Hawkins 1979 Implicational Universals As Predictors of WO Change

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 32

Linguistic Society of America

Implicational Universals as Predictors of Word Order Change


Author(s): John A. Hawkins
Source: Language, Vol. 55, No. 3 (Sep., 1979), pp. 618-648
Published by: Linguistic Society of America
Stable URL: http://www.jstor.org/stable/413320 .
Accessed: 27/03/2014 09:40

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Linguistic Society of America is collaborating with JSTOR to digitize, preserve and extend access to Language.

http://www.jstor.org

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS
PREDICTORS OF WORD ORDER CHANGE
JOHN A. HAWKINS
Universityof Southern California
Implicational universals of the form 'If a language has some word order P, then it
also has word order(s) Q' are widely used to explain word order change. A language
which develops word order P without having Q is said to be universallyinconsistent,
and it is predictedthat it will re-introduceconsistency by subsequentlyacquiring Q.
This paper, however, argues against the validity of such a paradigm. The current
assumptionthat the early western Indo-Europeandialects are inconsistent is founded
on inadequate formulations of the synchronic universals.First, more precise formu-
lations are given for some basic word order universalsin a large body of languages;
then their predictions are tested against all the early daughter languages of Indo-
European (using data from Friedrich 1975). None of these dialects exhibits incon-
sistency; hence we have no evidence for the reintroduction of consistency as a
valid mechanismof syntacticchange.It is shown, using historicaldata from Germanic,
that implicationaluniversalscan nonethelessmake correctpredictionsfor the relative
timing and mannerof word orderchanges, but only on the assumption that languages
in evolution obey synchronic universals. Some purely theoretical arguments are
provided to support the empirical findings, and an alternativeexplanatorymodel of
word order change is outlined.*

0. INTRODUCTION.Following the publication of Greenberg 1966 linguists


have become increasingly aware of the importance of implicational statements
within synchronic universal grammar. These statements are of this form: 'If a
language has some property (or properties) P, then it will also have some other
property (or properties) Q.' Twenty-three of the implicational statements proposed
by Greenberg involve word orders such as VSO, PREP, N+ADJ etc.1 More
* I would like to express my particular appreciation to Paul Friedrich, for his helpful
and careful comments on this paper. I would also like to thank Ed Keenan, who provided
the initial stimulus to research the synchronic-diachronicrelationship within the context of
Universal Grammar, in his seminar on language universals at Cambridge University, 1974.
Our many discussionssince then have also helped me to crystallizemy thinking. In addition, I
would like to thank Bob Stockwell and SandraThompsonfor their special help and encourage-
ment, both while I was a visiting professor at UCLA, 1976-7, and on so many occasions
since then. Since joining USC (Fall 1977) I have found an enormously helpful and supportive
environment in which to pursue this work. Among the faculty, I am particularlygrateful to
Larry Hyman, Bernard Comrie, and Ed Finegan in many ways; and I am indebted to the
students in my seminar on word order (Spring 1978) who helped me check the data of
Greenberg's Appendix II and who provided valuable feedback and comments: Saeed Ali
(UCLA), Linda Arvanites (UCLA), Alan Kim, Wei-Lin Lei, Galust Mardirussian(UCLA),
Charles Randriamasiamanana,Jack Roberson, Maner Thorpe, Yukiko Uchida, and Emily
Yarnall. Finally, I would like to thank Joseph Greenbergfor his encouragement,and David
Kilby and Talmy Giv6n, who gave me their criticismsof an earlier draft.
1
The following abbreviationsare used here for patterns of word order:
VSO Verb-Subject-Object
SVO Subject-Verb-Object
SOV Subject-Object-Verb
618

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 619

recently, historical linguists have adopted these synchronic word order universals
in the description and explanation of language change. The implicational universal
is of considerable potential usefulness in a historical context. A statement 'If P,
then Q' requires that languages must not have property P without property Q
(*P & -Q). If a language does have P, therefore, it must have Q as well (P & Q);
but in the absence of P, a language may still have Q (-P & Q), or else may have
neither property (-P & -Q). Three co-occurrence types are permitted, and just
one is disallowed. The implicational statement thus defines universal parameterson
types of linguistic variation, and this allows for the possibility of language change
from one type to another.
This paper is concerned with the relationship between such synchronic implica-
tional universals and language change in the area of word order. In ?1, I formulate
four hypotheses concerning the possible predictive power of synchronic word
order implications for historical change. In ?2, synchronic implicational universals
are carefully formulated for a group of properties in an extensive language sample,
taken primarily from Greenberg's Appendix II. The predictive potential of these
implicational statements for word order change is then tested against historical
data. In ?3, I analyse the same sample word order properties in the earliest dialects
of Indo-European (using data from Friedrich 1975); in ?4, I consider the evolution
of these word orders in the Germanic language family. I shall argue that synchronic
implicational universals do have considerable predictive potential in diachrony,
but that the most widely-held theory defining this potential (that associated with
Lehmann, Vennemann, and many others) has practically no empirical support. In
?5, I present purely theoretical arguments against Lehmann and Vennemann's
historical logic, and in support of the empirical findings. In ?6,I discuss the major
conclusions that can be drawn from comparison of synchronic and diachronic
data; and in ?7,I outline an alternative explanation for word order change, which is
founded on typological predictions and which avoids the empirical and theoretical
problems of the Lehmann-Vennemann theory.
A widely-held current view of the potential role of
1. FOUR HYPOTHESES.
synchronic implicational universals within diachronic syntax is represented by the
work of Lehmann and Vennemann (see references). We can paraphrase their
theory as follows, letting property P stand for VO order, Q for the other properties
of VO languages, - P for OV order, and - Q for the other properties of OV
languages. A language which has - P & - Q at some stage in its history, and which
then acquires P, thereby introducing universal inconsistency (*P & - Q), is required
by the implication 'If P, then Q' to acquire Q later; i.e., -P &-Q> *P &-Q>
P & Q. Hence the change from OV to VO serves as a trigger which sets in motion

PREP Preposition (before Noun)


POST Postposition (after Noun)
N + GEN Noun-Genitive
GEN + N Genitive-Noun
N + ADJ Noun-Adjective
ADJ + N Adjective-Noun
N + REL Noun-Relative
REL + N Relative-Noun

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
620 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

the subsequent evolution of N + ADJ (from ADJ + N), of N + GEN (from


GEN + N), of PREP (from POST) etc., all of which are implicationally consequent
upon VO. I refer to these as 'trigger-chain' theories. Their apparent merit is that
widespread changes in the syntax of a language through time-in this context,
extensive word order re-arrangements-are explained as the result of just one
initial change, that of - P to P. If that change alone can be independently explained,
then the synchronic implicational universals predict all the remaining changes. For
both Lehmann and Vennemann, the loss of case-marking in an SOV language
(through phonological reduction) brings about SOV/OSV ambiguities, which are
resolved by developing the new basic word order SVO. This structure is then the
trigger for the other word order changes.
Underlying the trigger-chain approach is the assumption that languages will
violate synchronic universals in the course of change-and, furthermore, that the
directionality (or drift) of change can be seen as an increasing approximation to
synchronic universal laws, following the initial violation. But however attractive
this schema might appear, there is another logically possible way in which implica-
tional universals might be invoked in a historical context. We might assume just
the reverse: that languages in evolution will remain consistent with synchronic
universal laws, and will obey throughout their histories whatever generalizations
can be set up on the basis of current cross-language synchronic comparison. This
alternative has the merit of remaining true to the synchronically observed facts,
although we now lose the ability to predict further word order changes after the
first one. If there is no universal inconsistency (*P & - Q), there will be no con-
comitant pressure on a language to relieve that inconsistency, and hence no
momentum for continued changes. A -P & -Q language which acquires Q,
becoming - P & Q, is perfectly consistent and is under no pressure to develop P
as well: it may do so but it does not have to.
We have, therefore, two hypotheses:
(1) THE UNIVERSALVIOLATION(or Trigger-Chain) HYPOTHESIS(UVH):
Languages violate synchronic implicational universals of the form 'If
P, then Q' by evolving *P & -Q co-occurrences. Such co-occurrences
trigger a chain of subsequent word order changes which re-introduce
consistency, as languages acquire whatever Q properties are observed
to co-occur with P on synchronic evidence.
(2) THE UNIVERSALCONSISTENCY HYPOTHESIS (UCH): At each stage in their
historical evolution, languages remain consistent with synchronic
universal implications.
These two hypotheses do not exhaust all the possibilities. Greenberg's word
order universals make crucial reference to 'basic' word order patterns. It is quite
possible for a language to have some SVO structures alongside basic SOV, or even
some VSO alongside basic SOV, or a minority of ADJ + N alongside N +ADJ; yet
Greenberg's implicational dependencies are defined almost exclusively with P and
Q as basic word orders. By contrast, word order change proceeds via a process of
'doubling'. A language which is developing from OV to VO does not suddenly
shift all its nominal objects from one side of the verb to the other. The incoming VO

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 621

exists first for a long period as a minority structure alongside older OV, gaining
gradually in frequency and grammaticalization until it replaces OV as the basic,
and perhaps eventually as the unique order, i.e. OV > OV/VO > VO. Now since
Greenberg's implicational universals are stated in terms of basic word orders they
do not, strictly speaking, make any predictions about incoming minority word
orders. Only when a minority P structure has achieved majority status does the
predictive power of synchronic word order universals come into effect in requiring
the co-occurrence of Q properties.
But if there is any reality to these synchronic word order dependencies, it is still
possible that they will translate into historical predictions about incoming minority
word order patterns in the course of change. If they do, and if we assume consistency
rather than violation in the course of change,2 then incoming minority P structures
should be accompanied by incoming minority Q structures; or else, minority
P structures will develop only after Q structures are already present (whether as
doublets with -Q or not). Thus minority P should be acquired either simulta-
neously with Q (-P & -Q > P & Q), or else after Q(-P & -Q > -P & Q >
P & Q), but not before it, in order to avoid innovating P structureswith no accom-
panying Q structures (-P &-Q > *P &-Q > P & Q). We might hypothesize
that minority P structures will not arise in the total absence of Q structures,
therefore:
(3) THE DOUBLINGACQUISITIONHYPOTHESIS
(DAH): Given a set of syn-
chronic implicational universals of the form 'If P, then Q', where P and
Q are basic word orders of certain specified types; then, at two
successive stages in the growth of a language,
IF: P is acquired as a doubling structurefrom the earlier uniquely
-P stage
THEN: EITHER Q must already be present at the earlier stage (whether
as a doublet with - Q or not), OR, if it is not present, Q
must be acquired as a doubling structure simultaneously
with P. But P will not be acquired in the total absence of Q.
This hypothesis refers to the first appearance of doubling structures at successive
stages in a language. But it makes no claims about the way in which minority P and
Q structures will further develop into basic structures. Since we now have reason-
ably extensive frequency studies of competing word orders in evolving IE languages,
we might use these in order to test the following, stronger, hypothesis:
(4) THE FREQUENCY (FIH): Given a set of synchronic
HYPOTHESIS
INCREASE
implicational universals of the form 'If P, then Q', where P and Q are
basic word orders of certain specified types; then, at two successive
stages in the growth of a language,
2
We could hypothesize that incoming minority P structureswill violate cross-language 'If
P, then Q' statements by developing minority P in the absence of Q, whereuponthis minority
P would trigger the subsequent acquisition of minority Q. However, there is no support for
this alternativein our data. And since hypotheses 3 and 4, which assume adherenceto the uni-
versals ratherthan violations, are confirmedby the data, we can save space by not formulating
the alternative hypotheses, merely noting instead that they are logically possible.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
622 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

