Vous êtes sur la page 1sur 38

Theta Theory and the Syntax of Applicatives in Chichewa

Author(s): Mark Baker


Source: Natural Language & Linguistic Theory, Vol. 6, No. 3 (Aug., 1988), pp. 353-389
Published by: Springer
Stable URL: http://www.jstor.org/stable/4047650 .
Accessed: 23/07/2011 15:30
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=springer. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

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

http://www.jstor.org

MARK BAKER

THETA

THEORY

OF

AND THE SYNTAX

APPLICATIVES

1. INTRODUCTION:
INSTRUMENTAL

IN CHICHEWA*

BENEFACTIVE AND
APPLICATIVES

Consider the following sentences from the Southern, lake dialect of


Chichewa, a Bantu language spoken in Malawi:1
(1)

Mavuto a- na-

umb -a

mtsuko.

Mavuto SP PAST mold ASP waterpot


Mavuto molded the waterpot.
(2)

Mavuto a- na-

umb -ir

-a

mfumu mtsuko.

Mavuto SP PAST mold APPL ASP chief

waterpot

Mavuto molded the waterpot for the chief.

* This paper could not have been written without Sam Mchombo,who has generously
sharedboth his native-speakerintuitionsand his linguistideas about Chichewasyntaxwith
me. My highest thanks go to him; he is not to be held responsiblefor my (mis?)interpretationsof what he has said. During the period when the researchfor this paper was
done, Mchombowas supportedby a grant from the System DevelopmentFoundationto
Kenneth Hale. In addition, I have benefited from comments and input from Noam
Chomsky, Ken Hale, Kyle Johnson, Richard Larson, Luigi Rizzi, Lisa Travis, and
audiencesat Brandeis,MIT, UC-Irvine,and the Universitedu Quebec 'aMontreal,as well
as the entirelinguisticscommunitiesof MIT and McGillUniversity.Finally,this articlehas
been improvedby the carefulattentionof the editorsand severalanonymousreviewersfor
NLLT.

Abbreviationsused in the glosses in this article are: A, gender agreement; ABS,


absolutive; APPL, applied affix; ASP, mood morpheme;EX, exclusive; HAB, habitual
tense; OP, object prefix;PASS, passive; PRES, present; RECIP, reciprocal;SP, subject
prefix;SUF, suffix.Glossesof pronominalagreementconsistof a numberindicatingperson,
a lower case letter indicating number (singular or plural), and an upper case letter
indicating grammaticalfunction (subject or object). Noun classes are not indicated in
the glosses, but certain agreement relationships are shown by highlighting. The
notation... *(B) ... means that the structure is grammaticalonly if B is included;
... (*B) ... meansthat it is grammaticalonly if B is not included.Tone is not marked.
I The judgmentsreportedhere differsubstantiallyfrom those given in Trithart(1977), an
importantpreviouswork on Chichewa.Trithart'sfield work was carriedout with speakers
froma differentregionof Malawi,and the dialectshe reportsdifferssystematicallyfromthe
one describedhere (sometimescalled 'Chinyanja')in some ways (Baker 1985). Her work
also containsa few importantmistakes,accordingto Bresnanand Mchombo(1987).
Natural Language and Linguistic Theory 6 (1988) 353-389
Publishers

? 1988 by Kluwer Academic

354

MARK

(3)

BAKER

-a mpeni mtsuko.
umb -ir
Mavuto a- naMavuto SP PAST mold APPL ASP knife waterpot

Mavutomoldedthe waterpotwitha knife.


Examples (2) and (3) illustratetwo types of APPLICATIVE constructionconstructionsin which a noun phrase that otherwisecould be expressed
only as an oblique phrase or not at all looks like a direct object of the
verb. I will refer to this new or 'promoted' argument as the APPLIED
OBJECT. In contrast,the argumentwhich correspondsto the direct object
in simple sentences like (1) (e.g. mtsuko 'pot') will be called the BASIC
OBJECT of the construction.In (2), the applied object is semanticallya
beneficiary;in (3) it is an instrument.In both constructions,there appears
on the verb an additional affix -ir,2 called the APPLIED AFFIX. These
applicative constructions are similar to dative shift constructions in
English, except for this morphological marking and the fact that the
process is productiveover a wider range of verb types. In RELATIONAL
GRAMMAR
(RG), applicatives are commonly known as OBLIQUE-TO-2
ADVANCEMENT clauses. They are a salient feature of the Bantu family,
and are found in many other languages as well.
Superficially,the only difference between the benefactive applicative
in (2) and the instrumentalapplicativein (3) is that the appliedobject is
animate in the former and inanimate in the latter. A closer look,
however, reveals importantasymmetries.For example,in the benefactive
construction,3only the applied object can be expressed by the object
prefixon the verb:4
2 Nonlow vowels in Chichewasuffixesundergovowel harmony.Thus, the appliedaffixis
[ir] after tense vowels [i, u, a], and [er] after lax vowels [e, o] or vowelless roots (Mtenje
1984). I use the tense vowel variantsas the citationforms.
The applied affix is used in certain other constructionsin Chichewa which are not
discussedin this article;some of these are mentionedin note 3 and in note 21.
3 Chichewaalso has goal applicativeconstructionssuch as (ib):

(i)a.

b.

Mavutoa- na- perek -a


chitseko kwa mfumu.
MavutoSP PAST hand ASP door
to chief
Mavutohandedthe door to the chief.
Mavutoa- na- perek -er
-a
mfumuchitseko.
MavutoSP PAST hand APPL ASP chief door

These have the same propertiesas the benefactiveapplicativesin (2).


4 Mchombo (1986) and Bresnanand Mchombo (1987) argue that the Chichewaobject
prefixis in fact a pronominalclitic, ratherthana simpleagreementmorpheme.Unlikesome
Bantulanguages,Chichewaoptionallyallowsthis clitic to be doubledby an NP matchingit
in noun class features;nevertheless,the doublingNP cannot appearinsidethe verb phrase,
as Bresnanand Mchomboshow. Among other things,this accountsfor the wordordershift
seen by comparing(4a) (withana 'children'included)to (2).

THE SYNTAX

(4)a.

OF APPLICATIVES

IN CHICHEWA

355

mtsuko (ana).
wa- umb -ir
-a
Mavuto a- naMavuto SP PAST OP mold APPL ASP waterpotchildren
Mavuto molded the waterpotfor them (the children).

b.

*Mavutoa- na-a
ana
(mtsuko).
u- umb -ir
Mavuto SP PAST OP mold APPL ASP children waterpot

Mavuto molded it (the waterpot)for the children.


In the instrumentalconstruction,however, either the applied object or
the basic object may be expressedas the object prefix:
-a
mitsuko (mpeni).
(5)a. Mavuto a- na- u- umb -ir
Mavuto SP PAST OP mold APPL ASP waterpotsknife
Mavuto molded the waterpbtswith it (the knife).
b.

-a
Mavuto a- na- i- umb -ir
mpeni (mitsuko).
Mavuto SP PAST OP mold APPL ASP knife waterpots

Mavuto molded them (the waterpots)with a knife.


A second difference appears in Wh-movement constructions such as
relative clauses. A simple example of a Chichewarelative is:
kuti
(6)
lyi ndi mfumuimene ndi- ku- ganiz -a
This is chief whom 1sS PRES think ASP that
on -a.
Mavuto a- naMavuto SP PAST see ASP
This is the chief whom I think that Mavuto saw.
In an instrumentalconstructioneither object may be relativized(8), but
in a benefactive constructionthe appliedobject cannot be (7):
kuti
(7)a. *lyi ndiyo mfumuimene ndi- ku- ganiz -a
This is
chief whom 1sS PRES think ASP that
-a
Mavuto a- na- umb -ir
mtsuko.
Mavuto SP PAST mold APPL ASP waterpot.
This is the chief whom I think Mavuto molded the waterpot
for.
b.

Uwu ndiwo mtsuko umene ndi- ku- ganiz -a


kuti
This is
waterpotwhich lsS PRES think ASP that
-a
Mavuto a- naumb -ir
mfumu.
Mavuto SP PAST mold APPL ASP chief
This is the waterpot which I think Mavuto molded for the
chief.

356

MARK

(8)a.

BAKER

Uwu ndi mpeni umene ndi- ku- ganiz -a


kuti Mavuto
This is knife which lsS PRES think ASP that Mavuto
a- naumb -ir
-a
mitsuko.
SP PAST mold APPL ASP waterpots
This is the knife which I think Mavuto molded the waterpots
with.

b.

ndi mistuko imene ndi- ku- ganiz -a


lyi
kuti
These are waterpotswhich lsS PRES think ASP that
Mavuto a- na- umb -ir
-a
mpeni.
Mavuto SP PAST mold APPL ASP knife
These are the waterpotswhich I think Mavuto molded with a
knife.

These two cases clearly have something in common; as a descriptive


generalizationcovering both, we can say that there seem to be two NPs
which can act like a direct object in instrumental applicative constructions,but only one in benefactive applicativeconstructions.
The contrasts in (4)-(5) and (7)-(8) are worthy of study for several
reasons.First,generative approachesto linguisticsseek to restrictthe set
of constructionsfound in naturallanguages in substantiveways. In light
of both the similarities and the differences between benefactive and
instrumentalapplicatives,the challenge is to define this set neither too
narrowly,thereby excluding one of the existing constructions,nor too
broadly, thereby making constructions that are never attested a
theoreticalpossibility.
The issue can be sharpened from the familiar language acquisition
perspective. Presumably,the Malawianchild sooner or later hears sentences like (2), (3), (4a), (5), (7b) and (8). Why then does he or she not
make the obvious generalizationthat (4b) and (7a) are also part of the
language?
Finally,the intuitiveidea that instrumentalapplicativeshave two direct
objects and benefactive applicatives have only one has a serious flaw:
neither of the postverbal NPs in benefactive constructions can consistently be identifiedas the sole object. Rather, the applied object has
the capacity to control the object prefix, while the basic object has the
potential of being relativized.This curious skewing of object properties
complicates the identification of grammaticalrelations and needs an
explanation.
In this article, I will give an analysis of Chichewa applicative con-

THE SYNTAX

OF APPLICATIVES

357

IN CHICHEWA

structions in the GOVERNMENT-BINDING (GB) theory framework of


Chomsky(1981, 1986a), which accounts for these patternsas well as for
other aspects of theirsyntax.In the process, I will show how the resultsof
the analysis bring importantlight to theoretical issues of how semantic
relationshipsare representedin syntax across languages.
2. A

THETA

THEORETIC

ASYMMETRY

The fundamentalclaim will be that the differencesbetween benefactive


and instrumentalapplicatives are due to a technical sense, based on
Theta theory, in which clauses with instrumentsdo indeed have two
objects, while those with benefactives have only one. In this, I develop
an insight of Marantz (1982, 1984). Marantzdiscusses differences between benefactive and instrumentalapplicativesin Fula, and attributes
them to a differencein how benefactive and instrumentalphrasesreceive
THEMATIC (6-) ROLES. Specifically,he claims (followinga suggestion of
Dick Carter) that benefactives actually receive their 0-roles from a
prepositional element, but instruments do not. The Chichewa applicatives show that this idea is correct in essence, but that Marantz's
analysisneeds to be revised and extended in importantways.
2.1. Marantz'sAnalysis
First, I brieflyreview Marantz'sproposal,to show why it does not fit the
Chichewafacts as it stands.The discussionwill be ratherinformal,and I
will translate some of Marantz's terminology into more familiar GB
terms; the interested reader should consult the cited references for
importanttechnical points.
Marantz assumes that benefactives are underlyingly VP modifiers
which are assigned a 6-role by a preposition(9a). Instrumentsare also
VP modifiers, but they receive their 6-role from the VP itself, the
prepositionserving only to transmitthis 6-role (9b):
VP

(9)a.

b.

