SPEC of IP and SPEC of VP: Two Subjects in Austronesian Languages

Author(s): Eithne Guilfoyle, Henrietta Hung, Lisa Travis

Reviewed work(s):
Source: Natural Language & Linguistic Theory, Vol. 10, No. 3 (Aug., 1992), pp. 375-414
Published by: Springer
Stable URL: http://www.jstor.org/stable/4047773 .
Accessed: 04/01/2012 15:57
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Natural Language & Linguistic
Theory.

http://www.jstor.org

EITHNE GUILFOYLE,

HENRIETTA

HUNG,

AND LISA TRAVIS

SPEC OF IP AND SPEC OF VP:

TWO SUBJECTS IN AUSTRONESIAN LANGUAGES*
In this paper we claim that certainfacts from four Austronesianlanguagesprovide
strikingsupportfor the claim that there is a VP-internalsubjectpositionin addition
to the traditionalsubjectpositionof SPEC of IP. We show (i) that subject-sensitive
propertiesmay be splitbetweenSPECof VP and SPECof IP, theta-sensitiveproperties (such as bindingand control) associatedwith the formerand structure-sensitive
properties(such as extractionand quantifierfloat) with the latter; (ii) that word
order facts about these four Austronesianlanguagesas well as variationsamongthe
languagesfollow from the assumptionthat the two subjectpositionsmay be licensed
simultaneouslyat S-structure.Lastly,we showthat the proposedanalysisof Austronesian phrasestructureallows the placementof Austronesianlanguageswithina typology of passive.

In the past few years, a number of studies have taken advantageof the
uniformityof the X' system of phrase structureto arguefor the existence
of two subject positions. More precisely, they have provided different
types of evidence for establishinga VP-internalspecifierposition in addition to the traditional SPEC of IP position. These include Kuroda
(1988), Kitagawa (1986), Fukui and Speas (1986), and Koopman and
Sportiche(1988).
In this paper we present furtherevidence for postulatingthe existence
of two subject positions using data from four Austronesian languages:
Malagasy,Tagalog,Cebuano,and Bahasa(Malaysiaand Indonesia).1The
main attractionsof assumingsuch an analysis are (i) that it suggests a
structuralexplanation for the split in subject-likeproperties which has
been a traditionalproblem for the analysisof subjectin these languages;
* The researchpresentedin this paper is the result of a trulycollaborativeeffort, and the
authorsowe a debt of gratitudeto manypeople duringthe writingof thispaper.The authors
wish to thank the audiencesat McGill University,BrandeisUniversity,WCCFLVIII and
GLOW 1989, where this materialhas been presented.We have receivedhelpfulcomments
from the other membersof the Malayo-Polynesiangroup at McGill, Mark Campanaand
Anna Maclachlan;and from SandraChung, Hans den Besten, Carol Georgopoulos,Eric
Hoekstra,RandyKamp,Paul Kroeger,BarryMiller, George Rebuschi,Paul Schachterand
three anonymousNLLT reviewers.We would also like to thank Zofia Laubitzand Ben
Shaerfor editorialhelp. All errorsand omissionsare the responsibilityof the authors.The
researchfor this paper was supportedby FCAR grant #88EQ3630, and SSHRCCgrant
#410-87-1071.
' 'Bahasa' literallymeans
'the language (of)' so the terms Bahasa Malaysiaand Bahasa
Indonesiarefer to the standarddialects of these two countries.Although there are some
differencesbetween the two dialects,the propertieswe discusshere are commonto both.

NaturalLanguageand LinguisticTheory10: 375-414, 1992.

? 1992 KluwerAcademicPublishers.Printedin the Netherlands.

376

EITHNE GUILFOYLE

ET AL.

(ii) that it accountsfor the facts of word order (and word ordervariation)
among these four languages;and (iii) that it provides a place for these
languageswithin a typologyof passive.
At the heart of this analysisare the assumptionsthat both IP and VP
have specifierpositions, that all argumentsare base-generatedwithinVP,
and that the SPEC of IP is a non-thetaposition. If an argumentis Casemarkedin its theta position, it may remain there; if not, it must raise to
the SPEC of IP position where it is Case-markedvia SPEC-HeadAgreement with INFL.
In addition to providingan account for word order facts, this analysis
provides strong supportfor our original assumptionthat all theta assignment is carried out within VP at D-structure. Finally we extend this
discussionby examininghow Austronesianlanguages fit into the phrase
structuretypology of naturallanguages.
1.

THEORETICAL

BACKGROUND

1.1. The CanonicalPosition of the SubjectNP

WithinGovernment-Binding(GB) theory, one way of accountingfor the
behavior of subjects as opposed to objects has been to represent the
subjectas being 'external'to the verbphrase.The subjectreceivesits theta
role from the VP (ratherthan the verb), and is immediatelydominatedby
IP (or S in earlierversions of the theory). This kind of analysis,which is
assumed in Chomsky (1981) and most recent work in GB theory, is
illustratedin (1) below:
IP

(1)
SPEC

I'

Agentj

IN~FL

VP.

[te]

NP

emei

The subjectpositionreceivesits theta role from VP via predication(see,

e.g., Williams (1981)), while the object is directly theta marked by the
verb. As the subject position is not necessarilytheta markedby the VP,

TWO SUBJECTS

IN AUSTRONESIAN

LANGUAGES

377

it does not always received a theta role. Thus, unlike any other position
within IP, the subject position may be the target of NP movement, or
may be occupied by an expletive NP.

1.2. The VP-InternalSubjectHypothesis

While the phrase structureshown in (1) above is still widely accepted, a
numberof researchershave proposedthat subjectsare actuallygenerated
within the projection of the verb phrase, and raise to Spec of IP at Sstructure.This proposal is often termed the VP-internalsubject hypothesis.
Kitagawa(1986), extending an idea of Kuroda's(publishedin Kuroda
(1988)), proposes this kind of analysisfor English. Using evidence from
extraposition,quantificationfacts, and constituencytests, Kitagawaproposes that English sentences have the structureshown in (2) below:
max

(2)

imed

NPi

Vmax

INFL

Vmd

NP
ti

NP

In (2) the verb phraseis strictlyhead-initialand the subjectis generated

as the sister to Vmed. The subject raises to the Spec of IP at S-structure
in order to receive Case from INFL. A 'subject-raising'approach to
Englishclause structurehas also been adoptedby Koopmanand Sportiche
(1988) and Fukui and Speas (1986). Koopman and Sporticheshow that
various types of extraction and quantificationphenomena can be accounted for under the VP-internalsubject analysis.Although the various
analysesproposedby the authorsmentioned above differin many details,
they all assume that English subject NPs raise from a position inside VP
to the Spec of IP position, and that this movement is motivated by the
Case requirementsof the subjectNP.
SPEC of IP receives Case at S-structure,and thereforethe subjectmust

378

EITHNE GUILFOYLE

ET AL.

occupy this position at S-structurein order to receive Nominative Case

and so avoid violatingthe Case Filter. However, Kuroda, Kitagawa,and
Koopmanand Sporticheall propose that subjectraisingis not obligatory
in all languages.Kurodaand Kitagawasuggestthat Japanesesubjectsmay
remain within the VP at S-structure.Koopman and Sportiche make a
similar claim for Italian and VSO languages. The analyses proposed so
far, then, assume that languagesvary with respect to whether the VP or
IP subjectposition is filled at S-structure.
In this paper we explore anotherlogical possibilityprovidedby the VPinternalsubjecthypothesis;namely, that languagesmay allow both subject
positionsto be filled at S-structuresimultaneously.The assumptionof this
possibilityallows us to explainmanypropertiesof subjectsand wordorder
in these languageswhichotherwisedefy readyexplanation,and to sharpen
our notion of the differencesbetween these two positions;thus it provides
strong evidence for the VP-internalsubjecthypothesis.
1.3. SPEC of IP, SPEC of VP, and ThetaRole Assignment
Before consideringthe nature of these two subjectpositions, we outline
our three majorassumptionsabouttheta assignment,whichare the following: First, all theta roles of the verb are assigned at D-structurewithin
the VP. This means that the SPEC of IP is a non-thetaposition. Second,
theta roles are assignedin a uniformmanner,as stated in the Uniformity
of Theta AssignmentHypothesisof Baker (1988). Thus, morphologically
relatedverb forms or, more specifically,the variouscomplexesformedby
affixingdifferent 'voice' markersto a given stem, will assign their theta
roles in an identical fashion. Third, hierarchicalprominence relations
within the theta grid (as discussedin Grimshawand Mester (1988)) are
mirroredin the structuralrelationswithin the VP. Thus, the most prominent theta role will be assignedto the SPEC of VP in such a way that it
c-commandsthe rest of the arguments. (We might also point out here
that while we will be using theta role labels such as Agent and Theme,
these are probablya simplificationof the actualmappingfrom theta roles
to syntacticpositions. For instance, although'most prominent'theta role
might be a more accurateterm than Agent, we will continue to use these
terms for the sake of convenience.)
Given these three assumptions,the picturewhichemergesis one where
the SPEC of IP is not filled at D-structure.At S-structure,an NP which
is not Case-markedin its base-generatedposition within VP may raise to
SPEC of IP where it receives Case via SPEC-Headagreementwith INFL.
We propose, then, that the D-structure common to the Austronesian

IN AUSTRONESIAN

TWO SUBJECTS

LANGUAGES

379

languageswe are discussing(Malagasy,Bahasa, Cebuano and Tagalog)

has the following representation(modulo the orientationof the SPEC of
IP).
(3)

The Proposed D-Structure2

IP

SPEC

IS
INFL

VP

SPEC
Agent
V

NP

Theme
Accordingto this tree there are two subjectpositions, the SPEC of VP,
to which a theta role is assigned, and the SPEC of IP, to which an NP
may move to receive Case. Which NP stays in the VP and which must
raise are determinedentirely by the Case-assigningpropertiesof both V
and INFL. In order to complete the picture, we assume that the verb
moves to INFL in all four languages. This is evident in the three Vinitiallanguages,Malagasy,Tagalogand Cebuano,and assumedin Bahasa
althoughnothing hinges on this assumption.
1.4. Malagasy:NP Licensingand NP Movement
In the previous section, we presented the assumptionsthat guide us in
determiningD-structure. In this and following sections we will see what
mechanismsact on the D-structureto create the S-structurerepresentations. We begin by looking at the active, passive, and circumstantial
constructionsin Malagasy,where it can be clearly seen how the presence
of a particularmorphemein V or INFL can license materialin a particular
NP positionvia Case-assignment.A changein the morphologicalcharacter
of the V + INFL complex is thus accompaniedby a change in its Case2

Note that SPEC of IP is to the rightand SPECof VP to the left. We assumethat this is
a distinctionbetween specifiersof functionalcategoriesand specifiersof lexical categories.
Unfortunately,since there is no evidencefor SPECof CP (questionstake the formof clefts
or predicatestructures(Seiter 1975)) or SPEC of DP (we assumethat possessorsremainin
the SPEC of NP), we do not have strongsupportfor this claim.

