You are on page 1of 16

Transitivity pairs, markedness,

and diachronic stability*


BERNARD COMRIE

Abstract

Utilizing the classification of formal relations in causative–inchoative verb


pairs developed by Haspelmath (1993), this article examines two further
ramifications. First, the 30 verbs in his list are divided into two groups of
fifteen, one where crosslinguistically a causative pairing is most likely and
the other where the anticausative pairing is most likely. Individual lan-
guages are then tested to see whether their verbs conform to this ordering,
with overall very positive results. Second, the diachronic stability of lan-
guages with predominance of one or other of the individual types within
the classification, in particular of the causative and anticausative types, is
examined, with some results suggesting stability at least over millennia.
Thus, Classical Arabic and Maltese have very similar profiles, even down
to individual lexical concepts (even where the two languages have noncog-
nate lexical items). Uralic languages, despite their close proximity to Indo-
European languages with predominance of the anticausative type, still show
predominance either of the causative type (Finnish, Udmurt), or of the
equipollent type (Hungarian), the latter perhaps reflecting a compromise
between the putative predominance of the causative type in Proto-Uralic
and contact with Indo-European.

1. Introduction

In a recent important contribution to our understanding of the relation


between intransitive and transitive members of a verb pair, and more spe-
cifically of verb pairs consisting of inchoative and causative, Haspelmath
(1993), itself an elaboration of Nedjalkov (1969), proposes a methodology
for investigating such pairs and draws a number of typological conclu-
sions. The aim of the present article is to investigate further data using

Linguistics 44–2 (2006), 303–318 0024–3949/06/0044–0303


6 Walter de Gruyter
Brought to you by | Brown University Rockefeller Library (Brown University Rockef
Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
304 B. Comrie

Haspelmath’s methodology, attempting to fine-tune some of the details,


and to test further hypotheses.
Haspelmath classifies the morphological markedness relation within
inchoative/causative pairs as follows. First, a distinction is made between
anticausative, causative,1 and nondirected. In an anticausative pair, the
intransitive member is morphologically more marked than the transitive,
as in the German and Japanese examples in (1).
(1) Derived intransitive (anticausative, A)
a. umdrehen ‘turn (vt)’ German
sich umdrehen ‘turn (vi)’
b. or-u ‘break (vt)’ Japanese2
or-e-ru ‘break (vi)’

In a causative pair, the transitive member is morphologically more


marked than the intransitive, as in the examples in (2).
(2) Derived transitive (causative, C)
a. enden ‘finish (vi)’ German3
be-enden ‘finish (vt)’
b. ak-u ‘open (vi)’ Japanese
ak-e-ru ‘open (vt)’

In nondirected pairs, there is no greater morphological marking of either


member, as in the examples in (3)–(5). Nondirected pairs can be further
subdivided into labile, where the same form is used as both intransitive
and transitive (as in example [3]); equipollent, where both members of
the pair are marked morphologically, but in di¤erent ways (as in example
[4]); and suppletive, where di¤erent intransitive and transitive roots are
used (as in example [5]).
(3) Nondirected I: labile (L)
a. kochen ‘boil’(vi/vt) German
b. hirak-u ‘open (vi/vt)’ Japanese
(4) Nondirected II: equipollent (E)
a. versinken ‘sink (vi)’ German
versenken ‘sink (vt)’
b. kowa-re-ru ‘be destroyed’ Japanese
kowa-s-u ‘destroy’
(5) Nondirected III: suppletive (S)
a. sterben ‘die’ German
töten ‘kill’
b. sin-u ‘die’ Japanese
koros-u ‘kill’

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
Transitivity pairs 305

Table 1. Maltese inchoative/causative pairs

Intransitive Transitive Ditransitive

1. boil g¡ela g¡alla C


2. freeze reża¡ reżża¡ C
3. dry nixef nixxef C
4. wake up (¼ 21) qam qajjem C
5. go out/put out n-tefa tefa A
6. sink g¡ereq g¡erreq C
7. learn/teach t-g¡allem g¡allem A
8. melt dieb dewweb C
9. stop waqaf waqqaf C
10. turn dar dawwar C
11. dissolve tera¡ terra¡ C
12. burn n-¡araq ¡araq A
13. destroy n-qered qered A
14. fill n3t4ela5 mela A
15. finish n-temm temm A
16. begin beda beda L
17. spread n-firex firex A
18. roll t-gerbeb gerbeb A
19. develop żviluppa żviluppa L
20. get lost/lose n-tilef tilef A
21. rise/raise (¼ 4) qam qajjem C
22. improve tieb tejjeb C
23. rock t-bandal bandal A
24. connect r3t4abat rabat A
25. change t-bidel bidel A
26. gather n-ġabar ġabar A
27. open n-feta¡ feta¡ A
28. break n-kiser kiser A
29. close nt-g¡alaq g¡alaq A
30. split n-feraq feraq A
31. die/kill miet qatel S