PP

VP

NP
pat

obj

VP

P
\

NP
ben

obj

PP

VP

NP

\S,tt

obj

NP

ist

358

MARK

BAKER

Marantz identifies the P elements in these structureswith the applied


affixesin (2) and (3). Applicativeconstructionsarise because these affixes
MERGE with the V to form a single word. When this happens, the new
relationshipsbetween the NPs and the derived verb are determinedby
principlesof morphologicalfeature percolationbased on Lieber (1980).
According to these principles, the properties of an affix always take
precedence over the propertiesof a root in determiningthe propertiesof
the combination.Therefore, since the benefactive phrase is the object
(=semantic role assignee) of the affixin (9a), it will be the object of the
combination, the basic object of the verb thus being relegated to a
secondaryfunction (see (lOa)). In contrast,the instrumentalphraseis not
the object of the prepositional affix in (9b). This allows the root to
determinethe object of the combination,and it is the instrumentalphrase
that is left to appear in a secondary function in the derived structure
(lOb):
(10)a.

VP

V + P NP
ben
obj

b.

NP
pat

VP

V + P NP
gat
obj

NP
instr

In this way, the thematic difference between benefactives and instruments projects into a difference in the syntax of the two types of
applicatives.
The basic generalizationderived from Marantz'sassumptionsis that in
benefactive constructionsthe applied object will have distinctive direct
object properties,whereas in instrumentalconstructionsthe basic object
will have such properties.5 The empirical basis of this claim comes
primarilyfrom the Niger-Congo language Fula: Sylla (1979) states that
in this language when a benefactive applicative verb is passivized, the
benefactive NP becomes the subject; when an instrumentalapplicative
verb is passivized,the patient NP becomes the subject.
Marantz'sempirical generalizationdoes not extend to the Chichewa
5 In fact, Marantz(1984, pp. 248-51) also allows for the possibilitythat instrumental
prepositionscan assignthematicroles in some languages,makingthem essentiallyidentical
to benefactives.In such languages,the instrumentratherthan the basic object will be the
direct object of the verb. Thus, the Chichewa instrumentscould be analyzedas being
ambiguousbetweenthese two possibilities,therebycapturingthe fact thateitherNP can act
like the object of the verb in essentiallyfree variation.I will develop a strongerposition,in
which there is no ambiguityin how thematicroles are assigned.

THE

OF APPLICATIVES

SYNTAX

IN CHICHEWA

359

applicatives,however. Half of the contrastin (4), (5) is consistentwith his


analysis:it is expected that the basic object can control the object prefix
in instrumentalconstructions,but not in benefactive constructions.It is
not expected that the appliedobject can also control the object prefixin
instrumentalconstructions,however.6Moreover,while Marantzdoes not
explicitlydiscuss Wh-movementin applicativeconstructions,the facts in
(7) are the opposite of what one might expect given the representationin
(lOa); in resisting extraction,benefactive applied objects behave unlike
direct objects rather than like them. Thus, more needs to be said about
these applicativeconstructions.
2.2. An AltemativeProposal
Rather than developing Marantz'sapproach per se to account for the
Chichewa data, I will explore an alternativeproposal,which differsboth
in how benefactives and instrumentalsare represented at D-structure,
and in the mechanisms that derive a surface structure from these
representations.
First,I assumethat benefactive and instrumentalphrasesare (optional)
arguments of the verb, rather than VP modifiers (cf. Baker, 1985).
Hence, at D-structure these phrases will be daughtersof the VP rather
than sisters of it. However, I maintain the core idea that in this new
location, benefactive NPs depend on a prepositionfor their 6-roles, while
instrumentalNPs depend only on the verb. Instrumentalprepositions,if
present at all, appearfor other reasons,such as Case theory (see section
3).
(11)a.
BENEFACTIVE
b.
INSTRUMENT
VP

PP

P
6

VP

NP

)n
ben

NP

iP)nP
instr

More precisely,the fact that the verb can agree with the instrumentis surprisinggiven
that it cannotagree with the basic object in a benefactiveapplicative:see (4b). If both (4b)
and (5a) were grammatical,one could simplysay that verbs in the languagecan have two
objects - i.e. two NPs which are assigned accusative Case. Then there would be few
differencesbetweenthe two NPs in eitherapplicativeconstruction,whichis consistentwith
Marantz'sassumptions.This state of affairsis found in some other Bantulanguages,such as
Kinyarwanda(Kimenyi1980; Marantz1984; Baker 1985), but not in Chichewa.

360

MARK

BAKER

I will call the relationshipbetween the circled NP and the verb in (1la)
one of INDIRECT 0-ROLE ASSIGNMENT, since the verb in fact 0-marksthe
PP and not the NP itself; the analogous relationshipin (1lb) I will call
DIRECT 0-ROLE ASSIGNMENT. Presumablythese two structures(together
with the familiarcase of PP adjuncts,in which the P assigns a 0-role to
the NP but receives none from the verb) exhaustthe possible D-structure
configurationsthat can correspondto S-structurePPs, given that the NP
must receive a 0-role by the Theta Criterionof Chomsky (1981). I will
show that the two must be distinguished,and that benefactives (and
goals; cf. note 3) are alwaysassociatedwith (lla), while instrumentsare
associatedwith (1lb).7
The mechanismsfor combiningthe verb and the appliedaffixare also
different from those assumed by Marantz. Rather than employing a
structure-destroyingprocess of merger, which combines the projections
of the merged elements, I will assume that combinationis the result of
INCORPORATION as in Baker (1985, 1988). Incorporationin this sense is
simply an instance of the generalizedtransformationMove a (Chomsky
1981, 1986a) that moves a word ratherthan a whole phrase (i.e. an X?
insteadof an Xm"), adjoiningit to anotherword. As with other instances
of Move a, the Projection Principle of Chomsky (1981) requires that
moved Xos leave tracesbehindwhen necessaryto preservethe categorial
expressionof argumentrelationships.Where and when such an X? can
move will then be restricted,because any trace it leaves behind must be
PROPERLY GOVERNED, in accordancewith the EMPTY CATEGORY PRINCIPLE (ECP) of Chomsky (1981) and subsequent work. For concreteness, I assume the following formulationsof these notions:

(12)

(ECP)
Every trace B must be governed by some element A, where
either
i. A is a lexical category that assigns B a 8-role, or
ii. A is coindexed with B by movement.

(13)

A GOVERNS B if and only if:


i. A c-commandsB, and

THE EMPTY CATEGORY PRINCIPLE:

This hypothesismust ultimatelybe stated somewhatmore abstractly.Observationally,


benefactives crosslinguisticallyappear with adpositions, with applicative verbs, with
semanticcase endings,or with serialverbs. Baker (1988) and the currentarticleshow how
the first two of these can be unified theoreticallyby the notion of Incorporation,but
ultimatelythe second two mustbe integratedinto an accountas well. This is left to further
research.

THE SYNTAX

(13)

OF APPLICATIVES

IN CHICHEWA

361

ii. no more than one maximalprojectiondominatesB but not


A, and

iii. any maximal projection that dominates B but not A is


0-markedby a lexical category (cf. Chomsky 1986a).
Principle(12) is a standardformulationof the ECP (cf. Chomsky 1986a,
p. 17); (13) is based on Johnson(1987), who develops certainideas found
in Kayne (1984).8 Given that 0-roles are alwaysassigned to phrasesand
not to heads, the trace of an X? can never be sanctionedby clause (i) of
(12); hence, it must be governed by its antecedent. In order for this to
happen, the X? must move from the head position of a complementof a
given verb to adjoin to that verb. If insteadit moves out of the subject of
the verb (or out of any phase in a higherclause), it will fail to c-command
its trace, violating (13i). If it moves out of a more deeply embedded
position within a complement,then (13ii) will be violated. Finally,(13iii)
makes it impossiblefor the X? to move out of an adjunctphrase. Thus,
the following statementfollows from (12) and (13) (see Baker (1988) for
details and discussion):
(14)

An X? category B may adjoin to anotherX? category A only


if A 0-marks the smallest maximal projection containing B
(cf. the HEAD MOVEMENT CONSTRAINT of Travis (1984) and
Chomsky(1986a)).

Crosslinguistically,this explains many facts. For example, in languages


with productive noun incorporation,the head of the direct object may
incorporateinto the verb, but the head of the subject, the head of the
object of a locative preposition,or the head of an NP adjunct may not
(see Baker (1988) and referencescited there).
These mechanismsapply to applicativeconstructionsin the following
way. When the prepositionin (1la) is moved, it must leave a trace of
category P, which will assign the benefactive 0-role to the strandedNP at
LF. The Projection Principle requires that this trace be present at
S-structureas well, since by assumptionthe verb 0-marksa benefactive
8

In fact, (13) combinesfor ease of expositiontwo differentnotionsof governmentwhich


Johnson (1987) argues should be kept distinct: normal (lexical) government,for which
(13ii) is relevant, and antecedent government,for which (13iii) (and more generallythe
mechanismsof Chomsky(1986a))are relevant.Johnsonfurtherassumesthattracesmustbe
governed in both of these ways. For XO movement,the two conditionscollapse into one,
however, because the antecedent is attached to the potential lexical governor, the two
therebyhaving the same structuralposition.
More or less the same distinctionsare made by the more complex definitionsgiven in
Baker (1988, section 7.1); the empiricaldifferencesbetweenthe two are not relevanthere.