380

EITHNE GUILFOYLE ET AL.

assigningproperties,resultingin a changein the wordorderand structural

analysis.
Suppose we take the active and the passive of a regulartransitiveverb
such as 'wash', which has two arguments,an Agent and a Theme. The
relevant sentences are given in (4) below. Traditionally,the NP in sentence-finalposition has been called either the Topic or the Focus NP, and
the particularmorphologyon the verb has been called either Topic or
Focus morphology.As will become clear below, we assumethat the Topic
NP is in the SPEC of IP.3
(4)a.

M-an-sasa(manasa)ny lamba amin'nysavony ny zazavavy.

the clothes withthe soap the girl
AT-wash
The girl washes the clothes with the soap.

b.

Sasa-na(sasan') ny zazavavyamin'nysavony ny lamba.

withthe soap the clothes
the girl
wash-TT
The clothes are washedwith the soap by the girl.

In (4a) the Agent is Topic, as indicatedby the sentence-finalposition of

the Agent, as well as by the particularmorphologyon the verb.4 In (4b)
the Theme is Topic. Assumingthat these two verb forms have a common
D-structure, we can derive the word order differences in the following
way. Hung (1988) claimsthat the prefixan- of manasain (4a) assignsCase
to the Theme and hence licenses the Theme NP in its theta position. Since
the Agent is not Case-markedin the SPEC of VP position, it is forced to
move to the SPEC of IP where it is assigned Case via SPEC-Head
Agreementwith INFL. Hung claimsfurtherthat the suffix-na on the verb
sasana (in 4b) is in INFL and assigns Case to the Agent in the SPEC of
3 In these examples,the Topic is underlinedin the data, and the Agent is in boldface.The
labelsthat we use maydifferfromthe labelsused by Schachterandothers.We will, however,
indicateverbal morphologyas being AT (Agent-Topic,since Agent is in SPEC of IP) and
TT (Theme-Topic,since Theme is in SPECof IP). The abbreviationsused in the paperare
given below:

ACC
BT GEN
LK OBL
RP -

- accusative
Benefactive-Topic
- genitive
linker
- oblique
recent past

TT - Theme-Topic

AT - Agent-Topic
DT - Directional-Topic
IT - Instrumental-Topic
LT - Location-Topic
PL - plural
TOP - topic

XT - non-Agent/non-ThemeTopic (Malagasy)

Furthertests for Topics are discussedin 2.1.1 and 2.1.2.

TWO SUBJECTS IN AUSTRONESIAN

381

LANGUAGES

VP position, hence licensing the Agent NP in its theta position. The

Theme, however, is not assignedCase and is forced to move to the SPEC
of IP. The S-structuretrees correspondingto (4a) and (4b) are shown in
(5a) and (5b).
(b)

(5) (a)
IP

IP
YS

SPEC

PEC

Themei

Agcnti
INFL

VP

INFL

SPEC

VI

SPEC

Agent

V'

4,

ti

NP

V
an-

Theme

--.--

L-

The strongest evidence in favor of the claim that the presence of a

particularmorphemelicenses materialin a particularNP position comes
from the existence of a verb form that combinesthe elements of (4a) and
(4b). When the verb bears the two affixes an- and -na, as in anasana
(an + sasa + na), both Agent and Theme are licensed in their respective
theta positions, and a third NP appearsin the specifierof IP position.6
This has been traditionallycalled the circumstantialform of the verb (see,
e.g., Keenan (1976)). An example of this is given in (6) (we will call the
circumstantialtopic morphologyXT since it may range over several theta
roles).
(6)

An-sasa-na(anasan')ny zazavavyny lamba ny savony.

XT-wash
the girl
the clothesthe soap
The soap was washed (with) the clothes by the girl.

Hung's analysisof (6) is given in (7). Since both Agent and Theme are

s In these and subsequenttrees, the solid arrowsindicatemovementand the dotted arrows

indicatecase assignment.
6 For each verb one particulartheta role is usually the unmarkedchoice for this third
argument;this would be encoded in the lexical conceptualstructureof the verb (see Hung
1988).

382

EITHNE GUILFOYLE

ET AL.

Case-markedwithinVP, a thirdNP, in this case an instrumentalNP, must

appearin the SPEC of IP position.7
IP

(7)

