS les livres que tu rn1as donnes.

The case Resistance Principle explains, arocng other things, the

fX)sitioning of SiS sentence peripherally. S1:.cIY1ell claims that SiS (or any

category that ex>ntai.ns a case-assigning feature) cannot be assigned case.

Altrough I take issue with this profX)sal later (see Chapter S) I let us rrM

assume that 5' is sentence peripheral am not in an arglJllent FOsition. 6

(19) a. I told Joan softly that you were there.b. *I told Joan that you were there softly.c. I told Joan a story softly.d. *1 told Joan softly a story.

(20) a. That she is a writer convinced me to see her.b. *Did that she is a writer convince you to see her?c. Her writing ability convinced me to see her.d. Did her writing ability convince you to see her?

In (19) \E see that thl~ anbedded sentence appears after the matrix

adverb \\bile the NP object must appear adjacent to the verb in order to be

assigned case. In (20) 1Ne see that a sentential subject cannot undergo

6. StONell argues against this novement as being a case of Heavy NP orFocus NP shift. 'lhe Gennan examples belat.' argue further against thisanalysis.

SUbject-Awe Inversion, While a subject NP may.

In Gennan, the di.splacement of the 8' is even trore obvious.

(21) . Ich hebe es gesagt.I have it said

•I have said it. I

(22 ) Ich habe gesagt, daB sie gekatmen ,adil.I have said that she cane 'want

•I have sai.d that she wants to cane. I

36

In order to avoid being assigned case am thereby violating the case

Resistance Principle, the S I ha.s noved to an A' p::>sition to the right of

the verb.

Pesetsky (1982) takes St~ll's case Adjacency Parameter and extends

it to acc:ount for the p::>e;ition of the subject NP in English. Notice that

if English is truly head-initial as its VP and PP suggest, it should be

V-o-S like Malagasy. We need to explain, then, the S-V-O order of

English. Pesetsky suggests that since case adjacency is relevant in

English, and since the subject is assigned case by !NFL, the subject must

be adjacent, am. therefore to the left of INFL. He assumes a structure

given below.

(23) I"~

I' NP(no case)

~I VP

AV NP

case

I"

NP~£-->"-. / ~

I VP~

V NP

Presunably, larguages such as Malagasy \«>uld not have adjacency as a

condition on case assigrment otherwise we lNOuld expect them, too, to be

S-V-o. In fact, it may be true that Malagasy has no adjacency condition

37

on case assignnent, as shJwn in the follC7lling example.

(24) Ividianan-dRakoto ny vary ny ankizybuy IN Pakoto the rice the children

•The children were oought the rice by Rakoto. I

'!he verb ividianana is able to assign case to the object ~ Y2£l.. in

spite of the intervening NP, Rakoto. 7 La~k of an adjacency condition

allows sentence ini.tial INFL to assign case to the sentence final

subject.

x' theory and headedness, lexically specified subcategorization

frames, and conditions on case assigrment go a long way 1:cJY1ards tile

accurate description of \fJOrd order. In this chapter, ho\tJever, we will

lex>k at sane languages that do not confonn canpletely to these

prrameters.

In section 1, I will use evidence fran \\Ord order changes in CLi.nese

to argue for the existence of t'NO additional parameters: the direction of

case assigrment and the direction of 6-IOle assignnent (Travis 1983. see

alSJ &x:>pnan 1983a, 1983b for a slightly different prop:>sal). In this

sect.ion I will 1Je principally concerned with the C()nstituents of the VP.

In section 2, I canpa.re my analysis of tNOrd order in Chinese to

Huang I S ( 1982) analysis. Since the analysis he profOses is dependent on

7. I will suggest 1Jelow that it is direction of predication, not caseadjacency, Which accounts for the placement of the subject NP in English.If this is true, then we are not forced to say that Mcllagasy does not havean adjacency condition on case assigrment. 'Ihis seems rrore accurate sincein Malagasy, like in English and in :French, PPs may not intervene betweenVa am their direct objects.

38

binary branching, I investigate, in section 3, different notions of

branching. My conclusion is that there is not need for any level of

constituency bet\lleen maxi.mal projections and lexical itans.

In section 4, I examine the const.ituents of I I in light of the

parameters of section I'. I claim that case assigrment am o-role

assigrment within I I Cb not mirror exact!y the same relations within the

VP, am I argue for another parameter, the direction of predication.

An assumption that rtms through these sections is that diachroni.c

study offers insights into parameters in the same marmer in Which

synchronic study does. 'l\«> gramnars, whether divided by time or space,

ma.y be differentiated by setting the parametric values differently.

I examine in sections 1 and 4, and in nore detail in section 5, the

characterization or classification of languages by tNOrd orders such as

S-V-o, S-o-V, etc. In sect.ion 1, I show that a 'NOrd order controversy

dissolves When one allO\-1S o-V versus V-o order to be further detailed by

adding other c:onst.i.tuents to the word order specification, such as

s~ategorized PPs .(PPl) and non-subcategorized PPs (PP2). In, section 4 I

shaN that the addition of INFL to the S constituents greatly eases the

strain of accounting for the word order at the level of I I. In section 5,

I directly confront the problem of typology using Steele (1978) as a

springboard. '!he \\Ord order parameters of sections 1 arrl 2 provide a nore

restricted view of typology of d-structures. Restrictions on r-bve-a

constrain the PJssible \'tUrd order. variations within a larguage. Taking

languages within Steele I s language sample, I look at sane apparent

39

colIDterexamples and sb:M b:M they, in fact, confonn to the system I

pr0lX'se, dIld thereby not only CDnfinn that system, but also provide a rrore

enlightening analysis of the language in question.

2.1' Constituents of the Verb Phrase

In this section I use evidence fran Arc'haic Ol.i.nese (N:.) and two

stages of M:Xlern Mandarin (MMI am ~2) to argue for t\tJO word order

parameters: the direct.i.on of case assignnent and the direct.ion of a-role

assignment (see l(Dopnan 1983a,b for a similar analysis).

Evidence canes fran Archaic O1.i.nese and M:Xlern tvlandarin as described

by Li and Thanpson (1973a, 1973b, 1974, 1975) and L.ight (1979). A

controversy arises concerning the word order of M:Xlern. Mandarin: Li and

Thanpson (L&T) believe that Mandarin is IIaJ..nost II &JV While Light believes

that it is strictly 5\0. since there is di.sagreement as to what this

final \\Ord order is after the change fran AC has taken place, there is

disagreenent as to what change it is that has to be characterized. I

first discuss the facts of Archaic arinese and MJdern M2lrrla.rin \\Ord

order. I then discuss the \'lOrd order of MJdern f\1andarin, paying

partiCUlar attention to the placanent of objects. I..&T, in saying that

MJdern Mandarin is ..a1mc>st OOV", are cacmenting on a change they see

taking place. I take this as a further change beyond that described by

Li.ght. rrhere are t\\O changes, then, that I characterize: (i) Archaic

Chinese to r-tdern Mandarin (AC to r+tl), and (ii) MJdern Mandarin to a

40

projected stage of Mcmdarin (t+11 to r+12).

2.1.1 Facts

2.1.1.1 Archaic Chinese

(fran 11th century B.c. to 4th century B. C. )

w. In Archaic Chinese (Ae), objects and prePJsitional rhrases

always appear after the verb as shown in (25 ) .

(25) qu yue yu ciplay music prep here

[VOPP]

(Light, 162)

There is one construction that I&T mention, l1aNever, which they claim

is a relic of an earlier f!lJV order. '!his construction appears when the

object is pronaninal. In thi.s instance, the object appears before the

verb.

(26) min xian you shi. fu yu yi (L&T, 1973a, 208)people sage have ten persons I assist

I Ten of the wisemen axocng the people assisted me. '

[ S 0 V ]

Yu, I I I, appears to the left of the verb yi, I assist', even tlnugh other

objects appear to the right of the verb as shown in (25) aOOve.

HI? NPs are all head-final in AC. According to I&T I this is another

relic of an earlier rov stage. '!his is typified by the fact that relative

clauses, m:difiers, aOO genitive NPs all precede the head noun.

PP. PPs are head-init ..ial as shown in (25) and in the example belCM.

(27) chu yu you gu (L&T, 1973a, 200)anerge fran dark valley

[ V [ P NP ]]

The generalization, then, is that Archai.c allnese has head-initial

VPs arx1 PPs (altrough pronaninal objects appear pre-verbally), arx1

head--final NPs.

2 •1. 1. 2 r-bdern MaIrlarin

VP. It is in the VPs of M:xiern Mandarin (r-M) that one can see the

greatest change in \\Ord order. r-bst ~ly, IIOst PPa nr::M appear

preverbally where in :PC they appeared IXlst-verbally.

(28) zai zher yanzou yinyue (L.ight, 162)prep here play music

'played I1Ulsic here'

(29) cong you gu chulai (:L&T, 1973a, 200)fran dark valley anerge

'anerge fran dark valley I

[ PP V ]

NP. '!here seans to be little change, ho\eJever, in the order of the

constituents within NPs. Light indicates that they are still head final,

i .e ., the head notm is preceded by its m:::rlifiers.

PP. L&T claim that PPs now can be ~IXlsitional, a claim that I

reject below. Their evidence canes fran constructions such as the one

given in (30).

41

42

(30) \\0 zai chuang-shang shuijiao (L&T, J.973a, 204)I at bed· on sleep

•I sleep on the bed. •

Concerning these PPs, L&T write:

Post-Insitions in M.:rlern Chinese are derived frannaninals in Archaic Crlnese. L.ike aspect markers, they carryneutral tones in speech, denotin:;J their suffixal status.(L&T, 1973a, 204)

The difference between these Inst{X>sitions and the pre{X>sitions

already seen is obvious. '!he fOst{X>sitions are naninal in nature while

the preInsitions are true adInsitions. I discuss the :i.rrp:>rtance of this

distinction below.

CIljects. Objec:ts in M-1 deserve a special cxmnent since they are the

main cause of the contrO\Tersy between Light arrl L&T. '!he controversy

centers around a a)nstruction which all~ the object of a verb to be

placed preverbally if it is preceded by what L&T call an object marker,

ba. Examples are given in (31) and (32) bel()\t/.

(31) haizi ba fan chi-diao Ie (L&T, 1975, 179)child ba food eat-down ASP

I The chi.ld ate the food. I

(32) haizi ba shu mai Iechild ba 1xx:>k buy ASP

'The child ooU3ht the 1xx:>k. I

To sumnarize the non-c.x:>ntroversial Inints of~, PPs now appear

before the vf.!rb whereas in AC they always appE?4red after i--he. verb, and NPs

are still head-final. '!he controversial {X>ints concern (i) 'lJhether the

object appears before or after the verb (Ba objects appear before the

verb I bare objects after the verb) i and (ii) whether the PPs are

prep:>sitional or p:>st{X>sitional (many PPs are still prefOsitional, yet it

43

appears that there has been an influx of p:>stp:>sitions).

2 .. 1. 2 The Controvers,y

Light sums up the CX)ntroversial issues as follO'NS::

There is no question that between Old Chinese am ~ooern

Mandarin there was a shift fran standard order of' SV(O)PP toSppv(O). NJr is there any question that preposing of objects,especially between subject and verb, was less carm::>n in OldChinese than nt:M. rn-te analytical questions are: 1'WJW dil1 thischange cane a1x:>ut, aYld \I41at does it mean? (Light, 163)

Li am~. L&T claim that I~ is ·very close to beiD,j an OCJV

language. Although the change has taken t\\O m.Ulenia to 'hag,:>en, they

suggest that the process is reM nearly eatplete. Their arguuent is as

follO'NS. If Olinese is IlDVing fran S\U to fJJV, preverbal objects should

becare IlDre camon. Also, any distinction that held bet'Ne!en preverbal and

p:>stverbal objects should now be encoded by sanething other than the

object •s p:>sition relative to the verb since soon only the preverbal

p:>sition "ill be possible. Finally, since the IlDVe is to a head-final VP,

we \\Quld expeCf. head-final reflexes elsewhere in the granmar I such as

placement of PPs. E-"idence that L&T give for exactly these changes is

that 1) the ba construction is becanirg 110re and nore eamon, 2) a

defin.i.te versus iOOefinite distinct.ion that used to hold of preverbal

versus p:>stverbal objects is beiD,j lost am replaced by the developnent of

indefinite and definite detenni.ners, 3) a semantic distinct.i.on between the

preverbal and p:>stverbal PPs is beirg lost and all PPs are appearing

preverbally, and 4) IDstp:>sit.ions begin to appear in the language.

Light. Light disagrees with I&T's suggestion and irJ.Stead claims that

44

r+f is strictly svo. His main reason for saying this is that, th:>ugh there

is a strategy for prep:Jsing Objects, this is not the urnnarked order of

constituents. He further prop:Jses a principle to aCCX)unt for the fact

that NPs are head-final. (I will not discuss his account here since it is

not relevant to my prop:Jsal. For details, see Light, 1979.)

In the sect.i.ons below, I show that the contro-\Tersy is easily resolved

simply by assuning a finer di.stinction of VJ? constituents than the

traditi.onal V-o, o-V. Also, L&T and Li.ght are describin~ slightly

different dialects or stages of Chinese that are easily distinguished

through the setting of parametric valu~s.

2.1.3 Parameters in Chinese

2.1.3.1 Archaic Chinese

In tenns of parameters, l1£, can be described using the parameters

already prop:Jsed. SJ?eaking only of the VP (the constituent.s of I I will be

dealt with in section 2,-4), "P£, is head initial and has str:i.ct adjacency

conditions on case assigrment. '!here are, h:Mever, tw:> p:>ssible

problems.

NPs and VPs. As was mentioned earlier, L&T elm. that there are

certain facts of AC that indicate an earlier OOV order in the language.

Q'le of these relics is the head final order of the NPs. In tenns of

parameters, we ~uld have to allON VPs am NPs to be able to specify

sepa.z-ately whether they are head-initial or 11ead-final. Perhaps [+N]

45

projections will have different s}?eCifications fran [-N] projections.

PnrJcminal. objects. As shown in (26), pronaninal NPs appedI before

the verb, suggesting an OVorder. Assuning, h:7Never, that these pronotms

are, in fact, clitics, we 'NOuld expect thE!Ill to clit.:icize to their governor

just as they do in Fanance laJ1(JUages. In the Ranance languages, elitics

appear to the left of the verb even though objects appear to the right of

the verb.8 Whatever the correct analysis for clitic placement in these

laJ1(JUages is, 'Ne assune that such an analysis \\QuId extern to Archaic

Orlnese. Let us assume, then, that VPs in AC are llead initial. ~

2.1.3.2 rblern Mandarin

It is with MJdern Mandarin (l-M) that our problems begin. First, let

us take the stage as I.&T am Light first describe it (l+1l), am not the

stage that L&T see the language drifting to (l-M2). At this point we

basically have S-PP-V-o. Ia:>king rrore closely, h:7Never, we see that sane

PPs appear after the verb. In this section, I lCX)k at these PPs to

detennine first, which type of PPs appear post-verbally (PPl) as opposed

to those that appear preverbally (PP2), and second, to find sane

hi.storical irnication as to Why this might be the case. Finally, I

propose a pararnetrical characterization of this distinct.i.on.

rrhere are t\\O prepositions that appear postverbally as well as

preverbally: gei Ito/fori and zai lat'. What is interesting in their

8. 5emitic clitics, hc:7.Never, appear to the right of the verb (Borer 1981).

46

distribution is that with sane verbs, the meaning of the preIX>sition

changes depending on whether the PP is pre-verbal or p:>st-verbal.

(33) ta gei we mai Ie chezi Ie (L&T, 1975, 180)he for me sell asp car ASP

'He sold a car for me.'

(34 ) ta mai ge.i wo chez.i. Iehe sell to me car ASP

I He sold a car to me. '

Generally, gei NP before a verb is benefactive while gei NP after the

verb is dative.

With zai there are t'NO distinctions. First, deperrling on where the

PP is placed, a different question might be answered. J30th (35b) and

(3Gb) can be translated as I Zhang-san is sleepiDJ on the flCX)r' but (35b)

answers (35a) While (36b) answers (36a).

(35) a. What is Zha.rgsan doing on the flcx:>r? (I.&T, 1975, 181-182)'b. Zhang-san zai di-shang shui

Z. at floor-on sleep

(36) a. Where is Zhangsan sleeping?b. Zhang-san shui zai di-shang

Z. sleep at flex>r-on

'!he second distinction is with rrotion verbs \\tiere pJstverbal zai is

direct.ional. \tkdle preverbal zai is locational.

(37) Zhang-san ti/.tO zai zhlDzi-shangZ. je'mp at table-on

'Zhang-san junped onto the table.'

(I&T, 1975, 182)

(38) Zhang-san zai zhuozi-shang tiaoz. at table-on jmp

'Zhang-san is junping (up and daMn) en the table.'

47

(39) \\U ba ta tui-dao zai shafa-sha.r¥JI SA 3s~ pl.lsh-fall at sofa-on

•I pushed·,him/her onto the sofa.'

(40) \\0 zai shafa-shang ba ta tui-dao IeI at sofa-on SA 3ag push-fall ASP

IOn the sofa, I pushed hi.m/her down. I

Chao's (1968) a<X.Utmt is a bit different fran I.&T's. Chao analyzes

verbs with double objects as being either v-v NP NP or V NP V NP, and

analyzes };X)st-verbal PPs as v-V NP. '!he relevant examples are given

9belC7ll •

v-v NP NP (B? 316-317)(41) a. sonq-geei ta i-fell Iiisend-to him a gift ASP

b. sonq i-fell iii geei tasent a gift ASP give him

(42) a. shuey. tzay chwang. sr1ClI1qsleep-at bed-on

I sleep in bed'

b. tzay ·ctMang.shanq shueyat bed-on sleep

I sleep on the bed I

VNPVNP

v-v NP

VNPV

(p. 353)

In (41a) and (42a), geei and tzay (gei and zai) are cliticized to the

verb and in (41b) and (42b), they appear as independent verbs.

I will assune I.&T ' s analysis in which the shi.ft fran co-verb to

pre};X)sit.i.on is canplete (see below for details), and that gei and zai are

preJ:X)sitiona. '!his \\Ould explain their apparent ambigtnus character

between clitic and verb. When adjacent to a verb, they may reanalyze as

in English (Weinberg and rbrnstein 1981) While still being able to appear

9. The orthography used in Clao diffars slightly fran that used by Lightan:1 I&T.

48

free of the verb in other constructions.

We have seen three distinctions between pre-verbal and post-verbal

PPs.

1) Preverbal PPs with gei are benefactive \tihile the same PPs are dative

when post-verbal, 2) preverbal PPs are not suitable answers to questions

while IX>st-verbal PPs are, and 3) preverbal PPs are l()Cational while

postverbal PPs are directional.

What these distinctiOI1S have in camon is that part of the meaning of

IX>stverbal preIX>sit,ions is derived fran the verb, whereas the meaning of

the preverbal prepositions is independent of the verb. Benefact.i.ves may

be added to any verb, wh.i.le datives can only be part of the argunent

structure of certain verbs.

The distinction of locat.ive PP versus directional PP is the same.

I.ocational PPs ma.y be found with any verb while di.rectional PPs require

llDt.ion verbs. In Gennan, as in other langlBges such as Russian and latin,

this difference is expressed in the preIX>sitions \tihich can take either

accusative (if the PP is direct.i.onal) or dative (if the PP is

locational) •

(43)

(44)

at

Sie geht ins Geschaft •she goes in-the(ace.) store

I She goes into the store. I

Sie arbeitet im GescMftshe works in-the(dat .) store

I She \\Urks in the store. I

The conclusion, then, is that direct.ional PPs are subcategorized by

the V while locat.ional PPs, generally, are not.

49

The di!?tinction between (35) and (36) is rrore difficult to

characterize. Basically, what I claim for the examples already discussed

is that pre-verbal PPs are not subcatE!C3orized for while IX>st-verbal PPs

are. In the case of (35) and (36), I make the same claim. In English, if

b ubc - fa - - -hI - 1- 10 thaver 5 atE!C30rJ.ze5 r a PP, J.t 15 not IX>SSl. e to pronam.na lze e

V in the fonn of do sanething leaving behind the PP. If the PF is not part

of the canplement structure, it may c:::ccur with the pro-fonn•

.(45) Bob is talking aOOut Chris.*Bob is doing sanething about Olris.*What is Bob doing aOOut Chris?

(46 ) lbb is lotmging about the house.Bop is doing sanething about the muse.What is Bob doing aOOut the house?

(47) Bob is jmping onto the table.*Bob is doing sanething onto the table.*What is Bob doing onto the table?

(48) Bob is junping on the table.Bob is doing sanething on the table.What is Bob doing on the table?

I have said above that locational PFs are not SUbcategorized.. ~re

cnrrectly, they are not subcategorized. for by notion verbs. With stative

verbs, hQ\4Jever, they may be part of the argllllent structure. This is

obvious in the facts belOIt.

(49) Into the city drove three carriages.(50) *In the city drove three carriages.(51) In the city ~tood three large ~s.

The claim is that only PP argunents of the verb can appear in the

10. I will use the tenn I pronaninalize • in spite of the fact that the'pro-fonn is of the catE!C3ory [+V].

50

inversion structure above. Chinese has a similar inverted CX)nstruction,

and again, locational PPs may appear preverbally with stative verbs.

(52) chuanJshang tangzhe bingrenbed on lie patient

IOn the bed lies the patient. I

'Ibe {X>sitioning of the PPs in (35) and (36) ma.y have to do with the

};X)sition where new infonnation, or focused ela:nents are fourrl11 • But this

dist.inction can only be achieved in cases where the PP nay be ambigtnllS

bet'N'eeIl being a canplanent or not, as is the case with locative PPs.

Basic,ally, I pro};X)se that, in all of these cases, IX>st-~erbal PPs in

Chinese add to the verb I s meaning, i.e., are part of the verb I s arguruent

structure an¢! get their Q-IOle fran the verb.

Li (1981) writes:

In conclusion, we observe t.1'.at the locative phrase mayoccur in either the preverbal p:lsition or the p:lstverbalIX>sition. In the preverbal fOBition, it has a generallocational meaning ani is essentially unconstrained withrespect to the verbs with \tihich it can occur: accx:>rdingly, thepreverbal locative phrase is called zhl.1an3YU I adverbial' byChinese gramnarians.

Post verbal locative phrases, on the other hand, arerestricted to certain types of verbs, just as direct objectsare, and are designated by the tenn~, ' cx::mplanent I, \tihichis also used for object, in traditional Chinese gramnar. 'lhisdistinction between these ·tlNO gramnatical tenns captures thedifference bet'N'eeIl the relatiVely free preverbal locativephrase and the IOOre tightly restricted postverbal locativephrase .in temlS of semantic .. intimacy" • (p. 409)

It is interesting to note that PC had an all-puqx>se IX>stverbal

11. 'Ibis was suggested to me by Iuigi Rizzi.

51

pre{X>sition ~ which was used to irrl.i.cate location, goal, dative, source,

object, instrunent, agentive, and canparative (see Huang 1978, Lin 1981).

It has subsequently been replaced by different p:everbal pre{X>sitions,

each with its own function. '!his suggests that the postverbal preposition

was used as a dtmny case-marker since the verb did not assign case and the

NP was given its 8-IOle by the verb. Na--l, ho\t,ever, in the case of the

preverbal pre{X>sitional object NPs, 9-roles are assigned by the

pre{X>sitions themselves.

'!he change fran verb to co-verb to pre{X>sition has received a lot of

attention (I.&T, 1973b, 1974; HuaI'g, 1978). Generally, serial verbs in

construc::tions such as in (53)ha.ve been reanalyzed as prep:lsitions as in

CX)nstructions like (54).

(53) S Vl 0 V2 0

(54) S P NP V2 0

For example, the pre{X>sition zai •at' derives fran the verb mecming

Ito be at I. '!he VP given in (28) \«>uld then have been translated as l was

here and played music I or •was here to play I1UJsic' whereas nt:M the

translation is I played music here I. Whether or not the second verb was

co-ordinate or sub-ordinate to the first verb is contrO\1ersial (see I.&T,

1973b). L&T give an example where the meaning is still ambigu::>us bet'Neen

that of a verb am a prep:lsition.

(55) ta na chanzi tiao le yige wu (L&T, 1974, 269)she take/with fan dance asp a dance

(i) 'She did a dance with a fan. (prep) .(ii) She tcx:>k a fan am did a dance. (co-verb)

52

rrhe original (co-verb) structure can be analyzed as either (56) or

(57) • (56) represents a control structure Where the second verb is

subordinate to the first verb. (57) illustrates a coordinate structure.

(56)

(57)

S. Vl. 01 [PRO. V2 02] (subordinate)]. 1

S [[ Vl 01 ] [ V2 02 ]] (co-ordinate)

rrhe final structure after the historical reanalysis is prepositional,

as in (58).

(58) 5 PP V2 02

I.£x:>ki.ng at these structures, it is easy to see why the e-role of the

NP within the PP is iOOepement of th~ secom verb. 'lb say that the

e-role is dependent on this verb \tJOuld be suggesting that in 'AC, V2

assigned a &-role to 01. In the structures represented in (56) and (57),

this is clearly i.rnp::>ssible since V2 is not in a position to govern 01. In

either of these analyses of the structure, 01 _is assigned its e-role by VI

not by V2. It is IX>t unreasonable, then, to assume that when this

structure is reanalyzed as PP V, the NP· within the PP is assigned its

e-role by the P and not by the following V. As for the post-verbal PPs, we

know for both historical and structural evidence that it is possible for

the verb to be the 9-role assigner.

Let us say, then, that the difference between J.X)stverbal and

preverbal PPs is that the preverbal PPs are thEmatically imepement of

the verb, i.e., do not get 9-marking by the V. The NP within the PP is

assigned its 9-role by the preJ.X)Sition. As sbJwn above, the J.X)st-verbal

PPs are not independent of the V. '!hey are argunents of the verb and get

53

their 6-markirg, at least in part, fran the verb. What does this mean in

tenns of 'NOrd order change and parameters? We have sai.d that PC was

head-initial. Is thi.s still true of MM?,

We will call PPs that are dependent on the V for e-rnarking PPls, and

t.OOse PPs that are irrlependent of the V, i.e., those where the P itself

assigns the e-IOle, Pl?2s. Assmn.:lng that PPls get 9-r1arking fran the verb,

while PP2s get e-marking independently of the verb (fran the prefX'sition

itself), a change fran S-V-<l-PPI-PP2 to 5-PP2-V-o-PPl means that the PP

that is not dependent on the verb is nt::M to the verb I s left. SUpfX'se tN'9

say that r-M is head final, thus accotmting for the fact that those

constituents that are not dependent on V are fotmd preverbally. First 1Ne

want independent evidence and notivation for this charge fran head-initial

to head-final, and then we must acC'Ount for the CXJnstituents that do

appear fX'st-verbally.

Independent evidence for supposing that r+1 is head-final can be found

in the p:>sition of aspect if we assune that aspect is realized on the INFL

rxXie, Aspect appears to the right of the verb « 33 ), (34), (55» which

\\Ould be expected if MM is head-final. ("!here are still questions to be

anSWtared concerning exactly where aspect is placed after the verb, but

this requires nore study.)

M:>tivation for' our claim can be fotmd in the structure of NPs. We

suggest that since NPs are already head final in AC, the graI11l1C\r will be

simpler if VPs are also head final.

:NOw, what of the p:>st-verbal constituents? We prop:>se that the

54

direction of e-role assigrment is another parameter \tJhich detennines \\Ord

order in languages. We can then claim that while r+tl is head final, it

assigns e-roles to the right t' If we l(X)k at the t\\O categories that

assign e-roles, preIX>sitions and verbs, we see that toth of them appear to

the left of the NPs to \tJhich they assign 8-roles. We will assume that

within NPs, e-roles are assigned by the preposition and not by the head N.

It may, at first, look odd that the ba NP phrase appears to the left

of the verb. We have seen above that only non-argunents of the verb

appear to the verb's left. Yet in the case of the ba CXXlstruction, the NP

argunent which appears to the right of the V without ba, now appears to

the left of the verb with ba..

A lCXJk at passive cnnstructions in Olinese, h~er, gives us a clue

as to what is happening in these cases. Olinese has no verbal }?Clssive

IlDrJ?hology. Passive is created, instead, by having the d-structure

subject in a bei NP ('by NP') <Xlnstruction.

(59) ta ba neizhi rna qi de hen leihe BA that h:>rse ride till very tired

•He rode that horse t.i.ll it got very tired. '

(60) Neizhi rna. bei ta qi de hen leithat horse by hIin ride till very tired

''!hat h:>rs~ was ridden by him till it got very tired. '

Again we have a case where it appears that an argunent of the verb

appears to its left. I pro{X)se, lnwever, that pre{X)sitions like ba and

12. '!here is evidence that perhaps }?Clssive fonnation in general consistsof the absorption of the external Q-role rather than the absorption ofcase. see Clapter 5.

55

bei serve to absorb a e-role of the verb12;. bei absorbs the agent e-role,

and ba absorbs the patient e-role. 'rhus these pre1X>sitions nr:M assign the

e-role independently of the verb, and the PP appears to the left of the

verb.

It is interesting that absorption of e-roles only happens to agents

and patients. It appears to be restricted to tOOse a:rgutents that do not

rely on the canp:>sitional 8-role assigrment of a verb and a prePJsition.

'!his process, then, can only subSlIle the e-assigning properties of the V,

and not the joint 6-role assigning pr~J?ert_i.es of a verb and a

pre1X>sition.

In sumnary, we have described a case where the word order of a

lanJUClge ma.y be determ.i.ned by settin:J the direction of 6-markin:J

independently of the head-initial/head-final parameter. This entails

viewing the head-initial/final parameter as a default specification of

direction; i.e., if there is a constituent which does n:>t fall within the

danain of an already specified parameter, its placanent is detennined by

the headedness par~eter. Kiparsky's Else\tbere Condition (1973) will

insure that the direction of 6-markin:J will affect the d-structure before

the head-initial/final parameter since the fanner acts on a subset of the

latter. Canplanents of the verb, then, will appear to the right of the

verb in order to be assigned 6-roles. All of the constituents that may be

assigned 6-roles in arx>ther way, as far as the verb is concerned, appear

according to the default condition of head-final. OJjects have a cooice

of either bein:J assigned a 9-role directly by the verb, in which case they

appear to its right, or by the Object marker ba in a pre-verbal PJsition.

56

2.1 e 4 Li & 'l.b:npson· s Speculations

'1lle evidence that L&T use to claim that r+1 is becaning an S-O-V

language is actually an irrlication of a further parametric chanJe.

2.1.4.1 PrelX'sed Cbjects

Let us first look at the prelX'sing of objects. Given what was said

above about 9-role assignnent, \fJe have to assune that prelX'sed objects get

their 9-role fran the ba that appears to their left. SUppose now that the

independent 9-role assignnent properties of ba are lost, and that the

verb, at least eat1pJsitionally, also assigns the 9-role OJ the object of

ba. 'lllis mt9aIls that the e-role is nt:M being assigned to the left. If

this is the case, then we also expect to find datives and other PP

argunents to the left of the verb. '1ll.is is exactly What L&T describe as

an indication of the nnve to S-o-V. If this change is as UicT describe it,

then datives soould be fotmd preverbally (as they are) and all semantic

distinction of preverbal versus lX'stverbal NPs sho\lld be lost (as they

suggest is happening). Why, then, as Light claims, is the unnarkErl order

still S-V-<l, and ~y is the preverbal object al'llaYS in a ba c:onstruction?

2.1.4. 2 Case Parameter

I proJ:X)se here another parameter Which accotmts for this synchronic

description. '1ll.is parameter is used to describe a lan:JUage Which has

every element of the verb phrase preverbally, except for the "bare"

57

object, Which apJ?eCirs {X>st-verbally. Also, any time the object appears

preverbally, it appears after an object marker. Obviously the verb can

assign 6-roles leftward since 6-marked PPs can appear on the left:.. Let us

say, then, that r+12 is head-final but assigns case ri'3htward. 'l'hiS

accoLUlts for the fact that all the canplanents of the verb ~ppear ~ its

left except for the element wh:i.ch requires caseIi.e ., the object. (Other

examples of case assigrment will be discussed belCM.) '!he onl.y way of not

having the object on the right of the verb is to put tiLe durmy case marker

ba ir. front of it.

An intere8t·tng bit of confinnation for this conclusion is found in

dative C'Onstructions. Li I S gramnar gives tltree different classes of verI)£;

Which take dD.tive NPs. sane verbs may take either the NP gei NP or the NP

NP construction, sane must take th~ NP gei NP construction, and sane must

take the NP NP c.'Onst.ruction. Examples, taken fran Li (pp.374-379) I are

given belOl/.

(61 ) gei obligatory:

a. ta dai-le yi baa tang gei Zhangsanhe bring-ASP one bag carrly to Zha.n:Jsan

IS/he brought a bag of candy to Zhangsan.·

b. ta da.i gei Zhargsan yi baa tar¥J

c. *ta dm-Ie Zhangsan yi bao tang

(62) gei optional:

a. \t4O song-Ie yi ping J1U gei taI give-ASP one bottle wine to 3ag

I I gave a bottle of wine to him/her. •

b. \'to soD3 gei ta yi ping jiu

c. we song-le ta yi ping j iu

58

(63) gei forbidden:

a. *w:> wen-Ie J1-ge wenti gei taI ask-ASP several-cL problan to 3sg

I I asked him/her several questions. I

b. *\\t> wen gei ta j i -ge wenti

c. \«) wen-Ie ta ji-ge wenti

If the shift of datives to preverbal 'posi,t.i.on can be accotmted for

through this cl1.aD3e of parameters, ~ \\Quld expect the NP NP constructions

to remain, bltt the gei NP phrases to be fotmd preverbally. Li writes:

the indirect object marked by gei has 'begun to aJ::1?ear inthe preverbal IX'sition ••• the appearance of the imirectobject in the preverbal POSit.:tOll, 'ho\tJever, is confined to onlya few verbs •• that is, to tOOse groups for Which gei iseither obligatory or optional. (pp. 386-387)

This is just the result that we would expect. Those NPs Which arec

assigned case imeperrlently of the vero, i.e., by a preIX'sition, ma.y

appear preverbally, while in the double object constructions, NP NP, roth

11U1st appear IX'stverbally in order to be assigned case.

What of other categories? NPs are still head final as one would

· N' do · 13. · ·expect S1Ilce s not ass1gn case • SlIlce preIX's1t10ns are case

assigners, PPs remain head-initial.

The "IX'stpositions" that L&T discuss crucially do not assign case.

In fact, they are derived fran NPs and therefore axe expected to be head

final.

13. But see cases of case-assigning adject.ives discussed in section2.2.1.1.

59

(63) \«) zai chuang-shanJ shuijiao (l&T, 1973a, 204)I at bed on sleep

I I sleep on the bed. I

(64) Zhang-san t.i.ao zai zhuozi-shangz. jl.llp at table-on

'Zhang-san junped onto the table.'

(L&T, 1975, 182)

(65) Zhang-san zai zhlDZi-shang tiaoz. at table-cn junp

'Zhang-san is junping (up and down) on the table. I

In the examples giv~ in (63), (64) and (65), we can see that the

IX>stpJsition shang acts as a noun canp::>und. \\hich is assigned case by a

true case assigner, the preJ.X)sition zai. Basically, -shang simply makes a

14place notm out of a camon noun •

2 ~ 1. 5 Conclusion

We can see that confusion can arise when languages are characterized

simply as either SVO or rov. If, l'aNever, we talk of larguages as being

S-V-o-PPI-PP2 or S-PP2-V-o-PPl or S-PP2-PPI-V-o or S-PP2-PPI-Q-V, such

problans might be avoided. Light writas:

•••as a careful reading of Greenberg I s fanDUS essay onyord order universals (1963) will iniicate, except for pure ornearly pure types, 'NOrd order differences are not discrete butcontin\.D\lS. On the line bet'N'een rov and SVO larguages, thereare many J.X)ints at which one will find a great many - or roc>st- of the larguages of the \\Orld. If one looks only at theordering of subject, verb, and object and universal w::>rd ordercorrelates, the discovery of a 'ItOm order continuun will notbe very ill1Jlli.nating. But if one takes into accotnltmarkedness, relative meaniIl3'fulness of \\Ord orders, topicpraninence versus subject praninence, as well as the canplexrole of affixation, 'it is likely that this continuun will bemult.i.variable and quite illun.i.nating in pointing to

14. Thanks to Wayne 0 'Neil for IX>inting this out to me.

60

di.fferences am similarities arrong the \\Orld I s languages. Or,in short, the statement that a language is &:JVor SVO or CEVis by itself a nearly meaningless statanent. (Light, 175)

I prop::>se that one should define langua.ges as 'being XI?

head-initial/final, yP head-initial/final (wh.ere X ani Y are variables

that range over all categories), leftward/rightward e-rnarking,

left\t1ard/right\t1ard casanarking. '!his way the order of all the VP internal

constituents can be characterized.. On a feature systan., PPI, PP2, and 0

\\Ould be as in (66) •

(66)Q-marking

case-marking

PPI+

PP2 o+

+

??

+

'!his means, for example, that PPls are 6-marked but not case-rrarked

by the verbIS. '!he divisions follow fran other principles. '!hat PPs are

never case-rrarked by the verb, am objects are always case-rrarked by the

verb follONS fran case theory. '!hat objects are alW3.YS e-rnarked by the

verb folla«s fran the 6-criterion am the Projection Principle. 'lhe oole

in this diagram is [-e-rnarking, +case-marking]. This is not surprising

since this feature grouping appears only in the very m3rked configuration

of Exceptional Case Marking and never between a verb and a manber of the!

same s.

We concllrle, then, that the change that Light describes involves a

15. Object NPs of unaccusative and passive verbs will also fall into thiscategory, since they are assigned a 9-role by the verb, but are notassigned case. 'lhese, h~ver, will pattern differently fran PPls since,as NPs, they need to be case-marked. Different ways of gettiD3case-marking will be d.'.scusued below: for alinese, sect.i.on 2.5.2.2, forGeI1l1C1I1, English, French, Irish, Chapter 5.

61

shifting of nore than just objects. It includes the IlDVanent of PPls and

PP2s • The shifting of these elements is best described by the resetting

of the parameters mentioned above. With these parameters, changes are

hetter described and restricted. L&T seem puzzled that the shi.ft fran

s-v-o to S-<l-V is taking so long (t\\O millenia) especially since, in their

view, the langua.ge has been in an unstable S-o-V/S-V-o condition the whole

time. '!his view also implies that there exist goals in language change

that may take generations of gramnars to achieve. It is i.rnp:lssible to

inex>rp::>rate such a view of lan:JUage into current thec>ry since no grann:ar

has access to another granrnar. In tenns of parameters I hCftNever, there is

nothing unstable. Chinese is not bet\tJeen stages, but rather at a specific

stable point.

2.1.6 Postscript

Langua.ges fran the Niger-COngo offer possible evidence for a m.i.rror

image of the changes described above. 'll1ere is disagreanent as to wtlat

the 'NOrd order of Proto-Niger-COngo was (Gival 1975, Heine 1980, H¥man

1975), but the variation in the present day languages is still best

described using the parameters of &-role assignnent and case. Givon

decribes larguages such as Kpelle. He writes that this lan:JUage has

retained some S-<l-V characterist:lcs (he believes that Proto-Niger-COngo

\fIaS S-<l-V) since objects are preverbal. ibWever, he says that S-V-O

orders are also evident since locative, instrumental, manner, benefactive,

and dative phrases fol~r:M the verb. It seems strange to say that the

langua.ge, therefore, is both S-O-V and s-V-o, since the object never

62

appears after the verb. Only PPs appear after the verb. In my tenns,

this is not a variation of S-<l-V versus S-V-<l, but a case of a langua.ge

which assigns case to the left but has a default head-in!tial parameter.

Where r+12 is PP2-PPI-V-<l, Kpelle is o-V-PPI-PP2. (see I<Dopnan 1983b for a

similar and IlDre detailed analysis of M:lh::>u, a l:brthern Mande larguage.)

~an adds to this description of Kpelle the fact that "there is gaXl

evidence that datives once preceded the verb" (Hyman, 1975, 128). This

suggests that at one pc:lint the order of the constituents of the VP in

Kpelle used to be PPI-Q-V-PP2. At thi.s stage, then, we lAOuld characterize

the language as still having the head final parameter, but with leftward

o-role assigrment. Where ml is PP2-V-o-PPl, this earlier fonn of Kpelle

was PP1-<l-V-PP2. Finally, +j9, a language in the same family, is strictly

S-o-V, i.e., strictly head-final.

~e i.nq:nrtant thing to understand is that a language like Kpelle is

no less "pure" (t.h::lugh perhaps IlDre narked) in its \\Ord order, than is

:J:j9. Its "variation" between S-V-O and S-O-V is simply a result of

diverse parameters.

2.2 Huang's Analysis

The type of Olinese that Hlang (1982) .describes rrost closely

resal't>les t+12. His discussion, b:Mever, is much rrore detailed. I \\Quld

like to argue here against his prop:>sed analysis, or roore correctly, take

his analysis and redo it in terms of the propc:lsed parameters. He, in

63

fact, suggests just such a p:>ssibility (fn. 10, p. 93). First I will

di.scuss his solut.i.on to the tNOrd order problem. 'lhen I will lex>k at his

data in tenns of the parameter of direction of case assignnent, brirgirg

up problans and suggesting solutions.

Huang prop:>ses a ~ filter Which states:

(67) a. [:>fl tJ-1 yP*] iff n=l and X#N (p.41)

_n-lb: [tt yP* x ] otherwise

Simply put, (67) states that Chinese is a head final language except

for the l~st levels of A", V" I a.rrl pUs, yielding representation like t4le

following.

(68) a. All

/\y" AI

~A NP

V"~

y" VI

~V NP

pll

~y" pi

AP NP

b. Nil

~y" N '

ANP N

2. 2.1 Problems for the case Analysis

2.2.1.1 Adjectives

We have already discussed the instances of VPs and PPs, and we have

sh:7Nn that the question of \\Ord order in these examples reduces to the

direction of case. B::7.\ever, this solution does not obviously extend to

APs. we krx:M that in Erglish, adjectives do not assign case. nus is

clear in the following examples.

(69) a. *proud Johna I. prooo of Johnb. appreciate John

.'

64

b - •*appreciative Johnb ll

• appreciative of John

'!he durmy pre{X)sition of llU.1st be inserted before an adjectival

canplanent, otherwise the NP will not be assigned case and the case Filter

will be· violated. In the (b) examples we can see that the verb assigns

case, but the adject.ival fonn of the verb does not and of must be inserted

to save the construction. (see Elrorrls (to appear) for a different

analysis of these facts.)

If this is the same for arinese, our analysis for the 'NOrd order will

not \tI:Jrk. '!here \\Quld be no reason for adjectivcll canplenents to be, .

p:lst-head if they do not require case fran the adjective. - As it turns

out, th:>ugh, adjectives do assign case in ~\inese.