IF: (a) there is an increase in the frequency of P structures


relative to their doublets - P, and if(b) the frequency of the
Q structures at the earlier stage, prior to the increase in P,
was less than 100%7(i.e., -Q doublets existed),
THEN:the implied Q structures will also have gained in frequency
along with P by the later stage.3
As it is formulated, the FIH predicts frequency increases both for newly-acquired
minority P and Q structures and for majority structures with less than 100%7
frequency at the earlier stage of comparison.
Note that hypothesis 3 follows logically from hypothesis 4, though not vice
versa. Imagine, in the extreme case, that a language at some stage in its history
has only -P & -Q structures, with no minority P or Q doublets. If, at the next
stage, that language develops some P structures (and has, say, 10% P to 90%
-P) then the FIH predicts that the percentage of Q structures can no longer be
zero; i.e., the acquisition of P doublets would imply in this case the acquisition of
Q doublets, which is what the DAH also requires. Conversely, imagine that, at the
next historical stage of inspection, there are 25% P structuresto 75% -P; then the
FIH predicts that the percentage of Q structuresmust also have increased, whereas
the DAH makes no predictions in this respect. After a presentation of some
synchronic implicational universals,I shall test each of these four hypotheses against
historical data.
2.1. SYNCHRONICWORD ORDER UNIVERSALS.An important prerequisite for
assessing the predictive potential of synchronic universals in a diachronic context is
that we first have an accurate statement of the cross-language synchronic word
order generalizations. In Hawkins 1978I argue that Greenbergmisses some interest-
ing word order patterns in his data. Furthermore,the reformulation of Greenberg's
universals by Lehmann and Vennemann introduces an intolerable number of
exceptions, misses more patterns, and misrepresents the careful balance between
non-statistical and statistical statements in Greenberg's work. Most of Greenberg's
3 Note that an increase in P's frequency relative to - P means, of course, a corresponding
decrease in -P's frequency. As a result, the FIH is equivalent to a requirement that any
decreasein -P's frequencymust be matchedby a decreasein - Q, where - Q is more than 0%.
However, althoughboth the DAH and FIH make predictionsfor the relativetimingof structures
ACQUIRED, no predictionsare being made by either hypothesis concerning structuresLOST. For
example, if P increasesto 100%(whereupon- P decreasesto 0% and is therebylost altogether),
the FIH requires only that Q should increase relative to - Q, not that Q should necessarily
increase to 100% (and -Q decreaseto 0%). Thus it is possible that -P could be lost before
-Q, and equally that -Q could be lost before -P. In the former case (i.e. P & -Q/Q), the
UCH would requirethat the - Q structuresco-existing with Q should not be basic, in order to
avoid a *P & - Q co-occurrence.The laws of word order loss are thus claimed to be partly
different from those of word order acquisition. This seems reasonable, since we know that
archaic word orders are often retained in a language to performcertain grammaticalfunctions
(cf. Li & Thompson's 1975 explanationfor the retardationof the SVO > SOV shift in Chinese).
By contrast, the actual innovation of a new word order pattern is here claimed to be deeply
determinedby the typological shape of the language at the historical stage in question, in the
form of the prior availability(or simultaneousacquisition) of universallyrequiredQ structures
as prerequisitesfor P's acquisition (where PO Q).

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 623

implicational statements are, contrary to current belief, formulated as NON-


STATISTICALgeneralizations.4 I believe that some exceptionless implicational uni-
versals can be formulated to cover the extensive data of Greenberg's Appendix II;
and in ?7 below I shall outline a distributionalgeneralizationpredicting the relative
quantities of languages of the attested types. There are 24 mathematically possible
co-occurrences of VSO/SVO/SOV, PREP/POST, N+GEN/GEN+N, and N+
ADJ/ADJ+N in Greenberg; of these, only 15 are attested. My implicational
universals collectively predict this balance. Their exceptionlessness is made possible
by the use of multi-valued correlations. Instead of correlating just two properties
with one another, as Greenberg generally does (if a language has some single word
order P, then it also has some single word order Q), my statements involve at least
three properties: e.g., if a language has some word order P, and if it has word order
Q, than it will also have word order R. By, in effect, increasing the conditioning
property P from one property to at least two, it becomes possible to predict the
co-occurrence of a third property in a straightforwardway, and to reveal significant
generalizations about co-occurring word orders.
It seems significant that 9 out of 24 possible word order co-occurrences in
Greenberg's large sample are totally unattested. Since we want to define as restric-
tively as possible the attested word order co-occurrence patterns, and to distinguish
them from the unattested ones, this fact is compatible with the use of stronger,
exceptionless universals, rather than with tendency-based statistical statements.
These latter necessarilyallow the offending *P & - Q language types to exist, albeit
as a minority type, even though no such languages have been found.5 Yet it is a
familiar finding in word order studies that the attested co-occurrence types are
significantly fewer than the mathematically possible co-occurrences. Thus Steele
1975 finds, using a genetically broad sample of languages, that the only positions
which a modal element can occupy in SVO, VSO, and SOV languages are the
following:
(5) SModVO
ModVSO VModSO
SOVMod SModOV ModSOV
There are four mathematically possible positions which a modal can occupy in
each of these three language types (e.g. ModSVO, SModVO, SVModO, SVOMod
4 The implicationaluniversalspresentedas NON-STATISTICALby Greenbergare the ones which
he numbers as follows (where a universalembodies two distinct claims, I distinguish them by
'a' and 'b'): 3, 5, 6, 7, 8, 11, 12, 13, 15, 16, 19b, 20, 21, 22a, 23a, 24, 25, 26, 27, 28, 29, 30, 31,
32, 33, 34, 35a, 36, 37, 38, 40, 43, 44, 45. The following are given as STATISTICAL:2, 4, 9, 10, 17,
18, 19a, 22b, 23b, 35b, 39, 41. The only universalsnot included in these two lists are three non-
implicational (or absolute) universals: 1, 14, 42.
5 In the event that exceptions are discovered to some implicational universal, then the
universal must be reformulatedto account as generallyas possible for the attested word order
co-occurrences, while ruling out the unattested ones; or else, if the exceptions are extremely
limited, we might simply change the status of the universal from non-statistical to statistical.
However, the over-use of tendency-basedstatements has resulted in the loss of many syn-
chronic generalizations,as I demonstratein this paper; and it has also given apparentlegitimacy
to a form of historical explanation, the trigger-chaintheory, which (as I hope to show) is both
empiricallyand theoreticallyillegitimate.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
624 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

in SVO languages); but only 6 of the 12 possibilities are actually attested. And more
generally, if we calculate the possible combinations of all verb positions (within the
sentence) with all noun positions (within the noun phrase), and if we then add in
the adposition orders (within the adposition phrase) and adjective orders (within the
adjective phrase), the mathematical co-occurrence posibilities that the word orders
of a whole language might potentially exhibit are quite enormous. Yet the attested
word order patterns that we find whole languages actually exhibiting are not nearly
so large. In fact, both my own research, and my reading of the research of others,
have convinced me that, out of the total number of these mathematical co-
occurrence possibilities, many more areuN-attested than attested. Such consider-
ations provide a strong motivation for avoiding statistical in favor of non-statistical
universals.
By formulating exceptionless implications for the available synchronic word
order data (or else nearly exceptionless implications where necessary or desirable,
e.g. in order to avoid complexity), it becomes a simple matter to test whether or not
the available diachronic data violate these synchronic implications, and whether or
not incoming minority (and majority) doublets obey the predictions made in terms
of basic structures. To this we shall now turn.
2.2. IMPLICATIONAL PATTERNS. The distribution of basic order types as given in
Greenberg's Appendix II is as follows:6
1. VSO/PREP/N + GEN/N + ADJ. Celtic languages; Hebrew, Aramaic, Arabic, Ancient
Egyptian, Berber; Nandi, Masai, Lotuko, Turkana,Didinga; Polynesian languages
6 What is important about Greenberg'sAppendix II is the wide genetic and areal coverage.

Many of the entriesrefer to whole families(e.g., Celtic, Bantu, Dravidian),and the individually
named languages are geneticallydiverse; the result is that the majorityof the world's language
families are represented.In a seminaron word order which I organizedat USC in the Spring of
1978, one of the collective projectswas to check as many as possible of these entries,particularly
the families and groups, and to add to the sample. Our researchprovided strong confirmation
for the implicationalpatternsderivablefrom Appendix II. We could find NO counter-examples
to the non-statistical implications(I), (II), (III') and (IV), although we did find it necessaryto
reclassifysome of Greenberg'sentriesamong the alreadyattested languagetypes. Ourexpanded
sample was compiled and classified too late to be included in the present paper, but I have
incorporatedthe following changes into the Appendix as reproducedhere:
Type 9: 'almost all Bantu languages' replaces'all ... ' (Tunen and Bandem are type 17;
Larry Hyman, p.c.) Type 10: Modern Greek was mistakenlyclassifiedas type 9 (contrary
to the classificationin Greenberg'sown 30-languagesample). Type 15: Rutulian and other
Daghestan languages in the Caucasus have been moved from type 14 (ManerThorpe, p.c.)
Type 23: Ijo has been moved from type 15 (cf. Givon 1975).Type 24: most Mandelanguages
have been moved from type 16 (cf. Givon 1975 and LarryHyman, p.c.)
I have also deleted the following entries in which verb, adposition, or adjective order are of
undecidablebasicness:
German (type 10) has both SOV and SVO orders (cf. fn. 22). Chinese (type 15) also has
both SOV and SVO (cf. Li & Thompson 1975). Nupe (type 16) is both pre- and post-
positional (Larry Hyman, p.c.) Burmese(type 24) has an indeterminatebasic adjective
order (Jack Roberson, p.c. cf. also Greenberg1966:108, note 1 to AppendixI).
Since my synchronicimplicationaluniversalsare definedin termsof basic word orders, 'If a lan-
guage has basic orderP, then it also has basic order Q', they make no predictionsfor languages
which do not clearly have basic order P. Importantdiachronicpredictionsare, however, made
for these mixed word order languages by the DAH and F1H (cf. ?4), and these predictionsare

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 625

and other Austronesian languages; Chinook, Tsimshian; Zapotec, Chinantec,


Mixtec, and other Oto-Manguean languages.
2. VSO/PREP/N+ GEN/ADJ + N. Tagabili and probablyother PhilippineAustronesian
languages; Kwakiutl, Quileute, Xinca.
3. VSOIPREP/GEN+ N/ADJ + N. Milpa Alta Nahuatl.
4. VSO/PREP/GEN+ NN + ADJ. No examples.
5. VSO/POST/N+ GEN/N + ADJ. No examples.
6. VSOIPOST/N+ GEN/ADJ + N. No examples.
7. VSO/POSTIGEN+ N/ADJ + N. Papago.
8. VSO/POST/GEN+ N/N + ADJ. No examples.
9. SVO/PREP/N + GEN/N+ADJ. Romance languages, Albanian; West Atlantic
languages, Yoruba, Edo group, most languages of Benue Congo group including
almost all Bantu languages; Shilluk, Acholi, Bari, most languages of Chad group
of Hamito-Semitic but not Hausa; Neo-Syriac, Khasi, Nicobarese, IKhmer,
Vietnamese, all Thai languages except Khamti; many Austronesian languages
including Malay; Subtiaba.
10. SVOIPREPIN+ GEN/ADJ/N. Modern Greek, Dutch, Icelandic, Slavonic, Efik,
Kredj, Maya, Papiamento.
11. SVO/PREPIGEN+ N/ADJ + N. Swedish, Danish.
12. SVOIPREP/GEN+N/N + ADJ. Arapesh (New Guinea).
13. SVO/POST/N+ GEN/N + ADJ. No examples.
14. SVO/POST/N+GEN/ADJ+N. No examples.
15. SVO/POST/GEN+ N/ADJ+ N. Finnish, Estonian, Algonquian, Zoque, Rutulian,
and other Daghestan languages in the Caucasus.
16. SVO/POST/GEN+N/N+ADJ. Most Voltaic languages, Kru, Twi, Ga, Guang, Ewe,
Songhai, Tonkawa, Guarani.
17. SOVIPREPIN+GEN/N+ADJ. Persian, Iraqw (Cushitic), Khamti (Thai), Akkadian.
18. SOV/PREP/N + GEN/ADJ + N. No examples.
19. SOVIPREP/GEN+ N/ADJ+ N. Amharic.
20. SOVIPREP/GEN+ NIN + ADJ. No examples.
21. SOV/POST/N+GEN/N+ADJ. Sumerian, Elamite, Galla, Kanuri, Teda, Kamilaroi
and other southeasternAustralian languages.
22. SOV/POST/N+ GEN/ADJ + N. No examples.
23. SOV/POST/GEN+ N/ADJ + N. Hindi, Bengali, and other Aryan languagesof India;
Modern Armenian, Finno-Ugric except Finnish group; Altaic, Yukaghir, Paleo-
Siberian, Korean, Ainu, Japanese, Gafat, Harari, Sidamo, Chamir, Bedauye,
Nama Hottentot; Khinalug, Abkhaz and other Caucasian languages; Burushaski,

derivedfrom languages which do have clear basic word orders on currentsynchronicevidence.


