Вы находитесь на странице: 1из 50

Lingua 82 (1990) 15 !-200.

North-Holland 151


A Universal-Tylmiogica| Perspective

Jae Jung SONG*

Department of Enghsh Language and Literature. National University of Singapore.
!0 Kent Ridge Crescent. Singapore 0511

Received August 1990

This paper provides a universal typology of causative constructions, on the basis of

which it further projects a diachronic model of causative affixes. The model is able to
predict two more potential sources o ' " e r than lexical verbs of cause from which causative
affixes may originate, since the universal typology sanctions only the compact, AND, and
PURP types of causative construction, it gains empirical support from cross-linguistic:
evidence, by helping to trace the hitherto unknown origins of causative affixes. It also
outlines the d~achrenic path through which causative affixes come into being. Finally, the
model offers an e~plana:]on of the co-occurrence of causative affixes and some purposive
elements in certain languages.

1. Introduction

Since the classic paper of Greenberg (1966) on word orders and their
correlations, a number of linguists have attempted to incorporate typological
facts into diachronic explanations (see Greenberg (1978) for his own epitomiza-
tion of the interface between diachrony, synchrony and l~nguage universals).
The most prominent type of such research has been the use of synchronic
word order universals in the reconstruction of word order, as found in
Lehmann (1974), Vennemann (1973, 1975), Friedrich (1975), and Hawkins
(1983) (however, cf. Watkins (1976)). Although these studies have andoubtedly

* I would like to record my gratitude to Edith Bavin, Hilary Chappell, Rabin Hardjadibrata,
Iohn Newman, and Peter Paul for their comments and suggestions at various stages of the
production of this paper. I am especially grateful and indebted also to Keith Allan, Barry Blake,
R.M.W. Dixon, 3ohanna Nichols, and Graham Mallinson for reading a much larger version of
this paper (i.e. Song (1989)) and for making numerous insightful comments and suggestions. 1
alone am reponsib!e for the errors and misinterpretations that, despite the above persons' efforts,
may still remain in the paper.

0024-3841/9l/$03.50 © 1991 - - Elsevier Science Publishers B.V. (North-Holland)

152 J.J. Song / The rise of causative affixes

kindled much interest in relating word order typology and reconstruction of

word order, what they claim to have excavated in their diachronic research is
anything but conclusively established. The fundamental weakness of this kind
of research is the assumption that languages are consistent in terms of word
order correlations, when the majority of the world's languages are known to
be inconsistent (Comrie (1981" 205)). On the other hand, some linguists (e.g.
Giv6n (1971a, b), "W~.~r
"q (1986), Hall (1987, 1988), Craig and Hale (1988)),
under the linguistic archaeologist's manifesto: 'Today's morphology is yester-
day's syntax', have attempted to either reconstruct earlier word order on the
basis of synchronic distribution of morphological affixes (cf. Comrie (1980),
Mallinson and Blake (1981), and Kefer (!985) for some pitfalls of this kind of
approach)., or trace origins of morphological affixes to earlier lexical sources.
Although most of the word order related works, in particular, s~m to laave
failed because there is such a large number of counterexamples, there are no
inherent contradictions in the embracing of the typological methodology in
diachronic linguistics (Comrie (! 98 l" 212)).
Jn this paper~ I will undertake a diachronic investigation of the latter type
of linguistic archaeology. I will attempt to propose a diachronic model of
causative affixes from a universal-typological perspective (see section 4) and
thus to trace possible sources of causative affixes in addition to those already
suggested in the literature (see sections 2 and 5.1). Such an attempt is made
possible on ,~h~basis of the universal typology of causative constructions that
I will present in section 3; the universal typology can be of great assistance in
setting limits to the potential for sources of causative affixes, since it predicts
or sanctions only particular types of causative construction. Conversely, the
validity of the typology of causative constructions put forward in this paper
can be strengthened if it can be shown to be indirectly involved in predicting
the possible sources of causative affixes: afortiori major sources of causative
affixes can be discovered due to the particular typological perspective that
has motivated the typology. Finally, on the basis of the diachzonic model of
causative affixes, [ will also account ,gor the restrictions on the occurrence of
causative affixes and some purposi've morphemes in some languages (see
section 5.2).
Wilt it be possible to propose such a diachronic model when the majority
of the world's languages lack historical documentation? Indeed, Lighffoot
(1979: 6-7) warns that languages without such ampJe historical documents as
Chinese, Tamil, and those of Indo-European and Semitic famil!es cannot
fon'n an appropriate basis for work on syntactic change. If one strictly
follows Lightfoot's advice, one cannot venture into any serious diachronic
J.J. Song / The rise of causative affixe~ 153

research on languages that do not ha~.r~earlier historical recordings. I will not

accept Lightfoot's pessimistic warnings, because I think that a serious din-
chronic investigation can be carried out under the assumption that a universal
typology delimits the variation of a given construction, e.g. the causative
construction. In particular, it will be shown that on the basis of the universal
typology of causative constructions put forward in this paper, one can indeed
trace affixes to their origins. In other words, lack of historical documentation
can be offset by argument for and justification of a given diachronic analysis.
In fact, even Lightfoot (1979: 380) himself acknowledges, albeit in a sligb',iy
different context, that grammatical re-analyses require argument and justifi-
cation, even where there is an abundance of (historical) data. For exaiaple, if
one wants to establish that form y has historically derived from form x, an
abundance of data regarding these *u'o... fo r.....
m ~ .o;
. n. o t. a. too;..., o^..A:,: . . . .
all for proving that form x is the ancestor of form y. This is because the data
itself does not justify the origin of form y. What is needed here is a
conceptual frame of reference that justifies the diachronic relationship be-
tween the two forms. In fact, such a procedure is the basis for distinguishing
poiysemy from homonymy. Therefore, in the case of lack of histolicai
recordings, the attitude taken by Lightfoot does not advance our general
knowledge of diachronic changes, except for noting that the formal similarity
between fo=m x and form y is accidental. If one, however, makes the same
assumption as I have done here, what one can at least establish is that form x
and form y cannot possibly have any diachronic relationship. What I will
pro,ride in this paper is then such a general conceptual frame of reference that
can show linguists where to look for origins of causative affixes in individual
languages in which they specialize. I have to make it clear at the outset that,
although I present a number of languages as providing evidence for the
diachronic model, I do not lay claim to the actual history of each of these
languages. This is a matter for language specialists to confirm or disconfirm,
as the case may be. What I present here is my cross-linguistic observation built
on a few other linguists' sporadic observations (see below how these !inguists
have noted the formal similarity of causative affixes to other non-causative
elements in their specialized language groups or languages) that in many
languages causative affixes have strong formal resemblance to certain coherent
groups of linguistic elements that ! will later characterize as PURP. 1 At the

T h e following a b b r e v i a t i o n s a r e used in the glosses: A B S = a b s o l u t i v e , A C C - - a c c u s a t i v e ,

A C T = a c t o r , A G = agent, A L L = allative, A P P L -- a p p l i c a t i v e , A R T -- article, A S P = aspect,
A U X = auxiliary, B E N = benefactive, C L = classifier, C O M • = c o m p l e m e n t i z e r , C P = c a u s a t i v e
prefix, CS = causative suffix, DA = different actor, DAT = dative, DEC = declarative,
~,54 J.J. Song / The rise of causative affixes

same time, unJ~ke these linguists, I will provide a diachronic model that can
argue for and justify the more-than-chance relationship that may exist be-
tv, een causative affixes, and non-causative elements including PURP.

2. Previous studies on origins of causative affixes

To the best of my knowledge, the earliest explicit reference to possible

relationships between causative affixes and other (noncausative) morphemes
is a comparative morphological study of Tibeto-Burman languages by Stuart
Wolfenden (1929). Even though it is the oldest of the studies that I will review
here, Wolfenden's classic work is definitely not the least insightful of them in
ff.zt he appeals to metaphorical extension to account for the observed
similarity between causative affixes and directive or benefactwe affixes in
some of the Tibeto-Burer, an languages.
Wo!fenden (!929: 4_6-4.8) notes that the causative affix s- in Tibetan may
ha ce originated from the general directive element~ since the latter also
appears in a group of verbs in the form of s-; e.g. s-pro-ba "to make go out'
vs. s-neg(,~)-pa 'to run to or towards'. Wolfenden regards this general
d~rl:ctive element as indicating either (a) general direction into the condition
or :;/ate denoted by the verb or (b) action to, towards, or for an entity; this
dir~ctive element is then metaphorically extended to '~'u.,,.. .,,,,~o,.,.,,,,,;
. . . . . . . . ~,...,-,-,;""~ ~a.,.,~,,.
the causee in the causative construction, the object of the directive element
undergoes the same kind of change in condition or state, or is subject to the
same kind of movement by another entity. Wolfenden (1929: 199-200) postu-
late~ for Burmese the same development from a directive element to a
causative affix on the basis of the fact ti'at the same regular aspiration of the
initia~ consonants of verb roots recognized as being characteristic of an
old directive element occurs also in causativized verbs. A more interesting
case comes from Ao, which has a causative affix dg~k-tsa- (Wolfenden
(1929: 139, 152)). What is noteworthy is the fact that in Ao the causative affix
cons~ts of two components, dfik and tsa (see section 5.2). As Wolfenden

D E P = d e p e r i d e n t , D E T -= d e t e r m i n e r . D I R = d i r e c t i o n a l , D P S T = d i s t a n t past, D S = d i f f e r e n t
subject, D T R = ditransitivizer. E ( R G ) = ergative, E M P ( H ) = e m p h a t i c , F I N = finite. F U T = fu-
!ure, [ N D =: indicative, N E G = negative, N F = non-finite, N M L = n o m i n a | i z e r , N O M = n o m i n a -
:ive, N U = n u m b e r , P E R F = perfective, P L = plural, P R E S = present. PREV = preverb. PRT =
)article, P S T = past, P U N T = p u n c t u a l , P U R P = p u r p o s i v e . P U R P ( o s e ) . Q = q u e s t i o n . R P S T =
"ecent past, $ A = s a m e a c t o r , S E Q = s e q u e n t i a l , S G = singular, S R = s w i t c h r e f e r e n c e . S T A =
atus, S U B = s a b j e c l , S U B J = s u b j u n c t i v e , T O P = topic, T P S T = t o d a y ' s past.
J . J . S o n g / The rise o f causative affixes 155

(1929: 137--139) himself notes, -tsa- is a dative-directive affix denoting that

the action is accomplished for the beneft of some entity other than the actor.
The other component d6k- is traced back to the same root as the Tibetan
verb ajug-pa "to cause, to compel'. However, Wolfenden (1929: 139) is not
completely sure whether the second component of the causative affix is
indeed the same as the benefactive element in Ao:

"The second element -tsa- ... is perhaps comparable to the B.[urmese] causative suffix -~e,
though, on the other hand, it [the causative -tsa-] may be nothing but the Ao [dative-] directive
element -tsa ... indicating that the (causative) verbal action falls upon an external object and
not on the speaker .., If-tsa- is this directive element, it is not a little interesting that here in Ao
association with roots carrying a causative sense has elevated it from its originally purely
directive function to an apparently "'causative" one, just as it did Tibetan -s-.'

Wolfenden is here relying on the same kind of metaphorization mechanism

that is reminiscent of the so-called localist theory (e.g. Anderson (1971), Diehl
(lC}7~ Radden / i Q R ~ " m~tnnh~rlo~l fr~nele~re c~o~,,r f~'am ...... ta

(less abstract) notions to less concrete (more abstract) notions° However, his
explanation is far short of being of the kind of general conceptual framework
that I alluded to in section 1. It is therefore not at all ct~ar in Wolfenden's
explanation on what grounds the directive element, rather than the causative
element, is basic: one can easily argue for the opposite direction of change,
i,e, from causative to directive (not to mention why the dative-directive affix
is used in addition to the causative affix; see 5.2 for my own explanation).
What is lacking in Wolfenden's account is an independent parameter to
determine the direction of change. The universal typology of causative
constructions that I will present below provides such a parameter, because it
clearly indicates the direction of change regarding these two functions: from
directive to causative.
Giv6n (1971a) is a classic paper that has piomulgated the linguistic
archaeologist's manifesto cited earlier. Giv6n has demonstrated with great
plausibility grammatical mechanisms through which independent lexical ele-
ments change into affixes. Although he discusses causative affixes originating
primarily from higher lexical ve-bs "to cause', Giv6n's main concern is to
establish the correlation between word order on the one hand and prefixes
and suffixes on the other. In OV languages, lexical causative verbs become
suffixes, whereas in VO languages, they become prefixes z (the following
diagrams are taken from Giv6n (1971a)).
2 Vennemann (1973: 21-22) suggests thal the prerequisite for grammaticalization of lexicai
causative verbs (i.e. into affixes) is that the higher and lower verbs should be contiguous.
156 J.J. Song / The rise of causative affixes

OV (syntax) Suflixation (- -,arphology)


NP V V NP V - [sufflX]V

VO (syntax) Prefixation (morphology)
/ S
/ \\
V NP V v[prefix] - V NP

Therefore, Giv6n is not of much help at all in tracking down causative

a~xes, except, that he notes (197ia: 4~2) that one primary source for causa-
tive affixes is lexical verbs of cause. Unfortunately, Giv6n does not tell us
what other nonprimary sources of causative affixes there are.
Coracle (1981: 176) cites in passing as evidence for the causative affix as an
indicator of increase in valency (due to the introduction of the causer) the
fact that in r~any languages the same morphology is used both in morpholo-
gical causatives and ditransitivization of monotransitive verbs: in WoJof, the
¢~m× -a! functions as a causative marker as in (3) and as a ditransitivizing
marker as in (4):

Di naa toog-al nenne bi
FUT ~SG sit-CS child the
'I will make the child sit.'

Mallinson and Blake (1981:423--424) refute Vennemann's suggestion, arguing that there is
no such prerequisite: noncontiguous lexical causative verbs can be grammaticalized by
gravitating to the lower verb. Heine and Reh (1984) provide ample evidence in African
languages for Mallinson and Blake's position. Bybee (1985: 39--41) also provides evidence in
support of Mallinson and Blake's position: morphology is not immovable fossilized
J.J. Song / The rise of causative affixes 157

(4) Wolof
Mungi dyeing-a1 eleew yi t~r6-6m
he read-DTR pupil the-PL book-his
"He is reading his book to the pupils."