(70) ta hen gaoxing zhei j ian smhe very hawy this matter

-He is very hawy about this matter. - (p. 27)

(71) ta dui zheijian am hen gaoxinghe t.o\erds this matter very happy

-He is very happy about thi.s IllCltter.-

In (70), the canplanent of the adjective appears after the adjective

am requires no pre{X)sition, while in (71) we can see that if the

canplanent appears before the adjective, a preIXJsition is required. These

examples not only shM us that adjectives can assign case in Chinese; they

also confinn our earlier proIXJsal for VPs and PPs, i.e., that case can

only be assigned to the right •

65

2.2.1.2 PF Adjacency

Huang •s X' template acts as a tilter of PF. He gives many gcxrl

argunents \tJhy this Imlst be the case: that is, why it llUlst apply after

IIOVanent rules. In cases where the object moves fran the original

p:>st-verbal p:>sition of Huang's analysis (as in "ba transfonna.tions ll

(72a), pa~sives (72b), topicalization (72c), and object prep:>sing (72d»

or does not appear (as in intransit:'~ves), another constituent may be found

p:>st-verbally. 'lhese other constituents may be Subjects16 (73a), extent

phrases (73b), double objects (73c), result phrases (73d), and predicative

phrases (73e).

(72) a. ta ba Lisi pian-Iehe M L. cheated

I He cheated Lisi. I

b. Neizhi mel bei ta qi de hen leithat horse by he ride IE very tired''!hat mrse was ridden by him till it got very tired. I

c. Neizhi ma, ta qi de hen leithat mrse he rode DE very tired'That mrse, he rode it lBltil he got very tired. I

d. Ta neizhi ma qi de hen lei17he that horse rode lE very tired

•He rode that m~se lmtil he got very tired. I

16•. This will occur in intransitives. We will discuss thi.s in detailbelOli.

17. Obj~t prep:>sing must be a scrambling rule such as those described insaito (1983a). Since the object is in an A I p:>sition, it need not beassigned case.

(73 ) a. xia-gtX> le yu Iefall-ASP ASP rain ASP

•It has rained.'

(p. 46)

66

b. ta ba neizhi. rna qi de hen lei (p. 53)he BA that oorse ride DE very tired

•He rode that horse until it got very tired. •

c. ta ba wuge pingguo chidiao-le liangge (p. 42)he BA. five apple eat-ASP t\\O

•Of the five apples, he ate t\\O.·

d. ta ba zhimen t:l-le yige dong (p. 42)he BA paJ:)er-door kick-ASP one oole

•He kicked a hole in the paJ:)er-dcx>r. •

e. tNanen ba ta dang sbaguawe Bl\ he treat-as £CX)1

•We reg'ard him as a fool.'

This rai.ses .~questions for our analysis. First, why can extent

phrases appear p::>st-verbally at all? Q1.e w:>uld not supp::>se that they need

case. seccndly, why can they only appear if no object is present

p::>st-verbally at s-structure?

Huang aCCOlRlts for this by having a restructuring rule which allows a

structure like (a) at s-structure to 'be reanalyzed as (b) at PF.

(74) a. V"

~VI X"

v~

b. V",=> V'

1\V X"

OUr analysis will not allow us this solution since X" may only appear

to the right if it is assigned case by the veI:b. Unless lNe can sbJw that

67

extent phrases are assigned case by the V, we cannot accntmt for their

fOst-verbal fOsition. In fact, there is evidence fran Fnglish that extent

phrases may get case fran the verb (thanks to Beth Levin for this fOint).

'!he relevant examples are given below.

(75) a.

c.

d.

I read a h:x:>k before goiIX] to bed.I read an hour before going to bed.

*I read (an oour a h:x:>k} before going to bed.l a h:x:>k an hour

I read a bc:xJk for an lDur before goiIX] to bed.

It seans in these cases that read assigns case to an hour. If the

verb also takes a direct object, then the case is aseignec1 to the direct

object and the extent phrase IlUlst be assigned case by a preJ.X>sition.

In Olinese, l1I:>t only can the verb assign case to the extent phrase,

it seans that it nust assign case to the extent phrase, unlike the English

example given Where the preJ.X>sition for may be used to assign case t~ the

extent phrase. '!he direct object, then, nnlSt get case by sane other

means. '!here are three ways that this may happen. (1) It nay ITC'Ve to

subject fOsitiori Where it will be assigned naninative case (there is no

passive verbal norphology in alinese) as in (72b). (2) It ma.y be assigned

case by ba, as in (73b). (3) '!he verb may be reduplicated with one verb

assigning case to the direct object, and the second verb assigning case to

the extent phrase, as shown belQIW.

(76) \\0 qi rna qi de hen lei (Huarg, 1982 p. 47)I ride oorse ride till very tired

•I rode the oorse \mtil I got very tired. I

~t.:lce that we are assunting that objects are base generated

68

pre-verbally. Only the elanent which is case rrarked by the verb, which in

these instances is the extent phrase, appears post-verbally.

2. 2.2 Problens for Huang I s Analysis

What are the differences bet\'leen the tw:> profOsals, then, or are they

only notational variants? '!he question reduces to whether or not

branching exactly mirrors case assignnent. We already see one reason why

they are not exactly the same. NPs, in not assigning case, are head

final. While Huang needs to stipulate the split bet\'leen N' s and the other

categories, we need only state that As, Va, arrl Ps assign case.

Another distinct.i.on is that Hlang is forced to have different

branching properties for NP arrl PP canplenents of a verb or adjective.

'!here are, then, t\leO different levels for subcategorized constituents. If

the C'Onstituent is an NP, it is a daughter of Vi and sister of V. If the

constituent is a PP, it is a daughter of V", and sister to V'. I see t\\JO

problans with this.

The first is that parallelism is lost roth within a category and

cross-categorically. Within a category, the fOst-head canplement \\Ould be

sister to V while the same canplanent, but in a pre-head position, would

be sister of V I •

(77) a. ta pian-Ie Lisihe cheat-ASP L.

•He cheated Lisi.·

b. ta ba Lisi pian-Iehe BA L. cheat-ASP•He cheated Li.si. I

(78) a. I"

~NP I'

I ~he V" I'

IV'~

V NPI I

cheat Lisi

b. III

~NP I'

I ~he V" I

~

~ r/ k V

B1\ Lisi cheat

69

Cross-categorially, the generalization again \\Quld be lost. '!he

pre-head canplements of nouns will al\\a.YS 'be prep::>sitional phrases (or

S' s) since the mUll cannot assign case. 'Ibis prslX)sitional phrase can

either be a sister to N «79a), or a sister ti:> N' «SOb).

(SO) a. N'I

/'.yll N'

APP N

I will assmne that these ·PPs will be sister of N (as in (a», even

toough they are preIOsitiona! phrases since this at least captures the

generalization that they are canplanents of the head. Also if these PPs

are not sisters to N, it is hard to imagine what constituents \\Quld ever

aRJear as sister to the head. If, ho\t.ever, we assmne that dui PPs in NPs

(illustrated belC1ll) are sisters of tP, dui type PPs \\Ould be on one level

in VPs and APs, and another level in NPs, even though, in roth cases, they

are subcutegorized canplanents.

70

(81) a. ta dui guoJ~a de re-aihe t.o\tards COtmtry IE rot-love'his enthusiastic love of CXJUlltryl