Thus, the clear cases synchronically, which our research suggests are the majority, make
precise predictions for both non-basic and basic orders in diachrony (and hence also for their
synchronicreflexes; but cf. fn. 3). In addition, Norwegian has been deleted from type 11; and
Papago (POST & VSO) has been entered into type 7 following the information given in
Greenberg'sadditional note (p. 107). Ed Keenan has confirmedfor me (p.c.), on the basis of his
work on verb-initiallanguages, that POST & VSO is indeed an existing, though limited, type.
Its occurrenceis in fact predictedby the existence of languages of the opposite type: PREP &
SOV (types 17 and 19). The limited distributionof both POST & VSO and PREP & SOV is
predictedby the principleof Cross-CategoryHarmony(cf. ?7). The smallernumberof POST &
VSO languages relative to PREP & SOV languages follows from the smaller number of VSO
languagesin general,comparedto SOV languages(197%of our expandedsample are VSO, 47%7
SOV). The chances of a limited distributiontype surfacing in a random sample are, therefore,
much less for VSO than for SOV. Finally, our expanded sample permitsme to delete the word
'probably' occurring before: 'other Austronesian languages' (type 1); 'other Philippine
Austronesian languages' (type 2), and 'Algonquian' (type 15).

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
626 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

Dravidian; Newari and other Sino-Tibetan languages; Marind-Anim, Navajo,


Maidu, Quechua; Ijo.
24. SOV/POST/GEN+N/N+ADJ. Basque, Hurrian, Urartian, Nubian, Kunama, Fur,
Sandawe, Lushai, ClassicalTibetan, Makasai, Bunak (Timor), Kate (New Guinea),
most Australian languages, Haida, Tlingit, Zuni, Chitimacha, Tunica, Lenca,
Matagalpa, Cuna, Chibcha, Warrau, most Mande languages.

2.21. Consider first SOV languages (types 17 through 24). The adjective and
genitive co-occurrences permit the following implication:
(6) If the adjective precedes the noun, then the genitive precedes the noun;
i.e., ADJ+NN GEN+N.
SOV languages possess the following co-occurrences:
(7) ADJ+N & GEN+N (P & Q): types 19 and 23.
N+ADJ & GEN+N (-P & Q): type 24.
N+ADJ & N+GEN (-P & -Q): types 17 and 21.
*ADJ + N & N + GEN (*P & -Q): no examples (potentially types 18 and
22).
Our first implicational universal is, therefore:
(I) If a language has SOV word order, and if the adjective precedes the noun,
then the genitive precedesthe noun; i.e., SOV v (ADJ + N v GEN + N).
2.22. VSO languages (types 1 through 8) exhibit the exact mirror-imagepattern:
(8) If the adjective follows the noun, then the genitive follows the noun;
i.e., N+ADJ - N+GEN.
VSO languages possess the following co-occurrences:
(9) N+ADJ & N+GEN (P & Q): type 1.
ADJ+N & N+GEN (-P & Q): type 2.
ADJ+N & GEN+N (-P & -Q): types 3 and 7.
*N + ADJ & GEN + N (*P & - Q): no examples (potentially types 4 and 8).
Our second implicational universal is, therefore:7
(II) If a language has VSO word order, and if the adjective follows the noun,
then the genitive follows the noun; i.e., VSO v (N+ADJ D N+ GEN).
7 Note that SVO languages pattern very differentlyfrom VSO languages. The co-occurrence
*N+ADJ & GEN+N, which is unattested with VSO, is attested in SVO languages (types 12
and 16). Similarly, SVO languages exhibit the co-occurrence *ADJ+N & N+GEN (type
10) which is ruled out by universal (I) for SOV languages. SVO languages are therefore not
constrained with respect to their noun modifier co-occurrencesas VSO and SOV languages
are. This arguesagainstcollapsing VSO and SVO into a common type, VO, as in the Lehmann-
Vennemannreformulationof Greenberg'suniversals.
Furthermore,in addition to languages with basic VSO, there are languages with basic VOS,
e.g. Malagasayand Gilbertese(CharlesRandriamasiamanana,personal communication),and
also languageswhich are best classifiedas verb-initialon account of the frequencyof both VSO
and VOS, e.g. Fijian, Samoan, and Tongan (cf. Hsieh 1976). The five languages illustratedall
have PREP & N+ADJ & N+ GEN, and so satisfy one of the co-occurrencepossibilities for
VSO languagesas definedby universal(II). We can thereforeconvert (II) into the more general
statement (Il') by changing the antecedentpropertyVSO into V-initial:
(II') If a language has verb-initialorder, and if the adjective follows the noun, then the
genitive follows the noun; i.e., V-initial - (N+ADJ = N+GEN).

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 627

2.23. The property PREP seems to have, throughout Greenberg's appendix, the
same conditioning effect on the co-occurrence of adjective and genitive orders as
does VSO. Languages with PREP have the following co-occurrences:
(10) N+ADJ & N+GEN (P & Q): types 1, 9, and 17.
ADJ+N & N+GEN (-P & Q): types 2 and 10.
ADJ+N & GEN+N (-P & -Q): types 3, 11, and 19.
*N +ADJ & GEN+ N (*P & -Q): no examples of types 4 and 20.
However, SVO languages show just one isolated example of N + ADJ & GEN + N
co-occurring with PREP (Arapesh, type 12). Bearing in mind this single counter-
example, we have the following (statistical) universal:8
(III) If a language has PREP order, and if the adjective follows the noun, then
the genitive follows the noun; i.e., PREP v (N+ADJ > N+GEN).
This implicational universal generates interesting predictions in conjunction with
implications (I) and (II). Thus implication (I) permits an SOV language to have
ADJ+N & GEN+N, N+ADJ & GEN+N, or N+ADJ & N+GEN. The first
and third of these co-occurrences, but not the second, overlap with the permitted
co-occurrences of the mirror-image implication N+ADJ v N+GEN within
universal (III), as shown in Table 1.

IMPLICATION
(I) IMPLICATION
(III)
ADJ+N & GEN+N ADJ+N & GEN+N
N+ADJ & GEN+N *N+ADJ & GEN+N
N+ADJ & N+GEN N+ADJ & N+GEN
*ADJ+N & N+GEN ADJ+N & N+GEN
TABLE1.

It follows that, if a language has both SOV and PREP, then it can have neither of
the starred co-occurrences. This is what we find. The only two attested SOV &
PREP co-occurrences are types 17 and 19, which have N+ADJ & N+GEN and
ADJ + N & GEN + N, respectively.
Universals (II) and (III) both define the same noun modifier co-occurrences
(N + ADJ v N + GEN), and hence do not further restrict the adjectiveand genitive
co-occurrences for a language with both VSO and PREP.
2.24. The property POST exerts the same co-occurrence requirements on
adjective and genitive orders as does SOV. Languageswith POST have the following
co-occurrences:
(11) ADJ+N & GEN+N (P & Q): types 7, 15, and 23.
N+ADJ & GEN+N (-P & Q): types 16 and 24.
8
A non-statistical version would be
(III) If a language has PREP order and either VSO or SOV, and if the adjective follows
the noun, then the genitive follows the noun; i.e., PREP & (VSO V SOV) D
(N + ADJ D N+GEN).
Since (III) has only the single exception, (III') would need to be supplemented with a distri-
butional statement to the effect that the odds against even an SVO language combining PREP
with *N+ADJ & GEN+N are extremely high. I shall employ (III) rather than (III') in the
main text because of its greater simplicity.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
628 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

N+ADJ & N+GEN (-P & -Q): type 21.


*ADJ + N & N+ GEN (*P & -Q): no examples (potentially types 6, 14,
and 22).
We therefore have the following universal:
(IV) If a language has POST order, and if the adjective precedes the noun,
then the genitive precedes the noun; i.e., POST - (ADJ+N -
GEN+ N).
This universal defines the same noun modifier co-occurrences as universal (I), and
hence does not restrict the range of possibilities in a language with both SOV and
POST. The collective predictions of universals (IV) and (II), however, are shown in
Table 2.
IMPLICATION
(II) IMPLICATION
(IV)
N+ADJ & N+GEN N+ADJ & N+GEN
ADJ+N & N+GEN *ADJ+N & N+GEN
ADJ+N & GEN+N ADJ+N & GEN+N
*N+ADJ & GEN+N N+ADJ & GEN+N
TABLE2.

If a language has both VSO and POST, therefore, it cannot have either of the
starred co-occurrences. It is significant that the one attested VSO & POST type
(type 7) has one of the permitted orders, ADJ+N & GEN+N.9
2.25. For our fifth implication we must go beyond Greenberg's Appendix II.
It seems that there is an implicational relationship between N + GEN and N + REL
in prepositional languages (cf. in particular Antinucci et al. 1976 and Peranteau
et al. 1972 for relative clause data). Co-occurrences in PREP languages are as
follows:
(12) N+GEN & N+REL (P & Q): types 1, 2, 9, 10, 17.
GEN+N & N+REL (-P & Q): types 3 and 11.
GEN+N & REL+N (-P & -Q): type 19.
*N+GEN & REL+N (*P & -Q): no examples.
This yields the implication:
(V) If a language has PREP order, and if the genitive follows the noun, then
the relative clause follows the noun; i.e., PREP - (N+GEN v
N+REL).
9 The other permitted order N+ADJ & N+GEN is therefore predicted to be a possible
co-occurrence with VSO & POST (type 5), even though unattested in a sample of this size
(recall fn. 6). Type 13 (SVO & POST & N+ ADJ & N + GEN) is also predictedto be a possible
postpositional language type, although unattested for the same reason as type 5. Of the nine
unattestedco-occurrencesin Appendix II, therefore,seven are ruled out by our non-statistical
implications: Type 4, by implications (II) and (III'); Type 6, by implication (IV); Type 8, by
implication (LI); Type 14, by implication (IV); Type 18, by implication (I); Type 20, by
implication(III'); and Type 22, by implications(I) and (IV). One furthertype (type 12) is ruled
out by the statisticalimplication(III). This makes eight excludedco-occurrencesin all, of which
seven are unattested, while one (type 12) is marginallyattested. The other (fourteen) attested
co-occurrencesare all permittedby universals(I)-(IV).