Comrie does not c'iscuss this interesting observation a,~y further: the use of
the same morphc, logy is purely syntactically motivz!ed in that the only
function of the morphology in both cases is to in,.rease the valency of
the aasic verbs, since there is no necessary connection with the semantic
parameters of causative constructions, e.g. control or semantic causative
types, i.e. direct vs. indirect causation. However, a serious question remains
unanswered: why none other than the dative marker (I will later recognize
this as an instantiation of the metaterrn PITXP) is rased to function as a
causative affix? In other" words, if a lan~'.,,ge like Wolof has more than one
valency increasing affix (e.g. comit~'.~ve, instrumental as well as benefactive),
why does the causative affix h,~ve the same form as whatever marker is an
instantiation of the metaterm PURP, and not comitative, for instance? Unless
this is answered, Comrie's putative syntactic explanation is nothing more
than a~ interesting piece of observation.
Finally, i't:ggy (1987) is an ambitious piece of work to account fo~" a
striking feature of (Uto-)Aztecan languages, the use of verbal suffixes to
function both as causa'~ves and as applicatives (cf. Langacker (1977: 144-
147); cf. a similar case in Wolof above). Indeed, anyone who has studied
•rammatical descriptions of the Aztecan languages will have undoubtedly
n ,:iced this interesting and pecaliar phenomcnon. Taggy attempts to demys-
tify t[~:, Aztecan puzzle by using Cognitive Grammar developed by Lang-
acker (1983, i987) However, I will show below that he simply fails to fulfil
his promise. His accot,:' does not lend any of the much needed insight into
the cognitive nature of c:~usation, not to mention the diachrony of causative
affixes, al~,hough I have to point ~,,:~ that Tuggy's account is basically
synchronic; it is not clear at all in Tuggy's ,.nosition what the diachronic
relatic.m of causative and applicative affixes is (which L ~*rived from which).
Tuggy notes that in Tetelcingo Nahuatl (TN), as in o~her Aztecan languages,
the same suffix -liya functions as both causative and applicative as shown in (5):

(5) Tetelcingo Nahuatl

(a) ni-k-mewi-liya
'I raise him.'
158 J.J. Song / The rise oJ causative affixes

(b) ni-k-tesi-liya3
I-her-grind = corn-APPL
"I grind corn for her.'

The basic motivation that Tuggy (1987: 598) promotes for the similarity
between the causative and applicative suffixes (which share the same form in
(5)) is that the applicative function in (5b) involves the causing of possession:
the process of grinding corn by the causer (or the grinder) results in someone
possessing a quantity of dough. In other w o r d s , the possessive relation is
caused by the process of grinding itself. This causation relation in (5b) is of
the same kind of relation found in (5a); the process of the causee's rising is
caused by the causer's (unspecified) action. This similarity is then the basic
motivation for the use of the same suffix in both causative and applicative in
TN. Therefore, in Tuggy's words (1987: 603); 'we have here a direct and
powerful expression of the close relationship of causatives to applicatives'.
This kind of analysis is not new at all, pace Tuggy, who claims that it is
Cognitive Grammar that can make such an analysis possible (1987: 587). In
fact, Tuggy's analysis is very reminiscent of Generative Semantics: Seuren
(1973: 37-38) proposes a similar analysis of indirect objects or datives in the
Generative Semantic vein (e.g. the sentence He gave the patient advice is
derived via the predicate raising rule from an underlying structure like He
caused the patient to have advice; also Comrie (1974: 25)). Tuggy's contribu-
tion is important in that it addresses the long overdue issue of the Aztecan
phenomenon. However, it fails to produce any insightful analysis of this
puzzling phenomenon, not to mention, of the diachronic mechanism of
causative affixes. It is thus silent on the diachronic path of the affixes in
quest,~on: what is the direction of the development, from the causative to the
applicative or vice versa? (see section 5).

3. A universal typolagy of causative constructions

I have set up a database of some 600 languages to determine the formal

variation of causative constructions. 4 A convenience sample of some 400
languages has emerged; s the descriptions of some 200 languages do not even

3 I do not know whether this sentence has an additional meaning: I cause her to grind corn.
4 For a full list of these languages and relevant references, see Song (1989).
s By a convenience sample, I mean that the sample has been constructed on the basis of
bibliographical availability of grammatical descriptions
J.J. Song / The rise of causative a~xes 159

deal with causative constructions. I have deliberately opted for a convenience

sample, since my aim is to formally characterize the causative construction,
but not to find any correlations between causative constructions and some
other linguistic features. I have abstained from constructing a sample statisti-
cally unbiased in genetic and geographical terms (cf. Bell (1978), Perkins
(1980), and Tomlin (1986) among others), for I know of no other way of
enhancing the possibility of delineating the formal variety of a given
function than by ever increasing the number of the languages that one studies
(witness the 'discovery" of object-initial languages and its subsequent effect on
linguistic theories). I have consulted the grammatical descriptions of as many
languages as possible. In sum, the more languages, the better typology, if the
typologist's goal is to establish the formal variety of languages with respect tc
given function. Further, since even genetically related or geographically

different constructions for the same function, I cannot always agree with
Tomlin (1986: 17-18), who argues that the validity and reliability of any
work in syntactic typology can be enhanced only when its language sample is
statistically unbiased in genetic and geographical terms. In this context, Pike's
sheer astonishment at how differently the languages within the same Gur
group in the Niger Congo family express causation is worth quoting
(1970: ll):

"This material from Bariba in Dahomey seems so different from Dagaari, Vagala, and Kasem
of Ghana and from the Mbembe and Degema of the lower part of Nigeria, that I re-checked ...
Somehow, the cultural universals of causation ... would have to be expressed in them also. Had
the Bariba type of data been overlooked in these other languages, or did it in fact not exist?"

One should not be under an impression here that I am discounting recent

developments in language sampling techniques (there are still serious
problems with these techniques, e.g. see Dryer (1989) for one such problem 6);
I find that e v e n cross-linguistic generalizations reached on the basis of genetic-
ally and geographically unbiased samples are to be tested against more and
more languages after all. Hence I have decided to start off with as many
languages as my economic and bibliographical resources allow.
I have identified three prototypical categories for my typology of causative
6 Another serious problem with current sampling techniques ns demonstrated by a hypothetical
situation in which some language families may not be represented in a given sample, because of
its small size, and these families may use a certain syntactic device not found in any other
language families. In such a situation, one may hardly expect to obtain a factually adequate
160 J.J. Song / The rise of causative affixes

constructions: the compact, A N D and PURP types. Due to limitation of

space, I will not here discuss the cognitive and pragmatic basis of the above
universal typology of causative constructions; interested readers are referred
to Song (1989). l will introduce a few metaterms to facilitate discussion of the
three types of causative construction. The term metaterm should be under-
stood as an abstract term, not as a linguistic element. In other words, a given
metaterm is a cover term for all instantiations ,of its prototype; a metaterm is
in a sense schematic of all its instantiations (Langacker (1987: 68)). An
analogy can be of help here. There ar ,nany kinds of tree. Although all these
trees have their own distinct minc :ures that warrant their different
names, e.g. maple, oak, pine, etc., th,; essential properties common to all of
them remain the same, treeness. So the terra tree is a superordinate category,
whereas terms like maple, oak, and pine, etc., are subordinate categories. It is
this sense in which a metaterrn is schematic of its instantiations.
The metaterms that | will use to account for my typology of causative
constructions are" S, V, N, 1, 2, cause, effect~ A N D , P U R P . There are also a
¢.,, cu,,~,,n,~,,~,,, u, L , , ~ metaterms u:_a:
l u l ~ .s .i a. t.g u., L..
uy ... : [ocuus{:], [Nffec;.,
[Vcause], [Veffectl, [Ncause] and [Neffecq. It is now in order to define these
metaterms. S represen,.s a s:-n*ence or clause. The numerals 1 and 2 represent
clause levels. Hence SI stands for the higher clause level, while S2 stands for
the lower clause level, or the clause of the deeper level o f embeddedness. V
stands for a verbal or predicate element. N stands for a nominal element.
Cause and effect need ne explanation. [Scause] means the clause whose
subject NP is the causer of a given causative situation, while [Seffect] is the
clause in which the subject NP is the causee. [Vcause] stands for all verbal
elements of cause, i.e., the elements that denote the causer's causing action.
[Veffect] stands for al~ verbal elements of effect i.e., the element that denotes
the caused action or state of the causee brought about by the causer. [Ncause]
means the causer NP, whereas [Neffect] means the causee NP. For instance,
in the following English sentence, made is an instantiation of [Vcause], and
cry is an instantiation of [Veffect]; John is an [Ncause]. whereas M a r k, is an

(6) English
John made Mary cry.

On the other hand, in the following Japanese example, the causative suffix
-use- is an instantiation of Wcause].
J.J. Song / The rise of causative aj~ixes 161

(7) Japanese
Taroo-ga Ziroo-o tomar-ase-ta
Taro-NOM Ziro-ACC stop-CS-PST
"Taro made Ziro stop."

Note that despite the actual surface differences, both the independent lexical
verb made and the suffix -ase- are instantiations of the same metaterr:~
[Vcause]. A N D refers to any elements that connect two clauses to indicate
that the two clauses, that is [Scause] and [Seffect], represent two sevarate
incidents or events that occurred in temporal sequence. So AND can be
schematic of either overt or covert markers, although it can be said that overt
markers indicate the kind of temporal sequence or connectedness in a more
forceful way than covert markers. P U R P represents an) elements tha"
indicate that the event denoting [Seffect] is yet to be realized or to come into
effect and that the other clause, [Scause], represents an event that occurred
with the purpose of realizing the event of [Seffect], hence PURP(ose).
Using this set of metaterms, I will construct three prototypical categories in
my typology of causative constructions. Note that some of the metaterms,
[Ncause] and [Neffect] are excluded from the prototypical schemata for the
sake of simplicity.
The most widely studied type of causative constraction is represented by
the first prototypical category in the form of:

(8) Sl (... [Vcause]+ [Veffect] ...) S! or

SI (... [Veffect]+ [Vcause] ...) S1

The exact order of the metaterms [Vcause] and ~reffect] is language particu-
lar. Hence the disjunctive formula for the first type of causative construction.
The essential property of this well-studied category is the contiguity of
[Vcause] and [Veffect]. In other words, no other elements can intervene
between these metaterms. Note that the sequence of the metaterms is bound
within a single clause. I will call this type the compact type of causative
construction. A typical example of this type of causative construction is
illustrated by the Japanese sentence in (7): so-called morphological causatives
belong to this category. At the extreme point of the contiguity of D/cause]
and [Veffect] lie lexical causatives, i.e. the same form used for both noncausa-
rive and causative functions, e.g. melt in English, or no formal resemblance
between noncausative and corresponding causative verbs, e.g. die vs. kill in
162 J.J. Song / The rise of causative affixes

English. The metaterm [Vcausc] does not have to be less than a free
morpheme, as in French:

(9) French
Je feral lire le iivre ~ Nicole
I will = make read the book (ACC) Nicole (DAT)
'I'll make Nicole read the book.'

Jacaltec also uses the same compact type of causative: 7

(1 O) Jacaltec
xa' ija' ix ya" tawet
made to carry CL/she her water to you
"She made you carry her water.'

The second type of causative construction is called the AND type of

causative construction and it is represented in the form of:

(11) SI (32 (...[Vcause]...) $2 + AND + $2 (...[Veffect]...) $2) S1

Note that the formula for the second category is not disjunctive unlike the
one for the compact t y w There are two clauses connected by the metaterm
AND; the two clauses are coordinated with each other and the ordering of
the clauses cannot be reversed. This is so, because this particular type of
causative represents such constructions that press the temporal sequence of
[$cause] and [Seffect] into service to express causation. The metaterm AND is
~chematic of such coordinating devices as sequential marking, conjunctions,
switch reference marking, or even zero marking. The function of A N D can be
carried out by zero marking, provided that the order of [Scause] and [Seffect],
as in (i l) is strictly adhered to. The mere juxtaposing of t.~- clauses can
iconically perform the function of registering the temporal sequence of the
events denoted by the clauses (e.g. Haiman (1985)). Mianmin, a Mountain
Ok language spoken in West Sepik Papua New Guinea, provides an example
of the AND type of causative construction, where the metaterm A N D is
realized by sequential marking (Smith and Weston (1974:138-139)).3

Craig (1977: 319) notes that in J~caltec the contiguily of two verbs is possible only in
C~HSatB". eS.
8 By and large, ] have adopted the genetic classification provided by the authors of the
grar0,matical descriptions used in this r~aper, If no such genetic classification is given by the
J.J. Song / The rise of causative aj~xes 163

(12) Mianmin
awok-o men-e ki-mab-o-a aai-e
mother-CL child-CL command-FUT-she-SEQ water-CL
fuela-n-a-mab-e bo
"The mother will make the child bathe.'

In (12) the sequential suffix -(i)a clearly demarcates the boundary between
the two clauses; each verb has its own full panoply of tense and personal
desinences. According to Dimmendaai (1982: 294-295), some Nilotic lan-
guages such as Turkana, Lango, and Noni use sequential marking to register
the metaterm AND in causatives. There are also languages that exploit the
switch referencing system to indicate the presence of AND in causatives.
Amele, spoken just south of the town of Madang, Papua New Guinea, is
such a language (Roberts (1987: 222)),

(13) Amele
ija od-ude-ce-min na qete-i-a
I SG do-3SG-SR(DS)- 1SG tree cut-3SG-TPST
'I made the man cut the tree.'

Note that in contrast to the Mianmin example in (12), one tense marker on
[Veffect] has its scope over the two clauses. Other Papuan languages like
Oroke.iva (Healey et al. (1969: 57), Waskia (Ross and Pa~i (1978: 18-22)),
and Gahuku (Deibler (i 976: 97-102)), and an Amerindi0~ language, Diegue-
fio (Langdon (1970: 153)), press the switch reference system into service to
register the metaterrn AND. Zero marking of AND is evident in Kinyar-
wanda (Kimenyi (~980: 160-163)). There are two free lexicai causative verbs
in this Bantu language, -teer- and -turn-. When the second verb is
chosen as [Vcause], the AND type of causative construction comes into

author, I have decided to follow the genetic classification of Ruhlen (1987). This does not
necessarily mean lhat I adopl his classification. Rather, it is because it is the latest overall genetic
classification of the world's languages available on the market. Therefore, I expect that some
linguists may not agree with my geuetie classification of their specialized languages or language
groups. Nevertheless, I will follow Ruhlen~ work, since the decision to follow it dc~s not have
any bearing on what | put forward in this paper.
164 J J. Song / The rise of causative affixes

(14) Kinyarwanda
umukoobwa y-a-tum-ye ~n-a-andik-a amabaruwa meenshi
girl she@ST-cause-ASP I-PST-write-ASP letters many
'The girl caused me to write many letters.'

