Malagasy Clause Structure (2000) - Ileana M. Paul

Malagasy clause structure
Ileana M. Paul
Department of Linguistics
McGill University
Montreal, Canada
A thesis submitted to the Faculty of Graduate Studies

in partial fulfillment of the requirements
of the degree of Doctor of Philosophy
January 2000
2000 Ileana M. Paul
voasary amoron-dalana
haka aho
tsy tompony
tsy haka aho
maha te-hihinana
'a lemon by the road
if I take it
I will not be its master
if I don't take it
I will want to eat it'
-Malagasy proverb
Abstract
This thesis explores the nature of voice in Malagasy, a language spoken in
Madagascar. I n c h apter 2, it is claimed that di fferent passives promote
arguments from different structural positions. Ev idence is provided for a
particular p o sition, [Spec, v2P], w h e r e a c e r t ai n c l ass o f e l e m e n t s
("displaced themes") may b e generated. O n e p a r ticular passive, the ap refix, pr omotes t o s u b ject elements i n t h i s p o sition . In ch a p t e r 3 ,
a rguments a r e pr e s ented i n fav o u r of a str uc t u ra l a n a l y si s o f
circumstantial topic ( CT). C T m o r p h o logy l i censes all arguments of t h e
verb. Du e to a requirement that all clauses have a subject (the Extended
Projection Principle), some element other t ha n a D P s t r u cturally C a se
marked by the verb must raise to subject. Finally, chapter 4 addresses the
left periphery in the Malagasy clause, in particular the structural positions
of topic and focus.
Resume
Dans cette these, il est question du statut de la voix en M a lgache, langue
parlee a Madagascar. D ans chapitre 2, il est avance que des arguments
d ifferents montent a l a p o s i tion d u s u jet l o r sque l e v e r b e p o r t e l e s
differents affixes passifs. C e r t ains arguments (des " t hemes d eplaces")
peuvent etre generes dans [Spec, v2P]. Si le verbe porte le prefixe a-, un
argument genere dans [ Spec, v2P] monte a la p o sition d u s u jet . D a n s
chapitre 3 , i l s' a g i t d' u n e a n a l ys e s t r u cturale d e l a con s t r u ction
"circumstantial t o p i c" ( CT).
L a m orp h o l o gi e C T li c e nse t o u s l e s
arguments du v e r be . A c a u s e d u " E x t ended Projection Principle", un
e lement sans cas structural doit assumer l a f o n ction d u s u j ect. Da n s
chapitre 4, i l e s t q u e stion d e l a p e r i p h erie g auche d e l a c l a use, en
particulier, les positions "topic" et "focus".
Acknowledgments
So many people touch one's career as a graduate student. Since I cannot
mention everyone, I will err on the side of brevity.
I first want to thank Lisa Travis, my advisor. Lisa has been a source of
inspiration, from my first baby steps in syntax to the last line of this thesis.
I have b e nefited n o t o n l y f r o m h e r g r e a t s e nse o f r e a soning a n d
knowledge of syntax and Malagasy, but also from her friendship.
All the professors, students and staff at the department of linguistics at
McGill deserve thanks for their help along the way. I w a n t to express my
special appreciation t o
M a r k B a ke r f o r m a n y s t i m u l ating. Jonathan
Bobaljik very g e nerously o f fered comments o n v e r sions o f c h a pter 2

during his summer holiday.
A warm thank you to Ed Keenan for offering me the chance to do field
work
w i t h hi m
in
M ada g a scar i n 19 9 5 a n d for
hi s
con t i n u ed
encouragement over t h e y e ars. F u r t he r a p p r eciation t o R o g er-Bruno

Rabenilaina and the other linguists at the Un iversite d'Antananarivo for
their hospitality during my stay in Madagascar.
T his thesis would not e xist w i t h out th e p a tience and in sight of m y
Malagasy c onsultants i n
M on t r e al :
Saho l y H an i t r i niaina, M i c h ele
Ratovonony and Julie Rabemananjara. Mi s oatra betsaka, tompoko! I tha nk

McGill and FCAR for funding which made this research possible.
In the greater linguistics community, several people have p r o v ided
i nvaluable feedback. I
conferences for
w o u l d l i k e t o t h ank th e a u diences at the A F L A
t h ei r i n p u t a n d s u p p o rt, especially Bill D a v ies, Diane
Massam, Matt Pearson, Peter Cole and G abriella H e rm on.

Norvin Richards for hi s co mments and fo r
Sa l a matto
k i n dl y a n swering m y l a s t-
minute, panic-ridden email messages.

Finally, I want to thank my friends, the Biomass (Martyna Kozlowska,
L ara Riente and Jila Ghomeshi), for r eminding m e t o e njoy l ife and f o r
teaching me how to d r in k v o d ka .
their support. M y
T h a nk s also to m y e x i n -laws for all
f a m i ly, especially my p arents, have been wonderfully
generous with love and understanding.

I dedicate this thesis to Felix, my ray of sunshine.
List of Abbreviations
1-first person
2-second person
3-third person
abs-absolutive
acc-accusative
asp-aspect
AT-Actor Topic
AV-Actor Voice
ben-benefactive
BT-Benefactive Topic
C-complementizer
CT-Circumstantial Topic
def-definite determiner
det-determiner
erg-ergative
ex-exclamative
excl-exclusive
foc-focus particle
fut-future
gen-genitive
incl-inclusive
IT-Instrumental Topic
LT-Locative Topic
neg-negation
nm-nominalizer
nom-nominative
NPI-negative polarity item
obl-oblique
opt-optative
OV-Object Voice
P-preposition
part-partitive
pass-passive
perf-perfective
pl-plural
pres-present
pst-past
Q-question marker
rel-relative marker
sg-singular
sp-subject prefix
top-topic particle
trans-transitivizer
TT- Theme Topic
Table of contents
Abstract.
Acknowledgements
Abbreviations
Table of contents
Chapter 1: Introduction
1 Voice
2 Voice in Austronesian
2 .1
S t r u cture
2.2
S e m a n tics.............
2.3
T he t a roles.
3 Malagasy
3 .1
W or d o r d e r
3.2
V oi c e ....................
3.3
M o r p h ology
4 Theoretical assumptions.
5 Organization of the thesis.
Chapter 2: Theme Topic
1 Introduction.
2 Passive in Malagasy
2.1
Pa s s ive formation .
2.2
a - p a ssive vs.- Vna passive: basic distribution.
2.2.1 O n e p assive.
2.2.2 T w o p a ssives
2.2.3 W h i c h passive~
2.3
L oc a t ive alternation and instrumental advancement....
2.3.1 B a s ic distribution
2.3.2 O p t i o n ality
2 .3.3 C l e f t s .
2.3.4 P r e p o sitions.
2.3.5 A rg u m e n ts: Word order .
2.3.6 C l e f t s again.
2.3.7 P r e p o sitions again
2.3.8 A p r e d i ction.
2.4
D at i v e verbs.
.1
.1
4
....5
.6
.7
.8 12
.13
.16
.17
.19
.19
20
23
25
25
26
31
.....33
34
36
37
38
39
41
42
43
44
3 Pearson (1998a).
48
4 A unified analysis of passive
54
4.1
S t r u c t ure ............................................................................... .....54
4.2
S e m a n tics.............................................................................. .....56
4.3
Cha n g e of location.
57
5 Discussion.
59
5.1
Ca s e and transitivity.
59
5.2
Con s t i t u ency.
.63
5.3
Ba s e generation vs. movement
.66
5.4
Cr o s s-linguistic evidence
.68
5.5.1 I n s t r u m ental advancement
.69
5.5.2
5.5.3
L o c a t ive alternation.
A pp l i c ative
6 Aspect
6.1
6.2
Kimaragang Dusun.
Malagasy .
6.2.1 L o c a t ive alternation.
6.2.2 D a t i v e shift
6.2.3 O n t h e i m p o rtance of roots..
6.2.4 F u r t h e r data.
7 Conclusion.
7.1
Pa s s ive
7.2
[Spec, v2P].
7.3
7 .4
Le x i cal semantics
T T v s . CT .
Chapter 3: Circumstantial Topic

1 Introduction.
2 CT clauses.
2.1
T h e c ore data
2 .2
T ar g e t s .
2.3
S u b j e cts.
2 .4
Cle f t s .
2.4.1 A T c l e fts.
2.4.2 C T c l efts.
2.5
A no t e on CPs..
2.6
W he r e are we?.
3 C T = promotion of adjuncts.
3.1
T he t a - agreement
3.2
A dj u n c ts vs. arguments ...........
3.3
P P a r g u m ents
3 .4 EC T M
3 .5
T h e o ry .
4 C T = preposition incorporation.
4.1
4.2
4.3
5 Analysis.
6 ECTM.
6.1
Gui l f o y le, Hung and Travis.....

A ppl i c atives..
Pre p o sitions .
6.2
Partitives in Malagasy
6.1.1 P e r i p h rastic partitives
6.1.2 S u b j ect position.
6.1.3 E x i s tentials
6.1.4 E x c e ptional Circumstantial Topic Marking.....
ECTM: An analysis..
6.3
Data.
6.4
6.5
6.3.1 P r e p o sitions and partitivity

6.3.2 T h e m e or not a theme?
Floating quantifiers: a problem?
Malagasy madness?.
.70
.71
73
.74
.76
.77
.79
.80
82
.85
.86
.86
.89
90
91
91
94
94
96
.100
.103
.107
.108
.110
.111
.111
....113
.116
.117
.117
.118
....118
.119
122
124
128
128
129
.130
.130
....131
.132
.134
.134
.135
.138
.141
6.5.1 T a galog
6.5.2 I n i b a l oi.
6.5.3 P a r t i t ive Case
6.6
Co n c l u sion
7 Circumstantial Topic: Other languages.
8 Conclusion.
Chapter 4: The left periphery
1 Introduction.
2 The left periphery.

2.1
T op i c and focus.
2.2
T op i c > focus
2.3
di a t o p ics..
2 .4
Fo c u s .
2.5
T h eb o d y gu ard.
2.6
M al a g asy clause-structure
3 The structure of clefts.
3.1 It-clefts.
3.2
Ps e u d o-clefts.
3 .3
Pi v ot =predicate
= headless relative
3.4
Pr e s u p positional clause
3.5
Int e r p r etation
3.6
Int e r l u de: ve
4 Cleft as focus.
5 Multiple clefts
5.1
Coo r d i n a tion
5.2
A m a l g a m ation.
5.3
M ul t i p l e specs
5.3.1 S l a v ic multiple wh-fronting
5.3.2 B o d y g u ard ~ focus
5.3.3 B o d y g u ard= topic
5.4
W he r e are we~
5.5
S p e c ulations .
5.6
W h - bodyguard.
5 .6.1
5 .6.2
6 Conclusion.
Or d er
D - l i nking
.142
.142
.143
.144
.145
.146
.148
.148
.148
.150
.151
.152
.157
.159
.161
.163
.163
.167
.169
.172
.175
.176
.181
.185
.188
.189
.193
.193
.196
.199
204
205
207
208
210
211
Chapter 5: Conclusion .
213
Appendix .
216
222
References
Chapter 1: Introduction
V o i ce
One of the most striking properties of western Austronesian languages is

t heir verbal m o r p h ology .
C e r t ai n m o r p h emes ap pear t o e n c ode t h e
relation that the subject NP bears to the rest of the clause.' T h i s can be
observed in the M alagasy example in (1), where the voice morph ology
differs depending on which DP appears in the subject position.
(I)
a.
Nanapaka
ity h azo ity
tam i n 'n y
ants y i Sahondra.
pst.AT.cut t
h i s tree this p s t .P.gen.det knif e S a hondra
'Sahondra cut this tree with the knife.'
b.
Nota p ahin'i Sahondra
tamin'ny
ants y i t yhazo ity.
pst. TT.cut.gen.Sahondra
pst. P .gen.det knif e t h i s tree this
'This tree was cut by Sahondra with the knife.'
Nanapahan'i Sahondra
i t y hazo ity
ny a n tsy.
pst.CT.cut.gen.Sahondra this tree this d e t knife

'The knife was used by Sahondra to cut the tree.'
I n (1a), the agent i s t h e s u bject an d t h e v e r b b e ar s A c t orTopic (A T )

marking. I n ( 1b), it is the theme that appears in the subject position and
the verb is marked with ThemeTopic (TT). Finally, in (1c) the instrument
is the subject when the verb takes CircumstantialTopic (CT) morphology.
I will refer to this morphology as the voice system. Du e to its prevalence
i n this l anguage family, th e w e stern A u s t r onesian v o ice sy stem h a s
received a certain amount of attention by grammarians and linguists alike.
The question arises as to th e
m orphology
p r o pe r t r e atment o f v o i ce .
Is the
s e nsitive t o t h e t a-roles, category, p o sition, f u n ction, o r
s omething else? Before addressing the voice system of M a l agasy, I wi ll

begin with some background on voice and passive. This is a topic that has
long captured the interest of linguists. Since it is not m y
g o a l t o g iv e a
' Throughout I will mark the (matrix) grammatical subject with a dotted underline. Other
constituents may be marked with brackets or boldface.
Introduction
comprehensive literature review, which would take us too far afield, I will
provide but a brief overview of the issues before turning to the Malagasy
data.
As a starting point, let us consider past treatments of the active-passive
a lternation.
M a n y n a t ural languages display alternations similar t o t h e
ones in (2) and (3).

(2)
(3)
a.
Felix cut the bread.
b.
The b r ead was cut by Felix.
a.
Felix a coupe le pain.
b.
L e painete
a coupe par Felix.
The two active sentences in (a) express basically the same meaning as the
p assive ones in (b). Nevertheless, the surface forms are quite different: i n
(2a) and (3a), the agent is the subject; in (2b) and (3b), the theme is the
subject. Much syntactic research has centered on formalizing the relation
between the active and the passive.
Drawing on d at a f ro m a r a n g e o f l a n g u ages, Relational G r am m ar
characterizes passive with two universal properties (Perlmutter and Postal
(1977)):
(4)
a. A direct object of an active clause is the (superficial) subject of the

'corresponding' passive.
b. The subject of an active clause is neither the (superficial) subject nor the
(superficial) direct object of the 'corresponding' passive.
Due to o t her l aw s o f g r a m ma r ( e .g. th e Stratal Un iqueness Law), the

a ctive subject i s
d e m oted i n a pa s s iv e c l ause, often s u r f acing a s a
" chomeur" (oblique). T hus passive has a dual nature: th e p r o m otion o f

the object to subject and the demotion of the subject.
A standard Government-Binding analysis claims that the theme moves
into the subject position because the passive verb can no l o nger assign
accusative Case (e.g. Chomsky (1981)). F u r t hermore, th e p assive verb
lacks the ability to project an external argument. (See Baker, Johnson and
R oberts (1989) for a slightly d i f ferent GB account.)
a ctive verb ma y
T h e s u bject of t h e
o p t i onally a p pear as an o b l i que adjunct. T h e s e t w o
properties are related in what is known as Burzio's Generalization (Burzio
(1986)):
Chapter 1
A verb which lacks an external argument fails to assign accusative
Case.
A verb which fails to assign accusative Case fails to theta-mark an
external argument.
As in RG, GB passive involves two components, here linked to Case and

theta-role assignment.
T here are some d i sadvantages to th e ab ove characterizations.
For
example, both RG and GB treat passive as consisting of two-parts. But do

these two parts always correlate? There are languages where promotion
of the logical object to subject does not lead to d emotion of th e l ogical
subject (see section 3.2 on Malagasy). There are also languages where the
demotion of the logical subject does not coincide with the promotion of a
logical object ( e.g. I t elmen
O o nathan B o baljik, p . c.) an d i m p e r sonal
passives in general (Comrie (1977)). Therefore, (5a) and (5b) are arguably
independent of each other, as are the tw o p a rts of th e RG definition of
passive. Due to these considerations, (4) and (5) have been modified by
subsequent research in both the RG and GB literature.
Another limitation of both the RG and the GB approaches is that they
only discuss the active-passive alternation. Some languages exhibit mor e
than the one-way distinction between active and passive illustrated in
(2)
and (3).' S p ecifically, passive is defined as a transformation that affects

arguments of the verb (objects in RG or accusative Case-marked NPs in
GB). Austronesian voice alternations, however, are not limited to passive
and hence often involve not only arguments, but also adjuncts. M a lagasy
is typical of such a system and is commonly assumed to h ave a t h r eevalued v o i c e
d i s t inction: a c tive, p a ssive, c i r cumstantial (sometimes
referred to as "relative"
).
In this thesis, I will show that not only does Malagasy enjoy more than
the active-passive alternation, it also benefits from distinct types of passive
f or different internal arguments.
F o r t h e m o s t p a rt, I w i l l i g n or e t h e
active sentences, concentrating on the passive and circumstantial and on

' Perlmutter and Postal (1977) fully acknowledge this limitation. B e l l ( 1976) provides an RG
analysis of Cebuano, which displays voice alternations similar to M alagasy. See also Gerdts
(1988) on Ilokano. Chomsky (1981) also points out the variation in passive-like constructions
cross-linguistically. For an overview of this variation, see Shibatani (1985).
Introduction
the NPs that appear in the subject position of t h ese clauses. Clearly, a
proper characterization of voice will depend on a careful investigation of
each alternation.
2
Vo i ce in Austronesian
Let us now t ur n t o e xisting analyses of voice in w e stern A u s tronesian.

These analyses will be discussed in more detail in chapters 2 and 3. Here, I
simply introduce the basic principles of each.
2.1 Structure
Guilfoyle, Hung and Travis (1992) (henceforth GHT) propose a purely

structural account of voice alternations. Consider the voice paradigm in
(1), repeated in (6).

a.
(6)
Nanapaka
ity h azo ity
tam i n 'n y
ants y i Sahondra.
pst.AT.cut
thi s tree this p s t .P.gen.det knife S a hondra
b.
tamin'ny
i t y hazo ity
ny a n tsy.
pst.CT.cut.gen.Sahondra this tree this

d e tknife
For GHT, the active morpheme assigns accusative case to the direct object
and the passive morpheme assigns genitive to the agent in [Spec, VP]. In
(6a), therefore, the agent does not receive Case and must raise to [Spec,
IP] for nominative. I n
( 6 b), on th e other h and, the theme is not Case-
m arked in i t s b ase p osition an d r a i ses to t h e s u bject p o sition . On e

advantage of this proposal is that it neatly captures certain properties of
CT.
As noted b y
G H T a n d o t h e rs, i n o n e r e s pect, th e c i r cumstantial
construction resembles both the active (AT) and the passive (TT). Like in
t he active, a transitive CT v erb m a y t a k e a d i r ect object m arked w i t h
accusative Case. L i k e in th e p assive, the agent appears in the genitive.
Chapter 1
These two properties are illustrated in (6c): Sahondrabears genitive Case;
ity hazo ity 'this tree' is in (unmarked) accusative. This combination of AT
and TT is also present to some degree in the morphology of CT: the verb
bears both an active prefix and a passive suffix.
( U n l ike in th e passive,
however, this suffix is invariably -ana and never -ina. ) The morphology of
the forms in (6) is given in (7).

(7)
a.
AT:
m-an - t apaka
manap a k a
pres-AT-cut
TT:
tap a ka-ina
tapahina
cut-Vna
CT:
a n-t a paka-ana
anapahan a
AT-cut-ana
T he above p arallels have le d s om e r e searchers t o c l ai m t h a t t h e

circumstantial indeed combines the active and the passive in the syntactic
structure.
T h i s i s t r u e o f b o t h t r a d i tional g r a m m ars (e.g. Rajemisa-
Raolison (1966)) and in the Principles and Parameters literature (e.g. GHT).
Recall that GHT state that active morphology assigns accusative Case to
the theme, while passive morphology licenses genitive Case for the agent.
P utting th e t w o t o g e ther c r eates CT , w h i c h i s c h a racterized b y t h e
availability of both accusative and genitive.
F u r t h ermore, GH T suggest
t hat CT i n v o l ves p r eposition i n corporation, along t h e l i nes o f B a k e r

(1988). Following preposition incorporation, the object of the preposition
(ny antsy'the knife' in (6c)) no longer receives Case and must raise to the
matrix subject position for nominative. It is via P-incorporation that GHT
integrate CT into the traditional Case analysis of voice.
2.2 Semantics
Keenan (i n
p r e ss) p r o poses a s e m antic a n alysis o f M a l a gasy v o i ce.
Simplifying somewhat, voice morphology denotes a function. For CT, for

e xample, the function is true iff the n ominal i n s u bject position can b e
r elated to th e p r edicate via some p r eposition. I n o t h e r w o r d s , w h i l e
K eenan argues against p r e position i n c orporation i n t h e s y n t ax , t h e
Introduction
prepositional meaning is encoded in the semantics of the CT morphology.
In this way, both Keenan and GHT base their analyses of CT on the link
between CT and prepositions.
2.3 Theta roles
Other linguists have focussed on th e l in k b e t w een t h ematic roles and
voice morphemes.
A t f i r s t g l ance, the v oice system i n M a l agasy and
many other western Austronesian languages appears to encode the theta

role of the NP in subject position.
(8)
a.
A cto r Topic (AT): agent subject
b.
The m e Topic (TT): theme subject
c.
Circ u m stantial Topic (CT): oblique subject
Due to the correlation between voice marking and th ematic roles, some
researchers have proposed a system of " t heta agreement" (e.g. Sityar (in
press))
(9) The nominal features of VoiceP must agree with the thematic features of the
topic.
The voice morphology on the verb is therefore a special form o f s u bject
agreement. F o ll ow ing verb raising, the subject NP and the verb are in a
l ocal, spec-head relation which mediates agreement. I n
s u p p or t o f t h i s
approach, current research suggests that theta-roles are in f act features

that must be checked, similar to Case (Boskovic (1994); Hornstein (1999)).
With theta-roles as features, theta-agreement is parallel to other types of
agreement, for example number and person.
S umming up, w e see that t h e M a l a gasy v o ice alternations can b e
viewed from a wide range of perspectives. One of the goals of chapters 2
a nd 3 i s t o
a s sess these di fferent analyses. A l t h o u g h I a r g u e f o r a
structural analysis of voice similar to GHT, I offer some modifications and

r efinements to t h eir p r o p osal. B e f or e d e l v in g i n t o t h e v o i c e d at a i n
c hapter 2, I
w i l l p r o v i d e s om e b a ckground o n M a l a gasy syntax a n d
' Although Sityar's proposal is for Cebuano, it can easily be adapted to any language with similar
voice alternations. She refers to as "topic" the position that I am calling "subject".
Chapter 1
morphology.
3
Mala as
The object o f
s t u d y i n t h i s d i s sertation i s t h e s y n ta x o f M a l a gasy.
Malagasy is a
w e s tern A u s t r onesian l an guage sp oken o n t h e i s l and
M adagascar b y
o v e r 1 2 m i l l ion people. D a h l ( 1951) suggests that t h e
c losest relatives ar e t h e S .E. B arito l a n guages, such a s M a a n jan i n

Kalimantan (Borneo). T h ere are various dialects in Madagascar; all th e
data cited are from th e M e r ina dialect of the capital and central plateau
r egion.'
U n l ess otherwise indicated, the data in t h i s d i ssertation w e r e
collected from n ative speaker consultants in M adagascar in 1995 and in

Montreal from 1993 to 1999. Malagasy data from other sources have been
checked with th e M o n t real consultants. A l l i n s tances of d i sagreement
among consultants have been noted in the text.
The grammar of M alagasy has been studied in the Western tradition
since the arrival of missionaries in the last century.
A M a l a gasy-French
dictionary was published in 1888 (Abinal and Malzac (1888)) and remains a
standard reference. Since then, several grammars w r i t ten by M a l agasy
linguists h av e
a p p eared ( e .g. R a hajarizafy ( 1 960); Rajemisa-Raolison
(1966); Rajaona (1972)). Due to ties with France, current linguistic research
i n Madagascar adopts th e t r a nsformational t h e ories o f G r o s s (1975);
Harris (1976). Several theses have appeared on different aspects of clause
structure (e.g. verbal complementation (Rabenilaina (1985)), adjectives
(Rahalalaoherivony
R abaovololona
( 1995)),
( 1 991))).
te m p ora l
adv e r b s
(Raharinirina-
Thus M al a g asy s y n t a x h a s be e n w el l -
documented for over a century.

In this section, I
w i l l g i v e s om e b a ckground o n M a l a gasy clause-
structure and morphology w h ich will be helpful to the reader unfamiliar

w ith thi s l a n guage.
I n t h e m a i n c h a p ters o f t h i s d i ssertation, I w i l l
therefore refer back t o
t h e i n t r o duction fo r c e rtain p o i nts not d i r ectly
related to the analysis at hand. The following is nevertheless not meant to

be a
c o m p r ehensive o v e r view o f M al a g asy g r a m m ar .
Inst e ad , I
concentrate on structures that will be relevant to the main content. For a

detailed discussion of Malagasy morphology, I refer the reader to Keenan
One of my consultants speaks the Betsileo dialect, which is also from the central plateau region
and is very similar to Merina. All data have been verified across consultants and differences in
judgements are mentioned.
Introduction
a nd Polinsky (1996). References to w o rk s o n
M a l a gasy syntax w il l b e
given throughout.
3.1 Word order
Malagasy is a VOS language with clear evidence for a constituent break
between the subject and the rest of th e clause, as will be shown b elow .
F or simplicity, I refer to this constituent as VP, although it may b e som e
larger XP.'
I m p o r t antly, the subject is VP-external. E x a m ples of b asic
M alagasy sentences are g i ven i n
( 1 0), w h er e t h e V P i s i n d i cated b y
bracketing.'
[Dokotera] i Bakoly.
d octor
Bako l y
'Bakoly is a doctor.'
[Hendry] ny ankizy.
w ise
det c h i l d
'The children are well-behaved.'
[Mividy
m o f o h o a n - dRabe] ity yehiyayy ity.
AT.buy
b r ead for a cc-Rabe this woman this
'This woman buys bread for Rabe.'
[Tany
an- t sena] izy.
pst.there at-market 3sg(nom)

'She was at the market.'
In each case, whether the predicate is nominal, adjectival, verbal or other

a nd despite the lack of a copula, I assume there t o b e a
indicated.
m a t ri x V P a s
A l l c a tegories are generally head-initial. T h u s d e t erminers
precede head nouns as in (10b) and prepositions are initial in PP (10d).

Somewhat unusually, demonstratives frame N Ps, as in
analysis
of
DP
stru c t u r e i n
( 10c). For a n
M ala g a sy , s e e Z r ib i - H e rt z
Mbolatianavalona (1999).
' Keenan (in press) refers to this constituent as PredP.
' All proper names in Malagasy include a determiner,i (10a), Rn (10c) or ry for plural proper
names.
and
Chapter 1
Keenan (1976) provides several tests that support the division between
VP and the subject, indicated by th e brackets in (10). Fo r
e x ample, the
question particle ve an d th e n egative polarity item in t sony 'n o l o n g er'

appear between the VP and the subject.'
( 11) a .
Mividy m ofo ho an'ny ankizyve i Bakoly?

A T.buy bread for
acc'det child
Q Ba k o l y
'Does Bakolybuy bread for the children?'
b.
T sy m i v id y
mof o ho an'ny ankizy
neg AT.buy bread foracc'det child
ints o ny i B a koly.
NPI
B akol y
'Bakoly no longer buys bread for the children'

Malagasy thus appears to be a
h i g hl y c o n figurational language, unlike
s ome of the Philippine languages, for example, which have f reer w o r d

order.
As well as appearing in a well-defined structural position, the subject in
M alagasy is subject to a
d e f i niteness/specificity restriction.
Inde f i n i t e
subjects are ungrammatical, as illustrated in (12) (cf. (10b))."

( 12) * Mihir a
A T.sing
ank t z y .
c hild
'Some children are singing.' or 'A child is singing.'

T his restriction on
t h e s u b ject p osition w i l l b e come i m p o r t ant i n t h e
discussion of CT in chapter 3.
I n order to a ccount fo r t h e V O S w o r d o r d er, G u i l foyle, H un g a n d
T ravis (1992) propose the following structure, where [Spec, IP] is to t h e
right.
' In chapter 4, section 3.6, I show that in fact ve is a second position clitic.
" As indicated by the translations of (12), there is no overt number marking on common nouns.
Examples such as (12) are therefore typically ambiguous. To simplify the text, I will give a
single translation.
Introduction
The verb raises to I and the subject DP raises to [Spec, IP] for nominative
Case. Other DPs are assigned Case in their base positions.
Recent research on phrase structure calls into question the analysis in
(13). K ayne (1994), for example, claims that p h rase structure is hi ghly
r estricted, only
a l l o w in g s p ecifier-head-complement u n d erlyin g o r d e r .
Thus the rightward [Spec, IP] position in (13) is problematic. A d o p t ing a

Kaynian framework, some r esearchers have s u ggested that
VOS arises due to ( some f or m
M a l agasy
o f ) V P - f r onting ( e .g. Pensalfini (1995);
Rackowski (1998); Rackowski and Travis (to appear)). This line of analysis
has been pursued in detail in various papers by Pearson (Pearson (1996b;
1998b; to appear)). Although these analyses differ in detail, the basic idea
c an be illustrated in th e f o l l o w in g e x ample . A s s h o w n i n ( 1 4 b), t h e
g rammatical subject raises to a position c-commanding the VP .
(or similar projection) moves to [Spec, TP].

( 14) a .
[ppep~cQte Milalao
baolina ] [ z Ubject
AT.play b a ll
ny zazavavy ].
det girl
'The girls are playing ball.'
10
The VP
Chapter 1
AgrSP
predicate
NP
vp
subject
For the most part, I adopt the GHT analysis of Malagasy. In certain cases,
however, I discuss the implications for a predicate-fronting analysis (see in
particular chapter 4). I also remain agnostic as the articulated structure of
IP (Pollock (1989)), as it does not bear on the issues in this thesis.
As mentioned above, the VOS order is fixed; Malagasy does not have
the free w or d o r de r c o m m only a v a ilable in th e Philippine and M a l ay
languages. In certain contrastive contexts, however, SVO is possible.
(15)
N y m pianatra mamaky teny, ny mpampianatra mihaino.

det student
A T xe a d w ord det teacher
AT.listen
'The students read aloud, the teacher listens.'
T he subject can also b e f r o n te d f o r t o p i c alization o r c l e f ting, t o b e

discussed in chapter 4.
( 16) a .
[ Ny mpianatra ] dia mamaky teny.

det student
top A T . r ead word
'As for the students, they are reading.'
[ Ny mpianatra ] no mamaky teny.

det student
foc A T . r e ad word
'It's the students who are reading.'
Other subject-initial "clauses" are given below.
( 1 7a) is an example of a
temporal adjunct, headed by th e p r eposition amin'.

clausal complement to a perception verb.
11
( 1 7b) illustrates the
( 1 7c), finally, is an instance of
Introduction
raising to object.
(17)
a.
Nalah e l o i Sahondra [tamin-dRasoa
an-dRabe].
nan o r oka
pst.sad
S a h ondra pst.P.gen.Rasoa p s t .AT.kiss a c c -Rabe
'Sahondra was sad when Rasoa kissed Rabe.'
Mijery
[ny namany
mikap o k any alika]
i Sahondra.
AT.watch det friend.3(gen) AT .beat det dog
Sahondra
'Sahondra watches her friends beat the dog.'

Mihevitra [ an-dRakoto h o hanasa
A T.think acc-Rakoto
l amba ]
C fu t . A T.wash c l o t h
R a be.
Rabe
'Rabe thinks that Rakoto will wash clothes.'
In all the above cases, the bracketed constituent appears to b e c lausal

r ather t h a n
n o m i n a l i n na t u r e ( e.g. v e rbal
morphology
o n t h e ve r b).
r a t h e r t h a n no m i n a l
Cl ea r l y , t h e s t r u c t ur e o f the cl a u sal
complements in (17) is debatable, but on the surface all exhibit SVO order.
I will not address these issues in this thesis.
3.2 Voice
In section 1, I outlined some properties of the Malagasy voice system. The
examples in (18) repeat the standard voice alternation.
( 18) a .
Nanapaka
ity h azo ity
tam i n 'n y
ants y i Sahondra.
pst.AT.cut
b.
tamin'ny
i t y hazo ity
ny a n tsy.
pst.CT.cut.gen.Sahondra this tree this d e t knife

12
Chapter 1
In an active (AT) clause, the agent is the subject, while in the passive (TT)
and circumstantial (CT) constructions, the agent surfaces immediately to
t he right of the verb, bearing genitive case. GHT t ake advantage of th e
VP-internal subject hypothesis to account for th e p osition of t h e a gent.
They claim that in TT and CT, the agent is licensed by genitive case in
[Spec, VP]. (In AT, genitive is not available and the agent raises to [Spec,
IP].) The difference between English and M alagasy therefore lies in the
f act that the passive agent is not demoted to an oblique in the latter.
[Spec, VP], the agent remains syntactically "active"."
d iscuss TT and
In
C h a p ters 2 and 3
C T , r e spectively, w i t h a f o c u s o n c h a r acterizing t h e
elements that are promoted to subject.
3.3 Morphology
Malagasy is predominantly a prefixing language. Suffixing is nevertheless
common, with a few infixes and circumfixes. Much of the morphological
s ystem is verb-based, as exemplified by the voice markers above. A c t i v e
verbs ar e
a l m ost al l d e r i ve d f r o m r o o t s , w h i c h a r e p r e d o m i n antly
nominal or adjectival. Th ere are some verbal roots, but these are almost
all unaccusative.
( 19)
a.
tonga
'arrive'
b.
avy
'come'
Other active verbs take one or more prefixes. The most common are anand i-. Bo t h d e r ive intransitive and transitive verbs, as seen in (20a,b).
When a single root has an intransitive and a transitive form, i- ma rks the
former and an- the latter (20c,d). Under an- p refixation, the nasal of the
prefix "fuses" with the initial consonant of the root, as in (20a,c); see Paul
(1996a) for details. In addition, the verb is marked with a temporal prefix:
m- (present), n- (past), h- (future)."
' Standard evidence against the obliqueness of genitive agents includes binding and control. Since
binding and control can be defined at D-structure (or argument structure; see e.g. Wechsler and Arka
(1998) on Balinese), this is not a knock-down argument against demotion. Note, however, that
the genitive agent occupies a fixed position in the clause, unlike obliques, which tend to display
less restricted distribution.
'" The past and future tense prefixes also occur in non-active voices, but the present tense m- is
not.
13
Introduction
( 20) a .
b.
m+an+hovitra = mangovitra
'(to) shiver'
m+i+kapoka = mikapoka
'(to) beat'
m+an+sasa = manasa
'(to) wash' (transitive)
m+i+sasa = misasa
'(to) wash' (intransitive)
Other active prefixes are listed below; both amp and if a re affixed ont o
active verbs, while aha attaches directly to a root .
A l l t a k e th e temporal
affixes.
( 21) a .
amp- (causative)
m+amp+an+sasa = mampanasa
aha-
'to make s.o. wash s.t.'
(abilitative/causative)
'to be able to wash'
m+aha+sasa = mahasasa
if- (reciprocal)
m+if+an+sasa = mifanasa
For
d i s cussion o f
the
'to wash each other'
cau s a tive, s e e R a n d r i amasimanana (1986);
Andriamierenana (1996); Travis (in press). Phillips (in press) provides a

thorough analysis of aha-.
The passive and the circumstantial voices will be the object of chapters
2 and 3, so I do not discuss them in detail here. A s m entioned above, the
agent in TT and CT clauses appears marked with genitive case. Genitive
case is widely present in Malagasy, marking passive agents, possessors as
well as the complements of some prepositions and adjectives; see Paul
(1996c) for a range of examples. G e nitive involves insertion of a
n a sal
consonant between the verb and the NP. M o r eover, the final -na, -ka, ortra of t h e v e rb is d r o pped. T h e se morpho-phonological processes are
i llustrated in (22); (22a) shows that /n/+ / r / = Endrj.' '

( 22) a .
sasa+ana+N+Rakoto = sasan-dRakoto
wash-Vna-gen-R
'washed by Rakoto'
" To be precise, the sequence ndr is a single prenasalized segment ['dr].
14
Chapter 1
toro+N+ny olona = toron'ny olona
shown-gen-det people
'shown by the people'
T he other cases, accusative and n o m inative, are no t m a r ke d o n m o s t
nouns. Thus only word order distinguishes (23a) from (23b).
( 23) a .
Nahita
ny l e hilahy ny yehiyayy.
pst.AT.see det man
de t w o m an
'The woman saw the man.'
Nahita
ny v e hivavy n y l ehilahy.
p st.AT.see det woman
det m a n
'The man saw the woman.'
Accusative case does mark p r oper n ames (24a), interrogative p r onouns

(24b) and demonstratives (24c).
( 24) a .
Nahita
an- d Rabe ny yehiyayy.
pst.AT.see acc-Rabe det woman

'The woman saw Rabe.'
Nahita
an' i z a
p st.AT.see acc'who
ny le h i lahy?
d e tm a n
'Who did the man see?'

N ahita
an' i t y v e h ivavy i t y
pst.AT.see acc'this woman
t his
ny lehilahy.
det man
'The man saw this woman.'
O therwise, th e
o n l y c o n s istent c ase m o r p h o logy s h o w s u p in the
pronominal system.
15
Introduction
nominative
accusative
genitive
1st singular
aho
ahy
-ko/-o
2nd singular
ianao
anao
-nao/-ao
-ny
3rd singular
1st plural incl
isika
antsika
-ntsika/-tsika
1st plural excl
izahay
anay
-nay/-ay
2nd plural
ianareo
anareo
-nareo/-areo
3rd plural"
izy (ireo)
azy (ireo)
-ny/izy ireo
T h e oretical assum t i o n s
T his thesis is w r i tten i n
t h e P r i n ciples and P arameters f r amewor k o f
Chomsky (1981). NP movement is motivated for Case assignment (under

government or spec-head agreement). Fo r th e most part, however, I do
not discuss the details of this movement.
I n s tead, I focus on the p h rase
structure of Malagasy.
This thesis also d r aw s
o n c e r t ai n a ssumptions abou t c o n nections
b etween lexical items and sy ntactic structure.
I n p a r t i cular, verbs a r e
a ssociated with lexical conceptual structure, which d etermines how
the
arguments are projected into the syntax.

5
Or an i z a t ion of the thesis
This chapter has provided an introduction to Malagasy syntax as well as

setting up the question of voice alternations. Let m e a d mi t at t hi s point
that this thesis has a clear empirical bias. Since my main goal is to arrive at
a proper characterization of voice in Malagasy, I focus almost exclusively
on Malagasy. N e v ertheless, I also draw on d ata from o t her l anguages,
both from
w i t hi n t h e s ame f amily ( e.g. Tagalog) and f r o m u n r e l ated
languages (e.g. Japanese). I s ho w t h a t m any o f t h e p h e nomena that I

d iscuss are not unique to M a l agasy. D u e t o
t h i s " n a r r ow " f o cu s o n a
s ingle language, I address a broad range of t o p ics.
A s a r e s ult, m a n y
" The third person pronouns are ambiguous between singular and plural. The addition of ireo
forces a plural reading.
16
Chapter 1
i ssues I will mention only in passing and leave for further research. T h e
overarching goal is to gain insight into the "genius" of Malagasy.
I n chapter 2, I address the nature of passive in Malagasy. I look at tw o
different passive affixes and argue that each affix p r o m otes arguments
from d i f ferent structural p o sitions.
M o r e o v er, I p r o v i d e e v i d ence in
favour of a position, [Spec, v2P], that hosts a special class of arguments,

"displaced themes" or "locata". In t his way, I use voice morphology as a
probe into the syntactic structure associated with three argument verbs.
The analysis of passive in chapter 2 leads directly into the investigation
of CT in chapter 3. W h ile passive promotes DP arguments to subject, CT
is less easily defined. I a r gue that CT is a kind of " elsewhere" voice that
promotes anything bu t
s t r ucturally Case-marked D Ps . T h e d i scussion
covers the nature of adjuncts and PPs in Malagasy, as these have been the
focus of previous analyses of CT. Taken together, the analyses in chapters
2 and 3 indicate that voice morphology can have distinct "functions", even
within a single language. For example, I show that one passive promotes
DP arguments from a particular position ([Spec, v2P]) and another passive
promotes DP arguments from a particular domain (the lower VP). CT, on
the other hand, promotes elements from a wide range of positions.
C hapters 2 an d
3 f o c u s o n t h e b a s e p o sition o f a r g u m ents ( a n d
a djuncts) as well as the phrase structure within the verbal projection.
In
chapter 4, I examine the positions of arguments and adjuncts when they

appear in the "left periphery" of the clause. Dr awing on data from topic
and focus constructions, I pr opose an articulated CP, along the l i nes of
Rizzi (1997). The core of chapter 4 is devoted to the cleft construction, due
to its relevance in both chapter 2 and 3 as a test for the argument-adjunct
d istinction.
A l t h o ug h chapter 4 d o e s no t a d d r ess precisely th e s a m e
questions as chapters 2 and 3, it i l l ustrates one way i n w h ich the A - b ar

system (extraction) interacts with the A system (voice) in Malagasy.
17
Chapter 2: Theme Topic

1 I n t roduction
This chapter focusses on the connections between argument structure and
phrase structure. In order to probe the syntactic structure of the clause, I
will look at the verbal morphology of v o ice alternations. In c h a pter 1, I
pointed out that Malagasy has a basic three-way voice distinction: AT, TT
and CT. I w ill now l ook m or e closely at TT, which in fact encompasses a
range of passive-like transformations. By analyzing tw o d i stinct passive
formations which promote different internal arguments, I will investigate
the internal structure of t h e v e rbal p r o jection. I n tu r n , t h e s y n tactic
structure provides an insight into argument structure.
In particular, I w i l l a r gu e fo r a n a d d i tional verbal p r o jection ( v 2P)
between the base positions of agent and theme. Certain DPs may be base
generated in [Spec, v2P]. Th e use of t his position is lexically restrained,
however, and only available with a special class of verbs and a restricted
class of DPs. I w ill pr ovide an informal characterization of the semantics
of the verbs and the DPs involved. U l t i m ately, a proper lexical semantic
s tudy will elucidate this issue. Fo r
t h e p u r p oses of t hi s chapter, I w i l l
concentrate on the syntactic structure. T h e d ata I w i l l
c onsider include
a lternations that have often been analyzed as i n v olving some f or m

s yntactic movement o r
of
" p r o m o t i on" , f o r e x a m ple d ative shift an d t h e
locative alternation. Du e to the restrictive nature of these alternations in

Malagasy, I will argue that there is no movement, per se. I n stead, verbs
that encode a change of location allow certain DPs (roughly, "displaced
themes" or "locata") to appear in [Spec, v2P]. Crucial evidence will come
from passive morphology.
v 2P] to subject.
O n e passive affix (a-) promotes DPs in [Spec,
A n o t her p assive affix (-Vna) pr omotes DPs from t h e
lower VP. The basic VP structure is given below.
Chapter 2
DP
vl'
~g
0
a- passive < l o catum>

0
-Vna passive
<th>
V'
DP/PP
<goal>
-Vna passive
A secondary goal will be to contrast passive with circumstantial topic in

chapter 3.
Section 2 p rovides an overview of t h e d a ta to be c onsidered in t h i s
chapter. I n
p a r ticular, I w ill i n vestigate verbs that show a n a l t ernation
between two di fferent passive affixes, a- and -Vna. In s ection 3, I discuss

a n analysis of t h e d a t a i n t e r m s o f l e x ical semantics. I s h o w s o m e
limitations of this approach and propose a different analysis in section 4.
Section 5 i s
d e v oted t o f u r t he r d i scussion of t h e d a ta . Du e t o the
importance of aspect in word order alternations, I consider the aspectual

properties of the Malagasy verbs in section 6. I h i g h light the importance
of the roots underlying these verbs in determining affectedness relations.
Section 7 concludes the chapter.
2 P assive in Mala a s
The introduction to voice in Malagasy in chapter 1 presented a simplified
version of the range of d ata.
E ach " v o ice" requires separate discussion
and explication. The differences between circumstantial and passive will

be crucial to my analysis, both in this chapter and the next.
Before beginning, how ever, i t i s
n ecessary to a n swer th e q u estion:
what is passive? This is, as mentioned in chapter 1, a thorny issue and the
subject of much linguistic research. For the purposes of this chapter, I will
c haracterize passive as a clause where an internal argument of a t w o
19
(or
Theme Topic
more) argument verb appears as the grammatical subject. T h e external
argument is either not present or surfaces in some non-subject, non-object
position.' As discussed in chapter 1, in Malagasy the external argument is
not realized as an oblique, unlike English. I n
o t her w o r ds, th e p assive
agent remains syntactically "active", although it is optional.
T h e l ack of
demotion to chomeur of p assive agents in A u stronesian has long been

recognized as one of the distinguishing properties of voice alternations in
these languages.
Note that my use of the term " p assive" is not wi t hout controversy in
the Austronesian literature. The lack of demotion m entioned above has
led some researchers to conclude that what I am calling "passive" is in fact
an ergative construction (e.g. Gerdts (1988) on I l o k ano; W e chsler and
Arka (1998) on Balinese). M oreover, some Austronesian languages have
two " p assive" constructions: one that is l ik e English and i n v o lves an
oblique agent and on e t hat i s
l i k e M a l agasy (e.g. I n d onesian (Chung
(1976b)) and Balinese (Wechsler and Arka (1998))). M y
v ery i n formal
definition of passive covers both types.

In fact, I do not consider Malagasy to be an ergative language (and I
am not aware of any ergative analyses of M alagasy). T h e subject of a
single argument verb and the subject of w hat appears to be a t r ansitive
verb are both marked with the same Case (nominative ). Note that in the
examples in (2), the verbal morphology is the same (man-, glossed as AT).
(2)
a.
Manasa
l a mba izy.
AT.wash cloth
3 ( n om)
'She is washing clothes.'

b.
M and i h y i z y .
AT.dance 3(nom)
'She is dancing.'
In fact, the genitive agent in M a lagasy (and other Austronesian languages) bears some
resemblance to (indefinite) objects: it surfaces right-adjacent to the verb and no elements can
intervene between the verb and the agent. H o wever, in M alagasy at least, genitive agents and
accusative objects have very different phonological properties. Fo r example, as mentioned in
chapter I, section 3.3, genitive involves pre-nasalization, not present in bare NP accusative.
Moreover, as also illustrated in the same section, a distinct set of p ronouns is available for
genitive and accusative.
20
Chapter 2
Mangovitra izy.
AT.tremble 3(nom)
'She is shivering.'
Under an ergative analysis, AT is a kind of antipassive. N o te, h o w ever,
that the logical object in (2b) is not marked as an oblique. H i storically, the
lack of demotion of the agent to oblique in what I am calling "passive" is
in part related to ergativity (see Chung (1978)). In other w o rds, I assume
there to be some influence on Malagasy from ergative languages in this
family. Au stronesian languages exhibit ergative characteristics to varying
d egrees, creating a
c o n t inuu m o f e r g a t i vi ty . M ac l a chlan ( 1996), f o r
e xample, a r gues t h a t T a g alog h a s b o t h no m i n a t ive-accusative a n d

ergative-absolutive properties and provides a "hybrid" analysis. Finally, I
believe that the ergativity issue is tangential to the data discussed in this
c hapter.
W h e t he r o n e a d o pt s a p a s sive o r a n e r g a t iv e a n alysis o f
M alagasy, there r emains th e q u estion o f h o w t o a c c o un t f o r w h i c h

arguments appear as the nominative/ absolutive DP. I h enceforth ignore
the ergativity debate and use the term "passive" as defined.
I begin with a brief description of passive formation in M a l agasy and
then examine two different affixes in detail. I w il l show that these affixes
are passive as defined above.
P r o b l ematic fo r t h i s claim ar e a p parent
adjuncts (e.g. instruments) that can passivize to subject position with the aprefix. If these examples truly involve adjuncts, then, by definition, the aprefix cannot be passive. I w ill p r ovide evidence, however, that in t hese
cases, the instrument has been base generated as an internal DP argument
of the verb. H e nce the a- passive is always from an a r g ument p osition.
Moreover, passive promotes DP and not PP arguments. I conclude with a
unified analysis of passive constructions that posits a special position for
certain base generated elements. T h e analysis of passive will pr ovide a
p oint o f
d e p a r ture f o r t h e d i s cussion o f c i r c umstantial t o pi c i n t h e
following chapter.
2.1 Passiveformation
There are essentially four t y pes of passive in M alagasy:
r o o ts, voa/tafa
passives, -Vna passives and a- passives. I provide here a brief overview of

all four, but for the remainder of this chapter, I focus on the last two.
M any b ar e
r o ot s c o r respond t o p a s sives: a n un d e r l y in g t h e m e
21
Theme Topic
appears in the subject position. Th e agent may appear as a genitive DP,
which forms a phonological unit with the verb, as in (3b,c).
a.
(3)
Hita ny sary.
seen det picture
'The picture is seen.'
Hitako
izy.
seen.lsg(gen) 3(nom)
'She is seen by me.'
Hitan-dRamatoa
aho.
seen.gen.Ramatoa
1s g (nom)
'I am seen by Ramatoa.'
R oot p assives are t h e m o s t c o m m o n p a s sive f o rm s i n t e x t c o u n t s
(Manorohanta (1998)).
The three other passives involve affixation.
T h e p r efixes voa and tafa
attach directly to a root to form a passive or inchoative verb.

a.
(4)
Voavory n y mpiasa.
voa.gather det worker
'The workers were gathered.'
Tafavory ny mpiasa.
tafa.gatherdet worker
'The workers gather.'

B oth affixes have additional aspectual import and I
r e fe r t h e r e ader t o
Travis (1996) for some discussion. Roughly, voa is available for transitives
and tafa is for intransitives. A c c ording to R ahajarizafy (1960), roots and

voa/tafa passives all express a resulting end state and thus form a n a tural
class (telic). Thus voa/tafa passives contrast with a-/ - V na passives, as the
latter are used to promote different arguments of the verb, as we will see
below.
Turning now to the passives that will be the focus of this chapter, most
22
Chapter 2
roots take the -ina or -ana suffix to f o rm t h e p a ssive.' S i n ce the choice
between -ina and -ana is lexicalized, I will refer to t he t wo a s -V n a."
In
general, the theme is promoted to subject.

laza+ ina = lazaina
'said'
sasa+ ana = sasana
'washed'
-Vna i s the most commonly o c curring passive affix ( as opposed to t h e

affixless roots mentioned above ) in texts (Manorohanta (1998)). F i nally,
there is the a- prefix, which also attaches directly to roots.
(6)
a.
a-tao
b.
a-didy
'done'
'cut'
In contrast to voa/tafa passives, neither -Vna nor a- passive are telic (see
section 6 for some discussion of telicity in Malagasy.) I n
will characterize the sy ntactic conditions u n der
w h a t f o l l ows, I
w h i c h t h e -V n a su f fix
alternates with the a- passive.

2.2 a- passive vs. -Vna passive: basic distribution
Verbs fall into two main classes with respect to passive affixes: those that
have only one of the passives and those that have both.
2.2.1
On e passive
In the first group we have the following verbs which take the a- prefix to
' A small number of verbs keep the active prefix in the passive. Examples are in (i).
(i)
root
active verb
meanin
halatra
man alatra
'to steal
'to re are'
voatra
mamboatra
This is also true for causatives in general.
an alarina
amboarina
CT
an alarana
amboarana
(ii)
verb
mam iasa
mankar
meanin
'to make work'
'to make sick'
CT
am iasana
ankarariana
am iasaina
ankararina
' Some roots may take both -ina and -ana, with a slight difference in meaning. These roots ate
very rare, however, and the distinction in meaning is disappearing. My consultants either reject
one form or consider the two to be synonymous.
In many cases, an "epenthetic" consonant is inserted between the root and the suffix. See Erwin
(1996) for arguments that this consonant is present in the underlying representation of the root and
not in the suffix.
23
Theme Topic
promote a theme to subject.'
root
active verb
meaning
subject of
a- passive
traka
mandraka
'raise'
theme
orina
manorina
'erect'
theme
idina
midina
'lower'
theme
verina
mamerina
'return'
theme
Some examples are illustrated in (8).'

a.
(8)
Aoriko
ny trano.
(a-orina-ko)
a.build. 1 sg(gen) det house

'I'm building a house.' (lit. 'The house is being built by me.')
Haveriny
ny boky.
(h-a-verina-ny)
fut.a.return.3(gen) det book

'She will return the book.' (lit. 'The book will be returned by her.')
These verbs have only one passive form and thus lack a -Vna variant.
Similarly, there are verbs that have only the -Vna passive. Some verbs
in this category are given in (9).
root
active verb
meaning
subject of
-Vna passive
haja
manaja
'respect'
theme
'need'
theme
lemy
mandemy
'weaken'
theme
vaky
mamaky
'read; break'
theme
In what follows, I provide illustrative examples in the tables. See the appendix for more
extensive tables.
For the remainder of this chapter, I wil l tr y t o p rovide glosses that are the most natural
translations of the Malagasy. Where necessary, I will also give literal translations. Only in this
subsection do I provide a morphological decomposition of the verbs.
24
Chapter 2
Typical clauses with -Vna passives are in (10).