SPEC

~~~XPI<
VP

INFL
-na

SPEC
Agent
V
anI

NP
Theme

NP
tq

Under this account, then, Topic morphologyconsists simply of Caseassigningmorphemes.8AT an- assigns Case to the Theme, forcing the
Agent to move to the SPEC of IP, TT -na assigns Case to the Agent,
forcing the Theme to move. Circumstantial Topic, an-. . . -na is a combi-

nation of these two, meaning that both Theme and Agent receive Case
within the VP, and a third NP moves to the SPEC of IP.
In all three constructions,verb movement to INFL also occurs. This
verb movement, together with our analysisof Agent as being generated
in the SPECof VP, also explainswhy only the Agent is allowed to appear
in a position immediately to the right of the verb, as in (4b) and (6),
violatingthe strict adjacencyconditionwhichnormallyholds between the

7 At this point we assume that the disappearanceof the prepositionin (6) is due to PrepositionIncorporation(as in Baker's(1988)analysisof applicatives)in additionto the appearance of the two Case-assigningmorphemes,an- and -na. By assumingthat preposition
incorporationis obligatorywith the circumstantialform of the verb, we can explain why
there must be an NP that moves to the SPEC of IP to get Case. We have representedthe
structurewith ternarybranching;however,we do this only for the sake of conveniencesince
the actualstructurewithinthe V' is not relevantto our discussion.
8 Note that this analysisof Topic morphologyhas an advantageover other analysessuch as
those that analyseTopic markersas agreementmarkersor incorporatedpronouns.These
analysesrequirethat the Topic NP be indexedwith the morphologyon the verb and cannot
neatly account for the combined effect of the morphologyobserved in the circumstantial
construction.

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

383

verb and the Theme. As we can see in example (8) below, no other
maximalprojectionmay intervenebetween a verb and its object.9
(8)a.

*Nanasa
omaly ny lamba ny zazavavy.
past-AT-washyesterdaythe clothesthe girl
The girl washed the clothes yesterday.

b.

*Nanasa
ho an'ny ankizy ny lamba ny zazavavy.
past-AT-washfor the childrenthe clothesthe girl
The girl washed the clothes for the children.

In this analysisof Malagasy,even thoughthe Agent appearsto interrupt

the adjacencycondition on structuralCase assignment,the trace of the
verb is still adjacentto its object. This providesadditionalsupportfor the
assumptionthat verb movementto INFL takes place.10
1.5. Consequencesof the Analysis
If we compare the NP positions that are, respectively,string adjacentto
the verb and sentence-final,we see that the formercan be occupied only
by the NP bearing the most prominenttheta role - usually the Agent as long as it is licensed there. On the other hand, the NP occupyingthe
sentence-finalpositionmay bear one of a numberof theta roles. The theta
consistencyof the first position is explainedby the fact that this position
is filled at D-structure,in contrast to the theta variableTopic position,
which is filled at S-structure.As we have said, the position adjacentto
the verb at S-structureis the SPEC of VP, which is assigned the most
prominenttheta role by the verb at D-structure.The sentence-finalposition is the SPEC of IP, whichis filled by NP movementfor reasonsrelated
to Case Theory. In a way, both NPs are structuralsubjects.The SPEC of
VP is the externalargumentof the V' and receives its theta role from the
constituentwhich containsthe verb and the other argumentsof the verb.
The SPEC of IP is also a subject, being in an argumentposition external
9 Adverbs, however,may appearin this positionas in Manasamafy ny lambany zazavavy
'The girl washed the clothes hard'. This can be accountedfor by assumingthat adverbs
adjointo VP (see, e.g., Pollock (1989)), or that adverbsare heads adjoinedto INFL (see
Travis(1988)). Thanksto George Rebuschifor pointingout this possibilityto us.
10 We need the additionalassumptionhere that internalargumentsand adjunctsare generated to the right of the V, as in English. This again shows that 'demoted'Agents in a
languagelike Malagasyare quite differentfrom their counterpartsin English, since they
appearin a positionin whichno other argumentor adjunctmay appear.

384

EITHNE GUILFOYLE ET AL.

to the VP where it is assigned case by INFL. As we see below, these two

subjectpositionscorrespondto differentsyntacticphenomena.
2. SUBJECTS IN AUSTRONESIAN

LANGUAGES

In some Austronesianlanguages, no single NP consistentlyhas all of the

properties often associated with subjects. Some subject properties are
characteristic of the NP which bears the theta role Agent, while others

are characteristicof the NP which is identifiedby verbal morphologyas

the Topic of the sentence. This division of propertieshas led some researchersto claim that the Agent NP is the 'true' subject (Byma 1987,
Carrier-Duncan1985, Gerdts 1988), while others argue that the Topic is
the subject (Travis and Williams 1983). Schachter(1976) concluded his
discussion of subject properties by saying that the notion of subject is
irrelevantto any analysisof Philippinelanguagesbecause subjectssimply
do not exist in these languages;thus the existence of subjectscannot be
considereda universalin any sense. Within GB theory, given issues such
as the ExtendedProjectionPrinciple,the derivativenotion of subject, the
problem of pro-droplanguages, etc., it is not clear what it means to say
that a languageis without a subject. In this section we considerhow our
analysisof word order in Austronesianallows us to explain the 'unusual'
properties of Austronesiansubjects without being forced to claim that
these languages'lack a subject'.
2.1. Some Propertiesof Subjects
In his discussionof the subject in Philippinelanguages,Schachter(1976)
outlines a numberof propertiesassociatedwith subjectsin the literature.
These propertiesare, in some cases, tied to a certainsyntactictheory but
might be seen as guidelines for determining the subject in any given
language. Four of these propertiesand their relevancefor the languages
investigated here are discussed below. Two syntactic phenomena, Whextractionand quantifierfloat, are sensitive to the Topic of the sentence
in all four languagesunderdiscussion.In Tagalogand Cebuano,however,
reflexivizationis sensitiveto the Agent NP, and Equi-NPdeletion to either
the Topic or the Agent.
2.1.1. PropertyI: Extractionfacts
In (9) to (15) below, we see that the Wh-questioningof an NP requires
the verb to bear Topic morphology'agreeing'with the theta role class of

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

385

the extracted NP. In (9) there are two examples from Bahasa where
Agent-Topics have been extracted, but only one is grammatical.In the
grammatical(9a), the Agent-Topic form of the verb membacaappears,
whereasthe ungrammatical(9b) has the Theme-Topicverb form dibaca.
(9)
a.
b.

Bahasa
Siapa yang membacabuku itu?
who COMPAT-read book the
*Siapayang buku itu dibaca?
who COMP book the TT-read
Who read the book?

Similarly,in the Tagalog" and Malagasyexamples shown in (10) and

(11), only the Agent-Topic forms of the verbs are grammatical,because
the Agent has been extractedin each case.12
(10)
a.

Tagalog
Sino ang
bumili
ng damit para sa bata'?
who COMPAT-boughtACC-dressfor OBL-child

b. *Sino ang
binili
para sa bata'
g damit?
who COMP TT-boughtfor OBL-child TOP-dress
c. *Sino ang
ibinili
ng damit B bata'?
who COMP BT-boughtACC-dressTOP-child
Who bought the dress for the child?
(11)

Malagasy

a. Iza no
mividy ny vary ho an'ny ankizy?
who COMPAT-buy the rice for the children
b. *Iza no
vidina ho an'ny ankizy ny va?
who COMP TT-buyfor the childrenthe rice
"

Informationquestionsin the Philippinelanguagesmaybe analyzedas cleftsor as nominalizations,as suggestedby Seiter (1975). This is immaterialto the discussionat hand since, no
matterwhat the analysis,it is alwaysthe Topic NP whichis targeted.
12 In the Tagalog examples, ng is glossed as ACC and as GEN. We assume that it is the
same Case-marker,but that in the formercase it is assignedby the V to its object, and in
the latterby morphologyin INFL to the SPECof VP. GenitiveCase (assignedby D to the
SPEC of NP) is also ng. It is importantto note, however, that we considerit a structural
Case-markerand not a preposition.

386

EITHNE GUILFOYLE

c. *Iza no

ET AL.

ividianany varyny aliy?

who COMPXT-buy the rice the children

Who bought the rice for the children?
In the examples shown in (12) to (14), a Theme-Topichas been extracted. Here the verb must bear Theme-Topicmorphologyfor the sentence to be grammatical.Thus, only the (b) examplesbelow are grammatical, since only they containverbs bearingTheme-Topicmorphology.
Bahasa
(12)
a. *Apa yang All membaca?
what COMPAli AT-read
b.

yang dibaca Ali?

what COMP TT-readAli
M

What was read by Ali?

(13)

Tagalog

a. *Ano ang
para sa bata' angtao?
bumli
what COMPA T-boughtfor OBL-child TOP-man
b. Ano ang
binili
ng tao
para sa bata'?
what COMP TT-boughtGEN-manfor OBL-child
bata'?
c. *Ano ang
ibinili
ng tao
what COMP BT-boughtGEN-man TOP-child
What was bought for the child by the man?
(14)

Malagasy

a. *Inona no
mividy ho an'ny ankizy ny lehilahy?
what COMP A T-buyfor the childrenthe man
b. Iona no
vidin' ny lehilahyho an'ny ankizy?
what COMP TT-buythe man for the children
c. *Inona no
ividianan'ny lehilahyny aky?
what COMP XT-buy the man the children
What was bought for the childrenby the man?
In Malagasyand Tagalog, the verb may have Topic morphologythat

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

387

designatesa Topic other than the Agent or the Theme. In Tagalog, the
verb has a numberof differentaffixeswhichcan identifyLocations,Goals,
Benefactives, and other relations as the Topic. Malagasyhas circumstantial morphologywhich appearson the verb whenever any element other
than Agent or Theme appears in Topic position. In (15) and (16), a
Benefactive has been extracted;in each case only the (c) examples have
verbs bearingthe appropriatemorphology,and thus only these are grammatical.
(15)

Tagalog