Note that both [Vcause] and [Veffect] carry their own separate tense/aspect
markings as well as cross-referencing marking. Other languages that have
zero marking of A N D are Kobon (Davies (1981: 165)), Yapese (Jonson
(1977: 287)), and Patep (Lauck (1976: 18)).
The last type of causative construction so far simply subsumed under the
so-called syntactic or analytic type by linguists is in the following form:

(15) $1 ($2 (...[Veffect]...) $2 + P U R P ... [Vcause]...) SI or

Sl (...[Vcause]... $2 (... [Veffect]...) $2 + PURP) SI

The disjunctive formula for this last category indicates that the embedding
position of [Seffect] is language particular. The essential property o f this
category is that the [Veffect] or [Seffect] as a whole is somehow marked by
the metaterm PURP. Hence [ will call this type the P U R P type o f causative
constr;~:ion. The metaterm PURP represents any grammatical elements that
signal some sense of goal or purpose and the like. The model is schematic of
the type of construction consisting of two clauses, one representing an event
carried out for the purpose of realizing the other event represented by the
other clause. In other words, [Seffect] is not factually substantiated; it
remains a goal or purpose and nothing more. Case markers including such
adposidons and verbal affixes as locative, dative, directional, allative, purpo-
sive0 and benefactive are frequently used as instantiations of PURP. 9 There is
ample discussion of metaphorical extension regarding these markers in the
literature (e.g. Anderson (1971), Diehl (1975), Lakoff and Johnson (1980),
Radden (1985), Genetti (I986), Heine and Ciaudi (1986)): the general idea is
that metaphorical transfers proceed from more concrete (less abstract)
notions to less concrete (more abstract) notions. Therefore, spatial]directional
notions (e.g. allative) serve as a vehicle for metaphorical extension to dative/
benefactive or even on to highly abstract notions such as goal or ;.,urpose. In

The list may seem a random, evenmiscellaneouscollectionofease markers. However,all these

markers one way or another signalsome senseof goal or purpose. Randomnessof the list is more
apparent than real; we know that languages differ in coding the same semantic role by using
different case markers, but within certain boundaries. In the present case, the boundary is the
sense of goal or purpose.
J.J. Song / The rise of causative affixes 165

other languages the metaterm PURP is indicated by such verbal markings as

future tense, subjunctive, irrealis, or even aspect. All these markings can be
abstracted to some sense of non-factuality; this sense becomes most conspic-
uous when [Veffect] is used with these verbal m~rkings in conjunction with
[V¢~.":.:] in past tense, or some functional equivalents to past tense in
ten eless languages. Suppose [Scause] and [Seffect] are contrasted in terms of
tense, the former having already occurred for the purpose of carrying out the
latter. The event of [Scause] is encoded in a clause wit|: past time reference,
while the second event of [Seffecq is encoded with one of the non-factual
PURP markings. This asymmetry of temporal reference, despite the absence
of some explicit marker, can give rise to some sense of goai or purpose In
other words, there is some kind cf pragmatic inference involved ir extracting
the sense of goal or purpose from the asymmetry of tempordl reference within
[Scause] and [Seffect];~,Finally other" languages may ,use some independent
marker of PURP (i.e. free morphemes that may not be directly associated
with either of [Scause] or [Seffect]; see the Greek example in (23) below).
In the Lhasa dialect of Tibetan the a!!ative marker is used to mark the
metaterm PURP in its causative construction (Goldstein and Norn~ng
(1970)). The allative marker -ru performs its normal function to mark the
case of direction toward something (Hannah (~,973: 78-79)).

(16) Tibetan
n ~ qh6 lee qa che-ru cfifi-payE
I ham work do-ALL cause-lSGPST
'I made him work."

Basque is another language that uses its aUative marker to register PURP
(Saltareili (1988: 221)). In this isolate ianguage, when tWeffect] is ditransitive,
the PURP type of causative construction is used in lieu of the normal
compact type.

(l 7) Basque
Mikel-ek Jori Edume-fi liburu-a eros-te-ra
Mikel-ERG Jon-ABS Edurne-DAT book-SG/ABS buy-NML-,~ LL
beh~r-tu z-u-en
force-PERF 3sgE-AUX-PST
'Mikel forced ion to buy a book for Edurne?

Note that in this language [Seffect] or [Veffect] is nominalized prior to hosting

166 J.J. Song / The rise of causative affixes

PURP. The allative marker -ra performs its normal function as in the
tbllowing: marking the allative role (Saltareili (1988:2)).

(18) Basque
bihar ez n-a-u
parke-ra eraman-go
park-ALL carry-FUT
'She will n~,t take me to the park tomorrow.'

Gbeya, a Niger-Congo language, exploits its benefactive case marker for the
purpose of signalling PURP in its causative construction (Samarin (1966: 64--
65)). Compare the two following sentences in (19).

(i 9) Gbeya
(a) hfi w~-r~ t~: s~
put BEN people come first
'Let people come first.'
(b) nd6 r6 tt; kof~a hfi bisa k6 r6
they come bride BEN boy the
'They would come for the bride for the sake of the boy.'

Kunjen is a language that presses its purposive case marker into service to
function as PURP (Sommer (1972: 44, 118)).

(20) gunjen
(a) abm ay egrj-a$ igun ay lalarjan
person .~ food-PURP go-PRES I uncle-DAT
'~ am going for food for my uncle.'
(b) arrjg ug gan_iyar ud il irjun abm a_tan-a¥ ambar
child ffighten-SUB-AG dog he him person bite-PURP cause-RPST
'The frightened child cat, red the dog to bite him.'

In this Australian language, the purposive marker is attached to [Veffect],

when it is either a transitive or middle verb. 1° Incidentally, the marker ~a~"
also encodes dative, directional as well as purposive; the metaphorical
lo Sommer (1972:118) finds the appearance of the allative marker in causatives unexpected.
However, given the typology p~esen;ed in this paper, it is anything but unexpected.
Y.J. Song / The rise of causative affixeA 167

extension that I alluded to above can be seen to be prevalent in this language.

Swahili is a language that exploits subjunctive mood to express PURP in its
causative construction; [Veffect] is in subjunctive mood ~Vitale (1981" 78-80,
152-! 68)).

(2 !) Swahili
Ahmed a-li-m-fanya mbwa a-l-e samaki mkubwa
Ahmed he-PST-him-make dog he-eat-SUBJ fish large
'Ahrned made the dog eat a large fish."

Agaw (Hetzron (1969: 16)), and Tzotzii (Aissen (i987: 214-215)) exploit
subjunctive mood for marking PURP. Jacaltec, a Mayan language, is a
language that uses future tense marking on [Veffect] in order to register
PURP, when [-Vcause] is iptze (Craig (1977: 239)). 11

(22) Jacaltec
x-~-(y)-iptze ix xo' ~-s-tx'ah-_a'
ASP-ABS-ERG3-force CL/she CL/her ABS3-ERG3-wash-FUT
xil kape s-ti' ha'
clothes ERG3-mouth water
'She forced her to wash the clothes by the fiver.'

Greek is a PURP type language that uses an independent PURP element.

According to Joseph and Philippaki-Warburton (1987: 171), a separate sub-
junctive particle na is used to mark PURP.

(23) Greek
6kana to~ jfini na fiji
made-ISG/ACT the -~-John-ACC PRT leave-3SG
'I made John leave.'

Kanakuru (Newman (1974: 111-112)), Russian (Maltzoff (1985: 175)) and

ix Jacaltec is very interesting in that there seems to be some variation betweLn the PURP type
of causative as in (22) and the A N D type of causative. Craig (1977: 271) notes that some speakers
use a coordinating suffix -(n)i in lieu of the future te~;se suffix (i.e. the metaterm PURP), as ill the
xc-ach-w-iptze hin haw-echma-ni
ASP-ABS2-ERGI-force ABSi ERG2-waiband
q forced you to wait for me."
i68 J.J. Song / The rise of causative aJ~xes

Ewondo (Redden (1980: ! 13-116) are like Greek in that they use independent
particles ~o represent the presence of PURP.
As their schematic formulae stand, shown in ( l l ) and (15), the two
prototypical types both involve two clauses. When languages have either the
AND or PURP type of causative construction, they are found to deviate
from the prototypical AND or PURP type in terms of formal reduction. Thet
is, languages deviate from the prototypicai types in terms of reduction of the
two clauses involved. So in some languages tense or aspect marking on
[Veffect] will have its scope over the sentences of [Scause] and [Seffect] (e.g.
see (13) above), whereas in other languages [Scause] and [Seffect] each have
their own tense markings (e.g. see (12) above). The ultimate degree of
reduction will then be the juxtaposition of the metaterm [Vcause] and
[~'effect]; the AND and PURP types on the one hand and the compact type
on the other constitute a kind of continuum of compactness of [Vcause] and
[Veffectj. One may suggest that this observation then motivates the reduction
of the thrt:efold typology into a twofold one: AND and PURP types of
causative construction. Indeed, such formal reduction is not unique to the
causative cor~,struction per se; formal reduction is everywhere :., ,he language
system, phonology, syntax, semantics etc. (cf. Haiman (1985. 102-147) for
many exampl,:s of formal reduction in diverse grammatical phenomena; see
also Bybee (1985)). In sum, the prototypi,~al compact type schematized as
~'cause] + [Veffect] or [Veffect] + [Vcause] can be regarded as the final stage
to which the other two types of causative construction ultimately arrive
through formal reduction. 12 However, in this paper I will teavc ~:~en the
possibility of ':educing the threefold typology into the twofold one, although I
am interested particularly in cases where the metaterm [Vcause] is materiv!-
ized as a cau'J;ar:ve affix or less than an independent lexical item, i.e. cases t,:
[Vcause] + [~,effect] or [Veffect] + try'cause].
Finally, it should be mentioned, albeit briefly, that the prototypical PURP
causative construction '~s non-implicative because of the inheient nature of the
metaterm PURP: the truth of [Seffect] is not entailed by [Scause]. This is so,
because the metaterm PURP marks non-factuality of [Seffect]. As the proto-
typical schema for the construction in (15)clearly indicates, the PURP type
construction is originally intended to express purposive function. Neverthe-
less, it is pressed into service to express causation in a number of languages. 13

~2 Tl'is is basically the main theme that runs throughout Bybee (1985). Bybee (1983: l i-[3i
propeses that expression types ranging from lexical to syntactic form a continuum of fusion, very
muc[,, in the sense of the reduction of [Vcause] and [Vcffect] developed in this paper.
~3 A similar comment applies to the AND type of causative ~onstruction. The schema in (i !) is
Y.J. Song / The rise of causative affixes 169

In the literature this PURP type of causative has not been cegarded as a
genuine causative construction. Indeed, the fact that the non-implicative
PURP construction is used to express (implicative) causation relation may
come as a surprise to linguists who are heavily influenced by the western
philosophical logic of causation (e.g. Lewis (1973), McCawley (1976) among
others): it is maintained that [Scause] entails the truth of [Seffect]. My cross-
linguistic survey strongly suggests that the western logic of causation simply
does not fit in ~ith the natural language fact: the semantic entailment relation
between [Scause] and [Seffect] is not a necessary condition for being a
causative construction (see Song (1989) for how pragmatic inferencing works
in both AND and PURP causative types to obtain the causative interpreta-
tion and how implicativeness is propositionalized in the PURP type of
causative construction via the strengthening of the pragmatic inferencing).
Perhaps, due to this philosophical logic of causation (i.¢. cause entails effect),
the PURP t y ~ of causative construction has not yet been recognized as a
causative construction par excellence. In section 5.2, i will present some
striking diachronic evidence from some languages for the existence of the
PURP type of causative construction: it will be shown that the PURP type
causative is fossilized in morphological causatives (see also Song (1988a) for
Korean, in which the purposive construction is in the process of changing
into a causative construction).

4. A diachronic model of causative affixes

Where do causative affixes come from? In order to answer this question, it

is first necessary to see what types of causative construction are available in
the world's languages. And the [universal typology provided in section 3 does
just that: the compact, AND, and PURP types. The r~ext thing to do is to
determine what potential sources these causative typ,~s may contribute for
causative affixes. To put it in another way, it is most i~nportant to relate these
causative types to causative affixes, since (at least intuitively) they seem to be
the natural providers of causative affixes.
Now what components is the typical morphol,jgical causative verb made
of?. The metaterm [Veffect] will form an essential part of the morphological
causative verb: one of the two essential compor, ents of causatives is ~h~:basic

intended to represem any coordinated construction that expresses sequential meaning. This
particular construction is also exploited to express carsative relation in a number of languages.
170 J.J. Song / The rise cf causative affixes

verb denoting the action or state of the causee. The other essential component
is, of course, whatever elements represent the metaterm [Vcause]. The typical
morpho!ogical causative verb is then a verb in which the [Vcause] element is
morphologically fused with the [Veffect] element. The compact type of
causative construction must be excluded from the present consideration,
because it is this very type of cau~a:,ve construction that is schematic of the
typical morphological causative. This leads to the other two causative types;
the AND and PURP types should be taken into account in order to see
who*her they have anything to contribute to the sources of causative affixes.
The schemata for the prototypical AND and P U R P types of causative
construction are presented below for ease of e.',.positi3n:

(24a) AND Type

Sl ($2 (...[Vcause]...) S2 + AND + S2 (...V[effect]o .) $2) S1
(24b) PURP Type
Sl (S2 (...[Veffect]...) S2 + PURP ...[Vcause]...) S1 or
SI (... [Vcause]... $2 (...[Veffect]...) $2 + PURP) SI

The most obvious source of causative affixes in both types in (24'~ is the
element representiHg p~'cause]: independent causative verbs are reduced to
affixes, which are then attached to basic verbs of [Veffect]. The diachronic
mechanism of causative affixes originating from independent lexical ~'cause]
eleme. ~ is that of formal reduction° widely discussed and documented in the
literature. So I will not say any more about this particulai" source. One may
wonder ~t this point what will happen to the metaterms A N D and PURP in
the proccess of the amalgamation of the [Vcause] and [Veffecq elements. I
will come back to this question shcrtly.
The other prominent participants besides [Veffect] in the above schemata
are the rnetaterms AND and PURP. When they are frequently used in
causative constructions, they will be st~rongly associated with causation per se.
This functional association is increasingly strengthened via frequent use of
either the AND or the PURP type of causative construction (see Bybee
(1985: 119-123) on the role of frequency in various morphological phe-
nt, mena). If it is further assumed tha~t the function of [Vcause] can be
somehow taken over by either the metaterm AND or the metaterm PURP,
then these metaterms may also be excellent sources of causative affixes. The
verb of [Vcause] tends to be highly semantically bleached in the sense that it is
not specific in terms of the type of causing action; in many languages tke
J.J. Song T;:e rise of causative ai%-res 171

(independent) causative verb is semantically general with the meaning of "do',

"make', or "cause', etc. In functional terms, such generality in meaning is
tantamount to nothing more than the signalling of causative function. And if
the schemata in (24) are by now identified as functioning as causatives, the
presence of either AND or P U R P will be sufficient to signal causative
function by being strongly associated with the causative. In other words, the
semantically bleached [Vcause] is now large|y predictable and even redun-
dant, thus very much subject to complete loss, as Kiparsky (1982: 67) notes:

"Morp~ol,,sical materizl which is prettic~;~!e on the surface lends to be more susceptible to loss
than morphological material which is r~et predictable or~ lhe surface."