Vakin-dRasoa
ilay boky.
(vaky-ina-Rasoa)
read.Vna.gen.Rasoa def book

'Rasoa is reading that book.'
H ajain-dRasoa
i Bako l y .
(haja-ina-Rasoa)
respect.Vna.gen.Rasoa Bakoly
'Rasoa respects Bakoly.'
I now turn to three instances where there is an alternation between the
two passive forms. T h ese alternations will be the focus of the remainder
of this chapter.
2.2.2 Tw o passives
The verbs that allow both passives can be subdivided into three groups.'
In the first group, the verb takes the a-passive to promote an instrument
to subject. The instrumental use of the a- passive is quite wide-spread.
(11)
root
Ca se I
active verb
meaning
subject of
subject of- subject of
a- passive
Vn,a
CT
passive
dldy
mandidy
'cut'
instr
theme
instr
fehy
mamehy
'tie'
instr
theme
instr
fefy
mamefy
'fence in'
instr
theme
instr
rakotra
mandrakotra
'cover'
instr
theme
instr
The examples in
( 12) i l l u strate that f o r C a s e I v e r bs, t h e a - p a ssive
promotes an instrument, while the -V na p assive is used to p r o m o te t h e

theme. (12a) illustrates the basic active clause and (12b,c) are the passive
c ounterparts. T h e C T clause in (12d) is p r o v ided fo r
c o m p arison." I n
Pearson (1998a) proposes four classes of verbs that take the n- passive, three of which align
with my three groups. The fourth allows only the n- passive. I discuss this class of verbs in 4.3.
As noted in chapter 1, CT morphology is in one sense a combination of A T a nd -V na.
Moreover, recall that in a CT clause, a wide range of elements may be promoted to subject. In the
25
Theme Topic
(12b-d), I have omitted the agent for clarity; a genitive agent is alw ays
possible, however.
(12)
a.
Nandi d y
ny hen a
tam i n 'n y
ants yRasoa.
pst.AT.cut
de tmeat p s t .P.gen.det knife Rasoa
'Rasoa cut the meat with the knife.'
Adidy ny hena
a .cut
n y a ntsy.
instrument
d e t meat d e tk n i f e
'The knife is used to cut the meat.'
Didiana
a m i n 'n y a n t sy ny hena.
cut.Vna
P . g en.det knife det meat
theme
'The meat is cut with the knife.'
Andidiana ny hena
n y antsy.
CT.cut
det k n i f e
d et m e at
instrument
Hence the verbs in (11) take both t y pes of affix, each used for d i f ferent
arguments: instruments (a-) or themes (-Vna). As illustrated in (12d), CT
is also possible for instruments. The alternation between CT and passive
will be discussed below.
Parallel to the examples above, there are verbs that take a (material)
theme" and a goal (or location) that allow for the a-/-Vna alternation.
tables below, I therefore only indicate which of the relevant arguments of the verb are promoted to
subject with CT.
R appaport and Levin (1988) refer to this argument as "locatum". I n w h at follows, I w i l l
continue to use "material theme" in connection with Case II verbs, reserving "locatum" for a
broader class of arguments. See section 2.3.1.
26
Chapter 2
(13)
Ca se II
root
active verb meaning
subject of subject of-
subject of
a- passive
CT
Vn,a
passive
fatratra
mamatratra
'stuff'
mat theme
goal
mat theme
lafika
mandafika
pad
mat theme
goal
mat theme
raraka
mandraraka
'scatter'
mat theme
goal
mat theme
tototra
manototra
'fill'
mat theme
goal
mat theme
( 14a) provides the st andard a ctive clause for C ase I I v e r b s w i t h t h e

p assives given i n
( 1 4b,c) and c i r cumstantial i n ( 1 4d) . O n c e a g ain, a
genitive agent is possible in (14b-d), but omitted to simplify the examples.
( 14) a .
Namatratra ny haronatamin'ny
va r y i Bakoly.
p st.AT.stuff d e t basket pst.P.gen.det r i c e

'Bakoly stuffed the basket with rice.'
Nafatratra ny harona ny yary.
Bak o l y
material theme
pst.a.stuff det basket det rice

'The rice was stuffed into the basket.'
Nofatrarana
va r y
pst.stuff.Vna rice
ny ha r o n a.
goal
det b asket
'The basket was stuffed with rice.'
Namatrarana ny harona ny yary.

pst.CT.stuff
material theme
d e tb asket det rice
'The rice was stuffed into the basket.'
For the verbs in (13), the a- passive promotes the material theme, while
the -Vna p assive promotes the goal. T h e m a t e rial t h e me ( but n o t t h e
goal) may also appear in the subject position of a CT clause, as in (14d). In
passing, note t ha t
C ase I I v e r b s a r e s i m i lar t o t h e E n g l ish " l o cative
alternation" class. In distinction to English, however, the goal or location

is always realized as a DP, never a PP. I return to this below.
A s a first attempt to simplify th e d ata, I w i l l c o n flate Case I an d I I
27
Theme Topic
verbs. With this goal in mind, I refer to instruments and material themes
as "locata". These elements share the following properties, which will be
discussed in detail below: they may appear either adjacent to the verb or
in a PP; they are promoted to subject with either the a- passive or CT; they
a re (usually) op tional i n
a n a c t iv e cl ause." I am not cl a i m in g t h a t
instruments and material themes are identical in all respects. For example,
there are interpretational differences. As shown in (15), material themes
are compatible with an instrument, but two in strumental phrases are not
permitted in a single clause.
Nameno n y
s i nibe tamin'ny
pst.AT.fill det pitcher pst.P.gen.det
rano tamin'ny tavoahangy

water pst.P.gen.det bottle
i Soa.
Soa
'Soa filled the pitcher with water with the bottle.'
Nandidy a ntsy ny hena tamin'ny
laf in y m a r anitra i Soa.
pst.AT.cut knife det meat pst.P.gen.det s i d e

sh a r p
'Soa cut the meat with the knife with its sharp edge.'
Soa
T his is true i n
E n g lish as w e ll, a s seen b y t h e c o n t rast b etween t h e
translations of
(15a,b) (similar facts are discussed in Lakoff ( 1968)).
H owever, d e spite t h ese s emantic d i f f erences, th e v o i c e s y s te m o f

M alagasy treats material themes and instruments on a p ar .
I ta k e t h e
voice data as initial evidence that when realized as DPs, instruments and
material themes appear in the same structural position. I w i l l address the
parallels between instruments and material themes further in section 2.3.
At this point, the reader may w o n der w h y I
a m c a lling the a- prefix
"passive". The preceding data suggest that the a- prefix is mainly used to
promote underlying PPs (material themes and i n struments) to su bject.
Furthermore, these PPs appear m or e
a d j unct-like than a r g ument-like.
T hus the a- prefix appears to resemble CT rather than TT .
I n o r der to
motivate following the traditional analysis of a- as a passive prefix, I w i ll

show that material themes and instruments can be realized as arguments.
I first, however, turn to the third class of alternating verbs.
'" In English, both take the preposition with.
28
Chapter 2
There is another large class of verbs that allows an alternation between
-Vna and a-. In t hese cases, the -Vna indicates a goal, while a- marks the
theme. Clearly, themes are arguments and not adjuncts, hence this class
of verbs provides some initial motivation for t r eating the a- prefix as a
passive morpheme.
(16)
Cas e III
root
active verb meaning
subject of subject of-
subject of
a- passive
CT
Vn,a
passive
roso
mandroso
'serve'
theme
goal
goal
tolotra
manolotra
'offer'
theme
goal
goal
seho
manaseho
'show'
theme
goal
goal
toro
manoro
'point out'
theme
goal
goal
(17) illustrates the voice alternations for a typical dative verb.
( 17) a .
M anolotra sary
an a o aho .
AT.offer p i cture 2g(acc) 1 sg(nom)

'I offer you a picture.'
b.
Tolo r a na
sary
i a n ao.
offer.Vna
pict u re 2sg(nom)
goal
'You are offered a picture.'

ny sary.
A tolotra
an a o
a.offer
2sg ( acc) d e t picture
theme
'The picture is offered to you.'

Anolorana sary
i a n ao.
goal
CT.offer picture 2 s g.nom

Comparing (17b) and (17d), we see that i n
alternates with CT.
29
t h i s case, the -V na pa ssive
Theme Topic
2.2.3 W h ich passive?
At this point, a confusing p attern emerges.
passives from roots.
Both -Vn a a n d a - f o r m
S om e r o ots allow b oth a f fixes. I n C ase I, -Vna is
used for themes. In Case III, a- is used for th emes.
B ot h a f fixes can be
used for different and distinct thematic relations, however, such as goals
(-
Vna) ( Case III) and instruments or m a terial themes (a-) (Case I and I I ) .
The table below summarizes the distribution of voice morphology across
theta-roles.
(18)
theme
instrument/
goal/locative
material theme
-Vn,a
aCT
Clearly, it is not p ossible to map d i rectly from
voice forms.''
t h e ta-roles to p a rticular
T h i s may be due in part to the difficulty of independently
determining the theta-role labels involved. It is also odd that instruments,

normally treated as adjuncts, can passivize. Finally, CT, typically related
to adjuncts, overlaps with p assive, which deserves some explanation."
Based on the above facts, Rajaona (1972) concludes that there are tw o aprefixes: a1 for objects and a2 for instruments.
M o r e r ecently, how ever,
P earson (1998a) has proposed an analysis of a- p assive that d r aw s o n

parallels in event structure between the various cases we have seen.
will discuss Pearson's analysis in more detail in section 3 below.
Similarly, I w il l a r gue t hat i n
(I
s p it e o f t h e v a r i ation i n t h e d i f f erent
passives, there is one unified passive phenomenon.
I n a l l i n stances the
targeted DPs are internal arguments of the verbs in question.
T h e basic
clause structure is given in (19).
This lack of one-to-one mapping between theta-roles and voice morphology will be crucial to
the discussion of voice and CT in chapter 3.
" To account for the overlap between passive and CT, Rabenilaina (1991) distinguishes between
voice (the verbal morphology) and diathesis (the role of the element appearing in subject position).
For example, (17d) is a goal diathesis with relative voice.
30
Chapter 2
vlP
(19)
"A
vl'
v2P
DP
Z
v2'
V
DP
V
V'
DP/PP
<goal>
I will show that instruments, material themes and themes of dative verbs
can be generated in [Spec, v2P]. The a- passive targets elements in [Spec,
v2P] and the -Vna p assive promotes elements from the lower VP. I t u r n
to evidence in favour of this analysis below.
A s i m ilar s t r u cture h a s b e e n p r o p o sed b y M ar a n t z ( 1 9 93) f o r
a pplicatives in general (including English dative shift)."
he argues for a
I n o t h e r w o r d s,
p o sition c-commanding th e t h eme w h ere i n struments
(and other things such as benefactives) can be base generated. I
b e l i eve
that the assumptions that underlie my analysis and his are very di fferent,
h owever.
I n s e c tion 5.3, I w i l l p r e sent a r g uments i n f a v ou r o f b a s e
g eneration over syntactic movement.
I m p o r t antly, in M a l agasy only a
subset of instruments can be generated in [Spec, v2P]. With applicatives,

however, " advancement" to o bject of i n struments appears to b e m u c h
f reer. I t ake this to indicate that applicatives involve movement and n o t
base generation. I n o t her w o r ds, I adopt Marantz's analysis to a certain
degree, but would not necessarily use it for his data.
2.3 Locative alternation and instrumental advancement
As mentioned above, material themes and instruments pattern together.
In this section, I w il l e x plore the syntactic behavior o f
t h ese elements.
Baker (1992) (fn. 2) suggests that instruments may be structurally subordinate to agents but
superordinate to themes.
31
Theme Topic
Crucially, I will show that both can be arguments of the verbs in question.
This may appear to be an u n justified claim fo r
standardly t r eated a s
i n s t ruments, which ar e
a d j u ncts. I w ill s h o w , h o w e v er, t h a t c e r t ain
i nstruments may b e b ase generated in a n a r g u m ent p o sition. "
Only
those instruments that appear overtly in th e i n ternal argument position,

right-adjacent to the verb, can passivize to subject. I conclude that the apassive targets this argument position rather than the adjunct instrument
p osition. T hus although DPs with d i f ferent thematic roles appear in t h e
subject position of these different passive clauses, the DPs are arguments,
not adjuncts. M o reover, the elements targeted by both t y pes of p assive
are realized as DPs, not PPs. Therefore, these verbs are all " p assive" as
defined earlier and contrast with circumstantial topic, which is limited to
objects of prepositions."
O n t h e o t her h and, instruments and m a terial
themes can alternatively be realized as PP adjuncts, in which case they are

p romoted to subject with CT .
I n o w t u r n t o e v i d ence in favour o f t h i s
analysis. In other w o rds, I will provide data in favour of th e dual status

(adjunct-argument) of instruments and material themes.
2.3.1
Ba sic distribution
Case I and Case II verbs exhibit an alternation between the two passives
(a- and -Vna) and also allow alternations in word order. I w i l l f i r st discuss
Case II constructions, as they resemble the locative alternation verbs in
English (sometimes referred to as the spray/load class)." L i k e E n g l ish,
there is no m o r phological marking o n
possible orderings o f
t h e v er b t o s i g nal th e d i f f erent
t h e " a r g u m ents". "
For Ca s e I I v e r b s , e i t h er
element can appear in the canonical direct object position, adjacent to the
verb, as shown in (20). I n
( 2 0a) the material theme is wi thin a PP that
follows the goal, while in (20b), it is a DP that precedes the goal.
" See Baker (1988) for discussion of the argument-like properties of instruments. He points out
that noun incorporation of instruments is fairly common and claims that the canonical structural
realization of an instrument is NP, rather than PP. A s w il l be discussed below, instruments ate
among the class of a djuncts that pattern i n s om e r espects with a rguments in M a l agasy.
Importantly, however, only a subset of instruments behave like arguments.
" This informal characterization of CT will be refined in the next chapter.
A s pointed out earlier, the locative alternation in Malagasy is not identical to English. F o r
example, the locative/goal is not realized as a PP.
" Since some alternations involve instruments, which are sometimes adjuncts, "argument" is not
precisely the correct term.
32
Chapter 2
(20)
Cas e II: locative alternation

a.
goal D p > material themepp
N amafy
ny tany tamin'ny v o a
Rasoa.
pst.AT.sow
d e t land pst.P.gen.det seed Rasoa
'Rasoa sowed the land with seeds.'
m aterial themeDp >
goa l D p
N amafy
ny ta n y
voa
Ras o a .
p st.AT.sow
s e e d d e t land Ra s o a
'Rasoa sowed seeds in the land.'
I n what f o ll ows, I w i l l a r gu e t hat a s a D P , t h e m a t erial t h em e i s a n
argument of the verb.
Turning now to Case I verbs, these are unusual from th e p erspective
of English, but also allow d i f ferent w or d
o r d e rs.'" P a r allel t o m a t erial
themes, instruments can appear right-adjacent to the verb i n
t h e d i r ect
object position, as shown in (21b). Following Seiter (1979), I will call this
process "instrumental a d v ancement" (although I w i l l a r g u e a g a i nst
syntactic movement ) . A s
i l l u strated below, Case I v e rb s p a ttern w i t h
Case II in allowing alternative word orders between a theme and another

element.'"
T h e i n strumental is realized as a PP in (21a) and as a DP in
(21b) with a subsequent change in word order."

(21)
Ca s e I: instrumental advancement
a.
theme D p
>
i nst r u m entpp
Nandidy
ny he n a tam i n 'n y
ants yRasoa.
pst.AT.cut
de tmeat p s t .P.gen.det knife Rasoa
'" Case I does bear some similarity to the following alternation in English.
(i)
Pat b e a t the table with the stick.
(ii)
Pat b e at the stick against the table.
I will return to this similarity in section 4.4.
Baker (1996b) revises earlier claims that applicatives are only possible with transitive verbs
( Baker (1988)). I n t erestingly, Malagasy instrumental advancement appears to be l i mited t o
transitives. I will discuss this issue further in section 5.1.
'" Instruments are obligatorily indefinite in this "advanced" position: they may not appear with a
determiner or a demonstrative (although adjectival modification is permitted). For Case II and III
verbs, however, the appearance of the determiner is less restricted . In fact, the overall distribution
of the determiner for objects in Malagasy is complex and not fully understood.
33
Theme Topic
instrumentDp >
the m eDp
Nandidy a n tsy ny hena Rasoa.

pst.AT.cut knife det meat Rasoa
'Rasoa cut the meat with a knife.'
I t is this alternate realization of i n struments as DPs that is crucial to th e

application of passive. Recall that instruments would appear to resist a
standard analysis of passive; they tend to pattern with adjuncts, as I will
show in the following sections. The data in (21), however, indicate that in
Malagasy, instruments can be generated in an argument position (I argue
against a movement analysis in 5.3 ). I
c l ai m t ha t i t i s p r e cisely those
i nstruments which can appear as arguments which can be pr om oted t o

subject with passive.
In sections 2.3.2-2.3.7, I will show that material themes and instruments
can be either adjunct PPs or argument DPs. For ease of reference, in what
f ollows, I
w i l l r e fe r t o i n s t r u m ents an d m a t erial t h emes i n t h ei r D P
position as "locata", since they undergo a change of location. I
p r e sent
three arguments in favour o f a d junct status of instrument and m aterial

theme PPs: they are optional, they may be clefted, they can be promoted
with CT." A s DPs, on the other hand, locata are arguments: they surface
i mmediately adjacent to the verb, they cannot b e c lefted and t hey a r e
promoted t o
s u b ject w i t h p a ssive. H e n c e i n s t ruments an d m a t e rial
themes have two d i stinct syntactic realizations: a d j u nct and a r g u m ent.

(22) illustrates these two possibilities.
( 22) a .
b.
verb
verb
DP l
D P2
[PP(adjunct) P DP2 ]
DPI
Since locata are arguments and since they can be promoted to subject with
the a- prefix, I believe that th e a- p r efix i s correctly characterized as a
passive affix. Passive always targets internal DP arguments of the v e rb,
never adjuncts.
In fact, instruments and material themes as PPs do not enjoy precisely the same distribution
within a clause. For example, as in English, there is a strict ordering between the two: m aterial
themes always precede instruments.
34
Chapter 2
2.3.2
Op t ionality
What are the tests to distinguish arguments from adjuncts? In an intuitive

s ense, arguments are obligatory an d a d juncts are optional.
k nown, h o w ever, t ha t
I t i s w ell-
c e rtain v e rb s h av e " o p t i onal" a r g u m ents a n d
h ence the "optionality" test is not clear-cut. N e v ertheless, I refer to t h i s

test (with caution) in w hat f o ll ows.
A s s h ow n i n ( 23), instruments and
material themes are optional and therefore pattern with adjuncts.
( 23) a .
Namafy
tany (ta m i n'n y
voa)
Rasoa.
pst.AT.sow
la n d ps t . P.gen.det seed Rasoa
'Rasoa sowed the land (with seeds).'
b.
Nand i d y
ny hen a
(tamin'ny
ants y ) R asoa.
pst.AT.cut
de tmeat ps t . P.gen.det knife R a soa
'Rasoa cut the meat (with the knife).'
I no w t u r n t o a M al a g asy-particular c onstruction t h a t s u p p o rts t h e
classification of locata as adjuncts.
2.3.3 Cl e
f ts
C lefts can be used as a test to di stinguish adjuncts from ar guments.
In
M alagasy, clefts are formed by f r onting an XP, which is followed by t h e

focus particle no. (The syntax of the cleft construction is discussed in detail
i n chapter 4.) I n
g e neral, only subjects and adjuncts can cleft when t h e
verb is active (an AT cleft). To cleft an object, it must first be promoted to

subject via passive, as shown in (24a,b). (24c) shows that an adjunct can
AT cleft freely.
( 24)
a. *
[ Ny lamba ] n o
d et cloth
f oc
mana s a
Rak o t o.
A T .w a s h Ra k o t o
'It is the clothes that Rakoto washes.'
tavoahangy i Soa
N ame n o
n y sin i b e t ami n ' n y
r ano t a m i n'ny
Soa
pst.AT.fill
d e t pitcher p s t .P.gen.det water pst.P.gen.det bottle
'Soa filled the pitcher with water with the bottle.'
b. * Nameno
ny s i n i b e tam i n ' n y
tavo a hangy t a min'n y r an o i Soa.
pst.AT.fill
d e t pitcher p s t .P.gen.det bottle
pst.P . g en.det water Soa
* 'Soa filled the pitcher with the bottle with water.'
" In fact, as will be seen in chapter 3, not all adjuncts can AT cleft in this manner. However, if a
non-subject can cleft, it is an adjunct.
(i)
a.
35
Theme Topic
[ Ny lamba ] n o
det cloth
foc
sasa n -dRakoto.
wash . V na.gen.Rakoto
'It is the clothes that are washed by Rakoto.'
[ Amin'ny Talata ]
no
P.gen.det Tuesday f o c
mana s a
lam b a R akoto.
AT .w a s h c l ot h R a k oto
'It is on Tuesday that Rakoto washes clothes.'
N ote that it is not simply the case that any PP may AT cleft freely. T r u e
PP arguments, such as the goal of dative verbs, cannot AT cleft."
( 25) * [ Hoan'ny zaza ]
f or.gen.det child
no
nandr o s o
vary
Rakoto.
foc
pst. A T .serve r i c e
Rakoto
'It's to the child that Rakoto served rice.'
AT clefting therefore distinguishes between arguments and adjuncts.

As (PP) adjuncts, instruments and material themes may AT cleft.
(26) a.
Tam in'ny
r a n o n o nameno ny tavoahangy i Sahondra.
pst.P.gen.det water
f o c pst.AT.fill det bottle
Sahondra
'It's with water that Sahondra filled the bottle.'
b.
T ami n ' n y
ants y n o n a ndidy h e n a i B akoly.
pst.P.gen'det knife
f o c pst.AT.cut meat B a k oly
'It's with a knife that Bakoly cut meat.'
Like the optionality test in 2.3.2, AT clefting points to the adjunct status of
these PPs. Similarly, the contrast between (25) and (26) is further evidence
t hat the AT clefting of m aterial themes and instruments occurs from an
adjunct position and not from an argument PP position.
" L ocative PP arguments can cleft, however.

( i)
Teo
amboni n ' ny latabatra no nametraka b oky i K o t o .
pst.there on.gen.det t a bl e
f oc p s t .AT.put b o o k K oto
'It was on the table that Koto put the book.'
Moreover, locative arguments are generally optional (there is some variation in judgement both
within and across speakers). Hence, they may best be classified as adjuncts.
36
Chapter 2
2.3.4
Pr e posi tions
In the preceding sections, I have argued that i n struments and m a terial

themes can be realized as adjuncts. I will now p r esent further evidence in
favour of the PP status of these adjuncts. As has been exemplified in the
above data, material themes and instruments commonly su rface as the
object of a preposition, amin'. Consider now CT clauses. Instruments and
material themes can be promoted t o
s u bject w it h CT , w h i ch has been
t raditionally linked t o a d j uncts and PPs. D e s criptively, w he n a P P i s

promoted to subject with CT, the P is dropped because there are no PP
s ubjects in Malagasy (see (27a)). On the other hand, if the PP is in a CT
cleft, the P can surface (although it is optional).
( 27) a .
Andidiana ny hena
n y antsy.
CT.cut
det k n i f e
d et m e a t
[ (Amin')ny
a n tsy ] no andidiana ny hena.
(P.gen.)det
k n i f e f o c CT.cut
det meat
'It is the knife that is used to cut the meat.'
I will discuss the interaction between circumstantial topic and th e cleft

construction in more detail in chapter 3. C r u cially, the data in (27b) show
t hat when a n i n strument i s p r o m oted t o s u bject w it h C T , i t i s b e i n g
promoted from th e PP position.
T h e same is true for m a terial themes.
This will contrast with passive, to be discussed immediately below.

Summing up, w e
h av e seen that m a terial themes and i n s t ruments
p attern with adjuncts when they are realized as PPs. I w i l l no w t u r n t o

e vidence that when DPs, they are arguments. Evidence comes from w o r d
o rder, clefting an d t h e l a c k o f p r e p o sitions i n c l e fts . Due to th e i r
argument status, they will be promoted w it h p assive. T hu s instruments
and material themes have a dual status, realized either as arguments or
adjuncts.
2.3.5
Ar g u ments: Word order
I n examples (20)-(21) at the beginning o f
s ection 2.3.1, I s h o we d t h a t
instruments and material themes can appear as DPs, right-adjacent to the

v erb.
I n t h i s p o sition, they b ehave l ik e i n ternal arguments A s w it h
37
Theme Topic
"normal" direct objects (i.e. non-alternating), an adverb cannot intervene
between the verb and an indefinite DP locatum."
(28) illustrates adverb
ordering for a verb and object.

(28)
a.
M ana s a lam b a
AT.wash
tsara Rakoto.
cl o thes g o o d R akoto
'Rakoto washes clothes well.'

b. *
Manasa
ts a ra la m ba Rakoto
AT.wash good c l o t hes Rakoto

(29) shows t hat t h i s s ame o r d ering r e striction ap plies t o " a d v anced"
instruments.
( 29) a .
Mandidy antsy tsara
ny h e na Rasoa.
AT.cut knife good d
et mea t Rasoa
'Rasoa cuts the meat well with a knife.'

b. *
Mandidy
ts araantsy n y
AT.cut good knife d
h e na Rasoa.
et m eat Rasoa
The above data indicate that when locata appear as DPs next to the verb
t hey pattern w it h d i r ect objects w it h r e spect to a d v erb p l a cement. I
conclude that as DPs, locata are in an argument position (not adjunct).
D ue to t h e s t r ict a djacency, it m a y a p p ear a s i f t h e l o c atum h a s
"incorporated" into the verb or that the examples are a kind of verb-noun
compound. I w ill show, however, that the instrumental advancement and
locative alternation illustrated above are neither noun i n corporation no r
compounding.
I n n o n -active voices, for example, a genitive agent can
surface between the verb and the locatum, as shown i n (30a). It is not
possible for the locatum to separate the verb and the genitive agent, as in
(3Ob).
" Recall that locata are usually indefinite.
38
Chapter 2
Didian'i Bozy
( 30) a .
ant s y ny hena.
cut.Vna.gen.Bozy knifedet meat

'The meat is cut by Bozy with a knife.'
b.
Didiana antsin'i Bozy
ny h e n a.
cut.Vnaknife.gen.Bozy det meat

Furthermore, note that it is possible to coordinate the material theme and
the goal to the exclusion of the verb.
( 31) a .
Nandrakotra b odofotsy ny fandrinany sy l a mba ny latabatra

pst.AT.cover blanket
de tbed.gen.3 and cloth d et table
i Sahondra.
Sahondra
'Sahondra covered her bed with a blanket and the table with a cloth.'
b.
Nani n d rona a n t sy ny trondro s y l efona ny omby i Sahondra.

pst.AT.pierce knife det fish
and s p ear det ox
Saho n d ra
'Sahondra pierced the fish with a knife and the ox with a spear.'
Consider (31a).
fotsy ' b l anket' c o u ld p o t e ntially h a v e

A l t h ough bodo
incorporated i nt o
t h e v e r b na n d rakotra' cover', t h er e i s
no h ost f or
incorporation of lamba'cloth'. If incorporation (in some sense) occurs with

t hese verbs, it must therefore be o p t i onal (and somehow
(30b)) or obtains at LF.
r u l e d ou t i n
T h e d ata in (30)-(31) indicate that instrumental
advancement and the locative alternation are neither noun incorporation

nor compounding.
As further evidence against compounding, note t ha t
t h e a d v anced
instrument is referential: it introduces a DP into the discourse which may

be referred to by a pronoun, as in (32).
(32)
T s y nandidy a n t sy ny hena Rabe satria t s ymaranitra i lay izy.

neg pst.AT.cut knife det meat Rabe because neg sharp
def 3 ( nom)
'Rabe didn't cut the meat with the knife because it wasn't sharp.'
Since words tend to be r eferentially opaque (Postal (1969; DiSciullo and
39
Theme Topic
Williams (1987)), (32) suggests that the verb and the advanced instrument
do not form a lexical unit."
Summing up, the data in this section indicate that as DPs, locata have
the same distribution as other direct objects. I t ake this as initial evidence
i n favour of analyzing locata as arguments of the verb .
I now turn to
further evidence in favour of this analysis.
2.3.6 Cle
fts again
Recall that
o n l y s u b jects an d a d j u ncts can A T c l e f t , n e ve r i n t e r n al
arguments.
M o r e o ver, i n section 2.3.3, we saw t h a t a s P Ps, m aterial
themes and instruments can AT cleft, which we took as evidence in favour

of adjunct status. As DPs, on the other hand, neither the material theme
nor the instrument may b e clefted in a n
a ctive clause, as illustrated in
(33)."
( 33)
a. * [
( N y) rano ] n o nameno
(det) water
n y t avoahangy i S ahondra.
f o c pst.AT.fill det bottle
Sahondra
'It's water that Sahondra filled into the bottle.'
b. *
[ ( N y) antsy ] no nandidy h e n a i B akoly.

(det) knife
f o c pst.AT.cut meat B a k oly
'It's a knife that Bakoly cut meat with.'
The ungrammaticality of (33) indicates that in their DP positions, both the

material theme and the instrument are arguments of the verb and hence
c annot AT cleft. T hus locata not only show th e same basic word o r d e r
distribution as a r g uments, they
a r e a l s o s u bject t o s i m i lar e x t raction
constraints.
2.3.7
Pr e posi tions agai n
In section 2.3.4, I claimed that as PPs, instruments and material themes are
promoted to subject with CT. Some evidence for the PP status came from
the CT cleft construction, where the preposition may be overt (see (27b)).
" The opacity of words has been challenged in the literature, however (Sproat (1988)). Thus on
its own, (32) presents weak evidence at best against a lexical analysis.
" Clefts of bare NPs are in principle possible. Hence (i) is grammatical.
(i)
L amb a n o s asan-dRakoto.
cloth
foc T T .wash.gen.Rakoto
'It's clothes that are washed by Rakoto.'
Thus it is not the presence or absence of a determiner in (33) that determines grammaticality.
40
Chapter 2
If locata are promoted to subject with passive from the DP p osition, not
from the PP position, we expect prepositions to be unavailable in a cleft
construction in a passive clause. In fact, the preposition is prohibited from
surfacing in a cleft of an instrument with the a- passive. This is shown in
(34b).
( 34) a .
[ Ny antsy ]
n o adidy n y hena.
det knife
f oc a . c u t de t m eat
b. * [
A m i n 'ny antsy ] no adidy ny hena.

P.gen.det knife
The ungrammaticality o f
f o c a .cut det meat
( 3 4b ) s u ggests that t h e a - p a s sive i n v o lv es
promotion from an internal DP argument position, not from the object of

the preposition.
T he fact that material themes and i n struments can be p r o m oted t o
subject via either passive or CT is indicative of their dual status. They can
be arguments, in which case there is no preposition and passive applies.
T hey can equally b e
a d j u ncts, b e s e lected b y a pr e p o sition an d b e
promoted to subject with CT.

2.3.8
A pr e diction
The analysis of instrumental and l ocative advancement feeding passive

makes a clear prediction: i f
a n e l ement cannot appear in an a r g u m ent
position, the a- passive should be impossible. Let us consider instruments.

Not all verbs allow an instrument to appear in the direct object position.
C rucially, in these cases, there is no a- passive form o f t h e v e rb . ( 3 5 a )
shows that like mandidy 'cut', mihinana 'eat' may take a DP theme and a PP
instrument. U n l ike with mandidy ' to cut', however, the instrument ma y
not appear right-adjacent to the verb, as shown by the ungrammaticality
of (35b). Moreover, mihinana has no a- passive, as in (35c).
( 35) a .
Nihinana h en a
ta m i n'n y
ant sy R a soa.
pst.AT.eat meat p s t .P.gen.det knife Rasoa

'Rasoa ate meat with the knife.'
41
Theme Topic
b.
Nihinana antsy ny hena Rasoa.

pst.AT.eat knife det meat Rasoa
Nahinan-dRasoa ny hena ny antsy.

a.eat.gen.Rasoa det meat det knife
The proposed dependency between advancement and passive remains to

be verified against a
l a rg e n u m be r o f v e r bs, bu t i n i t ial i n v estigation
suggests that the former feeds the latter."

Unfortunately, this correlation is not perfect. I t seems that instrument
a dvancement i s
understood.
s u b j ect t o le x i cal r e s t rictions t h a t r e m a i n p o o r l y
F o r e x a m ple, the o nl y i n s t rument t hat can appear as an
argument of mandidy 'cut' is antsy 'knife' (it is also the only instrument that
can be used with the a- passive of mandidy). Mo r e over, some verbs allow
instrument advancement but have lost the a- passive form. T h i s in f a ct
may be a tendency across the language to reduce the number of passive
forms per verb.
S e v eral speakers, for example, reject the a- passive of
mandidy'to cut' in favour of CT. This pattern was noted by Keenan (1976)
and appears to be still valid. I cl aim, nevertheless, that if a verb has an ap assive, it m ust allow i n strument a d v ancement. I n o t h e r w o r d s , t h e
availability of the a- passive for i n s truments is l in ked to t h e a r g u m e nt
status of the instrument.
Summing up, w e
h av e seen that m a terial themes and i n s t ruments
share certain properties. As DPs, they appear in argument positions and

cannot AT cleft .
A s P P s , t he y a r e a d j uncts and h e nce can A T c l e ft.
Furthermore, t h ese tw o
d i f f e rent r e alizations ( DP-PP) correlate w i t h
different voice morphology :

P Ps.
p a s sive p r o m otes DPs and C T p r o m o t es
I w i l l c o m p ar e t h ese p r o p erties w i t h t h o s e o f da t i v e v e r b s
immediately below.
2.4 Dative verbs
We now turn t o th e t h ird class of verbs, datives. I n
m a n y w a ys, these
resemble their English counterparts, the most w e l l -known b e ing ' g i v e'.
(manome 'give' does not allow the a- j-Vna alternation, however, and w i l l
" Some examples aregiven inthe appendix, under Table 3.
42
Chapter 2
therefore not be discussed.") L i k e Case I and II v e rbs, Case III (dative)
verbs have tw o p assive forms .
T h e r e ar e some d i f f erences, however,
b etween Case I and II on the one hand and Case III on the other.
show that these differences are due to the nature of
I will
t h e a r g u ments of
these verbs.
I repeat th e v o ice p a radigm f o r d a t i v e v e rb s i n ( 3 6 ) f o r e ase o f
reference.
( 36) a .
M anolotra sary
aho.
ana o
AT.offer p i cture 2sg(acc)
1 sg(nom)
'I offer you a picture.'

b.
Tolo r a na
sary
offer.Vna picture
i a n ao.
goal
2sg ( nom)
A tolotra
an a o
a.offer 2sg(acc)
ny s a r y .
theme
d e t picture
'The picture is offered to you.'

Anolorana sary
CT.offer
i a n ao.
goal
pi cture 2sg(nom)

A s illustrated i n
( 3 7), t h er e a p p ear t o b e th r e e b a si c w o r d o r d e r
possibilities with these verbs in the active.

(37)
Cas e III: dative shift

a.
Nand r o so
(ny) vary hoan'ny zaza i Bakoly.
pst.AT.serve (det) rice f o r 'det child

'Bakoly served rice to the child.'
B a k oly
'" This gap clearly shows that the availability of the n- passive is lexically determined. There is
no thematicdifference between the theme of manome 'give' andthe theme of manolotra 'offer'.
Nevertheless, the -Vna passive promotes the former and the n- passive promotes the latter.
43
Theme Topic
Nandroso
(ny) vary ny zaza i Bakoly.
pst.AT.serve (det) rice d et child Bakoly

Nandroso
ny z a z a
ny v a r y i B a k oly.
p st.AT.serve det child d e t rice

'Bakoly served the child the rice.'
In (37a), the theme i s

p reposition ma y
Bak o l y
r e alized as a D P a n d t h e g o a l a s a P P . The
b e o m i t t ed, a s i n ( 3 7 b ). (37b) is a d ouble object
construction that maintains the theme>goal word order. T his order may
be reversed as in (37c).
The dative shift in (37c) appears puzzling when compared to the Case I
a nd II verbs discussed above. D oes (37c) involve base generation of t h e
goal in an object position along the line of instrumental advancement? I
suggest that this is in fact not the case. The shifted goal in (37c) does not
pattern with instrumentals or material themes. For example, the goal in
this position must b e d e f inite, in c ontrast to i n struments and m a terial
themes, which tend to be indefinite.
( 38) * N androso
zaz a
(ny) vary i Bakoly.
pst.AT.serve chil d ( d et) rice B akoly

'Bakoly served a child rice.'
Furthermore, the goal passivizes with the -V na s uffix, as shown i n t h e
example in (36b). This contrasts with locata, which are promoted with th e
a- prefix. Finally, as pointed out by Pearson (forthcoming) and confirmed
with other n a tive M a l agasy speakers, the or der
g o al>theme i s h i g h ly
marked and not possible with all dative verbs. To account for the change
in word or der in (37c), I assume that M alagasy has limited VP-internal
"scrambling" of arguments. Thus (36c) is not representative and I will not
further discuss this ordering.
(37a,b) indicate that the goal of a dative verb may be realized either as
a DP or as a PP. Are these argument or adjunct positions? To account for
these data, Pearson (1999; to
a p p ear) p r o p oses a s y n tactic d i f ference
b etween the DP-PP and th e D P-DP constructions.
44
H e c i t es data f r o m
Chapter 2
binding and coordination t o
s u p p ort t hi s d i stinction. M y co n s u ltants,
however, do not agree with the judgements. I n stead, I assume that goals
are freely generated as DPs or PPs.
optional arguments.
M o r e o ver, I s h o w t h a t g o als are
R e c all t h e t w o t e s t s f o r t h e a r g u m ent-adjunct
d istinction: o p t i onality and A T clefting. I n
g e n eral, the g oal o f d a t i v e
verbs is optional.
(39)
N a n droso
vary i B a koly.
p st.AT.serve r ic e
Bak o l y
'Bakoly served rice.'
On the other hand, the goal cannot AT cleft, either as a DP or a PP.
(40) * [ (Hoan')ny
z a za ] no nandroso v a r y i B akoly.
(for.gen) det child foc pst.AT.serve rice Bakoly

'It's to the child that Bakoly served rice.'
Clefting shows that although optional, the goal patterns with arguments.
I t is therefore not s u r p rising that i t
p a ssivizes to subject (as shown i n
(36b)). Finally, since the goal can be a PP, it can be promoted t o subject
with CT ( i l lustrated in (36d)).
I c o n c lude that th e g oal i s a n o p t i onal
argument of these verbs and can be generated either as a DP or a PP.

T urning next to th e themes of d ative verbs, these are p r om oted t o
subject with th e a- p assive, like th e
discussed above.
i n s t ruments an d m a t erial t h e m es
Unl i k e i n struments and material themes, themes are
obligatory arguments of dative verbs and cannot be omitted.'"

(41) * Nandroso
ny z a z a i Bakoly.
pst.AT.serve det child Bakoly

'Bakoly served the child.'
A s seen by t h e w o r d o r d e r i n ( 3 7), themes surface next t o t h e v e r b .

Moreover, themes are only realized as DPs, never PPs.
" (41) is grammatical, but means 'Bakoly welcomed the child'.
45
Theme Topic
( 42) * Nandroso
ny z a z a tam i n 'ny vary
i Ba k o l y .
pst.AT.serve det child p s t .P.gen.det rice Bakoly

I n this w ay ,
t h e mes ar e p u r el y a r g u m ental, u n l ik e i n s t ruments a n d
material themes, which have a dual status. I n
o t her w o r ds, they cannot
cleft and cannot be promoted to subject with CT."

( 43)
a. *
[ N y vary ]
det rice
no n a n droso
ny zaza i Bakoly.
foc p s t .AT.serve det child B a k oly
b.
Nandrosoan'i Bakoly
ny z a za ny yary.
pst.CT.serve.gen.Bakoly det child det rice

Not surprisingly, themes of dative verbs are arguments in the syntax.

The core difference between Case I, II verbs and Case III verbs lies in
the argument/ adjunct status and the category of the elements associated
with these verbs. For Case I and II v erbs, the instruments and material
themes can be realized either as DPs or PPs, as arguments or adjuncts.
F or Case III v e rbs, th e g o als ar e e i t her D P s o r P Ps, bu t a r e a l w a y s
arguments (albeit optional ones). Themes are uniformly D P ar guments.
(44) sums up the different classes.
( 44)
a.
b.
Case I , I I i . ve r b
Cas eI I I
DP l
ii. verb
D P2
DP l
verb
DP l
(P) DP2
[PP(adjunct) P DP2 ]
N evertheless, these verbs all share certain p r operties.

passive forms, one o f
T h e y h av e t w o
w h i c h ( th e a- p assive) is u sed fo r i n s t r u ments,
material themes and themes of d ative verbs.
W h a t u n i fies this group ?
The next sections answer this question.
(43b) is grammatical under an irrelevant reading where the theme is understood as partitive.
These partitive readings arise when a theme is promoted to subject with CT. S ee chapter 3 for a
discussion of the connection between partitivity and CT.
46
Chapter 2
3 P e arson 1998a -'-'
In order to provide a unified analysis of passive, Pearson (1998a) suggests
that th e
e l ements p r o m oted w i t h t h e a - p a s sive a r e a l l "d i s p laced
themes"."
Rapp a p or t and L evin (1988) refer to t his type of o bject as
"locatum" .
T h e -V n a p a ssive, on t he o t h er h a n d, i n d i cates a g o al o r
endpoint of the action described. Pearson translates the event structure of

such verbs di rectly i nt o p h r ase structure.
T h e l o catum ( th e d i splaced
t heme) is the specifier of a s m all clause headed by G O .
T he g o a l i s
contained in a resultative small clause complement of TO.
(44)
CAUS
SC
[locatum]
GO
VP
PathP
state/event
The following examples illustrate Pearson's analysis.
( 4S) a .
Nandroso
vary n y z azai B akoly.
pst.AT.serve r ice det child B akoly

[SC vary GO [pathp TO [SC ny zaza ]]]
rice
det child
This section draws on a conference presentation given by Matt Pearson. Since then, he has
further refined his theory of these alternations, which will appear as part of Pearson (forthcoming).
Since this thesis is still forthcoming, my discussion is limited to an early stage of his work.
" Kroeger (1990) describes an affix in K imaragang Dusun which is used for the promotion of
displaced themes. Thus these languages appear to be syntactically sensitive to this classification.
47
Theme Topic
Nandidy a n tsy ny hena Rasoa.
( 46) a .
pst.AT.cut knife det meat Rasoa

[sc antsy; GO [pathp TO [sc PRO;

knife
didy
n y hena ]]]
cutting det meat
In (45) and (46), vary 'rice' and antsy 'knife' are locata and hence appear in
the specifier of the GO small clause. Note that the embedded small clauses
a re distinct. I n
p a r t i cular, there is a PRO i n ( 46b), controlled by t h e
locatum antsy 'knife'. Intuitively, the instrument is both a locatum and the
external argument of the lower verb. T h ere is no such control relation in
(45b). This difference between instruments and other locata will become
important in the following discussion.
Overall, I
ag r e e w i t h P e a r son's o b servations a n d t h e a n a l y sis
p resented in section 4 p o sits a p articular ph rase structure position f o r

locata. I di ffer, however, in th e overall p h rase structure representation.
In particular, I see problems with the resultative small clause structure he
assumes. Let us first examine Pearson's motivation for the PRO in (46b).
P earson suggests that the availability of th e a- passive is tied t o t h e
type of instrument inv olved:
t o o l s v ersus aides. Th e distinction is one
that draws on o b servations by L a k of f ( 1 968); Wojcik ( 1976); M arantz

(1984). Consider the following pairs.
(47)
(48)
a.
Sigou r ney cut the salami with this knife.
b.
Thi s k n ife cut the salami.
a.
Sigou r ney ate the salami with this knife.
b.
This knife ate the salami.
In (47a), this knif e is a "tool" as shown by the grammaticality of (47b). I n

(48a), however, this knife is an "aide" and (48b) is ungrammatical." R e call
See Brunson (1992) for arguments that the distinction between (47) and (48) is not due to
different types of instruments. Instead, she suggests that 'eat' requires its subject to be sentient,
w hile 'cut' does not. Hence an instrument can be the subject of the latter but not the former. M y
48
Chapter 2
from example (35) in section 2.3.8 that in Malagasy mandidy 'cut' has an ap assive while mihinana 'eat' does not .
a ppear t o
T h e e x a m ples i n ( 49) an d ( 5 0)
s u p p or t t h e c o n n ection b e t w een t h e a - p a s sive an d t h e
possibility of the instrument appearing as the subject of an active verb.
(49)
mandidy 'cut'. instrument Jsubject of AT verb, Jsubject of a-passive

a.
M and i d y t s ara ny hena ity antsy ity.

AT.cut good det meat this knife this
'This knife cuts meat well.'
Adidy ny hena
a .cut
n y a ntsy.
d e t meat d e tk n i f e
(S0)
mihinana 'eat'. instrument *subject of AT verb, *subject of a-passive

a. * Mihinana tsara ny hena ity antsy ity.
AT.eat
go o d det meat this knife this
'This knife eats meat well.'
b.
Nahinan-dRasoa ny hena ny antsy.

a.eat.gen.Rasoa det meat det knife
Hence, cutting seems to involve a tool and eating an aide in Malagasy, as

in English. P earson's phrase structure encodes the semantic di fference
b etween th e t w o i n s t r u m entals. In oth e r w o r d s , t h e a b i l it y o f a n
instrumental to act as an "argument" of the verb is linked to its semantic
interpretation.
Ideally, the correlation in (49) and (50) should hold for all instances of
the a- passive. This is not the case, however.
F o r e x ample, themes that
normally take the a- passive cannot be the subject of an active verb.
analysis of Malagasy is similar in spirit: whether or not an instrument can be "advanced" to object
is determined in part by the semantics of verb. See section 4.2 for discussion.
49
Theme Topic
rnanondraka 'water'. material theme *subject of AT verb, Ja-passive

a. * M anondraka t s ara ny voninkazo ny ra n o .
AT.water
g ood d e t flo wer
det wate r
'Water waters flowers well.'
Natondraka ny voninkazo ny rano.

p st.a.water det flower
det w a t e r
'The water was watered on the flowers.'
(52)
rnandroso 'serve'. theme *subject of AT verb, Ja-passive

a. * Mandroso tsara ny zaza i t y yary ity.
AT.serve good det child this rice this

'This rice serves to children well.'
Naroso
n yz az a
it yy ary ity.
pst.a.serve det child t h i s rice this

'This rice was served to the child.'
It is therefore unclear whether such elements could control the PRO i n
Pearson's structure.
T h e refore, Pearson must posit tw o d i f f erent small
clause types associated with change of location (compare (45b) and (46b)).
According to Pearson, material themes and themes do not control PRO
a nd the u n grammaticality o f
( 5 1a) and (52a) is no t u n e xpected. T h i s
"solution" im m ediately raises the question of the difference between the

two small clause types. Is there any independently motivated distinction
between the two? M o reover, it is not clear how the DP ny zaza 'the child'
can be a state or event in the SC in (45b).
More im p ortantly, h o w ever, even
Pearson's correlation does not h o ld .
w i t hi n t h e c l ass of i n s t ruments,
In ot h e r w o r d s , t h e c o n nection
between an instrument appearing as the subject of an a- passive clause

and as the subject of an AT verb is limited. I n p assing, Pearson does not
state whether this correlation is a one-way implication (as in (53a,b)) or a
bi-directional (as in (53c)).

( 53) a .
b.
instrument subject of a- passive ~ i n s t rument subject of AT verb

instrument subject of AT verb ~ i n strument subject of a- passive
50
Chapter 2
instrument subject of AT verb ~
i n strument subject of a- passive
Let us consider (53a). For this to be false, we need to find an example of a

verb that allows the promotion of an instrument with th e a- passive but
does not allow the instrument to be the subject of an AT verb. A s show n
by (54b), ny fahita lavitra 'binoculars' is promoted with the a- passive. (54c),
h owever, w h er e t h i s i n s t r ument i s t h e s u b j ect o f t h e A T v e r b , i s
ungrammatical.
(54)
mj i e ~ 'w a tch'. instrument *subject of AT verb, Ja-passive

a.
Nij e r y
fahita
lavit r a ny vorona R a k o to.
p st.AT.watch nm.AT.see far
de t b ir d
Ra k o t o
'Rakoto watched the birds with the binoculars.'
Najerin-dRakoto
ny vorona ny f ahita l
avit r a .
pst.a.watch.gen.Rakoto d e t bir d de tnm.AT.see far

'The binoculars were used by Rakoto to watch birds.
c.
Mijery
tsa r a (ny vorona) n y fahita l
AT.watch good (det bird)
avitr a .
d e tnm.AT.see far
'Binoculars watch birds well.'
( 54c) indicates that the instrument 'binoculars' patterns semantically w i t h

"aides" or "facilitating instruments" rather than with " t o ols" . Y e t t h e apassive is grammatical.
T o falsify (53b), we n eed a v er b t h a t a l l ow s an i n s t rument a s t h e
subject when the verb is active, but does not promote the instrument t o
subject with the a- passive. As shown in (55a,b), mamoha 'wake' can take
an instrument as a PP or as the subject of th e A T
v e rb . In s t r u m ental
advancement is impossible (55c), however, and there is no a- passive to

promote the instrument to subject (55d).
(55)
mamoha 'wake'. instrument Jsubject of AT verb, *a-passive

a.
N amoh a
an'i Koto tamin'ny lakolosy Rasoa.
p st.AT.wake acc.Koto pst.P.gen.det bell

'Rasoa woke Koto with the bell.'
51
R a soa
Theme Topic
Namoha
an'i Koto ny lakolosy.
pst.AT.wake acc.Koto det bell

'The bell woke Koto.'
c.
Namoha
l akol o s y a n ' i K oto Rasoa.
pst.AT.wake bel l
d.
Navoha
acc.Ko t o R a s oa
a n ' i K oto ny lakolosy.
pst.a.wake acc.Koto det bell

Pearson's correlation clearly does not hold in either direction (and clearly
not bi-directionally). I
t h e r efore conclude that his test fo r t o ol s v ersus
aides is not helpful in determining the range of the a- passive. Th e fact

that some instruments can be AT subjects is tangential to the question of
passive. A s w i l l become clear in t h e n ext section, it i s th e i n strument
a ppearing in an internal argument position that determines passive. I n
other words, the grammaticality of the instrumental advancement in (54a)
i s crucial to th e g r a mmaticality o f t h e p a ssive in (54b). S i m i l arly, t h e
ungrammaticality of (55c) determines the ungrammaticality of (55d).
I agree, h o w ever,
w i t h P e a rson's o bservation t h a t t h e e l e m ents
t argeted by t h e a - p assive are locata. T h i s c a n b e m a d e e v i d ent b y

comparing the grammatical a- passive in (54b) w it h
t h e u n g r ammatical
(56).
(56) * Najerin-dRakoto
ny vorona ny solomaso.
pst.a.watch.gen.Rakoto d e t bir d de t glasses

'The glasses were used by Rakoto to watch birds.'
When asked about the difference between (54b) and (56), my consultant
replied that you have to adjust binoculars, while glasses just sit on one's
face. This impressionistic response indicates that the a- passive is in fact
s ensitive to certain semantic features of th e p r o m oted element. I wi l l
elaborate further on this common interpretation in the following section.
52
Chapter 2
4 A
u n i f ied anal sis of a s s ive
4.1 Structure
Although our basic observations are the same, I w il l p r o pose a slightly
different phrase structure from Pearson's. I adopt a VP-shell structure a la
Chomsky (1995), augmented with an additional v2P between the hi gher
v 1P (where agents are generated) and t h e
l o w e r V P ( t h e d o m ai n o f
internal arguments such as themes and goals)."

vlP
DP
l'
<ag> A
0
<locatum>
R
V'
DP/PP
<goal>
I suggest that i n struments, material themes and th e t h emes of
dative
verbs ( = locata) may be base generated in the specifier position of v 2P.

The label of this projection is not crucial to the argument.
I m p o r t antly,
the locatum is in a specifier (argument) position in the verbal projection."