a'. N"

~~~ ~N~ r

dui cx>untry rot-love

b. A"

~PP AI

dui~S ~matter I

very happy

c. V"

~pp V'~ ,

BA NP V

d. N'IIN'

. ~N"Adui NP

'!here is another problem that could arise when NPs and PPs are

generated on different levels. We want direct object NPs to c-ca11tlal1d the

PP canplements of a verb in order to get the binding facts given below.

,(82) a.b.c.d.

*I told John. a'bout him.•I told John7 a'bout ~elf.•

*! sold the §lave. to him.• 1

I sold the Slave~ to h:im§elfi o

71

Assuni[):J Reinhart· s (1976) definit.ion of c-ccmnand, or Chansky

(1981), we must represent the relationsil.i.p between the diJ:'ect. object and

the PP as in (83) belCM, rather than as in (84).

(83) V,nr

Vi

~V NP pp

(84)

..'Huang relies crucially on binary branching for his account.. Asst.nn.ing

Reinhart I s or Clnnsk~{t:3 notion of c-cannand, l'r>\\rever, it appears that

binary branching is not co.crect since direct objects, then, will not

c-cannand canplanent PPs.

This raises the cxmplicated issue of what notion of branching we

need, tNhich raises the further question of \tvhlt definition of c-cannand we

need. I prop:>se in the following section that, not only is binary

branching an tmnecessary and even undesirable proliferation of ncxies, but

that only t"-~ levels, xnax and Y?, are required.

2. 3 Branching

In this section we investigate the consequences of maximally\

articulated types of branching systans versus minimally articulated. types

72

of branching systems. I will claim that with the developnent of case

theory and 6-theory, many facts that have been explained through branching

ma.y ncM be explained tllrOugh these theories. Once these facts have been

factored out of the structural ,representation of strings, branching may 1:>e

simplified to represent only tl10se const.i.tuents which are necessary for

operations which apply to phrase structures, i.e. rocwement,

pronaninalization, and. co-ordination.

Jackendoff pro};X)ses a triple bar level for every category - an XI

level which contains Y? and its functional arguments I an XII category that

contains X', X· specifiers, and. restrictive nrdifiers, and. an X"' category

with XII I XII specifiers I and non-restrictive mcx1ifiers. Huang (and Kayne

1981 for different reasons) proJ.X>ses binary branching so that each

additional CXXlstituent within a maximal projection will add a d:ifferent

level of branching (see also Hoji 1982 for Japanese). We have argued

above against binary branching, but what of Jackendoff I s proJ.X>sal?

let us first lOJk at What generalization may be captured by this

specification of branching18• Jackendoff shows that ordering of non-heads

can f011011 if \E assure that canplanemts appear within the XI level and

adjuncts within the XII level. We would, then, not expect adjuncts to

appear bet'Neel1 the head of a category and. that head I s canplanents.

'!he second reflex 'Ne \'JOuld expect is that ccnsti.tuents, whether they

18. As in the first part Gf this chapter, I leave specifiers aside.Perhaps this obscures the major clue to the problan, but for the timebeing I assume that this is not the case.

73

are X" or X· could be treated as discrete tmits for purp:>ses of Affect-a,

such as m::wanent, deletion, pronaninalization, and co-ordination.

2.3.1 Orderirg

As far as ordering is concerned, Jackendoff •s predict.ions are oorne

out.. Adjln1cts do not appear bet\4Jeen a head and its canplanents.

(85) a. Mary decided on it in her early teens.

b. *Mary decided in her early teens on it.

(86 ) a. the king of France fran England

b. *the kiD3 fran EhJlanc1 of France

In fact, ordering in many languclges is even stricter than the t\\O

different levels within the Vp (VI and VII) \\Quld imply. '!he

generalization is th3.t, within the VP, the order is object first, then

other canplanents, and finally the adjtmcts, as illustrated belCM.

(87) Verb Cbject PP-canplanents Mj\IDcts

we have argued that, unlike Huang's binary branchirg analysis, all

COt1f>lements must be at the same level of branching. '!his means, then,

that the orderirg of the object versus the PP-cauplements must follCM fran

sanething other than branching. '!he an.swer to this problan is obvious.

Since t.he object is dependent on the verb for case, it !1lUst appear

adjacent to the verb. PP adjtmcts, on the other hand, have no such case

requirement, am therefore no such adjacency requiranent. Evidence is

found in VPs whose internal argunents are all contained within PPs. Here,

74

since none of the argunents is dependent. on the verb for case, order is

fr 19ee.

(88) a. I talked to M.ichael about the bcok.b. I talked about the 1:xxJk to Michael.

(89) a. Who did you talk to t about the 'OOok?b. Who did you talk about the bcok to t?

(90) a • What did you talk to Mi.chael about t?b. What did you talk about t to Michael?

'!be discussion above has shQ\lJn that a relationship betlNeen a head and

a maximal projection may predict the ordering of the maximal projection

relative to others that do not enter into a similar relationship with the

head. In this instance, the orderiD3" follorNS not fran a branching

configuration but fran this relationship. rrhe example above uses the

relationship of case assignnent, and it opens up the question of whether

there are other relationships that can predict the relative ordering of

constituents. I pr0IX>se that the assignnent of e-roles is aLbU

condit..i.oned by adjacency. Obviously t\\U argllllents cannot be adjacent. to a

verb. I suggest instead that there exist certain datains and that these

danains cannot be interrupted by elanents fran outside of them. '!bere is

a case dorain and a 6-role or canplanent datain.

(91 ) a. Deoorah [1x>ught a b:x:>k] yesterday.b. *Deoorah oought yesterday a bcok.

CASE IXMAIN

(9~1) a. De1xlrah [p.1t a b:x:>k on the table] yesterday.*I:eoorah put a b:x:>k yesterday on the table.

19. I am assuming in these examples that IOOVement of a VP constituent to aVP-peripheral PJsition is always IX>ssible for the pUIpJse of focusirg, butthat extraction, then, is not IX>ssible. 'lhe fact that extraction isFOssible in the examples given imicates that all of the const.:ltuents arein IX>ssible d-structure IX>sitiona and have not undergone IOOVemenJ~.

75

. (93) a II Deoorah [put a J::xx)k under the table] quickly.b. Deborah put a J::xx)k quickly under the table.c. What did JRborah put the J::xx)k under quickly?d. *What did D:IDorah put the J::xx)k quickly under?

(91b) is \\Orse that (92b) because it violates both case adjacency and

cnnplement adjacency. In (93) I we see that the position of the VI? adverb

quickly is fairly free; 'OOwever, the extraction facts of (93b) and (93c)

indicate a difference in the PP depetlding on whether or not it is in the

canplement danain. Extraction is pJssible only if it is within the

canplanent danain of the verb I suggesting that in (93b) the PP has been

IlDVed and adjoined to the VP.

Let us tentat.ively define this as a condition on danain adjacency as

below.

(94) lXJnain Mjacency Condit.i.on (me) (tentative):

If a node X is in a direct relation R with a ncxle W,where R involves sisterl'lCXXl, if there is another nodeY such that X does not enter into R with Y, then

*[ ••• X ... Y ... W ... ]

Direct relations are e-role assigrment (canplementation)and case assignnent.

'!he following example raises a problan for the me.

(95) Joy considers Daan silly.;

Considers takes the predicate silly as its canplement which I have

claimed is a direct relati.on. IbNever, Il:!an is not a canplanent of

consider. I will argue in O1.apter 4 that a subject NP adjacent to its

predicate is a canplanalt of that predicate. '!his means that lA3an is a

canplement of silly. If we revise the definition of the D:main Mjacency

Cond.i.tion gi.ven ab:JVe, we will correctly predict the gratmlClticality of

(95).

(96) Ihnain Mjacency Condit.:lon (1ll\.C) (revised):

Given a d::'rect relation R between a node Wanda node X,where R involves sisterb:Jod, if there is another nc:rle Ythat does not enter into R with either W or X, then

*[ ••• X ... Y ... W ... ]

Direct relations are 6-role assignnent (canplanentation)and case assigrment ~

Since the relation of canplementat.i.on ooIds bet\\een consider and

silly, and it also ooIds bet~n Daan and silly, Dean may appear bet~

consider and silly.

If ordering can be predicted with this notion of danains, adjtmcts

may rr:M be at the same branching level as canplanents. rrhis conclusion

has advantdges within Ihthstein' s notion of predication, which specifies

that predication is a relation between nutually C-COlmarrliDJ

constituents. GiVE!1 Ieinhart I s (and Clansky 1981 ) definition of

c-cannand, only sisters are in a llUltual c-cannand relationship. If we

assume, as discussed in Olapter 1, that adjtmcts are predicated of

canplanents, then they IlU.1st be sisters te these canplanents and therefore

on the same branching level, as illustrated below.

(97) Teresa met I):)nna exhau3ted.

Vi

~APiI . I .

J. ].root O:lnna exhausted

76

77

2.3.2 Const.i.tuency

If ull the recursive menbers of a maximal proje::tion are sisters,

i .e. every maximal projection is only an r I h:>w does our propJsed

structure hold up under other argunents for a more hierarchical branching

systan? I row discuss the cases of MJve-a., pronaninalize-a, and

conjoin-a, where a is a const.:ltuent other than xnax or Y?

2.3. 2.1 ~ve-a

Rouveret & Vergnaud (1980) have argued for a case of M:>ve-a where a

is not .,;nax or vP, but VI. '!his lOOVenent rule is used to aCCOlRlt for the

array of facts evidenced in French causative constructions. '!heir rule is

given in (98) below.

(98) VP PrelXlsing: (p .130)

Chansky-adjoin *V to 5, where *V is sane project.ton ofcategory v.

A footnote explains that the A-<Ner-A condition "does not apply when

the analyzed e1anent in the string correspJoos to a category in the

structural descriptiin of the rule that is specified for a variable number

of bars" (p. 130). BalON are examples fran R&V where different bar-levels

have been rcoved.

(99 ) V: en a fai.t sortir Jeanne du bureau.one has made go-out Jeanne fran-the office

'we make Jeanne go out of the office. I

d-structure: On a fait [ Jeanne sortir du bureau ]

a-structure: en a fait ·[V sortir ]. [ Jeanne t. du bureau]1. 1

78

(100) VI: Anne fera lire ce livre a Claire.Anne will-make read thi.s book to Claire

•Anne will make Claire read this book.'

d-str~ture:s-structure:

Anne fer? [ Claire lire ce livre] 20Anne fera [VI lire ce livre]. [a Claire t.]

1 :I.

If this analysis is correct and VI is subject to M:lve-a, this is an

argunent for a structural nooe that is neither xnax nor Y? fbNever, it is

not so clear that this is the correct analysis of the construction.

First, as R&V point out, V::J can lOCJ\Te only subject to certain constraints.

An example of a structure \tihich is rule1 out is given below.

(101) *Anne fera lire (a) Claire ce livre.

If the verb is transitive, the direct object must lOCJ\Te wi~th the

verb. 'Ibis follows fran R&V's filter (83) given below in (102).

(102) *NF, lmless (a) NP is gCJ'.lerne1 by Tense(b) NP is governed by -wEi or +WH(c) NP is gO'Verned by A non-distinct fran [-N]

Focusing our attention on (a) and (b), this filter basically

recapitulates the case filter • 'll1e way it is being use1 in the examples

of the French causatives, however, is to prevent the case assigner fran

lOCJ\Ting away fran the elenent Which is assigned case. 'll1e cnnclus:ton,

thtm, is that V::J can move only when. it does not strand its case-marked

object. '!his, bJwever, does not explain the lmgramna.ticality of the

follOfling exaIuple.

20. ~-insertion applies presumably to assign case to Claire. see rJianzini(1982) for an analysis of this fact.

79

(103) *Eve1yne fera mettre Ie livre a Jean sur la table.Evelyne will-make put the book to Jean on the table

•Evelyne will make Jean put the book on the table.'

It also will not explain the lack of ambiguity in (104).

(104) Marie fera ~rire \IDe lettre ~ EMelyne ~ Jean-Ioup= Marie will make Jean-IDup write a letter to Evelyne ..,. Marie will make ~elyne write a letter to Jean-roup.

A rrore accurate generalization .i.s that all canplanents must move with

the verb. 'lhis does not imnediately argue against the rnovanent of V'. We

may extend the idea of the filter in (102) and say that both case and

e-roles Imlst be assigned under goveITJ:rent at s-structure. '!his \\Quld have

the effect of keeping a e-role assigner in a government relation with its

argments or with the traces of tlnse argunents.

'll1ere is an alternative analysis for these facts, ~ver. In

Manzin.i. (1983), it is assuned that these structures are base-generated,

and not created by novanent. Assum..i.ng base-generation and a notion of

danain adjacency, 'tA9 \\Ould expect case-marked ar,d e-rnarked itans to be

adjacent to the nead, in this case, the verb. 'll1ose constituents which

are not dependent on the verb, i.e ., a:i juncts, will appear sentence

finally.

In the absence of clear evidence to the contrary, I cn.lclooe that

there are no rules which troVe elanents which are neither InaXimal

projections nor lexical items. 'lhe absence of such rules tNOuld be

surprising in a systan where a in M:Jve-a \liaS free to pick any constituent,

and constituents proliferated through increasingly detailed branching

80

structures.

2.3.2.2 Pronaninalization

Another argllllent for a level of branching that is neither maximal nor

minimal is pronaninalization. In the sentences below, ~ see first a case

of an r:f pronoun, (105), then a case of an N' pronoun (106).

(105) I saw a picture of Debbie in the living roan and one ofKonrad in the dining roan.

(106) I saw a picture of Cindy in the living roan and one in thedining roan.

Assuming that of NP is the canplement phrase of picture, in (105) one

replaces just the lexical category leaving 1:ebind the canple:'lent. In

(106), though, both the r:f and the canplanent phrase are replaced

suggesting that one replaces N I •

Let us assume, as Jackendoff (1977) does, that do so can 1:e used to

di.stinguish subcategorized elements fran non-subcategorized elements in

the VP. His examples (p. 65) are given below.

(107) a. John talke::I to Bill aOOut Harry.b. *John talked to Rtli aOOut Ha.rry, but he didn I t do so aOOut Fred.c. *John talked aOOut Harry to Bill, but he didn I t do so to Fred.d. John talked to Bill aOOut Harry on &mday, but he didn I t do so

on 'lhursday ~

Fran this we might concltrle that do so is a verbal pro-fonn fQr V' or

vnax, but not for ...p. I extend this use of do so to do sanething based on

the similar pattern t'hat these tw::> constructions have, as s~ in the

following excmples.

81

(108) a. John did sanething.b. *John did sanething about Harry.c. *John did sanething to Bill.d. John did saneth.i.ng on SUnday.

Whatever the analysis is which accx:>unts for the differences between

these pro-fonns, it seans clear that they can pronaninalize VI and vnax,but not v".

'!he naninal pro-fonn one seans to pronaninalize rP and N I, and the

verbal pro-fonn do so(mething) seans to pronaninalize V' and ~. Both,

then, support the claim that there is a branching node that is not a

maxi.mal projection.

I ProfOse a different view of these pro-fonns. One question that one

might raise concenrlng these is \tJhy the verbal pro-fonn differs fran the

nc:rninal pro-fonn. Q'le may be follOJ#1ed by canplE!llE!1ts while do so may not

be. '!he other instance \tJhere we h~ve seen canplements that may not be

left 1::>ehind was in the case of French causatives. '!here \E said that the

canplanents or their traces had to be governed by their 6-role assigner,

the verb. we might also say this in the case of verbal pro-fornlS, but

then \tJe tNOuld have to explain \tJhy the canplanents of nolIDS do not fall

under the same restriction.

What we would 1::>e claiming in the case of the verbal pro-fonn is that

it is the V' that is pronaninalized. It is not for reasons of structure,

then, that the canplanents do not appear, but for reasons of e-role

assignnent.

In order to be <x>nsistent, \\'e have to assume that argunents of nOlIDS

..can get their 6-roles independent of the noun. 'Ibis is not a ne,..r idea

(Rappap:>rt 1983, Higginlx>tham 1983). '!he argunents that are used to

supp:>rt this claim are:

1) Argunents of nOlIDS (as opposed to argunents of verbs) are

optional.

(i) the order (to the men) (to leave to\tJn)(ii) Robin ordered * (the men) *(to leave town).

2) '!he o-roles assigned and prep:>sit.ions used have a one-to-one

rnappiBJ·

82

(i)(ii)(iii)

(iv)

SUsan ordered Andy to behave.SUsan saw Amy.

the order l*~ \ Andy

the sight t*~ ~ Amy

3) OnJ.y NPs themat.i.cally related to the head rnUI1 may a.PJ?ear in the

argunent structure.

( i) Beth appears to be happy.(ii) *Beth I S appearance to be happy(iii) I believe Beth to be happy.(iv) *the belief of Beth to be happy

Let us assume, then, with &:lppap:>rt and Higginlx>tharn, that argunents

of an NP are assi'3I1ed Q-roles by the prefOsitions thanselves. rnris will

explCl.in why proncmi.nalizat..ion·of tiP does not hinder the appearanc~ of the

canplanents of that llOlm. A near minimal pair is given below.

(109) a. *We drove to r-hntreal and they did so to New York.b. '!he drive to M:>ntreal was prettier than the one to New York.

83

Jackendoff claims that of canplanents may not be left behind in

pronaninalization and gives the example of *the king of France and the one

of Ehg1and as evidence. '!he picture cases given above I however I seem to

argue otherwise.

'!here seem to be t\\O types of of NP phrases. Ibrnstein and Lightfoot

(1980) differentiate a stooent with long hair and a stooent of physics by

stating that the fanner denotes t'ltO properties ('being a student and having

long hair) while the second denotes one property (beiD,j. a stooent of

physics). '!he type of PPs that can ~cur with the pronolID. one are of the

first type, i.e., separate predicates. BelCM we can see that the facts of

of NP phrases patte:rn the same way."

(110) the picture of Julia and the one of SUzanne(Ill) '!his picture is of Julia.(112) *the student of physics and the one of chemistry(113 ) *'!his sttrlen1': is of physics.

The generalization appears to be that if the prcIX>sition can assign a

o-role independently of the head N, then it ma.y. rana.in when the N is

pronaninalized. While making RappaIX>rt I s and Higginbotham I S analyses less

clear I these data do supp:>rt our analysis. '!hus the claim is that sane of

NP PPS within an NP ma.y be dependent on the head N for their 8-roles (va.

PappaIX>rt I s claim) and that these Ws crucially may not appear with the

pronanina1 one.

My conclusion, then, is that in both cases I it is ¥? which is being

pronaninalized. In the case of the verbal pro-fonn, the canplanE!1ts may

not appear because they will not be assigned 8-roles. In ~ case of the

naninal pro-fonn, canplanents may appE'ar sL,,~e they are assigned 8-roles

84

indeperldently of the head notm.

2.3. 2. 3 Co-ordination

Co-ordination is traditionally used as a test for constituency. In

the examples belew, 'Ne see that there appear to be bar levels betwraen vnax

and ifJ.

(114) Mark wrote and directed the play in 1984. V(115) Mark wrot..e the play and directErl the nnvie in 1984. Vi(116) Mark wrote the play in 1983 and directed the rocwie in 1984. V"(117) Dianne placErl and arranged the flowers on the table.(118) Dianne placed the daisies in the vase and arranged the tulips on the table'.(119) *Di.anne placErl the daisies and arra.rged the tulips on the table.

(114), (115), and (116) are examples of conjoined Va, V· s, ffi"ld ',/"s

respectively. (117)-(119) suggest that non-constituents cannot be

conjoined. '!he argunent is not clear, hatJever. Put the flO'l~rs may act

as a corstituent as sh:Jwn belCM.

(120) '!he children put the flowers ard placed tile presents on the table.

If it IlUlst therefore be a constituent, then arrargErl the tulips \'tOuld

be the same sort of constituent since 00th consist of V-NP(obj). (119) is.\

not ruled out, then, because the conjoirlts are not like ccnstituents.

Rather it seans that the example is :culErl out becaus~ the constit'~lsnt on

the table has a different relation to the first conjoint than it has to

the second conjoint. It is a canplanent of put and an adjlU1Ct to see.

Another reason 'Why it is not clear that co-ordination ia a test for

constituent str~tUre is given in the following examples.

85

(121) Claire ate tha carrots raw am. th~ peas cooked.

(i22) Lytle ~ent a letter be Hasriet and a book to aill.

If co-ordination is an argunent for c:onst.i.tuent structure, then the

carrots raw and tha peas ~'<X)ked are const.i.tuer.lts in (121) and a letter to

Ban....let and a beck 1r Bill are constitUeIlts in (122).

Rather than claim ~t these are ccnstituents, one might look for

CU'X)ther account of co-ordination (see Q:x:dall 1984 for discussion).

~e aOOve section is intended to raibc the question of what notion of

brancrdng ia Ileeded. By r'estrictirg b:.anching p:>ssibilities 'Ne may also

restrict other parts of the granmar. For instance, by not allC1tt1inq the

existence of nodes other than xnax and Y?, we also do not allow any rules

that crucially refer to any intEl:lllediate projections.

Any restriction on the levels of branching will also have an effect

on r.otions of c-carmand. 'Ihe main distinction bet\tJe€.Il kltm & SfX'rtiche' s

defini,tion of c-camtand, anc Reinhart's defini.tion is that A&S allow

c-ccmna.nd up to the first maximal ;Toje~,onrJ

(123) . X"

~A 'Zll

X Y"

For NicS, then, bu,~ nClt for Ieinhart, X c-c:xmnands Z". Chansky (1981)

adopts Reinhart dXcept that heads extend theil- c-connand danain to the .

highest n<::>G.e in a chain of nod.es which Sl~~S the same features. For

86

Chansky, then, X \\Quld c-canna.rrl Z". HJwever, A&S differ fran Chcmsky in

that for A&S, Y" c-ccmnands Z" \\bile for O1ansky, it does not.

In the view of branching presented in this thesis, ITOst of these

definitions fall together since the structure in (123) \\Quld now be as

bel~/.

(124)

x

'!he first branching node will al'l«lys be a maxirna! Pl.·oj~t.ion. The

only difference nDII will be betwean Chansky on one hand arrl A&S and

Reinhart on the other; since O1ansky I S defini.tion of c-cnmtarld will allow

a bead to govern out of a maxi.mal projection in a Chansky-adjoined

stru::ture •

(125)

In (125 ) I for Chansky, but not for Reinhcu:t or 1\&5, X will c-carmand

wnax. We will argue in O1apter 4 against just such a notion of c-carmand,

and therefore will assune a catIDir!erl A&S and Reinhart definition, together

with the notion of branching presented here.

r.Ihi.s issue of branchirg and c-cannand is a fertile area for

discussion (see, aIta1g others, saito 1983b), and a direction for future

research.

\.

87

2.4 Constituents of ]NFL'

Now let us look at the level above VP, i.e., the constituents of S.

ihe S-V-o type of tYfXJICX3Y suggests that the relevant constituents are the

subject and the predi.cate (NP, vp). '!his means that the VP either follows

the subject or precedes the subject. IbIIever, 1Ne can quickly see that the

distinctions are not so clearcut. Examples are given belC7N' of Gennan and

Welsh constructions.

(126) Ieh bin ins Kino gegangenI am in-the cinana gone

•I have gone to the cinema.'

(127) Y ma.e Sian yn mynd adrefPVP is John YN go(VN) lone

I John is goirg bane. I

(Sproa.t, 1983)

In (126), the verb appears to be split bet\\leen bin and gegangen and

in (127) bet~n~ and~. Gennan presents a further problan in that

the V is unarcibig\Dusly final in embedded clauses,

(128) daB i ..ch ins Kino gegangen binthat I in-the cinena gone am

•that I went to the cinema'

suggesting a simple s~-v underlying order. Nevertheless, even if one

accepts this analysis (see Belch .1962, Bierwi..sch 1963, FOster 1975,

ihiersch 1978) I a rigid S-o-V/S-V-o typology does not facilitate the

expression of the verb novement rule that \\Ould be necessary for Gennan

root clauses, nor does it account for the Welsh facts. ihe problem in

Gennan is that only the tensed !=Ortion of the verb lOOVSS. 'lhiersch' s rule

88

is given b;!lCM:

(129) RI:R2:

... ,... ,V[+tense] -> V, •••XII [ -verb] -> XII, •••

(Thiersch, p.38)

Sim.ilarly in welsh, it is the sente:n.ce-initial part of the verb which

is marked for tense. If we assure that INFL is the elanent which

expresses tense, we can account for this apparent split of the verbal

constituent with the INFL node (see safir for Gennan, Impnan for Illtch,

Sproat for Welsh). Accept.i.ng this argunent, then, \tie add !NFL to the

s-v~ tYIJe systan. 'lhe relevant c:onstituents are T:JCM S, V, 0, and INFL

(I), and the relevant manbers of II (or S) are S, I, and VP.

Benefiting fran the previous section where we discavered that the

order of constituents of the verb phrase deIJend upon the relations they

bear to one another, let us start this investigation by looki.ng at what

rela·tions might exist bet".l!.;..,.en the S ccnstituents . '!here are three

pairings: NP and INFL, INFL and VP, and NP and VP.

2.4.1 NP and !NFL

!NFL assigns naninative case to the subject NP. If case is assigned

by INFL ill the Sam:! rranner in which case is assigned by, P and V, \\e \\Quld

eX}?E!ct certain pro}?E!rties to hold. For instance I we \'JOuld expect !NFL to

gO\Tern the subject NP. Also, if the language has an adjacency requirenent

on case assignnent or a specified direction of case assigrment I a~ we have

seen for Chinese, we \\Quld expect these C'Onditions also to hold for INFL

and the subje:.~t NP.

\ !,\

89

I will suggest here, lxMever, that case assigrment by INFL is

different fran case assigrment by P or V. First, we can see that the

adjacency condition does not hold in the same way for INFL arrl the subject

NP as it does for V and the obje::t NP.

(130) *1 like very much books.(131) I certainly like books.

Mverbs may intervene be1;:.\\1een !NFL and the subject though they carmot

intervene between the V and the object.

Also, 'He can see in Clinese, where the direct.i.on ()f case assignnE!1t

IlUlst be specified, ttat INFL assigns case in a direction different fran

the direction specified for the P and t1 ~e V.

(132) ta gei \\0 mai Ie chezi Ie (I.&T, 1975, 180)he for me sell asp car !NFL

'He sold a car for me.'

(133) ta mai gei \VO chezi Iehe sell to me car INFL

'He sold a car to me.'

I will assume here, and discuss in more detail in Cllapter 5, that

INFI· nulst be c:>i.ndexed with the subject NP, and that it Imy be via this

coindexation that nan.inative case is assigned. There may be no conditions

on this coindexation beyond the stipluation that the INFL arrl the subject

NP must be sisters. We will see below that thi.s sister r~lationshipwill

fol1011 fran the requiranents on INFL and VP, and VP arrl the subject NP.

2.4.2 !NFL and VP

INFL is the head of the !NFL' project.i.on and it takes VP as its

90

canplanent. 1.his means that I and VP are in a head/canplaY¥3nt relation.

'Ibis is th relation which we 'l1.ave seen in our discussion of the

constituents of VP, i.e ., that between a V and its internal argunents.

'Ibe relative ordering of these t\\O constituents may be aCCOLnlted for

through the head-initial/final parameter. As \..e saw in Chinese, this

IJarameter may vary for different categories. III Archaic OLinese, NPs were

head-final while VPs ~re head-initial.

Another observation concerning thi.s relationship is that we do not

expect the canplanentation datain to be broken. '!hat is, 'Ne expect the I

and the VP to appear adjacent to one another. This tendency will be

discussed further in O1apter 3. 5.

2. 4. 3 NP and VP

We know that the VP assigns a e-role to the SUbject NP, though it

does not assign it case. Again the question is \'Jhether this 6-role,.

assigrment is the same as the e-role assignnent of a V or a P. Above, we

detennined that the case assignnent pro}?erties of INFL differed fran the

case assigrment properties of V and P. Here we will be assurrting that the

6-role assigning pro}?erties of VP are differer,;.t. fran the 9-role assigning

pro}?erties of V and P.

The asyrrmetry bet\\een subject 8-role assigrnnent and object &-role

assigrment follOlt'S fran Williams' (1981) distinction of internal versus

external 8-role assigrment. External 8-roles must be realized on

constituents that are external. to a maximal projection, in this case

91

external to tl~e VP. Within the VP, the V, as head of the projection, takes

certalll constituents (NP and pp) as canplements and assigns these

canplenents 6-roles. An object NF, then, receives its internal e-role

fran the head of which it is a canplement. 'lhe· subject NP receives its

6-role cOlqX)sitionally (see Marantz 1981 for details on this) fran the

maximal projection of VP. 'lhe subject is neither subcategorized for by the

VP, nor is it a canplement of the VP, since by definition

subcategorization and canplementation are properties of heads not phrasal

categories. (In Olapter 4 'Ne will suggest, however, that an NP Subject

may be the canplanent of a VP, contrary to the present definition of

canplements.)

If the external a-IOle is not assigned through canplemaltation, then,

h:Jw is it assigned? .Again, followirg Williet:1S (1981), I will assune that

this external 6-role is assigned through predication. It is fran this

relation of predication that the tri-partite division of I I will follcw.

!NFL and VP nust be sisters since VP is the canplanent of the head of I I

which is INF'L. VP and NP must be sisters because of tlleir predication

.celation.

(134)

~erefore, thE:= structure of I' is as shown below.

'!his confinns our conclusions concerning branchingness. Q1ce agai'.1

we are forced to collapse all project.ions bet~en lexical categot-ies and

maxi.mal projections •

Since the e-rnarking relation bet\\een 'Che VP and the subj~t NP is

92 .

different fran Jc.he one bet\fJeen the V and its argunents, \..e \\Quld expect

other differences to appear. For instance, the direction of predication

!nay be different fran the direction of direct e-rnarking; i.e., internal

arguments may appear to the right of the V vihile ext.ernal argtments appear

to the left~ of the VP. Again, in Chinese, \tJe see that thi.s is the case.

Even in ~l, where a-marking has to be specified as being to the right,

the Q-role assignErl by prErlication is to the left.

(135) [ta]i [mai gei \\0 chezi]i Iehe sell to me car INFL

•He sold a car to 1llE!. •

English also presents a clear example. '!he V assigns a e-role to the

right, but at the level of prErlication, the E)-role is assignErl to the

left. In tenns of the parameters so far discussed, I would say that

English is head-initial, but prErlicates occur on the right of the

ub ·ect21S J •

2. 5 Word Oro.er and Typol~

The Pl1Ip:)se of this section is to describe a restrictive means for

developing a \-tOrd order typc>103}' usirg the parameters that have been

prop:>sed above. 'Ibis method will be contrasted with the restrictions of

Steele (1978) which TNere· also prop:>sErl to account for the rarge of

possible variatiorts that a langtBge can evidence. I will discuss Steele I s

21. A clearer picture of the forces at v.ork at the level of I I is given insection 3.5.

93

constraints and. their use within a typ:>l03ical theory. '!hen I will

can}?ClIe the tw:> methods, first on a theoretical level, and then on an

atpirical level. In this last part. I will give data fran Steele which at

first look problanatic for the GB account, but I will SI10W h<.M these data,

under a different analysis, FOse no problem. In fact, the neN accounts

provide llot only ITDre illuninating analyses of the individua.l languages I

but also confinnation for the restrictions that the parameters iroFOse on

possible gramnars.

2.5.1 Steele

Steele (1978) looks at the basic word ol:der and \\Ord order variations

in sixty-three languages with the plJI1X)se of determining what constraints

play a role in restricting the range 0:[' variaJcions . She proFOses three

constraints which in turn enter into her definition for rigid, mixed, arid

free \\Ord order languages., '!he survey is basically stat.i.stical. For each

base w:>rd order, she li!;ts its FOssible 'Tariations as 'being very eamon,

CClI:uOll, not carm:n, uncamnn, and non-existent depending on the relati~ re

nlJllber of languages within that group that evidence that alternate \\Ord

order. Her table is given 'below..

94

(136)

Very Canton

Ccmron

N:)t Uncamon

Uncarrron

N:)nexistent

\lOS VSO sov SVU

VS:J va; a:NS\U

svo

ViS

ovs ovs vasOOV \lOS VSOSOV VSO SOVSVi ViS \Y3V

ovs&:IV00\7osv (sic OVS?)

(N:) examples are given of <l3Vor OVS)(Vi = intransitive verb)

The three constraints that Steele prop:>ses to explain the clustering

of carrron and uncarrron variations are as follQ\tJS:

A: A variation on the 'basic \\Ord order in which the ~lerb cx::curs in

other than its pJsition in the 'basic \\Ord order is to 1Je avoided. (p.

602) (e.g. SVU languages will not 1Je expected to have a vas variation)

A': A variation on the 'basic \\Ord order in which the verb occurs

either initial or final to the clause is to be avoided, if the verb was

neitller initial nor final respectively in the ba.sic order. (p. 602)

(e.g. VSO langua.ges will not be expected to have &)\7 variations)

B: A variation on the 'basic \\Ord order in which the Object precedes

and tile subjec:t follows the verb is to be avoided. (p. 604) (e.g. no

lan:JUage is eJC"J?eCted to have an OVS variation)

Her claim is tl1at V-peripheral langlsges (VSO, \OS, 80\') ~nd to obey

95

C'Onstraints AI and B, while verb medial larguages (she only gives examples

of S\U) tend to o"bey A and B. By looking at which languages violate which

constraints, Steele claims that the lal'lguages can "be put into three

groups: rigid \'tOrd order languages violate neither of the constraints

relevant to than, mixed \\Ord order languages violate only one relevant

constraint, and fr~ \\Ord order larguages violate roth constraints.

'!he aCcolIDt of typ:>logy presented in this thesis differs fran that

presented in Steele in three respects.

1) No langtlClge will have a catalog-ue list of lx>ssible variations.

All variations IlUlst follCM fran a cbJice of parameters that must be

Sl1pp:>rted by inde}?endent evidence.

2) Principles and parameters make strong predictions that cannot 'be

violated. A violation will not r:ean that the larguage is a m.i.xed \\Ord

order language as it does for Steele. Father it means that the system of

parameterd I'CD.lst "be r~lised so that the recalcitrant language can beI

. described.

3) Free tNOrd order langUClges are not langlBges that break the most

constraints. -Father they are larguages that set the parameters

differently. For instance, case may "be assigned in the lexicon so that

adjacency is never ne-:~ASal.~Y for case assigrment.

2.5.2 Possible counterexamples and the solution

'Ib make ~he difference bet'Neen the t\\O kinds of typ:>logy clearer, let

96

us look at sane of Steele' s data. '!here are three languages that she

labels "mixed 1hOrd order" because they each violate a single constraint.

Initially, these languages appear problenatic for the GB aC(X)tmt.

2.5.2«1 Diola-Fbgny

Diola-FcxJny is a Congo-l<'Drdofani.an langtlage, listed as having a ba.se

\\Ord order of S-V-O with an S-O-V variation. '!his violates Steele's

constraint A since the malia! verb now appears last.. !b\twever I for. Steele I

since her aca:>unt is descriptive rather than predictive, this is not

problanatic. It simply indicates that the language is of a pcrrticular

type, tha.t is, a mixed \\Ord order l~age.

This does, h0W3ver, present a {X)ssible count.erexample for the

approach to typ:>lcxy presented here. '!here is no novanent rule which will

allCJfN a {X)st-verba.l Object NP to appear in a sentence internal pre-verbal

p)sition.

rrhis stat..ement incllrles many assumptions whi..ch, while still

tmexplained, are fairly standard. 1) Only CCMPs on the left are p:>ssible

landing sites for JTK:JVallent. 2) Argunents ("f the V may move to the right,

subject to the Right lbof Constraint. 3) NPs may no-t move witt-Lin tho VP

to the left if the VP is specified as being head-init.i.al. 1lle variation

fran a ba.sic S-V-O to S-O-V \\Ould involve either JTK:JVanent of the object to

t."'1e left of the verb violating (3), or movanent of the V to sentence final

p:>sition. Altlntgfl' V-fronting is p:>ssible because of CCMP, V-backing is

not because of assumption (1 ) and ( 2) •

97

If the idea of paraIMters is ri.ght, there is also no way that t\\O

separate d-str\X:tures can be generated without saying that the verb

sanetimes assigrls case to the left and sanetirnes to the right. If the

verb only assigns case to the right, an object to the left of the verb

will violate the Ca.se Filter. Since Diola-Fogny is not a free \\Ord order

langua.ge, we knQ\tl that Ca.se ~'"s not assigned in the lexicon but must

receive it fran "a case-assigl1er is the syntax.

A review of the granmar of Diola-Fogny which Steele cites (sapir

1965) reveals that the OOVorder is very restricted. It is only p:lssible,

in fact, if the object is a concord pronotn and the subject is what 8a.pir

~alls a "disjtmet subjectll• An example is given belCM.

(1.37) injt no nikakat (sapir p. 101)I this left

'mris I left (\\On't have anything to do with) • '

'!he form of the first person prOIX>UIl is the independent fonn, and

presumably in subject p:lsition. We may, then, asstnlle that m), the concord

pronoun, acts like a clitic on the verb as in the French example given

below.

( 138) Subj pron-V ec (ec--anpty category)

Marie la voit.Mary it(FF1~) sees

'Mary sees it.'

What is crucial is that for the obj~t tc appear bet\tJeen t..he subject

and the verb, it IlUlst be pronaninal and, therefore, can "be analyzed as a

clitic.

rrhere is another means for fronting full NP objects for the pu.q:ose

98

of :fc)cusing the NP.

(139) ebe nijukE.CCIIl ls-Stlw

•I saw a CC1Y1. •

(sapir p. 36)

~ver, if this structure co-occurs with a disjunct.i.ve subject

proI"X)UIl, the object precedes the pronolUl suggesting that the Object is in

the aMP position.

(140) ebe inje ijukE­CCIIl I saw

I I saw a CCIIl. I

(sapir p. 36)

Diola-Fogny, then, under this new analysis, does not violate Steele's

constraint any m:::>re than French does. '!he verb is not final in these

stroctures since it is follO\\ed by an anpty category. Under Steele's

classification, Diola-Fogny is no longer a mixed \\Ord order language. In

tenns of this thesis, Diola-Fogny is no longer a counter-example to the

parameter of direction of· case-assignnent. lexical NP objects can only be

assigned case in the p:>st-verbal p:>sition. 22

Chinese and Olorti are different kinds of cotmterexamp1es for the

parametric account. Steele considers then 'both mixed 'I.Ord order languages

solely because they appear to "have pre-verbal subjects when the verb is

transitive, all(] lX'st-verbal subjects \'Jhen the verb is intransitive. 'Ibis,

-3.gain, is a violation of constraint A since the verb will be sentence

initial in an SVO language.

22. For accotnts of how clitics might be gellerated, see Jaeggli 1980, PDun1981, Ebrer 1981, am the references cited therein.

99

In tenns of parameters this variation \\Ould Illt.an that the direction

of 6-:role assignnent of the VP would have to be sensit.ive to features of

the verb. I claim that neither Chinese nor Chorti neErls to resort to this

sort of sensitivity to account for the variation of \\Ord order.

2.5.2.2 Chinese

In Chi.nese there are three imJ.X>rtant facts to I"ot.:i.ce. '!he first

involves the set of verbs that allow J.X>st-verbal subjects, the second is

that inverted subj-=cts are indefinite, and the third is the fact that

Chinese has no pleonastic elanent (Huang 1982).

t.baccusative verbs. In C".inese, not all intransitive verbs allCYtI

what Chao (1968) calls "inverted subjects" (i.e., subjects in J.X>st-varbal

position). '!he list of verbs that he gi.~"es whi.ch allQ\al this inversion is

a subset of his list of intransitive verbs. 'lhe verbs that allOli inverted

subjects are verbs of caning and appearance, going and disappearance.

1his senantic class correlates with Perlmutter's (1978) sanantic

characterization of tnlaccusat.i.ve verbs.

Once we have established that post-verbal subjects only occur in the

cases of unaccusative verbs, ~ can use Perlmutter's analysis for Oltch,

and 8.lrzio' s analysis (Burzio 1981) for Italian and assume that the single

argunents of these verbs are d-strtX!ture objects of the verb.

(141) ec [vp V NP]ec arrivel a man

The verb assigns the NP its 6-:role in obj~t fOsition. How the NP is

100

assigned case is c..untr0\7er~ial. Unaccusative verbs are, by definition,

verbs that do not assign accusative case to their logical direct objects.

I aSSllle, lDw'ever, several analyses that do allCM these verbs to

optionally assign case to an argunent whi.ch they govern (see Levin 1983,

lbthstein 1983, am::>n;J others, for details) am -will leave further

discussion of this issue till Cllapter 5. We will asstnne, then, that the NP

follO\\'ing the verb gets case by direct assignnent fran the verb.. '!his is

necessary because of our earlier account of (J1.i.nese. Cnly elements that

are assigned case by the verb can appear fOEst-verbally.

Imefi.ni.te subjects. Li & 'lhanp;on (1975) IDte that fOst-verbal

subjects are preferably iniefinite. 'lhis further confinns the above

analysis since certain definiteness effects are often found in these

unaccusative ex>nstructions. '!his is obvious in the Erglish cases given

be10'11. 'lhe first is an example of a grarrmatical str~ture with an

unaccusative verb and an indefinite NP It 'lhe second is out because of t..'1e

defini.te NP, and the third because the verb is not unaccusative. 23

( 142) 'lhere came into the house an erxJIlllOUS bro\tJn dog.

(143 ) *'!here cane into the oouse the erx:>mous braNn dog.

(144) *'!here cried into the handkerchief a tired old man.

Pro-dI:cp. 'lhe obvious difference between the English examples above

and the Chinese excmples leads us to the third :i.mp:)rtari.t fact, Chinese has

lX) pleonastic elanents. 'lhis is .imp:)rtant 'because if the subject fOsltion

23. For roc>re details on these effects, see Safir 1981 and the 'NOrks cited.therein.

101

al'Nays had to be filled, it \-.Quld be difficult to argue that, when Chinese

n.ppears to have a V~ sequence with an intransitive verb, there still is

a pre-verbal subject p:>sition. ~D3 that ati.nese is pro-drop, \\Ie nay

assume that what Steele has described as an S\O IV ViS alternation is

actually a pro V NP instantiation of svo. Stated this way, the problem

disappears. The alternation simpiy consists of a difference in whether or

not the single argunent of the verb is base-generated internally or

externally to the VP.

2. 5. 2. 3 Onrti.

Chorti appears to fall under a different analysis alth0U3h more data

are needed to be sure of these results. '!he imp::>rtant fact about Clnrt.:i

is that it has an ergative system for marking agreemmt with the verbal

argunents. Transitive subjects are marked on the verb by a set of

prefixes, while objects and intransitive subjects are marked by a set of

suffixes (~ 'I', -et 'you', -# 'he', -on 'wei I, -ox 'you' pl. I -op 'they'

(Oakley 1966 p. 244».

(145)

(146)

uy-alq' u-enhe-give-me

/lo?k/vi/enleft-I

Oakley, p. 245

Fotght, p. 68

FOu;Jht (1973) writes:

Chorti ..• is fundama,1tally ergative even in itssuperficial strt.rture. '!he subjects of transitive expressionsare concordially and syrltactically dist.i.nct fran the subjectsof intransitive expressions, \tIhile these latter are equivalentto the object of transi·t.ives.

102

(Fbught, p. 66)

'!his suggests that the assigrrnent of syntactic realizations of

argunE!'lts to thanatic roles goes according to the Ergative Hypothesis

prop:>sed by Marantz(1981) (see also I.eJin 1983). Marantz suggests that

sane languages cl'oose to map patients of transitive verbs onto the

syntactic subject p:>sition and agents to the syntactic object p:>sition.

(147) ACCt5ATIVE J~GFSi ject Ob1ect

Agent Patient

ER~TlVE I.AN3UAGFSSub ·ect Object

hJenX~entIf this ~e the case for Onrti, as is suggested by the case-marking

system, it \aK)uld 1Je an OVS language rather than an S'\,t) langlBge, and the

fact that the intransitive subject in found p:>st-verbally \\Ould be no

surprise.

For both O1.i.nese and Olorti, IDW, there is no ananaly in the

p:>sitioning of the subjects of intransitive verbs, altlDugh each larguage

accounts for this by different means. In the table 1Jelow we see, in

coltrnn I, Wny Steele's characterization of 1:oth the Chinese type of

langua.ge and the O1orti type of langua.ge look as if they are

CX)tmter-examples to the predictions made by the \\Ord order parameters.

Yet when we offer a differerlt analysis of the langua.ges, as shO\lJn in

colum II of the table,· the facts corroborate the prediction.

(148) I Steele I s II a. ChineseChinese & Onrti

b. Olorti.

S V 0

Vi S

S V 0

pro V 0

o V S

Vi S

103

In Chinese, VPs assign &-roles to the left, and in Cl'x>rti, VPs assign

e-roles to the right, whether the verb is transit.i.ve or int,!-ansitive.

'!he claim of this section is that the llDre restrictive view of

typ:>lCX3Y offered by advancements within the gramnatical theory of GB,

pushes the researcher towards profitable reanalyses of languages that

appear problanatic rather than allO\lling problems to be hidden lUlder a

mixed word order heading.

2.6 Discussion

2.6.1 SUnmary

'1tle PUI}X)se of this chapt~~r was to set up an account of \tJOrd order

which factored another task of Phrase Structure Rules into other

eat1J:X>nents of the granmar. By assuming the the direction of pred:i.cation,

e-role assignnent, case assigrrnent, and headedness may be set through

language specific parametric values, one may predict certain \\Ord orders.

In Archaic OUnese, VPs appear to be head-ini.tial. All non-head

elanents in the VP appear after the V, inclLrling the subcategorized and

non-subcategorizecl PPs.

'1tle first fonn of r-tXlern Mandarin described (M-11) has reanalyzed

certain verbs as prep:>sitions with the result that non-subcategoriZed PPs

(PP2s) appear preverbally. N:7#I only canplements of the verb appear after

the verb indicatirg that the verb assigns e-roles to the right, and the

104

default parameter is head final.

The second fonn of Mcu:ldarin (~2) has only bare object 1'lPs following

the verb while all PPs, sU1:x::ategorizerl am non-subcategorized, ap:Pear

before the verb. We can now say that the verb assigns case to the right.

'!he default parameter is still head-final.

(149) \\Ord order I headednessI

AC: V 0 PPI PP2 l1nitial1

r+11: PP2 V 0 PPI I filmI

r+12: PP2 PPI V 0 I finalI

O-~les case

right

right

2.6.2 CUestions

CUestions arise concerning this sort of system. 'lWt> that I shall