This means tha~: the metaterm A N D and the metaterm PURP are now
legitimate proxies for the metaterm [Vcause]. Further, when the metaterms
AND and PURP are insti*,utionalized as the [Vcause] element, it seems
plausible enough to suggest that ~hcy become (derivational) affixes through
formal reduction which is also shown :~bove to affect independent lexical
causative verbs. ~ It is not unusual in languages that such a take-over occurs.
A good example th:3t immediately comes to mind is the nasal vowels in
French: the nasal stop is known ~to have dropped out after the preceding
vowel took over nasality, e.g./ban/ ~ / b S / .
Having suggested that th,= functional take-over of the kind that ~ have
illustrated above is possible, ! will now propose a diachronic model to
adumbrate how the metaterms A N D and P U R P end up as causative affixes.
The initial stage in the model is: the originally noncausative constructions, i.e.
juxtaposition of two clauses (AND Type) and complex purposive construc-
tions (PURP Type), begin to be used to express causation. At this stage, the

a4 One may point out that the use o f "formal reduction' here is largely inappropriate, because
these metaterms are already morphologically bound with [Veffect] in the majority of the cases
cited in this paper (see examples from (|2) to (22~). However, ! do not think that the use is
inappropriate, because th; par~meler is inherently a continuous notion: derivational affixe~ ~adi
as causatives are morpho-phc, nologically more bound to theh" hosts than such grammatical
markers as case, tense, mooS. etc. The rnetaterms A N D and PURP are closer to these
grammatical affixes than to dezivational affixes. Therefore, it is completely appropriate 1o speak
of "formal reduction' when these gramrnatical metaterrns are changing into deriva~ional affixes,
despite the fact that they are ~]ready bound to the [Veffect] element. Cf. Bybee (1985), who
explores the continuum of fusio a ranging from iexical to syntactic and finds on the basis of cross-
linguistic evidence that derwaticnal affixes are more tightly bound to their hosts than inflectional
or grammatical affixes are. The term "formal reduction' ~bou|d be interpreted precisely in this
172 J.J. Song / The rise of causative affixes

presence of the [Vcause] element is absolutely necessary. The next stage will
be: the streogthened association of the AND and PURP constructions with
causation is such that the metaterm [Vcause] can be optionally deleted. The
penultimate stage is a situation it, which the [Vcause] element no longer
appears in the AND and PURP constructions; in other words, the originally
noncausative constructions have now become genuine causative constructions
via a complete take-over of [Vcause] by the metaterms A N D or PURP (there
may remain some grammatical information that indicates that these recta-
terms are not ofginally [Vcause] elements (see Yaqui below)). At the final
stage, the AND and PURP elements become formally (i.e. morpho-phonolo-
gicaily) reduced into affixes, which are then attached to the [Veffect] element
through the usual morphological process; they now become fully derivational
causative affixes (unlike the penultimate stage, there may not be any gram-
matical infermation aw.i~ablc that indicates that these metaterms are not
originally [Vcause] elements). At this particular stage, one expects these "new"
causative affixes to behave like usual derivational affixes: they may come to
lose productivity or generality (i.e. they may have limited applicability of
being added to only a restricted number of stems) and further become prone
to lexicalization, just like other derivational affixes (on such a grammatical
behavior of derivational affixes, especially causative ones, see also Bybee
(1985: 16-19)). Clearly, this 'usual defivational behavior' will differentiate the
final stage from the penultimate stage. Note that the diachronic model l',ere is
only an approximation of what may happen ir~ the functional take-over of the
causative function by the, AND and PURP elements and subsequent reduc-
tion of them into affixes; it is subject to further modification in the light of
more 'transitional languages', some examples of which I will review shortly. I
do not impute much theoretical importance to each of the stages of the
model, since for one thing there will surely be languages intermediate between
any two of the stages. These four stages are to be regarded only as some
guideposts; the stages mark off only conspicuous points (largely by impres-
sion) on the continuum from the originally noncausative syntactic construc-
du~,~ ~.v morphological causatives. Further, it should be borne in mind that
the model does not claim that aH languages that have either AND or PURP
type of causative in use must go through all these four stages; the model only
outlines how causative affixes m a y have arisen out of these metaterms, if
there is evidence that there is a diachronic relation between causative affixes
on the one hand and the metaterms on the other. Therefore, it is quite possible
that languages use the PURP type of causative without the metaterm PURP
taking over the function of [Vcause] and further being reduced to an affix.
J.J. Song / The rise of causative affixes 173

So far I have depicted the metaterms AND and PURP as having an equal
chance of changing into causative affixes. However, I should note that in my
database I could not find languages of the A N D type for any of the non-
initial diachronic stages (i.e., of course, except for the initial stage). I have
some suspicion that this is not accidental, e.g. due to lack of data. Although I
do not wish to completely rule out the possibility that the metaterm AND
ends up as a causative affix, I think that in comparison with the metaterm
PURP, the metaterm AND seems to be an extremely poor source of causative
affixes. There seem to be two complementary reasons for this state of affairs.
First, more often than not, the function of the metaterrr AND is carried out
by mere juxtaposition of two clauses (i.e. zero marking), rather than by some
explicit markers of AND. In contrast, the metaterm PURP is always marked
by some linguistic elements (i.e. non-zero marking). Secondly, there is a
semantic differential between the metaterm A N D and the metaterm PURP.
In other words, the metaterm AND is not semantically loaded in that its
presence is not a necessary condition for the interpretation of temporal
sequence of [Scause] and ~Seffect]: the linear order of the two clauses alone
iconically promotes such an interpretation. In fact, this is why it is the
metaterm AND, not the metaterm PURP, that can be realized as zero (see
example (14) above). Note that even non-zero markings of the metaterm
AND are almost void of any semantic content or even redundant owing to
the iconic interpretation of the linear order of the two clauses involved. On
the other hand, the metaterm PURP is semantically loaded in that it
specifically contributes the semantic content of some goal or purpose to the
construction that contains it. Wiu:out the presence of the metaterm PURP, it
is extremely difficult to obtain the meaning of goal or purpose. 15 Therefore, I
believe that semantic differential (aed/or difference in formal realization)
between AND and PURP plays a crucial role in the diachronic outcomes of
these two metaterms. I have already pointed out that non-zero markers of the
metaterm AND are also redundant and ,,~,M ~f semantic content. This aspect
of these non-zero markers of the metaterm A N D makes them very sus,=eptible
to complete loss, even before they are given a chance to compete with the
[Vcause] element, and to take over causative function from it. It is also quite
possible that, being void of semantic content, these markers may be 'squeezed

is One iaay argue that the metaterrfi A N D is not semantically loaded, because it is more often
realized as zero, and the metaterm P U R P is semantically loaded, because it is less often realized
as zero. I do not wish to involve myself in a potential egg-or-chicken debate, because it can be
argued that, for instance, the metaterm A N D is more often realized as zero, because it is not
semantically loaded. This is why I regard the two reasons here as complementary.
174 J.J. Song / The rise of causative affixes

out" into oblivion when the originally independent !exical [Vcause] element is
transformed into a derivational affix (i.e. when the originally biclausai A N D
type causative is formally reduced into a single clause). Whichever scenario
may turn out to be right, it is the semantic redundancy or lack of semantic
content of the metaterm AND that seems to facilitate its own demise.
Therefore, the functional take-over of the kind that I have depicted above
will not occur at all; rather the original [Vcause] element will retain its
causative function and further it will be formally reduced to an affix. In a
sense, the metaterm AND is no match for the lexical [Vcause] element as a
candidate for being a causative affix. The metaterm A N D is naturally a poor
source of causative affixes. In contrast, the metaterm PURP is a v c r y
competitive source of causative affixes: (a) it is very resistant to obliteration
due to its semantic loadedness; hence (b) it can stay around long enough to
take over the causative function of the metaterm [Vcause]. This preliminary
observation will be confi..rm.ed on the basis of cross-linguistic evidence in the
next section.
Further, the aspect of the metaterm PURl:' described in (a) - resistance to
obliteration - suggests a need to have another serious look at the diachronic
path of the original [Vcause] element being reduced to causative affixes fi la
Giv6n (1971a, b). Recall that I earlier posed a question about the fate of the
metaterms AND and PURP in the amalgamation of the. [Yeause] and
[Veffect] elements. I believe that thdr semantic differential has a lot to do
with their behaviors in the amalgamation process as well. If the P U R P
element is resistant to obliteration owing to its semantic Ioadedness, it is quite
likely that it may be trapped in between the [Vcause] and [Veffect] elements.
The metaterm PURP is more likely to survive the process of amalgamation of
~,he ~'eau~e] and [Veffect] elements. In contrast tc the resilient PURP, the
metat¢~m AND is more likely to be dropped or squeezed out in the process of
the amalgamation of [Vc~-use] and [Veffect] due to its semantic redundancy or
lack of semantic content. In section 5.2 I will cite a few languages as
providing evidence for the resistance of the metaterm PURP in the amalga-
mation process. Existence of such languages is not only important in confirm-
ing the differential characteristics of the metaterms PURP and AND (i.e. I
have not so far found a single language where the metaterm AND is trapped
between the [Vcause] and [Veffect] elements). But it also has far more
important implications for any synchronic theory of causative constructions.
In section 3 I have identified the existence of the PURP type of causative
construction, calling for a need to recognize it as a genuine causative
construction. • have a!ze mentioned that it is probably due to the western
J.J. Song / The rise of causative affixes 175

logic of causation (i.e. cause entails effect and the causative sentence must
have this entailment relatk3a~ that linguists have so far failed to recognize this
type of causative as a real causative. Evidence provided by such languages
which have PURP elements trapped in between the [Veause] and [Veffect]
elements in morphological causatives clearly strengthens my claim that the
PURP type of causative is indeed a genuine causative construction and
should be recognized as such in any synchronic theory of causative construc-
tions. If the PURP type construction (i.e. originally a purposive construction)
had not been exploited to express causation in these languages, there would
be no other way to account for the obfigatory presence of the metaterm
PURP in their morphological causatives.
The best way to demonstrate that the kind of diachronic path that I have
depicted so far is real in natural languages is to make reference to languages
which display each stage of the diachronic model. Since this is not possible at
present for the AND t y ~ of causative for the reasons mentioned, I will talk
about only PURP type languages.
Korean seems to have just begun to exploit a noncausative purposive
construction to express causation. In Korean there is a purposive element -ke
that marks subordinate purposive clauses as in (25):

(25) Korean
Kim ssi-ka ai-til-i korjpu-ha-ke pa~-es
Kim Mr.-NOM children-PL-NOM s~udy-do-PURP room-from
na = o-ass-ta
get = out-PST-IND
'Mr. Kim left the room so that the children would study.'

The same purposive marker appears between the [Vcause] element and the
[Veffeet] element in the so-called syntactic causative construction in Korean:

(26) Korean
Kim ssi-ka ai-fil-i korjpu-ha-ke ha-ass-ta
Kim Mr.-NOM child-PL-NOM study-do-PURP cause-PST-IND
'Mr. Kim made the children study.'

The presence of the [Vcause] element in cdu~etives like ~n~,.,,,in Korean is

obligatory. The causative sentence in (26) is non-implicative. Therefore, the
speaker is not committed to the truth of the em~.,edded clause in (26); it is
possible that the r,hildren did not study at all despit-, their father's attempt to
176 J.d. Song / The rise of causative affixes

make them do so. x~ In other words, (26) is still very much of a purposive
sentence. Note that in modem Korean the morphological causative suffix -l-
cannot be applied to verbs like kogpu-ha as in (26): in this case only the
PURP type ot causative construction is allowed. For further evidence that the
Korean syntactic causative is in the process of evolving from the purposive
construction, see Song (1988a).
Thai is a language d~at seems to be in the second stage ef the diachronic
model. According to Noss (1964: 177) there is a conjunction that marks
purposive subordinate clauses: hfiy ('~he following examples are from Vichit-
Vadakan (! 976)).

(27) Thai
kh~w khi~n c6tm~ay hfiy khun t6op
he write letter PURP you answer
'He wrote a letter so that you would answer.'

The conjur, cfion h@ also appears in the causative construction, where the
[Vcause] verb is tham 'do' or 'make'.

(28) TLai
Sa~a~.na~a tham h~y nisaa fi[ chin
Saka cause PURP Nisa hit l
'Saka made Nisa hit me."

However, there are also causatives without [Vcause] tham, but only with the
PURP marker h@.

(29) T/,a~
Sahkha'a hfiy d6k wi~
Saka PURP child run
'Saka %ad a child run.'
xo ~zvidence t!, .~yntactic causative sentenoes in Korean like (26) are nonqmplic~tive comes
from grammalic~i ~entences like the following:
Kim ssi-ka ai-t~l-i korjpu-ha-ke ha-oss6na ai-fil-i
Kim Mr.-NOM chiid-PL-NOM study-do-PURP cause-PST-but child-PL-NOM
koopu-ha-ci = en-~ss.ta
"*Mr. Kim wade his childre, study, but they did not.'
The ungramr-tica] English translation indicates tha~ the causative sentence in Engl,~sh is
J.J. Song / The rise of causative affixes 177

In Thai then, the PURP element can function optionally as the [-Vcause]
element on its own and it appears also as a genuine PURP element it: both
causative and noncausative constructions. This suggests that Thai seems to
have grammaticalized the noncausative construction as a causative construc-
tion to a greater extent than Korean.
.IjO, a Niger-Congo language, is also a language of the second stage: the
metaterm PURP is used in both causatwe and noncausative constructions
and at the same time the [Vcause] is optional, t7 While arguing that the word
order in Niger-Congo can be reconstructed as 2OV, Giv6n (1975: 95, 99)
points out that the causative suffix - t o o in .Ij9 can be traced back to a full
causative verb "make/came" fo!!owir,g the lower clause at some earlier stage
of its history, i do not accept Giv6n's claim that the causative suffix-mo can
be traced back to an independent lexical verb of [Vcause]. Instead, I argue
that the suffix - t o o is a PURP element, which sometimes appears with the
[Vcause] element in the PURP type of causative construction. Indeed, the
causative suffix -m.o is the same as the directional suffix -m.o marking nominal
arguments of directional roles.

(a) tob.o.u w.eni-m.6
child walk-DiR
'Walk towards the child.'
(b) ~iru-bi fiki tin kaka-m9
canoe-the take tree tie-DIR
'Tie the canoe to a tree."

The directional suffix - t o o is attached to what Williamson (1965:35) calls

transitive directional verbs and it has a meaning of 'towards' or 'in regard to'.
The causative suffix has exactly the same form as this directional suffix,
although neither Wiliiamson nor Giv6n makes this observation.

(31) .lj.o
(a) eri fir.u-bi bile-mo-mi
he canoe-the sank-CS-ASP
'He sank the canoe.'