Haspelmath then proceeds to take a list of 31 concepts typically ex-


pressed crosslinguistically by means of verbs, identifying the translation
equivalent (or equivalents) of each such concept in each language under
investigation, and thus establishing tables like Table 1 for Maltese, that
is, indicating for each concept which type of morphological relation is in-
stantiated in the language in question.4 In this way, it is possible to estab-
lish a profile for each language, showing the number of verbal concepts
for which that language makes use of each of the five types. Table 2 gives
the language profiles for the languages of Haspelmath (1993), supple-
mented by Swedish, Tsez, and Maltese.
The source of the statistics in Table 2 is primarily Haspelmath (1993),
to which reference should be made for the raw data and individual-

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
306 B. Comrie

Table 2. Language profiles

A C E L S

English 2 0 1 25 3
German 15.5 1 4 9.5 1
[Haspelmath 14.5 0 4 11.5 1]
Swedish 19 4 3 4 1
French 20.5 2 0 7.5 1
Rumanian 24 1 0 3 2 þ1 unclear
Russian 23 0 5 0 3
Lithuanian 17.5 6 6 0.5 1
Greek, Modern 13.5 0 0 16.5 1
Armenian 16 8.5 5.5 0 1
Hindi-Urdu 7.5 14 7.5 2 0
Finnish 4.5 14 6.5 0.5 1.5 þ4 unclear
[Haspelmath 12 13.5 0.5 0.5 1.5 þ3 unclear]
Udmurt 10.5 12.5 4.5 2.5 1
Hungarian 7 9 12 0 3
Turkish 9 17.5 2.5 0 1
Mongolian 6 22 2 0 1
Georgian 9 4.5 15.5 0 2
Tsez 4 18 8 0 1
Lezgian 8 12 6 5 0
Arabic, Standard 19 7.5 1 1 1.5
Maltese 17 11 0 2 1
Hebrew 20.5 7.5 2 1 0
Japanese 3.5 5.5 20.5 0.5 1
Indonesian 0 14 17 0 0
Swahili 11 11 8 0 1

language tables for the languages he considers. The Swedish data are
given in Comrie (2005), the Tsez data in Comrie (2000), the Maltese
data in Table 1. I have made some small changes to Haspelmath’s analy-
sis of German, in particular the consideration of enden — be-enden to
constitute a causative pair, and of er-löschen — löschen as an anticausa-
tive pair. Changes made to the analysis of the Finnish data are more sub-
stantial, and are discussed in more detail in Comrie (2005); they hinge pri-
marily on the analysis of pairs like parant-a-a — parant-u-a ‘improve’,
which I treat as having an opposition between intransitive su‰x -a and
transitive su‰x -u (the final -a is the infinitive marker). Haspelmath’s
original statistics are given in square brackets in Table 2.
It will be noted that the profiles of individual languages can di¤er
quite considerably. Considering only languages where the most popular
type has more than twice as many members as the next most popular,

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
Transitivity pairs 307

overwhelmingly anticausative languages are Rumanian, Russian, French,


and Lithuanian; overwhelmingly causative is Finnish (at least on my
reanalysis); overwhelmingly equipollent is Japanese; overwhelmingly
labile is English. It should be noted that the languages listed in Table 1,
and likewise the larger language sample in Haspelmath (1993), are not
claimed to be a balanced representation of the world’s languages, being
biased towards major languages of Eurasia (for which data are more
readily accessible). As noted by Nichols (1993) and others, the global
crosslinguistic tendency is to prefer causative pairs, with Europe being
anomalous in its high incidence of languages preferring anticausative
pairs.
The choice of the 31 verbal concepts is, however, principled. It is de-
signed to include both concepts that show up more frequently crosslin-
guistically as anticausative pairs and those that show up more frequently
as causative pairs, as well as concepts intermediate in terms of their fre-
quency of being encoded by one type or another. This is a point to which
we return in Section 2. Moreover, the pair ‘die’ — ‘kill’ was selected as
one that is crosslinguistically most likely to be encoded suppletively
among the languages of the world, although there are languages with
other possibilities for this concept, such as Turkish öl — öl-dür, using the
productive causative su‰x -DIr.