Passive results from the loss of accusative Case. For transitive verbs,
t he internal argument t hen m o ves t o
ISpec, IP] fo r
n o m i n ative C ase.
" On the relative position of themes and goals, I assume theme>goal, following Baker (1996a).
R ecall that in Malagasy, both NP and PP goals follow the theme, with a few exceptions. T h e
position of themes and goals is subject to debate in the literature, a topic I touch on in section 7.2.
As pointed out by Jonathan Bobaljik (p.c.), the n- passive could be relativized, rather than
defined over a particular position. I n o t her words, then- passive would promote the highest
argument below the agent. Note that this approach still requires instruments and material themes
to be generated in a position c-commanding the theme/goal. A l though attractive in current views
of"shortest move", I believe that this analysis misses certain generalizations. For example, as I
will discuss below, if locata all appear in one particular position we can easily capture facts about
transitives that only have the n- passive. F or t h ese verbs, their single internal argument will
appear in [Spec, v2P].
53
Theme Topic
Recall that some verbs will t ake the a- passive and others the -V na. I n
cases where there is no alternation, the choice between the two affixes is
lexically determined (but see below for discussion). On the other hand, if
a verb has two " i n t ernal" D P
a r g u ments, then tw o p a ssive forms w i l l
generally be available. For these ditransitives, potentially either argument

can raise to subject. T h e ar gument i n [Spec, v2P] will be promoted to
subject with the a- passive form (i.e. instruments, material themes, themes
of dative verbs). In other words, with the a- passive, no accusative Case is
available in [Spec, v2P]. The argument in the lower VP w ill take the -Vna
passive. -Vna passive therefore signals the lack of Case within VP."
A s for th e m o r p h ology, t hi s analysis is consistent w it h t h e s t r o n g
lexicalist view, where the fully formed verb w ith all its affixes is inserted
directly into the syntactic structure. I a m m o r e sympathetic, however, to
the view that verbal morphemes head distinct projections. I
t e n t atively
suggest that the a- prefix is the head of v2P (e when the verb is active).
The passive suffix, -Vna, on the other hand, is in a different head, outside
the verbal projection. That one passive is a prefix and the other is a suffix
is interesting and worth further investigation.
4.2 Semantics
In the discussion of Pearson's analysis, I pointed out that semantics seem
to play a role in d etermining w h ich elements are promoted w it h th e apassive. I w ill now c l arify this question. I n
t h e t r aditional literature on
Malagasy, grammarians have suggested that semantics directly di ctates

t he range o f
t h e a - p a s sive. A cc o r d in g t o R a h ajarizafy ( 1960), t h e
instruments which are promoted w it h th e a- passive are in some sense

necessary for th e a ction described by t h e v e rb . F o r e x a m p le, cutting
necessarily involves some cutting i m p l ement, w h ereas eating does not
r equire the use of an y
i n s t r ument ( other t han th e eater's ow n b o d y ) .
Hence the instrument of 'cut' will be promoted with the a- passive, but not
the instrument of 'eat'. W e h ave already seen this to be the case. In th e
discussion of the data, I have also mentioned the relevance of change of
location, and therefore have used the term " l ocatum" fo r th e ar guments
promoted with the a- passive. Is this notion of locatum sufficient? No, fo r
" Instruments, material themes and goals may also be generated within a PP. I n t hat case, they
are promoted to subject with circumstantial topic. I leave aside discussion of CT until chapter 3.
54
Chapter 2
many instruments undergo change of l o cation w i t h ou t a l l ow ing th e apassive. Recall the above discussion of mihinana 'eat' in (50). I
t h e r efore
f ollow R ahajarizafy's in tuition t h a t c e rtain l o cata ar e i n h erent i n t h e

meaning of the verb. Furthermore, I assume that these locata are present
in the LCS of the verbs and therefore can be realized as arguments. As
arguments, locata can then be promoted to subject with passive. In o t h er
words, the semantics determines the structural position of th e l o catum.
This structural position is crucial in the application of passive. In this way,
I concede the importance of semantics in determining the range of the ap assive. Clarification of these notions will come w it h a careful study o f
the lexical semantics of these verbs. Such a study is beyond the scope of
this thesis (and beyond the competence of this researcher), however, and I
leave it to a future lexical semantic investigation.
4.3 Change o
f location
The crucial data thus far have come from tw o p assive verbs. N ecessarily,
these verbs have two internal DP arguments. I
h a v e a r gued that the a-
passive promotes DP arguments in [Spec, v2P], while the -Vna passive

promotes DP arguments in the lower VP. W hat about verbs with a single
internal DP argument?
W e h a v e already seen that transitive verbs fall
into two classes: those that take a- and those that take -Vna. Since I have
l inked [Spec, v2P] t o
a p a r t i cular i n t erpretation ( l ocatum), w e w o u l d
expect transitive verbs that only have the a- passive to express a change of
location. This is in fact true:
t h e t h emes of these verbs undergo some
change of location or orientation."
( 58) a .
mandraka 'raise'
b.
midina
'lower' (transitive)
c.
mandavo
'spill' (transitive)
d.
mamindra 'move, displace'
Unfortunately, there is no one single syntactic test for locata. T h e r efore
Traditional Malagasy grammarians have discussed the factors determining the n- passive with
single argument verbs such as those in (58). Rahajarizafy (1960) claims that verbs that express a
situating or positioning take the n- passive. Rajaona (1972) suggests that verbs which transform
their objects take the n- passive. From the discussion, it should be clear that I am formalizing
Rahajarizafy's intuition.
55
Theme Topic
w e must rely on i n d ividual intuitions to d e termine w h ether t hi s i s t h e
correct characterization. A s
m e n t ioned above, I leave the semantics of
these verbs for further investigation.

I nterestingly, change of location does appear to l i cense word o r d e r
alternations in languages other than Malagasy. English change of location
v erbs, for
e x ample, a l lo w a l t e rnations similar t o t h e M a l a gasy d a t a
discussed in t hi s ch apter.'"
W e l l - k n ow n i s t h e sp r a y/l oad alternation
( Anderson (1971); Rappaport an d

( 1991)).
L e vi n ( 1 988); H oekstra an d M u l d e r
M o r e over, a s d i scussed by F i l l m or e ( 1970), certain v e rbs o f
physical contact also exhibit a change in word order. Typical of this class is
'hit', but also 'slap', 'strike', 'stroke', 'bump'.
(59)
a.
K im h i tthe table with the stick.
b.
K im hi t t h e stick against the table. (=
These verbs bear a s t r ong s i m i larity t o
(59a))
t h e i n strumental advancement
verbs of Malagasy. Thus both M a l agasy and English change of location

verbs allow word order alternations with minimal interpretational effects.
Interestingly, Malagasy also singles out the change of location class with
different passive affixes.
Verbs of change of state (e.g. 'break', 'bend', 'fold', 'shatter', 'crack'),
however, show a radical shift in meaning with word order change.
( 60) a .
b.
Kim broke the table with the stick.
Kim broke the stick against the table. (~ (60a))
The same holds true in Malagasy. The change in meaning is all the more
dramatic as the preposition remains the same. Thus it is purely position
that determines which argument u n d ergoes the change of state.
world kn ow ledge determines just ho w
( R e al
t h e c h ange of s t ate is realized;
(61a) could m ean 'Ketaka broke t he stick wi th t he t a ble', but t h is is a n
Goldberg (1995) discusses the "caused motion construction", which subsumes the spraylload
alternation. I have not found Malagasy equivalents to the intransitive swarm class.
(i) a .
B e e s are swarming in the garden.
b. T h e garden is swarming with bees.
This may be accidental or indicate that the swarm class should not be grouped together with the
spraylload class, as suggested by Jackendoff (1996) and contra Hoekstra and Mulder (1991).
56
Chapter 2
unlikely interpretation.)
(61)
a.
Nama ky
ny lat a batra ta m i n 'n y
pst.AT.break det table
langi l angy i Ketaka.
pst. P .gen.det stic k
Ketaka
'Ketaka broke the table with the stick.'
Namaky
ny la n g ilangy tamin'ny
p st.AT.break det stick
l ata b atra i Ketaka.
pst .P . gen.det tabl e
K etak a
'Ketaka broke the stick against the table.'
Moreover, none of the Malagasy verbs that allow the a-I-V na alternation
are pure change of state verbs like mamaky 'break'. T hus change of state
neither licenses word order alternations nor distinct passives.
Summing up, the class of change of location verbs have a special status
in the syntax. A l t ernating verbs in Malagasy and English all incorporate
some change of location. (Change of state may of course also arise, as in
the locative alternation. ) In Malagasy, this class is further distinguished by
the availability of different passive affixes. Again, an understanding of the
p recise relevance of c h ange o f l o c ation an d h o w t h i s i n t e racts w i t h
syntactic structure remains to be determined.
5
D i s cussion
I now discuss certain details of the above analysis. In particular, I examine

how Case and t r ansitivity i n t eract in th e a l ternating v e rbs c onsidered
a bove. I
t h e n l o o k f o r e v i d ence from c o o r d ination i n f a v ou r o f t h e
p roposed s t ructure .
So m e da t a a r e p ro m i s i ng, w h i l e o t h er s a r e
problematic due to the unusual coordination patterns in Malagasy. I al so

discuss the motivation for p o siting base generation in
ESpec, v2P] over
syntactic movement. This leads to a comparison with similar alternations

in other languages. I
p r o v i d e p o tential criteria for d i stinguishing base
generation from movement in word order alternations.

5.1 Case and transitivity
I assume that the verbs that allow the a- j-Vna alternation can assign more
than one accusative Case in the active form. This can be directly observed
in the following example.
57
Theme Topic
(62)
a.
Nanon d raka r a n o an - d Rasoa i S ahondra.

p st.AT.water
w a te r a c c . Rasoa
Saho n d r a
'Sahondra sprayed water on Rasoa.'

b.
Nano n draka a n ' i o a z y
i Sa h o ndra.
pst.AT.water acc.this 3(acc) Sahondra

'Sahondra sprayed this on her.'
Nihinana an'io i Sahondra.
pst.AT.eat acc.this Sahondra
'Sahondra ate this.'
Overt accusative Case is m arked o n
p r o pe r n a mes bu t n o t o n l e x ical
nouns, as seen in (62a). If we replace the noun with a pronoun, however,

overt Case-marking surfaces. This is illustrated in (62b). ((62c) shows that
"normal" accusative pronouns have the same surface form. ) Under the
p roposed analysis, passive involves the loss of one of these Cases. N o t e
that multiple accusative Case marking is in general possible in Malagasy,
as illustrated by the causative below.'"
(63)
N a m panasa an ' i K oto an-dRasoa
aho.
pst.cause.wash acc'Koto acc-Rasoa
Isg(nom)
'I made Rasoa wash Koto.'
Thus no special Case marking mechanisms are required for th e m u l tiple

passive verbs under discussion. See Baker (1988) for a similar conclusion
for applicatives in l a n guages wit h
m u l t i pl e structural Case. Since the
a lternating v e rbs ar e l e xically d e termined, i t i s n o t u n r e asonable t o

s uggest that they ar e also lexically marked a s assigning/ checking t w o
" Both the causee and the embedded theme can passivize to subject. U n like the alternating verbs
discussed in this chapter, however, only one passive form is used, -Vna. Hence, (i) is ambiguous.
an-dRasoa i Koto.
(i)
Nampa n asako
pst.cause.wash.Vna.lsg(gen) a c c-Rasoa i Koto
'Koto was made by me to wash Rasoa.'
or 'Koto was made by me to be washed by Rasoa.'
That only the -Vna passive is available is not surprising. Neither the causee nor the embedded
theme qualify as displaced themes. For me, both arguments would be generated in the lower VP
and hence targeted by the -Vna passive. Note that other than causatives, all "double accusative"
verbs (e.g. datives) are precisely those under investigation in this chapter.
58
Chapter 2
Cases. Structural accusative Case is available in [Spec, v2P] and in the
lower VP.'" W hen the instrument or material theme appears in a PP, v2P
is not projected.
There remains another possibly Case-related question. A s m entioned
e arlier, the only v erbs that license these alternations in w or d o r der a r e
t ransitive.
T h e r e ar e n o e x a m p les o f i n s t r umental a d v ancement, f o r
example, with a n i n t r ansitive verb ( this appears to b e t r u e f o r m a n y

i nstances of applicatives cross-linguistically, but see footnote 19) .
(64)
indicates schematically the unattested alternation.

(64)
a.
V PP( in s trument)
b. *
V D P(instrument)
T herefore these verbs d eserve some d i scussion. I n w h a t f o l l o w s , I

speculate on the reasons for the transitivity effect. I
t hat it i s du e t o
t e n t atively suggest
t h e s emantics of t h ese v erbs r a ther t h a n t h ei r C a s e
properties, per se.

F irst, recall that n o t a l l v e rb s t h a t h a v e t h e a - p a ssive ar e t h r e e
argument verbs. Some verbs only have the a- passive, which pr om otes
the sole internal argument (see (58) above for some examples). This class
o f verbs encodes a change of location, as mentioned earlier and I w o u l d
claim that the internal argument is generated in [Spec, v2P]. Hence there
i s no restriction per se o n
a rguments in the l ower VP .
h a v in g a n e l e m ent i n [Spec, v2P] and no

B u t t h e v e rb s i n (58) are not a l ternating
verbs; hence they do not show w h e ther or no t a l ternations are possible

with single argument verbs.
Similarly, certain locative alternation verbs (e.g. mamafy 'sow') have the
goal as an optional argument and t hus have the appearance of being a
s ingle argument verb .
W h e n t h e g o a l i s no t e x p ressed, however, t h e
material theme must surface as a DP, no t a s a PP; hence the contrast
between (65a) and (65b) (compare with (64) above).
" Marantz (1993) assumes that structural Case is assigned in all [Spec, VPj positions. I f urther
assume that structural Case is assigned to the goal, sister to V'.
59
Theme Topic
( 6S) a.
Mamafy voa ny mpamboly.

AT.sow seed det farmer
'The farmer sows seeds.'
b. *
Mamafy amin'ny vo a
ny m p amboly.
AT.sow P.gen.det seed d e t farmer

With the -Vna passive, however, either the DP or the PP is possible.
( 66) a .
Fafazan'ny
m pam b oly voa ny tany.
sow.Vna.gen.det farmer
se e d de tland
'The land is sown seeds in by the farmer.'
b.
Fa fazan'ny
m pam b oly amin'ny voa n y tany.
sow.Vna.gen.det farmer P.gen.det seed det land

'The land is sown with seeds by the farmer.'
(65) indicates that there is no alternation available when the verb takes a
single argument. O nce we add another argument, as in (66), the DP-PP
alternation reappears. Again, it appears that the DP-PP alternation is only
permitted with transitives. This is shown schematically in (67).
(67)
a.
V DP PP ( i nstrument/material theme)
b.
V DP(i n strument/material theme)

DP
(67a,b) correspond to the structures underlying (66a,b). Is the d i f ference

between (64) and (67) a general pattern?
What we would be looking for is a verb like 'walk' (or, even better, an
unaccusative ) which could take either a PP instrument or a DP instrument,
but no other internal arguments. And this DP instrument would passivize
with the a- prefix. (There would be no -Vna alternate since 'walk' takes no
(other) internal DP arguments.) C r ucially, the verb w o ul d h ave to b e a
change of location verb to allow the instrument to be generated in [Spec,
v 2P]. A s f a r a s I c a n t ell, th ere are n o v e rb s l ik e t h is."
' There is one possible example of this, illustrated in (i).
60
Due t o t h e
Chapter 2
restricted nature of " i n strumental advancement" i n
M a l a gasy, this gap
could simply be accidental. There are other possibilities. For example, the
class of verbs discussed herein encode a directed c hange of location. I n
other words, the motion is directed toward an endpoint. I f w e
notion a s
i m p o r t ant, t h e n i t is not
t ak e th is
su r p r i s in g t h a t t h e r e is n o
"advancement" with unergatives. Unergatives, by definition, will lack the

e ndpoint of
m o t i on .
T h e s e ar e t e n tative speculations, however, a n d
require a better understanding of the lexical semantics of verbs of motion.

I leave this for f u t ure research. I t
i s w o rt h n o t in g t ha t M a ssam (1998)
shows that Niuean instrumental advancement requires an agentive verb.

Thus similar word order alternations display similar dependencies on the
semantics of the verb.
5.2 Constit uency
To provide ev idence in
f a v ou r o f t h e p r o p osed st ructure, I c o n sider
coordination, a classic test fo r
c o n stituency. U n f o r t u n ately, M a l agasy
coordination f acts are q u it e c o m plex .
M or e o v er, w i t h t h e p r e d i cate
fronting analysis of word o r der m entioned in chapter 1, it is difficult to

d etermine just what constituents are being coordinated.
T h e r efore t h e
results may be inconclusive.

I n the proposed structure, the locatum fo rm s a constituent wit h t h e
goal, to the exclusion of the verb (assuming verb raising). C o o r d i nation
facts suggest this to be true.
( 68) a .
Nandrakotra b odofotsy ny fandrinany sy l a mba ny latabatra

pst.AT.cover blanket
de t bed.gen.3 and cloth d et table
i Sahondra.
Sahondra
'Sahondra covered her bed with a blanket and the table with a cloth.'
M a n d eha amin'ny f iara i S o a

AT.go P . g en.det car Soa
'Soa goes by car.'
b. Mandeha fiara i Soa.
AT.go c a r Soa
'Soa goes by car.'
(ia), however, is marked and (ib) has the status of a fixed expression. In this instance, Para 'car' is
promoted to subject with CT and not passive.
(i) a .
61
Theme Topic
b.
Nani n d rona a n t sy ny trondro s y l efona ny omby i Sahondra.

pst.AT.pierce knife det fish
and s p ear det ox
Saho n d ra
'Sahondra pierced the fish with a knife and the ox with a spear.'
V [[ 2plocatum goal ] and [,2p locatum goal ]]

(68c) illustrates the basic structure. U n der the proposed analysis, the data
in (68) represent v2P c o ordination .
S i m i l a rly, VP s can b e c o n joined,
excluding the verb and the locatum.
( 69) a .
Nandidy a ntsy ny hena teo
ambonin'ny latabatra
pst.AT.cut knife det meat pst.there on.gen.det table
sy ny mofo tao
an-dakozia i Koto.
and det bread pst.there
at- k i t c he n K ot o
'Koto cut the meat on the table and the bread in the kitchen
with the knife.'

V locatum [[vp goal PP] and [vp goal PP ]]
A s shown s chematically i n
( 6 9 b), ( 69a) i s a n e x a m pl e o f l o w e r V P
coordination.
P roblematic, however, are th e f o l l ow in g d a ta, w h ere th e v er b
and
locatum are conjoined.
( 70) a .
Nandidy a n tsy sy nandraraka sira ny hena Rabe.

pst.AT.cut knife and pst.AT.pour salt det meat Rabe
'Rabe cut with a knife and poured salt on the meat.'
Nandroso
var y sy nanolotra
labi e ra ny vahiny i Ketaka.
pst.AT.serve r ice and pst.AT.offer beer

d e t guest K e t aka
'Ketaka served rice and offered beer to the guests.'
[[~ V locatum ] and [~ V locatum ]] goal
In the clause structure proposed in (57), the verb and the locatum do not
form a constituent to the exclusion of the goal. D o
62
t h e examples in (70)
Chapter 2
involve Right Node Raising? If, however, I assume predicate fronting, the
data in (70) can be accounted for .
P r i o r t o p r e dicate fronting, the goal
moves to an XP position above v1P.
T h e n th e coordinated v1P (which
contains the verb and locatum and trace of the goal ) raises to [Spec, TP].
TP
(71)
xp
DP, /
V locatum t;
V locat u m t
D espite its rather baroque appearance, this kind o f
m o v e m ent i s q u i t e
common in coordinated structures in Malagasy. C o n sider the followin g

examples, cited by Keenan (in press):
( 72) a .
Nividianako
sy namaky
ilay b o k y i a nao
pst.CT.buy. 1 sg(gen) and pst.AT.read d e f book
2 s g(nom)
'You [[were bought+for by me and read] that book]'

b.
Nano n droako
ka naka
ilay toerana ianao
pst.CT.point-out. 1 sg(gen) and pst.AT.take d e f place 2 s g (nom)

'You were pointed out by me and took that place.'
In (72a), for example, the object ilay boky 'the book' h as undergone ATB
extraction to an object position. Under the predicate fronting analysis, the
coordinated VPs (or similar constituent ) subsequently raise to [Spec, TP]."
' What is unusual in these examples is that one verb is active and the other is CT.
English equivalent would be the following:
(i)
Th at b o at was bought by Eve and sails well.
63
T h e closest
Theme Topic
More work is clearly required on coordination in Malagasy.
5.3Base generation vs. m ovement
In the analysis presented, I
h a v e p o s ited t h a t c e r tain e l ements ( e .g.
instruments ) may be base generated in [Spec, v2P]. U n der an alternative

approach, the same element could be generated elsewhere and move into
a similar position. I n fact, either approach captures the basic facts about
the alternations. I
w i l l n o w t u r n t o s om e ar guments in f avour o f b a se
generation.
This issue arises with similar alternations in English: dative shift and
l ocative a l ternation .
D o the w or d or d e r di f f e r ences o b t ai n f r o m
movement or from different base structures? A full r ange of answers has

been offered in the literature. Since this topic has been treated by several
researchers and a complete discussion would be beyond the scope of this
chapter, I will l i mit m y self t o a
b r i e f d i scussion here. R e c ently, Baker
(1997) has reviewed th e l i t e rature o n b o t h d a t i v e s h if t a n d l o c ative

alternations. B aker argues that the former, but no t th e l atter, in v olves
s yntactic movement .
H e of f e r s s y n tactic evidence in f a v ou r o f t h i s
position, but he also notes that while the locative alternation results in a
clear change in aspectual meaning (affectedness), dative shift d oes not .
For Baker, the affected argument corresponds to a structural position, the
true theme position. T h erefore, he claims that the locative alternation is
b ase generated as two d i f ferent u n d erlying structures.
section 6 that i n
W e w i l l se e i n
M a l agasy, none o f t h e a l t ernations discussed change
affectedness relations. Following Baker's argumentation, we would t h en

e xpect the different word o r d ers to be d erived via m o v ement .
In this
section, however, I will argue against movement.

A lthough
t h e l o c a tive, i n s t r umental a n d da t i v e a l t e r nations i n
M alagasy d o
n o t c h a ng e t h e a f f ectedness relations, t h er e a r e b o t h
semantic and syntactic differences between the two structures. I w ill take
this as indicative of base generation rather than m ov ement.
F i r st, recall
t hat it is possible to semantically define the class of elements that m a y

appear in an internal DP argument position; all are displaced themes or
"locata".
S uch a
c o n sistent p a t tern r e m a in s u n e xplained u n d e r a
m ovement account. I t w o ul d be d ifficult to restrict movement t o
[Spec,
v2P] to a certain semantic class. Syntactic positions, however, tend to b e
64
Chapter 2
associated with i n t erpretation
(e.g. theta-assignment, op erator-variable
constructions ) . It i s t h erefore not u n r easonable to assume that o nl y
certain semantic class of elements may be generated in [Spec, v2P]."

Second, there are d i f ferences between the alternations that suggest
base generation rather than m ovement.
A s h a s been noted for English
(Gruber (1965); Lakoff (1968) and others), instrumental phrases require a

truly "agentive" subject. I n animates are not possible subjects in clauses
with instrumental phrases.
(73)
* The explosion killed the workers with a rock.
This is true for both instruments and material themes in Malagasy."
(74)
a. *
Namono ny mpiasa tamin'ny
va to ny fipoahana.
pst.AT.kill det worker pst.P.gen.det rock det explosion

'The explosion killed the workers with a rock.'
b.
Nanototra ny saha tamin'n y
vovok a ny riyotra.
pst.AT.fill det field pst.P.gen.det dust

'The wind covered the field with dust.'
d e tw i nd
O n the o t her h a nd , i f t h e m a t erial t h em e a p p ears as a n o b ject, t h e

sentence in (74b) becomes grammatical."
(75)
N a n ototra vovoka ny saha ny riyotra.

pst.AT.fill dust
d e t f ield d e t wind
'The wind covered the field with dust.'
Therefore, whatever renders (74b) unacceptable is not p r esent in ( 75).

This distinction between movement and base generation must be applied with c aution,
however. For example, not all verbs in English passivize (*Kim is resembled by Sandy). Do es
this entail that passive involves base generation rather than movement? I w ould claim that for
verbs like 'resemble', the apparent object is not in a true direct object position. H e nce passive
cannot apply.
' The grammaticality of the English translation of (74b) shows that in English, material themes
do not require an agentive subject. It is not clear why Malagasy and English differ in this respect.
' I have not been able to construct the equivalent of (75) with an advanced instrument. This may
be due in part to the restricted nature of instrumental advancement. Moreover, inanimate agents are
instrument-like and it is impossible to have two instruments in a clause.
65
Theme Topic
Since the incompatibility between an inanimate subject and a PP material
t heme is semantic in n a t u re, I s u g gest i t r e sults f ro m t h e t h et a r o l e
assigned to the material theme w i t hin a PP.
W h e n g enerated in [Spec,
v2P], the material theme receives a different theta role (locatum), one that
i s compatible w i t h a n i n a n i m at e s u bject .
A gai n , t h e c o n t r ast i n
grammaticality between (74) and (75) is indicative of d i stinct structures

underlying the two examples.
F inally, r ecall t h e d i f f e rence b e tween p a ssive an d C T cl e f t s o f
i nstruments and material themes, discussed in sections 2.3.4 and 2.3.7. I f
the verb is passive, a preposition is not p ossible in the cleft position, as
shown in (76a). If the verb is CT, however, a pr eposition is possible, as
given in (76b).
( 76) a .
[(*Amin')ny antsy ] no adidy ny hena.

(P.gen.)det knife
foc a . cut det meat
[ (Amin')ny
a n tsy ] no andidiana ny hena.
(P.gen.)det k n if e
f o c CT.cut det meat
I f the a- passive were t o i n v o lv e m o v ement f r o m t h e P P p o sition v i a

[ Spec, v2P], it i s u n clear ho w
t o a c c ount f o r t h e i m p o ssibility o f t h e
preposition in (76a). On the other hand, if the locatum in [Spec, v2P] is

a lways generated as a DP, the lack of a preposition is not surprising. W e
will see in chapter 3 that when a PP is promoted to subject in a CT clause,
the preposition is always permitted in a cleft.
In this section I
h a v e s u g gested some c r i teria t o d i s t inguish base
generation from m o v ement. I
w i l l n o w see how t h ese criteria apply t o
word order alternations in other languages.

5.4 Cross-linguistic evidence
As I have pointed out several times, the word order alternations that arise
in M alagasy ar e
s i m i lar t o E n g l i sh . Eng l i s h h a s b o t h t h e l o c a t ive
a lternation and d a tive shift .
I w il l n o w b r i e fl y c o n sider som e o t h e r
languages which also exhibit these alternations.
66
A l t h o ugh I do not have
Chapter 2
e nough data to d raw an y f i r m c onclusions, I will suggest that i n s o m e
c ases, the alternation is syntactic and therefore derived via movement.
In
other cases, a base generation analysis appears to be more appropriate.

5.5.1
In s t rumental advancement
Recall that in Malagasy, certain verbs allow a certain class of instruments

to appear as direct arguments. T his " i nstrumental advancement" occurs
in other Austronesian languages (e.g. Niuean (Seiter (1979)) and Madurese
(William
D a v i es, p . c.)), u s u ally a c c ompanied b y
morphology.
cha n g e s i n v er b
I n N i u e an, an ergative VSO language, the preposition aki
appears cliticized onto the verb.
( 77) a .
Kua hele tuai e
S i one e f a l aoa aki

[ [ e t it i p i
haa n a ]].
p erf cut p e r f erg Sione abs bread with a b s knife h i s

'Sione has cut the bread with his knife.'
Kua hele ak i tuai e
Sion e [ e t i t i pi h a a na ] e f a l aoa.
perf cut wi t h perf e r g Sione abs knife hi s

'Sione has cut the bread with his knife.'
abs b read
The former object of the preposition appears between the ergative agent
and the absolutive t h eme .
M ad u r e se, SVO, u ses a g e n eral-purpose
valency-extending affix, -aghi." In ( 7 8 a), the instrument appears in a PP.

In (78b), the instrument surfaces in the object position and the verb bears
extra morphology.
(78)
a.
A li no t o p
cend e l a
Ali A V.close window
biq k o r ten.
wi t h curtain
'Ali covered the window with a curtain.'
b.
A li n o t op-aghi
korte n daq cendela.
Ali AV.close-aghi
cur t ain to w i ndow
'Ali covered the window with a curtain.'
In contrast to N i u ean and M a d urese, the form o f
t h e v erb i n M a l agasy
' -aghi is also used to add a causee or benefactive (William Davies, p.c.).
67
Theme Topic
remains constant in these alternations.
Instruments ma y
N iuean).
a l s o b e p r o m o ted t o s u b j ect i n M a d u r e se (and
In the s e c a ses, th e v e r b m o r p h o l og y i n d i cates t ha t t h e
i nstrument has first been pr omoted t o o bject and then to subject.

M adurese example i n
The
( 7 9 ) i l l u strates that t h e v e r b b e a r s b o t h t h e
instrumental advancement suffix (aghi) and the passive prefix (e).

(79)
K o r ten jhuwa e-totop-aghi A l i daq cendela.

curtain det
OV . c l ose.aghi Ali to w i ndow
'The curtain was used by Ali to cover the window.'
I speculate that instrumental advancement in these languages is syntactic.

Looking at Niuean, for example, it appears that the range of instruments
that can be "advanced" is quite broad, less restricted than in Malagasy.
(80)
N e hopo aki e
ia e
kav e toua
pst jump aki erg she abs cord rope

'She jumped with a rope.'
I would take this as an indication of syntactic movement.
(Massam (1998),
however, assumes base generation.)

5.5.2 L ocative alternation
Drawing on data from Tagalog, Voskuil (1996) argues for a " t w o - stage"
analysis of passive that resembles in many respects the account given here
for Malagasy. A s in M a lagasy, some Tagalog verbs have more than one
passive alternate. For example, the verb tanim 'to plant', has two passive
f orms, each of w h ic h p r o m o tes a d i f f erent a r g ument o f t h e v e r b t o
subject. The examples in (81) are from Voskuil.
( 81) a .
Nagtanim
siya
ngbulaklaksahardin.
perf.mag.plant3sg.nom acc flowers obl garden

'She planted flowers in the garden.'
68
Chapter 2
I tinanim
ni y a
ang bulaklak sahardin.
perf.i.plant 3sg.gen n o m flowers obl garden

'The flowerswere planted by her in the garden.'
c.
Ti na m na n
niya
ng bulaklak ang hardin.
perf.plant.an 3sg.gen acc flowersnom garden

'The garden was planted by her with flowers.'
In (81b), the theme of the verb (ang bulaklak'the flowers') is the subject,
while (81c) has as subject the location of planting (ang hardin 'the garden').
Moreover, in (81b), the verb has a passive prefix (i-) and in (81c), the verb
has a passive suffix (-an)."
V o s k uil argues against an analysis where the
affixes pick out a DP with a particular theta role of the verb (agent, theme,
location) and promote it to n o m i native (i.e. a theta-agreement analysis).
Instead, he suggests that the alternation in (81) derives from the locative
alternation, which modifies the verb's argument structure. Subsequently,
the verb is passivized. V o skuil's label "tw o stage passive" is somewhat
misleading. The first stage is a change in argument structure which allows
either th e m a t erial t h em e o r t h e g o a l t o b e p r o j ected a s t h e d i r e ct
argument of the verb. T h e second stage is passive. In this way, Tagalog
has only
o n e p a s sive, w h ic h p r o m o tes t h e d i r ect o b ject t o s u b j ect.
Although details of Voskuil's analysis differ from m i n e (in p articular, he

assumes there is a single direct object position), the general conclusion is
the same: the different passives are the result of different base generation
structures."
5.5.3
Ap p l icative
Outside of A u s t r onesian, there ar e l a n guages which exhibit a s i m i l ar

m anipulation o f
i n t e r nal a r g u m ents .
languages are w e l l-know n
M or e sp e c i fically, t h e B a n t u
f o r t h e s o -called applicative construction.'"
Applicatives involve the promotion of the object of a preposition to direct

object, accompanied by m o r p h ological marking on th e v e rb .
C o n s i d er
' Interestingly, Tagalog and Malagasy both use a prefix to promote the locatum and a suffix to
promote the goal.
Unlike Malagasy, however, Tagalog does not show the locative alternation in th e active.
Hence there is no 'plant the garden with flowers' variant of (81a). I am not familiar enough with
the Tagalog data to speculate on the reasons for this gap.
Applicatives are of course not limited to Bantu languages. W i t hin Austronesian, Chamorro
(Gibson (1980)) and Indonesian (Chung (1976a)) are argued to have applicative constructions.
69
Theme Topic
the following pair from Chichewa, discussed in Baker (1988).
( 82) a .
Msangalatsi a - k u-yend-a
entertainer
ndi ndodo.
sp- p res-walk-asp with stick
'The entertainer is walking with a stick.'
Msangalatsi a - k u-yend-er-a
entertainer
ndodo.
sp- p res-walk-with-asp stick
'The entertainer is walking with a stick.'
In (82a), the DP ndodo 'stick' appears in a PP, while in (82b) there is no

p reposition an d
t h e v e r b b e a r s t h e s u f f i x -ir.
Cros s - linguistically
applicative constructions involve d atives, benefactives, instruments and

locatives.
I m p o r t antly f o r c o m p a rison w i t h M a l a gasy, th e C h i chewa
applicative feeds passive, hence the instrument in (82b) can become the
subject of a passive verb, as shown in (83).
(83)
N d o do i-ku-yend-er-edw-a.
stick s p -pres-walk-with-pass-asp
'The stick is being walked with.'
As in the Madurese example in (79), the Chichewa verb in (83) bears both
t he applicative an d
t h e p a s sive m o r p h emes .
M or p h e m e o r d e r i n g
s uggests that applicative has applied first, promoting th e i n strument t o

object. From t his position, passive may further p r om ote the instrument
to subject.'" Here I speculate that applicative in Chichewa is syntactic and
involves movement. F r o m th e available data, it appears that applicative
applies freely to different types of instruments (see example (82b) above).
Summing up, Malagasy is not unusual in allowing certain PP adjuncts
to appear as DP arguments. The main difference between Malagasy and
the languages mentioned in t hi s sub-section is that M a l agasy does not
mark t h es e
a l t ernations m o r p h o logically .
In this w a y , M al a g asy
'" As noted by Marantz (1993), however, it is not always the instrument that is promoted via
passive in these contexts. For my purposes, it is crucial that in order for the instrument to be
promoted to subject, it must first be promoted to object. W h ether further promotion occurs is
irrelevant. Although all speakers of Malagasy allow antsy 'knife' to be an argument of mandidy
'cut', not all allow the passive.
70
Chapter 2
r esembles English:
t h e l o cative alternation is no t a ssociated wit h a n y
particular morphology .
I do n o t c o n sider th e p r esence or absence of
morphemes to be an issue. As suggested by M arantz (1993), English and

M alagasy may s i m pl y h a v e n u l l m o r p h emes t ha t c o r respond t o t h e
applicative m a r k er s
i n ot h e r l a n g u ages ( p e rhaps i n v 2 ) .
M ore
i mportantly, h ow ever, although all t h ese languages appear to u s e t h e

same construction ("instrumental advancement", dative shift, applicative,
etc.), the actual derivations may be different. I h ave suggested that some
are cases of movement and others are base generated structures.
T he data in this section have provided fu r ther evidence in favour o f
the proposed analysis. I n
p a r ticular, I have discussed Case assignment,
constituency and the difference between base generation and movement.

I now turn to aspectual properties of alternating verbs.
6~ As ect
Much work on locative alternation verbs has focussed on their aspectual
properties (Anderson (1971); Rappaport and Levin (1988); Dowty (1991);

Hoekstra and Mulder (1991); Tenny (1994); Jackendoff (1996)). It has been
noted that when the goal is the direct object, it is interpreted as "w h olly
affected".
(84)
a.
Hann a h loaded the apples onto the wagon.
b.
Hann ah loaded the wagon with the apples.
I n (84a), the wagon may o r m a y n o t b e c o m pletely full . O n t h e o t h e r

hand, (84b) implies that the wagon is full. There is some debate, however,
over the status of t h e m a t erial theme (the apples). Tenny (1994) and
Dowty (1991) claim that i n
b o t h e x amples in (84) the d i rect argument
delimits the event (Tenny's "measuring out" an d D o w t y' s " I n cremental

Theme"). In (84a), it is the set of apples as they get used up that measures
o ut the progression of l o ading .
I n ( 8 4 b), on t h e o t her h a nd, i t i s t h e
wagon as it gradually fills that measures out the event.

F urthermore, b ot h
T e nn y a n d D o w t y a r g u e t h a t i t i s t h e d i r e ct
a rgument that determines the telicity of th e V P. "
I n o t h e r w o r d s, t h e
Jackendoff (1996) disagrees with this conclusion, but I set aside his criticisms as they ate
largely tangential to the Malagasy facts. Dowty (1991) provides a careful discussion of telicity
with consideration of the pragmatic effects of different verbs.
71
Theme Topic
direct argument controls the acceptability of the temporal adverbial in the
e xample below .
Te m p o r a l m o d i f i ers i n dicate telicity; 'i n a n h o u r ' i s
possible with telic events and 'for an h o ur' i s possible with atelic events
(Dowty (1979)).
( 85) a .
John sprayed subway cars with this can of paint *in/for an hour.
b.
John sprayed this subway car with paint in/*for an hour.
C.
John sprayed this can of paint onto subway cars in/?for an hour.
d.
John sprayed paint onto this subway car *in/for an hour.
If the direct argument is definite (as in (85b,c)), the predicate is telic and
compatible with 'in an hour' and not 'for an hour'."
O n t h e other hand, if
the direct argument is a bare plural or a mass noun, the predicate is atelic
and only the durative adverbial 'for an hour' is acceptable. By definition,
o nly incremental themes determine telicity; the d i rect argument i n
the
locative alternation is therefore an incremental theme.

As we w i l l s ee, similar r esults apparently o b t ain f o r t h e l o c a tive
alternation in K i m aragang Dusun. I n
M a l agasy, however, the data are
less clear and require careful discussion. I will show that [Spec, v2P] is not
a n affected object position pe r
se . In s t e ad, th e a f fected argument i s
d etermined by th e r oot t hat u n derlies the verb."

change in w or d
In o t h e r w o r d s, t h e
o r d e r i n t h e l o cative alternation d oes not c h ange th e
affectedness relations.
I t h e r efore d i sagree with M a r antz (1993), who
claims that [Spec, v2P] is an affected object position.

6.1 Kimaragang Dusun
In Kimaragang Dusun there is ev idence for th e d e r ived object having
specific aspectual import. "
A s d e s c ribed b y K r o e ger ( 1 990), di fferent
elements may surface in the derived object position, where they receive
a n "affected" interpretation.
descriptive label fo r
K r o e ger u ses the term " U n d e r goer" a s a
t h i s p o sition .
Int e r estingly, v e r bal m o r p h o l o gy
encodes the thematic relation of the "Undergoer".

" A s noted by Dowty (1991), the activity reading is marginally possible in (85c).
In my discussion of the Malagasy data, I will use the term "affected object" to refer to the
argument that gets "used up" by the action of the verb. See below for more discussion of aspect in
Malagasy.
Kroeger does point out that affectedness in Kimaragang Dusun cannot be defined in terms of
change of state or telicity.
72
Chapter 2
( 86) a .
g-po-suwang oku h
ditih
sada s i d pata'an.
AV-po-enter 1 sg.nom t h i s(acc)
f i s h loc basket
'I will put this fish in the basket.'

M onuwang
okuh
do pat a ' a n
AV-poN-enter lsg.nom a cc basket
do sada.
acc fish
'I will fill the basket with fish.'
Just as in the English translations, either the theme (sada 'fish') or the goal
( pata'an 'basket') can surface as the Undergoer, adjacent to the verb. T h e
Undergoer is understood as being "affected": i n
( 8 6b), for example, the
basket must be completely full. The difference in interpretation is marked

not only by word order, but also by the morphology on the verb: po- vs.
po
W he n t h e endpoint of the action is the Undergoer, the verb t akes
p o
W hen s o m e t hing o t her t h a n th e e n d p oint i s t h e U n d e r goer ( a
material theme in (86a)), the prefix is po-.

Kroeger mentions a n u m ber o f v e rb s w h ich allow t h i s alternation,
including 'give', 'throw' and 'split'. I n th e latter case, the Undergoer can
b e an instrument, similar to th e in strument advancement p r oposed f o r
Malagasy. For instruments, the reading is one of adversely affected; the
implication is that the action will be harmful to the instrument. H ence, the
instrument i s t h e p r i m ar y a f f ected object i n t h e U n d e r goer p o sition.
Kimaragang Dusun therefore resembles English:
t h e d i r ect argument,
whatever its role, is the affected argument.
6.2 Malagasy
In contrast to English and Kimaragang, the di fferent positions of DPs in
t he Malagasy alternations do not affect the affectedness interpretation.
will show that in the locative alternation only one argument (the goal) is
interpreted as the affected object. W it h d ative shift verbs, however, both
the theme and th e g oal ar e a f fected.
I w i l l a r g u e t hat t hi s d i f ference
b etween the verbs stems fro m i n d e pendent p r operties of th e r o ot s o f

these verbs.
I n o r de r
t o te s t f o r aff e c tedness an d t e l i city, w e ne e d s o m e
background on aspect in Malagasy. A s discussed in detail by Phillips (in
73
Theme Topic
press), most verbs in Malagasy are non-telic: the end result is implied but
n ot entailed.
I n o t h e r w o r d s , a l t h ough ( 87a) i s m os t o f te n u se d i n
situations where the dog is in fact captured by th e child, it is possible to

negate this implication, as in (87b).
( 87) a .
Nisambotra
n y alika ny zaza...
pst.AT.catch det dog det child

'The child caught the dog...'
... nefa faingana l o atra i lay alika.

b ut quic k
too
def d o g
'... but the dog was too quick.'

In order to m ake the v erb t elic, a different active prefix i s u sed:
With aha-, the end result is entailed.
a h a-.
H e n ce (88a) cannot felicitously be
followed by (88b).
( 88) a .
N ahasambotra
ny alika ny zaza...
pst.aha.catch det dog det child
'The child caught the dog...'
b.
.. . nefa faingana l o atra i lay alika.

b ut quic k
too
def d o g
'... but the dog was too quick.'

T he above data indicate that i n o r de r t o i n v e stigate telicity, i t i s f i r s t
necessary to use the prefix aha-. In other words, it is not the nature of the
object DP that d etermines telicity i n M a l agasy, in c o n trast t o E n g lish.
I nstead, it is th e v e rbal p r efix w h ich f o rces a t elic r eading, similar t o
perfective verbal affixes in Slavic languages (see Wierzbicka (1968) on
Polish). How, then, to test for affected objects? I will refer to the element
that undergoes the change of state entailed by th e v erb as the affected
o bject. In other w o r ds, I am u sing the term " a f fected object" in a v e r y
narrow sense. Due to the connection between telicity and affectedness in
74
Chapter 2
Malagasy, in what follows, I use telic verbs to test for affected objects."
6.2.1
Lo cative alternation
L et us now l ook at th e l ocative alternation in M a l agasy. W h e t her t h e

material theme appears in the direct object position (89) or in a PP (90), the
t he w a t e r n e e d n o t b e "wholly
i nterpretation r emains u n changed:

affected" by the action of filling.
( 89) a .
Nahafeno ny rano tao anatin'ny
tavoahangy ny sinibe
pst.aha.full det water pst.there in.gen.det
bottle
det pit c h er
Rabe ...
Rabe
'Rabe filled the water in the bottle into the pitcher.'
... nefa mbola misy

b ut still
A T .h a v e
rano t avela ao
w at e r left t h e r e
anatin'ilay tavoahangy.
in.gen.def bottle
'... but there is still water left in the bottle.'
( 90) a .
Nahafeno ny sinibe tamin'ny rano
tao a natin'ny
pst.aha.full det pitcher pst.P.gen.det water pst.there in.gen.det

tavoahangy Rabe...
bottle
Rabe
'Rabe filled the pitcher with water in the bottle.'
" Since nhn- verbs are unambiguously telic, they are compatible only with the equivalent of 'in
an hour', independent of the quantitative nature of the NPs.
(i) a. Nahafeno rano
n y tavoahangy tao anatin'iray ora
R abe.
pst.aha.full w a te r
de t b ottle
p st.there in.gen.one h ou r
Rabe
'Rabe filled water into the bottle in one hour.'
b. *Nahafeno rano ny tavoahangy nandritran'iray ora Rabe.
pst.aha.full w a ter det bottle
dur in g . gen.one hour R abe
'Rabe filled water into the bottle for one hour.'
The judgements in (i) are not changed if the arguments appear in the NP-PP order. Therefore, the
temporal adverbial test is not applicable.
75
Theme Topic
... nefa mbola misy
b ut still
rano t avela ao
A T .h a v e
w at e r left t h e r e
anatin'ilay tavoahangy.
in.gen.def bottle
'... but there is still water left in the bottle.'
Since a change of state in the m aterial theme is not entailed, I conclude

t hat material themes are not affected, whatever their syntactic position. I n
o ther w o r ds,
[ Spec, v2P] ( t h e p o s i tion f o r m a t e r ial t h e m es) i s n o t
associated with affectedness.

O n the other h a nd, th e g oal i s i n t erpreted as affected, either a s a
secondary object (91) or direct object (92).

(91)
a.
Nahaf e n o r a n o n y tavoahangy Rabe ...

pst.aha.full water det bottle
Rabe
'Rabe filled water into the bottle.'
b. ... nefambola misy
but still
toe r ana azo anasivana rano
AT . h ave p l ac e a ble CT.put w a t er
ilay tayoahangy.
def bottle
'... but there is still room to put water into the bottle.'
(92) a.
N ah afeno
ny tavoahangy tamin'ny r a n o R a be ...
pst.aha.full
de tbottl e
pst. P .gen.det water R a be
'Rabe filled the bottle with water....'
b. ... nefambola misy
but still
toe r ana azo anasivana rano
AT . h ave p l ac e a ble CT.put w a t er
ilay tayoahangy.
def bottle
'... but there is still room to put water into the bottle.'
The above data indicate that the verb mahafeno 'fill' is indeed telic, but that
only the end result of the bottle being full is entailed. The results in (89)(92) are the same with all the locative alternation verbs (e.g. mahatototra
76
Chapter 2
'fill', maharakotra'cover', etc. )."
Summing up, th e l ocative alternation in
M a l agasy is not associated
with any changes in meaning that have been n oted fo r

English and K i m aragang Du sun .
l a n guages like
O n l y t h e g oa l i s a n a f f ected object,
whether or not it is adjacent to the verb .
I n o w t u r n t o d a t i ves, which
differ from the locative alternation.

6.2.2
Da t ive shi
ft
Let us now consider the other large class of verbs that have two passives:
dative verbs. The data in (93) indicate that for these verbs, the theme is
interpreted as an affected object."
( 93) a .
Naharoso
ny v a r y tao
anatin'ny vilia
pst.aha.serve det rice pst.there i n .gen.det plate
hoan'ny
to.gen.det
ankizy ny mpiasa...
child d et worker
'The workers served the rice in the dish to the children...'
b.
... n e f a mbola misy

but still
vary
AT . have r i c e
tav e l a.
lef t
'... but there is still some rice left over.'
Similarly, the goal is affected, whether realized as a DP or PP.
( 94) a .
Naharoso
(ny) vary ny ankizy
n y m piasa...
pst.aha.serve (det) rice det child

det w o rker
'The workers served the children rice...'
b.
.. . nefa tsy ampy i l a y yary.

but neg enough def rice
'... the rice wasn't sufficient.'
Instrumental advancement verbs also pattern this way. T his is less surprising as one is hard
pressed to imagine just how a n i n strument could get "used up" t o m easure out the event.
Malagasy thus does not have the "adversely affected" reading associated with advanced instruments
in Kimaragang Dusun.
" (93b), (94b) and (95b) all remain impossible if nefa 'but' is replaced by kn 'and'.
77
Theme Topic
(95)
a.
Nahar o s o
(ny) v ary hoan'ny ankizy ny mpiasa...
pst.aha.serve (det) rice to.gen.det child det worker

'The workers served the rice to the children...'
b.
.. . nefa tsy ampy i l a y yary.

but neg enough def rice
'... the rice wasn't sufficient.'
For dative verbs, both the theme and the goal are affected arguments.
In section 6.2.1, I showed that with locative alternation verbs, there is
only one affected object. The above data indicate that with dative verbs,
there are two.
(96)
a f f ected object
locatum
goal
locative alternation
no
yes
dative shift
yes
yes
What might account for this difference? In w h a t f o l l ows, I argue that it

stems from differences between the roots that these verbs are built from.
6.2.3 On the importancef oroots
W ith very fe w
e x ceptions, all active verbs in M a l agasy are built f r o m
roots. These roots are either adjectival or nominal in nature and often can
be the matrix predicate of a
a rguments I
c l ause. I n t e restingly, th e split i n a f fected
n o t e d b e t w een l o c ative a l t ernation an d d a t i v e s h i f t i s
paralleled by another difference at the level of th e r o ot .
I n p a r t i cular, I
will show that the element that can be the external argument of the root is
interpreted as the affected object of the aha verb.
For locative alternation verbs, o nl y
argument of the root.
( 97) a .
Feno (rano) ny sinibe.

full (water) det pitcher
'The pitcher is full (of water).'
78
t h e g o a l m a y b e t h e e x t ernal
Chapter 2
b. *
Feno (ny sinibe) ny rano.

full (det pitcher) det water
Fatatra (vary) ny kitapo.
( 98) a .
stuffed (rice) det bag

'The bag is stuffed (with rice).'
b.
Fatatra (ny kitapo) ny yary.

stuffed (det bag) det rice
As shown by the ungrammaticality of (97b) and (98b), the material theme

cannot be the external argument of the root.
I n contrast, dative verbs allow e i ther th e t h eme o r
external argument of the root.
t h e g oal a s t h e
(As mentioned earlier, with m ost d ative
verbs the goal is optional and the theme is obligatory.)
( 99) a .
Roso (ny ankizy) ny yary.

served (det child) det rice
'The rice is served (to the children).'
Roso vary
n y a nkizy.
served rice det child

'The children are served rice.'
(lOO) a.
Toro
(ny mpandeha) ny lalana.
pointed-out (det traveller) det road

'The road is pointed out (to the travellers).'
Toro
lala n a ny mpandeha.
pointed-out road
d e t traveller
'The travellers are pointed out the road.'

Thus we have the following distinction between locative alternation roots
and dative roots.
79
Theme Topic
(101) external argument of root
Iocatum
goal
locative alternation
no
yes
dative shift
yes
yes
Comparing the table in (101) with the one in (96), we see that there is a
correlation between the external argument of th e r oot and th e affected
object. I therefore suggest that it is the external argument of the root that
u ndergoes the change of state in an telic verb."
I n o t h e r w o r ds, at t h e
l evel of t h e r o o t a f f ectedness relations ar e d e t ermined an d a r e n o t

changed b y
t h e di f f e r ent p o s i tions w h e r e e l e m ents a r e g e n erated.
Clearly, however, the change of state reading for the affected object is not
a lways present, as in active verbs m entioned at t h e b e ginning o f
section.
this
D i f f e rent verbal affixes (e.g. aha) are necessary to r e alize th e
change of state encoded by the root.

6.2.4
Fu r t her data
Above, I have claimed that the resulting end state is encoded lexically by
the root.
F o r l o cative alternation verbs, it is the goal that u n dergoes a
change of state. For dative verbs, it is both the theme and the goal. Is this
d ifference due to some underlying semantic distinction between the t w o
types of verb? I n o t her w o r ds, is there something inherent about dative
verbs as a
c l ass t ha t a l l ow s t h e m t o ha v e t w o di f f e r en t e x t ernal
arguments? I suggest that this is not the case. Instead, the only test for
affected object is the one given above: w h i ch element can appear as the
e xternal a r g u m ent
o f t h e ro o t .
Ev i d e n c e c o me s f r o m a l oc a t i v e
alternation verb mahafafy'sow', wh ich patterns with dative verbs for both
affectedness and at the root level.
A s shown by t h e examples below, either the material theme o r
the
goal may be the external argument of the root fafy 'sown'.
'" It is not clear at what level of representation affectedness is encoded. Moreover, at the level of
argument structure, certain roots (e.g. datives) appear to have two external arguments. Under
standard assumptions (e.g. Williams (1981)), only a single external argument is possible for a
lexical entry. Positing two different lexical entries for these roots opens up the question of why
the telic verb has two affected arguments simultaneously.
80
Chapter 2
(102) a.
Fafy n y yoa.
sown d e t seed
'The seeds are sown.'

b.
Fafy
ny s a ha.
sown d e t field
'The field is sown.'