362

MARK

BAKER

PP, and thus must have a PP sister at every syntacticlevel. This tracewill
be properlygoverned by its antecedent,since the configurationdescribed
in (14) is satisfied.9In contrast, if the P in (llb) is moved, it need not
leave a trace, since it is not part of the lexically determined thematic
structureof the clause; here the verb 0-marksan NP rather than a PP,
and the P does not assign a thematicrole of its own. Thus, the Projection
Principleand the Theta Criterionguaranteethat the structuraldifference
between instrumentaland benefactive phrases shown in (11) will be
preservedat S-structureand LF, as follows:
(15)a.

BENEFACTIVE

b.

INSTRUMENTAL

VP

VP

PP

V+Pi
tj

NP
pat

V + P NP NP
instr pat

NP
ben

The VP in (15b) contains two NPs which are directly 0-markedby the
verb, whereas the one in (15a) has only one NP which is directly
0-marked. Thus, there is a technical sense in which instrumentalconstructions have two objects while benefactive constructionshave only
one: the relevant sense of object is a structuraland ultimatelya Thetatheoretic one. This suggests that this system is well suited to derive the
observed syntactic differences between the two Chichewa applicatives
described in section 1. Before showing in detail how this can be done,
however,we provide independentempiricalevidence for the assumptions
about 0-role assignmentlaid out in (11) and (15).
2.3. Evidence from Noun Incorporation

Given that generalization(14) has solid theoreticalmotivationand broad


empirical support (Baker 1988), it can be used as a test for argument
structurein unclearcases. In particular,suppose that the head of a given
9 Note that the incorporabilityof benefactive Ps is theory-internalevidence for the
assumptionthat benefactivePPs are complementsof the verb. This distinguishesthemfrom
locative and temporal adjuncts, whose prepositionalheads cannot incorporateinto the
verb, as discussedin Baker (1985).
In (lSa) we must assumethat the complexverb does not count for c-command,to allow
the P to governits trace.This is truefor manycurrentdefinitionsof c-command(see Borer
1984; Baker 1988).

THE SYNTAX

OF APPLICATIVES

IN CHICHEWA

363

type of nominal can incorporate directly into the verb in some set of
languages. It then follows that (at least for those languages) such
nominalsare direct argumentsof the verb. Supposethat, in contrast,the
head of anothertype of nominaldoes not incorporateinto the verb in any
language. It would then be reasonableto assumethat this second type of
nominal is not directly 0-markedby the verb; its failure to incorporate
would then follow from knownprinciples.
With this in mind, I observe that there is an asymmetrybetween
instrumentsand benefactives/goalswith respect to noun incorporationin
polysyntheticlanguages. Instrumentsdo incorporatein some such languages, just like direct objects. The following data illustratethis in the
Austronesianlanguage Niuean, and in Nahuatl:
NIUEAN: (Seiter 1980)
(16)
kai tiimau a mautolu
a. Ne
fa

aki e

tau lima.

PAST HAB eat always ABS-we (EX) with ABS PL hand

We would alwayseat with the hands.


b.

Ne

fa

kai lima tiimau a

mautolu.

PAST HAB eat hand always ABS we (EX)

We would alwayseat by hand.


NAHAUTL: (Merlan1976)
(17)
a. Ne? 0- panci -tete?ki ika kocillo.
he

3sS bread cut

with knife

He cut the breadwith a knife.


b.

Ya7 kiko'cillo -tete?ki panci.


he 3sS130 knife cut
bread

He cut the bread with a knife.


Benefactives and goals, however, do not productively undergo noun
incorporation,even when animacy factors can be controlled for. Examples (18) and (19) illustratethis for Niuean and for SouthernTiwa:
(18)
NIUEAN: (Seiter 1980)
a. Ne
tutala a
au ke he tau tagata.
PAST talk

ABS I

to

PL person

I was talking to (the) people.


b. *Ne

tutala tagata a

PAST talk

au (ke he).

person ABS I

I was talking to people.

to

364

MARK

(19)
a.

BAKER

SOUTHERN TIWA: (Allen, Gardiner, and Frantz 1984)


Ta'u'u- wia -ban
hliawra -de.
woman SUF
baby
give
PAST
lsS:AIA
I gave the woman the child.

b. *Tahliara- ('u'u) -wia -ban.


lsS:A/A woman baby give PAST
I gave the woman it (the baby).
Example (19b) shows that the benefactive/goal cannot be incorporated,
whether or not the theme argument is also incorporated.
This asymmetry seems to be universal. Mithun (1984, p. 875) states the
following generalization about the possible semantic roles of incorporated elements, based on her extensive cross-linguistic investigation of
Noun Incorporation:
If a language incorporatesonly two types of argumentsthey will be patientsof transitive
and intransitiveVs - again, regardlessof the basic case structureof the language. The
majority of incorporatinglanguages follow this pattern. Many languages additionally
instruments
incorporate
[emphasismine] and/orlocations,such as Nahuatl(Andrews1975);
Takelma,a languageisolate of Oregon (Sapir1922); and Sora,a SouthMundalanguageof
India (Ramamurti1931).

Benefactive and goal arguments are conspicuously absent from this list of
possible incorporated elements. The same researcher states explicitly that
benefactive/goal incorporation is impossible in Tuscarora (Iroquoian;
(Mithun) Williams 1976, p. 56). Thus, I conclude that instrumental
incorporation is not uncommon,10 whereas benefactive incorporation
and goal incorporation do not exist in natural language.
This difference in incorporation possibilities between superficially
similar phrases supports the hypothesis about theta marking stated in
(11): instruments incorporate, showing that they receive their thematic
roles directly from the verb; benefactives do not incorporate because the
projection of the P required for 0-role assignment blocks movement.
Example (19) is particularly instructive in this regard, inasmuch as the
benefactive phrase looks superficially like a direct object: it appears
without adposition or Case marking and it triggers object agreement on
the verb. Nevertheless incorporation is still impossible. Our fundamental
10 A stronger statement- that instrumentscan incorporatein every language in which
noun incorporationis productive- is false; cf. SouthernTiwa (Allen, Gardiner,and Frantz
1984). I have no explanationfor this, but speculate that the reason may be due to Case
theory,along the lines of Baker(1988, sec. 3.4.4). A less attractivealternativewouldbe to
parameterize0-role assignmentto instruments;see note 5 and section 6.2.

THE SYNTAX

OF APPLICATIVES

IN CHICHEWA

365

hypothesis implies that there must be a benefactive P in this structure


even though none is apparenton the surface;the object propertiesof the
benefactive then result from the fact that this P is incorporatedinto the
verb. P incorporationforces the verb to assign structuralCase to the
benefactive NP (for reasons discussed below); the V governs this NP
since there is only one maximalprojection that contains the NP and not
the V (namelythe PP node), and that projectionis 0-markedby the verb.
Nevertheless, the head of the benefactive still cannot incorporate:
(20)

*Tahliawra1-wia -0i
-ban [pp ti[Np tiJ].
lsS:AIA woman give APPL PAST

The moved N hli"wr 'woman'does not govern its trace, since there are
two maximalprojections (PP and NP) that contain the trace but not the
moved N. Hence, the structureis ruled out by the ECP. Our assumptions
about Theta theory guarantee that this will always be the case with
benefactives, unlike instruments.
Note that this difference in incorporationis not immediatelyexplained
by Marantz'sapproach.For him, instrumentsare like subjectsratherthan
like objects inasmuchas they are sisters of the VP rather than the V.
Structuralsubjects never incorporate(Baker 1988), and it is not obvious
why instrumentsshould be different.On the other hand, benefactivesare
structurallyidentical to direct objects after merger (cf. (lOa)), and it is
unclear why they do not incorporate like direct objects.t Thus, the
pattern of polysyntheticphenomena supports the revision of Marantz's
hypothesis over the original, as well as over various other imaginable
possibilities.12
3.

CASE

ASSIGNMENT

AND

OBJECT

PREFIXES

IN APPLICATIVES

With this information about benefactive and instrumentalphrases in


hand, we have the notions that we need to find an explanationof the
1I1Marantzcould solve this problemby a type of extrinsicrule ordering:e.g. by sayingthat
N-V merger takes place at D-structure and P-V merger at S-structure.Unless other
motivationcould be foundfor these assumptions,such a solutionwould be ad hoc.
12 In fact, Englishhas a similarcontrast
betweenbenefactivesand instrumentals:the latter
can appearin deverbalcompoundsbut the formercannot (hand-madescarves, laser-cut
diamonds,fork-splitEnglish muffins,but *child-made scarves (=scarves made for children), *church-givencontributions(=contributionsfor/to the church)). Presumablythe
benefactive compounds are impossible because no preposition can be generated in a
word-internalstructure;thus, the benefactive 0-role cannotbe assigned.

366

MARK

BAKER

differencesbetween the two types of applicativesdescribedin section 1.


In this section, I discuss the two types of applicativesfrom the point of
view of Case theory, addressingthe most importantquestionsabout their
derivationleft unansweredin section 2.2. This will yield a simple account
of the asymmetriesbetween the two constructionsinvolving the object
prefix.
Both S-structuresin (15) pose a problemfor Case theory, in that there
are two bare NPs in the verb phrasewhich must receive abstractCase in
order to pass the Case Filter. Chichewa,however, is like most languages
in that its verbs can only assign structuralCase (i.e. accusative,objective
or absolutive)to one NP.13 The obvious questionis what about the other
NP? The issue is very similarto that of how the second NP receives Case
in English double object constructionslike (21):
(21)

John gave Marya book.