discuss are (i) What sort of paranetric (X)nfigurations srould \t.e expect,

and (ii) \tIhere do the parameters take effect?

N:>tice in the table abcNe, \lie never specify IlOre than t\ttO parameters

althoU3h conceivably there could be a langua.ge with the VP word order of

PP2-Q-V-PPI. Presunably the V in this larguage \\Quld assign o-roles to

the rigllt, case to the left, and the VP would be head-final. Given the

sma.ll nunber of larguages involveci in this stu:iy, I hesitate to deny the

existence o~ such a language. }b,.ever, ~ the interest of being as

restrictive as J:X)ssible, I will prornse that only on~ parameter ma.y be set

outside of the default case. '!his means that if the direct.ton of case

. assignnent rmJst be specified, then the direction of e-assignnent may not

105

be specified, and "ice versa.

TlLis can be taken one step further I in fact. Let us say, without

reference to the default parameter, that only one \\Ord order parameter may

be stipulated for any projection. How then can we accotmt for r+11 and

~2, where a value for the headedness parameter and one other parameter

must be indicated? let us assume that we have a langlBge that, li.ke M-11,

assigns e-roles to the right, but, tmlike MMl, is head-initial. Instead

of being PP2-V-o-PP1, it 'NOuld be V-o-PPI-PP2. 'Ihi.s, ho\\ever, is

indistirguishable fran a simple head-initial \\Ord order. 'nle same applies

to the direction of case-marking. If case-marking 'ltere to the right and

the VP 'ttere head-initial, the outcane still 1AOuld be V-o-PPI-PP2. let us

assume, then, that if the direction of e-rnarking, or case-marking i~ set,

then the default parameter \\Orks in s~h a way that all th:>se elenents

which do ·.. lOt fall under the already sfleCified parameter (PP2 in the case

of e-rnarking, PPI and PP2 in the case of case-marking) will be on the

opIDsite side of the head. With this assllllption \..e can say that only one

parameter need be specified.

The second question concerns the effects of such a view of

parameters. Where do these paraneters take effect? '!he ans\\er divides

bet~en d-~ture and a-structure if \tie assume that d-str~ture is the

"pure representation of GF-eu• If this is the case, then the di.rection of

9-m3rking should ha.ve an effect at d-~ture, but not the direction of

case-rrarking. I proIDse the following d-structures for the VPs of Archaic

Chinese, t+11, and M-12.

106

(150) a. ArchaLi(~ Chinese:[ V NP PPI PP2]

b. r+11:[ PP2 V NP PPl]

c. r+12:[ PP2 PPI NP V]

Notice that the direction of e-rnarking follows the head-initia1/ final

parcmeter tmless otheiwise specified. In a head-ini.tial VP (148), all the

canplanents (as 'Nell as non-canplanents) follow the verb. In a head-final

VP (150), all the canplanents (as \tJell as non-canplanents) precede the

verb. In (149) \tee can see that, thoU3h the VP is head-final, the

canplements f011011 the verb because the direction of e-role assignnent is

explicitly set.

In Ehglish \¥e see an example where only head-initial/final need be

specified and the direction of e-role assignnent will follOit'. '!he VP is

head-initial so canplements follatl the verb (151). 'Y?s, }-x)wev-er, are

head-final, so ~lanents must precede the verb.

(150) keep bcx>ks

(151 ) bookkeeping, bookkeeper

We can also see in the prop:>sed d-structures for PC, rvt11, and r+t2

that the direction of case-assignnent has no effect on the d-structure.

'!he effects of the direction of case assigrment are seen only at

s-structure (just as the effect of passive verbs and lJnaccusative verbs

appears at s-strooture). I am assuming, then, that the following sentence

in t-f.12 is derived as sh:>\tm belOll.

(152) d-structure:s-stru::ture:

~

ta Lisi pian-Ieta t. pian-Ie Lisi.

~ :L

107

'Ia. pian-Ie Lisi.•He cheated Li.si.·

Since headedness is not the specified parameter, the obj~t may move

to the right of the verb.

Thotgh 'both direction of case-marking and direction of o-marki.ng can

be seen as deriviDJ "'Ord order, they do so because of di.fferent effects at

different levels of the grammar.

The rarity of languages whi.ch set parameters other than

head-initial/final slXJgests that the default parameter is the least

marked. In the next chapter we will lCXlk at the 'NOrd order of a specific

larguage fa:nily am speculate on the historical derivation of \'wOrd orders,

and a1 the problan of markedness.

108

Chapter 3

In this chapter I investigate certain issues of word order in

Gennanic languages. '!hese langua.ges present s:PeCial problems to any

accx:n.mt of \\Ord order because of the variations evidenced within one

langua.ge. In the first section I give an accotmt of the problem using

Gennan arrl llltch as the languages of investigation. Here I describe what

is generally assumed to be the correct analysis of these variations (Bach

1962, Bierwi.sch 1963, den Besten 1977, 'lhiersch 1978, 5a.fir 1982, l(copnan

1983). Il.1tch and German "-Ord order are presumed to be tmderlyingly S 0 V

I. Fronting of !NFL and sane maximal projection in root clauses will give

the well-krlaNn verb-second (V2 ) effects.

(1 ) [ Xu fl. ••• t I ••••

1.. J 1.

1t.]J

In the second section I canpare Yiddish word order with German word

order and sl'rM b:Jw the accepted analysis for Gennan is not adequate for an

account of Yiddish. First, it is clear that the VP in Yiddish is

head-initial (V-o) wile in Gennan it is head-final (O-V). Secondly, I

1. '!here is disagreement as to where, exactly, these elements front to.For this I·eason I leave out any details of bracketing lIDtil later in thediscussion.

109

give t\\O argunents that. preverbal subjects nn.1st be distinguished fran

pre-verbal ron-subjects in Yiddish, and that therefore Yiddish must be

S-I-VP rather than I-S-VP.

In section 3 I argue that there is empirical evidence that supports

an analysis of Gennan and Dutch \tlh.:lch trore closely resembles the analysis

of Yiddish. I claim that German and L\1tch, like Yiddi.sh, are S-I-VP, but

Gennan VPs are head-final, \tlhile Yiddish VPs are head-initial.

In section 4 I not.i.vate the novement rules that I have proposed in

the previous secti.ons. 'Ihis involves a discussion of the rrovement of

heads since IO is fronted, not max. 'Ihis also involves an investigation

of the structur~ of CCle1P. I will clai.m that the head of a root CCMP must

be filled by the rrovement of !NFL, while the heads of other m1Ps exist

either tlu:ough subcategorization and feature specification or by the

lexical spell-out of the features of the head.

In section 5, I specUlate on the developnent of Gennanic \-JOrd order

and propose 1) that 1NCJrd order change can develop out of a contradiction

in the danands on the p:>sition of !NFL, and 2) that rules in syntax, as in

phono1o:JY, tend to ItDVe fran surface levels (PF) toward the lexicon

(d-structure) .

Finally~ in section 6, I speculate on an overview of word order facts

with the intent of explaining Why certain orders appear to be rarer than

others. I also discuss the :lmp::>rtance of a separate !NFL node and a

tripartite structure of I'. 'Ihis section is included with an eye to future

research.

110

3.1 Gennan and rlltch: the accepted, analysis

In Gennan, tlnugh \fJOrd order is fairly rigid (as oPPJsed to languages

such as Warlpiri), there is a variation which is inmerliately not.i.ceable

between rc:xJt' sentences where the inflec:ted verb is always second (in the

literature thi.s is calle the V2 effects) (2) and embedded sentences which

are always inflected verb-final (3).

(2) a. Die Frau hat das Buch gelesenthe 'NCIlla1l has the 1:x:ok read

I • The 'NCIlla1l has read the 1:x:ok. •

b. Das Buch hat die Frau gelesenthe l:xx>k has the 'NCIlla1l read

•The 'NCIlla1l has read the 1:xxJk. •

(3) Ieb glaube, daB die Frau das Buch gelesen hat.I believe that the 'NCIlla1l the 1:xxJk read has

•I believe that the~ has read the l:xx>k. •

In (2) we can see that the inflected vert? is in second PJsition

whether the subject NP, (2a), or the object NP, (2b), is S-ini.tial. In

(3) the inflected verb is clause-final in the embedded clause.

111

The generally accepted analysis is that Gennan an:l Dutch are underlyingly

verb-final (see Bach an:l Bierwisch for the earliest l1Otivations) 2. This

will account for the fact that even in rcx:>t clauses where the inflected

verb is in second p:>sition, the uninflect.ed vel"b (4) or the separable

prefix (or particle) (5) is still sentence-final.

(4) Sie hat das Buch gelesenshe has the bcx:>k read

•She has read the 1:x:x:>k w •

(5) Sie macht das Fenster zushe close the wirrlow

t She closes the window. t

(zunachen = to close)

Assuming that the verb is sentence final, we can say that the

D-structure 'NOrd order of Gel:man, witlDut !NFL, tttOuld be S-O-V (i.e., S-VP

with a head final VP). Now the question is \tJhere !NFL is base-generated,

an:l b:M the variations in its placanent are accounted for.

I will be assun:i.ng that the s-structure p:>sition of INFL can be

detennined by the p:>sition of the inflected verb. In sane laD3UClges, tlU.s

inflected verb may be a pleonastic verb such as do in English. In Getman,

the inflect.i.on appears on a main· verb. fbtI INFL and V cane to be t03ether

will be discussed in detail belCJW.

'!here are three cOOices, then. (1) Inflection may lTC\Te at

2. I will be using roth German and Il.1tch examples in this section. Thiscarries with it the asstmption that the t\tJO lan:JUages have the same wordorder in the relevant constructions wch may not be the case. In thisthesis, b:Mever, I will be making this assumption. Also, the analysisthat I present here is a synthesis of several analyses (den Besten,Thiersch, safiz-, Kcx:>pnan). Since I em up arguiD3 against sanethiD3 whichis ba.sic to al.l of these accounts, I will continue to group them into one.

112

a-structure to a p:>sition adjacent to the verb. This sort of rrovement

will leave a trace and therefore fall tmder the binding conditions and

EI:P. (2) The verb may rrove to !NFL, ani again must confonn to

restrict.ions on novanent. (3) If rnFL is left stranded., it must be

lexicalized by sane dUIrITIY inflectional elanent such as •do· in English.

I will assume with den Besten (1977) that rnFL is in a:MP when it is

fronted (see 5afir 1982 for a diffa·ent analysis). Den Best.en I s argument

is that subject cl.itics can appear either to the right of CCMP in embedded.

S •s ani to the right of !NFL (the inflected. verb) in rCXJt S' s •

(6) Heeft gisteren Piet/*ie die film 1"XX3 kunnen z.ien?has yesterday Peter/*IE the film still be able to see

·Was Peter still able to see the film yesterday? 0

(7) dat gisteren Piet/-*ie die film nag heeft kunnen zienthat yesterday Peter/*IE the film still has be able to see

•that Peter was still able to see the llDV'ie yesterday'

(8) Heeft ie gisteren di.e film oog kunnen zienhas IE yesterday the film still be able to see

'Was he still able to see the film yesterday?'

(9) Gisteren heeft ie die film nag kurmen zienyesterday has IE the film still be able to see

'Yesterday he was still able to see the film'.'

(10) dat ie gisteren die film rx:q heeft kunnen z.i.enthat IE yesterday the film still has be able to see

I that he \tRiS able to see to see the film yesterday. I

In (7) and (8), we can see that if sanethiD3 inteIVenes between the

subject and either !NFL or CXMP, the 1Neak fonn of the prOrx:lUll is not

p:>ssible. EiJwever, if the SUbject aPJ?P-ars next to nm·'L or CCMP, as sl'otm

by (9) and (10), the ocmi.native prOrDun cliticizes. '!he rule of

113

cliticization \\Ould 'be much neater if the inflected verb were actually in

CCMP. It could then be stated that nan.i.nat.i.ve pronouns cliticize to an

adjacent CCMP but to no other adjacent category.

(11)

(12)

[COMP-c!. [ e. 0 V I]]J oJ

[ X" I.-cl. [ e. 0 V t.]]1 J J 1

dat ie

heeft ie

we might then say that in Gennan and in I:Utch, INFL is generated .

finally, making them head-final languages. The V2 effects are created by

m:JVement of INFL to sentence init..i.al p:>sition foll<:J'Ned by the rrovement of

sane max.i.mal projection. Thiersch achieves thi.s in a rule oriented

frante'ltOrk by t1NO ordered fronting rules. let us assume for the m::ment

that there is sane means by which a V will m:JVe into INFL thereby creating

a V(+tense]. '!he m:JVanent of INFL, then, will be the same as the IOOVement

of V[+tense] •

(13) Rl:R2:

... ,... ,v[+tense] --> V, •••X"[-verb] -> X", •••

(Thiersch, p.38)

If 'Ne assume that 'Ihiersch I s front..ing rules actually neve elements

into CCMP, they will 'be blocked in embedded sentences because CCMP is

already filled and cannot 'be doubly filled. '!his analysis avoids the

problem of a rule of SUbject Awe Inversion. Sines the inflected verb will

always IlDve to the front of a root sentence, if the subject is not the

constituent. ~ch topical.i.zes, then. the subject will altNays follow' the

inflected verb. In other \fJOrds, a Subject, in ita d-struC"ture IX'sition,

will always follCM the inflected verb.

In a GB fr~rk, where IOOVetlents must be notivated and restricted

by principles of gramnar, Safir and Kcopnan prolX'se notivations for

114

fronting rrovanents (see Safir 1982 and F'a:>pnan 1983 for details). At thi.s

p:>int, oo\tiever, I am concerned only with the descriptive levels of the

analysis. The theoretical issues will be di.scussed in section 3.4.

3.2 Yiddish

In this section I first c.unpare the \ttOrd order of Yiddish to that of

Gennan. It is quickly obvious that a Thiersch-type of analysis is not

p:>ssible for Yiddish in spite of the fact that Yiddish also has V2

effects. The secom part of this section will be concerned with fiming

an adequate analysis. The conclusion I reach is that Yiddish must 'be

S-I-VP and it must have a Subject-Awe-Inversion rule which is triggered in

sane way by topicalization.

3.2.1 Yiddish is tnt INFL-final

The first obvious difference in Yi.ddi.sh is that the verb phrase is

head-initial (see Hall 1979, hJw'ever, for arx:>ther dialect).

(14) ikh oob gekoyft dos bukh in kranI have bought the book in store

I I bought the book in the store. I

(15) Ikh heyb on mayn heymarbet.I start- my hanert.Ork

•I start 'my hc:me\\ork. •

(onheytm = start)

In canparing these to the Gennan examples Where tWt:> part verbs IlDVe

only the inflected~ of the verb to sentence-secom p:>sition, while the

115

<Jther, uninflected part ranains sentence finally, we can quickly see that

there is no reason to FOsit head-final verb phrases. 'lherefore, let us

conclude that Yiddish is S-V-O (as 0PFOsed to Gennan S-o-V). N:1N, again,

the qllestion is where !NFL is base-generated. Gennan INFL appears in t\Yl'.J

places, in second. FOsition and sentence finally. Yiddish INFL, h::1Never I

only appears in one place, in secorrl FOsition. 'As \Ye see below, 'both root

am anbedded clauses sb:Jw V2 effects.

(16) Die froy hot gekoyft des bukh in der kranthe \«JllClll has bought the 'book in the store

I '!he 'NOllaIl bought the 1:x:xJk in the store. I

(17) In der kran h:>t die frey gekoyft dos bukhIn the sto!.""2 has the \\Ol1aIl l:x:>ught the 1:x:xJk

'In the store, the \\Ol1aIl bought the 1:x:xJk. '

(18) Ikh meyn az di froy hot gekoyft dos bukh i .. kron.I think that

I I think that the 'NOtlaIl bought the bcx:>k in store.'

(19) Ikh meyn az in kran hot di fray gekoyt dos bukh.

Besides the fact that Gennan INFL can surface sentence finally, we

are led to suspect this IX>sition for !NFL since the VP is head-final. If

INFL were base-generated sentence finally, then both vmx am max tNOuld

be head-final. Ei::1Never, for Yiddish, neither the surface IX>sition of rnFL

nor the 'NOrd order of the VP suggest sentence final IX>sition for !NFL.

'!his, then, leaves us with t\ttO ch:>ices: INFL is sentence initial (I-S-VP),

or INFL is sentence second. (S-I-VP).

3. 2. 2 Yiddish is not INFL-initial

116

let us assune that Yiddish is urrlerlyingly I -S-VP. This is

attractJ.ve since it makes Yiddish similar to Gennan3• German, after the

application of INFL front.ing, is also I-S-VP. All that is needed to

accolUlt for the V2 effects is the fronting of a maximal categ'ory sU.ch as

'Ih..i.ersch· s Ru.le 2. Just as in the analysis for German, with thi.s account

for Yiddish there is no need for a rule of Subject Awe Inversion, and V2

can be reduced to the fact that CCMP cann:>t be doubly filled. Below we

see sane ,IX)ssible derivations. I have placed the inflected verb, hob, in

the ,IX)sition of !NFL to show' the J"OCl'VEment rrnre clearly.

(20) D-structure:

[s 1rb ikh [vpgekoyft des bukh in kran]]]have I -oought the 'book in store

S-structure:

b. [8' i.kh [8 1rb t [vpgekoyft dos bukh in kran]]]

c • [5' dos bukh [5 1rb i.kh [vpgekoyft t in kran]]]

d. [s' in ](ran [s 1rb i.kh [vpgekoyft dos bukh t]]]

All of these ,IX)ssible derivations yield grarmat.i.cal Yiddish sentences

am the analysis appears to be the correct one.

'!here is one way, In\tJever, that an analysis of thi.s sort mi.ght prove

to be unsatisfactory. t'bte that preverbal subjects am preverbal objects

cann:>t be distingui.shed structurally since, in roth cases, they are in

CG1P. If, for sane reason, 'tie wanted to distin:Ju.i.sh them, it \\Ould not be

3. As explained in Chapter 2, I am asstming that rmx is a maximalprojection whi.ch inclmes I, its canplement Vp, and the subject NP whichis licensed by a predication relationship with the VP.

117

IX>ssible under this analysis. Below I give t'ltO reasons 1tJhy we do want to

distinguish preverbal subjects fran preverbal nonsubjects in Yiddish. (he

argument invelves proncnms, the other invelves extraction fran embedded

SiS.

3.2.2.1 Pronouns

'!bere are certain pronouns that cannot appear preverbally. Personal

pronouns are IX>ssible only with heavy stress, and es, I it', which cannot

tak - - -bi 4e stress, 18 ~SS1 e.

(21) a. Di kimer hobn im gezen.the children have him seen

''!be children saw him. '

b. *Im hobn di kinder gezen. (witbJut stress)

(22) a. Di froy hot es geleyent.the 'ttOllall has it read

''!be \\Otlall read it.'

b. *Es hot di frey geleyent.

we might 'llapt to say that these pronouns are clitics and therefore

carmot topicalize. Ib.-lever, \tJe~ that all subject prornuns nay appear

preverbally.

( 23) Es hot gegesn dos broyt... it has eaten the bread

I It ate the bread.. I

In order to save the r-s-vp accrn,mt, \JJe \'iOuld have to say that only

subject prornuns can IlDVe to CG1P perhaps because of the phonological

4. Clitics fran the Vp appear to the left of the main verb.

118

distinctions of nan.:inat.i.ve versus accusat.ive case. This is hard to argue

especially since subject ~ and object ~ are b::moJ?honous. '!here must be

sane di.stinction, then, between pre-verbal subjects and pre-verbal

non-subjects.

3.2.2.2 Extraction

A secorrl reason Why we want to distinguish preverbal subjects fran

preverbal rx:>nsubjects concerns extraction out of errbedded clauses.

~stanm (1977) provides evidence that extraction can only occur out of

an embedded clause if the subject is preverbal. If any other maximal

category is preverbal, then extraction is blC)Cked (the examples below are

based on examples in I.cMenstanm).

(24) a. der yid velkhn zey hobn geharget in vilnethe Jew which they have killed in Vilna

I the Jew which they killed in 1li.lna I

b. *der yid velkhn in vilne zey hobn geharget

c. *der yid in vilne velkhn zey hobn geharget

d. *der yid velkhn in vilne oobn zey geharget

(25) a. Ikh veys nit vemen zi bJt gezen zuntikI kno\t1 not Whan she has seen Sunday

I I don't krrIw wb:> she saw Surrlay. •

b. *Ikh veys nit vemen zunt:tk zi "tnt gezen

c. *Ikh veys nit ztmtik venen z.i. "tnt gezen

d. *Ikh veys nit vemen zuntik hot sie 'gezen

5

s. 'lhi.s structure is granmatical only if dar~ in vilne 'the Jew inViIna , is taken as one constituent. OtheOOse, itis out.

119

The exanples above s}x)w that extraction ma.y ~ur out of clauses when

it is the subject wh.i.ch is pre-verbal (see (24a) and (25a», hc>\vever, if

any other maximal category is preverbal (see (24b-d) and (25b-d» then

extraction is blocked l'X) matter what the order of CCMP and 'Ibpic. (examples

(b) and (c» or whether SA! has applied (examples (d». Constructions

with pre-verbal subjects, then, are not the same as construct.i.ons with

other preverbal elements.

3.2.3 Yiddi.sh is S-I-VP

Let us sup!X'se, then, that Yiddish is S-I-VP. If this is the case,

SUbjects can surface preverbally in their d-structure !=Osition. All

non-subjects, l'xJ\..ever, must tlDVe to cn.1P in order to appear preverbally.

N:Jw we can make the follCMing descriptive generalization fourrl.

(26) Restrict.:lon on 'Ibpicalization:

Unstressed Prorx:>UllQ nay not topicalize.

Since ~ can never bear stress, it can never topicalize. Other

prorx:nJns can only tlDVe to CCMP if they are heavily stressed.

The extraction facts may also be captured in an S-I-VP analysis.

First, it is i.m};x:>rtant to rx:>t.ice that it is not only \tAl-elements created

by extraction that CanlX)t co-occur with topics. A better' generalization

is that topicalization camnt co-occur with a iWh CCMP tht.Jugh it may

co-occur with a -\Iih CG1P.

(27) a. Ikh veys nit tsi er oot:i.rn gezen in ParisI kn:Jw not whether he has him seen in Paris

•I don' t know 'lihether he saw him in Paris. I

120

b. *Ikh veyz ni.t tsi in Paris hJt er im gezen

(28) a. Zi rot gezcqt az zi lDt im gezen in Parisshe has said that she has him seen in Paris

I She said that she saw h:im in Paris. I

b. Zi rot gezo:rt az in Paris hot zi im gezen.

'!hese example's show that extract.i.on is not blocked by a previous

extraction but rather by a CCMP which is already filled by a ~ elanent.

'!his element may have been created by extraction as in (24b-d) and

(25b-d) I or it may be base generated as in (27). let us simply say fer

the time being that both +wh and topicalizat.i.on invelve the same fOsition

in CCMP, and therefore, these t'NO c~t co-occur. BelC1ll I will discuss

the nature of <XMP in nore detail, but at this fOint of the discussion it

is sufficient to recogni.ze that preverbal subjects are rot in CCMP.

I have argued alxNe that Yiddish is tmderlyingly S-I-VP. 'lhis leaves

us with a problem that the r-s-vp aCCDunt aVoided, and that is the problem

of~. I leave the solution of this to a later section and row turn back

to Gennan.

3. 3 Gennan revisited

Having looked at Yiddish, we have raised questions that may be

applicable to German as well. In this section 'Ne will look at sim.i.lar

quest:lons in Gennan and argue for a S-I-VP structure similar to the one \JJe

have argued for in Yiddish. we will then test this analysis against sane

121

old argunents for an S-VP-I structure.

In the tradit.i.onal account for German, as in the case of the

suggested. r-s-vp structure for Yiddish, there is IlC) way to di.stinguish

between pre-verbal subjects and pre-verbal rr>n-subjects • Again, the

question is whether we need this distinction. Of the tw:> argunents used

for Yiddi.sh, only one is relevant to German. Since there are no V2

effects in anbedded. sentences in German6 , we cannot use extraction as a

di.agrx>stic for a subject/IXln-subject asynmetry. ~ver, it is the case

that premuns differ in their distribution deperrling- on whether they are

subjE!Ct or Inn-subject pronotms. BelOll \tJe see that ~-subject may appear

preverbally, but ~-object ma.y rot. Also, all personal prorx:nlns may

appear preverbally, but -if they are non-subjects, they must be heavily

stressed..

(2.9) a. Er hat das Brat gegessenhe has the bread eaten

'He has eaten the bread. '

b. *Es hat er gegessen.it has he eaten

'He has eaten it.'

c. Ich babe ibn geseher~

I have him seen'I have seen hi.m. I

d. *Ihn habe ich gesehen (with ibn tmstressed.)

'lhi.s suggests that Gennan also has the restriction given in (26) and

6. '!here' are embedded. sentences which do I'Dt contain a CCMP and 'lkLi.ch dosl'xM V2 effects. Because extraction fran these structures triggers SA.I,however, the facts that \tie find in Yi.ddish are obscured (see '!hiersch1978) •

122

that we do want to distingui.sh sUbjects fran rnn-subjects.

'nlese facts also argue for the existence of NP llDVement. 'lhiersch,

in his thesis, argues against NP rrovanent in German. He claims that the!

d-structhre object of a pa.ssive verb remains in object p:>sit.i.on and is

assign~ nan.i.nat.i.ve case by a special case-assigrment mechanism (see

Thierscry 1978 flor detai.ls).

(30) Heute wurde der Vater gesehenToday was the father seen

'The father was seen tOOay.'

ti. [S I Heute wurde [s e [vp der Vater gesehen ]]

'!his \\Ould mean that a preverbal D:Itl.ina.tive NP is fronted fran theI

object, p:>sition in a pa.ssive construction but fran Subject p:>sition in anI

active construction.

(31) I a. Der Vater wurde gesehenthe father \tlaS seen

I The fatller was seen.'I a I. [8' Der Vateri wurde [s e [vp t i gesehen]]

: b. D3r Vater hat das Buch gesehenthe father has the bcx>k seen

'The father has seen the bcx>k. I

b'. [s' Der Vateri hat [5 t i [vp das Buch gesehen ]]

In tenns of the present analysis \fJhere German is tmderlyingly S-I-VP,

'Nei \\Ould expect naninat.i.ve NPs in passive constructions to 1Jehave likei

!accusative NPs in active constructions. 'nlis \IJOuld mean that like the

aex::usative NPs of active verbs, tNe \tJCuld expect the rnn.i.native NPs of

:Passive verbs not to topicalize if they are unstressed proron.i.nals (see

(26». ~ever, we can see belCf.ll that this is not the case. Nan.i.native

123

unstressed proIX>lJIlS may appear preverbally whether they are in an active

or a pa.ssive construction.

(32) Es hat das Brat gegessenit has the bread eaten

•It has eaten the bread. '

(33) Es wurde gegessenit was eaten

•It was eaten. I (it = referential)

(34) *Es hat sie gegessen.it has she eaten

I She has eaten it. I

If, 'hC'Never, we assume that Gennan has NP l'lDVement and that

ncminative NPs in pa.ssive constructions are assigned case in the subject

J.X)sitien, the above configuration of judganents is predicted. In (32) and

(33), ~ is in its d-structure J.X)sition, while in (34), es has Il'OVed fran

the VP to CG1P violating (26).

If tNe J.X)sit a d-structure order for Gennan of s-r-vp, we can, again,

say that prooouns that CanI'X)t })ear stress canIDt topicalize. Like

Yiddish, to aCCOlmt for SA! we must say that topicalizati.on of a

rnn-subject constituent triggers Il'OVement of !NFL to CG1P.

'!his d-structure "-Ord order will also account for the lack of

ani:>iguity in the follOlling example.

(35) a. I "Die Tochter hat die M.1tter geki.lBt.the daughter has the nother kissed

b. = ''!he daughter has kissed the rrother.'c. + I '!he nother has kissed the daughter. I

Since the feminine article, die, is the same \\hether it is nan.'inative

or accusative, (35) s'hould be ambigtDus between the reading \\here the

124

daughter ~has kissed the IlDther ani the rrother has ki.ssed the daughter.

!bNever, only the fanner reading is avai.lable. Given an S-I-VP

d-structure for Gennan, we can assune that in any pJssibly ambiguous

structure, . the d-structure \-tUrd order is assumal. If a readll1g is desired

where the direct object has been topicalized and the m::>ther is kissing the

daughter, th'3 first NP may be stressed to irrlicate its topic p:>sition7•

NJw, what of den Besten' s argunent for supposing that !NFL is in

CCM?? Basically he states that since subject elitics attach to !NFL in

root SiS and to <:n-iP in embedded SiS, by saying that mFL is in <:n-iP in

rc:ot S's, the generalization is that subject clitics attach to CG1P.

'!here is only one root construction in our analysis that does not have the

!NFL in CG1P. 'Ibis is the construction where subject is sentence

init..i.al. Q.1r analysis of this construct.ion is given in (a) below, and the

traditional analysis in (b).

(36) a. [s NP I vp]

b. [?NP. I. [5 t. VP t.]8:I. J ]. J

Notice that in exactly these constructions, the subject is not in the

correct p:>sition to clit.i.cize in either analysis. The data, then, are

consistent with both analyses and thereby argue for neither one nor the

7. It has been p:>inted out to me by Wigi Rizz..i. that in a 'Ibiersch-typeanalysis, the unarribigtDus sentence in (35) still reflE!Cts d-structure \I.Ordorder even though the subject will have nrJVed fran its d-structurepJsition.

8. I leave the? category tmlabeled because of differences in thetheories. As will be discussed, KcxJpnan believes that it is 8" ani that!NFL alone occupies the <:n-iP position.

125

other.

Another argunent given by I<Dopnan (l983b) for this sort of analysis

might prove problanat"ic for nty analysis. K£:x)pncut argues, following

I<bster, that (37) below is derived fran (38) by an optional D-\tJord

. deletion in CCMP.

(37) Jan heeft hem ontm:>etJam has him met

'John ha.s met him. '

(38 ) Jan die heeft hem OIltnDet

John D has him met'John has met him. I

Kcx:>pnan uses examples such as these to argue that subjects in Illtch

main clauses such as in (37) occur in a 'lbpic p:>sitiOD outside of s' .

If it can be shown that subjects must be in this sort of 'lbpic

p:>sition, my analysis canrx>t be correct. H:Jwever, the data below sh:1N

that not all elements that appear sentence ini.tially can appear in 'lbpic

p:>sition.

I first question the nature of the rule \'Jhich optionally deletes

D--...ords in CCMP. 'lben I p:>int out that the inventory of elements that ma.y

appear preverba1l~l are not the same as the elements that nay appear in

D-\tJord constructions.

'!he argunent that D-\tJords may delete optionally in CXMP canes fran

examples of the followin:J type.

(40 ) (Oat) gelCX)f ik n.i.et'Iha.t believe I not

•I don I t believe that.'

(41) Ik gelex>f *(dat) metI .believe that rx>t

•I don I t believe that.'

(from KOopman, p. 221)

126

se"eral ccmnents should be ma..ie about these constructions. First,

the dat that is deleting is the topicalized object itself. Presunably I

then, as the topicalized object, it too should be outside of the S' just

as ~arie is in the example below.

(42) Marie (di.e) ken ik r.:i.etMary that-one 'kn::1fl I tnt

'Mary I don't lanw'. I

(frant KOopman, p. 221)

5ecoooly, examples such as (40) are marked witb:>ut the ~rd and

nust be set into a specific discourse cxmtext. whereas the NP-D-\\Ord-S'

sentences are less marked witb:>ut the ~rd. '!his makes it doubtful that

the same .tilenanenon which relates the t\\O options of (40) (with and

witoout the D...liItOrd) also relates the t\\O cooices of (42).

Beyond problems with the effects of the the ~rd deletion rule,

there is clear evidence that the oonstructions witoout the D-\\Ord allow

for different sentence initial NPs than do the construct.ions witrl the

D-\\Ords. N:>tice that if I argue for two di.fferent coristructions for

pre-verbal topics, one witil I>-\-.ords, one ~tb:>ut, I am arguing for three

different preverbal NP p:>sitions. 'lhe c.'Onstructions are given belOtti.

127

(43) [8 NP I vp]

Jan heeft het gegetenJohn has it eaten

IJohn has eaten it.'

(44) [51 NP. I. [s NP e. [vp ••• t .... ]]]1 J J 1

Oat broodje heeft Jan gegetenthat sandwich has John eaten

•John has eaten that sarrlwich. I

(45) [S.. NPi [S. o-c,..Ordi I j [S NP e j [vp

03.t brocxlje dat heeft Jan gegetenthat saniwich D has John eaten

''!hat san::1w:ich, John has eatal.·

t .•.• ]]]]1

I will s~ that all three p:>sitions are justified. In structures

such as (43), all NPs may appear (i.eft, tmstressed proI'X)UIlS as well as

other NPs). In structures such as (44), stress·ed, but not lUlstressed.

proOOln1S ma.y appear, as can reflexives, am finally in constructions such

as (45) I no prorx)\ms may appear at all.

(46 ) Preverbal Subject (= (43 ) )a. Jan heeft dat broodje gegeten

John has that san::1w:ich eaten~ John has eaten that sandwich. I

b. Bet heeft dat broodje gegetenIt has that sandwich eaten

I It has eaten that san::1w:ich. I

(47) Preverbal N:>n-subjects (= (45))a. Oat broodje hE!b ik geg-eten

that sandwich have' I eatenI I have eaten that san::1w:ich. I

b. Mr:!zelf heb ik nooit verloochendmyself have I never canpranised

I r- have never canpran:i.sed myself. I

128

c. ~t heb ik gegetenit 'have I eaten

I I have eaten it.'

(48) Pre-D-\ttUrd (= (47) )a. I:a.t brocXlje dat heb ik gegeten

that saniwich D have I eaten''!hat saniwich, I have eaten.'

b. *Mazelf die ik heb l'XX)it verlcxx:hend

c. *Bet dat heeft dat broodje gegetenIt D has that sandwich eaten

'It, it 'has eaten that saniwich. I

d. *Het dat heb ik gegetenIt 0 'have I eaten

'It, I have eaten.'

My claim is that there is IX) optional rule of I>-\tJord deletion in a:MP

except for the di.scourse oriented rule exatplified above. In structures

such as (45 ) I hc:J\t.ever, the D-\-Jord is required for coihdexation with the

extra-sentential NP. '!his imexation makes this construction lex>k very

much like a relat.i.ve clause, and, interestingly, D-\-Jords are not

optionally deleted in relative clauses.

(49) de jongen *(die) het gegeten heeftthe bJy D it eaten 'has

'the }:x)y \fJho ate it I

'!he main di.fference bet1Neen the main clause D-\ttUrd construction and

the relative clause construction is that !NFL does rnt ITOVe to CCMP in the

latter. It is this ITOVEment, presumably, which ident.i.fies the structure

as a main clause.

My conclusion is that neither the argllnet1t that INFL must be in a:J.1P,

ror the argunent t'hat sentence-ini.tial NPs must be outside of S' provide

strong et!idence against the S-I-VP structure which I 'have proposed for

129

Gennan. First, altmugh I agree that INFL may appear in CG1P, it is rnt

required. by the evidence to be in a::MP when subjects are sentence

ini.tial. Secondly, I do rx:>t believe that all sentence initial NPs are

outside of S·, but rather I only if a D--Nord appears is there an argtment

for this structure. '!here is, h'Jr..lever, D:)thing in my analysis that says

that subjects may not topicalize or coindex with a presentent.i.al NP. I

claim only that they are not required to do so.

3 •4 '!heoretical M:>t.i.vations

I have argued that l:xJth Yiddish and Gennan are S--I-VP langua.ges

differing only in the headedness of their VPs. Yiddish VP is

head-initial, Gennan VP is head-final. 'lhis entails that both languages

have StU in the event of topicalization (or a Yes/rib question). In this

sect.i.on I address this issue of INFL ltDVaTIent directly and exam.ine the

structure of CCMP.

I will claim that heads (¥fs) can only lOClVe into the cq,tegory that

governs than. 'Ihis appears to be true of ooun incoq::oration (see Baker

1983), of va IIDVanent, and IO novanent. '!his sort of novenent will

explain 'IAly 1° may nove into a:MP. '!he reason I O~ IlC\Te into a::MP, I

claim, is that rcx:rt:. SiS may have I'D CCMP. If there is CCMP, b:Jwever,

then, by X'_theory, this c:x:MP nust have a legit.i.ma.te head since all

maximal projections must have heads. If the head of a category is anpty,

then it falls un']er the ECP and must be properly governed (or identified

130

in tenns of C1apter 4).

In the case of c.l:MP, we will argue that the head may be licensed in a

nll1lber of \alays. (1) It may be properly gO\/erned (ident.i.fied) by the verb

whi.ch subcategorizes for its maxi.ma.l projection. (2) It may be lexical as

in the case of that, or because in English, or (3) it may be filled by .

rrovenent of IO. It is this type of rrovans1t which is discussed in the next

section.

3.4.1 r-bvenent of Heads

In my analysis of Gennan, I am forced to allOli !NFL to appear in

three p:>sit:ions: (1) in CCMP (SO), (ii) between the subject NP an] the VP

(51), and (iii) sentence finally (52).

(SO) [S' X"i INFLj [S NP t j [vp t i ••• ]]]

(51) [8 NP!NFL vp]

(52) [s I daB ~ [s NP t i [vp NP V+INFLi ]]]

'lhis, at f;irst , appears very messy. '!he traditional analysis needs

only t'NO p:>sitions for !NFL: sentence final which is base generated, and

V2 which occurs through !NFL fronting to c.l:MP. My analysis needs not only

a fronti03 rule but a lOOVement to the right. This means tw:> rules instead

of one, and rrovement to th~ right which is an uncamon type of rule9 • In

order to IIDtivate these t1NO lOOVanent..c;, I first discuss lTOVE!l1ent of heads,

am secorrlly discuss the relationship of. INFL and va.

9. Wh- rrovenent is universally to the left. cnly Focus NP shi.ft and PPand S' extrap:>sition rrove to the right.

131

In discussions of t-bve-a, IlDst often the object being IlDVed is a

maximal projection as in NP-novement and wh-rrovement. !NFL IlDVatlent,

b:Mever, involves the IlDVement of an >f. Baker (1983) investigates a

similar sort. of ltOVanent in a discussion of I'X)UI1 incoqoration. He claims

that in sane cases of ncnm incoqoration, the head of the NP which is

governed by the verb undergoes syntactic 11O\1anent leaving a trace.

(53) ne Oterontonni.· a' t on- '~ji-a' ens wa' -tha ' -tcan-akwe 'sapling pre-earth-suf PRr aor-3-1-harrlful-pick

I sapling \\Ould custanarily take up a handful of dirt ••• '(M:>haWk; fran Hewitt 1903)

Tean, 'harrlful', is the head of the logical object, 'handful of

dirt', and it IlDVes into the verb akwe' leaving 'behind the rest of the NF,

'of dirt' (fran Baker 1983, p. 14).

BelaA' 'Ne can canpare the structure of noun-incoqoration and INFL

fronting.

(54) a. ~un Incoqoration b. INFL-fronting

CCMP'

cmP+I~K.t! ...............V"

J.

I will fX)sit the following restriction on the lTOVEment of heads.

(55 ) Head. t-bvanent Constraint:

An Y? may only IlDVe into the the Y? tNhich properlygoverns it.

Another example of this type of IlDVanent is VO 11O\1anent in Gennanic

laBJUages •

132

(56) I'~

I VPI ~

hat. +present VP t.1. ~ 1.

A YA N gelesen

vi~le aJeher

'!he question of mi.ch verb in a series of Gennan verbs is inflected

is easily solvED if \tee assune that the verb has to l1DVe to INFL. If we

assune the structure of (55) I the only verb that INFL governs is the I top I

verb. 'Iherefore it is only the 'top' verb that can nrNe to INFL.

We can nr:JW see that \tJe account for the appearance of the inflected

verb in sentence SecDm PJsitiOD by t1NO rrovanents of heads into their

goverrDrs. First vO rroves into !NFL0

I then !NFL0

I which nr:JW contains

1°+VO, rroves into CCMPo.

(57) a. II

NP~~

VP V

AVP V

~v

b. II

~NP I VP

I 1\V.+I AVP t.

1. / \ 1.

NP1\ V

133

c. ea-tP 1

aaMP II~ /~~

X" [V.+I]. NP t. VP1. J :J ~

vP 't.~ 1

VP V

~V

I am asSun.ing that any m::wanent of !NFL or the V will leave an empty

category that must be properly governed. In these cases where ¥? moves

into its governing category, I will assune that the trace" is antecedent

governed.

Baker has {X)inted out to me (p.e.) I there are other incorp:lration

pherxmena that are parallel to ~un Irxx>rp:lration. '!he ones he mentions

are given below.

(58) VP~

V NP

AN •••

(59)

/"'"V VP

~V •••

(60)

(58) is exanplified by the case of ~un Incnrporation that \'Ie have

seen above. (59) is ex~ul-'lified by Japanese causatives and (60) by

applied verbs in Baha.sa IIrlonesian. Examples, taken fran Marantz (1981 )

are given belOll.

(61 ) Japanese causativa: (p. 310)Tar<x> ga Hanako ni okasi 0 tabe-sase-taTaro Di Hanako MT cake N:C eat-cAt.5E-PAST

'Taro 'let/made Hanako eat the cake.'

134

(62) Bahasa Imonesia: (p. 274)a. Saja man-ba\\la surat itu kepada Ali.

I ~-bring letter the to AliI I brought the letter to Ali. I

b. Saja man-bawa-kan Ali surat itu.I TRMS-bring-APPL Ali letter the

I I broU3ht Ali the letter.'

Baker makes the i.mp:>rtant observation (see alse> Marantz 1981) that

this sort of IOCJVement is constrained by the norpholCX]ical rules of a given

language and the interaction of these rules with the syntactic canp:>nent.

tbtm Incoqx:>ration can occur only in languages that have rxJt only the

proper canp:nmding rules but alED the proper interaction bet1Neen these

canp:>uniing rules am the syntax. EB31ish has V IOCJVEment into !NFL but

does not 'P..ave N l1D\1ement into V, V ItDVement into V, nor P ItDVanent into v.

I assune that this type of IOCJVement is a case of EhDrrls (1976) IDeal

Transfonnation. His definition is given belCJ#l (p.4).

local Transfamation: A transfonnation or atransfonnational operation that affects only an input sequenceof a single lX)nphrase rx:xie C and of one adjacent t:Onst.i.tuentC· that is specified without a variable, such that theoperation is J:Dt subject to any cond.ition exterior to c amC', is called a "1~al transfonnat.i.onu (or a localtransfonnational operation).

fie has prop:>!;)ed a recent revision (Em:>nds I to appear) of this

definition which rt:M includes the head 'of CI. The p:>ssible structural

description, then, \«luld be:

(63) a Y'Iy

· 135

If a is a head category, then this c:onfiguration translates into the

one \tJe have been di.scussing above, the relation bet'Neen a lexical category

(y) am the head (a) which governs its maximal projection (Y').

3.4.2 !NFL llDVanent

3. 4. 2. 1 !NFL llDVanent: Left

We might say that the V ItIJSt llDVe to INFL because the material in

!NFL must appear on a ver1:al. element. 'Ibis does nJt explain, l"ow'ever, why

INFL llDVes to a:MP. I will assume t\\O things. First, following

X' -theory, CG1P must have a head am INFL, by being able to llDVe to CG1Po,

10can appear as the head of a:MP • Secorxl I asslltle that operators appear

in CCMP, but nJt in the head IX'sition. 'lhe cc>nsequence of these t\\O

assumptions is that a:MP nust be headed in order to be able to contain an

operator. If the head of CCMP is anpty, it must be properly governed. In

a root clause, where CXMP is IDt properly governed, the head must be

filledl1 .

Different operators may appear in a:MP as shown below.

10. see Jaspers (1984) for a different analysis of INFL llDVanent intoCCMP. He assunes that !NFL llDVenent is needed in order to extern theextraction danain.

11. Speas (1984) accounts for this by stating that operators must begoverned. Within the analysis presented here, being governed is the sameas stating that every category must have a head. '!he difference in t..het\ttO analyses is that, for Speas, embedded operators are governed by thematrix verb. Ebr my analysis, the head of the embedded a:MP is identifiedby the matrix verb, a.rrl this head, in turn, governs the anbedded operator.

136

1) CG1P = XP anCl !NFL: If CG1P consists of a maximal projection and

!NFL, the reading is one where the }(J? is focused and binds a variable

within the I-. lilt is for x: x = ..., ... x ......

(64) [Den Vater hat [ das Kind t gekUBt ]]. the father has the child t kissed

-The child has kissed the father.-it is for x: x the father, the child kissed x

2) cx:MP = iwh and mFL: If CXMP consists' of a 'Ith..-wurd and mFL, the

readin3 is one where the wh-'ttOrd is a question operator birrlirg a variable

within the I-. uEbr \tthi.ch x: x = ••• , •••• x •••

(65) [Wen hat [ das K.ini t ge~ ]]\tk1o has the chi.ld t kissed

-Who has the child kissed? I

for which x: X a person, the child kissed x

3) CG1P = e and INFL(indicative): If the first position of CCMP is

enpty, we can still assune the existence of sane enpty operator depending

on the rocxrl of !NFL as the head of CXMP. If !NFL is indicat.i.ve, then the

anpty operator is a Yes/No quest-ton operator.

(66 ) [e hat [ das Kind den Vater geklffit]]has the child the father kissed

I Has the child kissed the father? I

4) CXMP = e and mFL(subjtmct.ive): If the empty position in CCMP

coex.i.sts with a verb in the Subjunctive, the operator is a corrlitional.

(67) [ e ~tte [ das Kind den Vatet:" geldffit]]had the child the father kissed

I Had the child kissed the father, ••• -

The distilx!tion 'bet'Neel1 the irrlicative and the subj~tive INFL in

CXMP is clear. Taking Fnglish examples, \tie can see that there is another

option for the o::>nditiona! clause which is to spell out the head of CCMP

137

as if.

(68) If I were a bit younger, I 'N\:)uld •••

When the head is explicitly conii.tional, the verb nay be either in

the subjunctive or the indicative.

(69 ) If I was a bit younger, I \\Otlld

That option, hc:1Never, is not available for the inverted version.

(70) *Was I a hit YOUB3er, I \\QuId •••

5) a:MP =0: If there is IX) CCMP, there is no o,Perator and \..e assume

the default illocutionary force of a declarative.

(71) ras Kind hat den Vater gekUBt..the child has the father kissed

I The child has kissed the father. I

There are t\ttO 'Nays that CCMP may be lexically fillai. One \tJe have

seen above, where !NFL neves into CCMP. The secorrl way that CCMP may be

filled is by the lexical insertion at d-structure of su1x>rdinating

<::X)njt.mcti.ons such as \filen, ~, because, etc. This seems IXJt to be a

p::>ssible option for a root S. let us draw another conclusion fran this

which is that +Tense rmJst be the head of E (Expression, as in Banfield

1973). In a nonnal declarative with no CCMP, -tTense will be in !NFL and

therefore be the head of I I ani therefore of E. If E \\'as an S' (CCMP') I

then the head of a:MP' nust have a +Tense elanent. This \\1Ould insure that

the head of CCMP \teOuld be filled by !NFL ani rx>t by a s,Pell-out of sane

sort.

This 'NOuld als:> explain various already noted facts. First, matrix

138

verbs cannot be infinitivals since INFL, as the head of I', will rx:>t

contain [-tJrense].

for s:xne rearon, exclamations do l'X)t need [+tense] in their heads:

(72) ~ crazy she is!~ crazy is she1What· an erK)I1I[)US cat she has 1

*What an C4"X)I'1OO\.1S cat has she 1

'!hts \\Ould explain \fAly strirgs such as OX>i fairel ('What to do' )

can only be used as exclamations and nev~ as questions.

(73) *PRO to go to school.

5ecordlyI the matrix CCMP cannot contain a -wh canplanent.tzer. E

will rr:M be CCMP I I but -\tAl CCMP does lDt trigger irwersion and the head of

CCMP' will r¥Jt conta.i.n +Tense.

(74) '!hat June will go to Ergland.

Also, this will explain the fact noted above concerning the

di.fference between tk\ord ma.trix constructions and relative clauses :in

Il1tch.

(75 ) D-fNOrd cc>nstruction:Jilt broodje edat heb [ik gegeten]]that sandwich [that have [I eaten]]

''!hat sarrlwich, I have eaten.

(76) Relat.i.ve Clause:dat broodje [dat [ik gegeten heb]]that sardwich [that [I eaten have]]

'the sandwich I have eaten'

In (75), the head of E is the matrix CCMP whi.ch contains dat and

hebe '!hi.s, then, qualifies as an E since the head contains a fTense

elanent. (76), ~ver, cannot be an E since the head of CXMP' C'Ontains

139

only dat ani I'X) elanent that is marked for -tJrense. (76), then, can serve

only as an argunent.

3. 4. 2. 2 !NFL nrNErnent: Ri.ght

tet:>vanent of mFL to the right is rrore problematic. I will proFOse,

usirg ideas fran Fabb (1984) am. Roberts (fortbeaning), that it is rx:>t so

much that !NFL rroves, but that !NFL, under certain corrlit.:tons, it allowed

to remain e:npt.y, ani that inflectional IlDrpholcgy may be generated

directly on the verb.

'll1e idea that I will use fran Fabb and Roberts is that inflectional

affixes are rarrlanly generated but must be checked by sane rreans. For

instance, in Gennan, a notm will be generated with accusative case, but if

it is IDt governed by a verb which can assign accusat.i.ve case, the