2. Markedness

As noted by Haspelmath (1993), it is possible to order 30 of the 31 verbal


concepts in his list in terms of their likelihood of being an anticausative or
a causative pair crosslinguistically. This ordering, which I have adopted
in Table 1, seems to correspond cognitively to the likelihood of the in-
transitive or transitive member of the pair being considered as basic. For
instance, it is more usual for entities to freeze, dry, or wake up of their
own accord, so the inchoative is typically less marked than the causative.
Conversely, it is more usual for entities to be opened, broken, closed, or
split by an external force than to open, etc., spontaneously, so here the
causative is typically less marked than the inchoative. While the ordering
of particular verbs may give rise to questions, the overall ordering seems
intuitively correct. (Note that the pair ‘die’ — ‘kill’ is not included since it
is typically expressed by means of suppletion.)
The empirical validity of the ordering, reflecting the cognitive marked-
ness of the inchoative or causative member of each pair, can be further
tested against the distribution of di¤erent types of pairings in the individ-
ual languages. Of course, this is not entirely an independent test since the

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
308 B. Comrie

data from the individual languages form part of the crosslinguistic data
pool that initially informed Haspelmath’s hypothesis. But nonetheless,
if the ordering is valid not only for the pooled data but also for each
individual language, or almost so, then the hypothesis is strengthened.
Table 3 provides the relevant information.
This table gives, separately for anticausatives and for causatives, and
for each language in Table 2, the number of verbs falling in the top half
(1–15) of the list of verbs as given in Table 1 versus the bottom half (16–
30) — item 31 is omitted since it usually involves suppletion.6 Since Indo-
nesian has no anticausatives, and English, Modern Greek, and Russian
no causatives, these combinations do not figure in Table 3. If our hypoth-
esis is correct, then the overall pattern should be that anticausatives have
fewer verbs in the top half of the verb list than in the bottom half, and
this is expressed in the right-hand column of Table 3 by dividing the num-
ber found in the top half of the verb list by the number found in the bot-
tom half. Conversely, causatives should have fewer verbs in the bottom
half of the verb list than in the top half, and this is expressed in the
right-hand column by dividing the number found in the bottom half of
the list by the number found in the top half. Doing the calculations this
way around almost avoids having to divide by zero. If the hypothesis
held in its strongest form, all the numbers in the right-hand column of
Table 3 would be less than zero.
Examination of the right-hand column shows that the hypothesis is al-
most borne out in its strongest form. Among the anticausatives, all lan-
guages but one fall in the range 0.00–0.80, while among the causatives,
all languages but two fall in the range 0.00–0.55. Two of the exceptions
involve very small numbers of instances, namely English with only two
anticausatives (both in the top half of the table, thus requiring division
by zero), and Swedish with only three causatives (one in the top half,
two in the bottom half ). For consistency, if one excludes these on the
basis of low numbers, one should also exclude confirmatory instances in-
volving three or fewer instances, namely causatives in French, German,
and Rumanian. The remaining exception is provided by causatives in
Swahili, with slightly more in the bottom half than in the top half (6
in the bottom versus 5.5 in the top, to give a ratio of 1.09), although
anticausatives in Swahili are well behaved (4 in the top versus 7 in the
bottom, to give a ratio of 0.57). Overall, the hypothesis is very strongly
confirmed.
Given that possibilities for future work have been introduced, it is
worth considering some further possibilities, in additional to tightening
up the statistical analysis. One could test the ordering by means of psy-
cholinguistic experiments, to see if it is indeed the case that subjects are

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
Transitivity pairs 309

Table 3. Anticausative and causative expression of particular concepts in particular