A standardGB analysisof (21) is that the verb can assign anothertype of


Case, inherent Case, in addition to its one structuralCase (Chomsky
1980). I extend this idea to Chichewa by assumingthat all verbs in the
language may assign an inherent Case, as well as (perhaps)a structural
Case. Thus, verbs in applicativeconstructionshave enough Case-assigning potential to satisfythe Case Filter.
Next, we must investigate the question of which object in an applicative construction receives the structuralCase, which receives the
inherent Case, and why. Studies of inherent Case assignmentin English
nominals (Chomsky, 1986b) and in inversion constructionsin Romance
(Belletti, 1988) make it clear that inherent Case differs from structural
Case in systematic ways. Most importantly,inherent Case is assigned
under government at D-structure and the assigning head must 6-mark
the relevant NP. In contrast, structuralCase is assigned under government at S-structure,and there need not be any direct thematicrelationship between the assigning head and the NP. These differences account
for the well-known fact that Vs can EXCEPTIONALLY
CASE-MARK
the
subject of an infinitivalcomplementclause, but Ns cannot (from Chomsky, 1986b):
(22)a. I believe [John to be the winner].
b. *The belief (of) [John to be the winner].
(cf. The belief that John is a winner.)

13 The pattern of facts which would be expected if Chichewaverbs could (like


Kinyarwanda)assign two structuralCases is describedbrieflyin note 6.

THE SYNTAX

OF APPLICATIVES

IN CHICHEWA

367

Here, the subject of the infinitive is governed but not 0-markedby the
head in question. Example (22a) is thus grammaticalbecause the V
assignsstructural(accusative)Case, but (22b) is blocked: the N can only
assign inherent Case (here genitive, realized as the preposition of), and
the embeddedsubject does not meet the thematic condition on inherent
Case assignment. Significantly,even verbs cannot exceptionally Casemarka second NP, as shown in (23b):
(23)a. I told John [NPthe answerto question three].
b. *1 told John [s [NPMary]to have won the prize].
(cf. I told John that Maryhas won the prize.)
Here the verb's structuralCase is assigned (necessarily)to the adjacent
object John, leaving only inherent Case available. This inherent Case
can be assigned to an NP argumentof the verb (23a), but it cannot be
assigned to an NP which is not an argumentof the verb, as in (23b).
Example (23b) is thus ruled out parallel to (22b), confirmingthe hypothesis that ditransitiveverbs assign an inherentCase.
Now considerChichewabenefactive applicatives.These are associated
with the D-structurein (1la) and the S-structurein (15a). At S-structure,
the trace of the prepositiongoverns the applied object, but traces have
no lexical properties.In particular,they have no Case features and thus
cannot assign Case to the applied object."4 Furthermore,this applied
object is not 0-marked by the verb at D-structure; hence it cannot
receive inherent Case from the verb either. Therefore, benefactive
applicative constructions will only be possible if the benefactive NP
receives the verb's structuralCase. StructuralCase assignmentto this NP
is legitimate,since the only requirementis that the verb govern the NP at
S-structure;governmentholds because only one maximalprojection(PP)
contains the NP and not the V, and this category is 0-markedby V. 15 It

14 The traces of verbs which incorporateinto INFL do appear to assign Case to their
complements(cf. Chomsky1986a; Travis 1984). Baker (1985, 1988) shows that this is the
exception (possiblyrelatedto the nonlexicalnatureof INFL),ratherthan the generalrule.
Baker also shows that traces of moved XOscannot transmitCase from the Xo itself to an
NP governedby the trace.These facts can be relatedto a PF conditionon Case assignment
relationships.
15 This accountmustbe complicatedslightlyif Chomsky(1986a, p. 42) is correctin adding
a.MINIMALITY
CONDITION
tp a statementof governmentsuch as that given in (13). The

idea of this condition is that a given head prevents its complements from being governed by
any other head. If so, a verb does not usually govern the object of a preposition. Structure
(24) is potentially an instance of this configuration, with the preposition trace blocking the

368

MARK

BAKER

then follows that the basic object in this constructionmust receive the
inherentCase. This is permittedbecause the basic object is 0-markedby
the verb at D-structure. Thus, the S-structure for benefactive applicatives is (24), with Case relationshipsas indicated:

(24)
S

NP

VP

Mavuto

NP

PP

Pj

umb

-ir

t,

NP

mfumu mtsuko

Next, considerthe instrumentalapplicativeconstruction,whose D- and


S-structuresare (llb) and (15b). These differ from the structuresassociated with benefactive applicatives in one very relevant respect: the
instrumentalNP as well as the patient NP is governed and 0-markedby
the verb at D-structure.Hence, both meet the requirementsfor inherent
Case assignment.Both are also governed by the verb at S-structure,and
meet the requirementsfor structuralCase assignment.Therefore, there
are two ways of assigningCase in instrumentalapplicatives,as indicated
in (25):

verb from governingthe appliedobject even at S-structure.Baker(1988), however,shows


that trace heads differ minimallyfrom lexical heads in this regard; trace heads do not
trigger the MinimalityCondition(the GovernmentTransparancyCorollary),because they
form a chain with a more distanthead as a resultof XOmovement.Thus, the complexverb
in (24) does govern the applied object NP at S-structure,and structuralCase may be
assigned.

THE

SYNTAX

369

IN CHICHEWA

OF APPLICATIVES

(25)a.
S

NP

VP

Mavuto

NP

NP

umb

ir

mpeni

mtsuko

structural

nherent

OR
b.
S

VP

NP

Mavuto

NP

NP

umb

ir

mpeni

mtsuko

/------inherent

stuctural

Instrumentalconstructionsthus have a degree of freedomwith respect to


Case theory that benefactive constructionsdo not have.
Confirmationfor this analysis comes from the word orders found in
applicativeconstructions.English requiresthat an NP assignedstructural
Case by the verb be immediatelyadjacent to that verb (Stowell 1981);
the same is true for Chichewa (Mchombo1986):

370

MARK

BAKER

*Mdyerekezia- ku- namiz -a


tsopano abusa.
SP PRES deceive ASP now
devil
priests
The devil is deceiving the priestsnow.
Inherent Case, however, is not subject to as strict an adjacency
requirement;(21) demonstratesthat at least the structurallyCase-marked
NP may intervene between the inherentlyCase-markedNP and the V.
Now, in instrumentalapplicatives,the objects can appearin either linear
order, in more or less free variation:
(27)a. Anyani a- na-a
kwapul-ir
agalu ndodo.
baboonsSP PAST whip APPL ASP dogs sticks
The baboonswhippedthe dogs with sticks.
(26)

-a
Anyani a- na- kwapul-ir
ndodo agalu.
baboonsSP PAST whip APPL ASP sticks dogs
In benefactive applicatives, however, the order is fixed, the applied
object necessarilycoming before the basic object:
-a
atsikanamauta
(28)a. Anyani a- ku- pang -ir
baboonsSP PRES make APPL ASP girls
bows
The baboons are makingbows for the girls.
b.

b. *Anyani a- ku- pang -ir


-a
mauta atsikana.
baboons SP PRES make APPL ASP bows girls
The adjacencyrequirementimplies that in each sentence the NP closest
to the verb is the NP that receives structuralCase. Thus (27) shows that
both of the options displayedin (25) are in fact possible,while (28) shows
that the benefactive NP must have structuralCase, in accordance with
(24).
The puzzle about the distributionof object prefixesfrom section 1 can
now be explained. Mchombo (1986) and Bresnanand Mchombo (1987)
show that these object prefixes in Chichewa are in fact incorporated
object pronouns,ratherlike the object clitics in the Romance languages;
these prefixescan then optionallybe doubled by an NP outside the verb
phrase that matches them in noun class. The fact to be accountedfor is
that these object prefixes can encode either the applied object or the
basic object in instrumentalapplicatives,but can only encode the applied
object in benefactive applicatives,as in (29) and (30):
(29)a. Ndi- na- wa- phik -ir
-a
nsima
(ana).
lsS PAST OP cook APPL ASP cornmush children
I cooked cornmushfor them (the children).

THE SYNTAX

OF APPLICATIVES

IN CHICHEWA

371

-a
ana
(29)b. *Ndi- na- i- phik -ir
(usima).
1sS PAST OP cook APPL ASP childrencommush
I cooked it (cornmush)for the children.
zi- bay -ir
-a
(30)a. Asilikaria- nasoldiers SP PAST OP stab APPL ASP
mikondo (njovu).
spears elephants
The soldiersstabbed them (the elephants)with spears.
b.

Asilikaria- na- i- bay -ir


-a
soldiers SP PAST OP stab APPL ASP
njovu
(mikondo).
elephants spears
The soldiersstabbed the elephantswith them (spears).

Following Borer (1984, pp. 36-37) and related GB work, the object
prefixes, like other clitics, can be taken to be spell-outs of the Caseassigning features of the verbs that they are attached to. More
specifically,it is naturalto stipulate that they are manifestationsof the
structuralCase features of the verb, rather than of the inherent Case
features, just as lo 'him' and la 'her' are manifestationsof accusative
Case but not dative Case in Spanish.Now, given that the structuralCase
of the verb is used up by the object prefixin (29) and (30), the sentences
will only be grammaticalif the remainingNP can receive inherentCase
from the verb. We have just seen that in instrumentalapplicativesthe
verb can assigninherentCase to either NP; thereforeboth possibilitiesin
(30) are acceptable. In contrast, the applied object in benefactive constructions cannot get inherent Case because it is 0-marked by a preposition and not the verb; hence (29b) is ungrammatical.This difference
in object prefixpossibilitiesis thus a second visible type of evidence that
Case assignmentis freer in instrumentalapplicativesthan in benefactive
applicatives, confirming the Theta-theoretic distinction between benefactives and instrumentalsproposedin section 2.
Before leaving this section, it is desirable to clarify the assumptions
about the role of prepositionalelements in instrumentalconstructions.
The basic hypothesis is that they are not 0-role assigners. The other
major reason for having a P appear in the structure in GovernmentBinding Theory is to satisfy Case Theory requirements.Specifically,Ps
can have the following functions:

372
(31)

MARK

BAKER

Ps may i. assign a 6-role


ii. assign Case to an NP
iii. REALIZE (i.e. spell out) an inherentCase
assignment.