~" However, unlike Thai, !j.o does not seem t o use the metaterm P U R P to mark purposive
subordinate clauses, bu; rather it uses the metaterm to mark the directional arguments /or
directional NPs), as illustrated below. Whether Ijo may use the P U R P element to mark the
subordinate clause is yet to be determined by further research.
178 J.,L Song / The rise of causative affixes

(b) w6ni uru aki-ni u-bgu-m6-m[

we wine take-ASP him-drink-CS-ASP
'We made him drink the wine.'

Clearly, the metaterm PURP is here being used in place of the [Vcause]
element in .Ijo. It can be said with certainty that the functional take-over of
[Vcause] by PURP has taken place. But has the take-over been completed in
this Niger-Congo language? That is, is .Ii9 a language of the penultimate stage
of the diachronic model? It seems that it is not yet. As both Williamson
(1965: 54-57) and Giv6n (1975-95) note, in certain cases a separate lexical
[-Vcause] element m/e can be used in conjunction with the so-called causative
SUffix -too.

(31) /j.o
~ e"i ~-',, ~ bi!e-mo-mi
he canoe-the sank-CS-ASP
'He sank the canoe.'

(32) ./Lo
(a) eri firu-bi mi.¢ biie-mo-mi
he canoe-the make sink-CS-ASP
'He made the canoe s~nk.'
(b) firfi.6 t9bg0 mi.e b'.unu-mg-mi
she child make sleep-CS-ASP
~She soothed the child to sleep."
(c) firfi.6 t9bp.fi b~nu-mo-mi
she child sleep-C$-ASP
~5he laid the child down to sleep.'

It seems that the metaterm PURP has not been cempletely institutic|nalized
as a genuir~e [Vcause] element, since the original [Vcause] element can appear
in its compan) as well.t8 Without the diachronic model of causative affixes
put forward in this paper, it :~ difficult to explain why a separate lexical
~8 Both Edith Bavin and Hilary Chappell pointed out to me that the causative with the
metaterm [Vcause] and the one without it are specialized for different semantic causative types,
i.e. indirect vs. direct. I acknowledge this kind of semantic specialization, but ] argue that prior to
the specialization, the optional deletion of the metaterm [Vcause] has to be accounted for. In
other words, the issue of the semantic specialization is a totally different matter from what I am
concerned with in the presenl investigation, i.e. ~he types, of causative cc,nstr-ct, ion ~ - ~ ~.he
functional take-over of [Vcause] by PURP.
d.J. Song / The rise of causative a~'xes 179

[Vcause] verb is used in addition to the causative suffix which is already

attaciled to the [Veffect] verb. In (32a) and (32b), the suffix -m.o is used to
register the presence of the metaterm PURP and it is not a genuine causative
suffix at all (as evidenced by the presence of mie, the [Vcause] element). On
the other hand, in (31a) and (32c) the suffix-too functions in lieu of the
[Vcause] element; hence the absence of the original [Vcause] element m ie. I
conclude that the so-called causative suffix -rn.o is not either fuily-fledged or
the old causative suffix in .ljg, pace Giv6n (1975: 99), who claim~ that the
causative suffix is a reduction of an old iexical causative verb; instead it is
still a genuine PURP element, since it still marks nomina~ directional
arguments as in (30), and it is in the process ot taking over the causative
function from the original [Vcause] element, since it can appear on its own
without the company of the [Vcause] element, as in (31a) and (32c).
The penultimate stage of the model in which the 0/cause] element is
permanently and obligatorily left out is exemplified by a causative construc-
tion from Yaqui, an Uto-Aztecan language (Lindenfeld (1973)). In this
language, the metaterm PURP functions both as a fully-fledged ~cause]
element and a purposive marker in noncausative constructions; the original
[Vcause] element has completely dropped out of use in the causative construc-
tion. Lindenfeid (1973: 103-104) notes that what she calls 'the command
markers' -sac and -~i?a are freel~ interchangeable particles that mark depen-
dent clauses of sentences that express an order:

(33) Yaqui
hu-ka ill usi-ta ne tehwa-k aman a wee-sae-kai
this-DEP little chBd-DEP I tell-STA there him go-PURP-DEP
'I told the little child to go there.'

These command markers or PURP markers in my theory function as if they

were genuine causative elements; no [Vcause] element can appear at all in
causative sentences:

(34) Yaqui
in kuuna baci-ta ho?ara-po nee hipu-?i2a
my husband corn-DEP ho~_,seoin I have-PURP
'My husband makes me keep corn in the house."

Lindenfeld (1973: 104, 105) points out that the higher verb, i.e. the [Vcause]
element, cannot occur in sentences like (34) and that there is no verb ~ear.4ng
180 J.J. Song / The rise of ca~ative affixes

'to cause' in Yaqui. One may argue that these command markers may have
been genuine causative suffixes all along, and that they may not have
originated from the metaterm PURP. However, evidence that this is r.ot true
comes from the fact that the subject of the lower clause (i.e. the causee NP) is
marked for dependency, which can be rega,ded as characteristic of subject-
hood in the embedded clause (LindenfekJ (1973: 65, 104)). In sum, the P U R P
element in Yaqui, while functioning as a genuine PURP marker in noncausa-
tive construct_ions, has completely taken over the function of the [Vcause]
element, which is now completely dead and extinct from the language.
As to the final stage of the model, the metaterm PURP now functions as a
fully-fledged [Vcause] element, but in morphologically reduced forms, i.e.
affixes morphologically bound to [Veffect] elements. This particular stage
then marks the rebirth of the metaterrn PURP as a defivational causative

5. Causative a~xes: A diachronic track-down

In this section | will present evidence bearing on the final stage of the
diachronic path that I hwe depicted above: causative affixes which may be
diachronically related to PURP elements. I wish to reiterate that the evidence
that I am presenting here is not conclusive by any means. What I do provide
is a general conceptual framework in which an attempt can be made to relate
,.~,,oo,:~,~,~,,,,,.,~,,,,,,om"'~oto what i! have described as instantiafions of the metaterm
PURP. Such a framework will do ei "or of two things: (a) relate causative
affixes to earlier sources ir~ languages for which no such attempt has so far
been made (the majority of the languages cited in this paper fall into this
ca~e~ory) or (b) reexamine hitherto claimed sources of causative affixes in
languages for which such an attempt has already been made (e.g. Ij9 in
section 4). Therefore, in v-'hat follows, ! do not claim that the causative affix
must be related to the ~etaterm PURP in a given language, but I suggest
that, given the general diac~ronic model, the causative affix m a y be related to
the metaterm PURP. One should not forget this important point.
It is n o w clear from the a gove discus.;ion of the diachronic model that it is
the m~taterm PURP (more accurately instantiations of PURP, as iiiustrated
earlier) in addition to [Vcause] that proves to be an excellent source of
causative affixes. The other metaterm AI~TD does not seem to L~. such a good
source for the reasons explained in section 4. Whether there are indeed any
languages, the causative affixes of which are historicaiiy related to the
J.J. Song i The rise of causative affixes 18 !

metaterm AND, can be determined only by further research. I will, therefore,

concentrate on those languages in which various instantiations of PURP may
have become petrified as causative affixes.

5.1. From P U R P markers to causative a ~ : , ~

In section 3, I looked at some languages whose causative construclions

contain the metaterm. PURP. This me*~erm is schematic of various instantia-
tions, such as dative, locative, allative: directional, goal, and benefactive
markers. Some of these markers express some sense of purpose or goal
through metaphorical extension; all register the presence of the metaterm
PURP. On the basis of the diachronic model depicted in section 4, I then
expect varying degrees of formal resemblance between these mm'kers and
causative affixes. For instance, it is no longer a theoretical problem to
account for the sceming!y disparate observations made by Wo!fenden and
Tuggy amov.g others (see section 2): in some Tibeto-Burman language~
causative suffixes havz a o,,~k:r~g resemblance to directional suffixes, in
Aztecan languages, the causative suffixes resemble the applicative ~u~xes, an
important pzrt o" the meaning of which is benefactive. It can be inferred that
in these lauguages either the directional or the ,~,~,,,,,.---"-~+:.......
~,. ~,,.~",,..~.=:, ~ m e
the causative suffix proper after its service as PURP in the causative
constru,:tioa, and its subsequent take-over of the function of [Vcause]. I will
cite below more languages like the Tibeto-Burman and Aztecan languages,
thus adding further to the observation already made by Wolfenden and
Tuggy. I believe that a uniform account of the formal resembk~nce bc.tweea
the causative affix and various instantiations of PURP has been long
overdue. I hope that the diachronic model in section 4 and what follows will
do it full justice. By that, I redan that the model will enable language
specialists (especially historical linguists) to check whether causative affixes
~.an be historically related tc~ PURP, c~r that it wiii enable them to reexamine
the hitherto claimed sources of caus~tive affixes in the light of the general
conceptual framework provided by this paper.
In Lama~g, the Chadic language spoken in northeastern Nigeria and
northern Cameroon, there is a causative suffix in use~ - ~ : this causative
suffix was originally restricted to motion verbs, but in Modern Lamang i~:
applies to at least one nonmotion verb (Wolff (1983: 123-124)). Wolff"
(1983: 105~ !24) suggests that this rather restrictive causative suffix is diachro-
~icaily related to the benefactive preposition ¢g& (Wolff (1983: 243). As
evidence for this relationship, Wolff (1983: 124) refers to the fact that the
182 J.J. Song / The ri.~e of causative affixes

causative suffix -q~ has a variant in the form of-~g~ in the imperative. In
addition, this language has a morphological process that derives causative
verbs from intransitive verbs by the addition of the causative suffix -Or'
(' {ff (1983: 114)). This particular suffix is also used to (a) convey the idea
f ~ ' cing the object which is affected by the action of the verb; (b) convey
t!{e I.~ea of doweward rnoveme~'~t. Wolff (1983: ll4) notes that there is some
kind of locative-directional notion attached to the two other functions carried
out by the su~x -Ov': the locative-directional meanirlg is indicative of one of
its original meanings.
K6hler (1981:507) reports that in Kxoe there is a causative suffix in the
form of k3. Again, the directional preposition in this Central Khoisan
language is none other than k~. Indeed, Heine and Reh (1984: 137) think that
in Kxoe the causative is derived from the directional preposition.
Pomo (Southeastern) is an American Hokan language in which the causa-
tive suffix -q- can be used i~~ lieu of a separate iexical causative verb ?xe
(Moshinsky (1974: 74)). There are two vertical directional morphemes which
contain -q-: -qla- ~downward' and -qlo-, -ql- 'upwards, uphill, up off the
ground' (Moshinsky (1974: 57, 58)). It is not implausible to infer that the
same form -q- is shared by ~he causative suffix and directional morphemes.
This inference is further bolstered by the fact that many forms containing -q-
are ambiguous between causatives and direction-related meanings (Mosh-
insky (1974: 56)):

(35) Porno
(a) ciqat: 'carry a lot of things away from here' or ~hand someone a
bowl or a glass full'
(b) 7t&~at: 'take two away flora here' or "give two to someone'

The morpheme -q- has two functions in (35): causative and direction to
somewhere away from the speaker. Again, Pomo points to a possible
common semantic denominator between causative and direction.
So,,~thern Agaw, a Cushitic language, is reported to have a causative suffix
°s- in addition to -c-, which is ~he more established causative suffix par
excellence (Hetzron (1969:61-63)). According to Hetzron, the causative
suffix -s- must have been the causative suffix in Southern Agaw, as it stillis in
other Cushit]c languages. And the dative (-instrumental) suffix in Southern
Agaw is-s-(Hetzron (1969: 21)).
Khasi, a Men-Khmer language, is also interesting in that the same prefix
pyn- functions as both causative and benefactive (note that .#he verb carries
Y.J, Song / The rise of ca--retire affixes 183

the benefactive prefix: in Relational Grammar, this is known as advance-

ment) (Henderson (1976: 501-502))"

(36) Khasi
(a) Causative
iap 'm die': oyniap 'to kill'
long 'to be': oynlong 'to create'
(b) Benefactive
kren 'to speak': pynkren 'to speak for another'
repair (loan from English): pynrepair 'to repair for someone'

It seems that this causative prefix is very productive to the extent that it
applies to recent loanwords such as repair, as Henderson notes (1976: 501),
although in a n e a r l i e r d e ~ e r i n t i c m n f the_ l a n e ~ m o a R a h ~ l ( I q g l • 1fig~ P l ~ i m ¢ .
that it 'plays a decidedly minor role in Khasi'.
Southern Sierra Miwok, an Amerindian language spoken in centra! Cali-
fornia, is another language in which a close cognitive relation between
causative and benefactive functions is formally apparent (Broadbent (1964)).
,There is a set of causative suffi~es in this Miwok language~, two of which
concern us here: -na and -nY. The first svgfix -ha has two functions: ~nefactive
(marked on the verb as well) and causative (Broadbent (|964: 74-75)):

(37) Southern Sierra Miwok

(a) Causative
takp- 'to thirst': takypna,- 'to make one thirsty'
cyt?yt-.e- "to like it': cyte.na- 'to make someone like it'
(b) Ber~efactive
?enh- 'to make': ?enyhna- 'to make tbr someone'
Kose.-n x 'to cook': kosenna t u ~ u u a i u l s o m e o n e '
myl. i- 'to sing': myl. ina- 'to sing for someone'

A piece c, ~vidence for a common semantic thread between th~ causative and
bcnefactive suffixes comes from the fact that the variants of present (ira)per-
fect and imperative modal s~ffixes following either the causative suffix or ~he
benefactive suffix are exactly the same: the present (ira)perfect has -na- ov
-nak-, and the imperative has -ni-, -no, or-X- (Broadbent (1964: 74--75)). The
second suffix in questioe - n Y also has a causative or benefactive meaning
(Broadbent (!964: 76))"
184 J.J. Song / 7"1'erise of causative affixes

(31g) Southern Sierra Miwok

(a) Causative
7yw.y- 'to eat': ?ywy,nY-'to feed'
(b) Benefactive
kala. -rj- ~to dance': kalar.~nY- 'to dance for'
liw.a- 'to talk': liwa.aY- 'to talk for'

Swahali, a Bantu language, has a causative suffix -ish-, which 3plalies to

both intransitive and transitive verbs (Ashton (~947), Comrie (1976a), Dric
vet (1976), Vitale (1981)). The following is taken from Driever (1976: 43). x9

(39) Swahili
baba a-li-m-fung-ish-a mtoto mlango
father SUB-PST-OBJ-ciose-CS-IND child door
'The father made the child close the Ooor.'