languages

Language 1–15 16–30 Ratio

Distribution of anticausatives:
Tsez 0 4 0.00
Mongolian 0.5 5.5 0.09
Udmurt 1 9.5 0.11
Finnish 1 4.5 0.22
Arabic 3.5 13.5 0.26
Turkish 2 7 0.29
Lezgian 2 6 0.33
German 4 11.5 0.35
Hungarian 2 5 0.40
Japanese 1 2.5 0.40
Georgian 3 6 0.50
Greek, Modern 4.5 9 0.50
Lithuanian 6 11.5 0.52
Armenian 5.5 10.5 0.52
Russian 8 15 0.53
Maltese 6 11 0.55
Swahili 4 7 0.57
Hebrew 7.5 13 0.58
Swedish 7 12 0.58
French 8.5 12 0.71
Rumanian 10 14 0.71
Hindi–Urdu 4 5 0.80
English 2 0 —
Distribution of causatives:
Arabic 8.5 0 0.00
French 1 0 0.00
Lithuanian 6 0 0.00
Rumanian 1 0 0.00
Hebrew 6 0.5 0.08
Hindi–Urdu 9.5 2 0.21
Maltese 9 2 0.22
Udmurt 10 2.5 0.25
Georgian 3.5 1 0.29
Hungarian 7 2 0.29
Lezgian 9 3 0.33
Japanese 4 1.5 0.38
Finnish 10 4 0.40
Indonesian 8.5 3.5 0.41
Armenian 6 2.5 0.42
Turkish 11 5.5 0.50
Mongolian 14.5 7.5 0.52
Tsez 11 6 0.55
Swahili 5.5 6 1.09
Swedish 1 2 2.00

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
310 B. Comrie

more likely to treat concepts with lower numbers in the list as more typi-
cally inchoative, those with higher numbers as more typically causative.
A di¤erent kind of test would involve searching for the relevant verbs in
a large corpus and seeing whether the hypothesized cognitive markedness
also has correlated frequency e¤ects, with the lower numbers occurring
more frequently as inchoatives, the higher numbers more frequently as
causatives.
In Japanese, it would also be possible to carry out such tests against a
more specific set of data. One of the striking characteristics of inchoative–
causative pairs in Japanese is that the su‰x -e is used with some verbs to
mark the inchoative, with other verbs to mark the causative. Comprehen-
sive lists of 36 pairs where -e marks the anticausative and 57 where it
marks the causative are given in Tables 4 and 5 respectively.
The lists are adapted from Jacobsen (1982: 197–199) and encompass
only verbs with the -e su‰x, whereas many verbs in Japanese, including
some in Haspelmath’s list, use other su‰xes to mark causative or anticau-
sative.7 In general, only those pairs are listed where the formal relation-
ship is absolutely transparent and the semantic relationship at least plau-
sible; the only exception to morphological transparency is the pair hairu
— ireru, where the intransitive member has an additional initial ha, his-
torically a separate morpheme.
With respect to the data in Tables 4 and 5, one can now ask the same
question as gave rise to the data in Table 3, namely, whether anticausa-
tive pairings tend to occur lower down in Table 1 and causatives higher
up. Unfortunately, the number of concepts in Tables 4 and 5 that corre-
spond closely to the verbal concepts identified by Haspelmath is small, so
that a detailed execution of the proposed task may not be particularly
valid. The fact that there are more verbs in Table 5 than in Table 4 might
suggest that causative pairings are more normal in Japanese than anticau-
sative pairings, a generalization that finds some support from the Japa-
nese entry in Table 2. If we start with the causative pairings of Table 5,
then my assessment is that two concepts from the top half are found
(‘sink’, ‘stop’) and three from the bottom half (‘connect’, ‘open’, ‘close’);
the prediction that the numbers should decline as we progress from lower
to higher decades is not borne out, but the number of items is small and
causative pairings do seem to be less marked in Japanese than anticausa-
tive pairings. Turning now to the anticausative pairings of Table 4, I find
three concepts from the top half (‘boil’, ‘melt’, ‘burn’) and four from
the bottom half (‘rock’, ‘open’, ‘break’, ‘split’); the prediction is that the
number of items should increase as we progress from lower to higher de-
cades, and this is rather weakly confirmed, in that the number in the third
decade is higher than in the lower decades. But probably one should

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
Transitivity pairs 311

Table 4. Japanese inchoatives in -e

Inchoative English gloss Causative English gloss

hag-e-ru come o¤ hag-u peel o¤


hirak-e-ru open hirak-u open
hodok-e-ru come untied hodok-u untie
hor-e-ru be hollowed hor-u dig
hur-e-ru shake hur-u shake
kak-e-ru lack kak-u lack
kir-e-ru be cut o¤ kir-u cut
kudak-e-ru be smashed kudak-u smash
kuzik-e-ru be discouraged kuzik-u break
makur-e-ru be tucked up makur-u tuck up
mi-e-ru be seen mi-ru see
mog-e-ru come o¤ mog-u pluck o¤
mom-e-ru be wrinkled mom-u wrinkle
muk-e-ru peel muk-u peel
nezir-e-ru get twisted nezir-u twist
ni-e-ru boil ni-ru boil
nom-e-ru (be good to) drink nom-u drink
nug-e-ru come o¤ nug-u take o¤
nuk-e-ru come o¤ nuk-u remove
or-e-ru break or-u break
sabak-e-ru be sold sabak-u sell
sak-e-ru tear sak-u tear
sir-e-ru become known sir-u come to know
sog-e-ru be hollow sog-u diminish
sur-e-ru rub sur-u rub
tigir-e-ru be torn o¤ tigir-u tear
tok-e-ru melt tok-u melt
tor-e-ru be produced tor-u take
tuka-e-ru be usable tuka-u use
ur-e-ru sell (well) ur-u sell
war-e-ru split war-u split
yabur-e-ru get broken yabur-u break
yak-e-ru burn yak-u burn
yom-e-ru be legible yom-u read
yozir-e-ru get twisted yozir-u twist
yur-e-ru shake yur-u shake