Following the reasoning in the preceding sections, I would expect both the
material theme and the goal to be affected objects. This is in fact the case.
When the verb bears the aha active prefix, both t he m a terial theme and
the goal are interpreted as reaching the end state described by the verb.
Let us first consider the goal. The results are not so surprising as they
pattern w it h v e rb s l i k e ma hfaeno ' fill'. In oth e r w o r d s , t h e g o a l i s
interpreted as affected by the action of sowing.
(103) a.
Nahafafy n y saha tamin'ny v o a tao

pst.aha.sow d e t field pst.P.gen.det seed pst.there
anatin'ny kitapo ny mpamboly...
in.gen.detbag det farmer
'The farmer sowed the field with the seeds in the bag...'
b.
... n e f a tsy ampy
ila y y oa.
but neg enough def seed

'...but the seeds were not sufficient.'
(104) a.
N ahafafy
ny vo a
tao
pst.aha.sow d e t seed
anatin'ny kitapo ny saha
pst.there i n .gen.det bag
ny mpamboly...
det farmer
'The farmer sowed the seeds in the bag in the field...'
b.
... n e f a tsy ampy

but neg enough
ila y y oa.
d e f seed
'...but the seeds were not sufficient.'
81
de t f i eld
Theme Topic
Turning now t o
t h e m aterial theme, we see that it m ust be c om pletely
used up by the action of sowing when it is realized as a DP. H ence it is an

affected object. This is a surprising result when maha
fafy is compared with
other locative alternation verbs such as mahafeno 'fill', where the material
theme is never an affected object (see examples (93) and (94) in section
6.2.1).
(ioS) a.
N ahafafy
ny vo a
pst.aha.sow d e t seed
anatin'ny kitapo ny saha
tao
pst.there i n .gen.det bag
de t f i eld
ny mpamboly...
det farmer
'The farmer sowed the seeds in the bag in the field...'

b.
.. . nefa mbola misy

but still
voa ta v ela ao
anatin'ilay kitapo.
AT . h ave s e ed left t h er e i n . gen.def bag
'...but there are still seeds left in the bag.'

When the material theme surfaces as a PP, however, it is not affected, as
shown by the data in (106).

(l06) a.
Nahafafy n y saha tamin'ny v o a tao

pst.aha.sow d e t field pst.P.gen.det seed pst.there
anatin'ny kitapo ny mpamboly...
in.gen.detbag det farmer
'The farmer sowed the field with the seeds in the bag...'
... nefa mbola misy

but still
voa
tave l a a o a natin'ilay kitapo.
AT . h ave s e e d le f t
the r e i n . gen.def bag
'...but there are still seeds left in the bag.'

To account for the difference between (106) and (105), I suggest that only
arguments can be affected. In (106), the material theme is in an adjunct
p osition, as argued i n
s e ction 2 .3 . T h e r e fore, w e d o n o t e x p ect t h e
material theme to act as an affected object in this position.
maha
fafy 'sow' p r o v i d es fur ther e vi dence in s up port o f
I m p o r t antly,
t h e c o r r e lation
between the external argument of a root and the affected argument of the
82
Chapter 2
telic verb.
A range of
d at a i n dicate that th e D P t h a t a p p ears as th e e x ternal
argument to a root is the argument that u n d ergoes the change of state.

This change of state is realized when the telic prefix aha is used. There are
n o i n t eractions w i t h
t h e q u a n t i tative n a t ur e o f t he D P s in the s e
alternations. M o r e over, w o r d
o r d e r d oes no t c h ange the affectedness
relations encoded by th e r o ot .
T h e o nl y cases where word o r der d o es
have an effect is when an element (e.g. a material theme) is realized in an

adjunct position. A s an adjunct, a material theme can not be i n terpreted
as affected, not a surprising result. I n su m, [Spec, v2P] is not an affected
argument position. I n o t her w o r ds, affectedness in Malagasy is lexically
rather than structurally d etermined.
F u r t h er, th e d a t a i n t h i s s ection
i llustrate the importance of roots in th e syntax of M a l agasy.
The root
encodes information that is only realized when certain affixes are added.'"
7
C o n c l u sion
T he focus of the present chapter has been passive constructions. I
have
argued that d i f ferent passive affixes target distinct structural positions.

The passive d at a
h av e p r o v i de d e v i d ence i n f a v ou r o f a st r u c t u ral
position, [Spec, v2P], between the positions for agent and other i n ternal
a rguments of the verb .
g enerated i n
I n M a l a g asy, only c ertain arguments may b e
[ S pec, v2P]. D r a w i n g o n a r a n g e o f d a t a f r o m t h r e e
argument verbs, I have shown that this position is for locata or displaced
themes.
In the following subsections, I discuss possible extensions of this line of
research.
7.1 Passive
The above analysis draws on a classical GB view of passive: unavailability
of Case forces DP movement.
M y a n a lysis crucially requires that o n l y
t hose DPs that are not m a rked fo r C ase can raise to subject. I n
other
words, the head that drives movement to subject position (e.g. T ) cannot
attract any Case-marked DPs in the structure.'" I n
a n a ctive clause, the
internal arguments of a t r ansitive verb are assigned Case; the external

argument must raise to get nom inative Case. I n
p a ssives, the different
" Phillips (in press) comes to a similar conclusion in her study of the nhn affix.
'" See chapter 3, section 5 for some discussion of the features that motivate subject movement.
83
Theme Topic
affixes signal th e
l oss o f C a se t o d i f f erent st ructural p o sitions. F or
example, in the -V n a p a ssive of a d a t i ve v e rb, a ccusative Case is n o t

available in V P
an d t h e g oal r aises over th e C ase-marked t h eme an d
agent.
While in English, passive can be used to promote v arious arguments,
M alagasy signals the di fferent arguments w it h d i f f erent affixes. S o m e
researchers take this to indicate that voice in Austronesian languages is a
type of " t h eta agreement".
D e s p ite the attractiveness of this approach,
we have already seen that it runs into problems with M alagasy passives.
The -Vna passive, for example, promotes themes and goals. Thus there is
n o one-to-one mapping between theta role labels and voice affixes." T h e
a nalysis proposed in t hi s chapter posits a basic split between th e t w o
p assive affixes. O n on e h and, the a- passive promotes arguments in a
particular position, [Spec, v2P]. O n
p romotes arguments in a
t h e o t he r h a nd, th e -V n a pa s sive
p a r t i cular d o m ain, th e l o w e r V P .
r esearch will d e termine w h e ther b ot h
Fu r t h e r
t y p e s o f p a ssive (position an d
d omain) are a v ailable cross-linguistically an d
t h e c o n straints o n t h e i r
distribution.
7.2 jSpec, v2P]

Matsuoka (1999) proposes for a very similar analysis to the one presented
in this chapter for certain dative arguments in Japanese. He argues for a
b ase generated position for a subset of d ative-marked arguments.
For
M atsuoka as well as for me, t his position is between the agent and t h e
theme. He discusses the following alternations.
(107) a.
John-ga h o n- o
Mary- n i wasas-ta.
John-nom book-acc M a r y-dat pass-pst

'John passed a book to Mary.'
Mary-ga J ohn-ni
pe n k i- o ab i s e-ta.
Mary-nom John-dat p aint-acc p our-pst

'Mary poured paint over John.'
Theta agreement will be discussed again in the next chapter, in the context of circumstantial
topic.
84
Chapter 2
M atsuoka p r o p oses t ha t
w i t h 'p a s s'-type v e r b s ( 1 0 7a), t h e d a t i v e
argument is a goal and generated below the accusative theme. For 'pour'type verbs (107b), on the other hand, the dative argument is generated
between the agent and the theme (he calls this dative a "possessor").
In support of his analysis, Matsuoka shows that with 'pass'-type verbs,
only the theme can be the subject in an inchoative construction.
(108) a.
Hon-ga
Ma r y -ni wasas-ta.
book-nom Mary-dat pass-pst

'The book passed to Mary.'
b. *
Mary-ga h o n-o
watar-ta.
Mary-nom book-acc pass-pst

Interestingly, with 'pour'-type verbs, only the dative argument can be the
subject of the inchoative verb.
(109) a.
John-ga p e nki- o
ab i - t a.
John-nom paint-acc pour-pst

'John got paint poured over him.'
b. *
Penki-ga John-ni
ab i - t a.
paint-nom John-dat
p o u r-pst
Matsuoka follows Baker (1996a) and argues that t hi s
d i f f erence stems
from the d i fferent structural positions of th e d a t ive arguments in each
type.
F urther crucial data in s u p port o f
M a t s u oka's analysis come f r o m
quantifier scope interactions. U n f o r tunately, the parallel Malagasy data

a re far from clear.
F o r i n d ependent reasons, the only w a y t o t e s t f o r
scope asymmetries is with bound p r o n ouns. A s
s h ow n i n ( 110a), a DP
goal can bind a DP theme. On the other hand, (110b) indicates that a goal
cannot bind out of a PP.
85
Theme Topic
(O) a.
N anolotra
ny b o k i ny
ny mpianatra r e hetra i Ketaka.
p st.AT.offer d e t book.3(gen) d e t student

'Ketaka offered his book to every student.'
b. ?? Nanolotra ny bokiny
ev e r y
Ket a k a
h o an'ny mpianatra
rehetra i Ketaka.
p st.AT.offer d e t book.3(gen) for.det student e v er y

'Ketaka offered his book to every student.'
K eta k a
Further data, however, cast doubt on t hi s conclusion. I n stead, there is

something "funny" about PPs in these structures that blocks binding both
i nto and out o f t h em .
( 1 1 1a) illustrates binding of a D P g oal b y a D P
theme. Binding is not possible when the goal is a PP, as shown by (111b).
(111) a.
N anolotra
ny b o k y rehetra n y t ompony
i K eta k a .
pst.AT.offer detbook every det owner.3(gen) Ketaka

'Ketakaoffered every book to its owner.'
b. ?? Nanolotra
ny b o k y rehetra h o a n'n y t o m p ony
pst.AT.offer d e t book every for . det owner.3(gen)

'Ketakaoffered every book to its owner.'
i Ketaka.
Ketaka
Moreover, there is much variation in judgements both wi thin and across

speakers. I t h e refore leave binding for f u r t her research. N e v e rtheless,
looking at parallels between the Malagasy and Japanese data, I conclude
that U n i v ersal G r a mma r
a l l ow s f o r a sp e c ial p o s i tion, [Spec, v2P].
Languages differ as to which elements may be generated in this position.

In this chapter, I argue that in M a l agasy [Spec, v2P] is restricted to
certain elements, roughly di splaced themes or locata. I n
position hosts dative "possessors". A t
J a panese, this
t h e same time, [Spec, v2P] is not
simply an open position. A s discussed above, base generation im plies a

restricted phenomenon.
I n t e r estingly, since[Spec, v2P] is s t ructurally
superior to themes, this position can obscure theta-hierarchy effects. For

example, if goals can be base generated in [Spec, v2P], they can appear to
act "higher" than themes. Most researchers acknowledge the difficulty of
determining th e
h i e r archy o f V P - i n t ernal a r g u ments. F o r e x a m p l e,
although Collins and T h r ainsson (1996) assume that goals c-command
86
Chapter 2
themes in Icelandic, they point out that this could be the result of syntactic
movement. I leave it for future work t o d etermine just which arguments
may appear in [Spec, v2P] cross-linguistically.
7.3 Lexical semantics
In section 4.2, I briefly mentioned that a lexical semantic study w il l shed
further light on the alternations discussed in this chapter. The connection
between argument structure and syntactic structure that I am arguing for
here is similar t o
t h e o n e p r e sented i n R a p p aport an d L e v i n ( 1 988).
According to m y
a n a lysis, the alternating verbs h ave tw o d i s tinct but
related underlying lexical semantic representations. This lexical conceptual

structure (LCS) then maps to predicate argument structures (PAS). There
a re linking r u les that d e termine w h ich v a r i ables are l i n ked t o w h i c h
positions. I believe that the Malagasy verbs discussed in this chapter all
s hare a similar LCS, probably encoding a change of location.
In other
words, all these verbs have the substructure in (112)."

(112) .... [x come to be at LOCATION] ...
A l i n k in g
r u l e t h e n s t a tes t ha t v e rb s w i t h t h i s c h a ng e o f l o c a t ion
component allow the "locata" (x) to be realized as a direct DP argument.

P assive is sensitive to the syntactic structure that is p r ojected from
the
LCS. The locative alternation and instrumental advancement verbs also

h ave a d i f f erent L CS, w h er e t h e
m a t erial t h em e an d i n s t r ument a r e
represented in a MEANS clause and are syntactically realized as adjuncts.

The details of this approach I leave for further research.
7.4 TT vs. CT
I n all the passive constructions discussed above, I have show n
a rguments
m o v e to
sub j e c t p o s i t io n f o r nom i n a t i ve .
that DP
A p par e n t
c ounterexamples, such as the passives of instrumentals, were shown t o

i nvolve base generation of the "adjunct" in an argument position. W e a r e
now in a position to contrast passive with CT, which does not promote DP
arguments and instead is used for adjuncts, PPs and other non-structurally
Case marked DPs. CT is therefore the topic of the next chapter.
" Or, as suggested by Pearson (1998a):
(i)
... [ x GO TO yi...
87
Chapter 3: Circumstantial Topic

1 I n t roduction
In chapter 1, I briefly discussed the notion of voice alternations. I p o i nted
o ut that although m an y l a n guages show a t w o - wa y a l t ernation
( e .g.
active-passive), Malagasy enjoys a three-way voice system: active-passivecircumstantial. In chapter 2, I discussed the nature of passive and briefly
contrasted it with the circumstantial (CT). I n t h i s chapter, I address the
s yntax o f
t h e c i r c umstantial v o i ce . W h a t i s of i nt e r est i s t h a t i n
circumstantial clauses, a wide range of elements appear in subject position.

It is the goal of this chapter to characterize the nature of t hese elements
and hence to provide a unified analysis of CT.
To begin, consider the standard voice alternation given in chapter 1,
repeated below.'
(I)
a.
Nanapaka
ity h a zo ity ta m i n 'n y
ants y i Sahondra.
pst.AT.cut
Notapahin'i
Sa h ondra tamin'ny
ants yitg hazo ity.
pst.TT.cut.gen.Sahondra pst.P.gen.det k n i f e this tree this

i t y hazo ity ny antsy.
pst.CT.cut.gen.Sahondra this tree this det knife

I n the previous chapter, we saw that T h eme Topic (TT) covers a w i d e
r ange of
p a ssive-like c onstructions. N e v e r t heless, I a r g ue d t h a t a l l
passives in Malagasy involve p r o m otion t o

never a n
a d j u nct, n e ve r a P P .
s u b ject of a D P a r g u m ent,
W e al s o s a w th a t e l e m ents (e.g.
instruments ) that can be realized either as DPs or PPs are promoted t o

subject either with TT or Circumstantial Topic (CT). I showed that in these
cases, TT promotes DPs, while CT p r o m otes PPs. I
In this chapter, I will gloss the various passive forms as TT.
d i d n o t , h o w e v er,
Chapter 3
offer a analysis of CT to capture this observation.
What is CT? It is difficult to arrive at a precise characterization of the
elements that are promoted to subject with CT.
S om e researchers focus
on the function of the CT subject. For example, Rajemisa-Raolison (1966)

claims that some "circumstance" of the action becomes the grammatical
subject (hence the l abel " C i r cumstantial T opic") .
A lo n g s i m i lar l i nes,
under a t h eta-agreement analysis of v o ice (e.g. Sityar (i n p r e ss)), CT

would be adjunct agreement. Others take the underlying form of the CT
subject to be crucial. K e enan (1976) suggests that it is an element t h at
would b e
m a r ke d w i t h o b l i qu e case that i s p r o m o ted .
K eenan's i dea, G u i l foyle, H u n g
For m a l i z i n g
a n d T r a v i s ( 1992) p r o pose t h a t C T
involves P-incorporation. These descriptions fall into two main categories,

f unction: CT promotes adjuncts (Rajemisa-Raolison and Sityar); form: C T
promotes objects of prepositions (Keenan and GHT).
Instead of treating CT as tied directly to prepositions or theta-roles, I
w ill argue that a C T
c l ause in v olves th e p r o m o tion t o s u b j ect o f a n
element that is not a structurally Case-marked DP. For the most part, this
will target PPs, hence the connection with prepositions. It will also include
adjuncts in general, hence the appearance of theta-agreement. U n l ike the
passive affixes discussed in the previous chapter, CT is sensitive neither to
position (cf. a- passive) nor to domain (cf. -Vna p assive). In f a c t, we w i l l
see that CT can promote an element from almost any structural position
in the clause.'
' The one position that is never associated with CT is the agent. This gap remains unexplained.
89
Circumstantial Topic
(2)
DP
vl'
~g
0
CT
a- passive < l o catum>
0
-Vna passive
<th>
V'
DP/PP
<goal>
-Vna passive
Under this vi ew, C T i n
M a l a gasy is a k i n d o f " e l s ewhere" v o i c e an d
contrasts with languages such as Tagalog and Cebuano, where CT is subdivided into di fferent types (benefactive, instrumental, locative ) . I w i l l
briefly discuss the Tagalog voice system in section 7.
I will begin by looking at the nature of elements that are promoted to
s ubject with C T m o r p h o logy . S i n c e CT i s q u i t e d i f f erent f ro m v o i c e
a lternations in well-known l anguages such as English, in section 2 I w i l l
p rovide a w ide range of examples. C r u cial will be the question o f t h e
theta-role an d
c a t egory o f t h e e l e m e nt s t a r geted b y C T.
In the
subsequent sections, I will consider two possible accounts for CT. F i r st, I
will evaluate the connection between CT and adjuncts. The discussion will
touch o n
a d j u n cts i n M a l a g asy c l a use-structure a n d t e s t s f o r the
argument-adjunct distinction. D r a w ing on data from PP arguments and

partitive themes, I will show that CT promotes not only adjuncts, but also
certain arguments. I
t h e r efore r eject the adjunct-agreement approach.
Second, I consider the link between CT and prepositions. A l t h ough many

e xamples involve PPs, the data presented in section 2 i n d icate that n o t
only the objects of prepositions are promoted to subject with CT; instead,
a range o f
c a t egories ar e i n v o l v ed .
H en c e C T i s n o t pr e p o sition
i ncorporation. Section 5 is devoted to th e p r oposed analysis of CT .
In
section 6, I consider a special use of CT that falls outside most accounts of
90
Chapter 3
CT. I w ill show, however, that m y
c onstruction.
a n alysis easily captures this unusual
I n s e ction 7, I b r i e fly c onsider th e e q u i valent o f C T i n
Tagalog before concluding in section 8.

2 C T cl auses
L et us now turn to the distribution of C T clauses. A s m e ntioned at t h e
b eginning of t his chapter, in a C T
c l ause it i s o f ten a n a d j unct that i s
promoted to subject. I will now explore the range of elements that can be
promoted. We will see that although adjuncts are indeed prevalent as CT
subjects, certain arguments may b e
p r o m o ted t o s u bject w it h C T . A
secondary goal will be a discussion of the cleft. M any CT clauses have the
form of a cleft construction. I w ill show that the category of the promoted
element determines whether it can be a subject or must appear in a cleft.
In other words, restrictions on the subject position account for obligatory
clefts.
2.1 The core data
In his grammar of M alagasy, Rajemisa-Raolison (1966) lists the range of
elements that can be promoted via CT :
p l a ce, time, goal, cause, means,
manner, instrument, p rice, benefactive, locative, etc.
S o m e i l l u strative
examples appear in (3).

(3)
a.
i nstr u m e nt
Anapahany
b o z aka ny antsiny.
CT.cut.3(gen) grass det knife.3(gen)

'Her knife is used by her to cut grass.'
locative adverbial
Itoeranay
i ty trano
ity.
CT.live.lpl(gen) t h is house this

'This house is lived in by us.'
causal adverbial
[ Noho
ny a n dr o a l ina ] no odiako
haingana.
because det day

nig h t f o c CT.go-home.l sg(gen) quickly
'It's because it is getting late that I am going home quickly.'
91
temporal adverbial
[ Tamin'ny zoma ]
vary.
no nivi d i anako
pst.P.gen.det Friday foc pst.CT.buy. 1 sg(gen) rice

'It's on Friday that I bought rice.'
Yet not all of these CT clauses act the same. A s i n
t h e examples above,
some circumstantial clauses are clefts while others are not.' I n
f a ct, while
(3a,b) have the option of being expressed as a cleft, (3c,d) are examples of
obligatory clefts.' This difference is illustrated in (4).
(4)
a.
[ Ny antsiny ]
no a n apahany
boz a k a.
det knife.3(gen) foc CT.cut.3(gen) grass

'It's her knife that is used by her to cut grass.'
[ Ity trano ity ] no itoeranay.

this house
t h i s foc CT.live.1pl(gen)
'It is this house that is lived in by us.'
c.
Nodiako
haingana no ho
ny andro alina.
p st.CT.go-home. 1 sg(gen) quickly

be c ause d e t da y nig h t
'I went home quickly because it was getting late.'
d.
Nividianako
vary
pst.CT.buy.1 sg(gen) rice

'I bought rice on Friday.'
The grammaticality o f
tamin'ny zoma.
pst.P.gen.det Friday
( 3a,b) and (4a,b) shows that b ot h l o catives and
instruments can be either in the subject position subjects or in a CT cleft.
Crucially, the "gap" in the clause corresponds to the subject position. T h u s although these
elements cannot surface as clause-final subjects, they nevertheless have been promoted with CT.
See section 2.4 for more discussion of clefts.
On the surface, obligatory clefting is reminiscent of the fact that CT-like constructions in some
western Austronesian languages only surface in extraction contexts (e.g. Tagalog: Foley (1976)).
In other words, the equivalent of CT is not used in simple clauses, but only when "necessary" for
A-bar movement (e.g. relativization, wh-questions, etc.). I believe there is a difference, however.
In Malagasy, clefting is truly obligatory for PPs, while the markedness of CT-like voices in other
languages is not ungrammaticality. I n fact, Malagasy grammarians point out that CT verbs are
most often used as noun modifiers (Malzac (1960); Rahajarizafy (1960); Rajemisa-Raolison
(1966)). This appears to be the same pragmatic (not grammatical) restriction as in Tagalog.
92
Chapter 3
C ausal and temporal adverbials, however, cannot be s u bjects and a r e
therefore clefted, as illustrated by the contrast between (3c,d) and (4c,d).
With CT clauses, therefore, several questions arise. W h ich elements are
p romoted?
W h a t t h e t a r o l e s an d c a t egories ar e i n v o l v ed ?
W hi c h
elements may appear in the clause-final subject position and w h ich mu st

be clefted? T hi s section will be d evoted t o
a n sw ering t h ese questions.
Once these questions are addressed, we will be in a position to assess the

various analyses of CT that have been proposed.
2.2 Targets
If we take voice morphology as "targeting" elements to be promoted, we
can ask: w hat is the target of a particular voice? Fo r th e a- passive, for
example, I suggested in chapter 2 that the target is the element in [Spec,
v2Pj. For CT, we have just seen that the range is quite w i de, including
different adjunct-like roles (locative, temporal, instrumental, etc. ). These
adjuncts arguably occupy different positions in the clause: some adjoined
to VP, others to IP (see, for example, Jackendoff (1972); Travis (1988) on
a dverbs).
M o r e o v er, a s w e w i l l s e e i n t h e f o l l o w in g e x a m ples, t h e
category of these elements varies widely.

Quite commonly, th e el ement p r o m oted t o
s u bject in a C T c l a u se
originates as the object of a preposition. The (a) examples below illustrate

the base position of the element that is the subject of the CT clause in the
(b) counterparts. (5) is a locative and (6) is a goal.

Mitoetra [pp amin'ity
ATIive
t r an o i t y ] r y Ratsimba.
P.g e n .this house this det Ratsimba
'The Ratsimbas live in this house.'
I toeran-dry
Rat si m b a ity t r ano i t y .
CT.live.gen.det Ratsimba this house this

'This house is lived in by the Ratsimbas.'
(6)
a.
Nand r os o
var y [ pp hoan'ny v a hiny ] R a k oto.
p st.AT.serve ric e
for . g en.det guest
'Rakoto served rice to the guests.'
93
Rak o t o
b.
Nand r osoan-dRakoto
vary
ny y a hiny.
pst.CT.serve.gen.Rakoto rice det guest

'The guests were served rice by Rakoto.'
T he following i l l u strate categories other t h a n
PP : ad v e r b s ( o f ten o f
adjectival status), CPs, VPs and DPs.' (N ote that all of these examples are
clefted; I will address this issue directly below.)
(7)
a.
Miteny
ma f y i B ozy.
A T.speak hard
B ozy
'Bozy speaks loudly.'
[,d, Mafy ] no itenenan'i Bozy.

hard f o c CT.speak.gen.Bozy
'It's loudly that Bozy speaks.'
(8)
a.
Mianatra mafy aho
m ba hahazo
kar a m a
AT.study hard 1 sg(nom) opt fut.AT.get salary

'I'm studying hard to earn a big salary.'
[cp Mba hahazo karama be ] no ianarako
be.
big
mafy.
o pt fut.get
sa l ar y bi g
foc C T .study. 1 sg(gen) h a r d
'It's in order to earn a big salary that I'm studying hard.'
M itsangana mihinana akoho
Rab e .
A T.stand A T .eat
Ra b e
chi c ke n
'Rabe stands while eating chicken.'

[vp Mihinana akoho ] no itsanganan-dRabe.
AT.eat chicken
f o c CT.stand.gen.Rabe
'It's while eating chicken that Rabe stands.'
' In certain cases, it is difficult to determine the category of the clefted XP. I n (9b), for example,
the clefted XP could be a DP rather than a VP (zero nominals are very common in M alagasy).
Nevertheless, the examples in (7) (10) do illustrate a range of categories.
94
Chapter 3
T onga io m araina
i o izy .
arrived this morning t h is 3(nom)

'She arrived this morning.'
[DpIo maraina i o ] no nahatongavany.
this morning t his foc pst.AT.arrive.3(gen)
'It's this morning that she arrived.'
In all the above examples, although o f
v a r y in g c ategory, th e elements
p romoted t o s u bject correspond t o a d v erbial m o d i fiers, similar t o t h e

English translations. A t
f i r s t g l ance, therefore, th e a d j u nct-agreement
a nalysis of CT appears attractive. I
w i l l c o nsider this analysis in m o r e
detail in section 3 and show that it is untenable, however.
T h e d ata also
illustrate that CT does not target elements of a single category. This range
o f categories will become i m p o r tant i n
s ection 4 .
Ca t e g or y w i l l a l s o
d etermine w h ether an e l ement ma y b e
a s u b ject, a q u e stion w h i c h I
address in the following section.

B efore concluding t hi s section, h o w ever, I
p o int ou t a s o m ewhat
exceptional use of CT.

(11)
A n a san-dRakoto
amin'ny
s a vony ny loyia.
CT.wash.gen.Rakoto P.gen.det soap d e t d ishes

'Rakoto washes some of the dishes with the soap.'
(11) is exceptional in two respects. First, the theme of th e v erb (ny lovia
' the d i shes') a p pears i n s ubject p o sition, b u t t he ve r b be a r s C T
morphology, not TT .
S econd, this subject DP is interpreted as partitive
( 'some of the dishes'), a reading not normally available for subjects. F o r

this reason, I r e fer t o
e x amples like (11) as Exceptional Circumstantial
Topic Marking (ECTM). On e of the challenges will be to integrate ECTM

into an analysis of CT .
I n s e ction 6, I p r o v id e a t h o r ough a n alysis of
ECTM.
2.3 Subjects
We can now turn to the question of which elements appear in the clausefinal subject position. As mentioned in chapter 1, Malagasy subjects must
95
be specific DPs.' T h i s restriction accounts for th e r a nge o f p e r m issible
subjects in CT clauses. For example, PP subjects are not p e r m i tted, as
shown by the ungrammaticality of (12a).' This is also true when the verb
is CT, as in (12b). For the preposition to co-occur with CT, the PP must be
in a clefted position not the subject position, as in (12c). (See the examples
in (3c,d) and (4c,d) above for a similar contrast).
( 12)
a. *
Maloto eo ambonin'ny l
dirty
at a b atra.
t h er e u n der.gen.det table
'Under the table is dirty.'

b. *
Anapahany bozaka amin'ny antsiny.

CT.cut.3(gen) grass P.gen.det knife.3(gen)
[ Amin'ny antsiny ]
no ana p ahany
boz a k a.
P.gen.det knife.3(gen) f o c CT.cut.3(gen) grass

'It's with her knife that she cuts grass.'
Interestingly, both p r ice and temporal DPs are barred from t h e s u bject
position, but must be clefted (as DPs).
( 13)
a. *
Nividianany
hena yaiopolo.
pst.CT.buy.3(gen)
me a t 80
[ Valopolo ]
80
n o n i v idianany
hena.
foc pst.CT.buy.gen.3meat
'It was for 80 (ariary) that she bought meat.'
( 14)
a. *
Nividianany
hena o maly.
pst.CT.buy.3(gen) meat yesterday
' This raises the question of CP subjects. See section 2.5 for discussion.
' (12a) is grammatical if the PP is preceded by a determiner. In other words, the subject in (i) is a
zero nominalization, not a PP.
(i)
M alot o ny e o ambo n i n'ny latabatra.
dirty
det t here under.gen.det table
'Under the table is dirty.'
96
Chapter 3
[ Omaly ] no nividianany
hena.
yesterday foc pst.CT.buy.3(gen) meat

'It was yesterday that she bought meat.'
Extending work by
C i n que (1990); Rizzi (1990), I suggest that price and
t emporal ad verbials are n o n -referential.
S i n c e o nl y s p ecific DP s a r e
a llowed in subject position, (13a) and (14a) must be expressed with t h e

cleft constructions in ( 13b) and (14b).
S i m i lar considerations apply t o
manner adverbs, VPs and CPs, which all fail as specific DPs and are barred
f rom th e subject position (see examples (7)-(10) above, w h ich ar e
clefts). H e n ce th e i n d ependently m o t i v ated r estriction o n
all
t h e s u b ject
position easily accounts for certain patterns in CT clauses. As illustrated in

many examples above, w h e n a n
e l e m ent i s b a r red f r o m t h e s u b ject
position, it surfaces in a cleft. I therefore now turn to clefts.
2.4 Clefts
W e have seen that a range of elements are promoted to subject in C T
clauses: PPs, adverbs, CPs, VPs, DPs. Restrictions on the subject position,
h owever, make clefting obligatory in many cases. What is a cleft? On t h e
surface, clefts involve a displaced element: th e clefted XP is clause-initial,
followed by a focus particle no. The structure and meaning of clefts will be
d iscussed in detail in chapter 4.
F o r t h i s chapter, I s i m pl y a ssume th e
structure below.
(I>)
[ F o p X P i [cpOPi [ Ip "ec " ] ] ]
T he XP is the clefted element, which receives a focus interpretation.
In
chapter 4 I provide arguments that XP is in fact the main predicate and the
subject is a headless relative clause. The details of this structure are not
important t o
t h e c u r r en t d i s cussion, h o w ever .
W ha t i s i mp o r t a nt,
however, is the operator-variable relation.

Let us now consider the properties of the clefted element. I n g eneral,
if what is being clefted was originally a PP, when the verb is in CT, either a
DP or a P P cleft is p ossible (e.g. prices do no t
therefore will never surface as PPs).
97
t a k e p r e p ositions and
(16)
[ ( A m i n ')ny antsiny ]
no an a pahany
boz a k a.
(P.gen.)det knife.3(gen) foc CT.cut.3(gen) grass

'It's with her knife that she cuts grass.'
See below for some discussion of these two possibilities.

Complementizers, however, are never optional. For example, mba, an
optative m a r k er,
i s o b l i g atory .
c omplements (not D P) .
N ote th a t mb a c a n o n l y ta k e IP
Th u s ( 1 7a) c o ntrasts w i t h ( 1 7b), w h er e t h e
complement of mba is a DP.
( 17) a .
Naka
i ty boky ity
pst.AT.take
t h i s book this 1 sg(nom)
m ba h a mpianarako
opt
aho
ana o .
fut . C T.teach.l sg(gen) 2sg(nom)
'I took this book to teach you with.'
b.
Naka
i ty boky ity
aho
pst.AT.take
t h i s book this 1 sg(nom)

anao.
mba ny fampianarako
opt
de t n m.CT.teach. 1 sg(gen) 2sg(nom)
In (17b), I have used the f- nominalization (glossed nm) of the CT verb in

(17a) (Paul (1996b); Hanitriniaina an d
T r a vi s ( 1998)). T h e s e d e r i v ed
nominals are event nominals and hence in principle are compatible with
the meaning of mba. Th e u n g rammaticality of ( 17b) is therefore purely
due to selectional restrictions.
Supporting data come from a n other p repositional-like element, noho
'because'.
Unlike mb a, noho i s com p a t i b le b o t h w it h I P an d DP
complements.
( 18) a .
Tsy afaka handeha a n y
A mb o s itra i Koto
neg free f u t .AT.go there
A m b o sitra i Koto
noho [tp izy
mbola k e l y
because 3(nom) s t il l
ta o n a l o atra ].
sma l l y e a r too
'Koto cannot go to Ambositra because he is still too young.'
98
Chapter 3
b.
Nand o sitra izy
noho [Np ny tahony ].
pst.AT.flee 3(nom)
b e c ause d e t fear.3(gen)
'He fled because of his fear.'
When it t akes an IP, noho a cts like a c o m p l e mentizer a nd c a n not b e
omitted.
W h e n i t t a kes a DP, no ho p atterns wi th p r e p ositions and i s
optional in the cleft position. This difference is illustrated in (19).
( 19) a .
[ *(Noho) iz y
mbola k e l y
because 3(nom) s t il l
tao n a l o atra ]
sma l l y e a r too
no tsy
af a h a n'i Kot o
han d eh a a n y
A mbo s itra.
foc neg
C T . f ree.gen.Koto fut.AT.go t here A m b ositra
'It is because he is still too young that Koto can't go to Ambositra.'
[ (Noho) ny tahony ]
no n a n dosirany.
(because) det fear.3(gen) foc pst.CT.flee 3(nom)

'It is because of his fear that he fled.'
As we w il l see in
unclear. I
s ection 2.5, th e b e haviour o f C P a d j u n cts remains
t h e r efore leave the n o n-omissibility o f C
as a s t i p u l ation.
Prepositions, on the other hand, are always optional in the cleft position.
I now look m ore closely at some restrictions on the cleft construction
itself.
I n t h i s s ection, I w i l l e x a m in e t h e r a ng e o f o p e r ator-variable
c onstructions that are available in clefts, with a focus on adjuncts. I
will
s how that certain restrictions on clefts can be d e rived q u ite simply b y

requiring the category of XP and the operator to match.
2.4.1 AT cle
fts
Recall that when the verb bears active morphology clefting is restricted to
subjects and adjuncts. The data in (20) show that internal arguments of
the verb, whether DP or PP, cannot be AT clefted.
( 20)
a. * [
M p i anatra ] no mikapoka i san'andro Rabe.

s tudent
foc A T . hi t
eac h .day R a be
'It's students that Rabe hits every day.'
99
b. *
[ H o an'ny ankizy ]
n on anolotra
mofo Rasoa.
for.gen.detchild foc pst.AT.offer bread Rasoa

'It's to the children that Rasoa offers bread.'
I will take this ban on the clefting of i n ternal arguments as a fact about
Malagasy grammar and not attempt an explanation. Descriptively, VP is a
barrier t o
A - b a r e x t r action . F or o n e f o r m a l a c count o f e x t r a ction
restrictions, see Nakamura (1996).

On the other hand, AT clefting of adjuncts is in general possible. This is
true for l o cative, temporal, an d
i n s t rumental a d verbials, as i l l ustrated
below.'
( 21) a .
[ Tao aorian'ny mpampianatra ] no nipetraka

pst.there after.gen.det teacher foc pst.AT.sit
ah o .
1s g (nom)
'It's behind the teacher that I sat.'
[ Taorian'ny mpampianatra ] no niteny
aho.
pst.after.gen.det teacher foc pst.AT.speak 1 sg(nom)

'It's after the teacher that I spoke.'
[ Amin'ny penina ]
P.gen.det pen
n o m anorotra aho.
foc AT . w r ite 1 sg(nom)
'It's with a pen that I write.'
The data in (21) show that AT clefting is possible with a b r oad r ange of
adjuncts. I therefore conclude that AT clefting of adjuncts (as opposed to
a rguments) is not ruled out by constraints on movement and now t ur n t o
some restrictions on what can appear in a cleft.
When an adjunct appears in an AT cleft, the preposition (if there is one)
must be overt. E x amples (22a,b) below are parallel to (21b,c) above but
are ungrammatical due to the missing preposition.
" There is some variation in judgements with benefactives. This may be due to the fact that they
are marked with the same preposition as goals and hence pattern with arguments.
(i) ?
[H oa n 'ny ankizy] no mividy m o f o aho.
for.gen.det child foc AT.buy bread I sg(nom)
'It's for children that I buy bread.'
100
Chapter 3
( 22)
a. *
[ N y mpampianatra ] no niteny
aho.
foc pst.AT.speak 1 sg(nom)
det teacher
'It's after the teacher that I spoke.'
b.
* [
N y penina ] no manorotra aho.

det pen
foc A T . write 1 sg(nom)
'It's with a pen that I write.'
F ollowing Wi lliams (1980), I suggest that the focussed element and t h e

o perator must agree in categorial features. I n
features, but
t h e f o c ussed element i s a D P .
(22), the operator has P

(2 3 ) g i v e s a s c h ematic
representation of the examples in (22).

(23) * DP [ OPpp ...AT.verb ... tpp ... ]
This mismatch leads to ungrammaticality.

Interestingly, the matching requirement explains why th e restrictions
o n relativization hold strictly:
o n l y a s u b ject can be r elativized. T h u s
while clefting applies freely to subjects and adjuncts, for an adjunct to be

the head of a relative clause, the verb must bear CT morphology.
( 24)
a. *
ny antony izay nandeha R akoto

det reason rel pst.AT.go Rakoto
'the reason that Rakoto left'
ny antony izay nandehanan-dRakoto

det reason rel pst.CT.go.gen.Rakoto
'the reason that Rakoto left'
Since the head of the relative clause is necessarily a DP and never a PP, the
ungrammaticality of (24a) is due to a mismatch, diagrammed in (25).
(25) * [ DP [ OPpp ...AT.verb ... tpp ...]]]
101
In other words, the ungrammaticality of (24a) is parallel to that of (22a,b).

Let us now turn to some cases of adjuncts that cannot cleft in the active
v oice (i.e. without f i rst b eing p r o m o ted t o s u b j ect vi a C T ) .
M an n e r
adverbs resist AT clefts."
( 26) a .
[ Haingana ] no milomano i Soa.

q uickly
foc AT . swim S o a
'It's quickly that Soa swims.'
b.
[ M afy ]
n o m i t en y
i Boz y .
h ard
foc A T . speak B o z y
To account for the u n grammaticality in (26), I assume that adverbs are
heads that do not project (Travis (1988)). By definition (Chomsky (1981)),

an operator must bind an XP, not an X .
have operator m ovement f r o m
T h e r efore, it is not p ossible to
a n a d v er b p o sition. T h i s i s i l l u strated
schematically below.
(27) * [ adverb [ OP; [ ... X'; ...]]]

As well as X manner adverbs, Malagasy has PP manner adverbials. Since
these are XPs, clefting is possible. (28) therefore contrasts with (26).
(28)
[ A m - pitiavana ] n o m anoroka an'i Koto Rasoa

P.gen.love
foc A T . k is s a c c .Koto Rasoa
'It's with love that Rasoa kisses Koto.'
Importantly, all of the above adjuncts can appear in clefts when the verb
bears CT voice morphology. Thus (29) and (26b) are a minimal pair.
' Note that these examples do not improve with the addition of another adverb, unlike the English
equivalents.
(i) *
Ten a maf y n o m i t eny i B o zy .
very hard
foc A T .speak Bozy
'It's really loudly that Bozy speaks.'
In (i), we have one adverb head adjoined to another, still an X', not an XP. See Heggie (1993) for
a discussion of the contrast in English.
102
Chapter 3
(29)
[ M a f y ] no i t e nenan'i Bozy.
hard
foc C T .speak.gen.Bozy
F or adverbs, this resolves the ungrammaticality since the operator w i l l

correspond to the subject, an XP position.
(30) [ adverb [ OP; [ ... X'; ... XP; ]]]

Descriptively, the A
m o v e m ent o f t h e a d v er b t o t h e s u b ject p o sition
changes the X to an XP. The operator then binds the XP, as illustrated in
(30). The matching condition holds for o p erator m o v ement and clearly
does not apply between the subject and the base position of th e adverb.
Examples of CT clefts like (29) will be discussed further in the next section.
Summing up, I take clefting as a test that distinguishes the category of
d ifferent adjuncts. I n
g e n eral, clefting o f a n a d j u nct i s p o ssible, w i t h
c ertain restrictions. These restrictions on clefting fall out directly from t h e

matching requirement. P r epositions are obligatorily realized in the cleft
position to match with th e category of th e o p erator (if i t d oes not pass
through th e
s u bject p osition). W e h a v e s e en, m o r eover, t h a t h e ads
cannot cleft, which follows from the reasonable assumption that a variable
must be an XP, not an X .
2.4.2 CT cle
fts
In the preceding section, I provided an analysis of certain restrictions on
the cleft position when the verb is active. I
adjuncts when th e v er b
n o w t u r n t o t h e clefting of
b e ars C T m o r p h o l ogy .
Unl i k e i n A T c l e f ts,
mismatches between the clefted XP and the operator are tolerated in CT

clefts. I provide some tentative suggestions to account for this difference.
We have already seen that i n
c o n trast to A T c l e fts, i t i s g e n erally
p ossible for either a DP or a PP to appear in the focus position of a C T

c left. I n o t her w o r ds, the matching requirement is not observed. T h e
preposition is thus "optional" in some respect. R ecall that I i n v o ked an
account based on a categorial mismatch to explain the obligatory presence
of prepositions in adjunct clefts (example in (21) and (22), diagrammed in
103
(31a,b)). I
s t il l m u s t explain w ha t a l l ows th e m i smatch in C T c l auses,
shown in (31d). Schematically, we have the following pattern.

( 31)
a. *
DP [OPpp ...AT.verb ...tpp ...]
b.
PP [ OP pp ...
AT.verb ... tpp ... ]
e.
D P [ OP D p ...CT.verb ... tpp ... tDp ]
d.
PP [ OPp p a ...
CT.verb ... tpp ... tDp ]
The contrast between ( 31a) and (31b) indicates that the category of t h e
focussed element and th e o p erator m u s t
m a t ch . D e s criptively, (31c,d)
indicate that the focussed element can reflect either the DP or PP status
when the verb is CT because of the intermediate step through the subject
position. T h i s is not p ossible wit h A T , w h i ch ha s operator m o v e m ent
d irectly from th e base PP position.
H e n c e th e o b l igatory m a t ching i n
( 31b). T his di fference between AT clefts and CT clefts may s tem f r o m

d istinctions between A and A - bar m o v ement.
I t i s t h e A m o v e m ent t o
subject in (31c,d) that distinguishes these examples from

only in v olve A - bar m o v ement .
must obey
r equirement.
( 3 1a,b), which
De s c r iptively, p u r e A - ba r m o v e m ent
t h e m a t c hing c o n d ition w h i l e A m o v e m en t e s capes this

I f t h e se tentative suggestions are c o rrect, they p r o v i d e
further evidence in favour of t r eating the subject in M a l agasy as an A

rather than an A-bar position (in the spirit of Travis (to appear) and contra
Pearson (to appear)).

2.5A note on CPs
I have focussed most of the discussion on DPs and PPs as they appear in
c lefts. I now turn to CPs, which do not pattern with other adjuncts. I w i l l
describe the distribution of CP subjects and clefts but leave a complete
a nalysis for f u t ur e w o rk .
In o t h e r w o r d s , f o r t h e r e m a i nder o f t h i s
chapter, I concentrate on DP and PP adjuncts. Certain CPs appear in the

subject position, while others cannot. I n t erestingly, those CPs which can
be subjects, cannot A T
c l eft . C P s w h i c h ar e b arred f r o m t h e s u b ject
position, on the other hand, do undergo CT clefting.

F or comparison, let us first consider CP arguments and then turn t o
CP adjuncts. In (32a), the complement CP has been passivized to subject,
104
Chapter 3
as shown by th e n egative polarity p l acement in (32b).'"
T h e same CP,
however, is ungrammatical in the cleft position in (32c).

Heveriko
( 32) a .
fa handeha i Soa.
TT.think.lsg(gen) C fut.AT.go Soa

'That Soa will go is thought by me.'
b.
T sy h everiko
intsony fa handeha i Soa.
neg TT.think.1 sg(gen) NPI
C fut. A T .go Soa
'That Soa will go is no longer thought by me.'
c.
* [
Fa handeha i Soa ] no heveriko.

C fut.AT.go Soa
foc T T .think. 1 sg(gen)
(lit.)'It is that Soa will go that is thought by me.'

Similar facts hold for CP subjects of active verbs.
( 33) a .
Mahasosotra an'i Sahondra fa nanoroka a n - d Rabe i Bakoly.

AT.annoy
acc . Sahondra C pst.AT.kiss acc-Rabe Bakoly
'That Bakoly kissed Rabe annoys Sahondra.'

b. * [
F a nanoroka an-dRabe i Bakoly ] no mahasosotra an'i Sahondra.

C pst.AT.kiss
acc-Rabe Bakoly foc AT.annoy acc.Sahondra
T hese data show that a CP can appear in subject position, but not i n t h e
c left position."
N o t e t h a t i f w e a ccount for th e acceptability o f C P s i n
subjectpositionby claiming they are really DPs, we are even further from
understanding the impossibility of CP clefts, as DP subjects freely occur in
the cleft position.
Turning now to adjunct CPs, first note that no adjunct CP can appear
in subject position.
Alternatively the CP in ( 32a,b) could be extraposed. Since I do not intend to provide an

analysis for CPs, I do not explore this possibility.
" Looking at similar data from English, Stowell (1985) and Heggie (1993) account for the ban on
CP clefts by stipulating that functional categories cannot antecede an operator. In today's world of
rampant functional categories, it is clear that this account must be modified.
105
( 34)
a. *
Nandehanako
satria
n i h i n ana v o a nkazo ma n t a.
unripe
pst.CT.go. 1 sg(gen) b ecause pst.AT.eat fruit

b.
Ianarako
mafy mba hahazo
C T.study. 1 sg(gen) hard opt f u t .AT.have
kar a ma be.
s a l ary b i g
What explains this restriction? We have already seen in (32a) that CPs are
n ot in p r i nciple barred f ro m
b e in g s u bjects. I l e a v e u n explained t h i s
difference between the CP in (32a) and the ones in (34).

S econd, no a d j unct CPs can c left w h e n
t h e v e r b i s a c t i ve . ( 3 5 a )
illustrates a causal adjunct and (35b) a purpose clause.

( 35)
a. * [
N oh o
izy
mbol a kel y
because 3(nom) still

no tsy afaka handeha
tao n a l o atra ]
s m a l l y e a r too
an y
foc neg free f u t . AT.go there
A mb o s itra i Koto.
A m bositra i Koto
'Because he is still too young, Koto cannot go to Ambositra.'
b.
* [ Mba hahazo karama be ] no mianatra mafy aho.

opt f u t . AT.get salary bi g foc A T . study har d 1sg(nom)
Since we have already seen that most adjuncts can undergo AT clefting,
this is a somewhat surprising result. Does it follow from the impossibility
of CP clefts mentioned earlier (see (32b) and (33b))? N o , b ecause unlike
argument CPs, these clausal adjuncts can be clefted when the verb bears
CT morphology.
mafy .
(36) [ Mba hahazo karama be ] no ianarako
opt f u t . AT.get salary bi g foc C T .study.l sg(gen) hard
Clearly, it is not th e clausal status which p r events these adjuncts from

clefting when the verb is active. Thus the description of clefting must be
modified; subjects and non-CP adjuncts can cleft in AT.
t hese adjuncts PPs only
M o r e over, calling
i n c reases the complications. I n ge n e r al, P P
106
Chapter 3
adjuncts can cleft. In other w o r ds, adjunct CPs do not pattern uniformly
with adjunct PPs. This rather confusing array of f acts is diagrammed in
(37)
(37)
CP subject
CP adjunct
PP adjunct
subject
AT cleft
CT cleft
Summing up, no adjunct CP can cleft in the active voice, but they allow
clefts when the verb is CT. CPs as arguments and CPs as adjuncts have
different properties. I do not have an explanation for this division among
CPs. As mentioned above, I leave a in-depth study of clausal adjuncts to
future research.
2.6 Where are we?
T his section ha s
s h ow n t h e r a n g e o f c o n s t ructions a ssociated w i t h
circumstantial morphology.
W e h a v e seen t hat C T i s o f ten associated
with adjuncts, although there w er e som e c o u nterexamples (goals and

p artitive themes).
v arious categories.
W e h a v e a lso seen that C T p r o m o tes elements o f

S i n ce m an y C T c l a uses are clefts, I p r o v i ded a n
overview of the cleft construction and some restrictions that apply.

Now that we are familiar with CT clauses, we can assess the various
a nalyses that have been proposed to account for th e d a ta .
consider th e
F i rst I w i ll
c o n nection b e t w een C T a n d a d j u n cts, so-called t h e t a-
agreement. Second I will assess the preposition-incorporation approach,

which treats CT as an applicative construction. I w i l l show that neither of
these analyses can account for the full r ange of C T clause and therefore
reject them as inadequate.
3
CT=
romotion of ad'uncts
As I have pointed out several times, there is an obvious link between CT

and adjuncts. In many cases, CT promotes an adjunct to subject. This is in
fact the traditional description of CT ( M alzac (1960); Rahajarizafy (1960);
Rajemisa-Raolison ( 1 966); R ajaona
( 1 972)).
W ith i n t he ge n e r a t iv e
tradition, Sityar (in press) has analyzed v o ice as t h eta-agreement.

107
If
adjuncts do not receive a theta-role (or bear an o v er-arching " a djunct"
r ole), then Sityar's analysis also associates CT w it h a d j uncts." L e t u s
consider this approach carefully and see how it w o uld apply to M alagasy.
I will then attempt to distinguish arguments from adjuncts. Once we have
a clear distinction, I will show that CT is not limited to the p r om otion o f
adjuncts.
He n c e , I w i l l r e j ec t t h e t r a d i t i onal a n alysis an d S i t y ar's
formalization of it.
3.1 Theta-agreement
As mentioned in chapter 1, Sityar (in press) proposes a system of " t h eta
a greement" t o
c a p t u r e t h e c o r r elation b e t w een v o i c e m a r k in g a n d
thematic roles:
(38)
T h e nominal features of VoiceP must agree with the thematic features of the
topic.
The voice morphology on the verb is therefore a special form o f s u bject

agreement. In support of this approach, certain authors claim that thetaroles are in fact features that must be checked, similar to Case (Boskovic
(1994); Hornstein (1999)). With theta-roles as features, theta-agreement is
parallel to other types of agreement, for example number and person.
Let us now look at little more closely at the theta-agreement analysis
of voice. In (39), I show schematically how this approach maps arguments
to verbal morphology.
M a l a gasy only has three voices, so the mapping
could proceed as below.

(39)
V o i ce Marking I
A,
<~
t h e me, goal> o ther (benefactive, instrumental, locative, etc.)
AT
TT
TT
CT
As signaled in (39), the external argument is linked to A T an d all ot h er

subcategorized elements (e.g. themes and goals) are externalized with TT.
N on-subcategorized e l ements u s e C T .
A t fi r s t gl a n ce, t h e t h e t a -
" Note that Sityar's analysis is for Cebuano, which splits the Malagasy CT into different voices.
I am therefore modifying Sityar's analysis to fit the Malagasy data.
108
Chapter 3
agreement analysis distinguishes between arguments and adjuncts.
A s an initial complication, recall from
c h apter 2 t h a t t h ere are t w o
passives in Malagasy. In (40), I give a possible modification of (39) that

takes into account the two passives.
(40)
V o i ce Marking II
A,
<~
AT
t h e me, goal> o ther (benefactive, instrumental, locative, etc.)

a-
-Vna
CT
However, in the chapter on passives, we saw that there was no one-toone mapping between th e p assive affix and t r aditional thematic roles.
Some themes externalize with the a- passive, others with -Vna. Mo r e o ver,
instruments appear in the subject position with the a- prefix. In c hapter 2,
I accounted for t hi s r ange b y
p r o p o sing t h a t " d i s placed themes" a r e
promoted with the a- passive. Im p o r tantly, I concluded that both t he apassive and the -V na p a ssive promote i n t ernal (DP) a rguments of t h e
verb. Let us then return to the basic insight of (39):
(41)
V o i ce Marking III
external argument internal arguments
AT
adjuncts
CT
I n the following sub-sections, I
d i scuss data f ro m P P a r g u m ents a n d
exceptional CT (ECTM) which are problematic for the analysis in (41).