a. *Sino ang
bumili ng damit ang tao?
who COMPAT-buyACC-dressTOP-man
a damit?
b. *Sino ang
binili ng tao
who COMP TT-buyGEN-man TOP-dress
c.

Sino ang
ibinili ng tao
ng damit?
who COMP BT-buy GEN-manACC-dress
Who was bought the dress (for) by the man?

(16)

Malagasy

a. *Iza no
mividy
vary ny lehilahy?
who COMPAT-boughtrice the man
b. *Iza no
vidin'
ny lehilahyny vary?
who COMP TT-boughtthe man the rice
c.

Iza no
ividianan' ny lehilahyny vary?
who COMP XT-boughtthe man the rice
Who was bought rice (for) by the man?

The examples of Wh-movementgiven in (9)-(15) thus demonstratethat

in these Austronesianlanguages,only the Topic can be the target of such
movement.13

2.1.2. PropertyII: FloatingQuantifiers

Floating quantifiersalso single out the Topic NP since they must be
associated with the Topic regardless of its theta role. In the Cebuano
13

Relativizationand cleftingworkin a similarmanner.

388

EITHNE GUILFOYLE

ET AL.

examples in (17), we see two possible non-floated forms of the phrase

meaning 'all the children',in which the quantifiertanan appears, respectively, before (as in (17a)) and after (as in (17b)) the head noun bata'
'children'. In (17c), however, the quantifierhas 'floated' to a position
outside the NP.14
(17)
a.

Cebuano
Ningdaganang tanan nga bata'.
A T-run TOP all LK child

(Bell 1976, (38a))

All the childrenran off.

b. Ningdaganang batang tanan.
A T-run TOP child-LKall

(Bell 1976, (38b))

All the childrenran off.

c.

Floated Form15
Ningdagantanan ang bata'.
AT-run all
Top child

(Bell 1976,(38c))

The childrenall ran off.

In (18) below, the quantifierhas also floated out of its noun phrase which in this example is the Theme-Topic. So, no matter whether the
topic is the Agent, as in (17c), or Theme, as in (18), the quantifieris
associatedwith the Topic.
(18)

Cebuano
Gibasa tanan sa mga istudienteang mga libro ni Rizal.
TT-readall pI
student TOP PL book Rizal
Rizal'sbooks were all being read by the students.
(Bell 1976, (39b))

Schachterpresentsthe case for identifyingTopic NPs as subjects,based

on the evidence from extraction and quantifierfloat as presented here,
and the relation of these phenomena to certain claims about subjects.
These are that subjectswill be affectedby relativizationif any grammatical
relation is (Keenan and Comrie 1977); and that, in the frameworkof
Relational Grammar,only 'terms' may launch floating quantifiers(Bell
14

See Bell (1976) for a more detaileddiscussionof the data presentedhere.

15 This quantifieris assumedto be positionedoutside the NP since it precedesthe Topic

Markerang. This is clearestin examplessuch as (18) where there is interveningmaterial.

TWO SUBJECTS

IN AUSTRONESIAN

LANGUAGES

389

1976). These observations,Schachtersuggests,may be handled"by identifying the topic as the subject" (Schachter1976, p. 501). Schachterobserves, however, that the picture becomes more complicatedwhen reflexivizationand Equi-NP deletion are included. As we discussbelow, in
a subset of the languages, reflexivizationis sensitive to the Agent, and
Equi-NP deletion to both the Topic and the Agent.
2.1.3. PropertyIII: Reflexivization
In Tagalog and Cebuano, only the Agent can always be the antecedent
of a reflexive (whetheror not it is the Topic) and the Agent itself never
bears reflexive morphology. In the examples in (19) below we see that
the reflexive form kanilangsarili takes the Agent NP as its antecedent
whetheror not this NP is the Topic.
(19)
a.

Tagalog(Schachter1976, (22) and (23))

sila
Nagiisip
sa kanilangsarili.
A T-think-aboutTOP-theyOBL-theirself
They think about themselves.

b.

nila
Iniisip
ang kanilangsarili.
TT-think-aboutGEN-theyTOP-themselves
They think about themselves.

c. *Iniisip
ng kanilangsarilisila.
TT-think-aboutGEN-themselvesTOP-they
*Themselvesthink about them.
Since reflexivizationis often sensitive to subjecthood(e.g., zibun in Japanese), one might conclude that the Agent in Tagalog is always the
subject. While this alreadycomplicatesthe picture that was beginningto
emerge in which the Topic NP appearedto have subject-likeproperties,
more complicationsarise when Equi-NP deletion is investigated.
2.1.4. PropertyIV: Equi-NP Deletion
In Tagalog and Cebuano, Equi-NP deletion may act on either the Topic
or the Agent of the embeddedclause. When the Topic of the embedded
clause is dropped, it may bear a numberof theta roles. This is shown in
(20) below.

390

EITHNE GUILFOYLE

(20)
a.

ET AL.

Tagalog16

sa akin.
Natatakotang batang maipakilala mo
TOP child-LKTT-introduceGEN-you OBL-me
afraid
The child is afraidof being introducedby you to me.

b.

Natatakotkang
ipakilala
ang bata' sa akin.
afraid
TOP-youAT-introduceTOP child OBL-me
You are afraidof introducingthe child to me.

When the Agent of the lower clause is dropped, it makes no difference

which NP is Topic. In the sentencesin (21) below we find that the Agent
is alwaysnull regardlessof whetherthe embeddedverb bearsAgent-Topic
(21a), Theme-Topic(21b) or Direction-Topic(21c) morphology:
(21)
a.

Tagalog(Schachter1976, (24a-c))
sa bangko.
Nagatubili siyanghumiram ng pera
AT-borrowACC-moneyOBL-bank
AT-hesitatedhe
He hesitated to borrowthe money from the bank.

b.

sa bangko.
Nagatubili siyanghiramin a pera
TT-borrowTOP-moneyOBL-bank
AT-hesitatedhe
He hesitated to borrowthe money from the bank.
(Literally:He hesitated the money to be borrowed from the
bank by

c.

.,)

Nagatubili siyanghiraman ng pera

angbangko
AT-hesitatedhe
DT-borrowACC-moneyTOP-bank
He hesitatedto borrowthe money from the bank.
(Literally:He hesitatedthe bank to be borrowedmoney from
by

*..)

Within GB theory, the phenomenon of Equi-NP deletion - which is

describedin terms of the presence of PRO in control structures- follows
from governmenttheory. Since PRO may appear only as the subject of
an uninflectedINFL, its presence is, in effect, restrictedto the SPEC of
16
Thanksto a reviewerfor supplyingthese data fromDell (1981), whichwere also accepted
by our informants.See Kroeger(1991)for an interestingalternativeanalysisfor the control
facts in Tagalog.

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

391

IP position. In Tagalog, however, PRO appearsin two differentpositions

which, in our account, are SPEC of VP and SPEC of IP.

2.2. Towardsan Account of the Split in SubjectProperties

If we attempt to use the four subject propertiesdiscussedabove as our
only criteriafor determiningsubjecthood,we are faced with a dilemma.
According to properties I (extraction) and II (quantifierfloat), it is the
Topic which is the subject;accordingto propertyIII (reflexivization)it is
the Agent, and accordingto property IV (Equi-NP deletion) it is both.
This dilemmamightbe solved if we appealto the analysisof Austronesian
phrase structurediscussedin Sections 1 and 2.
The split in Austronesiansubject propertiesreflects the difference in
properties associated with the SPEC of VP and SPEC of IP positions.
Since Agents are always base-generatedin the SPEC of VP, they will
alwaysexhibitthe propertiesassociatedwith that position. Likewise,Topics will exhibit the properties associated with the SPEC of IP position
which they occupy.
It is interestingto observe, moreover, that such a split is not unique to
the Austronesianlanguage family. Noting again that reflexivizationand
control in certainAustronesianlanguagesmay refer to the theta sensitive
position SPEC of VP, and extraction and quantifierfloat,to the theta
variable position SPEC of IP, we can see that a similar split has been
observedin more well-studiedlanguagessuch as English. Williams(1987,
p. 151), in a work which "explore[s]the possibilitythat the bindingtheory
(plus control) applies not to NPs, but to theta roles that are assignedto
NPs", comes to the conclusionthat "the pictureof grammarthat emerges
is one in whichthere are two components,one concernedexclusivelywith
relations between theta roles (theta role assignment,the theta criterion,
and bindingtheory) and one concernedwith positions (quantificationand
Wh-movement)."
While the match between Williams'proposalfor control of theta roles
and our proposalfor control of SPEC of VP is not perfect, it is significant
that in both cases bindingand controlfall on one side of the dividingline,
and Wh-movementand quantificationon the other side. Unlike Williams,
however, we wish to claim that both sets of relations are subject to
structuralpositions, one to SPEC of VP and the other to SPEC of IP.

392

EITHNE GUILFOYLE

ET AL.

2.2.1. PropertiesAssociatedwithSPEC of VP
The propertiesthat we can associatewith SPEC of VP are propertiesIII
and IV, i.e., reflexivizationand control. Within GB theory these are
properties that are subject to binding theory, and binding phenomena,
accordingto the suggestion of Williams(1987), as just described, sometimes appear to be theta sensitive. This analysis is compatiblewith the
data in (19) to (21). Thus we note that Agents are alwayspossible antecedents of reflexives, and that they may appear as PRO.17 However, as
Agents are alwaysbase-generatedin SPECof VP, it is possible to maintain
a purely structuralaccount of these binding facts, with PROs and the
antecedents of reflexives being associated with the structuralposition
SPEC of VP.18

2.2.2. PropertiesAssociatedwithSPEC of IP
The propertiesthat we associatewith SPEC of IP are propertiesI and II,
i.e., extractionand quantifierfloat. In orderto accountfor the interpretation of floatingquantifiers,we assume that they are licensed in the same
way as adverbs - in particular, subject-orientedadverbs (see, e.g.,
McConnell-Ginet(1982)).'9 Travis (1988) proposes that adverbslike reluctantlyin example (22) below are adjoinedto INFL and licensed by the
AGR featurein INFL. Since this featureis coindexedwith the NP (which
is the subject)in the SPEC of IP position, it has the effect of relatingthe
adverb to this NP, and thereby giving subject-orientedreadings for the
adverbsin (22). Thus, it is Marywho is reluctantin (22a), and the students
in (22b).

17