rather say that with so few items, the relevant hypotheses cannot really be
tested.

3. Diachronic stability

As already noted in passing, there are striking di¤erences in the statistical


distribution of the five di¤erent types of pairings in di¤erent languages.

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
312 B. Comrie

Table 5. Japanese causatives in -e

Inchoative English gloss Causative English gloss

ak-u open ak-e-ru open


aradat-u be aggravated aradat-e-ru aggravate
dok-u get out of the way dok-e-ru remove
ha-ir-u enter ir-e-ru put in
hikkom-u draw back hikkom-e-ru pull back
hisom-u lurk hisom-e-ru conceal
hukum-u include hukum-e-ru include
hus-u lie down hus-e-ru lay down
iradat-u be excited iradat-e-ru excite
isam-u be encouraged isam-e-ru admonish
itam-u hurt itam-e-ru injure
kagam-u bend kagam-e-ru bend
karam-u twine round karam-e-ru bind
katamuk-u lean katamuk-e-ru lean
katazuk-u be in order katazuk-e-ru tidy up
kizutuk-u get hurt kizutuk-e-ru hurt
kom-u be crowded kom-e-ru load
kurusim-u su¤er kurusim-e-ru torment
matiga-u be wrong matiga-e-ru mistake
muka-u face muka-e-ru meet
muk-u face muk-e-ru face
nagusam-u cheer up nagusam-e-ru cheer
narab-u line up narab-e-ru line up
natuk-u become attached to natuk-e-ru win over
nok-u get out of the way nok-e-ru remove
nurum-u become lukewarm nurum-e-ru make less hot
sakadat-u bristle sakadat-e-ru ru¿e up
sirizok-u retreat sirizok-e-ru expel
sizum-u sink sizum-e-ru sink
sitaga-u go along with sitaga-e-ru take along with
sodat-u grow up sodat-e-ru bring up
somuk-u turn away somuk-e-ru turn away
soro-u match soro-e-ru arrange
so-u go along with so-e-ru add to
subom-u get narrower subom-e-ru narrow
sukum-u crouch sukum-e-ru duck (head)
susum-u advance susum-e-ru promote
taga-u deviate taga-e-ru break (one’s word)
tat-u stand tat-e-ru raise
tawam-u be bent tawam-e-ru bend
tiga-u di¤er tiga-e-ru alter
tikazuk-u approach tikazuk-e-ru allow to come near
tizim-u shrink tizim-e-ru reduce
todok-u arrive todok-e-ru deliver
tubom-u close tubom-e-ru close

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
Transitivity pairs 313

Table 5 (Continued )

Inchoative English gloss Causative English gloss

tuga-u mate tuga-e-ru fix (an arrow to a string)


tuk-u adhere to tuk-e-ru attach
tumu be stu¤ed tum-e-ru stu¤
tutau go along tuta-e-ru convey
tuzuku continue tuzuk-e-ru continue
ukabu float ukab-e-ru set afloat
utumuku look down utumuk-e-ru cast down
yamu stop yam-e-ru stop
yasumu rest yasum-e-ru rest
yawaragu soften yawarag-e-ru soften
yugamu bend yugam-e-ru bend
yurumu loosen yurum-e-ru loosen