structure is ruled out.

In the same way, a verb fonn may be generated with a [+past] affix,

but if it is rxJt governed by an mFL bearirg the [+past] feature, the

structure will be ruled out. At d-structure, then, we may have:

(n) I'

NP!NFL VPI I ~

Dieter [+past] NP V. I I

das Buch kaufte[+past]

'lbe effect is that !NFL may be generated with only features. As we

have said before, heads must be ident.i.fied. In the structure given above,

!NFL may be license:1 either by nrNanent of the V, or by beirg governed by

140

'f'ed /"V"lUn12a Spec1 1 ~Y·~ •

!NFL ma.y also be generated with an inflectional affix. '!his will be

sufficient to identify the head. fbwever, since the affix must appear on

a verbal elanent, either the V Il'I.lSt rrove into !NFL, or, as in the case of

dO-SUPPJrt in En:llish, the affix appears on a pleonastic type of !NFL.

5cmeth.i.ng di.fferent must be sai.d alx>ut certain constructi.ons in

English. We krDw that main verbs' do rx>t raise into the !NFL nOOe (78) I

though aux.i.liaries (79) and nrxials (80) are generated there (see· Li.ghtfCX)t

1979, Roberts 1983 for a historical view of this). 'Ihis is easy to see in

cases \lihere !NFL has lOCJVed into a:MP.

(78) *What oought Peter?(79) What is Peter buyirlg?(SO) What could Peter buy?

The generalizati.on i.s that inflectional affixes may appear unaffixed

\lihen they appear adjacSlt to the verb. 'Ihis is not unreasonable. let us

say that the attachnent of inflection to a verb happens at PF and must be

a local rule. 'nlat is, inflection can only attach to an adjacent verb.

'Ihis tlRy be part C?f a tTOre general fhenanenon \tJhich is that, if a node

must be realized nnrpllologically on another const.:ltuent, the node ani the

constituent must be adjacent at PF, and therefore, at s-structure.

we can see below that- as scon as string adjac~ is di.sturbed, INFr~

must be spelled out with a verbal pleOraCJtic.

12. See Webelhuth 1983 for a similar analysis \lihich PJF;:lts an I-NP-VP \fJOrdorder for Ger:mcm. He also uses proper gaverrment to predict the PJsitionof the inflected verb.

141

(81 ) a. Peter oo\.J3ht a bcnk.b. [8 Peter [+past] [vp 1:ought[+past] a book ]]

(82 ) a. Peter did not buy a lxx:>k.b. [s Peter [do+pastJ rnt [vp buy a bcnk J]

(83) a. Did Peter buy a bcnk?b. [6' [do+past]i [5 Peter t i [vp bUy a lxx:>k ]]

Traces do n:Jt seen to interfere with this adjacency. When !NFL and

the verb are se,parated by traces, !NFL ma.y still ranain anpty. Evidence

for this canes fran matrix 'Yh......~estions. It has often been noted that

INFL does rx>t nove to C'~ when the wh--...ord is subject.

(84) What did ane see?*What saw she?~tat she saw?

(85) *Who did see it?Who saw it?

Various explanat.i.ons have been given for this (KcxJpnan, Jaspers,

etc) I and to these I add arx>ther which assunes, bJwever, that INFL does

nove to <XMP. Below \-Je see the structures for (84- ) and (85 ) •

(86)

(87)

[5' wha.t. INFL. [8 she t. see t.]]1 ] J 1

[s' wb). !NFL. [8 t. t. see it]]1 J 1 J

In (86), !NFL is not adjacent to the verb at PF since she

intervenes. 'Iherefore!NFL must be spelled out with do-supp:>rt. In (87),

bJwever, !NFL is se,parated fran the verb only by traces which will rnt be

visible at PF13• !NFL may, in this case, appear directly on the verb

aCCX)unting for the lack of ~o-support.

13. I am asslDlli.ng Ibuchard IS aCCDunt fo~· wanna contraction.

142

'!here are three 'Nays that inflectional affixes ma.y attach to the

verb. If the affix has been generated in INFL, either ~i) the verb must

be adjacent to INFL at PF am therefore at s-structure, or (ii) the verb

must nove into INFL. When the inflectional affix is generated directly on

the verb, adjacency is ac~eved at d-structure. In this case, since the

head of !NFL I bears no phonet.i.c features, it must either be properly

governed, or the verb must llOVe into it. In Chapter 4, ~ will assume a

rx:>tion of proper goverrment in \fJh.i.ch c:n-1P properly gOV'erns I I and INFL

properly governs VP.

Below \tJe can see b::M the IXJssible IXJsitiona for INFL in Gennan will

follow fran t.h.e fact that empty heads must be ident.ified. In (88), four

p:Jssible a:MPs are given, and in (89), four IXJssible !NFIs.

(88) CCMPsa. l'X) CXMP (rcxJt sentence)

Fil.led:b. COMP I

COMP~It/'. ~

wh !NFL. NP t. VPXI 1. 1

c. CXMP

~CGW I'if ~

\fArile NP 'r Wbecause

w/head of canplement

lexical spell-out

d.

(89) !NFL:

a.

Vi

V COMP'

CO'JP~ IIe

II

NP~vpI ~

V+af. t.~ 1

Empty:

lexically governed

Filled:

w/head of canplanent

143

b. I'

NP~vpr /'\

do+af ••• V

w/pleonastic

e.

d.

COMP '

CCMP I'

~NP I VP

e ~••• V+af

CCMP '

CCMP I'!NFL. ~

1 NP I VPe.

. 1

Enpty:

lexically governed

antecedent governed

. In Gennan, when the inflected verb appears in front of the Subject,

it is in CCMP as sbJwn in (88b) am (89d). When it appears between the

subject and the VP, it is in its base generated position as in (89a) and

when it appears sentence finally, it is as in CDnstruction (8ge).

144

'!he a1:xJve explains when !NFL ma.y be anpty in certain configurations.

tiJ\tI \aA:~ waJ,1.t an eJ\.-planat:ton of why !NFL must be anpty, i.e., why the V

canrx>t 1'llC)'t/e into !NFL in an ert"Cedded clause. Th.i.s same question arises in

the 1'llC)'t/anent of !NFL into CCMP. We know why !NFL need l'Dt 1'llC)'t/e into CCMP

in ert>edded sentences, but we need an explanation as to why INFL may rot

troVe into CCMP. '!hough alternat.i.ves are avai.lable, such as saying that

features in properly geverned heads CDunt as having filled the head

prevent..ing further troVenent, I prefer usL~ the last Pesort Principle

(Olansky, fall lectures 1983) curl say that verbs will move int.o INFL curl

INFL will rrove into CCMP only \fJhen required.

An effect nore subtle than that of Gennan is fourXi in Sccurlinavian

langua.ges. 9iiedish is like Yiddish in that it is S\U and has V2 effects

in rcx>t curl aribedded clauses. Because of this I will assune that it is

S-I-VP. '!here is an interesting alternation, ho\t.ever, in enbedd-:d

clauses, given in Platzack (1983). Sentence adverbials (here I lise

Pla1:zack's examples with the negative particle) appear to the right of the

inflected verb in rex>t clauses curl to the left of the infl~ted verb in

embedded clauses.

(90) a. Jonas gillade inte oonanJonas liked oot him'Jonas didn't like him.'

NP V+I Adv NP

b. att Jonas .inte gillade lDnan NP Adv V+I NPthat Jonas rx>t liked him

'that Jonas didn't like him'

:As in the Gennan accounts, Pla1:zack explains this variation through

!NFL troVement a.~ topicalization. I claim, l'¥:7.t1ever, that the difference

145

in order arises fr~ the fact that the arbedded INFL is properly governed

and therefore ma.y remain anpty.

(91) RCXJI': II

NP I Adv VP

~~NP

NP V+af Adv [ t NP ]

EMBEDDED: CCMP'~,

CCMP ~NP I Adv VP

e A.V+af NP

3.4.3 Sunmary

1. (jennan and Yiddish are 5-I-VP

NP e Adv [V+af NP ]

2. Ii2ads can rrove only if they llO\Te into the category by \JkLich they are

governed.

3. Inflectional affixes are randanly generated and restricted by

checki.ng mechanisrm;.

4. Heads may remain anpty only if they are properly governed.

5. Affixes ma.y appear unattached at a-structure if they are adjacent to

the form to which they will be affixed.

An. advantage of viewiDJ Gennan as beiDJ S-I-VP rather than S-VP-I, is

that it is no lODJer pUZZling Wly Gennan does not behave more like other

verb-final languages, such as Japanese or Turkish. rrhis is purely at the

146

level of speculation, but will be discussed rrore in Chapters 4 and 5.

3. 5 Historical speculat..i.ons

we have just seen ample proof of a separate INFL ncrle • Without this

ncde, \\Ord order facts in Gennanic languages \ttOuld be ve.~ difficult to

capture. In this section I offer speculation as to OOW different Gennanic

larguages came to have different 'NOrd orders. In the discussion I will

touch on different subjects such as the contradictory danands made on

INFL, the pl..1I:J.X)se of the rcot canplenentizer, ani b:Jw thes~ effect

langua.ge change. I will also speculate Wny subjects ani INFL seem draWll

to sentence initial pJsition.

3. 5. 1 Assumptions

In order to account for variations and changes in the order of S

constituents, I start by making certain asslltlptions. 'll1ese may be

CDntroversial but I conclude that they take us a way t.o\erds accotmting

for certain repJrted pherx:mena.

First, I assune that there is a scale of markedness for certain

configurations that. deperrl on adjac~ for case assignnent, direction of

goverrment, etc. In particular, I am asslIn.:i.ng that there are three

unnarked constructions.

147

3.5.1.1 INFL and Vadjacenqy:

An unmarked configurat.i.on \\Ould be one with !NFL adjacent to the VP

am on the side of the VP that the V is, on (e.g. English: S-I-V-o). This

configuration \\Ould be unmarked since it means that I I and Vi are either

both head-final or head-ini.tial am that the subject does n:>t intervene

bet\aJeeIl I O and its canplement VP. Als:>, as \tJe have seen in Fnglish, since

!NFL is often IlOrpholcgically deperrlent on the V, if these t\\{) are

generated in adjacent lXJsit.i.ons, affixation may occur without resorting to

m::wanent.

3.5.1.2 case Mjacency:

'Ibis is not a new notion and it is given a detai.led account in

St:c1Ilell (1981). I am assuning that the unnarked cordi.tion on case

assigrment is adjacency. Given that !NFL assigns case to the subject, we

\tt1Ould expect these t\\O constituents to be adjacent, even though, as I have

claimed, INFL assigns case by way of coindexat.i.on. StO'ttell presents this

in detail for the instances of a verb am its object (Adjacency Corrli.tion

on case Assignnent, pg. 110), and Pesetsky (1982) extends this in an

interesting 'llay for !NFL am subject (pg. ~52-254) (see Chapter 2 of this

thesis) ..

3.5.1.3 Unity of case assignment direction:

Greenberg rntices that V-o languages terrl to be PREp:>sitional while

148

o-V laD3lJages teOO to be IUSTp:>sitional. 'Ihis terrlency can be seen as a

tendency to assign case in the same direction. We can now extend that to

INFL so the V-o/P-NP languages, will also be INFL-5, arrl O-V/NP-P

langua.ges will terrl to be S-INFL. I<£xJpnan prolDses that, this tendency is

a requ:lranent at least at a-structure. Lan:Juages, such as Intch, can be

sho\t1n to confonn to the requi.ranent when given a different analysis. I do

not take such a strong stand. In my analysis of Yiddish (arrl Eh:Jlish) I I

ass\.ltle that in the S-I-V-o structure, the verb assigns case to the right,

am the INFL rx:rle assigns case to the left.

3.5.1.4 Inconsistency

The i.mp:>rtant thing to l'Xltice aOOut these three tendencies is that

they cann::>t all occur at once. If INFL and V are adjacent to one arDther,

they cann::>t 00th be adjacent to the NPs that they casernark, and assign

case in the same direction (S-r-V-o), if they are adjacent to one an:>ther

and assign case in the same direct.ion, they cannot be adjacent to the NPs

\tJhich they casanark (S-o-V-I), am if INFL and V assign case in the same

direction to adjacel'lt NPs, they cannot be adjacent to one another

(I-S-V-o, S-I-o-V). I will conteOO that it is just this tension that

causes langua.ges to change. Since there is no endlDint, there is no

attainable ideal.

3.5.2 Ge~i.c languages

With these assumptions in mind, I now look at \\Ord order changes that

have occurred in Gennanic larguages. Many of the ideas expressed here are

149

speculat.'ive but provide a p:>int of departure for further study•.

stage I

'!he preferred \\Ord order for Iatin was V+L~L final preceded by 5-0

(i.e. S-o-V+INFL); bJwever, a camon literary style placed the verb

(inflected verb) first. Lehman (1974) writes:

'!he three sentence-ini.tial verbs in these [examplesgiven] lines of exh'ilarat.'ion illustrate the marked \\Ord orderwhich \tJe fim in the Ve3a, in Haner, ani in much else of theearliest surviving material. (p.21)

As explanation for this marked order, he later writea:

Elenents in sentences can be anphasized by marking; thechief device for such anphasis is initial p:Jsition.Arrangenent in this p:>sitiOD is brought about by stylisticrules, applied after structures have been generated by P-rules[fhrase structure rules] am transfonnational. rules. (p. 219)

.Al:Jain the examples he gives are verb initial. '!here SeatU3, then, to

be a stylistic rule of verb prep:>sing. '!his is not LU1C011l1Dn. Haiman

(1974) mentions such a rule for Icelaniic wtrlch is used for "lively

narration" (p.9S).

Marked stylist-i.e rules often seem to affect the very beginning of the

sentence. Erglish becanes pro-drop, am Gennan am rutch, topic-drop.

(92 ) (I) came in late last n.i.ght, then \t1eIlt strai.ght to bed.

(93 ) (D:ls) babe ich nicht gelesen •. '!hat have I not readI I haven I t read that.'

(94) (Da.t) heb i.k n.i.et gezegd(that) have I rot said•I haven I t said that.'

150

Stage n

<he way a granmar of this sort might change is by having this

stylistic rule which was part of the PF cartp:)nent move into the syntactic

eatlp)nent as a rule of syntax. Because it is a fronting rule, it may

easily be reinterpreted as a movanent to CG1P. Like the V movanent into

CCMP di.scussed above, this w:>uld entail the mo:/anent of V into !NFL and

the IlDVanent of !NFL into the head of CCMP. Prestrnably this movanent

\t,Ould be due to a focus operator in CCMP and \«luld have an ill~utionary

effect (Gueron 1981).

let us say, then, that Stage II has INFL movement at S-structure

because of a focus operator in CCMP. Stage I \\Ou1d have been S-o-V+I:NFL

with a stylistic rule r-s-o-v. Stage II 1NOuld be CCMP S-o-V+INFL with a

syntactic rule [ e !NFL [5 0 V]]. At this !Dint in the research, this

stage is purely hypothetical, offering a gentle transition between t\\O

other stages.

St.age In:

I am imagining reM that children of Stage II speakers are faced with

!NFL first sentences am !DSit thi.s as a deep structure order. Perhaps.

the effects of the focus operator in CCMP are lost, the fo:rm becanes trore

and rrore camcn, to the !Dint where it is considered to be unnar~ed.

Whatever the means, we Ja10rII that the result \t1aS achieved. G=>thic and Old

Icelaniic b:Yth appear verb-initial accordin3 to Haiman (p.92-93) which I

take to mean inflected verb init.ial. Lock\t.ood (1969) states that the \\Ord

order variations for Old High Gennan (CHG) were finite verb first, second

151

or final alt.h.Jugh he only gives example~ of one \\Ord verbs. A sign that

this change has taken place is that now the !NFL final order in main

clauses is ircre marked. IDck\tJOOd mentions that in OHG, such \'tUrd order

\tIaS found mainly in poetry - a tradition that extends into the present,

(Lock\tJOOd, p. 260) • ']he finite verb remains clause final, h;)wever, in

anbedded sentences (though still l'X)t as rigidly as in M:Xlern Gennan, see

IDck.wt::od) • '!his variation between root and anbedded clauses can be

acc:ounted for by the differences bet'Neen rOot and anbedded a::J.1Ps mentioned

in earlier sections of this chapter.

'Ibpicalization is very eamon, ho\t.ever, \tJe know' that it is not

obligatory. MJst of the finite verb initial cases given by 'IDck\tJOOd and

sHa.iman are cases \-Alere there either is IX) rxminative NP, (9Sa) and (95b) ,

or cases \fihere the verb can be interpreted as unaccusative, (96a) and

(96b) •

(95) a. Limphit m.irI It beb:xJves me. I

b. Tunchet m.ir rehtI It seems right to me. I

OHG: Haiman, p.l04

(96) a • Was liuto filu in flize\t2S people many in urgency

I Many people were troubled. I

b. see, quimit der brutigClDI Behold, the bridegroan canes. I

OHG: IDckw:x:x1, p.256

In both of these cases, an argunent can ~ made that there is an

anpty subject .r;osition and that, like Italian, since there is no

pleonastic element, the preverbal topic .r;osition can ranain empty, perhaps

filled by pro. rater, \fihen pleonastic elements are introduced, according

152

to~ analyses (5afir, 'Ihiersch, Haiman), it is to fill this (X)sition.

'!here is an example given by Ha..iman, llo\t.ever, that shows that

topicalization is rx:>t obligatory.

(97) wanther do ar anne wuntane bouga. Haiman, P .101'ttOund-he fran the ann \\Ourrl rings

•Fran his ann he UIl\\Otmd the coiled rings. I

Here there is a subject, the verb is transitive (therefore not

unaccusative) and further the subject is prorx:rn.inal, lessening further the

chances that there has been an inversion of the subject a.rXl the ,,'erb

{i'lrase. '!here \feOuld be no argunent, then, that an empty pleonastic

element was in topic {X)sitioD.

In surrmary, the first three stages are:

Stage I: '!he \ttOrd order is basically S-o-V-INFL toc>1.J3h scrambling is

fairly free. '!here are no adjacency requiranents for case assignnent

suggest:tng that NPs are inserted with case. V+INFL are both head-final.

Stage II: '!he structure is rnw llC~e defined with the creat.:ion of a

CGtP ncrle (i.e. [[S-o-V-INFL]]). We can rr:JW see IOOVanent of INFL into

a:MP. What was a stylistic rule in PF has becane a syntactic rule of

IOOVanent to CGtP.

Stage III: '!he \fJOrd order has 11O\tI changed to r-s-o-v which is less

marked since !NFL is adjacent to the Subject, and V to Clbject. MJvanent

rules such as topicalization reM becane camcn. Perhaps this is because a

structure has been set up that will allCM {X)sitiona to be governed. Free

'NOrd order langua.ges do n:>t seem to have .movanent rules.

153

A stylistic rule has becane a syntactic rule Which has been

interpreted as D-structure \\Ord order.

Stage IVa

Since Stage III is marked according to our assumptions (V and INFL

are not adjacent, and tOOugh they assign case to adjacent NPS, they do not

assign case in the same direction) 'Ne might expect a change to occur. One

'Way to resolve the problan w:>uld 'be to have the verb phrase becane head

initial (V-<): I-S-V-o). 'Ibis is the structure that Sproat (1983a,b)

argues for in welsh. SCrnething that I.cckwtx:x1 mentions suggests a way that

such a change could cane aOOut. When writing of t:h,e p:>sition of the

finite verb in subordinate clauses, he mentions that there are times when

<Dnstituents are fotmd after the finite verb in anbedded clauses.

Basically he states that the finite verb was IlDst camcnly followed by

-'-=-:'~::-:::~::':'::"<.:' prep:>sitional phrases but could 'be follalled by objects and other----

·....·~ements .especially if these canplanents were long, :followed by

canplements of their C'Mn, or by relative clauses. I interpret this as

saying that the VP of CHG was rL1t strictly final or initial but that it

assigned case and theta role to the left. rrhe rule that put the PPs

p:>stverbally was a scrambliD3 rule, and the one that put the ccmplanents

p:>st verbally' was a syntactic rule of Focus NP shift (see St.c:1IIe11 1981).

It is easy to see how this might be<:xme the unmarked order.

Stage !Vb

Another way to resolve the instability of ISOV is to reanalyze the

topicalized NP as the subject. In tenns of acquisition, one might imagine

154

that a child hearing only simple sentences and predaninantly ones with

subject topics, w:>uld analyze this as a D-structure string rather than

transfo:rma~ona!lyderived strin3. later when encountering the Stage III

object-~pic strings, he would account for these with two transfonnations

- 'Ibpicalization and IIDVanent of !NFL into CG1P. This means that a child

hearing (98) will analyze it as (99).

(98)

(99)

[s. &Jbject. I. [s t. Object V t.]]~ J 1 J

[s Subject !NFL Object V]

I claim that this is the J.X>int Where ~ insertion is introduced, and v.here

the V.(~ constraint can first be observed (this will be discussed in rrore

detail in O1apter 5) ~

I have provided. a lengthy al."gU1leI1t that Germanic It:dlguages are

underlyin.gly !NFL secorxl. l-ly conclllSicn, then, is that Stage !Vb is

NP-INb""L-VP. Gemaan differs fran Yiddish in that Gel:man· s VP is

head-final, while Yiddish' s VP is head-init.i.al (similar to wel~l)) .

Alth:>l.l3h the V and !NFL are not contigtDus in Gennan, they each assign. ,

case leftward to adjacent (see S1:.c::1.-Jell for different definitions of

..ad jacent") NPs. Yiddish forfeits unity of direct..ion of case assignment,

but gains adjacency of !NFL and V.

Stage V

English has a configuration similar to Yiddish; NP-INFL-V-NP. '!here

is one ranarkable difference, bJwever. Yiddish shows V/2 effects while

English, generally, does not. ~scriptively, this simply means that

English has lost the rule which m::wes !NFL into CCMP in these cases.

155

(99) 'Iha.t kirrl of bcx>k, Eai.th likes.(99) "That kind of bcok likes Faith.(99) Yesterday, Faith wrote me a letter.

One \\Ould like to rePhrase this in tenns of a reanalysis of structure

(see Ie.~~rt 1984b for a similar change :in Hebrew), but this is the

Subject of \\Ork in progress.

3.6 Conclusion arxl Typological Speculations

'!be spirit of this section is that changes in language sh:luld be

restricted along the same lines that gramnars are restricted. With the

theoretical shi.ft fran gramnars of rules to gramnars of principles arxl

parameters, there is also a shift in the way that historical syntax should

be vieweCl. Changes in gramnars can no longer be described in terms of

rule lo~s or rule ret:.)rdering, but rather in changes of parameters ~ \.lust

as rules used to be uncovered by examining different gramnars (be they

different stages of the same language OI different languages), so can

pararoeters be teased out by exami..nin:J h:M languages can differ.

Fbssibilities of \\Ord order change can also give us insight into the

relat."lve distribution of certain \\Oro orders. 'Ibis is dangerous \'later

since, as Pulllm (1982) ~ints out, language distribution may have

~litical explarations as well as markedness explanations. I,

nevertheless, offer speculations as to \fJhy, for instance, CSV languages

are so rare (Pullum (1981) lists Apurina, Nadib, Urubu, and Xavante, all

languages of Brazil). If tNe consider the mirror image of CSV, that is,

156

VSO, \4Ve firrl these languages to be nore aburrlant. 'lh.is follows fran the

historical explanat.i.on given above.

Assun.i.ng that V-S-O languages are actually r-s-v-o, the marked

elanent in these languages is that I and V are no larger contigtX)us. If

the developnent of Gennan languages given above is correct, the way that

such an order may OCCllI is t1u:ough the IOCWanent of 1° into CCMP. If it is

also true, as 'Ne have suggested above, that only sentence initial CCMP is

free for IOCWanent, this split of INFL and V will only cane about through

an INFL fronting rule, creating •• I ••V.. but not ••V•• I... 'Ihis offers a

syntactic-: explanation for a a:>ticeable irrbalance of \\Ord order.

'!here is a markedness argunent for another rx:>ted imbalance, i.e • the

disparity in the nllIlber of S-V-O lan:rauges versus OVS languages. I

suggest that there is a tendency for predicates to appear on the right.

Sane Malaya-Polynesian languages have urrlergone the change fran V-O-S to

s-v-o. 'Ihis could also occur th-w:-ough a fronting rule \tJhich is then

reanalyzed as a d-structure predication relation.

(103) [8' [S I V 0 S]] --->

reanalyzed as .. r~'s~ I V 0 ]

[s. S [s I V 0 t ]]

Because of the markedness corrlition described above, we \\Quld not

expect an object fronting rule to be susceptible to the same reanalysis

since the result..1.ng structure \t«)u1d be o-I-V-S which \tJOuld precllX1e the

fX)ssiblity of haviXl3 a W.

Historical change also argues for t\\O theoretical claims, one is the

exist.ence of a separate INFL ncrle ani the other for a triparti.te structure

157

of II.

As \\e have seen irl the suggested historical stages above, it is the

ability of !NFL t.o act as the head of CCMP which triggers l3analysis, am

it is the inconsistency of the danands on !NFL which allONS the

reanalysis. Witoout the separat.e INFL node, l'Dt only \\QuId synchronic

accolBlts of certain grarnnars (Gennan, Irish) be more canplex, but also

diachronic accounts \IeOuld be rrore puzzlin:J.

By having a t..ripartite structure for I I, independently motivated by

our discussion of branching in C1apter 2, !NFL is allaNed to appear in

rrore configUrations. BelCJ.t1 'Ne canpare In structures wi.th I I structt:tres.

(104) II' structures: I I structures:

a. In b.

~NP I'

~I VP

~n

NP' "II~

VP I

c.~

NP I VP

d.~

NP VP I

e. In f. In g. I' h. I'~ ~ ~ ~.,

II NP II ~ VP I NP I VP NP

VP/"-.....I~

I VP

i.~

j. II

~I NP VP VP NP I

The.tripartite structure allows the sUbject node to intervene between

the !NFL node and the VP. '!his is the structure I prop:lse for VSO

languages. Witrcut this p:lssibility, reanlysis of the INFI.. fronting rule

of Stage II into the d-structure of Stage III would rnt be p:lssible. It

is true that the other structure allowed by the tripartite analysis is rnt

158

evidenced by any language (VP-NP-I), b:lwever, as I have explained above, I

feel that there is a historical explanation for this gap.

'!he above discussion is a sketch of h::J\.I \\Ord order parameters am

cordit.i.ons on d-structure m.i.ght lead once again to a rrore enlightened view

of language typolO3Y, roth synChronically and diachronically.

159

In thi.s chapter I argue for a ne\tl fonn of proper goverrment, that of

VP ~ernment. It has been noted that the Enpty category Principle, in

effect, is a recoverability corrlition on traces (see tbuchard 1982 am

references cited t.herein). let us say that every empty category must. be

identified am that this identification ha.s t'ttO parts. First the gap

i tsel f must. be ident.i.fied, i.e ., the fact that there is a gap. rrhis can

be done either by proper governnent ~.i an >f or by the presence of a lcxal

antecedent. 5ecorrlly the content of the gap must be identified, Which is

done by the transnission of features througll coiOOexation.

I am primarily concerned with the ident.i.fication of a gap, and

specifically trose gaps with no local antecedentsl • I will be assun.i.ng

that !NFL is capable of transmitting features but not capable of

identifying a gap, i.e., it neither properly gO\1erns t'¥)r cotmts as local

antecedent for the subject l,X)sition. 'lherefore, I will be concerned with

t'ttO types of gaps: toose left by extraction out of t.he local danain, am

trose that are coindexed with INFL fOI' transmission of features (pro for

1. For the identification of cnntent I will use the frama-.ork described inBouchard(1982) (see Chapter 1).

160

Italian and Irish, pleonastic pro for Ge.nnan and Malagasy).

In t.he first section I m:>t.i.vate an account of Gennan and Yiddish

~-insertiondi.fferent fran that prol,X>sed by Safir (1982). My claim is

that the pleonastic elanent, ~, is not in CG1P but in subject position,

follOlling the analysis of Chapter 3, and that its distribution can be

explained. through the presence or lack of proper gov~ent. Tl-J.s,

b:>\t.ever, 'Nt>uld require a different notion of goverrnnent fran the one

assuned in current frame\tJOrks (lGB, Aoun & Sp:>rtiche) since I am

suggest..ing that subject l,X>sition can be properly governed. by the VP.

'!he secx>nd section investigates nore closely this mtion of VP

goverrnnent. I prol,X>se that the relevant not.i.on is not a structural one

but one of canplanentation, like that of Jaeggli (1900), St.a-lell (1981),

Huang (1982). If an NP is in a ccmplanent pJsition \thi.ch can be assigned.

a 6-role dixectly, it is in a l,X>sition of proper goverrment. I further

claim that this includes not only canplements of a V within the VP, a

traditional dancdn of proper goverr:ment, but also a subject NP l,X>sition

which is adjacent to the VP. Since this involves a change in the

defini.tion of what a canplanent is, I discuss canplanentation am prol,X>se

a new definition.

In t.he thi.rd section of the chapter, I introouce evidence fran other

lan:JUages to support the notion of VP governnent. I first claim that

l,X>st-verbal subjects in Italian are properly governed. rDt beca\lSe they are

in the VP but because they are in a cnnplement IXJsition to trie VP am

t.hereby properly governed. by the VP. 'Ih.:ls accounts for the lack of thatr-wt

161

effects in Italian. ChaIrorro (Clung 1983), a V-S-O lanJUage, and

Malagasy, a V-o-S language, also have no that-t effects. I assume an

analysis for ChaIrorro along the lines of E1toms (1980) am Sproat

(1983a,b) which derives a V-S-O structure by V-fronting. I suggest tllat

the tmderlying "-Ord order of 'both languages is I-NP-VP am that the

subject p:>sition is properly governed by the VP. Finally I suggest that

pro-drop ent:ails proper goverrment of the subject p:>sition. '!he reason

that pro-drop characteristics in R:Jnance languages appear to coincide with

Subject-inversion is because subject p:>sition is properly governed in an

I-VP-NP order (subject-inversion) but not in an NP-I-VP order. 'Ihis

explains why Irish is pro-drop witrout having subject inversion since even

without subject inversion, the subject p:>sition in Irish, an I-S-VP

laIl3Uage, is properly governed by the VP.

In the last section, I discuss the distinct.i.on that must still be

made between properly governed pJsitions within the Vp am properly

governed p:>sitiona outside of the VP. Huang I s CCJrxlition on Extraction

n:ma.ins (CEO) states that properly governed constituents are p:>ssible

danains of extraction. IbNl:!ve.-.:, constituents that are properly governed

by XPs do not allCM extraction while constituents that are properly

governed by >fJs do. I then prop:>se a new fonnulation of the CEO to

account for thi.s distinction.

162

4. 1 Es Insert..ton

4.1.1 Facts

GeITllaD. has t\\O pleonastics which will be discussed in detail in

Chapter 5. It is sufficient, at thi.s p:>int, simply to di.stirgui.sh the t\-JO

by their superficial properties. A <:.'Onfusion arises because tmder certain

cirunstances, roth appear as ~.

(1) a. Es regnetit rains

•It is ra.in.in3.'

b. Es ist klar, daB die Frau das Buch gekauft hat.it is ·clear that the~ the 'OOok oought has

I It is clear that the 'NOnaIl has l:ought the l:x:ok. I

(2} a. Es siOO drat Ktrrler gekarmenIt are three children cane•'!here have cane three children.'

b. Es wird getanztit becanes danced

* ''lbere \eS danced.'

HcJ\Ever, tmder other conditions s1'lo'Nn belCM, the difference is

obvious. One fann of es which I will call ES is able to appear in these

environments. '!he other fonn of as which I will label FS/O is not able to

appear except in its 0 fonn.

(3) ESa. Heute regnet es.

Today -

b. Heute ist !! klar, daB die Frau das Buch gekauft hat.

163

(4) ES/Oa. *Heute sind es drei Kinder gekcmnen.

Heute sind 0 drei Kinder gekam1en.b. *Heute wird es getanzt:..

Heute wird Ogetanzt.

In thi.s chapter I discuss only FS/O, holding off the di.scussion of ES

am the distinction between the t\\O lmtil Chapter 5. '1lle question I

address ~ is \\here does ES/O appear as ~ and \\here as 0, and why.

The facts CX)ncerning the distribution of ES/O in Gennan ani Yiddish

are given belo,{. Basically the pleonastic surfaces as ~ only in clauses

that (1) sb::w V2 effects, and (2) have no topicalized ex>nst.ituent.

4.1.1.1 Gennan

we have seen tllat, in Gennan, V2 effects occur only in clauses \\here

there is no +/-wH canplanentizer.

( 5 ) a. Die Frau bat das Buch gelesen.the \\QtlaI1 has the 1:xx>k read

•The 'NatlaIl has read the book.'

b. I:a.s Buch belt die Frau gelesen •the book has the 'NatlaIl read

I The 'NatlaIl has read the 1:xx>k. I

(6) Ieh glaube, daB die Frau das Buch gelesen hat.I believe that the \fJOtlall the l:xx>k read has

I I believe that the 'ttOtlaI1 has read the bcx>k. I

In (5) we can see that the inflected verb is :in secoIX1 fOsi.tion

whether the subject NP or the object NP is S-initial. In (6), the

eni:>edded clause with a -wH canplernentizer has the inflected verb clause

final, i.e ., there are no V2 effects.

164

If the descript.i.ve analysis of the appearance .of ~ is correct, i.e.,

!:! only ~urs' in V2 clauses with IX> topicalized cnnst.i.tuent, we expect to

find ~ only in the type of clause shown in (5), with the addit.i.onal

restrict_ion that no elenent has been topicalized.

(7) es wurde heute getanzt.- becane today danced

I There \\'as dancing tcrlay. I

(8) Heute wurde (*es) getanzt

(9) Ieb glaube, daB (*es) heute getanzt wurde

(7) exanplifies a case Where ~ appears. '!he clause is V2 and there

is IX) topic. (8) exanplifies the fact that ~ cannot co-occur with a

topic, and (9) exanplifies the fact that es cannot occur in a clause Which

does rx>t slD.t1 V2 effects.

4.1.1.2 Yiddi.sh

In Yi.ddish, Where we find V2 effects in clauses with, as well as

witlDut, canplanentizers, we expect to fim as in 1:x:>th environnents. As

\tJe see below, this is the case.

(10) a. Es kumt a kind in kron.- canes a child in store

I A child canes into the store. I

b. Ikh rneyri az es kumt a ki.nd in der kran.I think that

I I think that a chi.ld canes in the store. I

And, like 'Gen'nan, Yiddish does not have.::! co-occurring with a

topic.

165

(11) a. In kran kunt (*es) a kirrl.'Into the store canes a child.'

b. Ikh meyn az in der kran kunt (*es) a kirrl.

'!he generalizat.i.on may be st.at.ed, then, that ~ aJ?Pears (i)

sentence-initially in a (ii) V2 type clause that (iii) does OC)t cont.ain a

topic.

4.1.2 Safir , 5 analysis

safir (1982) is able to capture thi.s generalization by claiming t:h2lt

es is in C04P. 'lhis neatly explains \JJhy es ~ll never ~ur with either a

topicalized phrase or a canplanentizer. 2

In safir , s analysis, like 'Ih.i.ersch· s analysis presented in Olapter 3,

the D-structure of Gennan is SOVI. In V2 clauses, aCCX)rdirg to Safir , s

analysis, it is obligatory that 1) !NFL ITDVe, and that 2) CCMP be filled.

(12) a. D-structure:

b. n:JFL IIDVanent ~

c • X" IIDVaDent:c I. es Insertion:

[5 I [5 heute getanzt wurde]

[s'[s ~e. [8 heute getanzt t.]1 1

[s' he'Jte. [s wurde [s t. getanzt t.]]][Sf es [sJwurde . [ heuteJgetanzt t.1]]

1 1

In (12b) above we see a case \JJhere !NFL has m:JVed, in (12c) CCMP is

filled by the constituent heute 'today', and in (12c') \E see a case 'Nhere

CCMP is filled by~. '!he facts of es-insertion as explained in safir , s

analysis are given below.

2. N:Jte t..hat the Yiddish canplanent.i.zer does lX)t "count" for V2. I willasstme! here t.hclt YiCHish is [CCMP [mPIe [8]]]. In safir' s analysisextended t.o Yi.ddish, es would be in 'IDPIC posit.i.on ailO\dng it to co-occurwith a canplementizerbut not a fronted X".

16{.

(13) a. [s,es [5 Ii [5 VP t i ]]]

b. [8' X" [8 Ii [5 *es VP t i ]]]

c. [5' daB [ *es VP I]]]

If es is used only to fill an otherwise empty CCMP, we can see Why

(13a) above is granmat.:tcal While the ~ in (13b) and (13c) are ruled out.

In both (13b) and (13c), the es is in a p)Siti011 within S and not in

CCMP.

4.1. 3 VP Government

Argunents are given in Clapter 3 as to why Gennan and Yiddish are,

both s-r-vp, rather than S-VP-I, am neither INFL frontirv:J l'X)r

'Ibpicalization are obligatDry.

1.bis, per se, does not create problems for 5afir' s analysis. Q'le

cx>uld still sa.y that if a subject p)sition were empty, then !NFL \\Ould

have to front into CXMP (perhaps to properly govern the empty subject

p)sition) am then the pre-INFL p)sition must be filled in order to

prevent t.he string fran being interpreted as a questi.on. Fs, in thi.s

analysis, \\Orks like a pleonastic CCMP am it is crucially in Ca-1P

p:>sit.ton.

Q'le of the arguuents for the S-I-VP order of Yiddi.sh, 1'1aIJever,

prcwides evidence against this analysis. '!he extraction facts presented

by I.cAEnstaIml (1977) give us a way. of distinguishing bet'Neell elenents in

CCMP am elements in subject p)sition. Basically, a -tWH canplenent canrx>t

co-occur with an elanent in CXMP but it can co-occur with a pre-INFL

167

subject. '!he crucial examples are given ~low.

(14) a. Fs zaynen gekumen dray kinderare cane three children

•'!here came three children.'

b. Ha.ynt zaynen gekunen dray kinderrrtxiay are cane three children

•Tc:day three children cane.'

(15) a. ikh veys nit far ves es zaynen gekumen dl-ei kinderI krDw nX why are cane three children

•I don' t know why three chi.ldren came.'

b. *ikh veys nit far vas haynt zaynen gektmen drei kinderI know not why today are cane three chi.ldren

•I donIt krDw why three children cane today.'

(16) a. es iz mir kalt in taimer.IE is to-me cold in roan

I I am cold in the roan.'

b. In tsimer iz m.ix kalt .•In the roan I am cold. •

(17) a. Ikh veys nit, far -"as es iz mir kalt in tsimer.I know not ~or \tihat it is to-me cold in roan

•I don' t krDw why· I am cold in the roan.'

b. *Ikh Ve!:fZ nit far vas in tsimer iz m.i.r kalt.I krxJw I'¥Jt for what in roan iz to-me cold

As 'Ne can see in (16a), the ~ \\h.i.ch appears pre-INFL is' in Subject.

p:>sition otllerwise tlle extraction that we see in (17a) w:>uld I'X)t be

allCJJNed. In (1Gb) where topicalization has occurred in ~'le enDedderl S, IX)

further extraction is p:>ssible as sb:lwn in (17b). 3

'Ihis s~ that ~ is in subject p:>sition :in Yiddish. We can use no

such argtment for Gennan. Eh11:>edded clauses in Gennan do ll':)t show V2

3. '!here are sane cases lNhere es seems t.o co-occu.r wi.tll syntacticsubjects. '!his will 'be discussed in section 4.3.1 When 'tie can};aIe German"subject inversion" with Italian SUbject inversion.

168

effect unless there is I'X) CCMP FOsition.

(18) Er sagte, daB sie das Buc~ gelesen hat.'he said that she the bcx:>k read has

I He said that she read the 1:)(:)(:'k. I

(19) Er ,-;agte, sie 12tte das Buch gelesen.(20) *Er sagte, daB sie hat das Such g€lesen.

since extraction will always create a CG1P, thi.s cannot be 'lSed. as a

test for preverbal constituency as in Yiddish.

Titere is, however, another arg'.I1ld1t that the Gennan pleonastic ~ is

in subject.. FOsit".Oll r'.ther -chan CCMP. WE:: have seen ;in Chapter 3 that the

refe',rential. prorx:nm ~ can apJ?ear prevPxl)ally only if it is subject.

(21) Es hat das Brat gegessenit has the bread eaten'It has eaten the bread. I

(22) Der Ii.md tnt es gegessenthe dog has it e=iten

'The dog has eaten it.'

(23j *Es hat der Hund gegessenit has the dog eaten

'The dog has eaten it.'

In (21), ~ is the praverbal subject, while in (23) the preverbal ~

is the Q~ject of tlle verb and tilerefore the sentence is ungranmat.ical (see

~pt.ar 3 for details). Qi the assl:.mpti.on that the referential ~, cannot

be in CG1P, we wi.ll also say that_ the ple:>nastic ~ cannot apJ?ear in

CCMP. This is e~J?E!Cially c!(,ar if we give Sanalltic c:ontent to this

rest.ricti.on. If o.l1y elanents that are sernant.i.cally salient eIDugh to be

focused can appear in topic FOsition, it is clear Why ~, botll ~efel'ential

am pleonastic, may not be topicalized.

169

If, as \tJe have argued, es is in subject fX>sition in roth Gennan and

Yiddi.sh, we are left with problans that Safir I s analysis accounts for.

Below 'Ne see once again the ~ di.stribut.:lon facts, rrJW rebracketed in

tenns of the analysis presented in O1apter 3 (canpare with (21».

(24)

(25)

(26)

[s es I vp]

[5' X". I. [*ese. [vp ••• t .••• ]]]J ]. ]. J

[daB [ *es e [vp ••• V+I]]

1lle dist-libution of ~ is no longer clear. Es is in the same

posit.i.on in each structure, only the realization of INFL differs. I

claim, lDwever, that it is this change in n1FL that accounts for the

di.stribution of es. In (25) !NFL IYDVes into CCMP and in (26) it is empty,

but in each case it leaves no {ilonetic material bet\\een the subj'eCt

position and the verb phrase. I prop:>se that the adjacency of the VP and

the subject NP, where a phonetically realized INFL no longer intervenes,

sets up a special relationship that allows the subject NP to rana.in

empt.y. In the following sect.i.ons I discuss this relation claim.i.ng that it.

is one of proper government.

4. 2 Goverrment

Ll this sect.ion I rrot.ivate t.he r..ot.ion cf VP 9V"ernment first by

explaining what it might mean conceptually, 3Irl then by givirg evidence

fran various languages to support my clai.m.

170

4. 2.1 Identification

It. is not clear What VP government means. Goverqment recently has

been defined in structural tenns (LGB). It is rxJt only a relationship

tllat ooIds of a lexical ,-,~tegory and its canplements, b'It the relationship

that ooIds of a lexical category am any other category which it

c-ccmmands within the same maximal projection (ooun & Sporoehe). '!his

\t,Ould include rnt only (a) belor:" but also (b) and (e) if we do rx:>t

consider S a maximal projectton.

(27) a. VP

VJb. VP c.

v/'--.. s

~~VP

NP

ANP N~

(27a) repre~~ts a verb and its canplement, (27b) a raising or EX:M

structure, and (27c) a rx:>lU1 phrase. As in Jaeggli (1980), St-ONell (1981)

and Iasnik & sait.o (1984), I restrict proper government only to (27a)

claiming that the other structures may be accc:nmta:i for as suggested in

\\Ork such as Bouchard (1982), Iasni.k & Saito (1984) . If proper goverrment

is (27a) am rot (27b) or (27c), the rotion is no lOlXjer purely

structural, nor can it correlate with case assignne.nt. It is not

structural since (a) and (e) have the SaITlE! structural configuration. It

does n:>t correlate with case assignnent since in (a) and (b) case is

assigned to t.he NP, but Oil1y in (a) is the NP properly governed.

In Olapter 1, the not-ton of Full Interpret.at.i.on was intrcxiuced. It.

specifies that all elEments must be licensed am that this licensilXj can

171

be att.ained either by canplan~.ntat.i.onor predication. I..a:lkiD3 again at

the st-ructure of English, 'He can see the distinction bet\f.Jeen the t.\\O

methcds of licensin:J. Elements licenced by canplanentation are connected

to tJ1e tree by full lines, and th:>se licensed by predicat_ton are connect.ed

by dot..ted lines.

(28)

Taking the core examples used to explicate the lOtion of proper

govenment belCJW, we see that the main asynmetry is bet~ subjects and

objects.

(29) a. Who \\On't you say that Mary saw t?b. *Wtx:> \\On I t you say that t saw Mary?

With t..he range of data extended a bit, we can see that there is an

asymnetry even within the verb phrase between adjuncts am canplanents.

(30) a. 'lb \\han \'tOn I t you say that Mary gave tile book t?b. *fbw slowly \leOn' t you say Mary gave the book to Joan?

I have coosen to negate t.he matrix verb for the fo11aNing reason. As

Iasni.k & saito (1984) rx:>te, the foll~ exarnple is gramnatical.

(31) How slOllly did Sean say that Matthew gave ~ar~ the book?

I am assun:i.D;J, hJwever I that this is rx:>t a core case since, as 6CX)n

as t..he mat.rix verb is given roore content I t.he construction \\Orsens.

(32) *HJw slOfllly soould Sean have said that Matthew 'gave Jared the ~k t?(33) *How slC7l11y might Sean believe that Matthew gave Jared the book t?(34) *HJw slowly does sean regret. that Matthew gave ,Jared the book t?

172

Canpare these with

(35) What should sean have said that Ma.tmew gave Jared t?(36) What m.i.ght Bean believe that Matthew gave Jared t?(37) What does sean regret that Ma.t.t.hew gave Jared t?

The explanation for these nay lie in the content of CCMP. If it can

be argl1ed that the lmgramnat.i.cal cases abL,ve are similar to wh-i sland

effects,

(38) *How slOllly does Sean 'NOooer what Ma.t.tllew gave Jarecl?

am that negation, m::rlals, am factive verbs affect the embedda:l CCMP in a

way similar to a \Ith-};i1rase in CXMP, Iasn.:tk & saito's proper governnent

will suffice. At this p:>int, I simply p:>int out the di.fference between

adjtmcts and canplements. '!his di.s-t-..inction may also be acCOtmted for in a

di.fferent way as sU93ested by Cinque (fort:.b:nni.ng)4.

I will follow the prop:>sal of H-Jang (1982) and assume t.hat

canplenents are properly governed whi.le adjuncts are oot. Within the

tenns of Full Interpretation, thi.s translates into the claim t.hat those

elenents licensed by predication a't"e rnt properly governed whi.le toose

elements licensed by canplanentat.i.on are.

What is the intuition behioo proper goverrment? Let us say that. (i)

every gap must be identified and (ii) that the content of every gap must

be recoverable. IiJw are gaps ident.i.fied am b:Jw are contents

recoverable?

4. '!his was p:>inted out t.o me by IJJigi Rizzi.

173

4. 2.1.1 ~p identification

'!he presence of a gap may be identified in t\\O ways - either by being

requ.ired through canplenentation (lexical governnent), or by the presence

of a local antecedent (antecedent government). '!he claim is that empty

categc>ries must be identi.fied am that lexical governnent is only one "Way

of ident.i.fying the gap. BelOIT we can see t.he t\\O types of

identification.

(39) Antecedent governnent:a. I didn't say woo Mary saw t.b. I didn't say why Mary saw Joan t.c. I didn't say wlo t saw Joan.

(40) Lexical goverrment:a. Who didn't }IOU say that Mary saw t?b. *Who didn I t }IOU say that.. t saw Mary?c. ~y didn I t }IOU say that Mary saw Joan t?

In (39), the gap is ident.ified by a local antecedent. In (40) where

t.he antecedent is no longer local, we can see the di.fference bet\aJeen the

case of the (39a) example am the (39b) and (39c) examples. In (39a), the

gap is identified I"Dt only by the lcx::al antecedent, but als:> by lexical

goverrment since it is in a canplement p:>sit:ton to the verb.

St.o\t.ell (1981) unifies these t\fJO IDtions, that of lexical government..

am thclt of antecedent goverrment, urrler one rntion of coirrlexation. The

basic idea is that indexation of an argunent with the 9-grid of the verb

serves as a lC)Cal antecedent:. relation (p. 303) •

II

(41) [which book]i did you say that. Ben [VI [V read ] [e].][OBJ-e]. 1

1

174

Here the [e] is antecedent governed by the [OBJ-eJ in the G-grid of .

t.he verb, in the same way the [e] is antecedent governed by 'liho belOW'.

(42) I \'tOnier [8' [who]. [ [e]. read Po:Jer's bcx:>k]]1. 1

I keep these t\teO types of identificat:lon separate, as do Lasnik &

Saito, for conceptual reasons. 'Ih:>ugh 'both a local antecErlent and a

lexical governor in sane 'llay identify the presence of a gap, the method of

indentification is not the same. '!he identi.fication of a gap by way of

lexical goverrment follONS fran the Projecti.on Principle and the

e-criterion. If a lexical i tan requires an argunent and IX) argunent

appears, there IlUISt be an empty category. Antecedent gove.rrment follows

fran the i.nq:xJssibility of vacuous quantification. If there is a

\'h-operator in a:MP, there must be a variable within its danain"

Scmet.:i.mes that variable will be in a IX>sition which is already ident.i.fied

by the Projection Principle (i .e., lexically governed) • If not, however,

the operator nay itself identify an adjunct gap •

.4~ 2.1.2 Recoverability of ,features

rrhe prob:!.an of recoverability is inextricably tied to the problans of

chains, whether A'-chains or A-chains. rrhe c:ontent of a gap will be

recoverable in the cases lNhere it is coindexed with an element with the

proper features. 'Ibis elanent, in nost cases, will be the head of the

c~in. Variables will recover relevant features fran tlle A'-binder, and