The examplesin (21) suggest that the position SPEC of VP is not alwaysgovernedor
Case-markedby the verb in Tagalog, since PRO may appear in this position. We are,
therefore,forced to say that Case-assignmentto SPEC of VP is optional. A similarclaim
has also been made, however,for subjectsin non-finiteclausesin Irishby McCloskey(1984)
and Chungand McCloskey(1987).
18 We do not suggest,however,that our analysiswill also accountfor the factsthat Williams
(1987) presents, where it appears that the Theme theta role may be controlled within
nominals.
19 Note that Sportiche's(1988) accountof QuantifierFloat does not have an obvioususe in
these languages.Floated quantifiersalwaysappearimmediatelypost-verbally,independent
of what the theta role (and presumablyD-structureposition)of the relatedNP is.

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

393

(22)a. Maryreluctantlywas instructingthe students.

b.

The studentsreluctantlywere being instructedby Mary.

Given this analysis,it is not surprisingthat the quantifierin Austronesian languagesshould be sensitive to the NP in the SPEC of IP position,
since it is this NP with which AGR will be coindexed. In the Tagalog
example below, lahat 'all' is adjoinedto INFL and licensed by the AGR
feature in INFL. Since INFL is coindexed with the NP in the SPEC of
IP, the quantifierwill be interpretedwith the Topic NP mga libro and not
mga bata'.
(23)

Tagalog(Schachterand Otanes 1972, p. 148)

binasang lahat ng mga bata' ang mga libro.
TT-read-LKall ACC-pl child TOP-pl book

a.

The childrenread all of the books.

b. *AIIthe childrenread the books.

As for extraction,we want to explain why extractiononly occurs from
the SPEC of IP. Although an account of this is beyond the concernsof
this paper, the phrasestructurethat we have proposedfor these languages
offers a clear directionfor a solution. Extractionobviouslyrequiressome
type of localityto preventmovementout of the VP acrossa filled SPECof
IP. This locality might be determinedwithin the frameworkof relativized
minimality(Rizzi 1990) or barriers(Chomsky1986a);we leave this as an
open question for future research.
2.2.3. Summary
What emerges from our analysisof extraction,quantifierfloat, reflexivization and control is that these phenomena, though sensitiveto subjectsin
some languages,appearto be split between two NPs in some Austronesian
languages.Extractionand quantifierfloat targetTopic NPs, reflexivization
targets Agent NPs, and Equi-NP deletion targets either. Such a split of
propertiesis not surprisinggiven the phrase structurewe have proposed.
Nor is our claim that both sets of phenomena have structuralaccounts,
despite the apparentsensitivityof some of the propertiesto theta roles.

394

EITHNE GUILFOYLE ET AL.

3. WORD ORDER VARIATION AMONG

AUSTRONESIAN

LANGUAGES

In this section we show that the presence of non-topic Agents in SPEC

of VP can explaincertainword orderfacts. Consideragainthe D-structure
tree in (3), repeated below, which we claim to be basic to Malagasy,
Bahasa, Tagalogand Cebuano(abstractingawayfromminordifferences).
IP

(24)

SPEC

I'
INFL

VP
V

SPEC
Agent
V

NP

Theme
The discussionof Malagasyin Section 1.4 above suggestsan idea of how
the verbal morphologymight play a role in the licensingof NPs. Now we
will turn to a discussionof some of the propertiesof the other Austronesian languageswhich distinguishthem from Malagasy.
3.1. Tagalog
Tagaloghas fairlyfree word order, and has been claimedto be a nonconfigurationallanguage (Carrier-Duncan1985). Nevertheless, it seems that
it mustbe configurationalat some level if we are to explainthe constraints
on Wh-questions,relativeclauses, and cleft structuresdescribedin Section
2. We will assume that at D-structurethe order of constituentsis AgentV-Theme-X,as shownin (24). Althoughfactorssuch as the relativelength
of each word and the pronominalor full NP status of various arguments
may influence surface word order, preferred orders (as judged by our
informants)do emerge when these factorsare controlledfor. These orders
are shown in (25) (based on Schachter1976).2o
20 Differentinformantsallow differentamountsof variation.We assumethat there mustbe
some sort of scramblingrule and, in particular,PP extraposition,to accountfor the various
permutationsof the ordersof post-verbalelements. Across informants,however, non-topic

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

395

(25)a.(i) Mag-aalis ng bigas sa sako para sa bata'

AT-take-outACC-riceOBL-sackfor OBL-child
ang babae.
TOP-woman
The woman will take rice out of the sack for the child.
(ii) Mag-aalis angbabae ng bigas sa sako para
AT-take-outTOP-womanACC-riceOBL-sackfor
sa bata'.
OBL-child
The woman will take rice out of the sack for the child.
b.

sa sako para sa bata'

Aalisin
ng babae
TT-take-outGEN-womanOBL-sackfor OBL-child
ang bigas.
TOP-rice
The rice will be taken out of the sack for the child by the
woman.

As these examples show, Tagalog is very much like Malagasyexcept

for the particleang whichmarksthe Topicof the clause, and an interesting
word order variation which occurs in Tagalog, but not Malagasy. This
variationis shown above in (25ai,ii) where the verb has the same AgentTopic form in both instances.21 Accordingto our analysis, while V must
raise to INFL, the Agent may move to SPEC of IP, producingthe order
shownin (25ai), or remainin the SPECof VP, producingthe ordershown
in (25aii). The correspondingtrees are given in (26).

Agent is singledout as the preferredelement to follow the verb. (Thanksto an anonymous

reviewerfor inputon this question.)This is easilyexplainedif these are alwaysfoundwithin
the SPECof VP. Further,in cases where all the argumentsof the verb are lexicallyrealized,
the only Topic that can appearadjacentto the verb is the Agent Topic. Again this follows
fromthe fact that Agent is generatedin SPECof VP and is allowedto remainin thisposition
even if it is the designatedTopic.
21 Cebuano appears to allow only the word order shown in (25aii). Our thanks to an
anonymousreviewerfor pointingthis out to us.

396

EITHNE GUILFOYLE

(26)a. Agent moved to SPEC

of IP

ET AL.

b. Agent remainingin
SPEC of VP

I'P
I

IP
SPEC

SPEC

Agenti

VP

INFL
SPEC

VP

INFL
V'

SPEC

V'

Agent
V

NP
Theme

NP
Themc

The Theme, however, does not have a choice of positions when it is

Topic. The unmarkedwordorderis one in whichthe TopicNP has moved,
as shown above in (25b). Our explanationof this asymmetryhas to do
with the relation of the Agent to INFL. In (26a) INFL Case-marksits
specifiervia SPEC-HeadAgreement, as in Malagasy.In (26b), however,
INFL governs down into the SPEC position of its VP complement and
licenses the Agent NP in this position.22
Note that, after V has moved into INFL and before a Topic has moved
to the SPEC of IP position, constituentsexhibit the following order: VAgent-Theme-X.This suggeststhat a sentence containinga verb that does
not designatea Topic should also have this word order, which is what we
find. Below is an example of just such a sentence, where the verb is in
the 'recent past,' a form without any special topic-markingmorphology.
(27)

Kakakainng leon ng tigre.

RP-ate GEN-lionA CC-tiger
The lion ate the tiger.
*The tiger ate the lion.

If the word orderwere completelyfree, it ought to be possiblefor either

ng leon or ng tigre to be the Agent.23 In fact, ng leon must be the Agent,

and the sentence is ungrammaticalon the reading'The tiger ate the lion'.
22 The ability of INFL to assign Case to the SPEC of VP has alreadybeen proposed by
Guilfoyle(1988) for VSO languagessuch as Irish.
23 GenitiveCase assignedby the V to its SPECis homophonouswith ACC assignedby the
V to its complement.

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

397

These data suggest that the Agent precedes the Theme at an underlying
level.24

3.2. Bahasa Malaysiaand BahasaIndonesia

BahasaMalaysiaandBahasaIndonesiaare examplesof the SVO members
of the Austronesianlanguagefamily;they neverthelessresembleMalagasy
and Tagalog in also permittinga non-demoted Agent alongside a nonAgent Topic. In this section we will discusspropertieswhich are found in
both dialects, and refer to these dialects collectively as Bahasa. Among
its various sentence constructions,Bahasa has the active, the subjective
passive, the canonicalpassive with the Agent in a by-typephrase headed
by the preposition oleh, and the canonical passive with the Agent but
withoutoleh. In this section we will attemptto show how a unifiedanalysis
of all four of these constructionsfollows from a commonD-structurewith
two specifierpositions:a theta position in VP and a non-thetaposition in
IP.
Unlike Malagasy and Tagalog, Bahasa is a subject-initiallanguage.
Except for this minor difference, we assume the same D-structureproposed earlier. The active constructionis fairly straightforwardsince it
coincides with the Actor-Topicconstruction,and the analysis should be
familiarexcept for the SVO surface word order. (Data are from Hung
(1987) unless otherwiseindicated.)
(28)

'Ali/Saya/ Lelaki itu membacabuku itu denganteliti.

Alil II
boy the AT-read book the with care
Ali/I/the boy read the book carefully.

In accordancewith our analysis,the presenceof the verbalprefixmeNis associatedwith the appearanceof the Theme in VP and the Agent in
the SPEC of IP behavingmuch like an- in Malagasy.While the existence
of verb movement to INFL is not evident in the active construction,it
will be necessary to assume that it takes place in the formation of the
passive, as we will see. The S-structurerepresentationof the active constructionis shownbelow in (29). The Theme receives Case withinthe VP,
while the Agent mustmove to the SPECof IP to receive Case from INFL.

24
Whileit mightbe arguedthat, in free word orderlanguages,Agents precedeThemesfor
pragmaticreasons,we are assuminghere that there are also structuralreasonsfor this word
order preference.

398

EITHNE GUILFOYLE

(29)

ET AL.

IP
SPEC
Agenti

I'

INFL

VP
SPEC

V'

ti

NP
Themc

For a detailed descriptionof the passive constructionin Bahasa, we