The question that I wish to address in this section is whether there is any
principled basis to these di¤erences in terms of either the genealogical
a‰liation of the languages concerned or their geographical distribution.
This investigation is particularly important in light of claims made by
Nichols (1993 and elsewhere) about the historical stability of patterns of
inchoative–causative pairing in the lexicon.
In her work on this topic, Nichols notes that crosslinguistically there is
a preference for causative pairings across the languages of the world in
general, although much of Europe constitutes a significant exception,
with a preference for anticausative pairings. This feature of European
languages was already noted in Section 1. Many European languages, in
particular Indo-European languages, make widespread use of what is ety-
mologically a reflexive marker (e.g. French se, German sich, Russian -sja)
as a means of deriving intransitives from transitives, although other mor-
phological strategies are also found, such as the use of the middle voice in
Modern Greek. Before proceeding to a more detailed examination of the
data, it is necessary to clarify certain important points. First, the nature
of the morphological relationship between inchoative and causative mem-
bers of a pair is only one of a number of features taken into consideration
by Nichols in assessing relevant aspects of the typological profile of a lan-
guage, so the material to be presented is not an exhaustive confrontation
with her claims. Second, Nichols’ claims concerning diachronic stability
relate primarily to the anticausative and causative types, suggesting that
it requires a considerable time period for a language to develop either an
overwhelmingly anticausative or an overwhelmingly causative profile.
Changes leading towards at least some of the other types are less decisive.
In particular, it is always possible for a language to lose morphology, as

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
314 B. Comrie

has happened in the case of English, with its striking and areally atypical
predominance of labile pairs.
Three sets of data from among those discussed in this article are rele-
vant to the question of diachronic stability. The first concerns a restricted
set of languages, namely the relationship between Standard Arabic and
Maltese, two closely related Semitic languages — indeed, from a purely
linguistic point of view Maltese can be considered a variety of vernacular
Arabic. The time depth separating the two varieties from their common
ancestor is just over a millennium, with the proviso that Standard Arabic
has been maintained artificially close to that ancestor while Maltese has
undergone substantial influence from Sicilian and more recently Standard
Italian, Romance varieties with the typical European propensity for anti-
causative pairings. Over this time period, however, Maltese and Standard
Arabic have diverged little from one another in terms of their transitivity
profiles, both having a preference for anticausative formations, these out-
numbering causative formations by about 2 to 1, with little more than a
token presence of the other types. The numbers are perhaps too small to
make any more specific claims, in addition to which slight di¤erences in
decisions as to which lexical item best corresponds to the relevant verbal
concept might have led to slight di¤erences in the overall statistics. None-
theless, it is worth noting that the slight increase in causative formations
in Maltese vis-à-vis Standard Arabic actually goes against the dominant
European pattern, to which Italian conforms.
One can, however, take the comparison of Standard Arabic and Mal-
tese somewhat further, by examining particular concepts in the list. It
turns out that for 23 of the 31 items, Maltese and Standard Arabic agree
in the assignment of that item to a particular type. (More specifically,
in nineteen cases the type assigned by me for Maltese is the only type
assigned by Haspelmath for Arabic; in four cases one of the two types
assigned by Haspelmath for Arabic). In ten cases, shown in Table 6, the
two languages agree even though etymologically unrelated items are
given in the source lists — although this number should perhaps be re-
duced to seven, since in three cases Haspelmath gives alternative lexical
choices, belonging to di¤erent types, for Standard Arabic. But at least
a high degree of overall diachronic stability, extending down even into
some of the details, is manifested by these two languages over a period
of about a millennium.
One item in the Maltese list of Table 1 is worthy of specific comment,
namely the one Italian/Sicilian loan żviluppa ‘develop’, which has the
same form as inchoative and as causative. In Italian, the two are, or
at least can be, distinguished as intransitive sviluppar-si and transitive
sviluppare, the former having the etymologically reflexive su‰x to form a

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
Transitivity pairs 315

Table 6. Standard Arabic and Maltese pairs with the same type but di¤erent lexical choices

Standard Arabic Maltese

Inchoative Causative Inchoative Causative

3. ja¤a ja¤afa nixef nixxef


4. sahaa sahhaa qam qajjem
˙ ˙
st-ayqaz a ˙ ˙˙ a
’a-yqaz
8. saaha ˙ sayyah˙a dieb dewweb
˙ ara
n-sah sahara˙
11. ˙ ˙
¶aaba ˙ ˙
’a-¶aaba tera¡ terra¡
17. n3t4ašara našara n-firex firex
18. ta-dahraja dahraja t-gerbeb gerbeb
20. fuqida˙ ˙
faqada n-tilef tilef
daa‘a xasira
23. ta-’arjaha ’arjaha t-bandal bandal
26. l3t4amma˙ lamma ˙ n-ġabar ġabar
30. n-šaqqa šaqqa n-feraq feraq