Significantly,a P may in a particularstructure have function (iii) (or


perhapsfunction (ii)) without having function (i). In these instances, the
P is not present at D-structure,which representsonly thematicrelationships; rather,it is insertedin the syntax. The best-knownexample is the
of found in English derived nominals (e.g. my removal *(of) the cyst

compared with I removedthe cyst), which on the analysisof Chomsky


(1986b) is merely a realizationof genitive Case.
I suggest that the prepositionalelements found in instrumentalclauses
in Chichewa are similar to these uses of of. Thus, no P is present at
D-structurein such constructions.If there are two NPs in the VP, then
the V must assign one of them inherentCase. When this occurs, a P must
be inserted at S-structureas a realizationof the inherent Case assignment, just as in English nominals.16 However, Chichewa differs from
English in that it has Ps which are affixesmorphologically,such as -ir, as
well as Ps which are roots. We may assume that either type of P can in
principle be inserted to realize the inherent Case relationship. This
option leads to the two surface realizationsof clauses with instruments
found in Chichewa, both of which are derived from the D-structurein
(32a):17

16 It is not clear why no P is needed to realize inherent Case in English dative shift
sentences like (21). Unlike the Chichewa applicatives,the English constructionis only
possiblewith a semanticallyrestrictedclass of predicates;these restrictionsmay make the P
dispensibleat PF. Indeed, even in Chichewa the applied affix does not appear with a
handfulof canonicaldative shift verbs, such as -pats- 'give'.
Probablyrelated is the problem, raised by an NLLT reviewer, of why English verbs
cannot assign inherentCase to the patient NP in instrumentalconstructions,making the
prepositionwithoptional:

(i)

*I carved the puttyknifea figurine.


(compare:I carved Bill a figurine.)

Here one might say that (for some reason) in English there are no canonicalinstrument
verbs whose lexical semanticsmake the realizationof inherentCase by a P unnecessary,
even though there are canonicaldative verbs. This gap is probablynot accidental,since
Chichewaalso lacks verbs which (parallelto -pats-) allow the appliedaffixto be omitted
when an instrumentNP appears. As to why with cannot realize Case assignmentto a
figurineinsteadof to theputtyknifein (i), see note 17.
17 There is a furtherdifferencebetweenndi and -ir beyondthe fact that one is an affix:ndi
but not -ir is limitedto realizinginherentCase on the instrumentNP. Thus, (i) is ungrammatical:

THE SYNTAX

(32)a.

OF APPLICATIVES

IN CHICHEWA

373

Mavuto [vp umb mpeni mtsuko]

Mavuto

mold knife waterpot

(orderof NPs irrelevant)


b.

Mavuto a-na-umb-amtsuko [ndi mpeni]


with

c.

Mavuto [a-na-umb-ir-a]mtsukompeni
APPL
Mavuto molded the waterpotwith a knife.

Insertingthe P affixflags the Case assignerat S-structure;insertingthe


independent P flags the Case assignee. Thus, we need not say that in
applicativeconstructionsan instrumentalP exists withoutany function at
D-structure and subsequently incorporates without leaving a trace.
Rather, the applied affix is added directly to the verb in the syntax to
satisfy requirements of Case theory. Benefactive constructions are
similar, except that -ir assigns the benefactive 0-role (and hence is
presentat D-structure)in additionto realizingthe assignmentof inherent
Case. In other words, -ir has both property (i) and property (iii) in
benefactive constructions,just as for in English benefactives has both
property (i) and property (ii). On this account, the applied affix in
Chichewa has essentially the same range of Case and Theta theory
functionsas prepositionsin English have. 18
(i)

*Mavutoa- na- umb -a


mpenindi mtsuko.
MavutoSP PAST mold ASP knife with waterpot
Mavutomolded the waterpotwith a knife.

In this regard,ndi is like English with. This differenceI attributeto the well-knownfact
that prepositionsvary in how broador narrowa rangeof thematicroles they are consistent
with, some languagesmakingmore lexical distinctionsthan others. This is independentof
whetherthe P appearswith the N or on the V; in Eskimo,instrumentalCase can appearon
patientsas well as on instruments,unlikendi.
The Fula data analyzedin Marantz(1982, 1984) may be taken as the conversesituation.
Fula is unlike Chichewa in that it has a distinct instrumentalapplied morpheme,which
appearsonly in instrumentalconstructions.Supposethat this morphemeis like Chichewa
ndi rather than Chichewa -ir; it can realize the assignment of inherent Case to an
instrumentbut not to a patient. The effect will be that with respect to grammatical
processesinvolvingstructuralCase, such as passive,the patientand not the instrumentwill
act like a canonicalobject - which is true.This showshow the currenttheorycan subsume
the data that motivatedMarantz'stheory.
18 Baker (1985, 1988) suggests an alternativeto the analysisin this section, in which there
is no inherentCase and NPs pass the Case Filterwhen theirheadsincorporateinto the verb
at LF. The theoryof incorporationimpliesthat thereare two waysin which this can happen

374

MARK BAKER

4. EXTRACTION POSSIBILITIES IN APPLICATIVES


Next, I turn to the second asymmetry between the two applicative types,
which involves how Wh-movement affects applied objects. An additional

example of the relevant contrastis:


(33)

Ndi- nanen -a
kuti Mavuto a- nathyol -er
lsS PAST say ASP that MavutoSP PAST break APPL
-a
mfumu/ndodo mpando.
ASP chief stick chair
I said that Mavuto broke the chair for the chief/withthe stick.

(34)

*lyi ndiyo mfumuimene ndi- na- nen -a


kuti Mavuto
this is
chief whom lsS PAST say ASP that Mavuto
a- nathyol -er
-a
mpando.
SP PAST break APPL ASP chair
This is the chief whom I said that Mavuto broke the chair for.

(35)

lyi ndi ndodo imene ndi- nanen -a


kuti
this is stick which lsS PAST say ASP that
Mavuto a- na-a
thyol -er
mpando.
Mavuto SP PAST break APPL ASP chair
This is the stick which I said Mavuto broke the chair with.

Example (33) is a typical applicative,in which the applied object could


be either a benefactive (mfumu,'chief')or an instrument(ndodo, 'stick').
When a moved relative pronoun inene 'which/whom' bears an instrumentalrole (35), the result is acceptable;when it bears a benefactive
role (34), it is not. Significantly,whatever constraint is involved has
nothing to do with there being two unmarkedNPs after the V in (33),
because the same effect shows up in the restricted instances where
applicativescan be formed from intransitiveverbs (cf. note 23):
in instrumentalconstructions,but only one in benefactive constructions(cf. (16)-(19)
above), again explainingwhy there is more freedomin Case assignmentin the formerthan
in the latter.
An NLLT reviewer points out that the text analysisas it stands predicts that -ir can
appear on simple transitive verbs when the verb assigns its inherent rather than its
structuralCase to its complement.If this predictionis to be avoided,we muststipulatethat
verbs which can assign both types of Case in fact only assign inherentCase in a given
sentence if they also assignstructuralCase. In other words,structuralCase assignmenthas
priority.

THE SYNTAX

OF APPLICATIVES

IN CHICHEWA

375

vin -ir
-a
mfumu/mikondo.
Atsikanaa- naSP PAST dance APPL ASP chief spears
girls
The girls danced for the chief/withspears.

(36)

a- na*lyi ndi mfumuimene ndi- ku- ganiz -a


this is chief whom 1sS PRES think ASP SP PAST

(37)

-a
vin -ir
dance APPL ASP
This is the chief whom I think they danced for.
a- nalyi ndi mikondo imene ndi- ku- ganiz -a
this is spears which lsS PRES think ASP SP PAST

(38)

-a
vin -ir
dance APPL ASP
These are the spears that I think they danced with.
Our analysisof Chichewaapplicativesimpliesthat there is a systematic
structuraldifferencebetween (34) and (37) on the one hand, and (35) and
(38) on the other: the trace of the relative pronounis governed by a null
prepositionin the former,but not in the latter:
(39)a.

[NP

chief [s, whichj [s...

[s Mavuto break-APPLi [pp ei tj]

chair]]]]
b.

[NP

stick [s, whichj [s ... [s Mavuto break-APPL tj chair]]]]

Thus, if the trace of Wh-movementis, for some reason, ungrammatical


when it is the complementof an empty head, the observed contrastwill
follow immediately.
In fact, a proposal to this effect already exists in the literature, in
Czepluch (1982). The ungrammaticalityof (34) recalls the well-known
fact that dative-shifted goals and benefactives cannot readily be Whmoved in English (from Stowell 1981):
(40)

*[Whosemother]i did Greg bake [ep ti] a birthdaycake?

Czepluch (following Kayne 1984) claims that there is a phonologically


empty preposition(represented'ep' in (40)) associatedwith the benefactive object in the English construction,parallelto the one I have argued
for in Chichewa;indeed, if section 2 is correct, this conclusionis forced
by Theta theory. Czepluch then blocks extractionof the benefactiveby a
general constraint against configurationswith embedded empty categories - a *[e[t]] Filter, in effect. Baker (1988) develops this somewhat,

376

MARK

BAKER

showing that a more precise statementof the conditionis somethinglike


(41):
(41)

NonobliqueTrace Filter:19
*[?Pi ... V+Xj
[-V] (N or P).

...

[xp tj tij...]

at S-structure, where X is

Here 'Op' stands for an operator phrase in COMP. From this perspective, the contributionof the Chichewadata is primarilyone of contrast:it
shows that precisely when Theta theory and the ProjectionPrincipledo
not imply the existence of an empty P - namely in instrumentalconstructions- the difficultyin extractingan appliedobject disappears.This
is exactly what an analysisalong the lines of Czepluch (1982) predicts.20
This type of analysis also explains the fact that the basic object is
extractablein both type of applicatives,since in neitheris it governed by
a null P (cf. (39)):
(42)

kuti
nen -a
Uwu ndi mpandoumene ndi- naThis is chair which 1sS PAST say ASP that
-a
Mavuto a- namfumu/ndodo.
thyol -er
Mavuto SP PAST break APPL ASP chief stick
This is the chair which I said that Mavuto broke
{forthe chief.l
with a stick. S

19 Baker(1988) pointsout that the variousstipulationsin this filtersuggest thatit shouldbe


derivedfromCase Theorysomehow.However,the fact that violationsof (41) can be rather
mild (see below)suggeststhatit shouldnot be directlylinkedto the impossibilityof preposition strandingin Chichewa; all cases of strandingovert prepositionsare very strongly
ungrammatical.
Whetheror not (41) is universalremainsunresolved.Certainly,apparentcounterexamples appearin the literature;Kisseberthand Abasheikh(1977) is one example.On the other
hand, the effect is ratherdelicate, so that the attested examplesmay not be sufficientto
decide the point. Thus, in both English and Chichewa,violations of (41) are much less
severe if the Wh-phraseappearsin the local COMP- i.e. in the COMPof the clause that
immediatelycontainsthe empty head. Furthermore,the object prefixin Chichewacan act
as a resumptivepronounfor the relative pronoun,and sentences like (34) and (37) become completely grammaticalwhen it appears.This is not surprising,since the conventional islandeffects also disappearin such constructions(see Barker1988). Thus, the status
of (41) must be left to furtherresearch.
20 Accounts of the deviance of (40) in English which have no null prepositiongenerally
assumethat it is ungrammaticalto extract the innermostNP of a V-NP-NP configuration
for some reason (Jackendoffand Cullicover 1971; Oehrle 1975; Stowell 1981; Kayne
1984, ch. 5). Such approachesdo not extend to Chichewa;both the acceptabilityof (35)
and the ungrammaticality
of (37) are problematicfor them.