referthe readerto Chung(1976) and Sie (1989). They reportthree typesof
passive:the subjectivepassive, the canonicalpassivewithout the agentive
prepositionoleh, and the canonicalpassive with the agentive preposition
oleh. On this third type of passive, speakers are divided between the
conservativeand the liberal camps. The conservativespeakersallow only
a third person NP to appear as the object of oleh. The liberal speakers
impose no such constraints. Let us now turn to each type of passive,
noting the particularrestrictionswhich applyto each.
We begin with the subjectivepassive, whichhas the followingcharacteristics: the surface word order is Theme-Agent-Verb,the verb bears no
morphology, and the Agent is restrictedto being a pronoun, clitic, or
propername.
(30)a.

kupukul.
Anjingit-uAli/ saya/
dog the Alil 1 sg. pronounl/ sg. clitic hit
The dog was hit by Ali/me.

b. *Anjingit-ulelaki itu pukul.

dog the boy the hit
The dog was hit by the boy.
The canonicalpassive withoutoleh has the followingcharacteristics:the
surfaceword order is Theme-di-Verb-Agent,the verb bearsthe prefixdi-,
and there is a strict adjacencyrequirementwhich holds between the verb
and the Agent.

TWO SUBJECTS IN AUSTRONESIAN

(31)a.

LANGUAGES

399

Budak-budakitu telah ditangkappolis. (Othman1981,p. 103)

child-PL
the ASP TT-catch police
The childrenhave been caughtby the police.

b.

Buku itu dibaca lelaki itu.

book the TT-readboy the
The book was read by the boy.

c.

Ahmad dipukulAlNdengan kayu.

Ahmad TT-hit Ali with stick

(Lewis1969,p. 49)

Ahmad was hit by Ali with a stick.

itu dikirim kepada Tomo All.
(32)a. *
TomoAli
money the IT-send to

(Sie 1989,p. 62)

Uang itu dikirim All kepadaTomo.

Tomo
money the TT-sendAli to

(Sie 1989,p. 62)

b.

The money was sent to Tomo by Ali.

The canonical passive with oleh has the following characteristics:the
the verbbearsthe prefix
surfacewordorderis Theme-di-Verb-oleh-Agent,
di-, and the object of the preposition oleh is interpretedas the Agent.
Note that the entire PP may be absent, and that interveningmaterialmay
appearbetween the verb and the PP, if it is present. Conservativespeakers
requirethat the object of oleh be in the third person; liberalspeakersdo
not.25
(33)a.

Buku itu dibaca.

book the TT-read
The book was read.

b.

Buku itu dibaca oleh Ali/ lelaki itu.

book the TT-readby Alil boy the
The book was read by Ali/the boy.

25 An anonymousreviewerhas pointedout to us that the constraintsobservedby the set of

conservativespeakersholds true of Javanese.

400

EITHNE

c.

GUILFOYLE

ET AL.

Sepuluhdolar sudah di-bayarkepadatukang rumput

workergrass
ten
dollarsASP TT-pay to
oleh saya.26
by lsg
Ten dollars has been paid to the gardenerby me.

It has been suggestedthat the presence versus absence of the agentive

prepositionmight be due to a rule of oleh deletion or the generationof
an empty preposition.277 We reject these proposals on the grounds that
neithercan explainwhy interveningmaterialcan appearbetween the verb
and the PP as in (33c) but not between the verb and the Agent as in
(32a).
The analysiswe propose to account for these differenttypes of passive
assumes that they are all derived from a common D-structurein which
the verb assigns an Agent theta role to the NP in the SPEC of VP, and
a Theme theta role to the NP in the complement position. Under our
account, the canonicalpassive with or without oleh (excludingthe dialect
spoken by the more liberal speakers) is analyzedin the same way as the
subjectivepassive. This is accomplishedfirstby postulatingthat the prefix
di- is the realizationof the set of features representingthe third person.
Supportfor this comes from the fact that the thirdperson singularsubject
pronounis dia. The second assumptionis that NPs are, in fact, dominated
by DP, the maximalprojectionof the functionalcategoryDET, and that
the Case-markingrelationbetween a DP and a V0 may be createdby the
movement of the head of the DP to a V in INFL.28Of course, due to
locality constraints,not all DPs in the VP are eligible to enter into this
relationsipwith INFL. The DP in the SPECof VP bears a special relation
to INFL, as we have seen in Malagasyand Tagalog.
The subjectivepassive constructionis derivedfrom a similarstructure.
In a constructionlike Anjing itu ku-pukul'I beat the dog', what is present
at D-structurein the SPEC of VP is a set of features in the head of DP
representingthe first person singularAgent, ku-. These features, if realized in the SPEC of VP, will not be case-marked,since there are no
potential Case-markersnearby. Instead, the head of DP, or DET, may
raise to INFL and be realizedthere as ku- on the verb whichhas also been

26

This example was provided to us by an anonymous reviewer.

Thanks to Georges Rebuschi for pointing this out to us.
28 We will henceforth refer to NPs as DPs unless referring to traditional concepts such as
NP-trace or the function of an NP.
27

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

401

raised. In this way, a Case-markingrelationhas been createdbetween the

Agent and Verb,29and the Theme moves to the SPEC of IP.
(34)

SubjectivePassive (see 30a)

IP

anjingiLuk

I
kun

VP
V

pukuli DP

D?

V0

tn

ti

Two commentsneed to be made. First of all, notice that in this analysis

the verb is never truly 'bare'. It is always associated with some head,
lexical (meN-) or functional(D). Perhapsthis explainswhy we never find
sentences like the ones below:
(35)a. *Buku itu PRO baca.
book the PRO read
b. *Buku itu PRO baca oleh Ali.
book the PRO read by Ali
Second, we stated that only pronouns, clitics, and proper names like Ali
could appearas the Agent in the subjectivepassiveconstruction.This can
be accounted for if we assume the claim made in Postal (1969) that
pronounsand propernames are located in the position of the determiner.
The same thing happens in a canonicalpassive constructionlike Buku
itu di-baca lelaki itu 'The boy read the book'. Lelaki itu 'the boy' is
assigned the role of Agent in the SPEC of VP. The contents of the
head of the DP are the features for the third person singular.(There is
independent evidence that the determineritu is only a modifier, not a
functionalhead.) The head of the functionalcategoryDP, or DET, then
raises to INFL, leaving behind any materialthat might be in the lexical

29
This is similarto Baker's (1988) views on Case-assignmentand incorporation.If a noun
has been incorporatedinto a verb, it does not need to be Case-markedin the usualmanner.

402

EITHNE GUILFOYLE

ET AL.

projectionNP. In INFL, DET is spelled out as di- and combineswith the

verb which also raises to INEL, giving di-baca. Again, a Case-marking
relation has been established between the Agent and the verb.30The
Theme, buku itu, moves to the SPEC of IP in the usual manner.
(36)

Canonicalpassivewithoutoleh (see (31b))

IP
bukuiluk

I?
din

VP
bacaiDP

V
Np V0

tn

lelaki

lj

ilu

The adjacencyrequirementthat holds between the verb and the Agent is

by now a familiarone; it is a requirementthat holds between INFL and
the SPEC of VP.
We turnnext to the canonicalpassivewith the agentiveprepositionoleh
of the conservativespeakers. This constructionoffers an alternativeway
in which to add to the reference of an unspecifiedthird person Agent,
namely with the addition of an oleh-phrase.As with the cases described
above, the set of third person featuresin DET raises to INFL where it is
realizedas di- on the verb which has also raised. The oleh-phrasemay or
may not be present. Oleh itself assigns an Agent theta role to its object,
and Case-marksit as well. This specified Agent serves to double the
unspecifiedAgent of the verb, so the two must be coindexed, and thus
must agree in person and number.

30 Note that a propername can also be 'stranded'as an NP, indicatingthat a propername

may or may not be analyzedas being located in the head of DP, e.g. Buku itu di-bacaAli,
'The book was read by Ali'.

TWO SUBJECTS IN AUSTRONESIAN

(37)

LANGUAGES

403

Canonicalpassivewitholeh (conservative)(see (33b))

IP
Y

Themck
10

din

VP

V; DP
DIVtk
tn

V
PP
ti

DP
Agent,

In this case, interveningmaterial may occur between the verb and the
Agent. This is because the Agent is in a prepositionalphrase under VP
or V' and bears no Case-markingrelationto the verb which is in INFL.
Given this account, it is reasonable to say that the liberal speakers
seem to have lost the originalcharacterizatonof the morphemedi- as an
expressionof the third person.31For them, the morphemedi- is like the
Englishpassive morpheme-en, and we propose that the analysisis therefore much the same. Following Fukui and Speas (1986) and Guilfoyle
(1990), we assume that the passive morpheme is in V, that it is not a
Case-marker,and that the only thing that can appearin the SPEC of VP
is thereforePRO (which cruciallymust not be Case-marked).

31 If this is true, these same speakerswould be expected to have also lost the canonical
passivewithoutoleh.

404
(38)

EITHNE GUILFOYLE

ET AL.

Canonicalpassivewitholeh (liberal) (see (33c))

'P

Tbemek
I?
[di-ViI

VP
V

PROn

tk pp

ti

DP
Agent,,

That the various types of passive in Bahasa can be derived from a

common D-structureprovides further support for our initial hypotheses
concerningphrase structureand theta assignment.We have shown that
Bahasa differs from the other languages in that it does not have direct
Case-markingof the SPEC of VP. Instead, it makes use of DET to INFL
movement,an X0 movementrulewhichhas also been proposedto account
for cliticizationfacts in Romance (Kayne, 1989). This rule interactswith
the V-to-INFL rule to establish a Case-markingrelation between the
Agent and the verb, as shown in (34), (36), and (37). In these three cases,
the morpheme associated with the verb is the expression of the person
and number features of the Agent. In the liberal dialect the origins of
the morpheme di- have been lost, and instead di- is analyzed on a par
with the Englishpassive morpheme.

3.3. Summary
By assumingtwo subjectpositions, the SPEC of IP and the SPEC of VP,