characteristic European anticausative pairing. Despite the heavy influence


of Sicilian/Italian on Maltese, including the development of ways of
conjugating Romance-origin verbs in Maltese and the adoption of some
Romance nominal plural formations (e.g. in -i), Maltese has neither bor-
rowed nor calqued the Romance reflexive pattern, providing another spe-
cific instance of diachronic stability.
The second set of data concerns the overall preference for anticausa-
tive pairings in the Indo-European languages of Europe, running clearly
across Germanic, Romance, and Slavic, to which we might add Modern
Greek (which has 13.5 anticausative pairs and no causative pairs, al-
though its most frequent type is the labile, at 16.5). The predominance of
anticausative pairs diminishes somewhat as one moves further east, with
Lithuanian and Armenian showing a rise in the number of causative
pairs, although both still have anticausative pairs by far the most fre-
quent. But the one clearly non-European Indo-European language in the
sample, Hindi-Urdu, reverses the pattern, having causative pairs as its
most frequent type.
Proto-Indo-European had both productive causative and anticausative
formations, in particular the causative in *-y and the middle voice, the
latter having inter alia the function of detransitivization. In the west,
causativization has clearly lost out, although it was still productive as late
as Proto-Germanic, with its *-j causative (where the di¤erence between y
for Proto-Indo-European and j for Proto-Germanic is notational) surviv-
ing in nonproductive formations like English fall — fell, the latter for
*fall-j-. In the east, the same is true of the middle voice. However, the

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
316 B. Comrie

middle voice has also fallen out of use across most of the Indo-European
languages of Europe, surviving intact only in Modern Greek; in Ger-
manic, the middle voice was intact in Gothic, but is found only in fossil-
ized traces in the other oldest Germanic languages; it was alive in Latin
deponent verbs (though with a di¤erent formation), but does not survive
into the Romance languages; it is not found in Balto-Slavic. Rather, the
European languages, with few exceptions (such as English) have partici-
pated in what must have been an areally conditioned renewal of the se-
mantics of the middle voice by means of reflexive morphology. Only at
the periphery do we find languages like Modern Greek that have not lost
the inherited middle voice or English that have not developed a reflexive-
based anticausative formation. (In the case of English, the early loss of
the inherited reflexive, cognate with German sich, Danish sig, may have
been a contributory factor. Even in the earliest Old English the only sur-
viving descendant of the Common Germanic reflexive is the possessive
sin, and even this is only sporadically attested in verse.)
In the case of the Indo-European languages of Europe, then, we are
looking at a period of perhaps two millennia (more or less in the case
of individual languages and branches) in which an innovative system of
marking anticausatives arose, albeit at least in part as a replacement for
the earlier middle voice. The predominance of anticausative pairings in
European Indo-European languages nonetheless gives some indication of
the time span needed by a language to develop this crosslinguistically
marked pattern as its basic typological profile.
Third, we can turn to the Uralic languages that have been in close con-
tact with European languages, to see to what extent they have assimilated
to the predominantly anticausative pattern of their Indo-European neigh-
bors. The languages for which analyzed data are available from Haspel-
math (1993) are Finnish (in close contact with Balto-Slavic and Germanic
languages), Hungarian (in close contact with Slavic, Germanic, and Ro-
mance languages), and Udmurt (in close contact with Russian). I have ex-
amined this question in more detail elsewhere (Comrie 2005), and a brief
summary will be given here. These languages do indeed seem to occupy
an intermediate position between the globally preferred pattern of caus-
ative pairings and the locally preferred pattern of anticausative pairings
in Europe, although a detailed evaluation of this claim would require
comparison of other Uralic languages, in particular those whose contact
with European languages has been less long-lasting. Udmurt is a clear
case of a language with roughly equal numbers of causative and anti-
causative pairings, with a slight preference for the former in the con-
cepts selected by Haspelmath. By my reanalysis, Finnish is overwhelm-
ingly characterized by causative pairings, although Haspelmath’s original

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
Transitivity pairs 317

assignments (which are closer to those assigned by traditional grammar-


ians of Finnish) would have Finnish with roughly equal numbers of anti-
causative and causative pairings. Whatever the precise analysis, Udmurt
and Finnish clearly make significantly greater use of causative pairings
than do their Balto-Slavic or Germanic neighbors, so that whatever their
starting point, they have at best partially assimilated to the general Euro-
pean pattern. Given the heavy Indo-European influence on Finnish espe-
cially, with its adoption of subject — verb — object word order and of
finite subordination, this would suggest that the transitivity profile of a
language is indeed relatively stable diachronically.
Hungarian provides an interesting case, in that its most frequent pair-
ing in the data considered is the equipollent type, a type that is otherwise
not particularly frequent in Europe. Indeed, the only other language
considered here with a predominance of the equipollent type is Japanese.
(The language sample used by Haspelmath [1993] also includes Georgian
and Indonesian with predominance of the equipollent type.) If it is the
case that the original Uralic pattern preferred causative pairings, then it
is possible that the Hungarian preference for equipollent pairings repre-
sents a compromise between this pattern (whence the morpheme marking
the causative member of the pair) and the typical European pattern
(whence the morpheme marking the inchoative member of the pair), but
this must for now remain a speculation to be tested in future research.