NP-traces will recover relevant features fran A-biniers. In other cases,

\'here empty categories are base-generated in &-marked IX>sitiona, the

features ma.y be transnitted by arx>ther element. In the case of anaphoric

175

PRO, this will be the antecedent of the anaphor. Arbitrary PRO, according

t.o lbuchard, is assigned t.he R-index aEbitrary. In the case of pro, this

will be INFL. 'lhis coindexation with INFL will be discussed in IlDre

detail in later sect.i.ons of thi.s chapter I and in Clapter :'.

4. 2. 2 VP Goverrment

If gaps may be idenr_i.fied through canplementat.i.on, we can explain Why

a gap in the canplanent p:>sition of a V may be id\~ti.fied, but what of a .

subject? If the subject position is alW3.YS licensel through predicati.nn

and oc>t canplanentation, it can never be lexically gove:med. I proi;ose

that there are tw:> ways that a subject NP is distinguished fran a

canplanent NP within the VPo Cile stens fran -the fact that a VP, unlike a

V, when it assigns a 9-role to a subject, may assiS4~ it in t\\O different

\tJays, as \'Je will see belOW". '!he other stens fran the fact that the VP,

unlike a V, is a phrasal category rather tl1an a lexical category.

UOOer certain ex>nditionc;, we can say the &-role is transni.tted fran

the VP to its extel-nal argunent by coindexation of predicat:ion. In thi.s

way we e:cpect subjects to behave like adjuncts (Huang, 1982) since

adjt.mcts, too, are licensed by predicati.on (Chansky, fall lectures 1983).

I ProfOse that there is arnther way in which a &-rol~ may be assigned to

an external argunent. '!his secorrl 'Way is direct &-marking, the same 'Ila.y

that a V assigns 9-roles to its canplenents. My ProIDsal, tllen, will 1Je

t.hat subjects ma.y be canplements of VPs and \'then they are, they are:

properly governed by the VP.

176

Obviously we do n:>t want all subject:s to be canplanents otherwise we

'NOuld expect no subject/cbjec:=t asymnet..ries. We must, then, determine ~t

corrlitions must l:e put on this sort of canplanentation.

We saw in O1apter 2 that the O:main Mjacency Corrlition will not

allO\tl the danain of canplementation to l:e interrupted. We \\QuId expect

the same sort of condit.i.ons to mld for a VP and its canplanent NP. In t11e

VP, if a canplanent \tIa.S not within the canplenent danain of the verb, it

c:ould IX)t be assigned its e-role and the structure \'as ruled out. At the

level of I I, lrJrNever, there is an alternative since the VP can assign tha

subject NP its 9-role via predication. Fbr the mcment I will say that the

relevant element for the di.sturbance of adjacency of the VP and the

subject NP is the head of the projection, i.e., INFLo. '!he ProfOsal is

that if VP and NP are generated adjacent to one another, or if INFL is

empty I then the NP can be C'Onsic1ered to 'be the canplanent of the VP.

Taking the structures relevant to ES/O once again, we can see that in

every case \ti1ere !! is not fOssible, the VP is adjacent to the subject

fOsition, and therefore thesubj~t is in a canplanent fOsition. N:>tice

that. empty categories are not visible for this notion of adjacency.

(43) es I VP

(44) X" I *es e VP

(45) daB *es e VP

Since FS/O can surface as either ~ or 0, we have a choice of \ti1et.her

'He take ~ or 0 as the urrlerlying fonn of the pronotm. In order to

distinguish thi.s plEOnastic fran 111e ES of example (1 ), I will assume that

177

it is o. Within the theory of GB, empty pronaninals of this type that ma.y

appear in governed {X)sitions are considered to be pro. Ebr the rest of

this thesis, then, I will be referring to tile FS/o pleonastic as pro.

· Unlike previous a~unts of pro-drop languages (Jaeggli 1980, Rizzi 1982),

I will 'be assuming that pro has to be properly governed. Like any other

empty category, PE2. must be identi.fied, am since it has no l~al

ant.ecedent, ident.i.ficatfon of the gap will entail proper government. In

thi.s section I will discuss the requiranent of proper gO\Terrment arrl. leave

tlle details of feature recoverability to Chapter 5.

'!he structures \fJe are rrJW concerned with are given belOl/...

(46) *pro I VP

(47) XU I prrl e VP

(48) daB pro e VP

The ungranmat_i.cal string (46) is the one in which the gap ~ts not

identified because it is rxX properly gavenled rx:>r does it have a local

antecedent. In these cases, the least marked prorx:)lm ~ must be inserted

tD act as identification. I will asstllle that ~ is D:>t only not required

in (47) and (48), but i.mpJssible because of the last Resort Principle

(Oonsky class 1983) which states that ~e-a occurs only when required.

I ext.em thi.s to Affect-a which will include es-insertion. In pro-drop

la1'lC3l1ages, prorntnls are only inserted to fcx::us than. 1hi.s makes 00 sense

in tlle case of a pleonast.ic elanent (pointed out by Mannru saito). In

· (47) and (48), the empty category pro is identified by beiD3 in a

canplement position to the VP.

178

We do 1Dt want to say in these constructions that the VP actually

assigns a 9-role to the subject p:>sit.i.on, since ~ is a pleonastic element

that CanBJt receive a 6-role. '!his bri.nc:Js us to the second distinction

bet~en the relationship of a V and its canplanent.s and the relationship

of a VP am. its external argunent. Since V is a lexical category, the

relati.onship bet\tJeeIl the V and its canplEments can depend on the lexical

characteristics of the V, i.e. I Whether or 1Dt the V assigns case or

assigns e-:roles , etc. '!he relationship between a VP and its syntact_1.cally

external argunent CamDt be sensitive to such characteristics. This means

that t.he e-role a'3signing properties of the V are not visible at the level

of the VP. Whether or rx:>t the VP contains a V with an external thenatic

role to assign, the NP may be in a canplement relationship witil the VP. It

is thisrelationsmp that allO'NS proper goverrment of the subject NP

position. Because of this distinction bet'Neen a V and a VP, let us keep

separate the mtions of lexical goverrment ani canplement goverrment.

Both will insure proper goverrment for the ~P, end lexical government

will be a type of canplanent goverrment. we will see in section 4. 4 the

need for this distinction.

4. 2. 2.1 Ebnnal D3finition

Using a fonnalism sim.Uar to S~ll's, .~ can see the notion of VP

gOVernnerlt nore clearly. Below tNe see a case where 9-roles are ass:i.gned

to t.he internal argunents of the verb.

(49) VP'

~V NP. PP.

OBJ= i 1 Je

PP = je

Let US assume t.hat every VP also has a e-grid. '!he -way which this

179

9-grid will di.ffer fran that of the V is that its 6-role must be

determined by its head. In other ~rds, the external e-role must

percolate up fran the >f level to the xnax level. We might imagine, then,

a struct:.ure as belOll.

(50) I'~

NPk)~k

V e ~.E 1

eOBJ = i

ePP = je

P.J

'!be external 9-role (E), if there is one, will percolate t..o the VP

am be assigned to the NP tNhich is syntactically external to the VP.

~ 'Ne can say that the subject, NP is tCI the VP What the object NP is

to the V. In each case, an index has bE=en register~ in the 9-grid of a

sister oode. '!his leaves t\«> questions: (i) \fJhat happens in a predication

relation and (ii) what happens with a pleonastic.

lbtllst.ein (1983) defines a predicate as

180

an open one-place synt-actic function requiringsaturation, or closure·by an argunent. The syntactic unitwhi.ch may be a predicate is a. maximal. projeCtton (XP). Allmaximal projections are one-place functions which requireclosure, but an XI? is a EE-rl.icate oilly when it is closed by anexternal argunent, i.e., an argtmE!1'1t which is not c-camnandedby the head of t.he XP.

In l:x:lt1l R)thstein I s and Williams I \\Ork, all maximal. projections are

capable of being pred.:i.cates. An NP or an f11 must have an open argunent,

b:Jf...ever, to fLmction as a predicate (see Williams, 1980).

VPs which assign e-roles externally obviously satisfy this defini.tion

of predicate since they have an open argunent. My claim is FJimply that

there are t\\O wa.ys to saturate this open argun~t. Above \t.e have seen~

a canplanent subject may saturate the open IX>sition. '!his, presunably, is

only IX>ssible urrler the cordi-tions of adjac~ which \\e mentioned above.

Predication, I assune, creates a different structure denot.ing a

di.fferent relationship. '!his is given below.

(51) II

I VJte

~~.V NP~

Ee

OBJ = ie

PP = je

j "i

When the external 8-role percolates to the VP, the VP will still have

an open argunent which will only be saturated at the level of

predication. It is only where a constituent is inco:rp:>rated into tlle

181

8-grid of a nOOe tbat it is canplanent governe3.

Pleonastics present a different problem. I have suggest.e3 above that.

anpc.y pleonastics IlU.1st be canplanent gCV'ern~. This appears inconsistent

with what 'Ne have said above that canplement.s enter into e-grids. It is

hard to see b:lw an elanent that does rnt receive a 6-role can enter into a

~id.

We have assumed above t"lt.at VPs are generated with empt.y 9-grids.

Half of the grid is fille3 by the percolation of the external argunent

fcan the V, and the other half is either fillt:rl. ~. the index of an

adjacent NP, or it is left open ani the VP enters into a pl'e3~.~tion

relation. In the case of a pleonastic I the verb has no external ~role to

pass onto its maximal projection. '!he question is: d~s this mean that

the VP has no 9-grid at all? RJthstein writes that all XPs are one-place

pred.icates that ItUlst be saturate3. 'Ibis \.ould mean that all XPs are

generated with [ = ] 6-grids, independent of the lexical specification of

their heads (see lligginOOtham 1984 for a diSC\lSsion of saturation of

NPs) • '!he Gennan pleonastic construction beION' 'NOuld have the given

9-grid on the VP.

(52)

CCMP

~heute kannt

CCHP'

182

Altmugh kannt has only one &-role to assign, and it assigns it

internally, the pleonastic is still allOited to saturate tile Q-grid of the

VP by direct irrlexation. Becal1.se of this relation of direct irrlexation, I

claim that tlle stlbject l:X>Sit.i.Okl is properly goveI:ned by the VP even when

it i."s not ~5signed a 9-role.

In sUll1llaIY, a VP may assign a 6-role to a Subject in t\\O ways. '!he

first involves predication coirrleY.ation. '!here are 00 aqjacency

COrrlitiOllS in t.h'l~ instance as shown in tlle case of adjuncts.

(53)

(54)

[John]i will [vp buy C\ book]i Nyl. I Wi

[JO~]i will [wP1ay the go.1le]i t.cm:>:crow [barefcx:>t]i . .NY- I vpl. Adverb Adjunct).

(54) is a case of an adjl.l11ct: which is licensed by being coirrlexed

with John, and Jo~ in turn is licensed by being <Di.rxiexe:1 with the VP, as

shown in (53).

'!he second wa.y a Vp ma.y assign a &-role is hy direct indexation wi.th

tlle 9-9Tid which ckJes require adjacellCY. '!here is no sarantic difference

obvious bet\tJeen tlle t'ltO types of 8-role assignnent. In other \\Ords, ,in

t.he Gennan senterlCes below, wet:: expect. the sanantic cont~t of the subject-

9-role to be the sam~ wh~ -her It is a~signed tl1rough predication as in

(55) or tlu:ough canplEmentat..ion as in (56).

(~5) PrOO.l.ca'tiC,1; ~ I v~[sDi.e Frau hat das Such gelesen.]

(56) C'anplanen-c.at.i.on: I NP e VP[ .~s Buell. 'hi1t. [LI die Frau t; t. gelesen ]]~ J). ~ ~ J

fle can see a case in Fnglish a.s \VeIl. In (57), the result gets its

,Ij

;*.

183

e-role directly fran the verb, whereas in (58), the 6-role is assigned via

pred.ication.

(57) We did not expect thi.~ result.(58) 1his result was unexpected..

When an NP is in a p::>sit.ion which allC7NS direct indexation with the

e-grid, i.e., a canplanent p::>sition, it is in a p::>sition of proper

government. It is t.his notion of proper government which can explain the

facts of es distribution in German arrl Yiddish. '!he gap of the enpt.y

pleonast.ic is only properly ident.i.fied. when it iJ in a canplana1.t p::>sition

of the VP. Ii:J\.lever, when \:he intervention of INFL makes this relationship

imp::>ssible, The pleonastic must be identified by the insertion of es.

C!le nore restri~tion must. put on canplanent governnent. Only

predicat-es which are themselves licensed by canplanSltat.i.on may, in turn,

license a subject through canplanenteition. A relevant example is given

below.

(59) Clifi:ard, tired. of painting, left the studio early"

, Tired of paintin;J is predicated of, am adjacent to the NP1

Clifford. 'Ih.is adjac~ of a pred.icate, h:Jw'ever, is rot sufficient to

pr.operly govern the subjE:..lCt p::>sit.i.on, since the predicate itself is not

licensed. through canplanent:ation. In an r-s-vp structure, the predicate

"VP is the canplement of INFL. It. t.hen may take .. ~he subject. as its

canp1E!Tien·~. arrl properly govern it. Canplanent governnent will be defineJ.

as belOt\':

184

(60 ) Conplanent Goverrment:

If an XI? is in the canplanent danai.n of a e-assigner, andthe e-assigner itse~f is canplanent governed, then XP iscanplenent governed •

4. 2. 2. 2 Ident.i.ficat..i.on vs. Features

'Ib provide sUpp:lrt for the claim that proper ident.ificar_ton involves

t\lA:> subparts, ident.ification of the gap am ident.i.fication of the features

of a gap, we will see belaN' examples of the four p:>ssibilities set up by

this systan. In Yidlish \\Ie fim all four p:>ssibilities since INFL carries

features for identification of all pleonastic elements6 but oot for

referential elanents. Let. us say for the time being that the IIrichnessll

of INFL 'has to do with what features it is able to carry (see Chapter 5).

Yiddish also 'has a subject p:>sition adjacent to the VP which is pro}?erly

identified am a subject p:>sition which is licensed by prErlication and,

therefore, will rnt be ident.i.fied. We ~uld exp3Ct, then, the follOYJing

distribution •

(61) Yiddi.sh ident_ificat.ion:Features Ga.p

a. + +b. +c. +·d.

pleonastic in I-S-VPpleonastic in S-I-VPreferential in I-S-VPreferential in S"'I-VP

we prec1.ict fran this table that YJe will find ga'ps in subject p:>sition

5. I assune that I' is the canplanent of CCMP and VP is the canp,lanent ofINFL, too1lgh it is l'X>t clear what it means for CQ\1P to assign a Q-role toI', or INFL to VP.

6. Whereas Ciennan INFL only identifies FS/O and not ES.

185

in Yiddish only when the subject is a pleonastic ani when the subject is

in a canplement Insit.i.on, i.e., adjacent to t.he VP. In any of the otller

three Inssibilities, \\Ie \\QuId expect to have to fill the subject

Insit_ton • '!his, we can see belC1ll, is the case.

(61) a ' • Ha.ynt iz mir kalt.today is to-me cold'I am cold tcrlay. I

b ' • Es iz m.i.r kalt haynt.

c' • Zi iz gekunen haynt.she i's cane today

I She came tcrlay.·

d l• Ha.ynt iz z.i gekumen •

A clearer example is found ttAlen one canpares dir~t versu~\ indirect

questions. Ebr Yiddish (ani Italian, but for a different analysis see

Calabresi, forthcaning), unlike Ehglish, I' is a barrier for

ident.i.fication of a gap by a l~al antecedent. '!his is clear in the case

of an indirect question.

(62 ) Ikh veya nit ver * (es) mt gekoyft dos bukh.I krow r:ot wl-o it has oought the 1::x:x:>k

I I don I t know \tJho bought the bcx:>k. I

(63 ) Ikh veys ni.t vas zi mt gekoyft t.I know not what she has bought

I I donIt krON what she oought. •

In (63) \tJe can see that \tJhen a Insition is properly ident.ified by

canplanent.ation, a gap is Inssible. HJwever, in (62) sinc:e the anbeddej S

is an NP-I-VP structure, the s(Jbject Insition is not licensed by

canplane.ntation. In this case, then, the Insition must 'be identified by a

-j

· -t 7Wha - - t- - th 1..._- fprOI'X)UI1 m-51 U. t 15 1nteres ...1I1g 1S e C1Ll1ce 0 pron:>un.

186

Es is the

prorx>un least marked for features. If it were a resunptive proroun, we

\\Ould expect it. to bear the relevant features of nunber, gender, etc. In

fact, Yiddish MS such resunpt.i.ve pro:OOlRlS as shJwn below (see I.t:Menstanm

1977 for more examples).

(64) di froyen vas mir veysn nit vas zey tuenthe \ttOt1en which 'IJe kro:l rx:>t what they do

•t.he \\Ot\en that YJe don' t know \ttlat they dr) I

What we can say, towever, in the case of (63) is that the function of

tlle prOIlOlnl ~ is simply to identi.fy the IXlsition, not to supply the

features. '!hese features will re supplied through coindexat.i.on to the

elanent in CG1P. Here 'Ne have a case of (61h) • Direct questions are

examples of (61a) since in these cases, due to rrovanent of ~IFL into a:MP,

the IOsition of the gap will re identified via canplementation. Features

will re supplied in direct questions, as in indirect que~.:..ionsI by

CX)irXiexation with the wh--...ord in CCMP.

(65) Vas hot z.i. gekoyft t?What MS she oolr:Jht t

'What did she buy?'

(66) VPI mt (*e::;) gekoyft dos bukh?\mo has it bought 'the bcok

'Who oought the bcok?'

fO'·

In (65) \'Je can see t.hat in direct questions, unlike indirect

questions, INFL must rrove into CCMP. In (66) we can see t-hat this

7. I am assuning that es in this instance is not the same as the es is lEakumt a froy·, i.e., itdoes not indicate t.hat the logical subject~s inthe VP. My reason for this is that inversion is very restricted in Yiddish(nore so than in Italian or Gennan) • I assume, then, that logicalsubjects can never appear within the VP in Yiddish.

187

ITOVanent obviates the necessity for identifying the {Xlsition of the gap

via prorx>un insertion since now the gap will be identified by being in a

canplanent relat.i.on with the VP.

In the chapters that follow, we will see further cases which separate

identification of a {Xlsition ani recOV'erability of features. Malagasy

idenT_i.fies all of its subject {Xlsit.i.ons but only has features to recover

pleonastics (see Olapter 5), the 'lbscan dialect of Italian has {Xlst-verbal

subject {Xlsition weh is identified, but it. must add features to !NFL in

order to have pro-drop. '!he rx:>rmal representation of INFL is too

i.mpJverished. In Irish, though there is pro-drop indicating that the

subject {Xlsition is properly identified, n:>t every INFL in the paradigrt1

has enough fei=itures to recover the material in the {Xlsition of subject

(Olapter 5).

4. 2. 3 Structural Goverrment

'!he main difference bet~en the not.i.on of gO\/errJnent pro{Xlsed in e1.i.s

\ttOrk ani t~t pro{Xlsed by Chansky (1981) is that governnent by

canplanentat..i.()n is defined ftmct.:tonally. '!his is even true at the level

of t.he VP 'Nhere the argunent whicl\ saturates the VP by direct indexation

into t.he 9-grid of the VP is properly governed. '!he not_ion presanted by

Chansky is StructllIal. In this section I will offe!."' argl.3l1ents against the

notion of structural goverrment, ani concltrle, as have Jaeggli (1980),

St:.aNell (1981) and Iasmk and Sa:lto (1984) that pr.Jper goverrment'must be

defined funct_i.onally. '!his section will examine both goverrmalt, and

proper goverrment as a subset of goverrment.

188

Olcmsky does I'X)t igrnre the functional flavor of goverm,ent

As is often true, the "core notion" of governnent hasclear thatatic content, but the operat.i.ve n::>tion involvesstructural cr>nfigurations generalizing the core notion. (p.163)

Note that goverrment is closely related tosubcategorization. We m.i.ght (a1m::>st) say that thesubcategorizat.i.on features of V are actually governors andthat the category V inherits governneilt fran these features.(p.l64)

A natural, if tentative conclusion, then, is that aproperly governs b only if b is in the canplement of a. (p.274)

While Clansky reccqnizes the functional definition of the core cases,

he claims that there are purely structural extensions of the lDtion. Ebr

instance ," for the puI1X)ses of case-marking v.hich, according to C1ansky I

can only happen urrler the corrlitions of goverrment, !NFL must govern the

subject. NP (67) and an EI:M verb IllUSi·. gCVet'1l the subject };Osition inside an

enbedded S (68). Fbr the explanati::>i:l of raising facts, the acccnmt in LGB

also assumes that a raisL,g verb must govern into t.he embedded S (68)"

(67) II

~NP I VP

"'---'

Olcmsky writes t"hat

(68) VP

v/~s

~/

the notion of government must meet se',eral kinds of corrlitions:( i) comi.tions on cb:>ice of govenx>r(ii) conditions on governed tenns(iii) structural conditions on the relation of governnent.

These corrlitions are clarified in the I.GB defin.i.tion of goverrnnent

given balON":

189

(69) [a governs 9 in]

[b ••• g ••• a ..• g ••• ], Where\i) a =>f(ii) Wlere b is a maximal projection,

if b daninates g, then b dan.i.nates a(iii) a c-ccmmands 9 (p. 165)

The goverrx>r must be a lexical category, and the corrlitions on the

tenus that are governed an:l the structure of the relation deperrl crucially

on the notions of maximal projection and c-canmand. A· lexical category

governs any elanent \tJhich it c-cannarrls as long as the governed elanent is

rot contained in a maximal projection wh.:lch does not also contain the

governing' elanent. Direct mention is made of governirg into a MaXimal

projection Which is ruled out by (ii). G::wernment out: df a maximal

projection will be restricted by the definition of c-cannarrl.

G:>vernment deperrls on c-cc:mnand, and, in Cllapter 2, we nave seen row

the I'Dtion of branchingness that is pror:osed in th,i"s thesis affects the

notion of c-ccmmand. We need, then, to look at t.he nor.ion of c-ccMrand

used in C1.ansky· s definition of goverrment. '!hi.s is given belCM.

(70) a c-ccmmands b if and only if( i) a does rnt contain b(ii) supp:lse that 91' ••• , % is the maximal sequence such that

(a) 9 =a.(b) gr; =a J

(e) g"7 i.mne1iate1y dan.i.nates g. +1men if d daninat..es a, theft either

(I) d daninates b, or(II) d =~ and 91 dan.inates b. (p.l66)

The way this defini,tion of c-carrnand differs fran Paun and

S}?Ortiche' s is that there is no ment_i.on of maximal projection. '!he basic

idea is that a sequence of daninatirg ncxles can be considered one

projection if t.hey match in features. Given an adjoined construct_ion as

190

the one shown bela-l, X cx:>uld c-cannard arrl therefore govern yP in

tllcmskyl s definition, even trough there is an intervening maximalI

!

projection. In Olapter 2 \tA9 have argued. against this ootion to

c-ccmnand.

(71) XP

/".........XP yP

~X zp

The question is Wheth.er or I'X)t \tJe need this notion ofc-camnand for

the issue of governnent. Before we discuris this, we must investigate the

notion of eroper governnent as a subset f")f gavernnent.

The structural rx:>tion of proper governnent hi..nges on what >fJ is in

t..he governing relation. .According to the structural definition of

goverrment., >fJ can be anythi.ng. Given the st.cuctures bela-!, we should

fim at least all the governnent relati,')ns in:1icated since 'fee fim case

assignment in every example.

(72 ) I'

NP~vp(73) CCMP'

CCMP~II

NP~

(74) VP

V~---rl~

NP

We do net f l'xJwever, want all of these governed. elanents to be also

properly governed.. Presunably, they are governed because they can be

case-marked., but rr>t all of these fOsitions allow extraction. Therefore,

oot all of these fOsit.i.ons are properly gCNerned. In oIder to dist.ingui.sh

proper gaverrment fran governnent, comi.tiona must be put on >fJ.

Basically I >P canrx:>t equal !NFL. Aithough mI"L assigns case to t11e

191

subject NP, j t does rot propArly govern it. What \tJe are doin:J with this

distinct_l.on is se,Parat.i.ng case assigners fran proper governors.

'!he ITOS\.'" obvious case where the defini.tion of c-ccmnarrl presented in

LGB (i.e., where a h~ad can c-a::mmand every elemait wit.."'1in an adjoined

structure as sh::1Nn in (71» is in the Italian or Spanish subject inversion

sentences e1at ~ will discuss in the next section. As 'Ne have briefly

explained in Olapter I, traces must be properly governed. Since

extract.ion fran a j:X)st-verbal subjec:t p:lsit.i.on in Italien shows no that-t

effects, it has rl3el1 assuned (Chansky 1981, Rizzi 1982) that the

p:lst-verbal subject is within the VP as shown belOYl.

(75) VP

~VP NP

~If, as the account presented in Rizzi (1982) suggests, the V is able

to properly govern (am therefore govern) the p:lst-verbal subject, the

extraction facts can be explained.

'Ibis proper gCNernnent by the V, hO\\ever, creates other problans. If

goverrme..'1t is purely structural, ani if the V is able to govern the

p:lst-verbal subjf9Ct,. then the V should be able "'..0 govern all of the

constituents withUl the VP. Am further, if V is a proper goverrx:>r, which

it must. be for p:lst-verbal subjects as well as for Objects, the V nust

properly govern all of the constituents within the VP.

Hc::l\..ever, we know that adjuncts in Italian and Span.i.sh are not

properly' governed. Belatl we can see the satre asynmetry bet'fleent

canplenents of the verb am adjuncts that \4Je have in En:Jlish.

(76 ) Ha t..ana:1o el alimento frior.as (35) taken the foc:rl cold

'S/he has taken the food cold.'

(77) *cCrro te pre.guntas si ha tanado el alimentob:Jw you W'jooer \\hether has (3s ) taken the food

I Ibw do you \\Ond.er whether s/he has taken the fOOt'i?'

(78) ()I~ alimento te preguntas si ha tanado friowhat foc:rl you WJooer whether has (38) taken cold

'What food do you \\Oooer \\hether s/he has taken cold?'

Given the structure belON, it is hard to see h:lw any structural

definition of the facts will predict this outcane, ever1 if we assume an

articulated VP.

192

(79) VP

/'~NP

V •••

(80) AVI Adjuncts

AV •••

In fact.. if \\e had. to predict any asynmetry, \\Ie WJuld expect V to

g~(ern the adjunct and rnt the !X>st-varbal subjec:t.

Sane questions that need to ::'9 answered are:

1) When, if ever, can a head (properly) govern into a maximal

projection?

2) When, i-F ever, can a head (properly) govern out of its maximal

projection?

3) Within the same maximal projection of the head, \tJhat can be

governed?

193

4) What elanents properly govern?

(1) As far as g~/errma1t into a maximal project.i.on is concerned, we

want to have the head of the projection governed if am only if the

maximal projection itself is governed. Bellett..i. & Rizzi (1983) argue that

the head of a maximal projection is properly governed if the maximal

project.i.on is prop3rly governed. '!his notion is needed indeJ?endent of

their dat.a for my account of Gennanic \\Ord order. I have claimed that

INFL (or CCMP) may be anptyonly if I' (or CCMP') are prop3rly governed.

'Illerefore, in one way, a maximal projection does rx>t block

goverrment. IblAever, a maximal projecT_ion will blcx:k government \Ihlen rJle

elanent to be governed is tnt a head. 'Ihis ba.sically means that a maximal

projection will be a barrier to goverrment. '!he fact t:hat the head is

goven'1ed sh:>uld foll~ fran a rx>tion of head am the percolation of

features.

mus creat..es problans for EI:M and raising if I I is a maximal

projection. Ebr these cases, I assune B::>uchard's analysis (also adopted

by Iasnik and Saito 1984) 'Nhich reduces both raising and Exceptional

Case-marking t.o instances of antecede."lt goverrment.

(2) As far as governnent outside of a maximal projection is

concerned, let us say that l'X> such goverrment is fX)ssi1?le am leave the

probl€!ll of Italian fX)st-verba.l subje:::!ts for the rocment. 'As lNe see below,

there is a di.fferent solution for this problem.

(3) Within the maximal projectj~on of the head, we do rot ",ant

194

adjuncts to be pro:f)erly governed, but we do want to have canplements

governed. 'Ih.i.s could be done structurally on an articulated brarlching

structure, but we have argued against such a structure in C'ha.pter 2. Given

r.he branchm,j ..:»trocture prop:lsed in Clapter 2, we must functionally define

(by creating danains) the consr_i.tuents that are to be governed.

'!he structural definition of gO\Ternment is both tcx:> strong and too

~. Using the definition of c-carrnarrl of k>un ani Sportic1:le, adjuncts

will be properly governeCl since they are in the same maximal projection as

V, a pro:f)er govertr>r. !:bst-verbal NPs ill Italian, hJwever, will rot be

properly governed since they are not in the same maximal projection as a

proper goveIl'X)r. '!he problan of adjuncts ma.y be solved by restricting the

notion of c-a:mmand so that only canplE!nmts a-.:e c-cxmnanded. by the verb.

'Ibis does rnt, hJwever, solve the problan of p:lst-verbal subjects. '!he

problen of mst-'.7PIbal subj~ts may be solved by extending the notion of

c-carrnarrl to chain c-cannand so that the verb will row c-cannand the

mat-verbal subject. 'Ibis, hO\\ever, will not solve the problen of the

adjuncts. Either problan IllCly be solved by either restricting or lcx:>sening

the notion of goverrment, but iT. is not obvious how a change in the

structural defini.tion of governnent can account for 00th problans at

once.

(4) A final question I°ana..inS: can only heads be gaverrDrs? mus is a

hold over fran the structural defini.tion of goverrment. '!he question is,

t.hen, whether phrasal nc:xies can have carrplanents. o:>viously I we aSSlnne

that phrasal ncrles can have canplanents since we claim that VP can t?~ce an

[NP, S] canplement.

195

'Ib sunnarize the argunents for a functional definition rather that a

stroctural definit.i.on of proper gO\Te.rTmlmt:

Within the VP, if 'lie want to define proper governnent s·tructurally I

we must. have (i) an articulated branchi.ng system different.i.a't.i.ng 'bet\\een

canplanents am adjuncts, and (ii) Peinhart I S definition of c-cannand so

that the V will on!y e-caunand its canplanents.8

(81) v..

A~.A l'1L.lJuncts.

V Canplenents

'Ihis raises problans if 'Ne asslIlle that (i) aCljmcts must be sisters

of the constituents they are predicated of and (ii) if tNe VIallt a e-cannarrl

relationship for anaphors in sentences such as

(82) I canforted Jeri prooo of herself for havirg done so well.

Also, it is hard to see h::1N such a structural definition could extend

to inclooe fOst-verbal subjects in Italian witoout inclt:rling adjuncts.

(83) V"

VIl~NP

VI ~juncts~

V eatlplenents

Finally, if structurally defined, all heads will govern all maximal.

projections \tJhi.ch they e-cannarrl but a distinction must be made 'between

s. A version of Reinhart.· s definition of c-cannand must be adopted wherean intransitive verb, tootgh it has no canplements, \f.Ould oot C-C011l1arrl anadjunct.

196

goverrment am proper goverrment. '!he contrast is given belOfi.

(84 ) proper goverrment vs. government

a. VP

~V NP

b. I'

~I

We must, t..hen, stipulate t11at sane it:e.ms are proper governors (V),

sane are just governors (INFL[+tense] ) •

I contend that t..he true split is bet'l.een INFL and V because the

subject NP is rnt a canplement of INFL while the object NP is the

canplement of V. '!he distinct.:lon bet\Een INFL[+tense] and INFL[-tense] has

rnthi.ng to do with goverrment I rather it has to do with case ( see

Boochard) •

AltOOugh I agree with 01ansky as far as the co.t-e cases of proper

goverrment are corcernErl, I cla:h-n tlat the extension ~s alon:3' functional

lines, not st.~tural lines. 'lhe object NP is the canplement of the "v and

thert.!by is properly governErl. '!he cases of proper goverrment that I want

to account for are belaA'.

(85)

(87)

~V~pp

C

(86)

(88)

I

I'

(89) V'

v

197

vs.

(90) I'

~VP(91)

v pp M~J

In all cases, the governed elements are canplanalts of the governors,

or are the heads of the canplements.

4. 3 EmPirical Consequences

If VI? goverrment has sane conceptual content, we \\Ould expect t.o find

effects in languages where SlJbjectS are fourrl adjacent to VPs. In this

sect-_i.on I investigate scme other languages, in particular Italian,

ChanDrro, am Irish.

4.3.1 It:alian

In Italian, subjects may appear either before or after the verb

phrase.

(92)

(93)

,Anna parleraAnna will talk

'Anna will talk

Parlera Anna.

'lhese t\f.O p:Jsitions, .h<:J\\ever, aheM differ.ent syntactic effects.

Rizzi(1982) has rnterl that extraction of subject witlDut. ~ violations is

p:Jssible only fran the J l~'_-VP p:Jsit.i.on. 'Ih.i.s suggests that t.he p:Jst-VP

p:Jsition is. properly gO\1erned.

198

Given a notion of VP government, this is exactly what we \\QuId

expect. let us assume that the structure for (92) and (93) are (94) and

(95) respectively.

(94) II

~N I I Vi

IV

(95)

~/1r ..... VI I

iV

In (94), the subject NP is in a predication relation to the VP, 'Nhile in

(95), the subject NP is in a canplement relation to the VP. It is the

structure in (95) that \tJe \\Ould expect to sb:.lw no that-t effects, as is

the case9 •

'Ihe i.mp:>rtant addition to the theory that allONS us this analysis is

that languages 'Nhich allC1fll subject NPs to directly index into the Q-grid

of the VP, also allC7.t1 proper government of this D-EXJsition. 'Ihe Italian

extraction facts follow directly since the EXJst-VP Subject EXJsition is

properly govenled by the VP While the pre-VP subject EXJsition is not.

An alternative analysis (Chansky 1981, Rizzi 1982) is that EXJst-VP

NPs are actually in the VP as sb=1Nn below.

(96)

9. Constituents may ,appear bet\\een the VP and the EXJst-verbal subject,h:1Never, like within the VP in French, these do not disturb the adjacencyrelation. Q1ly!NFL as the head of SI will prevent direct indexation.

199

The subject NP is lowered and adjoined to the VP leaving an enpty

t · th 11T"l ub· .. 10 rn... • d f· eelca egory Ul e pre-vr S ]ect FOSl.t~on. .L.llen government 1.5 e ill

structurally. Using O1.ornsky·s definition of goverrunent and c-eamnand, V

will c-ccmnand the FOst-verbal subject am, therefore, will properly

govern it. '!his analysis brings with it the problems discussed al::xJve. If

a V can govern outside its maxi.mal prc,jection, then it certainly will

govern everything within its maximal projection. we do not want this

result because of adjlUlcts which we do not want to say are properly

governed.

There is a further curiosity concerning this structure. A striking

difference bettNeen t-.l1e structure I FOsit for FOst-verbal subject

<.X>nstructions, and the one that Rizzi J:X>sits is that I have no pre-verbal

subject FOsitior~ while Rizzi does. It is hard to test this in a language

where pleonastics are enpt.y. What is r€!11Clrkable is that subject inversion

o:curs only in languages with enpt.y pleonastics.

It has been suggested (Jaspers forthcaning) that. Gennan is a case of

a language which has subject inversion with overt pleonastics. We have

seen examples of this sort of structure, and 110re are given bel<Jlll.

(97) Es wurde eirle Frau gesehenwas a woman seen

'A wcman was seen.'

(98) Es stiess ihn ein Soldat von der Brtickepushed him a soldier fran the bridge

I A soldier pushed him fran the bridge.'

10. I use the In and I' structure in these examples as Cnnsky and Rizzido, trough it is not crucial to sh::1N' the difference betlNeen the t\«Janalyses.

200

'1b:>ugh this pleonastic is underlyingly empty, when it is not in a

position of proper govemment, it is filled with es, indicating that a

separate subject position exists. I contend, 'ho\tIever, that there is a

difference bet'Neen the German l..."Onstruction and the Italian construction.

Both constructions have certaj.n restrictions on t..'1em, but these

restrictions are different. '!he German C'Onstructions appear to be similar

to English there-constructiorlS and the restrictions are similar. AItb::>ugh

definite subjects are allONed (see (97) and (98», they are not preferred,

and the constructions 'I.Orsen with proper names (99) and are canpletely out

with proIlOlDls (100).

(98) *Es sind die Kinder gekatmenare the children cane

I '!he children cane. I

(99) *Es ist Hans gekamlen

(100 ) *Es ist er gekatmen.

I will maintain, here, that the German construction and the Italian

construction are not. the sane. In the case of German, the logical subject

is in the VP and the syntactic subject position is empty. In Italian, the

inverted subject is still [NP,S] and there is no pre-verbal subject

position.

4. 3. 2 INFI.rl-V Peripheral lan:JUB.ges: C1arrorro

'!here seans to be a generalization aoout INFL+V peripheral languages

which is that subject positions appear to be properly governErl (see h)t.m

201

1983.for a different analysis for Chinese). 'lhis has been ,p:>inted out for

verb f.i.nal langua.ges such as arinese (Huang 1982) and Japanese (saito

1984). In these languages, where there is no overt rrovanent, the facts are

lX>t as striking as in INFLW init.i.al langua.ges were there is

wh-fronting. Below I di.scuss Ch.anorro, a V-S-o lan:JUage, am Malagasy, a

V-o-S langua.ge.

Chung (1983) describes a problem presented by C1aIoorro, a \150

lan:JUage. Ebr sane phenanena in ChanDrro, it appears that \tJe sh:>uld ,p:>sit

a flat strocture, and for others, a hierarchical structure. A flat

structure is preferable when one considers \\Oro order, extraction facts,

and the EI::P. Since <J1aIoorro is V-S-O, it is not easy to see b::)w one could

argue for a VP since it '#tOuld have to be discontinlDus. cne might assune,

then, that OlaIoorro has a flat structure as sho\-Jn below (see Barss 1983

for a different analysis).

(101) S

~NP2 NPl=Subj~t NP2=Object

Als:>, since C1anorro does not show that-t effects, we may have

aIDther reason to assune a flat structure.

(102) Hay:i. na palao'an ma'a'nao-11U.1 [na u-kahatwho? L 'ttOIlaI1 INFL(WH)+afraid-your [CGiP INFL(3s)-lift

t esti na dangkulu-n kahunt thi.s L big-L box

* 'Which girl are you afrai.d that t will pick up this big box?'

If the subject NP and the object NP '/Jere 00th in the same relation to

the V, we \«)u1d rot expect than to behave asyrranetrically in relation to

extraction. Ihth \tJJuld 'be properly governed by the V.

202

A problan arises, b::1Never, when one lcx:>ks at other gOV'errment facts

such as case assignnent or binding. In both instances, subject and object

behave di.fferently fran one arnther. SUbject is assigned naninative case,

ooject accusative case. PRO app9ars only in subjec:t txJsit.ion suggesting

tha.t this fOsition is ungoverned. Rrlsirg only occurs fran subject

{Xlsit.i.on showing that a matrix verb can govern only the subject {Xlsition

of an arbedded sentence, rx:>t the object fOsition.

(103) PROa. Na IInClgUf [mal mai PRO gi bisikletaJ

fun INF-(s)-ride PRO IDe bicycle•It •s fun PRO to ride on bicycles. I

b. *Ma'a'nao si Rita [um-insutta (i mediku) PRO]INFL(s)-afraid lhn Rita INF-insult (the doctor) PRO

*'Rita is afraid (for the doctor) to insult PRO. I

(104) Rrlsinga. Ha-tutuhun si JOse [fu-un-a'maolik i karet nigap]

!NFL( 3s )-begin Unm Jose INF--repair the car yesterdayI Jose began t to repair the car yesterday.'

Clung argues, then, tha.t for governnent as it is relevant for case

and binding, we need a hierarchical structure with a VP.

Ckle result of Clung's analysis is tha.t ChanDrro has t\\O coex.i.sting

stnrtures, one (the flat structure) that she suggests is language

specific, a.rrl the other (the hierarchical structural tha.t she suggests is

universal. Mother result is tha.t t.b:>ugh there is only one definition of

governnent, it is used. in t'_'JO different danains, t\ttO danains which

coincide with the t..-.o coexisting stroctures. Governnent as it concerns

t-races of extraction applies to the language specific flat structure, a.rrl

goverrment~ as it concerns the non-suverrment of PRO a.rrl the government

needed for case-assigrment, applies to the uni""ersal hierarchical

203

structure.

I claim that there is only one strl.lC:ture for C1.aIrorro. fur thi.s

analysis I rely on a variation of Elroms IS (1980) am Sproat IS (1983a., b)

analysis of V-S-O languages, Bo~hard's (1982) analysis for PRO and

raising, am the notion of VP goverrment.

aroms (1980) accounts for VSO langua.ges by IX'sit.i.ng a V-fronting

rule fran an SVO st..ructure. Sproat (1983) develops this analysis by

asstming that this V-nevanent is actually a tOOVE!nent of !NFL into C'CMP.

IkJth of these analyses, then, will have VPs at every level of

representation (see (106». Unlike either of these accounts, I will

assune that the INFL is sentence initial at d-structure. In a frame\leOrk

\fkdch incltrles a t-xipa.rtite I I (see section 3.6) I an I-NP-VP 'NOrd order is

p:>ssible. As I have claimed in my analysis of Gennan, !NFL must be filled

._..... · · 1 ~11Wilen 1 t J.S IDt proper y governt:U •

In Irish ~ can see cases were an analytic verb splits over t\\O

p:>sitions (Vl-S-V2-o). In ChanDrro, bJwever, since the verb is

non-analytic, the \fAlole foITn llD'Jes to sentence initial fX)sition. 'nlis

accounts for the vso appear~.

G.i.ven Ebuchard IS framert.Ork as described in C1apter 1, the

11. I assune that the proJ?E!r goverrment of CGtP' anP I' differs franlarguage to larguage. We can see that in Yiddis11, INFL must tOOVe intoCGiP even in embedded clauses. 'lhis suggests that the head of CGtP is notproperly governErl by the matrix verb as it is in Gennan. Presunably, Vrroves into rnFL even in clauses with an overt CGtP in VSO langua.ges. '!hisis an avenue for further research.

204

di.stribution of PRO in ChanDrro does oot COtlC:enl us. Whether or n:>t. it is

governed is irrelevant. Ib~ver, the binding of this PRO by an element in

t-he rcatrix S ie of c:oncern, just as the birrlirv:J of the subject 1X>sition by

an elanent in the matrix S in the raising cases is a concern. If Charrorro

llCls a flat st..r\X:ture, Wny can only the subject beat· this relatiorl to the

matrix 51

What I propJse is that \eJe adopt tl1e suggested variat.i.on of Elnorrls' s

am Sprcat I s VP structure thereby avoidirg the problans of birrlirg am

case assigrment ment.i.oned above. NJw only the problen of proper

governnent of the gap in subject:. pJsition ranains. We can see in the

stru::ture below, 'hc:J\..ever, that VP c;overrment will insure that subject

pJsition will al'Nays be properly gOVenled.

(105) S

I+V~VP1 ~

I NP

t.1

'!he subject NP is adjacent to the VP and a gap in this IX'sit.i.on will

be properly identified. '!hus with Sprcat' s str\X:ture, Ih~hard's birrlirv:JI

and VP goverrment, we no longer have to IX'sit t\\O co-ex.isting str~tures

for ChanDrro, rnr t\\O crucially different uses of goverrment.

we may \eIlt to say that it is the lOOVEment of the verb that allows

extraction fran the subject pJsition. '!he propJsal \\Ould be that the V in

CCMP 'NOuld prorer1y govern the subject pJsit.i.on under a structural

definition of gc.wernnent. 'Ibis solution is n:>t available to us because

205

the subject is l'X)t a canplanent of the V, therefore canrnt be properly

governed by the V. A way of distingui.shing bet~en the t\\C> analyses \YOuld

be with a larguage which had no verb movenent, yet still s~ rx:> that-t

effects.

Malagasy, a V-O-S langua.ge, stJPlX>rts the claim that VP government can

ao:ount for apparent ~ violations. Malagasy is a rigid \\Ord order,

non-pro drop V-o-S langua.ge, which, like Charoc>rro, allC)fNS extract.i.on of

subjects out of erbedded clauses.

(106) heverin-dPabe fa efa nanasa ny lamba. R3k0tobelieved-by R. that alrecrly 'lash-p3.st the clothes R.

I It is believed by Rabe that Ie.koto has already \\ashed the clothes. I

(107) lza l'XJ heverin-dPabe fa efa nanasa ny laniba\-Iho INK believed-by R. that already wash-p3.st the clothes

.Iwtx> is it believed by Pabe that t has already '£shed the clothes. I

Ma.lagasy presents no problans conc:erni.ng d-str\X:ture \\Ord order. We

nay assune that it is I-V-o-S ani that the existence of a verb phrase will

account for PRO and rai.sing fran subject fX)sit:ion just as in Olamorro.

And once again we can appeal to VP 9O'lerrment to account for the lack of

that-t effects. '!he fact that Malagasy, a \QS language, \tJOrks in a way

similar to Ol.artDrro irrlicates that it is not the 1lUVement of the verb

which accounts for the proper goverrment of the subject. !Osition since in

Malagasy 'Ne have no reason to assune any V-rcovanent.

Evidence fran Mcl1agasy also weakens arDther way of accolmtirg for the

ex-r.raction facts in Italian. A ne\lJE!r prOfX)sal (see Slito 1983b who

credits Jaeggli per personal camtllntcation) is that the anpty pre-verbal.

subject fX)sition antecErlent governs fX)st-verbal gap accomt.i.ng for the

206

extraction facts.

(108)~

pro. I VP

1 ~.

~ 1

This, ho\t.ever, does not account for Malagasy Which certainly has the

urxlerlying 'ItOrd order of V-Q-S. Because of the similarity of \\Ord order

in I,X)st-verbal subj~t constructions in Italian am regular Ma.lagasy

constructions, one m.i.ght suspect that a similarity in extraction facts

falls out fran this similarity in \fJOrd order. In fact, I take thi.s one

step further. I..ar:guages that are the mirror image of this \t,Ord order,

i.e. S-Q-V-I should ala:> sllo\tl the same effect, as they do. '!he fact that

\'tOrd order facts ani extraction facts fall into such a generalization,

then, is no accident.

4. 3. 3 Irish ani Null Subject Iarguages

It has, in the past, been claimEd (Jaeggli 1980, O1ansky 1981, but

see Rizz:l 1982, Safir 1982) that subject ixwersion aln pro-drop are part

of one pherx:menon. Irish is an interesting c:ounter-example to this

claim. Irish is an 1-6-VP language tNhich does lDt have a productive use

of subject inversion, yet it appears to be pro-drop (MjCloskey & Hale, to

appear) •

(109) ra gcuirfea isteach ar an phost sin gheobhfa eif put(<l:lNDIT 52) in on that jci:>. get(CCNDIT 52) it

•If you applied for that job, you \«>uld get it.'

207

I claim that the only reason that pro-drop seans to co-occur with.

Subject-inversion is that JOOst pro-drop languages being sttrlied are

S-I-V-o Iarguages am S-I-V-o Iarguages witb::>ut subject-inversioIl do oot

have a properly governed subject {X)sit:lon. Without such a {X)sit..i.on,

pro-drop \\QuId leave an unidenti.fied gap. If an S-I-V-o larguage has an

I-V-o-S variant, ho.-.ever I there is the {X)ssibility of having an identified

gap in subject {X)sition am features can be supplied to the gap by INFL.

Irish, being I-S-o-V, need rx:>t undergo subject inversion in order to

have an ident.i.fied gap. Subject l,X)sition is adjacent to the vp am in a

p:>sit.i.on of direct e--assigrment.

It has been aSSLmed that pro need rDt be properly gOlerned. In fact,

originally it was assuned t-..hclt the empty category in pro-drop languages

\eS PRO (see Chansky 1981) am therefore n:>t g0\7erned at all. 'lhis

involVED a proI:Dsal that lble R, a rule \thi.ch l~red INFL onto the verb,

applied in the syntax in pro-drop larguages. 'lhis create:! the structure

belCJW ~.:tch accounted for why PRO was alleJAed to appear in subject

l,X)sition since this l,X)sition \fIaS 1'X) lOB3eI: gOV'erned by INFL.

(110) II

NP~VP~

V+I NP

'Ibis analysis also acco\Dted for the p:>stp:>sing of subjects since

these subjects lNere assuned to be iIi VP. If INFL lawers, it reM can assign

case to an NP in the VP.

(111) . II

~t. t. VP

J 1 ~

VP NP.~ J

V+I. NPJ.

208

Folloong the direction of Fol£hard, I claim that all empty

categories have to be ident.:i.fie:1, which for my analysis, means properly

governed if there is no local antecedent. '!he exception is arbitrary PRO

which is proncminal am therefore free (see Ebuchard 1982 for details).