and by furtherassumingthat they may be filled simultaneously,we have
accountedfor the following facts:
(i) The Agent is the only maximalprojectionwhich may appearbetween a verb and its object at S-structure.
(ii) No prepositionis needed to license the 'demoted'subjectin Malagasy and Tagglog.

TWO SUBJECTS IN AUSTRONESIAN LANGUAGES

405

(iii) There are two possible positions for an Agent-Topicin Tagalog.

(iv) There are three forms of the passive in Bahasa.

4. THE PHRASE-STRUCTURE

OF AUSTRONESIAN
AND A TYPOLOGY OF PASSIVE

LANGUAGES

We have shown that our analysis can explain the split in subject properties

which has long challengedresearchersof Austronesianlanguages,as well

as the word order facts in a number of these languages. We will now
consider the implicationsof our analysisfor the typology of naturallanguages. In particular,we will consider the non-Agent-Topic(i.e., nonactive) forms found in Austronesianlanguages and compare them with
passive structuresin languageslike English, showingwhy a languagelike
English expressesthe Agent argumentof a passive as a by-phrase,while
the Austronesianlanguages examined here - with the sole exception of
Bahasa, as just described- lack by-phrasesaltogether.
For this analysiswe drawon Guilfoyle's(1991) argumentthat the form
of passivizationin a givenlanguageis closelyrelatedto its phrasestructure,
and, more specifically,on her (1990) argumentthat languagespermitting
the SPEC of VP to be Case-marked,and hence filled by any lexical DP
at S-structure, will lack Agent by-phrases, whereas languages which, in
passives, can only Case-mark SPEC of IP will have Agent by-phrases.
Whether or not SPEC of IP receives Case depends on the directionality
of Case assignment in the language and on the value of the setting of the

head-initial/head-finalparameter.
To see how this analysisworks for English, considerthe tree below.
(39)

IP
SPEC

I'

John INEL
F
Nom

VP
/

DP

ti

V
sees
English

DP
Mary

406

EITHNE GUILFOYLE

ET AL.

In English INFL assigns Case to the left only; therefore, the only DP
types that can occupy the SPEC of VP at S-Structureare PRO and NP
trace, which do not appearin Case-markedpositions. A lexicallyrealized
subjectDP will alwaysraise from its D-structureposition.
In Section 1.3 we argued that, in Austronesian languages, the nonAgent DP that has moved to the SPEC of IP and the Agent DP that has
been base-generatedin the SPEC of VP are both licensed at S-structure.
If we considerTheme-Topicconstructionsin particular,we note that they
are equivalent to English passives except for having their Agent in the
SPEC of VP rather than having a by-phrase. Thus the Theme-Topic
sentences in (40) have the S-structurerepresentedin (41).

(40)a. Tagalog
sa sako para sa bata'
Aalisin
ng babae
TT-take-outGEN-womanOBL-sackfor OBL-child
TOP-rice
The rice was taken out of the sack for the child by the woman.
b. Malagasy
Vidin' ny lehilahyho an'ny ankizy ny vary.
TT-buythe man for the childrenthe rice
The rice was bought by the man for the children.

TWO SUBJECTS IN AUSTRONESIAN

(41)

LANGUAGES

407

IP
I'

SPEC

Themci

INFL'

VP

via

VI

SPEC
Agent
V

DP

tj

ThemcTopic(Passive)
MalagasyandTagalog

Accordingto the analysisof Guilfoyle (1990, 1991),32the expressionof

Agent by means of a by- phrase in the English passive stems from the
inability of the SPEC of VP to receive Case. The SPEC of VP receives
an Agent theta role, and may only be occupiedby a PRO coindexedwith
an adjunctby-phrase.The S-structurerepresentationof an Englishpassive
is given in (42) below:
(42)

IP
SPEC
John

I'

INFL

VP

: was/

Nom
V

SPEC
PROj

by
V
seen

PP
Maryj

DP

In contrast, Austronesian does permit the SPEC of VP to be Case-

marked.Bahasafalls somewherebetweenEnglishand the other Austrone32

This view of Englishpassivehas also been proposedby Fukui and Speas (1986).

408

EITHNE

GUILFOYLE

ET AL.

sian languages. The Bahasa canonicalpassive, as we have seen, consists

of two subtypes:one in which the Agent in the SPEC of VP is licensed
only because DET can move into INFL (and not because of direct Casemarking,as in Malagasy),and one in which there is no such option and
the Agent can only be PRO. Our analysisof the latter subtypeis similar
to that of English passive structures;the Agent theta role is assigned to
PRO, which may be coindexedwith an adjunctPP.
The analysisof passivepresentedabove has importanttheoreticalimplications, as it relates to two proposals in the literature:Burzio's (1986)
generalizationregarding Case and theta role assignment, and Baker's
(1988) UTAH.
Baker (1988) proposes that the assignmentof theta roles is mediated
by UTAH, given in (43) below:
(43)

The Uniformityof Theta AssignmentHypothesis(UTAH)

Identicalthematicrelationshipsbetween items are represented
by identicalstructuralrelationshipsbetween those items at the
level of D-structure.
(Baker 1988, p. 46)

This hypothesis predicts that particularphrase-structurepositions are

associated with particulartheta roles. Under most theories of passive
formationthe assignmentof the Agent theta role presentsa difficultyfor
UTAH. While Themes are always associatedwith the complementof V,
Agents are associatedwith the subject position only in active sentences.
In passive structuresthe Agent is assumed to be suppressedand is not
assigned to the subject position. This problem does not arise under the
analysispresentedhere, because the Agent is assignedto SPEC of VP in
both active and passive structures.In passive structuresit may appear as
PRO or as a lexical DP in a given languagedependingon whetheror not
SPEC of VP can receive Case in that language. Adopting the proposed
analysis of passive formation in English and Austronesian allows us to
maintain a strong version of UTAH, because the Agent theta role is
always associatedwith the SPEC of VP.
Moreover, our analysis is incompatible with Burzio's generalization,
which links the assignmentof Case to the object DP with the assignment
of a theta role to the subjectDP. Burzio's generalizationis given in (44)
below:
(44)

Burzio's Generalization(adaptedfrom Burzio 1986, p. 178)

TWO SUBJECTS

IN AUSTRONESIAN

LANGUAGES

409

A verb may assign a theta role to its subject iff it assigns

(accusative)Case to its object.
Burzio'sgeneralizationis undesirableon both theoreticaland empirical
grounds. On theoreticalgroundsit is objectionablebecause it is not clear
how the mechanismsof Case and theta role assignmentare related, seeing
that they occurat differentlevels of the grammar,and, therefore,how they
conspire to associate external theta roles with accusativeCase. Burzio's
generalizationis inadequateon empiricalgroundsbecause it cannot account for the facts of passivizationin Austronesianlanguageslike Tagalog
and Malagasy. In these languages both the Agent and the Theme are
expressed in passive structures. Thus, while the verb does not assign
accusativeCase to its object, it does theta mark its subject.
Under our analysisBurzio'sgeneralizationis eliminated,as passivization
has no effect on theta assignmentwhatsoever;rather, it affects only the
ability of the verb to assign accusative Case to the Theme. The Agent
theta role is assignedto the SPEC of VP in the usualway; however, some
languageswill not permitthis DP to be lexicallyrealizedbecause they do
not permit the SPEC of VP to receive Case.
Finally, strongpredictionscan be made about the presence or absence
of adjunctby-phrasesin variouslanguagesdependingon the status of the
SPEC of VP and the SPEC of IP in them. More specifically,one should
never find a languagewhere, in a given construction,the SPEC of VP is
Case-marked,and yet an adjunctby- phrase may still appear. We relate
the presence of these adjunctby-phrasesdirectly to the inability of the
SPEC of VP to be licensed in English. In languageslike Malagasyand
Tagalog, 'passive'morphologyand the verb licenses a lexical DP in the
SPEC of VP at S-structure.In languages like English, SPEC of VP is
never licensed and thus can be occupied only by PRO. Again, Bahasa
falls somewhere between the two language types. In Bahasa, the SPEC
of VP is not licensed by direct Case-marking.In this sense it is like
English, in that the by-phraseis availablebut only to specifythe reference
of a third person Agent. Unlike English, Bahasa can also express an
Agent in a Theme-Topicconstructionwithout the use of the adjunctPP.

5.

CONCLUSION

In this paper, we have proposed that the two NPs in Austronesianthat

behave like subjects- the Agent and the Topic - are associatedwith two
different SPEC positions: the SPEC of VP and the SPEC of IP, respectively. This proposal, as we have shown, allows for a straightforward

410

EITHNE GUILFOYLE

ET AL.

solution, within the Government-Bindingframework,to the problemsof

wordorderand the notion of subjectthat have perplexedAustronesianists
for many years, and provides a place for Austronesianlanguageswithin
a typologyof passives.

APPENDIX

The foregoing analysis of Austronesianlanguages has attempted to account for specific word order patterns and passivization forms while
achieving the greatest generalityin its conclusions about the 'shape' of
these languages. One specificpattern, however, that was not treatedwas
the alternateorderof TagalogAgent-Topicstructures,in whichthe Agent
DP remainsin the SPEC of VP (see (25aii)). We will argue here that this
order raises difficult but certainly not unanswerablequestions for our
analysis. One question is simply an empiricalone. When the Topic has
not moved to the SPEC of IP, do the propertiesclaimed to be sensitive
to the SPEC of IP still pick out the Topic? The answerappearsto be yes,
if we examine the one instance in which it occurs - namely, when the
Topic is the Agent and remains in the SPEC of VP. What is required
for well-formednesshere is the appearanceon the V of Agent-Topic
morphology:with this condition met, only the Agent DP may extracted
(45a-c), or may launch a floated quantifier(45d).
(45)
a.

Tagalog
Magsalis
angbabae ng bigas sa sako para