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 629

Implication (III) established the validity of N+ADJ = N+GEN for pre-


positional languages, and implication (V) established N+GEN - N+REL also
for prepositional languages. This yields an implicational chain, or hierarchy:
(N + ADJ v N + GEN) & (N + GEN v N + REL). I shall refer to this as the Pre-
positional Noun Modifier Hierarchy (PrNMH). The PrNMH defines the follow-
ing four (out of eight mathematically possible) co-occurrence possibilities:'?
(13) a. N+ADJ & N+GEN & N+REL: types 1, 9, and 17.
b. ADJ+N & N+GEN & N+REL: types 2 and 10.
c. ADJ+N & GEN+N & N+REL: types 3 and 11.
d. ADJ+N & GEN+N & REL+N: type 19.
2.26. The fourth co-occurrence possibility defined by the PrNMH is very rare in
prepositional languages. This is the result of our sixth implication. Only in SOV
languages does PREP co-occur with REL+ N (type 19), and even here there is only
a single entry, Amharic.1 Our sixth implicational universal therefore reads:
(VI) If a language has PREP order and either VSO or SVO, then it does not
have REL+N; i.e., PREP & (VSO V SVO) - not REL+N.
3. TESTINGTHEUVH AND UCH. According to the Universal Violation Hypo-
thesis, synchronic implicational universals provide a momentum for change
following some initial (and independently explained) violation. By contrast, the
Universal Consistency Hypothesis claims that universal violation is not a valid
mechanism of change, and that languages in evolution will remain consistent with
independently established universal laws. We might test these two hypotheses by
investigating the earliest IE daughter languages. Do the earliest records violate the
six synchronic implications of ?2, or do they not? If they do, the UVH becomes
plausible, and we might investigate subsequent developments in the respective
languages to see if any later chain of changes proceeds as predicted by the initial
violation trigger. But if they do not, there is no evidence for the UVH, and the UCH
is supported.
Friedrich gives a careful account of word order patterns in the earliest IE dialects,
including all those which figure in our six synchronic implications. There are many
instances of doubling structures in the early dialects (N+ADJ together with
ADJ + N, N + GEN together with GEN + N, etc); and Friedrich discusses both the
relative frequencies and the grammatical differences between competing orders
before deciding which is basic. In a small number of cases, he concludes that two
orders are equally basic. Friedrich's findings are summarized in Table 3, with the
most important page references given with each entry. All subfamilies are repre-
sented except for Germanic, which is discussed in detail in ??4.3-4.5.
Consider first the relative orders of the noun modifiers: the adjective, genitive,
10 The
eight mathematically possible co-occurrences are:
N+ADJ & N+GEN & N+REL N+ADJ & GEN+N & N+REL
ADJ+N & N+GEN & N+REL N+ADJ & N+GEN & REL+N
ADJ+N & GEN+N & N+REL N+ADJ & GEN+N & REL+N
ADJ+N & GEN+N & REL+N ADJ+N & N+GEN & REL+N
11It is interesting in this regard that Amharic, according to Little 1977, is moving historically
toward postpositions and away from prepositions.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
630 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

YOUNGER
CELTIC(pp. 58-9) SLAVIC(P. 62) AVESTAN(p. 45)
/N + ADJ (adj + n)/N + ADJ adj + n/N + ADJ
/N + GEN (gen + n)/N + GEN gen+n/N+GEN
/N + REL /N + REL (rel + n)/N+ REL
/PREP /PREP (post)/PREP
rsvo)
/VSO sov/VSO SOV/{ osv
Covs)
ALBANIAN OLD ARMENIAN HOMERIC GREEK
(pp. 25-6) (PP. 41-2) (pp. 23-6)
ADJ+N/ ADJ + N/n + adj ADJ+N/n+adj
/N+GEN gen+n/N+GEN gen + n/N + GEN
/N + REL /N + REL (rel + n)/N + REL
/PREP /PREP /PREP
/SVO vso/SVO sov/svo
ITALIC OLD PERSIAN TOCHARIAN
(pp. 52-8) (P. 43) (pp. 49-50)
ADJ + N/n + adj ADJ + N/n + adj ADJ + N/
gen + n/n + gen GEN + N/n + gen GEN + N/
(rel + n)/N+REL (rel + n)/N + REL rel+n/N+REL
/PREP (post)/PREP POST/prep
SOV/svo SOV/svo SOV/
GATHIC AVESTAN SANSKRIT HITTITE
(pp. 44-5) (pp. 26,29,33,37) (pp. 25,26,29,36)
adj + n/n + adj ADJ+N/n+adj ADJ + N/
GEN+N/n+gen GEN + N/n + gen GEN + N/n + gen
rel+n/N+REL REL+N/n+rel REL + N/n + rel
post/PREP POST/prep POST/
{svo}
SOV/< osv S SOV/ SOV/
LovsJ
TABLE 3. Indo-European word order data (from Friedrich 1975). Capital letters refer to word
orders which are basic or unique. Lower-case letters refer to non-basic orders; those in paren-
theses designate word orders which are even more limited.12 Word orders written to the right
of the slash are operand-before-operator orders (as defined by Vennemann; see references);
those to the left of the slash are operator-before-operand orders. For word orders involving the
verb, the BASICverb position has been situated according to this criterion; but any non-basic
verb positions are then placed on the opposite side of the slash, regardless of their operator and
operand ordering.

and relative clause. Of the eight mathematically possible co-occurrences of these


three modifiers, only four basic order co-occurrences are attested in IE. It is
significant that these are exactly the four defined by the PrNMH of ?2.25.13Table 4
summarizesthe four classes, ignoring now the minority word orders for the sake of
clarity.
12 This summary does not aim for completeness in presenting the minority word orders of the

respective languages. Thus, the fact that some minority word order is not listed under a
particular entry does not mean that it is totally unattested.
13 This fact is most suggestive for the reconstruction of Proto-Indo-European. Any recon-
struction is problematic when, as in the present instance, the earliest daughter languages are
fairly evenly divided among certain variation types. The age of the respective daughter lan-

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 631

CLASS 1 CLASS 2 CLASS 3 CLASS 4


/N + ADJ ADJ + N/ ADJ + N/ ADJ + N/
IN + GEN /N + GEN GEN + N/ GEN + N/
/N + REL /N + REL /N + REL REL + N/
Celtic Albanian Old Persian Sanskrit
(VSO & PREP) (SVO & PREP) (SOV & PREP) (SOV & POST)
Slavic Old Armenian Tocharian Hittite
(VSO & PREP) (SVO & PREP) (SOV & POST) (SOV & POST)
Younger Avestan Homeric Greek
(SOV & PREP) (sov/svo & PREP)
TABLE 4.

guages, the possible influence of neighboring languages, and morphological pointers to


earlier syntactic order can then all be taken into account. But perhapsmost importantof all,
whichevervariation type is chosen as the prototype should be that which is most compatible
with all the subsequentvariation. In this case, the earliest IE noun modifierco-occurrencesare
exactly those definedby implicationaluniversals(III), PREP D (N + ADJ D N + GEN), and (V),
PREP D (N+GEN D N + REL). If we assume that the parent language had prepositions, we
therefore account naturally for this variation. The daughter languages have simply divided
themselves up into the four available noun modifier co-occurrenceoptions. But if we assume
that the parentlanguagehad postpositions, we fail to accountfor the following ratherinteresting
fact. In general, postpositional languagescan have noun modifierco-occurrenceswhich contra-
dict those permittedfor prepositionallanguages, such as N+ADJ & GEN+N. According to
universal (IV), POST D (ADJ + N D GEN + N), the co-occurrence N + ADJ & GEN + N is
well-formed, and occurs in many postpositional languages, e.g. Basque (SOV & POST),
Chibcha(SOV & POST), Guarani(SVO & POST),and Songhai(SVO & POST).Similarly,many
postpositional languageshave the co-occurrenceN + ADJ & REL + N, which is also unattested
in prepositional languages; Basque, Chibcha, and Guarani are examples, while Songhai has
N+ADJ & N+REL. Thus, postpositional languages are attested with these over-all co-
occurrences:
N+ADJ & GEN+N & REL+N
N+ADJ & GEN+N & N+REL
But neither of these is permittedby the prepositionalimplications.If we assume that PIE had
postpositions, it therefore becomes an accident that the earliest postpositional daughter lan-
guages exhibit ONLYnoun modifier co-occurrenceswhich overlap with those permittedby the
prepositional implications (III) and (V), namely ADJ+N & GEN+N & N+REL (class 3 of
Table 4) and ADJ + N & GEN + N & REL + N (class 4); and that they have no co-occurrences
which are uniquely characteristicof postpositional languages. However, if we assume that PIE
had prepositions, then not only do we account exactly for the four noun modifier variation
types which are attested, but we also have a ready explanation for the noun modifier co-
occurrences of the postpositional IE languages. These languages, which are in the minority,
have changed from prepositionsto postpositions at preciselythat point in their evolution when
their noun modifierco-occurrenceswerecompatiblewith both prepositionsand postpositions-
and when postpositions were in fact to be preferred,judging by the quantitative synchronic
evidence of Greenberg'ssample: the co-occurrenceADJ+N & GEN+N, which is shared by
all the postpositional IE daughters,occurs synchronicallymuch more often with postpositions
than with prepositions, even though both are permittedby co-occurrences(cf. ?7). In this way
the postpositional IE languagesremain consistent with all the universalimplications(and with
synchronic quantitativepredictions). Such consistency in the course of change is supportedby
abundant evidence in my main text.
But the argument that PIE was prepositional now has important consequences for the
reconstruction of PIE verb position. I calculate that only 87o of prepositional languages in
Greenberg'sAppendix II co-occur with SOV word order. Hence, if PIE was prepositional, the
chances that PIE was an SOV language (cf. Lehmann 1974) are less than 1 in 10 on current
synchronicevidence; either SVO or VSO is much more plausible.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
632 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

ITALIC GATHIC AVESTAN


ADJ + N/ adj + n/n + adj
gen + n/n+ gen GEN + N/
/N + REL /N + REL
SOV & PREP SOV & PREP
TABLE5.

Two entries from Table 3 not listed in Table 4, Italic and Gathic Avestan, are
shown in Table 5.
Italic has no basic genitive order, and Gathic Avestan no basic adjective order.
In effect, Italic is a mixture of classes 2 and 3, since it has ADJ+ N & N+ REL,
while having both pre- and post-nominal genitives. And Gathic Avestan is closest
to class 3, the only class having GEN + N & N + REL. I shall consider these equally
basic orders in relation to the synchronic predictions.
Let us now check each of the early IE dialects to see how they fare against the
six synchronic implicational universals of ?2.2. I repeat the universals for con-
venience:

(I) SOVD (ADJ + N DGEN+N)


(II) VSOD (N + ADJ N +GEN)
(III) ' (N + ADJ
PREP ' N + GEN)
(IV) POSTD(ADJ+NDGEN+N)
(V) PREPD (N+ GEN DN+REL)
(VI) PREP & (VSOV SVO) D not REL + N

Below I list, for eachIE language or subfamily, whatever synchronic implications


make relevant co-occurrence predictions for that entry. For example, Celtic and
Slavic both have VSO & PREP. By universal (II), since they have VSO, then if they
have N+ADJ, they must have N+GEN. They do have N+ADJ, so they are
predicted to have N + GEN as well, which they do. Since they have PREP, then by
universal (III) they are again required to have N+GEN, given that they have
N + ADJ. By universal (V), the presence of PREP and N+ GEN conditions the co-
occurrence of N + REL. This prediction is fulfilled. By universal (VI), PREP & VSO
cannot co-occur with REL+N, and this also holds.
Consider first the Class 1 languages:

(14) Celtic and Slavic (VSO & PREP & N+ADJ & N+GEN & N+REL):
(II): no violation.
(III): no violation.
(VI): no violation.
(VI): no violation.