3.2 Adjunctsvs. arg uments
What are adjuncts? U n der one tr aditional characterization, adjuncts are
non-obligatory elements in a
o bject, for example, an
g rammaticality o f a
c l ause. T h u s i n d i s tinction t o t h e d i r e ct
a d j unct can b e o m i t ted w i t h ou t a f f ecting t h e
s e ntence. A s m e n t i oned i n c h apter 2, t his test i s
somewhat limited. M any verbs are "optionally" transitive. Typical of this

group is 'eat'. T hus if something is an adjunct, it is optional, but not all
optional elements are adjuncts. T hi s test therefore needs to be applied
with caution, but I will refer to it in the following discussion.
109
In chapter 2, we saw that Malagasy has another test that distinguishes
adjuncts from ar guments.
section 2.4. I n
C o n sider A T clefts, which were discussed in
g e n eral, only s u bjects and adjuncts can un dergo A - b ar
movement in M alagasy. Let us see how the op tionality and the clefting
tests apply to the verb manome'to give'.
(42)
manome 'give'
a.
M an o me boky hoan'i
Koto iSahondra.
AT.give book for.gen.Koto Sahondra
'Sahondra gives a book to Koto.'
b. *
Manome boky i Sahondra.

AT.give book Sahondra
'Sahondra gives a book.'
c.
Manome hoan'iKoto iSahondra.

AT.give for.gen.Koto Sahondra
'Sahondra gives to Koto.'
d. * [
B oky ]
n o m anome hoan'i Koto i Sahondra.
book
f oc AT.give
for.gen.Koto Sahondra
'It's a book Sahondra gives to Koto.'
e.
* [
H o an'i Koto ] no manome boky i Sahondra.

for.gen.Koto focAT.give book Sahondra
'It's to Koto that Sahondra gives a book.
N either
t h e t h e m e n o r the go a l is op t i o n al , a s s h o w n by
ungrammaticality o f
the
( 42b,c), and neither can A T c l eft, as i l l ustrated in
(42d,e). The two tests clearly correlate and show that both the theme and
the goal of manome'give' are arguments. Hence we can classify manomeas
ditransitive.
The results for mamaky 'break' are shown in (43).
110
Chapter 3
(43)
m a m aky 'break'
a.
M ama k y ny tavoahangy amin'ny l a n g ilangy i Noro.

A T.break det bottle
P.gen . det stic k
Noro
'Noro breaks the bottle with the stick.'
Mamaky ny tavoahangy iNoro.

A T.break det bottle
Noro
'Noro breaks the bottle.'
c.
Mamaky a m in'ny
l a n g ilangy i Noro.
AT.break P.gen.det stick
Noro
'Noro breaks with the stick.'
d.
* [
N y tavoahangy ] no mamaky a m i n ' n y la n g ilangy i Noro.

det bottle
foc AT.break P.gen.det stic k
Noro
'It's the bottle that Noro breaks with the stick.'
[ Amin'ny langilangy ] no mamaky n y tavoahangy i Noro.

P.gen.det stick
foc AT . b reak det bottle
Noro
'It's with the stick that Noro breaks the bottle.'
A gain, the op tionality an d
t h e c l e fting t ests correlate. T h e t h e m e i s
obligatory and cannot AT cleft, as shown in (43c,d). I t
i s t h e r efore an
argument. The instrument, on the other hand, is optional and can AT cleft
(see (43b,e)). It is an adjunct. H e nce mamaky'break' is transitive. T aken
together, the optionality and the AT clefting tests make a clear argumentadjunct distinction.
Now w e can turn t o s ome m or e d i f ficult cases. C o n sider mandroso
'serve', which, like manome'give', has both a theme and a goal. In c hapter
2 , I claimed that both th e t h eme an d t h e g oal ar e a r guments. L e t u s
review the data.
(44)
ma n d r oso'serve'
a.
M and r oso vary hoan'ny v ahiny Rasoa.

AT.serve rice for.gen.det guest R a soa
'Rasoa serves rice to the guests.'
111
Mandroso vary Rasoa.

AT.serve rice Rasoa
'Rasoa serves rice.'
c.
Mandroso hoan'ny
v a h iny Rasoa.
AT.serve for.gen.det guest R a soa

'Rasoa serves to the guests.'
d. * [
V ary ] no mandroso hoan'ny v a h iny Rasoa.

rice
f o c AT.serve for.gen.det guest R asoa
'It's rice that Rasoa serves to the guests.'
[ Hoan'ny vahiny ]
n o m androso vary Rasoa.
for.gen.det guest foc AT.serve rice Rasoa

'It's to the guests that Rasoa serves rice.'
The theme of
ma ndroso'serve' is a r g ument-like by b o t h
t e s ts ( 44c,d).
However, the optionality test and the clefting tests give different results
for the goal. I t is optional (44b), but it cannot AT cleft (44e). To account
f or thi s d i screpancy, I
s u g gest t hat t h er e ar e o p t i o nal a r g u m ents i n
Malagasy (just as in English). T h u s the goal is an optional argument of

mandroso 'serve'. In other words, I take the clefting test as being relevant
for the adjunct-argument distinction."
3.3 PP arguments
Recall the
t h e ta-agreement a p p r oach t o v o i ce . M od i f i e d t o f i t the
Malagasy system, this analysis states that internal arguments (irrespective

of category) w il l b e
p r o m o ted w i t h T T . T h e d a t a f r o m g o als i n itially
suggest that the theta-agreement approach is correct. G o als are internal

arguments (as shown in (44)) and are promoted with TT, as shown in (45).
As in English and all other languages, there will i n the end always be exceptions to this
generalization. Recall, for example, that CP adjuncts cannot AT cleft. Y e t I w o uld hesitate to
classify them as arguments. I n e vertheless take the AT cleft test as distinguishing between
arguments and adjuncts for the purposes of this thesis.
112
Chapter 3
(45)
vary n y yahiny.
N o t o loran-dRasoa
pst.offer.Vna.gen.Rasoa rice
de tguest
'The guestswere offered rice by Rasoa.'
The theta-agreement analysis also states that CT i s l i m i ted t o
a d j uncts.
I nterestingly, w h il e (45) i s g r a mmatical, th e g oal ca n a l t ernatively b e

promoted with CT.
(46)
N a n drosoan-dRasoa
vary n y yahiny.
pst.CT.offer.gen.Rasoa r i c e
de tg uest
'The guestswere offered rice by Rasoa.'
Since I have argued that goals are arguments, (46) shows that it is clearly
incorrect to characterize CT in terms of the argument-adjunct distinction.
CT can promote certain arguments, such as goals.
I now
t u r n t o f u r t h e r d a t a w h i c h ar e p r o b lematic fo r t h e t h e t a-
a greement approach to CT. A s above, these data involve pr omotion t o

subject of a non-adjunct with CT.
3.4 ECTM
A s mentioned briefly i n
t h e f i rst section of t h i s chapter, CT ha s w h a t
appears to be an exceptional use, illustrated in (47).

(47)
A n a san-dRakoto
ny loyia.
CT.wash.gen.Rakoto det dishes

'Rakoto washes some of the dishes.'
I will discuss this case in detail in section 6. For the m oment, I p o int o u t
t hat the element promoted to subject is a theme, clearly an argument o f
the transitive verb manasa'wash'. Ne v e rtheless, the voice morphology is
C T. A s w i t h PP arguments, the datum i n (47) indicates that CT i s n o t
limited to the promotion of adjuncts.
3.5 Theory
A bove, I
h a v e g i v e n s o m e e m p i r ical a r g u m ents a g ainst t h e t h e t a-
a greement analysis of voice. I n p a r ticular, I have shown t hat CT i s n o t

uniquely for the promotion of adjuncts. In certain cases, arguments (such
113
as goals and partitive themes) can be promoted with CT .
t heoretical reasons to
c onsensus on
T h ere are also
q u e stion t h eta-agreement. F i r s t , t h er e i s l i t t l e
t h e s e t o f t h e t a-roles av ailable t o t h e g r a m m a r a n d
definitions of particular theta-roles are typically v ague.
I t i s n o t c l ear,
m oreover, that syntax is ever sensitive to the content of theta roles. Fo r

e xample, the t h eta-criterion e n sures a
arguments in
p r o p e r m a p p in g b e t w een D P
s y n tax an d t h e t a p o s i tions i n t h e p r e d i cate a r g u m ent
structure. But the theta-criterion is not concerned with t h eta role labels,
per se. Rappaport and Levin (1988) discuss in more detail the issues that
surround the role of thematic relations in syntactic theory. I conclude that
thematic features are not a desirable addition to the theory.
Summing up, this section has evaluated a particular analysis of voice
that links CT to a d juncts.
M oreover,
A r a n g e o f d a t a show t h i s t o b e i n c o rrect.
t h e t a-agreement a s s u c h s u f f er s f r o m th e o r etical s h o r t -
c omings due t o t h e d i r ect reference to t h e c o n tent o f t h e t a r o l es . I

therefore reject the theta-agreement analysis.
4
CT =
re osition incor oration
As a possible alternative to the adjunct agreement analysis above, let us

consider t h e
c o n n ection t h a t h a s b e e n dr a w n be t w e e n C T and
prepositions. In his early paper on Malagasy, Keenan (1976) remarks that

CT promotes elements that w o uld be m a r ked w i t h
preposition)."
o b l i que case (i.e. a
In the i r c l assic w or k o n A u s t r o nesian subjects, GH T
propose that CT is a type of preposition incorporation (a la Baker (1988)).

I will first outline the GHT analysis and then explore two aspects of their
a ccount:
t he p a r a llel d r aw n w i t h a p p l i catives and t h e p r e sence o f
prepositions. I show that CT is not an applicative construction and does

not always involve prepositions. I t h erefore reject the GHT approach to
CT."
4.1 Guilfoyle, Hung and Travis

GHT state that active morphology assigns accusative Case to the theme,
w hile passive morphology licenses genitive Case for the agent.
Putting
the two together creates CT, which is characterized by the availability of
" As I discuss below, Keenan (in press) rejects his (1976) description of CT.
" To be precise, I reject the P-incorporation part of their analysis. I agree with their observations
about Case assignment in the different voices.
114
Chapter 3
both accusative and genitive.
(48)
DPnom
I'
vp
V
DPgen
V
V
PP
DPncc P
Furthermore, GHT suggest that CT i n v o l ves preposition i n corporation,

along the lines of Baker (1988). Following preposition incorporation, the
object of the preposition no l o nger receives Case and must raise to th e
matrix subject position for nominative. It is via P-incorporation that GHT
integrate CT into the traditional Case analysis of voice.
4.2 Applicatives
The P-incorporation analysis of GHT equates CT and applicatives, which
Baker's analysis was designed to a ccount f or .
c onstruction?
W h a t i s a n a p p l i cative
C h u n g ( 1 976a) discusses the f o ll ow in g e x amples f r o m
Bahasa Indonesia, which illustrate a classic case of an applicative.

(49)
a.
Orang i t u masak ikan untuk perempuan itu.

man
t h e cook f i s h fo r
wom a n
'The man cookedfishfor the woman.'
115
the
Orang itu me-masak-kan
per e mpuan itu ikan.
m an
w om an
t h e t r ans-cook-ben
the f i s h
'The man cookedthe woman fish.'
In (49a), perempuan i tu 'the wo m a n' is t he object of t he p r e position. I n

(49b), however, it is a direct argument of the verb. M o r eover, the verb in
( 49b) bears special m o r phology .
Pa s s ive can t he n a p p ly , p r o m o t i n g
perempuan itu 'the woman' to subject.

(50)
P e rempuan itu di-masak-kan ikan oleh orang i tu.

woman
t h e pass-cook-ben fish by man
the
'The woman was cooked fish by the man.'
(50) resembles a CT construction: a D P t hat was originally in a PP is in

subject position.
Crucially, applicative takes an oblique and promotes it to object; hence
the object position plays an important role. I n f act, Baker (1988) refers to
this property of a p p licative as "M arantz's Generalization":
w h e n ever a
verb appears with both extra morphology and an additional DP argument

b earing some oblique thematic role, that ad ditional DP ar gument w i l l
behave like a surface direct object of t h e c o m p lex v erb . I n t e r estingly,
Malagasy does not seem to h ave an applicative construction parallel to
(49b), as shown by the ungrammaticality of (51b).

(51)
a.
Nahan d r o
ny tr o n dro hoan'ny v e h ivavy n y lehilahy.
pst.AT.cook d e t fish
for . gen.det woman d e t man
'The man cookedfishfor the woman.'
b. *
Nahandro
ny v e hivavy n y t r ondro ny lehilahy.
pst.AT.cook d et woman
Nandrahoan'ny
de t f is h
lehi l a h y
de tm an
ny t r ondro ny yehiyayy.
pst.CT.cook.gen.det ma n
det fi s h
'The woman was cooked fish by the man.'
det w oman
Instead, the object of the preposition is promoted directly to subject, as in
116
Chapter 3
(51c), parallel to (50).
D e s criptively, it a p pears that th e d e rived object
position is not available in Malagasy.

P ursuing thi s p ossibility, T r avis ( 1999); Rackowski an d
T r a vi s ( t o
appear) claim that AgrO in Malagasy is "defective" in nature and link this
t o th e n a t ur e o f m o v e m en t i n t h i s l a n g u age. "
Beg i n n in g w i t h a
discussion of verb-initial wor d o r d er, they p r o v ide suggestive evidence

that Malagasy is a predicate fronting rather than an argument f r o n t i n g
l anguage.
T h e y s h o w t h a t t h i s h a s s e v eral c o n sequences fo r t h e
r ealization of a r g uments.
T h e y s u g gest t hat a r g u ment m o v e m ent i n
Malagasy is driven by reasons other than feature checking. A g r O can be

inserted at variable points in th e d e r ivation and i s i n v isible to f u r t h er
movement. A s evidence in support of the unusual nature of AgrO, they
point out t h e l ack o f m o v e m ent t o o b j ect p o sition i n a p p l i catives, as
mentioned above.
unavailability o f
T h e y a l s o n ot e o t her p r o cesses which indicate the

t h e d e r i v e d o b j ect p o s i tion .
For exa m p l e , c r o ss-
linguistically possessor raising usually targets the possessor of an object.

In M a l agasy, p o ssessor r a i sin g
i s un i q u el y f r o m s u b j ec t p o s i t i on.
M oreover, apparent raising to object in M a l agasy is arguably no t t o
derived object position (see Travis (1997)). Thus we could conclude that
CT is applicative, but that some independent property o f
t h e l a n guage
bars the promoted element from s u r facing in a d erived object position.

Instead, it raises all the way to subject." I
s u g gest, however, that this is
not the correct conclusion.

Before we turn to CT, recall that in chapter 2 I have argued that certain
e lements may be base generated in a p r o jection c-commanded by t h e
agent and c-commanding t h e
t h e m e, [ Spec, v2P].
I am r e f e r r in g t o
instruments and material themes. I argued that this is not a derived object
position as it is limited to a subset of these elements. I then concluded that
the apparent "advancement" of instruments and material themes was not
a kind of applicative due to its restrictive nature. What about CT? Since it
does promote a w id e r ange of adjuncts, is it a k i n d o f
a p p l icative that
" For completely different reasons, Maclachlan and Nakamura (1997) argue that AgrO is "inert"
in active clauses in Tagalog. Thus there indeed appears to be something unusual about the object
position in these languages.
" This property of CT falls out directly under an ergative analysis of Malagasy. I do not,
however,believe Malagasy to be an ergative language. See chapter 2 for an extremely brief
discussion of ergativity.
117
promotes directly to subject, by-passing the object position?
I s u g g est
that the resemblance between CT and applicative is spurious. In contrast

to standard examples of applicative, CT is too b road in scope. W e h a v e
seen that CT involves the promotion of almost any adjunct (time, cause,
manner, etc. ). As discussed in Baker (1988), applicative is restricted to a
subset of obliques, the most common being benefactives, instruments and
l ocatives. A
r a n g e of f acts thus leads to th e c onclusion that C T i s n o t
applicative.
At t hi s
p o i nt , w e a r e c o n f r o nted w i t h t h e m a t t e r o f r e d e f i n i ng
applicative. Early work benefited by examining the similarities between a

r ange of constructions, under the heading "applicative". I t
w o u l d no w b e
important to look at each construction separately to determine if there is

i ndeed a single applicative.
I ne v e r t heless believe that C T h a s v e r y
different properties from what has traditionally been called an applicative

construction.
4.3 Prepositions
Although it may be incorrect to label CT as applicative, this does not rule
o ut P-incorporation, pe r
se.
Let us th e r e fore e x amine t h e r o l e o f
prepositions in CT constructions.
K e enan (in press) criticizes the GH T
a nalysis, in p a rticular th e P - i ncorporation account o f C T .
Ins t ea d o f
positing prepositions in the syntax, Keenan proposes that CT encodes the

prepositional meaning in the semantics. He points out that in many cases,
t he element that i s p r o m o ted w i t h C T n e v e r a p p ears w it h a n o v e r t
preposition. Keenan cites the following illustrative examples.
(52)
a.
Tonga o m al y
arrive
Rabe
y e sterday R a be
'Rabe arrived yesterday.'
b.
[Omaly ] no nahatongavan-dRabe
yesterday foc pst.CT.arrive.gen.Rabe
'It was yesterday that Rabe arrived.'
118
Chapter 3
(53)
a.
Mivarotra mo ra
ny
mIIiy a rotra amin'io a nd r o io
A T.sell
det
me r c hant P . gen.this day t h i s
che a p
'The merchants sell cheap on that day.'
b.
[ Mora ] no i v a r otan'ny m
c heap f o c
pivar o tra amin'io a
ndr o io
CT . sell.gen.det m e rchant P . gen.this day t h i s
'It is cheap that the merchants sell on that day.'
It would b e p o ssible to p o sit u n d erlying n ull p r epositions d o m inating

these adjuncts, but such covert prepositions have little other motivation in
the grammar. M o r e over, Larson (1985) argues against such an approach
to bare DP adverbials in English. He points out that the class of adverbials
i n English is b r o ad, i n cluding PP, A d vP, C P
and DP .
A re a l l t h e s e
dominated by a common invisible node?

Similar considerations apply to M alagasy, where a range of categories
may serve as adverbials and are promoted to subject with CT due to their
n on-DP status.
S e v eral examples were discussed in section 2.2: n o n -
referential DPs, CPs and VPs.

[ Valopolo ]
80
n o n i v idianany
hena.
foc pst.CT.buy.3(gen) meat
'It was for 80 (ariary) that she bought meat.'
[ Mihinana akoho ] no itsanganan-dRabe.

AT.eat chicken f o c CT.stand.gen.Rabe
'It's while eating chicken that Rabe stands.'
Yet it is unreasonable to posit empty structure simply to capture the voice

alternations.'" I n sum, we cannot link CT to prepositions, because of the
data in (52), (53) and (54).

There are
a ls o t h e oretical p r o b l ems i n h e rent t o t h e pr e p o sition
'" Emonds (1987) and McCawley (1988) both criticize Larson's analysis and argue in favour of an
empty preposition in DP adverbials. Their arguments would not extend to all the Malagasy cases,
h owever. M c C awley, for example, shows that bare DP adverbials act like PPs and not l i k e
adverbs. In M a lagasy, however, all of these elements pattern together in being promoted to
subject with CT, despite other distributional differences. For example, CP adverbials appear in a
post-subject position, unlike most other adverbials. Nevertheless, all adverbials are promoted with
CT.
119
incorporation approach. First, recall that in many cases, the element that
i s being p r o m oted i s
g e n erated i n a n a d j u nc t p o s ition .
H en c e t h e
preposition would be moving out of an adjunct, in violation of the Head

Movement Constraint (Travis (1984)), which states that a head may o n l y
move to a head that properly governs it. Since a verb does not properly
govern a PP adjunct, head movement of the preposition should be ruled
out. S e cond, in t y p ical cases of p r eposition i n corporation (e.g. Bantu
applicatives), Case is still available for the object of th e p r eposition (the
derived object). According to Baker (1988), for example, the verb governs
t he d e r i ve d
o b j ect a f t e r P - i n c orporation d u e to the G ov e r n m e n t
Transparency Corollary (GTC). A c cording to the GTC, a lexical category

t hat ha s a n
i t e m i n c o r p orated i n t o i t g o v e rn s e v e r y thin g t h a t t h e
i ncorporated item governed in its original position.
S i nce the verb w i t h
the incorporated preposition governs the object, it in principle can assign

Case.
T o a c c o un t f o r C T w i t h p r e p o sition i n c orporation t h e r efore
requires the additional stipulation that Case is lost.

S umming up, a good p r o p ortion of examples of CT d o
promotion of the object of a preposition.
i n v o lv e t h e
N e v e r theless, there are many
e xamples that do not fall under this generalization.
A n y a c count of C T
t hat relies on the presence of a preposition is therefore inadequate. I n o w

suggest an alternative analysis.
5
Anal sis
The data in section 2 illustrated the range of elements that are promoted
to subject in a CT clause. Most adjuncts can be promoted to subject, some
obligatorily appearing in a cleft position due to categorial restrictions on
the subject position (e.g. adverbs). What is common to all these elements?
Is CT to promote adjuncts? No; we saw in section 3 that PP arguments
(e.g. goals and partitive themes) can take CT.
H e n c e CT is no t p u r e ly
sensitive to the argument-adjunct distinction. Does CT promote PPs? No;

as shown in section 4, not all adjuncts are marked by p r e positions, yet
they are associated with CT morphology.
T o resolve this problem, I suggest that GHT's basic insight into CT i s
correct. As mentioned at the beginning of this chapter, GHT propose that
CT morphology satisfies the Case requirements of all the arguments of
the verb. A transitive CT verb will assign accusative Case to its theme and
120
Chapter 3
genitive Case to its agent. Some other element must raise into the subject
position. The other element will be anything that is not structurally Casemarked by the verb.'" T his will include all objects of prepositions as well
a s adjuncts in general, whether they take a preposition or not .
We now
need to understand what determines movement to the subject position.

The previous chapter presented an analysis of passive where the lack
of accusative Case forced DP movement t o subject position.
T h e same
cannot be true for CT. The elements that raise to subject position often do
not need Case, for example adverbials. Instead, raising is forced by other
c onsiderations.
H e r e I i n v o k e t h e EPP. M a l a gasy has a v er y s t r o n g
requirement that clauses have a subject.

A s initial motivation f o r
empty o r
t h i s p o sition, n ot e t h a t M a l agasy has n o
d u m m y s u b j ects. K e e na n ( 1976) points ou t t h a t w e a t h er
expressions use an overt DP in subject position.
(SS) a.
Mafana ny andro.
hot
de td ay
'It's hot.' (lit. 'The day is hot.')
Mandrivotra n y andro.
A T.wind
det d a y
'It's windy.' (lit. 'The day winds.')
Avy
n y o rana.
come det rain
'It's raining.' (lit. 'The rain comes.')

Similarly, there are no cases where a dummy replaces a sentential subject.
Instead, the entire CP appears in subject position."
" T his analysis of CT forces the conclusion that complement CPs are Case-marked by the verb
since they take passive and not CT.
Keenan (1976) suggests that in examples parallel to (56) but with a passive verb, raising to
s ubject optionally occurs: the embedded subject may raise into the matrix subject position. I n
either case, some element (CP or DP) occupies the matrix subject position. In (56) and (57), there
is still the possibility that the CP is extraposed.
121
(56)
M a z ava fa e fa
clear
l asa
ny mp i a natra.
C alr e ady g one d e t student
'That the students have already left is clear.'
That the CP is in subject position can be shown by the placement of V Pfinal particles.
(57)
T s y mazava
intsony fa efa l
neg clear N P I
asa
ny mp i a natra.
C alre a d y g on e d e t student
'That the students have already left is no longer clear.'

Finally, the other common construction for dummy subjects is existentials.
In Paul (in press), I show that existential constructions may take an overt
subject, as in (58b,c). Moreover, I argue that a null DP occupies the subject
position in apparently null-subject existentials, as indicated in (58a).
( 58) a .
Misy
zaza m itomany pro.
AT.have c h ild AT.cry

'There is a child crying.'
Misy
mito m any ny zaza.
AT.have A T . cr y
det c hild
'Some ofthe children are crying.'

Misy
molo t r a
A T.have l i p
ny a k oho.
det chicken
'Chickens have lips.'
S ome m o t i v ation
f o r t he nu l l s u b j ect i n
(58a) c o me s
f r o m the
interpretation of existential constructions. Th e existential verb sets up a

part-whole relation between the VP-internal material and the VP-external
subject. Consider first (58c), which has a clear possessive interpretation,
licensed by the i n alienable possession relation.
T u r n i n g n o w t o ( 5 8b),
w hich has a p a rtitive reading, I s u ggest it m e ans something l i k e ' t h e

children have crying ones among them'. E x tending this to (58a), I posit a
null subject that corresponds to the universe of discourse. The reading is
122
Chapter 3
' the w o rl d
h a s c r y in g c h i l d ren i n i t ' .
M al a g asy t h u s h a s a s t r o n g
requirement that the subject position be filled."
I n G B t erms, this is due
to the EPP.
I n the context of the p r edicate-fronting approach to M a l agasy w o r d
order discussed in chapter 1, t hi s a nalysis raises interesting q uestions
about the position of the subject and the features that trigger movement.
Under a typical predicate fronting analysis, the predicate raises to [Spec,
TP] and the subject surfaces in a lower [Spec, XP]. If the EPP motivates the
predicate raising, as suggested by M a ssam and Smallwood (1996), what
motivates subject raising? In most cases, the subject will be forced to raise
to be assigned Case.
H o w e v er, as just d i scussed, Case considerations
clearly do not drive subject raising in CT clauses since the raised element
typically does not require Case (e.g. adverbs).
Massam and Smallwood (1996) propose two ty pes of EPP to account
for the difference between predicate fronting languages such as Niuean
and argument fr onting languages like English. I n
N i u e an, the EPP is a
strong [T] feature which attracts either a V or a XP predicate."

o btains vi a
C h e cking
or v ia m ov e m en t t o [Spec, TP],
h ea d a d j u nction t o T
depending on the X /X P status of the predicate. In English, on the other

hand, the EPP is a strong [D] feature, which attracts a DP. The [D] feature
is located on T in clauses, but also on the head of small clauses. Following
this distinction, I tentatively suggest that Malagasy has both: a strong
feature on T
a n d a s t r ong [D] feature in X .
predicate and the latter attracts the subject.

Niuean, M alagasy does appear t o
[T]
The f o r m e r a t t r acts the

U n l i k e l a n guages such as
h a v e a c l e a rly i d e n tifiable subject
position.
A s d i s cussed by M a s sam (t o a p p ear), although s u bjects c-
command
o b j ects i n N iu e an , t h e y a r e n o t oth e r w i s e s t r u cturally
distinguished (e.g. by extraction, raising, quantifier float). In M alagasy, on

the other hand, the subject is associated with p a r ticular pr operties, for
example position (recall the particle placement facts from chapter 1) and
extraction.
H en c e t h e r e i s s o m e m o t i v a t io n f o r tr e a t in g s u b j ect
movement as standard argument movement. I leave this issue for further

research.
" M alagasy does have limited "topic-drop" of NPs in certain discourse contexts.
" M assam (to appear) modifies this account slightly. She proposes that in Niuean the EPP is a
strong [pred] feature that uniformly attracts an XP to [Spec, IP], where TP dominates IP.
123
Before continuing, I point out t hat Sells (1998) criticizes the mapping
approach to voice.
A l t h o ug h h e f o cusses on Philippine languages, his
criticisms apply t o
M a l agasy. S e lls claims that v o ice m arkers i n t h e se
languages ignore any argument structure hierarchy. I n

a rgues against both t h eta agreement an d
o t her w o r ds, he
s t r u ctural analyses of v o i c e
alternations in these languages. Draw ing on data from exceptional voice

marking (see section 6), he shows that there is no one-to-one mapping
between v oice an d a r g u m ents .
Ins t e ad, h e p r o p o ses lexical l i n k ing
between a verb's arguments and v oice morphemes (similar to C a r r ierD uncan (1985)). I n
o t her w o r ds, arguments and v o ice m o r phemes f o r
e ach verb are linked in the lexicon. I
b e l i eve, how ever, t hat t h ere a r e
regularities to the voice system that are not captured by a lexical linking
approach. Moreover, I believe that it is not by i gnoring the "exceptional"
u ses o f
v o i c e m o r p h o l og y th a t w e
characterization. O n
w ill
ar r i v e
at
the
cor r e c t
t h e contrary, these "exceptions" can p r o v ide n ew
insight into the true nature of voice alternations. I
t u r n t o o n e o f t h e se
e xceptional instances of CT in the next section and show ho w i t
follows
directly from the present analysis.

6
ECTM
If the proposed analysis of CT is correct, we would expect CT (and not TT)

to promote t h emes that ar e ( fo r s om e
m arked.
r e ason) no t s t r u cturally C ase-
In th i s s e c tion, I a n a l yz e i n d e t ai l o n e p a r t i cular u s e o f
circumstantial topic, Exceptional Circumstantial Topic Marking (ECTM). I

will show t h at in E CTM c onstructions, the verb takes a QP ( "q uantifier
phrase") theme. The DP in this QP is not Case-marked by the verb and is
therefore p r o m oted
t o s u b j ect w i t h C T m or p h o l o gy .
ECTM thu s
provides further evidence for dissociation between voice and theta roles
and for the "elsewhere" nature of CT.
6.1 Partitives in Malagasy
For the p u r p oses of c o m p arison, I
strategies employed i n
b e gi n b y e x p l o r in g f o u r d i f f e rent
M a l a gasy t o e x p r ess p a rtitivity : per i p h r astic
partitives, subject position, existentials, ECTM. In the subsequent sections,

I will concentrate on ECTM, due to its relevance for the analysis of CT. I n
order t o
s i m p l if y d i s cussion, I w i l l u s e t h e f o l l o w i n g t e r m i n o l ogy:
p artitives express a relation between a part o r
124
" s u bset" and a w h ol e o r
Chapter 3
"superset".
6.1.1
Pe riphrastic partitives
Malagasy has a wide range of prepositions, but the m ost p o lysemous is

amin'. According to Rajemisa-Raolison (1966), 'at', 'with', 'in', 'of', 'about',
' from', and 't ow ard' are just a few o f t h e r elations expressed. Fo r t h i s
r eason, I
g l o s s a min' a s P f o r p r e p o sition . am i n ' i s a l s o u se d f o r
periphrastic partitives, as in (59).

Nihinana [ telo
( 59) a .
t a m i n 'n y
akon d r o ] a ho.
pst.AT.eat three p st.P.gen.det banana
1s g ( nom)
'I ate three of the bananas.'
b.
Nana s a
lovia [ ny telo t
ami n ' n y
lehila h y].
pst.AT.wash d is h
d e t three p s t.P.gen.det man
'Three of the men were washing clothes.'
Note the surface similarity between the M alagasy partitive and English:
as in (59a), the subset telo 'three' precedes the preposition and the superset
ny akondro 'the bananas' follows. I assume the structure of these partitives
to be the following:
(60)
D'
QP
Q'
telo
PP
P'
three
amin'
ny akondro
of
the bananas
The preposition amin' assigns case to the DP. Du e to semantic constraints

on partitives, the head Q
i s r e s tricted to w e a k q u a ntifiers (Jackendoff
(1977); Ladusaw (1982); Hoeksema (1984)).
125
T h i s st ructure w ill b ecome
important below.
6.1.2
Su b ject position
As pointed out b y
P e arson (1996c), in c ertain d i scourse contexts, the
subject DP can be interpreted as partitive.
( 61) a .
Nisy
akondro enina teo
pst.AT.have
ba n an a s i x
ambonin'ny
la t a batra.
pst . t here on.gen.det table
'There were six bananas on the table.'
Nohaniko
ny akondro roa.
pst. TT.eat.gen. 1 sg de t banana two

'I ate two (of the) bananas.'
Nihinana akondro
r o a aho.
pst.AT.eat banana
t w o 1 sg(nom)
'I ate two bananas.'
(61b) is felicitous in a context where (61a) had just been uttered. (61c), on
the other hand, sounds strange. A s explained by Pearson, the subject in
(61b) is specific: the sentence presupposes the existence of some bananas,
of which t w o
b a n anas f or m a s u b set . In th e s e c o n texts, t h erefore,
p artitivity a p p ears.
p rovided by
T h e s u b j ect e x p resses a s u bset; th e s u p erset i s
t h e d i scourse."
( 6 1 b ) i s equivalent t o ( 62), w h ich ha s a
periphrastic partitive.
( 62)
N i h i nana roa tamin'ny
akon d r o ah o .
pst.AT.eat two pst.P.gen.det banana
1s g (nom)
'I ate two of the bananas.'
Although interesting, this type of partitive reading will not b e d i scussed

further in this chapter.
6.1.3
Ex i stentials
The existential construction in M a lagasy has been studied most recently
by Polinsky (1994); Pearson (1996); Paul (in press). These authors point
" Similar effects arise in Turkish with accusative Case marked objects. See Enq (1991).
126
Chapter 3
out the following unusual use of the existential verb.
(63)
Misy
main t y ny akondro folo.
AT.have
bla c k de t b anana ten
'Some ofthe ten bananas are black.'

H ere again, partitivity a r ises in
t h e s u bject p osition ( compare w i t h a
"standard" existential in (61a)). This partitive contrasts with th e s ubset

partitive in (61b), however, as the subject DP in (63) sets up a superset not
a subset.
T h e s u b set a p pears t o b e r e p r esented b y t h e e m b e d d ed
predicate, in thi s case, mainty 'b l ack'. I n
co n t r a st to t h e p e r i p h rastic
partitives (see (59) and (62)), no (overt) preposition is present in (63). See
above references for more detailed discussion of existentials.
6.1.4
Ex ceptional Circumstantial Topic Marking
T he final partitive strategy w il l b e t h e f o cus of t h i s section due t o i t s

connection to CT. A
s t andard CT clause appears in (64) for reference in
the following discussion.

(64)
A n a san-dRakoto
lovi a
ny s ayony.
CT.wash.gen.Rakotodishes det soap

'The soap is used by Rakoto to wash dishes.'
Consider now ( 65):
t h e v er b b ears CT m o r p h ology, but th e t h eme n y
lovia 'the dishes' is in the subject position.

(65)
A n a san-dRakoto
amin'ny
s a vony ny loyia.
CT.wash.gen.Rakoto P.gen.det soap d e td ish

'Rakoto washes some of the dishes with the soap.'
(65) is exceptional in two respects. The theme is promoted to subject, but

the verb bears CT, not TT m o r p h ology .
partitive, a reading not
S e cond, the subject is superset
a v ailable in (64), and distinct from
t h e s u b set
subject partitive mentioned in 6.1.2. I will therefore refer to constructions

like (65) as Exceptional Circumstantial Topic Marking (ECTM).
M a l agasy
ECTM is possible with a wide range of transitive verbs (e.g. activities: maka
127
'take', mandoko 'paint', m fi oka 'smoke', mamono 'kill'; s tates: mitia 'l i ke',
mahafantatra 'know')." I n k e eping with the above descriptions, I point out

that the subject DP expresses a superset. The subset is not expressed as an
overt DP. I n t his way, the ECTM construction resembles the existential:
compare (63) w it h
I mportantly, t h e
( 6 5) . I n bo t h c a s es, th e s u b ject i s t h e s u p e rset.

p a r t i t iv e r e ading i n ( 6 5 ) a r i ses w i t h ou t a n y o v e r t
marking, such as a preposition.

S umming up, t h ere are f ou r
Malagasy. Fo r
s t r ategies for ex pressing partitivity i n
t h e r e m ainder o f t h i s section, I c o ncentrate on E C TM ,
arguing in favour of a QP theme."

6.2 ECTM: An analysis
T he ECTM
c o n struction h a s n o t r e c eived a n y a n a l y sis w i t h i n t h e
g enerative literature as fa r a s
I a m a w a r e . T h e e x ac t s t atus o f t h i s
construction is the subject of some d i sagreement w i t hin th e

Malagasy grammars.
t r aditional
F o r e x a m p le, according t o R a h ajarizafy ( 1960),
Malzac (1960) and Rajaona (1972), examples such as (65) illustrate one of
the normal uses of circumstantial voice. For these grammarians, 'part of'
is therefore comparable to the various oblique relations of circumstantials.
R ajemisa-Raolison (1966), however, lists the p artitive use of C T
as an
exception. In the following section, I will examine ECTM in m or e d etail

and evaluate the traditional claims. In fact, as will be discussed below, I
conclude thatboth are correct: ECTM is a normal CT construction, with a
special twist.
R ecall my analysis of CT :
s tructurally C a se-marked D P .
p r o m o t io n t o s u b ject of a n y t h ing bu t a
I ther e f or e p o s i t t h a t i n a n EC T M
construction, the theme is generated as a QP. The tree in (66) shows the
structure underlying (65), omitting the i n strumental PP
(amin'ny savony
'with the soap') for simplicity."

" There is some speaker variation in the precise range of verbs that allow ECTM.
In fact, the partitive reading can arise in other positions (e.g. goal). T h ese examples ate
therefore ambiguous since CT normally promotes goals (see section 3.3).
vary ny yahiny.
(i)
Nandr o soan-dRakoto
pst.CT.serve.gen.Rakoto r i ce det guest
'The guests were served rice by Rakoto.'
or 'Some of the guests were served rice by Rakoto.'
For the purposes of this discussion, I concentrate on themes, as they constitute the most
surprising instance of CT.
" I n the tree in (66), the QP theme is projected in [Spec, VP]. N o t hing in this analysis hinges
on this particular position, as opposed to the sister to V, for example. I also assume the partitive
128
Chapter 3
(66)
I'
vP
VP
Rakoto v'
QP
Q'
PP
P'
DP s a sa
wash
ny lovia
the dishes
In (66) the verb takes a QP complement. The head of this QP is an abstract
quantifier corresponding t o
' s o me ' ( c rucially, a w e a k q u a n tifier) .
c ertain contexts, the pr eposition i n
t h e P P c o m p l ement t o Q
o vertly realized, as we w ill see below .
In
ca n b e
T h e D P c o m p l ement t o t h e P
raises to [Spec, IP] for n om inative Case. T h e fi nal structure is given in
(67)
theme to be a bare QP rather than a full partitive DP (cf. (60}}. Extraction out of DPs is not in
general possible.
(i} *
[Momban'ny alikaji n o namakian-dRasoa
(ny} boky t;.
about.gen.det dog
foc p s t.CT.read.gen.Rasoa (det} book
'It's about dogs that Rasoa read a book.'
129
(67)
D
ny lovia
the dishes
anasana DP
wash
Rakoto
v'
QP
Q'
P
P'
DP
I now turn to some facts that provide evidence in support of th e abov e

analysis. The crucial characteristic of my analysis of ECTM is the presence
of a QP theme (i.e. both category and position are important)."
6.3 Data
In the following subsections, I argue for th e p resence of an u n d erlying
quantifier in the ECTM construction.
F u r t h er, I show t hat the p artitive
subject originates in a theme position

6.3.1
Pr e positions and partitivity
T he proposed structure has a Q P t h eme, w h er e t h e h ea d o f t h i s Q P

e xpresses partitivity o r
' s om e o f' . M a l a g asy d oes no t h a v e a n o v e r t
quantifier that corresponds to this meaning; I nevertheless assume a null

weak quantifier to be present in these structures in order t o account fo r
the interpretation. Recall that the preposition amin' is required to assign
The above analysis bears some superficial similarity to Kayne's account of de NP in French
(Kayne (1975)). He posits an underlying null head [@ de carottesj that corresponds to a quantifier.
Importantly for Kayne, however, this is not a partitive NP as partitives and de NPs do not share
the same distribution.
130
Chapter 3
Case to the DP, along the lines of regular partitives discussed in section
6.1.1. This is illustrated in (68a) below. We have already seen that when a
P P is promoted to subject with CT, the preposition may b e o v ert i n t h e
cleft position (an illustrative example is given in (68c)). This is also true in
ECTM, as illustrated in (68b). A s w i t h m ost C T clefts, the preposition is
optional in both (68b,c).

( 68) a .
Manasa [ p
a m i n 'n y lo v i a ] Rakoto.
AT.wash
P.ge n .det d is h Ra k o to
'Rakoto washes some of the dishes'
[ (Amin')ny lovia ]
no a n asan-dRakoto.
( P.gen.)det dish
foc CT . w ash.gen.Rakoto
'Rakoto washes some of the dishes'
[ (Amin')ny
s a vony ] n o a n asan-dRakoto
( P.gen.)det soa p
lovi a .
foc CT . w ash.gen.Rakotodish
'It is with soap that Rakoto washes clothes.'
(68b) shows that the partitive DP in ECTM originates as the complement

o f a preposition. I
l i n k t h e p r esence of t hi s p r eposition t o t h e u s e o f
prepositions in periphrastic partitives, discussed earlier.

6.3.2 Theme or nota theme?
I now examine the theme status of the partitive subject in ECTM .
note that the verbs that are compatible with E CTM
t ransitives.
F i rst
ar e no t " o p t i onal"
In oth e r w o r d , t h e t h e m e a r g u m en t m u s t b e ov e r t l y
expressed.
(69) * Manasa
R a k oto.
AT.wash Rakoto
'Rakoto washes.'
If the argument structure of the verb is to be satisfied, there must be some

element in the theme position.
A s a second test for theme-hood, I turn to secondary predication. T h e
131
data in (70) illustrate that depictive secondary predicates are possible in
AT, TT, CT and in ECTM.
( 70) a .
m an g atsiaka ny kafe
Misotro
AT.drink c old
R a s oa.
det coffee Rasoa
'Rasoa drinks coffee cold.'

b.
Sotr o i n y
mangatsiaka ny kafe.
T T.drink.3(gen) c o l d
det coffee
'She drinkscoffee cold.'
[Ao
a n - dakozia] no isotroany
there acc-kitchen foc CT.drink.3(gen) cold

'It's in the kitchen that she drinks coffee cold.'
det coffee
~ 'It's in the kitchen cold that she drinks coffee.'
d.
I sot r oany
C T.drink.3(gen) c o l d
det coffee
'She drinks some of the coffee cold.'

Note that in the examples in (70), the secondary p r edicate, mangatsiaka
'cold', is always predicated of the theme of the verb. T hi s restriction can
b e accounted for by W i l l i ams (1980)'s mutual c-command condition o n
predication, if we assume the secondary predicate to be adjoined to V' at
the (lower) VP-level (Baker (1997)).
132
Chapter 3
(71)
ny kafe
AP
V'
mangatsiaka
the coffee
cold
sotro
drink
The grammaticality of (70d) therefore suggests that the partitive subject in
a n ECTM c onstruction ha s t h eme s t atus.
W h a t i s c r u cial i s n o t t h e
particular analysis of secondary predication. I n stead, it is important that

p artitive subjects in E CT M
c o n structions p a ttern w i t h t h e m es, a s i n
(70a,b), and not with adjuncts (cf. (70c)).

As further support fo r
g e nerating th e p a r titive subject as a t h eme,
c onsider the following examples. (72) reveals that the theme role of t h e
verb is saturated by the partitive DP and cannot be overtly expressed with
an independent DP. In other words, both the superset and the subset of a
partitive cannot cooccur in ECTM.
( 72)
a. *
Nihinanako
(ny)
mainty ny akondro.
pst.CT.eat. 1 sg(gen) de t
bla c k d e t banana
'Of the bananas, I ate some black ones.'
b. *
Nihininako
akondro n y voankazo.
pst.CT.eat.1 sg(gen) b anana

'Of the fruit, I ate some bananas.'
133
de tf r uit
The data in (72) indicate that the partitive DP blocks the p r esence of a
theme. T h e p r o posed structure in (66) accounts for these cooccurrence
restrictions by generating a QP in theme position. Clearly, another theme
cannot be projected.
6.4 Floating quanti fiers: a problem?
In (72), we saw that in ECTM constructions, the theme position cannot be
occupied by an independent nominal. Interestingly, there are cases where
the theme position can apparently be filled by a numeral.'" I s u ggest that
this numeral is the overt realization of the head of QP. This is illustrated in
(73a), an ECTM construction ( c f . ny l ovia telo ' the three d ishes'). S u c h
"floating quantifiers" are ungrammatical in all other voices: TT (73b); AT
(73c); and CT (73d).

A nasan-dRakoto
( 73) a .
telo
CT.wash.gen.Rakoto three
ny lo y i a .
d e t dishes
'Rakoto washes three of the dishes.'

b. *
Sasan-dRakoto
telo
TT.wash.gen.Rakoto three
n y l o yia.
d et dishes
'Three dishes are washed by Rakoto.'
c.
Manasa t e l o
lov i a n y a n kizy.
AT.wash three
d i shes det children
'The three children wash dishes.'
d.
Nihirako
telo
n y a nkizy.
pst.CT.sing. 1 sg(gen) three det children

'I sang for the three children.'
There arise two questions about the above data. F i r st, wh y ar e fl oating
'" As well as numerals, the weak quantifiers betsaka 'many' and vitsy 'few' may appear in this
"floated" position in ECTM constructions.
(i)
A nasa n -dRakoto
bets a k a n y loyia.
CT.wash.gen.Rakoto many d e tdish
'Rakoto washes many of the dishes.'
134
Chapter 3
quantifiers only available in ECTM constructions?'" Second, what is the
position of the floating quantifier?
To better understand the data, compare the examples in (74).
( 74) a .
Anasan-dRakoto
telo
CT.wash.gen.Rakoto three
ny J o y i a.
d e t dish
'Rakoto washes three of the dishes.'
Anasan-dRakoto
ny loyia telo.
CT.wash.gen.Rakoto det dish three

'Rakoto washes some of the three dishes'
W hether the n u m eral appears w i t hi n t h e D P o r " floated" a f fects th e
meaning of the sentence. (74a) is not equivalent to (74b). In the form er,
the floating quantifier specifies the cardinality of the set of washed dishes
(='t hree of the dishes'). In the latter, the numeral specifies the cardinality
of the set of dishes under consideration (but not the washed ones).
To solve this puzzle, I propose that the weak quantifier in (74a) is an
overt realization of the head of QP .
T h e D P h a s r aised to t h e s u bject
position, stranding the Q telo 'three'.
" There are also strong "floating quantifiers" in Malagasy, which quantify strictly over the subject
position. Two examples are given below.
(i) a .
No h a nin'ny m p i anatra da h o l ony yoankazo.
p st.TT.eat.gen.det student all
det f r u i t
'The students ate all the fruit.'
b. Nihinana voankazo
da h o lony mpianatra.
al1
det s t udent
p st.AT.eat f r u i t
'All the students ate the fruit.'
Since these quantifiers are subject-sensitive, I tentatively suggest that they are subject-oriented
adverbs rather than truly "floated" quantifiers (Doetjes (1997)).
135
IP
ny lovia
vP
the dishes
anasana DP
wash
Rakoto
v'
t elo
P'
DP
three
The structure in (75) ensures the correct reading for the quantifier. Recall
f rom above t hat t h i s i s t h e s t r u cture o f a p a r t i t iv e D P . H e n c e , t h e
interpretation will be 'three of the dishes'.
I n (74b), on th e o t her h a nd, th e n u m eral o r i ginates within th e D P
embedded under QP. Like other nominal modifiers, the numeral appears
after the N ,
p e r h aps in the head of N u m P . T h e e n t ir e embedded D P
raises to subject position and the null Q i s interpreted as 'some'.
136
Chapter 3
(76)
QP
p'
Dp
ny lovia telo
the dishes three
From th e st ructures in ( 75) and (76), w e
c a n see t ha t t h er e ar e t w o
p ositions for numerals in Malagasy: p r e n ominal and postnominal.
The
following example illustrates that it is possible for both these positions to

be filled.
(77)
A n a san-dRakoto
telo
ny lo y i a folo.
CT.wash.gen.Rakoto three det dish ten

'Rakoto washes three of the ten dishes.'
In (77), telo 'three' is the Q and folo 'ten' is Num .

T his structure also correctly r u les out " q u a ntifier f l oat" f r o m o t h e r
D Ps. The numeral will only be stranded when a DP embedded in a QP
moves to the subject position. In the case of regular themes, for example,
an entire partitive DP (which includes the QP) moves to subject position
and no stranding will be possible. Finally, note that the only element that
can appear in the "object" position of an ECTM construction is a numeral.
DPs or N s such as in (72) above will not be generated in the head of QP.
Thus the floating quantifier data do not create a problem for the proposed
a nalysis. I n
f a ct, they are evidence in f av our o f t h e u n d e rlying Q P i n
ECTM.
6.5 Malagasy madness?
The data f r o m
E C T M i l l u s trate a c l ear m i s m atch b e t ween t h e m atic
features and voice. D r aw ing on similar data from T agalog and Inibaloi,
both Ballard (1974) and Foley (1976) argue against a
approach to voice.
t h e t a-agreement
T h us, a range of A u s tronesian languages require a
137
structural approach to voice. The Tagalog and Inibaloi data are discussed
directly below, but I w ill not offer an explicit analysis of these languages.
A lthough the d ata are s i m ilar t o
M a l a gasy ECTM, I b e l i eve t ha t t h e
u nderlying structures are different.
N e v e rtheless, the conclusion to b e
d rawn from t h ese data is the same:
v o i c e m o r p h ology d oes no t m a p
directly to particular theta-roles, contra the theta-agreement analysis. W e

w ill see
b e lo w t h a t t h e s e l a n g uages us e v o i c e t o s i g n a l s h i ft s i n
interpretation. Th e exact meaning change is different in each language,

but this is not surprising given the d i fferent range of C T -like v oices in
P hilippine languages. I n t erestingly, similar shifts can be observed w i t h
partitive Case marking in Finno-Ugric, to be discussed in section 6.5.3.
6.5.1 Tagalog
Unlike Malagasy, Tagalog makes fine-grained distinctions among adjuncts
in verbal morphology: e.g. Benefactive, Locative, Instrumental. O f t hese,
the Locative is used for
t h e e q u ivalent o f t h e E C T M c o n struction, as
shown in (78b). (78a) is the Theme Topic (TT) counterpart. (Data from
Foley (1976).)
(78) a .
Bina s a
ng l a l ake a ng libro.
TT.read
g e n man n o m book
'The man read the book.'

Binasahan ng lalake ang libro.
LT.read
ge n man n o m book
'The man readfrom the book.'

T he partitive subject in (78b) is understood as the argument of a P ,
as
encoded by the verbal morphology. M o r e over, the Tagalog data suggest

that this preposition is locative in nature.
A s i n M a l a gasy, the n o r m al
"passive" (TT) is possible with definites only. I f th e theme DP is partially

affected by the action of the verb, an "adjunct" voice is used, in this case
locative voice. The Tagalog data thus closely resemble the Malagasy.
6.5.2
In i b aloi
Ballard (1974) discusses the different verbal affixes that appear in Inibaloi,
another Philippine language. He notes that a single verb may appear with
different affixes but with no apparent grammatical change in the clause.
138
Chapter 3
( 79) a .
Bedatbaten
t o 'y
pi n g k an.
l ine-up
3sg
plate s
'She will line up the plates.'

b.
I bal a tbat t o ' y
pi n g k an.
l ine-up
plat e s
3sg
'She will line up the plates.'

I n both cases, the theme o f t h e v e rb, p ingkan 'plates', appears in t h e
subject position. But the verbal affixes differ. (79a) uses the usual object
focus (patient oriented) marker -en, while the i- prefix in (79b) is a locative
marker. Ballard notes a semantic shift between the two examples.
d escribes the r esult o b tained: all t h e
(79a)
p l a tes w er e a f fected. I n (7 9 b ) ,
h owever, it is the position o f t h e p l a tes that is being affected. I n t h e

former, therefore, we have a typical passive construction, with no special
semantic effects. In the latter, the difference in meaning is not clear from
Ballard's description.
e quivalent of
H o w e v er, if w e t h i n k o f ( 7 9b) as expressing the
' Sh e w i l l d o l i n i n g-up o f t h e p l a t es', th e p a r t i t iv e ( o r
indefinite) reading emerges. In o t her w o r ds, the focus is on the p r ocess

described by the verb rather than the fact that all the plates were lined up.
Although the Inibaloi data do not align perfectly w it h th e
and Tagalog ECTM
M a l agasy
d i scussed above, I b e l i eve t h ere i s a n i m p o r t ant
s imilarity between the use of v o ice in these languages. A l l

regular "passive" to p r o m ote a t h eme to subject.
t h ree have
A l l h av e other v o i ce
alternations available, usually reserved for oblique elements.
A n d t h e se
l anguages can use these oblique voices for themes with similar (but n o t
i dentical) s emantic
m orphology i s no t
e f f ects .
Im por t a n t ly , t h e di f f e r ence i n v oi c e
m a t ched by a d i f f e rence in t h e m atic r o le, pe r s e .
Hence, we have further evidence that theta-agreement cannot account for

the full range of voice data in Austronesian. Further, the data suggest that
voice, in particular CT-like voice, may require distinct analysis in different
languages. I address this question briefly in section 7.
6.5.3
Pa r titive Case
Interestingly, partitive Case in Finnish and Mordvinian (both Finno-Ugric
139
languages) shows a pattern similar to the exceptional Austronesian voice,
but in a slightly different way (Kiparsky (1995)). In Finnish, partitive Case
has both a DP-related function and an aspectual function. Partitive marks
quantitatively indeterminate DPs (independent of aspect) and objects of
atelic predicates. (80a) is an example of a telic predicate with a bare plural
object, marked with partitive Case. (80b) illustrates that the object of an
atelic
p r e d i cate i s m ar k e d
w ith
p art i t i ve , i r r e s pective of
the
quantificational properties of the object.
( 80) a .
Saa-n karhu-j-a.
get- 1 sg bear-pl-part
'I'll get bears.'
Etsi-n
karh u -j-a.
seek- 1 sg bear-pl-part
'I'm looking for (the) bears.'
The Mordvinian partitive Case ending is cognate with the Finnish, but is
p urely DP-related.
I t o c c urs on i n d efinite bare p l urals o r m a s s n o u n
objects, as shown in (81). (Data from Kiparsky (1995).)