'!he empty subject in a pro-drop language, then, must be properly gOV'erned,

i .e • in a canplanent };Dsition •

'1his is a change fran W1a.t is currently asslI1lErl. I am pro};Dsing that

the empty subject is actually in the };Dstverbal pJs:ttion in Italian,

wereas Rizzi assunes that pro is in the preverbal };Dsition. His argunent

.is based on evidence fran the 'lbscan dialect of It-alian. If the stlije:t

is preverbal, there is agreement on the verb. Ibweve.l7, if the subject is

};Dstverbal, there is no agreement.

(112) Wi. gllha parlatoI He sp:>ke. I

La llha parlatoI She sp:>ke. I

(113)

loro l'hanrx> parlatol'1hey s};Dke .. I

GI'ha parlato f~ I(10m)

l,f the subj~t prolDun is droR?E!d, the agreenent has to appear on the

verb suggestirg that the empty sulc>ject is pre-verbal.

209

(114) Gl I ha parlato. (= I Ii:! sp:>ke. I )

L lha parlato.L I hanrx> parlato.

AsSlITLing, however, that 'both the p:>SitiOl1 of the' gap and the contents

of the gap must be ident.:tfied, 1Ne can say that the enpty p:>sition is

p:>st-verba.l but that agreenent must appear in order to identify the

C'Olltents of the gap. 'Ibis, then, is rx>t necessarily an argunent that the

enpt.y subject J:X)sition has to be pre-verbal.

4.4 Revision of the CEO

we have seen argunents above fran Italian, <J1aIoorro, and Irish that

the [NP,S] p:>sition ma.y be properly governed. In this section, I argue

that proper governnent of the subject p:>sition, ho\e.ever, does not mirror

exactly the proper governnent of the object p:>sition12• '1hJlJ3'h subject

p:>sit.i.on appears to be properly gcwerned for the EI:P, for the CEO the

subject p:>sition does rx:>t appear to be pzoperly gO\1erned. Ebr ttds reason

we need a distin.ct.i.on bet\teen canplemE!lt governnent, which tray be

governnent by an XI? or an >fJ, and lexical govetnnent, whi.ch is governnent

only by an >f.

In this section we will see that in Spanish, C1.inese, and We!lsh,

12. '!his first came to my attention throU3h Pap:>p:>rt (1984a). Iap:>p:>rtargues that extrap:>sition is Inssible out of properly governed p:>f3itionswithin the VP, but r¥)t properly governed subject p:>SitiOllS. f.Or thisreaaon, she distinguishes bet'Neell proJ?E!r gO'lernnent (canplanent goverrment.in this thesis) and T-governnent (lexi~al goverrmE!lt in this t:lesis).

210

either the lack of that-t effects or the existence of pro-drop argue that

the subject p:>sition is properly governed. '!he fact that the subject is

not a p:>ssible extraction danain in tenns o~ Huang's CEO I b:>wever I argues

that this p:>sition is not properly governed.

4.4.1 Spanish

We can see in the Spanish example belOll that constituents in subject

p::>sition act as islands to extraction in the same way that adjuncts do.

(115) Juan escribio tres libros a:>bre la literatura hispano-americanaJohn writes three 1:xx>ks aJ:x:)ut the literature hispano-american

'John writes three books about hispano-american literature. I

(116) Sobre que excribio Juan tres libros t?About \that writes John three 1:xx>ks t

I AOOut what does John write three 1:xx>ks?'

(117) Tres 1ibros sabre la literatura hispano-americanathree lxx:>ks about the Iiterature hispano-american

provocaron muchas discusionesprovoked much - discussion

I Three 1:xx>ks about hispano-american literature provoked much discussion. I

(118) *Sobre que provocaron tres libros muchas discusiones?*I AOOut \tJhat did three 1:xX'Jks provoke much discussion? I

(119) Provc::x:aron muchas discusiones tres libros sabre la literaturahispano-americana •

(120) *Sobre que provocaron muchas discusiones tres libros t?*I Al:xJut what did three 1:xX'Jk t provoke mu~h discussion? I

In (116) above 'Ne see a ca~e Where extraction is p:>ssible out of an

NP in object p:>sition. In (118), Where extraction occurs Ollt of the

preverbal subject, the string is UBJramnatical. '!his is tnt surprising

since pre-verbal subjects are not properly governed and this CXJnstruction

211

slDuld" be ruled out by the CED. Where extraction occurs out of a

p:Jst-verbal Subject as in (119), ho\-.ever, we m.i.ght eXp3Ct different

results. '!his is the p:Jsition which is argued in the literature as being

properly governed. '!he CEO 'NOuld then predict thi.s as a possible

extraction site. Yet \fJe can see in (120) above that this is rnt the

case.

4.4.2 Chinese

We will see row t.hat the ~e distinct.i.on which held for spanish also

holds for Chinese. '!he subject p:Jsition appears to 'be properly gOJerned,

yet elements Ca.nrDt 'be extracted out: of a constituent in the subject

p:Jsition at s-str~ure.

I have argued in Olapter 2 that Orlnese is an S-VP-I langua.ge. Given

this \\Ord' order am the lX)tion of VP goverrment, 'tte 'ttOuld predict that

Clinese \tJOuld have I¥) that-t effects. B.1ang (1982) argU3s, in fact, that

this is the case. His examples are given 'bel~ (p.479, ex. (61».

(121) zhejian 8hi [5' gen [8 shei lai ]] zui. you guanxithi.s matter wi.th 'Nbc cane lOOst have relation

'Who is the person x such that this ma.tter has ItDst relationto do wi.th x t s caning? t

As Il1ang p:Jints out, the trace of the LF rrovanent of shei 'who'1

appears to be properly governed since the structure is not ruled out. He'

accolm~s for the lack of a subject/object asymnetry in Cl.L~ese by claimirg

that !NFL may t>e a proper goverrxJr of the subject posit.i.on. '!his solution

is IDt available for the analysis presented in thi.s thesis since pro,Per

212

goverrment holds only of c:onst_i.tuents arrl their canplenents. 'Ihe

<:'allplanent of I is VP not the subje::t NP. M:>re imp:>rtan.tly; we do not need

this solution since the adjac~ of the VP to the subject NP lXlsition

predicts tha.t the subject will be properly governed independently.

'!he problan arises, lDw'ever, when the subject lXlsition is tested as

an extraction danain. Ibang' s CEO would predict that extraction fran a

subject consti.tuent sb::>uld be lXlssible since this lXlsition is properly

governed. 'Ihe exanple bel.:7tl argues otherwise •

.(122) ??Neige ren., Lisi da-le t. sm \\0 hen bU3aoxingtha.t. man1 Lisi hit-asp ]. make I very unhappy

* I that man, [that Lisi hit t] merle me very unhappy •

Like Spanish, C1inese ha.s a subject lXlsition which appearss properly

gO\Terned for B:P but not for CEO.

4.4.3 welsh

Finally, in welsh, Sproat (1983b) gives evidence that subjects are

rx:rt:. lXlssible extraction danains. His examples are of parasitic gaps.

Assuning that the subject/ooject asynmetry in parasit.:ic gaps is due to t..he

~ (I<ayne 1983), Sproat' s examples argue that the subject 'p:>sition· is oot

properly governed.

(123) a. ?I>jma'r llyfrau. [S. o. [S-~ brynasant:. l1'i,y e. [s heb wybcXihere-the books 1 ]. PRT theY-OOU3ht tlley 1. wi ut know

[5 as byddai rha.id idynt. [s PRO. ddarllen e. ]]]]]if w=>uld-be necessity to-th*" J read 1

? I Here are the books. 'lhi.ch they botght e. witb:>ut knowin whetherthey had to reaA e. • 1.

].

213

b. *Dyma.'r llyfrau. [5' o. [s a brynasant hwy e. [5 heb wybod). ~ 1

[s os byddai [s PRO darllen e.] yn syniad da ]]]]). PRT idea geed

* I Here are the 1xx:>ks. Which they ocn.ght e. \tlitoout krx:>wirg whetherreadin:J e. \\QuIa 'be a gccd idea. I l.

:L

Since welsh is, in my accomt an I-S-VP langua.ge like Irish, we \\Ould

expect the subject p:>sition to 'be properly governed. Further, since Welsh

h~s si.m.Uar pro-drop effects, this eXI;eCtat.i.on is realized (exx. fran

M=Closkey & Hale to appear).

(124) a. ~lais efsee(~T 51) him

'I saw him.'

b. gtNelais i efsee(PAST 51) I himII saw him.'

4.4.4 CED-revised

It. appears, then, that filang's CEO must be revised. '!he

generalization is I'Dt that properly governed consti.tuer.-ts are lX)ssible

extraction sites. we have seerl above that in ~ish, ari.nese, and Welsh,

were subjects are properly governed, rnthing may be extracted fran within

t.he subject lX)sition. '!he CEO must be restrict.ed to constituents Which

are lexically governed, . i .e. properly governed by a >f category as

13opp:>sed to an XP category •

13. '!his revision of the CEO will be changed again slightly in Olapter s.

214

(125) OJrrlition' on Extraction Ihnain (revised):

A phrase A may be extracted out of a danai.n B only if B islexically governed.

4. 5 OJnclusion

In this chapter I have argued that t'ttO aspects of missing linguistic

material must be ident:lfiedi p:Jsition am content. 'Ihe p:Jsition of the

gap will alw:lYS be clear if it is in sane direct requirement relation to

arx>ther elanent. I have argued t'hc:tt there are t\«> \tI3.ys requiranent can

occur: (i) a verb can require an internal argut\ent and (ii) a VP can

require a syntactically external cu:gunent for PUIp:)ses of saturation as

claimerl in IOthstein (1983). If these needs are satisfied wi..thin a

certain direct: o-marking danain, the gaps will be clearly irrlicated. 'Ihis

is al\\a.YS the case with internal cu:gunE!lts. With external argunents, the

requiranent of an external cu:gunent may be (in sane languages like

Italian) or must be (in sane languages like Ehglish) satisfied indirectly

by ex>irrlexation. It also may be (in cases like Italian), or must be (in

cases like ClaIoorro) directly satisfied. In the case of direct

subcategorizat.ion, the gap will be properly identified, or properly

governei. rnris relation explains different phenanena in different

languages slX:h as the di.stribution of anpty pleonastics in Gennan am

Yi.ddish, the extraction facts in Italian, absence of that-t effects in

O1anono am Mal:lgasy, am the ex.i.stexx:e of pro-drop witoout free subject .

inversion. in. Irish. Witmut a notion of VP governnent, this range of

phenanena renain as separate problems with idiosyncratic solutions.

215

216

Qlapter 5

This chapter examines constructions \tJhich use pleonastic elanents. I

begin by claiming that there is a certain hierarchy of ex>nstructions \tklich

require pleonastic elanents am then I sh:Jw \tJhich pleonastics appear in

\tJhich constructions in a number of different langua.ges. 'Ihe clustering of

pleonastics coosen by these various larguages offer further CDnfinnation

for the hi.erarchy itself.

Further investigation of the inventory of pleonastics offers us

different cx>ntrasts. One involves the split between lexical pleonastics

and anpty pleonastics. I pr<JIX>se an accotmt of this split \tJhich involves

a feature specification of INFL. This gives a clearer accotmt of the

intuitive n:>tion of the "richness" of inflection wch can extend to the

explanation of the pro-drop phenanerx:>n.

A second split WUch occurs arrcng the languages studied is

exEmElified in Erglish by the use of it versus there. As \aJe have seen in

Chapter 3, Genna.n has a sim.Uar split between ES am FS/O. I will prop:>se

that these t'NO types of pleonastics, the it-type (I-type) am the

there-type (T-type) I also differ in their features. I-type pleonastics

are a spell-out of the features [+#], [ +case] I \tJhile T-type ple:>nastics are

217

~i.mI?ly a spell-out of [+case].

Finally I will review sane previous analyses of the use of pleonastic

elements.. '1b:)ugh these are sufficient for the larguages for which they

1Nere proposed, I will argue that a new analysis should be sought 'lihich can

be used cross-lirguistically. This represents \\Ork in prcgress. A

solution is sketched for the split bet\¥een I-type and T-type pleonastics

in Gennan am n.ttch versus Eh31ish, am problems are raised for future

research.

s. 1 The hierarchy

BelOli is the hierarchy WUch I propose for constructions containing

pleonastic elements. I will discuss each of the splits separately and

give reasons for the ranking using evidence fran the langta.ges being

studied am fran stages of larguage chan:Je. Much of this discussion will

be at T.he level of specul~tion but will be sUPPJrted later by the

clusterirg of c:ross-lirgui.stic facts.

218

It will eat anything in sight.It rained all day.It is tcxJ wirrly to sail.It is clear that she will do it.It is believed that she will do it.It seans that she will do it.'!here 'liaS fourrl urrler the tree a great treasure.'!here appeared out of the ShadaNS a large dog.

(1) I. Referential :II. Weather V:

Wea.ther Mj:III. V-Mj-S':IV. V(pa.ssive)-5' :

V(rai.sing)-S I:v. V(pa.ssive)-NP:

V(tmaccusative)-NP:

VI. V(pa.ssive) ••• : Heute wurde 0 getanzt. (Gennan)Tcrlay was - danced'It \\laS danced tOOay. '

V(unaccusative) ••• : Chua.i.gh 0 de sholas an lae (Irish)went - of light the day

'O:lylight faded. I

5. 1.1 Justification for the hierarchy

Referential~ the rest (I). '!his split bet\\leell referential NPs

and ron-referential NPs is a.lnost too ciJvious to mention. It will,

lnwever, disti.n:Ju-i.sh Italian fran Yiciiish, am further distirguish Italian

and Yiddish fran German. Italian all<:7h'S referential prOn:>UIlS to drop (2),

\tIhile Yiddish does rnt (3). Fbwever, Yiddish will require all types of

pleonastics to drop (4), \thi.le Gennan all<:7h'S only T-type pleonastic to

drop (5). In other \\lOrds, \fJe already see laD3Uages dividing into three

groups.

(2) Arriva.arrive (he/she/it)

I He/she/i t arrives. I

(3 ) Haynt rot *(es) alts gegesn.Today has it all eatenlIt has eaten everything else. I

(4) Haynt geyt (*es) a regn.Tcday goes rain

, It I S raining today.'

219

(5) a. Heute regnet *(es).Today rains

I It •s rai.ni.n; tcrlay _ I

b. Heute sind (*es) zwei Kinder gekarmenr..crlay are t\\1O children cane

•Today there came t"-O children _•

Italian, then, allCIWS "a 0 referential, Yiddish does rot allow a 0

referential, but does all~ a 0 T-type and ~-type pleonastic, while Gennan

only allCIWS a 0 T-type pleonastic.

weather Predicates (II)

(1) Nea.t:her V VB. weather Mj- '!he reason for includL.ig the break

between tNeather V c.'Onstructions am. weather Mj constructions within

category II canes fran historical evidence. N:> language in our sample

makes a division between these t\t.O cx:>nstructions, but I include both to

allC7#1 for further specification. Haiman (1974 p.102-103) writes that

there 'NaB a stage of Gennan where weather verbs required overt pleonastics

while \tJeather adjectives did rx>t.

Even in CIiG, impersonal VERBAL predicates requiredimpersonal subjects... This continues to 'be the case today

'Ibis \t«lS rnt true of impersonal adjectival and rxm.:i.nalpredicates •

(6) Ube tag ist, licht istwhen day is light is

'When it is day, it is light.'

(2) Nea.t:her pred.icat:es va. rest. '!he split bet1Neell the pleonasticsI

which appear with weather predicates am. toose that appear in other

constructions has been considered the split bet'Neen quasi-argunents and

non-argunents (IGB, p. 325) • Evidence for such a split has cane fran

220

ex>ntrol C.'Onstructions wh..i.ch seem fOssible in the case of quasi-arguments,

but not in the case of non-argunents. 'lhese examples are given below.

(7) It rained witbJut PRO soowing for days and days.(8) *It 'NaS clear tl1at she could do it without PRO being obvious.

'!he analysis of these facts is rnt so clear, b:Jwever, since the

following example seems better than (a).

(9) It \alaS clear witbJut PRO being obvious that she could do it.

This may be a split between III and IV of the hierarchyl.

(i) *It is obvious without seeming tl1at S I •

(ii) *It \tIaS believed witbJut being expected that S·.

Be-Adj ~ rest (III). Evidence for the split bet\tJeen

cx:>pU1a-adjective constructions and the constructions belC711 it on the

hierarchy canes fran IA.1tch which uses bet for the be-Mj cx:>nsi:ruction and

all the C'Onstructions higher in the hierarchy while using ~ (at least

optionally) for all the constructions lower in the hierarchy. An example

is given below.

(10) *Er is dui.delijk dat Eric zijn huiswerk niet heeft ganaaktthere is clear t'hat Eric his l1ane\tA:)rk rnt has done

•It is clear tl1at Eric has l"Dt done hi.s haneNt>rk. I

(11) Er \tK)rdt bew'eerd dat Eric zijn huiswerk niet heeft genaakt'!here was claimed tl1at Eric his haneNt>rk not has done

I It \fIaS· claimed t'hat Eric 'has rnt done his hc:Jne\,.ork. I

'!he JOOst obvious difference that can be suggested is that those

constructions above this division on the hierarchy all have a o-role

1. '!his \liaS fOinted out: to me by tbam Chansky and credited t..o IlligiBurzio.

221

assigned to the subject PJsition (see Safir to appear). For this sort of

analysis we have to assume a quasi-type of a-role for ~ather predicates.

V(pa.ssive/raising)-S' (IV)

(l) V(passi.ve)~· vs. V(rai.siD.:J)~·. '!he J.X)stulation of the split

between passive verbs am raising verbs is speculative, am it is rx:>t

clear that any language splits its inventory of pleonastics at this

PJint.

Cile reason that one might place these tlNO constructions in this order

has to do with the similarity of NP is Mj am NP is V(pa.ssive) in

English. ~t only do they have the same superficial structure, but there

is evidence for an adjectival passive as 'Nell as a verbal passive (wasow

1977, 1978), the fanner having the fonn is Mj, the latter is V(pa.ssive)

(is Mj: is surprised, is discovered; is V(passive): is pranised).

In a larguage with adjectiVal passives, 'Ne \ftOuld expect V(passive)-S'

t..o pattern like V-Mj-S I assigning a &-1:'01e to the subject PJsition. In a

lan:JUage with only verbal passives, V(passive)-S I sb:>uld pattern with

V(raising)-S I.

I am assuning that raising verbs assign an internal, but not an

external e-role. Verbal passivee are the satre, assigning an internal, but

not an ext.ernal ~role (see Williams 1981). AdjectiVal pa.ssives, 1lO'.Never,

assign an external 9-role. In Williams I (1981) characterization of

lexical rules, addition of rooqilologica,l material may affect the arglll\ent

structure of a lexical i tern. Passive nDrpholCX;W which is added by a

222

lexical rule, t~en, 11&.1 externalize an argunent.

(12) '!be cake \-JaS eaten. (adj ect.i.val )'!he cake was tmeaten.

(13) '!he boys tHere given t the bcx>k. (verbal)-IrtIhe ooys were llnJiven the 1:x:x>k.

'!he adjectival prefix~ (meaning 'lDt') may affix to an adjectival

but l'X)t a verbal pa.ssive. '!his distinction \ttOuld accotmt for a p::>ssible

differ~e bet\t1eell V(pa.ssive)-S' and V(raising)-8' since raising verbs

will always assign an internal 8-role mi.le pa.ssi.ves, if adjectival and

therefore derived by a lexical rule, ma.y assign all external o-role (see

Fabb, 1984, hoNever, for a different analysis).

(2) V'(passive/raisi.D3)-S' VS. V(passive/tmaCCUSative)NJ? 'Ihis split

is a major one and is a p::>int \ti1ere lan;uages do make a di.vision bet\4Jee!l

t\lJO types of pleonastic. elanents. 'Ihis is exemplified in Eh:Jlish with the

division bettNeen it and there.

(14)[ It L seems that she can do it.-IrtIhere ~

(15) ~It l. \tS.s found under the tree a great treasure.( '!here~

'!he true pleonastics ( i .e., non-argunents) alxlVe this p::>int in the:

hierarchy are in chains with S' s, if anythi.n:J at all. The pleonastics

below this p:>int are in chains with NPs, if anything at all.

V(passive/unaccusative)-NP (V)

(1) V(plSSive)-HP VB. V(1WJCKY:IJSiltive)-HP. 'the di.stinction bet'Neen.

pa.ssive verbs am unaccusative verbs may simply mirror the di.~3tinction

223

·given above 1Jet~ V(passive)-S' am V(rai.sing)-S', the fanner 1Jeing

rooqilolc.:x:Jically derived, the latter being lexical. V(passive)-NP, being

IlDqii101c.:x:Jically derived, may have its flrgunent structure affected. This

split, l1o\tIever, alg) lacks confinnation am again, I simply assume that

the is V(passive) cx)Dstruction is closer to the is Mj construction.

Th.i.s assumption causes a different p:coblem fr~ the V(passive)-S'

cases, l:'lc:1Never. If the passive is lexi ~ally (rather than syntact.i.cally)

derive..i, then the 6-role is assigned externally to the subject IX'sition.

1.he NP III.1St 1Je generated in the subject IX'sition thus making the structure

similar to arrt NP V Mj constructic:n. Ii:Jwever, dince, SiS but lot NPs

freely extrapose, \IJe \I.Ould expect a split bet~en the V(p:lssive)NP

structures, am the V(p:lssive)S' structures.

(i6) a. A great many doors \Ere oJ?eIled.b. A great many dcnrs were open.c. '!here \Ere opened a great many dex>rs.d. '*'!here 1Nere open a great rreny dc:x:>rs.

(17) a. '!hat sara 'NOuld finish her degree 1It'aB expacted.b. '!hat sara tNOuld finish her degree was clear.c. It \¥as expected that sara ta.Ould finish her dSJree.d. It was clear tha.t sara w:>uld finish her degree.

Subjects may be extraposed only if they are sentential. lc;)gical NP

subjects that. are fourrl !X>st-verbally are actually ba.se-generated within

the VP as objects ei-"ller to passive verbs or unaccusat.ive verbs.

The result of the above discussion is that \tJe predict that larguages

which have only lexically der:lved passives will rnt have the

pleonastic-V(passive)NP c..'Or.tstructions unless it can be independently

argued that d-structure subjects may appear in the VP. '!he reasoning

224

behi.n] this is that V(pa.ssive) in such ~ laJl3Uage will assign a o-role to

the subject J.X)sition, therefore the subject J.X)sition must be filled at

d-struct:ure. In this way, such V(pa.ssive)-type verbs will be

indistingui.shable fran other verbs \tbich assign 9-roles to the subject

J.X)sition.

Malagasy, in fact, sUPJ.X)rts this prediction. Tra.vis and Williams

(1983) argue that Malaya-Polynesian laJl3Uages use lexical rules 'llhich

externa~ize argunents of the verb. In these langua.ges, then, we 'IJOuld

expect a pa.sRive-type construction to assign the pa.tient 9-role directly

to the subject.

(18) Mividy vary Ri.na.AT-bIy rice Rina

I Rina buys rice. I

(19) Vidin-dIe.kot:o ny vary•I?l'-buy-by-Pakoto the rice

I '!he rice is bought by Pakoto. t

In (19), ~~ 'the rice' gets assigne:i its 8-role by the VP. We

pre:iict that Malagasy will IXlt have a V(pa.ssive)NP construction.

'!he act:.ua~ analysis is IXlt obvious. First, Malagasy does rnt have

overt pleona.stics. Secom, since it is \100, it is difficult to sa.y, in a

V-NP structure, whether the NP is in the VP or in subject J.X)sition.

Ho\\ever, in Ma.lagasy, NPs within the VP are a.llowed to be indefinite,

while NP subjects are rrJt.

(20) M.ividy vary ny lehilahy.AT-buy rice the men/man

I The men/man b~/ S l"ice.·

225

(21) *Mividy vary lehilahy.AT-DIy rice men/man

'Men/a man buy/s rice. I

We may take the fact that the NP in a pclssive construction may not be

irrlefini.te as an irrlication that it may B:>t be within the VP.

(22) *Vidin_ varyPl'-buy-by Rakoto rice

I Rice \tt6S 'bought by Rakoto. I

(2) V(plSSive/tmaCCl1Sa'tive)RP VB. V(passive/unaccusative).... 'lh.i.s

split does rnt exi.st in many larguages s~ impersonal pclssives do rot

occur in many of the languclges studi.ed. IiJ\tJever I the difference in these

t\tJO CX)nstructions is clear. One OXlstruction concerns a p:>ssible chain

bet\tleeIl the subject p:>sition and an NP, and the other I I'X) ella.in at all

aince the sentence does rx:>t contain arrrt:.her NP.

V(passi',e) •• ". vs. V(unaccusative).. (VI)

In our language sample, only Irish contai.ns thi.s last fonn. Agaj.n,

the fonner fonn is norphologically derived am has adjectiVal terrlencies

\tbile the la.tter is not derived and is not adjectival. Bere, hO'.\ever I it

is even harder to argue for the existence of an a.djectival fonn where

there is no NP in subject p:>sition at any level. '!he pred.iction is tha.t

la.n:JUages of this type, such a.s Irish, must have syntactic pa.ssives.

M.1ch of the above discussion is specula.tive am descriptive. It is

meant simply a.s B. heuristic for a closer investiga.tion of cross-lirguistic

pleonastic invenbories.

5. 2 Pleona.stics: language specific

In this section tNe look IOOre closely B.t the hierarchy we ha.ve

prop:>sed B.S it rela.tes to B. certa.in nU11ber of largua.ges to 1:>e studied.

BelC7A' is a. chart of such pleonastic elements. 1he 0 I s indica.te tha.t in

these constrlJCtions, the ple:>na.stic element is pro. - irrlicates tha.t

the larguage does n:rt:. M.ve such a construction.

226

(23)English:Yiddish:Irish:n.ttch:Gennan:French:Ma~agasy:

lta.lian:

T--typethereooeroi1oo

I-qpeitosehetesi1ao

Fr Ehg Ir OJ Ger M3~ Yid Ita.lI. referentia~ il it se het es izy es 0

II. W1eB.ther-V il it se het es 0 0

'Nea.ther-Adj il it se het es 0 0

III. V-Mj-S' il it se het es 0 0 0

IV. V(pa.ssive)-S I i1 it (se) er/het 0 0 0 0

v(raising) -5 I il it (se) er 0 0 0 0

v. V(pa.ssive)NP il there 0 er 0 0 0

V(una.ccusa.tive) NP i1 there 0 er 0 0 0

VI. V(pa.ssive) ••• 0 er 0 0

V(unaccusa~ive) ••• 0

227

I ha.ve inc1trled referential prol'X)UllS in the ab.Ne table beca.use they

are often h:m:>pholXll1S with the I-type ple:>na.stic and are i.mp:>rtant for the

pro-drop fa.eta.

'!here are severa~ ,IXlints tha.t can be studied fran this da.ta.. Q;le

issue involves empty versus lexical NPs, am a.IX)ther issue involves the

split bet\tJeen t\\O types of pleonastics, the T-type and the I-type. Gennan

am tutch, for insta.n=e fa~l ~ether on the secorrl question since they

both split 'bebNeen the t\\O types of pleona.stics B.t the same ,IXlint.

HJwever, Gennan di.ffers fran tutch on the first question si~e Gennan

a.llaNS empty pleona.stics while rutch genera.lly does not.

'!here are uevera.l questions, then, which are raJ.soo.

1. Why are sane pleona.stics empty?2. Why are sane pleona.stics optiona~?

3. Why do scme langua.ges ha.ve t\\O pleona.stics, and Wla.t is the difference'bet.ween the t'NO?

'!hese questions will 'be addressed in the sections 'below.

5. 2~ 1 Hnpty subjec+--s

There are five langua.ges in our sample tha.t a~lC711 empty subjects but

these fall into three groups. lta.lian a.llows referential a.s well a.s

non-referential empty subjects. Ma~a.ga.sy and Yi.ddish ha.ve empty

l'X)n-referential subjects, am Iris~ am GeIman a.llow only T-type

pleona.stics to 'be empty. As 'lara.ldsen (1978) implies, !NFL can be "rich"

to di.fferent extents. I pro,IXlse tha.t if !NFL is very rich, referential

prOrK)uns ma.y drop, if !NFL is quite rich, .a.ll pleona.stics drop, and if

228

!NFL is only slightly rich, only T-type pleona.stics can drop.

'!bere are several p:>ints that should be made here. sane of them will

be answered in further sections of this chapter, sane will be left: for

future research.

1) Why does the appearance of the empty subject £0110/1 the

hierarchy? In OL er \\Ords, \tJhy are there I'X) cases \tJhere T-type

pleonastics are lexical \tthile I-type pleonastics are empty? '!he answer is

su.ggested a1:xNe in the discussion of the richness of INFL and will be

fleshed out bit by bit in the following paragraphs. '!he solution, I feel,

is in what features are necessary in referential and pleonastic elements.

'!his will give us an idea of which features !NFL must carry in order to

properly identify di.fferent types of ernpty subjects.

2) '!bere appears to be a gap in the p3radigm as shown below.

(24) Referential

o1111

I-type

oo111

T-type

ooo21

ItalianMalagasy, YiddishGennan, Irish, rutchEnglishFrench

A larJ3Uage will lX)t use the same pleonastic for roth the I-type and

the T-type unless 1) it also uses that item for the referential prorx:>lD1,

or 2) the pleonastic is o. 'lhis may be a false generalization due to the

languages cOOsen but I assume that this is not the case. Another way of

stating this is that the I-type pleonastic must be the same as the least

marked referential prOlDun, unless the I-type (and therefore the T-type)

229

pleonastic can be properly ident.i.fied by !NFL. This question I leave

tmanS\ttlered.

3) Unlike referent.i.al pro-drop, pleonastic pro-drop appear to be

obligatory. As mentioned earlier (suggested by MallDru saito) I asslIlle

that in both cases, pro-drop is optional. Referential pror.ouns may be

used for anphasis. Since this is senseless in the case of pleonastics,

they will never appear.

5.2.1.1~ features

In Olapter 4, we have argued that pro-drop must be sensitive to

proper goverrment of the subject lDsition. This means that in Italian,

the empty subject, pro, is in the lDst-verba1 subject lDsition. '!he

secom part of identification of gaps had to do with feature

specification. In the case of gaps left by IlOlanent, a correctly fanned

chain recovers the features of the gap fran the head of the chain. '!he

types of gaps that \tJe are discussin; in this chapter, l'r:Jwever, are rx:>t

created l'Jy IlOlanent so it is less obvious ...mat will supply the necessary

features.

I suggest t.hat there are certain features that ~"'L carries that

identify the content of a properly identified gap. As we have seen above,

these features I1U1St distinguish four types of !NFL, Italian type, Yiddish

type, Gennan type, am Fn31ish type. In Italian, ~ must be rich erDugh

to identify referential NPs as 'Nell as 00th I- and T-type plec>nastics. In

Malagasy am Yi&lish, !NFL must identify both I- am T-type pleonastics.

230

In Gennan am Irish, !NFL identifies only T-type pleonastics, am in

English, !NFL identifies no NPs at all.

let us begin by .assun.:i.ng that the relevant feature for Italian !NFL

is [+person] 2. 'Ibis means that \\hen INFL is coindexed with the subject

p:lsitial, !NFL may supply the feature [+person] to the chain that is

created. ret us further asslIne that e-r'~les must be assigned t..o chains

that have the feature [+person]. In Getman, even tmugh subject p:lsition

can be properly governed in SU am embedded CDnStructions (see O'lapter

4), .i.t is IX>t a pro-dIOP lan:JUage since !NFL does lX)t carry the feature

[+person] • '!he existence of a feature in !NFL is in no way obvious fran

the appearance of !NFL itself. Inflection in Gennan certainly appears

rich enough to designate the person of the subject. Features of !NFL must

be considered a nore abstract entity that can only be teased out through

other Iitencmena in the language.

'lhe feature [+person] describes the division between Italian and the

rest of the lan:JUBges surveyed here. I sU93est further that the feature

\\hich divides Malagasy and Yiddish frein Gennan, Irish, Fnglish and French

is that of [+#] am the feature that separates Gennan am Irish fran

Ehgl.:lsh and French is [+cJ 'lthere NP[+c] is an NP which may bear case

witmut 'being plnnetically' realized. Peasons for chcDsing these features

will ~e clearer belOll. '!hese features form their Q\IJn hierarchy. If

INFL has person, then it must have! am £. If it has !, it must have c.

2. '!he cooice of [+person] is arbitrary. Rizzi (fall l~tures, 1983) uses[+argunent] tmUC3h in a slightly different way.

231

(25) !NFL feature hierarchy:

[+person] -> [ +#] -> [+c]

5.2.2 Optional pleonastics

Anot.her quest.ion \thich \tJe have raised is 'YIny sane pleonastics appear

to be optionally deleted in certain c:onstructions. Below \ttle see cases in

Irish 'YInere a" pleonastic mayor may not be expressed.

(26) Irish:Bhi (se) creidte ariamh againn go dtiocfadh an slanuitheoir ar aia\ttlaS i1: believed ever a1:-us CCMP cane«(l)NDrr) the saviour back

'we had al\eYS believed that the ~viour \tJCuld cane again.'

In a way, this is an inqx>rtant confinnation for the a'bove hierarchy.

'!he fOint in the hierarchy \fklere t.he pleonastic is "optional" is alS) the

fOint where the split 'between the t'AO pleonastics canes. Similar

constructions in D.ttch give us a clue as to the reason for this

optionality.

(27) nItch:t:~is gebleken dat Eric zi jn huiS'Nerk niet heeft ganaakt.

is slDwn that Eric his haner...ork not has done•It is sl'nwrl that Eric has done his hcrne\t.ork. I

As 'Ne see a'bove, it is not that the pleonastic is optional, but

rather that the cooice of which pleonastic is used is variable. What is

crucial is that languages that have "optional" pleonastics, have a 0

pleonastic :in their inventory. 'therefore, when the pleonastic appears to

be optionally expressed, actually it is a cooice be~ the lexical

pleonastic am the 0 plE!Onastic.

232

'nlere is a llDre canplicated case in D.1tch where the T-type of

pleonastic also appears to be optional (fran D. Jast=ers)

(28) *(Er) 'IeOrdt be\t1eerd dat Eric zijn huiswerk niet heeft genaakt\tes claimed t.hat Eric hi.s hane'ttOrk not. has done

I It was claimed that Eric had n::>t done his 1lanew:>rk. I

(29) Dany heeft:: me verteld dat (er) be\Eerd \\Ordt dat Jan ••••I:Snny has rre told that claimed \-laS that John

'Danny has told me that it \flaS claimed that John ••• '

Here is a case where a larguage witbJut a 0 pleonastic does have an

lIq*.ional" element. My analysis is that the dialect of D.1tch \fA'lich does

allC1fl this has a variant of !NFL which is [+c] like the German INFL. 'this

means that t.he T-type of pleonastic is empty \fA'len INFL [+c] is cbJsen and

filled when INFL[-0] is coosen. Crucially, the T-type pleonastic can only

be 0 when it is in a fXlsition of proper government as defined in O1apter

4. Altoough these ant>iguities make it very difficult to arrive at a'

rest-..rictive analysis with clear predictions, the analysis presented here

will nevertheless predict that D.1tch will never drop a het that does rot

have an er alternant. 'Ihis is rome out below.

(30) Is *(het) betreurd dat hij het hoek heeft gekocht?is regretted that he the bcx>k has rought

'Is it regretted that he has rought the bcx:>k? I

Basically, languages that have a [+c] INFL may have a variant where

!NFL is alSo [+#] (German), or a language with an !NFL with no [+]

features may have a variant with a [+c] lNFL.

5.2.3 I-type va. T-type

233

'llle issue 'Ne have been sk.i.rtiD3 is Why there are t\ttO pleonastics at

all. And W'ly all languages do rnt split the hierarchy in the same way.

'lhis is a krotty problem am is di.fficult to Lmw:im in an organized

fashion. My first step to\\Iards a solution wil~ l:>e through Irisl... At

first, any coherent description of pleonastic distribution in Irish lc::oks

imp:)ssible. I claim, hch.ever, that by allowing verbs to specify o-rrarking

am case-marking properties in the lexicon, sanething which is needed

independently, one can at least group the facts systematically. 'llLi.s

grouping also sug:Jests a way of parti.tioning the hierarchy.

en top of this prelim.i.nary partitioning, I will p::>sit another

di.vision that can be made follow:inJ Peulam (1983). 'lhis SecDrrl division

deperrls on the sort of chain that is created, i.e., 1Nhether the pleonastic

is coirrlexed with an 5' or an NP. AltbJaJh I eventually disagree with

Reuland' s analysis, it offers an i.mp:>rtant starting p::>int.

Having divide1 the hierarchy into subparts, I then investigate the

lCMer p::>rtion, mainly T-type constructions Which involve p::>st-verbal NPs

or I'D VP internal argunent at all. I argue against any theory which

relies crucially on chains (Pollock (1982) I SBfir (1982) I and Ieuland

(198~» am offer a different explanation for the presence of a

pleonastic. My analysis is based on the int.eraction of case assignnent to

the p'stverbal NP, am agreanerit of the verb with the NP.

Finally I return to the problem of the split bet\EeIl T-type and

I-type plt90nastics presenting a direction of investigation suggested by

sane facts in n.ttch, and questions for further research.

234

5.2.3.1 Irish

'!he data fran Irish3, at first glance, seem very confusing and

haphazard. Predicates appear arbitrarily to take I'X) plElJrlastic, an

obligatory pleonastic, or an optional pleonastic. Ql closer

investigation, h::Mever, we can see that sane of the predicates fall into

clusters.

v...,. Anytime the "logical subject" is an NP, whether it is marked

ac:cusative (31), or it is in a PP (32), there can be I'X) pleonastic.

(31) Marbhadh (*se) ar an bmthar areir ekill(IMPER PAST) on the road last night him(acc)

I He was killed on the road last night. I

(32) OUaochlaigh (*se) ar a neartdeteriorate on his strength

I His strergt:h waned.'

Als:>, if there is no logical subject at all, the pleonastic must be

absent.

(33) Tostadh (*se) sealbe-quiet (PAST IMPER) a-time

''!here \..as a silence. I

V-S'. Where the construction contains an S·, there are three

PJssibilities. In sane cases, the ~ is required, in sane it is optional,

and in ~e it is rv:>t PJssible. Qle regularity holds, hc::7t.ever. Whenever

the verb is passive, the ~ is optional. 'lb.is makes Irish look like IlJtch

-------3. Infonnation on Irish pleonastics canes fran J. r-t::Closkey.

235

Where the ple::mast.ic is either bet or ere As \\Ie have discussed earlier,

then, in Irish it is I'X)t so much that the pleonastic can be dropped

opt:ionally, but rather that either one or the other ple:>nastic may be

used, i.e., either se or o.

(34) Ta se beartaithe againn a dhul go Meiriceao

is it detel:m.i,ned at-us go(-FIN) to America'We have detennined to go to Mlerica. I

1) se/O (either I-type or T-type)

Let us BSSUIle that these verbs, being passive, assign a 6-role but no

case to the object p:>sition. We can say, then, that all predicates which

take SiS and either ~ or 0, all assign Q-IOles but no case to the object

p:>sition.

2) se (I-type)

Mony of the cop.1la-Mjective-S' constructions take an obligatory ~.

h:Jain this lCX)KS very IlUlCh like the Gennan facts. We will assune in these

cases that a &-role is assigned to the subject p:>sition.

(35 ) Ta sa fior gur dhuirt: me sinis it true a::MP said I that'It's true that I said that. I

Let's sllpp:>se that this can be extended to all the predicates that

take an obligatory .!:..

3) 0 (T-type)

Finally, let us aSSlDle that tl'x>se predicates Which C8r1lX)t take !!:. are

like the V-PP or ~J~1P(acc) counterparts in that they assign both a-role

236

ani case (either directly or through a PP) to the internal argunent.

~ 'Ne can characterize the Irish facts. se is required of

predicates which assign a 8-role to the subject fOsition. It is optional

with predicates which assign Q-role to the object fOsition but no case,

ani it is obligatorily absent with predicates which assign 9-role am case

to the object fOsition.

(36) a. *(se)b. (se)c. (*se)

a <- VP (V Adj) (assuming Irish is ~)V -> e (V passive)V ->fa 1

(case)

'lhis analysis is not as stipulatory as it may first oR'E!ar. We know

that passives can assign 9-roles to objects witrout assigni.rKJ case. Arrl

we know that passives with SiS take an optional~. 'Ihis implies that it

may have to do with 9-role ani case assignnent. An .i.rnpJssible case \\Ould

'be a language like Irish, but:. \\here impersonal passives also had an

optiorlal pleonastic. Since there is ID internal argtment in an impersonal

passive, 'Ne WJuld rnt say that t.he verb assigned an internal e-role.

'!hen, even with the i.diosyn::racies of the lexicon, we could rx>t generalize

the class of verbs that took an o}?tional ple:>nastic. '!he fOint is, that

trough the lexicon can specify many idiosyncratic properties, it restricts

c~in possibilities.

What \'Ita have gained fran this investigation of Irish is that the

9-role ani case rnarkin:J abilities of the verb are i.mp:)rtant to the

distribution of pleonastics. We will see this idea developed 'belOli.

237

5. 3 T-type pleonastics

In 1:his section I review sane analyses of T-type pleonastics, and how

the pres~e or lack of agreanent may interact with the type of pleonastic

that a construction anploys. My reviews are brief and I suggest reading

the originals for further detail.

5. 3. 1 Revie'tlS

Reulaoo (1982). '1lle basic conclusion of Reuland's paper that we will

discuss is that chains \thich act as external argunents must contain one

alii only one NP. In a chain a •••S·, since S· is B)t an NP, a must be. In

a chain a •• uNP, a may rx>t be an NP. Reuland further ~ltrles that it,

het, aoo ~ are all NPs, wh.tle ther:e, er and 0 are rot, in English, Cutch,

am Gennan respectively.

I am assuning, cn the contrary, that all plecJnastics are NPs,

incltrlirg tba T-type mentioned above \thich Feulam assunes are not NPs ..

An argunent for 1:his canes fran case theory. '!he distribution of there is

4. 'Ibis may not be the correct gene::alization since, aE p:>inted out to meby~ Chansky, PPs s'h:J\.i the ~me distributional facts.

(i) en the grass is a nice place to sit.(ii) *I bJped on the grass to be a nice place to sit.(iii) I expected on the grass to be a nice place to sit.

238

predictable if \\le asst..m:! that it is an NP and, like other NPs, nee:ls

4case •

(37) 'nlere tNere three \Ollen in the group.

(38) *I h:>ped there to be three \«]11en in the group.

(39) I hoped for there to be three \«]11en in the group.

(40) I expected there to be three \«]11en in the group.

Pollock(1982). Ebllock discusses only th@ cases \IJh.:ich contain

p:>st-verbal NPs. Ii:! argues that AGR mul3t be coind~ea with an NP which. is

[+/ -nunber] I [+n::m:inat.i.ve]5• 'D1is is me2lnt to account for sane

differences bet'Neen Eh;Jl;.sh am French that have to do with ch:>ice of

pleonastics and agreanent on the verb. '!he relevant data are given

belOil.

(41)

(42)

(43)

(44)

(45)

(46)

(47)

'nlere (??is") three people in the roantare '\

There (*are~ a cat in the roan.[is)

then there l*:es~into the roan an enonoous do:J

then there r*:es1 into the roan three enorroous dogs

I1 y ( *ont?trois personnes dc;lnS la chambre(a )

I1 Y[ a I un chat dans la chambre.l*ont\

Il[est ~ arriv6 trois femnesl*sont)

5. R:>llock 's [+/-nunber] represents [plural (+nunber) /singular( -nllTlber) j.I have used [+#] in a very different way meanirg that an NP which is [+# ]is either singular or plural but at least carries a feature for nunber.

(48) n ) est( arriv~ une femne.(*sont)

Ebll<JCk's analysis makes several assllllptions:

239

1. Certain pleonastics Imlst be coWexed with AGR because of an

intrinsic charetct-eristic of 'CDhesion forte' (strong c=ohesion) I for

instance, il am ce.

2. 'lhose elanents \\h.:i.ch are gO\1erned by and cnindexed with AGR are

assigned lDlli.native case.

3. Cet+..ain plec>nastics are inherently [+/-nllnber]. 'lhese are iI, and

ca.

4" A chain may only have one NP with [+/-nllnber].

5. A chain may be assignec.1 case only once.

6. French unaccusatives assign case to the NPs 'lthich they govern in the

VP.

7• Fnglish cop.iJ.as assign case to the NP wi.thin the VP.

8. AGR must be coimexed with an NP 'lthich is [+ncmi.native] am

[+/ -nllnber] •

N:7tI let I s l<:x:)K at the follOlling examples.

(49) •• then there cane in-t:o the roan three \IJCl\en.

(50) I1 est arriv' trois harmes.

(51) n y ~a <trois personnes dans 1a chambre.<"rot)

240

In (49), three 'ttOtlen does rx>t get case fran the verb since English

unaccusatives do n:>t assign case. '!he NP therefore must be coindexed with

the subject {X)sition to get case. Si~e three 'ttOtlen has [+/-ntrnber] but

no case , am there llas cose but does nJt:. have [+/ -nllTiber] , the ehain is

\\ell-fonne:l arid AGR may be coirnexEd with the cllain. 'lhis aCCOtmts for

tpe agreEment of the verb with the {X)st-verbal NP.

In (50) and (51), est arrive am ~ as·sign case to trois hCmnes and

trois persormes respectively. Sirx:e AGR is coirnexed with il by 'cohesion

forte', and since lillaa [+/-nunber] am will be assigned rx:m..i.native case

.by coindexation with AGR, the sentence is 'Nell-fonnEd.

'!he example belat'l \'tOrks a bit differently fran toose already

discussed.

(52) a •??'Ihere' s three people in the rcxmb. '!here are three people in the rcxm.

Since Ehglish copula assigns naninative case to the NP which it

governs (per assunption 7), three people will get case directly fran the

verb. '!here are t'NO {X)ssible irrlexations then in Poll()Ck' s aCC'Ount.

(53) there AGR be three people in the rcxma. i i jb. i j j

In (538) (\thich produc~s (528» AGR is coindexEd with t.here. FbIlock

must rr:M say tllat an NP, when governed by AGR,' may be assigned the feature

[-n\J11ber] (i.~., singular). '!he string will be gramnatical sinc\~ AG< '~':i ..ll

be coirrlexed with an NP (there) which is [+rx:minativeJ an:! [~}i-i"1t,JTber].

241

In (53b) (which prcrluces (52b» AGR is coimexed with the naninative

NP in the VP. 'Ihis satif.i.es assumption (8). But nt:JW I:bllcxk must allow

there to be generated with case (I un argunent muni d'un cas inherent' )6.

5.3.2 T-type pleonastics are Case

Witrout going into lOOre detail on either of these analyses, I suggest

that any analysis which deperrls on chains is m.i.sguided. B:Jth Reulam aOO

FOllcxk concltrle that !NFL (AGR) Im..1St be linked with scmething stronger

than just a T-type pleonastic. If the pleonastic is in a chain with an S'

(Reulam) or with nothing else at all (FOll()Ck), the pleonastic must be

sufficiently "strong" to bear the irrlex of AGR. Fbr Feulam, this means

cb:x>sing it or het rather than there or~. Fbr l:bllock it means cbJosing

il (rather than ~ which does IDt have [+/-nllTber]). Fbr Peulam, being

•strong enough' means being an· NP. Fbr Ebllock, it means having

[+/-nllli:>er] am BEinative case. Asslllli.n:.:J that the strong pleonastic is

the I-type, am the \\leak plE!Onastic is the T-type, we 'NtJuld al\tlays expect

the I-type pleonastic \tihenever there was IX) chain fotrnErl. '!his certainly

is the case in French as sl"1o'Nn al:xJve.

'lhis is rnt the case, towever, in other laD3Uages. Qle

counterexample canes fran Gennan impersonal passives. Below we see that

the pleonastic is 0 which is the GeI:man T-type pleonastic.

6. We have, hC1\ftever, already seen examples \tihich argue that there must beassigned case (see examples (37)-(40).

242

(54) He!ute wurde 0 getanzt.today was danced

.' It \tes danced tcrlay. I

'!here is no obvious NP to coWex the 0 pleonastic with yet tNe still

have the 'Neak fonn, the T-type, of pleonastic. Reulam suggest:.s that

impersonal passives have enpt:y NP objects that suffice for coindexation.

AltlDllg'h Poll~k does l'X)t incltne Gennan in his stu:1y, in his tenns, these

NPs 'NOuld presunably also have to have [+/-nunber]. 'lhis raises the

question of \thy this enpt:y NP \\Ould al\\aYs have to 1:e singular. '!he

example belc::JII is tmgramnatical.

(55) *Heute wurden 0 getanzt.today W'ere danced

Beyond this, 'h<:JwIever, there are further problems. In Irish there are

cases which are harder to argue against. In the examples where the only

internal argllllent is assigned case by a preposition, the existence of a

chain is less tenable.

(56) Laguigh ar an stoinn'lteaken on the stoJ:Il\

''!he stann \\eakened. •

'!here is no agreement, first of all, and secorrlly, the chC'in 'ItOuld be

assigned case twice; once by the preIX>sition and once by coWexation with

AGR. '!he conclusion, then, is t.hat there is no chain, AGR is coindexed

only with the plec.nastic. ~er, the pleonastic is the T-type, the \fJ8ak

fonn, suggesting that the restrictions that lEuland am J:bllock have

p:lsited for coirrlexation with AGR are i.rxx>rrect.

Perhaps the clearest characteristic of the T-type pleonastic is that

243

it is assigned case by INFJ:.,. It does rnt need to transfer this case or to

fonn a chain. We have seen counterexamples to 'both of these claims in

Gennan am Irish. let us say, then, that T-type pleonastics are simply c3

manifestation of case assigrment. '!hey cannot be assigrled o-roles because

they do rnt have the feature [+person]. '!hey therefore must. be in oon-Q