(15) Younger Avestan (SOV & PREP & N+ADJ & N+GEN & N+REL):
(I): no violation (N +ADJ & N+ GEN is one of the three co-occur-
rences permitted by ADJ + N v GEN + N).
(III): no violation.
(V): no violation.
(VI): no violation.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 633

Consider next the Class 3 languages:


(16) Old Persian (SOV & PREP & ADJ + N & GEN + N & N + REL):
(I): no violation.
(II): no violation.
(V): no violation (GEN + N & N+ REL is one of the three co-occur-
rences permitted by N + GEN v N + REL).
(VI): no violation.
(17) Tocharian (SOV & POST & ADJ+N & GEN+N & N+REL):
(I): no violation.
(IV): no violation.
The Class 4 languages have identical basic word orders:
(18) Sanskrit and Hittite (SOV & POST & ADJ + N & GEN + N & REL + N):
(I): no violation.
(IV): no violation
Consider now the Class 2 languages:
(19) Albanian and Old Armenian (SVO & PREP & ADJ+N & N+GEN &
N+ REL):
(V): no violation.
(VI): no violation.
(III): does not necessarily apply to SVO & PREP languages (since there
is just one exception in Greenberg's appendix); yet Albanian
and Old Armenian are still consistent (having ADJ+ N & N+
GEN) with the implication PREP - (N+ADJ v N+GEN).
(20) Homeric Greek (sov/svo & PREP & ADJ + N & N + GEN & N + REL):
(V): no violation.
(VI): does not strictly apply, because of the absence of SVO as a basic
word order; yet Greek is consistent, having N+REL.
(III): does not strictly apply; yet Greek is consistent with PREP v
(N+ADJ v N+GEN).
What emerges is that, for the basic word order co-occurrences of the ten entries
in Table 4, not a single implicational universal derivablefrom Greenberg'sappendix
is violated. Evidently, impossible word order co-occurrence types, on current
synchronic evidence, are also impossible in diachrony. And whatever subsequent
word order changes these IE languages underwent are NOT attributable to the re-
introduction of universal consistency, because there was no violation to begin with.
Consider now Italic and Gathic Avestan. Italic was excluded from Table 4, since
it has no basic genitive order, and Gathic Avestan since it has no basic adjective
order. How are we to interpret these equally frequent genitive and adjective orders
in the light of the synchronic implications? It is perhaps most plausible to argue
that neither order is basic in these cases. With this interpretation, Italic in fact
exhibits a violation, though Gathic does not:
(21) Italic (SOV & PREP & ADJ+N & gen+n/n+gen & N+REL):
(I): violation (since basic ADJ+N requires basic GEN+N, yet
neither gen + n nor n + gen is basic).

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
634 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

(III): no violation (only basic N+ADJ requires basic N+GEN, and


Italic has ADJ + N).
(V): no violation (N+REL may occur whether a language has basic
N + GEN or not).
(22) Gathic Avestan (SOV & PREP & adj+n/n+adj & GEN+N &
N + REL):
(I): no violation (SOV does not necessarily require the co-occurrence
of basic ADJ + N).
(III): no violation (since n + adj is not basic, N + GEN is not required).
(V): no violation.
However, the significance of the inconsistency of Italic with respect to one implica-
tion depends crucially on this interpretation of equally frequent word orders. If we
were to adopt a different interpretation-allowing, e.g., that either of the equally
frequent word orders could be regarded as basic, then neither Italic nor Gathic
Avestan would exhibit any violation.l4
It seems, therefore, that the UVH receives no convincing support from Friedrich's
analysis of the early IE dialects, and that the UCH is the correct theory. And
without universal violations in the early stages of these subfamilies, the whole logic
of trigger-chain theories falls.

4. TESTINGTHEDAH AND FIH. The above validation of the UCH indicates that
the synchronic implications of ?2 are strong constraints on languages in evolution.
Evidently, languages will change historically only within the parameters of syn-
chronically attested variation types. This finding should encourage us to see whether
even finer predictions for language change can be made, assuming consistency
rather than violation. We can accordingly test the Doubling Acquisition Hypo-
thesis and the Frequency Increase Hypothesis, both of which impart to incoming
doubling structuresthe same co-occurrence implications which hold between basic
word orders on synchronic evidence.

4.1. DOUBLING IN AVESTAN. The facts from Avestan (Iranian branch) are
interesting in this regard (cf. Friedrich, 44-5). Records of Gathic Avestan date
from around 900 B.C., while records of Younger Avestan are probably from the 5th
or 4th century B.C.-several hundred years later, at any rate. The quantities for
competing word orders in Friedrich's sample are shown in Table 6.
In ?2.25, we established the Prepositional Noun Modifier Hierarchy: (N + ADJ =
N + GEN) & (N + GEN v N + REL). It is interesting that the percentage increases
from Gathic to Younger Avestan involve prepositions and all three postnominal
modifiers. Thus the ratio of prepositions to postpositions jumps from 3/1 to 6/1.
The equal distribution of adjective orders goes in favor of N+ADJ, the minority
14
Thus, if gen+n could count as basic in Italic, this would save implication (I), SOV O
(ADJ+N D GEN+ N). And if either adj+n, or neither adj+n nor n+adj, count as basic in
Gathic Avestan, this will save implication (III), PREP O (N + ADJ O N + GEN). if n + adj were
basic, this would create a *P & -Q co-occurrence with GEN+N. The major point about all
the early IE dialects, however, is that there is not a single clear counter-example in which a basic
P structure co-occurs with a basic - Q structure.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 635

GATHICAVESTAN (SOV) YOUNGER AVESTAN (SOV)


adj + n/n + adj adj+ n/N + ADJ
21 20 28 38
GEN + N/n + gen gen n/N GEN
+ +
27 20 8 11
rel + n/N + REL (rel+ n)/N + REL
11 20 2 13
post/PREP (post)/PREP
4 12 4 26
TABLE6.

n+gen becomes the majority N+GEN, and the older dominant N+REL (by
2/1) increases its dominance to 6/1, almost removing rel+n from the language.
This is suggestive evidence for the FIH, which requires that, if there is a universal
implication P >Q, then any increase in P (relative to -P doublets) should be
accompanied by increases in Q (relative to -Q doublets, if any). Given that
prepositions and postnominal adjectives increase in frequency from Gathic to
Younger Avestan, it follows from the FIH, and from the implications (N + ADJ v
N+GEN) & (N+GEN v N+REL), that postnominal genitives and relative
clauses must also increase in frequency-which they do.15
4.2. SOMEMORESYNCHRONIC DATA. Before considering more diachronic pre-
dictions for change, from Germanic, let us add more noun modifier properties to the
15
Actually, the interpretationof the FIH in relation to universals(III), PREP O (N +ADJ D
N+GEN), and (V), PREP O (N+GEN D N+REL), is slightly more complex than it is
presented in the main text. Both these universals are of the form P D (Q O R). Yet the FIH
formulates predictions only for implicational statements of the simple form P D Q. The con-
sequent Q in universals(II1) and (V) is thereforecomplex, consisting of a furtherimplication.
Thus (III) is to be read: 'If a language has prepositions, and if it has N + ADJ, it will also have
N + GEN'; this permits prepositions to co-occur with N +ADJ & N+ GEN, ADJ+ N & N+
GEN, or withADJ + N & GEN + N, but not with *N + ADJ & GEN + N. Derivatively,therefore,
the FIH requiresthat 'If a languageincreasesits prepositions(relativeto postpositions),and if it
increases N+ADJ (relative to ADJ + N), it will also increase N+ GEN (relative to GEN+N,
if any).' Thus, if prepositions increase, it does not necessarily follow that N+ADJ must
increase; but if it does, N+GEN must increase as well (if it stands at less than l007o). And if
N + ADJ does not increase, N + GEN may still increase, or it may not. If prepositions gain in
frequency, therefore, there are three basic possibilities regarding frequency increases in the
noun modifiers; and these possibilities correspond to the three noun-modifier co-occurrences
which are found for basic word orders in languages with prepositions:
(a) Both N+ADJ and N+GEN increase (cf. PREP & N+ADJ & N+GEN languages).
(b) N + GEN increaseswhile N + ADJ does not (cf. PREP & ADJ + N & N + GEN languages).
(c) Neither N + ADJ nor N + GEN increases(cf. PREP & ADJ + N & GEN + N languages).
What is ruled out, therefore, is:
(d) N + ADJ increases, while N + GEN does not (cf. *PREP & N + ADJ & GEN + N).
We can therefore predict, on the basis of the impossibility of the co-occurrence *PREP &
N+ADJ & GEN+ N for basic word orders,that there will be no increasein both prepositions
and N + ADJ in a language which is not matched by an increasein N + GEN. Where, as in the
Avestan data, both PREP and N + ADJ do increasein frequency,N + GEN must also increase.
By parallel reasoning, a complex formula such as P D (Q & R) would predict, by the FlH,
that 'If P increases in frequency, then both Q and R will increase'; and P O (Q v R) would
predict 'If P increasesin frequency, then either Q or R (or both) will increase.'(See also fn. 19.)

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
636 LANGUAGE,VOLUME55, NUMBER 3 (1979)

PrNMH on the basis of further synchronic evidence. The data from Greenberg's
30-language sample, coupled with data from Hsieh 1976, allow us to set up two more
implications for prepositional languages. The first is this:
(VII) If a language has PREP word order, and if the possessive adjective
follows the noun, then the descriptive adjective follows the noun; i.e.,
PREP - (N+POSS v N+ADJ).
Thus prepositional languages have N + POSS & N + ADJ, POSS + N & N + ADJ,
or POSS+N & ADJ+N, but never *N+POSS & ADJ+N. Relevant data are:
(23) PREP - (N+POSS v N+ADJ):
a. N+POSS & N+ADJ: Arabic (VSO & PREP), Hebrew (VSO &
PREP), Indonesian (SVO & PREP), Marshallese (SVO & PREP).
b. POSS+N & N+ADJ: Fijian (V-initial & PREP), Tongan (V-initial
& PREP), Samoan (V-initial & PREP), French (SVO & PREP),
Irish (VSO & PREP).
c. POSS+N & ADJ+N: English (SVO & PREP), Danish (SVO &
PREP), Norwegian (SVO & PREP), Swedish (SVO & PREP).
d. *N+POSS & ADJ+N: No examples.
The other universal implication is this:
(VIII) If a language has PREP word order, and if the demonstrative adjective
follows the noun, then the descriptive adjective follows the noun; i.e.,
PREP - (N+DEM v N+ADJ).
Thus prepositional languages have N+DEM & N+ADJ, DEM+N & N+ADJ,
or DEM+N & ADJ+N, but never *N+DEM & ADJ+N. Relevant data are:
(24) PREP2 (N+DEM D N+ADJ):
a. N+DEM & N+ADJ: Arabic (VSO & PREP), Berber (VSO &
PREP), Fijian (V-initial & PREP), Fulani (SVO & PREP),
Hebrew (VSO & PREP), Indonesian (SVO & PREP), Irish
(VSO & PREP), Ulithian (SVO & PREP), Welsh (VSO & PREP),
Yoruba (SVO & PREP), Zapotec (VSO & PREP), Malay (SVO &
PREP), Maori (VSO & PREP), Marshallese (SVO & PREP),
Rotuman (SVO & PREP), Swahili (SVO & PREP), Takuu (vso/svo
& PREP), Thai (SVO & PREP), Tongan (V-initial & PREP).
b. DEM+N & N+ADJ: Masai (VSO & PREP), French (SVO &
PREP), Spanish (SVO & PREP), Italian (SVO & PREP).
c. DEM+N & ADJ+N: Greek (SVO & PREP), Maya (SVO &
PREP), Serbian (SVO & PREP), English (SVO & PREP), Danish
(SVO & PREP), Norwegian (SVO & PREP), Swedish (SVO &
PREP).
d. *N+DEM & ADJ+N: No examples.