(81)
vina-do.
M o n , a d a, sim-t-tan
ok fine drink-cause- I sgsubj/2sgobj vodka-part

'Ok, so I'll let you drink vodka.'
In other words, Mordvinian partitive Case is restricted to partitive objects.
It does not mark the objects of atelic verbs. A l t h ough partitive Case has
b oth aspectual (VP) and D P
f u n c tions i n F i n n ish, i t o n l y h a s t h e D P
function in Mordvinian. In Malagasy, the exceptional voice (CT) is used to

mark partitivity, a DP function. In Inibaloi, the exceptional voice is used to
mark a shift in aspect, a VP function.
6.6 Concl usion
This section has been a "case study" in how th e proposed analysis of CT
accounts for a very particular use of this voice, ECTM. I h ave shown that
t he unusual p a r titive r e ading i n C T c o n s t ructions a r ises du e t o a n
140
Chapter 3
u nderlying QP in theme position. The quantifier encodes partitivity. T h e
DP complement to this Q i s p r om oted to subject via CT because it is not
structurally Case-marked by the verb. I n t erestingly, other languages use
voice to mark similar effects.
7
C i r c u mstantial To i c : Other lan u a e s
In the discussion of passive in chapter 2, it was suggested that the different

passive types are sensitive to the different structural positions of internal
arguments. The a- passive targets arguments in [Spec, v2P] while the -Vna
passive promotes DP arguments in VP. T hus voice in M alagasy appears
t o make fine-grained distinctions between syntactic positions. O n ce w e
look at circumstantial topic, however, these distinctions disappear. Recall
that circumstantial topic is used t o
p r o m ot e a n y t h ing b u t s t r u cturally
Case-marked DPs to subject. This heterogeneous class includes adjuncts

and objects of prepositions. C l e arly, these elements are generated in a
range of structural positions.
A m o n g a d juncts, it is often assumed that
there are different levels of adjunction VP, IP, etc. M o r eover, as shown
in section 2.2, adjuncts are of differing categorial status: adverbs (heads),
PPs, CPs, VPs. Yet all adjuncts use the one CT. Similarly, I have provided
arguments that ECTM inv olves a partitive theme.
T h e b ase position of
the subject in an ECTM construction is thus in the canonical direct object

position, not an adjunct. In sum, CT is much more of an all-encompassing
voice alternation.
In c o n t rast,
s o m e A u s t r o nesian l a n g u ages m a k e fi n e r - grained
distinctions a m on g
a d j u n cts .
For
b enefactive and instrumental voices."
exa m p l e , T a g alog h a s l o c a tive,

T h e d at a below ar e f r o m F o l e y
(I976).
(82)
a.
Ti nir h a n n g l a l ake ang bukid.

LT.buy
g e n man n o m farm
'The man lived on the farm.'
The locative voice is used for locations, sources and goals (and partitives). Foley notes that
i nstrumental focus clauses are grammatical but r are .
A c c o rding t o K r o eger ( 1990), t he
instrumental voice in Kimaragang Dusun is mainly used in extraction contexts (relative clauses,
clefts, wh questions). This is probably due to pragmatic constraints against having an instrument
as the subject.
141
Ibinili
ng la l ak e n g i sd a a n g b ata.
BT.buy gen man acc fish nom child

'The man bought some fish for the child.'
Ipinutol
n g l a l ake ng isd a a n g k u tsilyo.
IT.cut
gen m a n a c c fish n o mk nife
'The man cut the fish with the knife.'
I n Tagalog and other Philippine languages, voice morphology no t
only
indicates the presence of a preposition, but also encodes the nature of that
p reposition.
In the s e l a n g u ages, t h e v o i c e m o r p h o l og y a n d the
underlying preposition are directly linked, unlike what we have seen for
Malagasy. It therefore appears that the preposition-incorporation analysis
is appropriate for Philippine languages. For these reasons, I suggest that
the Philippine prepositional voices are a form of applicative. The range of
"adjuncts" that can be promoted to subject is limited, much in the same
way as Bantu applicatives, as mentioned in section 4."
T he discussion of M a l agasy voice (TT i n c h apter 2 a n d C T i n t h e
p resent ch apter)
h a s i n d i cated t h a t i n or d e r to unde r s tand v o i c e
a lternations in a particular language, an in-depth study is necessary. F o r

example, it was only by considering the full range of passive constructions
that an analysis of CT wa s p ossible.
T h e r efore, m y c o n clusions about
Tagalog are tentative in nature. Th e Tagalog equivalents of CT given in

( 82) may i n f act r equire a n e v en t s t r u cture analysis, as p r oposed b y
Naumann and Latrouite (1999). I leave for future research a comparison
of v oice sy stems w i t hi n
w e s tern A u s t r onesian. T h e p r e s ent t h e sis
indicates the direction that such a comparative study would take.

8
C o n c l u sion
In this chapter, I have examined the circumstantial topic construction in

Malagasy. From the perspective of English, this voice is highly u nusual:
in many cases, an adjunct is promoted to subject. I
e x p l o red a r ange of
d ata which il lustrated th e h e terogeneous qu ality o f
CT.
One o f th e
One common criticism of the P-incorporation analysis of Bantu applicative is that despite
different prepositions, the applicative morphology is c onstant. T a g alog presents the reverse
situation. A single preposition, sn, is used to mark elements which will be promoted to subject
with distinct voices.
142
Chapter 3
p rincipal challenges of any analysis of CT lies in accounting for this w i d e
range. I pointed out two main approaches to CT that have been proposed
in the literature. T h e f i rst li nks CT t o adjuncts; the second exploits the
connection between CT and p r epositions. D e spite the attractiveness of
t hese analyses, I showed that neither accounts for th e f ul l a r ray o f
CT
clauses. Under my analysis, CTis an " elsewhere" voice. In a CTclause, all

the arguments of the verb r eceive Case. S i nce M alagasy has a st r ong
requirement that the subject position be filled, any element other than a
structurally Case-marked DP is promoted to subject.
O n e a d vantage of
this analysis is that it easily accounts for what appears to be an exceptional

use of CT, the ECTM construction.
We are now in a position to compare CT with TT, which was discussed
in chapter 2. In chapter 2, I showed that two passive affixes, a- and -Vna,
promote distinct arguments. I a r gued that a- promotes DPs from
v2P], while -V na p r o m o tes D Ps f r om t h e l o w e r V P .
[Spec,
The fo r m e r i s
therefore sensitive to position and the latter to domain. Th e discussion of

C T has shown that this voice is again different from TT .
C T d oes not
target a particular position nor is it restricted to a particular domain. Since

adjuncts are projected at various levels in the ph rase structure, CT can
promote an element from almost any structural position.
In both the present chapter and chapter 2, I have argued that voice can
s erve as a p r obe i nt o p h r ase structure. I n p a r t i cular, I l o o ked a t t h e
p osition of internal arguments and the argument-adjunct distinction. A l l
o f these elements are situated within the extended verbal projection.
In
t he next chapter, I t u r n m y a t t e ntion t o c o m p l ementizer l ayer o f t h e

clause. Since I often use the cleft construction as a test for structure, in
c hapter 4 I will provide an explicit analysis of the syntax of clefts. I
will
also contrast clefts with topicalization constructions. In the same way that
t he voice system "feeds" A -bar m ovement, chapters 2 and 3 set up t h e
necessary background to the discussion of topic and focus in chapter 4.
143
Chapter 4: The left periphery

1 I n t roduction
The previous chapters have investigated the projection of arguments and
adjuncts in the VP and IP layers. I
h a v e p r i m arily f ocussed on the base
position of these elements and the im p lications for v oice alternations. I

now turn my attention to cases where arguments and adjuncts appear in
the left periphery. W e h ave already touched on the cleft strategy, here I
will provide an analysis. I wi ll c onsider the Malagasy data in light of t h e
" split CP" h y p othesis of Rizzi (1997). I w i l l b egin w i t h an o v erview o f
Rizzi's proposal and how it applies to Malagasy topic and focus. I w ill be
concerned with the relative position of these elements and conclude that
like Italian, Malagasy has topic>focus>topic ordering.
In Malagasy, the cleft construction is associated with a particular focus
interpretation. S i nce the cleft has been im p o rtant i n
p r e v i ous chapters
(especially for the discussion of circumstantial topic in chapter 3), the core
of the present chapter is devoted to the cleft construction.
F i r st, I show
that the cleft in Malagasy is an equative construction: th e clefted element

is a predicate and a headless relative is the subject. N ext, I argue that the
clefted element, which carries a special focus interpretation, appears in the
specifier of a
f u n c tional p r ojection, FocusP. I th e n a d d r ess apparent
multiple clefts and show
t h a t t h ese i n v o lv e t w o s e p arate p r ojections:
F ocusP and TopicP, as proposed in the first part of th e chapter. I
argue
against other possible analyses of m u l t iple clefts, namely coordination,

amalgamation and multiple specifiers. Finally, I look at some interesting
properties of multiple wh-movement.
2
T h e left e r i h e
Drawing mainly on d ata from
I t a lian, French and English, Rizzi (1997)
a rgues for the following projections in the "complementizer layer" of t h e

clause. (All the "Comp" projections are in bold.)
Chapter 4
VP
ForceP
TopicP*
FocusP
TopicP*
FinP
IP
Roughly speaking, the complementizer system connects a p r o p osition

( the lower IP ) and a h i g her clause (or d i s course, in t h e c ase o f r o o t
clauses). The matrix verb selects a clause of a particular type or
( e.g. declarative, interrogative). A t
" f o r ce"
t h e o t her en d o f t h e C P l a y er, t h e
c omplementizer s e lects a n I P of a cer t a i n te n s e s p e cification o r

"finiteness". For Rizzi, the force and finiteness distinctions are encoded on
heads of
s e p arate p r ojections, ForceP an d F i n i teP, r espectively.
In
between, we have topic and focus positions. I w ill be concerned with the
topic and focus projections in Malagasy and will not address the issue of
the different Comp-like heads. Rizzi's "splitting" of Comp i n t o d i f f erent
projections parallels work o n I nfl b y P o l l ock (1989) and the VP-shell of
Larson (1988). This body of w o r k a t t empts to r epresent finer semantic,
syntactic and morphological distinctions in phrase structure.
As shown in the above tree, Rizzi argues for two topic positions, which
sandwich the focus projection.
T h e f o l l ow ing Italian example illustrates
this ordering.
( 2)
[ > , , A Gianni], [fo,,

to Gianni
Q U E STO], [>,, domani], g l i d ovrete dire.

tomorrow him should tell
this
'To Gianni, THIS, tomorrow, you should tell him.'
145
The le
ft periphery
Moreover, Rizzi claims that topics may iterate (marked by a * on e a c h

TopicP), although focus is limited to a single position.' T h e M alagasy data
will support Rizzi's proposal, for
M a l agasy also has t o pi c) f ocus)topic
word order in the left periphery. In the remainder of this section, I discuss
topic and focus, concentrating on p ositional facts. I n
sections 3-5, I will
look in more detail at the structure of the focus construction. I
l e ave a
proper study of topicalization for further research.
2.1 Topic andfocus

This is not the place to review th e l arge body o f l i t erature on topic and
focus. Instead, I w ill ou tline some basic characteristics of these notions;
see Rooth (1996) for an overview of focus and Erteschik-Shir (1997) for an
a ccount of both topic and focus in terms of " f ocus structure".
Following
J ackendoff (1972), I assume that f o cus can be d e fined in t e rm s o f t h e

d iscourse notion of p r esupposition.
P r esuppositions are what b ot h t h e
speaker and hearer assume to be the case at the point when the sentence
i s uttered.
sentence. I n
F o c u s c o r responds t o t h e n o n - p resupposed p ar t o f t h e
o t her w o r ds, focussed elements are in some sense "n ew "
information. A s i m ple sentence is ambiguous; depending on the context,

it can have different presuppositions. I t ake the following example fr om
Zubizarreta (1998) to illustrate this point. In each of (3a,b,c,d) we have the
same sentence that i s
a s signed d i f f erent f o cu s i n t erpretations.
The
focussed constituent is marked wi th [F]. A constituent not marked [F] is

interpreted as the presupposition or as part of the presupposition.
E a ch
example is associated with an (implicit) context question.
(3)
a.
[F John [ate [the pie]]]].

[What happened?]
[ John [F ate [the pie]]]].
[What did John do?]
[ John [ate [F the pie]]]].
[What did John eat?]
' The limit on focus derives from semantic restrictions. See Rizzi (1997) and Zubizarreta (1998)
for some discussion. This restriction will be important when we turn to apparent multiple foci in
section 5.
146
Chapter 4
[[FJohn] [ate [ the pie]]]].

[Who ate the pie?]
The examples in
(3 ) d o n o t e x h a ust al l t h e d i f f e rent p o ssible f o cus
readings, but illustrate a few.

Topics, on the other hand, are what a statement is about. T hey are old
information that is contextually salient. The topic is set off from the rest of
the clause: th e comment.
T h e c o m m ent i s predicated of the topic and
introduces new i n f o rmation .
U s i n g t h e t e r m i n ology o f E r t eschik-Shir
(1997) (who adopts Reinhart (1981)'s file metaphor), a topic signals the
l istener to locate an existing card and add the new i n f o rmation f rom t h e
comment to that card.
Before continuing, I note that in this chapter I will solely be concerned
w ith topic and f o cus as associated with d i splaced elements.
I n ot h e r
words, I w i l l i g n ore th e f o cus interpretation that a rises w it h p r o sodic

p rominence. I
w i l l a lso not i n v estigate the t opic-like properties of t h e
clause-final subject position (see Pearson (1996a)). I n stead, I concentrate

on th e
c l ause-initial p o sitions r eserved f o r t o p i c alized an d f o c u ssed
elements. I n
t h i s w ay, M a lagasy resembles English, Italian, H u ngarian
and many other languages. In the next sections, I look at the hierarchical
relations among these positions.
2.2 Topicofocus
As discussed by Keenan (1976), Malagasy has a topic position, which is
followed by dia, and a focus position, which is followed by no, at the left
edge of the clause. In the following examples, we see that topics precede
f ocussed elements and that the inverse order is ungrammatical. I n
(4a),
the topicalized element is the object of the verb and hence is "doubled" by
the resumptive pronoun azy (see below for discussion of the resumptive
pronoun strategy).' '
' I translate Malagasy topicalization with English left dislocation and with "as for", depending on
which seems the most natural in context. I make no claims about the relation between Malagasy
and English topicalization, however.
D ue to the complex structure of topic and focus constructions, in this chapter, I wil l n o t
underline the subject. I n stead, where necessary, I will use square brackets to indicate relevant
constituents.
147
The le
ft periphery
(4)
a.
[Ity radara ity]pic dia [ny Rosiana]fo,, no nanao
this radar this
top d e t Russian
azy.
foc p s t .AT.do 3(acc)
'As for this radar, it's the Russians who made it.'
b. *
[ N y Rosiana]fo,, no [ity radara ity]pic dia nanao

det Russian
foc th i s radar this
azy.
top p s t.AT.do 3(acc)
'It's the Russians who, this radar, made it.'
I will refer to the topic ity radara ity 'this radar' in (4a) as a "dia topic".
Drawing on similar data from M a ori, Pearce (1999) argues for a single
t opic position t hat c -commands f ocus.'
I w ill s h o w , h o w e v er, t h a t
Malagasy has a s econd t o pi c p r o jection b elow
f o c us . A n il l u s t rative
example is given below.

(5)
[Ny l o v ia maloto]pic dia [isan'andro]fo,, [Rabe]tppno manasa azy ireo.

det dish dirty
top each'day
Rabe
foc A T . w ash 3(acc) pl
'As for the dirty dishes, it's every day that Rabe washes them.'
In (5), Rabe i s in t h e l o w e r t o p i c p o s ition. B ef o r e a r g u i ng f o r th i s

a dditional TopicP, I w il l g i v e s ome b ackground o n t o p i c an d f o cus i n
Malagasy.
2.3 dia topics
I will not provide an in-depth discussion of dia topics. I n s tead, I give an
o verview o f
s o m e p a r t icularities of t o p icalization as r elevant f o r t h e
discussion of f o cus.
T h e m a i n c o n clusion i s t h at di a to p i cs occupy a
p osition structurally superior t o f o cussed elements. I al s o s h o w t h a t

topicalization can be derived either via movement or vi a base generation.
For the former, only subjects and adjuncts may be topicalized, as is usual
for extraction in M a lagasy. T h e l atter strategy allows a w i der r ange of
elements to appear in the topic position.
In passing, note that w ha t I
a m c a l ling "d ia to picalization", Keenan
(1976) refers to as "weak topicalization". H e claims that these are not true
Tagalog also has topic>focus order (Kroeger (1993)), as do Spanish (Zubizarreta (1998)) and the
Mayan languages discussed in Aissen (1992). N o ne of these languages allows a topic below
focus. Rizzi (1997: fn 17) points out that some Italian speakers find the lower topic marginal. I
return to this issue in section 5.5.
148
Chapter 4
topics of conversation, as evinced by t h eir common use w ith p l ace and
time adverbials. Strong topicalization, where raha 'if' precedes the topic,
defines a topic of i m p o rtance for the conversation.
T h e c o n trast in
(6)
illustrates that in certain cases of strong topicalization, such as (6a), there is

no "gap" corresponding to th e t opic; this is not possible with the w e ak
topicalization (dia topic) in (6b).

(6)
a.
strong topicalization
Raha ny voankazo dia tiako
if
det f ru i t
ny akondro.
top like. 1 sg(gen) det banana
'As for fruit, I like bananas.'

b.
weak topicalization
Ny voankazo dia tiako
det fruit
ny akondro.
top like.lsg(gen) detbanana
'Fruit, I like bananas.'
I do not discuss strong topicalization any further.'

T opicalization involves a f r o nted element w h ich i s f o l l owed b y
topic particle dia.'
the
Inter e s tingly, t h e t o p i c h a s d i f f e r ent p r o p e r ties
depending on whether there is a focussed constituent in the clause or not.

Looking first at simple topic structures, the range of elements is limited to
s ubjects and most adjunct PPs.
O b j ects and PP arguments cannot b e
topicalized if the verb bears active morphology.
' Keenan's strong vs. weak topicalization may be parallel to what Aissen (1992) calls external and
internal topics.
' The particle dia also occurs in the following construction, a type of pseudo-cleft.
( i)
Ny man a sa lamba dia Rako t o .
d et AT.wash cloth top
Rak o t o
'The one who is washing clothes is Rakoto.'
I leave this for further research, but suggest that in (i) a headless relative has fronted from the
subject position. Ra k oto is a nominal (equative) predicate (it can take negation, for example).
Note that the interpretation of (i) is parallel to a cleft construction.
(ii)
Rako t o no m anasa l a m b a.
Rakoto foc AT.wash cloth
'It's Rakoto who is washing clothes.'
See below for more discussion of clefts. T h e difference between (i) and (ii) deserves further
investigation. Clearly unrelated to topicalization are the uses of dia as a coordinating conjunction,
a temporaladverb meaning 'and then',and a superlative marker.
149
The le
ft periphery
(7)
a.
Rasoa dia nanapaka bozaka omaly.

Rasoa top pst.AT.cut grass yesterday
'As for Rasoa, she cut grass yesterday.'
b.
O m aly
d ia nanapaka bozaka Rasoa.
yesterday top pst.AT.cut grass Rasoa

'Yesterday, Rasoa cut grass.'
c.
T a m in'ny
a n tsy dia nanapaka bozaka Rasoa.
pst.P.gen.det knife
t o p pst.AT.cut grass Rasoa
'With the knife, Rasoa cut grass.'

d. *
Ity radara ity dia nanao
hain g ana ( a zy )
this radar this top pst.AT.do quickly
R a s oa.
(3(acc)) Rasoa
'This radar, Rasoa built (it) quickly.'

e. *
Hoan'ny v a hiny dia nanolotra
vary
f or.gen.det guest top pst.AT.offer r i c e

'To the guests, Rakoto offered rice.'
These data
R a k oto.
Rak o t o
i n d i cate t h a t t o p i calization s h ow s t h e us u a l e x t r action
restrictions, common to many western Austronesian languages.

If there is a focus in the clause, however, topicalization is much less
restricted. (8) illustrates topicalization of a subject, object, benefactive and
instrumental.'
( 8) also shows that w i t h t o p i c>focus, the topic m ust b e
doubled by a resumptive pronoun unless it is the subject ((8a)).
Recalling the discussion in chapter 2, the question arises whether material themes and
instruments topicalize from the DP position or the PP position. ( 8d) shows topicalization from
the PP position, which seems to be the only possibility. S i m i lar results obtain for material
themes.
a zy
ny tr a n o .
(i) a . * Ny bozaka d i a ny mpiasa no nanafo
det grass
top det worker foc pst.AT.roof 3(acc) d e t roof
'As for the grass, it's the workers who roofed it onto the house.'
b. Ny bozaka dia ny mpiasano nanafo
n y trano
tami n y .
det grass
top det worker foc pst.AT.roof det roof
p st .P . 3(gen)
'As for the grass, it's the workers who roofed the house with it.'
A proper investigation of topics should explain the contrast in (i).
150
Chapter 4
(8)
a.
I Ketaka
d i a tany Betafo
n o nipetraka ( * i z y )
Ketaka
t op p s t .there Betafo
foc pst.AT.live (3(nom))
t amin'n y
t aona
p st.P.gen.det year
l as a .
g one
'As for Ketaka, it's in Betafo that she lived last year.'
Ny vary dia izaho no nanolotra
*(an'io)
hoan'ny vahiny.
det rice top 1 sg(nom) foc pst.AT.offer (acc.it)
f o r .gen.det guest
'As for the rice, it's me who offered it to the guests.'

Ny vahiny dia izaho
vary *(ho azy).
no nanolotra
d et guest top 1 sg(nom) foc pst.AT.offer
rice
(for 3(acc))
'As for the guests, it's me who offered them rice.'

d.
Ny a n t sy dia Rakoto no nandidy
ny h e n a * (taminy).
det knife top Rakoto foc pst.AT.cut d et meat
(pst.P.3(gen))
'As for the knife, it's Rakoto who cut the meat with it.'
N ote, however, that the topicalized elements are all D Ps.

grammatical, in contrast to the examples in
(9)
Tany
P P s ar e n o t
(7).
Fiana r a ntsoa dia ny mahantra no mipetraka
pst.there F i anarantsoa top det poor

amin'ny t r an o t o m boka ( t any).
foc A T .live
P.gen.det house earth

(pst.there)
'As for F, it's poor people who live in dirt houses there.'
b. *
Tamin'ny antsy
d i a Rasoa no nanapaka b ozaka (taminy).
pst.P.gen.det knife top Rasoa foc pst.AT.cut grass
(pst.P.3(gen))
'As for the knife, it's Rasoa who cut grass with it.'
Summing up, we have the following d i f ferences between a simple topic

and topic combined with focus.
151
The le
ft periphery
(10)
(11)
X P dia ....
a.
X P m u s t be "extractable" (e.g. subject or adjunct)
b.
*resumptive pronouns
c.
X P c a n be a range of categories (e.g. DP, PP)
X P dia YP no ...
a.
X P n e ed not be extractable (e.g. may be object)
b.
resumptive pronouns allowed (for objects)'
c.
X P m us t be a DP
To account for this difference, I suggest that only in the cases of simple
t opicalization is there actual m o v ement o f
t h e t o p i calized element t o
[Spec, TopicP]. Descriptively, movement to the topic position is blocked

by the presence of an X P i n
violation.
[ S pec, FocusP] a Relativized M i n imality
A s w e l l a s th e m o v ement s t r ategy, h o w ever, t h ere i s b ase
generation in [Spec, TopicP]. Base generation requires the p r esence of

r esumptive p r o n ouns
( w i t h t h e e x c eption o f s u b j ects). W i th ba s e
g eneration, the r ange o f
t o p i cs i s m u c h f r eer s i nce n o m o v e m ent i s
involved. On the other hand, since resumptive pronouns in Malagasy are

only DPs, the base generation strategy is limited to DPs. A
similar split i n
s u p erficially
t o p i calization strategies is ap parent i n E n g l i sh.
(12a)
exhibits movement ch aracteristics, while (12b) arguably i n v o l ves base

generation of the topicalized constituent, Eve (Chomsky 1977).
( 12) a .
b.
Eve, Felix really likes.

Eve, Felix really likes her.
N ote, f i n ally,
t h a t m o v e m en t i n M a l a g asy i s " preferred"
t o bas e
generation. In other words, if movement is possible (as when there is no

f ocus), base generation i s n o t a n o p t i on . H en c e b ase g eneration i s
something o f
a " l a s t r e sort" s t r ategy an d n o t a v a i lable i n s e ntences
equivalent to (12b)."
" Even DP adjuncts are never associated with resumptive pronouns.
(i) I o maraina i o di a i K etaka no nanapaka
boza k a ( * t a min'izany fotoana izany).
t his morning this top Ketaka foc pst.AT.cut
gr a s s ( pst.P.gen.that time t h a t )
'That morning, it was Ketaka who cut grass.'
This strategy is limited to topicalization, however. As far as I am aware, no other extraction
process in Malagasy allows free extraction in combination with resumptive pronouns.
152
Chapter 4
2.4 Focus
I n Malagasy, focus is expressed by a cleft. (In section 4 below, I p r o v i d e
arguments for the label " f ocus".)
fronted an d
L i k e t o p i cs, the focussed element is
f o l l o we d b y a par t i c le, i n t h i s c a se no .
topicalization, focus is one of a
A lon g w ith
n u m b e r o f m o v e m ent t r a n sformations
that are sensitive to subjects. Objects may not directly cleft. I n stead, the
object is promoted to subject with passive and then fronted for a cleft.'"
( 13)
a. *
Lamba; no
m a n asa t; amin'ny savony R ak o to.
c loth f o c
A T .w a s h P . gen.det soa p
Rakot o
'It's clothes that Rakoto washes with the soap.'
Ny lamba; no sasan-dRakoto
t;
ami n ' n y
s a v ony t;.
det cloth f o c TT.wash.gen.Rakoto P.gen.det soap

'It's the clothes that are washed by Rakoto with the soap.'
Unlike internal arguments, most adjuncts can be clefted in any voice (AT,
CT)
11 12
( 14) a .
A min'ny savony; n o
m ana s a
P .gen.det soap
A T .w a s h c l ot h
f oc
lamb a t ;
Rako t o .
Rakot o
'It's with the soap that Rakoto washes clothes.'
b.
A mi n' n y s a vony; no s a san-dRakoto t;

P.gen.det soap
ny l a mba.
foc T T . w ash.gen.Rakoto det cloth
'It's with the soap that the clothes are washed by Rakoto.'
'" The judgements for (13a,b) are not affected by the presence or absence of the determiner. (13b),
for example, is equally grammatical with or without the determiner.
" In the previous chapter we saw some exceptions to this generalization.
" Malagasy differs from other languages in Austronesian by clefting adjuncts. Most commonly
(e.g. in Tagalog, Malay), adjuncts are merely fronted without a cleft marker. This is not possible
in M a lagasy; the c l ef t m a rker i s o b l i gatory ( b u t s e e ( 3 8 ) b e lo w f o r s o m e a p parent
counterexamples).
(i)
Tany A m b o sitra
*(no) mipetraka Rasoa.
pst.there Ambositra
( foc) AT.live
Ras o a
'It's in Ambositra that Rasoa lives.'
(ii)
A i za * ( no) manasa l a m b a Ra k o to?
w here (foc) AT.wash cloth
Rak o t o
'Where is Rakoto washing clothes?'
153
The le
ft periphery
c.
A mi n' ny
sa v o ny ;
P .gen.det soap
no
anasa n - d Rakot o
lamba t ; t;.
foc
CT .was h.gen.Rakotocloth
'It's the soap that is used by Rakoto to wash the clothes.'
In (14c), the adjunct is clefting as a subject. In other w o r ds, CT prom otes

the adjunct to subject and clefting occurs from t hi s position.
C h a p ter 3
provides an analysis of CT and o f th e occurrence of th e p r eposition in

c lefts such as (14c).
N o t e t h a t w h - m o v ement i s a s u b - t yp e o f f o c u s
movement. Thus (15) is parallel in all relevant respects to (13b).

(15)
I n o na; no sasan-dRakoto t;
ami n ' n y s a v ony t;?
what foc TT.wash.gen.Rakoto P.gen.det soap

'What is washed by Rakoto with the soap?'
In many l a n guages, wh-movement i s t o
a f o c u s p o sition ( e.g. I t alian,
Hungarian). In this chapter, I will concentrate on non-wh clefts, although

data from wh clefts will be considered (see section 5.6 in particular).
Unlike topics, focussed elements are never associated with resumptive
pronouns. This restriction is illustrated in the following examples.
( 16)
a. *
Ny trano tomboka no mipetraka (aminy)
n y m ahantra.
det house earth f o cA T .sit

(P.3(gen)) det poor
'It's earth houses that poor people live in (them).'
Amin'ny trano
to m b ok a n o mipetraka (*aminy) ny mahantra.
P.gen.det house eart h

foc A T . si t ( P .3(gen)) det poor
'It's in earth houses that poor people live.'
Ny mahantra no mipetraka amin'ny trano tomboka (*izy ireo).

det poor
foc AT . si t
P.ge n .det house earth (3(nom) pl)
'It's poor people who live in earth houses.'
A similar pattern obtains in Italian:
r e sumptive clitics are obligatory f o r
topics and incompatible with focussed constituents (Rizzi (1997)). Since I

associate the presence of resumptive p r o nouns in M a l agasy w it h b a se
generation, I believe this difference between topic and focus stems from
154
Chapter 4
the fact that the latter always involves movement."
Thus far we have seen clear evidence for TopicP dominating FocusP
(see examples in (8)). What evidence is there for another topic position?
2.5 The bodyguard
As discussed by Keenan (1976), there are instances of apparent multiple
topics and multiple foci. Examples are provided in (17).
( 17) a .
Omaly i Soa dia nandoko
ny t r anony.
yesterday Soa top pst.AT.paint det house.3(gen)

'Yesterday, Soa painted her house.'
b.
O m a ly i Soa nandoko
no
n y tranony.
yesterday Soa foc pst.AT.paint det house.3(gen)
'It was yesterday that Soa painted her house.'
He refers to these as "the bodyguard condition".

(18)
B o d y guard Condition (Keenan (1976)):

when a non-subject is fronted in a cleft or by topicalization, it can optionally
be accompanied by the grammatical subject
D escriptively, subject fronting b eing p r eferred, th e adjunct can " p i g gy

back" on t h e s u bject. T h e b o d y g u ard c o ndition i s l i m i ted t o a s i n g le
adjunct and the (matrix) subject.
(17a) and (17b) in fact have different properties and I w il l ar gue that
only the latter is a bodyguard construction. (17a) is an example of iterated
t opic projections, parallel to Italian. A s i n i tial evidence in favour o f t h i s
approach, note that the order of t o p ics in (17a) may be r eversed, as in
(19a), although the result is slightly degraded. This is not possible with the
focus construction in
(17b), as illustrated in (19b), which is completely
ungrammatical.
" Rizzi (1997) accounts for this difference by claiming that focalization is truly quantificational,
but topicalization is not. Note, however, that since the Italian and the Malagasy data do not align
perfectly (e.g. neither focus nor topic induce weak cross-over, quantified DPs can be focus or
topic), I cannot adopt Rizzi's account of the distinction between topic and focus, which I leave as a
stipulation.
155
The le
ft periphery
(19)
a . ? I S o a o m al y
Soa
di a n a ndoko
ny tranony.
y e sterday top pst.AT.paint d e t house.3(gen)
'Soa, yesterday, she painted her house.'

b. *
I Soa omaly
Soa
n o nandoko n y tranony.
y e sterday foc pst.AT.paint det house.3(gen)
'It was Soa yesterday who painted her house.'

A s fu r ther
e v i d ence t hat m u l t i pl e t o p ic s ar e p o s sible, c onsider t h e
following example.
(20)
R a h ampitso i t y antsy ity d i a i Soa no hanapaka bo z aka aminy.

tomorrow
this k nife this top Soa foc fut.AT.cut
gr a s s P . 3(gen)
'Tomorrow, this knife, it's Soa who will cut grass with it.'
Thus the dia topic may b e i t e r a ted ( a lthough o n ly o n e di a pa r t icle is
p ossible) and
t h er e a r e n o p a r t i cular r e strictions ( o ther t h a n t h o s e
mentioned in 2.3) on what may appear in these positions.
For (17b), on the other hand, I suggest that the "bodyguard" (i Soa) is
i n a l o w e r t o pi c p o sition, c-commanded b y t h e f o cu s p o sition . Th e
bodyguard, unlike the dia topic, is restricted to subjects. In section 5.3.3, I
will show that the bodyguard has topic properties; for the moment, let us
a ssume this to be the case.
( 2 1 ) i l l ustrates the topic) focus)topic w o r d
order."
(21)
[ N yl ovia maloto]Top dia [isan'andro]Foc [Rabe]Tpp no manasa azy ireo.

det dish dirty
top each'day
Rabe
foc A T . w ash 3(acc) pl
'As for the dirty dishes, it's every day that Rabe washes them.'
Thus Malagasy has both a
" h i g h " t o p i c ( w i t h di a ) a nd a " l o w " t o p i c
(bodyguard), providing support for the extended CP proposed by Rizzi.

In other w o r ds, the bodyguard corresponds to a
f i xed p o sition, rather
Analyzing the bodyguard as a topic clearly points away from treating no as a focus marker, per
se, since the XP that precedes it is not always focus. I return to the problem of labelling no in
3.4.
156
Chapter 4
than being an "extra" position licensed by topic or focus. D i stributionally,
the two topic positions have very di fferent properties. The "l ow " t o pic is
always the subject. The "high" topic, on the other hand, can be filled by
a lmost any e l ement i n t h e c l ause :
sub j e ct, o bject, ob lique, a s w a s
discussed in section 2.3. Moreover, I claim that the bodyguard is restricted

t o cleft constructions.
I w i l l a d d r ess th e u n u sual p r o p erties o f t h e
bodyguard in section 5.5.

2.6 Malagasyclause-structure
The preceding sections have shown
t h a t v a r i ou s elements (e.g. topic,
f ocus) may appear in the left periphery of th e clause in M a l agasy.
As
mentioned in chapter 1, recent work on p h rase structure in Malagasy has

adopted a " p r edicate fronting" a p p r oach to
V O S o r der ( e.g. Pensalfini
(1995); Pearson (1996b); Rackowski (1998); Rackowski and Travis (to

appear)). Under this analysis, the VP (or a similar XP) raises to a specifier
position above the subject DP. For the sake of clarity, let us say that the
subject raises to [Spec, AgrP] and that the VP raises to [Spec, TP], where
TP dominates AgrSP."
C o m b i n ing this analysis with the pr oposed view
of topic and focus, we get the following picture of the Malagasy clause."
" I am using AgrP as a convenient label for the subject position.

Below I will slightly modify this tree and argue that the XP in [Spec, FocusPj is in f act a
predicate. The lower topic position is limited to subjects and, therefore, is always a DP.
157
The le
ft periphery
(22)
cp
TopicP*
(xp)
FocusP
(YP)
TopicP
(DP)
TP
VP;
AgrP
DP
( subject)
VP
In a standard declarative clause, the VP (or some extended projection) will

be in the highest (filled) specifier position. In general, movement of XPs is
driven by feature-checking requirements of heads. For example, raising
of the predicate to [Spec, TP] satisfies the EPP features of T . "
Si m i l a rly,
f ocussed and t o p icalized elements satisfy th e f e atures o f F o cu s a n d

Topic, respectively. A l on g w ith Rizzi (1997), I assume that the topic and
focus positions are not projected unless "activated".
no topics or f o cussed elements i n
T h a t is, if there are
t h e c l ause, the T o picP and F o cusP
projections will be absent.

S umming up , t h i s s ection h a s p r o v i ded a n o v e r v ie w o f t h e l e f t
p eriphery in M a l agasy. I w i l l n o w c o n sider i n m o r e d e t ail th e f o c u s
construction, which has the form of a cleft.
" See section 5 in chapter 3 for some speculations on the EPP in Malagasy.
158
Chapter 4
3 T h e structure of clefts
T his section is devoted to the structure of th e cleft in M a lagasy. A s w e
have already seen, the clefted element appears to the left of the verb and
is followed by the particle no, glossed as 'foc' for 'focus'."
( 23)
R a k oto no
man a s a la m b a.
Rakoto foc
A T . w ash c l oth
'It's Rakoto who washes clothes.'
In the discussion below, I will use the following terminology t o d escribe

clefts: the "p i v ot" (Rakotoin (23)) and the " p r esuppositional clause" (the
remainder of the sentence ) Oackendoff (1972)).
T he structure of clefts has been investigated in many languages. I w i l l
d iscuss some of the p r oposals and their r elevance to M a l agasy.
s how that M a l agasy clefts have an e quative structure:
I will
th e pivot is a
predicate and the presuppositional clause is a headless relative in subject

position.
(24)
[ p i v ot; [ [Dp presuppositional clause] t; ]]
In other words, the structure bears some similarities to a pseudo-cleft. I

begin with some previous analyses of focus constructions.
3.1 It-clefts
Chomsky (1977) proposes the following structure for English clefts, where
XP, the pivot, is adjoined to CP.
(25)
a.
It is [ c p [this book] [cp OP that I really like t ]]
'" In fact, as I will argue below, no is not a focus marker, per se. However, since it is always
present in focus constructions, the label 'foc' seems most appropriate at this point.
159
The le
ft periphery
x p;
cp
that
Alternatively, th e
p i v o t c o ul d a p p ear i n t h e s p e cifier o f a s e p a r ate
projection, FocusP. T his has been suggested for I talian and H u n g arian
focus constructions (Rizzi (1997); E. Kiss (1998)).'"

(26)
FocusP
XP;
Foc'
CP
C'
T he difference between (25) and (26) lies in th e p o sition o f
the pivot:
adjoined or in a specifier position. M oreover, both Rizzi and E. Kiss argue

for d i rect A - bar m o v ement o f t h e p i v o t t o t h e f o cu s p o sition .
Chomsky, h o w e ver, o p erator
c lause accounts fo r
Fo r
m o v e m ent w i t h i n t h e p r e suppositional
t h e A - b a r m o v e m en t p r o p e r ties o f c l e ft s ( s e e
Chomsky (1977) for discussion).

" In Hungarian, the focus position is immediately pre-verbal. Hence, E. Kiss does not have any
projections between FocusP and VP.
160
Chapter 4
A dapting these proposal to M a l agasy clefts, we h ave th e f o l l ow i n g
trees.'"
( 27)
a.
Rakot o n o
man a s a lam b a.
Rakoto foc
A T . w ash c l oth
'It's Rakoto who washes clothes.'
Chomsky-tree
CP
D Pi
Rakoto
CP
O P;
C'
C'
IP
no
manasa lamba t;
Rizzi/E.Kiss-tree
FocusP
DPi
Rakoto
no
manasa lamba t;
I n (27b), the p i vo t i s
o perator-variable p a ir .
m ovement t o
b ase g enerated an d m o v e m ent o b t ains vi a t h e

Pos i t i n g a n a n a l y sis w h e r e t h e r e i s d i r e ct
t h e p i v o t p o s i tion, as i n ( 27c), is hard t o
m o t i v at e i n
Malagasy. If m o v ement w ere to occur directly from the subject position

t o the p i v ot , w e
w o u l d e x p ect th e p i v o t p o s i tion t o h a v e t h e s a m e
restrictions as the subject position. As already seen in chapter 3, however,

the elements that can appear in the pivot position are a superset of those
'" The position of the focus particle no is immaterial at this point, either in C" or Foc".
161
The le
ft periphery
that can be subjects. For example, while indefinite (nonspecific) subjects
are ungrammatical, it is quite common for the pivot to lack a determiner.
(28)
a.
M ilal a o baolina*(ny) zazavavy.

AT.play ball
(det) girl
b.
(Ny) zazavavy no milalao
bao l i n a.
( det) girl
foc AT . p la y b a l l
'It's (the) girls who are playing ball.'
S imilarly, the u n grammaticality o f P P s ubjects contrasts with t h e w e l l formedness of PP clefts. I n
c h a p ter 3, w e c o nsidered many e x amples
parallel to (29).
(29)
a.
A ndaf i h an'i Soa ny atody (*amin') ny bozaka.

CT.pad.gen.Soa det egg (P.gen) det grass
'Soa pads the egg with the straw.'
(Amin') nybozaka no andafihan'i Soa ny atody.

(P.gen) det grass foc CT.pad.gen.Soa det egg
'It's with the straw that Soa pads the egg.'
In fact, some CT clauses must be realized as a cleft due to restrictions on

the subject position. A s
d i scussed in chapter 3, (30a) is un grammatical
b ecause the p r ice n om inal ( a n o n - referential DP ) i s n o t p e r m i t ted i n

subject postion; once clefted, the example is acceptable, as in (30b).
( 30)
a. *
Nividianany
hena valopolo.
pst.CT.buy.3(gen) meat 80
V alopolo no nividianany
hena.
80
foc pst.CT.buy.3(gen) meat
'It's for 80 (ariary) that she bought meat.'
Thus in or der f o r
d i r ect m o v ement t o b e v a li d f o r M a l a gasy clefts, it
162
Chapter 4
would require the additional stipulation of allowing indefinites, PPs, and
non-referential DPs to move through the subject position as long as they
d o not r emain t h ere .
Th i s i s n o t a n u n r e asonable solution, bu t t h e
Chomsky analysis avoids this problem as it i s n o t t h e p i v o t i t self th at

moves, bu t
a n e m p t y o p e r a t or .
A l tho u g h t h i s a p p r o ach a p p ears
adequate, I will now consider a different cleft structure that is more easily
motivated for Malagasy.
3.2 Pseudo-clefts
The structures presented above equate Malagasy clefts to English it-clefts
and focus fronting in Romance and Hungarian. W i t hin the Austronesian
t radition, h ow ever, r esearchers have p o i nted ou t
t h a t c l efts i n t h e s e
languages bear some similarities to equational clauses with a h e adless

relative as the subject. For some analyses, see Kroeger (1993) and
Richards (1996) on Tagalog, Georgopoulos (1991) on Palauan and Cole,
H ermon and A m a n (to appear) on M alay."
F r o m t h i s p erspective, the
pivot is in fact the predicate and the presuppositional clause is the subject.
Consequently, clefts in Austronesian bear a close resemblance to (English)
pseudo-clefts (e.g. what Rakoto washed was the clothes). (31) is the structure
proposed by Georgopoulos (and Kroeger) and (32) is argued for by Cole,

Hermon and Aman (to appear) (henceforth, CHA).
(31)
G e o rgopoulos-tree
IP
DP
ptvot
" Pearson (1996c: fn 17) posits a similar analysis for Malagasy clefts.
163
The le
ft periphery
(32)
C H A - tree
CP
IP
p>vot
DP
VP
In both (31) and (32), there is a predication relation between the pivot and
the operator in the relative clause, similar to the Chomsky-style analysis
in (27b). The main difference between (31)/(32) and (27b) is that in the
former the pi vot is a predicate, not an argument. (31) and (32) can be
distinguished in the position of the pivot: in (31), the pivot remains in the
( clause-initial) predicate position, while in (32), it moves to [Spec, CP]. I
will adopt the analysis of clefts in (32), with some minor mo difications. In
particular, I argue that the pivot is a predicate that moves into the specifier
of a functional projection, FocusP (not CP). The presuppositional clause is
a headless relative in the subject position.
Interestingly, th e
s t r u cture i n ( 3 2 ) i s s t r i k i n gly p a r allel t o r e c ent
p roposals for d e r i v in g p r e dicate-initial w o r d o r d er , m e n t i oned i n t h e

introductory c h apter an d
a g ain i n s e c tion 2 .6 .
Un d e r t h i s a n a lysis,
structures along the lines of (32) are prevalent in M a l agasy. T o
c apture
interpretational differences (in particular, the focus reading fo r
c l efts), I
suggest that the difference between a cleft and a regular clause lies in the
final position of the VP/predicate: [Spec, FocusP] for clefts and [Spec, TP]
for "normal" clauses.
See section 4 for some discussion of th e special
focus interpretation for pivots. (33b) illustrates the structure of the cleft in
(33a): the predicate moves to [Spec, TP] for EPP and then to [Spec,
FocusP] for the focus interpretation.
164
Chapter 4
( 33) a .
[predicate Ny zazavavy ] [subject no milalao baolina ].
det girl
f oc AT.play
b all
(lit.) 'The ones who are playing ball are the girls.'
FocusP
VP;
VP
AgrP
VP
subject
(34b), on the other h a nd, r epresents a standard clause.
T h e p r e dicate
moves to [Spec, TPj, giving VOS word order.
( 34) a .
b aolina ] [ s Ubject
[predicQte Milalao
AT.play b a ll
ny zazavavy ].
det girl

TP
VP.,
AgrP
predicate
DP
VP
subject
I now turn to some evidence in favour of the structure in (33b) for clefts.
165
The le
ft periphery
3.3 Pivot =predicate
T urning first to the p i v ot, for b ot h G e orgopoulos and C H A , t h e p i v o t

originates as a predicate. In Georgopoulos' structure in (31), there is no
movement of the pivot; it remains in the clause-initial position, consistent
with Palauan predicate-initial word order.
M a l ay, on the other hand, is a
subject-initial language that also has clause-initial clefts. CH A
t h e r efore
propose that the pivot moves into [Spec, CP], as shown in (32)."
In these languages, the pivot does have predicate-like properties. Fo r
example, the Palauan examples in (41) below show that the pivot takes the
subject agreement marker ng. Tu r n ing to Ma lagasy, most pivots are DPs
or PPs, which are also possible as matrix predicates."
( 35) a .
Any Antananarivo no mipetraka i Ketaka.

t here Antananarivo foc AT.liv e
K eta k a
'It's in Antananarivo that Ketaka lives.'
Any Antananarivo i Ketaka.

there Antananarivo Ketaka
'Ketaka is in Antananarivo.'
M oreover, a clefted DP can be negated, unlike argument DP s and l i k e

predicates (verbal or other). Thus the pivot in (36a) can take negation and
patterns w it h
t h e n o m i nal p r e dicate in ( 36d). ( 3 6b,c) show that as
arguments, DPs cannot be negated.
( 36) a .
Tsy Rasoa no nanoroka a n-dRakoto.

neg Rasoa foc pst.AT.kiss acc-Rakoto
'It's not Rasoa who kissed Rakoto.'
b. *
Nanoroka
t sy a n - d Rakoto Rasoa.
pst.AT.kiss
n e g acc-Rakoto Rasoa
Due to predicate-initial order, in both Malagasy and Palauan the movement of the pivot to
[Spec, FocusP] is string-vacuous.
" In chapter 3 section 2.4, I provide an account for restrictions on what can be a pivot.
166
Chapter 4
c.
Nanoroka
an-dR a k oto tsy Rasoa.
pst.AT.kiss
a c c -Rakoto neg Rasoa
Tsy mpianatra Rasoa.

n eg student
Ras o a
'Rasoa is not a student.'
Similarly, the pivot may be preceded by the verbal particle toa ' seems'. As
shown in (37a), toa normally precedes the predicate.
(37)
a.
Toa
nano r ok a a n - dRakoto Rasoa.
seem pst.AT.kiss acc-Rakoto Rasoa

'Rasoa seems to have kissed Rakoto.'
b.
Toa
Ras o a no nanoroka a n - dRakoto.
seem Rasoa foc pst.AT.kiss acc-Rakoto

'It seems to be Rasoa who kissed Rakoto.'
T he above d at a i n d icate that th e
p i v o t h a s a s i m i l a r d i s t ribution t o
predicates."
F or further ev idence that th e p i v o t i s a p r e d i cate, recall that w h questions are formed by clefts (see section 2.4). Note that in certain cases,
however, wh-elements can be matrix predicates without clefting.
( 38) a .
I za
iana o ?
who
2 s g (nom)
'Who are you?'
In a cleft, both tsy and toa can also appear on the embedded predicate. Thus these are not
simply clause-initial particles.
(i) a .
Ra s oa no tsy nanoroka an - d Rakoto.
Rasoa foc neg pst.AT.kiss acc-Rakoto
'It's Rasoa who didn't kiss Rakoto.'
b. R asoa no toa
nan o r ok a an- d R akoto.
Rasoa foc seem pst.AT.kiss acc-Rakoto
'It's Rasoa who seems to have kissed Rakoto.'
167
The le
ft periphery
Inona ireto vato ireto?
what these stone these
'What are these stones?'
In order to account fo r
t h e d i f f erence between examples like (38) and
clefts, I suggest that in (38) the wh-elements are in th e r e gular f r o n ted
predicate position (e.g. [Spec, TP]), not in [Spec, FocusP] (the trees in (33)
above illustrate the two p o sitions ). Clearly, there is n o
i n c ompatibility
b etween wh-elements and t h e p r e d icate position . I t i s t h e r e fore n o t

u nreasonable t o
a r g u e t h a t w h e n in cl e f t s , w h - elements a r e a l s o
predicates.
Interestingly, in cases such as (38), clefting is not possible, as shown by
the ungrammaticality of (39).
( 39)
a. *
Iza no ianao?
who foc 2sg(nom)

'Who are you?'
b. *
Inona no ireto vato ireto?

what foc these stone these
'What are these stones?'
I do not have a complete explanation for th e contrast between (38) and

(39). I tentatively suggest that the ungrammaticality of (39) derives from
s ome incompatibility b e tween a
h e a d less relative an d a d e f i n it e D P
(names, pronouns, etc.). The meaning for (39a) would be 'the one who is
you is who?', a decidedly odd construction.
Summing up, a range of d ata show t hat the p i vot i n a cleft patterns
with predicates. I take this as evidence in favour of the structure in (33b),
w here th e p i v o t m o v e s t o [Spec, FocusP] fro m
t h e m a t ri x p r e d icate
position.
3.4 Presuppositional clause= headless relative
L et us now examine the cleft structures in more detail. I n b o t h (31) and
(32), the presuppositional clause has the structure of a headless relative. In
a wide range of A u stronesian languages, the cleft m a rker i s a r e l ative
168
Chapter 4
clause marker or an other nom inal-marker.
T h e f o l l ow ing i l lustrate this
for Malay, an SVO language (data from CHA).
( 40) a .
[predicate Siapa]i [subjectyang kau nampak] ti.

who
t hat
y o usee
'Who do you see?'
[subjectYang
that
ka u n ampak ] [predicate Siti ] (-lah).
y o u see
Siti (foc)
'The one you see is Siti.'

[Dp buku [cp ya ng Jo hn beli ]]
book
that
John b ought
'the book that John bought'
(40a) is a clefted question, marked by yang (glossed as 'that'). ( 4 0 b)

illustrates yang as a headless relative marker. T h a t yang can be used for
regular (headed) relative clauses is shown in (40c).
In Palauan, the morpheme a is an all-purpose DP marker and precedes
t he presuppositional clause i n
a c l e ft . T he d a t a i n ( 4 1 ) a r e f r o m
Georgopoulos (1991). (41a) illustrates the standard predicate-initial wor d

order while (41b,c) are clefts. Note the presence of a in all examples.
( 41) a .
[predicate
N g-mekelekolt] [subject a
agr-cold
r alm ].
R-water
'The water is cold.'
[predicute Ng-Basilia ]
[ s ubject
a mengaus er tia el tet ].
agr-Basilia
R-weave P dem L bag
'It's Basilia who's weaving this bag.'
[predicate Ng-te'a] [ subject
agr-who
a kileld-ii a
s u b] ?
R-pf-heat-3s soup
'Who heated up the soup?'

As described by
G e o r gopoulos, a al w a ys o c c urs b e f ore a D P .
169
She
The le
ft periphery
c oncludes that since the presuppositional clause in (41b,c) is marked w i t h
a, it is a free relative in the subject position.
The Malagasy no is somewhat mys terious by c o m parison as it is not
used elsewhere in the language."
H e n c e this type o f h e adless relative
does not surface except in clefts.
F r e e r e latives, for example, use izay
(which is also the relative clause marker).

( 42) Hahazo karama be izay miasa
f ut.AT.get salary big rel
m a fy.
A T .w o r k h a r d
'Whoever works hard will make lots of money.'