According to Dfiever (1976: 130), some of her informants who accepted it

with sligh~ reluctance interpreted (39) as the applicative: Tildefather closed the
door for the child. I take this as a piece of evidence that the causative stfffix
-ish- may taave been the metaterm PURP. In other words, the Swahili
causative su~x may have originated from the applicative suffix which is by
now ah'nost dead, but for some~ speakers it still retains ils original function.
Of course, ~hiz: is only speculative at the present stage of the research.
Incidental!' $wahili has a separate applicative suffix:

(40) Swahili
Hasar~i a-I~-m-bomo-le-a All ukuta
tq,-:san SUS-PST-OBJ-pull = down-APPL--IND All wall
'Hasan pulled down the wall for AlL'

As I hawe shown in section 3, the metaterm PURP is realized also as such

verbal markers as future tense, subjunctive mood, irrealis, and even incomple-
rive a,,~pect. All these markers share the essential p r o ~ r t y of non-factuaiity.
The diacl'~roi~k model of causative affixes also implies that these markers also
may end up as causative affixes. The model then provides a uniform
diachronic conceptual framework in which causative affixes can be traced
back to their origins.
~-~ I have changed ~;ome of Dr/ever's ,,;iosses in accordance with the con,ention that i used for
the earlier Swahi~i examples in section 3.
,LJ. Song / The rise of causative affixes 185

In Abkhaz, a Caucasian language, there is the causative prefix r-, which

applies te~ both intransitive and transitive verbs (Hewitt (1979: 170-171)):

(4 i) Abkhaz
(a) d-g~io-yt'
'He is standing, up.'
(b) d~-s~-r-g~lo-yt"
hi~ i-CP-stand-FIN
'I stand him up.'
(c) y~-q'a-s-~z'+-yt"
'I did it.'
(d) y¢-b-~-r-q'a-e'e-yt'
'I made you do it.'

The purposive clause is constructed by adding either -re, -raza, or -rana

(Hewitt (1979: 42)):

(42) Abkhaz
s-y%za d~-z-.b~-rc (or -r~$, rano) ~-kalak' (a-) ax's-co-yt'
my-friend him-I-see-PURP ART-town it-to-I-go-FIN
'I am going to town to see my friend."

It is noteworthy that the c,,mmor~ part of these three purposive markers is the
same as the ~ausative prefix. In fact, Hewitt (1979: 199-201) argues on the
basi.,~ of its co-occurrence restrictions of cross-referencing prefixes that the
con~aon form r in the pu~l~osive rnarker~ iis further related to the non-finite
future tense. At any rate, it demands an explanation why the ~ame form
appea~,s in both causatives and purposive cnauses.
In Wiyot, an Amerind:an language (Almosan-Keresiouan family), the
causative suffix -iy is attached to ~ntr~nsitive verbs with what Teeter
(1964:~'7 ~ ca|Is ~i-stems' to derive causative verbs:

(43) Wiyot
lag- 'go': lagiy 'cause to go'

Interestingly enough, but not unexpectedly, one of the subjunctive suffixes in

186 J.J. Song / The rise of causative affixes

Wiyot is none other than -(v. Evidence for the common origin of these two
suffixes comes from the fact that the subjunctive suffix -iy applies also to the
verbs of i-stem (Teeter (1964: 58)).
In Luisefio there is a set of causative suffixes (Kroeber and Grace (1960:
46-48)). One of the causative suffixes in this Uto-Azteean language is -pi, (or
occasionally -pa) (Kroeber and Grace (i 960: ! 40)):

(44) Luisego
(a) qar~pax 'to fall': qar:~pi 'to make to fall'
(b) laq~.pax 'to be smooth': laq:~pi 'to smoothen'

And the same form appears as a suffix marking the purposive clause
(Kroeber and Grace (!960: 147)):

(45) Luiseao
hungiax ya?a.ci po-rje-pi
'showed the man that he might leave.'

It is an unmistakable fact that the same form is used in t~oth causative and
purposive functions.
Now I will survey three large language groups in a summary fashion:
Amerindian, Australian and Austronesian. Since the phenomenon that 1 have
observed and accounted for is quite widespread in these three groups, for the
majority of these languages I will simply note that the phenomenon exists.
Earlier I noted Tuggy's work in Uto-Aztecan languages where the causa-
tive suffx and the applicative suffix share the same forms. However, my
izavestigatio~ shows that the phenomeon is not restricted to Uto-Aztecan
languages: other Amerindian languages also share the same tbrms for both
functions. I will thus survey these ~anguages under a single heading: Amerin-
In Classical Nahuatl, there are two main w~ys of morphologically deriving
causative verbs" (a) one involves vowel contrast and is restricted to intransi-
tive verbs; t~'J'~'~'~"~,,-other regular causativizing process involves the suffixing of
-tia, which applies to both intransitive and transitive verbs, and -lia and -huia,
which are generally restricted to intransitive verbs (Andrews (1975: 85-101)).
As Tuggy (1987) also notes, Classical Nahuatl (Aztecan) is no exception to
the phenomenon characteristic of the Aztecan languages: the same forms are
used for both causative and applicative functions. Indeed, in Classical
Nahuatl, there are three applicative suffixes: -ia, -lia and -huia. Note that the
J.J. Song / The rise of causative affixes 187

last two are the more frequent (Andrews (1975: 102-103)). The applicative
suffixes register in the verbal complex the semantic role of dative, directional,
benefactive t,,ter alia. Again, it is an unmistakable fact that virtually the same
forms are used Ior both ~ nctions.
Other Aztecan or Uto-Aztecan languages that behave like Classical
Nahuatl are Michoacan Nahuati (Sischo (1979: 354-355)), Huasteca Nahuatl
(Belier and Belier (1919: 281-282)), North Puebla Nahuatl (Brockway
(1979: 177)), Tetelcingo Nahuatl (Tuggy (1979: 105-108, 1987)), Western
Tarahumara (Burgess (1984: 107)), and Luisefio (Kroeber and Grace
(1960: 46, 144)).
Jacaltec, a Mayan language, also displays the phenomenon common in the
Aztecan languages. According to Day (1973: 44), the same forms, -~e, -tze
and -re, are used for both causative and applicative functions.
Maidu, a Penutiaa language, us~:~ the same f o ~ -ti for causative and
benefactive (or applicative?) functions (Shipley (1964:40, 44)).
Now I will move to the Australian family. (Cf. Dixon (1980) and Blake
(1987) for a general introduction to these languages.) Before I look at some of
the Australian languages, I urge the reader to, recall from section 3 the PURP
type causative construction in Basque: the [Seffect] nominalized by the suffix
-to is marked by the metaterm PURP, i.e. the allative marker -ra. In the
process of reduction of the [Vcause] element (originally a PURP element in
our case), the nominalizer may be trapped in between the [Vcause] and
~effect] elements. Some of the Australian languages that I will survey below
support this line of diachronic reasoning.
Gumbaynggir has the causative suffix -ygura, which derives causative verbs
from intransitive verbs (Eades (1979: 304)). Eades entertains two possible
analyses for this causative suffix: (a) y g u (purposive verbal marker that can
be traced back to dative/allative suffix) plus ra; (b) the present tense suffix y
and guru. I will prefer the first analysis to the second one, because the present
tense does not contribute to non-factuality, an essential property of some of
the PURP elements (see section 3), and because the second analysis also has
to account for the form ra. just as the first one does. At any rate, in
Gumbaynggir, the causative suffix is unmistakably related to the,' dative/
~llative suffix -gu.
In Pitta-Pitta, the causative suffix -la is used to form causative verb: from
intransitive ones (Blake (1979: 204)). The same suffix is used on the verb to
mark the semantic role of beneficiary (Blake (1979: 205)).
In Ngandi, there are two allomorphs of the 6ausative suffix: -guba and
-n?guba (Heath (1978: 9i-92)). The purposive case marker and dative case
188 J.J. Song / The rise of causative affixes

marker in Ngandi is -ku. It can be thus conjectured that the causative suffix
contains the cognate of the dative/purposive case marker. If this is the case,
the question to be answered is: What is the remaining bit -nT?. Heath suggests
that this may be related to the now obsolete nominalizing suffix -n? (Heath
(i978: 91,127)). Ngandi thus provides an example where the nominalizing
marker is trapped in the take-over and the subsequent reduction process of
the metaterm PURP.
Nunggubuyu is another Australian language with a similar causative suffix
development to Ngandi: part of the causative suffix -jga/-/gi, -j-, is possibly the
archaic nominalizing suffix (Heath (1984: 393)). Note that the remaining form of the
causative suffix can be traced back to the dafive-gu/-ku, one of the most widespread
morphemes in r~u~R,d,a~i
. . . . . . . ~ - ',~n~u,~s
. . . . . . . . . . (Blake ~,1~,87' 35)) via some vowel changes.
Walmatjari requires the causative suffix -kuji to co-occur with the nom-
inalizing suffix -u (Hudson (1978: 48)). Agair., note the formal resemblance of
t. h. .e. . .t-.qn~ati~,r~ •.,,.., t h e ,~,~
. . . . . . . . . . . . eufKv tr~ ,-,.,/u~ll~.,la.,~uv~/i-~UlVU~iV~:
A,~*~,,~,,iKo,,.,o¢~,.,*:,,,~I . . . . . . .
c a s e -".,~l Kt::r
: . . . . . . '--
In Yidiny, the causative suffix -~a-I is exactly the same form used on the
verb to reg,.'ster the dative/locative semantic roles (Dixon (1977:305--318)). 2°
The last language family that I will survey is Austronesiar~. It is an
unmistakable fact about the Austronesian family that the causative affix is
formally alike or similar to the affix of benefactive or directional role that
appears on the verb.
In Indonesian, the causati,,e suffix -kan converts both intransitive and
transitive verbs into causative ones: the same suffix appears on the verb to
add ioeative/benefactive meaning (MacDonald (1976: 54--55), Tampuboion
(1983: 44-51 )).
Javanese is, not surprisingly, similar to Indonesian in that the same form
-(q)ak~ is used for purposes of deriving causative verbs and registering
benefactive and directional roles (Horne (1961:207-208)); the form -(q)ake

zo Dixon (1977: 128) notes that Yidiny is unusual in that it haas ergative -~gu and locative -ta,
when almost all Australian languages show exact correspondence between locative and ergative
allomorphs, the only difference bring that locative ends in -a and ergative in -u and when the
frequent pat~tern is:
ergative locative
onto vowePfinal disyllabic stems -0gu -rjga
onto vowel-final trisyllabic stems -lu -la
Dixon further notes that in Yidiny lhere is no surface ~race at all of locative -~jga. However, it is
possible that the seemingly disappeared locative -~ga may have its trace in the locative/dative
verbal mark~ .,m-!, ",~ifich i~ a~so a causa ive suffix, althougt~ I have to admit that 1he !oeative
NP governed by this marker should refer to only some language or speech style (Dixon
(1977: 306)). i owe this ob~ervati,~n on Yidiny to Rarry Blake.
J.J. Song / The rise of causative a~xes 189

is a Ngoko (informal style) form, the counterpart in Krama (formal style) is

-(q)aken (Home (1961 • 228)).
Winstedt (1914/1957.89-90, 98-99) reports that Malay has the causative
prefix p ~ and the causative suffix -kan. Winstedt (1914/1957: 98) points ou~
that the causative suffix is related to the directional preposition akan.
Walker (1976: 30-31) reports that in Lampung the causative suffix -ko is
used to form causative stems and it is used also to give the basic stems
benefactive meaning.
Sundanese (Ha~djadibrata (1985" 18-20)), Melayu Betawi (lkranagara
(1975: 175, 179)), and Balinese (Barber (1977: 156-157, 159, 164)) are similar
to the above Austronesian languages in that the same form is used for
causative and benefactive functions.
The languages to follow all have reflexes of the Proto-Oceanic causative
prefix *paka- (Pawley (1972), Bradshaw (1979), Schfitz (1986)). Bikol has the
causative -p-r c- t"t x. . .p.a - ~lvtlntz
"~'-". . . . .~,~,t
. " . .~. . .!.o:~)1.
. . . . ! ne same prefix is used to indic~te
direction in Bikol (Mintz (1971: 180-181)). Maori is another language which
~ems to have a reflex of *paka- as the causative prefix whaka-. According to
Biggs (1969: 83), the same prefix is attached to locatives and adds the
meaning of 'in the direction of'. In Tagalog, the causative prefix that De
Guzman (1978" 336) claims is the most productive of all dcrivational affixes
n Tagalog is none other than pa-. According to Blake (1925: 264-265), the
prefix is used also to derive verbs of rn~::ion from locative nouns, etc.:
directional function. Other languages that belong to this group are CeL'uano
(Wolff (1966: 479; 1967: 261)), Haroi (Goscnnick (1977:110-111)), Hilig,~v-
non (Wolfenden (1975" 102-142), Motus (I971" 286-287)), Kaliai-Kove
(Counts (1969: 68, 76--77, 11 l)), Marshailese (Bender (1969: l 11, 112), Zewen
(1977: 60--61)) (these languages use the same forms for both causative and
directional functions), and Casiguran Dumagat (Headland and Healey
(1974: 32-37, 39)) (this particular language uses the same prefix pa- for both
causative and benefactive functions).
So far I have surveyed a number of languages that display striking
resemblances between causative affixes and PURP markers. These resem-
blances have gone largely unnoticed or even ignored, nor has there been any
serious attempt to establish some kind of diachronic relationship between
them. I have made here such an attempt to tie the loose ends around the
formal resemblance between fhe causative affix and the PURP marker. This
attempt has become possible on the basis of the typology of causative
constructions and the related diachronic model of'causative affixes developed
in this paper.
190 J.J. Song / The rise of causative affixes

5.2. Some irregular cas~s: Trapped PUP, P

I have come across some languages in which there are some additional
affixes bound to the morphological combinations of [Vcause] and [reflect}:
co-occurrence restrictions of causative affixes with some seemingly unproduc-
tive elements. ~n such languages, certain affixes accompany causative affixes
in the morphological defvation of causative verbs; there do not seem to be
any obvious reasons why they have to co-occur with causative affixes without
making specific semantic contributions to the morphological derivation in
question. The fact that it has not yet been discussed at all in the literature
may illustrate well how elusive this phenomenon is. |n this section | will
account for such 'deviations' in the light of the diachronic model of causative
affixes. In sectiov 3, I have argued that the metaterm PURP is resistant to
obliteration (or reducuon for that matter) owing to its semantic loaded~ess
and/or to the fact that it is ~rw~ys realized as non-zero elements: there is
............ of its being ",,. .-~r,
much ~t.~ur.,,,,n . . veu"i in the reduction process of the [Vcause]
element in the PURP type of causative construction. Conseque~tly, the
sequence of the metaterm PURP and the [Vcause] element as a wno|e is
reanalyzed as a causative a;fix in the overall reduction process:

(46) ... [Veffect]-PUiRP-[Vcause] •••


... [Veffect]-PURP-v!ffix [Vcause] vaffi× ...


... ~effect]-vaffix [PURP-~'cause]] vaffix ...