4.3. THEEARLYGERMANIC
DIALECTS.
We shall now examine one language family
in more detail, testing to see whether subsequent developments from the earliest
records are constrained by our eight universalimplications operatingin conjunction

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 637

with the UCH, DAH, and FIH. Of particular importance will be changes in the
noun modifier and adposition orders. The relevant synchronic universals are:
(III) PREP > (N + ADJ D N + GEN)
(V) PREP D (N + GEN N + REL)
(VI) PREP & (VSO v SVO) D not REL + N
(VII) PREP D (N + POSS D N + ADJ)
(VIII) PREP O (N + DEM D N + ADJ)

Our earliest records of Proto-Germanic are from Runic inscriptions dating from
200-600 A.D., which Smith 1971 calls Late Common Germanic; and though the
evidence is sparse,he establishesthe word orderfiguresshown in Table 7 (overleaf).16
Late Common Germanic is universally consistent:
(25) Late Common Germanic (SOV & PREP & N+POSS & N+DEM &
N+ADJ & gen+n/n+gen & N+REL):
(I): SOV v (ADJ+NN GEN+N): no violation.
(III): whether there is violation depends on the interpretation of
gen + n/n + gen.
(V): no violation.
(VII): no violation.
(VIII): no violation.
From Late Common Germanic to Gothic, the next oldest dialect, there are some
interesting changes. Smith gives two sets of figures for Gothic (both taken from the
Skeireins), which are also given in Table 7.17
Compared to Late Common Germanic, Gothic has acquired two new word
orders: ADJ+N as a basic structure, and poss+n as a minority structure. In
addition, Gothic has increasedthe frequency of the following prenominal modifiers:
GEN+N (from LCGmc. 50% to Go. 57%, 77%); ADJ+N (from LCGmc. 07o to
Go. 737O, 83%); DEM+N (from LCGmc. 33% to Go. 100%); and poss+n
(from LCGmc. 0% to Go. 27%, 14%).
Consider these figures in relation to the FIH and universal (III), PREP - (N+
ADJ >N+ GEN). Gothic has PREP; hence universal (III) predicts, in conjunction
with the FIH, that if N + ADJ increases in frequency, so too must N + GEN. In fact,
N+ADJ does not increase in frequency in Gothic, relative to LCGmc., but
GEN + N does. Now, the implication N + ADJ 2 N + GEN is logically equivalent,
by contraposition, to GEN+NN DADJ+N.18 Hence, if GEN+N increases in
frequency, so too must ADJ+ N, and this indeed happens.
16
Although Smith's evidence is limited, the subsequentfiguresfor East and West Germanic,
presented in Tables 7-8 below, do make this state of affairs for LCGmc. very plausible (cf.
fnn. 17, 18).
17 The Skeireins consists of two texts: a late 4th century Bible translation,and a 6th century

commentary. Smith gives two sets of figures-(a), averagedfrom both texts, and (b), from the
commentaryalone. His purpose is to avoid the possible effects of classical influenceon Gothic
word order in translation. However, the differencesbetween the two sets of figures are not
great; and the increased figures for GEN+N and ADJ+N in the commentary are plausibly
accounted for by its later date. The preposingof noun modifiershas continued to a point where
it is more in line with the early West Germanic dialects (except for the possessive adjective).
18The law of contraposition can be represented as follows: (P D Q) _ (-QD -P). In
everyday terms: 'If Fred was eating an apple (P), then Fred was eating a fruit (Q)' is logically
equivalent to: 'If Fred was not eating a fruit (-Q), then Fred was not eating an apple (-P).'

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
638 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

LATE COMMON
GERMANIC GOTHIC
(East Germanic daughter)
(a) (b)
/N + POSS poss + n/N + POSS poss + n/N + POSS
100% 27%7 73% 14%7 86%
dem+n/N+DEM DEM+N/ DEM+N/
33% 67% 100% 100%
/N + ADJ ADJ + N/n + adj ADJ + N/n + adj
100% 73% 27% 83% 17%
gen + n/n + gen GEN + N/n + gen GEN + N/n + gen
50% 50% 57% 43% 77% 23%
/N + RELa /N + REL /N+REL
/PREP /PREP /PREP

SOV/{V } sov/{"jl} sov/{ IV}


TABLE7. Late Common Germanic and Gothic word order data (from Smith 1971). Capital
letters refer to word orders which are basic or unique. Lower-case letters refer to non-basic
orders; v-1 and v-2 designate verb-first and verb-second orders, as defined by Smith. The
criterionfor placing word ordersto the right or left of the slash is the same as for Table 3, above.
a Smith does not
give relative clause and prepositional figures, and the figures for verb
position are not relevant in the present context.
Consider further universals(VII), PREP v (N+ POSS v N + ADJ), and (VIII),
PREP - (N+DEM v N+ADJ). N+POSS v N+ADJ and N+DEM - N+
ADJ are logically equivalent to ADJ + N D POSS + N and ADJ + N D DEM + N
respectively. Since ADJ + N increases in frequency, so must POSS + N and
DEM + N; and they do. Only one nominal modifier fails to develop any prenominal
doublets in Gothic, and that is the relative clause. This is as predicted. Although
universal (V), PREP v (N + GEN v N + REL), could have requiredan increase in
postnominal relatives if postnominal genitives had increased, nothing requires an
increase in or development of prenominal relatives following the increase in pre-
nominal genitives.
The FIH therefore makes accurate predictions for Gothic19. Consider also the
19Note, however, that the Gothic data (like the rest of the Germanicdata to be considered)
differ from the Avestan data in that PREP-the ultimate antecedent of implications(III), (V),
(VII), and (VIII)-does not increasein frequency,although all the predictednoun modifiersdo.
These implicationsare all of the form PD (Q OR), which is logicallyequivalentto P O(-R - Q);
but in each case the ultimate antecedent, P, does not increase. However, in all the Germanic
languages, PREP is already the predominant, if not the exclusive, order (but cf. Early New
High German, ?4.5); hence it cannot reasonably increase further. Since property P cannot
increase further,it is only natural to assume (as I do in the main text) that no furtherincrease
in P is necessaryfor the FIH to be able to make predictionsfor the Q and R structures.To make
this possibility fully explicit, and to illustrate the predictions of the FIH for formulae of the
kind P D (Q D R), let us now redefinethe FIH as follows:
THE FREQUENCY
INCREASEHYPOTHESIS
(FIH):
Given a set of synchronicimplicationaluniversalsof the form P D (Q O R), whereP, Q, and R
are basic word ordersof certain specifiedtypes; then, at two successivestages in the growth of a
language, IF
(a) EITHER thereis an increasein the frequencyof P structuresrelativeto their - P doublets,
OR P is already 100%; and if

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 639

predictions of the DAH. The new doublet ADJ+N (which is already a basic
structure by the time of these figures) does not arise prior to the existence of
structures which are logically consequent upon it: it arises at the same time as the
new doublet poss + n (cf. ADJ + N - POSS + N), and after DEM + N (cf. ADJ + N
> DEM +N), which is already present in LCGmc. (as a doublet); but it does not
arise before either of these implied properties. No implication requires the rise of
prenominal relatives, though prenominal relatives can arise once prenominal
genitives are present; cf. the logical equivalence of N+GEN > N+REL and
REL + N > GEN + N in universal(V). Thus there is no instance in these data of a P
structure arising in the total absence of a Q structure, where P v Q.
There is, however, one anomaly in the Gothic data. Even though the DAH and
FIH correctly predict the rise of a preposed possessive, the postposed possessive is
still in the majority, and this creates an exception to implication (VII), PREP > (N
+ POSS - N + ADJ), since N + POSS exists alongside ADJ + N. This is the first,and
the only, clear counter-example to the UCH in our data; and it is significant that
the DAH and FIH are still being obeyed. The extent of the increase of poss+n
relative to LCGmc. has simply not been sufficient to satisfy the UCH as well.
Smith's figures for the early West Germanic dialects, shown in Table 8, reveal a
development from LCGmc. fundamentally similar to what we have seen for Gothic.
All the prenominal modifiers except the relative clause have again increased in

OLD SAXON OLD HIGH GERMAN OLD ENGLISH


(Heliand, 9th (750-1050 A.D.) Anglo-Saxon Chronicle
century A.D.) (entries for 1035 A.D.)
POSS+ N/n + poss POSS+ N/n + poss POSS+ N/
85% 15% 96% 4% 100%
DEM+N/n+dem DEM+N/ DEM+N/
97% 3% 100% 100%
ADJ+N/n+adj ADJ+N/n+adj ADJ+N/n+adj
83% 17% 85% 15% 99% 1%
GEN+N/n+gen GEN + N/n + gen GEN + N/n + gen
65% 35% 98% 2% 87% 13%
/N+REL /N+REL /N+REL
/PREP /PREP /PREP

SoVI{2} SOV/{2-I} IV{V-2}


TABLE8. Old Saxon, Old High German, and Old English word order data (from Smith 1971).
Capital letters refer to word orders which are basic or unique. Lower-caselettersrefer to non-
basic orders; v-1 and v-2 (or V-2) designate verb-firstand verb-secondorders, as defined by
Smith. The criterion for placing word orders to the right or left of the slash is the same as for
Table 3, above.

(b) there is an increasein the frequencyof Q structuresrelativeto their - Q doublets; and if


(c) the frequencyof the R structuresat the earlierstage, prior to the increasein Q, was less
than 1007%(i.e. if -R doublets existed); THEN
the R structureswill also have gained in frequencyalong with Q (and P) by the later stage.
What is ruled out, therefore, is a frequencyincrease in Q (and in P if it stands at less than
100%) which is not matched by an increasein R. (Recall also fn. 15.)

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
640 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

frequency, as predicted by the FIH and DAH; and POSS+N is now a basic
structure, as predicted by the UCH. Since Gothic is an East Germanic daughter,
while Old Saxon, Old High German, and Old English are West Germanic, we must
compare both groups with their common immediately preceding stage, Late
Common Germanic, rather than with one another.20
4.4. A BRIEFLOOKAT MIDDLEENGLISH.Middle English shows a marked increase
in postnominal modifiers as compared to Old English. According to figures from
Fries 1940, the postnominal genitive becomes basic and is used 84.57%of the time
by 1300 A.D. (compared to 13% in 1035; cf. Table 8.) Postnominal adjectives also
increase, though I lack exact figures. To quote from Lightfoot (1975:205):
'OE adjectival modifiers appeared characteristicallyin prenominal position, particularly
for uncoordinated adjectives and participles. However, in ME postnominal adjectives
become increasinglycommon. In this period most adjectivescan occur as pre- and post-
nominals, and the latter appear to represent the productive position, because all newly
borrowed adjectives are introduced as post-nominals.'21
Recall universals (III), PREP v (N + ADJ D N+ GEN), and (V), PREP > (N+
GEN - N+ REL). Since ME is exclusively prepositional, and since the frequency of
n+adj increases from OE, the FIH and (III) predict that n + gen will increase as
well, which it does. And by the FIH and (V), an increase in n+gen should either
be accompanied by an increase in N+REL, or else this structure must already
be 100%. This is also satisfied.

4.5. A BRIEFLOOKAT EARLYNEW HIGH GERMAN.In the 16th century there are
three parallel developments in the word order of standard (written) German: (1) a
number of postpositions develop alongside the (still predominant) prepositions, (2)
the rigid verb-final rule of the modern language becomes fixed, following an
increasing tendency in Middle High German to move the verb leftward in the
sentence, as compared to Old High German;22 and (3) the language acquires a
non-basic prenominal relative clause. Lehmann 1971 and Weber 1971 document
these changes.
Now consider universal (VI): PREP & (VSO V SVO) v not Rel + N. A language
20 The trend to prepose noun modifiersin the early
Germanicdialects has advanced further
in West Germanic than in East Germanic. This is readily explainable in terms of the later
date of the West Germanicrecords. The figuresfor Gothic are roughly what one would expect,
given the figures for LCGmc. and early West Germanic. This preposing pattern lends further
credibility to the LCGmc. figures, and also to the reliability of the Gothic data. Whatever
classical influenceswere exerted upon Gothic have not disruptedthis development.
21 Importantly,Lightfoot (fn. 9) demonstratesthat Middle
English could not have borrowed
the postnominal adjective order from French (the source of most of the borrowings),for the
simple reason that the great majorityof French adjectivesin this period were prenominal.The
postposing of adjectivesin Middle English is, therefore,internally motivated.
22 The verb-finalrule of Modern German places non-finite
verb forms (infinitives,participles,
particles, and other verbal satellites)at the end of main clauses (the finite verb being in second
position), and both finite and non-finite verb forms at the end of subordinateclauses. During
the Middle High German period, however, prepositional phrases and certain other NP's
(includingsome direct objects)became increasinglyfrequentto the right of these finite and non-
finite verb forms in both main and subordinateclauses, though German never became basic
SVO (cf. Lockwood 1968, ch. XI).