I simply stipulate that the headless relative marked by
equative clauses.
no is restricted to
D u e t o t h e s i m i l arities between clefts an d r e l ative
constructions, I will assume that no is a nominal marker." I t i s p r o bably

either a determiner or a complementizer. Th e tw o t rees in (43) illustrate
t wo possible structures of th e h eadless relative in q u estion.
S i n ce th e
precise position of no i s not crucial to my a n a lysis, I will n ot a ttempt t o

distinguish between these two possibilities.
DP
( 43) a .
D
NP
no
NP
cp
op...vbl
Clearly unrelated is the past tense marker no. An o ther use of no, likely related to the focus
construction is in the first clause of an if...then statement, when the second expresses a cause.
raharaha nalehako.
(i)
Izaho
no tsy t o n ga, nisy
Isg(nom) foc neg arrived pst.AT.exist b u siness pst.a.gone. 1sg(gen)
'If I didn't come, it's because business called me elsewhere.'
Finally, no appears in certain SVO contexts, where the subject is an indefinite pronoun.
(ii) Na iza na iza
(no ) t sy mamafa
l al a n a di avoasazy.
or who or who
(foc) neg AT.sweep r oa d
top v o a .punish
'Whoever doesn't sweep the road will be punished.'
In (ii), no is optional and in fact some speakers prefer to omit it.
In Sundanese, an Indonesian language, the relative clause marker is nu (Hardjadibrata (1985)).
Malagasy and Sundanese are related, but not closely enough for this to be conclusive evidence,
however. M a l zac ( 1960) mentions that certain M alagasy grammarians believeno t o be
diachronically related to the determiner ny. He does not provide any references, however.
170
Chapter 4
NP
cp
NP
op
no
vbl
The fact that no is not a focus marker p er se wi ll be i m p o r tant once we

examine multiple clefts. I
w i l l n e x t d i scuss the interpretation of t h e se
headless relative structures.

3.5 Interpretation
In this section, I address the nature of the headless relative clause in the
proposed structure of the cleft. In a cleft where the pivot corresponds to
t he subject, the headless relative means something l ik e ' th e o n e / t h i n g
who/that...'.
(44)
B a k oly no manapaka bozaka.

B akoly foc AT.cut
gr a s s
(lit.) 'The one who is cutting grass is Bakoly.'

When something other than a subject is the pivot, however, the meaning
shifts. Consider (45), which has a PP pivot.
(45)
A m i n ' ny antsy no manapaka bozaka i Bakoly.

P .gen.det knife foc AT.cut
gra s s B a k o l y
'It is with a knife that Bakoly is cutting grass.'
Here, the headless relative cannot mean 'the one who is cutting grass' for
two reasons. F i r st, the agent of cutting (Bakoly) is expressed within the
r elative.
S e c ond, assuming that clefts have an equative structure, it i s
somewhat odd to equate a PP with a nominal referring to an individual. I

therefore suggest that in (45), the headless relative is interpreted as an
171
The le
ft periphery
event nominal (like a gerund). I n o t her w o r ds, (45) means 'The event of
Bakoly cutting grass was with a knife'.
T his account may at f i rst appear stipulative. I n
nominals in
f a ct, h o w ever, z er o
M a l agasy f r eely h a v e e i t he r a n e v e n t o r a n in d i v i d u al
interpretation. Both readings are illustrated in (46).
( 46) a .
Faly n y manapaka bozaka.

h appy det AT.cut
gr as s
'The ones who are cutting grass are happy.'

Sarotra ny manapaka bozaka.
d ifficult det AT.cut
gr as s
'Cutting grass is difficult.'
In (46a), the zero nominal clearly receives an individual reading. (46b), on

t he other h a nd, i s
a g e r u n d -like z er o n o m i n al .
It i s the r e f or e n o t
unreasonable to suggest the event and the individual readings are also
available for the headless relatives in cleft constructions."
Summing up, in this section, I have provided arguments for analyzing
t he Malagasy cleft as an equative construction.
T h e p i vo t is in fact t h e
matrix predicate and th e p r esuppositional clause is a h eadless relative

clause in the subject position. I n section 4, I will m o t ivate the claim that
t he pivot is interpreted as focus. I
t a k e t his as evidence that the p i v o t
appears in the specifier of a functional projection, [Spec, FocusP].

3.6 Interlude: ve
At this point, I w il l consider a M a l agasy-particular test fo r
t h e s u bject-
predicate distinction. Looking at focus, this test initially points in favour of

the above analysis of clefts. I w i l l sh ow, h o w ever, that once a range of
data are considered, this test must be reconsidered. A l t h ough not directly
relevant f o r
t h e c l e f t c o n s t ruction, t h i s s e c tion r e v e als p r e v i ously
unnoticed data that warrant analysis.

Keenan (1976) noted that certain particles appear immediately before
the subject in a clause. Some examples are given below :
t h e q u e stion
particle ve (47a), the exclamative p article anie (47b) and t h e n e g a tive

" These two readings are available for active and circumstantial and passive verbs.
172
Chapter 4
polarity item intsony (47c).
( 47) a .
Manapaka bozaka ve i Bakoly?

A T.cut grass Q
Ba k o l y
'Is Bakoly cutting grass?'
Manapaka bozakaanie i Bakoly!

AT.cut grass e x
Bakoly
'Is Bakoly ever cutting grass! '
c.
T s y manapaka bozaka
intsony i Bakoly.
neg AT.cut
gr a s s N P I
Bakoly
'Bakoly is no longer cutting grass'

In what follows, I will consider the question particle ve." I
w i l l show th at
contrary to p r evious claims ve is not a p article that marks off th e r i g ht

edge of the VP. Instead, I will show that it is a second position clitic.
Due to its position, ve is often used as a t est fo r
indirectly, clause structure.
s u bject-hood and,
C o n sider now a c l efted clause. C o m p a rin g
the position of ve in (48) and (47a) appears to i n d icate that th e clefted

element is the predicate and the remainder of th e clause is the subject.
This is precisely the evidence that Pearson (1996a) mentions in favour of a
headless relative clause structure for clefts.
(48)
B a k olyve no manapaka bozaka?

B akoly Q foc AT.cut
gr as s
'Is it Bakoly who is cutting grass?'

L et u s
n e x t tu r n to an app a r en t m u l t i pl e c l e f t (the bodyguard
construction ) . In this case, ve surfaces between the first element and t h e

'" The NPI intsony and the hortative anie do not have the same distribution as ve. These particles
do appear to indicate the right edge of a VP, hence they can surface either with the pivot or in the
presuppositional clause, which contains an embedded VP.
(i) a .
T sy
i Bak o l y intsony n o manapaka bozaka.
n eg
B ako l y NP I
foc A T . c u t gra s s
'It's no longer Bakoly who is cutting grass.'
b. I B akoly no tsy manapaka bozaka i ntsony.
B akoly foc neg AT.cut g r as s NPI
'It's Bakoly who is no longer cutting grass.'
173
The le
ft periphery
second.
(49) Omaly
veBakoly no nanapaka bozaka?
yesterday Q Bakoly foc pst.AT.cut grass

'Was it yesterday that Bakoly was cutting grass?'
Following the same reasoning, this datum would suggest that the second
element, Bakoly, is contained within the subject position. In o t h er w o r d s ,
t he meaning of (49) would be something l ik e ' W as Bakoly's cutting o f
grass yesterday?'. Sofar,these are reasonable conclusions.
Things become
m o r e c o m p l i cated, h o w e v er , o n c e w e tu r n t o
t opicalization. Parallel to clefting, ve appears between the topic and th e

rest of the clause.
(50)
B a k olyve dia manapaka bozaka?

Bakoly Q top AT.cut grass
'Bakoly, is she cutting grass?'
Moreover, if there is both a topic and a cleft, ve still surfaces immediately

after the topicalized constituent.
(51)
I t y antsyity ve di a B akolyno nanapaka b o z akataminy?

this knife this Q top Bakoly foc pst.AT.cut grass p st.P.3(gen)
'This knife, did Bakoly cut grass with it?'
Should we conclude that in topicalization, the topic is a predicate?

I argue that this is not the correct conclusion. Topics, unlike focussed
elements, do no t p a t tern w i t h p r e dicates. A s p o i n ted ou t b y K e e n an
(1976), topics are incompatible with negation and with the verbal particle
toa 'seems'.
( 52)
a. *
Tsy Rasoa dia nanoroka a n -dRakoto.

neg Rasoa top pst.AT.kiss acc-Rakoto
'Not Rasoa who kissed Rakoto.'
174
Chapter 4
b.
Toa
R a soa dia nanoroka a n -dRakoto.
seem Rasoa top pst.AT.kiss acc-Rakoto

'It seems that Rasoa, she kissed Rakoto.'
I take these data as showing that topics, unlike foci, are not p r edicates.
Hence, ve does not consistently appear immediately following the matrix
predicate. A s should now b e clear from ex amples (48)-(51), ve surfaces
immediately following the first constituent in a clause, irrespective of the
nature of that constituent. In other words, ve is a second position clitic.
S upporting e v i d ence comes f r o m
S V O c o n structions. In ce r t a i n
c ontexts (like contrastive clauses), Malagasy allows SVO w or d o r d er .
In
these cases, the question particle appears between the subject and the VP.
(53)
N y m pianatra ve mamaky t
e n y , ny mpampianatra
ve mihaino?
det student
Q ATxead
wor d det teacher
'Are the students reading, the teachers listening?'
Q AT.listen
C learly, in this example ve does not m ark of f t h e r i ght edge of th e V P .

R ather, it is in second position, in t hi s case immediately follow ing t h e
subject DP. That the DP in initial position is not a predicate can be seen by
the impossibility of negation.
(54) * Tsyny mpianatra mamakyteny,tsy nympampianatra m i haino.
neg detstudent ATxead word neg det teacher
AT.listen
Thus it is incorrect to claim that ve can be used as a positional test for the
predicate.
To account for the positioning of ve, we could p r opose that ve is the
head of the highest C projection, ForceP. This would capture the fact that
the question particle marks the clause type (interrogative) (Cheng (1991)).
As a clitic, ve then appears after the highest XP (predicate, focus, topic,
etc.). Thus the placement of the question particle varies depending on the
clause structure. Since the highest XP is usually the predicate phrase, ve
does appear to mark the right edge of the predicate phrase. T his is not
t he place for a n
a n a lysis o f s econd p o sition c l iticization. I th e r e f o r e
assume that as a clitic, ve is sensitive to both syntax (it appears after an XP,
175
The le
ft periphery
not after a word ) and prosody (it appears after the first XP in the clause,
not in a particular syntactic position ). (see Halpern and Zwicky (1996) for a
recent collection o f
a r t i cles on t h i s t o p i c) . F i n a l ly , n ot e t h a t s e cond
p osition clitics are commo n
i n w e s t ern A u s t r onesian languages. S e e
Kroeger (1993) on Tagalog, where the Q particle ba is likely cognate with

v e. T h e
p r o p osal fo r v e i s t h e r efore c onsistent w it h f a cts f ro m t h i s
language family.
4 C l eft as focus
A bove, I
h a v e c l aimed t ha t c l efting i s a f o c u s c o n struction, w i t h o u t
motivation. I n t his section I wi ll show that the pivot in a cleft receives a

p articular interpretation, similar to f o cus in o t her l a n guages.
H e n c e it
seems reasonable to refer to the cleft as a focus construction. A l t h ough I

discuss some interpretational pr operties of f o cus, I
semantic analysis per se .
wi l l n o t p r o v i d e a
In s t ead, th e f o l l o w in g d i scussion is si mply
i ntended to i l l u strate the p r operties of clefts as evidence in f a v ou r
of
treating the cleft as a focus construction. I take this as indirect evidence in

favour of positing movement to a particular functional projection, FocusP.
The cleft c onstruction i n
a ssociated w it h
M a l a gasy h a s m a n y o f the p ro p e r t ies
f o cu s m o v e m ent i n o t h e r l a n g u ages. T h e p i v o t i s
u nderstood as contrastive and exhaustive. For th e f o ll ow ing, I d raw o n

d iscussion by E. Kiss (1998). She defines "identificational focus" as t h e
following:
(55)
T h e function of identificational focus: An identificational focus represents a

subset of the set of contextually or situationally given elements for which the
predicate phrase can potentially hold; it is identified as the exhaustive subset

of this set of which the predicate phrase actually holds.
Semantically,
c onstruction.
i d e n t i ficational f o c u s i nv o l v e s an
oper a t o r -variable
F o r H u n g arian, E. Kiss posits syntactic movement o f t h e
p ivot to [ Spec, FocusP] to set u p t h e p r o per b i n d ing r elation. I n t h e

s tructure I
a d op t f o r M a l a gasy, th e p i v o t i s n o t d i r e ctly i n v o l ved i n
binding a variable. Instead, the operator-variable pair is contained in the

headless relative.
E. Kiss provides some tests for d i stinguishing i d entificational focus
176
Chapter 4
(which involves movement t o a f u n ctional projection) from i n f o r m ation
focus (which does not involve movement). T hese tests will show that the
Malagasy cleft expresses identificational focus. In f ormation focus is often
associated with pitch accent and does not have the same semantic force as
identificational f o cus .
A s defi n e d , i d e n t i ficational f o cu s e x p r esses
exhaustive identification. Hence, the answers in (56b,c) to the question in

(56a) have different meanings.
(S6) a.
N andeha t a i za
ianao?
pst.AT.go pst.where 2sg.nom

'Where did you go?'
Nandeha tan y
A mbos i tra aho.
pst.AT.go pst.there
A m b o sitra 1sg.nom
'I went to Ambositra.'
c.
Tany
A mbos i t ra no nandeha a h o .
pst.there A m bositra foc pst.AT.go 1 sg.nom

'It was to Ambositra that I went.'
(56b) does not exclude the possibility that I went to other places as well as
Ambositra.
T h e c l eft construction in (56c), however, i s a n
e x h austive
answer; Ambositra is the only destination.

Similarly, consider the following pairs.'"
(57)
a.
Nov idi n ' i Bakoly
ny satroka sy ny kiraro.
pst.TT.buy.gen.Bakoly
d e t ha t
'Bakoly bought a hat and shoes.'
Novidin'i Bakoly
ny satroka.
'Bakoly bought a hat.'
d e t hat
and det shoe
" E . Kiss attributes this test to Szabolcsi (1981). The judgements in (57) do not change if the
verb is in AT . I u s e TT in (57) to provide minimal pairs with (58), where TT is necessary to
allow clefting of the logical object.
177
The le
ft periphery
( 58) a .
Ny satroka sy
ny k i r ar o n o n o vidin'i Bakoly.
det hat
det s ho e f o c pst.TT.buy.gen.Bakoly
and
'It's a hat and shoes that Bakoly bought.'
b.
Ny s a t roka no novidin'i Bakoly.

det hat
foc p st.AT.buy.gen.Bakoly
'It's a hat that Bakoly bought.'
As in the English equivalents, the sentence in (57b) is a logical consequence

of the one in (57a). On the other hand, (58b) is not a logical consequence
of (58a). In fact, (58b) contradicts (58a). H e nce, the cleft construction in
(58) passes the test of exhaustivity.
Exhaustivity is further illustrated with the following test (as above, this
t est was inspired by E . K i ss (1998)). C l e fts carry a p r e supposition o f
e xhaustivity .
In
(59 b ) , t h e ne g a t io n i s i nt e r p r eted d e n y i n g th e
exhaustivity introduced in (59a).

(59)
a.
A : Ny s a t roka no novidin'i Bakoly.

det hat
foc p st.TT.buy.gen.Bakoly
b.
B : Ts i a , n y k i raro no novidiny.
no
det s hoe foc pst. TT.buy.3(gen)
'No, it's shoes that she bought.'
B y (59b), speaker B corrects speaker A
a n d i n d i cates that Bakoly o n l y
bought shoes, not a h at . I n a r g u m en t p o sitions (such as subject and

object), there is no such presupposition and hence the negation in (60b) is
not appropriate in response to (60a). In what follows, '' marks examples
that are grammatical, but pragmatically odd in context.
(60)
a.
A : Novi d i n'i Bakoly
ny satroka.
'Bakoly bought a hat.'
d e t hat
178
Chapter 4
b.
B : T s ia , n o v i diny
no
ny kiraro koa.
pst . T T .buy.3(gen) det shoe also
'No, it's also shoes that she bought.'

This test will be used to investigate apparent multiple clefts in section 5.3.2.
Finally, there are certain distributional restrictions on the elements that
can appear in the focus position.
( 61)
a. *
Bakoly koa no
na n deha t a n y
A mbos i t r a.
Bakoly also foc
p s t .AT.go pst.there A m b o sitra
'It's also Bakoly who went to Ambositra.'
b. *
N a ny mpianatra
v o t savotsa aza
no na h az o
i sa
t sara.
or det student
weak
even foc p st.AT.get number good
'It's even the weak students who got good grades.'
Na iza na iza
n o m a havita izany.
or who or who foc AT.done that

'It's anyone who can do that.'
These elements appear t o
h a v e s om e s emantic clash w it h e x h austive
identification. For example, the import of ' also' in (61a) is to assert that
someone else went t o
A m b o sitra. H e n c e th e m e aning o f t h i s a d v erb
conflicts with the presupposition of exhaustivity.
S i m i lar considerations
hold for 'also'. I n definites like na iza na iza 'anyone' in ( 61c) also do not
express the exclusion necessary in a cleft. Since DPs with these semantic
f eatures are not p e r m i tted i n a
c l eft, w e see that th e c l eft p o sition i s
associated with a particular interpretation, in this case exhaustivity.

Note, however, that a clefted koa 'also' is possible in certain contexts,
just as in English. A cleft with koa 'also' identifies a member of the relevant
set in addition to the member previously identified, with the rest of the set
still excluded. (62b) is therefore a possible reply to the statement (62a).
(62)
a.
A : Ny s a t roka no novidin'i Bakoly.

det hat foc pst.AT.buy.gen.Bakoly
179
The le
ft periphery
b.
B : Ny k i r aro ko a
det shoe a ls o
no n o v i diny.
fo cp st.AT.buy.gen.3
'It's also shoes that she bought.'
Here, Speaker A identifies a hat as a member of the set of things bought

by Bakoly, excluding all other possible items. Speaker B adds shoes to this
set, again excluding other items.
The above data show that in a cleft, the displaced element receives a
special focus interpretation, "identificational focus". I
t h e r efore conclude
that the cleft position is associated with focus features. Furthermore, like
Hungarian, Malagasy clause structure includes a functional projection that
hosts focussed elements."
N o w t h a t w e h av e explored simple focus, I
turn to apparent multiple foci.

5
M u l t i I e clefts
A s an u n u sual t w is t t o
t h e c l e f t c o n struction, M a l agasy a l l ow s t w o
elements to be fronted.
(63)
O m al y
Ras o a n o nanoroka
Yesterday Rasoa foc pst.AT.kiss
an-dRakoto.
a c c -Rakoto
'It was yesterday that Rasoa kissed Rakoto.'
As mentioned in 2.5, this has been described by
K e enan (1976) as the
"bodyguard condition" (repeated below)."
(64)
B o d y guard Condition (Keenan (1976)):

when a non-subject is fronted in a cleft, it can optionally be accompanied by
the grammatical subject
I will continue to call the first element (the adjunct) the pivot and refer to
the second (the subject) as the bodyguard. I n t erestingly, the bodyguard
construction is not available in the other Austronesian languages that I am
aware of.
T he question i m m e d iately a r i ses as t o
w h e t he r t h e t w o fr o n t e d
'" Unlike Hungarian, however, Malagasy pivots are predicates, not arguments, as discussed above.
" Here I restrict the bodyguard condition toclefts. See section 2.5 for discussion.
180
Chapter 4
elements are both in a focus position (either together in the same one or
in two separate focus projections). Following the discussion in section 2.5,
I will argue that in fact only the first element (the adjunct) is a pivot in the
u sual sense.
T h e s e c ond e l ement ( t h e s u b ject) i s i n a l o w e r t o p i c
projection. The proposed structure is given in (65).
(65)
a djunct Foc'
Topic P
(pivot)
subject Top '
TP
(bodyguard)
I have argued elsewhere that in the bodyguard construction, the adjunct
adjoins to the subject and both front as a unit (Paul (1998)). Here, I reject
t hat analysis (see section 5.2) and argue that the structure in (65) m o r e
a ccurately captures interpretational an d
b odyguard construction.
s t r u ctural p e culiarities o f t h e
F o r t h e m o m e nt, I s e t a side the q uestion o f
whether the bodyguard is base generated or moved into [Spec, TopicP]. I

will address this issue in section 5.5.
I n the f o l l ow in g sections, I
b odyguard
d i scuss other p o ssible analyses of t h e
c o n s t ruction, n a m el y co o r d i n ation, a m a l g amation a n d
multiple specifiers. I will show that none of these alternatives adequately

a ccounts for the d ata.
I n t h e d i s cussion, I w i l l p r esent arguments f o r
treating the bodyguard as a topic.

B efore turning to t hese alternate analyses, however, I w il l p o int o u t
some potentially p r o blematic aspects of
181
t h e b o d y g u ar d c o n struction.
The le
ft periphery
First, it is often argued that a sentence can have only a single focus (Rizzi
( 1997); Zubizarreta (1998)). T his is for i n t erpretational reasons.
reconsider the d i v ision o f
Let us
a s e n t ence int o p i v o t a n d p r e suppositional
clause.
(66)
[ p i vot [ presuppositional clause ]]
The presuppositional clause is restricted to given information. It t h erefore

cannot contain a focussed element (new information). W e t hen conclude
that if the bodyguard construction expresses multiple focus, it could not
involve multiple focus projections. Instead, along the lines of multiple whquestions, the two elements would h av e t o
f o r m a u n i t a t s om e l e v el
("absorption"). Second, recall that I have argued that in a cleft, the pivot is
a predicate. If the bodyguard were truly an example of a m u l t iple cleft,
this would require there being more than one predicate. C l early, this is
not possible in a non conjoined clause. Thus we have indirect evidence for
t reating the bodyguard as something other t han a s econd p i v ot .
This
point will be important in the following sections.

5.1 Coordination
In the proposed tree in (65), the two parts of a bodyguard construction do
not form a constituent. That is, the adjunct and the subject are in separate
projections.
A n o t he r p o ssible analysis of t h e b o d y g u ard c o nstruction
would treat this as some form of covert coordination in the focus position.
T he bodyguard construction does not, h o w e ver, p a t tern w i t h o v e r t l y
coordinated pivots. W it h o v ert coordination, for example, two adjuncts
are possible in the pi vot (67a). M o r e over, it i s n o t p o ssible to o v e r t ly
conjoin an adjunct and the subject (67b).
( 67) a .
Omaly
ary t an y
an-ts e na no n i v i d y
vary R asoa.
yesterday and pst.there at-market foc pst.AT.buy rice Rasoa

'It was yesterday and at the market that Rasoa bought rice.'
b. *
Omaly
ar y R asoa no nanoroka
Yesterday and Rasoa foc pst.AT.kiss
182
an-dRakoto.
a c c -Rakoto
Chapter 4
Precisely the reverse is true for the bodyguard construction: t w o a djuncts
are ungrammatical and an adjunct-subject combination is grammatical."
Omaly
( 68) a .
ta ny
an-ts e n a n o n ivid y
vary R asoa.
yesterday pst.there a t -market foc pst.AT.buy rice Rasoa

'It was yesterday at the market that Rasoa bought rice.'
Omaly
Ras o a no nanoroka
Yesterday Rasoa foc pst.AT.kiss
an-dRakoto.
a c c -Rakoto
'It was yesterday that Rasoa kissed Rakoto.'
T hus (68b) and other i n stances of th e b o d y guard c onstruction are n o t

some form of covert coordination.
5.2 Amalgamation
I n Paul (1998), I ar gued t hat th e M a l agasy body guard c o nstruction i s
equivalent to the amalgamation of wh-elements in Japanese. T o see the
parallels, I briefly review the Japanese data. I
w i l l t hen show, h o w ever,
that this analysis is untenable.

S aito (1994) proposes that an a d junct wh-element can adjoin t o
an
argument wh-element. He points out the following contrast.
( 69) a .
John-ga
na n i - o
naze k a tta-no?
John-nom what-acc why bought-Q

'Why did John buy what?'
b.
John-ga
na z e nani-o
kat t a -no?
John-nom why what-acc bought-Q

Dare-ga
na z e nani-o
kat t a -no?
who-nom why what-acc bought-Q

'Why did who buy what?'
(69a), where the argument nani-o 'what' precedes the adjunct naze 'why', is
" (68a) improves with a distinct pause between the two adjuncts, indicating perhaps that the first
adjunct is in some clause-adjoined position. T h e bodyguard construction does not require this
marked pause between the pivot and the bodyguard.
183
The le
ft periphery
grammatical.
( 6 9 b ) s h ow s t h a t t h e r e v erse o r de r i s u n g r a m m atical.
Interestingly, adding another higher argument wh, as in (69c), improves

the grammaticality o f
( 6 9b). T h i s i s r e ferred t o a s t h e " a d d i t ional-wh
effect". Simplifying Saito's account somewhat, the adjunct wh adjoins to a

higher argument wh at LF. F rom t his position, it can license its trace and
escape an ECP violation.
w here
A d j u n ction obtains in (69a,c) but not in (69b),
t h e a d j u nc t c - c o m m and s t h e ar g u m e nt , r e s u l t in g in
the
ungrammaticality of (69b).
Pursuing thi s analysis, Tanaka (1998) claims that w h -elements can
a malgamate and then move as a constituent in overt syntax.
H e shows
that the ungrammatical long-distance scrambling of a wh-adjunct in (70a)

improves when an argument wh scrambles along with it, as in (70b).
( 70)
a. *
Naze; John-ga [ Bill-ga n a ni-o t; n aosita-kadooka ]

why
Jo h n - nom Bill-nom what-acc fixed-whether
siritagatteiru-no?
want-to-know Q
Nani-o; naze> John-ga [ Bill-ga t; tj
naosita-kadooka ]
what-acc why John-nom Bill-nom
fixed-whether
siritagatteiru-no?
want-to-know Q
'What does Bill want to know why Bill fixed?'
Together with th e d ata in (69), the examples in (70) illustrate a certain
dependency between an adjunct wh and an argument wh.
M o r e over, the
a malgamated wh-elements can surface in th e f o cus p o sition o f
a c l eft
construction.
(71)
John-ga Mary-ni ageta-no-wa nani-o naze
datta-no?
John-nom Mary-datgave-nm-top what-acc why
cop-pst-Q
(lit.) 'It was what why that John gave to Mary?'

Tanaka argues that (70) and (71) both i n d icate that amalgamation can
occur in the syntax as well as at LF.
Let us examine th e i m p o r t ant a spects of t h e a b ov e a n a lysis an d
184
Chapter 4
compare them with the Malagasy facts. Crucially, in Japanese an adjunct
wh adjoins to an argument wh fo r l i censing.
In ot h e r w o r d s, only an
argument can "host" an adjunct. Recall that the bodyguard construction

also involves an adjunct and an argument ( the subject).
M o r e o v er, as
s een in (70) and ( 7 1), amalgamation i n J apanese allows th e t w o w h e lements to m o v e t o g e ther a n d a p p ear i n a c l e f t . Re c al l t h a t t h e
bodyguard construction arises in clefts. D u e t o t h ese parallels between
the Japanese amalgamation and the M alagasy bodyguard, it i s p o ssible
that in Malagasy, the adjunct adjoins to the subject and both front as a unit
(as in the tree in (72), proposed in Paul (1998)).

(72)
FocusP
Focus'
adjunctj
subjecti
Foc u s
There are a n u m ber o f
IP
d i f f erences, however, between th e b o d y gu ard
construction in Malagasy and the Japanese amalgamation.

F irst, only wh-phrases may host other wh-phrases in Japanese. W e
have seen that the bodyguard construction is easily possible with non-wh
elements. Moreover, it is possible in Malagasy to have a combination of +
and -wh words in the bodyguard construction. (N ote, however, that (73b)
is only acceptable for t h ose speakers wh o a l low a
w h b o d y g u ard ( see
sections 5.3.3 and 5.6 for further discussion). The /o in (73b) indicates that
the example is subject to dialectal variation. )
( 73) a .
Taiza i
B a koly no nanasa
pst.where Bakoly foc pst.AT.wash

'Where did Bakoly wash clothes?'
b. %
T a min'i o
loko i o
lamba?
cloth
iza n o nandoko
trano?
pst.P.gen.this paint this w h o foc p st.AT.paint h o u se

'Who painted houses with this paint?'
185
The le
ft periphery
Thus Japanese amalgamation i s
b ot h m o r e r e s t ricted t h a n M a l a gasy
bodyguard (limited to wh-phrases) and less restricted (having a wh-phrase

as a bodyguard is in fact highly marked in M alagasy). T hese differences
could stem, however, from independent properties of the two languages.
In other w o r ds, amalgamation could be used by d i f f erent languages in
d ifferent w a ys .
I tur n t o d a t a w h i c h a r e m o r e p r o b l ematic fo r t h e
amalgamation approach.
In the tree in (72), the two fronted elements form a constituent. I n i tial
evidence against the constituency of the pivot and bodyguard comes from
clausal coordination.
(74)
Omaly
Ras oa no nivarotra he n a ary Rakoto no nividy
yesterday Rasoa foc pst.AT.sell meat and Rakoto foc pst.AT.buy
mofo.
bread
'It was yesterday that Rasoa sold meat and Rakoto bought bread.'
I n (74), the adjunct omaly 'yesterday' is i n t erpreted as m o d i fying b o t h

c lauses. I n
o t h e r w o r d s , i t a p p ears that w e h a v e c o n j unction o f a
constituent that i n cludes the subject but excludes the adjunct, perhaps
TopicP.
(75)
[ F 'pcp o maly [[q,zp Rasoa no ....] ary [q,zp

yesterday R a soa foc a n d
R a k oto no ...] ] ]
R ako to foc
( 74) is problematic under an a ccount w h ere b ot h t h e a d junct and t h e

subject appear amalgamated in a single position.
F urthermore, the pi vot and th e bodyguard can be separated by t h e
question particle ve and the parenthetical hono 'so they say'.
( 76) a .
Omaly
v e Ra s oa n o nanapaka b ozaka?
yesterday Q Rasoa f o c pst.AT.cut grass

'Was it yesterday that Rasoa cut grass?'
b.
Oma l y
hono
yesterday so-they-say
Rasoa no nanapaka b o zaka.

R a soa foc pst.AT.cut grass
'It was yesterday, so they say, that Rasoa cut grass'
186
Chapter 4
The data in (74) and (76) indicate that the pivot and the bodyguard do not
make up a single constituent.
Before concluding, recall that I have argued that the focussed element
is in fact an equative predicate. Given this analysis, the structure in (72) is
i mpossible to d e rive as there can only b e
a s i n gle p redicate in a n o n
conjoined clause. Thus despite initial similarities, Malagasy and Japanese

multiple clefts involve very different structures. I t h erefore conclude that
it is not possible to analyze the Malagasy facts along the same lines as the
Japanese ones.
I now t u r n t o a n o t he r a ccount o f m u l t i pl e f r o n ting . Du r i n g t h e
d iscussion, I will show t hat in a b o d y g uard c onstruction, only t h e f i r st
element receives a focus interpretation, not the second.
T h ese facts also
point away from t reating the bodyguard construction as amalgamation,

which places both elements in the focus position and hence both should
receive a focus reading.
5.3 Multiple specifiers

As is well-known in the literature on wh-questions, Slavic languages allow
multiple fronting of w h -w ords.
S i nce the Malagasy cleft construction is
a lso used for w h-questions, it m a kes sense to c ompare M a lagasy wi t h

Slavic." M u ch has been written on the Slavic multiple fronting, but I w i ll
draw mainly on the description in Rudin (1988).
5.3.1 Slavic multiple rohfronting

Rudin argues that there are two types of multiple wh-fronting languages.
In one, all the wh-words are adjoined to each other in [Spec, CP]. In the
other, only the first wh-word is in [Spec, CP] and the others are adjoined
t o IP (via scrambling).
T h e f i r s t t y p e o f l a n g u age i s exemplified b y
Bulgarian (77a) and the second by Serbo-Croatian (77b).

(77)
a.
Bulgarian
K oj ko g o v i zda?
who whom sees
'Who sees whom?'
" M oreover, Boskovic (1997) claims that wh-movement in Slavic languages is focus movement,
further strengthening the parallels between Malagasy and Slavic fronting.
187
The le
ft periphery
Ko
kog a vi d i ?
who
w h o m sees
Serbo-Croatian
'Who sees whom?'

Rudin argues that in Bulgarian, but not in Serbo-Croatian, all the fronted
wh-words form a constituent.
To support t his d i stinction, Rudin
d i scusses three m ai n d i f f erences
between Bulgarian and Serbo-Croatian: constituency, ordering and island

effects. I w ill concentrate on the first two .
Sh e shows that in Bulgarian,
nothing can separate the wh-words, as shown in (78a). In Serbo-Croatian,

on the other hand, clitics and parentheticals appear between the first whword and the others, as illustrated in (78b).
( 78)
a. *
Koj t i
w ho you i s
kakv o kazal?
Bulgarian
w hat t o ld
'Who told you what?'

Ko
mu j e sta d a o ?
who
h i m is what given
Serbo-Croatian
'Who gave him what?'

Rudin also shows that Bulgarian, but not Serbo-Croatian, imposes linear
ordering constraints on the fr onted wh-words.
Sh e gives the following
examples.
( 79) a .
Koj
ko g o e v i djal?
who
w h o m is seen
Bulgarian
'Who saw whom?
b.
Kogo koj e vidjal?

whom who is seen
The contrast between (79a) and (79b) shows that the nominative wh must
precede th e
a c c usative w h - element i n Bu l g a r ian .
however, allows either order.
188
Serb o - C r oatian,
Chapter 4
( 80) a .
Ko
je k oga
vid i o ?
w ho
i s who m
s een
Serbo-Croatian
'Who saw whom?

b.
K oga j e k o
vid i o ?
whom is who seen

Recently, Richards (1997) has accounted for t hese differences by slightly
m odifying Rudin's analysis. H e
a r g ues that i n B u l g arian, all t h e w h -
elements are in distinct [Spec, CP] positions (the "multiple spec" analysis).
For Serbo-Croatian, he suggests that only the first raises to [Spec, CP], the
rest are adjoined to IP via scrambling."
L et us no w
r e t u r n t o t h e M a l a gasy b o d y gu ard c o n struction a n d
c ompare it to m u l t iple fronting in Slavic.
O n e d i s tinction between t h e
Malagasy multiple fronting and Slavic is that Malagasy restricts fronting to

two elements. Slavic languages, such as Bulgarian, do not have this limit
as shown by (81).
( 81)
Koj
kak v o no kogo e
dal?
who
w h a t t o w ho m h a s given
Bulgarian
'Who gave what to whom?'

S etting this d i f ference aside fo r
t h e m o m e nt , i s M a l agasy m o r e l i k e
Bulgarian or Serbo-Croatian? Recall that in the bodyguard construction,

the wh-words are strictly ordered: adjunct)subject. This is like Bulgarian,
in that the linear sequence is fixed, but in t h e r e v erse order (see (79)).
Additionally, M a l agasy i s
Croatian.
n o t a s c r a m blin g l a n guage, u n l ik e S e rbo-
A s s uming for th e m o m ent M a l agasy to b e a B u l g arian-type
f ronting l anguage, but w i t h f u n n y w o r d o r d e r ( a f ter a ll, i t i s a V O S

language!), we would place both elements in the bodyguard construction
i n multiple specifiers of FocusP, as in the tree below." "
In several recent papers, Boskovic has criticized the Rudin/Richards analysis of the difference
between Bulgarian and Serbo-Croatian (Boskovic (1997; 1998)). A l t hough the two papers offer
different analyses, both posit a single head that attracts all the wh-elements. A s w e w i l l see
immediately, this is not a possible analysis for Malagasy multiple fronting.
A lternatively, Rudin's analysis would adjoin the adjunct to the subject in [Spec, CP]. T h i s
gives rise to a structure similar to the amalgamation discussed above and hence suffers from the
189
The le
ft periphery
(82)
FocusP
xp
YP
ptvotj
Focus '
bodyguard; Foc'
no
I will a r g ue, h o w ever, t ha t t h i s i s n o t t h e c o r r ect s t r u cture f o r t h e
bodyguard c o nstruction,
p articular, I
d r a w in g o n da t a f r o m in t e r p r etation .
In
w i l l s h o w t h a t t h e b o d y g u ard d o e s no t r e ceive a f o c u s
interpretation, which is unexpected if the bodyguard is ISpec, FocusP]. I

then show that on the contrary, the bodyguard has topic-like properties.
5.3.2 Bodyguard ~
focus
In this section, I argue that in fact only the pivot and not the bodyguard is
in a focus projection, drawing on differences in interpretation between the
pivot and the bodyguard. I n section 4, we saw that the cleft construction
is associated with a
p a r ticular focus interpretation. If t h e b o d y g u a r d
condition involves movement of both elements into a focus projection (as

in the multiple specifier analysis), we would expect both to receive a focus
reading. I will show that this is not the case."
Recall from the discussion of single clefts that the pivot is interpreted
a s exhaustive (seen in examples (59) and (60) in section 4) .
p resupposition of exhaustivity can be denied.
H e nce the
Let us turn now t o t h e
example in (83a), which has a bodyguard construction. The data indicate

that i n
(83 a ), om a ly 'y e s t erday' i s
focu s s ed a n d
ther e f o re the
presupposition of exhaustively behind t hi s f ocus can be n egated, as in

(83b). O n
t h e o t her h a nd, th e b o d y guard, Rasoa, is not f o cussed and
same drawbacks. Note also that the tree in (84) is not possible if focussed elements are predicates,
as we have seen earlier.
Malagasy differs from Bulgarian in allowing elements to appear between the pivot and the
bodyguard. Thus one could conclude that Malagasy is not like Bulgarian nor like Serbo-Croatian,
which in fact I believe to be the case.
In fact, the data in the following subsections also point away from the coordination and
amalgamation analyses, which also place both the pivot and the bodyguard in the focus projection.
190
Chapter 4
therefore cannot be negated, hence both (83c) and (83d) are inappropriate
follow-ups to (83a).
A: Omaly
( 83) a .
Ras o a no nijinja
vary.
yesterday Rasoa foc pst.AT.harvest rice

'It was yesterday that Rasoa harvested rice.'
B 1:
T s i a , a f a k 'omaly izy
no
no ni j i n j a
vary.
fre e ' yesterday 3(nom) foc pst.AT.harvest rice
'No, it was the day before yesterday that she harvested rice.'
c.
B2:
Tsia , af a k ' o maly R ak o to no nijinja

no
vary.
fre e ' yesterday Rakoto foc pst.AT.harvest rice
'No, it was the day before yesterday that Rakoto harvested rice.'
c.
B3:
Tsia , oma l y
no
Rakot o no nijinja
vary.
yes t erday Rakoto foc pst.AT.harvest rice
'No, it was yesterday that Rakoto harvested rice.'
Thus we see that the bodyguard has a semantic force that differs from
pivots in standard clefts.
Similar results obtain for negation. Recall that the pivot in a cleft can
take negation, cited earlier as evidence in f av our o f
t h e p r e d icate-like
nature of the pivot. Interestingly, the bodyguard cannot be negated. This

difference between the pivot and the bodyguard is illustrated in (84).
( 84) a .
T sy omaly
Ras o a no nandeha
f iar a .
neg yesterday Rasoa foc pst.AT.go c ar

'It wasn't yesterday that Rasoa went by car.'
b.
Omaly
tsy R asoa no nandeha f i a ra.
yesterday neg Rasoa foc pst.AT.go car

'It was yesterday that it wasn't Rasoa who went by car.'
N ote that in (84a), negation takes scope over omaly 'yesterday' and n o t
Rasoa, which remains part of the presupposition.
191
The le
ft periphery
Consider next negative conjuncts. For one of m y
s p eakers, negative
c onjuncts are possible in al l p o s itions, subject, object, p r edicate. F o r

another, however, the examples in (85) are all ungrammatical. I t h e refore
*
m ark (85) w it h ( ) to i n d icate this v a riation i n j u d g ement . It is t h e
restricted use of negative conjuncts by this second speaker that is crucial
to the following discussion.
(85)
a . ( * ) N a nas a
lamba R a naivo fa tsy Rabe.
pst.AT.wash cloth
R a naivo C neg Rabe
'Ranaivo and not Rabe washed clothes.'
b. (*) N anasa
lamba f a tsy lovia Ranaivo.
pst.AT.wash cloth C n eg dish Ranaivo

'Ranaivo washed clothes and not dishes.'
c. (*) N anasa
fa tsy nanjaitra
lam b a Ranaivo.
pst.AT.wash C neg pst.AT.sew cloth
R a naivo
'Ranaivo washed and not sewed clothes.'
I mportantly, fo r t h e s econd s peaker th e o n l y p o s sible p osition fo r a

negative conjunct is after the pivot, as illustrated in (86a,b). In (86b), the
negative conjunct appears between the pivot and the bodyguard.
( 86) a .
Omaly
fa t s y a n droany no nandoko
tran o i Bakoly.
yesterday C neg today

foc p s t .AT.paint h o use Bakoly
'It was yesterday and not today that Bakoly painted a house.'
b.
Oma ly
fa t s y a n d roany B akoly no nandoko
trano .
yesterday C neg today

Bako l y foc pst.AT.paint h o use
'It was yesterday and not today that Bakoly painted a house.'
Thus for the second speaker, a negative conjunct is only available after a
focussed element. For this speaker, as shown in (87) the bodyguard itself
cannot take a negative conjunct, unlike the pivot in (86)."
'" (87) is grammatical, however, for the speaker who has freer placement of the conjunct. What is
crucial is the contrast between (86) and (87) for the other speaker.
192
Chapter 4
(87) (*) Omaly
i B a k oly fa tsy Rabe no nandoko
trano.
y esterday B a koly C neg Rabe foc pst.AT.paint

hou s e
'It was yesterday that Bakoly and not Rabe painted the house.'
T he u n g r ammaticality
o f (8 7 ) f o r th i s s p e a ke r i n d i cates t h a t t h e
bodyguard is not focussed.

S umming up, in this section I have argued that the p r operties of th e
true pivot are different from those of the bodyguard. The impossibility of
negation indicates that the bodyguard is not a predicate. The impossibility
o f a negative conjunct (for one speaker) is likely tied to th e l ack o f
an
e xhaustive interpretation fo r t h e b o d y guard, m e n t ioned a b ov e i n t h e

discussion of example (83). For this speaker, negative conjuncts are only
p ossible when the first conjunct is interpreted as exhaustive. A
r ange of
data indicate that the bodyguard d oes not receive a focus interpretation
and therefore does not appear in ESpec, FocusP]. These results are clearly
i ncompatible with the m u l tiple specifier analysis of multiple fronting .
next look
m o r e c l osely a t t h e b o d y g u ard i t self, fo r f u r t he r e v i d ence
against the multiple spec approach.

5.3.3
Bo dyguard= topic
I n the structure proposed, the bodyguard appears in a topic position.

n ow support t hi s analysis by
s h o w in g t h a t t h e b o d y g u ard h a s t o p i c
properties. It is used for presupposed information.'" T hi s contrasts with

t he pivot, which is for n on-presupposed material. I
w ill draw on data
f rom definiteness restrictions and question-answer pairs to support t h i s

d istinction.
D u e t o t h e s h ar p d i f f erences between th e p i v o t a n d t h e
b odyguard, I
co n c l ud e t h a t t h e m ul t i p l e s p e cifier a n a lysis i s n o t
appropriate for M a lagasy multiple fronting.
I n o t h e r w o r d s, M a l agasy
multiple fronting should be treated differently from Slavic. A d d i t ionally,

this section provides evidence in favour o f th e p r o posed analysis of th e
left periphery, given in section 2.
As an initial observation, recall from section 2.4 that indefinite DPs can
appear in the focus position. The bodyguard, however, must be definite,
as shown by the contrast between (88b) and (88c).
" A l ternatively, the bodyguard could be part of the presuppositional clause. See section 5.5 for
discussion of this possibility.
193
The le
ft periphery
( 88) a .
Zazavavy no nilalao
baolina tany
an-tok o t any.
girl
foc ps t . AT.play b a l l pst. t here a t-yard
'It was girls who were playing in the yard.'
b. *
T any
an-t o k otany z a zavavy n o nilalao
pst.there at-yar d
c.
Tany
girl
baoli n a .
foc pst . A T .play ball
an-to k o tany n y zazavavy n o nilalao
baoli n a .
pst.there at-yar d
det gi rl
foc pst . A T.play ball
'It was in the yard that the girls were playing ball.'
Since topics refer to presupposed information, they m ust be d efinite (or
specific)." T his definiteness restriction on the bodyguard points toward it
being a t opic position.
F u r t h er, th e c ontrast between (88a) and ( 88c)
illustrates an important difference between the pivot and the bodyguard,

suggesting that the two are not both in a focus projection.
A s a further distinction, consider the distribution of p r o n ouns.
The
bodyguard i s f u ll y c o mpatible w it h p r o n ouns t hat r efer t o p r e v i ously

mentioned DPs. Thus one possible answer to (89a) is given in (89b).
( 89) a .
Q: Taiza
no nandeha
f i a r a i Soa?
pst.where foc pst.AT.go car Soa

'Where did Soa go by car?'
A : Tany
A nt ana n arivo iz y
no nan d eh a
f iar a .
pst.there A n tananarivo 3(nom) foc pst.AT.go car

'It's to Tana that she went by car.'
In (89a), i Soa is part of the presuppositional clause. Since the pronoun izy
i n th e b o d y g u ar d p o s i tion c a n r e f e r t o th i s p r e s u pposed D P , t h e
bodyguard cannot be a focus position. I n general, the true pi vot cannot
refer to p r esupposed information, showing t hat even i n
a b o d y g u a rd
construction it retains its focus interpretation, which is restricted to new

A lthough topic are presupposed, all presupposed information is not necessarily a topic.
example, objects in the presuppositional clause may be nonspecific, as illustrated in (88a).
194
For
Chapter 4
i nformation .
H enc e (90b), although
f u l l y g r a m m a tical, i s n o t an
appropriate answer to (90a).
( 90) a .
o maly i S o a ?
Q: Nividy bokytaiza
pst.AT.buy book pst.where y e sterday Soa

'Where did Soa buy books yesterday?'
b.
A: O m a ly izy no nividy
boky t any an-tsena.
yesterday 3(nom) foc pst.AT.buy
b o o k pst.there at-market
'It was yesterday that she bought books at market.'
The unacceptability of (90b) as an answer to (90a) is due to the presence of

omaly 'yesterday' in th e p i vot p osition.
o maly 'yesterday' is p art o f t h e
presuppositional clause in (90a) and hence is incompatible with th e focus

associated with the pivot. N o t e that the unacceptability of (90b) does not
lie with the answer to the wh-question being in-situ. T his is possible, even
in a clefted clause."
( 91) a .
Q: ISoa no nandeha fiara taiza?

Soa foc pst.AT.go car pst.where
'Where did Soa go by car?'
A: I Soa nonandeha
Soa foc pst.AT.go
fi ara tany
car
Ambositra.
pst. t here Ambositra
'Soa went to Ambositra by car.'
Thus the problem with (90b) is due to presupposed information appearing

in the pivot position.
Consider now
t h e u s e o f t h e b o d y g uard c o nstruction i n q u e stion-
answer pairs. I w il l f i rst pr ovide some background on q u estion-answer

pairs in Malagasy. In general, if the question has the form of a cleft, a cleft
answer is required. For i n-situ questions such as (92a), either a cleft (92c)
or an in-situ (92b) answer is possible, the former being slightly m a rked
due to the added focus interpretation.
At this point, it is not clear to me how a wh-in situ is interpreted in a presuppositional clause
(as in (91a)). I leave this for future work on wh-in situ.
195
The le
ft periphery
( 92) a .
Q: Nanasa
lamba t a m i n'inona R a k o to?
pst.AT.wash c loth
p s t .P.gen.what Rakoto
'What did Rakoto wash clothes with?'
Al: Nanasa
lamb a t a min'it y
savo n y ity
p st.AT.wash cloth p s t .P.gen.this soap t h i s

'Rakoto washed clothes with this soap.'
c.
A 2: Ta m i n 'it y
Rakoto.
Rakoto
sav o n y ity no nanasa la m b a R akoto.
pst.P.gen.this soap t h i s foc pst.AT.wash cloth
R a k o to
'It's with this soap that Rakoto washed clothes.'
Importantly, either a cleft or a non-cleft answer is appropriate.

Let us now turn to m u ltiple wh-questions. Bodyguard clauses are not
appropriate answers to mu l tiple wh-questions where one wh-element is
in-situ.
( 93) a .
Q: Iza
who
no n a n apaka bozaka oviana?

f o c pst.AT.cut grass when
'Who cut grass when?'

b.
A 1 : Omaly
Ras oa no nanapaka b o zaka.
yesterday Rasoa foc pst.AT.cut grass

'It was yesterday that Rasoa cut grass.'
c.
A 2: R a soa no nanapaka bozaka omaly.

Rasoa foc pst.AT.cut grass yesterday
'It was Rasoa who cut grass yesterday.'
I suggest that (93b) is an inappropriate answer because the body guard

p osition cannot be used for new in f ormation, Rasoa in this case. "
I link
this limitation on the bodyguard to it being a topic.
' Again, the problem with (93b) is not due tothe clefting of omnly 'yesterday'. (92c) shows that
the answer to an in-situ question can be realized as a cleft. N ote also that we cannot reverse the
order of omnly and Rnson, due to the ordering constraints in the bodyguard construction.
196
Chapter 4
I will discuss wh bodyguard in section 5.6. For the moment, how ever,
note that if the bodyguard is indeed a topic, it should be incompatible with
w h-elements,
w hi c h
i nterpretationally an d
ar e
usua l l y
as s o ciated w i th
foc u s
(bot h
p o s itionally). T h i s i s i n g e n e ral t r u e : ( 9 4a) i s
u ngrammatical for some of the Malagasy speakers I have consulted. F o r

one of my c onsultants, however, the b o dy guard can b e a
w h -element,
with a restricted interpretation. She requires that the bodyguard wh-word

be D-linked, in the sense of Pesetsky (1987). In other w o r ds, there must
b e a context-specified set of p e ople, kn ow n t o b o t h t h e s p e aker a n d
hearer, that restricts the range of p ossible answers. Fo r
t h i s speaker, a
multiple wh-question can take a bodyguard construction as the answer if

t he question itself has the form o f
a b o d y g u ard c onstruction. ( 9 4b ) i s
therefore an appropriate answer to (94a). (Recall that /o in (94a) indicates

that it is subject to dialectal variation. )
(94) a. % Q: Ovianaiza no nanapaka bozaka?
when who foc pst.AT.cut grass
'When did who cut grass?'
A: Omaly
Ras o a no nanapaka b o zaka.

Due to D-linking, the referent of iza 'who' in (94a) is restricted: t he set of

p eople under consideration is known to the speaker and hearer. A m o r e
literal translation of (94a) would be 'When did which of them cut grass?',
where 'them' refers to the set of possible people. The answer (Rasoa) is a
possible bodyguard since it is not completely new in formation.
T h i s use
of the bodyguard construction will be discussed again in section 5.6.

A range of data shows that in a bodyguard construction, only the first
e lement (the t r u e
properties. I
f ronting.
p i v ot ) i s f o cussed. T h e s e cond el ement h a s t o p i c
t h e r efore reject the multiple specifier analysis of m u l t i ple
Des p i t e i n i tial s i m i larities, M alagasy b o d y guar d h a s v e r y
d ifferent properties from Slavic wh-fronting.
T h u s an y a n alysis of t h e
latter will not apply to the former.