THE SYNTAX

OF APPLICATIVES

IN CHICHEWA

377

Important theoretical questions remain concerning the nature of the


prohibition expressed in (41). Nevertheless, the thematic properties
motivated in section 2 induce the exact structuraldistinction among
verbal objects that an account of Wh-movementdepends on.
It is interesting at this point to compare the solutions to the object
prefix puzzle and the extractionpuzzle. Section 1 observed that the two
could be unified with the intuitive observation that instrumentalconstructions are more like true double object structuresthan benefactive
constructionsare. The difficultywith this observationwas that it is the
benefactive appliedobject that behaves like the true object with respect
to the object prefix,but the basic object that behaves like the true object
with respect to extraction - a seeming contradiction.We now see how
the paradoxis resolved. The Wh-movementpuzzleproves to be the more
transparent:benefactive appliedobjects are distinguishedfrom canonical
direct objects in that they are governed by a null preposition; basic
objects and instrumentalappliedobjects are like canonicalobjects in this
respect. Thus, the benefactive applied object alone cannot undergo
Wh-movement,by the NonobliqueTrace Filter.The object prefixpuzzle
is similar, but has a twist. The benefactive applied object must get
structuralCase (therebyappearingas the object prefix),but not because
it is like a canonical object. On the contrary: it alone is structurally
unlike an object, and thus cannot receive inherentCase. Hence, it must
receive the verb's accusativeCase, a necessitywhich does not hold of the
other types of object. This resolutionto the paradoxis abstractin that it
cruciallyrelies on asymmetriesin the applicationof an invisible process
of Case assignmentapplyingat D-structure.However, this abstractnessis
not problematicfor language acquisition,because the analysisis derived
from a universal generalizationabout 0-role assignment,together with
familiarprinciples of Universal Grammarsuch as the Theta Criterion,
the ProjectionPrinciple,and the Case Filter.

5.

VERB

TRANSITIVITY

AND

APPLICATIVES

Significantly,there are additionaldifferences between benefactive and


instrumentalapplicatives in Chichewa which can also be explained in
terms of the basic Theta Theoretic difference. It has been observed in
the literature that in many languages one cannot form a benefactive
applicativebased on an intransitiveverb (e.g. Chung 1976; Aissen 1983;
Marantz 1984). This is generally true in Chichewa as well. The restriction holds for verbs which are purely intransitive and never take an

378

MARK

object, as in (43) and

BAKER

(44):21

(43)

-a
*Mkangou- ku- yend -eranyani.
SP PRES walk APPL ASP baboons
lion
The lion is walkingfor the baboons.

(44)

sek -er
-a
*Kalulua- naatsikana.
hare SP PAST laugh APPL ASP girls
The hare laughed for the girls.

It also holds for verbs which are usuallytransitive,but which may appear
without an object when the understoodpatient of the action is indefinite,
generic, or prototypical.Examplesof such verbs are:
(45)a. Mlimi a- ku- sem -a
(mtondo).
farmer SP PRES carve ASP mortar
The farmeris carving (a mortar).
b.

Amayi a- ku- umb -a


(mtsuko).
woman SP PRES mold ASP waterpot
The woman is molding (a waterpot).

Benefactive applicatives are systematicallypossible based on the transitive uses of these verbs but not on the intransitiveuses:
(46)

Mlimi a- ku- sem -er


-a
mbidzi*(mtondo).
farmerSP PRES carve APPL ASP zebras mortar
The farmeris carving for the zebras. (OK with 'a mortar')

21 The Chichewa affix -ir can also be a derivational transitivizingmorpheme, which


attachesto some verb roots in the lexiconwith semanticallyirregularresults.Example(43),
for instance, is grammaticalwith the idiomaticmeaning 'The lion is inspecting the baboons'. One can confirmthat there is no P incorporationin such examplesbecause the object NP anyani 'baboons'in the grammaticalinterpretationof (43) can be Wh-moved,in
contrastwith exampleslike (37). These idiomaticforms are thus syntacticallyas well as
semanticallydistinguishablefrom productiveapplicativeformation.
Example(43) can also be interpretedvery narrowlyas 'The lion is walkingin place of the
baboons'- which looks like a more standardbenefactivereading.It cannot, however,have
the usual range of benefactiveinterpretations,such as 'The lion is walkingto bring good
resultsto the baboons,to please the baboons,becausethe baboonstold him to', etc. In this
respect,(43) contrastswith the otherexamplescited in this article,including(36). I have no
account of this 'in place of' readingof (43); it may be relatedin some way to Chichewa's
REASON APPLICATIVES,
mentioned in Baker (1988).

THE SYNTAX

(47)

OF APPLICATIVES

IN CHICHEWA

379

Amayi a- ku- umb -ir


-a
mwana*(mtsuko).
woman SP PRES mold APPL ASP child
waterpot
The woman is molding for the child. (OK with 'a waterpot')

In contrast, the English translationsof (43), (44), (46), (47), while in


some cases semanticallyunusual,are acceptable given whatever circumstances are requiredto imagine the action named by the verb being done
to benefit someone. Example (48), for instance, is fine, given popular
methodsof fund-raising: 22
(48)

Tomorrow,I will walk for Jerry'skids.

Similarly,the gloss of (44) is appropriateif we suppose that the hare is an


entertainer,laughing on purpose for a dramaticeffect. Thus, the deviance of the Chichewa examples must be a result of the syntax of
applicativesratherthan the semanticsor pragmaticsof benefactives.
Interestingly,instrumentalapplicativesare different from benefactive
applicativesin these respects. They can be formed from monadic verbs
ratherfreely:
(49)

-a
ndodo.
Msangalatsia- ku- yend -er
entertainer SP PRES walk APPL ASP stick
The entertaineris walkingwith a stick.

(50)

Mbalamezi- ma- uluk -ir


-a
mapiko.
birds
SP HAB fly APPL ASP wings
Birds fly with wings.

Instrumentalapplicatives are also possible with both transitive and intransitiveinstancesof an optionallytransitiveverb root:
(51)

Mbuzizi- ku- dy -er


-a
mipeni (chinangwa).
goats SP PRES eat APPL ASP knives cassava root
The goats are eating (cassavaroot) with knives.

(52)

Mlimi a- ku- sem -er


-a
sompho (mtondo).
farmer SP PRES carve APPL ASP adze mortar
The farmeris carving (a mortar)with an adze.

This difference needs to be explained.


22

A colleague reportshaving received a publicityflyer with exactly this expressionon it


(althoughthe specificcharitywas different).

380

MARK

BAKER

In fact, the contrast follows from a simple extension of the ideas we


have developed. Since benefactives get their 0-roles from a P but
instrumentsdo not, the S-structureof (for example) (43) must be (53),
while that of (49) is (54):
(53)

Mkango [vp yend + eri [pp ti [NP anyani]]].

(54)

Msangalatsi[vp yend+ er [NPndodo]].

These are differentfrom applicativesof transitiveverbs in two ways: (i)


there is no basic object NP, and (ii) the V is plausibly not a structural
Case assigner. This second statement is confirmedfor the pure intransitive verbs in Chichewa by the fact that they almost never appearwith
cognate objects of the type sometimesfound in English (e.g. I walkeda
long walk, I walkedthe wholetrailby sunset.)23 The optionallytransitive
verbs certainlycan be Case assignerswhen they have an object, but we
may assumethat the propertyof assigningstructuralCase is linked to the
propertyof assigningan internal(patient) 0-role in the lexical entries of
these verbs, such that they never have the former propertyunless they
have the latter as well. Then, the question arises of how the VP-internal
NP in (53) and (54) can satisfythe Case Filter.In (54), there is a way: the
instrumentalNP is eligible to receive inherentCase from the verb, since
it is 0-marked by the verb at D-structure. This inherent Case is then
realizedby insertingthe applied affix,as with transitiveverbs. The same
mechanismscannot rescue the benefactive NP in (53), however; it is not
0-markedby the verb at D-structure,and thus cannot receive inherent
Case. No other Case is available,and (53) is ruled out by the Case Filter.
This explainsthe ungrammaticalityof (43), (44) and (46), (47) in contrast
with (49)-(52). The same principles that force a benefactive applied
23 In fact, a few intransitiveverbs do allow cognate
objects in Chichewa,and hence must
have accusativeCase-assigningfeatures.An exampleis -vin- 'dance':

(i)

Atsikanaa- na- vin -a


(chiwoda).
girls
SP PAST dance ASP chiwoda
The girls danced (the chiwoda- a tribaldance).

In fact, this verb also allowsbenefactiveas well as instrumentalapplicativesto be formed,


even withoutthe cognate object appearing;see (36). Thus, there is a correlationbetween
Case-assigningpropertiesand the ability to form benefactive applicatives,as the theory
predicts.
Note that cognate object verbslike -vin- 'dance'can be distinguishedon lexicalsemantic
groundsfrom the optionallytransitiveverbs like -sem- 'carve'that are discussedin the text.
FollowingRizzi (1986), one mightsay that the verbsin the lattergroupactuallyhave a null
object NP that receives structuralCase. This could explain the link between having no
overt object and havingno structuralCase to assignfor this class of verbs.

THE SYNTAX

OF APPLICATIVES

IN CHICHEWA

381

object to get structuralCase from a transitiveverb make it impossible


for such an object to appearwith an intransitiveverb.
This analysis makes a striking prediction about instrumental applicatives of optionally transitive verbs. When an instrumentalNP appears with an intransitive token of the verb, it necessarily receives
inherent Case, whereas with a transitivetolkenof that verb, it may get
structuralCase instead. Since the Chichewa object prefix is a reflex of
structural Case features (cf. the discussion of (29)-(30)), it will be
impossiblefor an object prefix to agree with an instrumentin indefinite
object deletion constructions,even though such agreementis possible in
other circumstances.This is exactly correct:
(55)

Mlimi a- ku- i- sem -er


-a
*(mtondo)sompho.
farmer SP PRES OP carve APPL ASP mortar adze
The farmer is carving with the adze. (OK if

....

carving a

mortar. . .')
(56)

Chatsaliraa- nali- khap -ir


-a
ChatsaliraSP PAST OP slash APPL ASP
*(njoka)khasu.
snake hoe
Chatsalira slashed with a hoe. (OK if

'..