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 641

which has prenominal relative clauses negates the consequent property (not
REL + N) of (VI), and hence cannot satisfy the antecedent by having both PREP
and either VSO or SVO. It may have either POST with VSO or SVO, or PREP with
SOV (i.e. neither VSO nor SVO), or else POST with SOV. Early New High
German develops minority rel + n structures. According to the FIH, therefore, the
acquisition of rel+ n must be accompanied by frequency increases in either post-
positions or SOV, or in both. In fact, both SOV and postpositions increase
simultaneously with prenominal relatives. By the DAH, the new minority structure,
rel+n, must be accompanied or preceded by the acquisition of either POST or
SOV, or both; and this is also fulfilled.
We see, therefore, that the Germanic languages, in their evolution from LCGmc.,
obey much finer predictions derived from synchronic implicational universals than
are incorporated in the UCH alone. The synchronic implications defined in terms
of basic word order patterns can be translated into historical constraints on the
acquisition and frequency increase of doubling structures, many of them minority
word orders.

ARGUMENTS
5. THEORETICAL AGAINSTTHEUVH. Once synchronic implicational
universals are more precisely formulated and tested against the early IE dialects,
we can see that there is practically no support for any universal violation or in-
consistency. This finding removes the cornerstone of Lehmann- and Vennemann-
type trigger-chain theories, according to which the re-introduction of universal
consistency is the primary driving force behind word order change. This lack of
empirical support for the UVH is particularly welcome in light of the following
theoretical objections.
First, language universals are by definition exceptionless, or at least the excep-
tions must be rare. If we invoke an exceptionless synchronic implicational universal
in a diachronic context, we thus have no grounds for assuming the possibility of
violation historically, given that the current synchronic evidence is supposed to be
predicting diachronic developments. But even with statistical universals (as in the
Lehmann-Vennemann theory), the likelihood of a violation arising historically
can presumably be no greater than the (necessarily small) percentage of exceptions
on synchronic grounds. Yet trigger-chain theories assume not just that universal
violation may SOMETIMES occur: violation is their very cornerstone, and hence
must ALWAYSoccur. So the more exceptionless the synchronic universal statements
are, the more implausible trigger-chaintheories become. And even when exceptions
exist, these theories will necessarily still be at variance with the synchronic facts.
Second, trigger-chain theories are internally inconsistent. They assume that
language universals can be both weak and strong constraints on language evolution.
Language universals are sufficiently weak, it seems, to allow an exceptional *P &
-Q co-occurrence to arise at some stage in a language's history. But they can then
become so strongly operative as to effect a total restructuringof the syntax of the
language, by introducing all the Q properties which are implicationally dependent
on P. But this subsequent strength is at variance with the initial weakness in
permitting inconsistency to arise, as is the initial weakness with the subsequent
strength.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
642 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

Third, Lehmannand Vennemann assume a progression of the form - P & - Q >


*P & -Q > P & Q. But the change from -P to P necessarily proceeds via a
doubling stage -P > -P/P > P. As a result, any pull which P (VO) exerts in
favor of the acquisition of Q structures, at the - P/P & - Q stage, will simply be
offset by the parallel implication -P = -Q, radiating from the co-existing (and
initially basic) -P (OV), in favor of the retention of -Q (i.e. the properties of OV
languages). And any attempted increase in the P structure, relative to -P, should
be offset by the implication -Q - P(which is logically equivalent to P Q),
forcing the co-occurrence and hence retention of - P. To claim, therefore, that the
existence of *P & -Q co-occurrences explains the historical rise of positive Q
word orders at the - P/P & - Q stage is to invoke a single implication, -P Q, as a
mechanism of change, when there are two other implications within the same
theory, -P - Q and -Q > - P, which contradict this development and which
force a language to retain existing properties.
Fourth, a crucial ingredient of the Lehmann-Vennemann theory seems to be the
growth of SVO word order. But SVO is typologically ambivalent. In both Green-
berg's implications and mine, VSO and SOV can be implicationally antecedent
properties, but not SVO. Nothing correlates with SVO in a unique and principled
way, and hence SVO has no trigger potential in a historical context: there are no
SVO word order universals to violate.

6. CONCLUSIONS. Four conclusions can be drawn from our comparison of


synchronic and diachronic data.
First, the UVH is to be rejected on empirical and theoretical grounds. The word
order co-occurrences of the earliest IE daughter languages do not violate the
(non-statistical and statistical) implicational universals derivable from current
cross-language synchronic evidence (cf. ??2-3). Hence the alternative theory, the
UCH, is supported. IE word order variation falls squarely within the universal
parameters of variation currently attested in a large sample of languages from
many different families. In addition, the UVH is both theoretically self-con-
tradictory and necessarily at variance with synchronic evidence (cf. ?5).
Second, the assumption of no violation in diachrony still enables us to exploit
synchronic implicational universalsin generatingpredictions for historical change-
predictions which receive strong support from Avestan and Germanic (cf. ?4). In
effect, the DAH makes predictions for the timing of word order changes relative to
one another within a language. Given a synchronic implicational universal 'If P,
then Q' and a language with - P & - Q, I claim that the acquisition of P must occur
either after that of Q (-P &-Q > -P & Q > P & Q), or simultaneously with it
(-P &-Q > P & Q), but not before it, in order to avoid a *P & - Q co-occurrence
(- P & -Q > *P &-Q > P & Q). If P is acquired, therefore, this guarantees
either the prior presence or the simultaneous acquisition of Q.23And by the FIH, if
23
There is an obvious moral here for all implicational universals when considered in a dia-
chronic context. For example, any language which acquires a new relativization strategy, or
which expands (or diminishes) the relativization possibilities of an already existing strategy,
can be expected to conform to the synchronic predictions of Keenan & Comrie's Accessibility
Hierarchy (1977).

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
UNIVERSALSAS PREDICTORS
IMPLICATIONAL 643

P increases in frequency, so must Q, although Q may increase in frequency without


P's doing so.
However, the DAH and FIH make extremely strong claims, and it would not
really surprise me if some counter-examples were found in other language families.
I hope that a suitable modification could still be proposed, preserving the basic
insights from the Avestan and Germanic data-their adherence to the DAH and
FIH appears principled, given the extensive support for the synchronic implica-
tional universals upon which they are based. Our IE data suggest that languages in
transition change from one synchronically-attested set of basic word order co-
occurrences to another; and since both the earlier and the later stages obey the
synchronic universals, it is reasonable that new minor and increasingly frequent
word orders will anticipate the basic word order co-occurrences of the later stage,
by innovating and increasing only those word orders which do not result in *P &
- Q co-occurrences.
Third, the data from early Indo-European match well the implicational pre-
dictions which we have derived from Greenberg's Appendix II. These data fit less
well into Greenberg's VSO, SVO, and SOV trichotomy (cf. the SOV & PREP
languages!), and they do not fit at all well into the Lehmann-Vennemann
VO/OV dichotomy. In fn. 13, I use the implicational universals to reconstruct
prepositions for Proto-Indo-European (not postpositions, cf. Lehmann 1974).
From there I argue on distributional grounds that the chances that PIE was an
SOV language are less than 1 in 10.
Fourth, the evidence against the UVH calls for a new form of explanation for
WHY word orders change, one which is consistent with the UCH. We cannot use
implicational universals to explain the continued momentum for word order
changes following some initial violation. Instead, we must seek motivations for
change within the parameters of synchronically-attested word order variation
types. The data of this paper reveal the complexity of this variation, and the kinds
of implicational statements which are needed to predict it. The success of these
implicational statements in predicting word order change is further evidence for the
reality of the constraints on word order variation which they define, and exposes
the need for a new paradigm of historical explanation. I shall conclude this paper
by briefly outlining the alternative paradigm which I have been developing.

7. TOWARDAN ALTERNATIVE PARADIGM. Two major explanatory questions need


to be answered. Why are the synchronic facts the way they are? In this context,
why do our implicational universals define the range of permissible word order
co-occurrences which they do, rather than any others? And why should a language
change any of its word orders within the permitted co-occurrence types?
In some work still unpublished, I propose some answers to the synchronic
question by explaining why the implicational universals derived from Greenberg's
Appendix II are reasonable. The range of co-occurrences permitted by an impli-
cation such as PREP - (N +ADJ v N + GEN) are explained on semantic and
syntactic grounds. I retain Vennemann's distinction between operator and operand
categories, but I exploit the semantic-syntactic differences between categories such
as adjective and genitive in order to account for the permissibleand non-permissible

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
644 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

co-occurrences. Note in the present context only that any synchronic explanation
for current word order co-occurrences simultaneously explains the constraints on
languages in evolution, given the consistency discovered between synchronic and
diachronic data. Hence, the explanatory principles that underlie our synchronic
implicational universals also underlie the UCH, DAH, and FIH.
The second explanatory question is the more important in this diachronic
context. If the majority of word order changes in a language are not a response to
some initial universal violation, what are they a response to, and why do they
occur?
Note first that there is no shortage of proposals in the literature explaining why
individual word orders may change, e.g. the verb in relation to the direct object
(cf. Lehmann and Vennemann), or the verb in relation to prepositional phrases
(cf. Li & Thompson 1974) and various adverbials (cf. Hyman 1975). Such explana-
tions for verb shift are in principle quite acceptable in this context, though I do not
accept that the new verb position should acquire the status of a trigger, converting
inconsistent word orders into consistent ones. But this now poses a problem, for it
is well known that a change in verb position is generally accompanied (or followed)
by at least some changes within the other categories as well: the noun phrase, the
adjective phrase, and the prepositional or postpositional phrase. The Lehmann-
Vennemann 'chain' of responses to the initial trigger is an attempt to account for
such cross-categorial generality in word order change. Since the trigger-chain
theory is invalid, we now need a new cross-categorial theory. It must be pointed
out, however, that the chain of cross-categorialchanges which is supposed to follow
the shift from SOV to SVO, according to Vennemann's theory, is rarely fulfilled in
its entirety, even after many centuries. For example, English has been a pre-
dominantly VO language since Old English times; yet most of the modifiers of the
noun in Modern English precede the noun, on the OV pattern (only relative clauses
and some genitives follow).
In place of the Lehmann-Vennemann implicational chain, I have developed a
distributional principle which makes more correct predictions concerning the extent
to which word order change involving one category will generalize to another. This
I call the principle of Cross-Category Harmony (CCH). Its primary purpose is to
predict the relative quantities of languages, on current synchronic evidence, which
have the various implicationally-permittedword order co-occurrence arrays. These
quantities vary considerably, as can be seen by comparing the 15 attested types in
Greenberg's Appendix II. It also makes important historical predictions.
CCH presupposes a distinction between modifier (operator) and modified
(operand) categories, essentially though not entirely as in Vennemann's theory.
Determiners, adjectives, and genitives are operators on nouns as operands; objects,
adverbials, and also subjects are operators on verbs as operands; and NP's are
operators on adpositions as operands. But in contrast to Vennemann's natural
serialization principle, CCH does not insist that languages tend to serialize all their
operators on a consistent side of their respective operands (all on the left, or all on
the right). Far too many (in fact, most) languages regularlyhave some operators on
the noun to the left and some to the right, some operators on the verb to the left
and some to the right, and so on. CCH asserts instead that there is a quantifiable