Note that this conclusion then suggests that no is not a f o cus marker
197
The le
ft periphery
per se, since the elements that precede it are not n ecessarily un i f ormly
f ocussed elements.
T h u s w e h a v e i n d i rect evidence in f a v ou r o f t h e
relative clause structure for clefts given in (33b) in section 3.2 and against
treating no as the head of a focus projection, as in (27) in section 3.1.
5.4 Where are we?
I have argued against three p ossible analyses of t h e b o d y g u ard .
In
particular, I have shown that the pivot and the bodyguard do not form a
constituent nor do they share focus properties. I therefore concluded that
they are i n d i stinct positions, [Spec, FocusP] for the pivot and [Spec,
TopicP] for the bodyguard. (95) provides the basic structure for a (matrix)
bodyguard clause, such as (94b). (This FocusP may also appear embedded
under TopicP or other complementizer projections.)
(95)
Fo usP
TopicP
pivot
omaly
"/
bodyguard
Rasoa
t;
AgrSP
DP
vp
op...vbl
no nanapaka bozaka t ;
In the discussion in section 5.6 below of wh-fronting, another instance of

the cleft construction, we will see more data that confirm the difference in
i nterpretation between the pivot and the bodyguard .
closely at how movement of the bodyguard obtains.
198
I now look more
Chapter 4
5.5 S peculations
In the preceding sections, I have argued for a topic position below focus:
t he bodyguard .
proposal. I f
T h e r e r e m ai n s om e u n e xplained p r oblems w i t h t h i s
w e c o n sider th e t r e e i n ( 9 5), th e b o d y g uard a p p arently
m oves out o f t h e h e adless relative clause to th e t o pi c p o sition . T h i s

m ovement i s
h i g h l y p r o b l ematic; g i ve n s t a n dard a s sumptions, t h e
b odyguard
m o v e m en t i l l u s trated i n ( 9 5 ) v i o l a tes t h e c o m p l e x D P
c onstraint.
S e cond, this topic position is l i m ited t o s u bjects. W e s a w
earlier that as w ell as subjects, adjuncts can be t o picalized, hence this

restriction is somewhat surprising. T h i rd, the bodyguard position is only
a vailable i n
c l ef t c o n structions. W hy w ou l d a top i c p r o j ection b e
d ependent on
a h i g h e r f o cu s p r ojection? F i n a l ly , a s p o i nted ou t i n
footnote 4, Rizzi's low topic position is highly marked in Italian and is not
a vailable i n
m a n y l a n g u ages ( e.g. Sp anish, M a o ri , T a g alog, M a y a n
languages). In this section, I speculate on how to resolve these difficulties

and consider three alternate analyses to direct movement.
A s a first solution, it is possible to claim t hat th e b o d y guard i s n o t
m oved to t h e t o pi c p osition, but i s b ase-generated there.
A n em p t y
operator binds a null resumptive pro in th e headless relative. A

account fo r
s i m i l ar
I t a l ian t o p icalization i s p r o p o sed b y R i z z i ( 1 9 97). A n
anaphoric operator (in the specifier of a projection below the TopicP and
above TP, perhaps FinP) is coindexed with the bodyguard topic and binds
a null constant.
instead it l i nk s
c ategory). I n
T h i s o p erator d oes no t assign a r a nge t o i t s b i n d ee,

i t s a n t ecedent (the topic) to th e
b i ndee (the empty
o t h e r w o r ds, the r elation between th e o p erator an d t h e
empty category is not one of quantification. M o r e over, null constants are

only DPs and no t
o t her c ategories, as argued b y C i n qu e (1990); Rizzi
(1997). Although this may account for the restriction of the bodyguard to
(DP) subjects, the movement of the operator from the headless relative to
[Spec, FinP] is still problematic.
The second solution is to drop the headless relative analysis of clefts.
W ithout th e
c o m p lex DP , t o p icalization can p r o ceed as i n I t a l ian, a s
outlined above. I b e lieve, however, that the arguments for th e headless

relative analysis of clefts are compelling. The sole advantage would be a
simple analysis of the bodyguard construction.
As a final possibility, the bodyguard could in fact be within the headless
199
The le
ft periphery
relative: a subject in [Spec, DP].
(96)
FocusP
vp
AgrP
pivot
DP
vp
DP
bodyguard D '
NP
Under this approach, the bodyguard is a kind of preverbal subject that is

only licensed in headless relative clauses (perhaps by the event nominal).
This analysis has some advantages. I t
b odyguards t o
a ccounts for the restriction of
s u b jects, the t o pi c p r o p erties of b o d y g u ards an d t h e
m arkedness of wh-bodyguards. Consider first the following example. A s

stated earlier, the headless relative in these instances is interpreted as an
event nominal.
(97)
O m aly Rasoa no nanapaka b ozaka.

Under the analysis in (96), the meaning of (97) is in fact 'Rasoa's cutting of
grass was yesterday', a promising result. Th e restriction to subjects falls
out from the analysis. Since by hypothesis, the bodyguard is a preverbal
subject, adjuncts will never be in t his position.
b odyguards d iscussed in 5 .3.3 follow
T h e t o pi c p r operties of
f rom t h e f act that
Malagasy have t o pic-like p r o p erties, as
su b j e cts i n
n o te d b y m a n y r e s earchers.
Finally, consider the marginality of wh bodyguards.
A l t h o ugh M a lagasy
has wh-in situ, wh-elements can never surface in the subject position.
200
Chapter 4
Nanapaka
(9g)
boz a ka iza?
pst.AT.cut grass who

'Who cut grass?'
If bodyguards are subjects, it is not su rprising that w h-elements are in
general banned from the bodyguard position. T r eating the bodyguard as
a preverbal subject in headless relatives is clearly a p r o m i sing analysis.
Nevertheless, I leave it for fu rther research to determine w h ether this is
the correct approach.
B efore turning to w h-bodyguards, I w o ul d l ik e to p o int ou t t hat t h e
speculative nature of this section does not call into question the conclusion
that the bodyguard patterns with topics and not with focussed elements.
The only uncertainty is the precise position that the bodyguard occupies.
5.6 Wh- bodyguard
As mentioned above, the bodyguard construction is not available for whelements for most speakers of Malagasy." I
i nvoking a n
consultant
i n c o mpatibility b e t w een w h - elements an d t o p i cs .

who
The
ac c e pt s w h - b o dyguar d i s v e r y con s i stent i n her
judgements, however. I
grammar.
e x p l ained this restriction by
w i l l t h e refore examine this construction in h er
T h e d a t a w i l l p r o v i d e f u r t he r s u p p or t f o r t h e p r o p o sed
a nalysis of th e b o d y g u ard c o nstruction. I n pa r t i cular, I w i l l l o o k a t

ordering and interpretational restrictions.
5.6.1
Or d e r
Parallel to the standard bodyguard construction, multiple wh -clefts have a

fixed word order: adjunct)subject. Placing the subject before the adjunct
I have recently discovered that the consultant who rejects bodyguard wh in matrix clauses,
accepts the same construction in embedded clauses.
i za no nivid y
i l ay bo k y ?
(i)
M anan t ena R a soa fa taiza
AT.hope
Ras o a C pst.where w h of oc pst.AT.buy de f book
'Who does Rasoa hope bought the book where?'
(i) is a matrix question, with both wh-elements taking wide scope. I must set aside this puzzle for
two reasons. First, in this chapter, I have mainly been concemed with mono-clausal examples.
Embedded clauses involve further complications. Second, I have not had the time to properly
investigate the differences between matrix and embedded wh-bodyguard constructions Nevertheless,
initial field work has shown that the properties discussed in this section, namely word order and Dlinking, also apply to embedded whbodyguard.
Interestingly, Rizzi (1997: fn 18) points out that some speakers of Italian allow for both a
wh-phrase and a focussed element in embedded contexts (in clear contrast with matrix clauses). He
speculates that wh-elements in embedded clauses appear in a position different from matrix wh (i.e.
not in [Spec, FocusP]).
201
The le
ft periphery
results in ungrammaticality. Th e examples in (99) illustrate this ordering
restriction.
( 99)
a . % Ov i ana iza
when wh o
no l a s a nod y ?
fo c l ef t
pst . A T .go-home
'Who went home when?'

b. *
I za
ovi a na no lasa n o d y ?
who
w h e n f o c lef t
pst . A T .go-home
As shown in section 5.1, it is not possible to account for multiple fronting

b y p r o p osing s om e f o r m o f c o v e r t c o o r d i nation . W he n a n ov e r t
conjunction i s a d ded, as i n
( 1 00), th e o r de r o f e l e m ents i s r e v ersed:
subjects precede adjuncts.
( 1 00 ) t h u s c o n t rasts w i t h t h e b o d y g u a r d
construction in (99a)."
( 100) Iz a
who
ary o v ian a
no l a s a nod y ?
a n d when
foc l e f t
pst. A T .go-home
'Who went home and when?'

Impossible, moreover, is an adjunct bodyguard, even in cases where the
adjunct has been promoted to subject with CT.
(101) * Taiza
man a o ahoana no nandokoan-dRabe
iret o trano ireto?
pst.where AT.do how

f o c pst.CT.paint.gen.Rabe these house these
'Where and how did Rabe paint these houses?'
Finally, note that a maximum of two wh-elements front.
' From the perspective of English, (100) is very odd. As noted by Browne (1972), in English,
only wh-adjuncts may be conjoined, not arguments.
(i) a .
W h e n and where did you see them?
b. * Who and with what broke the glass?
The astute reader may recall that this type of conjunction is impossible for non-wh clefts; (100)
thus contrasts with (ii).
(ii) *
Ras o a a r y o m al y no l a s a nody.
Rasoa and yesterday foc left pst.AT.go-home
I have no explanation for the difference in grammaticality between (100) and (ii).
202
Chapter 4
( 102) * Nahoana taiza
why
iza n o n i v id y
boky io?
io
pst.w h ere wh o fo c p st.AT.buy
thisbook this
As shown by (103), the ungrammaticality of (102) is not due to some kind

of parsing limitation: it is possible to have three wh-elements in a clause.
( 103) % Nahoana iza
why
who
io
no n i v i dy
boky i o
tai z a ?
foc p s t .AT.buy this b o o k this p st.where
'Why did who buy this book where?'

It thus appears that only two elements may be fronted.
What accounts for the word order and "max two" effect in bodyguard
constructions?
Co n s i de r f i r s t t h e a d j u n ct) subject o r d ering .
In the
proposed analysis, the first element (the pivot) is focussed and the second
( the bodyguard) i s a
b odyguard examples:
s p ecial k in d o f t o p i c . Th i s i s a l s o t r u e i n w h o n l y t h e f i r s t w h -element i s t r u l y i n a fo c u s
position. The second is in the topic position. W e h ave seen that this topic
position is limited to subjects (either by th e n ull constant analysis or by
treating it as a pre-verbal subject). Therefore an adjunct will never be a
b odyguard."
I n p r i n ciple, either an adjunct or th e subject can be in t h e
pivot position, but because the bodyguard is the subject and there is only
one subject, the pivot is limited to adjuncts in the bodyguard construction.
M oreover,
s i n c e F o c us P d o m i n a tes T o p i cP, t h e w or d
or d e r
is
adjunct)subject. Finally, as neither the focus position nor this special topic
can iterate, examples like ( 102) cannot b e
g e n erated. T h e p r o p o sed
analysis of th e b o d y guard c onstruction as focus plus topic p r ov ides a

simple account for the basic properties of this construction.
5.6.2 D - l i nking
In the following, I illustrate some interesting properties of wh bodyguard
constructions which w il l p r o v id e ad ditional evidence for t h e t o p i c-like
status of the bodyguard. Keenan (1976) notes that the best "bodyguards"
are active agents. N evertheless, the following, where the bodyguard is a
theme subject, is an acceptable sentence.
' An adjunct that has been promoted to subject with CT, as in (101), is also ruled out from the
bodyguard position. T hi s is likely due to the same constraint that prohibits non-referential DPs
from the subject position (see chapter 3).
203
The le
ft periphery
(104) % Nahoana iza

why
no no r o han-dRasoa?
who
foc p s t .TT.kiss.gen.Rasoa
'Why was who kissed by Rasoa?'

In contrast to
(104), an inanimate bodyguard i s u n g r ammatical u nless
preceded by the determiner ilay.

(105) a.
Taiza
i non a no novidin-dRasoa?
pst.where what foc pst. TT.buy.gen.Rasoa

'Where did Rasoa buy what?'
b. %
T aiz a
i lay i n ona no novidin-dRasoa?
pst.where def what f o c pst. TT.buy.gen.Rasoa

'Where did Rasoa buy which of the things?'
Recall that I
s u g gested that i n o r de r t o o v e r come t h e i n c ompatibility
between topic and wh, a wh-bodyguard must be D-linked. A n i m ates (iza

'who') are treated as D-linked in Malagasy, but inanimates (inona 'what')
require a determiner."
T h u s the topic status of the bodyguard p osition
restricts the elements that can surface there."

Summing up, a certain dialect of Malagasy allows the multiple fronting
o f wh-elements.
I ha v e s h o w n t h a t t h i s m u l t i pl e f r o n t in g c a n b e
subsumed under the analysis given in section 5.4 for apparent m u l tiple
clefts. In other w o rds, the first element (the pivot) is in a focus position,
while the second (the bodyguard) is a topic. Since it is in a topic position,
t he bodyguard w h i s o b l igatorily D - l i nked. A c o n t ext-specified set o f
e lements must be available to b ot h
s p eaker and h e arer t o r e strict t h e
' Even in English, animate wh-elements are more "referential" than inanimates: they more easily
admit a D-linked interpretation. One effect of this is the ability to escape from weak islands. The
following data are from Szabolcsi and Zwarts (1992-1993).
(i) a .
W h i c h man do you regret that I saw?
b.? Who do you regret that I saw?
c.?? What do you regret that I saw?
' All wh-elements, D-linked or other, are excluded from the dia topic position.
(i) *
Ny mp i anatra iza dia nahamarina ny asa m a r i ka?
det student w h o top pst.aha.true det work math
'Which student did well on the math assignment?'
I do not have an explanation for this difference between dia topics and bodyguards. As mentioned
earlier, topicalization in Malagasy awaits further research.
204
Chapter 4
r eference of th e b o d y guard w h -element.
T h u s a l t h ough m u l t i ple w h -
fronting is limited in distribution among Malagasy speakers, its properties

support the proposed analysis of topic and focus.
6 C o nclusion
This chapter has been devoted to the left periphery. I
h a v e a r gued that
M alagasy has topic and focus positions in the CP layer of th e clause. I n

other w o r ds, t o picalized and
f o cussed elements appear i n p r e - v erbal
positions, where topic)focus. These results provide evidence in support

of Rizzi (1997)'s expanded CP.
F u r t h ermore, Rizzi's proposed structure
neatly accounts for an u n u sual pr operty o f

c onstruction.
M a l a gasy: t h e b o d y g u ard
I s u g g ested that th e b o d y guard i s i n f act a t o p ic . L i k e
Italian, therefore, Malagasy apparently has an a d ditional topic position

b elow focus. I speculate on the nature of t his topic position toward t h e
end of the chapter.
I next looked more closely at the focus construction, which is formed
by clefting. U n der m y
a n alysis, a cleft has the structure of an equative
clause, involving a headless relative in the subject position and a preposed

predicate in t h e
f o cus p o sition . In t e r estingly, t h e p r e d icate f r o n tin g
p roposed for clefts may be m or e general, perhaps accounting fo r

word order. I
VOS
t h en showed that clefts in Malagasy are associated with a
p articular focus interpretation.
T h i s focus interpretation is linked to t h e
position of the clefted element in [Spec, FocusP].

The last part of the chapter was devoted to ap parent multiple clefts:
the bodyguard construction. I
a nalyses of m u l tiple clefts:
p r o v i ded evidence against three possible

c o o r d i nation, am algamation, an d m u l t i p l e
specs. A l t h ough none of t h ese approaches are valid fo r
M a l agasy, the
discussion illustrated the importance of considering a range of data when

looking at multiple fronting. Pursuing the ideas presented in the first part
of the chapter, I argued that the bodyguard construction is best analyzed
not as a multiple cleft, but as a combination of clefting and topicalization.
Finally, I considered multiple wh-fronting and showed that the particular
properties of this construction fall out from the analysis provided.
I consider the present chapter to be the first step in understanding the
l eft periphery in Malagasy. The focus of this chapter was the cleft for tw o
reasons: first, clefts were used as a test for structure in chapters 2 and 3;
205
The le
ft periphery
second, a thorough analysis of the Malagasy complementizer system lies
beyond the scope of a single chapter. F o r
e x ample, although I d i scuss
topicalization, I do not provide an analysis. C l early, this is an area that

m erits further work .
M o r e over, I have restricted my attention to m o n o -
c lausal clefts, thereby i g n oring a
w i d e r a ng e o f d a t a , i n cluding l o n g -
distance dependencies and wh-questions in general. I also left unanswered

questions about the bodyguard. Is it really a lower topic position or a prev erbal subject w i t hi n
a n e v e n t n o m i n al ?
W hy ar e em b e d de d w h
b odyguard c onstructions allowed w h i l e t h ei r m a t ri x c o u nterparts a r e

generally ungrammatical? Finally, it would be imp ortant to compare the
Malagasy facts with data from o t her A u s tronesian languages. K r o e g er
(1993) has analyzed several constructions in Tagalog where constituents
a ppear p r e-verbally: t o p icalization, c l efting, a d j unct
inversion.
f r o n t in g a n d a y
T o w ha t e x t en t t h e l e f t p e r i p h ery i n o t h e r v e r b - initial
languages is similar to Malagasy remains to be discovered.
206
Chapter 5: Conclusion
T he present thesis is about M a l agasy syntax.
I r e p eat th e o b v i ous t o
remind the reader of the initial goal of this thesis: to grasp the "genius" of
Malagasy. As should be equally obvious, I believe that the genius of this
l anguage lies in t h e v o i c e system . V o i c e a l t ernations are central t o
M alagasy grammar.
V o i c e p r o v i des insight into sy ntactic structure. I
argued in chapter 2 that different passive voices are indicative of different

s yntactic positions and d o m a i ns .
t raditional c o n ceptions o f
V o i c e ca n b e m o r e c o m p l e x t h a n
p a s s ive . A s sho w n by
t he d a t a fr o m
circumstantial topic constructions, for example, voice is not limited to the

promotion o f
a r g u m ents. V o i c e p r o m o tes elements to a s y n tactically
prominent position, whence they are accessible to A-bar extraction. V o i ce

morphemes can have different functions. Some target specific positions,
s ome target particular d o m ains, others are " elsewhere" i n n a t u re . I n
m any w ays, v oice plays a salient and c o m p lex r ol e i n
t h e s y n ta x o f
Malagasy.
T he discussion of passive in chapter 2 provided evidence in favour o f
treating voice as a purely syntactic phenomenon. In particular, I discussed
examples where semantics appears to be a factor (e.g. the a- passive) and
argued that semantics plays a role in determining w h ere arguments are
projected in the syntax. In other words, there is but an indirect connection
b etween semantics and passive. O n t h e o t her hand, I d o n o t m e a n t o
deny the pragmatic effects of passive on discourse. I have been primarily
concerned with passive at the sentence level, not in the context of a text.
F urther research wil l d e t erm ine ho w t o i n t e grate the syntax w i t h t h e
discourse.
In chapter 2, I argued for a p a r ticular syntactic position ( ISpec, v2P])
for b ase g e n eratin g
a c e r t ai n c l a s s o f a r g u m e n ts .
T hi s p o s i t i on,
s ubordinate to agents and superordinate to t h emes, has th e e f fect o f

obscuring the th eta hierarchy .
M or e o v er, I d i s c ussed the d i f ferences
between this base generated position and a true derived object position,
Chapter 5
with reference to other languages. One extension of this analysis would
b e to investigate this position in a r a nge o f l a n g uages. I s t h er e so m e
semantic similarity to the elements that are generated in [Spec, v2P] or is it
simply a " w i l d card" p o sition?
I s i t p o s sible to s u p p ort th e p r o p osed
difference between base generation and m o v em ent c r oss-linguistically?

How does this affect the analysis of applicative and other derived object
constructions?
C hapter 3 continued the syntactic view of v o i ce. I
f o c u ssed on t h e
circumstantial topic, wh ich is u n u sual from th e p e rspective of
E n g lish
active-passive alternations. The key to the proposed analysis is that CT is

a n "elsewhere" v o i ce .
E l s e w h ere c o n d i ti ons ar e w e l l -studied i n t h e
domain of p h onology and m o r p h ology, but i t

how this notion may be applied to sy ntax.
r e m a ins to b e s een ju st
O n e f u r t her consequence of
the proposed analysis is a rethinking of th e EPP in M alagasy compared

with other verb-initial languages. Ar e some verb-initial languages (like
Niuean) uniquely predicate-fronting, wh ile others (like M a l agasy) have
both predicate and argument movement? I s there a di fference between
subject and object movement? E v i d ence for these distinctions will come
from adverb p l acement and th e p o sitional p r operties of arguments in
these languages.
Just as chapter 2 argued for a n a r t i culated VP, chapter 4 p r esented
evidence in fav our o f a n a r t i culated CP .
including topic and focus, appear in
Ce r t a i n e x t r acted elements,
t h e C P p r o j ection . M or e o v er, I
argued that in focus constructions, it is a predicate and not an ar gument

that appears in the pivot position. I also presented evidence in favour of a
t opic position below focus, the " b o d y gu ard" .
I m p o r t a ntly, i t i s t h e A
movement of the voice system, outlined in chapters 1-3, that feeds the Abar movement of topic and focus. I n o r der t o i n v estigate the latter, we
must first understand the former. I
h o p e t hat t his thesis has provided
some insight into both the A and the A-bar systems of Malagasy.
Combining the results of chapters 2-4, we get the following picture of
Malagasy clause structure.
Conclusion
CP
TopicP*
(XP)
FocusP
(YP)
TopicP
(DP) ~
TP
AgrP
vlP
v2P
VP
Just as the tree in (1) represents only the bare bones of the clause, so this
thesis is but the first step tow ards a better u n d erstanding of M a l agasy
g rammar.
O n c e t h i s i n i t ia l g r o u n d w o r k i s i n p l a ce, h o w ever, m o r e
detailed analyses will follow.
Appendix
In chapter 2, I address the different passive forms, a- and -V na. I n t h e
following tables, I provide further examples of the verbs associated with
t hese affixes. Table 1 combines the other f ou r it gives a range of t h e
verbs that have the a- passive. Table 2 shows those verbs which just have
t he a- passive. Tables 3-5 illustrate the verbs that have both the a- and Vna passives. These tables are not exhaustive, but are intended to s how
as wide a range as possible to supplement the examples in chapter 2.
Table 1: a- passive verbs
root
active verb meaning
subject of
subject of-
subject of
a- passive
Vna
CT
passive
dldy
mandidy
'cut'
instr
theme
instr
doboka
mandoboka
'beat'
instr
theme
instr
dona
mandona
'beat'
instr
theme
instr
fafa
mamafa
'sweep'
instr
theme
instr
fafy
mamafy
sow
mat theme
goal
neither
fahana
mamahana
'feed'
mat theme
goal
mat theme
fatratra
mamatratra
'stuff'
mat theme
goal
mat theme
fehy
mamehy
'tie'
mat theme
goal
mat theme
fefy
mamefy
'fence in'
mat theme
goal
mat theme
feno
mameno
'fill'
mat theme
goal
mat theme
idina
midina
'lower'
theme
jery
mijery
'watch'
instr
joro
manajoro
'raise'
theme
kapa
mikapa
'cut'
instr
theme
instr
kapoka
mikapoka
'hit'
instr
theme
instr
lafika
mandafika
pad
mat theme
goal
mat theme
lahatra
mandahatra
'line up'
theme
goal
latsaka
mandatsaka
dlop
theme
goal
210
goal
theme
instr
loc
Appendix
root
active verb
meaning
subject of
subject of-
subject of
a- passive
Vna
CT
passive
lavo
mandavo
'spill'
theme
goal
lefa
mandefa
'send'
theme
goal
loko
mandoko
'paint'
instr
orina
manorina
'erect'
theme
loc
petraka
mametraka
'place'
theme
loc
rakotra
mandrakotra
'cover'
mat theme
goal
mat theme
raraka
mandraraka
'scatter'
mat theme
goal
neither
roso
mandroso
'serve'
theme
goal
goal
sasa
manasa
'wash'
instr
theme
instr
seho
manaseho
'show'
theme
goal
goal
solo
manolo
'change'
mat theme
goal
mat theme
tafy
manafy
'clothe'
mat theme
goal
mat theme
tafo
manafo
'roof'
mat theme
goal
mat theme
tolotra
manolotra
'offer'
theme
goal
goal
toro
manoro
'point out'
theme
goal
goal
tondraka
manondraka
'water'
mat theme
goal
mat theme
tototra
manototra
'fill'
mat theme
goal
mat theme
traka
mandraka
'raise'
theme
tsindrona
manindrona
'pierce'
instr
tsipy
manipy
'throw'
theme
valy
mamaly
'answer'
theme
goal
neither
velatra
mamelatra
'unroll'
theme
theme/goal
goal
verina
mamerina
'return'
theme
vono
mamono
'kill'
instr
211
theme
instr
loc
theme
instr
goal
goal
theme
instr
Appendix
Table 2: verbs with only a- passive
root
active verb meaning
subject of
subject of-
subject of
a- passive
Vn,a
CT
passive
idina
midina
'lower'
theme
loc
joro
manajoro
'raise'
theme
loc
lahatra
mandahatra
'line up'
theme
loc
latsaka
mandatsaka
dlop
theme
loc
lavo
mandavo
'spill'
theme
loc
orina
manorina
'build'
theme
loc
petraka
mametraka
'place'
theme
loc
traka
mandraka
'raise'
theme
loc
tsipy
manipy
'throw'
theme
loc
verina
mamerina
'return'
theme
goal
Table 3: Case I verbs (instrumental advancement)
root
active verb meaning
subject of
subject of-
subject of
a- passive
Vn,a
CT
passive
dldy
mandidy
'cut'
instr
theme
instr
doboka
mandoboka
'beat'
instr
theme
instr
dona
mandona
'beat'
instr
theme
instr
fafa
mamafa
'sweep'
instr
theme
instr
jery
mijery
'watch'
instr
theme
instr
kapa
mikapa
'cut'
instr
theme
instr
kapoka
mikapoka
'hit'
instr
theme
instr
loko
mandoko
'paint'
instr
theme
instr
sasa
manasa
'wash'
instr
theme
instr
tsindrona
manindrona
'pierce'
instr
theme
instr
vono
mamono
'kill'
instr
theme
instr
These verbs all allow the instrument to appear in an "advanced" position.
212
Appendix
(1)
Manasa savony nylamba Rakoto.

AT.wash soap d e t cloth R a k o to
'Rakoto washes the clothes with soap.'
(2)
Namono
lan gilangy voalavo

ny
Rasoa.
pst.AT.kil l
stick
det rat
Rasoa
'Rasoa killed the rat with a stick.'
(3)
Nikapoka langilangy ny rindrina i Bakoly.

p st.AT.hit stick
det wa l l
Bako l y
'Bakoly hit the wall with a stick.'
(4)
Nikapa famaky ny hazo iSoa.

pst.AT.cut ax
det t r e e
Soa
'Soa cut the tree with an ax.'
(5)
Nandoko
lo k o m e n a ny tranony i Sahondra.
pst.AT.paint c o l our red

det h ouse.3(gen) Sahondra
'Sahondra painted her house with red paint.'
213
Appendix
Table 4: Case II verbs (locative alternation)
root
active verb
meaning
subject of
subject of-
subject of
a- passive
Vn,a
CT
passive
fafy
mamafy
fahana
sow
mat theme
goal
neither
mamahana
'feed'
mat theme
goal
mat theme
fatratra
mamatratra
'stuff'
mat theme
goal
mat theme
fehy
mamehy
'tie'
mat theme
goal
mat theme
fefy
mamefy
'fence in'
mat theme
goal
mat theme
feno
mameno
'fill'
mat theme
goal
mat theme
lafika
mandafika
pad
mat theme
goal
mat theme
rakotra
mandrakotra
'cover'
mat theme
goal
mat theme
raraka
mandraraka
'scatter'
mat theme
goal
neither
tafy
manafy
'clothe'
mat theme
goal
mat theme
tafo
manafo
'roof'
mat theme
goal
mat theme
tondraka
manondraka
'water'
mat theme
goal
mat theme
tototra
manototra
'fill'
mat theme
goal
mat theme
Unlike instruments, material themes are compatible with an instrument

PP.
(6)
Manafo
b o z aka ny trano a m i n 'n y m a r i toa i Sahondra.
AT.roof
g r a s s d et house P.gen.det hammer Sahondra
'Sahondra roofs the house with straw with the hammer.'
(7)
Nanotora fasika ny hady tamin'ny
ang a dy Rasoa.
pst.AT.fill sand d e t ditch pst.P.gen.det shovel Rasoa

'Rasoa fill sand into the ditch with the shovel.'
(8)
N amafy
voa n y saha
t a m i n 'n y
tana n y
i Bak o l y .
pst.AT.sow
s e ed det field pst.P.gen.det hand.3(gen) Bakoly
'Bakoly sowed seeds in the field with her hand.'
214
Appendix
Table 5: Case III verbs (dative shift)
root
active verb meaning
subject of
subject of-
subject of
a- passive
Vna
CT
passive
roso
mandroso
'serve'
theme
goal
goal
seho
manaseho
'show'
theme
goal
goal
solo
manolo
'change'
theme
goal
goal
tolotra
manolotra
'offer'
theme
goal
goal
toro
manoro
'point out'
theme
goal
goal
valy
mamaly
'answer'
theme
goal
neither
velatra
mamelatra
'unroll'
theme
theme/goal
goal
215
References
Abinal, R.P. and S.J. Malzac. 1888. Dictionnaire Malgache-Frangais.
Fianarantsoa.
Aissen, Judith. 1992. Topic and focus in Mayan. Language 68: 43-79.
Anderson, Stephen. 1971. On the role of Deep Structure in semantic interpretation.
Foundations of Language 6: 197-219.
Andriamierenana, Clement Luc. 1996. Morphological causatives in Malagasy. In

UCLA Occasional Papers in Linguistics, l7:
Th e structure fo Malagasy, Eds.
M. Pearson and I. Paul: 58-75.

Baker, Mark. 1988. Incorporation: A t heory o f grammatical function changing.
Chicago: University of Chicago Press.

Baker, Mark. 1992. Thematic constraints on syntactic structures: Evidence from
locative applicatives. In Thematic structure: Its role in grammar, Ed. I. M.
Roca: 23-46. Berlin: Foris.

Baker, Mark. 1996a. On the structural position of themes and goals. In Phrase
structure and the Lexicon, Eds. L. Zaring and J. Rooryck: 7-34. Dordrecht:
Kluwer Academic Press.

Baker, Mark. 1996b. The polysynthesis parameter. Oxford: Oxford University
Press.
Baker, Mark. 1997. Thematic roles and syntactic structure. In Elements of
grammar, Ed. L. Haegeman: 73-137. Dordrecht: Kluwer.

Baker, Mark, Kyle Johnson and Ian Roberts. 1989. Passive arguments raised.
Linguistic Inquiry 20: 219-251.
Ballard, D. Lee. 1974. The semantics of Inibaloi verbal affixes. Lingua 34: 181-218.
Bell, Sarah. 1976. Cebuano subjects in two frameworks. PhD thesis, Cambridge,
MA, MIT.
Boskovic, Zeljko. 1994. D-structure, theta criterion, and movement into theta
positions. Linguistic Analysis 24: 247-286.

Boskovic, Zeljko. 1997. Superiority effects with multiple wh-fronting in SerboCroatian. Lingua 102: 1-20.
Boskovic, Zeljko. 1998. Wh-phrases and wh-movement in Slavic. Comparative
Slavic Morphosyntax, Bloomington, Indiana.
Browne, Wayles E. 1972. Conjoined question words and a limitation on English
surface structures. Linguistic Inquiry 3: 223-226.
Brunson, Barbara. 1992. Thematic discontinuity. PhD thesis, Toronto, University
References
of Toronto.
Burzio, Luigi. 1986. Italian syntax. Dordrecht: Kluwer.
Carrier-Duncan, Jill. 1985. Linking of thematic roles in derivational word formation.

Linguistic Inquiry 16: 1-34.
Cheng, Lisa. 1991. On the typology of wh-questions. PhD thesis, Cambridge, MA,
MIT.
Chomsky, Noam. 1977. On WH-movement. In Formal syntax, Eds. P. Culicover,
T. Wasow and A. Akmajian: 71-132. New York: Academic Press.
Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris.
Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press.
Chung, Sandra. 1976a. An object-creating rule in Bahasa Indonesia.Linguistic
Inquiry 7: 41-87.
Chung, Sandra. 1976b. On the subject of two passives in Indonesian. In Subject and
topic, Ed. C. Li. New York: Academic Press.
Chung, Sandra. 1978. Case marking and grammatical relations in Polynesian.
Austin: University of Texas Press.
Cinque, Guglielmo. 1990. Types of A-bar dependencies. Cambridge, MA: MIT

Press.
Cole, Peter, Gabriella Hermon and Norhaida Aman. to appear. Clefted questions in
Malay. In Malayllndonesian linguistics, Eds. D. Gil and J. Collins. London:
Curzon Press.
Collins, Chris and Hoskuldur Thrainsson. 1996. VP-internal structure and object
shift in Icelandic. Linguistic Inquiry 27: 391-444.
Comrie, Bernard. 1977. In defense of spontaneous demotion: The impersonal
passive. In Syntax and semantics 8: Grammatical relations, Eds. P. Cole and J.
Sadock. New York: Academic Press.

Dahl, Otto Christian. 1951. Malgache et maanjan: U ne comparaison linguistique.
Oslo: Egede-Instituttet.
DiSciullo, Anna-Maria and Edwin Williams. 1987. On the definition of word.

Cambridge, MA: MIT Press.
Doetjes, Jenny. 1997. Quantifiers and selection: On the distribution of quantifying
expressions in French, Dutch and English. HIL. PhD thesis, Leiden, University
of Leiden.
Dowty, David. 1979. Word meaning and Montague grammar. Dordrecht: Reidel.
Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67:
547-619.
217
References
E. Kiss, Katalin. 1998. Identificational focus versus information focus. Language
74: 245-273.
Emonds, Joseph. 1987. The invisible category principle. Linguistic Inquiry 18: 613632.
Enq, Murvet. 1991. The semantics of specificity. Linguistic Inquiry 22: 1-25.
Erteschik-Shir, Nomi. 1997. The dynamics of focus structure. Cambridge:
Cambridge University Press.
Erwin, Sean. 1996. Quantity and moras: An amicable separation. In UCLA
Occasional Papers in Linguistics, I7: Th e structure of Malagasy, Eds. M.
Pearson and I. Paul: 2-30. Los Angeles: UCLA.

Fillmore, Charles. 1970. The grammar of hitting and breaking. In Readings in
English transformational grammar, Eds. R. Jacobs and P. Rosenbaum: 120133. Waltham, MA: Ginn.
Foley, William A. 1976. Comparative syntax in Austronesian. PhD thesis,
Berkeley, University of California, Berkeley.
Georgopoulos, Carol. 1991. Syntactic variables: Resumptive pronouns and A'
binding in Palauan. Dordrecht: Kluwer.
Gerdts, Donna. 1988. Antipassives and causatives in Ilokano: Evidence for an

ergative analysis. In Studies in Austronesian linguistics, Ed. R. McGinn: 295-
322. Athens, OH: Ohio University Center for International Studies.

Gibson, Jeanne. 1980. Clause union in Chamorro and Universal Grammar. PhD
thesis, San Diego, UC San Diego.
Goldberg, Adele. 1995. Constructions: A construction grammar approach to
argument structure. Chicago: University of Chicago Press.
Gross, Maurice. 1975. Methodes en syntaxe. Paris: Herman.
Gruber, Jeffrey. 1965. Studies in lexical relations. PhD thesis, Cambridge, MA,
MIT.
Guilfoyle, Eithne, Henrietta Hung and Lisa Travis. 1992. Spec of IP and Spec of
VP: Two subjects in Austronesian languages.Natural Language and Linguistic
Theory 10: 375-414.

Halpern, Aaron L. and Arnold M. Zwicky, eds. 1996. Approaching second: Second
position clitics and related phenomena. Stanford: CSLI.
Hanitriniaina, Saholy and Lisa Travis. 1998. F-nominals in Malagasy and

unparsing. CLA, University of Ottawa.
Hardjadibrata, RR. 1985. Sundanese: A syntactic analysis: Canberra: Australian
National University.
218
References
Harris, Zellig. 1976. Notes du cours de syntaxe. Paris: Seuil.
Heggie, Lorie. 1993. The range of null operators: Evidence from clefting. Natural
Language and Linguistic Theory 11: 45-84.
Hoeksema, Jacob. 1984. Partitives. ms, Groningen.

Hoekstra, Teun and Rene Mulder. 1991. Unergatives as copular verbs; locational
and existential predication. The Linguistic Review 7: 1-79.
Hornstein, Norbert. 1999. Movement and control. Linguistic Inquiry 30: 69-96.
Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cambridge,
MA: MIT Press.

Jackendoff, Ray. 1977. X-bar syntax. Cambridge, MA: MIT Press.
Jackendoff, Ray. 1996. The proper treatment of measuring out, telicity, and perhaps
even quantification in English. Natural Language and Linguistic Theory 14:
305-354.
Kayne, Richard.1975. French syntax: the transformational cycle. Cambridge, MA:
MIT Press.
Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press.
Keenan, Edward. 1976. Remarkable subjects in Malagasy. In Subject and topic, Ed.
C. Li: 249-301. New York: Academic Press.
Keenan, Edward and Maria Polinsky. 1996. Malagasy morphology. In Handbook
of morphology, Eds. A. Zwicky and A. Spencer: Oxford University Press.
Keenan, Edward. in press. Morphology is structure: A Malagasy test case. In
Formal issues in Austronesian linguistics, Eds. I. Paul, V. Phillips and L.
Travis: 23-41. Dordrecht: Kluwer.

Kiparsky, Paul. 1995. Partitive case and aspect. ms, Stanford University.
Kroeger, Paul. 1990. The morphology of affectedness in Kimaragang Dusun. ms.
Kroeger, Paul. 1993. Phrase structure and grammatical relations in Tagalog.
Stanford, CA: Center for the Study of Language and Information.

Ladusaw, William. 1982. Semantic constraints on the English partitive construction.
WCCFL I, Stanford University.

Lakoff, George. 1968. Instrumental adverbs and the concept of deep structure.
Foundations of Language 4: 4-29.
Larson, Richard. 1985. Bare-NP adverbs. Linguistics Inquiry 16: 595-621.

Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19:
335-391.
Maclachlan, Anna. 1996. Aspects of ergativity in Tagalog. PhD thesis, Montreal,
Canada, McGill University.
219
References
Maclachlan, Anna and Masanori Nakamura. 1997. Case checking and specificity in
Tagalog. The Linguistic Review 14: 307-333.
Malzac, R.P. 1960. Grammaire Malgache. Paris: Societe d'Editions
Geographiques, Maritimes et Coloniales.

Manorohanta, Cecile. 1998. A quantitative study of voice in Malagasy. AFLA-V,
University of Hawai'i.
Marantz, Alec. 1984. On the nature of grammatical relations. Cambridge, MA:
MIT Press.
Marantz, Alec. 1993. Implications of asymmetries in double object constructions. In
Theoretical aspects of Bantu grammar, Ed. S. A. Mchombo: 113-150.
Stanford: CSLA Publications.

Massam, Diane. 1998. Instrumental aki and the nature of Niuean transitivity.
Oceanic Linguistics 37: 1-17.
Massam, Diane. to appear. VSO is VOS: Aspects of Niuean word order. In The
syntax of verb-initial languages, Eds. A. Carnie and E. Guilfoyle. Oxford:
Oxford University Press.

Massam, Diane and Carolyn Smallwood. 1996. Essential features of predication in
English and Niuean. NELS 27, McGill University, Montreal, GLSA.
Matsuoka, Mikinari. 1999. Two independent positions for dative arguments. CLA,
University of Sherbrooke.
McCawley, James. 1988. Adverbial NPs: Bare or clad in see-through garb?
Language 64: 583-590.
Nakamura, Masanori. 1996. Economy of chain formation. PhD thesis, Montreal,
Canada, McGill University.
Naumann, Ralf and Anja Latrouite. 1999. On the interpretation of Tagalog voice
affixes. AFLA VI, University of Toronto, Holland Academic Graphics.
Paul, Ileana. 1996a. The active marker and nasals in Malagasy. In UCLA
Occasional Papers in Linguistics, l7: Th e structure fo Malagasy, Vol. l, Eds.
M. Pearson and I. Paul: 49-57.

Paul, Ileana. 1996b. F-words and other (Malagasy) nominals. CLA, Brock
University, Calgary Working Papers in Linguistics.
Paul, Ileana. 1996c. The Malagasy genitive. In UCLA Occasional Papers in
Linguistics, l7: Th e structure o
f Malagasy, Vol. l, Eds. M. Pearson and I.
Paul: 76-91.
Paul, Ileana. 1997. Malagasy existentials: Specificity and partitivity. ms, McGill
University.
220
References
Paul, Ileana. 1998. Focus movement and wh-questions in Malagasy. WECOL,
Arizona State University.
Paul, Ileana. in press. Malagasy existentials: A syntactic account of specificity. In

Formal issues in Austronesian linguistics, Eds. I. Paul, V. Phillips and L.
Travis: 57-74. Dordrecht: Kluwer.

Pearce, Elizabeth. 1999. Topic and focus in a head-initial language: Maori. AFLA
VI, University of Toronto, Toronto Working Papers in Linguistics.
Pearson, Matt. 1996a. Domain phrases and topic arguments in Malagasy
existentials. In UCLA Occasional Papers in Linguistics, l7:
Th e structure fo
Malagasy, Eds. M. Pearson and I. Paul: 113-141.

Pearson, Matt. 1996b. Pied-piping into the left periphery. NELS 27, McGill
University, Montreal, Canada, GLSA.
Pearson, Matt. 1996c. Raising and restructuring in Malagasy exitentials. MA thesis,
Los Angeles, CA, UCLA.
Pearson, Matt. 1998a. Event structure and the syntax of themes and instruments:
The case of the Malagasy Translative Voice (a-passive). AFLA-V, University of

Hawai'i.
Pearson, Matt. 1998b. Predicate raising and VOS order in Malagasy. In UCLA
Occasional Papers in Linguistics, 20: The structure of Malagasy, Vol. 2, Ed. I.
Paul. Los Angeles: UCLA.

Pearson, Matt. 1999. Direct and inverse order languages. GLOW, Berlin.
Pearson, Matt. forthcoming. Word order and the syntax of events: The case of the
Malagasy predicate. PhD thesis, Los Angeles, UCLA.

Pearson, Matt. to appear. Two types of VO languages. In OV and VO, Ed. P.
Svenonius. Amsterdam: John Benjamins.
Pensalfini, Robert. 1995. Malagasy phrase structure and the LCA. In MITWPL 27:
Papers on minimalist syntax, Eds. R. Pensalfini and H. Ura.
Perlmutter, David and Paul Postal. 1977. Toward a universal characterization of
passivization. Proceedings of the Third Annual Meeting of the Berkeley

Linguistics Society, University of California Berkeley.
Pesetsky, David. 1987. Wh-in-situ: Movement and unselective binding. In The
representation o f (in)definiteness, Eds. E. J. Reuland and A. ter Meulen: 98-
129. Cambridge, MA: MIT Press.

Phillips, Vivianne. in press. The interactions between prefix and root: The case of
maha- in Malagasy. In Formal issues in Austronesian linguistics, Eds. I. Paul,
V. Phillips and L. Travis: 75-92. Dordrecht: Kluwer.
221
References
Polinsky, Maria. 1994. The existential construction in Malagasy. Seventh
International Congress of Austronesian Linguistics, Leiden.
Pollock, Jean- Yves. 1989. Verb movement, Universal Grammar and the structure of
IP. Linguistic Inquiry 20: 365-424.
Postal, Paul. 1969. Anaphoric islands. CLS 5, University of Chicago.
Rabenilaina, Roger-Bruno. 1985. Lexique-grammaire du malgache. Constructions
transitives et intransitives. These de doctorat d'etat, Paris, Universite Paris VII.
Rabenilaina, Roger-Bruno. 1991. Voix et diathese en malgache.Linguisticae

Investi gationes 15: 325-335.
Rackowski, Andrea. 1998. Malagasy adverbs. In UCLA Occasional Papers in

Linguistics, 20: The structure o f Malagasy, VoL 2, Ed. I. Paul. Los Angeles:
UCLA.
Rackowski, Andrea and Lisa Travis. to appear. V-initial languages: X or XP
movement and adverbial placement. In The syntax of verb-initial languages, Eds.
A. Carnie and E. Guilfoyle. Oxford: Oxford University Press.
Rahajarizafy,Antoine de Padoue. 1960. Essai de grammaire malgache.
Antanimena Tananarive: Imprimerie Catholique.

Rahalalaoherivony, B. S. 1995. Lexique-grammaire du malgache: constructions
adjectivales. PhD thesis, Paris, Paris VII.
Raharinirina-Rabaovololona. 1991. Lexique-grammaire des composes du malgache.
Les adverbes de temps. PhD thesis, Paris, Paris VII.
Rajaona, Simeon. 1972.Structure du malgache: Etude des formes predicatives.
Fianarantsoa: Librairie Ambozontany.
Rajemisa-Raolison, Regis. 1966. Grammaire malgache. Fianarantsoa: Librairie

Ambozontany.
Randriamasimanana, Charles. 1986. The causatives of Malagasy. Honolulu:
University of Hawaii Press.

Rappaport, Malka and Beth Levin. 1988. What to do with theta-roles. In Syntax and
Semantics 2l: Thematic relations, Ed. W. Wilkins: 7-36. New York: Academic
Press.
Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics.

Philosophica 27: 53-94.
Richards, Norvin. 1996. Syntax versus semantics in Tagalog wh-movement. In
UCLA Occasional Papers in Linguistics 2l: Recent papers in Austronesian
linguistics, Ed. M. Pearson: 259-275. Los Angeles: UCLA.

Richards, Norvin. 1997. What moves where when in which language? PhD thesis,
222
References
Cambridge, MA, MIT.
Rizzi, Luigi. 1990. Relativized minimality. Cambridge, MA: MIT Press.
Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of grammar,
Ed. L. Haegeman: 281-337. Dordrecht: Kluwer.
Rooth, Mats. 1996. Focus. In The handbook of contemporary semantictheory, Ed.
S. Lappin: 271-297. Oxford: Blackwell.
Rudin, Catherine. 1988. On multiple questions and multiple wh fronting. Natural
Language and LinguisticTheory 6: 445-501.
Saito, Mamoru. 1994. Additional-wh effects and the adjunction site theory. Journal
of East Asian Linguistics 3: 195-240.
Seiter, William. 1979. Instrumental advancement in Niuean. Linguistic Inquiry 10:
595-621.
Sells, Peter. 1998. The functions of voice markers in the Philippine languages. In
Morphology and its relation to phonology and syntax, Eds. S. Lapointe, D.
Brentari and P. Farrell: 111-137. Stanford: CSLI Publications.

Shibatani, Masayoshi. 1985. Passives and related constructions: A prototype
analysis. Language 61: 821-848.
Sityar, Emily. in press. The topic and y indefinite in Cebuano. In Formal issues in
Austronesian linguistics, Eds. I. Paul, V. Phillips and L. Travis: 131-153.
Dordrecht: Kluwer.
Sproat, Richard. 1988. On anaphoric islandhood. In Theoretical morphology:

Approaches in modern linguistics, Eds. M. Hammond and M. Noonan: 291-
301. New York: Academic Press.

Stowell, Tim. 1985. Null antecedents and proper government. NELS 16, McGill
University.
Szabolcsi, Anna. 1981. The semantics of topic-focus articulation. In Formal
methods in the study of language, Eds. J. Groenendijk, T. Janssen and M.
Stokhof: 513-541. Amsterdam: Matematisch Centrum.

Szabolcsi, Anna and Frans Zwarts. 1992-1993. Weak islands and an algebraic
semantics for scope taking. Natural Language Semantics 1: 235-284.
Tanaka, Hidekazu. 1998. Conditions on logical form derivations and

representations. PhD thesis, Montreal, Canada, McGill University.
Tenny, Carol. 1994. Aspectual roles and the syntax-semantics inter face. Dordrecht:
Kluwer.
Travis, Lisa. 1984. Parameters and effects of word order variation. PhD thesis,
Cambridge, MA, MIT.
223
References
Travis, Lisa. 1988. The syntax of adverbs. McGill Working Papers in Linguistics,
SpecialIssue on Comparative Germanic Syntax: 280-310.
Travis, Lisa. 1996. Achievements and external theta-role assignment: suggestive

data from Malagasy. AFLA-3, UCLA.
Travis, Lisa. 1997. Theta-positions and binding in Balinese and Malagasy. AFLA-4,
UCLA.
Travis, Lisa. 1999. Objects in Malagasy and in English. The role of grammatical
functions in transformational syntax, University of Illinois, Urbana-Champaign.
Travis, Lisa. in press. The 1-syntax/s-syntax boundary: Evidence from
Austronesian. In Formal issues in Austronesian linguistics, Eds. I. Paul, V.
Phillips and L. Travis: 151-175. Dordrecht: Kluwer.

Travis, Lisa. to appear. T-positions and binding in Balinese and Malagasy.
Canadian Journal of Linguistics.
Voskuil, Jan. 1996. Comparative morphology. Verb taxonomy in Indonesian,
Tagalog and Dutch. Leiden: HIL Dissertations.
Wechsler, Stephen and I Wayan Arka. 1998. Syntactic ergativity in Balinese: An

argument structure based account. Natural Language and Linguistic Theory 16:
387-441.
Wierzbicka, Anna. 1968. On the semantics of the verbal aspect in Polish. In To
honor Roman Jakobson: Essays on the occasion o f his seventieth birthday,
2231-2249. The Hague: Mouton.

Williams, Edwin. 1980. Predication. Linguistic Inquiry 11: 203-238.
Williams, Edwin. 1981. Argument structure and morphology. The linguistic review
1: 81-114.
Wojcik, R. 1976. Where do instrumental NPs come from? In Syntax and Semantics
6: The grammar of causative constructions, Ed. M. Shibatani. New York:

Academic Press.
Zribi-Hertz, Anne and Liliane Mbolatianavalona. 1999. Towards a modular theory
of linguistic deficiency: Evidence from Malagasy personal pronouns. Natural
Language and Linguistic Theory 17: 161-218.
Zubizarreta, Maria Luisa. 1998. Prosody, focus, and word order. Cambridge, MA:
MIT Press.
224

Malagasy Clause Structure (2000) - Ileana M. Paul

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Malagasy Clause Structure (2000) - Ileana M. Paul

Caricato da

Copyright:

Formati disponibili

Malagasy clause structure

A thesis submitted to the Faculty of Graduate Studies

2000 Ileana M. Paul

Bobaljik very g e nerously o f fered comments o n v e r sions o f c h a pter 2

M ada g a scar i n 19 9 5 a n d for

encouragement over t h e y e ars. F u r t he r a p p r eciation t o R o g er-Bruno

Saho l y H an i t r i niaina, M i c h ele

Ratovonony and Julie Rabemananjara. Mi s oatra betsaka, tompoko! I tha nk

w o u l d l i k e t o t h ank th e a u diences at the A F L A

t h ei r i n p u t a n d s u p p o rt, especially Bill D a v ies, Diane

Massam, Matt Pearson, Peter Cole and G abriella H e rm on.

minute, panic-ridden email messages.

T h a nk s also to m y e x i n -laws for all

f a m i ly, especially my p arents, have been wonderfully

generous with love and understanding.

Chapter 3: Circumstantial Topic

Gui l f o y le, Hung and Travis.....

6.3.1 P r e p o sitions and partitivity

2 The left periphery.

One of the most striking properties of western Austronesian languages is

C e r t ai n m o r p h emes ap pear t o e n c ode t h e

ity h azo ity

h i s tree this p s t .P.gen.det knif e S a hondra

'Sahondra cut this tree with the knife.'

Nota p ahin'i Sahondra

ants y i t yhazo ity.

pst.CT.cut.gen.Sahondra this tree this d e t knife

I n (1a), the agent i s t h e s u bject an d t h e v e r b b e ar s A c t orTopic (A T )

s e nsitive t o t h e t a-roles, category, p o sition, f u n ction, o r

s omething else? Before addressing the voice system of M a l agasy, I wi ll

M a n y n a t ural languages display alternations similar t o t h e

ones in (2) and (3).

Felix cut the bread.

The b r ead was cut by Felix.

Felix a coupe le pain.

a. A direct object of an active clause is the (superficial) subject of the

Due to o t her l aw s o f g r a m ma r ( e .g. th e Stratal Un iqueness Law), the

d e m oted i n a pa s s iv e c l ause, often s u r f acing a s a

" chomeur" (oblique). T hus passive has a dual nature: th e p r o m otion o f

o p t i onally a p pear as an o b l i que adjunct. T h e s e t w o

properties are related in what is known as Burzio's Generalization (Burzio

A verb which lacks an external argument fails to assign accusative

As in RG, GB passive involves two components, here linked to Case and

example, both RG and GB treat passive as consisting of two-parts. But do

O o nathan B o baljik, p . c.) an d i m p e r sonal

and (3).' S p ecifically, passive is defined as a transformation that affects

d i s t inction: a c tive, p a ssive, c i r cumstantial (sometimes

active sentences, concentrating on the passive and circumstantial and on

Let us now t ur n t o e xisting analyses of voice in w e stern A u s tronesian.

Guilfoyle, Hung and Travis (1992) (henceforth GHT) propose a purely

(1), repeated in (6).

ity h azo ity

thi s tree this p s t .P.gen.det knife S a hondra

'Sahondra cut this tree with the knife.'

Nota p ahin'i Sahondra

ants y i t yhazo ity.

pst.CT.cut.gen.Sahondra this tree this

( 6 b), on th e other h and, the theme is not Case-

m arked in i t s b ase p osition an d r a i ses to t h e s u bject p o sition . On e

G H T a n d o t h e rs, i n o n e r e s pect, th e c i r cumstantial

the forms in (6) is given in (7).

T he above p arallels have le d s om e r e searchers t o c l ai m t h a t t h e

T h i s i s t r u e o f b o t h t r a d i tional g r a m m ars (e.g. Rajemisa-