4. Conclusions

In this article, I have tried to explore some of the implications and


ramifications of the approach to inchoative–causative pairs developed by
Haspelmath (1993). In particular, I have examined the hypotheses that
the verbal concepts selected by Haspelmath can be arranged in order of
markedness of anticausative versus causative pairings, and that the tran-
sitivity profile of a language in terms of the contrast between predomi-
nance of anticausative and causative pairings is diachronically stable. I
hope to have provided some evidence in favor of both hypotheses, though
also to have indicated areas where more work is needed, for instance
in the psycholinguistic and corpus-linguistic testing of implications of
the markedness hypothesis and in the need for more detailed historical-
comparative research to test the diachronic stability hypothesis.

Received 20 January 2004 Max Planck Institute for Evolutionary


Revised version received Anthropology, Leipzig
11 May 2005

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM
318 B. Comrie

Notes

* Correspondence address: Max Planck Institute for Evolutionary Anthropology,


Deutscher Platz 6, D-04103 Leipzig, Germany. E-mail: comrie@eva.mpg.de.
1. Note that the term ‘‘causative’’ is thus used in two di¤erent senses: to refer to the more
transitive member of an inchoative/causative pair, and more specifically to refer to the
relation between members of such pairs in which the causative member is morphologi-
cally more marked. The context should clarify which usage is intended in particular
cases.
2. In the cited Japanese verb forms, the final -(r)u is the nonpast su‰x, also used as the
citation form.
3. Haspelmath does not consider German enden — be-enden to be a causative pair because
the prefix be- is not otherwise used in this function. I have somewhat loosened Haspel-
math’s criteria in order to consider only the distinction between more and less morpho-
logically marked members of a pair, an easier operational criterion.
4. Angled brackets are used in this table and elsewhere to indicate infixes. Where a given
concept has two roughly equally natural translation equivalents in a given language that
belong to di¤erent types, each is assigned a half point.
5. By assimilation for expected *m3t4ela.
6. This data analysis method follows that used in Dryer (1989).
7. The pair yureru — yuru is actually given by Jacobsen in the causative list; I assume that
this is an oversight and have reassigned it to the anticausatives.

References

Comrie, Bernard (2000). Tsez as an intransitive language: an aspect of the structure of the
Tsez verbal lexicon. In Slovo v tekste i v slovare: sbornik statej k semidesjatiletiju akade-
mika Ju. D. Apresjana, L. L. Iomdin and L. P. Krysin (eds.), 312–316. Moscow: Jazyki
russkoj kul’tury.
— (2005). La typologie des langues ouraliennes, comparées avec les autres grandes familles
d’Eurasie septentrionale. In Les langues ouraliennes aujoudh’hui, Jocelyne Fernandez-Vest
(ed.), 75–85. Paris: Librairie Honoré Champion.
Dryer, Matthew (1989). Large linguistic areas and languages sampling. Studies in Language
13, 257–292.
Haspelmath, Martin (1993). More on the typology of inchoative/causative verb alterna-
tions. In Causatives and Transitivity, Bernard Comrie and Maria Polinsky (eds.), 87–120.
Amsterdam: John Benjamins.
Jacobsen, Wesley (1982). Transitivity in the Japanese Verbal System. Bloomington: Indiana
University Linguistics Club.
Nedjalkov, Vladimir P. (1969). Nekotorye verojatnostnye universalii v glagol’nom slovo-
obrazovanii. In Jazykovye universalii i lingvističeskaja tipologija, Igor’ F. Vardul’ (ed.),
106–114. Moscow: Nauka.
Nichols, Johanna (1993). Transitive and causative in the Slavic lexicon: evidence from Rus-
sian. In Causatives and Transitivity, Bernard Comrie and Maria Polinsky (eds.), 69–86.
Amsterdam: John Benjamins.

Brought to you by | Brown University Rockefeller Library (Brown University Rockef


Authenticated | 172.16.1.226
Download Date | 7/17/12 5:45 PM

You might also like