.slashed

a snake

with.. .')
Examples (55) and (56) are surprisinginasmuchas it is in the structure
with less apparent competition for object status that the instrument
cannot have the object propertyof showing up as an object prefix.
More generally,this section shows that the Theta-theoreticdistinction
between benefactivesand instrumentsexplainsdifferencesin their syntax
which go beyond what would be expected given only the informal
generalizationthat clauses with instrumentsare more double-object-like
than clauses with benefactives.
6.

IMPLICATIONS

AND

CONCLUSIONS

In closing, I would like to draw from this analysis of Chichewa applicatives two specific implications about the nature of linguistic
representation:one concerning the role of grammaticalrelations, and
one concerning the role of thematic relationsin determiningunderlying
syntactic structure.

382

MARK

BAKER

6.1. GrammaticalRelations
Some researchers have claimed that the grammar must make direct
reference to grammaticalrelations such as subject, object, and indirect
object. For example, they take alternations in the expression of
arguments,such as those found in applicatives,to be explicit functions
defined over these grammaticalrelations(e.g. Perlmutter1983; Perlmutter and Rosen 1984; Bresnan 1982). In these terms, an applicative
process makes a benefactive oblique into a direct object (BEN-TO2-ADVANCEMENT in RG terminology;see Chung 1976; Aissen 1983).
The analysisof Chichewa calls this into question.
First, the Chichewa applicativessuggest that there is no single notion
of direct object that plays a central role in the grammar;ratherNPs are
simply grouped into naturalclasses in different ways depending on the
specific concerns of the different modules of the grammar,such as X'
theory, Government,and Case theory. Benefactive applied objects, for
example, are objects in the Case and Governmenttheory senses, but not
in the structuralsense. Instrumentalapplied objects, on the other hand,
are objects in the structuralsense and in the Governmenttheory sense,
but may or may not be in the sense defined by Case theory. The simple
term 'direct object' is thus not refined enough to be useful in this
situation,even if one distinguishesbetween deep and surfacesenses. This
suggests that 'object' is not a basic notion of linguistic theory,but rather
a convenient term which can be used in a variety of related senses
defined by the more fundamentalprinciplesof the grammar.
In addition, we have shown that there is no need to analyze the
Chichewa applicativesas involving functions defined over grammatical
relations.Instead, their propertiescan for the most part be derived from
an interplayof more general principles,including those of Case assignment and the movement of X? level categories. The current article is
thus a case studyof how the hypothesisthat such functionsdo not exist in
natural language (articulatedin Marantz (1984), Williams (1984), and
Baker (1985, 1988)) can be worked out over a range of complex data.
The two Chichewa applicatives are a particularlyinteresting test case,
because one must account for their differences without the luxury of
attributing them to two different rules of applicative formation. This
problem is solved by appealing to a universaldifference between benefactives and instrumentalsthat affects how principlesoperate.
This solution does not imply that exactly the same differencesbetween
benefactive and instrumentalapplicativeswill appearin every language;
for example, if languages differ in their Case-assigningproperties,these

THE

SYNTAX

OF APPLICATIVES

IN CHICHEWA

383

differenceswill affect the range of applicativeconstructionsfound in the


language. Nevertheless, I do predict that some differences between
benefactive and instrumentalapplicativeswill appearin every language
that has both, and that the differences will be in a consistent direction:
the instrumentalapplicativeswill be the more double-object-likeof the
two. Furtherassessmentof this predictionis left to future research.24
6.2. ThematicRole Assignmentand D-structure
As emphasized throughout, the key to the analysis is a difference
between how benefactive and instrumentalthematicroles are assignedin
Chichewa: the former are assigned by a P; the latter directly by the V.
Now, there is good reason to suppose that this difference must be an
unlearnedone, and hence part of Universal Grammar.Internalto Chichewa itself, the difference is heavily disguised on the surface: instruments often appearwith an overt P, even though that P is not necessary
for Theta theory (e.g. (32b)), while benefactives and goals generally
appearwithout an overt P even though one is needed for Theta theory.
Therefore, there is no way to deduce the distinction from the surface
distributionof PPs. Moreover, the same distinctionbetween benefactives
and instrumentalsshows up in a very different way in polysynthetic
languages, where benefactives never incorporate,while instrumentsdo.
Thus, the case for the 0-marking difference being part of Universal
Grammaris a strong one.
This result powerfullysuggests that underlyingsyntacticstructuresare
a projectionof (aspectsof) humansemanticand/orconceptualstructures,
as proposed in Marantz (1984). The central idea that, given constant
semantics, underlying syntactic structures are universal in important
respects has been entertained many times and rejected many times;
Rosen (1984) is a good recent example. Here, we have found reason to
accept such an idea, perhaps in the form of the UNIVERSAL THETA
ASSIGNMENT
HYPOTHESIS
of Baker (1988), which states that identical
thematicrole assignmentsmust have parallelstructuralrepresentationsat
D-structure.It is striking that several theories of lexical semantics (e.g.
Preliminaryevidence in supportof this predictioncomes from Kinyarwandaand Sierra
Popoluca.These languageshave differenttransitivitypatternsfrom Chichewa,but in both,
instrumentalapplied objects, like ordinaryobjects, allow their possessors to undergo
PossessorRaising.This is impossiblewith the possessorsof benefactiveand dative applied
objects (Kimenyi 1980, p. 110; Marlett 1986). If Possessor Raising is an effect of
incorporatingthe possessedN into the V at LF, as in Baker(1988), then these asymmetries
follow from the theoryin section 2.3 with no new complications.
24

384

MARK BAKER

Jackendoff 1987; Foley and Van Valin 1984) state that instrumentsare
like themes/patientsin that both are manipulatedby the actor in the
course of the action. Benefactives, on the other hand, are like locatives
in that they express the goal or destination(in an abstractsense) toward
which the action progresses.If somethinglike these statementsis true of
human conceptual structures, and if underlying syntactic structure is
projected from these conceptual structures,then it is natural that both
patient/themesand instrumentsshould be directly 0-markedat D-structure, while both benefactive/goalsand (many)locatives are 0-markedvia
a prepositionin additionto the verb.
A variety of work, particularlyin GB, has suggested that in basic
transitive clauses agent 6-roles are consistently assigned in one
configuration(outside the verb phrase),while patient roles are assigned
in another (inside the verb phrase), this asymmetrybeing the projection
of an asymmetrybetween the two in compositionalsemantics (Marantz
1984; Chomsky 1981). The difference between benefactives and instrumentalsargued for here is a second example of this same type. The
unclear cases are no doubt more numerousthan the clear ones; nevertheless, if universalstatementscan be made about how agents, patients,
instruments,benefactives,and goals are representedin syntax,we may be
optimisticthat the rest of the patternwill fall into place as well.25

APPENDIX:

APPLICATIVES

AND DETRANSITIVIZATION

Sections 3 and 5 of this article show that the Case theory propertiesof
verbs affect the expression of benefactive phrases and instrumental
phrases differently. It is natural to ask as well how detransitivizing

morphology,which absorbsthe Case-assigningpropertiesof the verb it is


attached to, affects the syntax of applicative constructions. Indeed,
becoming the subject of a passive is the third important signal of
direct-objecthood in most Bantu studies, along with appearing immediatelyafter the verb and controllingthe object prefix (cf. Kisseberth
and Abasheikh 1977).
Chichewa has at least two distinct detransitivizationprocesses which
are relevant: a passive voice of the familiar kind, and a reciprocal
25 Recent workon psychologicalverbs, such as
Belletti and Rizzi (1986), suggeststhat this
can be done for experiencerthematic roles as well. Earlier work by Relational Grammarianson inversionis based on the same intuition,althoughthey do not representthe
consistenciesin structuralterms.

THE SYNTAX

OF APPLICATIVES

IN CHICHEWA

385

voice.26 Exampleswith a simple transitiveverb are:

nkhumba.
meny -a
(57)a. Nkhuku zi- nachickens SP PAST beat ASP pigs
The chickens beat the pigs.
(ndi
b. Nkhumbazi- na- meny -edw -a
SP PAST beat PASS ASP with
pigs
PASSIVE

nkhuku).
chickens
The pigs were beaten (by the chickens).
c.

-a.
Nkhumbazi- na- meny -an
SP PAST beat RECIP ASP
pigs

RECIPROCAL

The pigs beat each other.


The familiar GB account of sentences like (57b) is that the passive
morpheme-idw absorbsthe verb's abilityto assign a thematicrole to the
subject position and its ability to assign structuralCase to its complement. The first of these effects allows this complementto move into the
subject position; the second forces it to do so in order to pass the Case
Filter. Marantz (1984) argues that essentially the same analysis is appropriate for reflexive and reciprocal voices as well, given the crosslinguisticsimilaritiesbetween the constructions.Thus, the S-structurefor
both (57b) and (57c) can be schematizedas:
(58)

[s NPi Infl [vP verb- Jrecipi ti]l


Ipass, J

The only difference between the two constructionslies in the nature of


the morphemeattached to the verb: passive morphologyguaranteesthat
the absorbedexternal 0-role is interpretedas referentiallydistinct from
the derived subject (cf. Baker, Johnson, and Roberts to appear),while
reciprocalmorphologyis anaphoric,in the sense that the external 0-role
it absorbsis referentiallylinked to the derived subject.
The text analysispredicts that the Case-absorbingpropertiesof these
morphemeswill affect the objects of the two applicative constructions
differently.Since the applied object in a benefactive applicativecannot
26 There is also a reflexive voice in Chichewa, indicated by the prefix dzi-. This constructionbehaves like the reciprocalsillustratedin the text, and should receive a similar
analysis.

386

MARK

BAKER

get inherent Case from the verb and the verb's structural Case is
eliminated by the detransitivizingmorpheme, the benefactive NP must
move to the subject position to get nominative Case. The basic object
remains in the VP, with inherent Case. In instrumentalconstructions,
however, either the instrumentor the basic object may receive inherent
Case, and either can move to the subjectposition and receive nominative
Case. Therefore, there should be two possible passives and two reciprocals with instrumentalapplicatives,but only one of each with benefactives.
The data only partially supports the theory in this domain. The
benefactive applicativeswork as expected. The examples in (59) show
that the benefactive alone can become the subject in a passive:
-a
(59)a. Ndi- na- phik -ir
ana
nsima.
1sS PAST cook APPL ASP childrencornmush
I cooked cornmushfor the children.
b.