but: casemarked {X)sitions. N:>n-o {X)sitions will al'NaYS be subject

{X)sitions (accordin:J to the Exterrled Projection Principle) so \Va expect

T-type pleonastics to appear in the same structural {X)sition as subject

NPs. '!he near canplanentary di.stribution of lexical NPs am T-type

pleonastic NPs will be detennined by the e-assigning abilities of the VP.

we will see below that M:we-a will account for the fact that the

canplementary distribution is not canp1ete.

(57) '!he 'NOllan bo~ht the lxok.

(58) *!'here 'bo~ht the book.

(59) "'The \\Ol1an \feeS discovered under the tree a great treasure.

(60) '!here was discovered urrler the tree a great treasure.

(61) A great treasure \tIaS discovered under the tree.

(58) is out by the e-criterion since the T-type pleonastic carunt

bear the 6-role assigned by the VP. (59) is also out by the e-criterion

since there are t\teO NPs for one e-role, the 6-role assigned by discover to

its object {X)sition. In (60), lDwever, ~ great treasure gets the Q-role ~

assigned to the object {X)siti.on of discover, and there is in a non-o

{X)sition which is assigned case. '!he canplanentary distribution of T-type

. pleonastics and referential NPs is disturbed only by M:>ve-a as ShONn in

(61) where a great treasure tooV'es into the ron-e casenarked {X)sition. '!he

244

T-type pleonastic, then, Il\3Y 'be a late spell-out rule of the feature

[+case] •

we have already seen that case does not necessarily cane fran INFL.

T-type plE!CX"1astics are also fouOO in F.X:M constructions.

(62) She 'believe3 tllere to 'be buried under the tree a great treasure.

'llle fact that 'I.e want to capture is that, in sane languages which

have a 0 T-type pleonastic, a [+tense] INFL does not require a lexical

subject.. '!he way I.have coosen to represent this is by a11owin:J INFL to

assic.;.tl a feature [+c] to tlle empty subject pJsition so that tlle subject is

row able to receive case witoout bei.n:J lexicalized.

We have said above that tlle feature [+c] \\hich has been assigned to

the subject pJsition by !NFL in the case of Getman allows the subject to

bear case without being phonetically realized. Q1e question might be

whether thi.s is true only of nani.native case, or whether it is true of all

cases. '!here is evidence fran Arabic that an empt:.y NP[+c] can only bear

ncminative case. h)un (1980) argues that there is an elanent

[-referent.ial][-I;hc>netic] \lAUch lIis to expletive elements (it, .•. ) \\hat

PRO is to referential prorxnJI1s (he, she, ••• ) II (p.l) • (Within the

framer.e.ork being assuned for this thesis, PRO is ~ in this context.)

What is interestir¥3 ab'.Jut this Sllpty elanent in Arabic is that it surfaces

when "it is in a I=Osition Which is assigned accusative case. 'lhe relevant

examples are 9iven 'bel<:7#1.

(63) mi.nal \\9 :diji ?annaprep-the clear that

I it is clear that ••• I

245

(64) ?inna lwalada ya19abu(ace)

•Indeed the boy(acc) 11e is playing.'

(65) ?innahu pUnal \..e :dihi ?annaindeed-it prep-the clear that

•Irrleed it is clear t.hat ••••

In (63) '.Ne see a construction with a missing pleonastic. In (64) wa

can see that the canplanentizer ?inna, \tJhich acts as a corroborative in

the matrix S, assigns accusative case to the adjacent NP. In (65), we see

that \tihen this adjacent NP is an anpt.y pleonastic, it must surface

phonetically as -hUe

Altb:nJgh the examples above use an I-type pleonastic, the argunent 'Ne

are making still goes thIough. Arabic is a pro-drop language, allowing

referential NPs to be empty. J\ccording to our analysis, then, !NFL

CDntains the feature [+person], and by our feature hierarchy given in

(25), !NFL must also contain [+#] and [+c]. '!his means that it may be

assigned a o-role and naninative case witoout being plx>netically

realized. Ait}x)ugh 'Ne have not yet discussed the feature [+#], let us for

rr:M assure that an NP wi~h this feature may fonn a chain with an S I •

Arabic !NFL bears the feature [+#J. 'lherefore, the chain of ([NP, S],

!NFL) may be coindexed with an extrap:>sed S· as in example (63). 'lhere is

no feature, hc1ftever, that allC7#1S an empty subject p:>sition to bear

accusative case. 'lhis is the result tNe want. In Arabic, \tihen the anpty

subject is assigned accusative case, it must be lexicalized.

Aoun I S article raises the question of pleonastic PRO. In pleonastics

appear in p:>sitions \tJhere PRO can appear? 'Ibis Broounts to asking \tJhether

246

PRO can appear in rnn-Q FOsitions. <Jbviously the environnent 'He are

looking for is one of arbitrary PRO rather than control PRo. Since

controllers are at least quasi-argunents, we will rot fim T-type

pleonastics in these constructions.

In positions of arbitrary PRO, it ells:> appears that T-type pleonastic

elenents CoIllX)t appear.

(66) a • It is difficult PRO to make baclava.b. It is difficult for us to rreke baclava.c. *It is difficult PRO to be lots of people in the oouse.d. It is difficult for there to be lots of people in the muse.

In the examples above, it appears that PRO call1X)t be a T-type

pleonastic. '!here are t\teO reasons \IA1.y this might be. '!he first reason is

that if the subject position is rnt assigned case, then the NP lots of

people canrnt get case tlu:ough the <nimexation with the subject

p::>sition. (66c) is ruled out, then, by the Case Filter.

lA secom reason Why (66c) is urgramnatical might be t"tli3t PRO has a

restricted interpretation. In the examples belaN, it appears that PRO has

to be hunan. 7

7. '!his restriction of PRO seans to be relevant only in infinitivals.Gerurrls a110\411 a wider raD3e of PRO interpretations incltrliD3 I-typepleonastics.

(i) '!he rocks rolled cbeJn the hill witb:>ut PRO hitting anyone.(ii) It rained all night witoout PRO~.(iii) It is possible, wi1::hout: PRO being obvious, that Jeremy will arrive tonight.

But. even this is rx>t. clear.

(iv) *('!here) having been too many people, we left:. early.

247

(67) '!hey rolled down the hill.a • they = rocksb. they = children

(68) It is difficult PRO to roll dONrl the hi.ll.a. *PRO = reeksb. PRO = children

'!hi.s is alSJ \\hat 'Ne w.Juld exp3Ct assuming +-J1e' characterization of

arbitrary PRO Which is that it is a variable ranging over irrlividuals in

danain D. Since it must pick out:. an individual in this danam, it must be

referential.

Gennan examples indicate that both case theory and the restriction on

the sanantic distribution of PRO rule out (65). BelCJW \Ie see an example

~ere there is no NP which relies on the subject pJsition for case

assignnent. I assure in this example that this is out because of the

[+human] restriction on PRO (see S3fir (to appear) for a different

analysis.)

(69) *Es ist sch\er getanzt zu \\erden.is difficult danced to becane

* I It is difficult to be danced. I

5.3.3 Analysis

let us say that a parameter is set 'Ilithin a languoge as to whether

LInaccusative am passive verbs may assign case of sane sort to their

internal argunents (see the discussion on Belletti (forthcaning) beleM).

Also, \tJe will assune that AGR is al\eYs CX)Wexed with [NP,S]. If the

internal argunent is assigned case, then there is no need for it t..o fonn a

chain with the subject p:lsition. In fact, it is n:Jt p:lssible for it to

248

fonn a chain with the subject p:lsition since the chain \teOuld be assigned

case twice.

If case is assigned, there is no chain and AGR is coindexed only with

the [NP,S] p:lsition. '!he consequence of this is that the verb will rx:>t

agree with the [NP, vp]. lack of agreenent, then, is crucially linked with

the fact that the NP can get case irrleperrlent of the subject p:lsition.

'!his analysis is easiest to accept \tihere it is obvious that the

[NP, vp] is assigned case by sane elenent other than INFL. '!his may be

either because the NP appears with accusative case marking on it or within

a PP. It is IlDre difficult to accept this analysis when, as in French,

there is no overt sign that case has been assigned to the NPs that do n:>t

appear within PPs.

(70) II est e-rrive' trois fenmes.

Est is singular 000 does l'X)t agree with the [NP, vp]. AJ.:cording to

our analysis \tJIE! have to assume that there is no chain and that case is

assigned to the p:lst-verbal NP by tlle V. '!his is true in Erglish as well.

In examples fran colloquial Ehglish such as toose given belC7N, there is no

agreanent bet\fleell the verb am the [NP, vp] •

(71 ) '!here I s a lot of books in the roan.(72) '!here seans t to be a lot of bcx:>ks in the roan.

What this means in 1:enns of our analysis is 1:hat. there is no chain

between the subject NP am the [NP, VP]. 'lhe NP, then, must get case fran

t.he V in order rx:>t to violate the case Filter. tbtice that these facts

are only true of copulas in Erglish (as already p:linted out in Pc>ll~k's

249

analysis above).

(73) *... then there canes a lot of people into the roan.(74) irtIhere seans to have arrived a lot of people fran r-Dntreal.

It is not .:irrmErliately clear \fJhich case is assigned. We have seen t\\O

different t.ypes of case that tle NP can get; (i) accusative, (ii) oblique

(within PP). AnoTller problem is also raised wiTll our analysis. Sa fir

(1982) assunes ~ial1y that FOst-verbal NPs in French are coinJexed with

the subject FOsition to account for the fact that the NP nust be

inJefini.te. As in ~lish there constructions, in the Frerch

constructions the [NP, vp] Imlst be irrlefinite.

(75)

(76)

II est arrive l*~"\b:mne

'!here arrivedt*t~e~man

5afir FOints out an interest.ing fact in French (p. 177) first

rx:>ticed by ~yne (1975) •

(77) Jean a t.ire' sur Ie bateauJam has soot on/at the lxlat

'Jam soot on/at the lxlat.'

(78) 11 a ete' tire' sur Ie ba+.:eau/un bateauhas been sOOt at the txJat/ a lxlat

''!here 'NaS soot at the lxlat/a lxlat.'

(77) is ambigtDus bet\tJeen t'ttO readings. Sur Ie bateau can be the-------locative phrase 'on the txJat' or tirer~ can be the idianatic verb

meaning 'to shcx:lt at'. With the lA3ssive constrtrticn, only the idianatic

reading is FOssible since passivization in French is sensitive to the

argunent structure of the verb. Only verbs with internal argunents Il\3Y be

250

passivized (as in Irish).

What is imp:>rtant in these constructions, ho\t.ever, is that the NP may

be definite. Safir uses this as evidence that there is no chain only in

the construction where the C'rgunent NP is contained within a PP. It is

crtX:ial for Bafir that French unaccusative am passive Va do rnt assign

case. If they did, \ft1e tNOuld expect no chBin and., therefore, no

definiteness effects. .

Belletti (forthcani.ng) offers an alternat..:ive to this analysis. Sle

sugjests thBt verbs may assign partitive case to adjacent aWs. Sin::e

partitive case will COmley the meaning I s:me', it. is necessary that. the NP

be ind.efini1:e. '!he only NPs, then, thBt will B)t. have to be ind.efinite

are trose tl1at. are assigned accusative case (as in Irish) or trose in

PPs.

'!he conclusion'Ne draw is thBt if the verb a.grees only with the NP in

subject p:lsition (i.e., the pleonastic) I it is because INFL is coind.exed

only with this p:l~ition, and. that the pleonastic, in turt:1, wi1:hin tl1at S'

is (X)indexed only with INFL. '!his means that (i) there is l"X) other NP as

in Gennanic impersonal ,Passives, or (ii) if 1:.here is an NP which appeers

to be the logical subject, it is assigne1 case by sane elenent within the

VP. 'lhis may be either the V itself, or a preIX>sition. '!he T-type of

pleonastic appears in constructions both with am witmut agreanent with

this logical subjec.~. In our analysis, this means that a T-type

pleonastic mayor may rx>t· be in a chBin. Its purp:>se is simply to

spell-out ca.se.

251

'll1e analysis presented here differs from Pollock's in the following

ways. Firstly, there is no need for 'cohesion forte'. I assume t-hat !NFL

(AGR) is alV«3ys coimexed with the [NP,S] IX>sition. 'T.his, in turn, means

that t.her~ will al\#teys be coimexed with AGR and will always be ossigned

case in a tensed clause throU;h this co~.ndexation. We therefore do rx:>t

need to say that there can have inherent case. Since AGR must be

coindexed with [NP, S], we c~nn::>t generate the folloong coindexed

struct.ure which is crucial to Pollock's analysis.

(79) there A(;R

.i jis three people in the roan

j

we will, hOY.ever, produce the t\\O structures below.

(8) there AGR be three pe:Jpla in the roana. i i jb. iii

I:Ollock rules out (SOb) since three people will be assigned....,.

ron.inative case by be, and there will be assigned case by coindexation

with AGR. rille (~hain will be ru.1ed out by the ex.i.stence of t\tJO case

assignm NPs.

'!his problem may be sol,'ed by assun.i.ng that be dues not necessarily. -assign case to an NP which it gO'1erns. In fact, in the UIma!:ked form,

(BOb), be does l1C)t assign case to the [NP, VP]. 1:.1 a marked c...onstruction,

l"Owever, be does Dssign case, am \tie fini the verb in the s:i.n;u1ar whether

or n:>t the [NP, VP] is singular.

N:>tice that, tmlike Pollcck wh:;) assunes that i1 ::'s essentially

different fran there, I am assum.:ing that i~ can play the role of both the

252

T-type pleonastic am the I-type pleonastic. In other \'tOrds, I assure

that il is, like there, a spell-out:. of case, but:. that it alB) has the

feature [+#]. '!he generalization may be that languages with subj'SCt

elitics will use these elitics for the T-type ple:>nastic. 'Ibis is oot

l.mreasonable in light of the characterization of clitics as the spell-out

of case features (B:>rer 1981).

5.4 I-type pleonastics

we have arguad al:x:we that T-type pleonastics are the spell-out of the

feature [+case], am \..e have hinted that I-type pleonastics are the

spell-out of [+#]. '!he clDice of the feature is fairly arbitrary as it is

in the case of [+peroon]. '!he intuition I want to capture is that the

I-type pleonastic is in sane way stronJer than the T-type pleonastic. In

this sect:.icn, \tJe will review sane of the data CX)ncernirg I-type

pleonastics and set up a problem for future research.

First, let us separate quasi-argunents fran l'X)n-argunents. I will be

assuning that quasi-arglltlents, tlDse argunents used with \\leather

predicates, IRUst be [+#] because of their quasi- status.

'!he other sort of I-type pleonastic ~s the one \\hich is coindexed

with an S·. BelOtti 'Ie will lcok first at an analysis ProiDsed by safir

(1984) and examine why this analysis will not accotnt for the data we have

presented. '!hen we will di.scuss an interestiBl arraBlanent of data in

D..1tch.

253

'!he problan. then renains that Gennan, Irish, am Il.1tch all use the

T-type pleonastic in construction Where F11g1ish uses only the I-k.ype.

5.4.1 ES VB. ES/O: Gennan

8afir (to appear). safir (to app:!ar) offers a solution to the ~/o

vs. ES dilE!l'l'l'Da. B2 basically prop:>ses that ES appears Where there is a

trace left that is IDt properly governed. His account canrnt be used

within the frame'ltOrk presented here as will be sho\tJn belOll.

Safir rrakes the generalization that ES is used in CDntexts \tJhe"':'e a

o-role is assigned to the subject p:>sition. '!his incltrles mainly

extrap:>sition type clauses. He CX)ntrasts the follow:i.r:g sentences.

(81) a. Er sagte, daB (?*es) :ibn erklart wurde, daB Hans den a.md getl:)tet hat.he said that him explained was that Hans the dog killed had

'He said that it \tIaS explained to him that Hans killed the dog. '

b. Gestern wurde (?*es) erklart, daB •••yesterday \tRJS explained that

•Yesterday it \\as explained that ••• '

(82) a. Er sagte, daB ?*{e~} klar ist, daB••• 8

he said that clear is that'He said that it is clear that ••• '

b. NUn ist. ?*(es) klar, daB •••~ is clear that

,N:Jw it is clear that ••• '

8. '!he pleonastic is an.:itted, 1lo\ever, if a dative NP is incllrled (I •••daB mir klar ist daB •••• ). I will assure W'rler certain circur.stancesthat t.hese dative NPs may appear in the subject p:>sition.

254

His generalization is "we might: say that predicates that do rot

assign external theta-roles permit s-ES to drop, but t.b:>se predicates that

do assign external theta-roles do rDt:. pennit S-ES to drop.lI(p. 27) •

Since subject position is assigned a e-role, it will follow fr~ the

Projection Principle am the e-criterion that the subject position will be

filled at d-str~ture. At s-structure, then, the posit.:i.on will contain

the trace of roovanent. 5afir claims that this trace will rx>t be properly

governed and therefore must be filled. withES. 'this will account for the

fact that ES will appear only in e-rnarked positions.

In Olapter 3 we have seen ~ fir I s account for the distribut:..i.on of

FS/O. Unlike our account, he does rx:rt:. use proper goverrrllent to explain

the es-insertion facts. Si.nce \'t'e do use proper governnent to explain the

ES-O distribution of the FS/O pleonastic, \tJe C8nrK)t use proper governnent

t.o explain the ES vs. ES/O distribution.

In Olapters 3 and 4, I have argued that a position wh..i.ch is adjacent

to a VP is pl:Operly governed. In the context of Gennan pleonastics, this

accounts for the facts of FS/O.

(83 ) [;:; \ ww:de [vp getanzt]

(84) Heute wuroefesL[vp getanzt]. <. 0\

In (84) the 0 pleonastic is properly governed and is identi.fied by.r#'

features on INFL. 'therefore it is allowed to surface in its 0 foIm...· In

(83), the position is not properly governed and must be fi~l~ by es./" -

,,,"

What \tee want to explain rr:M is 'Nhy the ES pleonastic (I-type) appears in

255

(82). AccoroiD3 to our analysis, the I-type ple::>nastic appears even in

1:he {X)sition of proper goverrme1t. 'lh:ls is what distingushes it fran the

T-type pleonastic •

.It nt:JW is clear why \tJWa CC'nBJt use S:t fir I s B:P account to explain

these facts. He must assune that subject {X)sition is never properly

governed in German. '!his ap,pears not to be the case as shown by the lack

of that-t effects in the dialects of Gennan which allO'N extraction out of

daB ~lauses.

Independent of our analysis of German, hO'itever, safir I s analysis does

IX)t sean 1:0 be the right one. In Irish, the subject {X)sition is al\\1aYs

properly governed (it has no that-t effects, see Olapter 4). Yet the facts

are the sane. se is obligatory \then the subject {X)sitio~is·-assigned

case, o1:herwi.se it uses the 0 T-type pleonastic.~ ." 'ltiis offers further

encouraganent to search for alXlther .~ut.ion.

5.4.2 HE'!' vs. ~; EUtCh",.".,...;

....

...... .- ... '!here are sane interesting data fran D.1tch that provide us with serne

clues as to the solution of the het/er constructions. As \\e have srown

above, sane Dltch examples can use either het or ere

(85) Het/Er 'ItOrdt betreurd dat. hij bet boek gekocht heeft.lJecane regretted that he 1:he boo~ bou;}ht has

I It \tIaS regretted that he ooU3ht the l:x:x>k. •

'!his is a V(~asive)S' construction wch allOhS either the I-type or

the T-type pleonastic. ~s \\e see by the F.n:Jlish translation, in English

only the I-type is all~.

-.,...../

256

As in Fn31ish, when the verb is active, the 8' can be right

dislocated leaving a copy it.

(86) ik betreur dot hij het boek ge~ht heeftI regret that .he the book boU:Jht hasII regret that he ha~.ool¥3ht the book. I

(87) ik betreUr het dat hij het~ boek gekocht heeft~/. regret it that he the book OOl¥3ht has

.. . II recJret it that he has bo~ht the book.'

Also, as in Fn31ish, extraction is J;Ossible only out of the S I which

is not right dislocated.

(88) Wat betreur je dat hij gekocht heeft?\-Jhat regret I that he l:x:lt.ght has

'What do I regret that he has ootght? I

(89) "Wat betreur je het dat hij gekocht heeft?what regret I it that he ooU:Jht has

* '"What do I recJret it that he has OOl¥3ht? I

rLhis raises a question of whether the passive is the passive of the

right dislocated strlrture or of the active structure without het. By

using the extraction facts as a diagn:>stic, \tie can see that the het fonn

of the passive appears to be a right dislocated constrtrtion mi.le the .!£

pa~sive' appear to be the true passivized fonn of the V-S' cons1:rlX:tion.

(90 ) Wat w:>rdt ex: betreurd dat hi j ge](c)cht heeft:what becane regret that he ool¥3ht has

Ilmat is it regretted that he bol¥3ht?'

(91) *Wat 'ItOrdt het betreurd dat hij gekocht heeft\tklat becane it regretted that he bol¥3ht has

'What is it regretted that he OOl¥3ht?'

(S2) Bet \\{)rdt geloofd dat hi j het boek gekocht: heeftbecane bEalievEd that he the book bol¥3ht has

'It was believed that he has ootght the book. I

257

(93) *Wat WJrdt het gelc::ofd dat hij gekocht heeft.?\\hat becane believed that he bou;Jht has

'What was it believed that he has OOtr3'ht?'

'!he prelimi.nary conclusion to draw, then, is that the het form of the

V(passive)~' con~tion is always a right dislocated form C'CC()untiD;l for

the i.n'tp)ssibility of extraction fran the S' s9

•

'Ibis analysis \\Ould involve assun..i.ng that all V-Mj-S· constructions

are also £ODDS of right dislocation. As \..e see belOll, extraction is I"X)t

possible in these constructions either.

(94) *Wat is het duidelijk dat hij gekocht heeft?What is it clear that he bc::>u3ht has

'What is it clear that he oou:Jht?'

BelOil is an analysis Wich relies on the fact that S' extraposition

fran subject creates a structural arri=>iguity bet\lleen the extrapJsed am the

right di.sl~ated strtrture.

5.4. 3 Analysis

'!he problan \Ie \\ent to solve is why English uses I-type pleonastics

9. '!he use of het is easier with factives weh is not surprising sincethis is also the case in direct object position in Ehglish as well asIl.1tch. '!he use appears to improve with lDn-factives in certainconstructions.

(i) \terd regretted it tret he didn't get to !:Oland.(ii) ??J.;anet believai it that IBn ccx>ked for Sarah.(iii) Janet couldn't believe it that ran ccx:jked for sarah.

258

in V(p6ssive)-6 1 const.Iu:*ions, mile Gennan am IXltch uses T-type

pleonasticslO• ']here are tw:> i.m.Fortant subparts to this problem. Qle

~ncerns the passive am one c:oncerns the role of argllllent. S' s. '!he

solution involves tying the fact that Ehglish differs fran Gennan and

Il.1tch in this question to the fa~ that English does rx:rt:. have impersonal

pC'ssives mi.le Gennan and Illtch do.

In the first part of the argunent, we will discuss qu.i.ckly the

constraints on passive in the lan:JUages beiBJ investigated. 'Ihen \tJe will

examine the p:>sition of argunent SiS. We will relate these 1:\\0

ooservations to a IX>ssible solution. Finally \tie will proIX>se an

explanation for the extraction facts mentioned in the previous section.

5.4.3.1 Passive

Generally, it is assuna:l within GB theory that passive movanalt is

tri93ered by the fact that passive ITOrphology absorbs the accusative case

assigning abilities of the verb. ']he object, unable to be assigned case

in the object IX>sition, must move to the subject IX>sition lNhere it is

assigned rxmi.native case. ']he subject p:>sition will be anpty since

passive verbs do rx:>t assign external 6-roles. What has been focused on in

this accolBlt, is the fact that passive verbs do rnt assign case.

let us instead assune that the central fact is that passive verbs n:>

10. I thank !any Jaspers in particular for help in thi.s problane

259

longer assign external o-roles (see lbthstein 1983 ~ for a similar

analysis) 11 • As 'Ne have seen in languages such as Gennan, Illtch, and

Irish, rx:>t all passivization involves vems which, in their active fonns,

assign case. By saying that passivization is the absorption of an

external 8-role12'Ne can sIS) capture the fact that tb::>U3h GeIman and

IUtch can passivize intransitives (95), they cal'UX)t passivize unac:cusative

verbs (96).

(95 ) Es \lJUI"de getanzt.there \telaS darx::ed

* ·'lbere \tIas danced.'

(96 ) *Fs wurde ge:Jsngen.there '113 S gone

*·'!here 'lIB S gone.'

Since lIDaccusative verbs do rot assign external 9-roles, it is not

surprisin:J that they do rot urrlert;P passivization.

Ot:.her restrict.i.ons must be placed on the application of

flC'ssivization, ~er. languages differ in interesting \\9YS. Gennan am

nJ.tch a~ar to have the freest fo~ of passive since any verb with an

external 9-role rrey passivize. Eh;lish appears very restricted. N:>t only

must the verb also assign an internal e-role, it must assign this 9-role

directly (i.e., IX> prePJsition). I will express this generalization by

stating that passivization in Ehglish requires that the V cpvern an

internal NP or S I argunent.

11. I have benefited greatly fran discussion with Mark Baker and MrianaBelletti on this question.

12. see lbberts (forthcaning) for an interesting account of this effect.

French, tNhile rnt as free as GeIman, is also rx:>t as restricted as

Ehglish. French may not passivize verbs with no internal argunents. It

may, 1:Dwever, passivize verbs which take prep:>sitional argunents. sane

relevant examples are given belOVl.

(97) lD internal argunents:Gennan:

Es lNUrde getanzt.there tIf9s . danced

* ''!here \..es danced.'

French:~ L /

*11 a e't:e danae.there \tes danced

* ''!here \tBS dailcm.·

Ph:31ish:*!bere wa 5 danced.

(98) pre};X)sitional internal argunentsGer:man:

Es wurde tiber diese Fragen diskutiert •there \ISS CNer these questions discussed

* ''!here \tees discussed about:. these questions. I

French:I1 a ete tire sur le bateau.there 'lteS s"OOt the \:oat

* ''!here \\as shot the boat. I

Ph:31ish:*!bere \ISS talked about the question.

(99 ) direct internal argunentsGennan:

IBs Buch wurde gekauft.the 1:x:Dk 'ISS bou:Jht

I '!he 'book 'lIBs botght. I

French:I.e livre a et1! achet6.the 1:x:Dk 'It9s botght

''!he book w:J S bo\.J3ht. •

Fnglish:'!he book 'IIa S botght.

260

261

The result of this is that Gennan/rutch differ fran Erglish in that

they can passivize verbs \\b..i.ch have no argunents while Ehglish can

passivize only verbs which take direct internal argunents. '!his leads us

to the question of what status SI s have in V(passive)-S I L"'Ot1Struct.i.ons,

aOO what status they have in general.

5.4.3.2 Argument SiS

Stowell (1981) claims in his Case Pesistance Principle that SiS must

trOVe out of case-marked IPsitions. Altlntgh this is, at first, very

appealing, there are instances \tilere case assignnent cannot account for

all the facts. In Gennan, S· s do trOVe out of the direct object IPsitien.

&tt, as \..e can see belOll, they trOVe even in passivized st~ures which

presunably do tnt assign case.

(100) Ieh glaube, daB Paula gekcmnen ist.I believe that Paula cane is

II believe tl1at Paula has cane.'

(101) Es wurde geglaubt:., daB Paula gekatmen ist.there \\laS believed that Paula cane is

*·There was believed tha't: Paula has cane.'

Let us say instead, that (for sane reason) S' s cannot ranain in

argunent IPsitions. This \\Ould mean thi3t at d-strtcture, the

configuration \«)uld be as in (102 ) and at. s-stru::ture, a s in (103 ) •

(102)

(103)

Gennan/IArt.ch

Erglish

Gennan/nJtch

Erglish

[VI? 8' V]

[vp V S']

[vp t V S']

[vp V t 5']

262

'!he structures for EZglish \\Quld be as in (104) aOO for Il.1tch and

Gennan, (105).

(104)a·NP~

b.~

NP I VP

V~SI ~~

V t S'

(105) a.~

b. I'

NP~VPNP I A ~5' V t V 5'

I am asstming that though the 8' m:wes out. of the'argunent I;Osition,

it is still in the canplement danain of the verb aOO is lexically governed

by the verb. 'Ibis will be i.mp:>rtant for our account of the extracti.on

facts below. I am also assuning that the internal argunent .[X)sition which

is cru::ial in the characterization of the passive is the pJsition adjacent

to the verb. It is this .[X)sitien \fJhere case will be assigned aOO this

.[X)sition in \fJhich direct internal argunents are generated at d-str~ture.

'!he restriction 1Ne have pJsited for passives in EZglish is that the V

must govern a direct internal argul1E!lt \fJh.i.ch ~uld me6n that only (104a)

w:>uld be able to urrlertF passivization. !n (104b), the direct internal

argl.ment I;Osition is empty.

N:M l~.: us say that passivization has the effect of forcing lTlClVanmt

out of this direct internal argunent. .[X)sition. Focus NPs lllC'y lTlClVe to a

focuSed NP pesitien.

(106) '!here \..as discovered e. un:1er the tree [a great treasure].1 . 1.

263

Since this {X)sition is l'X)t available for S' s, they move to Subject

IDsition.

(107) '!hat Gideon \tJOuld sail the boat 'NaS expected.

D.1td1, then,. has the cooice of either IlDV'in;J the 8' out of the

argunent IDsition creating a {X)ssible structure for an :iJnI:ersonal passive

usiD3 er, or, like in Fnglish, the S' IlDV'es to the subject IDsition.

Nlen the SiS IlDV'e to the subje::t IDsition, the stnrtures are as

given below.

(f108) En9liSh:~

S'. I ~

1. / "-.....V t.

1.

Il1tch: AS'. I VP

1. I ~t. V

1.

N:1N they lex>k like their V-Mj COtJrlterparts at d-strt.X:ture.

(109) English: II

~S' I A

V An·\ -_,Jis clear

Intch:~

S' I Av' "Mj

ist duldelijk

We need t\tO rrore assunpt:.ions to alse> account for the extract.ion

facts. let us say that \tk1en an S' extraIDses fran subject IDsition, it

must 99 to the periphery of that 5'. Where it attaches depends on the

goverrment (or in terms of this thesis, the headedness) of the laD:JUage.

I pro{X)se that the S I \Illll ~ttach to the laest node \#th.:ich is al101.ed by

goverrment. we see belqw what this Illt':!ans.

(110) Eh31ish: II

~NP I VP

V~SI

264

J:Utch: I I

Nt'~--w--.-S I

~NP V

'the BrrcJ\tJS sl'ow the direction of governnalt of each canstituent • In

Il.1tch, since only I is head-initial, the S' must attach to I I. In English,

b:JINever, since VP is also head-initial, S I rooy attach to the VJ?13.

FurthE:r, let us assune 'that an S I III.1St be lexically governed by the V

in order to be 8n extraction danain. '!his revises our view of le)Cical

govenment in C1apter 4. let us lo:>k at three types of constituen"t:s within

the VP: S· s that are internal argunents to the V (111), s•s that are

eX1:rG'J.X)sed external argunents of the VP (112), and adjuncts that are

licensed by predication (113).

(111) sara said 'that lew put M&3x in the car.

(112 ) It is clear t:hat r.ew put Max in the car.

(113) Donna ate the fish ste\ed in curry sauce.

Tne first tw:>, but rxrt. the last, are J.X)ssible extraction danains •

(111 I ) In ~t car did S3ra say that Lew put Max?

(112 I ) In \ffhat car is it clear that I.ew put M3x?

(113') *In ~t sauce did DJnna eat the fish ste\ed?

All of the constituents are st..nx:turally governed by t~ V, am only

the SI in (111) is both canplement governed and lexically governed by the

v. Neither canp1anent governnent rx>r a struct.ural rntion of lexical

13. Peinhart (1976) argues that extrap:>sed SiS attach to the VP.

265

govarnnent makes the cx>rrect prediction. we want (~.12) to pattern with

(Ill) rather ~..han with (113). lhe diff~rence bet\\een (112) and (113) is

that the S· ill (112) is assignErl a 9-role by the rnE!OOers of the VP t;;hile

the adjunct in (113) is a preiicate licensed by pra:lication coindexation

with a canplanent~ '!his is slnwn in (114) belCM.

(114) a. Extrap:>ss1 5'~I

VP

~Sle:lear that s

b. AdjlJl1Ct

As v.e see in (114a) al:xJve, the S· is governErl both by the constituent

which has the O-:role to assign and by a lexical category. Unlike the case

of an internal argunent S, the\E t\\O functions ar~ rot carriErl by the same

elanent.

As we have saw. in Clapter 4, for ~p effects, the empty category

must be canplanent governErl, while for CEO effects, the danauJ must be

both canplement governed and lexically governed. lhe refinana1t \Ire have

made here is that in the case where an elanent is both lexically gcwerned.

and canplement governed, it is not r:.ecessarily the same corstituent which

is resp:>nsib1e fox: 00t1l.

lhe above sketch of an analysis seeks to answer t\\O quest.ions.

266

(1) Why does I:utch all~ T-type pleonastics, while English must use

I-type pleonastics, with constr~tions \'Jhere the S I is base-generated VP

~ternally? 'Ihese <Dnstructions are the V(passive)-5 1 and the

V( raising) -5 I constr~tions •

(2) \'by does Il.1tch not allow extraction fran any const~~tion using

an I-type pleonastic, whi.le Englisll does?

(1) In the analysis, I have related the 9resence of the T-type

pleorastic in Gennan and IUtc~ to the fact t'llat these languages allOl/

impersonal passives. If the 8' is not in the argunent fXJsit.i.on, English

verbs may lXlt be passivized si.rx:e they l1UJst have internal argunents to

passivize • GeIman and Ilttch verbs may passivize, 'ho\ever, and since no

argunent will be in a chain with the subject fXJsition, the T-type

pleonastic is used.

If the S' is in argunent fXJsit.i.on (as it must be in Ehglish and may

be in Gennan or IlJ:t:.ch), then the S I will rocwe to subject fXJsition in the

passivization. process. tbw the passive st.r~ture will be Ifr~ the Mj

structures si.rx:e lJoth will involve extrBfXJsition fran the subject

[Ceit.i.on.

(2) In the extraIXJsed stroctures (t.lx>se that use the I-type

pleonastic), because of the di.ffererx:es in headedness, GeIm3n and DJtch

S •s will attach to I' Wrlle Ehglish S' s will attach to the ,: nus will

account for the di.fference in the extraction fXJssibilities. AsslJlling that

right. dislocated SiS at.tach to I I, the fact. UlBt the extraction

fXJssibilities of extraIXJsed SiS in GeIm3n reflect the e."rtraction

267

p:>ssibilities of right dislocated structures canes as rx> surprise.

M3ny questions are left lmanS\lJered· in this chapter. Why do Ehglish

raising verbs take I-type pl8?nastics? fbw does Irish fit into this

typology? \'bat are the facts of German concerning the use of I-type

pleonastics ani extraction? Ibw do case assignnent, }?Ossivization, S I

mJVanent, and pleonastic cbJice interact? '!hese I leave for furtheL~ stooy

rather than \dlder specula1:ion.

268

In this chapter, while I review several issues raised in earlier

chapters of this thesis, I also give an overview of sane ideas that appear

scattered througoout.

6.1 Ibnains

In Olapter 2, the IXrnain Adjacency Condition (DAC) waa L~troduced to

account for certain generalizations a1x>ut \\Ord order at d-structure. We

also saw in Olapter 4, however, that the me reappears in the issue of

proper governnent at both s-structure and LF. lXJnains, then, play an

i.rrqx>rtant role at all levels of syntax.

At d-structure, by positing the me and using the notion of danains,

tNe find we no 'longer need to account for \ItOr~ order througl, phrase

strJ.cture rules. If a CX)nstituent is depement on another ccllStituent for

case or o-role, then a danain is set up and the D\C canes into effect. We

have also Seell that directional parameters involve danains. In fact, they

may 'be reworded using only the definition of danain and the

269

head-final/head-initial parameter. 'Case assignment tCJ the left· ma.y be

rephrase:1 as 'ease danain head-final'. D:mains, then, are established at

d-structure by lexical specifications of heads, or, as we have seen in the

case of VPs, by the 6-gI:id of a nBxi.mal projection.

'!hese dana.ins are also crucial at a-structure, as the Projection

Principle \\Ould predict. As lexical properties, th.~y must be satisfied a~

every level. ~f the gramnar. '!he dana.in relations, then, must be

transparen~ at s-structure as 'Nell. If a case-marked or &-markedf

constitu~t is ItDV'ed, it must be coindexed with a trace in the necessary

danain. '!his coindexation is necessary for t'ttO reasons.. '!he constituent

itself must be able to retrieve the case and./or &-role that was assigned

to it, and the empty category that is left, though properly governed by

virtue of 'baing in the danain, must be able to retrieve the feat\lres of

its content.

In Olapter 4 we have seen lnw danains are i.InI:ortant for the notion of

proper goverrment. An empty category within a canplanent dana.in is

properly governed. '!he distinction of being within a ckma..i.n or not is

clearest with subjects which may receive their 8-role by coindexation ll1ith

a predicate, or by being in the canplement danain of the VP. Only in the

latter case is that subject !XJsition properly goveITled.

'!he same B:P effects appear at IF, as s11o'wn in Olinese. Since

subjects in Chinese are within the canplement dana.in of the VP, they may

be extracted at IE without violating the EI:P.

IXmains are rrost clearly outlined in VPs and. NPs. At the level of I'

270

(S), CroSSlllg dependencies are established, creating a tension which

affects \\Ord order and government properties. In C1apter 2, \tie discussed

the different relationships anong the three members of I I: 1°, NP, and VP.

Unlike the head of the VP, the head of I' does not roth assign case and

8-role. IO assigns case to the NP, while the VP assigns the 8-role to the

NP. '!be canplanent of the head, la, is the VP, not the NP. Obviously not

all of these relationships can require adjacency, arxi, in fact, there are

alternatives • INFL may assign case through the Obligatory coindexation of

!NFL with the subject NP, and the VP may assign its o-role through the

coindexation of predication. Since neither of these relations requires

adjac,-'ncy, all of the requiranents nay l:Je met. In Chapter 3, ~.Mever, we

shatl that it is exactly these crossing deperXiencies that e11courage

reanalysis of s-stt·uctures as d-structures. It is also the difference in

relations at the level of I' that create the distinction l:Jet\tJeen properly

governeq. subject p:>sitions (those in the canplanent danain of the vp) I and

not properly gO'Verned subject fX1sitions (those which receive their 9-rolp

~'u:ough coindexation \,nth the vp).

6. 2 Imp?rtance of !NFL I I

/

we have just discussed the role of !NFL in the l'J.storical developnent

of lar¥3Ua9es. Since so many demands are placed on !NFL, a-structure

variations are easily reanalyzed as d-structure w:>rd orders. '!his

reanalysis, b:M'ever, must l:Je triggered by an s-structure variation and

!NFL plays an i.mp::>rtant role here as \tIell. By the Head tebvanent

271

Constraint of Olapter 3, INFLa may front into CCMPo. If the structure

created by this fronting rule is able to satisfy sane of the danands on

!NFL, i t ~cay then be reanalyzed as ad-structure.

!NFL also plays an important role in 'ttOrd order typology. With only

S, 0, and V, Gennan and Japanese have the same word order (S-o-V) and

Fngliscl and Irish have the same 'ItOrd order (S-V-o). With the addition of

INFL to the inventory of canstituents, obvious distinctions in these

larguages may be accotmted for. Gennan is S-I-o-V, while Japanese is

S-o-V-I. English is S-I-V-o while Irish is I-S-V-o.

'!his leads to other questions. '!here appear to be similarities

bet\tJeen verb-final (Japanese, Turkish) and verb-initial (Clarrorro,

Malagasy) lan:JUa9es. '!hey all tend to be pro-drop lan:JUages, without

pleonastics, and with IX) that-t effects. With the addition of !NFL to the

list of relevant categories, one question is: does V-peripheral

(V-final/V-initial) or INFL-peripheral nore accur2.tely describe this class

of languages? '!he secorrl question is: Why does this generalization

exist?

we can anS\\er the first question by looking at the four languages 1Ne

have just discussed: Gennan, Japanese, English and Irish. Japanese and

Gennan are V-peripheral (in an S, 0, V typology) while Japanese and Irish

are INFL-peripheral (in an S, 0, V, I typology). It is the latter class

that has the characterisitcs of being pro-drop, having rn pleonast.ics, and

no that-t effects. Gennan confenns only in its I-S-Q-V s-structure

variant, which is also INFL-peripheral. '!he correct characterization,

272

then, depeOOs on the IXlsition of INFL, not of V.

In answer to the second question, the i.rnp:lrtance of the IXlsition of

!NFL foll~ fran the notion of VP goverrment presented in Chapter 4 •. If

!NFL is on the periphety of I', then the subject IXlsition is adjacent to

the VP. In this IXlsition, the subject IXlsition is properly governed (by

C'atlplement government), and the characteristics will follow. Proper

goverrment will allow the appearance of pro, and the trace of rocwanent.

While head-initial languages have wh-m::wanent at s-stI11cture (01arrorro,

Malagasy, Irish), and head-final langua.ges have wh-novenent at IF

(Japanese, Chinese), since the EX:P is a condition of 1:xJth s-stru,-~ure a11d

IE, INFL-peripheral effects will appear in both groups.

6.3 ~rd Order Effects

A final observation is that \ttOrd order, independently of \tihether it

is base generated or created by IlDVanent, or whether it is semantically. .

marked or not, will have an effect in other canponents of the gt.-amnar. We

have seen that sane larguages such as Irish am ChamJrro base generate

subjects within the canplement danai.n of the VP. Such a \\Ord order is

unmarke:1 in these languages, both structurally and semantically. In

Italian, IXlst-verbal subjects are base generated within the canplement

danain of the VP, but the configuration is sanantically marked ~ince it is

. used to focus the subject NP. In Gennan, the subject NP nay appeat- within

the canplanent danain of the verb at a-structure either because !NFL has

273

IOOVed to CCJ.1P, or because INFL is 1X)t phc>netically realized. What is

i.mp:>rtant, thoU3h, is not how or \tJhy the \\Ord order came about, but rather

that, at s-structure, the subject NP is within the canplement danain of

the VP, and is, thereby, properly governed.

Syntax is the stuiy of h:Jw elanents are ordered, \tJhere "order ll means

both sequence (precedence) and organization (dcminance). '!his thesis

discussed h:Jw la.R3\lage. specific \\Ord orders might be acquired, and h:Jw

grannars might change. '!his involved pJsiting a system of directional

parameters and universal constraints on word orders that restricted

fXJssible d-structures. '!he thesis also discussed the effect of \\Ord order

on other eatqX.)nents of the gramnar such as f.bcle-a and the EX:P. '!he

concl~ion is that \\Ord order is I'X)t a stipulated surface filter with no

effect outsiGc of PF, nor is it a simply a catalCXJUe of d-structure phrase

structure rules. '!he claim is that directional pa.rameters which create

d-structure 'NOrd orders, and llO\7ement rules which create s-structure word

orders both have effects that go far beyooo. a simple linearization of

elEJIlP..nts for the prcxiuction of speech.

ABBREVIATIONS

Glosses:

3M - 3rd person marker

3S - 3rd perscn, singular

N:-C - accusative

aor - aorist

APPL - applied verb

ASP - aspect

AT - Actor 'lbpic (Malagasy)

SA - Chinese, object marker

CAl5E - causative verb

CL - classifier (Chinese)

OONDIT - CX)nditiona!

o (~rd) - D.1tch canplanentizer

DA.T - dative

IE - Chinese, canplementizer

-FThl - infinitival (Irish)

ERG - ergative (warlpiri)

IE - DItch, subject clitia 'he'

IMPER PAST - impersonal past

INF - infinitival

L - linker (Olanorro)

IN - linker (Malagasy)

IDe - IDeative

to! - naninative

NPAST - oon-past (warlpiri)

274

PAST IMPER - impersonal past

pre- - prefix

PRES - present

PRr - p:1rticle

PI' - Patient topic (Malagasy)

PVP - pre-verbal-particle (Welsh)

s~ - 1st person, singular

52 - 2nd J?erscn, sinJular

-suf - suffix

'l'RAb5 - transitive

VN - verbal notm (Welsh)

YN - Welsh, pro:Jressive particle

'!heary:

[+c] - an NP in a chain with [+c] I'lBy be case-marked without beinglexicalized (see Olapter 5)

[+#] - an NP in a chain with [+#] nay license an S· argur ent throughcoindexation (see Chapter 5)

[ +nllt1ber/ -nllt1ber] - pIural/singular, fran Pc>llock (1982) (see Olapter 5)

[+person] - an NP in a chain with [+person] may receive a o-role

A position - argunent position

A' position - non-a.rgunent position

A&S - .Aolnl and SpJrtiche

AS: _. Archaic Chinese

af - affix

CEO - Condition on Extraction D:mai.ns (Huang 1982)

C1 - elitie

CCMP' - is the same as S I in I.GB

275

DAC - IXJnai.n Mjacency Condition

E - extelnal argunent

e I ec - anpty category

EX:M - Exceptional case Marking

EI:P - Empty Category Principle (ctansky 1981)

LF - lII.cc3ical Ebnn"

1GB - Lectures on Goverrment am Biming I C'hansky 1981.

GB - Government and Biming (Theory)

GF - gramna.tical function

GF-8 - GF-t,.~eta, the gramnat..ical function \fthich bears a o-role

I' - is the~ as S in LGB (except in a system with a t'NO-bar level ­-then III is the same as S)

I-type - I-type pleonastic (like English •it')

l&T - Li aI".d 'lhanpson

r+1 - MJdern Mandarin (What is camon bet\tJeen r+11 and r+12)

r-fA1. - 1st stage of r+1, described -by Light, and L&T

r+12 - 2rrl stage of r+1, described by L&T and Huang

o - Object (as in S-o-V, Subject-Verb-Object)

e - ·theta ., as in &-role, e-rnarking

o - null, as in • a 0 T-type pleonastic, i.e., pro.

oa:; - Old High Gennan

PF - J:i'lonolCXJical canp::>nent

R&V - R:>uveret and Vergnaud

S - subject (as in S-o-V, SUbject-Verb-Chject)

S - sentence (as in [NP,S] )

SA! - Subject Awe Inversion

276

subscripting - used to irrlicate the trace of I'OClVanent or coreference

superscripting - used for predication, coJ.ndexation of !NFL with [NP, S],and coindexation of a VP internal argunent with the subject p:>sition.

t - trace of I'OClVanent

T-type - T-type pleonastic (like 'there' in English)

V2 - verb-Sec.Dnd effect, such as in Gennan

277

278

BmLIcx;RAEHY

.Aotm, Y. (1982) "'!he Ebnnal Nature of Anaphoric Relations,lI PhDDissertat.icn, MIT.

(1983 ) "Gene:r:alized Binding in O1inese II InS. tEe.----an:l D. Srx>rtiche (1983) liOn the farnal theory of government II

--Lin~·-gm.-.-oIIIlI:-stic Review, 2. 3.

Bach, E. (1962) "'!t1e Order of Elements in 3. Transfonnational Grammar ofGennan," I.angu3.ge 38.

Baker, M. (1983) "NJun Incorporation in Iroquoian" ms. MIT.

Banfield, A.. (1973) "Narrative Style and the Granmar of Direct andIndirect Speech" Foundations of Iangua.ge 10.

Barss, A. (1984) tlClain Biming" syntax generals paJ?er, MIT.

Belletti, A. ( forthcaning) IllS. MIT.,\

and L. Rizzi (1981) "'!he syntax of I ne': sane theu..::aticaliraplicatiOns," '!he Linguistic Revi~ 1. 2.

Besten, H. den (1977) "Ch the interaction of r<X>t transfomations wi!lexical deleti""e rules," in W. Abraham, ed. en the Ebnnal Syntax of theWestge~ John Benjami.ns, lYnsterdam.

Bierwisch, M. (1963) Gramna.tik des Deutschen Verbs, Studia GraIl1l\9.tica II,Berlin.

B:lrer, H. (1981) "Parametric Variations in Clitic Constructions," PhDDissertatial, MIT.

Bouchard, D. (1982) "On the Content of Ehlpty categorieo," PhDDissertatiCXl, MIT.

Burzia, L. (1981) "Intransitive Verbs and Italian Auxi.liaries, II PiillDissertation, MIT.

calabresi, A. ( forthcaning) InS. MIT.

Chao, Y.R. (1968) A Granmar of Spoken Olin~~, University of ca.l;lforniaPress, IDs An3eles, CA.

01ansky, N. (1965) Aspects of the rrheory of Syntax, MIT Press, cambridge I

MA.

(1970) "Remarks on N:mina1ization, II in R. Jacobs and P.Ibsenbaum, eds. Readirgs in English Transfonnational Granmar, Ginn,Waltham, MA.

279

(] 980) "Ql Biming" rJ.n:Jll"lsti~ Inquiry 11, 1-46.

(1981) Lectures on Govennent and Binding, Feris Publications,J:':>rdrecht, fbllaM.

(1982) sane Concepts and Consequences cf the 'l'leory of--Go-iVe-rnm--en-t am BWIiij:, MIT Press, cambridge, MA,.