AT-will-take-outTOP-womanACC-riceOBL-sackfor
sa bata'.
OBL-child
The woman will take rice out of the sack for the child.

b.

Sin ang
magsalis
ng bigas sa saka para
who COMPA T-will-take-out
ACC-riceOBL-sackfor
sa bata'?
OBL-child
Who will take rice out of the sack for the child?

c. *Ano ang
magsalis
angbabae sa sako
what COMPAT-will-take-outTOP-womanOBL-sack

TWO SUBJECTS IN AUSTRONESIAN

LANGUAGES

411

para sa bata'?
for OBL child
What will the woman take out of the sack for the child?
d.

Magsalis
lahat ang mga babae ng mga libro
AT-will-take-outall TOP-pl womanACC-pl book
sa sako para sa bata'.
OBL-sackfor OBL-child
i. All the women will take books out of the sack for the
child.
ii. *The women will take all the books out of the sack for the
child.

We might say that the Agent is simply behaving as if it had moved, a

suggestionwhich can be made more precise in the following claim: even
when there is no movement, the topic morphologyof the verb still Casemarks the SPEC of IP, licensing a pleonastic pro in this position.33 This

pleonasticwill be coindexedwith the Topic, preventingmovementof any

other DP.
While this sort of pleonasticcoindexationmay seem very differentfrom
CHAIN formationbetween pleonasticsand VP-internalDPs in languages
such as English (see, e.g., Chomsky (1986b)), there are at least two
argumentsin favor of such an analysis.First, it explainsa minimaldifference between the alternateAgent-Topicand the Recent Pastconstructions
in Tagalog, which consists in the absence in the latter of any Topic
morphologyon the verb. Accordingto our analysis,the latterconstruction
would not license the SPEC of IP throughCase-marking,and thus would
have no pro in the SPEC of IP and no coindexationbetween it and the
SPEC of VP. This would, in turn, lead to the prediction that any VPinternalDP may extractthroughthe trulyempty SPECof IP (whichwould
behave very similarlyto SPEC of CP). This predictionis in fact borne out
in the examples below (based on McGinn 1988):
(46)

Kabibililang ni Pedro
ng tela
RP-buy just GEN-PedroACC-cloth
Pedro just bought some cloth.

33 It has been suggested (Lasnik 1989) that in there constructionsin English, both the
pleonasticand the DP in the VP are assignedCase. This is also what we would be saying
for these constructionsin Tagalog.

412

EITHNE

GUILFOYLE

ET AL.

(47)a. Sino ang kabibililang ng tela?

Who has just bought some cloth?
b. Ano ang kabibililang xi Pedro?
What has just been bought by Pedro?
c.

Sino ang kabibililang ni Pedro ng tela?

Who has just been bought some cloth by Pedro?

A second argumentin favor of such an analysisis based on data from

Kalagan, another Philippine language. In this language, Topic DPs,
though marked in the same fashion as those in Tagalog and Cebuano,
must remainin the VP whether or not they are Agents. Collins (1970, p.
4)34 briefly describesthe word order facts of Kalagan, claiming that the
underlying order is: V-Agent-Object-Instrument-Benefactive-LocativeTime. She adds that "the one regularexceptionis that when the ya-phrase
[Topic]is not the agent, it immediatelyfollows the agent, all other phrases
keeping their places." In our account this implies that the Agent Topic
remains in SPEC of VP, and all other Topics would appear in object
position, i.e., adjacentto the trace of the verb. There are two questions
that need to be answered here: (i) why are the circumstantialTopics
(instrumental,benefactive, locative, time) found in object position, and
(ii) why do all Topics remainwithin the VP?
As an answer to the first question, we would propose that the Pincorporationwhichis part of oblique topic constructionsforces the Topic
DP to appear adjacent to the trace of the verb, much like applicative
constructionsin Chichewa(Baker 1988). To answerthe second question,
we would claim that the coindexationthat we have arguedfor in Tagalog
between SPEC of IP and SPEC of VP is extended in Kalaganto include
coindexationbetween SPEC of IP and object position. This suggeststhat
languagesdifferwith regardto such coindexationpossibilities.In Tagalog,
the Agent optionally remains within the VP. In Cebuano, the Agent
always remainswithin the VP. In Kalagan, all Topic DPs remain within
the VP. In every case, however, the Topic will act as if it were in the
SPEC of IP position due to coindexationwith the pleonasticpro.
REFERENCES
Baker, Mark:1988,Incorporation:
A Theoryof Grammatical
FunctionChanging,University
of ChicagoPress, Chicago.

3 We would like to thankan anonymousreviewerfor directingus to this work.

TWO SUBJECTS IN AUSTRONESIAN LANGUAGES

413

Bell, Sarah:1976, CebuanoSubjectsin Two Frameworks,unpublishedPh.D. dissertation,

MIT.
Burzio, Luigi: 1986, ItalianSyntax:a Government-Binding
Approach,Reidel, Dordrecht.
Byma, Gary: 1987, Governmentand Bindingin Tagalog:an ErgativeAnalysis,unpublished
M.A. Thesis, Universityof Calgary.
Carrier-Duncan,Jill: 1985, 'Linkingof ThematicRoles in DerivationalWord Formation',
LinguisticInquiry16, 1-35.
Chomsky,Noam: 1981, Lecturesin Governmentand Binding,Foris, Dordrecht.
Chomsky,Noam: 1986a,Barriers,The MIT Press, Cambridge,MA.
Chomsky,Noam: 1986b,Knowledgeof Language,Praeger,New York.
Chung, Sandra:1976, 'On the Subjectof Two Passivesin Indonesian',in C. N. Li (ed.),
Subjectand Topic,AcademicPress, New York, pp. 57-99.
Chung, Sandra and Jim McCloskey:1987, 'GovernmentBarriersand Small Clauses in
Modem Irish',LinguisticInquiry18, 173-238.
Collins, Grace C.: 1970, Two Viewsof KalaganGrammar,unpublishedPh.D. dissertation,
IndianaUniversity.
Dell, Francois:1981, 'On CertainSententialComplementsin Tagalog',PhilippineJournal
of Linguistics12, 11-28.
Fukui, Naoki and MargaretSpeas: 1986, 'Specifiersand Projections',MIT WorkingPapers
in Linguistics8, 128-172.
Gerdts, Donna: 1988, 'Antipassivesand Causativesin Ilokano: Evidence for an Ergative
Analysis',in R. McGinn(ed.), Studiesin AustronesianLinguistics,Ohio UniversityPress,
Athens, OH, pp. 295-321.
Grimshaw,Jane and Armin Mester: 1988, 'Light Verbs and Theta-Marking',Linguistic
Inquiry19, 205-232.
Guilfoyle, Eithne: 1988, 'Parametersand FunctionalProjection', Proceedingsof NELS
XVIII.

Guilfoyle,Eithne:1990,FunctionalCategoriesand PhraseStructureParameters,unpublished
Ph.D. dissertation,McGillUniversity.
Guilfoyle,Eithne:1991,'Phrase-Structure
andPassive',in D. Bouchardand K. Leffel (eds.),
Viewson Phrase-Structure,
KluwerAcademicPublishers,Dordrecht,pp. 137-156.
Hung, Henrietta:1987, Functionaland LexicalCategoriesin BahasaMalaysia,unpublished
M.A. thesis, Universityof Toronto.
Hung, Henrietta:1988,'DerivedVerbsandNominalsin Malagasy',unpublishedms., McGill
University.
Kayne,Richard:1989, 'Null Subjectsand CliticClimbing',in 0. Jaeggliand K. Safir(eds.),
TheNull SubjectParameter,KluwerAcademicPublishers,Dordrecht,pp. 239-261.
Keenan, Edward:1976, 'RemarkableSubjectsin Malagasy',in C. N. Li (ed.), Subjectand
Topic,AcademicPress, New York, pp. 247-301.
Keenan, Edward and Bernard Comrie: 1977, 'Noun Phrase Accessibilityand Universal
Grammar',LinguisticInquiry8, 63-99.
Kitagawa,Yoshihisa:1986,Subjectsin Japaneseand English,unpublishedPh.D. dissertation,
UMass, Amherst.
Koopman,Hilda and DominiqueSportiche:1988, 'Subjects',unpublishedms., UCLA.
Kroeger,Paul: 1991, PhraseStructureand GrammaticalRelationsin Tagalog,unpublished
Ph.D. dissertation,StanfordUniversity.
Kuroda,S.-Y.: 1988, 'WhetherWe Agree or Not: a ComparativeSyntax of English and
Japanese',in W. Poser (ed.), Papersfrom the SecondInternationalWorkshopon Japanese
Syntax,CSLI, Stanford,pp. 103-143.
Lasnik, Howard: 1989, 'Case and Expletives:Notes Towarda ParametricAccount', talk
givenat the SecondPrincetonWorkshopon ComparativeGrammar,PrincetonUniversity.
Lewis, M. Blanche:1969, SentenceAnalysisin ModernMalay,CambridgeUniversityPress,
Cambridge.
McCloskey,James:1984, 'Raising,Subcategorization,
and Selectionin ModernIrish',Natural Languageand LinguisticTheory1, 441-485.

414

EITHNE GUILFOYLE

ET AL.

McConnell-Ginet,Sally: 1982, 'Adverbsand LogicalForm', Language58, 144-184.

McGinn, Richard:1988, 'Governmentand Case in Tagalog',in R. McGinn(ed.), Studies
in AustronesianLinguistics,Ohio UniversityPress, Athens, OH, pp. 275-293.
Othman,Arbak:1981, TatabahasaBahasaMalayasia,SarjanaEnterprise,KualaLumpur.
Pollock, Jean-Yves:1989, 'Verb Movement,UniversalGrammarand the Structureof IP,'
LinguisticInquiry20, 365-424.
Postal, Paul: 1969, 'On So-CalledPronounsin English',in D. Reibel and S. Schane(eds.),
ModernStudiesin English,Prentice-Hall,EnglewoodCliffs, NJ, pp. 201-224.
Rizzi, Luigi: 1990, ReladvizedMinimality,The MIT Press, Cambridge,MA.
Schachter,Paul: 1976, 'The Subjectin PhilippineLanguages:Topic, Actor, Actor-Topicor
None of the Above', in C. Li (ed.), Subjectand Topic, AcademicPress, New York, pp.
491-518.
Schachter,Pauland Fe Otanes: 1972, TagalogReferenceGrammar,Universityof California
Press, Berkeley.
Seiter, William:1975, 'InformationQuestionsin PhilippineLanguages',unpublishedms.,
Universityof California,San Diego.
Sie, Ing Djiang:1989, TheSyntacticPassivein BahasaIndonesia,unpublishedPh.D. dissertation, Universityof Amsterdam.
Sportiche, Dominique: 1988, 'A Theory of Floating Quantifiersand its Corollariesfor
ConstituentStructure',LinguisticInquiry19, 425-450.
Travis,Lisa: 1988, 'The Syntaxof Adverbs',McGill WorkingPapersin Linguistics,Special
Issue on ComparativeGermanicSyntax,280-310.
Travis,Lisa and EdwinWilliams:1983, 'Externalizationof Argumentsin Malayo-Polynesian
Languages',The LinguisticReview2, 57-78.
Williams,Edwin:1981,'ArgumentStructureand Morphology',TheLinguisticReview1, 81114.
Williams, Edwin: 1987, 'ImplicitArguments, the Binding Theory and Control', Natural
Languageand LinguisticTheory5, 151-180.
Received 21 November1989
Revised 1 August 1991
Eithne Guilfoyle
Departmentof Linguistics
Universityof Calgary
2500 UniversityDrive NW
Calgary,Alberta
CanadaT2N 1N4
HenriettaHung
LinguisticsProgram
Departmentof Psychology
BrandeisUniversity
Waltham,MA 02254
Lisa Travis
Departmentof Linguistics
McGillUniversity
1001SherbrookeWest
Montreal,PQ
CanadaH3A 1G5