Ana
a- na- phik -ir
-idw -a
nsima.
childrenSP PAST cook APPL PASS ASP commush
The childrenwere cooked cornmush.

c. *Nsima i- na- phik -ir


-idw -a ana.
cornmushSP PAST cook APPL PASS ASP children
Cornmushwas cooked for the children.
In the reciprocal voice the benefactive but not the patient can be
interpretedas coreferentialwith the agent of the action, showing that
only the benefactive can move to the subject position to bind the
anaphoricmorphemeon the verb:
(60)

Ana
a- na- meny -er
-an
-a
zigawenga.
childrenSP PRES hit
APPL RECIP ASP ruffians
a. OK The childreniare beating the ruffiansfor each otheri.
b. *The childrenare beating each other for the ruffians.

This much follows from the theory.


Difficultiesarise with instrumentalapplicatives,however. In passives,
instrumentscan move to the subject position, but it is unacceptablefor
the patient to do so:
(61)a. Fisi a- na- dul -ir
-a
mpeni chingwe.
hyena SP PAST cut APPL ASP knife rope
The hyena cut the rope with a knife.

THE SYNTAX

b.

OF APPLICATIVES

IN CHICHEWA

387

Mpeni u- na- dul -ir


-idw -a
chingwe
knife SP PAST cut APPL PASS ASP rope
(ndi fisi).
by hyena
The knife was used (by the hyena) to cut the rope.

c. *Chingwechi- na- dul -ir


-idw -a
mpeni
rope
SP PAST cut APPL PASS ASP knife
(ndi fisi).
by hyena
The rope was cut with a knife (by the hyena).
The ungrammaticalityof (61c) is not expected. The opposite pattern is
found with the reciprocalvoice:
-a
-an
(62)a. *Asilikalia- ku- bay -ir
ngombe.
soldiers SP PRES stab APPL RECIP ASP cows
The soldiersare stabbingthe cows with each other.
b.

Asilikali a- ku- bay -ir


-an
-a
mipeni.
soldiers SP PRES stab APPL RECIP ASP knives

The soldiersare stabbingeach other with knives.


This time the patient NP can bind the anaphoricmorpheme from the
subject position (62b), but, contrary to expectation, the instrumental
appliedobject cannot (62a). Note that the patternin (62) is the reverseof
the one found with benefactive applicatives.
This range of data is compatible with the theory developed in this
work, but is not fully explainedby it. All the structureswhich violate the
system of constraintswe have discussedare in fact ungrammatical((59c)
and (60b)). Two other sentences- (61c) and (62a) - are also impossible.I
must assume that these are deviant because of the interventionof other
factors, either pragmatic/functionalor syntactic. For (62a), this is very
plausible;Trithart's(1976, p. 16) observationthat instrumentalphrases
cannot refer to humansis sufficientto rule out (62a) and similarexamples. The ungrammaticalityof (61c) is more mysterious,and must be left
to furtherresearch.27In the meantime, however, we can conclude that
27

Note that (i) is grammatical and synonymous with (61c):


Chingwe chi- nadul -idw -a
ndi mpeni (ndi fisi).
Rope
SP PAST cut PASS ASP with knife with hyena

(i)

The rope was cut with a knife (by the hyena).


This may be an important hint to the nature of (61c)'s deviance.

388

MARK

BAKER

the behavior of detransitivizingprocesses in Chichewa is generallyconsistent with the basic structuraland thematic distinctionbetween benefactive and instrumentalNPs which is established in the body of this
article.
REFERENCES
Aissen, Judith: 1983, 'IndirectObject Advancementin Tzotzil', in D. Perlmutter(ed.),
Studiesin RelationalGrammar1, Universityof Chicago Press, Chicago, pp. 325-366.
Allen, Barbara, Donna Gardiner, and Donald Frantz: 1984, 'Noun Incorporationin
SouthernTiwa', IJAL 50, 293-311.
Andrews,J. Richard:1975, Introductionto ClassicalNahuatl, Universityof Texas Press,
Austin.
A Theoryof GrammaticalFunctionChanging,unpubBaker, Mark: 1985, Incorporation:
lished Ph.D. dissertation,MIT.
A Theoryof GrammaticalFunctionChanging,Universityof
:1988, Incorporation:
Chicago Press.Chicago.
Baker, Mark, Kyle Johnson, and Ian Roberts: to appear, 'Passive ArgumentsRaised',
LinguisticInquiry.
Belleti, Adriana:1988, 'Unaccusativesas Case Assigners',LinguisticInquiry19, 1-34.
Belletti, Adriana and Luigi Rizzi: 1986, 'Psych-Verbsand TH-Theory', LexiconProject
WorkingPapers# 13, Center for Cognitive Science, MIT; publishedin NLLT 6, 3 (this
issue).
Borer, Hagit: 1984, ParametricSyntax: Case Studiesin Semiticand RomanceLanguages,
Foris,Dordrecht.
Bresnan, Joan: 1982, The Mental Representationof GrammaticalRelations,MIT Press,
Cambridge,Mass.
Bresnan,Joan and Sam Mchombo:1987, 'Topic, Pronoun,and Agreementin Chichewa',
Language63, 741-782.
Chomsky,Noam: 1980, 'On Binding',LinguisticInquiry11, 1-46.
:1981, Lectureson Govemmentand Binding,Foris,Dordrecht.
1986a, Barriers,MIT Press,Cambridge,Mass.
1986b, Knowledgeof Language:Its Nature,Origin,and Use, Praeger,New York.
Chung,Sandra:1976, 'An Object-CreatingRule in BahasaIndonesia',LinguisticInquiry7,
1-37.
Czepluch,H.: 1982, 'Case Theory and the Dative Construction',The LinguisticReview2,
1-38.
Foley, Williamand Robert Van Valin: 1984, FunctionalSyntaxand UniversalGrammar,
CambridgeUniversityPress,Cambridge.
Jackendoff,Ray: 1987, 'The Statusof ThematicRelationsin LinguisticTheory',Linguistic
Inquiry18, 369-412.
Jackendoff,Ray and Peter Culicover: 1971, 'A Reconsiderationof Dative Movements',
Foundationsof Language7, 397-412.
Johnson, Kyle: 1987, 'Clausal Gerunds, the ECP, and Government',unpublishedms.
UCLA.
Kayne, Richard:1984, Connectednessand BinaryBranching,Foris,Dordrecht.
Kimenyi, Alexandre: 1980, A Relational Grammar of Kinyarwanda, University of Cali-

forniaPress,Berkeley.
Kisseberth,Charlesand MohammadAbasheikh:1977, 'The Object Relationshipin ChiMwi:ni,a BantuLanguage',in P. Cole and J. Sadock (eds.), Syntaxand Semantics,Vol.
8: GrammaticalRelations,AcademicPress,New York, pp. 179-218.

THE SYNTAX OF APPLICATIVES

IN CHICHEWA

389

Lieber,Rochelle: 1980, On the Organizationof theLexicon,unpublishedPh.D. dissertation,


MIT.
Marantz, Alec: 1982, 'Affixationand the Syntax of Applied Verb Constructions',in
Proceedingsof theFirstWestCoastConferenceon FormalLinguistics,StanfordUniversity,
pp. 330-340.
: 1984, On the Natureof GrammaticalRelations,MIT Press,Cambridge,Mass.
Marlett,Steve: 1986, 'SyntacticLevels and Multiattachment
in SierraPopoluca',IJAL 52,
359-87.
Mchombo,Sam: 1986, 'The Nonexistenceof Verb Object Agreementin Bantu', unpublished ms., San Jose State University.
Merlan,Francesca:1976, 'Noun Incorporationand DiscourseReferencein ModernNahuatl', IJAL 42, 177-191.
Mithun,Marianne:1984, 'The Evolutionof Noun Incorporation',Language60, 845-895.
Mtenje,A.: 1984, 'An AutosegmentalAnalysisof ChichewaVowel Harmony',unpublished
ms., UniversityCollege, London.
Oerhle, Richard:1975, The GrammaticalStatusof the EnglishDative Altemation,unpublished Ph.D. dissertation,MIT.
Perlmutter,David (ed.): 1983, Studiesin Relational Grammar1, Universityof Chicago
Press, Chicago.
Perlmutter,David and Carol Rosen (eds.): 1984, Studies in Relational Grammar2,
Universityof Chicago Press,Chicago.
Ramamurti,G. V.: 1931, A Manual of the So:ra: (or Savara) Language, Government
Press,Madras.
Rizzi, Luigi: 1986, 'Null Objects in Italianand the Theory of pro', LinguisticInquiry17,
501-557.
Rosen, Carol: 1984, 'The Interface between Semantic Roles and Initial Grammatical
Relations', in D. Perlmutterand C. Rosen (eds.), Studies in Relational Grammar2,
Universityof Chicago Press,Chicago, pp. 38-77.
Sapir, Edward:1922, 'The Takelma Languagesof SouthwesternOregon', in FranzBoaz
(ed.), Handbookof AmericanIndianLanguages(BAE Bulletin40) 2, pp. 1-296.
Seiter,William:1980, Studiesin Niuean Syntax,Garland,New York.
Stowell,Timothy:1981, Originsof PhraseStructure,unpublishedPh.D. dissertation,MIT.
Sylla, Y.: 1979, GrammaticalRelationsand Fula Syntax, unpublishedPh.D. dissertation,
UCLA.
Travis, Lisa: 1984, Parametersand Effectsof WordOrderVariation,unpublishedPh.D.
dissertation,MIT.
Trithart, Lee: 1977, Relational Grammarand Chichewa Subjectivization,unpublished
Ph.D. dissertation, UCLA, distributed by Indiana University Linguistics Club,
Bloomington.
Williams,Edwin: 1984, 'GrammaticalRelations',LinguisticInquiry15, 639-674.
Williams,MarianneMithun:1976, A Grammarof Tuscarora,Garland,New York.
Received 25 February1986
Revised 7 December 1987
Departmentof Linguistics
McGill University
1001 SherbrookeStreetWest
Montreal,PQ H3A iG5
Canada

Vous aimerez peut-être aussi