Then the additional affix bound to the morphological combination of the

[Vcause] and ~4effect] elements may be the metaterm PURP trapped in the
reduction process. So it becomes worthwhile to examine languages in which
some unproductive affixes are obligatorily required in their morphological
derivations of causatives. I will show below that such additional affixes are
indeed instantiations of the metaterm PURP trapped in the reduction process
of the [Vcause] element A la Givbn. 21
z~ As shown earlier in section 5.1. nominalizers may be also :rapped between the IVcausel and
[reflect} elements.
J.J. Song / The rise of causative affixes 191

~'inally, I emphasize that the existence of such languages adds very strong
support to the reality of what I have characterized as the PURP type of
causative construction in section 3. Recall that this particular type of causa-
tive has not at all been recognized as a causative construction par excellence.
The evolutional path of this irregular kind of causative (i.e. from the syntactic
PURP type of causative construction to a morphological causative) can be
retrieved with reasonable certainty from examination of the reanalysis of the
sequence of PURP and [Vcause] as a causative affix. 2~
In Bashkir, a Turkic language, the contiguity of the [Veffect] element (i.e.
the basic verb) and one of the causative suffixes is broken up by the
additional suffix -G)I'. Poppe (1964: 70) notes that the causative suffix -GA'r
contains one of the causative suffixes -r and that the remaining form -G~f is
an unproductive suffix. However, it seems to me that this unproductive suffix
is an instantiation of the metaterm PURP trapped in between the [Vcause]
and [Veffecq elements: in Bashkir the d~.tive suffix marking ,qr~=tion, goal
etc., is none other than -G ~1"(Poppe (1964: 36)).
In Songish, a dialect of Straits Salish spoken in British Columbia, Canada,
a similar thing can be found. In this language, there is a causative suffix -txw
(Ratio (1972: 151)). And the purposive element (or rJURP) -(a)s appears in
between the basic verb and the causative suffix:

(47) Straits Salish

(a) l~w~n-s-txw tu? nil
'Cause him to see."
(b) ?~le? it~ ?u? xifi-s-txw
to:be + here IPL ASP feed-PURP-CS
We cause (someone) to feed them.'

In Kanud, a Saharan language spoken in northeastern Nigeria, what Hutcl-i-

son (1976:43) calls the +ngin class verbs must be followed by the applied
morpheme -k when they are m,.~sphologically causativized by the prefix y#b:

(48) Kanuri
(a) ffi-ngin
'I wake up.'
(b) yitr-ffi-#-kin
'I wake up someone.'
Again, the actual/surface sequence of PURP and [Vcause] is language particular.
192 J.J. Song / The rise o} causative a2t~xes

Note that the aHomorph gk of the applied morpheme follows the causative
prefix in (48b). According to Hutchison t1976: 42), this applied morpheme
functions as an indicator of dative, benefactive, directional semantic roles.
Again, the applied morpheme in Kanuri is an instantiation of the metaterm
PURP: the applied morpheme that appears in Kanuri causative vert~s ~s a
fossilized PURP. z3
Bandjalang is an Australian language that adds support to my diachronic
m,~.de| of caus~.tive affixes. According to Crowley (1978: 88), this language
has about fifteen irregular verbs. When the ca,sa~.i--,e suffix -ma is added to
some twelve of these irregular verbs, the roots should be in their purposive
forms (italicized in the following examples) (Crowley (1978: 88)):

(49) Bandjalang
(a) dung 'cry': dung-bi.'B-ma "cause to cry'
.yuai-o,~.jff'mun~ h, al!.l.i~.. Ibg,r ~ga¢

(c) bala: n 'die': b a h :~-bin-ma 'cause to die'

In the light of the tendency that irregular verbs are remnants of the old
language system (cf. English irregular verbs, which are carried over from its
old verb conjugation system; see Crowley (1978: 99-101) for arguments for
the irregular verbs as :~art of the older Bandjalang language), the fact that
only irregular verbs have to combine the purposive su~x with the ~deffect]
and [Vcause] elements ~,uggests that (a) in its earlier history Bandjalang had a
PURP type of causative construction; (b) t~e language has now lost the
construction; and (c) p~rt of the construction is tbssilized or petrified in some
twelve causativized irregular verbs. [n Bandja~ang, the purposive suffix is
trapped in t}~e t ~ e caps~Jle of the PURP type of causative construction.
Finally, | find it appropriate to end this rather long "diachronic journey' by
going back to the q~estion that evidently puzzled Wolfenden six decaoes ago.
I noted in section 2 that Wolfenden (1929: 139, 152) was not certain whether
-tsa- in the Ao causative affix ddk-tsa- is related to the benefactive element.
Now it can be seen that this is indeed the benefactive element that was used
:o serve as a PUP P clement, which is now reanalyzed as part of the causative
amx prope~.
The metate~m PURP ~s likely to be caught in ~,e time capsule of the PUR~-

z.~ Of course, this e~ample from Kanuri is not e×actly a case o f a trapped PURP, but it
ilJustrates how the metz:terrn can be encapsulated along with the [ V c a ~ ] element in morphologi-
cal causative verbs: in effect, Lhe causative prehx and the P U R P element work as a causative
cireumfix in Kanuri.
J.J. Song / The rise of causative affixes 193

t y ~ causative construction: it may become fossilized during the amalgama-

tion process of the [Vcause] and [Veffect] elements. And the diachronic model
of causative affixes put forward here has now dug up the time capsule and
'detrapped' the fossilized form in a reveali~,g way. No serious mention of
these 'deviations' has been made in the literature, nor has any diachronic
model attempted to unravel the mystery of the P U R P element trapped in
between the [Veffect] element and the causative affix. Further, ~.he 'trapping'
of the metaterm PURP in the reduction process of [Vcause] and [Veffect]
clearly demonstrates the need to recognize the hitherto ignored type of
causative construction, namely the PURP type.

6. Conclusion

In this paper, I have proposed a universal typology of causative construc-

tions. Further, I have put forward a rather powerful diachronic modeI of
causative affixes on the basis of the universal-typological perspective assumed
in the typology. This diachronic model makes a very strong claim that it can
delimit possible sources of causative affixes only to [Vcause], PUP P, and
AND, although I have shown that the metaterm A N D may not be as
competitive a source for causative affixes as the other two. The model also
provides a plausible explanation for erstwhi!: mysterious :rid elusive co-
occurrence restrictions of causative affixes with PURP elements found in
some languages (e.~, i,~ Bana~atang).
1 am fully aware that what I have claimed so far should be verified in each
of the languages discussed in this paper (and in as many languages as possible
in the future) on the basis of historical evidence, where this ig available. And I
have said repeatedly that what the present paper proposes is a general
conceptual framework in which possible origins of causative affixes can be
identified, and hitherto claimed origins of causative affixes can be re-exam-
ined. The paper does not at all have any pretense in laying claim to the actual
histories of the languages that it ha~ discussed. The kind of diachronic model
as presented in this paper should be, most welcome to diachronic linguistics,
since the majority of the world's languages lack ~ c i e n t historical data
2!ong tbe lines aemznded by Lightfoot (1979), and since verification in each
language of the world is evidently impossible. The diachronic model of
c.~usative affixes proposed here at least provides a ~'~.l~ifia~le conceptual
framework oI ~anguage change and certainly forms a very strong foundation
on which the universal typology of causative constructions can be rigorously
194 J.J. Song / The rise of causative affixes

tested, and at worst it sets the agenda for future diachronic research in
individual languages.
Finally, I should point out that the diachronic model developed in this
paper deals with various instantiations of the metaterm P U R P and A N D
used in lieu of or as causative affizes, be they in full or reduced forms. In
other words, my model indicates the direction of change from PURP or
A N D to causative affixes; it is not concerned with the extended direction of
change: from causative affixes to noncausative markers. I have in mind Bantu
languages, in which causative affixes seem to be extended to cover instrumen-
tal meaning. In Kinyarwanda the causative suffix -iis/:- is also used to mark
instrumental case in the verbal complex (Kimenyi (1980: 32-33, |64--172)).
Evidence that these sutfixes are conceptually related comes from the fact that
the instrument~ suffix undergoes the same allomorphic changes as the
causative suffix ;Kimenyi (1980: 238)). In Nkore-Kiga there is the causative
auffix osa (Ta~'Ic,~
• . . . . . ;48))° The
~,ro., _ same fo__rrn_is again used as an ;,~t,-,,,-,,,,,~.
tal verbal suffix (Tayior (1985: 98-99)). Compare:

(a) ~-aa-mu-bo-orek[yes]a amaino

LTOP-him-them-show-CS teeth
'I made him show them his teeth.'
(b) y-aa-gi-hindu[z]a enkoni
he-TOP-it-turn-CS stick
'He turned it round with a stick.'

The semantic extension of the causative suffix to the instrumental case

marker ira the Bantu languages can be accounted for if it is assumed that the
instrument (e.g. s t c L in (50b)) is anth-omorphized as a human intermediary,
just like a human eausee argument: an instrurfient is regarded as an
intermediary agent whom the causer uses to bring about his/her "desired
effect. The same phenomenon can be observed in some Australian languages
(Blake (1987:69)): the causative affix can be used also to encode the
instrumental role. This kind of extension has nothing to do with the causa-
tion per se: the diachronic model of causative affixes is not concerned with
such a second order or subsequent extension of c~:usative affixes. It shews
that the sources of causative affixes are: [Vcause], P U R P and potentially
J.J. Song / The rise of causative a~xes 195


Aissen~ Judith, 1987. Tzotzil clause structure. Dordrecht: Reidel.

Anderson, John, 1971. The grammar of c~se: Towards a l~2a!ist theory. London: Cambridge
University Press.
Andrews, Jam~ R , 1975. Introduction to Classical Nahuatl. Austin, TX" University of Texas
Ashton, E.O., 1947. Swahili grammar. London: Longman.
Barber, Charles G., 1977..A grammar o~' the Baiinese ianguage. Ab~rdeet~: Uaivcrsi~y of
Bell, Alan, 1978. Language sample. |n: J.H. Greenberg et al. (eds.), 123-156.
Belier, Richard and Patricia Belier, 1979. Huasteca N,~huatl. in: R.W. Langacker (ed.), 199-306.
Bender, Byron W., I969. Spoken Mershallese. Honolulu, HI: University of H~waii Press.
Biggs, Bruce, 1969. Let's learn Maori. Wellington: A.H.&A.W. Reed.
Blake, Barry J., 1979. Pitta-PRta. In: R.M.W. Dixon ~cl B.J. Blake (¢ds.), i82-242.
Blake, Barry J., 1987. Australian Aboriginal ~ammar. London: Croom Helm.
Blake, Frank R., 1925/1967. A grammar of the Tagfilog language. New York: American Oriental
Bradshaw, Joel, 1979. Serial causative constructions a~d word order change in Papua New
Guinea. Working Papers in Linguistics l I (2), ! 3-34.
Broadbent, Sylvia M., 1964. The Southern Sierra Miwok language. Berkeley, CA: University of
Califoruia Press.
Brockway, Earl, 1979. North Puebla Nahnatl. In: R.W. Langacker (ed.)o 141-198.
Burgess, Don, 1984. Western Tarahumara. in: R.W. Langaeker (ed.), 1-149.
Bybee, Joan L., 1985. Morphology: A study of the relation between meaning and forr~.
Amsterdam: Benjamins.
Bybee, Joan L., 1988. The diaehronic dimension in explanation. In: J.A. Hawkins (ed.), 350-379.
Comrie, Bernard, 1974. Causatives and universal grammar. Transactions of the Philological
Society, 1-32.
Comrie, Berna,d, |976a. The syntax c~fcausative censtruetions: Cross-language similarities and
#Avergvnces. In: M. Shib~_tani(ed.), 261-312.
Comrie, Bernard, 1976b. Review of Xolodovie (ed.), Tipologija kauza~ivnyx kons:rukcij. Lan-
guage 52, 479--488.
Comrie, Bernard, 1980. Morphology and word order reconstruction: Problems and prospects. In:
J. Fisiak (ed.), Historical morphology, 83-96. The Hague: Mouton.
Comtie, Bernard, 1981. Language universals and linguistic typ31o~y. O~ford: Biaekweii.
Counts, David R., 1969. A grammar of Kalial-Kove. Honol du~ HI: University of Hawaii Press.
Craig, Colette, 1977. The structure of Jaca|tec. Austin, TX: University ~f Texas Press.
Craig, Colette and Kenneth Hale, !988. Relational preverbs m some languages of the Americas:
Typological and historical perspectives. Language 64, 312-344.
Crowley, Terry, 1978. The middle Clarence dialects of Bandjalang. Canberra: Australian Institute
of Aboriginal Studies.
Davies, John, 1981. Kobon. Amsterdam: North-Holland.
Day, Christopher, 1973. Th~ JacaRec language. Bloomington, IN: Indiana University Press.
De Guzman, Videa P., 1978. Sy~,'tacticderivation of Tagalog verbs. Honolulu, H~: University of
Hawaii Press.
196 d.J. Song / The rise of causative affixes

Deibler, Ellis W., 1976. Semantic relationships of Gahuku verbs. Norman, OK: S~L/University
of Oklahoma.
Diehl, Lor~, 1975. Space case: Some principles and their implications concerning linear order in
natural languages. SIL Working Papers 19, 93-150.
Dimmendaal, Gerrit J., 1982. The two morphological verb classes in Nilotic. In: Rainer Vossen
and Marianne Bechhaus-Gerst (eds.), Nilotic studies, 269-309. Berlin: Dietrich Reimer Verlag.
Dimmendaal, Gerrit J., 1983. The Turkana language, DordrecM: Forts.
Dixon, Robert M.W., 1977. A grammar of Yidin,. C a.,,briuge:
~ "-' Cambridge University Press.

Dixon, Robert M.W., 1980. The languages of Australia. Cambridg_':: r~,.,,~,~,4g~ t l,; ..... --,.,
Dixon, R,3bert M.W. and Barry J. Blake (eds.), 1979. Handbook of Australian !anguag¢~.
Canberra: Australian National University.
Dfiever, Dorothea, 1976. Aspects or a case grammar of Mombasa Swahili. Hamburg: Helmut
Buske Verlag.