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL
UNIVERSALSAS PREDICTORS 645

preference, across languages, for the ratio of preposed to postposed operators


within one operand category to generalize to the other operand categories. For
example, rigid verb-final languages are generally noun-final within the NP, while
non-rigid verb-final languages (which have at least one operator regularlyfollowing
the verb) are generally not noun-final; e.g., they regularly have the adjective
operator following the noun, thereby matching the operator(s) after the verb.
Conversely, VSO languages are significantly more noun-initial within their NP's
than are SVO languages-in which one operator on the verb, the subject, regularly
precedes the verb, and at least one of the smaller operators on the noun, e.g.
determiners, numerals, and possessive adjectives, regularly precede the noun.
Note also the regular decline in the numbers of attested languages as the adjective
and genitive operators depart from the predominant serializationpattern defined by
the operators on the verb and on the adposition. Universals (I), SOV v (ADJ + N -
GEN+N), and (IV), POST - (ADJ+N v GEN+N), permit the following co-
occurrence possibilities in the types of Greenberg's Appendix II:24
(26) 23. SOV & POST & ADJ+N & GEN+N 29 Igs <L
24. SOV & POST & N+ADJ & GEN+N 24 IgsI
21. SOV & POST & N+ADJ & N+GEN 7 Igs4
-
Universals (II), VSO (N + ADJ - N+GEN), and (III), PREP D (N + ADJ v
N+GEN), permit the following:
(27) 1. VSO & PREP & N+ADJ & N+GEN 19 lgs g
2. VSO & PREP & ADJ?N & N+GEN 5 lgs
3. VSO & PREP & ADJ+N & GEN+N 1lg ,
The same distributional pattern can be seen in prepositional languages with SVO:
(28) 9. SVO & PREP & N+ADJ & N+GEN 20 lgs =
10. SVO &PREP &ADJ+N&N+GEN 81gs
11. SVO & PREP & ADJ+N & GEN+N 2 lgs4,"
As the operators on the noun are increasingly postposed in SOV & POST lan-
guages, and increasingly preposed in VSO and SVO prepositional languages, the
number of languages decreases. This postposing of noun operators in SOV &
POST languages conflicts with the predominant preposing of operators on the verb
and on the adposition, as does the preposing of noun operators in VSO/SVO &
PREP languages with the predominant postposing of operators on the verb and
24 Some of
the entries in Greenbergrefer to language subfamilies or groups, rather than to
individual languages. In calculatingthe language quantitiesfor these co-occurrencecells, I have
had to calculatethe groups as individual languagesonly, since Greenberggives no information
on how many membersof each group he actually checked. For language types 23 and 24, this
results in a disproportionatelyhigh figure for type 24, since type 23 is actually geneticallyand
areally more diverse. The figures for my expanded sample (cf. footnote 6) are approximately:
type 23: 80 lgs [
type 24: 50 lgs c)
type 21: 11 lgs4,
This gives a decline somewhat more in line with the VSO & PREP and SVO & PREP figures.
Otherwise the relative proportions of all language types in the expanded sample are very
similar to those in Greenberg'soriginal sample.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
646 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

adposition. CCH predicts that the more similar the balance in operator preposing
and postposing across the different operand categories, the greater the number of
languages-and the more dissimilar, the fewer the languages. Further details are
presented elsewhere.
The explanation proposed for CCH is twofold. First, I see in it an analogical
preference for like semantic elements-operators and operands-to be treated in a
like manner. The order of the operand in relation to all its operators is preferably
comparable across the different operand categories. Second, I suggest, using some
of the insights of the X theory of generative grammar (cf. Jackendoff 1977), that a
language with a comparable balance of operators to operands across the different
operand categories permits the formulation of more cross-categorial rules of gram-
mar than one with disharmonic orderings. Disharmonic cross-categorial word
orders preclude the collapsing of individual rules into more general cross-categorial
rules, with the result that the quantitatively preferred languages actually have
simpler grammars.
The historical implications of CCH are, I believe, significant. In the word order
co-occurrence preferences defined by CCH, I see a strong internal motive for any
language either to remain within, or to move toward, a preferredtype. Therefore, if
there is an independently-motivated need for, e.g., the verb to shift position, it
is expected that this verb shift will be matched by cross-categorial word order
re-adjustmentsto the extent predicted by CCH, in order to simplify the grammar.
Thus if a rigid SOV language becomes non-rigid, it is expected that at least one
noun operator will simultaneously move to the right of the noun. If a VSO language
becomes SVO, at least one noun operator should simultaneously move to the left of
the noun. Similarly, if a (post-positional) SOV language develops into a (pre-
positional) SVO language, the formerly non-preferred N + ADJ & N + GEN
co-occurrence becomes a highly preferred co-occurrence with SVO & PREP, and
hence a most likely accompanying change (the extent of its likelihood being, in fact,
quantifiable on the basis of the word order preferences on current synchronic
evidence). Thus a change in one operand order should be accompanied by re-
orderings in the other categories. However, to the extent that any re-ordering of
operands within one category is not simultaneously matched by re-orderingsin the
other categories in accordance with quantitative preferences,the language becomes
a non-preferred and more complex type, and is under pressure to re-introduce
cross-category compatibility-the extent of this pressure being again reflected in
the quantities of the relevant languages on current evidence.25
25
Not only do the quantitative word order correlations for currently-attestedlanguages
make predictions for the relative probability of simultaneous, accompanying changes; they
can also predict, we may assume, the relative amounts of time during which any language will
have non-preferredor disharmonic word order co-occurrences in the course of its history.
Thereis scarcelya languagefamily, even in the limitedtime-depthavailableto us for observation
and reconstruction,which does not show at least some word order change; and there are many
language families which reveal extensiveword order re-arrangementsthrough time. Given this,
the decreasingquantitiesof languageswith progressivelydisharmonicword orderco-occurrences
can be seen as a natural consequenceof the fact that every language, in the course of its evolu-
tion, will have had harmonic and preferredword order co-occurrencesfor greater periods of
time than non-preferredones, with length of time correlatingwith degree of CCH. Thus the

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
IMPLICATIONAL UNIVERSALS AS PREDICTORS 647

The general explanatory model of word order change which I would propose can
therefore be set out as follows. All languages in their evolution are constrained by
implicational universals such as have been defined, and can change only relative to
the co-occurrence possibilities which these permit. Some operator-operand re-
orderings, particularly involving the verb, can be explained as responses to gram-
matical changes elsewhere in the grammar. Such categorial changes, however, are
then simultaneously or subsequently matched by cross-categorial word order
re-adjustments, to the extent predicted by CCH, as the syntactic and semantic rules
of the grammar accommodate the word order changes which are independently
motivated for a single category.
This model therefore preserves Vennemann's distinction between operator and
operand categories; but I reject both his natural serialization principle and his
historical trigger-chain theory. Implicational universals are not the appropriate
device with which to explain any chain of word order changes following some
independently-motivated individual change. The main purpose of this paper has
been to argue that implicational universals are preferably exceptionless (or nearly
so), and that languages in evolution can be expected not to violate them. Implica-
tional statements thus define constraints on the timing of changes relative to one
another-'traffic-ways' of change, in effect, within the universally-attested para-
meters of variation. I have suggested instead that current distributional evidence
can predict, with varying degrees of probability, which other word order changes
will accompany or follow independently-motivatedindividual changes, as languages
remain among or move toward the more preferred types.

REFERENCES
ANDERSON,JOHNM., and CHARLES
JONES(eds.) 1974. Historical linguistics. Amsterdam:
North-Holland.
ANTINUCCI,F.; A. DURANTI;and L. GEBERT.1976. Relative clause structure, relative
clause perception,and the change from SOV to SVO. MS.
FRIEDRICH,PAUL. 1975. Proto-Indo-European syntax. (Journal of Indo-European
Studies,monograph1.) Butte: MontanaCollege of MineralScience& Technology.
FRIES,CHARLES
C. 1940. On the development of the structural use of word order in
Modern English Lg. 16.199-208.
GIVON,TALMY.1975. Serial verbs and syntactic change: Niger-Congo. In Li, 49-112.
JOSEPHH. 1966. Some universals of grammar with particular reference to
GREENBERG,
the order of meaningfulelements. Universalsof language, 2nd ed., ed. by J. H.
Greenberg,73-113.
HAWKINS,JOHNA. 1978. Theoretically significant word order patterns in Greenberg's
Appendix II. Universityof Essex, Dept. of Languageand Linguistics,Occasional
papers 20.87-150.

co-occurrencepreferenceson currentsynchronicevidencecan be expected to have a direct time


correlate. In this way, a synchronicsample of languagesat any one point in time will naturally
reflect, in number of languages, the amounts of time during which any one language is respec-
tively harmonic or disharmonic.Among the testablepredictionsmade by such a theory are the
following: the currentlyextremelydisharmoniclanguagesshould not have been in a disharmonic
state for a long period of time; and those members of a language family in evolution which
introduce disharmonyfirst, or to a greaterextent than others, should resolve it soonest. I expect
to present evidence on these matters elsewhere.

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions
648 LANGUAGE, VOLUME 55, NUMBER 3 (1979)

HSIEH,HSIN-I. 1976. Reconstructability and predictability in syntactic change. Paper


read at the Symposium on the Mechanisms of Linguistic Change, Santa Barbara.
HYMAN,LARRYM. 1975. On the change from SOV to SVO: evidence from Niger-
Congo. In Li, 113-47.
JACKENDOFF, RAY. 1977. X syntax: a study of phrase structure. (Linguistic Inquiry,
monograph 2.) Cambridge, MA: MIT Press.
KEENAN,EDWARDL., and B. COMRIE.1977. Noun phrase accessibility and universal
grammar. Linguistic Inquiry 8.63-99.
LEHMANN,WINFREDP. 1971. On the rise of SOV patterns in New High German.
Grammatik, Kybernetik, Kommunikation (Festschrift Alfred Hoppe), ed. by K. G.
Schweisthal, 19-24. Bonn: Dummler.
--. 1972. Proto-Germanic Syntax. Toward a grammar of Proto-Germanic, ed. by
Frans van Coetsem & Herbert L. Kufner, 239-68. Tuibingen: Niemeyer.
--. 1974. Proto-Indo-European syntax. Austin: University of Texas Press.
LI, CHARLES N. (ed.) 1975. Word order and word order change. Austin: University of
Texas Press.
--, and SANDRAA. THOMPSON. 1974. Historical change of word order: a case study
of Chinese and its implications. In Anderson & Jones, 199-218.
--, --. 1975. The semantic function of word order: a case study in Mandarin. In Li,
163-95.
LIGHTFOOT, DAVID.1975. Diachronic syntax: extraposition and deep structure reanaly-
sis. Proceedings of the 5th Annual Meeting, N. E. Linguistic Society, 200-14.
LITTLE,GRETAD. 1977. Linguistic typology and syntactic change in Amharic. Paper
read at the Modern Language Association.
LOCKWOOD, W. B. 1968. Historical German syntax. Oxford: University Press.
PERANTEAU, PAUL M., et al. (eds.) 1972. The Chicago Which Hunt: Papers from the
Relative Clause Festival. Chicago: CLS.
SMITH,JESSEROBERT.1971. Word order in the older Germanic dialects. (University of
Illinois dissertation.) Ann Arbor: University Microfilms.
STEELE, SUSAN.1975. On some factors that affect and effect word order. In Li, 197-268.
VENNEMANN, THEO.1972. Analogy in generative grammar: the origin of word order.
Paper read at the 11th International Congress of Linguists.
. 1974a. Topics, subjects and word order: from SXV to SVX via TVX. In Anderson
& Jones, 339-76.
. 1974b. Theoretical word order studies: results and problems. Papiere zur Lin-
guistik 7.5-25.
-. 1975. An explanation of drift. In Li, 269-305.
WEBER,HEINRICH. 1971. Das erweiterte Adjektiv- und Partizipialattribut im Deutschen.
Munchen: Hueber.
[Received 18 July 1978.]

This content downloaded from 157.193.151.132 on Thu, 27 Mar 2014 09:40:28 AM


All use subject to JSTOR Terms and Conditions

You might also like