,_~es, Diana, 1979. Gumbaynggir. In: R.M.W. Di×on and B.L Blake (eds.L 244-361.
kOley, William A. and Robert D. Van Valin, 1984. Functional syntax and universal gramnmr.
Cambridge: Cambridge Ur, iversity Press.
Fri,~rich, Paul, 19)5. Proto-lndo-Euro~:ean syntax. Butte, MT: Montana College of Mineral
Science and Technology.
Ge~etti, Carol. !9~6. The developme;nt of subordinators from postpositions in Bodic languages.
BLS 12, 387--4(;0.
Giv6n, Talmy, 1971a. Historical sy~itax and synchronic m~rphology: An archaeologist's field
trip. CLS 7, 394--415.
Giv6n, Talmy, 1971b. Some historical changes in the noun-ciass system of ~ . . . . *~: . . . . . ;~,~,~
causes and wider implications. In: Chin-W. K~m and Herbert Stahlke (eds.), Papers in African
linguistics, 33-54. Edmonton: Linguistic Research.
Givon, Talmy, 1975. Serial ,'er~s and syntactic change. In: Charles Li (ed,~ ~ ' - i i ~
Giv6n, Talmy, 1979. On understanding grammar. New York: Acader~ic Press.
Givrn, Talmy, i984. Syntax: A functional-typological im:~,,crcn, VoL I. Amsterdam: Benjamins.
Go~dstefir~ Me~vyn C and Nawang Nornang, 1970. Modern spoken Tibetan: Lhasa dialect.
Seat:le, WA: University of Washington Press.
Gosehnick, Hell~,, 1977. Haroi clauses. Papers in South East As;an Linguistics 4, 105--124.
Greenberg, Je,seph H., 11966.Some universals of grammar with particular reference to the order
of mean~ngfu! dements. In: Joseph H. Greenberg (ed.L Universals of language, 73-.113.
Camb:~dge, MA: MIT Press.
~3reenberg, Joseph H., 1978. Diachrony, synchrony and language universals. In: Joseph H.
Greenberg, Ch~fle~ A. Ferguson and Edith A. Moravcsik (ed.L Universa|s of human language,
Vol. t: 61-9~. Stanford, CA: Stanford University Press.
Greenberg, Joseph H., 1987. Language in the Americas. Stanford, CA: Stanford University Press.
Haiman, John, 1985. Natural syntax. Cambridge: Cambridge University Press.
H.dl, Christopher J., 1987. Language structure and explanation: A case from morphology. Ph. D.
dissertation, University of South,,i, "-':~", '-
Hall, Christopher J., 1,~8..~.tegradng diachronic and processing principles in explaining the
suffixing preference. ~n: J A. Hawkins (ed.L 32!-349.
J.J. Song / The rise of causative affixes 197

Hannah, Herbert B., i973. Grammar of the Tibetan language. India: Literary and CoUoquia]
Hardjadibrata, R. Rabinranat, 1985. Sundanese: A syntactical analysis. Canberra: Australian
National University.
Hawkins, John A., 1983. Word order universals. New York: Academic Press.
Hawkins, John A. (eel.) :988. Explaining language universals. Oxford: B!ackwell.
Headland, Thomas N. and Alan Healey, 1974. Grammatical sketch of Casiguran Dumagat.
Papers in Philippine Linguistics 6, !-54.
Healcy~ A~ A. Isorc~mbo and M. Chittleborough, 1969. Preliminary notes on Orokaiva grammar.
Papers in New Guinea Linguistics 9, 33--64.
Heath, Jeffrey, 1978. Ngandi grammar, texts and dictionary. Canberra: Australian Institute of
Aboriginal Studies.
Heath, Jeffrey, 1984. Functional grammar of Nunggubuyu. Canberra: Australian Institute of
Aboriginal Studies.

from Man. Berlin: Dietrich Reimer Verlag.

:-~eine, Bemd
. . . . . .and
. . . . .Mcchehad
. Rch, !984. G-. . . . ... . . . .~-';-
. . . . z~at;on ,~u
~-'~ ,--~-~o,,a~v~'-~:-in African languages.
Hamburg: Helmut Broke Verlag.
Henderson, Eug6nie J.A, 1976. Vestig¢~ of morphology in modem standard Khasi. In: P.N.
Jenner, L.C. Thompson and S. Staros~ (ads.), Austroasiatic studies, 477-522. Honolulu, HI:
University of Hawaii Press.
Hetzron, Robert, 1969. The verbal systerc: of Southern Agaw. Los Angeles, CA: University of
Califorrfia Press.
Hewitt, B.G., 1979. Abkhaz. Amsterdam: North-Holland.
Hoffmann, Carl, 1963o A grammar of the Marg~ language. London: O~ford University Press.
Home, Elinor C., 1961. Beginning Javanese. Ne~ Haven, CT: Yale University Pr~s.
Hudson, Joyce, 1978. The core of Wahna~jari grar~maro Canberra: Aus~¢alian Institute of
Aboriginal Studies.
Hutchison, John P., 1976. Aspects of Kanur~ syntax. Ph. D. Dissertation, Indiana University.
Ikranagara, Kay, 1975. Melayu Betaw/graramar. Ph.D. dissertation, University of Hawaii.
Jenscn, John T., 1977. Yapcs¢ reference grammar. Honolulu, HI: Uni-ersity Press of Hawaii.
Joseph, Brian and Irene Philippaki-Warburton, 1987. Mocl:~ Greek. Londo,~: Croom Helm.
Kefer, Michel, 1985. What -~yntaxcan we reconstruct from morphology? Lingua 66, 151-175.
Kimenyi, Alexandre, 1980. A relational grammar of Kinyarwanda. Berkeley, CA: University of
California Press.
Kiparsky, t'aul, 1962. E~ph~nation ia phonology. Dordrecht: Foris.
K6hler, O ~ n , 1981. La langue Kxo~. In: .l. Pcrrot (ed.), Les |~ngues duns le monde ancien et
moderne, ~83-555. Paris: Centre National de la Recherche Scie~tifique.
gr~ber, Alfred L. and George W. Grace, 1960. The Sparkman gr~n~mar of Luisefio. Berkeley
CA: University of California Press.
I~koff, George and Mark Jchnson, 1980. Metaphors we live by. Chicago, IL: University of
Chicago Press.
Langacker, Ronald W., 19'77. Studies in Uto-Aztecan grammar, Vol. 1: An overview of Uto-
Aztecan grammar. SIL and University of Texas at Arlington.
Langacker, Ronald W. (ed.), 1979. Studies in Uto-Aztccan grammar, Vol. II: Modern Aztec
grammatical sketches. SIL and University of Texas at Arlington.
198 J.J. Song / The rise of causative affixes

Langacker, Ronaid W, 1983. Foundations of cognitive grammar. Bloomington, IN: indiana

University Linguistic Club.
Langa~,ker, Ronald W., t984. Studies in Uto-Aztecan grammar, Vol. IV: Southern U~o-Aztecan
grammatical sketches SIL and University of Texas at Arlington.
Langacker, Ronald W., 1987. Foundations of cognitive grammar. Stanford, CA: Stanford
University Press.
Langdon, Margaret, Ic~70. A grammar of Dieguefio. Berkeley, CA: University of California
Lauck, Linda M., 1976. Patep sentences. Workpapers in P~pua New Guinea Languages |7. 5-
Lehmann, Winfred P., 1974. Proto-lndo-European syntax Austin, TX: University of Texas
Lewis, David, 1973 Cattsation. The Journal of Philosophy 7"), 556--567.
Li, Charles (ed.), 1975. Word order and word order change. ~ustin, "IX: University of Texas.
Lighttoot, D., 1979. Principles of diach~onic syntax. ~:ambncIge: Cambridge University Press.
Lindenfeld, 3acquel~ne, L973. Yaqui synta:c P~erkeley CA: University of California Press.
Ma~:Donald, Rodcrick F'., 1976. Indonesian reference grammar. Washington, DC: Georgetown
Uifiversity Press.
Mallin,~on, Graham and. Barry J. Blake, 1981. Language t)pology: Cross-linguistic studies in
syntax. Amsterdam: North-Holland.
Maltzoff, Nicholas, 1985. Essentials of Russian grammar. Lincolnwood: Passport Books.
McCawley, lames D., 1976. Remarks on what can cause what. [~: M. 3hibatani (ed.), 117-12c~.
Merlan, Francesca, 19[;3. Ngalakan grammar, texts and vocabulary. Canberra: Austl-alian
National University.
Mintz. Mal~'olm W., 19"rl. Biko] grammar notes. Honolulu, HI: University of :,-hwaii Pre,;s.
Moshinsky, Julius, 1974. A grammar o~" Southeastern Pomo. Berkeley, CJ~: Uni~,ers~y of
California Pres3.
Mctus, Cecile L., 1971. Hdig Lynon les~ons. Honolulu, HI: University of Hawaii Press.
Newman, Paul, 1974. The Kz auk uru language. Leeds: U r iversity of Leeds.
Nichols, Johanna, 1986. Head-marking and dependent-marking grammar. Language 62, 56-
Noss, R~chatd B., 1964. Thai reference grammar. Washington, DC. Foreign Service Institute.
k'awie~, A,,:~ew ~ , !~77 Or~ the internal relationships ol Eastern Oceanic laa3uages. In: R.C.
Green and M. Ke!ly (eds.L Studies in Oceanic Culture H~stor~, i-~42. ;iu~ol*lu, HI: BerrAce
P. Bishop Museum.
PeJkius, Revere D., 1980. "['he evolu:iou of culture and gzammar. Ph.D. dis:ertat~on, SUNY,
Pike Kenneth, !970. Tag~ernic and matrix linguistics applied to selected African languages.
Norman, OK: Sl~L/University of Oklahoma.
Poppe, Nikolai, 1964. Bashkir manual. Bloomington, IN: Indiana University Press.
Rabel, Lili, 1961. Kh~si: A language of Assam. ;,~at,~n Rouge, LA: Louisiana State University
Radden, G., 1985. Spatial metaphors underlying preposifior.s of ca~.,salily. ~n: W. Papro~l~ and
R. Di~'en Ieds.~, The ubiquity of metaph3r, 177-207. Amsterdam: Benjamins.
Raffo, Yolanda, 1972. A phonology and a~orphology of Songish, a diaiect of Straits Salish.
?h. D. dissertation, University of Kansas.
J.J. Song / The rise of cau.:ative affixes 199

Reclden, James E., 1980. A descriptive grammar of Ew,~ndo. Carbondale, IL: Southern Illinois
Ro~rLs, John R , ~91,7
Amele. London: Croom Helm.
Ross Malcolm an,] John N. Paol, 1978. A Waskia graramar sketch and vocabulary. Canberra:
Australian National University.
guhlen, Merritt, 1987. A guide to the world's languages, Vol. !: Classification. S ~m%~d, CA:
$taafo~ Universi,'y Press.
Saltarelli, Marie, 1988. Bas':lue. London: Croom Helm.
Sanmrin, Wiiliam J., 1966. The Gbeya language: Grammar, texts and vocabularies. Berkeley,
CA: University of Californi~ Press.
Schfitz, Albert J., I969. Nguna grammar. Honolulu, HI: University of Hawaii Press.
Schfitz, Albert J., t~86. The Fijia1: language. Honolulu, HI: University Press of Hawaii.
Scur,:n, l~ter A.M.~ 1973. predicate raising and dative in French and sundry languages.
University at "frier: Linguistic Age,~cy.
Shibatani, Masayoshi, |9,'6. Syntax a~:~dsemamics, Voi. 6: The grammar of causative construc-
tions. New York: Acaden~ic Press.
Shipley, Wi!!i~m F., 1963. Mmdu grammar Berkeley, CA: University of California Press.
Shipley, William F., 1964. Maidu texts and dictionary. Berkeley, CA: University of California
Sischo, Willia~ R., 1979. Michoacfin ~"ahuatl. In: R.W. Langacker (~.), 33-127.
Smi'th: Jean and Pare Weston, 1974. No~es on Mianmin grammar. Workpapers in Papua New
Guinea Languages 7, 35-142.
Sommer, Bruce A., 1969. Kunjen phonology: Synchronic and diachro~ic. Canberra: Australian
National Uni,/ersity.
Sommer, B~lce A., 1972. Kunjen syntax: A gea,'rati*o view. Canberra: Australian Insti~'.ute of
Abongin~l Studies.
Song, Jar L, 1988a. Chase linkage in Korean perip:~rastic cau~~," nv¢~ and purposive constructions.
Language Research 24, 583-606.
Song, Jar J.. 1988b Review of The cau~tives of Malagasy b) C. Randriamasimanana.
Australian Journal of Linguistics 8, 340-343.
Song, Jar J., 1989. A universal typology of causative constructions: Functional and diachronic
perspectives. Ph. D. thesis, Monash University.
Tampubolon, Daulat P., 1983. Verbal aff~ation in Indonesian: A semantic enplanation. Can-
berra: Australian National University.
Taylor, Charles, |9g5. Nkore-Kiga. London: Croom Helm.
Teeter, Karl V., 1964. The Wiyot language. Berkeley, CA: Umvers~ty of California Press.
Tom!in, Russell S., 1986. Basic word order: Funet'.'o~-~! p~nei.'~!cs. Londot~: Croom Helm.
TullY, David H., 19"~9.Tetelcingo Nahuatl. ~n: R.W. Langacker (~i.)~ 1-140.
TullY, David H., ~9~7. Nfihuatl causative/applicatives in cognitive grammar. In: 1~ Rudzka-
Ostyn (ed.), Topics in co~itive linguistics, 587-618. Philadelphia, PA: Benjamins
Vennemann, Thee, 1973. Explanation in synh~. In: J.P. Kimball (ed.), Syntax an~ semantics,
Vol 2, 1-50. New York: Seminar Press.
Vennemann, Thee, |975. An explanation of drift In: C. Li (ed.), 269-305.
Vichit-~Padakan, Rasami, 1976. The concept of inadvertence in Thai periphsastic causative
constructions. ~ : ~ . Shibatani (ed.), ~.59-4.76.
Vitale, Anthony J., 1981. Swahili syntax. Dordrecht: Foris.
200 J.J. Song / The rise of causative affixes

Walker, Dale F., 1976. A grammar of the Lampung language: The Pesisir dialect of Way Lima.
~karta: Badan Penyelenggar S~-=iN* JSA, Universitas Atma Jaya.
Watkins, Calvert, !976. Towa.d Prolo-lndo-European syntax: Problems and pseu-Jo-pr,)b]ems.
I," S.B. Steerer, C.A. Walker and S.S. Mufwene (eds.), Diachronic syntax, 305--326. Chicago,
Weir, E.M. Helen, 1986. Footprints of yesterday's syntax: Diachronic development of certain
verb prefixes in an OSV language (Nad6b). Lingua 68, 291-316.
Williamson, Kay, 1965. A grammar of the Kolokuma dialect of .Ij.o. London: Cambridge
University Press.
Winstedt, Richard O., 1914/1957. Malay grammar. Oxford: Clarendon.
Wolfenden, Elmer, 1971. Hiligayr~on reference grammar. Honolulu, HI: University of Hawaii
Wolfendcn, Elm~,ro 1975. A description of Hiligaynon syntax. Norman, OK: SlL/University of
Wolf.~nden, Stuart N., 1929. Outlines of Tibeto-Burman linguistic morphology. London: Royal
As:,atic Society
Wolff, Ekkehard, 1983. A grammar of the Lamang (Gwfid Lfim~) language. Gliickstadt: Verlag
J.J. Augustia.
Wolff, John V., !966-1967. Beginning Cebuano. New Haven, CT: Yale University Press.
Zcwen, Francois X.N., 1977. The Marshallcse language: A study of its phonology, morphology
and syntax. Berlin: Verlag yon Dietrich Reimer.